writeback, cgroup: fix null-ptr-deref write in bdi_split_work_to_wbs
Subsystems (4)
PATCH v2 1/1 writeback, cgroup: fix null-ptr-deref write in bdi_split_work_to_wbs
KASAN report null-ptr-deref:
==================================================================
BUG: KASAN: null-ptr-deref in bdi_split_work_to_wbs+0x5c5/0x7b0 Write of size 8 at addr 0000000000000000 by task sync/943 CPU: 5 PID: 943 Comm: sync Tainted: 6.3.0-rc5-next-20230406-dirty #461 Call Trace: <TASK> dump_stack_lvl+0x7f/0xc0 print_report+0x2ba/0x340 kasan_report+0xc4/0x120 kasan_check_range+0x1b7/0x2e0 __kasan_check_write+0x24/0x40 bdi_split_work_to_wbs+0x5c5/0x7b0 sync_inodes_sb+0x195/0x630 sync_inodes_one_sb+0x3a/0x50 iterate_supers+0x106/0x1b0 ksys_sync+0x98/0x160
[...]
==================================================================
The race that causes the above issue is as follows:
cpu1 cpu2
-------------------------|------------------------- inode_switch_wbs
INIT_WORK(&isw->work, inode_switch_wbs_work_fn) queue_rcu_work(isw_wq, &isw->work)
// queue_work async
inode_switch_wbs_work_fn
wb_put_many(old_wb, nr_switched)
percpu_ref_put_many
ref->data->release(ref)
cgwb_release
queue_work(cgwb_release_wq, &wb->release_work)
// queue_work async
&wb->release_work
cgwb_release_workfn
ksys_sync
iterate_supers
sync_inodes_one_sb
sync_inodes_sb
bdi_split_work_to_wbs
kmalloc(sizeof(*work), GFP_ATOMIC)
// alloc memory failed
percpu_ref_exit
ref->data = NULL
kfree(data)
wb_get(wb)
percpu_ref_get(&wb->refcnt)
percpu_ref_get_many(ref, 1)
atomic_long_add(nr, &ref->data->count)
atomic64_add(i, v)
// trigger null-ptr-derefbdi_split_work_to_wbs() traverses &bdi->wb_list to split work into all wbs. If the allocation of new work fails, the on-stack fallback will be used and the reference count of the current wb is increased afterwards. If cgroup writeback membership switches occur before getting the reference count and the current wb is released as old_wd, then calling wb_get() or wb_put() will trigger the null pointer dereference above.
This issue was introduced in v4.3-rc7 (see fix tag1). Both sync_inodes_sb() and __writeback_inodes_sb_nr() calls to bdi_split_work_to_wbs() can trigger this issue. For scenarios called via sync_inodes_sb(), originally commit 7fc5854f8c6e ("writeback: synchronize sync(2) against cgroup writeback membership switches") reduced the possibility of the issue by adding wb_switch_rwsem, but in v5.14-rc1 (see fix tag2) removed the "inode_io_list_del_locked(inode, old_wb)" from inode_switch_wbs_work_fn() so that wb->state contains WB_has_dirty_io, thus old_wb is not skipped when traversing wbs in bdi_split_work_to_wbs(), and the issue becomes easily reproducible again.
To solve this problem, percpu_ref_exit() is called under RCU protection to avoid race between cgwb_release_workfn() and bdi_split_work_to_wbs(). Moreover, replace wb_get() with wb_tryget() in bdi_split_work_to_wbs(), and skip the current wb if wb_tryget() fails because the wb has already been shutdown.
---V1->V2:
Use RCU instead of wb_switch_rwsem to avoid race.
fs/fs-writeback.c | 17 ++++++++++------- mm/backing-dev.c | 12 ++++++++++-- 2 files changed, 20 insertions(+), 9 deletions(-)
fs/fs-writeback.c +10 -7
@@ -978,6 +978,16 @@ static void bdi_split_work_to_wbs(struct backing_dev_info *bdi,continue ; } +/* +* If wb_tryget fails, the wb has been shutdown, skip it. +* +* Pin @wb so that it stays on @bdi->wb_list. This allows +* continuing iteration from @wb after dropping and +* regrabbing rcu read lock. +*/ +if (!wb_tryget (wb)) +continue ; +/* alloc failed, execute synchronously using on-stack fallback */ work = &fallback_work; *work = *base_work;
@@ -986,13 +996,6 @@ static void bdi_split_work_to_wbs(struct backing_dev_info *bdi, work->done = &fallback_work_done;wb_queue_work (wb, work); - -/* -* Pin @wb so that it stays on @bdi->wb_list. This allows -* continuing iteration from @wb after dropping and -* regrabbing rcu read lock. -*/ -wb_get (wb); last_wb = wb;rcu_read_unlock ();
mm/backing-dev.c +10 -2
@@ -507,6 +507,15 @@ static LIST_HEAD(offline_cgwbs);static void cleanup_offline_cgwbs_workfn (struct work_struct *work); static DECLARE_WORK(cleanup_offline_cgwbs_work, cleanup_offline_cgwbs_workfn); +static void cgwb_free_rcu (struct rcu_head *rcu_head) +{ +struct bdi_writeback *wb =container_of (rcu_head, + struct bdi_writeback, rcu); + +percpu_ref_exit (&wb->refcnt); +kfree (wb); +} +static void cgwb_release_workfn (struct work_struct *work) { struct bdi_writeback *wb = container_of(work, struct bdi_writeback,
@@ -529,11 +538,10 @@ static void cgwb_release_workfn(struct work_struct *work)list_del (&wb->offline_node);spin_unlock_irq (&cgwb_lock); -percpu_ref_exit (&wb->refcnt);wb_exit (wb);bdi_put (bdi);WARN_ON_ONCE (!list_empty (&wb->b_attached)); -kfree_rcu (wb, rcu); +call_rcu (&wb->rcu, cgwb_free_rcu); }static void cgwb_release (struct percpu_ref *refcnt)
Signature · 1 line
2.31.1
Andrew Morton
maintainerQuoted 11 lines · On Mon, 10 Apr 2023 21:08:26 +0800 Baokun Li <[email protected]> wrote:
...
To solve this problem,
percpu_ref_exit()is called under RCU protection to avoid race betweencgwb_release_workfn()andbdi_split_work_to_wbs(). Moreover, replacewb_get()withwb_tryget()inbdi_split_work_to_wbs(), and skip the current wb ifwb_tryget()fails because the wb has already been shutdown.Fixes b817525a4a80 ("writeback: bdi_writeback iteration must not skip dying ones")Fixes f3b6a6df38aa ("writeback, cgroup: keep list of inodes attached to bdi_writeback")
Two Fixes: is awkward. The Fixes: serves a guide to which kernel versions should be patched, but those two commits are six years apart.
So... how far back should this fix be backported?
Quoted 2 lines
fs/fs-writeback.c | 17 ++++++++++------- mm/backing-dev.c | 12 ++++++++++--
Jens, which tree do you think should carry this?
Quoted 10 lines · Andrew Morton
Quoted 8 lines · On Mon, 10 Apr 2023 21:08:26 +0800 Baokun Li <[email protected]> wrote:
...
To solve this problem,
percpu_ref_exit()is called under RCU protection to avoid race betweencgwb_release_workfn()andbdi_split_work_to_wbs(). Moreover, replacewb_get()withwb_tryget()inbdi_split_work_to_wbs(), and skip the current wb ifwb_tryget()fails because the wb has already been shutdown.
Quoted 6 lines
Quoted 2 lines
Fixes b817525a4a80 ("writeback: bdi_writeback iteration must not skip dying ones")Fixes f3b6a6df38aa ("writeback, cgroup: keep list of inodes attached to bdi_writeback")Two Fixes: is awkward. The Fixes: serves a guide to which kernel versions should be patched, but those two commits are six years apart.
So... how far back should this fix be backported?
This issue was introduced in v4.3-rc7 by commit b817525a4a80 ("writeback: bdi_writeback iteration must not skip dying ones"), so anything that has this commit incorporated is problematic.
Another fix tag patch invalidates a previously unintentional fix, and then the problem becomes more easily reproducible. This fix tag can actually be removed, and is added here so that people who see the patch will know what happened.
Signature · 3 lines
With Best Regards, Baokun Li .
Quoted 78 lines · Baokun Li
KASAN report null-ptr-deref:
==================================================================BUG: KASAN: null-ptr-deref in bdi_split_work_to_wbs+0x5c5/0x7b0 Write of size 8 at addr 0000000000000000 by task sync/943 CPU: 5 PID: 943 Comm: sync Tainted: 6.3.0-rc5-next-20230406-dirty #461 Call Trace: <TASK> dump_stack_lvl+0x7f/0xc0 print_report+0x2ba/0x340 kasan_report+0xc4/0x120 kasan_check_range+0x1b7/0x2e0 __kasan_check_write+0x24/0x40 bdi_split_work_to_wbs+0x5c5/0x7b0 sync_inodes_sb+0x195/0x630 sync_inodes_one_sb+0x3a/0x50 iterate_supers+0x106/0x1b0 ksys_sync+0x98/0x160[...]
==================================================================The race that causes the above issue is as follows:
cpu1 cpu2
-------------------------|-------------------------inode_switch_wbs
INIT_WORK(&isw->work, inode_switch_wbs_work_fn)queue_rcu_work(isw_wq, &isw->work)
//queue_workasync
inode_switch_wbs_work_fn
wb_put_many(old_wb, nr_switched)percpu_ref_put_many ref->data->release(ref) cgwb_release queue_work(cgwb_release_wq, &wb->release_work) // queue_work async &wb->release_work cgwb_release_workfn ksys_sync iterate_supers sync_inodes_one_sb sync_inodes_sb bdi_split_work_to_wbs kmalloc(sizeof(*work), GFP_ATOMIC) // alloc memory failed percpu_ref_exit ref->data = NULL kfree(data) wb_get(wb) percpu_ref_get(&wb->refcnt) percpu_ref_get_many(ref, 1) atomic_long_add(nr, &ref->data->count) atomic64_add(i, v) // trigger null-ptr-deref
bdi_split_work_to_wbs()traverses&bdi->wb_listto split work into all wbs. If the allocation of new work fails, the on-stack fallback will be used and the reference count of the current wb is increased afterwards. If cgroup writeback membership switches occur before getting the reference count and the current wb is released asold_wd, then callingwb_get()orwb_put()will trigger the null pointer dereference above.This issue was introduced in v4.3-rc7 (see fix tag1). Both
sync_inodes_sb()and__writeback_inodes_sb_nr()calls tobdi_split_work_to_wbs()can trigger this issue. For scenarios called viasync_inodes_sb(), originally commit7fc5854f8c6e("writeback: synchronize sync(2) against cgroup writeback membership switches") reduced the possibility of the issue by addingwb_switch_rwsem, but in v5.14-rc1 (see fix tag2) removed the "inode_io_list_del_locked(inode, old_wb)" frominode_switch_wbs_work_fn()so thatwb->statecontainsWB_has_dirty_io, thusold_wbis not skipped when traversing wbs inbdi_split_work_to_wbs(), and the issue becomes easily reproducible again.To solve this problem,
percpu_ref_exit()is called under RCU protection to avoid race betweencgwb_release_workfn()andbdi_split_work_to_wbs(). Moreover, replacewb_get()withwb_tryget()inbdi_split_work_to_wbs(), and skip the current wb ifwb_tryget()fails because the wb has already been shutdown.
Quoted 7 lines
Fixes b817525a4a80 ("writeback: bdi_writeback iteration must not skip dying ones")Fixes f3b6a6df38aa ("writeback, cgroup: keep list of inodes attached to bdi_writeback")Signed-off-by Baokun Li <[email protected]>---
V1->V2:
Use RCU instead ofwb_switch_rwsemto avoid race.
The cgwb shutdown code is really messy. But your change looks good to me and I don't see an easier way around this race. Feel free to add:
Honza
Quoted 76 lines
fs/fs-writeback.c | 17 ++++++++++------- mm/backing-dev.c | 12 ++++++++++-- 2 files changed, 20 insertions(+), 9 deletions(-)fs/fs-writeback.c +10
-7@@ -978,6 +978,16 @@ static void bdi_split_work_to_wbs(struct backing_dev_info *bdi,continue ; } +/* +* If wb_tryget fails, the wb has been shutdown, skip it. +* +* Pin @wb so that it stays on @bdi->wb_list. This allows +* continuing iteration from @wb after dropping and +* regrabbing rcu read lock. +*/ +if (!wb_tryget (wb)) +continue ; +/* alloc failed, execute synchronously using on-stack fallback */ work = &fallback_work; *work = *base_work;@@ -986,13 +996,6 @@ static void bdi_split_work_to_wbs(struct backing_dev_info *bdi, work->done = &fallback_work_done;wb_queue_work (wb, work); - -/* -* Pin @wb so that it stays on @bdi->wb_list. This allows -* continuing iteration from @wb after dropping and -* regrabbing rcu read lock. -*/ -wb_get (wb); last_wb = wb;rcu_read_unlock ();mm/backing-dev.c +10
-2@@ -507,6 +507,15 @@ static LIST_HEAD(offline_cgwbs);static void cleanup_offline_cgwbs_workfn (struct work_struct *work); static DECLARE_WORK(cleanup_offline_cgwbs_work, cleanup_offline_cgwbs_workfn); +static void cgwb_free_rcu (struct rcu_head *rcu_head) +{ +struct bdi_writeback *wb =container_of (rcu_head, + struct bdi_writeback, rcu); + +percpu_ref_exit (&wb->refcnt); +kfree (wb); +} +static void cgwb_release_workfn (struct work_struct *work) { struct bdi_writeback *wb = container_of(work, struct bdi_writeback,@@ -529,11 +538,10 @@ static void cgwb_release_workfn(struct work_struct *work)list_del (&wb->offline_node);spin_unlock_irq (&cgwb_lock); -percpu_ref_exit (&wb->refcnt);wb_exit (wb);bdi_put (bdi);WARN_ON_ONCE (!list_empty (&wb->b_attached)); -kfree_rcu (wb, rcu); +call_rcu (&wb->rcu, cgwb_free_rcu); }static void cgwb_release (struct percpu_ref *refcnt)Signature · 1 line
2.31.1
Signature · 2 lines
Jan Kara <[email protected]> SUSE Labs, CR
Quoted 78 lines · Baokun Li
KASAN report null-ptr-deref:
==================================================================BUG: KASAN: null-ptr-deref in bdi_split_work_to_wbs+0x5c5/0x7b0 Write of size 8 at addr 0000000000000000 by task sync/943 CPU: 5 PID: 943 Comm: sync Tainted: 6.3.0-rc5-next-20230406-dirty #461 Call Trace: <TASK> dump_stack_lvl+0x7f/0xc0 print_report+0x2ba/0x340 kasan_report+0xc4/0x120 kasan_check_range+0x1b7/0x2e0 __kasan_check_write+0x24/0x40 bdi_split_work_to_wbs+0x5c5/0x7b0 sync_inodes_sb+0x195/0x630 sync_inodes_one_sb+0x3a/0x50 iterate_supers+0x106/0x1b0 ksys_sync+0x98/0x160[...]
==================================================================The race that causes the above issue is as follows:
cpu1 cpu2
-------------------------|-------------------------inode_switch_wbs
INIT_WORK(&isw->work, inode_switch_wbs_work_fn)queue_rcu_work(isw_wq, &isw->work)
//queue_workasync
inode_switch_wbs_work_fn
wb_put_many(old_wb, nr_switched)percpu_ref_put_many ref->data->release(ref) cgwb_release queue_work(cgwb_release_wq, &wb->release_work) // queue_work async &wb->release_work cgwb_release_workfn ksys_sync iterate_supers sync_inodes_one_sb sync_inodes_sb bdi_split_work_to_wbs kmalloc(sizeof(*work), GFP_ATOMIC) // alloc memory failed percpu_ref_exit ref->data = NULL kfree(data) wb_get(wb) percpu_ref_get(&wb->refcnt) percpu_ref_get_many(ref, 1) atomic_long_add(nr, &ref->data->count) atomic64_add(i, v) // trigger null-ptr-deref
bdi_split_work_to_wbs()traverses&bdi->wb_listto split work into all wbs. If the allocation of new work fails, the on-stack fallback will be used and the reference count of the current wb is increased afterwards. If cgroup writeback membership switches occur before getting the reference count and the current wb is released asold_wd, then callingwb_get()orwb_put()will trigger the null pointer dereference above.This issue was introduced in v4.3-rc7 (see fix tag1). Both
sync_inodes_sb()and__writeback_inodes_sb_nr()calls tobdi_split_work_to_wbs()can trigger this issue. For scenarios called viasync_inodes_sb(), originally commit7fc5854f8c6e("writeback: synchronize sync(2) against cgroup writeback membership switches") reduced the possibility of the issue by addingwb_switch_rwsem, but in v5.14-rc1 (see fix tag2) removed the "inode_io_list_del_locked(inode, old_wb)" frominode_switch_wbs_work_fn()so thatwb->statecontainsWB_has_dirty_io, thusold_wbis not skipped when traversing wbs inbdi_split_work_to_wbs(), and the issue becomes easily reproducible again.To solve this problem,
percpu_ref_exit()is called under RCU protection to avoid race betweencgwb_release_workfn()andbdi_split_work_to_wbs(). Moreover, replacewb_get()withwb_tryget()inbdi_split_work_to_wbs(), and skip the current wb ifwb_tryget()fails because the wb has already been shutdown.
Quoted 4 lines
Fixes b817525a4a80 ("writeback: bdi_writeback iteration must not skip dying ones")Fixes f3b6a6df38aa ("writeback, cgroup: keep list of inodes attached to bdi_writeback")Signed-off-by Baokun Li <[email protected]>
Thanks.
Signature · 1 line
tejun
Andrew Morton
maintainerQuoted 72 lines · On Mon, 10 Apr 2023 21:08:26 +0800 Baokun Li <[email protected]> wrote:
KASAN report null-ptr-deref:
==================================================================BUG: KASAN: null-ptr-deref in bdi_split_work_to_wbs+0x5c5/0x7b0 Write of size 8 at addr 0000000000000000 by task sync/943 CPU: 5 PID: 943 Comm: sync Tainted: 6.3.0-rc5-next-20230406-dirty #461 Call Trace: <TASK> dump_stack_lvl+0x7f/0xc0 print_report+0x2ba/0x340 kasan_report+0xc4/0x120 kasan_check_range+0x1b7/0x2e0 __kasan_check_write+0x24/0x40 bdi_split_work_to_wbs+0x5c5/0x7b0 sync_inodes_sb+0x195/0x630 sync_inodes_one_sb+0x3a/0x50 iterate_supers+0x106/0x1b0 ksys_sync+0x98/0x160[...]
==================================================================The race that causes the above issue is as follows:
cpu1 cpu2
-------------------------|-------------------------inode_switch_wbs
INIT_WORK(&isw->work, inode_switch_wbs_work_fn)queue_rcu_work(isw_wq, &isw->work)
//queue_workasync
inode_switch_wbs_work_fn
wb_put_many(old_wb, nr_switched)percpu_ref_put_many ref->data->release(ref) cgwb_release queue_work(cgwb_release_wq, &wb->release_work) // queue_work async &wb->release_work cgwb_release_workfn ksys_sync iterate_supers sync_inodes_one_sb sync_inodes_sb bdi_split_work_to_wbs kmalloc(sizeof(*work), GFP_ATOMIC) // alloc memory failed percpu_ref_exit ref->data = NULL kfree(data) wb_get(wb) percpu_ref_get(&wb->refcnt) percpu_ref_get_many(ref, 1) atomic_long_add(nr, &ref->data->count) atomic64_add(i, v) // trigger null-ptr-deref
bdi_split_work_to_wbs()traverses&bdi->wb_listto split work into all wbs. If the allocation of new work fails, the on-stack fallback will be used and the reference count of the current wb is increased afterwards. If cgroup writeback membership switches occur before getting the reference count and the current wb is released asold_wd, then callingwb_get()orwb_put()will trigger the null pointer dereference above.This issue was introduced in v4.3-rc7 (see fix tag1). Both
sync_inodes_sb()and__writeback_inodes_sb_nr()calls tobdi_split_work_to_wbs()can trigger this issue. For scenarios called viasync_inodes_sb(), originally commit7fc5854f8c6e("writeback: synchronize sync(2) against cgroup writeback membership switches") reduced the possibility of the issue by addingwb_switch_rwsem, but in v5.14-rc1 (see fix tag2) removed the "inode_io_list_del_locked(inode, old_wb)" frominode_switch_wbs_work_fn()so thatwb->statecontainsWB_has_dirty_io, thusold_wbis not skipped when traversing wbs inbdi_split_work_to_wbs(), and the issue becomes easily reproducible again.
Quoted 8 lines
To solve this problem,
percpu_ref_exit()is called under RCU protection to avoid race betweencgwb_release_workfn()andbdi_split_work_to_wbs(). Moreover, replacewb_get()withwb_tryget()inbdi_split_work_to_wbs(), and skip the current wb ifwb_tryget()fails because the wb has already been shutdown.Fixes b817525a4a80 ("writeback: bdi_writeback iteration must not skip dying ones")Fixes f3b6a6df38aa ("writeback, cgroup: keep list of inodes attached to bdi_writeback")
Cc Roman for this second commit.
Quoted 1 line
Having two Fixes: is awkward. These serve as a guide to tell -stable maintainers which kernels need the fix. Can we be more precise?
Quoted 38 lines
fs/fs-writeback.c +10
-7@@ -978,6 +978,16 @@ static void bdi_split_work_to_wbs(struct backing_dev_info *bdi,continue ; } +/* +* If wb_tryget fails, the wb has been shutdown, skip it. +* +* Pin @wb so that it stays on @bdi->wb_list. This allows +* continuing iteration from @wb after dropping and +* regrabbing rcu read lock. +*/ +if (!wb_tryget (wb)) +continue ; +/* alloc failed, execute synchronously using on-stack fallback */ work = &fallback_work; *work = *base_work;@@ -986,13 +996,6 @@ static void bdi_split_work_to_wbs(struct backing_dev_info *bdi, work->done = &fallback_work_done;wb_queue_work (wb, work); - -/* -* Pin @wb so that it stays on @bdi->wb_list. This allows -* continuing iteration from @wb after dropping and -* regrabbing rcu read lock. -*/ -wb_get (wb); last_wb = wb;rcu_read_unlock ();mm/backing-dev.c +0
-0
mm/backing-dev.c static LIST_HEAD(offline_cgwbs); · line 513
Quoted 7 lines
mm/backing-dev.c +4
-0@@ -507,6 +507,15 @@ static LIST_HEAD(offline_cgwbs);static void cleanup_offline_cgwbs_workfn (struct work_struct *work); static DECLARE_WORK(cleanup_offline_cgwbs_work, cleanup_offline_cgwbs_workfn); +static void cgwb_free_rcu (struct rcu_head *rcu_head) +{ +struct bdi_writeback *wb =container_of (rcu_head, + struct bdi_writeback, rcu);
nit:
struct bdi_writeback *wb; wb = container_of(rcu_head, struct bdi_writeback, rcu);
looks nicer, no?
Quoted 20 lines
mm/backing-dev.c +5
-2+percpu_ref_exit (&wb->refcnt); +kfree (wb); +} +static void cgwb_release_workfn (struct work_struct *work) { struct bdi_writeback *wb = container_of(work, struct bdi_writeback,@@ -529,11 +538,10 @@ static void cgwb_release_workfn(struct work_struct *work)list_del (&wb->offline_node);spin_unlock_irq (&cgwb_lock); -percpu_ref_exit (&wb->refcnt);wb_exit (wb);bdi_put (bdi);WARN_ON_ONCE (!list_empty (&wb->b_attached)); -kfree_rcu (wb, rcu); +call_rcu (&wb->rcu, cgwb_free_rcu); }static void cgwb_release (struct percpu_ref *refcnt)
Roman Gushchin
Acked-byQuoted 80 lines · Andrew Morton
Quoted 78 lines · On Mon, 10 Apr 2023 21:08:26 +0800 Baokun Li <[email protected]> wrote:
KASAN report null-ptr-deref:
==================================================================BUG: KASAN: null-ptr-deref in bdi_split_work_to_wbs+0x5c5/0x7b0 Write of size 8 at addr 0000000000000000 by task sync/943 CPU: 5 PID: 943 Comm: sync Tainted: 6.3.0-rc5-next-20230406-dirty #461 Call Trace: <TASK> dump_stack_lvl+0x7f/0xc0 print_report+0x2ba/0x340 kasan_report+0xc4/0x120 kasan_check_range+0x1b7/0x2e0 __kasan_check_write+0x24/0x40 bdi_split_work_to_wbs+0x5c5/0x7b0 sync_inodes_sb+0x195/0x630 sync_inodes_one_sb+0x3a/0x50 iterate_supers+0x106/0x1b0 ksys_sync+0x98/0x160[...]
==================================================================The race that causes the above issue is as follows:
cpu1 cpu2
-------------------------|-------------------------inode_switch_wbs
INIT_WORK(&isw->work, inode_switch_wbs_work_fn)queue_rcu_work(isw_wq, &isw->work)
//queue_workasync
inode_switch_wbs_work_fn
wb_put_many(old_wb, nr_switched)percpu_ref_put_many ref->data->release(ref) cgwb_release queue_work(cgwb_release_wq, &wb->release_work) // queue_work async &wb->release_work cgwb_release_workfn ksys_sync iterate_supers sync_inodes_one_sb sync_inodes_sb bdi_split_work_to_wbs kmalloc(sizeof(*work), GFP_ATOMIC) // alloc memory failed percpu_ref_exit ref->data = NULL kfree(data) wb_get(wb) percpu_ref_get(&wb->refcnt) percpu_ref_get_many(ref, 1) atomic_long_add(nr, &ref->data->count) atomic64_add(i, v) // trigger null-ptr-deref
bdi_split_work_to_wbs()traverses&bdi->wb_listto split work into all wbs. If the allocation of new work fails, the on-stack fallback will be used and the reference count of the current wb is increased afterwards. If cgroup writeback membership switches occur before getting the reference count and the current wb is released asold_wd, then callingwb_get()orwb_put()will trigger the null pointer dereference above.This issue was introduced in v4.3-rc7 (see fix tag1). Both
sync_inodes_sb()and__writeback_inodes_sb_nr()calls tobdi_split_work_to_wbs()can trigger this issue. For scenarios called viasync_inodes_sb(), originally commit7fc5854f8c6e("writeback: synchronize sync(2) against cgroup writeback membership switches") reduced the possibility of the issue by addingwb_switch_rwsem, but in v5.14-rc1 (see fix tag2) removed the "inode_io_list_del_locked(inode, old_wb)" frominode_switch_wbs_work_fn()so thatwb->statecontainsWB_has_dirty_io, thusold_wbis not skipped when traversing wbs inbdi_split_work_to_wbs(), and the issue becomes easily reproducible again.To solve this problem,
percpu_ref_exit()is called under RCU protection to avoid race betweencgwb_release_workfn()andbdi_split_work_to_wbs(). Moreover, replacewb_get()withwb_tryget()inbdi_split_work_to_wbs(), and skip the current wb ifwb_tryget()fails because the wb has already been shutdown.
Quoted 4 lines
Quoted 2 lines
Fixes b817525a4a80 ("writeback: bdi_writeback iteration must not skip dying ones")Fixes f3b6a6df38aa ("writeback, cgroup: keep list of inodes attached to bdi_writeback")Cc Roman for this second commit.
Thanks for the heads up!
The patch looks good to me.
Thanks!
Quoted 80 lines · Andrew Morton
Quoted 78 lines · On Mon, 10 Apr 2023 21:08:26 +0800 Baokun Li <[email protected]> wrote:
KASAN report null-ptr-deref:
==================================================================BUG: KASAN: null-ptr-deref in bdi_split_work_to_wbs+0x5c5/0x7b0 Write of size 8 at addr 0000000000000000 by task sync/943 CPU: 5 PID: 943 Comm: sync Tainted: 6.3.0-rc5-next-20230406-dirty #461 Call Trace: <TASK> dump_stack_lvl+0x7f/0xc0 print_report+0x2ba/0x340 kasan_report+0xc4/0x120 kasan_check_range+0x1b7/0x2e0 __kasan_check_write+0x24/0x40 bdi_split_work_to_wbs+0x5c5/0x7b0 sync_inodes_sb+0x195/0x630 sync_inodes_one_sb+0x3a/0x50 iterate_supers+0x106/0x1b0 ksys_sync+0x98/0x160[...]
==================================================================The race that causes the above issue is as follows:
cpu1 cpu2
-------------------------|-------------------------inode_switch_wbs
INIT_WORK(&isw->work, inode_switch_wbs_work_fn)queue_rcu_work(isw_wq, &isw->work)
//queue_workasync
inode_switch_wbs_work_fnwb_put_many(old_wb, nr_switched) percpu_ref_put_many ref->data->release(ref) cgwb_release queue_work(cgwb_release_wq, &wb->release_work) // queue_work async &wb->release_work cgwb_release_workfn ksys_sync iterate_supers sync_inodes_one_sb sync_inodes_sb bdi_split_work_to_wbs kmalloc(sizeof(*work), GFP_ATOMIC) // alloc memory failed percpu_ref_exit ref->data = NULL kfree(data) wb_get(wb) percpu_ref_get(&wb->refcnt) percpu_ref_get_many(ref, 1) atomic_long_add(nr, &ref->data->count) atomic64_add(i, v) // trigger null-ptr-deref
bdi_split_work_to_wbs()traverses&bdi->wb_listto split work into all wbs. If the allocation of new work fails, the on-stack fallback will be used and the reference count of the current wb is increased afterwards. If cgroup writeback membership switches occur before getting the reference count and the current wb is released asold_wd, then callingwb_get()orwb_put()will trigger the null pointer dereference above.This issue was introduced in v4.3-rc7 (see fix tag1). Both
sync_inodes_sb()and__writeback_inodes_sb_nr()calls tobdi_split_work_to_wbs()can trigger this issue. For scenarios called viasync_inodes_sb(), originally commit7fc5854f8c6e("writeback: synchronize sync(2) against cgroup writeback membership switches") reduced the possibility of the issue by addingwb_switch_rwsem, but in v5.14-rc1 (see fix tag2) removed the "inode_io_list_del_locked(inode, old_wb)" frominode_switch_wbs_work_fn()so thatwb->statecontainsWB_has_dirty_io, thusold_wbis not skipped when traversing wbs inbdi_split_work_to_wbs(), and the issue becomes easily reproducible again.To solve this problem,
percpu_ref_exit()is called under RCU protection to avoid race betweencgwb_release_workfn()andbdi_split_work_to_wbs(). Moreover, replacewb_get()withwb_tryget()inbdi_split_work_to_wbs(), and skip the current wb ifwb_tryget()fails because the wb has already been shutdown.
Quoted 3 lines
Quoted 2 lines
Fixes b817525a4a80 ("writeback: bdi_writeback iteration must not skip dying ones")Fixes f3b6a6df38aa ("writeback, cgroup: keep list of inodes attached to bdi_writeback")Cc Roman for this second commit.
Thanks! I forgot to cc to Roman!
Quoted 4 lines
Quoted 2 lines
Having two Fixes: is awkward. These serve as a guide to tell -stable maintainers which kernels need the fix. Can we be more precise?
Yes! I'm very sorry about that. As I replied earlier, this issue was introduced in v4.3-rc7.
Quoted 47 lines
Quoted 38 lines
fs/fs-writeback.c +10
-7@@ -978,6 +978,16 @@ static void bdi_split_work_to_wbs(struct backing_dev_info *bdi,continue ; } +/* +* If wb_tryget fails, the wb has been shutdown, skip it. +* +* Pin @wb so that it stays on @bdi->wb_list. This allows +* continuing iteration from @wb after dropping and +* regrabbing rcu read lock. +*/ +if (!wb_tryget (wb)) +continue ; +/* alloc failed, execute synchronously using on-stack fallback */ work = &fallback_work; *work = *base_work;@@ -986,13 +996,6 @@ static void bdi_split_work_to_wbs(struct backing_dev_info *bdi, work->done = &fallback_work_done;wb_queue_work (wb, work); - -/* -* Pin @wb so that it stays on @bdi->wb_list. This allows -* continuing iteration from @wb after dropping and -* regrabbing rcu read lock. -*/ -wb_get (wb); last_wb = wb;rcu_read_unlock ();mm/backing-dev.c +0
-0mm/backing-dev.c static LIST_HEAD(offline_cgwbs); · line 513
Quoted 7 lines
mm/backing-dev.c +4
-0@@ -507,6 +507,15 @@ static LIST_HEAD(offline_cgwbs);static void cleanup_offline_cgwbs_workfn (struct work_struct *work); static DECLARE_WORK(cleanup_offline_cgwbs_work, cleanup_offline_cgwbs_workfn); +static void cgwb_free_rcu (struct rcu_head *rcu_head) +{ +struct bdi_writeback *wb =container_of (rcu_head, + struct bdi_writeback, rcu);nit:
Quoted 5 lines
struct bdi_writeback *wb; wb = container_of(rcu_head, struct bdi_writeback, rcu);looks nicer, no?
Yes, it looks neater.
Quoted 20 lines
Quoted 20 lines
mm/backing-dev.c +5
-2+percpu_ref_exit (&wb->refcnt); +kfree (wb); +} +static void cgwb_release_workfn (struct work_struct *work) { struct bdi_writeback *wb = container_of(work, struct bdi_writeback,@@ -529,11 +538,10 @@ static void cgwb_release_workfn(struct work_struct *work)list_del (&wb->offline_node);spin_unlock_irq (&cgwb_lock); -percpu_ref_exit (&wb->refcnt);wb_exit (wb);bdi_put (bdi);WARN_ON_ONCE (!list_empty (&wb->b_attached)); -kfree_rcu (wb, rcu); +call_rcu (&wb->rcu, cgwb_free_rcu); }
Quoted 1 line
mm/backing-dev.c static void cgwb_release_workfn(struct work_struct *work) · line 547
Quoted 1 line
mm/backing-dev.c +0
-0static void cgwb_release (struct percpu_ref *refcnt)
Thanks!
Signature · 3 lines
With Best Regards, Baokun Li .