[PATCH] mm/vmalloc: Fix return value check for vb_alloc

[PATCH] mm/vmalloc: Fix return value check for vb_alloc

[刘海龙(LaoLiu)]

If vm_map_ram(page, 0, 0) would cause panic by vmap_pages_range_noflush, so
change IS_ERR to IS_ERR_OR_NULL to fix this.

Signed-off-by: Hailong.Liu <liuhailong@oppo.com>
---
 mm/vmalloc.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/mm/vmalloc.c b/mm/vmalloc.c
index d12a17fc0c17..109732006cf7 100644
--- a/mm/vmalloc.c
+++ b/mm/vmalloc.c
@@ -2387,7 +2387,7 @@ void *vm_map_ram(struct page **pages, unsigned int
count, int node)

        if (likely(count <= VMAP_MAX_ALLOC)) {
                mem = vb_alloc(size, GFP_KERNEL);
-               if (IS_ERR(mem))
+               if (IS_ERR_OR_NULL(mem))
                        return NULL;
                addr = (unsigned long)mem;
        } else {
--
2.34.1

Re: [PATCH] mm/vmalloc: Fix return value check for vb_alloc

[Barry Song]

On Wed, Feb 28, 2024 at 9:51 PM 刘海龙(LaoLiu) <liuhailong@oppo.com> wrote:
>
> If vm_map_ram(page, 0, 0) would cause panic by vmap_pages_range_noflush, so
> change IS_ERR to IS_ERR_OR_NULL to fix this.
>
> Signed-off-by: Hailong.Liu <liuhailong@oppo.com>
> ---
>  mm/vmalloc.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/mm/vmalloc.c b/mm/vmalloc.c
> index d12a17fc0c17..109732006cf7 100644
> --- a/mm/vmalloc.c
> +++ b/mm/vmalloc.c
> @@ -2387,7 +2387,7 @@ void *vm_map_ram(struct page **pages, unsigned int
> count, int node)
>
>         if (likely(count <= VMAP_MAX_ALLOC)) {
>                 mem = vb_alloc(size, GFP_KERNEL);
> -               if (IS_ERR(mem))
> +               if (IS_ERR_OR_NULL(mem))

it seems the only case for vb_alloc to return NULL is size = 0, isn't
it a bug of
caller?

>                         return NULL;
>                 addr = (unsigned long)mem;
>         } else {
> --
> 2.34.1

Thanks
Barry

Re: [PATCH] mm/vmalloc: Fix return value check for vb_alloc

[Barry Song]

On Wed, Feb 28, 2024 at 10:34 PM Barry Song <21cnbao@gmail.com> wrote:
>
> On Wed, Feb 28, 2024 at 9:51 PM 刘海龙(LaoLiu) <liuhailong@oppo.com> wrote:
> >
> > If vm_map_ram(page, 0, 0) would cause panic by vmap_pages_range_noflush, so
> > change IS_ERR to IS_ERR_OR_NULL to fix this.
> >
> > Signed-off-by: Hailong.Liu <liuhailong@oppo.com>
> > ---
> >  mm/vmalloc.c | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > diff --git a/mm/vmalloc.c b/mm/vmalloc.c
> > index d12a17fc0c17..109732006cf7 100644
> > --- a/mm/vmalloc.c
> > +++ b/mm/vmalloc.c
> > @@ -2387,7 +2387,7 @@ void *vm_map_ram(struct page **pages, unsigned int
> > count, int node)
> >
> >         if (likely(count <= VMAP_MAX_ALLOC)) {
> >                 mem = vb_alloc(size, GFP_KERNEL);
> > -               if (IS_ERR(mem))
> > +               if (IS_ERR_OR_NULL(mem))
>
> it seems the only case for vb_alloc to return NULL is size = 0, isn't
> it a bug of
> caller?

what about the below?

diff --git a/mm/vmalloc.c b/mm/vmalloc.c
index 25a8df497255..640157221c95 100644
--- a/mm/vmalloc.c
+++ b/mm/vmalloc.c
@@ -2834,6 +2834,9 @@ void *vm_map_ram(struct page **pages, unsigned
int count, int node)
        unsigned long addr;
        void *mem;

+       if (unlikely(count == 0))
+               return NULL;
+
        if (likely(count <= VMAP_MAX_ALLOC)) {
                mem = vb_alloc(size, GFP_KERNEL);
                if (IS_ERR(mem))


>
> >                         return NULL;
> >                 addr = (unsigned long)mem;
> >         } else {
> > --
> > 2.34.1
Thanks
Barry

Re: [PATCH] mm/vmalloc: Fix return value check for vb_alloc

[刘海龙(LaoLiu)]

On 2024/2/28 17:34, Barry Song wrote:
> On Wed, Feb 28, 2024 at 9:51 PM 刘海龙(LaoLiu) <liuhailong@oppo.com> wrote:
>>
>> If vm_map_ram(page, 0, 0) would cause panic by vmap_pages_range_noflush, so
>> change IS_ERR to IS_ERR_OR_NULL to fix this.
>>
>> Signed-off-by: Hailong.Liu <liuhailong@oppo.com>
>> ---
>>  mm/vmalloc.c | 2 +-
>>  1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/mm/vmalloc.c b/mm/vmalloc.c
>> index d12a17fc0c17..109732006cf7 100644
>> --- a/mm/vmalloc.c
>> +++ b/mm/vmalloc.c
>> @@ -2387,7 +2387,7 @@ void *vm_map_ram(struct page **pages, unsigned int
>> count, int node)
>>
>>         if (likely(count <= VMAP_MAX_ALLOC)) {
>>                 mem = vb_alloc(size, GFP_KERNEL);
>> -               if (IS_ERR(mem))
>> +               if (IS_ERR_OR_NULL(mem))
> 
> it seems the only case for vb_alloc to return NULL is size = 0, isn't
> it a bug of
> caller?
vb_alloc had already checked the size == 0, so it should be return NULL
to caller or not panic here.

In fact, we encounter z_erofs_lz4_decompress issue.

[54032.383633][T25392]  vmap_pages_range_noflush+0x790/0x8f8
[54032.383637][T25392]  vm_map_ram+0x1c8/0x10b0
[54032.383642][T25392]  z_erofs_lz4_decompress+0x60/0x1e8
[54032.383648][T25392]  z_erofs_decompress_pcluster+0x624/0x9fc
[54032.383653][T25392]  z_erofs_decompress_kickoff+0x18c/0x224
[54032.383658][T25392]  z_erofs_decompressqueue_endio+0x1a8/0x1e0
[54032.383663][T25392]  bio_endio+0x188/0x47c
[54032.383667][T25392]  clone_endio+0x1a0/0x550
[54032.383674][T25392]  bio_endio+0x14c/0x47c
[54032.383678][T25392]  verity_work.60258+0x7c/0x13c
[54032.383682][T25392]  process_one_work+0x1b8/0xa98
[54032.383687][T25392]  worker_thread+0x160/0x6c0
[54032.383691][T25392]  kthread+0x15c/0x1d0
[54032.383696][T25392]  ret_from_fork+0x10/0x20

z_erofs_lz4_decompress has checked the return value, so it's reasonable
to return NULL if size == 0.

Brs,
Hailong.
> 
>>                         return NULL;
>>                 addr = (unsigned long)mem;
>>         } else {
>> --
>> 2.34.1
> 
> Thanks
> Barry

Re: [PATCH] mm/vmalloc: Fix return value check for vb_alloc

[Barry Song]

On Wed, Feb 28, 2024 at 11:02 PM 刘海龙(LaoLiu) <liuhailong@oppo.com> wrote:
>
> On 2024/2/28 17:34, Barry Song wrote:
> > On Wed, Feb 28, 2024 at 9:51 PM 刘海龙(LaoLiu) <liuhailong@oppo.com> wrote:
> >>
> >> If vm_map_ram(page, 0, 0) would cause panic by vmap_pages_range_noflush, so
> >> change IS_ERR to IS_ERR_OR_NULL to fix this.
> >>
> >> Signed-off-by: Hailong.Liu <liuhailong@oppo.com>
> >> ---
> >>  mm/vmalloc.c | 2 +-
> >>  1 file changed, 1 insertion(+), 1 deletion(-)
> >>
> >> diff --git a/mm/vmalloc.c b/mm/vmalloc.c
> >> index d12a17fc0c17..109732006cf7 100644
> >> --- a/mm/vmalloc.c
> >> +++ b/mm/vmalloc.c
> >> @@ -2387,7 +2387,7 @@ void *vm_map_ram(struct page **pages, unsigned int
> >> count, int node)
> >>
> >>         if (likely(count <= VMAP_MAX_ALLOC)) {
> >>                 mem = vb_alloc(size, GFP_KERNEL);
> >> -               if (IS_ERR(mem))
> >> +               if (IS_ERR_OR_NULL(mem))
> >
> > it seems the only case for vb_alloc to return NULL is size = 0, isn't
> > it a bug of
> > caller?
> vb_alloc had already checked the size == 0, so it should be return NULL
> to caller or not panic here.
>
> In fact, we encounter z_erofs_lz4_decompress issue.
>
> [54032.383633][T25392]  vmap_pages_range_noflush+0x790/0x8f8
> [54032.383637][T25392]  vm_map_ram+0x1c8/0x10b0
> [54032.383642][T25392]  z_erofs_lz4_decompress+0x60/0x1e8
> [54032.383648][T25392]  z_erofs_decompress_pcluster+0x624/0x9fc
> [54032.383653][T25392]  z_erofs_decompress_kickoff+0x18c/0x224
> [54032.383658][T25392]  z_erofs_decompressqueue_endio+0x1a8/0x1e0
> [54032.383663][T25392]  bio_endio+0x188/0x47c
> [54032.383667][T25392]  clone_endio+0x1a0/0x550
> [54032.383674][T25392]  bio_endio+0x14c/0x47c
> [54032.383678][T25392]  verity_work.60258+0x7c/0x13c
> [54032.383682][T25392]  process_one_work+0x1b8/0xa98
> [54032.383687][T25392]  worker_thread+0x160/0x6c0
> [54032.383691][T25392]  kthread+0x15c/0x1d0
> [54032.383696][T25392]  ret_from_fork+0x10/0x20
>
> z_erofs_lz4_decompress has checked the return value, so it's reasonable
> to return NULL if size == 0.

I agree. but there is no reason to activate a WARN_ON in vb_alloc as
obviously it doesn't like it. so fix it earlier.

I even feel z_erofs_lz4_decompress is a better place than vm_map_ram
according to your description. but at least vm_map_ram is better than
checking a vb_alloc's ret after it gives a parameter obviously hated and
causes complaints.

>
> Brs,
> Hailong.
> >
> >>                         return NULL;
> >>                 addr = (unsigned long)mem;
> >>         } else {
> >> --
> >> 2.34.1
> >

Thanks
Barry