iomap_dio_rw: Allocate AIO completion queue before submitting dio

Executing xfs/104 test in a loop on Linux-v4.13 kernel on a ppc64 machine can cause the following NULL pointer dereference,

Unable to handle kernel paging request for data at address 0x00000100
Faulting instruction address: 0xc0000000000fece8
Oops: Kernel access of bad area, sig: 11 [#1]

SMP NR_CPUS=2048
DEBUG_PAGEALLOC
NUMA
pSeries

Modules linked in:
CPU: 2 PID: 0 Comm: swapper/2 Not tainted 4.13.0 #1

task: c00000063849c200 task.stack: c0000006384f0000
NIP: c0000000000fece8 LR: c0000000000ff2cc CTR: c000000000383380
REGS: c00000063ffdb400 TRAP: 0300 Not tainted (4.13.0)
MSR: 8000000000009032 <SF,EE,ME,IR,DR,RI>
CR: 28002222 XER: 20000000
CFAR: c0000000000227a4 DAR: 0000000000000100 DSISR: 40000000 SOFTE: 0
GPR00: c0000000000ff2cc c00000063ffdb680 c000000001252e00 0000000000000800
GPR04: 0000000000000000 c0000006387b8730 c000000633d5db70 0000000fffffffe1
GPR08: c0000003aa89ed68 0000000000000000 0000000000000000 c000000633d5db70
GPR12: 0000000028002422 c00000000fd80a00 c0000006384f3f90 c00000063ffd8000
GPR16: 0000000000000000 c0000006387b8730 c0000000012475e0 c0000000012475e0
GPR20: 0000000000000002 0000000000000000 0000000000000001 0000000000000002
GPR24: 0000000000000010 c00000063e155800 0000000000000800 0000000000000800
GPR28: 0000000000000000 c000000633d5db00 0000000000000000 0000000000000000 NIP [c0000000000fece8] .__queue_work+0x68/0x600
LR [c0000000000ff2cc] .queue_work_on+0x4c/0x80

Call Trace:

[c00000063ffdb680] [c00000063ffdb720] 0xc00000063ffdb720 (unreliable) [c00000063ffdb770] [c0000000000ff2cc] .queue_work_on+0x4c/0x80 [c00000063ffdb7f0] [c00000000038343c] .iomap_dio_bio_end_io+0xbc/0x1f0 [c00000063ffdb880] [c0000000006cdc18] .bio_endio+0x118/0x1f0 [c00000063ffdb910] [c0000000006d9e30] .blk_update_request+0xd0/0x470 [c00000063ffdb9b0] [c0000000006e8284] .blk_mq_end_request+0x24/0xc0 [c00000063ffdba30] [c00000000082ba50] .lo_complete_rq+0x40/0xe0 [c00000063ffdbab0] [c0000000006e5e48] .__blk_mq_complete_request_remote+0x28/0x40 [c00000063ffdbb20] [c00000000018e674] .flush_smp_call_function_queue+0xc4/0x1e0 [c00000063ffdbbb0] [c00000000003f94c] .smp_ipi_demux_relaxed+0x8c/0x100 [c00000063ffdbc40] [c000000000081054] .icp_hv_ipi_action+0x54/0xa0 [c00000063ffdbcc0] [c0000000001505a4] .__handle_irq_event_percpu+0x84/0x2c0 [c00000063ffdbd90] [c000000000150808] .handle_irq_event_percpu+0x28/0x80 [c00000063ffdbe20] [c000000000156858] .handle_percpu_irq+0x78/0xc0 [c00000063ffdbea0] [c00000000014ee70] .generic_handle_irq+0x40/0x70 [c00000063ffdbf10] [c000000000015a88] .__do_irq+0x88/0x200 [c00000063ffdbf90] [c000000000027a70] .call_do_irq+0x14/0x24 [c0000006384f3810] [c000000000015c84] .do_IRQ+0x84/0x130

This occurs due to the following sequence of events,

1. Allocate dio for Direct I/O write.
2. Invoke iomap_apply() until iov_iter_count() bytes have been submitted.
- Assume that we have submitted atleast one bio. Hence iomap_dio->ref value

     will be >= 2.

- If during the second iteration, iomap_apply() ends up returning -ENOSPC, we would break out of the loop and since the 'ret' value is a negative number we end up not allocating memory for super_block->s_dio_done_wq.
3. Meanwhile, iomap_dio_bio_end_io() is invoked for bios that have been submitted and here the code ends up dereferencing the NULL pointer stored at super_block->s_dio_done_wq.

This commit fixes the bug by allocating memory for super_block->s_dio_done_wq before iomap_apply() is invoked.

---
 fs/iomap.c | 14 +++++++-------
 1 file changed, 7 insertions(+), 7 deletions(-)

index 59cc98a..c91abc1 100644
@@ -993,6 +993,13 @@ iomap_dio_rw(struct kiocb *iocb, struct iov_iter *iter,
 	WARN_ON_ONCE(ret);
 	ret = 0;
 
+	if (iov_iter_rw(iter) == WRITE && !is_sync_kiocb(iocb) &&
+		!inode->i_sb->s_dio_done_wq) {
+		ret = sb_init_dio_done_wq(inode->i_sb);
+		if (ret < 0)
+			goto out_free_dio;
+	}
+
 	inode_dio_begin(inode);
 
 	blk_start_plug(&plug);

@@ -1015,13 +1022,6 @@ iomap_dio_rw(struct kiocb *iocb, struct iov_iter *iter,
 	if (ret < 0)
 		iomap_dio_set_error(dio, ret);
 
-	if (ret >= 0 && iov_iter_rw(iter) == WRITE && !is_sync_kiocb(iocb) &&
-			!inode->i_sb->s_dio_done_wq) {
-		ret = sb_init_dio_done_wq(inode->i_sb);
-		if (ret < 0)
-			iomap_dio_set_error(dio, ret);
-	}
-
 	if (!atomic_dec_and_test(&dio->ref)) {
 		if (!is_sync_kiocb(iocb))
 			return -EIOCBQUEUED;

2.9.5

I tested this patch on x86_64 host. Without the patch, I could crash the host by running xfs/104 within 50 iterations, and host survived 500 iterations of xfs/104 run with this patch applied. So

Thanks!
Eryu

Executing xfs/104 test in a loop on Linux-v4.13 kernel on a ppc64 machine can cause the following NULL pointer dereference,

Unable to handle kernel paging request for data at address 0x00000100
Faulting instruction address: 0xc0000000000fece8
Oops: Kernel access of bad area, sig: 11 [#1]

SMP NR_CPUS=2048
DEBUG_PAGEALLOC
NUMA
pSeries

Modules linked in:
CPU: 2 PID: 0 Comm: swapper/2 Not tainted 4.13.0 #1

task: c00000063849c200 task.stack: c0000006384f0000
NIP: c0000000000fece8 LR: c0000000000ff2cc CTR: c000000000383380
REGS: c00000063ffdb400 TRAP: 0300 Not tainted (4.13.0)
MSR: 8000000000009032 <SF,EE,ME,IR,DR,RI>
CR: 28002222 XER: 20000000
CFAR: c0000000000227a4 DAR: 0000000000000100 DSISR: 40000000 SOFTE: 0
GPR00: c0000000000ff2cc c00000063ffdb680 c000000001252e00 0000000000000800
GPR04: 0000000000000000 c0000006387b8730 c000000633d5db70 0000000fffffffe1
GPR08: c0000003aa89ed68 0000000000000000 0000000000000000 c000000633d5db70
GPR12: 0000000028002422 c00000000fd80a00 c0000006384f3f90 c00000063ffd8000
GPR16: 0000000000000000 c0000006387b8730 c0000000012475e0 c0000000012475e0
GPR20: 0000000000000002 0000000000000000 0000000000000001 0000000000000002
GPR24: 0000000000000010 c00000063e155800 0000000000000800 0000000000000800
GPR28: 0000000000000000 c000000633d5db00 0000000000000000 0000000000000000 NIP [c0000000000fece8] .__queue_work+0x68/0x600
LR [c0000000000ff2cc] .queue_work_on+0x4c/0x80

Call Trace:

[c00000063ffdb680] [c00000063ffdb720] 0xc00000063ffdb720 (unreliable) [c00000063ffdb770] [c0000000000ff2cc] .queue_work_on+0x4c/0x80 [c00000063ffdb7f0] [c00000000038343c] .iomap_dio_bio_end_io+0xbc/0x1f0 [c00000063ffdb880] [c0000000006cdc18] .bio_endio+0x118/0x1f0 [c00000063ffdb910] [c0000000006d9e30] .blk_update_request+0xd0/0x470 [c00000063ffdb9b0] [c0000000006e8284] .blk_mq_end_request+0x24/0xc0 [c00000063ffdba30] [c00000000082ba50] .lo_complete_rq+0x40/0xe0 [c00000063ffdbab0] [c0000000006e5e48] .__blk_mq_complete_request_remote+0x28/0x40 [c00000063ffdbb20] [c00000000018e674] .flush_smp_call_function_queue+0xc4/0x1e0 [c00000063ffdbbb0] [c00000000003f94c] .smp_ipi_demux_relaxed+0x8c/0x100 [c00000063ffdbc40] [c000000000081054] .icp_hv_ipi_action+0x54/0xa0 [c00000063ffdbcc0] [c0000000001505a4] .__handle_irq_event_percpu+0x84/0x2c0 [c00000063ffdbd90] [c000000000150808] .handle_irq_event_percpu+0x28/0x80 [c00000063ffdbe20] [c000000000156858] .handle_percpu_irq+0x78/0xc0 [c00000063ffdbea0] [c00000000014ee70] .generic_handle_irq+0x40/0x70 [c00000063ffdbf10] [c000000000015a88] .__do_irq+0x88/0x200 [c00000063ffdbf90] [c000000000027a70] .call_do_irq+0x14/0x24 [c0000006384f3810] [c000000000015c84] .do_IRQ+0x84/0x130

This occurs due to the following sequence of events,

1. Allocate dio for Direct I/O write.
2. Invoke iomap_apply() until iov_iter_count() bytes have been submitted.
- Assume that we have submitted atleast one bio. Hence iomap_dio->ref value

     will be >= 2.

- If during the second iteration, iomap_apply() ends up returning -ENOSPC, we would break out of the loop and since the 'ret' value is a negative number we end up not allocating memory for super_block->s_dio_done_wq.
3. Meanwhile, iomap_dio_bio_end_io() is invoked for bios that have been submitted and here the code ends up dereferencing the NULL pointer stored at super_block->s_dio_done_wq.

This commit fixes the bug by allocating memory for super_block->s_dio_done_wq before iomap_apply() is invoked.

---

Changelog:
v1->v2:
1. Fix indentation for the continuation of the if conditional.

 fs/iomap.c | 14 +++++++-------
 1 file changed, 7 insertions(+), 7 deletions(-)

index 269b24a..d4f526a 100644
@@ -993,6 +993,13 @@ iomap_dio_rw(struct kiocb *iocb, struct iov_iter *iter,
 	WARN_ON_ONCE(ret);
 	ret = 0;
 
+	if (iov_iter_rw(iter) == WRITE && !is_sync_kiocb(iocb) &&
+	    !inode->i_sb->s_dio_done_wq) {
+		ret = sb_init_dio_done_wq(inode->i_sb);
+		if (ret < 0)
+			goto out_free_dio;
+	}
+
 	inode_dio_begin(inode);
 
 	blk_start_plug(&plug);

@@ -1015,13 +1022,6 @@ iomap_dio_rw(struct kiocb *iocb, struct iov_iter *iter,
 	if (ret < 0)
 		iomap_dio_set_error(dio, ret);
 
-	if (ret >= 0 && iov_iter_rw(iter) == WRITE && !is_sync_kiocb(iocb) &&
-			!inode->i_sb->s_dio_done_wq) {
-		ret = sb_init_dio_done_wq(inode->i_sb);
-		if (ret < 0)
-			iomap_dio_set_error(dio, ret);
-	}
-
 	if (!atomic_dec_and_test(&dio->ref)) {
 		if (!is_sync_kiocb(iocb))
 			return -EIOCBQUEUED;

2.9.5

Can you please trim down the oops stack for the commit message? All that we really need is the stack trace - the arch specific register dumps don't add anything of value to the commit message. i.e:

.queue_work_on
.iomap_dio_bio_end_io
.bio_endio
.blk_update_request
.blk_mq_end_request
.lo_complete_rq
.__blk_mq_complete_request_remote
.flush_smp_call_function_queue
.smp_ipi_demux_relaxed
.icp_hv_ipi_action
.__handle_irq_event_percpu
.handle_irq_event_percpu
.handle_percpu_irq
.generic_handle_irq
.__do_irq
.call_do_irq
.do_IRQ

Cheers,

Dave.

Dave Chinner
[email protected]

Executing xfs/104 test in a loop on Linux-v4.13 kernel on a ppc64 machine can cause the following NULL pointer dereference,

.queue_work_on+0x4c/0x80
.iomap_dio_bio_end_io+0xbc/0x1f0
.bio_endio+0x118/0x1f0
.blk_update_request+0xd0/0x470
.blk_mq_end_request+0x24/0xc0
.lo_complete_rq+0x40/0xe0
.__blk_mq_complete_request_remote+0x28/0x40
.flush_smp_call_function_queue+0xc4/0x1e0
.smp_ipi_demux_relaxed+0x8c/0x100
.icp_hv_ipi_action+0x54/0xa0
.__handle_irq_event_percpu+0x84/0x2c0
.handle_irq_event_percpu+0x28/0x80
.handle_percpu_irq+0x78/0xc0
.generic_handle_irq+0x40/0x70
.__do_irq+0x88/0x200
.call_do_irq+0x14/0x24
.do_IRQ+0x84/0x130

This occurs due to the following sequence of events,

1. Allocate dio for Direct I/O write.
2. Invoke iomap_apply() until iov_iter_count() bytes have been submitted.
- Assume that we have submitted atleast one bio. Hence iomap_dio->ref value

     will be >= 2.

- If during the second iteration, iomap_apply() ends up returning -ENOSPC, we would break out of the loop and since the 'ret' value is a negative number we end up not allocating memory for super_block->s_dio_done_wq.
3. Meanwhile, iomap_dio_bio_end_io() is invoked for bios that have been submitted and here the code ends up dereferencing the NULL pointer stored at super_block->s_dio_done_wq.

This commit fixes the bug by allocating memory for super_block->s_dio_done_wq before iomap_apply() is invoked.

---

Changelog:
v1->v2:
1. Fix indentation for the continuation of the if conditional.
v2->v3:
1. Remove architecture specific register dumps from the stack trace.

fs/iomap.c | 14 +++++++-------

 1 file changed, 7 insertions(+), 7 deletions(-)

index 269b24a..d4f526a 100644
@@ -993,6 +993,13 @@ iomap_dio_rw(struct kiocb *iocb, struct iov_iter *iter,
 	WARN_ON_ONCE(ret);
 	ret = 0;
 
+	if (iov_iter_rw(iter) == WRITE && !is_sync_kiocb(iocb) &&
+	    !inode->i_sb->s_dio_done_wq) {
+		ret = sb_init_dio_done_wq(inode->i_sb);
+		if (ret < 0)
+			goto out_free_dio;
+	}
+
 	inode_dio_begin(inode);
 
 	blk_start_plug(&plug);

@@ -1015,13 +1022,6 @@ iomap_dio_rw(struct kiocb *iocb, struct iov_iter *iter,
 	if (ret < 0)
 		iomap_dio_set_error(dio, ret);
 
-	if (ret >= 0 && iov_iter_rw(iter) == WRITE && !is_sync_kiocb(iocb) &&
-			!inode->i_sb->s_dio_done_wq) {
-		ret = sb_init_dio_done_wq(inode->i_sb);
-		if (ret < 0)
-			iomap_dio_set_error(dio, ret);
-	}
-
 	if (!atomic_dec_and_test(&dio->ref)) {
 		if (!is_sync_kiocb(iocb))
 			return -EIOCBQUEUED;

2.9.5