device-mapper fixes for 2.6.29
Alasdair G Kergon
maintainerPlease pull from:
master.kernel.org:/pub/scm/linux/kernel/git/agk/linux-2.6-dm.git
to get the following device-mapper fixes for 2.6.29:
Huang Ying (1):
dm crypt: fix kcryptd_async_done parameter
Mikulas Patocka (2):
dm table: rework reference counting fix
dm io: respect BIO_MAX_PAGES limit
Milan Broz (2):
dm ioctl: validate name length when renaming dm crypt: wait for endio to complete before destruction
drivers/md/dm-crypt.c | 43 +++++++++++++++++++++++++++++++------------ drivers/md/dm-io.c | 2 ++ drivers/md/dm-ioctl.c | 7 ++++--- drivers/md/dm.c | 32 +++++++++++++++++++------------- 4 files changed, 56 insertions(+), 28 deletions(-)
Thanks,
Alasdair
Signature · 1 line
[email protected]
Mikulas Patocka
maintainerQuoted 4 lines · Alasdair G Kergon
Please pull from:
master.kernel.org:/pub/scm/linux/kernel/git/agk/linux-2.6-dm.git
Quoted 8 lines
to get the following device-mapper fixes for 2.6.29:
Huang Ying (1):
dm crypt: fixkcryptd_async_doneparameterMikulas Patocka (2):
dm table: rework reference counting fix
dm io: respectBIO_MAX_PAGESlimit
Hi
Also look at:
dm-io crashes when signal is sent:
http://people.redhat.com/mpatocka/patches/kernel/misc/2.6.29-rc8/dm-dmio-fix-signal-crash.patch
that partial-read dm-raid1 + dm-multipath crash:
http://people.redhat.com/mpatocka/patches/kernel/misc/2.6.29-rc8/dm-record-bio-vector.patch
http://people.redhat.com/mpatocka/patches/kernel/misc/2.6.29-rc8/dm-raid1-read-record-slab.patch
crash or data corruption in snapshots (callbacks are called concurrently and shouldn't be):
http://people.redhat.com/mpatocka/patches/kernel/misc/2.6.29-rc8/dm-kcopyd-fix.patch
failure to check pending exception after we dropped a lock:
http://people.redhat.com/mpatocka/patches/kernel/misc/2.6.29-rc8/dm-snap-break-__find_pending_exception.patch
http://people.redhat.com/mpatocka/patches/kernel/misc/2.6.29-rc8/dm-snap-move-allocation-to-the-caller.patch
http://people.redhat.com/mpatocka/patches/kernel/misc/2.6.29-rc8/dm-snap-check-complete-exceptions.patch
--- all the patches fix crashes or data corruptions.
--- all these patches except the first one were already committed to RHEL 4.8, so you should put them upstream, at least for testing.
Mikulas
Quoted 14 lines
Milan Broz (2):
dm ioctl: validate name length when renaming dm crypt: wait for endio to complete before destructiondrivers/md/dm-crypt.c | 43 +++++++++++++++++++++++++++++++------------ drivers/md/dm-io.c | 2 ++ drivers/md/dm-ioctl.c | 7 ++++--- drivers/md/dm.c | 32 +++++++++++++++++++------------- 4 files changed, 56 insertions(+), 28 deletions(-)Thanks,
AlasdairSignature · 1 line
[email protected]
Alasdair G Kergon
maintainerQuoted 1 line · Mikulas Patocka
http://people.redhat.com/mpatocka/patches/kernel/misc/2.6.29-rc8/dm-kcopyd-fix.patch
Needs submitting to patchwork - split into 4 patches perhaps?
The others are now queued for linux-next and possibly still 2.6.29.
Alasdair
Signature · 1 line
[email protected]
Mikulas Patocka
maintainerQuoted 10 lines · Alasdair G Kergon
Quoted 1 line · Mikulas Patocka
http://people.redhat.com/mpatocka/patches/kernel/misc/2.6.29-rc8/dm-kcopyd-fix.patch
Needs submitting to patchwork - split into 4 patches perhaps?
The others are now queued for linux-next and possibly still 2.6.29.
Alasdair
Signature · 1 line
[email protected]
I think it is already in patchwork, but maybe under some weird name (it didn't have [PATCH] header when I sent it). Anyway, I'm resending it, split to 3 pieces.
Mikulas
Mikulas Patocka
maintainerPATCH 1/3 kcopyd+snapshots race condition
In reply to Alasdair G KergonUse variable to point to kcopyd client. This will be needed for further patches.
--- drivers/md/dm-kcopyd.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)
drivers/md/dm-kcopyd.c +2 -1
@@ -461,6 +461,7 @@ static void segment_complete(int read_ersector_t progress =0 ;sector_t count =0 ;struct kcopyd_job *job = (struct kcopyd_job *) context; +struct dm_kcopyd_client *kc = job->kc;mutex_lock (&job->lock);
@@ -490,7 +491,7 @@ static void segment_complete(int read_erif (count) {int i; -struct kcopyd_job *sub_job =mempool_alloc (job->kc->job_pool, +struct kcopyd_job *sub_job =mempool_alloc (kc->job_pool,GFP_NOIO ); *sub_job = *job;
Mikulas Patocka
maintainerPATCH 2/3 kcopyd+snapshots race condition
Test for job->pages being zero. This will be needed for the next patch.
--- drivers/md/dm-kcopyd.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)
drivers/md/dm-kcopyd.c +2 -1
@@ -297,7 +297,8 @@ static int run_complete_job(struct kcopydm_kcopyd_notify_fn fn = job->fn;struct dm_kcopyd_client *kc = job->kc; -kcopyd_put_pages (kc, job->pages); +if (job->pages) +kcopyd_put_pages (kc, job->pages);mempool_free (job, kc->job_pool);fn (read_err, write_err, context);
Re: [PATCH 2/3] kcopyd+snapshots race condition
Quoted 1 line · Mikulas Patocka
Test for
job->pagesbeing zero. This will be needed for the next patch.
Queued to linux-next (+stable).
Alasdair
Signature · 1 line
[email protected]
Mikulas Patocka
maintainerPATCH 3/3 kcopyd+snapshots race condition
In reply to Mikulas PatockaUnder specific circumstances, kcopyd callback could be called from the thread that called dm_kcopyd_copy not from kcopyd workqueue.
dm_kcopyd_copy -> split_job -> segment_complete -> job->fn()
This code path is taken if thread calling dm_kcopyd_copy is delayed due to scheduling inside split_job/segment_complete and the subjobs complete before the loop in split_job completes.
Snapshots depend on the fact that callbacks are called from the singlethreaded kcopyd workqueue and expect that there is no racing between individual callbacks. The racing between callbacks can lead to corruption of exception store and it can also cause that exception store callbacks are called twice for the same exception --- a likely reason for crashes inside pending_complete() / remove_exception().
When I reviewed kcopyd further, I found four total problems:
1. job->fn being called from the thread that submitted the job (see above).
2. job->fn(read_err, write_err, job->context); in segment_complete reports the error of the last subjob, not the union of all errors.
3. This comment is bogus ---- the jobs are not cancelable. It is leftover from some time when developers thought about adding cancel possibility.
/*
* To avoid a race we must keep the job around
* until after the notify function has completed.
* Otherwise the client may try and stop the job
* after we've completed.
*/job->fn(read_err, write_err, job->context);mempool_free(job, job->kc->job_pool);
4. Master jobs and subjobs are allocated from the same mempool. Completion and freeing of a master job depends on successful allocation of all subjobs -> deadlock.
This patch moves completion of master jobs to run_complete_job (being called from the kcopyd workqueue). The patch fixes problems 1, 2, 3.
Problem 4 will need more changes and another patch.
--- drivers/md/dm-kcopyd.c | 12 ++++-------- 1 file changed, 4 insertions(+), 8 deletions(-)
drivers/md/dm-kcopyd.c +4 -8
@@ -510,14 +510,8 @@ static void segment_complete(int read_er }else if (atomic_dec_and_test (&job->sub_jobs)) { -/* -* To avoid a race we must keep the job around -* until after the notify function has completed. -* Otherwise the client may try and stop the job -* after we've completed. -*/ - job->fn (read_err, write_err, job->context); -mempool_free (job, job->kc->job_pool); +push (&kc->complete_jobs, job); +wake (kc); } }
@@ -530,6 +524,8 @@ static void split_job(struct kcopyd_job {int i; +atomic_inc (&job->kc->nr_jobs); +atomic_set (&job->sub_jobs,SPLIT_COUNT );for (i =0 ; i <SPLIT_COUNT ; i++)segment_complete (0 ,0u , job);
Mikulas Patocka
maintainerPATCH 3/3 kcopyd+snapshots race condition ... updated
Under specific circumstances, kcopyd callback could be called from the thread that called dm_kcopyd_copy not from kcopyd workqueue.
dm_kcopyd_copy -> split_job -> segment_complete -> job->fn()
This code path is taken if thread calling dm_kcopyd_copy is delayed due to scheduling inside split_job/segment_complete and the subjobs complete before the loop in split_job completes.
Snapshots depend on the fact that callbacks are called from the singlethreaded kcopyd workqueue and expect that there is no racing between individual callbacks. The racing between callbacks can lead to corruption of exception store and it can also cause that exception store callbacks are called twice for the same exception --- a likely reason for crashes inside pending_complete() / remove_exception().
This patch fixes two problems:
1. job->fn being called from the thread that submitted the job (see above).
- Fix: hand over the completion callback to the kcopyd thread.
2. job->fn(read_err, write_err, job->context); in segment_complete reports the error of the last subjob, not the union of all errors.
- Fix: pass job->write_err to the callback to report all error bits (it is done already in run_complete_job)
--- drivers/md/dm-kcopyd.c | 17 +++++++++++------ 1 file changed, 11 insertions(+), 6 deletions(-)
drivers/md/dm-kcopyd.c +11 -6
@@ -511,13 +511,16 @@ static void segment_complete(int read_er }else if (atomic_dec_and_test (&job->sub_jobs)) {/* -* To avoid a race we must keep the job around -* until after the notify function has completed. -* Otherwise the client may try and stop the job -* after we've completed. +* Queue the completion callback to the kcopyd thread. +* +* Some callers assume that all the completions are called +* from a single thread and don't race with each other. +* +* We must not call the callback directly here because this +* code may not be executing in the thread. */ - job->fn (read_err, write_err, job->context); -mempool_free (job, job->kc->job_pool); +push (&kc->complete_jobs, job); +wake (kc); } }
@@ -530,6 +533,8 @@ static void split_job(struct kcopyd_job {int i; +atomic_inc (&job->kc->nr_jobs); +atomic_set (&job->sub_jobs,SPLIT_COUNT );for (i =0 ; i <SPLIT_COUNT ; i++)segment_complete (0 ,0u , job);
Re: [PATCH 3/3] kcopyd+snapshots race condition ... updated
Alasdair G Kergon
maintainerRe: [PATCH 1/3] kcopyd+snapshots race condition
In reply to Mikulas Patocka