kernel_optimize_test/security/selinux/ss/sidtab.c

629 lines
14 KiB
C
Raw Normal View History

License cleanup: add SPDX GPL-2.0 license identifier to files with no license Many source files in the tree are missing licensing information, which makes it harder for compliance tools to determine the correct license. By default all files without license information are under the default license of the kernel, which is GPL version 2. Update the files which contain no license information with the 'GPL-2.0' SPDX license identifier. The SPDX identifier is a legally binding shorthand, which can be used instead of the full boiler plate text. This patch is based on work done by Thomas Gleixner and Kate Stewart and Philippe Ombredanne. How this work was done: Patches were generated and checked against linux-4.14-rc6 for a subset of the use cases: - file had no licensing information it it. - file was a */uapi/* one with no licensing information in it, - file was a */uapi/* one with existing licensing information, Further patches will be generated in subsequent months to fix up cases where non-standard license headers were used, and references to license had to be inferred by heuristics based on keywords. The analysis to determine which SPDX License Identifier to be applied to a file was done in a spreadsheet of side by side results from of the output of two independent scanners (ScanCode & Windriver) producing SPDX tag:value files created by Philippe Ombredanne. Philippe prepared the base worksheet, and did an initial spot review of a few 1000 files. The 4.13 kernel was the starting point of the analysis with 60,537 files assessed. Kate Stewart did a file by file comparison of the scanner results in the spreadsheet to determine which SPDX license identifier(s) to be applied to the file. She confirmed any determination that was not immediately clear with lawyers working with the Linux Foundation. Criteria used to select files for SPDX license identifier tagging was: - Files considered eligible had to be source code files. - Make and config files were included as candidates if they contained >5 lines of source - File already had some variant of a license header in it (even if <5 lines). All documentation files were explicitly excluded. The following heuristics were used to determine which SPDX license identifiers to apply. - when both scanners couldn't find any license traces, file was considered to have no license information in it, and the top level COPYING file license applied. For non */uapi/* files that summary was: SPDX license identifier # files ---------------------------------------------------|------- GPL-2.0 11139 and resulted in the first patch in this series. If that file was a */uapi/* path one, it was "GPL-2.0 WITH Linux-syscall-note" otherwise it was "GPL-2.0". Results of that was: SPDX license identifier # files ---------------------------------------------------|------- GPL-2.0 WITH Linux-syscall-note 930 and resulted in the second patch in this series. - if a file had some form of licensing information in it, and was one of the */uapi/* ones, it was denoted with the Linux-syscall-note if any GPL family license was found in the file or had no licensing in it (per prior point). Results summary: SPDX license identifier # files ---------------------------------------------------|------ GPL-2.0 WITH Linux-syscall-note 270 GPL-2.0+ WITH Linux-syscall-note 169 ((GPL-2.0 WITH Linux-syscall-note) OR BSD-2-Clause) 21 ((GPL-2.0 WITH Linux-syscall-note) OR BSD-3-Clause) 17 LGPL-2.1+ WITH Linux-syscall-note 15 GPL-1.0+ WITH Linux-syscall-note 14 ((GPL-2.0+ WITH Linux-syscall-note) OR BSD-3-Clause) 5 LGPL-2.0+ WITH Linux-syscall-note 4 LGPL-2.1 WITH Linux-syscall-note 3 ((GPL-2.0 WITH Linux-syscall-note) OR MIT) 3 ((GPL-2.0 WITH Linux-syscall-note) AND MIT) 1 and that resulted in the third patch in this series. - when the two scanners agreed on the detected license(s), that became the concluded license(s). - when there was disagreement between the two scanners (one detected a license but the other didn't, or they both detected different licenses) a manual inspection of the file occurred. - In most cases a manual inspection of the information in the file resulted in a clear resolution of the license that should apply (and which scanner probably needed to revisit its heuristics). - When it was not immediately clear, the license identifier was confirmed with lawyers working with the Linux Foundation. - If there was any question as to the appropriate license identifier, the file was flagged for further research and to be revisited later in time. In total, over 70 hours of logged manual review was done on the spreadsheet to determine the SPDX license identifiers to apply to the source files by Kate, Philippe, Thomas and, in some cases, confirmation by lawyers working with the Linux Foundation. Kate also obtained a third independent scan of the 4.13 code base from FOSSology, and compared selected files where the other two scanners disagreed against that SPDX file, to see if there was new insights. The Windriver scanner is based on an older version of FOSSology in part, so they are related. Thomas did random spot checks in about 500 files from the spreadsheets for the uapi headers and agreed with SPDX license identifier in the files he inspected. For the non-uapi files Thomas did random spot checks in about 15000 files. In initial set of patches against 4.14-rc6, 3 files were found to have copy/paste license identifier errors, and have been fixed to reflect the correct identifier. Additionally Philippe spent 10 hours this week doing a detailed manual inspection and review of the 12,461 patched files from the initial patch version early this week with: - a full scancode scan run, collecting the matched texts, detected license ids and scores - reviewing anything where there was a license detected (about 500+ files) to ensure that the applied SPDX license was correct - reviewing anything where there was no detection but the patch license was not GPL-2.0 WITH Linux-syscall-note to ensure that the applied SPDX license was correct This produced a worksheet with 20 files needing minor correction. This worksheet was then exported into 3 different .csv files for the different types of files to be modified. These .csv files were then reviewed by Greg. Thomas wrote a script to parse the csv files and add the proper SPDX tag to the file, in the format that the file expected. This script was further refined by Greg based on the output to detect more types of files automatically and to distinguish between header and source .c files (which need different comment types.) Finally Greg ran the script using the .csv files to generate the patches. Reviewed-by: Kate Stewart <kstewart@linuxfoundation.org> Reviewed-by: Philippe Ombredanne <pombredanne@nexb.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2017-11-01 22:07:57 +08:00
// SPDX-License-Identifier: GPL-2.0
/*
* Implementation of the SID table type.
*
selinux: overhaul sidtab to fix bug and improve performance Before this patch, during a policy reload the sidtab would become frozen and trying to map a new context to SID would be unable to add a new entry to sidtab and fail with -ENOMEM. Such failures are usually propagated into userspace, which has no way of distignuishing them from actual allocation failures and thus doesn't handle them gracefully. Such situation can be triggered e.g. by the following reproducer: while true; do load_policy; echo -n .; sleep 0.1; done & for (( i = 0; i < 1024; i++ )); do runcon -l s0:c$i echo -n x || break # or: # chcon -l s0:c$i <some_file> || break done This patch overhauls the sidtab so it doesn't need to be frozen during policy reload, thus solving the above problem. The new SID table leverages the fact that SIDs are allocated sequentially and are never invalidated and stores them in linear buckets indexed by a tree structure. This brings several advantages: 1. Fast SID -> context lookup - this lookup can now be done in logarithmic time complexity (usually in less than 4 array lookups) and can still be done safely without locking. 2. No need to re-search the whole table on reverse lookup miss - after acquiring the spinlock only the newly added entries need to be searched, which means that reverse lookups that end up inserting a new entry are now about twice as fast. 3. No need to freeze sidtab during policy reload - it is now possible to handle insertion of new entries even during sidtab conversion. The tree structure of the new sidtab is able to grow automatically to up to about 2^31 entries (at which point it should not have more than about 4 tree levels). The old sidtab had a theoretical capacity of almost 2^32 entries, but half of that is still more than enough since by that point the reverse table lookups would become unusably slow anyway... The number of entries per tree node is selected automatically so that each node fits into a single page, which should be the easiest size for kmalloc() to handle. Note that the cache for reverse lookup is preserved with equivalent logic. The only difference is that instead of storing pointers to the hash table nodes it stores just the indices of the cached entries. The new cache ensures that the indices are loaded/stored atomically, but it still has the drawback that concurrent cache updates may mess up the contents of the cache. Such situation however only reduces its effectivity, not the correctness of lookups. Tested by selinux-testsuite and thoroughly tortured by this simple stress test: ``` function rand_cat() { echo $(( $RANDOM % 1024 )) } function do_work() { while true; do echo -n "system_u:system_r:kernel_t:s0:c$(rand_cat),c$(rand_cat)" \ >/sys/fs/selinux/context 2>/dev/null || true done } do_work >/dev/null & do_work >/dev/null & do_work >/dev/null & while load_policy; do echo -n .; sleep 0.1; done kill %1 kill %2 kill %3 ``` Link: https://github.com/SELinuxProject/selinux-kernel/issues/38 Reported-by: Orion Poplawski <orion@nwra.com> Reported-by: Li Kun <hw.likun@huawei.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: most of sidtab.c merged by hand due to conflicts] [PM: checkpatch fixes in mls.c, services.c, sidtab.c] Signed-off-by: Paul Moore <paul@paul-moore.com>
2018-11-30 23:24:08 +08:00
* Original author: Stephen Smalley, <sds@tycho.nsa.gov>
* Author: Ondrej Mosnacek, <omosnacek@gmail.com>
*
* Copyright (C) 2018 Red Hat, Inc.
*/
selinux: overhaul sidtab to fix bug and improve performance Before this patch, during a policy reload the sidtab would become frozen and trying to map a new context to SID would be unable to add a new entry to sidtab and fail with -ENOMEM. Such failures are usually propagated into userspace, which has no way of distignuishing them from actual allocation failures and thus doesn't handle them gracefully. Such situation can be triggered e.g. by the following reproducer: while true; do load_policy; echo -n .; sleep 0.1; done & for (( i = 0; i < 1024; i++ )); do runcon -l s0:c$i echo -n x || break # or: # chcon -l s0:c$i <some_file> || break done This patch overhauls the sidtab so it doesn't need to be frozen during policy reload, thus solving the above problem. The new SID table leverages the fact that SIDs are allocated sequentially and are never invalidated and stores them in linear buckets indexed by a tree structure. This brings several advantages: 1. Fast SID -> context lookup - this lookup can now be done in logarithmic time complexity (usually in less than 4 array lookups) and can still be done safely without locking. 2. No need to re-search the whole table on reverse lookup miss - after acquiring the spinlock only the newly added entries need to be searched, which means that reverse lookups that end up inserting a new entry are now about twice as fast. 3. No need to freeze sidtab during policy reload - it is now possible to handle insertion of new entries even during sidtab conversion. The tree structure of the new sidtab is able to grow automatically to up to about 2^31 entries (at which point it should not have more than about 4 tree levels). The old sidtab had a theoretical capacity of almost 2^32 entries, but half of that is still more than enough since by that point the reverse table lookups would become unusably slow anyway... The number of entries per tree node is selected automatically so that each node fits into a single page, which should be the easiest size for kmalloc() to handle. Note that the cache for reverse lookup is preserved with equivalent logic. The only difference is that instead of storing pointers to the hash table nodes it stores just the indices of the cached entries. The new cache ensures that the indices are loaded/stored atomically, but it still has the drawback that concurrent cache updates may mess up the contents of the cache. Such situation however only reduces its effectivity, not the correctness of lookups. Tested by selinux-testsuite and thoroughly tortured by this simple stress test: ``` function rand_cat() { echo $(( $RANDOM % 1024 )) } function do_work() { while true; do echo -n "system_u:system_r:kernel_t:s0:c$(rand_cat),c$(rand_cat)" \ >/sys/fs/selinux/context 2>/dev/null || true done } do_work >/dev/null & do_work >/dev/null & do_work >/dev/null & while load_policy; do echo -n .; sleep 0.1; done kill %1 kill %2 kill %3 ``` Link: https://github.com/SELinuxProject/selinux-kernel/issues/38 Reported-by: Orion Poplawski <orion@nwra.com> Reported-by: Li Kun <hw.likun@huawei.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: most of sidtab.c merged by hand due to conflicts] [PM: checkpatch fixes in mls.c, services.c, sidtab.c] Signed-off-by: Paul Moore <paul@paul-moore.com>
2018-11-30 23:24:08 +08:00
#include <linux/errno.h>
#include <linux/kernel.h>
selinux: cache the SID -> context string translation Translating a context struct to string can be quite slow, especially if the context has a lot of category bits set. This can cause quite noticeable performance impact in situations where the translation needs to be done repeatedly. A common example is a UNIX datagram socket with the SO_PASSSEC option enabled, which is used e.g. by systemd-journald when receiving log messages via datagram socket. This scenario can be reproduced with: cat /dev/urandom | base64 | logger & timeout 30s perf record -p $(pidof systemd-journald) -a -g kill %1 perf report -g none --pretty raw | grep security_secid_to_secctx Before the caching introduced by this patch, computing the context string (security_secid_to_secctx() function) takes up ~65% of systemd-journald's CPU time (assuming a context with 1024 categories set and Fedora x86_64 release kernel configs). After this patch (assuming near-perfect cache hit ratio) this overhead is reduced to just ~2%. This patch addresses the issue by caching a certain number (compile-time configurable) of recently used context strings to speed up repeated translations of the same context, while using only a small amount of memory. The cache is integrated into the existing sidtab table by adding a field to each entry, which when not NULL contains an RCU-protected pointer to a cache entry containing the cached string. The cache entries are kept in a linked list sorted according to how recently they were used. On a cache miss when the cache is full, the least recently used entry is removed to make space for the new entry. The patch migrates security_sid_to_context_core() to use the cache (also a few other functions where it was possible without too much fuss, but these mostly use the translation for logging in case of error, which is rare). Link: https://bugzilla.redhat.com/show_bug.cgi?id=1733259 Cc: Michal Sekletar <msekleta@redhat.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> Reviewed-by: Paul E. McKenney <paulmck@kernel.org> [PM: lots of merge fixups due to collisions with other sidtab patches] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-26 21:57:00 +08:00
#include <linux/list.h>
#include <linux/rcupdate.h>
#include <linux/slab.h>
selinux: overhaul sidtab to fix bug and improve performance Before this patch, during a policy reload the sidtab would become frozen and trying to map a new context to SID would be unable to add a new entry to sidtab and fail with -ENOMEM. Such failures are usually propagated into userspace, which has no way of distignuishing them from actual allocation failures and thus doesn't handle them gracefully. Such situation can be triggered e.g. by the following reproducer: while true; do load_policy; echo -n .; sleep 0.1; done & for (( i = 0; i < 1024; i++ )); do runcon -l s0:c$i echo -n x || break # or: # chcon -l s0:c$i <some_file> || break done This patch overhauls the sidtab so it doesn't need to be frozen during policy reload, thus solving the above problem. The new SID table leverages the fact that SIDs are allocated sequentially and are never invalidated and stores them in linear buckets indexed by a tree structure. This brings several advantages: 1. Fast SID -> context lookup - this lookup can now be done in logarithmic time complexity (usually in less than 4 array lookups) and can still be done safely without locking. 2. No need to re-search the whole table on reverse lookup miss - after acquiring the spinlock only the newly added entries need to be searched, which means that reverse lookups that end up inserting a new entry are now about twice as fast. 3. No need to freeze sidtab during policy reload - it is now possible to handle insertion of new entries even during sidtab conversion. The tree structure of the new sidtab is able to grow automatically to up to about 2^31 entries (at which point it should not have more than about 4 tree levels). The old sidtab had a theoretical capacity of almost 2^32 entries, but half of that is still more than enough since by that point the reverse table lookups would become unusably slow anyway... The number of entries per tree node is selected automatically so that each node fits into a single page, which should be the easiest size for kmalloc() to handle. Note that the cache for reverse lookup is preserved with equivalent logic. The only difference is that instead of storing pointers to the hash table nodes it stores just the indices of the cached entries. The new cache ensures that the indices are loaded/stored atomically, but it still has the drawback that concurrent cache updates may mess up the contents of the cache. Such situation however only reduces its effectivity, not the correctness of lookups. Tested by selinux-testsuite and thoroughly tortured by this simple stress test: ``` function rand_cat() { echo $(( $RANDOM % 1024 )) } function do_work() { while true; do echo -n "system_u:system_r:kernel_t:s0:c$(rand_cat),c$(rand_cat)" \ >/sys/fs/selinux/context 2>/dev/null || true done } do_work >/dev/null & do_work >/dev/null & do_work >/dev/null & while load_policy; do echo -n .; sleep 0.1; done kill %1 kill %2 kill %3 ``` Link: https://github.com/SELinuxProject/selinux-kernel/issues/38 Reported-by: Orion Poplawski <orion@nwra.com> Reported-by: Li Kun <hw.likun@huawei.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: most of sidtab.c merged by hand due to conflicts] [PM: checkpatch fixes in mls.c, services.c, sidtab.c] Signed-off-by: Paul Moore <paul@paul-moore.com>
2018-11-30 23:24:08 +08:00
#include <linux/sched.h>
#include <linux/spinlock.h>
#include <asm/barrier.h>
#include "flask.h"
#include "security.h"
#include "sidtab.h"
selinux: cache the SID -> context string translation Translating a context struct to string can be quite slow, especially if the context has a lot of category bits set. This can cause quite noticeable performance impact in situations where the translation needs to be done repeatedly. A common example is a UNIX datagram socket with the SO_PASSSEC option enabled, which is used e.g. by systemd-journald when receiving log messages via datagram socket. This scenario can be reproduced with: cat /dev/urandom | base64 | logger & timeout 30s perf record -p $(pidof systemd-journald) -a -g kill %1 perf report -g none --pretty raw | grep security_secid_to_secctx Before the caching introduced by this patch, computing the context string (security_secid_to_secctx() function) takes up ~65% of systemd-journald's CPU time (assuming a context with 1024 categories set and Fedora x86_64 release kernel configs). After this patch (assuming near-perfect cache hit ratio) this overhead is reduced to just ~2%. This patch addresses the issue by caching a certain number (compile-time configurable) of recently used context strings to speed up repeated translations of the same context, while using only a small amount of memory. The cache is integrated into the existing sidtab table by adding a field to each entry, which when not NULL contains an RCU-protected pointer to a cache entry containing the cached string. The cache entries are kept in a linked list sorted according to how recently they were used. On a cache miss when the cache is full, the least recently used entry is removed to make space for the new entry. The patch migrates security_sid_to_context_core() to use the cache (also a few other functions where it was possible without too much fuss, but these mostly use the translation for logging in case of error, which is rare). Link: https://bugzilla.redhat.com/show_bug.cgi?id=1733259 Cc: Michal Sekletar <msekleta@redhat.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> Reviewed-by: Paul E. McKenney <paulmck@kernel.org> [PM: lots of merge fixups due to collisions with other sidtab patches] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-26 21:57:00 +08:00
struct sidtab_str_cache {
struct rcu_head rcu_member;
struct list_head lru_member;
struct sidtab_entry *parent;
u32 len;
char str[];
};
selinux: sidtab reverse lookup hash table This replaces the reverse table lookup and reverse cache with a hashtable which improves cache-miss reverse-lookup times from O(n) to O(1)* and maintains the same performance as a reverse cache hit. This reduces the time needed to add a new sidtab entry from ~500us to 5us on a Pixel 3 when there are ~10,000 sidtab entries. The implementation uses the kernel's generic hashtable API, It uses the context's string represtation as the hash source, and the kernels generic string hashing algorithm full_name_hash() to reduce the string to a 32 bit value. This change also maintains the improvement introduced in commit ee1a84fdfeed ("selinux: overhaul sidtab to fix bug and improve performance") which removed the need to keep the current sidtab locked during policy reload. It does however introduce periodic locking of the target sidtab while converting the hashtable. Sidtab entries are never modified or removed, so the context struct stored in the sid_to_context tree can also be used for the context_to_sid hashtable to reduce memory usage. This bug was reported by: - On the selinux bug tracker. BUG: kernel softlockup due to too many SIDs/contexts #37 https://github.com/SELinuxProject/selinux-kernel/issues/37 - Jovana Knezevic on Android's bugtracker. Bug: 140252993 "During multi-user performance testing, we create and remove users many times. selinux_android_restorecon_pkgdir goes from 1ms to over 20ms after about 200 user creations and removals. Accumulated over ~280 packages, that adds a significant time to user creation, making perf benchmarks unreliable." * Hashtable lookup is only O(1) when n < the number of buckets. Signed-off-by: Jeff Vander Stoep <jeffv@google.com> Reported-by: Stephen Smalley <sds@tycho.nsa.gov> Reported-by: Jovana Knezevic <jovanak@google.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: subj tweak, removed changelog from patch description] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-22 17:33:06 +08:00
#define index_to_sid(index) (index + SECINITSID_NUM + 1)
#define sid_to_index(sid) (sid - (SECINITSID_NUM + 1))
int sidtab_init(struct sidtab *s)
{
selinux: overhaul sidtab to fix bug and improve performance Before this patch, during a policy reload the sidtab would become frozen and trying to map a new context to SID would be unable to add a new entry to sidtab and fail with -ENOMEM. Such failures are usually propagated into userspace, which has no way of distignuishing them from actual allocation failures and thus doesn't handle them gracefully. Such situation can be triggered e.g. by the following reproducer: while true; do load_policy; echo -n .; sleep 0.1; done & for (( i = 0; i < 1024; i++ )); do runcon -l s0:c$i echo -n x || break # or: # chcon -l s0:c$i <some_file> || break done This patch overhauls the sidtab so it doesn't need to be frozen during policy reload, thus solving the above problem. The new SID table leverages the fact that SIDs are allocated sequentially and are never invalidated and stores them in linear buckets indexed by a tree structure. This brings several advantages: 1. Fast SID -> context lookup - this lookup can now be done in logarithmic time complexity (usually in less than 4 array lookups) and can still be done safely without locking. 2. No need to re-search the whole table on reverse lookup miss - after acquiring the spinlock only the newly added entries need to be searched, which means that reverse lookups that end up inserting a new entry are now about twice as fast. 3. No need to freeze sidtab during policy reload - it is now possible to handle insertion of new entries even during sidtab conversion. The tree structure of the new sidtab is able to grow automatically to up to about 2^31 entries (at which point it should not have more than about 4 tree levels). The old sidtab had a theoretical capacity of almost 2^32 entries, but half of that is still more than enough since by that point the reverse table lookups would become unusably slow anyway... The number of entries per tree node is selected automatically so that each node fits into a single page, which should be the easiest size for kmalloc() to handle. Note that the cache for reverse lookup is preserved with equivalent logic. The only difference is that instead of storing pointers to the hash table nodes it stores just the indices of the cached entries. The new cache ensures that the indices are loaded/stored atomically, but it still has the drawback that concurrent cache updates may mess up the contents of the cache. Such situation however only reduces its effectivity, not the correctness of lookups. Tested by selinux-testsuite and thoroughly tortured by this simple stress test: ``` function rand_cat() { echo $(( $RANDOM % 1024 )) } function do_work() { while true; do echo -n "system_u:system_r:kernel_t:s0:c$(rand_cat),c$(rand_cat)" \ >/sys/fs/selinux/context 2>/dev/null || true done } do_work >/dev/null & do_work >/dev/null & do_work >/dev/null & while load_policy; do echo -n .; sleep 0.1; done kill %1 kill %2 kill %3 ``` Link: https://github.com/SELinuxProject/selinux-kernel/issues/38 Reported-by: Orion Poplawski <orion@nwra.com> Reported-by: Li Kun <hw.likun@huawei.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: most of sidtab.c merged by hand due to conflicts] [PM: checkpatch fixes in mls.c, services.c, sidtab.c] Signed-off-by: Paul Moore <paul@paul-moore.com>
2018-11-30 23:24:08 +08:00
u32 i;
selinux: overhaul sidtab to fix bug and improve performance Before this patch, during a policy reload the sidtab would become frozen and trying to map a new context to SID would be unable to add a new entry to sidtab and fail with -ENOMEM. Such failures are usually propagated into userspace, which has no way of distignuishing them from actual allocation failures and thus doesn't handle them gracefully. Such situation can be triggered e.g. by the following reproducer: while true; do load_policy; echo -n .; sleep 0.1; done & for (( i = 0; i < 1024; i++ )); do runcon -l s0:c$i echo -n x || break # or: # chcon -l s0:c$i <some_file> || break done This patch overhauls the sidtab so it doesn't need to be frozen during policy reload, thus solving the above problem. The new SID table leverages the fact that SIDs are allocated sequentially and are never invalidated and stores them in linear buckets indexed by a tree structure. This brings several advantages: 1. Fast SID -> context lookup - this lookup can now be done in logarithmic time complexity (usually in less than 4 array lookups) and can still be done safely without locking. 2. No need to re-search the whole table on reverse lookup miss - after acquiring the spinlock only the newly added entries need to be searched, which means that reverse lookups that end up inserting a new entry are now about twice as fast. 3. No need to freeze sidtab during policy reload - it is now possible to handle insertion of new entries even during sidtab conversion. The tree structure of the new sidtab is able to grow automatically to up to about 2^31 entries (at which point it should not have more than about 4 tree levels). The old sidtab had a theoretical capacity of almost 2^32 entries, but half of that is still more than enough since by that point the reverse table lookups would become unusably slow anyway... The number of entries per tree node is selected automatically so that each node fits into a single page, which should be the easiest size for kmalloc() to handle. Note that the cache for reverse lookup is preserved with equivalent logic. The only difference is that instead of storing pointers to the hash table nodes it stores just the indices of the cached entries. The new cache ensures that the indices are loaded/stored atomically, but it still has the drawback that concurrent cache updates may mess up the contents of the cache. Such situation however only reduces its effectivity, not the correctness of lookups. Tested by selinux-testsuite and thoroughly tortured by this simple stress test: ``` function rand_cat() { echo $(( $RANDOM % 1024 )) } function do_work() { while true; do echo -n "system_u:system_r:kernel_t:s0:c$(rand_cat),c$(rand_cat)" \ >/sys/fs/selinux/context 2>/dev/null || true done } do_work >/dev/null & do_work >/dev/null & do_work >/dev/null & while load_policy; do echo -n .; sleep 0.1; done kill %1 kill %2 kill %3 ``` Link: https://github.com/SELinuxProject/selinux-kernel/issues/38 Reported-by: Orion Poplawski <orion@nwra.com> Reported-by: Li Kun <hw.likun@huawei.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: most of sidtab.c merged by hand due to conflicts] [PM: checkpatch fixes in mls.c, services.c, sidtab.c] Signed-off-by: Paul Moore <paul@paul-moore.com>
2018-11-30 23:24:08 +08:00
memset(s->roots, 0, sizeof(s->roots));
for (i = 0; i < SECINITSID_NUM; i++)
s->isids[i].set = 0;
selinux: fix race between old and new sidtab commit 9ad6e9cb39c66366bf7b9aece114aca277981a1f upstream. Since commit 1b8b31a2e612 ("selinux: convert policy read-write lock to RCU"), there is a small window during policy load where the new policy pointer has already been installed, but some threads may still be holding the old policy pointer in their read-side RCU critical sections. This means that there may be conflicting attempts to add a new SID entry to both tables via sidtab_context_to_sid(). See also (and the rest of the thread): https://lore.kernel.org/selinux/CAFqZXNvfux46_f8gnvVvRYMKoes24nwm2n3sPbMjrB8vKTW00g@mail.gmail.com/ Fix this by installing the new policy pointer under the old sidtab's spinlock along with marking the old sidtab as "frozen". Then, if an attempt to add new entry to a "frozen" sidtab is detected, make sidtab_context_to_sid() return -ESTALE to indicate that a new policy has been installed and that the caller will have to abort the policy transaction and try again after re-taking the policy pointer (which is guaranteed to be a newer policy). This requires adding a retry-on-ESTALE logic to all callers of sidtab_context_to_sid(), but fortunately these are easy to determine and aren't that many. This seems to be the simplest solution for this problem, even if it looks somewhat ugly. Note that other places in the kernel (e.g. do_mknodat() in fs/namei.c) use similar stale-retry patterns, so I think it's reasonable. Cc: stable@vger.kernel.org Fixes: 1b8b31a2e612 ("selinux: convert policy read-write lock to RCU") Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Signed-off-by: Paul Moore <paul@paul-moore.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2021-04-07 15:24:43 +08:00
s->frozen = false;
s->count = 0;
selinux: overhaul sidtab to fix bug and improve performance Before this patch, during a policy reload the sidtab would become frozen and trying to map a new context to SID would be unable to add a new entry to sidtab and fail with -ENOMEM. Such failures are usually propagated into userspace, which has no way of distignuishing them from actual allocation failures and thus doesn't handle them gracefully. Such situation can be triggered e.g. by the following reproducer: while true; do load_policy; echo -n .; sleep 0.1; done & for (( i = 0; i < 1024; i++ )); do runcon -l s0:c$i echo -n x || break # or: # chcon -l s0:c$i <some_file> || break done This patch overhauls the sidtab so it doesn't need to be frozen during policy reload, thus solving the above problem. The new SID table leverages the fact that SIDs are allocated sequentially and are never invalidated and stores them in linear buckets indexed by a tree structure. This brings several advantages: 1. Fast SID -> context lookup - this lookup can now be done in logarithmic time complexity (usually in less than 4 array lookups) and can still be done safely without locking. 2. No need to re-search the whole table on reverse lookup miss - after acquiring the spinlock only the newly added entries need to be searched, which means that reverse lookups that end up inserting a new entry are now about twice as fast. 3. No need to freeze sidtab during policy reload - it is now possible to handle insertion of new entries even during sidtab conversion. The tree structure of the new sidtab is able to grow automatically to up to about 2^31 entries (at which point it should not have more than about 4 tree levels). The old sidtab had a theoretical capacity of almost 2^32 entries, but half of that is still more than enough since by that point the reverse table lookups would become unusably slow anyway... The number of entries per tree node is selected automatically so that each node fits into a single page, which should be the easiest size for kmalloc() to handle. Note that the cache for reverse lookup is preserved with equivalent logic. The only difference is that instead of storing pointers to the hash table nodes it stores just the indices of the cached entries. The new cache ensures that the indices are loaded/stored atomically, but it still has the drawback that concurrent cache updates may mess up the contents of the cache. Such situation however only reduces its effectivity, not the correctness of lookups. Tested by selinux-testsuite and thoroughly tortured by this simple stress test: ``` function rand_cat() { echo $(( $RANDOM % 1024 )) } function do_work() { while true; do echo -n "system_u:system_r:kernel_t:s0:c$(rand_cat),c$(rand_cat)" \ >/sys/fs/selinux/context 2>/dev/null || true done } do_work >/dev/null & do_work >/dev/null & do_work >/dev/null & while load_policy; do echo -n .; sleep 0.1; done kill %1 kill %2 kill %3 ``` Link: https://github.com/SELinuxProject/selinux-kernel/issues/38 Reported-by: Orion Poplawski <orion@nwra.com> Reported-by: Li Kun <hw.likun@huawei.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: most of sidtab.c merged by hand due to conflicts] [PM: checkpatch fixes in mls.c, services.c, sidtab.c] Signed-off-by: Paul Moore <paul@paul-moore.com>
2018-11-30 23:24:08 +08:00
s->convert = NULL;
selinux: sidtab reverse lookup hash table This replaces the reverse table lookup and reverse cache with a hashtable which improves cache-miss reverse-lookup times from O(n) to O(1)* and maintains the same performance as a reverse cache hit. This reduces the time needed to add a new sidtab entry from ~500us to 5us on a Pixel 3 when there are ~10,000 sidtab entries. The implementation uses the kernel's generic hashtable API, It uses the context's string represtation as the hash source, and the kernels generic string hashing algorithm full_name_hash() to reduce the string to a 32 bit value. This change also maintains the improvement introduced in commit ee1a84fdfeed ("selinux: overhaul sidtab to fix bug and improve performance") which removed the need to keep the current sidtab locked during policy reload. It does however introduce periodic locking of the target sidtab while converting the hashtable. Sidtab entries are never modified or removed, so the context struct stored in the sid_to_context tree can also be used for the context_to_sid hashtable to reduce memory usage. This bug was reported by: - On the selinux bug tracker. BUG: kernel softlockup due to too many SIDs/contexts #37 https://github.com/SELinuxProject/selinux-kernel/issues/37 - Jovana Knezevic on Android's bugtracker. Bug: 140252993 "During multi-user performance testing, we create and remove users many times. selinux_android_restorecon_pkgdir goes from 1ms to over 20ms after about 200 user creations and removals. Accumulated over ~280 packages, that adds a significant time to user creation, making perf benchmarks unreliable." * Hashtable lookup is only O(1) when n < the number of buckets. Signed-off-by: Jeff Vander Stoep <jeffv@google.com> Reported-by: Stephen Smalley <sds@tycho.nsa.gov> Reported-by: Jovana Knezevic <jovanak@google.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: subj tweak, removed changelog from patch description] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-22 17:33:06 +08:00
hash_init(s->context_to_sid);
spin_lock_init(&s->lock);
selinux: cache the SID -> context string translation Translating a context struct to string can be quite slow, especially if the context has a lot of category bits set. This can cause quite noticeable performance impact in situations where the translation needs to be done repeatedly. A common example is a UNIX datagram socket with the SO_PASSSEC option enabled, which is used e.g. by systemd-journald when receiving log messages via datagram socket. This scenario can be reproduced with: cat /dev/urandom | base64 | logger & timeout 30s perf record -p $(pidof systemd-journald) -a -g kill %1 perf report -g none --pretty raw | grep security_secid_to_secctx Before the caching introduced by this patch, computing the context string (security_secid_to_secctx() function) takes up ~65% of systemd-journald's CPU time (assuming a context with 1024 categories set and Fedora x86_64 release kernel configs). After this patch (assuming near-perfect cache hit ratio) this overhead is reduced to just ~2%. This patch addresses the issue by caching a certain number (compile-time configurable) of recently used context strings to speed up repeated translations of the same context, while using only a small amount of memory. The cache is integrated into the existing sidtab table by adding a field to each entry, which when not NULL contains an RCU-protected pointer to a cache entry containing the cached string. The cache entries are kept in a linked list sorted according to how recently they were used. On a cache miss when the cache is full, the least recently used entry is removed to make space for the new entry. The patch migrates security_sid_to_context_core() to use the cache (also a few other functions where it was possible without too much fuss, but these mostly use the translation for logging in case of error, which is rare). Link: https://bugzilla.redhat.com/show_bug.cgi?id=1733259 Cc: Michal Sekletar <msekleta@redhat.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> Reviewed-by: Paul E. McKenney <paulmck@kernel.org> [PM: lots of merge fixups due to collisions with other sidtab patches] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-26 21:57:00 +08:00
#if CONFIG_SECURITY_SELINUX_SID2STR_CACHE_SIZE > 0
s->cache_free_slots = CONFIG_SECURITY_SELINUX_SID2STR_CACHE_SIZE;
INIT_LIST_HEAD(&s->cache_lru_list);
spin_lock_init(&s->cache_lock);
#endif
return 0;
}
static u32 context_to_sid(struct sidtab *s, struct context *context, u32 hash)
selinux: sidtab reverse lookup hash table This replaces the reverse table lookup and reverse cache with a hashtable which improves cache-miss reverse-lookup times from O(n) to O(1)* and maintains the same performance as a reverse cache hit. This reduces the time needed to add a new sidtab entry from ~500us to 5us on a Pixel 3 when there are ~10,000 sidtab entries. The implementation uses the kernel's generic hashtable API, It uses the context's string represtation as the hash source, and the kernels generic string hashing algorithm full_name_hash() to reduce the string to a 32 bit value. This change also maintains the improvement introduced in commit ee1a84fdfeed ("selinux: overhaul sidtab to fix bug and improve performance") which removed the need to keep the current sidtab locked during policy reload. It does however introduce periodic locking of the target sidtab while converting the hashtable. Sidtab entries are never modified or removed, so the context struct stored in the sid_to_context tree can also be used for the context_to_sid hashtable to reduce memory usage. This bug was reported by: - On the selinux bug tracker. BUG: kernel softlockup due to too many SIDs/contexts #37 https://github.com/SELinuxProject/selinux-kernel/issues/37 - Jovana Knezevic on Android's bugtracker. Bug: 140252993 "During multi-user performance testing, we create and remove users many times. selinux_android_restorecon_pkgdir goes from 1ms to over 20ms after about 200 user creations and removals. Accumulated over ~280 packages, that adds a significant time to user creation, making perf benchmarks unreliable." * Hashtable lookup is only O(1) when n < the number of buckets. Signed-off-by: Jeff Vander Stoep <jeffv@google.com> Reported-by: Stephen Smalley <sds@tycho.nsa.gov> Reported-by: Jovana Knezevic <jovanak@google.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: subj tweak, removed changelog from patch description] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-22 17:33:06 +08:00
{
selinux: cache the SID -> context string translation Translating a context struct to string can be quite slow, especially if the context has a lot of category bits set. This can cause quite noticeable performance impact in situations where the translation needs to be done repeatedly. A common example is a UNIX datagram socket with the SO_PASSSEC option enabled, which is used e.g. by systemd-journald when receiving log messages via datagram socket. This scenario can be reproduced with: cat /dev/urandom | base64 | logger & timeout 30s perf record -p $(pidof systemd-journald) -a -g kill %1 perf report -g none --pretty raw | grep security_secid_to_secctx Before the caching introduced by this patch, computing the context string (security_secid_to_secctx() function) takes up ~65% of systemd-journald's CPU time (assuming a context with 1024 categories set and Fedora x86_64 release kernel configs). After this patch (assuming near-perfect cache hit ratio) this overhead is reduced to just ~2%. This patch addresses the issue by caching a certain number (compile-time configurable) of recently used context strings to speed up repeated translations of the same context, while using only a small amount of memory. The cache is integrated into the existing sidtab table by adding a field to each entry, which when not NULL contains an RCU-protected pointer to a cache entry containing the cached string. The cache entries are kept in a linked list sorted according to how recently they were used. On a cache miss when the cache is full, the least recently used entry is removed to make space for the new entry. The patch migrates security_sid_to_context_core() to use the cache (also a few other functions where it was possible without too much fuss, but these mostly use the translation for logging in case of error, which is rare). Link: https://bugzilla.redhat.com/show_bug.cgi?id=1733259 Cc: Michal Sekletar <msekleta@redhat.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> Reviewed-by: Paul E. McKenney <paulmck@kernel.org> [PM: lots of merge fixups due to collisions with other sidtab patches] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-26 21:57:00 +08:00
struct sidtab_entry *entry;
selinux: sidtab reverse lookup hash table This replaces the reverse table lookup and reverse cache with a hashtable which improves cache-miss reverse-lookup times from O(n) to O(1)* and maintains the same performance as a reverse cache hit. This reduces the time needed to add a new sidtab entry from ~500us to 5us on a Pixel 3 when there are ~10,000 sidtab entries. The implementation uses the kernel's generic hashtable API, It uses the context's string represtation as the hash source, and the kernels generic string hashing algorithm full_name_hash() to reduce the string to a 32 bit value. This change also maintains the improvement introduced in commit ee1a84fdfeed ("selinux: overhaul sidtab to fix bug and improve performance") which removed the need to keep the current sidtab locked during policy reload. It does however introduce periodic locking of the target sidtab while converting the hashtable. Sidtab entries are never modified or removed, so the context struct stored in the sid_to_context tree can also be used for the context_to_sid hashtable to reduce memory usage. This bug was reported by: - On the selinux bug tracker. BUG: kernel softlockup due to too many SIDs/contexts #37 https://github.com/SELinuxProject/selinux-kernel/issues/37 - Jovana Knezevic on Android's bugtracker. Bug: 140252993 "During multi-user performance testing, we create and remove users many times. selinux_android_restorecon_pkgdir goes from 1ms to over 20ms after about 200 user creations and removals. Accumulated over ~280 packages, that adds a significant time to user creation, making perf benchmarks unreliable." * Hashtable lookup is only O(1) when n < the number of buckets. Signed-off-by: Jeff Vander Stoep <jeffv@google.com> Reported-by: Stephen Smalley <sds@tycho.nsa.gov> Reported-by: Jovana Knezevic <jovanak@google.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: subj tweak, removed changelog from patch description] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-22 17:33:06 +08:00
u32 sid = 0;
rcu_read_lock();
hash_for_each_possible_rcu(s->context_to_sid, entry, list, hash) {
if (entry->hash != hash)
continue;
selinux: sidtab reverse lookup hash table This replaces the reverse table lookup and reverse cache with a hashtable which improves cache-miss reverse-lookup times from O(n) to O(1)* and maintains the same performance as a reverse cache hit. This reduces the time needed to add a new sidtab entry from ~500us to 5us on a Pixel 3 when there are ~10,000 sidtab entries. The implementation uses the kernel's generic hashtable API, It uses the context's string represtation as the hash source, and the kernels generic string hashing algorithm full_name_hash() to reduce the string to a 32 bit value. This change also maintains the improvement introduced in commit ee1a84fdfeed ("selinux: overhaul sidtab to fix bug and improve performance") which removed the need to keep the current sidtab locked during policy reload. It does however introduce periodic locking of the target sidtab while converting the hashtable. Sidtab entries are never modified or removed, so the context struct stored in the sid_to_context tree can also be used for the context_to_sid hashtable to reduce memory usage. This bug was reported by: - On the selinux bug tracker. BUG: kernel softlockup due to too many SIDs/contexts #37 https://github.com/SELinuxProject/selinux-kernel/issues/37 - Jovana Knezevic on Android's bugtracker. Bug: 140252993 "During multi-user performance testing, we create and remove users many times. selinux_android_restorecon_pkgdir goes from 1ms to over 20ms after about 200 user creations and removals. Accumulated over ~280 packages, that adds a significant time to user creation, making perf benchmarks unreliable." * Hashtable lookup is only O(1) when n < the number of buckets. Signed-off-by: Jeff Vander Stoep <jeffv@google.com> Reported-by: Stephen Smalley <sds@tycho.nsa.gov> Reported-by: Jovana Knezevic <jovanak@google.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: subj tweak, removed changelog from patch description] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-22 17:33:06 +08:00
if (context_cmp(&entry->context, context)) {
sid = entry->sid;
break;
}
}
rcu_read_unlock();
return sid;
}
int sidtab_set_initial(struct sidtab *s, u32 sid, struct context *context)
{
selinux: cache the SID -> context string translation Translating a context struct to string can be quite slow, especially if the context has a lot of category bits set. This can cause quite noticeable performance impact in situations where the translation needs to be done repeatedly. A common example is a UNIX datagram socket with the SO_PASSSEC option enabled, which is used e.g. by systemd-journald when receiving log messages via datagram socket. This scenario can be reproduced with: cat /dev/urandom | base64 | logger & timeout 30s perf record -p $(pidof systemd-journald) -a -g kill %1 perf report -g none --pretty raw | grep security_secid_to_secctx Before the caching introduced by this patch, computing the context string (security_secid_to_secctx() function) takes up ~65% of systemd-journald's CPU time (assuming a context with 1024 categories set and Fedora x86_64 release kernel configs). After this patch (assuming near-perfect cache hit ratio) this overhead is reduced to just ~2%. This patch addresses the issue by caching a certain number (compile-time configurable) of recently used context strings to speed up repeated translations of the same context, while using only a small amount of memory. The cache is integrated into the existing sidtab table by adding a field to each entry, which when not NULL contains an RCU-protected pointer to a cache entry containing the cached string. The cache entries are kept in a linked list sorted according to how recently they were used. On a cache miss when the cache is full, the least recently used entry is removed to make space for the new entry. The patch migrates security_sid_to_context_core() to use the cache (also a few other functions where it was possible without too much fuss, but these mostly use the translation for logging in case of error, which is rare). Link: https://bugzilla.redhat.com/show_bug.cgi?id=1733259 Cc: Michal Sekletar <msekleta@redhat.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> Reviewed-by: Paul E. McKenney <paulmck@kernel.org> [PM: lots of merge fixups due to collisions with other sidtab patches] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-26 21:57:00 +08:00
struct sidtab_isid_entry *isid;
u32 hash;
int rc;
if (sid == 0 || sid > SECINITSID_NUM)
return -EINVAL;
selinux: cache the SID -> context string translation Translating a context struct to string can be quite slow, especially if the context has a lot of category bits set. This can cause quite noticeable performance impact in situations where the translation needs to be done repeatedly. A common example is a UNIX datagram socket with the SO_PASSSEC option enabled, which is used e.g. by systemd-journald when receiving log messages via datagram socket. This scenario can be reproduced with: cat /dev/urandom | base64 | logger & timeout 30s perf record -p $(pidof systemd-journald) -a -g kill %1 perf report -g none --pretty raw | grep security_secid_to_secctx Before the caching introduced by this patch, computing the context string (security_secid_to_secctx() function) takes up ~65% of systemd-journald's CPU time (assuming a context with 1024 categories set and Fedora x86_64 release kernel configs). After this patch (assuming near-perfect cache hit ratio) this overhead is reduced to just ~2%. This patch addresses the issue by caching a certain number (compile-time configurable) of recently used context strings to speed up repeated translations of the same context, while using only a small amount of memory. The cache is integrated into the existing sidtab table by adding a field to each entry, which when not NULL contains an RCU-protected pointer to a cache entry containing the cached string. The cache entries are kept in a linked list sorted according to how recently they were used. On a cache miss when the cache is full, the least recently used entry is removed to make space for the new entry. The patch migrates security_sid_to_context_core() to use the cache (also a few other functions where it was possible without too much fuss, but these mostly use the translation for logging in case of error, which is rare). Link: https://bugzilla.redhat.com/show_bug.cgi?id=1733259 Cc: Michal Sekletar <msekleta@redhat.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> Reviewed-by: Paul E. McKenney <paulmck@kernel.org> [PM: lots of merge fixups due to collisions with other sidtab patches] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-26 21:57:00 +08:00
isid = &s->isids[sid - 1];
selinux: cache the SID -> context string translation Translating a context struct to string can be quite slow, especially if the context has a lot of category bits set. This can cause quite noticeable performance impact in situations where the translation needs to be done repeatedly. A common example is a UNIX datagram socket with the SO_PASSSEC option enabled, which is used e.g. by systemd-journald when receiving log messages via datagram socket. This scenario can be reproduced with: cat /dev/urandom | base64 | logger & timeout 30s perf record -p $(pidof systemd-journald) -a -g kill %1 perf report -g none --pretty raw | grep security_secid_to_secctx Before the caching introduced by this patch, computing the context string (security_secid_to_secctx() function) takes up ~65% of systemd-journald's CPU time (assuming a context with 1024 categories set and Fedora x86_64 release kernel configs). After this patch (assuming near-perfect cache hit ratio) this overhead is reduced to just ~2%. This patch addresses the issue by caching a certain number (compile-time configurable) of recently used context strings to speed up repeated translations of the same context, while using only a small amount of memory. The cache is integrated into the existing sidtab table by adding a field to each entry, which when not NULL contains an RCU-protected pointer to a cache entry containing the cached string. The cache entries are kept in a linked list sorted according to how recently they were used. On a cache miss when the cache is full, the least recently used entry is removed to make space for the new entry. The patch migrates security_sid_to_context_core() to use the cache (also a few other functions where it was possible without too much fuss, but these mostly use the translation for logging in case of error, which is rare). Link: https://bugzilla.redhat.com/show_bug.cgi?id=1733259 Cc: Michal Sekletar <msekleta@redhat.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> Reviewed-by: Paul E. McKenney <paulmck@kernel.org> [PM: lots of merge fixups due to collisions with other sidtab patches] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-26 21:57:00 +08:00
rc = context_cpy(&isid->entry.context, context);
if (rc)
return rc;
selinux: cache the SID -> context string translation Translating a context struct to string can be quite slow, especially if the context has a lot of category bits set. This can cause quite noticeable performance impact in situations where the translation needs to be done repeatedly. A common example is a UNIX datagram socket with the SO_PASSSEC option enabled, which is used e.g. by systemd-journald when receiving log messages via datagram socket. This scenario can be reproduced with: cat /dev/urandom | base64 | logger & timeout 30s perf record -p $(pidof systemd-journald) -a -g kill %1 perf report -g none --pretty raw | grep security_secid_to_secctx Before the caching introduced by this patch, computing the context string (security_secid_to_secctx() function) takes up ~65% of systemd-journald's CPU time (assuming a context with 1024 categories set and Fedora x86_64 release kernel configs). After this patch (assuming near-perfect cache hit ratio) this overhead is reduced to just ~2%. This patch addresses the issue by caching a certain number (compile-time configurable) of recently used context strings to speed up repeated translations of the same context, while using only a small amount of memory. The cache is integrated into the existing sidtab table by adding a field to each entry, which when not NULL contains an RCU-protected pointer to a cache entry containing the cached string. The cache entries are kept in a linked list sorted according to how recently they were used. On a cache miss when the cache is full, the least recently used entry is removed to make space for the new entry. The patch migrates security_sid_to_context_core() to use the cache (also a few other functions where it was possible without too much fuss, but these mostly use the translation for logging in case of error, which is rare). Link: https://bugzilla.redhat.com/show_bug.cgi?id=1733259 Cc: Michal Sekletar <msekleta@redhat.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> Reviewed-by: Paul E. McKenney <paulmck@kernel.org> [PM: lots of merge fixups due to collisions with other sidtab patches] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-26 21:57:00 +08:00
#if CONFIG_SECURITY_SELINUX_SID2STR_CACHE_SIZE > 0
isid->entry.cache = NULL;
#endif
isid->set = 1;
selinux: sidtab reverse lookup hash table This replaces the reverse table lookup and reverse cache with a hashtable which improves cache-miss reverse-lookup times from O(n) to O(1)* and maintains the same performance as a reverse cache hit. This reduces the time needed to add a new sidtab entry from ~500us to 5us on a Pixel 3 when there are ~10,000 sidtab entries. The implementation uses the kernel's generic hashtable API, It uses the context's string represtation as the hash source, and the kernels generic string hashing algorithm full_name_hash() to reduce the string to a 32 bit value. This change also maintains the improvement introduced in commit ee1a84fdfeed ("selinux: overhaul sidtab to fix bug and improve performance") which removed the need to keep the current sidtab locked during policy reload. It does however introduce periodic locking of the target sidtab while converting the hashtable. Sidtab entries are never modified or removed, so the context struct stored in the sid_to_context tree can also be used for the context_to_sid hashtable to reduce memory usage. This bug was reported by: - On the selinux bug tracker. BUG: kernel softlockup due to too many SIDs/contexts #37 https://github.com/SELinuxProject/selinux-kernel/issues/37 - Jovana Knezevic on Android's bugtracker. Bug: 140252993 "During multi-user performance testing, we create and remove users many times. selinux_android_restorecon_pkgdir goes from 1ms to over 20ms after about 200 user creations and removals. Accumulated over ~280 packages, that adds a significant time to user creation, making perf benchmarks unreliable." * Hashtable lookup is only O(1) when n < the number of buckets. Signed-off-by: Jeff Vander Stoep <jeffv@google.com> Reported-by: Stephen Smalley <sds@tycho.nsa.gov> Reported-by: Jovana Knezevic <jovanak@google.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: subj tweak, removed changelog from patch description] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-22 17:33:06 +08:00
hash = context_compute_hash(context);
selinux: sidtab reverse lookup hash table This replaces the reverse table lookup and reverse cache with a hashtable which improves cache-miss reverse-lookup times from O(n) to O(1)* and maintains the same performance as a reverse cache hit. This reduces the time needed to add a new sidtab entry from ~500us to 5us on a Pixel 3 when there are ~10,000 sidtab entries. The implementation uses the kernel's generic hashtable API, It uses the context's string represtation as the hash source, and the kernels generic string hashing algorithm full_name_hash() to reduce the string to a 32 bit value. This change also maintains the improvement introduced in commit ee1a84fdfeed ("selinux: overhaul sidtab to fix bug and improve performance") which removed the need to keep the current sidtab locked during policy reload. It does however introduce periodic locking of the target sidtab while converting the hashtable. Sidtab entries are never modified or removed, so the context struct stored in the sid_to_context tree can also be used for the context_to_sid hashtable to reduce memory usage. This bug was reported by: - On the selinux bug tracker. BUG: kernel softlockup due to too many SIDs/contexts #37 https://github.com/SELinuxProject/selinux-kernel/issues/37 - Jovana Knezevic on Android's bugtracker. Bug: 140252993 "During multi-user performance testing, we create and remove users many times. selinux_android_restorecon_pkgdir goes from 1ms to over 20ms after about 200 user creations and removals. Accumulated over ~280 packages, that adds a significant time to user creation, making perf benchmarks unreliable." * Hashtable lookup is only O(1) when n < the number of buckets. Signed-off-by: Jeff Vander Stoep <jeffv@google.com> Reported-by: Stephen Smalley <sds@tycho.nsa.gov> Reported-by: Jovana Knezevic <jovanak@google.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: subj tweak, removed changelog from patch description] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-22 17:33:06 +08:00
/*
* Multiple initial sids may map to the same context. Check that this
* context is not already represented in the context_to_sid hashtable
* to avoid duplicate entries and long linked lists upon hash
* collision.
*/
if (!context_to_sid(s, context, hash)) {
selinux: cache the SID -> context string translation Translating a context struct to string can be quite slow, especially if the context has a lot of category bits set. This can cause quite noticeable performance impact in situations where the translation needs to be done repeatedly. A common example is a UNIX datagram socket with the SO_PASSSEC option enabled, which is used e.g. by systemd-journald when receiving log messages via datagram socket. This scenario can be reproduced with: cat /dev/urandom | base64 | logger & timeout 30s perf record -p $(pidof systemd-journald) -a -g kill %1 perf report -g none --pretty raw | grep security_secid_to_secctx Before the caching introduced by this patch, computing the context string (security_secid_to_secctx() function) takes up ~65% of systemd-journald's CPU time (assuming a context with 1024 categories set and Fedora x86_64 release kernel configs). After this patch (assuming near-perfect cache hit ratio) this overhead is reduced to just ~2%. This patch addresses the issue by caching a certain number (compile-time configurable) of recently used context strings to speed up repeated translations of the same context, while using only a small amount of memory. The cache is integrated into the existing sidtab table by adding a field to each entry, which when not NULL contains an RCU-protected pointer to a cache entry containing the cached string. The cache entries are kept in a linked list sorted according to how recently they were used. On a cache miss when the cache is full, the least recently used entry is removed to make space for the new entry. The patch migrates security_sid_to_context_core() to use the cache (also a few other functions where it was possible without too much fuss, but these mostly use the translation for logging in case of error, which is rare). Link: https://bugzilla.redhat.com/show_bug.cgi?id=1733259 Cc: Michal Sekletar <msekleta@redhat.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> Reviewed-by: Paul E. McKenney <paulmck@kernel.org> [PM: lots of merge fixups due to collisions with other sidtab patches] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-26 21:57:00 +08:00
isid->entry.sid = sid;
isid->entry.hash = hash;
hash_add(s->context_to_sid, &isid->entry.list, hash);
selinux: sidtab reverse lookup hash table This replaces the reverse table lookup and reverse cache with a hashtable which improves cache-miss reverse-lookup times from O(n) to O(1)* and maintains the same performance as a reverse cache hit. This reduces the time needed to add a new sidtab entry from ~500us to 5us on a Pixel 3 when there are ~10,000 sidtab entries. The implementation uses the kernel's generic hashtable API, It uses the context's string represtation as the hash source, and the kernels generic string hashing algorithm full_name_hash() to reduce the string to a 32 bit value. This change also maintains the improvement introduced in commit ee1a84fdfeed ("selinux: overhaul sidtab to fix bug and improve performance") which removed the need to keep the current sidtab locked during policy reload. It does however introduce periodic locking of the target sidtab while converting the hashtable. Sidtab entries are never modified or removed, so the context struct stored in the sid_to_context tree can also be used for the context_to_sid hashtable to reduce memory usage. This bug was reported by: - On the selinux bug tracker. BUG: kernel softlockup due to too many SIDs/contexts #37 https://github.com/SELinuxProject/selinux-kernel/issues/37 - Jovana Knezevic on Android's bugtracker. Bug: 140252993 "During multi-user performance testing, we create and remove users many times. selinux_android_restorecon_pkgdir goes from 1ms to over 20ms after about 200 user creations and removals. Accumulated over ~280 packages, that adds a significant time to user creation, making perf benchmarks unreliable." * Hashtable lookup is only O(1) when n < the number of buckets. Signed-off-by: Jeff Vander Stoep <jeffv@google.com> Reported-by: Stephen Smalley <sds@tycho.nsa.gov> Reported-by: Jovana Knezevic <jovanak@google.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: subj tweak, removed changelog from patch description] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-22 17:33:06 +08:00
}
return 0;
}
selinux: sidtab reverse lookup hash table This replaces the reverse table lookup and reverse cache with a hashtable which improves cache-miss reverse-lookup times from O(n) to O(1)* and maintains the same performance as a reverse cache hit. This reduces the time needed to add a new sidtab entry from ~500us to 5us on a Pixel 3 when there are ~10,000 sidtab entries. The implementation uses the kernel's generic hashtable API, It uses the context's string represtation as the hash source, and the kernels generic string hashing algorithm full_name_hash() to reduce the string to a 32 bit value. This change also maintains the improvement introduced in commit ee1a84fdfeed ("selinux: overhaul sidtab to fix bug and improve performance") which removed the need to keep the current sidtab locked during policy reload. It does however introduce periodic locking of the target sidtab while converting the hashtable. Sidtab entries are never modified or removed, so the context struct stored in the sid_to_context tree can also be used for the context_to_sid hashtable to reduce memory usage. This bug was reported by: - On the selinux bug tracker. BUG: kernel softlockup due to too many SIDs/contexts #37 https://github.com/SELinuxProject/selinux-kernel/issues/37 - Jovana Knezevic on Android's bugtracker. Bug: 140252993 "During multi-user performance testing, we create and remove users many times. selinux_android_restorecon_pkgdir goes from 1ms to over 20ms after about 200 user creations and removals. Accumulated over ~280 packages, that adds a significant time to user creation, making perf benchmarks unreliable." * Hashtable lookup is only O(1) when n < the number of buckets. Signed-off-by: Jeff Vander Stoep <jeffv@google.com> Reported-by: Stephen Smalley <sds@tycho.nsa.gov> Reported-by: Jovana Knezevic <jovanak@google.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: subj tweak, removed changelog from patch description] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-22 17:33:06 +08:00
int sidtab_hash_stats(struct sidtab *sidtab, char *page)
{
int i;
int chain_len = 0;
int slots_used = 0;
int entries = 0;
int max_chain_len = 0;
int cur_bucket = 0;
selinux: cache the SID -> context string translation Translating a context struct to string can be quite slow, especially if the context has a lot of category bits set. This can cause quite noticeable performance impact in situations where the translation needs to be done repeatedly. A common example is a UNIX datagram socket with the SO_PASSSEC option enabled, which is used e.g. by systemd-journald when receiving log messages via datagram socket. This scenario can be reproduced with: cat /dev/urandom | base64 | logger & timeout 30s perf record -p $(pidof systemd-journald) -a -g kill %1 perf report -g none --pretty raw | grep security_secid_to_secctx Before the caching introduced by this patch, computing the context string (security_secid_to_secctx() function) takes up ~65% of systemd-journald's CPU time (assuming a context with 1024 categories set and Fedora x86_64 release kernel configs). After this patch (assuming near-perfect cache hit ratio) this overhead is reduced to just ~2%. This patch addresses the issue by caching a certain number (compile-time configurable) of recently used context strings to speed up repeated translations of the same context, while using only a small amount of memory. The cache is integrated into the existing sidtab table by adding a field to each entry, which when not NULL contains an RCU-protected pointer to a cache entry containing the cached string. The cache entries are kept in a linked list sorted according to how recently they were used. On a cache miss when the cache is full, the least recently used entry is removed to make space for the new entry. The patch migrates security_sid_to_context_core() to use the cache (also a few other functions where it was possible without too much fuss, but these mostly use the translation for logging in case of error, which is rare). Link: https://bugzilla.redhat.com/show_bug.cgi?id=1733259 Cc: Michal Sekletar <msekleta@redhat.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> Reviewed-by: Paul E. McKenney <paulmck@kernel.org> [PM: lots of merge fixups due to collisions with other sidtab patches] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-26 21:57:00 +08:00
struct sidtab_entry *entry;
selinux: sidtab reverse lookup hash table This replaces the reverse table lookup and reverse cache with a hashtable which improves cache-miss reverse-lookup times from O(n) to O(1)* and maintains the same performance as a reverse cache hit. This reduces the time needed to add a new sidtab entry from ~500us to 5us on a Pixel 3 when there are ~10,000 sidtab entries. The implementation uses the kernel's generic hashtable API, It uses the context's string represtation as the hash source, and the kernels generic string hashing algorithm full_name_hash() to reduce the string to a 32 bit value. This change also maintains the improvement introduced in commit ee1a84fdfeed ("selinux: overhaul sidtab to fix bug and improve performance") which removed the need to keep the current sidtab locked during policy reload. It does however introduce periodic locking of the target sidtab while converting the hashtable. Sidtab entries are never modified or removed, so the context struct stored in the sid_to_context tree can also be used for the context_to_sid hashtable to reduce memory usage. This bug was reported by: - On the selinux bug tracker. BUG: kernel softlockup due to too many SIDs/contexts #37 https://github.com/SELinuxProject/selinux-kernel/issues/37 - Jovana Knezevic on Android's bugtracker. Bug: 140252993 "During multi-user performance testing, we create and remove users many times. selinux_android_restorecon_pkgdir goes from 1ms to over 20ms after about 200 user creations and removals. Accumulated over ~280 packages, that adds a significant time to user creation, making perf benchmarks unreliable." * Hashtable lookup is only O(1) when n < the number of buckets. Signed-off-by: Jeff Vander Stoep <jeffv@google.com> Reported-by: Stephen Smalley <sds@tycho.nsa.gov> Reported-by: Jovana Knezevic <jovanak@google.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: subj tweak, removed changelog from patch description] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-22 17:33:06 +08:00
rcu_read_lock();
hash_for_each_rcu(sidtab->context_to_sid, i, entry, list) {
entries++;
if (i == cur_bucket) {
chain_len++;
if (chain_len == 1)
slots_used++;
} else {
cur_bucket = i;
if (chain_len > max_chain_len)
max_chain_len = chain_len;
chain_len = 0;
}
}
rcu_read_unlock();
if (chain_len > max_chain_len)
max_chain_len = chain_len;
return scnprintf(page, PAGE_SIZE, "entries: %d\nbuckets used: %d/%d\n"
"longest chain: %d\n", entries,
slots_used, SIDTAB_HASH_BUCKETS, max_chain_len);
}
selinux: overhaul sidtab to fix bug and improve performance Before this patch, during a policy reload the sidtab would become frozen and trying to map a new context to SID would be unable to add a new entry to sidtab and fail with -ENOMEM. Such failures are usually propagated into userspace, which has no way of distignuishing them from actual allocation failures and thus doesn't handle them gracefully. Such situation can be triggered e.g. by the following reproducer: while true; do load_policy; echo -n .; sleep 0.1; done & for (( i = 0; i < 1024; i++ )); do runcon -l s0:c$i echo -n x || break # or: # chcon -l s0:c$i <some_file> || break done This patch overhauls the sidtab so it doesn't need to be frozen during policy reload, thus solving the above problem. The new SID table leverages the fact that SIDs are allocated sequentially and are never invalidated and stores them in linear buckets indexed by a tree structure. This brings several advantages: 1. Fast SID -> context lookup - this lookup can now be done in logarithmic time complexity (usually in less than 4 array lookups) and can still be done safely without locking. 2. No need to re-search the whole table on reverse lookup miss - after acquiring the spinlock only the newly added entries need to be searched, which means that reverse lookups that end up inserting a new entry are now about twice as fast. 3. No need to freeze sidtab during policy reload - it is now possible to handle insertion of new entries even during sidtab conversion. The tree structure of the new sidtab is able to grow automatically to up to about 2^31 entries (at which point it should not have more than about 4 tree levels). The old sidtab had a theoretical capacity of almost 2^32 entries, but half of that is still more than enough since by that point the reverse table lookups would become unusably slow anyway... The number of entries per tree node is selected automatically so that each node fits into a single page, which should be the easiest size for kmalloc() to handle. Note that the cache for reverse lookup is preserved with equivalent logic. The only difference is that instead of storing pointers to the hash table nodes it stores just the indices of the cached entries. The new cache ensures that the indices are loaded/stored atomically, but it still has the drawback that concurrent cache updates may mess up the contents of the cache. Such situation however only reduces its effectivity, not the correctness of lookups. Tested by selinux-testsuite and thoroughly tortured by this simple stress test: ``` function rand_cat() { echo $(( $RANDOM % 1024 )) } function do_work() { while true; do echo -n "system_u:system_r:kernel_t:s0:c$(rand_cat),c$(rand_cat)" \ >/sys/fs/selinux/context 2>/dev/null || true done } do_work >/dev/null & do_work >/dev/null & do_work >/dev/null & while load_policy; do echo -n .; sleep 0.1; done kill %1 kill %2 kill %3 ``` Link: https://github.com/SELinuxProject/selinux-kernel/issues/38 Reported-by: Orion Poplawski <orion@nwra.com> Reported-by: Li Kun <hw.likun@huawei.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: most of sidtab.c merged by hand due to conflicts] [PM: checkpatch fixes in mls.c, services.c, sidtab.c] Signed-off-by: Paul Moore <paul@paul-moore.com>
2018-11-30 23:24:08 +08:00
static u32 sidtab_level_from_count(u32 count)
{
selinux: overhaul sidtab to fix bug and improve performance Before this patch, during a policy reload the sidtab would become frozen and trying to map a new context to SID would be unable to add a new entry to sidtab and fail with -ENOMEM. Such failures are usually propagated into userspace, which has no way of distignuishing them from actual allocation failures and thus doesn't handle them gracefully. Such situation can be triggered e.g. by the following reproducer: while true; do load_policy; echo -n .; sleep 0.1; done & for (( i = 0; i < 1024; i++ )); do runcon -l s0:c$i echo -n x || break # or: # chcon -l s0:c$i <some_file> || break done This patch overhauls the sidtab so it doesn't need to be frozen during policy reload, thus solving the above problem. The new SID table leverages the fact that SIDs are allocated sequentially and are never invalidated and stores them in linear buckets indexed by a tree structure. This brings several advantages: 1. Fast SID -> context lookup - this lookup can now be done in logarithmic time complexity (usually in less than 4 array lookups) and can still be done safely without locking. 2. No need to re-search the whole table on reverse lookup miss - after acquiring the spinlock only the newly added entries need to be searched, which means that reverse lookups that end up inserting a new entry are now about twice as fast. 3. No need to freeze sidtab during policy reload - it is now possible to handle insertion of new entries even during sidtab conversion. The tree structure of the new sidtab is able to grow automatically to up to about 2^31 entries (at which point it should not have more than about 4 tree levels). The old sidtab had a theoretical capacity of almost 2^32 entries, but half of that is still more than enough since by that point the reverse table lookups would become unusably slow anyway... The number of entries per tree node is selected automatically so that each node fits into a single page, which should be the easiest size for kmalloc() to handle. Note that the cache for reverse lookup is preserved with equivalent logic. The only difference is that instead of storing pointers to the hash table nodes it stores just the indices of the cached entries. The new cache ensures that the indices are loaded/stored atomically, but it still has the drawback that concurrent cache updates may mess up the contents of the cache. Such situation however only reduces its effectivity, not the correctness of lookups. Tested by selinux-testsuite and thoroughly tortured by this simple stress test: ``` function rand_cat() { echo $(( $RANDOM % 1024 )) } function do_work() { while true; do echo -n "system_u:system_r:kernel_t:s0:c$(rand_cat),c$(rand_cat)" \ >/sys/fs/selinux/context 2>/dev/null || true done } do_work >/dev/null & do_work >/dev/null & do_work >/dev/null & while load_policy; do echo -n .; sleep 0.1; done kill %1 kill %2 kill %3 ``` Link: https://github.com/SELinuxProject/selinux-kernel/issues/38 Reported-by: Orion Poplawski <orion@nwra.com> Reported-by: Li Kun <hw.likun@huawei.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: most of sidtab.c merged by hand due to conflicts] [PM: checkpatch fixes in mls.c, services.c, sidtab.c] Signed-off-by: Paul Moore <paul@paul-moore.com>
2018-11-30 23:24:08 +08:00
u32 capacity = SIDTAB_LEAF_ENTRIES;
u32 level = 0;
while (count > capacity) {
capacity <<= SIDTAB_INNER_SHIFT;
++level;
}
return level;
}
static int sidtab_alloc_roots(struct sidtab *s, u32 level)
{
u32 l;
if (!s->roots[0].ptr_leaf) {
s->roots[0].ptr_leaf = kzalloc(SIDTAB_NODE_ALLOC_SIZE,
GFP_ATOMIC);
if (!s->roots[0].ptr_leaf)
return -ENOMEM;
}
for (l = 1; l <= level; ++l)
if (!s->roots[l].ptr_inner) {
s->roots[l].ptr_inner = kzalloc(SIDTAB_NODE_ALLOC_SIZE,
GFP_ATOMIC);
if (!s->roots[l].ptr_inner)
return -ENOMEM;
s->roots[l].ptr_inner->entries[0] = s->roots[l - 1];
}
return 0;
}
selinux: cache the SID -> context string translation Translating a context struct to string can be quite slow, especially if the context has a lot of category bits set. This can cause quite noticeable performance impact in situations where the translation needs to be done repeatedly. A common example is a UNIX datagram socket with the SO_PASSSEC option enabled, which is used e.g. by systemd-journald when receiving log messages via datagram socket. This scenario can be reproduced with: cat /dev/urandom | base64 | logger & timeout 30s perf record -p $(pidof systemd-journald) -a -g kill %1 perf report -g none --pretty raw | grep security_secid_to_secctx Before the caching introduced by this patch, computing the context string (security_secid_to_secctx() function) takes up ~65% of systemd-journald's CPU time (assuming a context with 1024 categories set and Fedora x86_64 release kernel configs). After this patch (assuming near-perfect cache hit ratio) this overhead is reduced to just ~2%. This patch addresses the issue by caching a certain number (compile-time configurable) of recently used context strings to speed up repeated translations of the same context, while using only a small amount of memory. The cache is integrated into the existing sidtab table by adding a field to each entry, which when not NULL contains an RCU-protected pointer to a cache entry containing the cached string. The cache entries are kept in a linked list sorted according to how recently they were used. On a cache miss when the cache is full, the least recently used entry is removed to make space for the new entry. The patch migrates security_sid_to_context_core() to use the cache (also a few other functions where it was possible without too much fuss, but these mostly use the translation for logging in case of error, which is rare). Link: https://bugzilla.redhat.com/show_bug.cgi?id=1733259 Cc: Michal Sekletar <msekleta@redhat.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> Reviewed-by: Paul E. McKenney <paulmck@kernel.org> [PM: lots of merge fixups due to collisions with other sidtab patches] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-26 21:57:00 +08:00
static struct sidtab_entry *sidtab_do_lookup(struct sidtab *s, u32 index,
int alloc)
selinux: overhaul sidtab to fix bug and improve performance Before this patch, during a policy reload the sidtab would become frozen and trying to map a new context to SID would be unable to add a new entry to sidtab and fail with -ENOMEM. Such failures are usually propagated into userspace, which has no way of distignuishing them from actual allocation failures and thus doesn't handle them gracefully. Such situation can be triggered e.g. by the following reproducer: while true; do load_policy; echo -n .; sleep 0.1; done & for (( i = 0; i < 1024; i++ )); do runcon -l s0:c$i echo -n x || break # or: # chcon -l s0:c$i <some_file> || break done This patch overhauls the sidtab so it doesn't need to be frozen during policy reload, thus solving the above problem. The new SID table leverages the fact that SIDs are allocated sequentially and are never invalidated and stores them in linear buckets indexed by a tree structure. This brings several advantages: 1. Fast SID -> context lookup - this lookup can now be done in logarithmic time complexity (usually in less than 4 array lookups) and can still be done safely without locking. 2. No need to re-search the whole table on reverse lookup miss - after acquiring the spinlock only the newly added entries need to be searched, which means that reverse lookups that end up inserting a new entry are now about twice as fast. 3. No need to freeze sidtab during policy reload - it is now possible to handle insertion of new entries even during sidtab conversion. The tree structure of the new sidtab is able to grow automatically to up to about 2^31 entries (at which point it should not have more than about 4 tree levels). The old sidtab had a theoretical capacity of almost 2^32 entries, but half of that is still more than enough since by that point the reverse table lookups would become unusably slow anyway... The number of entries per tree node is selected automatically so that each node fits into a single page, which should be the easiest size for kmalloc() to handle. Note that the cache for reverse lookup is preserved with equivalent logic. The only difference is that instead of storing pointers to the hash table nodes it stores just the indices of the cached entries. The new cache ensures that the indices are loaded/stored atomically, but it still has the drawback that concurrent cache updates may mess up the contents of the cache. Such situation however only reduces its effectivity, not the correctness of lookups. Tested by selinux-testsuite and thoroughly tortured by this simple stress test: ``` function rand_cat() { echo $(( $RANDOM % 1024 )) } function do_work() { while true; do echo -n "system_u:system_r:kernel_t:s0:c$(rand_cat),c$(rand_cat)" \ >/sys/fs/selinux/context 2>/dev/null || true done } do_work >/dev/null & do_work >/dev/null & do_work >/dev/null & while load_policy; do echo -n .; sleep 0.1; done kill %1 kill %2 kill %3 ``` Link: https://github.com/SELinuxProject/selinux-kernel/issues/38 Reported-by: Orion Poplawski <orion@nwra.com> Reported-by: Li Kun <hw.likun@huawei.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: most of sidtab.c merged by hand due to conflicts] [PM: checkpatch fixes in mls.c, services.c, sidtab.c] Signed-off-by: Paul Moore <paul@paul-moore.com>
2018-11-30 23:24:08 +08:00
{
union sidtab_entry_inner *entry;
u32 level, capacity_shift, leaf_index = index / SIDTAB_LEAF_ENTRIES;
/* find the level of the subtree we need */
level = sidtab_level_from_count(index + 1);
capacity_shift = level * SIDTAB_INNER_SHIFT;
/* allocate roots if needed */
if (alloc && sidtab_alloc_roots(s, level) != 0)
return NULL;
/* lookup inside the subtree */
entry = &s->roots[level];
while (level != 0) {
capacity_shift -= SIDTAB_INNER_SHIFT;
--level;
entry = &entry->ptr_inner->entries[leaf_index >> capacity_shift];
leaf_index &= ((u32)1 << capacity_shift) - 1;
if (!entry->ptr_inner) {
if (alloc)
entry->ptr_inner = kzalloc(SIDTAB_NODE_ALLOC_SIZE,
GFP_ATOMIC);
if (!entry->ptr_inner)
return NULL;
}
}
if (!entry->ptr_leaf) {
if (alloc)
entry->ptr_leaf = kzalloc(SIDTAB_NODE_ALLOC_SIZE,
GFP_ATOMIC);
if (!entry->ptr_leaf)
return NULL;
}
selinux: sidtab reverse lookup hash table This replaces the reverse table lookup and reverse cache with a hashtable which improves cache-miss reverse-lookup times from O(n) to O(1)* and maintains the same performance as a reverse cache hit. This reduces the time needed to add a new sidtab entry from ~500us to 5us on a Pixel 3 when there are ~10,000 sidtab entries. The implementation uses the kernel's generic hashtable API, It uses the context's string represtation as the hash source, and the kernels generic string hashing algorithm full_name_hash() to reduce the string to a 32 bit value. This change also maintains the improvement introduced in commit ee1a84fdfeed ("selinux: overhaul sidtab to fix bug and improve performance") which removed the need to keep the current sidtab locked during policy reload. It does however introduce periodic locking of the target sidtab while converting the hashtable. Sidtab entries are never modified or removed, so the context struct stored in the sid_to_context tree can also be used for the context_to_sid hashtable to reduce memory usage. This bug was reported by: - On the selinux bug tracker. BUG: kernel softlockup due to too many SIDs/contexts #37 https://github.com/SELinuxProject/selinux-kernel/issues/37 - Jovana Knezevic on Android's bugtracker. Bug: 140252993 "During multi-user performance testing, we create and remove users many times. selinux_android_restorecon_pkgdir goes from 1ms to over 20ms after about 200 user creations and removals. Accumulated over ~280 packages, that adds a significant time to user creation, making perf benchmarks unreliable." * Hashtable lookup is only O(1) when n < the number of buckets. Signed-off-by: Jeff Vander Stoep <jeffv@google.com> Reported-by: Stephen Smalley <sds@tycho.nsa.gov> Reported-by: Jovana Knezevic <jovanak@google.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: subj tweak, removed changelog from patch description] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-22 17:33:06 +08:00
return &entry->ptr_leaf->entries[index % SIDTAB_LEAF_ENTRIES];
selinux: overhaul sidtab to fix bug and improve performance Before this patch, during a policy reload the sidtab would become frozen and trying to map a new context to SID would be unable to add a new entry to sidtab and fail with -ENOMEM. Such failures are usually propagated into userspace, which has no way of distignuishing them from actual allocation failures and thus doesn't handle them gracefully. Such situation can be triggered e.g. by the following reproducer: while true; do load_policy; echo -n .; sleep 0.1; done & for (( i = 0; i < 1024; i++ )); do runcon -l s0:c$i echo -n x || break # or: # chcon -l s0:c$i <some_file> || break done This patch overhauls the sidtab so it doesn't need to be frozen during policy reload, thus solving the above problem. The new SID table leverages the fact that SIDs are allocated sequentially and are never invalidated and stores them in linear buckets indexed by a tree structure. This brings several advantages: 1. Fast SID -> context lookup - this lookup can now be done in logarithmic time complexity (usually in less than 4 array lookups) and can still be done safely without locking. 2. No need to re-search the whole table on reverse lookup miss - after acquiring the spinlock only the newly added entries need to be searched, which means that reverse lookups that end up inserting a new entry are now about twice as fast. 3. No need to freeze sidtab during policy reload - it is now possible to handle insertion of new entries even during sidtab conversion. The tree structure of the new sidtab is able to grow automatically to up to about 2^31 entries (at which point it should not have more than about 4 tree levels). The old sidtab had a theoretical capacity of almost 2^32 entries, but half of that is still more than enough since by that point the reverse table lookups would become unusably slow anyway... The number of entries per tree node is selected automatically so that each node fits into a single page, which should be the easiest size for kmalloc() to handle. Note that the cache for reverse lookup is preserved with equivalent logic. The only difference is that instead of storing pointers to the hash table nodes it stores just the indices of the cached entries. The new cache ensures that the indices are loaded/stored atomically, but it still has the drawback that concurrent cache updates may mess up the contents of the cache. Such situation however only reduces its effectivity, not the correctness of lookups. Tested by selinux-testsuite and thoroughly tortured by this simple stress test: ``` function rand_cat() { echo $(( $RANDOM % 1024 )) } function do_work() { while true; do echo -n "system_u:system_r:kernel_t:s0:c$(rand_cat),c$(rand_cat)" \ >/sys/fs/selinux/context 2>/dev/null || true done } do_work >/dev/null & do_work >/dev/null & do_work >/dev/null & while load_policy; do echo -n .; sleep 0.1; done kill %1 kill %2 kill %3 ``` Link: https://github.com/SELinuxProject/selinux-kernel/issues/38 Reported-by: Orion Poplawski <orion@nwra.com> Reported-by: Li Kun <hw.likun@huawei.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: most of sidtab.c merged by hand due to conflicts] [PM: checkpatch fixes in mls.c, services.c, sidtab.c] Signed-off-by: Paul Moore <paul@paul-moore.com>
2018-11-30 23:24:08 +08:00
}
selinux: cache the SID -> context string translation Translating a context struct to string can be quite slow, especially if the context has a lot of category bits set. This can cause quite noticeable performance impact in situations where the translation needs to be done repeatedly. A common example is a UNIX datagram socket with the SO_PASSSEC option enabled, which is used e.g. by systemd-journald when receiving log messages via datagram socket. This scenario can be reproduced with: cat /dev/urandom | base64 | logger & timeout 30s perf record -p $(pidof systemd-journald) -a -g kill %1 perf report -g none --pretty raw | grep security_secid_to_secctx Before the caching introduced by this patch, computing the context string (security_secid_to_secctx() function) takes up ~65% of systemd-journald's CPU time (assuming a context with 1024 categories set and Fedora x86_64 release kernel configs). After this patch (assuming near-perfect cache hit ratio) this overhead is reduced to just ~2%. This patch addresses the issue by caching a certain number (compile-time configurable) of recently used context strings to speed up repeated translations of the same context, while using only a small amount of memory. The cache is integrated into the existing sidtab table by adding a field to each entry, which when not NULL contains an RCU-protected pointer to a cache entry containing the cached string. The cache entries are kept in a linked list sorted according to how recently they were used. On a cache miss when the cache is full, the least recently used entry is removed to make space for the new entry. The patch migrates security_sid_to_context_core() to use the cache (also a few other functions where it was possible without too much fuss, but these mostly use the translation for logging in case of error, which is rare). Link: https://bugzilla.redhat.com/show_bug.cgi?id=1733259 Cc: Michal Sekletar <msekleta@redhat.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> Reviewed-by: Paul E. McKenney <paulmck@kernel.org> [PM: lots of merge fixups due to collisions with other sidtab patches] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-26 21:57:00 +08:00
static struct sidtab_entry *sidtab_lookup(struct sidtab *s, u32 index)
selinux: overhaul sidtab to fix bug and improve performance Before this patch, during a policy reload the sidtab would become frozen and trying to map a new context to SID would be unable to add a new entry to sidtab and fail with -ENOMEM. Such failures are usually propagated into userspace, which has no way of distignuishing them from actual allocation failures and thus doesn't handle them gracefully. Such situation can be triggered e.g. by the following reproducer: while true; do load_policy; echo -n .; sleep 0.1; done & for (( i = 0; i < 1024; i++ )); do runcon -l s0:c$i echo -n x || break # or: # chcon -l s0:c$i <some_file> || break done This patch overhauls the sidtab so it doesn't need to be frozen during policy reload, thus solving the above problem. The new SID table leverages the fact that SIDs are allocated sequentially and are never invalidated and stores them in linear buckets indexed by a tree structure. This brings several advantages: 1. Fast SID -> context lookup - this lookup can now be done in logarithmic time complexity (usually in less than 4 array lookups) and can still be done safely without locking. 2. No need to re-search the whole table on reverse lookup miss - after acquiring the spinlock only the newly added entries need to be searched, which means that reverse lookups that end up inserting a new entry are now about twice as fast. 3. No need to freeze sidtab during policy reload - it is now possible to handle insertion of new entries even during sidtab conversion. The tree structure of the new sidtab is able to grow automatically to up to about 2^31 entries (at which point it should not have more than about 4 tree levels). The old sidtab had a theoretical capacity of almost 2^32 entries, but half of that is still more than enough since by that point the reverse table lookups would become unusably slow anyway... The number of entries per tree node is selected automatically so that each node fits into a single page, which should be the easiest size for kmalloc() to handle. Note that the cache for reverse lookup is preserved with equivalent logic. The only difference is that instead of storing pointers to the hash table nodes it stores just the indices of the cached entries. The new cache ensures that the indices are loaded/stored atomically, but it still has the drawback that concurrent cache updates may mess up the contents of the cache. Such situation however only reduces its effectivity, not the correctness of lookups. Tested by selinux-testsuite and thoroughly tortured by this simple stress test: ``` function rand_cat() { echo $(( $RANDOM % 1024 )) } function do_work() { while true; do echo -n "system_u:system_r:kernel_t:s0:c$(rand_cat),c$(rand_cat)" \ >/sys/fs/selinux/context 2>/dev/null || true done } do_work >/dev/null & do_work >/dev/null & do_work >/dev/null & while load_policy; do echo -n .; sleep 0.1; done kill %1 kill %2 kill %3 ``` Link: https://github.com/SELinuxProject/selinux-kernel/issues/38 Reported-by: Orion Poplawski <orion@nwra.com> Reported-by: Li Kun <hw.likun@huawei.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: most of sidtab.c merged by hand due to conflicts] [PM: checkpatch fixes in mls.c, services.c, sidtab.c] Signed-off-by: Paul Moore <paul@paul-moore.com>
2018-11-30 23:24:08 +08:00
{
/* read entries only after reading count */
u32 count = smp_load_acquire(&s->count);
selinux: overhaul sidtab to fix bug and improve performance Before this patch, during a policy reload the sidtab would become frozen and trying to map a new context to SID would be unable to add a new entry to sidtab and fail with -ENOMEM. Such failures are usually propagated into userspace, which has no way of distignuishing them from actual allocation failures and thus doesn't handle them gracefully. Such situation can be triggered e.g. by the following reproducer: while true; do load_policy; echo -n .; sleep 0.1; done & for (( i = 0; i < 1024; i++ )); do runcon -l s0:c$i echo -n x || break # or: # chcon -l s0:c$i <some_file> || break done This patch overhauls the sidtab so it doesn't need to be frozen during policy reload, thus solving the above problem. The new SID table leverages the fact that SIDs are allocated sequentially and are never invalidated and stores them in linear buckets indexed by a tree structure. This brings several advantages: 1. Fast SID -> context lookup - this lookup can now be done in logarithmic time complexity (usually in less than 4 array lookups) and can still be done safely without locking. 2. No need to re-search the whole table on reverse lookup miss - after acquiring the spinlock only the newly added entries need to be searched, which means that reverse lookups that end up inserting a new entry are now about twice as fast. 3. No need to freeze sidtab during policy reload - it is now possible to handle insertion of new entries even during sidtab conversion. The tree structure of the new sidtab is able to grow automatically to up to about 2^31 entries (at which point it should not have more than about 4 tree levels). The old sidtab had a theoretical capacity of almost 2^32 entries, but half of that is still more than enough since by that point the reverse table lookups would become unusably slow anyway... The number of entries per tree node is selected automatically so that each node fits into a single page, which should be the easiest size for kmalloc() to handle. Note that the cache for reverse lookup is preserved with equivalent logic. The only difference is that instead of storing pointers to the hash table nodes it stores just the indices of the cached entries. The new cache ensures that the indices are loaded/stored atomically, but it still has the drawback that concurrent cache updates may mess up the contents of the cache. Such situation however only reduces its effectivity, not the correctness of lookups. Tested by selinux-testsuite and thoroughly tortured by this simple stress test: ``` function rand_cat() { echo $(( $RANDOM % 1024 )) } function do_work() { while true; do echo -n "system_u:system_r:kernel_t:s0:c$(rand_cat),c$(rand_cat)" \ >/sys/fs/selinux/context 2>/dev/null || true done } do_work >/dev/null & do_work >/dev/null & do_work >/dev/null & while load_policy; do echo -n .; sleep 0.1; done kill %1 kill %2 kill %3 ``` Link: https://github.com/SELinuxProject/selinux-kernel/issues/38 Reported-by: Orion Poplawski <orion@nwra.com> Reported-by: Li Kun <hw.likun@huawei.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: most of sidtab.c merged by hand due to conflicts] [PM: checkpatch fixes in mls.c, services.c, sidtab.c] Signed-off-by: Paul Moore <paul@paul-moore.com>
2018-11-30 23:24:08 +08:00
if (index >= count)
return NULL;
selinux: support deferred mapping of contexts Introduce SELinux support for deferred mapping of security contexts in the SID table upon policy reload, and use this support for inode security contexts when the context is not yet valid under the current policy. Only processes with CAP_MAC_ADMIN + mac_admin permission in policy can set undefined security contexts on inodes. Inodes with such undefined contexts are treated as having the unlabeled context until the context becomes valid upon a policy reload that defines the context. Context invalidation upon policy reload also uses this support to save the context information in the SID table and later recover it upon a subsequent policy reload that defines the context again. This support is to enable package managers and similar programs to set down file contexts unknown to the system policy at the time the file is created in order to better support placing loadable policy modules in packages and to support build systems that need to create images of different distro releases with different policies w/o requiring all of the contexts to be defined or legal in the build host policy. With this patch applied, the following sequence is possible, although in practice it is recommended that this permission only be allowed to specific program domains such as the package manager. # rmdir baz # rm bar # touch bar # chcon -t foo_exec_t bar # foo_exec_t is not yet defined chcon: failed to change context of `bar' to `system_u:object_r:foo_exec_t': Invalid argument # mkdir -Z system_u:object_r:foo_exec_t baz mkdir: failed to set default file creation context to `system_u:object_r:foo_exec_t': Invalid argument # cat setundefined.te policy_module(setundefined, 1.0) require { type unconfined_t; type unlabeled_t; } files_type(unlabeled_t) allow unconfined_t self:capability2 mac_admin; # make -f /usr/share/selinux/devel/Makefile setundefined.pp # semodule -i setundefined.pp # chcon -t foo_exec_t bar # foo_exec_t is not yet defined # mkdir -Z system_u:object_r:foo_exec_t baz # ls -Zd bar baz -rw-r--r-- root root system_u:object_r:unlabeled_t bar drwxr-xr-x root root system_u:object_r:unlabeled_t baz # cat foo.te policy_module(foo, 1.0) type foo_exec_t; files_type(foo_exec_t) # make -f /usr/share/selinux/devel/Makefile foo.pp # semodule -i foo.pp # defines foo_exec_t # ls -Zd bar baz -rw-r--r-- root root user_u:object_r:foo_exec_t bar drwxr-xr-x root root system_u:object_r:foo_exec_t baz # semodule -r foo # ls -Zd bar baz -rw-r--r-- root root system_u:object_r:unlabeled_t bar drwxr-xr-x root root system_u:object_r:unlabeled_t baz # semodule -i foo.pp # ls -Zd bar baz -rw-r--r-- root root user_u:object_r:foo_exec_t bar drwxr-xr-x root root system_u:object_r:foo_exec_t baz # semodule -r setundefined foo # chcon -t foo_exec_t bar # no longer defined and not allowed chcon: failed to change context of `bar' to `system_u:object_r:foo_exec_t': Invalid argument # rmdir baz # mkdir -Z system_u:object_r:foo_exec_t baz mkdir: failed to set default file creation context to `system_u:object_r:foo_exec_t': Invalid argument Signed-off-by: Stephen Smalley <sds@tycho.nsa.gov> Signed-off-by: James Morris <jmorris@namei.org>
2008-05-08 01:03:20 +08:00
selinux: cache the SID -> context string translation Translating a context struct to string can be quite slow, especially if the context has a lot of category bits set. This can cause quite noticeable performance impact in situations where the translation needs to be done repeatedly. A common example is a UNIX datagram socket with the SO_PASSSEC option enabled, which is used e.g. by systemd-journald when receiving log messages via datagram socket. This scenario can be reproduced with: cat /dev/urandom | base64 | logger & timeout 30s perf record -p $(pidof systemd-journald) -a -g kill %1 perf report -g none --pretty raw | grep security_secid_to_secctx Before the caching introduced by this patch, computing the context string (security_secid_to_secctx() function) takes up ~65% of systemd-journald's CPU time (assuming a context with 1024 categories set and Fedora x86_64 release kernel configs). After this patch (assuming near-perfect cache hit ratio) this overhead is reduced to just ~2%. This patch addresses the issue by caching a certain number (compile-time configurable) of recently used context strings to speed up repeated translations of the same context, while using only a small amount of memory. The cache is integrated into the existing sidtab table by adding a field to each entry, which when not NULL contains an RCU-protected pointer to a cache entry containing the cached string. The cache entries are kept in a linked list sorted according to how recently they were used. On a cache miss when the cache is full, the least recently used entry is removed to make space for the new entry. The patch migrates security_sid_to_context_core() to use the cache (also a few other functions where it was possible without too much fuss, but these mostly use the translation for logging in case of error, which is rare). Link: https://bugzilla.redhat.com/show_bug.cgi?id=1733259 Cc: Michal Sekletar <msekleta@redhat.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> Reviewed-by: Paul E. McKenney <paulmck@kernel.org> [PM: lots of merge fixups due to collisions with other sidtab patches] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-26 21:57:00 +08:00
return sidtab_do_lookup(s, index, 0);
}
selinux: cache the SID -> context string translation Translating a context struct to string can be quite slow, especially if the context has a lot of category bits set. This can cause quite noticeable performance impact in situations where the translation needs to be done repeatedly. A common example is a UNIX datagram socket with the SO_PASSSEC option enabled, which is used e.g. by systemd-journald when receiving log messages via datagram socket. This scenario can be reproduced with: cat /dev/urandom | base64 | logger & timeout 30s perf record -p $(pidof systemd-journald) -a -g kill %1 perf report -g none --pretty raw | grep security_secid_to_secctx Before the caching introduced by this patch, computing the context string (security_secid_to_secctx() function) takes up ~65% of systemd-journald's CPU time (assuming a context with 1024 categories set and Fedora x86_64 release kernel configs). After this patch (assuming near-perfect cache hit ratio) this overhead is reduced to just ~2%. This patch addresses the issue by caching a certain number (compile-time configurable) of recently used context strings to speed up repeated translations of the same context, while using only a small amount of memory. The cache is integrated into the existing sidtab table by adding a field to each entry, which when not NULL contains an RCU-protected pointer to a cache entry containing the cached string. The cache entries are kept in a linked list sorted according to how recently they were used. On a cache miss when the cache is full, the least recently used entry is removed to make space for the new entry. The patch migrates security_sid_to_context_core() to use the cache (also a few other functions where it was possible without too much fuss, but these mostly use the translation for logging in case of error, which is rare). Link: https://bugzilla.redhat.com/show_bug.cgi?id=1733259 Cc: Michal Sekletar <msekleta@redhat.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> Reviewed-by: Paul E. McKenney <paulmck@kernel.org> [PM: lots of merge fixups due to collisions with other sidtab patches] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-26 21:57:00 +08:00
static struct sidtab_entry *sidtab_lookup_initial(struct sidtab *s, u32 sid)
{
selinux: cache the SID -> context string translation Translating a context struct to string can be quite slow, especially if the context has a lot of category bits set. This can cause quite noticeable performance impact in situations where the translation needs to be done repeatedly. A common example is a UNIX datagram socket with the SO_PASSSEC option enabled, which is used e.g. by systemd-journald when receiving log messages via datagram socket. This scenario can be reproduced with: cat /dev/urandom | base64 | logger & timeout 30s perf record -p $(pidof systemd-journald) -a -g kill %1 perf report -g none --pretty raw | grep security_secid_to_secctx Before the caching introduced by this patch, computing the context string (security_secid_to_secctx() function) takes up ~65% of systemd-journald's CPU time (assuming a context with 1024 categories set and Fedora x86_64 release kernel configs). After this patch (assuming near-perfect cache hit ratio) this overhead is reduced to just ~2%. This patch addresses the issue by caching a certain number (compile-time configurable) of recently used context strings to speed up repeated translations of the same context, while using only a small amount of memory. The cache is integrated into the existing sidtab table by adding a field to each entry, which when not NULL contains an RCU-protected pointer to a cache entry containing the cached string. The cache entries are kept in a linked list sorted according to how recently they were used. On a cache miss when the cache is full, the least recently used entry is removed to make space for the new entry. The patch migrates security_sid_to_context_core() to use the cache (also a few other functions where it was possible without too much fuss, but these mostly use the translation for logging in case of error, which is rare). Link: https://bugzilla.redhat.com/show_bug.cgi?id=1733259 Cc: Michal Sekletar <msekleta@redhat.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> Reviewed-by: Paul E. McKenney <paulmck@kernel.org> [PM: lots of merge fixups due to collisions with other sidtab patches] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-26 21:57:00 +08:00
return s->isids[sid - 1].set ? &s->isids[sid - 1].entry : NULL;
}
selinux: cache the SID -> context string translation Translating a context struct to string can be quite slow, especially if the context has a lot of category bits set. This can cause quite noticeable performance impact in situations where the translation needs to be done repeatedly. A common example is a UNIX datagram socket with the SO_PASSSEC option enabled, which is used e.g. by systemd-journald when receiving log messages via datagram socket. This scenario can be reproduced with: cat /dev/urandom | base64 | logger & timeout 30s perf record -p $(pidof systemd-journald) -a -g kill %1 perf report -g none --pretty raw | grep security_secid_to_secctx Before the caching introduced by this patch, computing the context string (security_secid_to_secctx() function) takes up ~65% of systemd-journald's CPU time (assuming a context with 1024 categories set and Fedora x86_64 release kernel configs). After this patch (assuming near-perfect cache hit ratio) this overhead is reduced to just ~2%. This patch addresses the issue by caching a certain number (compile-time configurable) of recently used context strings to speed up repeated translations of the same context, while using only a small amount of memory. The cache is integrated into the existing sidtab table by adding a field to each entry, which when not NULL contains an RCU-protected pointer to a cache entry containing the cached string. The cache entries are kept in a linked list sorted according to how recently they were used. On a cache miss when the cache is full, the least recently used entry is removed to make space for the new entry. The patch migrates security_sid_to_context_core() to use the cache (also a few other functions where it was possible without too much fuss, but these mostly use the translation for logging in case of error, which is rare). Link: https://bugzilla.redhat.com/show_bug.cgi?id=1733259 Cc: Michal Sekletar <msekleta@redhat.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> Reviewed-by: Paul E. McKenney <paulmck@kernel.org> [PM: lots of merge fixups due to collisions with other sidtab patches] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-26 21:57:00 +08:00
static struct sidtab_entry *sidtab_search_core(struct sidtab *s, u32 sid,
int force)
{
if (sid != 0) {
selinux: cache the SID -> context string translation Translating a context struct to string can be quite slow, especially if the context has a lot of category bits set. This can cause quite noticeable performance impact in situations where the translation needs to be done repeatedly. A common example is a UNIX datagram socket with the SO_PASSSEC option enabled, which is used e.g. by systemd-journald when receiving log messages via datagram socket. This scenario can be reproduced with: cat /dev/urandom | base64 | logger & timeout 30s perf record -p $(pidof systemd-journald) -a -g kill %1 perf report -g none --pretty raw | grep security_secid_to_secctx Before the caching introduced by this patch, computing the context string (security_secid_to_secctx() function) takes up ~65% of systemd-journald's CPU time (assuming a context with 1024 categories set and Fedora x86_64 release kernel configs). After this patch (assuming near-perfect cache hit ratio) this overhead is reduced to just ~2%. This patch addresses the issue by caching a certain number (compile-time configurable) of recently used context strings to speed up repeated translations of the same context, while using only a small amount of memory. The cache is integrated into the existing sidtab table by adding a field to each entry, which when not NULL contains an RCU-protected pointer to a cache entry containing the cached string. The cache entries are kept in a linked list sorted according to how recently they were used. On a cache miss when the cache is full, the least recently used entry is removed to make space for the new entry. The patch migrates security_sid_to_context_core() to use the cache (also a few other functions where it was possible without too much fuss, but these mostly use the translation for logging in case of error, which is rare). Link: https://bugzilla.redhat.com/show_bug.cgi?id=1733259 Cc: Michal Sekletar <msekleta@redhat.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> Reviewed-by: Paul E. McKenney <paulmck@kernel.org> [PM: lots of merge fixups due to collisions with other sidtab patches] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-26 21:57:00 +08:00
struct sidtab_entry *entry;
if (sid > SECINITSID_NUM)
selinux: cache the SID -> context string translation Translating a context struct to string can be quite slow, especially if the context has a lot of category bits set. This can cause quite noticeable performance impact in situations where the translation needs to be done repeatedly. A common example is a UNIX datagram socket with the SO_PASSSEC option enabled, which is used e.g. by systemd-journald when receiving log messages via datagram socket. This scenario can be reproduced with: cat /dev/urandom | base64 | logger & timeout 30s perf record -p $(pidof systemd-journald) -a -g kill %1 perf report -g none --pretty raw | grep security_secid_to_secctx Before the caching introduced by this patch, computing the context string (security_secid_to_secctx() function) takes up ~65% of systemd-journald's CPU time (assuming a context with 1024 categories set and Fedora x86_64 release kernel configs). After this patch (assuming near-perfect cache hit ratio) this overhead is reduced to just ~2%. This patch addresses the issue by caching a certain number (compile-time configurable) of recently used context strings to speed up repeated translations of the same context, while using only a small amount of memory. The cache is integrated into the existing sidtab table by adding a field to each entry, which when not NULL contains an RCU-protected pointer to a cache entry containing the cached string. The cache entries are kept in a linked list sorted according to how recently they were used. On a cache miss when the cache is full, the least recently used entry is removed to make space for the new entry. The patch migrates security_sid_to_context_core() to use the cache (also a few other functions where it was possible without too much fuss, but these mostly use the translation for logging in case of error, which is rare). Link: https://bugzilla.redhat.com/show_bug.cgi?id=1733259 Cc: Michal Sekletar <msekleta@redhat.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> Reviewed-by: Paul E. McKenney <paulmck@kernel.org> [PM: lots of merge fixups due to collisions with other sidtab patches] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-26 21:57:00 +08:00
entry = sidtab_lookup(s, sid_to_index(sid));
else
selinux: cache the SID -> context string translation Translating a context struct to string can be quite slow, especially if the context has a lot of category bits set. This can cause quite noticeable performance impact in situations where the translation needs to be done repeatedly. A common example is a UNIX datagram socket with the SO_PASSSEC option enabled, which is used e.g. by systemd-journald when receiving log messages via datagram socket. This scenario can be reproduced with: cat /dev/urandom | base64 | logger & timeout 30s perf record -p $(pidof systemd-journald) -a -g kill %1 perf report -g none --pretty raw | grep security_secid_to_secctx Before the caching introduced by this patch, computing the context string (security_secid_to_secctx() function) takes up ~65% of systemd-journald's CPU time (assuming a context with 1024 categories set and Fedora x86_64 release kernel configs). After this patch (assuming near-perfect cache hit ratio) this overhead is reduced to just ~2%. This patch addresses the issue by caching a certain number (compile-time configurable) of recently used context strings to speed up repeated translations of the same context, while using only a small amount of memory. The cache is integrated into the existing sidtab table by adding a field to each entry, which when not NULL contains an RCU-protected pointer to a cache entry containing the cached string. The cache entries are kept in a linked list sorted according to how recently they were used. On a cache miss when the cache is full, the least recently used entry is removed to make space for the new entry. The patch migrates security_sid_to_context_core() to use the cache (also a few other functions where it was possible without too much fuss, but these mostly use the translation for logging in case of error, which is rare). Link: https://bugzilla.redhat.com/show_bug.cgi?id=1733259 Cc: Michal Sekletar <msekleta@redhat.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> Reviewed-by: Paul E. McKenney <paulmck@kernel.org> [PM: lots of merge fixups due to collisions with other sidtab patches] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-26 21:57:00 +08:00
entry = sidtab_lookup_initial(s, sid);
if (entry && (!entry->context.len || force))
return entry;
}
return sidtab_lookup_initial(s, SECINITSID_UNLABELED);
}
selinux: cache the SID -> context string translation Translating a context struct to string can be quite slow, especially if the context has a lot of category bits set. This can cause quite noticeable performance impact in situations where the translation needs to be done repeatedly. A common example is a UNIX datagram socket with the SO_PASSSEC option enabled, which is used e.g. by systemd-journald when receiving log messages via datagram socket. This scenario can be reproduced with: cat /dev/urandom | base64 | logger & timeout 30s perf record -p $(pidof systemd-journald) -a -g kill %1 perf report -g none --pretty raw | grep security_secid_to_secctx Before the caching introduced by this patch, computing the context string (security_secid_to_secctx() function) takes up ~65% of systemd-journald's CPU time (assuming a context with 1024 categories set and Fedora x86_64 release kernel configs). After this patch (assuming near-perfect cache hit ratio) this overhead is reduced to just ~2%. This patch addresses the issue by caching a certain number (compile-time configurable) of recently used context strings to speed up repeated translations of the same context, while using only a small amount of memory. The cache is integrated into the existing sidtab table by adding a field to each entry, which when not NULL contains an RCU-protected pointer to a cache entry containing the cached string. The cache entries are kept in a linked list sorted according to how recently they were used. On a cache miss when the cache is full, the least recently used entry is removed to make space for the new entry. The patch migrates security_sid_to_context_core() to use the cache (also a few other functions where it was possible without too much fuss, but these mostly use the translation for logging in case of error, which is rare). Link: https://bugzilla.redhat.com/show_bug.cgi?id=1733259 Cc: Michal Sekletar <msekleta@redhat.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> Reviewed-by: Paul E. McKenney <paulmck@kernel.org> [PM: lots of merge fixups due to collisions with other sidtab patches] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-26 21:57:00 +08:00
struct sidtab_entry *sidtab_search_entry(struct sidtab *s, u32 sid)
selinux: support deferred mapping of contexts Introduce SELinux support for deferred mapping of security contexts in the SID table upon policy reload, and use this support for inode security contexts when the context is not yet valid under the current policy. Only processes with CAP_MAC_ADMIN + mac_admin permission in policy can set undefined security contexts on inodes. Inodes with such undefined contexts are treated as having the unlabeled context until the context becomes valid upon a policy reload that defines the context. Context invalidation upon policy reload also uses this support to save the context information in the SID table and later recover it upon a subsequent policy reload that defines the context again. This support is to enable package managers and similar programs to set down file contexts unknown to the system policy at the time the file is created in order to better support placing loadable policy modules in packages and to support build systems that need to create images of different distro releases with different policies w/o requiring all of the contexts to be defined or legal in the build host policy. With this patch applied, the following sequence is possible, although in practice it is recommended that this permission only be allowed to specific program domains such as the package manager. # rmdir baz # rm bar # touch bar # chcon -t foo_exec_t bar # foo_exec_t is not yet defined chcon: failed to change context of `bar' to `system_u:object_r:foo_exec_t': Invalid argument # mkdir -Z system_u:object_r:foo_exec_t baz mkdir: failed to set default file creation context to `system_u:object_r:foo_exec_t': Invalid argument # cat setundefined.te policy_module(setundefined, 1.0) require { type unconfined_t; type unlabeled_t; } files_type(unlabeled_t) allow unconfined_t self:capability2 mac_admin; # make -f /usr/share/selinux/devel/Makefile setundefined.pp # semodule -i setundefined.pp # chcon -t foo_exec_t bar # foo_exec_t is not yet defined # mkdir -Z system_u:object_r:foo_exec_t baz # ls -Zd bar baz -rw-r--r-- root root system_u:object_r:unlabeled_t bar drwxr-xr-x root root system_u:object_r:unlabeled_t baz # cat foo.te policy_module(foo, 1.0) type foo_exec_t; files_type(foo_exec_t) # make -f /usr/share/selinux/devel/Makefile foo.pp # semodule -i foo.pp # defines foo_exec_t # ls -Zd bar baz -rw-r--r-- root root user_u:object_r:foo_exec_t bar drwxr-xr-x root root system_u:object_r:foo_exec_t baz # semodule -r foo # ls -Zd bar baz -rw-r--r-- root root system_u:object_r:unlabeled_t bar drwxr-xr-x root root system_u:object_r:unlabeled_t baz # semodule -i foo.pp # ls -Zd bar baz -rw-r--r-- root root user_u:object_r:foo_exec_t bar drwxr-xr-x root root system_u:object_r:foo_exec_t baz # semodule -r setundefined foo # chcon -t foo_exec_t bar # no longer defined and not allowed chcon: failed to change context of `bar' to `system_u:object_r:foo_exec_t': Invalid argument # rmdir baz # mkdir -Z system_u:object_r:foo_exec_t baz mkdir: failed to set default file creation context to `system_u:object_r:foo_exec_t': Invalid argument Signed-off-by: Stephen Smalley <sds@tycho.nsa.gov> Signed-off-by: James Morris <jmorris@namei.org>
2008-05-08 01:03:20 +08:00
{
return sidtab_search_core(s, sid, 0);
}
selinux: cache the SID -> context string translation Translating a context struct to string can be quite slow, especially if the context has a lot of category bits set. This can cause quite noticeable performance impact in situations where the translation needs to be done repeatedly. A common example is a UNIX datagram socket with the SO_PASSSEC option enabled, which is used e.g. by systemd-journald when receiving log messages via datagram socket. This scenario can be reproduced with: cat /dev/urandom | base64 | logger & timeout 30s perf record -p $(pidof systemd-journald) -a -g kill %1 perf report -g none --pretty raw | grep security_secid_to_secctx Before the caching introduced by this patch, computing the context string (security_secid_to_secctx() function) takes up ~65% of systemd-journald's CPU time (assuming a context with 1024 categories set and Fedora x86_64 release kernel configs). After this patch (assuming near-perfect cache hit ratio) this overhead is reduced to just ~2%. This patch addresses the issue by caching a certain number (compile-time configurable) of recently used context strings to speed up repeated translations of the same context, while using only a small amount of memory. The cache is integrated into the existing sidtab table by adding a field to each entry, which when not NULL contains an RCU-protected pointer to a cache entry containing the cached string. The cache entries are kept in a linked list sorted according to how recently they were used. On a cache miss when the cache is full, the least recently used entry is removed to make space for the new entry. The patch migrates security_sid_to_context_core() to use the cache (also a few other functions where it was possible without too much fuss, but these mostly use the translation for logging in case of error, which is rare). Link: https://bugzilla.redhat.com/show_bug.cgi?id=1733259 Cc: Michal Sekletar <msekleta@redhat.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> Reviewed-by: Paul E. McKenney <paulmck@kernel.org> [PM: lots of merge fixups due to collisions with other sidtab patches] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-26 21:57:00 +08:00
struct sidtab_entry *sidtab_search_entry_force(struct sidtab *s, u32 sid)
selinux: support deferred mapping of contexts Introduce SELinux support for deferred mapping of security contexts in the SID table upon policy reload, and use this support for inode security contexts when the context is not yet valid under the current policy. Only processes with CAP_MAC_ADMIN + mac_admin permission in policy can set undefined security contexts on inodes. Inodes with such undefined contexts are treated as having the unlabeled context until the context becomes valid upon a policy reload that defines the context. Context invalidation upon policy reload also uses this support to save the context information in the SID table and later recover it upon a subsequent policy reload that defines the context again. This support is to enable package managers and similar programs to set down file contexts unknown to the system policy at the time the file is created in order to better support placing loadable policy modules in packages and to support build systems that need to create images of different distro releases with different policies w/o requiring all of the contexts to be defined or legal in the build host policy. With this patch applied, the following sequence is possible, although in practice it is recommended that this permission only be allowed to specific program domains such as the package manager. # rmdir baz # rm bar # touch bar # chcon -t foo_exec_t bar # foo_exec_t is not yet defined chcon: failed to change context of `bar' to `system_u:object_r:foo_exec_t': Invalid argument # mkdir -Z system_u:object_r:foo_exec_t baz mkdir: failed to set default file creation context to `system_u:object_r:foo_exec_t': Invalid argument # cat setundefined.te policy_module(setundefined, 1.0) require { type unconfined_t; type unlabeled_t; } files_type(unlabeled_t) allow unconfined_t self:capability2 mac_admin; # make -f /usr/share/selinux/devel/Makefile setundefined.pp # semodule -i setundefined.pp # chcon -t foo_exec_t bar # foo_exec_t is not yet defined # mkdir -Z system_u:object_r:foo_exec_t baz # ls -Zd bar baz -rw-r--r-- root root system_u:object_r:unlabeled_t bar drwxr-xr-x root root system_u:object_r:unlabeled_t baz # cat foo.te policy_module(foo, 1.0) type foo_exec_t; files_type(foo_exec_t) # make -f /usr/share/selinux/devel/Makefile foo.pp # semodule -i foo.pp # defines foo_exec_t # ls -Zd bar baz -rw-r--r-- root root user_u:object_r:foo_exec_t bar drwxr-xr-x root root system_u:object_r:foo_exec_t baz # semodule -r foo # ls -Zd bar baz -rw-r--r-- root root system_u:object_r:unlabeled_t bar drwxr-xr-x root root system_u:object_r:unlabeled_t baz # semodule -i foo.pp # ls -Zd bar baz -rw-r--r-- root root user_u:object_r:foo_exec_t bar drwxr-xr-x root root system_u:object_r:foo_exec_t baz # semodule -r setundefined foo # chcon -t foo_exec_t bar # no longer defined and not allowed chcon: failed to change context of `bar' to `system_u:object_r:foo_exec_t': Invalid argument # rmdir baz # mkdir -Z system_u:object_r:foo_exec_t baz mkdir: failed to set default file creation context to `system_u:object_r:foo_exec_t': Invalid argument Signed-off-by: Stephen Smalley <sds@tycho.nsa.gov> Signed-off-by: James Morris <jmorris@namei.org>
2008-05-08 01:03:20 +08:00
{
return sidtab_search_core(s, sid, 1);
}
selinux: sidtab reverse lookup hash table This replaces the reverse table lookup and reverse cache with a hashtable which improves cache-miss reverse-lookup times from O(n) to O(1)* and maintains the same performance as a reverse cache hit. This reduces the time needed to add a new sidtab entry from ~500us to 5us on a Pixel 3 when there are ~10,000 sidtab entries. The implementation uses the kernel's generic hashtable API, It uses the context's string represtation as the hash source, and the kernels generic string hashing algorithm full_name_hash() to reduce the string to a 32 bit value. This change also maintains the improvement introduced in commit ee1a84fdfeed ("selinux: overhaul sidtab to fix bug and improve performance") which removed the need to keep the current sidtab locked during policy reload. It does however introduce periodic locking of the target sidtab while converting the hashtable. Sidtab entries are never modified or removed, so the context struct stored in the sid_to_context tree can also be used for the context_to_sid hashtable to reduce memory usage. This bug was reported by: - On the selinux bug tracker. BUG: kernel softlockup due to too many SIDs/contexts #37 https://github.com/SELinuxProject/selinux-kernel/issues/37 - Jovana Knezevic on Android's bugtracker. Bug: 140252993 "During multi-user performance testing, we create and remove users many times. selinux_android_restorecon_pkgdir goes from 1ms to over 20ms after about 200 user creations and removals. Accumulated over ~280 packages, that adds a significant time to user creation, making perf benchmarks unreliable." * Hashtable lookup is only O(1) when n < the number of buckets. Signed-off-by: Jeff Vander Stoep <jeffv@google.com> Reported-by: Stephen Smalley <sds@tycho.nsa.gov> Reported-by: Jovana Knezevic <jovanak@google.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: subj tweak, removed changelog from patch description] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-22 17:33:06 +08:00
int sidtab_context_to_sid(struct sidtab *s, struct context *context,
u32 *sid)
{
unsigned long flags;
u32 count, hash = context_compute_hash(context);
selinux: overhaul sidtab to fix bug and improve performance Before this patch, during a policy reload the sidtab would become frozen and trying to map a new context to SID would be unable to add a new entry to sidtab and fail with -ENOMEM. Such failures are usually propagated into userspace, which has no way of distignuishing them from actual allocation failures and thus doesn't handle them gracefully. Such situation can be triggered e.g. by the following reproducer: while true; do load_policy; echo -n .; sleep 0.1; done & for (( i = 0; i < 1024; i++ )); do runcon -l s0:c$i echo -n x || break # or: # chcon -l s0:c$i <some_file> || break done This patch overhauls the sidtab so it doesn't need to be frozen during policy reload, thus solving the above problem. The new SID table leverages the fact that SIDs are allocated sequentially and are never invalidated and stores them in linear buckets indexed by a tree structure. This brings several advantages: 1. Fast SID -> context lookup - this lookup can now be done in logarithmic time complexity (usually in less than 4 array lookups) and can still be done safely without locking. 2. No need to re-search the whole table on reverse lookup miss - after acquiring the spinlock only the newly added entries need to be searched, which means that reverse lookups that end up inserting a new entry are now about twice as fast. 3. No need to freeze sidtab during policy reload - it is now possible to handle insertion of new entries even during sidtab conversion. The tree structure of the new sidtab is able to grow automatically to up to about 2^31 entries (at which point it should not have more than about 4 tree levels). The old sidtab had a theoretical capacity of almost 2^32 entries, but half of that is still more than enough since by that point the reverse table lookups would become unusably slow anyway... The number of entries per tree node is selected automatically so that each node fits into a single page, which should be the easiest size for kmalloc() to handle. Note that the cache for reverse lookup is preserved with equivalent logic. The only difference is that instead of storing pointers to the hash table nodes it stores just the indices of the cached entries. The new cache ensures that the indices are loaded/stored atomically, but it still has the drawback that concurrent cache updates may mess up the contents of the cache. Such situation however only reduces its effectivity, not the correctness of lookups. Tested by selinux-testsuite and thoroughly tortured by this simple stress test: ``` function rand_cat() { echo $(( $RANDOM % 1024 )) } function do_work() { while true; do echo -n "system_u:system_r:kernel_t:s0:c$(rand_cat),c$(rand_cat)" \ >/sys/fs/selinux/context 2>/dev/null || true done } do_work >/dev/null & do_work >/dev/null & do_work >/dev/null & while load_policy; do echo -n .; sleep 0.1; done kill %1 kill %2 kill %3 ``` Link: https://github.com/SELinuxProject/selinux-kernel/issues/38 Reported-by: Orion Poplawski <orion@nwra.com> Reported-by: Li Kun <hw.likun@huawei.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: most of sidtab.c merged by hand due to conflicts] [PM: checkpatch fixes in mls.c, services.c, sidtab.c] Signed-off-by: Paul Moore <paul@paul-moore.com>
2018-11-30 23:24:08 +08:00
struct sidtab_convert_params *convert;
selinux: cache the SID -> context string translation Translating a context struct to string can be quite slow, especially if the context has a lot of category bits set. This can cause quite noticeable performance impact in situations where the translation needs to be done repeatedly. A common example is a UNIX datagram socket with the SO_PASSSEC option enabled, which is used e.g. by systemd-journald when receiving log messages via datagram socket. This scenario can be reproduced with: cat /dev/urandom | base64 | logger & timeout 30s perf record -p $(pidof systemd-journald) -a -g kill %1 perf report -g none --pretty raw | grep security_secid_to_secctx Before the caching introduced by this patch, computing the context string (security_secid_to_secctx() function) takes up ~65% of systemd-journald's CPU time (assuming a context with 1024 categories set and Fedora x86_64 release kernel configs). After this patch (assuming near-perfect cache hit ratio) this overhead is reduced to just ~2%. This patch addresses the issue by caching a certain number (compile-time configurable) of recently used context strings to speed up repeated translations of the same context, while using only a small amount of memory. The cache is integrated into the existing sidtab table by adding a field to each entry, which when not NULL contains an RCU-protected pointer to a cache entry containing the cached string. The cache entries are kept in a linked list sorted according to how recently they were used. On a cache miss when the cache is full, the least recently used entry is removed to make space for the new entry. The patch migrates security_sid_to_context_core() to use the cache (also a few other functions where it was possible without too much fuss, but these mostly use the translation for logging in case of error, which is rare). Link: https://bugzilla.redhat.com/show_bug.cgi?id=1733259 Cc: Michal Sekletar <msekleta@redhat.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> Reviewed-by: Paul E. McKenney <paulmck@kernel.org> [PM: lots of merge fixups due to collisions with other sidtab patches] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-26 21:57:00 +08:00
struct sidtab_entry *dst, *dst_convert;
selinux: overhaul sidtab to fix bug and improve performance Before this patch, during a policy reload the sidtab would become frozen and trying to map a new context to SID would be unable to add a new entry to sidtab and fail with -ENOMEM. Such failures are usually propagated into userspace, which has no way of distignuishing them from actual allocation failures and thus doesn't handle them gracefully. Such situation can be triggered e.g. by the following reproducer: while true; do load_policy; echo -n .; sleep 0.1; done & for (( i = 0; i < 1024; i++ )); do runcon -l s0:c$i echo -n x || break # or: # chcon -l s0:c$i <some_file> || break done This patch overhauls the sidtab so it doesn't need to be frozen during policy reload, thus solving the above problem. The new SID table leverages the fact that SIDs are allocated sequentially and are never invalidated and stores them in linear buckets indexed by a tree structure. This brings several advantages: 1. Fast SID -> context lookup - this lookup can now be done in logarithmic time complexity (usually in less than 4 array lookups) and can still be done safely without locking. 2. No need to re-search the whole table on reverse lookup miss - after acquiring the spinlock only the newly added entries need to be searched, which means that reverse lookups that end up inserting a new entry are now about twice as fast. 3. No need to freeze sidtab during policy reload - it is now possible to handle insertion of new entries even during sidtab conversion. The tree structure of the new sidtab is able to grow automatically to up to about 2^31 entries (at which point it should not have more than about 4 tree levels). The old sidtab had a theoretical capacity of almost 2^32 entries, but half of that is still more than enough since by that point the reverse table lookups would become unusably slow anyway... The number of entries per tree node is selected automatically so that each node fits into a single page, which should be the easiest size for kmalloc() to handle. Note that the cache for reverse lookup is preserved with equivalent logic. The only difference is that instead of storing pointers to the hash table nodes it stores just the indices of the cached entries. The new cache ensures that the indices are loaded/stored atomically, but it still has the drawback that concurrent cache updates may mess up the contents of the cache. Such situation however only reduces its effectivity, not the correctness of lookups. Tested by selinux-testsuite and thoroughly tortured by this simple stress test: ``` function rand_cat() { echo $(( $RANDOM % 1024 )) } function do_work() { while true; do echo -n "system_u:system_r:kernel_t:s0:c$(rand_cat),c$(rand_cat)" \ >/sys/fs/selinux/context 2>/dev/null || true done } do_work >/dev/null & do_work >/dev/null & do_work >/dev/null & while load_policy; do echo -n .; sleep 0.1; done kill %1 kill %2 kill %3 ``` Link: https://github.com/SELinuxProject/selinux-kernel/issues/38 Reported-by: Orion Poplawski <orion@nwra.com> Reported-by: Li Kun <hw.likun@huawei.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: most of sidtab.c merged by hand due to conflicts] [PM: checkpatch fixes in mls.c, services.c, sidtab.c] Signed-off-by: Paul Moore <paul@paul-moore.com>
2018-11-30 23:24:08 +08:00
int rc;
*sid = context_to_sid(s, context, hash);
selinux: sidtab reverse lookup hash table This replaces the reverse table lookup and reverse cache with a hashtable which improves cache-miss reverse-lookup times from O(n) to O(1)* and maintains the same performance as a reverse cache hit. This reduces the time needed to add a new sidtab entry from ~500us to 5us on a Pixel 3 when there are ~10,000 sidtab entries. The implementation uses the kernel's generic hashtable API, It uses the context's string represtation as the hash source, and the kernels generic string hashing algorithm full_name_hash() to reduce the string to a 32 bit value. This change also maintains the improvement introduced in commit ee1a84fdfeed ("selinux: overhaul sidtab to fix bug and improve performance") which removed the need to keep the current sidtab locked during policy reload. It does however introduce periodic locking of the target sidtab while converting the hashtable. Sidtab entries are never modified or removed, so the context struct stored in the sid_to_context tree can also be used for the context_to_sid hashtable to reduce memory usage. This bug was reported by: - On the selinux bug tracker. BUG: kernel softlockup due to too many SIDs/contexts #37 https://github.com/SELinuxProject/selinux-kernel/issues/37 - Jovana Knezevic on Android's bugtracker. Bug: 140252993 "During multi-user performance testing, we create and remove users many times. selinux_android_restorecon_pkgdir goes from 1ms to over 20ms after about 200 user creations and removals. Accumulated over ~280 packages, that adds a significant time to user creation, making perf benchmarks unreliable." * Hashtable lookup is only O(1) when n < the number of buckets. Signed-off-by: Jeff Vander Stoep <jeffv@google.com> Reported-by: Stephen Smalley <sds@tycho.nsa.gov> Reported-by: Jovana Knezevic <jovanak@google.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: subj tweak, removed changelog from patch description] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-22 17:33:06 +08:00
if (*sid)
selinux: overhaul sidtab to fix bug and improve performance Before this patch, during a policy reload the sidtab would become frozen and trying to map a new context to SID would be unable to add a new entry to sidtab and fail with -ENOMEM. Such failures are usually propagated into userspace, which has no way of distignuishing them from actual allocation failures and thus doesn't handle them gracefully. Such situation can be triggered e.g. by the following reproducer: while true; do load_policy; echo -n .; sleep 0.1; done & for (( i = 0; i < 1024; i++ )); do runcon -l s0:c$i echo -n x || break # or: # chcon -l s0:c$i <some_file> || break done This patch overhauls the sidtab so it doesn't need to be frozen during policy reload, thus solving the above problem. The new SID table leverages the fact that SIDs are allocated sequentially and are never invalidated and stores them in linear buckets indexed by a tree structure. This brings several advantages: 1. Fast SID -> context lookup - this lookup can now be done in logarithmic time complexity (usually in less than 4 array lookups) and can still be done safely without locking. 2. No need to re-search the whole table on reverse lookup miss - after acquiring the spinlock only the newly added entries need to be searched, which means that reverse lookups that end up inserting a new entry are now about twice as fast. 3. No need to freeze sidtab during policy reload - it is now possible to handle insertion of new entries even during sidtab conversion. The tree structure of the new sidtab is able to grow automatically to up to about 2^31 entries (at which point it should not have more than about 4 tree levels). The old sidtab had a theoretical capacity of almost 2^32 entries, but half of that is still more than enough since by that point the reverse table lookups would become unusably slow anyway... The number of entries per tree node is selected automatically so that each node fits into a single page, which should be the easiest size for kmalloc() to handle. Note that the cache for reverse lookup is preserved with equivalent logic. The only difference is that instead of storing pointers to the hash table nodes it stores just the indices of the cached entries. The new cache ensures that the indices are loaded/stored atomically, but it still has the drawback that concurrent cache updates may mess up the contents of the cache. Such situation however only reduces its effectivity, not the correctness of lookups. Tested by selinux-testsuite and thoroughly tortured by this simple stress test: ``` function rand_cat() { echo $(( $RANDOM % 1024 )) } function do_work() { while true; do echo -n "system_u:system_r:kernel_t:s0:c$(rand_cat),c$(rand_cat)" \ >/sys/fs/selinux/context 2>/dev/null || true done } do_work >/dev/null & do_work >/dev/null & do_work >/dev/null & while load_policy; do echo -n .; sleep 0.1; done kill %1 kill %2 kill %3 ``` Link: https://github.com/SELinuxProject/selinux-kernel/issues/38 Reported-by: Orion Poplawski <orion@nwra.com> Reported-by: Li Kun <hw.likun@huawei.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: most of sidtab.c merged by hand due to conflicts] [PM: checkpatch fixes in mls.c, services.c, sidtab.c] Signed-off-by: Paul Moore <paul@paul-moore.com>
2018-11-30 23:24:08 +08:00
return 0;
/* lock-free search failed: lock, re-search, and insert if not found */
spin_lock_irqsave(&s->lock, flags);
selinux: sidtab reverse lookup hash table This replaces the reverse table lookup and reverse cache with a hashtable which improves cache-miss reverse-lookup times from O(n) to O(1)* and maintains the same performance as a reverse cache hit. This reduces the time needed to add a new sidtab entry from ~500us to 5us on a Pixel 3 when there are ~10,000 sidtab entries. The implementation uses the kernel's generic hashtable API, It uses the context's string represtation as the hash source, and the kernels generic string hashing algorithm full_name_hash() to reduce the string to a 32 bit value. This change also maintains the improvement introduced in commit ee1a84fdfeed ("selinux: overhaul sidtab to fix bug and improve performance") which removed the need to keep the current sidtab locked during policy reload. It does however introduce periodic locking of the target sidtab while converting the hashtable. Sidtab entries are never modified or removed, so the context struct stored in the sid_to_context tree can also be used for the context_to_sid hashtable to reduce memory usage. This bug was reported by: - On the selinux bug tracker. BUG: kernel softlockup due to too many SIDs/contexts #37 https://github.com/SELinuxProject/selinux-kernel/issues/37 - Jovana Knezevic on Android's bugtracker. Bug: 140252993 "During multi-user performance testing, we create and remove users many times. selinux_android_restorecon_pkgdir goes from 1ms to over 20ms after about 200 user creations and removals. Accumulated over ~280 packages, that adds a significant time to user creation, making perf benchmarks unreliable." * Hashtable lookup is only O(1) when n < the number of buckets. Signed-off-by: Jeff Vander Stoep <jeffv@google.com> Reported-by: Stephen Smalley <sds@tycho.nsa.gov> Reported-by: Jovana Knezevic <jovanak@google.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: subj tweak, removed changelog from patch description] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-22 17:33:06 +08:00
rc = 0;
*sid = context_to_sid(s, context, hash);
selinux: sidtab reverse lookup hash table This replaces the reverse table lookup and reverse cache with a hashtable which improves cache-miss reverse-lookup times from O(n) to O(1)* and maintains the same performance as a reverse cache hit. This reduces the time needed to add a new sidtab entry from ~500us to 5us on a Pixel 3 when there are ~10,000 sidtab entries. The implementation uses the kernel's generic hashtable API, It uses the context's string represtation as the hash source, and the kernels generic string hashing algorithm full_name_hash() to reduce the string to a 32 bit value. This change also maintains the improvement introduced in commit ee1a84fdfeed ("selinux: overhaul sidtab to fix bug and improve performance") which removed the need to keep the current sidtab locked during policy reload. It does however introduce periodic locking of the target sidtab while converting the hashtable. Sidtab entries are never modified or removed, so the context struct stored in the sid_to_context tree can also be used for the context_to_sid hashtable to reduce memory usage. This bug was reported by: - On the selinux bug tracker. BUG: kernel softlockup due to too many SIDs/contexts #37 https://github.com/SELinuxProject/selinux-kernel/issues/37 - Jovana Knezevic on Android's bugtracker. Bug: 140252993 "During multi-user performance testing, we create and remove users many times. selinux_android_restorecon_pkgdir goes from 1ms to over 20ms after about 200 user creations and removals. Accumulated over ~280 packages, that adds a significant time to user creation, making perf benchmarks unreliable." * Hashtable lookup is only O(1) when n < the number of buckets. Signed-off-by: Jeff Vander Stoep <jeffv@google.com> Reported-by: Stephen Smalley <sds@tycho.nsa.gov> Reported-by: Jovana Knezevic <jovanak@google.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: subj tweak, removed changelog from patch description] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-22 17:33:06 +08:00
if (*sid)
goto out_unlock;
selinux: fix race between old and new sidtab commit 9ad6e9cb39c66366bf7b9aece114aca277981a1f upstream. Since commit 1b8b31a2e612 ("selinux: convert policy read-write lock to RCU"), there is a small window during policy load where the new policy pointer has already been installed, but some threads may still be holding the old policy pointer in their read-side RCU critical sections. This means that there may be conflicting attempts to add a new SID entry to both tables via sidtab_context_to_sid(). See also (and the rest of the thread): https://lore.kernel.org/selinux/CAFqZXNvfux46_f8gnvVvRYMKoes24nwm2n3sPbMjrB8vKTW00g@mail.gmail.com/ Fix this by installing the new policy pointer under the old sidtab's spinlock along with marking the old sidtab as "frozen". Then, if an attempt to add new entry to a "frozen" sidtab is detected, make sidtab_context_to_sid() return -ESTALE to indicate that a new policy has been installed and that the caller will have to abort the policy transaction and try again after re-taking the policy pointer (which is guaranteed to be a newer policy). This requires adding a retry-on-ESTALE logic to all callers of sidtab_context_to_sid(), but fortunately these are easy to determine and aren't that many. This seems to be the simplest solution for this problem, even if it looks somewhat ugly. Note that other places in the kernel (e.g. do_mknodat() in fs/namei.c) use similar stale-retry patterns, so I think it's reasonable. Cc: stable@vger.kernel.org Fixes: 1b8b31a2e612 ("selinux: convert policy read-write lock to RCU") Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Signed-off-by: Paul Moore <paul@paul-moore.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2021-04-07 15:24:43 +08:00
if (unlikely(s->frozen)) {
/*
* This sidtab is now frozen - tell the caller to abort and
* get the new one.
*/
rc = -ESTALE;
goto out_unlock;
}
count = s->count;
selinux: overhaul sidtab to fix bug and improve performance Before this patch, during a policy reload the sidtab would become frozen and trying to map a new context to SID would be unable to add a new entry to sidtab and fail with -ENOMEM. Such failures are usually propagated into userspace, which has no way of distignuishing them from actual allocation failures and thus doesn't handle them gracefully. Such situation can be triggered e.g. by the following reproducer: while true; do load_policy; echo -n .; sleep 0.1; done & for (( i = 0; i < 1024; i++ )); do runcon -l s0:c$i echo -n x || break # or: # chcon -l s0:c$i <some_file> || break done This patch overhauls the sidtab so it doesn't need to be frozen during policy reload, thus solving the above problem. The new SID table leverages the fact that SIDs are allocated sequentially and are never invalidated and stores them in linear buckets indexed by a tree structure. This brings several advantages: 1. Fast SID -> context lookup - this lookup can now be done in logarithmic time complexity (usually in less than 4 array lookups) and can still be done safely without locking. 2. No need to re-search the whole table on reverse lookup miss - after acquiring the spinlock only the newly added entries need to be searched, which means that reverse lookups that end up inserting a new entry are now about twice as fast. 3. No need to freeze sidtab during policy reload - it is now possible to handle insertion of new entries even during sidtab conversion. The tree structure of the new sidtab is able to grow automatically to up to about 2^31 entries (at which point it should not have more than about 4 tree levels). The old sidtab had a theoretical capacity of almost 2^32 entries, but half of that is still more than enough since by that point the reverse table lookups would become unusably slow anyway... The number of entries per tree node is selected automatically so that each node fits into a single page, which should be the easiest size for kmalloc() to handle. Note that the cache for reverse lookup is preserved with equivalent logic. The only difference is that instead of storing pointers to the hash table nodes it stores just the indices of the cached entries. The new cache ensures that the indices are loaded/stored atomically, but it still has the drawback that concurrent cache updates may mess up the contents of the cache. Such situation however only reduces its effectivity, not the correctness of lookups. Tested by selinux-testsuite and thoroughly tortured by this simple stress test: ``` function rand_cat() { echo $(( $RANDOM % 1024 )) } function do_work() { while true; do echo -n "system_u:system_r:kernel_t:s0:c$(rand_cat),c$(rand_cat)" \ >/sys/fs/selinux/context 2>/dev/null || true done } do_work >/dev/null & do_work >/dev/null & do_work >/dev/null & while load_policy; do echo -n .; sleep 0.1; done kill %1 kill %2 kill %3 ``` Link: https://github.com/SELinuxProject/selinux-kernel/issues/38 Reported-by: Orion Poplawski <orion@nwra.com> Reported-by: Li Kun <hw.likun@huawei.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: most of sidtab.c merged by hand due to conflicts] [PM: checkpatch fixes in mls.c, services.c, sidtab.c] Signed-off-by: Paul Moore <paul@paul-moore.com>
2018-11-30 23:24:08 +08:00
convert = s->convert;
/* bail out if we already reached max entries */
rc = -EOVERFLOW;
if (count >= SIDTAB_MAX)
goto out_unlock;
selinux: overhaul sidtab to fix bug and improve performance Before this patch, during a policy reload the sidtab would become frozen and trying to map a new context to SID would be unable to add a new entry to sidtab and fail with -ENOMEM. Such failures are usually propagated into userspace, which has no way of distignuishing them from actual allocation failures and thus doesn't handle them gracefully. Such situation can be triggered e.g. by the following reproducer: while true; do load_policy; echo -n .; sleep 0.1; done & for (( i = 0; i < 1024; i++ )); do runcon -l s0:c$i echo -n x || break # or: # chcon -l s0:c$i <some_file> || break done This patch overhauls the sidtab so it doesn't need to be frozen during policy reload, thus solving the above problem. The new SID table leverages the fact that SIDs are allocated sequentially and are never invalidated and stores them in linear buckets indexed by a tree structure. This brings several advantages: 1. Fast SID -> context lookup - this lookup can now be done in logarithmic time complexity (usually in less than 4 array lookups) and can still be done safely without locking. 2. No need to re-search the whole table on reverse lookup miss - after acquiring the spinlock only the newly added entries need to be searched, which means that reverse lookups that end up inserting a new entry are now about twice as fast. 3. No need to freeze sidtab during policy reload - it is now possible to handle insertion of new entries even during sidtab conversion. The tree structure of the new sidtab is able to grow automatically to up to about 2^31 entries (at which point it should not have more than about 4 tree levels). The old sidtab had a theoretical capacity of almost 2^32 entries, but half of that is still more than enough since by that point the reverse table lookups would become unusably slow anyway... The number of entries per tree node is selected automatically so that each node fits into a single page, which should be the easiest size for kmalloc() to handle. Note that the cache for reverse lookup is preserved with equivalent logic. The only difference is that instead of storing pointers to the hash table nodes it stores just the indices of the cached entries. The new cache ensures that the indices are loaded/stored atomically, but it still has the drawback that concurrent cache updates may mess up the contents of the cache. Such situation however only reduces its effectivity, not the correctness of lookups. Tested by selinux-testsuite and thoroughly tortured by this simple stress test: ``` function rand_cat() { echo $(( $RANDOM % 1024 )) } function do_work() { while true; do echo -n "system_u:system_r:kernel_t:s0:c$(rand_cat),c$(rand_cat)" \ >/sys/fs/selinux/context 2>/dev/null || true done } do_work >/dev/null & do_work >/dev/null & do_work >/dev/null & while load_policy; do echo -n .; sleep 0.1; done kill %1 kill %2 kill %3 ``` Link: https://github.com/SELinuxProject/selinux-kernel/issues/38 Reported-by: Orion Poplawski <orion@nwra.com> Reported-by: Li Kun <hw.likun@huawei.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: most of sidtab.c merged by hand due to conflicts] [PM: checkpatch fixes in mls.c, services.c, sidtab.c] Signed-off-by: Paul Moore <paul@paul-moore.com>
2018-11-30 23:24:08 +08:00
/* insert context into new entry */
rc = -ENOMEM;
dst = sidtab_do_lookup(s, count, 1);
if (!dst)
goto out_unlock;
selinux: sidtab reverse lookup hash table This replaces the reverse table lookup and reverse cache with a hashtable which improves cache-miss reverse-lookup times from O(n) to O(1)* and maintains the same performance as a reverse cache hit. This reduces the time needed to add a new sidtab entry from ~500us to 5us on a Pixel 3 when there are ~10,000 sidtab entries. The implementation uses the kernel's generic hashtable API, It uses the context's string represtation as the hash source, and the kernels generic string hashing algorithm full_name_hash() to reduce the string to a 32 bit value. This change also maintains the improvement introduced in commit ee1a84fdfeed ("selinux: overhaul sidtab to fix bug and improve performance") which removed the need to keep the current sidtab locked during policy reload. It does however introduce periodic locking of the target sidtab while converting the hashtable. Sidtab entries are never modified or removed, so the context struct stored in the sid_to_context tree can also be used for the context_to_sid hashtable to reduce memory usage. This bug was reported by: - On the selinux bug tracker. BUG: kernel softlockup due to too many SIDs/contexts #37 https://github.com/SELinuxProject/selinux-kernel/issues/37 - Jovana Knezevic on Android's bugtracker. Bug: 140252993 "During multi-user performance testing, we create and remove users many times. selinux_android_restorecon_pkgdir goes from 1ms to over 20ms after about 200 user creations and removals. Accumulated over ~280 packages, that adds a significant time to user creation, making perf benchmarks unreliable." * Hashtable lookup is only O(1) when n < the number of buckets. Signed-off-by: Jeff Vander Stoep <jeffv@google.com> Reported-by: Stephen Smalley <sds@tycho.nsa.gov> Reported-by: Jovana Knezevic <jovanak@google.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: subj tweak, removed changelog from patch description] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-22 17:33:06 +08:00
dst->sid = index_to_sid(count);
dst->hash = hash;
selinux: sidtab reverse lookup hash table This replaces the reverse table lookup and reverse cache with a hashtable which improves cache-miss reverse-lookup times from O(n) to O(1)* and maintains the same performance as a reverse cache hit. This reduces the time needed to add a new sidtab entry from ~500us to 5us on a Pixel 3 when there are ~10,000 sidtab entries. The implementation uses the kernel's generic hashtable API, It uses the context's string represtation as the hash source, and the kernels generic string hashing algorithm full_name_hash() to reduce the string to a 32 bit value. This change also maintains the improvement introduced in commit ee1a84fdfeed ("selinux: overhaul sidtab to fix bug and improve performance") which removed the need to keep the current sidtab locked during policy reload. It does however introduce periodic locking of the target sidtab while converting the hashtable. Sidtab entries are never modified or removed, so the context struct stored in the sid_to_context tree can also be used for the context_to_sid hashtable to reduce memory usage. This bug was reported by: - On the selinux bug tracker. BUG: kernel softlockup due to too many SIDs/contexts #37 https://github.com/SELinuxProject/selinux-kernel/issues/37 - Jovana Knezevic on Android's bugtracker. Bug: 140252993 "During multi-user performance testing, we create and remove users many times. selinux_android_restorecon_pkgdir goes from 1ms to over 20ms after about 200 user creations and removals. Accumulated over ~280 packages, that adds a significant time to user creation, making perf benchmarks unreliable." * Hashtable lookup is only O(1) when n < the number of buckets. Signed-off-by: Jeff Vander Stoep <jeffv@google.com> Reported-by: Stephen Smalley <sds@tycho.nsa.gov> Reported-by: Jovana Knezevic <jovanak@google.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: subj tweak, removed changelog from patch description] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-22 17:33:06 +08:00
rc = context_cpy(&dst->context, context);
selinux: overhaul sidtab to fix bug and improve performance Before this patch, during a policy reload the sidtab would become frozen and trying to map a new context to SID would be unable to add a new entry to sidtab and fail with -ENOMEM. Such failures are usually propagated into userspace, which has no way of distignuishing them from actual allocation failures and thus doesn't handle them gracefully. Such situation can be triggered e.g. by the following reproducer: while true; do load_policy; echo -n .; sleep 0.1; done & for (( i = 0; i < 1024; i++ )); do runcon -l s0:c$i echo -n x || break # or: # chcon -l s0:c$i <some_file> || break done This patch overhauls the sidtab so it doesn't need to be frozen during policy reload, thus solving the above problem. The new SID table leverages the fact that SIDs are allocated sequentially and are never invalidated and stores them in linear buckets indexed by a tree structure. This brings several advantages: 1. Fast SID -> context lookup - this lookup can now be done in logarithmic time complexity (usually in less than 4 array lookups) and can still be done safely without locking. 2. No need to re-search the whole table on reverse lookup miss - after acquiring the spinlock only the newly added entries need to be searched, which means that reverse lookups that end up inserting a new entry are now about twice as fast. 3. No need to freeze sidtab during policy reload - it is now possible to handle insertion of new entries even during sidtab conversion. The tree structure of the new sidtab is able to grow automatically to up to about 2^31 entries (at which point it should not have more than about 4 tree levels). The old sidtab had a theoretical capacity of almost 2^32 entries, but half of that is still more than enough since by that point the reverse table lookups would become unusably slow anyway... The number of entries per tree node is selected automatically so that each node fits into a single page, which should be the easiest size for kmalloc() to handle. Note that the cache for reverse lookup is preserved with equivalent logic. The only difference is that instead of storing pointers to the hash table nodes it stores just the indices of the cached entries. The new cache ensures that the indices are loaded/stored atomically, but it still has the drawback that concurrent cache updates may mess up the contents of the cache. Such situation however only reduces its effectivity, not the correctness of lookups. Tested by selinux-testsuite and thoroughly tortured by this simple stress test: ``` function rand_cat() { echo $(( $RANDOM % 1024 )) } function do_work() { while true; do echo -n "system_u:system_r:kernel_t:s0:c$(rand_cat),c$(rand_cat)" \ >/sys/fs/selinux/context 2>/dev/null || true done } do_work >/dev/null & do_work >/dev/null & do_work >/dev/null & while load_policy; do echo -n .; sleep 0.1; done kill %1 kill %2 kill %3 ``` Link: https://github.com/SELinuxProject/selinux-kernel/issues/38 Reported-by: Orion Poplawski <orion@nwra.com> Reported-by: Li Kun <hw.likun@huawei.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: most of sidtab.c merged by hand due to conflicts] [PM: checkpatch fixes in mls.c, services.c, sidtab.c] Signed-off-by: Paul Moore <paul@paul-moore.com>
2018-11-30 23:24:08 +08:00
if (rc)
goto out_unlock;
/*
* if we are building a new sidtab, we need to convert the context
* and insert it there as well
*/
if (convert) {
rc = -ENOMEM;
dst_convert = sidtab_do_lookup(convert->target, count, 1);
if (!dst_convert) {
selinux: sidtab reverse lookup hash table This replaces the reverse table lookup and reverse cache with a hashtable which improves cache-miss reverse-lookup times from O(n) to O(1)* and maintains the same performance as a reverse cache hit. This reduces the time needed to add a new sidtab entry from ~500us to 5us on a Pixel 3 when there are ~10,000 sidtab entries. The implementation uses the kernel's generic hashtable API, It uses the context's string represtation as the hash source, and the kernels generic string hashing algorithm full_name_hash() to reduce the string to a 32 bit value. This change also maintains the improvement introduced in commit ee1a84fdfeed ("selinux: overhaul sidtab to fix bug and improve performance") which removed the need to keep the current sidtab locked during policy reload. It does however introduce periodic locking of the target sidtab while converting the hashtable. Sidtab entries are never modified or removed, so the context struct stored in the sid_to_context tree can also be used for the context_to_sid hashtable to reduce memory usage. This bug was reported by: - On the selinux bug tracker. BUG: kernel softlockup due to too many SIDs/contexts #37 https://github.com/SELinuxProject/selinux-kernel/issues/37 - Jovana Knezevic on Android's bugtracker. Bug: 140252993 "During multi-user performance testing, we create and remove users many times. selinux_android_restorecon_pkgdir goes from 1ms to over 20ms after about 200 user creations and removals. Accumulated over ~280 packages, that adds a significant time to user creation, making perf benchmarks unreliable." * Hashtable lookup is only O(1) when n < the number of buckets. Signed-off-by: Jeff Vander Stoep <jeffv@google.com> Reported-by: Stephen Smalley <sds@tycho.nsa.gov> Reported-by: Jovana Knezevic <jovanak@google.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: subj tweak, removed changelog from patch description] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-22 17:33:06 +08:00
context_destroy(&dst->context);
selinux: overhaul sidtab to fix bug and improve performance Before this patch, during a policy reload the sidtab would become frozen and trying to map a new context to SID would be unable to add a new entry to sidtab and fail with -ENOMEM. Such failures are usually propagated into userspace, which has no way of distignuishing them from actual allocation failures and thus doesn't handle them gracefully. Such situation can be triggered e.g. by the following reproducer: while true; do load_policy; echo -n .; sleep 0.1; done & for (( i = 0; i < 1024; i++ )); do runcon -l s0:c$i echo -n x || break # or: # chcon -l s0:c$i <some_file> || break done This patch overhauls the sidtab so it doesn't need to be frozen during policy reload, thus solving the above problem. The new SID table leverages the fact that SIDs are allocated sequentially and are never invalidated and stores them in linear buckets indexed by a tree structure. This brings several advantages: 1. Fast SID -> context lookup - this lookup can now be done in logarithmic time complexity (usually in less than 4 array lookups) and can still be done safely without locking. 2. No need to re-search the whole table on reverse lookup miss - after acquiring the spinlock only the newly added entries need to be searched, which means that reverse lookups that end up inserting a new entry are now about twice as fast. 3. No need to freeze sidtab during policy reload - it is now possible to handle insertion of new entries even during sidtab conversion. The tree structure of the new sidtab is able to grow automatically to up to about 2^31 entries (at which point it should not have more than about 4 tree levels). The old sidtab had a theoretical capacity of almost 2^32 entries, but half of that is still more than enough since by that point the reverse table lookups would become unusably slow anyway... The number of entries per tree node is selected automatically so that each node fits into a single page, which should be the easiest size for kmalloc() to handle. Note that the cache for reverse lookup is preserved with equivalent logic. The only difference is that instead of storing pointers to the hash table nodes it stores just the indices of the cached entries. The new cache ensures that the indices are loaded/stored atomically, but it still has the drawback that concurrent cache updates may mess up the contents of the cache. Such situation however only reduces its effectivity, not the correctness of lookups. Tested by selinux-testsuite and thoroughly tortured by this simple stress test: ``` function rand_cat() { echo $(( $RANDOM % 1024 )) } function do_work() { while true; do echo -n "system_u:system_r:kernel_t:s0:c$(rand_cat),c$(rand_cat)" \ >/sys/fs/selinux/context 2>/dev/null || true done } do_work >/dev/null & do_work >/dev/null & do_work >/dev/null & while load_policy; do echo -n .; sleep 0.1; done kill %1 kill %2 kill %3 ``` Link: https://github.com/SELinuxProject/selinux-kernel/issues/38 Reported-by: Orion Poplawski <orion@nwra.com> Reported-by: Li Kun <hw.likun@huawei.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: most of sidtab.c merged by hand due to conflicts] [PM: checkpatch fixes in mls.c, services.c, sidtab.c] Signed-off-by: Paul Moore <paul@paul-moore.com>
2018-11-30 23:24:08 +08:00
goto out_unlock;
}
selinux: sidtab reverse lookup hash table This replaces the reverse table lookup and reverse cache with a hashtable which improves cache-miss reverse-lookup times from O(n) to O(1)* and maintains the same performance as a reverse cache hit. This reduces the time needed to add a new sidtab entry from ~500us to 5us on a Pixel 3 when there are ~10,000 sidtab entries. The implementation uses the kernel's generic hashtable API, It uses the context's string represtation as the hash source, and the kernels generic string hashing algorithm full_name_hash() to reduce the string to a 32 bit value. This change also maintains the improvement introduced in commit ee1a84fdfeed ("selinux: overhaul sidtab to fix bug and improve performance") which removed the need to keep the current sidtab locked during policy reload. It does however introduce periodic locking of the target sidtab while converting the hashtable. Sidtab entries are never modified or removed, so the context struct stored in the sid_to_context tree can also be used for the context_to_sid hashtable to reduce memory usage. This bug was reported by: - On the selinux bug tracker. BUG: kernel softlockup due to too many SIDs/contexts #37 https://github.com/SELinuxProject/selinux-kernel/issues/37 - Jovana Knezevic on Android's bugtracker. Bug: 140252993 "During multi-user performance testing, we create and remove users many times. selinux_android_restorecon_pkgdir goes from 1ms to over 20ms after about 200 user creations and removals. Accumulated over ~280 packages, that adds a significant time to user creation, making perf benchmarks unreliable." * Hashtable lookup is only O(1) when n < the number of buckets. Signed-off-by: Jeff Vander Stoep <jeffv@google.com> Reported-by: Stephen Smalley <sds@tycho.nsa.gov> Reported-by: Jovana Knezevic <jovanak@google.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: subj tweak, removed changelog from patch description] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-22 17:33:06 +08:00
rc = convert->func(context, &dst_convert->context,
selinux: cache the SID -> context string translation Translating a context struct to string can be quite slow, especially if the context has a lot of category bits set. This can cause quite noticeable performance impact in situations where the translation needs to be done repeatedly. A common example is a UNIX datagram socket with the SO_PASSSEC option enabled, which is used e.g. by systemd-journald when receiving log messages via datagram socket. This scenario can be reproduced with: cat /dev/urandom | base64 | logger & timeout 30s perf record -p $(pidof systemd-journald) -a -g kill %1 perf report -g none --pretty raw | grep security_secid_to_secctx Before the caching introduced by this patch, computing the context string (security_secid_to_secctx() function) takes up ~65% of systemd-journald's CPU time (assuming a context with 1024 categories set and Fedora x86_64 release kernel configs). After this patch (assuming near-perfect cache hit ratio) this overhead is reduced to just ~2%. This patch addresses the issue by caching a certain number (compile-time configurable) of recently used context strings to speed up repeated translations of the same context, while using only a small amount of memory. The cache is integrated into the existing sidtab table by adding a field to each entry, which when not NULL contains an RCU-protected pointer to a cache entry containing the cached string. The cache entries are kept in a linked list sorted according to how recently they were used. On a cache miss when the cache is full, the least recently used entry is removed to make space for the new entry. The patch migrates security_sid_to_context_core() to use the cache (also a few other functions where it was possible without too much fuss, but these mostly use the translation for logging in case of error, which is rare). Link: https://bugzilla.redhat.com/show_bug.cgi?id=1733259 Cc: Michal Sekletar <msekleta@redhat.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> Reviewed-by: Paul E. McKenney <paulmck@kernel.org> [PM: lots of merge fixups due to collisions with other sidtab patches] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-26 21:57:00 +08:00
convert->args);
selinux: overhaul sidtab to fix bug and improve performance Before this patch, during a policy reload the sidtab would become frozen and trying to map a new context to SID would be unable to add a new entry to sidtab and fail with -ENOMEM. Such failures are usually propagated into userspace, which has no way of distignuishing them from actual allocation failures and thus doesn't handle them gracefully. Such situation can be triggered e.g. by the following reproducer: while true; do load_policy; echo -n .; sleep 0.1; done & for (( i = 0; i < 1024; i++ )); do runcon -l s0:c$i echo -n x || break # or: # chcon -l s0:c$i <some_file> || break done This patch overhauls the sidtab so it doesn't need to be frozen during policy reload, thus solving the above problem. The new SID table leverages the fact that SIDs are allocated sequentially and are never invalidated and stores them in linear buckets indexed by a tree structure. This brings several advantages: 1. Fast SID -> context lookup - this lookup can now be done in logarithmic time complexity (usually in less than 4 array lookups) and can still be done safely without locking. 2. No need to re-search the whole table on reverse lookup miss - after acquiring the spinlock only the newly added entries need to be searched, which means that reverse lookups that end up inserting a new entry are now about twice as fast. 3. No need to freeze sidtab during policy reload - it is now possible to handle insertion of new entries even during sidtab conversion. The tree structure of the new sidtab is able to grow automatically to up to about 2^31 entries (at which point it should not have more than about 4 tree levels). The old sidtab had a theoretical capacity of almost 2^32 entries, but half of that is still more than enough since by that point the reverse table lookups would become unusably slow anyway... The number of entries per tree node is selected automatically so that each node fits into a single page, which should be the easiest size for kmalloc() to handle. Note that the cache for reverse lookup is preserved with equivalent logic. The only difference is that instead of storing pointers to the hash table nodes it stores just the indices of the cached entries. The new cache ensures that the indices are loaded/stored atomically, but it still has the drawback that concurrent cache updates may mess up the contents of the cache. Such situation however only reduces its effectivity, not the correctness of lookups. Tested by selinux-testsuite and thoroughly tortured by this simple stress test: ``` function rand_cat() { echo $(( $RANDOM % 1024 )) } function do_work() { while true; do echo -n "system_u:system_r:kernel_t:s0:c$(rand_cat),c$(rand_cat)" \ >/sys/fs/selinux/context 2>/dev/null || true done } do_work >/dev/null & do_work >/dev/null & do_work >/dev/null & while load_policy; do echo -n .; sleep 0.1; done kill %1 kill %2 kill %3 ``` Link: https://github.com/SELinuxProject/selinux-kernel/issues/38 Reported-by: Orion Poplawski <orion@nwra.com> Reported-by: Li Kun <hw.likun@huawei.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: most of sidtab.c merged by hand due to conflicts] [PM: checkpatch fixes in mls.c, services.c, sidtab.c] Signed-off-by: Paul Moore <paul@paul-moore.com>
2018-11-30 23:24:08 +08:00
if (rc) {
selinux: sidtab reverse lookup hash table This replaces the reverse table lookup and reverse cache with a hashtable which improves cache-miss reverse-lookup times from O(n) to O(1)* and maintains the same performance as a reverse cache hit. This reduces the time needed to add a new sidtab entry from ~500us to 5us on a Pixel 3 when there are ~10,000 sidtab entries. The implementation uses the kernel's generic hashtable API, It uses the context's string represtation as the hash source, and the kernels generic string hashing algorithm full_name_hash() to reduce the string to a 32 bit value. This change also maintains the improvement introduced in commit ee1a84fdfeed ("selinux: overhaul sidtab to fix bug and improve performance") which removed the need to keep the current sidtab locked during policy reload. It does however introduce periodic locking of the target sidtab while converting the hashtable. Sidtab entries are never modified or removed, so the context struct stored in the sid_to_context tree can also be used for the context_to_sid hashtable to reduce memory usage. This bug was reported by: - On the selinux bug tracker. BUG: kernel softlockup due to too many SIDs/contexts #37 https://github.com/SELinuxProject/selinux-kernel/issues/37 - Jovana Knezevic on Android's bugtracker. Bug: 140252993 "During multi-user performance testing, we create and remove users many times. selinux_android_restorecon_pkgdir goes from 1ms to over 20ms after about 200 user creations and removals. Accumulated over ~280 packages, that adds a significant time to user creation, making perf benchmarks unreliable." * Hashtable lookup is only O(1) when n < the number of buckets. Signed-off-by: Jeff Vander Stoep <jeffv@google.com> Reported-by: Stephen Smalley <sds@tycho.nsa.gov> Reported-by: Jovana Knezevic <jovanak@google.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: subj tweak, removed changelog from patch description] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-22 17:33:06 +08:00
context_destroy(&dst->context);
selinux: overhaul sidtab to fix bug and improve performance Before this patch, during a policy reload the sidtab would become frozen and trying to map a new context to SID would be unable to add a new entry to sidtab and fail with -ENOMEM. Such failures are usually propagated into userspace, which has no way of distignuishing them from actual allocation failures and thus doesn't handle them gracefully. Such situation can be triggered e.g. by the following reproducer: while true; do load_policy; echo -n .; sleep 0.1; done & for (( i = 0; i < 1024; i++ )); do runcon -l s0:c$i echo -n x || break # or: # chcon -l s0:c$i <some_file> || break done This patch overhauls the sidtab so it doesn't need to be frozen during policy reload, thus solving the above problem. The new SID table leverages the fact that SIDs are allocated sequentially and are never invalidated and stores them in linear buckets indexed by a tree structure. This brings several advantages: 1. Fast SID -> context lookup - this lookup can now be done in logarithmic time complexity (usually in less than 4 array lookups) and can still be done safely without locking. 2. No need to re-search the whole table on reverse lookup miss - after acquiring the spinlock only the newly added entries need to be searched, which means that reverse lookups that end up inserting a new entry are now about twice as fast. 3. No need to freeze sidtab during policy reload - it is now possible to handle insertion of new entries even during sidtab conversion. The tree structure of the new sidtab is able to grow automatically to up to about 2^31 entries (at which point it should not have more than about 4 tree levels). The old sidtab had a theoretical capacity of almost 2^32 entries, but half of that is still more than enough since by that point the reverse table lookups would become unusably slow anyway... The number of entries per tree node is selected automatically so that each node fits into a single page, which should be the easiest size for kmalloc() to handle. Note that the cache for reverse lookup is preserved with equivalent logic. The only difference is that instead of storing pointers to the hash table nodes it stores just the indices of the cached entries. The new cache ensures that the indices are loaded/stored atomically, but it still has the drawback that concurrent cache updates may mess up the contents of the cache. Such situation however only reduces its effectivity, not the correctness of lookups. Tested by selinux-testsuite and thoroughly tortured by this simple stress test: ``` function rand_cat() { echo $(( $RANDOM % 1024 )) } function do_work() { while true; do echo -n "system_u:system_r:kernel_t:s0:c$(rand_cat),c$(rand_cat)" \ >/sys/fs/selinux/context 2>/dev/null || true done } do_work >/dev/null & do_work >/dev/null & do_work >/dev/null & while load_policy; do echo -n .; sleep 0.1; done kill %1 kill %2 kill %3 ``` Link: https://github.com/SELinuxProject/selinux-kernel/issues/38 Reported-by: Orion Poplawski <orion@nwra.com> Reported-by: Li Kun <hw.likun@huawei.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: most of sidtab.c merged by hand due to conflicts] [PM: checkpatch fixes in mls.c, services.c, sidtab.c] Signed-off-by: Paul Moore <paul@paul-moore.com>
2018-11-30 23:24:08 +08:00
goto out_unlock;
}
selinux: sidtab reverse lookup hash table This replaces the reverse table lookup and reverse cache with a hashtable which improves cache-miss reverse-lookup times from O(n) to O(1)* and maintains the same performance as a reverse cache hit. This reduces the time needed to add a new sidtab entry from ~500us to 5us on a Pixel 3 when there are ~10,000 sidtab entries. The implementation uses the kernel's generic hashtable API, It uses the context's string represtation as the hash source, and the kernels generic string hashing algorithm full_name_hash() to reduce the string to a 32 bit value. This change also maintains the improvement introduced in commit ee1a84fdfeed ("selinux: overhaul sidtab to fix bug and improve performance") which removed the need to keep the current sidtab locked during policy reload. It does however introduce periodic locking of the target sidtab while converting the hashtable. Sidtab entries are never modified or removed, so the context struct stored in the sid_to_context tree can also be used for the context_to_sid hashtable to reduce memory usage. This bug was reported by: - On the selinux bug tracker. BUG: kernel softlockup due to too many SIDs/contexts #37 https://github.com/SELinuxProject/selinux-kernel/issues/37 - Jovana Knezevic on Android's bugtracker. Bug: 140252993 "During multi-user performance testing, we create and remove users many times. selinux_android_restorecon_pkgdir goes from 1ms to over 20ms after about 200 user creations and removals. Accumulated over ~280 packages, that adds a significant time to user creation, making perf benchmarks unreliable." * Hashtable lookup is only O(1) when n < the number of buckets. Signed-off-by: Jeff Vander Stoep <jeffv@google.com> Reported-by: Stephen Smalley <sds@tycho.nsa.gov> Reported-by: Jovana Knezevic <jovanak@google.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: subj tweak, removed changelog from patch description] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-22 17:33:06 +08:00
dst_convert->sid = index_to_sid(count);
dst_convert->hash = context_compute_hash(&dst_convert->context);
convert->target->count = count + 1;
selinux: sidtab reverse lookup hash table This replaces the reverse table lookup and reverse cache with a hashtable which improves cache-miss reverse-lookup times from O(n) to O(1)* and maintains the same performance as a reverse cache hit. This reduces the time needed to add a new sidtab entry from ~500us to 5us on a Pixel 3 when there are ~10,000 sidtab entries. The implementation uses the kernel's generic hashtable API, It uses the context's string represtation as the hash source, and the kernels generic string hashing algorithm full_name_hash() to reduce the string to a 32 bit value. This change also maintains the improvement introduced in commit ee1a84fdfeed ("selinux: overhaul sidtab to fix bug and improve performance") which removed the need to keep the current sidtab locked during policy reload. It does however introduce periodic locking of the target sidtab while converting the hashtable. Sidtab entries are never modified or removed, so the context struct stored in the sid_to_context tree can also be used for the context_to_sid hashtable to reduce memory usage. This bug was reported by: - On the selinux bug tracker. BUG: kernel softlockup due to too many SIDs/contexts #37 https://github.com/SELinuxProject/selinux-kernel/issues/37 - Jovana Knezevic on Android's bugtracker. Bug: 140252993 "During multi-user performance testing, we create and remove users many times. selinux_android_restorecon_pkgdir goes from 1ms to over 20ms after about 200 user creations and removals. Accumulated over ~280 packages, that adds a significant time to user creation, making perf benchmarks unreliable." * Hashtable lookup is only O(1) when n < the number of buckets. Signed-off-by: Jeff Vander Stoep <jeffv@google.com> Reported-by: Stephen Smalley <sds@tycho.nsa.gov> Reported-by: Jovana Knezevic <jovanak@google.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: subj tweak, removed changelog from patch description] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-22 17:33:06 +08:00
hash_add_rcu(convert->target->context_to_sid,
&dst_convert->list, dst_convert->hash);
}
selinux: overhaul sidtab to fix bug and improve performance Before this patch, during a policy reload the sidtab would become frozen and trying to map a new context to SID would be unable to add a new entry to sidtab and fail with -ENOMEM. Such failures are usually propagated into userspace, which has no way of distignuishing them from actual allocation failures and thus doesn't handle them gracefully. Such situation can be triggered e.g. by the following reproducer: while true; do load_policy; echo -n .; sleep 0.1; done & for (( i = 0; i < 1024; i++ )); do runcon -l s0:c$i echo -n x || break # or: # chcon -l s0:c$i <some_file> || break done This patch overhauls the sidtab so it doesn't need to be frozen during policy reload, thus solving the above problem. The new SID table leverages the fact that SIDs are allocated sequentially and are never invalidated and stores them in linear buckets indexed by a tree structure. This brings several advantages: 1. Fast SID -> context lookup - this lookup can now be done in logarithmic time complexity (usually in less than 4 array lookups) and can still be done safely without locking. 2. No need to re-search the whole table on reverse lookup miss - after acquiring the spinlock only the newly added entries need to be searched, which means that reverse lookups that end up inserting a new entry are now about twice as fast. 3. No need to freeze sidtab during policy reload - it is now possible to handle insertion of new entries even during sidtab conversion. The tree structure of the new sidtab is able to grow automatically to up to about 2^31 entries (at which point it should not have more than about 4 tree levels). The old sidtab had a theoretical capacity of almost 2^32 entries, but half of that is still more than enough since by that point the reverse table lookups would become unusably slow anyway... The number of entries per tree node is selected automatically so that each node fits into a single page, which should be the easiest size for kmalloc() to handle. Note that the cache for reverse lookup is preserved with equivalent logic. The only difference is that instead of storing pointers to the hash table nodes it stores just the indices of the cached entries. The new cache ensures that the indices are loaded/stored atomically, but it still has the drawback that concurrent cache updates may mess up the contents of the cache. Such situation however only reduces its effectivity, not the correctness of lookups. Tested by selinux-testsuite and thoroughly tortured by this simple stress test: ``` function rand_cat() { echo $(( $RANDOM % 1024 )) } function do_work() { while true; do echo -n "system_u:system_r:kernel_t:s0:c$(rand_cat),c$(rand_cat)" \ >/sys/fs/selinux/context 2>/dev/null || true done } do_work >/dev/null & do_work >/dev/null & do_work >/dev/null & while load_policy; do echo -n .; sleep 0.1; done kill %1 kill %2 kill %3 ``` Link: https://github.com/SELinuxProject/selinux-kernel/issues/38 Reported-by: Orion Poplawski <orion@nwra.com> Reported-by: Li Kun <hw.likun@huawei.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: most of sidtab.c merged by hand due to conflicts] [PM: checkpatch fixes in mls.c, services.c, sidtab.c] Signed-off-by: Paul Moore <paul@paul-moore.com>
2018-11-30 23:24:08 +08:00
if (context->len)
pr_info("SELinux: Context %s is not valid (left unmapped).\n",
context->str);
selinux: sidtab reverse lookup hash table This replaces the reverse table lookup and reverse cache with a hashtable which improves cache-miss reverse-lookup times from O(n) to O(1)* and maintains the same performance as a reverse cache hit. This reduces the time needed to add a new sidtab entry from ~500us to 5us on a Pixel 3 when there are ~10,000 sidtab entries. The implementation uses the kernel's generic hashtable API, It uses the context's string represtation as the hash source, and the kernels generic string hashing algorithm full_name_hash() to reduce the string to a 32 bit value. This change also maintains the improvement introduced in commit ee1a84fdfeed ("selinux: overhaul sidtab to fix bug and improve performance") which removed the need to keep the current sidtab locked during policy reload. It does however introduce periodic locking of the target sidtab while converting the hashtable. Sidtab entries are never modified or removed, so the context struct stored in the sid_to_context tree can also be used for the context_to_sid hashtable to reduce memory usage. This bug was reported by: - On the selinux bug tracker. BUG: kernel softlockup due to too many SIDs/contexts #37 https://github.com/SELinuxProject/selinux-kernel/issues/37 - Jovana Knezevic on Android's bugtracker. Bug: 140252993 "During multi-user performance testing, we create and remove users many times. selinux_android_restorecon_pkgdir goes from 1ms to over 20ms after about 200 user creations and removals. Accumulated over ~280 packages, that adds a significant time to user creation, making perf benchmarks unreliable." * Hashtable lookup is only O(1) when n < the number of buckets. Signed-off-by: Jeff Vander Stoep <jeffv@google.com> Reported-by: Stephen Smalley <sds@tycho.nsa.gov> Reported-by: Jovana Knezevic <jovanak@google.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: subj tweak, removed changelog from patch description] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-22 17:33:06 +08:00
*sid = index_to_sid(count);
selinux: overhaul sidtab to fix bug and improve performance Before this patch, during a policy reload the sidtab would become frozen and trying to map a new context to SID would be unable to add a new entry to sidtab and fail with -ENOMEM. Such failures are usually propagated into userspace, which has no way of distignuishing them from actual allocation failures and thus doesn't handle them gracefully. Such situation can be triggered e.g. by the following reproducer: while true; do load_policy; echo -n .; sleep 0.1; done & for (( i = 0; i < 1024; i++ )); do runcon -l s0:c$i echo -n x || break # or: # chcon -l s0:c$i <some_file> || break done This patch overhauls the sidtab so it doesn't need to be frozen during policy reload, thus solving the above problem. The new SID table leverages the fact that SIDs are allocated sequentially and are never invalidated and stores them in linear buckets indexed by a tree structure. This brings several advantages: 1. Fast SID -> context lookup - this lookup can now be done in logarithmic time complexity (usually in less than 4 array lookups) and can still be done safely without locking. 2. No need to re-search the whole table on reverse lookup miss - after acquiring the spinlock only the newly added entries need to be searched, which means that reverse lookups that end up inserting a new entry are now about twice as fast. 3. No need to freeze sidtab during policy reload - it is now possible to handle insertion of new entries even during sidtab conversion. The tree structure of the new sidtab is able to grow automatically to up to about 2^31 entries (at which point it should not have more than about 4 tree levels). The old sidtab had a theoretical capacity of almost 2^32 entries, but half of that is still more than enough since by that point the reverse table lookups would become unusably slow anyway... The number of entries per tree node is selected automatically so that each node fits into a single page, which should be the easiest size for kmalloc() to handle. Note that the cache for reverse lookup is preserved with equivalent logic. The only difference is that instead of storing pointers to the hash table nodes it stores just the indices of the cached entries. The new cache ensures that the indices are loaded/stored atomically, but it still has the drawback that concurrent cache updates may mess up the contents of the cache. Such situation however only reduces its effectivity, not the correctness of lookups. Tested by selinux-testsuite and thoroughly tortured by this simple stress test: ``` function rand_cat() { echo $(( $RANDOM % 1024 )) } function do_work() { while true; do echo -n "system_u:system_r:kernel_t:s0:c$(rand_cat),c$(rand_cat)" \ >/sys/fs/selinux/context 2>/dev/null || true done } do_work >/dev/null & do_work >/dev/null & do_work >/dev/null & while load_policy; do echo -n .; sleep 0.1; done kill %1 kill %2 kill %3 ``` Link: https://github.com/SELinuxProject/selinux-kernel/issues/38 Reported-by: Orion Poplawski <orion@nwra.com> Reported-by: Li Kun <hw.likun@huawei.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: most of sidtab.c merged by hand due to conflicts] [PM: checkpatch fixes in mls.c, services.c, sidtab.c] Signed-off-by: Paul Moore <paul@paul-moore.com>
2018-11-30 23:24:08 +08:00
selinux: sidtab reverse lookup hash table This replaces the reverse table lookup and reverse cache with a hashtable which improves cache-miss reverse-lookup times from O(n) to O(1)* and maintains the same performance as a reverse cache hit. This reduces the time needed to add a new sidtab entry from ~500us to 5us on a Pixel 3 when there are ~10,000 sidtab entries. The implementation uses the kernel's generic hashtable API, It uses the context's string represtation as the hash source, and the kernels generic string hashing algorithm full_name_hash() to reduce the string to a 32 bit value. This change also maintains the improvement introduced in commit ee1a84fdfeed ("selinux: overhaul sidtab to fix bug and improve performance") which removed the need to keep the current sidtab locked during policy reload. It does however introduce periodic locking of the target sidtab while converting the hashtable. Sidtab entries are never modified or removed, so the context struct stored in the sid_to_context tree can also be used for the context_to_sid hashtable to reduce memory usage. This bug was reported by: - On the selinux bug tracker. BUG: kernel softlockup due to too many SIDs/contexts #37 https://github.com/SELinuxProject/selinux-kernel/issues/37 - Jovana Knezevic on Android's bugtracker. Bug: 140252993 "During multi-user performance testing, we create and remove users many times. selinux_android_restorecon_pkgdir goes from 1ms to over 20ms after about 200 user creations and removals. Accumulated over ~280 packages, that adds a significant time to user creation, making perf benchmarks unreliable." * Hashtable lookup is only O(1) when n < the number of buckets. Signed-off-by: Jeff Vander Stoep <jeffv@google.com> Reported-by: Stephen Smalley <sds@tycho.nsa.gov> Reported-by: Jovana Knezevic <jovanak@google.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: subj tweak, removed changelog from patch description] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-22 17:33:06 +08:00
/* write entries before updating count */
smp_store_release(&s->count, count + 1);
hash_add_rcu(s->context_to_sid, &dst->list, dst->hash);
selinux: overhaul sidtab to fix bug and improve performance Before this patch, during a policy reload the sidtab would become frozen and trying to map a new context to SID would be unable to add a new entry to sidtab and fail with -ENOMEM. Such failures are usually propagated into userspace, which has no way of distignuishing them from actual allocation failures and thus doesn't handle them gracefully. Such situation can be triggered e.g. by the following reproducer: while true; do load_policy; echo -n .; sleep 0.1; done & for (( i = 0; i < 1024; i++ )); do runcon -l s0:c$i echo -n x || break # or: # chcon -l s0:c$i <some_file> || break done This patch overhauls the sidtab so it doesn't need to be frozen during policy reload, thus solving the above problem. The new SID table leverages the fact that SIDs are allocated sequentially and are never invalidated and stores them in linear buckets indexed by a tree structure. This brings several advantages: 1. Fast SID -> context lookup - this lookup can now be done in logarithmic time complexity (usually in less than 4 array lookups) and can still be done safely without locking. 2. No need to re-search the whole table on reverse lookup miss - after acquiring the spinlock only the newly added entries need to be searched, which means that reverse lookups that end up inserting a new entry are now about twice as fast. 3. No need to freeze sidtab during policy reload - it is now possible to handle insertion of new entries even during sidtab conversion. The tree structure of the new sidtab is able to grow automatically to up to about 2^31 entries (at which point it should not have more than about 4 tree levels). The old sidtab had a theoretical capacity of almost 2^32 entries, but half of that is still more than enough since by that point the reverse table lookups would become unusably slow anyway... The number of entries per tree node is selected automatically so that each node fits into a single page, which should be the easiest size for kmalloc() to handle. Note that the cache for reverse lookup is preserved with equivalent logic. The only difference is that instead of storing pointers to the hash table nodes it stores just the indices of the cached entries. The new cache ensures that the indices are loaded/stored atomically, but it still has the drawback that concurrent cache updates may mess up the contents of the cache. Such situation however only reduces its effectivity, not the correctness of lookups. Tested by selinux-testsuite and thoroughly tortured by this simple stress test: ``` function rand_cat() { echo $(( $RANDOM % 1024 )) } function do_work() { while true; do echo -n "system_u:system_r:kernel_t:s0:c$(rand_cat),c$(rand_cat)" \ >/sys/fs/selinux/context 2>/dev/null || true done } do_work >/dev/null & do_work >/dev/null & do_work >/dev/null & while load_policy; do echo -n .; sleep 0.1; done kill %1 kill %2 kill %3 ``` Link: https://github.com/SELinuxProject/selinux-kernel/issues/38 Reported-by: Orion Poplawski <orion@nwra.com> Reported-by: Li Kun <hw.likun@huawei.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: most of sidtab.c merged by hand due to conflicts] [PM: checkpatch fixes in mls.c, services.c, sidtab.c] Signed-off-by: Paul Moore <paul@paul-moore.com>
2018-11-30 23:24:08 +08:00
rc = 0;
out_unlock:
spin_unlock_irqrestore(&s->lock, flags);
return rc;
}
selinux: sidtab reverse lookup hash table This replaces the reverse table lookup and reverse cache with a hashtable which improves cache-miss reverse-lookup times from O(n) to O(1)* and maintains the same performance as a reverse cache hit. This reduces the time needed to add a new sidtab entry from ~500us to 5us on a Pixel 3 when there are ~10,000 sidtab entries. The implementation uses the kernel's generic hashtable API, It uses the context's string represtation as the hash source, and the kernels generic string hashing algorithm full_name_hash() to reduce the string to a 32 bit value. This change also maintains the improvement introduced in commit ee1a84fdfeed ("selinux: overhaul sidtab to fix bug and improve performance") which removed the need to keep the current sidtab locked during policy reload. It does however introduce periodic locking of the target sidtab while converting the hashtable. Sidtab entries are never modified or removed, so the context struct stored in the sid_to_context tree can also be used for the context_to_sid hashtable to reduce memory usage. This bug was reported by: - On the selinux bug tracker. BUG: kernel softlockup due to too many SIDs/contexts #37 https://github.com/SELinuxProject/selinux-kernel/issues/37 - Jovana Knezevic on Android's bugtracker. Bug: 140252993 "During multi-user performance testing, we create and remove users many times. selinux_android_restorecon_pkgdir goes from 1ms to over 20ms after about 200 user creations and removals. Accumulated over ~280 packages, that adds a significant time to user creation, making perf benchmarks unreliable." * Hashtable lookup is only O(1) when n < the number of buckets. Signed-off-by: Jeff Vander Stoep <jeffv@google.com> Reported-by: Stephen Smalley <sds@tycho.nsa.gov> Reported-by: Jovana Knezevic <jovanak@google.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: subj tweak, removed changelog from patch description] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-22 17:33:06 +08:00
static void sidtab_convert_hashtable(struct sidtab *s, u32 count)
{
selinux: cache the SID -> context string translation Translating a context struct to string can be quite slow, especially if the context has a lot of category bits set. This can cause quite noticeable performance impact in situations where the translation needs to be done repeatedly. A common example is a UNIX datagram socket with the SO_PASSSEC option enabled, which is used e.g. by systemd-journald when receiving log messages via datagram socket. This scenario can be reproduced with: cat /dev/urandom | base64 | logger & timeout 30s perf record -p $(pidof systemd-journald) -a -g kill %1 perf report -g none --pretty raw | grep security_secid_to_secctx Before the caching introduced by this patch, computing the context string (security_secid_to_secctx() function) takes up ~65% of systemd-journald's CPU time (assuming a context with 1024 categories set and Fedora x86_64 release kernel configs). After this patch (assuming near-perfect cache hit ratio) this overhead is reduced to just ~2%. This patch addresses the issue by caching a certain number (compile-time configurable) of recently used context strings to speed up repeated translations of the same context, while using only a small amount of memory. The cache is integrated into the existing sidtab table by adding a field to each entry, which when not NULL contains an RCU-protected pointer to a cache entry containing the cached string. The cache entries are kept in a linked list sorted according to how recently they were used. On a cache miss when the cache is full, the least recently used entry is removed to make space for the new entry. The patch migrates security_sid_to_context_core() to use the cache (also a few other functions where it was possible without too much fuss, but these mostly use the translation for logging in case of error, which is rare). Link: https://bugzilla.redhat.com/show_bug.cgi?id=1733259 Cc: Michal Sekletar <msekleta@redhat.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> Reviewed-by: Paul E. McKenney <paulmck@kernel.org> [PM: lots of merge fixups due to collisions with other sidtab patches] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-26 21:57:00 +08:00
struct sidtab_entry *entry;
u32 i;
selinux: sidtab reverse lookup hash table This replaces the reverse table lookup and reverse cache with a hashtable which improves cache-miss reverse-lookup times from O(n) to O(1)* and maintains the same performance as a reverse cache hit. This reduces the time needed to add a new sidtab entry from ~500us to 5us on a Pixel 3 when there are ~10,000 sidtab entries. The implementation uses the kernel's generic hashtable API, It uses the context's string represtation as the hash source, and the kernels generic string hashing algorithm full_name_hash() to reduce the string to a 32 bit value. This change also maintains the improvement introduced in commit ee1a84fdfeed ("selinux: overhaul sidtab to fix bug and improve performance") which removed the need to keep the current sidtab locked during policy reload. It does however introduce periodic locking of the target sidtab while converting the hashtable. Sidtab entries are never modified or removed, so the context struct stored in the sid_to_context tree can also be used for the context_to_sid hashtable to reduce memory usage. This bug was reported by: - On the selinux bug tracker. BUG: kernel softlockup due to too many SIDs/contexts #37 https://github.com/SELinuxProject/selinux-kernel/issues/37 - Jovana Knezevic on Android's bugtracker. Bug: 140252993 "During multi-user performance testing, we create and remove users many times. selinux_android_restorecon_pkgdir goes from 1ms to over 20ms after about 200 user creations and removals. Accumulated over ~280 packages, that adds a significant time to user creation, making perf benchmarks unreliable." * Hashtable lookup is only O(1) when n < the number of buckets. Signed-off-by: Jeff Vander Stoep <jeffv@google.com> Reported-by: Stephen Smalley <sds@tycho.nsa.gov> Reported-by: Jovana Knezevic <jovanak@google.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: subj tweak, removed changelog from patch description] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-22 17:33:06 +08:00
for (i = 0; i < count; i++) {
entry = sidtab_do_lookup(s, i, 0);
entry->sid = index_to_sid(i);
entry->hash = context_compute_hash(&entry->context);
hash_add_rcu(s->context_to_sid, &entry->list, entry->hash);
selinux: sidtab reverse lookup hash table This replaces the reverse table lookup and reverse cache with a hashtable which improves cache-miss reverse-lookup times from O(n) to O(1)* and maintains the same performance as a reverse cache hit. This reduces the time needed to add a new sidtab entry from ~500us to 5us on a Pixel 3 when there are ~10,000 sidtab entries. The implementation uses the kernel's generic hashtable API, It uses the context's string represtation as the hash source, and the kernels generic string hashing algorithm full_name_hash() to reduce the string to a 32 bit value. This change also maintains the improvement introduced in commit ee1a84fdfeed ("selinux: overhaul sidtab to fix bug and improve performance") which removed the need to keep the current sidtab locked during policy reload. It does however introduce periodic locking of the target sidtab while converting the hashtable. Sidtab entries are never modified or removed, so the context struct stored in the sid_to_context tree can also be used for the context_to_sid hashtable to reduce memory usage. This bug was reported by: - On the selinux bug tracker. BUG: kernel softlockup due to too many SIDs/contexts #37 https://github.com/SELinuxProject/selinux-kernel/issues/37 - Jovana Knezevic on Android's bugtracker. Bug: 140252993 "During multi-user performance testing, we create and remove users many times. selinux_android_restorecon_pkgdir goes from 1ms to over 20ms after about 200 user creations and removals. Accumulated over ~280 packages, that adds a significant time to user creation, making perf benchmarks unreliable." * Hashtable lookup is only O(1) when n < the number of buckets. Signed-off-by: Jeff Vander Stoep <jeffv@google.com> Reported-by: Stephen Smalley <sds@tycho.nsa.gov> Reported-by: Jovana Knezevic <jovanak@google.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: subj tweak, removed changelog from patch description] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-22 17:33:06 +08:00
}
}
selinux: overhaul sidtab to fix bug and improve performance Before this patch, during a policy reload the sidtab would become frozen and trying to map a new context to SID would be unable to add a new entry to sidtab and fail with -ENOMEM. Such failures are usually propagated into userspace, which has no way of distignuishing them from actual allocation failures and thus doesn't handle them gracefully. Such situation can be triggered e.g. by the following reproducer: while true; do load_policy; echo -n .; sleep 0.1; done & for (( i = 0; i < 1024; i++ )); do runcon -l s0:c$i echo -n x || break # or: # chcon -l s0:c$i <some_file> || break done This patch overhauls the sidtab so it doesn't need to be frozen during policy reload, thus solving the above problem. The new SID table leverages the fact that SIDs are allocated sequentially and are never invalidated and stores them in linear buckets indexed by a tree structure. This brings several advantages: 1. Fast SID -> context lookup - this lookup can now be done in logarithmic time complexity (usually in less than 4 array lookups) and can still be done safely without locking. 2. No need to re-search the whole table on reverse lookup miss - after acquiring the spinlock only the newly added entries need to be searched, which means that reverse lookups that end up inserting a new entry are now about twice as fast. 3. No need to freeze sidtab during policy reload - it is now possible to handle insertion of new entries even during sidtab conversion. The tree structure of the new sidtab is able to grow automatically to up to about 2^31 entries (at which point it should not have more than about 4 tree levels). The old sidtab had a theoretical capacity of almost 2^32 entries, but half of that is still more than enough since by that point the reverse table lookups would become unusably slow anyway... The number of entries per tree node is selected automatically so that each node fits into a single page, which should be the easiest size for kmalloc() to handle. Note that the cache for reverse lookup is preserved with equivalent logic. The only difference is that instead of storing pointers to the hash table nodes it stores just the indices of the cached entries. The new cache ensures that the indices are loaded/stored atomically, but it still has the drawback that concurrent cache updates may mess up the contents of the cache. Such situation however only reduces its effectivity, not the correctness of lookups. Tested by selinux-testsuite and thoroughly tortured by this simple stress test: ``` function rand_cat() { echo $(( $RANDOM % 1024 )) } function do_work() { while true; do echo -n "system_u:system_r:kernel_t:s0:c$(rand_cat),c$(rand_cat)" \ >/sys/fs/selinux/context 2>/dev/null || true done } do_work >/dev/null & do_work >/dev/null & do_work >/dev/null & while load_policy; do echo -n .; sleep 0.1; done kill %1 kill %2 kill %3 ``` Link: https://github.com/SELinuxProject/selinux-kernel/issues/38 Reported-by: Orion Poplawski <orion@nwra.com> Reported-by: Li Kun <hw.likun@huawei.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: most of sidtab.c merged by hand due to conflicts] [PM: checkpatch fixes in mls.c, services.c, sidtab.c] Signed-off-by: Paul Moore <paul@paul-moore.com>
2018-11-30 23:24:08 +08:00
static int sidtab_convert_tree(union sidtab_entry_inner *edst,
union sidtab_entry_inner *esrc,
u32 *pos, u32 count, u32 level,
struct sidtab_convert_params *convert)
{
selinux: overhaul sidtab to fix bug and improve performance Before this patch, during a policy reload the sidtab would become frozen and trying to map a new context to SID would be unable to add a new entry to sidtab and fail with -ENOMEM. Such failures are usually propagated into userspace, which has no way of distignuishing them from actual allocation failures and thus doesn't handle them gracefully. Such situation can be triggered e.g. by the following reproducer: while true; do load_policy; echo -n .; sleep 0.1; done & for (( i = 0; i < 1024; i++ )); do runcon -l s0:c$i echo -n x || break # or: # chcon -l s0:c$i <some_file> || break done This patch overhauls the sidtab so it doesn't need to be frozen during policy reload, thus solving the above problem. The new SID table leverages the fact that SIDs are allocated sequentially and are never invalidated and stores them in linear buckets indexed by a tree structure. This brings several advantages: 1. Fast SID -> context lookup - this lookup can now be done in logarithmic time complexity (usually in less than 4 array lookups) and can still be done safely without locking. 2. No need to re-search the whole table on reverse lookup miss - after acquiring the spinlock only the newly added entries need to be searched, which means that reverse lookups that end up inserting a new entry are now about twice as fast. 3. No need to freeze sidtab during policy reload - it is now possible to handle insertion of new entries even during sidtab conversion. The tree structure of the new sidtab is able to grow automatically to up to about 2^31 entries (at which point it should not have more than about 4 tree levels). The old sidtab had a theoretical capacity of almost 2^32 entries, but half of that is still more than enough since by that point the reverse table lookups would become unusably slow anyway... The number of entries per tree node is selected automatically so that each node fits into a single page, which should be the easiest size for kmalloc() to handle. Note that the cache for reverse lookup is preserved with equivalent logic. The only difference is that instead of storing pointers to the hash table nodes it stores just the indices of the cached entries. The new cache ensures that the indices are loaded/stored atomically, but it still has the drawback that concurrent cache updates may mess up the contents of the cache. Such situation however only reduces its effectivity, not the correctness of lookups. Tested by selinux-testsuite and thoroughly tortured by this simple stress test: ``` function rand_cat() { echo $(( $RANDOM % 1024 )) } function do_work() { while true; do echo -n "system_u:system_r:kernel_t:s0:c$(rand_cat),c$(rand_cat)" \ >/sys/fs/selinux/context 2>/dev/null || true done } do_work >/dev/null & do_work >/dev/null & do_work >/dev/null & while load_policy; do echo -n .; sleep 0.1; done kill %1 kill %2 kill %3 ``` Link: https://github.com/SELinuxProject/selinux-kernel/issues/38 Reported-by: Orion Poplawski <orion@nwra.com> Reported-by: Li Kun <hw.likun@huawei.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: most of sidtab.c merged by hand due to conflicts] [PM: checkpatch fixes in mls.c, services.c, sidtab.c] Signed-off-by: Paul Moore <paul@paul-moore.com>
2018-11-30 23:24:08 +08:00
int rc;
u32 i;
selinux: overhaul sidtab to fix bug and improve performance Before this patch, during a policy reload the sidtab would become frozen and trying to map a new context to SID would be unable to add a new entry to sidtab and fail with -ENOMEM. Such failures are usually propagated into userspace, which has no way of distignuishing them from actual allocation failures and thus doesn't handle them gracefully. Such situation can be triggered e.g. by the following reproducer: while true; do load_policy; echo -n .; sleep 0.1; done & for (( i = 0; i < 1024; i++ )); do runcon -l s0:c$i echo -n x || break # or: # chcon -l s0:c$i <some_file> || break done This patch overhauls the sidtab so it doesn't need to be frozen during policy reload, thus solving the above problem. The new SID table leverages the fact that SIDs are allocated sequentially and are never invalidated and stores them in linear buckets indexed by a tree structure. This brings several advantages: 1. Fast SID -> context lookup - this lookup can now be done in logarithmic time complexity (usually in less than 4 array lookups) and can still be done safely without locking. 2. No need to re-search the whole table on reverse lookup miss - after acquiring the spinlock only the newly added entries need to be searched, which means that reverse lookups that end up inserting a new entry are now about twice as fast. 3. No need to freeze sidtab during policy reload - it is now possible to handle insertion of new entries even during sidtab conversion. The tree structure of the new sidtab is able to grow automatically to up to about 2^31 entries (at which point it should not have more than about 4 tree levels). The old sidtab had a theoretical capacity of almost 2^32 entries, but half of that is still more than enough since by that point the reverse table lookups would become unusably slow anyway... The number of entries per tree node is selected automatically so that each node fits into a single page, which should be the easiest size for kmalloc() to handle. Note that the cache for reverse lookup is preserved with equivalent logic. The only difference is that instead of storing pointers to the hash table nodes it stores just the indices of the cached entries. The new cache ensures that the indices are loaded/stored atomically, but it still has the drawback that concurrent cache updates may mess up the contents of the cache. Such situation however only reduces its effectivity, not the correctness of lookups. Tested by selinux-testsuite and thoroughly tortured by this simple stress test: ``` function rand_cat() { echo $(( $RANDOM % 1024 )) } function do_work() { while true; do echo -n "system_u:system_r:kernel_t:s0:c$(rand_cat),c$(rand_cat)" \ >/sys/fs/selinux/context 2>/dev/null || true done } do_work >/dev/null & do_work >/dev/null & do_work >/dev/null & while load_policy; do echo -n .; sleep 0.1; done kill %1 kill %2 kill %3 ``` Link: https://github.com/SELinuxProject/selinux-kernel/issues/38 Reported-by: Orion Poplawski <orion@nwra.com> Reported-by: Li Kun <hw.likun@huawei.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: most of sidtab.c merged by hand due to conflicts] [PM: checkpatch fixes in mls.c, services.c, sidtab.c] Signed-off-by: Paul Moore <paul@paul-moore.com>
2018-11-30 23:24:08 +08:00
if (level != 0) {
if (!edst->ptr_inner) {
edst->ptr_inner = kzalloc(SIDTAB_NODE_ALLOC_SIZE,
GFP_KERNEL);
if (!edst->ptr_inner)
return -ENOMEM;
}
selinux: overhaul sidtab to fix bug and improve performance Before this patch, during a policy reload the sidtab would become frozen and trying to map a new context to SID would be unable to add a new entry to sidtab and fail with -ENOMEM. Such failures are usually propagated into userspace, which has no way of distignuishing them from actual allocation failures and thus doesn't handle them gracefully. Such situation can be triggered e.g. by the following reproducer: while true; do load_policy; echo -n .; sleep 0.1; done & for (( i = 0; i < 1024; i++ )); do runcon -l s0:c$i echo -n x || break # or: # chcon -l s0:c$i <some_file> || break done This patch overhauls the sidtab so it doesn't need to be frozen during policy reload, thus solving the above problem. The new SID table leverages the fact that SIDs are allocated sequentially and are never invalidated and stores them in linear buckets indexed by a tree structure. This brings several advantages: 1. Fast SID -> context lookup - this lookup can now be done in logarithmic time complexity (usually in less than 4 array lookups) and can still be done safely without locking. 2. No need to re-search the whole table on reverse lookup miss - after acquiring the spinlock only the newly added entries need to be searched, which means that reverse lookups that end up inserting a new entry are now about twice as fast. 3. No need to freeze sidtab during policy reload - it is now possible to handle insertion of new entries even during sidtab conversion. The tree structure of the new sidtab is able to grow automatically to up to about 2^31 entries (at which point it should not have more than about 4 tree levels). The old sidtab had a theoretical capacity of almost 2^32 entries, but half of that is still more than enough since by that point the reverse table lookups would become unusably slow anyway... The number of entries per tree node is selected automatically so that each node fits into a single page, which should be the easiest size for kmalloc() to handle. Note that the cache for reverse lookup is preserved with equivalent logic. The only difference is that instead of storing pointers to the hash table nodes it stores just the indices of the cached entries. The new cache ensures that the indices are loaded/stored atomically, but it still has the drawback that concurrent cache updates may mess up the contents of the cache. Such situation however only reduces its effectivity, not the correctness of lookups. Tested by selinux-testsuite and thoroughly tortured by this simple stress test: ``` function rand_cat() { echo $(( $RANDOM % 1024 )) } function do_work() { while true; do echo -n "system_u:system_r:kernel_t:s0:c$(rand_cat),c$(rand_cat)" \ >/sys/fs/selinux/context 2>/dev/null || true done } do_work >/dev/null & do_work >/dev/null & do_work >/dev/null & while load_policy; do echo -n .; sleep 0.1; done kill %1 kill %2 kill %3 ``` Link: https://github.com/SELinuxProject/selinux-kernel/issues/38 Reported-by: Orion Poplawski <orion@nwra.com> Reported-by: Li Kun <hw.likun@huawei.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: most of sidtab.c merged by hand due to conflicts] [PM: checkpatch fixes in mls.c, services.c, sidtab.c] Signed-off-by: Paul Moore <paul@paul-moore.com>
2018-11-30 23:24:08 +08:00
i = 0;
while (i < SIDTAB_INNER_ENTRIES && *pos < count) {
rc = sidtab_convert_tree(&edst->ptr_inner->entries[i],
&esrc->ptr_inner->entries[i],
pos, count, level - 1,
convert);
if (rc)
return rc;
i++;
}
} else {
if (!edst->ptr_leaf) {
edst->ptr_leaf = kzalloc(SIDTAB_NODE_ALLOC_SIZE,
GFP_KERNEL);
if (!edst->ptr_leaf)
return -ENOMEM;
}
i = 0;
while (i < SIDTAB_LEAF_ENTRIES && *pos < count) {
rc = convert->func(&esrc->ptr_leaf->entries[i].context,
&edst->ptr_leaf->entries[i].context,
convert->args);
if (rc)
return rc;
(*pos)++;
i++;
}
cond_resched();
}
return 0;
}
int sidtab_convert(struct sidtab *s, struct sidtab_convert_params *params)
{
unsigned long flags;
u32 count, level, pos;
int rc;
spin_lock_irqsave(&s->lock, flags);
/* concurrent policy loads are not allowed */
if (s->convert) {
spin_unlock_irqrestore(&s->lock, flags);
return -EBUSY;
}
count = s->count;
selinux: overhaul sidtab to fix bug and improve performance Before this patch, during a policy reload the sidtab would become frozen and trying to map a new context to SID would be unable to add a new entry to sidtab and fail with -ENOMEM. Such failures are usually propagated into userspace, which has no way of distignuishing them from actual allocation failures and thus doesn't handle them gracefully. Such situation can be triggered e.g. by the following reproducer: while true; do load_policy; echo -n .; sleep 0.1; done & for (( i = 0; i < 1024; i++ )); do runcon -l s0:c$i echo -n x || break # or: # chcon -l s0:c$i <some_file> || break done This patch overhauls the sidtab so it doesn't need to be frozen during policy reload, thus solving the above problem. The new SID table leverages the fact that SIDs are allocated sequentially and are never invalidated and stores them in linear buckets indexed by a tree structure. This brings several advantages: 1. Fast SID -> context lookup - this lookup can now be done in logarithmic time complexity (usually in less than 4 array lookups) and can still be done safely without locking. 2. No need to re-search the whole table on reverse lookup miss - after acquiring the spinlock only the newly added entries need to be searched, which means that reverse lookups that end up inserting a new entry are now about twice as fast. 3. No need to freeze sidtab during policy reload - it is now possible to handle insertion of new entries even during sidtab conversion. The tree structure of the new sidtab is able to grow automatically to up to about 2^31 entries (at which point it should not have more than about 4 tree levels). The old sidtab had a theoretical capacity of almost 2^32 entries, but half of that is still more than enough since by that point the reverse table lookups would become unusably slow anyway... The number of entries per tree node is selected automatically so that each node fits into a single page, which should be the easiest size for kmalloc() to handle. Note that the cache for reverse lookup is preserved with equivalent logic. The only difference is that instead of storing pointers to the hash table nodes it stores just the indices of the cached entries. The new cache ensures that the indices are loaded/stored atomically, but it still has the drawback that concurrent cache updates may mess up the contents of the cache. Such situation however only reduces its effectivity, not the correctness of lookups. Tested by selinux-testsuite and thoroughly tortured by this simple stress test: ``` function rand_cat() { echo $(( $RANDOM % 1024 )) } function do_work() { while true; do echo -n "system_u:system_r:kernel_t:s0:c$(rand_cat),c$(rand_cat)" \ >/sys/fs/selinux/context 2>/dev/null || true done } do_work >/dev/null & do_work >/dev/null & do_work >/dev/null & while load_policy; do echo -n .; sleep 0.1; done kill %1 kill %2 kill %3 ``` Link: https://github.com/SELinuxProject/selinux-kernel/issues/38 Reported-by: Orion Poplawski <orion@nwra.com> Reported-by: Li Kun <hw.likun@huawei.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: most of sidtab.c merged by hand due to conflicts] [PM: checkpatch fixes in mls.c, services.c, sidtab.c] Signed-off-by: Paul Moore <paul@paul-moore.com>
2018-11-30 23:24:08 +08:00
level = sidtab_level_from_count(count);
/* allocate last leaf in the new sidtab (to avoid race with
* live convert)
*/
rc = sidtab_do_lookup(params->target, count - 1, 1) ? 0 : -ENOMEM;
if (rc) {
spin_unlock_irqrestore(&s->lock, flags);
return rc;
}
/* set count in case no new entries are added during conversion */
params->target->count = count;
selinux: overhaul sidtab to fix bug and improve performance Before this patch, during a policy reload the sidtab would become frozen and trying to map a new context to SID would be unable to add a new entry to sidtab and fail with -ENOMEM. Such failures are usually propagated into userspace, which has no way of distignuishing them from actual allocation failures and thus doesn't handle them gracefully. Such situation can be triggered e.g. by the following reproducer: while true; do load_policy; echo -n .; sleep 0.1; done & for (( i = 0; i < 1024; i++ )); do runcon -l s0:c$i echo -n x || break # or: # chcon -l s0:c$i <some_file> || break done This patch overhauls the sidtab so it doesn't need to be frozen during policy reload, thus solving the above problem. The new SID table leverages the fact that SIDs are allocated sequentially and are never invalidated and stores them in linear buckets indexed by a tree structure. This brings several advantages: 1. Fast SID -> context lookup - this lookup can now be done in logarithmic time complexity (usually in less than 4 array lookups) and can still be done safely without locking. 2. No need to re-search the whole table on reverse lookup miss - after acquiring the spinlock only the newly added entries need to be searched, which means that reverse lookups that end up inserting a new entry are now about twice as fast. 3. No need to freeze sidtab during policy reload - it is now possible to handle insertion of new entries even during sidtab conversion. The tree structure of the new sidtab is able to grow automatically to up to about 2^31 entries (at which point it should not have more than about 4 tree levels). The old sidtab had a theoretical capacity of almost 2^32 entries, but half of that is still more than enough since by that point the reverse table lookups would become unusably slow anyway... The number of entries per tree node is selected automatically so that each node fits into a single page, which should be the easiest size for kmalloc() to handle. Note that the cache for reverse lookup is preserved with equivalent logic. The only difference is that instead of storing pointers to the hash table nodes it stores just the indices of the cached entries. The new cache ensures that the indices are loaded/stored atomically, but it still has the drawback that concurrent cache updates may mess up the contents of the cache. Such situation however only reduces its effectivity, not the correctness of lookups. Tested by selinux-testsuite and thoroughly tortured by this simple stress test: ``` function rand_cat() { echo $(( $RANDOM % 1024 )) } function do_work() { while true; do echo -n "system_u:system_r:kernel_t:s0:c$(rand_cat),c$(rand_cat)" \ >/sys/fs/selinux/context 2>/dev/null || true done } do_work >/dev/null & do_work >/dev/null & do_work >/dev/null & while load_policy; do echo -n .; sleep 0.1; done kill %1 kill %2 kill %3 ``` Link: https://github.com/SELinuxProject/selinux-kernel/issues/38 Reported-by: Orion Poplawski <orion@nwra.com> Reported-by: Li Kun <hw.likun@huawei.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: most of sidtab.c merged by hand due to conflicts] [PM: checkpatch fixes in mls.c, services.c, sidtab.c] Signed-off-by: Paul Moore <paul@paul-moore.com>
2018-11-30 23:24:08 +08:00
/* enable live convert of new entries */
s->convert = params;
selinux: sidtab reverse lookup hash table This replaces the reverse table lookup and reverse cache with a hashtable which improves cache-miss reverse-lookup times from O(n) to O(1)* and maintains the same performance as a reverse cache hit. This reduces the time needed to add a new sidtab entry from ~500us to 5us on a Pixel 3 when there are ~10,000 sidtab entries. The implementation uses the kernel's generic hashtable API, It uses the context's string represtation as the hash source, and the kernels generic string hashing algorithm full_name_hash() to reduce the string to a 32 bit value. This change also maintains the improvement introduced in commit ee1a84fdfeed ("selinux: overhaul sidtab to fix bug and improve performance") which removed the need to keep the current sidtab locked during policy reload. It does however introduce periodic locking of the target sidtab while converting the hashtable. Sidtab entries are never modified or removed, so the context struct stored in the sid_to_context tree can also be used for the context_to_sid hashtable to reduce memory usage. This bug was reported by: - On the selinux bug tracker. BUG: kernel softlockup due to too many SIDs/contexts #37 https://github.com/SELinuxProject/selinux-kernel/issues/37 - Jovana Knezevic on Android's bugtracker. Bug: 140252993 "During multi-user performance testing, we create and remove users many times. selinux_android_restorecon_pkgdir goes from 1ms to over 20ms after about 200 user creations and removals. Accumulated over ~280 packages, that adds a significant time to user creation, making perf benchmarks unreliable." * Hashtable lookup is only O(1) when n < the number of buckets. Signed-off-by: Jeff Vander Stoep <jeffv@google.com> Reported-by: Stephen Smalley <sds@tycho.nsa.gov> Reported-by: Jovana Knezevic <jovanak@google.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: subj tweak, removed changelog from patch description] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-22 17:33:06 +08:00
/* we can safely convert the tree outside the lock */
selinux: overhaul sidtab to fix bug and improve performance Before this patch, during a policy reload the sidtab would become frozen and trying to map a new context to SID would be unable to add a new entry to sidtab and fail with -ENOMEM. Such failures are usually propagated into userspace, which has no way of distignuishing them from actual allocation failures and thus doesn't handle them gracefully. Such situation can be triggered e.g. by the following reproducer: while true; do load_policy; echo -n .; sleep 0.1; done & for (( i = 0; i < 1024; i++ )); do runcon -l s0:c$i echo -n x || break # or: # chcon -l s0:c$i <some_file> || break done This patch overhauls the sidtab so it doesn't need to be frozen during policy reload, thus solving the above problem. The new SID table leverages the fact that SIDs are allocated sequentially and are never invalidated and stores them in linear buckets indexed by a tree structure. This brings several advantages: 1. Fast SID -> context lookup - this lookup can now be done in logarithmic time complexity (usually in less than 4 array lookups) and can still be done safely without locking. 2. No need to re-search the whole table on reverse lookup miss - after acquiring the spinlock only the newly added entries need to be searched, which means that reverse lookups that end up inserting a new entry are now about twice as fast. 3. No need to freeze sidtab during policy reload - it is now possible to handle insertion of new entries even during sidtab conversion. The tree structure of the new sidtab is able to grow automatically to up to about 2^31 entries (at which point it should not have more than about 4 tree levels). The old sidtab had a theoretical capacity of almost 2^32 entries, but half of that is still more than enough since by that point the reverse table lookups would become unusably slow anyway... The number of entries per tree node is selected automatically so that each node fits into a single page, which should be the easiest size for kmalloc() to handle. Note that the cache for reverse lookup is preserved with equivalent logic. The only difference is that instead of storing pointers to the hash table nodes it stores just the indices of the cached entries. The new cache ensures that the indices are loaded/stored atomically, but it still has the drawback that concurrent cache updates may mess up the contents of the cache. Such situation however only reduces its effectivity, not the correctness of lookups. Tested by selinux-testsuite and thoroughly tortured by this simple stress test: ``` function rand_cat() { echo $(( $RANDOM % 1024 )) } function do_work() { while true; do echo -n "system_u:system_r:kernel_t:s0:c$(rand_cat),c$(rand_cat)" \ >/sys/fs/selinux/context 2>/dev/null || true done } do_work >/dev/null & do_work >/dev/null & do_work >/dev/null & while load_policy; do echo -n .; sleep 0.1; done kill %1 kill %2 kill %3 ``` Link: https://github.com/SELinuxProject/selinux-kernel/issues/38 Reported-by: Orion Poplawski <orion@nwra.com> Reported-by: Li Kun <hw.likun@huawei.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: most of sidtab.c merged by hand due to conflicts] [PM: checkpatch fixes in mls.c, services.c, sidtab.c] Signed-off-by: Paul Moore <paul@paul-moore.com>
2018-11-30 23:24:08 +08:00
spin_unlock_irqrestore(&s->lock, flags);
pr_info("SELinux: Converting %u SID table entries...\n", count);
/* convert all entries not covered by live convert */
pos = 0;
rc = sidtab_convert_tree(&params->target->roots[level],
&s->roots[level], &pos, count, level, params);
if (rc) {
/* we need to keep the old table - disable live convert */
spin_lock_irqsave(&s->lock, flags);
s->convert = NULL;
spin_unlock_irqrestore(&s->lock, flags);
selinux: sidtab reverse lookup hash table This replaces the reverse table lookup and reverse cache with a hashtable which improves cache-miss reverse-lookup times from O(n) to O(1)* and maintains the same performance as a reverse cache hit. This reduces the time needed to add a new sidtab entry from ~500us to 5us on a Pixel 3 when there are ~10,000 sidtab entries. The implementation uses the kernel's generic hashtable API, It uses the context's string represtation as the hash source, and the kernels generic string hashing algorithm full_name_hash() to reduce the string to a 32 bit value. This change also maintains the improvement introduced in commit ee1a84fdfeed ("selinux: overhaul sidtab to fix bug and improve performance") which removed the need to keep the current sidtab locked during policy reload. It does however introduce periodic locking of the target sidtab while converting the hashtable. Sidtab entries are never modified or removed, so the context struct stored in the sid_to_context tree can also be used for the context_to_sid hashtable to reduce memory usage. This bug was reported by: - On the selinux bug tracker. BUG: kernel softlockup due to too many SIDs/contexts #37 https://github.com/SELinuxProject/selinux-kernel/issues/37 - Jovana Knezevic on Android's bugtracker. Bug: 140252993 "During multi-user performance testing, we create and remove users many times. selinux_android_restorecon_pkgdir goes from 1ms to over 20ms after about 200 user creations and removals. Accumulated over ~280 packages, that adds a significant time to user creation, making perf benchmarks unreliable." * Hashtable lookup is only O(1) when n < the number of buckets. Signed-off-by: Jeff Vander Stoep <jeffv@google.com> Reported-by: Stephen Smalley <sds@tycho.nsa.gov> Reported-by: Jovana Knezevic <jovanak@google.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: subj tweak, removed changelog from patch description] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-22 17:33:06 +08:00
return rc;
selinux: overhaul sidtab to fix bug and improve performance Before this patch, during a policy reload the sidtab would become frozen and trying to map a new context to SID would be unable to add a new entry to sidtab and fail with -ENOMEM. Such failures are usually propagated into userspace, which has no way of distignuishing them from actual allocation failures and thus doesn't handle them gracefully. Such situation can be triggered e.g. by the following reproducer: while true; do load_policy; echo -n .; sleep 0.1; done & for (( i = 0; i < 1024; i++ )); do runcon -l s0:c$i echo -n x || break # or: # chcon -l s0:c$i <some_file> || break done This patch overhauls the sidtab so it doesn't need to be frozen during policy reload, thus solving the above problem. The new SID table leverages the fact that SIDs are allocated sequentially and are never invalidated and stores them in linear buckets indexed by a tree structure. This brings several advantages: 1. Fast SID -> context lookup - this lookup can now be done in logarithmic time complexity (usually in less than 4 array lookups) and can still be done safely without locking. 2. No need to re-search the whole table on reverse lookup miss - after acquiring the spinlock only the newly added entries need to be searched, which means that reverse lookups that end up inserting a new entry are now about twice as fast. 3. No need to freeze sidtab during policy reload - it is now possible to handle insertion of new entries even during sidtab conversion. The tree structure of the new sidtab is able to grow automatically to up to about 2^31 entries (at which point it should not have more than about 4 tree levels). The old sidtab had a theoretical capacity of almost 2^32 entries, but half of that is still more than enough since by that point the reverse table lookups would become unusably slow anyway... The number of entries per tree node is selected automatically so that each node fits into a single page, which should be the easiest size for kmalloc() to handle. Note that the cache for reverse lookup is preserved with equivalent logic. The only difference is that instead of storing pointers to the hash table nodes it stores just the indices of the cached entries. The new cache ensures that the indices are loaded/stored atomically, but it still has the drawback that concurrent cache updates may mess up the contents of the cache. Such situation however only reduces its effectivity, not the correctness of lookups. Tested by selinux-testsuite and thoroughly tortured by this simple stress test: ``` function rand_cat() { echo $(( $RANDOM % 1024 )) } function do_work() { while true; do echo -n "system_u:system_r:kernel_t:s0:c$(rand_cat),c$(rand_cat)" \ >/sys/fs/selinux/context 2>/dev/null || true done } do_work >/dev/null & do_work >/dev/null & do_work >/dev/null & while load_policy; do echo -n .; sleep 0.1; done kill %1 kill %2 kill %3 ``` Link: https://github.com/SELinuxProject/selinux-kernel/issues/38 Reported-by: Orion Poplawski <orion@nwra.com> Reported-by: Li Kun <hw.likun@huawei.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: most of sidtab.c merged by hand due to conflicts] [PM: checkpatch fixes in mls.c, services.c, sidtab.c] Signed-off-by: Paul Moore <paul@paul-moore.com>
2018-11-30 23:24:08 +08:00
}
selinux: sidtab reverse lookup hash table This replaces the reverse table lookup and reverse cache with a hashtable which improves cache-miss reverse-lookup times from O(n) to O(1)* and maintains the same performance as a reverse cache hit. This reduces the time needed to add a new sidtab entry from ~500us to 5us on a Pixel 3 when there are ~10,000 sidtab entries. The implementation uses the kernel's generic hashtable API, It uses the context's string represtation as the hash source, and the kernels generic string hashing algorithm full_name_hash() to reduce the string to a 32 bit value. This change also maintains the improvement introduced in commit ee1a84fdfeed ("selinux: overhaul sidtab to fix bug and improve performance") which removed the need to keep the current sidtab locked during policy reload. It does however introduce periodic locking of the target sidtab while converting the hashtable. Sidtab entries are never modified or removed, so the context struct stored in the sid_to_context tree can also be used for the context_to_sid hashtable to reduce memory usage. This bug was reported by: - On the selinux bug tracker. BUG: kernel softlockup due to too many SIDs/contexts #37 https://github.com/SELinuxProject/selinux-kernel/issues/37 - Jovana Knezevic on Android's bugtracker. Bug: 140252993 "During multi-user performance testing, we create and remove users many times. selinux_android_restorecon_pkgdir goes from 1ms to over 20ms after about 200 user creations and removals. Accumulated over ~280 packages, that adds a significant time to user creation, making perf benchmarks unreliable." * Hashtable lookup is only O(1) when n < the number of buckets. Signed-off-by: Jeff Vander Stoep <jeffv@google.com> Reported-by: Stephen Smalley <sds@tycho.nsa.gov> Reported-by: Jovana Knezevic <jovanak@google.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: subj tweak, removed changelog from patch description] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-22 17:33:06 +08:00
/*
* The hashtable can also be modified in sidtab_context_to_sid()
* so we must re-acquire the lock here.
*/
spin_lock_irqsave(&s->lock, flags);
sidtab_convert_hashtable(params->target, count);
spin_unlock_irqrestore(&s->lock, flags);
return 0;
}
selinux: move policy commit after updating selinuxfs With the refactoring of the policy load logic in the security server from the previous change, it is now possible to split out the committing of the new policy from security_load_policy() and perform it only after successful updating of selinuxfs. Change security_load_policy() to return the newly populated policy data structures to the caller, export selinux_policy_commit() for external callers, and introduce selinux_policy_cancel() to provide a way to cancel the policy load in the event of an error during updating of the selinuxfs directory tree. Further, rework the interfaces used by selinuxfs to get information from the policy when creating the new directory tree to take and act upon the new policy data structure rather than the current/active policy. Update selinuxfs to use these updated and new interfaces. While we are here, stop re-creating the policy_capabilities directory on each policy load since it does not depend on the policy, and stop trying to create the booleans and classes directories during the initial creation of selinuxfs since no information is available until first policy load. After this change, a failure while updating the booleans and class directories will cause the entire policy load to be canceled, leaving the original policy intact, and policy load notifications to userspace will only happen after a successful completion of updating those directories. This does not (yet) provide full atomicity with respect to the updating of the directory trees themselves. Signed-off-by: Stephen Smalley <stephen.smalley.work@gmail.com> Signed-off-by: Paul Moore <paul@paul-moore.com>
2020-08-07 21:29:34 +08:00
void sidtab_cancel_convert(struct sidtab *s)
{
unsigned long flags;
/* cancelling policy load - disable live convert of sidtab */
spin_lock_irqsave(&s->lock, flags);
s->convert = NULL;
spin_unlock_irqrestore(&s->lock, flags);
}
selinux: fix race between old and new sidtab commit 9ad6e9cb39c66366bf7b9aece114aca277981a1f upstream. Since commit 1b8b31a2e612 ("selinux: convert policy read-write lock to RCU"), there is a small window during policy load where the new policy pointer has already been installed, but some threads may still be holding the old policy pointer in their read-side RCU critical sections. This means that there may be conflicting attempts to add a new SID entry to both tables via sidtab_context_to_sid(). See also (and the rest of the thread): https://lore.kernel.org/selinux/CAFqZXNvfux46_f8gnvVvRYMKoes24nwm2n3sPbMjrB8vKTW00g@mail.gmail.com/ Fix this by installing the new policy pointer under the old sidtab's spinlock along with marking the old sidtab as "frozen". Then, if an attempt to add new entry to a "frozen" sidtab is detected, make sidtab_context_to_sid() return -ESTALE to indicate that a new policy has been installed and that the caller will have to abort the policy transaction and try again after re-taking the policy pointer (which is guaranteed to be a newer policy). This requires adding a retry-on-ESTALE logic to all callers of sidtab_context_to_sid(), but fortunately these are easy to determine and aren't that many. This seems to be the simplest solution for this problem, even if it looks somewhat ugly. Note that other places in the kernel (e.g. do_mknodat() in fs/namei.c) use similar stale-retry patterns, so I think it's reasonable. Cc: stable@vger.kernel.org Fixes: 1b8b31a2e612 ("selinux: convert policy read-write lock to RCU") Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Signed-off-by: Paul Moore <paul@paul-moore.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2021-04-07 15:24:43 +08:00
void sidtab_freeze_begin(struct sidtab *s, unsigned long *flags) __acquires(&s->lock)
{
spin_lock_irqsave(&s->lock, *flags);
s->frozen = true;
s->convert = NULL;
}
void sidtab_freeze_end(struct sidtab *s, unsigned long *flags) __releases(&s->lock)
{
spin_unlock_irqrestore(&s->lock, *flags);
}
selinux: cache the SID -> context string translation Translating a context struct to string can be quite slow, especially if the context has a lot of category bits set. This can cause quite noticeable performance impact in situations where the translation needs to be done repeatedly. A common example is a UNIX datagram socket with the SO_PASSSEC option enabled, which is used e.g. by systemd-journald when receiving log messages via datagram socket. This scenario can be reproduced with: cat /dev/urandom | base64 | logger & timeout 30s perf record -p $(pidof systemd-journald) -a -g kill %1 perf report -g none --pretty raw | grep security_secid_to_secctx Before the caching introduced by this patch, computing the context string (security_secid_to_secctx() function) takes up ~65% of systemd-journald's CPU time (assuming a context with 1024 categories set and Fedora x86_64 release kernel configs). After this patch (assuming near-perfect cache hit ratio) this overhead is reduced to just ~2%. This patch addresses the issue by caching a certain number (compile-time configurable) of recently used context strings to speed up repeated translations of the same context, while using only a small amount of memory. The cache is integrated into the existing sidtab table by adding a field to each entry, which when not NULL contains an RCU-protected pointer to a cache entry containing the cached string. The cache entries are kept in a linked list sorted according to how recently they were used. On a cache miss when the cache is full, the least recently used entry is removed to make space for the new entry. The patch migrates security_sid_to_context_core() to use the cache (also a few other functions where it was possible without too much fuss, but these mostly use the translation for logging in case of error, which is rare). Link: https://bugzilla.redhat.com/show_bug.cgi?id=1733259 Cc: Michal Sekletar <msekleta@redhat.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> Reviewed-by: Paul E. McKenney <paulmck@kernel.org> [PM: lots of merge fixups due to collisions with other sidtab patches] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-26 21:57:00 +08:00
static void sidtab_destroy_entry(struct sidtab_entry *entry)
{
context_destroy(&entry->context);
#if CONFIG_SECURITY_SELINUX_SID2STR_CACHE_SIZE > 0
kfree(rcu_dereference_raw(entry->cache));
#endif
}
selinux: overhaul sidtab to fix bug and improve performance Before this patch, during a policy reload the sidtab would become frozen and trying to map a new context to SID would be unable to add a new entry to sidtab and fail with -ENOMEM. Such failures are usually propagated into userspace, which has no way of distignuishing them from actual allocation failures and thus doesn't handle them gracefully. Such situation can be triggered e.g. by the following reproducer: while true; do load_policy; echo -n .; sleep 0.1; done & for (( i = 0; i < 1024; i++ )); do runcon -l s0:c$i echo -n x || break # or: # chcon -l s0:c$i <some_file> || break done This patch overhauls the sidtab so it doesn't need to be frozen during policy reload, thus solving the above problem. The new SID table leverages the fact that SIDs are allocated sequentially and are never invalidated and stores them in linear buckets indexed by a tree structure. This brings several advantages: 1. Fast SID -> context lookup - this lookup can now be done in logarithmic time complexity (usually in less than 4 array lookups) and can still be done safely without locking. 2. No need to re-search the whole table on reverse lookup miss - after acquiring the spinlock only the newly added entries need to be searched, which means that reverse lookups that end up inserting a new entry are now about twice as fast. 3. No need to freeze sidtab during policy reload - it is now possible to handle insertion of new entries even during sidtab conversion. The tree structure of the new sidtab is able to grow automatically to up to about 2^31 entries (at which point it should not have more than about 4 tree levels). The old sidtab had a theoretical capacity of almost 2^32 entries, but half of that is still more than enough since by that point the reverse table lookups would become unusably slow anyway... The number of entries per tree node is selected automatically so that each node fits into a single page, which should be the easiest size for kmalloc() to handle. Note that the cache for reverse lookup is preserved with equivalent logic. The only difference is that instead of storing pointers to the hash table nodes it stores just the indices of the cached entries. The new cache ensures that the indices are loaded/stored atomically, but it still has the drawback that concurrent cache updates may mess up the contents of the cache. Such situation however only reduces its effectivity, not the correctness of lookups. Tested by selinux-testsuite and thoroughly tortured by this simple stress test: ``` function rand_cat() { echo $(( $RANDOM % 1024 )) } function do_work() { while true; do echo -n "system_u:system_r:kernel_t:s0:c$(rand_cat),c$(rand_cat)" \ >/sys/fs/selinux/context 2>/dev/null || true done } do_work >/dev/null & do_work >/dev/null & do_work >/dev/null & while load_policy; do echo -n .; sleep 0.1; done kill %1 kill %2 kill %3 ``` Link: https://github.com/SELinuxProject/selinux-kernel/issues/38 Reported-by: Orion Poplawski <orion@nwra.com> Reported-by: Li Kun <hw.likun@huawei.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: most of sidtab.c merged by hand due to conflicts] [PM: checkpatch fixes in mls.c, services.c, sidtab.c] Signed-off-by: Paul Moore <paul@paul-moore.com>
2018-11-30 23:24:08 +08:00
static void sidtab_destroy_tree(union sidtab_entry_inner entry, u32 level)
{
selinux: overhaul sidtab to fix bug and improve performance Before this patch, during a policy reload the sidtab would become frozen and trying to map a new context to SID would be unable to add a new entry to sidtab and fail with -ENOMEM. Such failures are usually propagated into userspace, which has no way of distignuishing them from actual allocation failures and thus doesn't handle them gracefully. Such situation can be triggered e.g. by the following reproducer: while true; do load_policy; echo -n .; sleep 0.1; done & for (( i = 0; i < 1024; i++ )); do runcon -l s0:c$i echo -n x || break # or: # chcon -l s0:c$i <some_file> || break done This patch overhauls the sidtab so it doesn't need to be frozen during policy reload, thus solving the above problem. The new SID table leverages the fact that SIDs are allocated sequentially and are never invalidated and stores them in linear buckets indexed by a tree structure. This brings several advantages: 1. Fast SID -> context lookup - this lookup can now be done in logarithmic time complexity (usually in less than 4 array lookups) and can still be done safely without locking. 2. No need to re-search the whole table on reverse lookup miss - after acquiring the spinlock only the newly added entries need to be searched, which means that reverse lookups that end up inserting a new entry are now about twice as fast. 3. No need to freeze sidtab during policy reload - it is now possible to handle insertion of new entries even during sidtab conversion. The tree structure of the new sidtab is able to grow automatically to up to about 2^31 entries (at which point it should not have more than about 4 tree levels). The old sidtab had a theoretical capacity of almost 2^32 entries, but half of that is still more than enough since by that point the reverse table lookups would become unusably slow anyway... The number of entries per tree node is selected automatically so that each node fits into a single page, which should be the easiest size for kmalloc() to handle. Note that the cache for reverse lookup is preserved with equivalent logic. The only difference is that instead of storing pointers to the hash table nodes it stores just the indices of the cached entries. The new cache ensures that the indices are loaded/stored atomically, but it still has the drawback that concurrent cache updates may mess up the contents of the cache. Such situation however only reduces its effectivity, not the correctness of lookups. Tested by selinux-testsuite and thoroughly tortured by this simple stress test: ``` function rand_cat() { echo $(( $RANDOM % 1024 )) } function do_work() { while true; do echo -n "system_u:system_r:kernel_t:s0:c$(rand_cat),c$(rand_cat)" \ >/sys/fs/selinux/context 2>/dev/null || true done } do_work >/dev/null & do_work >/dev/null & do_work >/dev/null & while load_policy; do echo -n .; sleep 0.1; done kill %1 kill %2 kill %3 ``` Link: https://github.com/SELinuxProject/selinux-kernel/issues/38 Reported-by: Orion Poplawski <orion@nwra.com> Reported-by: Li Kun <hw.likun@huawei.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: most of sidtab.c merged by hand due to conflicts] [PM: checkpatch fixes in mls.c, services.c, sidtab.c] Signed-off-by: Paul Moore <paul@paul-moore.com>
2018-11-30 23:24:08 +08:00
u32 i;
if (level != 0) {
struct sidtab_node_inner *node = entry.ptr_inner;
if (!node)
return;
for (i = 0; i < SIDTAB_INNER_ENTRIES; i++)
sidtab_destroy_tree(node->entries[i], level - 1);
kfree(node);
} else {
struct sidtab_node_leaf *node = entry.ptr_leaf;
selinux: overhaul sidtab to fix bug and improve performance Before this patch, during a policy reload the sidtab would become frozen and trying to map a new context to SID would be unable to add a new entry to sidtab and fail with -ENOMEM. Such failures are usually propagated into userspace, which has no way of distignuishing them from actual allocation failures and thus doesn't handle them gracefully. Such situation can be triggered e.g. by the following reproducer: while true; do load_policy; echo -n .; sleep 0.1; done & for (( i = 0; i < 1024; i++ )); do runcon -l s0:c$i echo -n x || break # or: # chcon -l s0:c$i <some_file> || break done This patch overhauls the sidtab so it doesn't need to be frozen during policy reload, thus solving the above problem. The new SID table leverages the fact that SIDs are allocated sequentially and are never invalidated and stores them in linear buckets indexed by a tree structure. This brings several advantages: 1. Fast SID -> context lookup - this lookup can now be done in logarithmic time complexity (usually in less than 4 array lookups) and can still be done safely without locking. 2. No need to re-search the whole table on reverse lookup miss - after acquiring the spinlock only the newly added entries need to be searched, which means that reverse lookups that end up inserting a new entry are now about twice as fast. 3. No need to freeze sidtab during policy reload - it is now possible to handle insertion of new entries even during sidtab conversion. The tree structure of the new sidtab is able to grow automatically to up to about 2^31 entries (at which point it should not have more than about 4 tree levels). The old sidtab had a theoretical capacity of almost 2^32 entries, but half of that is still more than enough since by that point the reverse table lookups would become unusably slow anyway... The number of entries per tree node is selected automatically so that each node fits into a single page, which should be the easiest size for kmalloc() to handle. Note that the cache for reverse lookup is preserved with equivalent logic. The only difference is that instead of storing pointers to the hash table nodes it stores just the indices of the cached entries. The new cache ensures that the indices are loaded/stored atomically, but it still has the drawback that concurrent cache updates may mess up the contents of the cache. Such situation however only reduces its effectivity, not the correctness of lookups. Tested by selinux-testsuite and thoroughly tortured by this simple stress test: ``` function rand_cat() { echo $(( $RANDOM % 1024 )) } function do_work() { while true; do echo -n "system_u:system_r:kernel_t:s0:c$(rand_cat),c$(rand_cat)" \ >/sys/fs/selinux/context 2>/dev/null || true done } do_work >/dev/null & do_work >/dev/null & do_work >/dev/null & while load_policy; do echo -n .; sleep 0.1; done kill %1 kill %2 kill %3 ``` Link: https://github.com/SELinuxProject/selinux-kernel/issues/38 Reported-by: Orion Poplawski <orion@nwra.com> Reported-by: Li Kun <hw.likun@huawei.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: most of sidtab.c merged by hand due to conflicts] [PM: checkpatch fixes in mls.c, services.c, sidtab.c] Signed-off-by: Paul Moore <paul@paul-moore.com>
2018-11-30 23:24:08 +08:00
if (!node)
return;
for (i = 0; i < SIDTAB_LEAF_ENTRIES; i++)
selinux: cache the SID -> context string translation Translating a context struct to string can be quite slow, especially if the context has a lot of category bits set. This can cause quite noticeable performance impact in situations where the translation needs to be done repeatedly. A common example is a UNIX datagram socket with the SO_PASSSEC option enabled, which is used e.g. by systemd-journald when receiving log messages via datagram socket. This scenario can be reproduced with: cat /dev/urandom | base64 | logger & timeout 30s perf record -p $(pidof systemd-journald) -a -g kill %1 perf report -g none --pretty raw | grep security_secid_to_secctx Before the caching introduced by this patch, computing the context string (security_secid_to_secctx() function) takes up ~65% of systemd-journald's CPU time (assuming a context with 1024 categories set and Fedora x86_64 release kernel configs). After this patch (assuming near-perfect cache hit ratio) this overhead is reduced to just ~2%. This patch addresses the issue by caching a certain number (compile-time configurable) of recently used context strings to speed up repeated translations of the same context, while using only a small amount of memory. The cache is integrated into the existing sidtab table by adding a field to each entry, which when not NULL contains an RCU-protected pointer to a cache entry containing the cached string. The cache entries are kept in a linked list sorted according to how recently they were used. On a cache miss when the cache is full, the least recently used entry is removed to make space for the new entry. The patch migrates security_sid_to_context_core() to use the cache (also a few other functions where it was possible without too much fuss, but these mostly use the translation for logging in case of error, which is rare). Link: https://bugzilla.redhat.com/show_bug.cgi?id=1733259 Cc: Michal Sekletar <msekleta@redhat.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> Reviewed-by: Paul E. McKenney <paulmck@kernel.org> [PM: lots of merge fixups due to collisions with other sidtab patches] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-26 21:57:00 +08:00
sidtab_destroy_entry(&node->entries[i]);
selinux: overhaul sidtab to fix bug and improve performance Before this patch, during a policy reload the sidtab would become frozen and trying to map a new context to SID would be unable to add a new entry to sidtab and fail with -ENOMEM. Such failures are usually propagated into userspace, which has no way of distignuishing them from actual allocation failures and thus doesn't handle them gracefully. Such situation can be triggered e.g. by the following reproducer: while true; do load_policy; echo -n .; sleep 0.1; done & for (( i = 0; i < 1024; i++ )); do runcon -l s0:c$i echo -n x || break # or: # chcon -l s0:c$i <some_file> || break done This patch overhauls the sidtab so it doesn't need to be frozen during policy reload, thus solving the above problem. The new SID table leverages the fact that SIDs are allocated sequentially and are never invalidated and stores them in linear buckets indexed by a tree structure. This brings several advantages: 1. Fast SID -> context lookup - this lookup can now be done in logarithmic time complexity (usually in less than 4 array lookups) and can still be done safely without locking. 2. No need to re-search the whole table on reverse lookup miss - after acquiring the spinlock only the newly added entries need to be searched, which means that reverse lookups that end up inserting a new entry are now about twice as fast. 3. No need to freeze sidtab during policy reload - it is now possible to handle insertion of new entries even during sidtab conversion. The tree structure of the new sidtab is able to grow automatically to up to about 2^31 entries (at which point it should not have more than about 4 tree levels). The old sidtab had a theoretical capacity of almost 2^32 entries, but half of that is still more than enough since by that point the reverse table lookups would become unusably slow anyway... The number of entries per tree node is selected automatically so that each node fits into a single page, which should be the easiest size for kmalloc() to handle. Note that the cache for reverse lookup is preserved with equivalent logic. The only difference is that instead of storing pointers to the hash table nodes it stores just the indices of the cached entries. The new cache ensures that the indices are loaded/stored atomically, but it still has the drawback that concurrent cache updates may mess up the contents of the cache. Such situation however only reduces its effectivity, not the correctness of lookups. Tested by selinux-testsuite and thoroughly tortured by this simple stress test: ``` function rand_cat() { echo $(( $RANDOM % 1024 )) } function do_work() { while true; do echo -n "system_u:system_r:kernel_t:s0:c$(rand_cat),c$(rand_cat)" \ >/sys/fs/selinux/context 2>/dev/null || true done } do_work >/dev/null & do_work >/dev/null & do_work >/dev/null & while load_policy; do echo -n .; sleep 0.1; done kill %1 kill %2 kill %3 ``` Link: https://github.com/SELinuxProject/selinux-kernel/issues/38 Reported-by: Orion Poplawski <orion@nwra.com> Reported-by: Li Kun <hw.likun@huawei.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: most of sidtab.c merged by hand due to conflicts] [PM: checkpatch fixes in mls.c, services.c, sidtab.c] Signed-off-by: Paul Moore <paul@paul-moore.com>
2018-11-30 23:24:08 +08:00
kfree(node);
}
}
void sidtab_destroy(struct sidtab *s)
{
u32 i, level;
for (i = 0; i < SECINITSID_NUM; i++)
if (s->isids[i].set)
selinux: cache the SID -> context string translation Translating a context struct to string can be quite slow, especially if the context has a lot of category bits set. This can cause quite noticeable performance impact in situations where the translation needs to be done repeatedly. A common example is a UNIX datagram socket with the SO_PASSSEC option enabled, which is used e.g. by systemd-journald when receiving log messages via datagram socket. This scenario can be reproduced with: cat /dev/urandom | base64 | logger & timeout 30s perf record -p $(pidof systemd-journald) -a -g kill %1 perf report -g none --pretty raw | grep security_secid_to_secctx Before the caching introduced by this patch, computing the context string (security_secid_to_secctx() function) takes up ~65% of systemd-journald's CPU time (assuming a context with 1024 categories set and Fedora x86_64 release kernel configs). After this patch (assuming near-perfect cache hit ratio) this overhead is reduced to just ~2%. This patch addresses the issue by caching a certain number (compile-time configurable) of recently used context strings to speed up repeated translations of the same context, while using only a small amount of memory. The cache is integrated into the existing sidtab table by adding a field to each entry, which when not NULL contains an RCU-protected pointer to a cache entry containing the cached string. The cache entries are kept in a linked list sorted according to how recently they were used. On a cache miss when the cache is full, the least recently used entry is removed to make space for the new entry. The patch migrates security_sid_to_context_core() to use the cache (also a few other functions where it was possible without too much fuss, but these mostly use the translation for logging in case of error, which is rare). Link: https://bugzilla.redhat.com/show_bug.cgi?id=1733259 Cc: Michal Sekletar <msekleta@redhat.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> Reviewed-by: Paul E. McKenney <paulmck@kernel.org> [PM: lots of merge fixups due to collisions with other sidtab patches] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-26 21:57:00 +08:00
sidtab_destroy_entry(&s->isids[i].entry);
selinux: overhaul sidtab to fix bug and improve performance Before this patch, during a policy reload the sidtab would become frozen and trying to map a new context to SID would be unable to add a new entry to sidtab and fail with -ENOMEM. Such failures are usually propagated into userspace, which has no way of distignuishing them from actual allocation failures and thus doesn't handle them gracefully. Such situation can be triggered e.g. by the following reproducer: while true; do load_policy; echo -n .; sleep 0.1; done & for (( i = 0; i < 1024; i++ )); do runcon -l s0:c$i echo -n x || break # or: # chcon -l s0:c$i <some_file> || break done This patch overhauls the sidtab so it doesn't need to be frozen during policy reload, thus solving the above problem. The new SID table leverages the fact that SIDs are allocated sequentially and are never invalidated and stores them in linear buckets indexed by a tree structure. This brings several advantages: 1. Fast SID -> context lookup - this lookup can now be done in logarithmic time complexity (usually in less than 4 array lookups) and can still be done safely without locking. 2. No need to re-search the whole table on reverse lookup miss - after acquiring the spinlock only the newly added entries need to be searched, which means that reverse lookups that end up inserting a new entry are now about twice as fast. 3. No need to freeze sidtab during policy reload - it is now possible to handle insertion of new entries even during sidtab conversion. The tree structure of the new sidtab is able to grow automatically to up to about 2^31 entries (at which point it should not have more than about 4 tree levels). The old sidtab had a theoretical capacity of almost 2^32 entries, but half of that is still more than enough since by that point the reverse table lookups would become unusably slow anyway... The number of entries per tree node is selected automatically so that each node fits into a single page, which should be the easiest size for kmalloc() to handle. Note that the cache for reverse lookup is preserved with equivalent logic. The only difference is that instead of storing pointers to the hash table nodes it stores just the indices of the cached entries. The new cache ensures that the indices are loaded/stored atomically, but it still has the drawback that concurrent cache updates may mess up the contents of the cache. Such situation however only reduces its effectivity, not the correctness of lookups. Tested by selinux-testsuite and thoroughly tortured by this simple stress test: ``` function rand_cat() { echo $(( $RANDOM % 1024 )) } function do_work() { while true; do echo -n "system_u:system_r:kernel_t:s0:c$(rand_cat),c$(rand_cat)" \ >/sys/fs/selinux/context 2>/dev/null || true done } do_work >/dev/null & do_work >/dev/null & do_work >/dev/null & while load_policy; do echo -n .; sleep 0.1; done kill %1 kill %2 kill %3 ``` Link: https://github.com/SELinuxProject/selinux-kernel/issues/38 Reported-by: Orion Poplawski <orion@nwra.com> Reported-by: Li Kun <hw.likun@huawei.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: most of sidtab.c merged by hand due to conflicts] [PM: checkpatch fixes in mls.c, services.c, sidtab.c] Signed-off-by: Paul Moore <paul@paul-moore.com>
2018-11-30 23:24:08 +08:00
level = SIDTAB_MAX_LEVEL;
while (level && !s->roots[level].ptr_inner)
--level;
sidtab_destroy_tree(s->roots[level], level);
selinux: sidtab reverse lookup hash table This replaces the reverse table lookup and reverse cache with a hashtable which improves cache-miss reverse-lookup times from O(n) to O(1)* and maintains the same performance as a reverse cache hit. This reduces the time needed to add a new sidtab entry from ~500us to 5us on a Pixel 3 when there are ~10,000 sidtab entries. The implementation uses the kernel's generic hashtable API, It uses the context's string represtation as the hash source, and the kernels generic string hashing algorithm full_name_hash() to reduce the string to a 32 bit value. This change also maintains the improvement introduced in commit ee1a84fdfeed ("selinux: overhaul sidtab to fix bug and improve performance") which removed the need to keep the current sidtab locked during policy reload. It does however introduce periodic locking of the target sidtab while converting the hashtable. Sidtab entries are never modified or removed, so the context struct stored in the sid_to_context tree can also be used for the context_to_sid hashtable to reduce memory usage. This bug was reported by: - On the selinux bug tracker. BUG: kernel softlockup due to too many SIDs/contexts #37 https://github.com/SELinuxProject/selinux-kernel/issues/37 - Jovana Knezevic on Android's bugtracker. Bug: 140252993 "During multi-user performance testing, we create and remove users many times. selinux_android_restorecon_pkgdir goes from 1ms to over 20ms after about 200 user creations and removals. Accumulated over ~280 packages, that adds a significant time to user creation, making perf benchmarks unreliable." * Hashtable lookup is only O(1) when n < the number of buckets. Signed-off-by: Jeff Vander Stoep <jeffv@google.com> Reported-by: Stephen Smalley <sds@tycho.nsa.gov> Reported-by: Jovana Knezevic <jovanak@google.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> [PM: subj tweak, removed changelog from patch description] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-22 17:33:06 +08:00
/*
* The context_to_sid hashtable's objects are all shared
* with the isids array and context tree, and so don't need
* to be cleaned up here.
*/
}
selinux: cache the SID -> context string translation Translating a context struct to string can be quite slow, especially if the context has a lot of category bits set. This can cause quite noticeable performance impact in situations where the translation needs to be done repeatedly. A common example is a UNIX datagram socket with the SO_PASSSEC option enabled, which is used e.g. by systemd-journald when receiving log messages via datagram socket. This scenario can be reproduced with: cat /dev/urandom | base64 | logger & timeout 30s perf record -p $(pidof systemd-journald) -a -g kill %1 perf report -g none --pretty raw | grep security_secid_to_secctx Before the caching introduced by this patch, computing the context string (security_secid_to_secctx() function) takes up ~65% of systemd-journald's CPU time (assuming a context with 1024 categories set and Fedora x86_64 release kernel configs). After this patch (assuming near-perfect cache hit ratio) this overhead is reduced to just ~2%. This patch addresses the issue by caching a certain number (compile-time configurable) of recently used context strings to speed up repeated translations of the same context, while using only a small amount of memory. The cache is integrated into the existing sidtab table by adding a field to each entry, which when not NULL contains an RCU-protected pointer to a cache entry containing the cached string. The cache entries are kept in a linked list sorted according to how recently they were used. On a cache miss when the cache is full, the least recently used entry is removed to make space for the new entry. The patch migrates security_sid_to_context_core() to use the cache (also a few other functions where it was possible without too much fuss, but these mostly use the translation for logging in case of error, which is rare). Link: https://bugzilla.redhat.com/show_bug.cgi?id=1733259 Cc: Michal Sekletar <msekleta@redhat.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> Reviewed-by: Paul E. McKenney <paulmck@kernel.org> [PM: lots of merge fixups due to collisions with other sidtab patches] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-26 21:57:00 +08:00
#if CONFIG_SECURITY_SELINUX_SID2STR_CACHE_SIZE > 0
void sidtab_sid2str_put(struct sidtab *s, struct sidtab_entry *entry,
const char *str, u32 str_len)
{
struct sidtab_str_cache *cache, *victim = NULL;
selinux: fix sidtab string cache locking Avoiding taking a lock in an IRQ context is not enough to prevent deadlocks, as discovered by syzbot: === WARNING: SOFTIRQ-safe -> SOFTIRQ-unsafe lock order detected 5.5.0-syzkaller #0 Not tainted ----------------------------------------------------- syz-executor.0/8927 [HC0[0]:SC0[2]:HE1:SE0] is trying to acquire: ffff888027c94098 (&(&s->cache_lock)->rlock){+.+.}, at: spin_lock include/linux/spinlock.h:338 [inline] ffff888027c94098 (&(&s->cache_lock)->rlock){+.+.}, at: sidtab_sid2str_put.part.0+0x36/0x880 security/selinux/ss/sidtab.c:533 and this task is already holding: ffffffff898639b0 (&(&nf_conntrack_locks[i])->rlock){+.-.}, at: spin_lock include/linux/spinlock.h:338 [inline] ffffffff898639b0 (&(&nf_conntrack_locks[i])->rlock){+.-.}, at: nf_conntrack_lock+0x17/0x70 net/netfilter/nf_conntrack_core.c:91 which would create a new lock dependency: (&(&nf_conntrack_locks[i])->rlock){+.-.} -> (&(&s->cache_lock)->rlock){+.+.} but this new dependency connects a SOFTIRQ-irq-safe lock: (&(&nf_conntrack_locks[i])->rlock){+.-.} [...] other info that might help us debug this: Possible interrupt unsafe locking scenario: CPU0 CPU1 ---- ---- lock(&(&s->cache_lock)->rlock); local_irq_disable(); lock(&(&nf_conntrack_locks[i])->rlock); lock(&(&s->cache_lock)->rlock); <Interrupt> lock(&(&nf_conntrack_locks[i])->rlock); *** DEADLOCK *** [...] === Fix this by simply locking with irqsave/irqrestore and stop giving up on !in_task(). It makes the locking a bit slower, but it shouldn't make a big difference in real workloads. Under the scenario from [1] (only cache hits) it only increased the runtime overhead from the security_secid_to_secctx() function from ~2% to ~3% (it was ~5-65% before introducing the cache). [1] https://bugzilla.redhat.com/show_bug.cgi?id=1733259 Fixes: d97bd23c2d7d ("selinux: cache the SID -> context string translation") Reported-by: syzbot+61cba5033e2072d61806@syzkaller.appspotmail.com Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Acked-by: Stephen Smalley <sds@tycho.nsa.gov> Signed-off-by: Paul Moore <paul@paul-moore.com>
2020-02-03 16:50:23 +08:00
unsigned long flags;
selinux: cache the SID -> context string translation Translating a context struct to string can be quite slow, especially if the context has a lot of category bits set. This can cause quite noticeable performance impact in situations where the translation needs to be done repeatedly. A common example is a UNIX datagram socket with the SO_PASSSEC option enabled, which is used e.g. by systemd-journald when receiving log messages via datagram socket. This scenario can be reproduced with: cat /dev/urandom | base64 | logger & timeout 30s perf record -p $(pidof systemd-journald) -a -g kill %1 perf report -g none --pretty raw | grep security_secid_to_secctx Before the caching introduced by this patch, computing the context string (security_secid_to_secctx() function) takes up ~65% of systemd-journald's CPU time (assuming a context with 1024 categories set and Fedora x86_64 release kernel configs). After this patch (assuming near-perfect cache hit ratio) this overhead is reduced to just ~2%. This patch addresses the issue by caching a certain number (compile-time configurable) of recently used context strings to speed up repeated translations of the same context, while using only a small amount of memory. The cache is integrated into the existing sidtab table by adding a field to each entry, which when not NULL contains an RCU-protected pointer to a cache entry containing the cached string. The cache entries are kept in a linked list sorted according to how recently they were used. On a cache miss when the cache is full, the least recently used entry is removed to make space for the new entry. The patch migrates security_sid_to_context_core() to use the cache (also a few other functions where it was possible without too much fuss, but these mostly use the translation for logging in case of error, which is rare). Link: https://bugzilla.redhat.com/show_bug.cgi?id=1733259 Cc: Michal Sekletar <msekleta@redhat.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> Reviewed-by: Paul E. McKenney <paulmck@kernel.org> [PM: lots of merge fixups due to collisions with other sidtab patches] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-26 21:57:00 +08:00
/* do not cache invalid contexts */
if (entry->context.len)
return;
selinux: fix sidtab string cache locking Avoiding taking a lock in an IRQ context is not enough to prevent deadlocks, as discovered by syzbot: === WARNING: SOFTIRQ-safe -> SOFTIRQ-unsafe lock order detected 5.5.0-syzkaller #0 Not tainted ----------------------------------------------------- syz-executor.0/8927 [HC0[0]:SC0[2]:HE1:SE0] is trying to acquire: ffff888027c94098 (&(&s->cache_lock)->rlock){+.+.}, at: spin_lock include/linux/spinlock.h:338 [inline] ffff888027c94098 (&(&s->cache_lock)->rlock){+.+.}, at: sidtab_sid2str_put.part.0+0x36/0x880 security/selinux/ss/sidtab.c:533 and this task is already holding: ffffffff898639b0 (&(&nf_conntrack_locks[i])->rlock){+.-.}, at: spin_lock include/linux/spinlock.h:338 [inline] ffffffff898639b0 (&(&nf_conntrack_locks[i])->rlock){+.-.}, at: nf_conntrack_lock+0x17/0x70 net/netfilter/nf_conntrack_core.c:91 which would create a new lock dependency: (&(&nf_conntrack_locks[i])->rlock){+.-.} -> (&(&s->cache_lock)->rlock){+.+.} but this new dependency connects a SOFTIRQ-irq-safe lock: (&(&nf_conntrack_locks[i])->rlock){+.-.} [...] other info that might help us debug this: Possible interrupt unsafe locking scenario: CPU0 CPU1 ---- ---- lock(&(&s->cache_lock)->rlock); local_irq_disable(); lock(&(&nf_conntrack_locks[i])->rlock); lock(&(&s->cache_lock)->rlock); <Interrupt> lock(&(&nf_conntrack_locks[i])->rlock); *** DEADLOCK *** [...] === Fix this by simply locking with irqsave/irqrestore and stop giving up on !in_task(). It makes the locking a bit slower, but it shouldn't make a big difference in real workloads. Under the scenario from [1] (only cache hits) it only increased the runtime overhead from the security_secid_to_secctx() function from ~2% to ~3% (it was ~5-65% before introducing the cache). [1] https://bugzilla.redhat.com/show_bug.cgi?id=1733259 Fixes: d97bd23c2d7d ("selinux: cache the SID -> context string translation") Reported-by: syzbot+61cba5033e2072d61806@syzkaller.appspotmail.com Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Acked-by: Stephen Smalley <sds@tycho.nsa.gov> Signed-off-by: Paul Moore <paul@paul-moore.com>
2020-02-03 16:50:23 +08:00
spin_lock_irqsave(&s->cache_lock, flags);
selinux: cache the SID -> context string translation Translating a context struct to string can be quite slow, especially if the context has a lot of category bits set. This can cause quite noticeable performance impact in situations where the translation needs to be done repeatedly. A common example is a UNIX datagram socket with the SO_PASSSEC option enabled, which is used e.g. by systemd-journald when receiving log messages via datagram socket. This scenario can be reproduced with: cat /dev/urandom | base64 | logger & timeout 30s perf record -p $(pidof systemd-journald) -a -g kill %1 perf report -g none --pretty raw | grep security_secid_to_secctx Before the caching introduced by this patch, computing the context string (security_secid_to_secctx() function) takes up ~65% of systemd-journald's CPU time (assuming a context with 1024 categories set and Fedora x86_64 release kernel configs). After this patch (assuming near-perfect cache hit ratio) this overhead is reduced to just ~2%. This patch addresses the issue by caching a certain number (compile-time configurable) of recently used context strings to speed up repeated translations of the same context, while using only a small amount of memory. The cache is integrated into the existing sidtab table by adding a field to each entry, which when not NULL contains an RCU-protected pointer to a cache entry containing the cached string. The cache entries are kept in a linked list sorted according to how recently they were used. On a cache miss when the cache is full, the least recently used entry is removed to make space for the new entry. The patch migrates security_sid_to_context_core() to use the cache (also a few other functions where it was possible without too much fuss, but these mostly use the translation for logging in case of error, which is rare). Link: https://bugzilla.redhat.com/show_bug.cgi?id=1733259 Cc: Michal Sekletar <msekleta@redhat.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> Reviewed-by: Paul E. McKenney <paulmck@kernel.org> [PM: lots of merge fixups due to collisions with other sidtab patches] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-26 21:57:00 +08:00
cache = rcu_dereference_protected(entry->cache,
lockdep_is_held(&s->cache_lock));
if (cache) {
/* entry in cache - just bump to the head of LRU list */
list_move(&cache->lru_member, &s->cache_lru_list);
goto out_unlock;
}
cache = kmalloc(sizeof(struct sidtab_str_cache) + str_len, GFP_ATOMIC);
if (!cache)
goto out_unlock;
if (s->cache_free_slots == 0) {
/* pop a cache entry from the tail and free it */
victim = container_of(s->cache_lru_list.prev,
struct sidtab_str_cache, lru_member);
list_del(&victim->lru_member);
rcu_assign_pointer(victim->parent->cache, NULL);
} else {
s->cache_free_slots--;
}
cache->parent = entry;
cache->len = str_len;
memcpy(cache->str, str, str_len);
list_add(&cache->lru_member, &s->cache_lru_list);
rcu_assign_pointer(entry->cache, cache);
out_unlock:
selinux: fix sidtab string cache locking Avoiding taking a lock in an IRQ context is not enough to prevent deadlocks, as discovered by syzbot: === WARNING: SOFTIRQ-safe -> SOFTIRQ-unsafe lock order detected 5.5.0-syzkaller #0 Not tainted ----------------------------------------------------- syz-executor.0/8927 [HC0[0]:SC0[2]:HE1:SE0] is trying to acquire: ffff888027c94098 (&(&s->cache_lock)->rlock){+.+.}, at: spin_lock include/linux/spinlock.h:338 [inline] ffff888027c94098 (&(&s->cache_lock)->rlock){+.+.}, at: sidtab_sid2str_put.part.0+0x36/0x880 security/selinux/ss/sidtab.c:533 and this task is already holding: ffffffff898639b0 (&(&nf_conntrack_locks[i])->rlock){+.-.}, at: spin_lock include/linux/spinlock.h:338 [inline] ffffffff898639b0 (&(&nf_conntrack_locks[i])->rlock){+.-.}, at: nf_conntrack_lock+0x17/0x70 net/netfilter/nf_conntrack_core.c:91 which would create a new lock dependency: (&(&nf_conntrack_locks[i])->rlock){+.-.} -> (&(&s->cache_lock)->rlock){+.+.} but this new dependency connects a SOFTIRQ-irq-safe lock: (&(&nf_conntrack_locks[i])->rlock){+.-.} [...] other info that might help us debug this: Possible interrupt unsafe locking scenario: CPU0 CPU1 ---- ---- lock(&(&s->cache_lock)->rlock); local_irq_disable(); lock(&(&nf_conntrack_locks[i])->rlock); lock(&(&s->cache_lock)->rlock); <Interrupt> lock(&(&nf_conntrack_locks[i])->rlock); *** DEADLOCK *** [...] === Fix this by simply locking with irqsave/irqrestore and stop giving up on !in_task(). It makes the locking a bit slower, but it shouldn't make a big difference in real workloads. Under the scenario from [1] (only cache hits) it only increased the runtime overhead from the security_secid_to_secctx() function from ~2% to ~3% (it was ~5-65% before introducing the cache). [1] https://bugzilla.redhat.com/show_bug.cgi?id=1733259 Fixes: d97bd23c2d7d ("selinux: cache the SID -> context string translation") Reported-by: syzbot+61cba5033e2072d61806@syzkaller.appspotmail.com Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Acked-by: Stephen Smalley <sds@tycho.nsa.gov> Signed-off-by: Paul Moore <paul@paul-moore.com>
2020-02-03 16:50:23 +08:00
spin_unlock_irqrestore(&s->cache_lock, flags);
selinux: cache the SID -> context string translation Translating a context struct to string can be quite slow, especially if the context has a lot of category bits set. This can cause quite noticeable performance impact in situations where the translation needs to be done repeatedly. A common example is a UNIX datagram socket with the SO_PASSSEC option enabled, which is used e.g. by systemd-journald when receiving log messages via datagram socket. This scenario can be reproduced with: cat /dev/urandom | base64 | logger & timeout 30s perf record -p $(pidof systemd-journald) -a -g kill %1 perf report -g none --pretty raw | grep security_secid_to_secctx Before the caching introduced by this patch, computing the context string (security_secid_to_secctx() function) takes up ~65% of systemd-journald's CPU time (assuming a context with 1024 categories set and Fedora x86_64 release kernel configs). After this patch (assuming near-perfect cache hit ratio) this overhead is reduced to just ~2%. This patch addresses the issue by caching a certain number (compile-time configurable) of recently used context strings to speed up repeated translations of the same context, while using only a small amount of memory. The cache is integrated into the existing sidtab table by adding a field to each entry, which when not NULL contains an RCU-protected pointer to a cache entry containing the cached string. The cache entries are kept in a linked list sorted according to how recently they were used. On a cache miss when the cache is full, the least recently used entry is removed to make space for the new entry. The patch migrates security_sid_to_context_core() to use the cache (also a few other functions where it was possible without too much fuss, but these mostly use the translation for logging in case of error, which is rare). Link: https://bugzilla.redhat.com/show_bug.cgi?id=1733259 Cc: Michal Sekletar <msekleta@redhat.com> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Reviewed-by: Stephen Smalley <sds@tycho.nsa.gov> Tested-by: Stephen Smalley <sds@tycho.nsa.gov> Reviewed-by: Paul E. McKenney <paulmck@kernel.org> [PM: lots of merge fixups due to collisions with other sidtab patches] Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-11-26 21:57:00 +08:00
kfree_rcu(victim, rcu_member);
}
int sidtab_sid2str_get(struct sidtab *s, struct sidtab_entry *entry,
char **out, u32 *out_len)
{
struct sidtab_str_cache *cache;
int rc = 0;
if (entry->context.len)
return -ENOENT; /* do not cache invalid contexts */
rcu_read_lock();
cache = rcu_dereference(entry->cache);
if (!cache) {
rc = -ENOENT;
} else {
*out_len = cache->len;
if (out) {
*out = kmemdup(cache->str, cache->len, GFP_ATOMIC);
if (!*out)
rc = -ENOMEM;
}
}
rcu_read_unlock();
if (!rc && out)
sidtab_sid2str_put(s, entry, *out, *out_len);
return rc;
}
#endif /* CONFIG_SECURITY_SELINUX_SID2STR_CACHE_SIZE > 0 */