* [f2fs-dev] [PATCH 1/2] f2fs: fix race in concurrent f2fs_stop_gc_thread @ 2024-10-31 6:45 Long Li via Linux-f2fs-devel 2024-10-31 6:45 ` [f2fs-dev] [PATCH 2/2] Revert "f2fs: fix to avoid use-after-free in f2fs_stop_gc_thread()" Long Li via Linux-f2fs-devel 2024-11-01 2:56 ` [f2fs-dev] [PATCH 1/2] f2fs: fix race in concurrent f2fs_stop_gc_thread Chao Yu via Linux-f2fs-devel 0 siblings, 2 replies; 4+ messages in thread From: Long Li via Linux-f2fs-devel @ 2024-10-31 6:45 UTC (permalink / raw) To: jaegeuk, chao Cc: yi.zhang, lonuxli.64, yangerkun, linux-kernel, linux-f2fs-devel, leo.lilong In my test case, concurrent calls to f2fs shutdown report the following stack trace: Oops: general protection fault, probably for non-canonical address 0xc6cfff63bb5513fc: 0000 [#1] PREEMPT SMP PTI CPU: 0 UID: 0 PID: 678 Comm: f2fs_rep_shutdo Not tainted 6.12.0-rc5-next-20241029-g6fb2fa9805c5-dirty #85 Call Trace: <TASK> ? show_regs+0x8b/0xa0 ? __die_body+0x26/0xa0 ? die_addr+0x54/0x90 ? exc_general_protection+0x24b/0x5c0 ? asm_exc_general_protection+0x26/0x30 ? kthread_stop+0x46/0x390 f2fs_stop_gc_thread+0x6c/0x110 f2fs_do_shutdown+0x309/0x3a0 f2fs_ioc_shutdown+0x150/0x1c0 __f2fs_ioctl+0xffd/0x2ac0 f2fs_ioctl+0x76/0xe0 vfs_ioctl+0x23/0x60 __x64_sys_ioctl+0xce/0xf0 x64_sys_call+0x2b1b/0x4540 do_syscall_64+0xa7/0x240 entry_SYSCALL_64_after_hwframe+0x76/0x7e The root cause is a race condition in f2fs_stop_gc_thread() called from different f2fs shutdown paths: [CPU0] [CPU1] ---------------------- ----------------------- f2fs_stop_gc_thread f2fs_stop_gc_thread gc_th = sbi->gc_thread gc_th = sbi->gc_thread kfree(gc_th) sbi->gc_thread = NULL < gc_th != NULL > kthread_stop(gc_th->f2fs_gc_task) //UAF The commit c7f114d864ac ("f2fs: fix to avoid use-after-free in f2fs_stop_gc_thread()") attempted to fix this issue by using a read semaphore to prevent races between shutdown and remount threads, but itfails to prevent all race conditions. While upgrading to s_umount write lock in f2fs_do_shutdown() would fix the current issue, however, using s_umount lock requires extreme caution to avoid lock recursion. A better solution is to introduce a semaphore to prevent races between concurrent f2fs_stop_gc_thread calls. Fixes: 7950e9ac638e ("f2fs: stop gc/discard thread after fs shutdown") Signed-off-by: Long Li <leo.lilong@huawei.com> --- fs/f2fs/f2fs.h | 1 + fs/f2fs/gc.c | 9 +++++++-- fs/f2fs/super.c | 1 + 3 files changed, 9 insertions(+), 2 deletions(-) diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h index 3c6f3cce5779..7ae1e2a4789f 100644 --- a/fs/f2fs/f2fs.h +++ b/fs/f2fs/f2fs.h @@ -1679,6 +1679,7 @@ struct f2fs_sb_info { * race between GC and GC or CP */ struct f2fs_gc_kthread *gc_thread; /* GC thread */ + struct semaphore gc_clean_lock; /* semaphore for clean GC thread */ struct atgc_management am; /* atgc management */ unsigned int cur_victim_sec; /* current victim section num */ unsigned int gc_mode; /* current GC state */ diff --git a/fs/f2fs/gc.c b/fs/f2fs/gc.c index e40bdd12e36d..e1b8bf98b5fa 100644 --- a/fs/f2fs/gc.c +++ b/fs/f2fs/gc.c @@ -232,14 +232,19 @@ int f2fs_start_gc_thread(struct f2fs_sb_info *sbi) void f2fs_stop_gc_thread(struct f2fs_sb_info *sbi) { - struct f2fs_gc_kthread *gc_th = sbi->gc_thread; + struct f2fs_gc_kthread *gc_th; - if (!gc_th) + down(&sbi->gc_clean_lock); + gc_th = sbi->gc_thread; + if (!gc_th) { + up(&sbi->gc_clean_lock); return; + } kthread_stop(gc_th->f2fs_gc_task); wake_up_all(&gc_th->fggc_wq); kfree(gc_th); sbi->gc_thread = NULL; + up(&sbi->gc_clean_lock); } static int select_gc_type(struct f2fs_sb_info *sbi, int gc_type) diff --git a/fs/f2fs/super.c b/fs/f2fs/super.c index 80a53dbf1c38..47a15050ea9c 100644 --- a/fs/f2fs/super.c +++ b/fs/f2fs/super.c @@ -4419,6 +4419,7 @@ static int f2fs_fill_super(struct super_block *sb, void *data, int silent) /* initialize locks within allocated memory */ init_f2fs_rwsem(&sbi->gc_lock); + sema_init(&sbi->gc_clean_lock, 1); mutex_init(&sbi->writepages); init_f2fs_rwsem(&sbi->cp_global_sem); init_f2fs_rwsem(&sbi->node_write); -- 2.39.2 _______________________________________________ Linux-f2fs-devel mailing list Linux-f2fs-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel ^ permalink raw reply related [flat|nested] 4+ messages in thread
* [f2fs-dev] [PATCH 2/2] Revert "f2fs: fix to avoid use-after-free in f2fs_stop_gc_thread()" 2024-10-31 6:45 [f2fs-dev] [PATCH 1/2] f2fs: fix race in concurrent f2fs_stop_gc_thread Long Li via Linux-f2fs-devel @ 2024-10-31 6:45 ` Long Li via Linux-f2fs-devel 2024-11-01 2:56 ` [f2fs-dev] [PATCH 1/2] f2fs: fix race in concurrent f2fs_stop_gc_thread Chao Yu via Linux-f2fs-devel 1 sibling, 0 replies; 4+ messages in thread From: Long Li via Linux-f2fs-devel @ 2024-10-31 6:45 UTC (permalink / raw) To: jaegeuk, chao Cc: yi.zhang, lonuxli.64, yangerkun, linux-kernel, linux-f2fs-devel, leo.lilong This reverts commit c7f114d864ac91515bb07ac271e9824a20f5ed95. The race conditions between concurrent f2fs_stop_gc_thread() calls are now protected by a dedicated lock, making the additional s_umount lock protection unnecessary. Therefore, revert this patch. Signed-off-by: Long Li <leo.lilong@huawei.com> --- fs/f2fs/f2fs.h | 2 +- fs/f2fs/file.c | 11 ++--------- fs/f2fs/super.c | 2 +- 3 files changed, 4 insertions(+), 11 deletions(-) diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h index 7ae1e2a4789f..2143604ce416 100644 --- a/fs/f2fs/f2fs.h +++ b/fs/f2fs/f2fs.h @@ -3522,7 +3522,7 @@ int f2fs_setattr(struct mnt_idmap *idmap, struct dentry *dentry, int f2fs_truncate_hole(struct inode *inode, pgoff_t pg_start, pgoff_t pg_end); void f2fs_truncate_data_blocks_range(struct dnode_of_data *dn, int count); int f2fs_do_shutdown(struct f2fs_sb_info *sbi, unsigned int flag, - bool readonly, bool need_lock); + bool readonly); int f2fs_precache_extents(struct inode *inode); int f2fs_fileattr_get(struct dentry *dentry, struct fileattr *fa); int f2fs_fileattr_set(struct mnt_idmap *idmap, diff --git a/fs/f2fs/file.c b/fs/f2fs/file.c index 75a8b22da664..5d7b4fdae9c4 100644 --- a/fs/f2fs/file.c +++ b/fs/f2fs/file.c @@ -2318,7 +2318,7 @@ static int f2fs_ioc_abort_atomic_write(struct file *filp) } int f2fs_do_shutdown(struct f2fs_sb_info *sbi, unsigned int flag, - bool readonly, bool need_lock) + bool readonly) { struct super_block *sb = sbi->sb; int ret = 0; @@ -2365,19 +2365,12 @@ int f2fs_do_shutdown(struct f2fs_sb_info *sbi, unsigned int flag, if (readonly) goto out; - /* grab sb->s_umount to avoid racing w/ remount() */ - if (need_lock) - down_read(&sbi->sb->s_umount); - f2fs_stop_gc_thread(sbi); f2fs_stop_discard_thread(sbi); f2fs_drop_discard_cmd(sbi); clear_opt(sbi, DISCARD); - if (need_lock) - up_read(&sbi->sb->s_umount); - f2fs_update_time(sbi, REQ_TIME); out: @@ -2414,7 +2407,7 @@ static int f2fs_ioc_shutdown(struct file *filp, unsigned long arg) } } - ret = f2fs_do_shutdown(sbi, in, readonly, true); + ret = f2fs_do_shutdown(sbi, in, readonly); if (need_drop) mnt_drop_write_file(filp); diff --git a/fs/f2fs/super.c b/fs/f2fs/super.c index 47a15050ea9c..a720fb9ef196 100644 --- a/fs/f2fs/super.c +++ b/fs/f2fs/super.c @@ -2569,7 +2569,7 @@ static int f2fs_remount(struct super_block *sb, int *flags, char *data) static void f2fs_shutdown(struct super_block *sb) { - f2fs_do_shutdown(F2FS_SB(sb), F2FS_GOING_DOWN_NOSYNC, false, false); + f2fs_do_shutdown(F2FS_SB(sb), F2FS_GOING_DOWN_NOSYNC, false); } #ifdef CONFIG_QUOTA -- 2.39.2 _______________________________________________ Linux-f2fs-devel mailing list Linux-f2fs-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel ^ permalink raw reply related [flat|nested] 4+ messages in thread
* Re: [f2fs-dev] [PATCH 1/2] f2fs: fix race in concurrent f2fs_stop_gc_thread 2024-10-31 6:45 [f2fs-dev] [PATCH 1/2] f2fs: fix race in concurrent f2fs_stop_gc_thread Long Li via Linux-f2fs-devel 2024-10-31 6:45 ` [f2fs-dev] [PATCH 2/2] Revert "f2fs: fix to avoid use-after-free in f2fs_stop_gc_thread()" Long Li via Linux-f2fs-devel @ 2024-11-01 2:56 ` Chao Yu via Linux-f2fs-devel 2024-11-04 1:31 ` Long Li via Linux-f2fs-devel 1 sibling, 1 reply; 4+ messages in thread From: Chao Yu via Linux-f2fs-devel @ 2024-11-01 2:56 UTC (permalink / raw) To: Long Li, jaegeuk Cc: yi.zhang, lonuxli.64, yangerkun, linux-kernel, linux-f2fs-devel On 2024/10/31 14:45, Long Li wrote: > In my test case, concurrent calls to f2fs shutdown report the following > stack trace: > > Oops: general protection fault, probably for non-canonical address 0xc6cfff63bb5513fc: 0000 [#1] PREEMPT SMP PTI > CPU: 0 UID: 0 PID: 678 Comm: f2fs_rep_shutdo Not tainted 6.12.0-rc5-next-20241029-g6fb2fa9805c5-dirty #85 > Call Trace: > <TASK> > ? show_regs+0x8b/0xa0 > ? __die_body+0x26/0xa0 > ? die_addr+0x54/0x90 > ? exc_general_protection+0x24b/0x5c0 > ? asm_exc_general_protection+0x26/0x30 > ? kthread_stop+0x46/0x390 > f2fs_stop_gc_thread+0x6c/0x110 > f2fs_do_shutdown+0x309/0x3a0 > f2fs_ioc_shutdown+0x150/0x1c0 > __f2fs_ioctl+0xffd/0x2ac0 > f2fs_ioctl+0x76/0xe0 > vfs_ioctl+0x23/0x60 > __x64_sys_ioctl+0xce/0xf0 > x64_sys_call+0x2b1b/0x4540 > do_syscall_64+0xa7/0x240 > entry_SYSCALL_64_after_hwframe+0x76/0x7e > > The root cause is a race condition in f2fs_stop_gc_thread() called from > different f2fs shutdown paths: > > [CPU0] [CPU1] > ---------------------- ----------------------- > f2fs_stop_gc_thread f2fs_stop_gc_thread > gc_th = sbi->gc_thread > gc_th = sbi->gc_thread > kfree(gc_th) > sbi->gc_thread = NULL > < gc_th != NULL > > kthread_stop(gc_th->f2fs_gc_task) //UAF Hi Long, Thanks for catching this. Can we cover f2fs_stop_gc_thread() w/ write lock of s_umount in f2fs_do_shutdown()? Thanks, > > The commit c7f114d864ac ("f2fs: fix to avoid use-after-free in > f2fs_stop_gc_thread()") attempted to fix this issue by using a read > semaphore to prevent races between shutdown and remount threads, but > itfails to prevent all race conditions. > > While upgrading to s_umount write lock in f2fs_do_shutdown() would fix > the current issue, however, using s_umount lock requires extreme caution > to avoid lock recursion. A better solution is to introduce a semaphore > to prevent races between concurrent f2fs_stop_gc_thread calls. > > Fixes: 7950e9ac638e ("f2fs: stop gc/discard thread after fs shutdown") > Signed-off-by: Long Li <leo.lilong@huawei.com> > --- > fs/f2fs/f2fs.h | 1 + > fs/f2fs/gc.c | 9 +++++++-- > fs/f2fs/super.c | 1 + > 3 files changed, 9 insertions(+), 2 deletions(-) > > diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h > index 3c6f3cce5779..7ae1e2a4789f 100644 > --- a/fs/f2fs/f2fs.h > +++ b/fs/f2fs/f2fs.h > @@ -1679,6 +1679,7 @@ struct f2fs_sb_info { > * race between GC and GC or CP > */ > struct f2fs_gc_kthread *gc_thread; /* GC thread */ > + struct semaphore gc_clean_lock; /* semaphore for clean GC thread */ > struct atgc_management am; /* atgc management */ > unsigned int cur_victim_sec; /* current victim section num */ > unsigned int gc_mode; /* current GC state */ > diff --git a/fs/f2fs/gc.c b/fs/f2fs/gc.c > index e40bdd12e36d..e1b8bf98b5fa 100644 > --- a/fs/f2fs/gc.c > +++ b/fs/f2fs/gc.c > @@ -232,14 +232,19 @@ int f2fs_start_gc_thread(struct f2fs_sb_info *sbi) > > void f2fs_stop_gc_thread(struct f2fs_sb_info *sbi) > { > - struct f2fs_gc_kthread *gc_th = sbi->gc_thread; > + struct f2fs_gc_kthread *gc_th; > > - if (!gc_th) > + down(&sbi->gc_clean_lock); > + gc_th = sbi->gc_thread; > + if (!gc_th) { > + up(&sbi->gc_clean_lock); > return; > + } > kthread_stop(gc_th->f2fs_gc_task); > wake_up_all(&gc_th->fggc_wq); > kfree(gc_th); > sbi->gc_thread = NULL; > + up(&sbi->gc_clean_lock); > } > > static int select_gc_type(struct f2fs_sb_info *sbi, int gc_type) > diff --git a/fs/f2fs/super.c b/fs/f2fs/super.c > index 80a53dbf1c38..47a15050ea9c 100644 > --- a/fs/f2fs/super.c > +++ b/fs/f2fs/super.c > @@ -4419,6 +4419,7 @@ static int f2fs_fill_super(struct super_block *sb, void *data, int silent) > > /* initialize locks within allocated memory */ > init_f2fs_rwsem(&sbi->gc_lock); > + sema_init(&sbi->gc_clean_lock, 1); > mutex_init(&sbi->writepages); > init_f2fs_rwsem(&sbi->cp_global_sem); > init_f2fs_rwsem(&sbi->node_write); _______________________________________________ Linux-f2fs-devel mailing list Linux-f2fs-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel ^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: [f2fs-dev] [PATCH 1/2] f2fs: fix race in concurrent f2fs_stop_gc_thread 2024-11-01 2:56 ` [f2fs-dev] [PATCH 1/2] f2fs: fix race in concurrent f2fs_stop_gc_thread Chao Yu via Linux-f2fs-devel @ 2024-11-04 1:31 ` Long Li via Linux-f2fs-devel 0 siblings, 0 replies; 4+ messages in thread From: Long Li via Linux-f2fs-devel @ 2024-11-04 1:31 UTC (permalink / raw) To: Chao Yu, jaegeuk Cc: lonuxli.64, yangerkun, yi.zhang, linux-kernel, linux-f2fs-devel On Fri, Nov 01, 2024 at 10:56:59AM +0800, Chao Yu wrote: > On 2024/10/31 14:45, Long Li wrote: > > In my test case, concurrent calls to f2fs shutdown report the following > > stack trace: > > > > Oops: general protection fault, probably for non-canonical address 0xc6cfff63bb5513fc: 0000 [#1] PREEMPT SMP PTI > > CPU: 0 UID: 0 PID: 678 Comm: f2fs_rep_shutdo Not tainted 6.12.0-rc5-next-20241029-g6fb2fa9805c5-dirty #85 > > Call Trace: > > <TASK> > > ? show_regs+0x8b/0xa0 > > ? __die_body+0x26/0xa0 > > ? die_addr+0x54/0x90 > > ? exc_general_protection+0x24b/0x5c0 > > ? asm_exc_general_protection+0x26/0x30 > > ? kthread_stop+0x46/0x390 > > f2fs_stop_gc_thread+0x6c/0x110 > > f2fs_do_shutdown+0x309/0x3a0 > > f2fs_ioc_shutdown+0x150/0x1c0 > > __f2fs_ioctl+0xffd/0x2ac0 > > f2fs_ioctl+0x76/0xe0 > > vfs_ioctl+0x23/0x60 > > __x64_sys_ioctl+0xce/0xf0 > > x64_sys_call+0x2b1b/0x4540 > > do_syscall_64+0xa7/0x240 > > entry_SYSCALL_64_after_hwframe+0x76/0x7e > > > > The root cause is a race condition in f2fs_stop_gc_thread() called from > > different f2fs shutdown paths: > > > > [CPU0] [CPU1] > > ---------------------- ----------------------- > > f2fs_stop_gc_thread f2fs_stop_gc_thread > > gc_th = sbi->gc_thread > > gc_th = sbi->gc_thread > > kfree(gc_th) > > sbi->gc_thread = NULL > > < gc_th != NULL > > > kthread_stop(gc_th->f2fs_gc_task) //UAF > > Hi Long, > > Thanks for catching this. > > Can we cover f2fs_stop_gc_thread() w/ write lock of s_umount in > f2fs_do_shutdown()? > > Thanks, > Hi Chao, If you think holding write lock inf2fs_do_shutdown() is better, I'll send v2 with this change. Thanks, Long Li > > > > The commit c7f114d864ac ("f2fs: fix to avoid use-after-free in > > f2fs_stop_gc_thread()") attempted to fix this issue by using a read > > semaphore to prevent races between shutdown and remount threads, but > > itfails to prevent all race conditions. > > > > While upgrading to s_umount write lock in f2fs_do_shutdown() would fix > > the current issue, however, using s_umount lock requires extreme caution > > to avoid lock recursion. A better solution is to introduce a semaphore > > to prevent races between concurrent f2fs_stop_gc_thread calls. > > > > Fixes: 7950e9ac638e ("f2fs: stop gc/discard thread after fs shutdown") > > Signed-off-by: Long Li <leo.lilong@huawei.com> > > --- > > fs/f2fs/f2fs.h | 1 + > > fs/f2fs/gc.c | 9 +++++++-- > > fs/f2fs/super.c | 1 + > > 3 files changed, 9 insertions(+), 2 deletions(-) > > > > diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h > > index 3c6f3cce5779..7ae1e2a4789f 100644 > > --- a/fs/f2fs/f2fs.h > > +++ b/fs/f2fs/f2fs.h > > @@ -1679,6 +1679,7 @@ struct f2fs_sb_info { > > * race between GC and GC or CP > > */ > > struct f2fs_gc_kthread *gc_thread; /* GC thread */ > > + struct semaphore gc_clean_lock; /* semaphore for clean GC thread */ > > struct atgc_management am; /* atgc management */ > > unsigned int cur_victim_sec; /* current victim section num */ > > unsigned int gc_mode; /* current GC state */ > > diff --git a/fs/f2fs/gc.c b/fs/f2fs/gc.c > > index e40bdd12e36d..e1b8bf98b5fa 100644 > > --- a/fs/f2fs/gc.c > > +++ b/fs/f2fs/gc.c > > @@ -232,14 +232,19 @@ int f2fs_start_gc_thread(struct f2fs_sb_info *sbi) > > void f2fs_stop_gc_thread(struct f2fs_sb_info *sbi) > > { > > - struct f2fs_gc_kthread *gc_th = sbi->gc_thread; > > + struct f2fs_gc_kthread *gc_th; > > - if (!gc_th) > > + down(&sbi->gc_clean_lock); > > + gc_th = sbi->gc_thread; > > + if (!gc_th) { > > + up(&sbi->gc_clean_lock); > > return; > > + } > > kthread_stop(gc_th->f2fs_gc_task); > > wake_up_all(&gc_th->fggc_wq); > > kfree(gc_th); > > sbi->gc_thread = NULL; > > + up(&sbi->gc_clean_lock); > > } > > static int select_gc_type(struct f2fs_sb_info *sbi, int gc_type) > > diff --git a/fs/f2fs/super.c b/fs/f2fs/super.c > > index 80a53dbf1c38..47a15050ea9c 100644 > > --- a/fs/f2fs/super.c > > +++ b/fs/f2fs/super.c > > @@ -4419,6 +4419,7 @@ static int f2fs_fill_super(struct super_block *sb, void *data, int silent) > > /* initialize locks within allocated memory */ > > init_f2fs_rwsem(&sbi->gc_lock); > > + sema_init(&sbi->gc_clean_lock, 1); > > mutex_init(&sbi->writepages); > > init_f2fs_rwsem(&sbi->cp_global_sem); > > init_f2fs_rwsem(&sbi->node_write); > _______________________________________________ Linux-f2fs-devel mailing list Linux-f2fs-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel ^ permalink raw reply [flat|nested] 4+ messages in thread
end of thread, other threads:[~2024-11-04 1:18 UTC | newest] Thread overview: 4+ messages (download: mbox.gz follow: Atom feed -- links below jump to the message on this page -- 2024-10-31 6:45 [f2fs-dev] [PATCH 1/2] f2fs: fix race in concurrent f2fs_stop_gc_thread Long Li via Linux-f2fs-devel 2024-10-31 6:45 ` [f2fs-dev] [PATCH 2/2] Revert "f2fs: fix to avoid use-after-free in f2fs_stop_gc_thread()" Long Li via Linux-f2fs-devel 2024-11-01 2:56 ` [f2fs-dev] [PATCH 1/2] f2fs: fix race in concurrent f2fs_stop_gc_thread Chao Yu via Linux-f2fs-devel 2024-11-04 1:31 ` Long Li via Linux-f2fs-devel
This is a public inbox, see mirroring instructions for how to clone and mirror all data and code used for this inbox; as well as URLs for NNTP newsgroup(s).