erofs: fix use-after-free of on-stack io[]
PATCH 1/1 erofs: fix use-after-free of on-stack io[]
From: Hongyu Jin <[email protected]>
The root cause is the race as follows(k5.4):
Thread #1 Thread #2(irq ctx)
z_erofs_submit_and_unzip()
struct z_erofs_vle_unzip_io io_A[]
submit bio A
z_erofs_vle_read_endio() // bio A z_erofs_vle_unzip_kickoff() spin_lock_irqsave() atomic_add_return()
wait_event()
[end of function]z_erofs_submit_and_unzip() // bio B
wake_up_locked(io_A[]) // crash
struct z_erofs_vle_unzip_io io_B[]
submit bio B
wait_event()
Backtrace in kernel5.4:
[ 10.129413] 8<--- cut here --- [ 10.129422] Unable to handle kernel paging request at virtual address eb0454a4 [ 10.364157] CPU: 0 PID: 709 Comm: getprop Tainted: G WC O 5.4.147-ab09225 #1 [ 11.556325] [<c01b33b8>] (__wake_up_common) from [<c01b3300>] (__wake_up_locked+0x40/0x48) [ 11.565487] [<c01b3300>] (__wake_up_locked) from [<c044c8d0>] (z_erofs_vle_unzip_kickoff+0x6c/0xc0) [ 11.575438] [<c044c8d0>] (z_erofs_vle_unzip_kickoff) from [<c044c854>] (z_erofs_vle_read_endio+0x16c/0x17c) [ 11.586082] [<c044c854>] (z_erofs_vle_read_endio) from [<c06a80e8>] (clone_endio+0xb4/0x1d0) [ 11.595428] [<c06a80e8>] (clone_endio) from [<c04a1280>] (blk_update_request+0x150/0x4dc) [ 11.604516] [<c04a1280>] (blk_update_request) from [<c06dea28>] (mmc_blk_cqe_complete_rq+0x144/0x15c) [ 11.614640] [<c06dea28>] (mmc_blk_cqe_complete_rq) from [<c04a5d90>] (blk_done_softirq+0xb0/0xcc) [ 11.624419] [<c04a5d90>] (blk_done_softirq) from [<c010242c>] (__do_softirq+0x184/0x56c) [ 11.633419] [<c010242c>] (__do_softirq) from [<c01051e8>] (irq_exit+0xd4/0x138) [ 11.641640] [<c01051e8>] (irq_exit) from [<c010c314>] (__handle_domain_irq+0x94/0xd0) [ 11.650381] [<c010c314>] (__handle_domain_irq) from [<c04fde70>] (gic_handle_irq+0x50/0xd4) [ 11.659641] [<c04fde70>] (gic_handle_irq) from [<c0101b70>] (__irq_svc+0x70/0xb0)
--- fs/erofs/zdata.c | 12 ++++-------- fs/erofs/zdata.h | 2 +- 2 files changed, 5 insertions(+), 9 deletions(-)
fs/erofs/zdata.c +4 -8
@@ -782,12 +782,9 @@ static void z_erofs_decompress_kickoff(struct z_erofs_decompressqueue *io,/* wake up the caller thread for sync decompression */ if (sync) { -unsigned long flags; - -spin_lock_irqsave (&io->u.wait.lock, flags);if (!atomic_add_return (bios, &io->pending_bios)) -wake_up_locked (&io->u.wait); -spin_unlock_irqrestore (&io->u.wait.lock, flags); +complete (&io->u.done); +return ; }
@@ -1207,7 +1204,7 @@ jobqueue_init(struct super_block *sb, }else { fg_out: q = fgq; -init_waitqueue_head (&fgq->u.wait); +init_completion (&fgq->u.done);atomic_set (&fgq->pending_bios,0 ); } q->sb = sb;
@@ -1370,8 +1367,7 @@ static void z_erofs_runqueue(struct super_block *sb,return ;/* wait until all bios are completed */ -io_wait_event (io[JQ_SUBMIT ].u.wait, - !atomic_read (&io[JQ_SUBMIT ].pending_bios)); +wait_for_completion_io (&io[JQ_SUBMIT ].u.done);/* handle synchronous decompress queue in the caller context */ z_erofs_decompress_queue (&io[JQ_SUBMIT ], pagepool);
fs/erofs/zdata.h +1 -1
@@ -89,7 +89,7 @@ struct z_erofs_decompressqueue {z_erofs_next_pcluster_t head;union { -wait_queue_head_t wait; +struct completion done;struct work_struct work; } u; };
Signature · 1 line
2.25.1
Quoted 15 lines · Hongyu Jin
From: Hongyu Jin <[email protected]>
The root cause is the race as follows(k5.4):
Thread #1 Thread #2(irq ctx)
z_erofs_submit_and_unzip()
structz_erofs_vle_unzip_ioio_A[]
submit bio A
z_erofs_vle_read_endio()// bio Az_erofs_vle_unzip_kickoff()spin_lock_irqsave()atomic_add_return()
wait_event()
[end of function]z_erofs_submit_and_unzip()// bio B
Quoted 4 lines
wake_up_locked(io_A[]) // crashstruct
z_erofs_vle_unzip_ioio_B[]
submit bio B
wait_event()
Thanks, good catch!
Yet could you turn the race above into the current function names?
Quoted 53 lines
Backtrace in kernel5.4:
[ 10.129413] 8<--- cut here --- [ 10.129422] Unable to handle kernel paging request at virtual address eb0454a4 [ 10.364157] CPU: 0 PID: 709 Comm: getprop Tainted: G WC O 5.4.147-ab09225 #1 [ 11.556325] [<c01b33b8>] (__wake_up_common) from [<c01b3300>] (__wake_up_locked+0x40/0x48) [ 11.565487] [<c01b3300>] (__wake_up_locked) from [<c044c8d0>] (z_erofs_vle_unzip_kickoff+0x6c/0xc0) [ 11.575438] [<c044c8d0>] (z_erofs_vle_unzip_kickoff) from [<c044c854>] (z_erofs_vle_read_endio+0x16c/0x17c) [ 11.586082] [<c044c854>] (z_erofs_vle_read_endio) from [<c06a80e8>] (clone_endio+0xb4/0x1d0) [ 11.595428] [<c06a80e8>] (clone_endio) from [<c04a1280>] (blk_update_request+0x150/0x4dc) [ 11.604516] [<c04a1280>] (blk_update_request) from [<c06dea28>] (mmc_blk_cqe_complete_rq+0x144/0x15c) [ 11.614640] [<c06dea28>] (mmc_blk_cqe_complete_rq) from [<c04a5d90>] (blk_done_softirq+0xb0/0xcc) [ 11.624419] [<c04a5d90>] (blk_done_softirq) from [<c010242c>] (__do_softirq+0x184/0x56c) [ 11.633419] [<c010242c>] (__do_softirq) from [<c01051e8>] (irq_exit+0xd4/0x138) [ 11.641640] [<c01051e8>] (irq_exit) from [<c010c314>] (__handle_domain_irq+0x94/0xd0) [ 11.650381] [<c010c314>] (__handle_domain_irq) from [<c04fde70>] (gic_handle_irq+0x50/0xd4) [ 11.659641] [<c04fde70>] (gic_handle_irq) from [<c0101b70>] (__irq_svc+0x70/0xb0)Signed-off-by Hongyu Jin <[email protected]>--- fs/erofs/zdata.c | 12 ++++-------- fs/erofs/zdata.h | 2 +- 2 files changed, 5 insertions(+), 9 deletions(-)fs/erofs/zdata.c +3
-6@@ -782,12 +782,9 @@ static void z_erofs_decompress_kickoff(struct z_erofs_decompressqueue *io,/* wake up the caller thread for sync decompression */ if (sync) { -unsigned long flags; - -spin_lock_irqsave (&io->u.wait.lock, flags);if (!atomic_add_return (bios, &io->pending_bios)) -wake_up_locked (&io->u.wait); -spin_unlock_irqrestore (&io->u.wait.lock, flags); +complete (&io->u.done); +return ; }@@ -1207,7 +1204,7 @@ jobqueue_init(struct super_block *sb, }else { fg_out: q = fgq; -init_waitqueue_head (&fgq->u.wait); +init_completion (&fgq->u.done);atomic_set (&fgq->pending_bios,0 ); } q->sb = sb;
fs/erofs/zdata.c static void z_erofs_runqueue(struct super_block *sb, · line 1370
Quoted 6 lines
fs/erofs/zdata.c +1
-2@@ -1370,8 +1367,7 @@ static void z_erofs_runqueue(struct super_block *sb,return ;/* wait until all bios are completed */ -io_wait_event (io[JQ_SUBMIT ].u.wait, - !atomic_read (&io[JQ_SUBMIT ].pending_bios)); +wait_for_completion_io (&io[JQ_SUBMIT ].u.done);
Thanks, good catch!
What if pending_bios is always 0 (nr_bios == 0), is it possible?
Thanks,
Gao Xiang
Quoted 22 lines · Gao Xiang <[email protected]> 于2022年4月1日周五 14:55写道:
Quoted 15 lines · Hongyu Jin
From: Hongyu Jin <[email protected]>
The root cause is the race as follows(k5.4):
Thread #1 Thread #2(irq ctx)
z_erofs_submit_and_unzip()
structz_erofs_vle_unzip_ioio_A[]
submit bio A
z_erofs_vle_read_endio()// bio Az_erofs_vle_unzip_kickoff()spin_lock_irqsave()atomic_add_return()
wait_event()
[end of function]z_erofs_submit_and_unzip()// bio BQuoted 4 lines
wake_up_locked(io_A[]) // crashstruct
z_erofs_vle_unzip_ioio_B[]
submit bio B
wait_event()
Quoted 2 lines
Thanks, good catch!
Yet could you turn the race above into the current function names?
ok, I will chang it.
Quoted 56 lines
Quoted 56 lines
Backtrace in kernel5.4:
[ 10.129413] 8<--- cut here --- [ 10.129422] Unable to handle kernel paging request at virtual address eb0454a4 [ 10.364157] CPU: 0 PID: 709 Comm: getprop Tainted: G WC O 5.4.147-ab09225 #1 [ 11.556325] [<c01b33b8>] (__wake_up_common) from [<c01b3300>] (__wake_up_locked+0x40/0x48) [ 11.565487] [<c01b3300>] (__wake_up_locked) from [<c044c8d0>] (z_erofs_vle_unzip_kickoff+0x6c/0xc0) [ 11.575438] [<c044c8d0>] (z_erofs_vle_unzip_kickoff) from [<c044c854>] (z_erofs_vle_read_endio+0x16c/0x17c) [ 11.586082] [<c044c854>] (z_erofs_vle_read_endio) from [<c06a80e8>] (clone_endio+0xb4/0x1d0) [ 11.595428] [<c06a80e8>] (clone_endio) from [<c04a1280>] (blk_update_request+0x150/0x4dc) [ 11.604516] [<c04a1280>] (blk_update_request) from [<c06dea28>] (mmc_blk_cqe_complete_rq+0x144/0x15c) [ 11.614640] [<c06dea28>] (mmc_blk_cqe_complete_rq) from [<c04a5d90>] (blk_done_softirq+0xb0/0xcc) [ 11.624419] [<c04a5d90>] (blk_done_softirq) from [<c010242c>] (__do_softirq+0x184/0x56c) [ 11.633419] [<c010242c>] (__do_softirq) from [<c01051e8>] (irq_exit+0xd4/0x138) [ 11.641640] [<c01051e8>] (irq_exit) from [<c010c314>] (__handle_domain_irq+0x94/0xd0) [ 11.650381] [<c010c314>] (__handle_domain_irq) from [<c04fde70>] (gic_handle_irq+0x50/0xd4) [ 11.659641] [<c04fde70>] (gic_handle_irq) from [<c0101b70>] (__irq_svc+0x70/0xb0)Signed-off-by Hongyu Jin <[email protected]>--- fs/erofs/zdata.c | 12 ++++-------- fs/erofs/zdata.h | 2 +- 2 files changed, 5 insertions(+), 9 deletions(-)fs/erofs/zdata.c +3
-6@@ -782,12 +782,9 @@ static void z_erofs_decompress_kickoff(struct z_erofs_decompressqueue *io,/* wake up the caller thread for sync decompression */ if (sync) { -unsigned long flags; - -spin_lock_irqsave (&io->u.wait.lock, flags);if (!atomic_add_return (bios, &io->pending_bios)) -wake_up_locked (&io->u.wait); -spin_unlock_irqrestore (&io->u.wait.lock, flags); +complete (&io->u.done); +return ; }@@ -1207,7 +1204,7 @@ jobqueue_init(struct super_block *sb, }else { fg_out: q = fgq; -init_waitqueue_head (&fgq->u.wait); +init_completion (&fgq->u.done);atomic_set (&fgq->pending_bios,0 ); } q->sb = sb;@@ -1370,8 +1367,7 @@ static void z_erofs_runqueue(struct super_block *sb,return ;
Quoted 8 lines
fs/erofs/zdata.c static void z_erofs_runqueue(struct super_block *sb, · line 1370
Quoted 4 lines
fs/erofs/zdata.c +1
-2/* wait until all bios are completed */ -io_wait_event (io[JQ_SUBMIT ].u.wait, - !atomic_read (&io[JQ_SUBMIT ].pending_bios)); +wait_for_completion_io (&io[JQ_SUBMIT ].u.done);Thanks, good catch!
What if
pending_biosis always 0 (nr_bios== 0), is it possible?
The pending_bios isn't always 0. If bio is completed faster before io_wait_event() called, the value of pending_bios from 1 to 0, when enter io_wait_event(), it will not acquire lock and return immediately.
Quoted 3 lines
Thanks,
Gao Xiang
Quoted 90 lines · Henry King
Quoted 22 lines · Gao Xiang <[email protected]> 于2022年4月1日周五 14:55写道:
Quoted 15 lines · Hongyu Jin
From: Hongyu Jin <[email protected]>
The root cause is the race as follows(k5.4):
Thread #1 Thread #2(irq ctx)
z_erofs_submit_and_unzip()
structz_erofs_vle_unzip_ioio_A[]
submit bio A
z_erofs_vle_read_endio()// bio Az_erofs_vle_unzip_kickoff()spin_lock_irqsave()atomic_add_return()
wait_event()
[end of function]z_erofs_submit_and_unzip()// bio BQuoted 4 lines
wake_up_locked(io_A[]) // crashstruct
z_erofs_vle_unzip_ioio_B[]
submit bio B
wait_event()Quoted 2 lines
Thanks, good catch!
Yet could you turn the race above into the current function names?ok, I will chang it.
Quoted 63 lines
Quoted 54 lines
Backtrace in kernel5.4:
[ 10.129413] 8<--- cut here --- [ 10.129422] Unable to handle kernel paging request at virtual address eb0454a4 [ 10.364157] CPU: 0 PID: 709 Comm: getprop Tainted: G WC O 5.4.147-ab09225 #1 [ 11.556325] [<c01b33b8>] (__wake_up_common) from [<c01b3300>] (__wake_up_locked+0x40/0x48) [ 11.565487] [<c01b3300>] (__wake_up_locked) from [<c044c8d0>] (z_erofs_vle_unzip_kickoff+0x6c/0xc0) [ 11.575438] [<c044c8d0>] (z_erofs_vle_unzip_kickoff) from [<c044c854>] (z_erofs_vle_read_endio+0x16c/0x17c) [ 11.586082] [<c044c854>] (z_erofs_vle_read_endio) from [<c06a80e8>] (clone_endio+0xb4/0x1d0) [ 11.595428] [<c06a80e8>] (clone_endio) from [<c04a1280>] (blk_update_request+0x150/0x4dc) [ 11.604516] [<c04a1280>] (blk_update_request) from [<c06dea28>] (mmc_blk_cqe_complete_rq+0x144/0x15c) [ 11.614640] [<c06dea28>] (mmc_blk_cqe_complete_rq) from [<c04a5d90>] (blk_done_softirq+0xb0/0xcc) [ 11.624419] [<c04a5d90>] (blk_done_softirq) from [<c010242c>] (__do_softirq+0x184/0x56c) [ 11.633419] [<c010242c>] (__do_softirq) from [<c01051e8>] (irq_exit+0xd4/0x138) [ 11.641640] [<c01051e8>] (irq_exit) from [<c010c314>] (__handle_domain_irq+0x94/0xd0) [ 11.650381] [<c010c314>] (__handle_domain_irq) from [<c04fde70>] (gic_handle_irq+0x50/0xd4) [ 11.659641] [<c04fde70>] (gic_handle_irq) from [<c0101b70>] (__irq_svc+0x70/0xb0)Signed-off-by Hongyu Jin <[email protected]>--- fs/erofs/zdata.c | 12 ++++-------- fs/erofs/zdata.h | 2 +- 2 files changed, 5 insertions(+), 9 deletions(-)fs/erofs/zdata.c +3
-6@@ -782,12 +782,9 @@ static void z_erofs_decompress_kickoff(struct z_erofs_decompressqueue *io,/* wake up the caller thread for sync decompression */ if (sync) { -unsigned long flags; - -spin_lock_irqsave (&io->u.wait.lock, flags);if (!atomic_add_return (bios, &io->pending_bios)) -wake_up_locked (&io->u.wait); -spin_unlock_irqrestore (&io->u.wait.lock, flags); +complete (&io->u.done); +return ; }@@ -1207,7 +1204,7 @@ jobqueue_init(struct super_block *sb, }else { fg_out: q = fgq; -init_waitqueue_head (&fgq->u.wait); +init_completion (&fgq->u.done);atomic_set (&fgq->pending_bios,0 ); } q->sb = sb;fs/erofs/zdata.c static void z_erofs_runqueue(struct super_block *sb, · line 1370
Quoted 6 lines
fs/erofs/zdata.c +1
-2@@ -1370,8 +1367,7 @@ static void z_erofs_runqueue(struct super_block *sb,return ;/* wait until all bios are completed */ -io_wait_event (io[JQ_SUBMIT ].u.wait, - !atomic_read (&io[JQ_SUBMIT ].pending_bios)); +wait_for_completion_io (&io[JQ_SUBMIT ].u.done);Thanks, good catch!
Quoted 4 lines
Quoted 1 line
What if
pending_biosis always 0 (nr_bios== 0), is it possible?The
pending_biosisn't always 0. If bio is completed faster beforeio_wait_event()called, the value ofpending_biosfrom 1 to 0, when enterio_wait_event(), it will not acquire lock and return immediately.
nope, IMO, if no io submission, we could run into this case.
Thanks,
Gao Xiang
Quoted 3 lines
Quoted 3 lines
Thanks,
Gao Xiang
Quoted 2 lines · Gao Xiang
Quoted 1 line · Henry King
Gao Xiang <[email protected]> 于2022年4月1日周五 14:55写道:
...
Quoted 10 lines
Quoted 10 lines
Quoted 10 lines
fs/erofs/zdata.c static void z_erofs_runqueue(struct super_block *sb, · line 1370
Quoted 7 lines
fs/erofs/zdata.c +1
-2@@ -1370,8 +1367,7 @@ static void z_erofs_runqueue(struct super_block *sb,return ;/* wait until all bios are completed */ -io_wait_event (io[JQ_SUBMIT ].u.wait, - !atomic_read (&io[JQ_SUBMIT ].pending_bios)); +wait_for_completion_io (&io[JQ_SUBMIT ].u.done);Thanks, good catch!
Quoted 7 lines
Quoted 4 lines
Quoted 1 line
What if
pending_biosis always 0 (nr_bios== 0), is it possible?The
pending_biosisn't always 0. If bio is completed faster beforeio_wait_event()called, the value ofpending_biosfrom 1 to 0, when enterio_wait_event(), it will not acquire lock and return immediately.nope, IMO, if no io submission, we could run into this case.
Ok, after revisiting the code, I think it's impossible as well. Please help commit another version with commit message updated. I will test it later (it's a quite rare race I think.)
Thanks,
Gao Xiang
Quoted 6 lines
Thanks,
Gao XiangQuoted 3 lines
Quoted 3 lines
Thanks,
Gao Xiang