Skip to contentSkip to conversation

block: fix blk_aio_*() segfault when blk->root == NULL

PATCH Superseded Stefan Hajnoczi 4 participants 13 messages Feb 8, 2018 – Feb 12, 2018 Block layer

v1 of 3 Superseded Latest is v3 →

lore
Share & tools
lore
#0 lore raw

cover block: fix blk_aio_*() segfault when blk->root == NULL

Using bdrv_inc_in_flight(blk_bs(blk)) doesn't work since BlockBackend->root may be NULL.

This patch series solves the issue by adding an BlockBackend->in_flight counter so requests can be tracked even when there is no BlockDriverState.

This should fix the IDE and virtio-blk segfaults that have been encountered when there is no BlockDriverState.

The patch is based on work by Kevin Wolf.

Kevin Wolf (1):
block: test blk_aio_flush() with blk->root == NULL

Stefan Hajnoczi (2):
block: add BlockBackend->in_flight counter
Revert "IDE: Do not flush empty CDROM drives"

 tests/Makefile.include     |  2 ++
 block.c                    |  2 +-
 block/block-backend.c      | 59 +++++++++++++++++++++++++++++----
 hw/ide/core.c              | 10 +-----
 tests/test-block-backend.c | 82 ++++++++++++++++++++++++++++++++++++++++++++++
 5 files changed, 138 insertions(+), 17 deletions(-)
 create mode 100644 tests/test-block-backend.c
Signature · 1 line
2.14.3
#1 lore raw

PATCH 1/3 block: add BlockBackend->in_flight counter

BlockBackend currently relies on BlockDriverState->in_flight to track requests for blk_drain(). There is a corner case where BlockDriverState->in_flight cannot be used though: blk->root can be NULL when there is no medium. This results in a segfault when the NULL pointer is dereferenced.

Introduce a BlockBackend->in_flight counter for aio requests so it works even when blk->root == NULL.

Based on a patch by Kevin Wolf <[email protected]>.

---
 block.c               |  2 +-
 block/block-backend.c | 59 +++++++++++++++++++++++++++++++++++++++++++++------
 2 files changed, 53 insertions(+), 8 deletions(-)
block.c +1 -1
index a8da4f2b25..140f7919f8 100644
@@ -4716,7 +4716,7 @@ out:
 
 AioContext *bdrv_get_aio_context(BlockDriverState *bs)
 {
-    return bs->aio_context;
+    return bs ? bs->aio_context : qemu_get_aio_context();
 }
 
 void bdrv_coroutine_enter(BlockDriverState *bs, Coroutine *co)
block/block-backend.c +52 -7

104 lines not on this page · raw message

Signature · 1 line
2.14.3
Reviewed-by on v1 1/3
#2 lore raw

Re: [Qemu-devel] [PATCH 1/3] block: add BlockBackend->in_flight counter

Quoted 9 lines · Stefan Hajnoczi

BlockBackend currently relies on BlockDriverState->in_flight to track requests for blk_drain(). There is a corner case where BlockDriverState->in_flight cannot be used though: blk->root can be NULL when there is no medium. This results in a segfault when the NULL pointer is dereferenced.

Introduce a BlockBackend->in_flight counter for aio requests so it works even when blk->root == NULL.

Quoted 9 lines

Based on a patch by Kevin Wolf <[email protected]>.

---
  block.c               |  2 +-
  block/block-backend.c | 59 +++++++++++++++++++++++++++++++++++++++++++++------
  2 files changed, 53 insertions(+), 8 deletions(-)
Signature · 3 lines
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3266
Virtualization:  qemu.org | libvirt.org
#3 lore raw

Re: [Qemu-devel] [PATCH 1/3] block: add BlockBackend->in_flight counter

In reply to Stefan Hajnoczi

block/block-backend.c int blk_flush(BlockBackend *blk) · line 1551

Quoted 13 lines · Stefan Hajnoczi
block/block-backend.c +13 -0
+    BlockDriverState *bs = blk_bs(blk);
+
+    if (bs) {
+        bdrv_drained_begin(bs);
+    }
+
+    /* We may have aio requests like -ENOMEDIUM in flight */
+    while (atomic_mb_read(&blk->in_flight) > 0) {
+        aio_poll(blk_get_aio_context(blk), true);
+    }
+
+    if (bs) {
+        bdrv_drained_end(bs);

This only works if you are in the default AioContext, otherwise you can run handlers for one AioContexts in two threads.

bdrv_dec_in_flight uses bdrv_wakeup and BDRV_POLL_WHILE to ensure that this doesn't happen, so there would be more code that you have to copy into block-backend.c.

Paolo

#4 lore raw

Re: [Qemu-devel] [PATCH 1/3] block: add BlockBackend->in_flight counter

Am 09.02.2018 um 17:28 hat Paolo Bonzini geschrieben:

Quoted 9 lines · Stefan Hajnoczi
Quoted 8 lines
block/block-backend.c +8 -0
+    BlockDriverState *bs = blk_bs(blk);
+
+    if (bs) {
+        bdrv_drained_begin(bs);
+    }
+
+    /* We may have aio requests like -ENOMEDIUM in flight */
+    while (atomic_mb_read(&blk->in_flight) > 0) {
Quoted 8 lines

block/block-backend.c int blk_flush(BlockBackend *blk) · line 1551

Quoted 5 lines
block/block-backend.c +5 -0
+        aio_poll(blk_get_aio_context(blk), true);
+    }
+
+    if (bs) {
+        bdrv_drained_end(bs);

This only works if you are in the default AioContext, otherwise you can run handlers for one AioContexts in two threads.

Should aio_poll() assert that it's called in the right thread to make sure we obey the rules?

Quoted 3 lines

bdrv_dec_in_flight uses bdrv_wakeup and BDRV_POLL_WHILE to ensure that this doesn't happen, so there would be more code that you have to copy into block-backend.c.

Instead of copying, can't we generalise it into a POLL_WHILE(ctx, wakeup, cond) and make BDRV_POLL_WHILE() a wrapper for that?

Kevin

#5 lore raw

Re: [Qemu-devel] [PATCH 1/3] block: add BlockBackend->in_flight counter

Quoted 22 lines · Kevin Wolf

Am 09.02.2018 um 17:28 hat Paolo Bonzini geschrieben:

Quoted 9 lines · Stefan Hajnoczi
Quoted 8 lines
block/block-backend.c +8 -0
+    BlockDriverState *bs = blk_bs(blk);
+
+    if (bs) {
+        bdrv_drained_begin(bs);
+    }
+
+    /* We may have aio requests like -ENOMEDIUM in flight */
+    while (atomic_mb_read(&blk->in_flight) > 0) {
Quoted 8 lines

block/block-backend.c int blk_flush(BlockBackend *blk) · line 1551

Quoted 5 lines
block/block-backend.c +5 -0
+        aio_poll(blk_get_aio_context(blk), true);
+    }
+
+    if (bs) {
+        bdrv_drained_end(bs);

This only works if you are in the default AioContext, otherwise you can run handlers for one AioContexts in two threads.

Should aio_poll() assert that it's called in the right thread to make sure we obey the rules?

Quoted 6 lines
Quoted 3 lines

bdrv_dec_in_flight uses bdrv_wakeup and BDRV_POLL_WHILE to ensure that this doesn't happen, so there would be more code that you have to copy into block-backend.c.

Instead of copying, can't we generalise it into a POLL_WHILE(ctx, wakeup, cond) and make BDRV_POLL_WHILE() a wrapper for that?

Yes, or even move bdrv_wakeup to AioContext would do. We already have block layer-specific fields such as the linux-aio state(*).

(*) though now that linux-aio has been improved to do something
like epoll, there may be a better reason to place linux-aio state in AioContext.

Paolo

#6 lore raw

Re: [Qemu-devel] [PATCH 1/3] block: add BlockBackend->in_flight counter

Quoted 30 lines · Paolo Bonzini
Quoted 22 lines · Kevin Wolf

Am 09.02.2018 um 17:28 hat Paolo Bonzini geschrieben:

Quoted 9 lines · Stefan Hajnoczi
Quoted 8 lines
block/block-backend.c +8 -0
+    BlockDriverState *bs = blk_bs(blk);
+
+    if (bs) {
+        bdrv_drained_begin(bs);
+    }
+
+    /* We may have aio requests like -ENOMEDIUM in flight */
+    while (atomic_mb_read(&blk->in_flight) > 0) {
Quoted 8 lines

block/block-backend.c int blk_flush(BlockBackend *blk) · line 1551

Quoted 5 lines
block/block-backend.c +5 -0
+        aio_poll(blk_get_aio_context(blk), true);
+    }
+
+    if (bs) {
+        bdrv_drained_end(bs);

This only works if you are in the default AioContext, otherwise you can run handlers for one AioContexts in two threads.

Should aio_poll() assert that it's called in the right thread to make sure we obey the rules?

Quoted 6 lines
Quoted 3 lines

bdrv_dec_in_flight uses bdrv_wakeup and BDRV_POLL_WHILE to ensure that this doesn't happen, so there would be more code that you have to copy into block-backend.c.

Instead of copying, can't we generalise it into a POLL_WHILE(ctx, wakeup, cond) and make BDRV_POLL_WHILE() a wrapper for that?

Quoted 6 lines

Yes, or even move bdrv_wakeup to AioContext would do. We already have block layer-specific fields such as the linux-aio state(*).

(*) though now that linux-aio has been improved to do something
like epoll, there may be a better reason to place linux-aio state in AioContext.

Thanks for the ideas, will fix in v2.

Stefan

#7 lore raw

PATCH 2/3 block: test blk_aio_flush() with blk->root == NULL

In reply to Stefan Hajnoczi

From: Kevin Wolf <[email protected]>

This patch adds test cases for the scenario where blk_aio_flush() is called on a BlockBackend with no root. Calling drain afterwards should complete the requests with -ENOMEDIUM.

---
 tests/Makefile.include     |  2 ++
 tests/test-block-backend.c | 82 ++++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 84 insertions(+)
 create mode 100644 tests/test-block-backend.c
tests/Makefile.include +2 -0
index f41da235ae..1a457c29bc 100644
@@ -83,6 +83,7 @@ gcov-files-test-hbitmap-y = blockjob.c
 check-unit-y += tests/test-bdrv-drain$(EXESUF)
 check-unit-y += tests/test-blockjob$(EXESUF)
 check-unit-y += tests/test-blockjob-txn$(EXESUF)
+check-unit-y += tests/test-block-backend$(EXESUF)
 check-unit-y += tests/test-x86-cpuid$(EXESUF)
 # all code tested by test-x86-cpuid is inside topology.h
 gcov-files-test-x86-cpuid-y =
@@ -609,6 +610,7 @@ tests/test-throttle$(EXESUF): tests/test-throttle.o $(test-block-obj-y)
 tests/test-bdrv-drain$(EXESUF): tests/test-bdrv-drain.o $(test-block-obj-y) $(test-util-obj-y)
 tests/test-blockjob$(EXESUF): tests/test-blockjob.o $(test-block-obj-y) $(test-util-obj-y)
 tests/test-blockjob-txn$(EXESUF): tests/test-blockjob-txn.o $(test-block-obj-y) $(test-util-obj-y)
+tests/test-block-backend$(EXESUF): tests/test-block-backend.o $(test-block-obj-y) $(test-util-obj-y)
 tests/test-thread-pool$(EXESUF): tests/test-thread-pool.o $(test-block-obj-y)
 tests/test-iov$(EXESUF): tests/test-iov.o $(test-util-obj-y)
 tests/test-hbitmap$(EXESUF): tests/test-hbitmap.o $(test-util-obj-y) $(test-crypto-obj-y)
tests/test-block-backend.c new file +82 -0

82 lines not on this page · raw message

Signature · 1 line
2.14.3
Reviewed-by on v1 2/3
#8 lore raw

Re: [Qemu-devel] [PATCH 2/3] block: test blk_aio_flush() with blk->root == NULL

Quoted 17 lines · Stefan Hajnoczi

From: Kevin Wolf <[email protected]>

This patch adds test cases for the scenario where blk_aio_flush() is called on a BlockBackend with no root. Calling drain afterwards should complete the requests with -ENOMEDIUM.

---
  tests/Makefile.include     |  2 ++
  tests/test-block-backend.c | 82 ++++++++++++++++++++++++++++++++++++++++++++++
  2 files changed, 84 insertions(+)

create mode 100644 tests/test-block-backend.c

+++ b/tests/test-block-backend.c

tests/test-block-backend.c line 4

Quoted 4 lines
tests/test-block-backend.c new file +4 -0
@@ -0,0 +1,82 @@
+/*
+ * BlockBackend tests
+ *
+ * Copyright (c) 2017 Kevin Wolf <[email protected]>

Want to include 2018? Using a redhat.com address but a personal copyright? (Then again, Red Hat has some pretty cool policies on what it permits for code written on an employee's own time).

As long as Kevin is happy with the final copyright line that goes into the commit:

Signature · 3 lines
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3266
Virtualization:  qemu.org | libvirt.org
#9 lore raw

Re: [Qemu-devel] [PATCH 2/3] block: test blk_aio_flush() with blk->root == NULL

Quoted 27 lines · Eric Blake
Quoted 17 lines · Stefan Hajnoczi

From: Kevin Wolf <[email protected]>

This patch adds test cases for the scenario where blk_aio_flush() is called on a BlockBackend with no root. Calling drain afterwards should complete the requests with -ENOMEDIUM.

---
  tests/Makefile.include     |  2 ++
  tests/test-block-backend.c | 82 ++++++++++++++++++++++++++++++++++++++++++++++
  2 files changed, 84 insertions(+)

create mode 100644 tests/test-block-backend.c

+++ b/tests/test-block-backend.c

tests/test-block-backend.c line 4

Quoted 4 lines
tests/test-block-backend.c new file +4 -0
@@ -0,0 +1,82 @@
+/*
+ * BlockBackend tests
+ *
+ * Copyright (c) 2017 Kevin Wolf <[email protected]>

Want to include 2018? Using a redhat.com address but a personal copyright? (Then again, Red Hat has some pretty cool policies on what it permits for code written on an employee's own time).

Quoted 4 lines

As long as Kevin is happy with the final copyright line that goes into the commit:

This is how I got the test case from Kevin, I didn't change the copyright line.

Stefan

#10 lore raw

PATCH 3/3 Revert "IDE: Do not flush empty CDROM drives"

In reply to Stefan Hajnoczi

This reverts commit 4da97120d51a4383aa96d741a2b837f8c4bbcd0b.

blk_aio_flush() now handles the blk->root == NULL case, so we no longer need this workaround.

---
 hw/ide/core.c | 10 +---------
 1 file changed, 1 insertion(+), 9 deletions(-)
hw/ide/core.c +1 -9
index 5be72d41dc..6154bc6ad7 100644
@@ -1085,15 +1085,7 @@ static void ide_flush_cache(IDEState *s)
     s->status |= BUSY_STAT;
     ide_set_retry(s);
     block_acct_start(blk_get_stats(s->blk), &s->acct, 0, BLOCK_ACCT_FLUSH);
-
-    if (blk_bs(s->blk)) {
-        s->pio_aiocb = blk_aio_flush(s->blk, ide_flush_cb, s);
-    } else {
-        /* XXX blk_aio_flush() crashes when blk_bs(blk) is NULL, remove this
-         * temporary workaround when blk_aio_*() functions handle NULL blk_bs.
-         */
-        ide_flush_cb(s, 0);
-    }
+    s->pio_aiocb = blk_aio_flush(s->blk, ide_flush_cb, s);
 }
 
 static void ide_cfata_metadata_inquiry(IDEState *s)
Signature · 1 line
2.14.3
Reviewed-by on v1 3/3
#11 lore raw

Re: [Qemu-devel] [PATCH 3/3] Revert "IDE: Do not flush empty CDROM drives"

Quoted 11 lines · Stefan Hajnoczi

This reverts commit 4da97120d51a4383aa96d741a2b837f8c4bbcd0b.

blk_aio_flush() now handles the blk->root == NULL case, so we no longer need this workaround.

---
  hw/ide/core.c | 10 +---------
  1 file changed, 1 insertion(+), 9 deletions(-)
Signature · 3 lines
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3266
Virtualization:  qemu.org | libvirt.org
Tested-by on the v1 series
#12 lore raw
In reply to Stefan Hajnoczi
Quoted 9 lines · Stefan Hajnoczi

Using bdrv_inc_in_flight(blk_bs(blk)) doesn't work since BlockBackend->root may be NULL.

This patch series solves the issue by adding an BlockBackend->in_flight counter so requests can be tracked even when there is no BlockDriverState.

This should fix the IDE and virtio-blk segfaults that have been encountered when there is no BlockDriverState.

Quoted 8 lines

The patch is based on work by Kevin Wolf.

Kevin Wolf (1):

   block: test blk_aio_flush() with blk->root == NULL

Stefan Hajnoczi (2):
block: add BlockBackend->in_flight counter
Revert "IDE: Do not flush empty CDROM drives"

Tested by applying the series out of order (2, 3, 1) - the new test fails after 2 (so even our workaround was not robust), fails after 3 (expected there, because we revert the workaround), then finally passes after 1 (the correct fix with no workaround needed). So for the series, you can add:

Signature · 3 lines
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3266
Virtualization:  qemu.org | libvirt.org