[Devel] [PATCH] dm-qcow2: use kvmalloc for bvec allocation during rq splitting

[Andrey Zhadchenko]

On 6/25/25 13:05, Pavel Tikhomirov wrote:
> Looks good.
> 
> Reviewed-by: Pavel Tikhomirov <ptikhomirov at virtuozzo.com>
> 
> Stupid question: we have another bvec allocation in qcow2_rw_pages_sync, 
> should we also switch it to kvmalloc?

Should be fine there. Most uses are either single page or in case of 
header parsing is rather small

> 
> On 6/25/25 16:50, Andrey Zhadchenko wrote:
>> as we saw up to 32768 bios in a single request. kmalloc with
>> GFP_NOIO simply fails on a loaded system.
>>
>> [ 1351.057243] kworker/u9:8: page allocation failure: order:6, 
>> mode:0x40c00(GFP_NOIO|__GFP_COMP), 
>> nodemask=(null),cpuset=/,mems_allowed=0
>> [ 1351.058056] CPU: 2 PID: 12720 Comm: kworker/u9:8 ve: / Kdump: 
>> loaded Tainted: G               X  -------  ---  
>> 5.14.0-427.44.1.vz9.80.39 #1 80.39
>> [ 1351.058543] Hardware name: Acronis OpenStack Compute/Virtuozzo, 
>> BIOS 1.16.1-1.vz9.2 04/01/2014
>> [ 1351.058936] Workqueue: dm-qcow2 do_qcow2_work [dm_qcow2]
>> [ 1351.059214] Call Trace:
>> [ 1351.059390]  <TASK>
>> [ 1351.059560]  dump_stack_lvl+0x34/0x48
>> [ 1351.059762]  warn_alloc+0x138/0x160
>> [ 1351.059982]  ? __alloc_pages_direct_compact+0xa7/0x280
>> [ 1351.060259]  __alloc_pages_slowpath.constprop.0+0x76f/0x7e0
>> [ 1351.060543]  __alloc_pages+0x3bb/0x3f0
>> [ 1351.060759]  ? create_bvec_from_rq+0x8f/0x130 [dm_qcow2]
>> [ 1351.061011]  __kmalloc_large_node+0x79/0x100
>> [ 1351.061231]  __kmalloc+0xca/0x140
>> [ 1351.061430]  create_bvec_from_rq+0x8f/0x130 [dm_qcow2]
>> [ 1351.061679]  do_qcow2_work+0x3cd/0xbc0 [dm_qcow2]
>> [ 1351.061925]  ? trigger_cpulimit_balance+0x6b/0x180
>> [ 1351.062167]  ? _raw_spin_unlock+0xa/0x30
>> [ 1351.062391]  ? finish_task_switch.isra.0+0x8c/0x2a0
>> [ 1351.062625]  process_one_work+0x1e5/0x3b0
>> [ 1351.062838]  worker_thread+0x50/0x3a0
>> [ 1351.063044]  ? __pfx_worker_thread+0x10/0x10
>> [ 1351.063321]  kthread+0xe0/0x100
>> [ 1351.063516]  ? __pfx_kthread+0x10/0x10
>> [ 1351.063713]  ret_from_fork+0x2c/0x50
>> [ 1351.063935]  </TASK>
>>
>> Also update ploop to use kvmalloc for the same purpose.
>>
>> https://virtuozzo.atlassian.net/browse/VSTOR-109363
>> Signed-off-by: Andrey Zhadchenko <andrey.zhadchenko at virtuozzo.com>
>> ---
>>   drivers/md/dm-ploop-map.c | 6 +++---
>>   drivers/md/dm-qcow2-map.c | 6 +++---
>>   2 files changed, 6 insertions(+), 6 deletions(-)
>>
>> diff --git a/drivers/md/dm-ploop-map.c b/drivers/md/dm-ploop-map.c
>> index 35085a04ca5f..89f8d527150d 100644
>> --- a/drivers/md/dm-ploop-map.c
>> +++ b/drivers/md/dm-ploop-map.c
>> @@ -194,7 +194,7 @@ static void ploop_prq_endio(struct pio *pio, void 
>> *prq_ptr,
>>       struct request *rq = prq->rq;
>>       if (prq->bvec)
>> -        kfree(prq->bvec);
>> +        kvfree(prq->bvec);
>>       if (prq->css)
>>           css_put(prq->css);
>>       /*
>> @@ -1963,8 +1963,8 @@ static struct bio_vec 
>> *ploop_create_bvec_from_rq(struct request *rq)
>>       rq_for_each_bvec(bv, rq, rq_iter)
>>           nr_bvec++;
>> -    bvec = kmalloc_array(nr_bvec, sizeof(struct bio_vec),
>> -                 GFP_ATOMIC);
>> +    bvec = kvmalloc_array(nr_bvec, sizeof(struct bio_vec),
>> +                  GFP_ATOMIC);
>>       if (!bvec)
>>           goto out;
>> diff --git a/drivers/md/dm-qcow2-map.c b/drivers/md/dm-qcow2-map.c
>> index f7cb036bb416..80552e2ab734 100644
>> --- a/drivers/md/dm-qcow2-map.c
>> +++ b/drivers/md/dm-qcow2-map.c
>> @@ -3501,8 +3501,8 @@ static struct bio_vec 
>> *create_bvec_from_rq(struct request *rq)
>>       rq_for_each_bvec(bv, rq, rq_iter)
>>           nr_bvec++;
>> -    bvec = kmalloc_array(nr_bvec, sizeof(struct bio_vec),
>> -                 GFP_NOIO);
>> +    bvec = kvmalloc_array(nr_bvec, sizeof(struct bio_vec),
>> +                  GFP_NOIO);
>>       if (!bvec)
>>           goto out;
>> @@ -4188,7 +4188,7 @@ static void qrq_endio(struct qcow2_target *tgt, 
>> struct qio *qio,
>>       struct request *rq = qrq->rq;
>>       if (qrq->bvec)
>> -        kfree(qrq->bvec);
>> +        kvfree(qrq->bvec);
>>       /*
>>        * Here is exit point for rq, and here we handle ENOSPC.
>>        * Embedded qios will be reinitialized like they've just
>