[PATCH v2] fscrypt: fix left shift underflow when inode->i_blkbits > PAGE_SHIFT

[PATCH v2] fscrypt: fix left shift underflow when inode->i_blkbits > PAGE_SHIFT

[Yongpeng Yang]

From: Yongpeng Yang <yangyongpeng@xiaomi.com>

When simulating an nvme device on qemu with both logical_block_size and
physical_block_size set to 8 KiB, a error trace appears during partition
table reading at boot time. The issue is caused by inode->i_blkbits being
larger than PAGE_SHIFT, which leads to a left shift of -1 and triggering a
UBSAN warning.

[    2.697306] ------------[ cut here ]------------
[    2.697309] UBSAN: shift-out-of-bounds in fs/crypto/inline_crypt.c:336:37
[    2.697311] shift exponent -1 is negative
[    2.697315] CPU: 3 UID: 0 PID: 274 Comm: (udev-worker) Not tainted 6.18.0-rc2+ #34 PREEMPT(voluntary)
[    2.697317] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.16.3-0-ga6ed6b701f0a-prebuilt.qemu.org 04/01/2014
[    2.697320] Call Trace:
[    2.697324]  <TASK>
[    2.697325]  dump_stack_lvl+0x76/0xa0
[    2.697340]  dump_stack+0x10/0x20
[    2.697342]  __ubsan_handle_shift_out_of_bounds+0x1e3/0x390
[    2.697351]  bh_get_inode_and_lblk_num.cold+0x12/0x94
[    2.697359]  fscrypt_set_bio_crypt_ctx_bh+0x44/0x90
[    2.697365]  submit_bh_wbc+0xb6/0x190
[    2.697370]  block_read_full_folio+0x194/0x270
[    2.697371]  ? __pfx_blkdev_get_block+0x10/0x10
[    2.697375]  ? __pfx_blkdev_read_folio+0x10/0x10
[    2.697377]  blkdev_read_folio+0x18/0x30
[    2.697379]  filemap_read_folio+0x40/0xe0
[    2.697382]  filemap_get_pages+0x5ef/0x7a0
[    2.697385]  ? mmap_region+0x63/0xd0
[    2.697389]  filemap_read+0x11d/0x520
[    2.697392]  blkdev_read_iter+0x7c/0x180
[    2.697393]  vfs_read+0x261/0x390
[    2.697397]  ksys_read+0x71/0xf0
[    2.697398]  __x64_sys_read+0x19/0x30
[    2.697399]  x64_sys_call+0x1e88/0x26a0
[    2.697405]  do_syscall_64+0x80/0x670
[    2.697410]  ? __x64_sys_newfstat+0x15/0x20
[    2.697414]  ? x64_sys_call+0x204a/0x26a0
[    2.697415]  ? do_syscall_64+0xb8/0x670
[    2.697417]  ? irqentry_exit_to_user_mode+0x2e/0x2a0
[    2.697420]  ? irqentry_exit+0x43/0x50
[    2.697421]  ? exc_page_fault+0x90/0x1b0
[    2.697422]  entry_SYSCALL_64_after_hwframe+0x76/0x7e
[    2.697425] RIP: 0033:0x75054cba4a06
[    2.697426] Code: 5d e8 41 8b 93 08 03 00 00 59 5e 48 83 f8 fc 75 19 83 e2 39 83 fa 08 75 11 e8 26 ff ff ff 66 0f 1f 44 00 00 48 8b 45 10 0f 05 <48> 8b 5d f8 c9 c3 0f 1f 40 00 f3 0f 1e fa 55 48 89 e5 48 83 ec 08
[    2.697427] RSP: 002b:00007fff973723a0 EFLAGS: 00000202 ORIG_RAX: 0000000000000000
[    2.697430] RAX: ffffffffffffffda RBX: 00005ea9a2c02760 RCX: 000075054cba4a06
[    2.697432] RDX: 0000000000002000 RSI: 000075054c190000 RDI: 000000000000001b
[    2.697433] RBP: 00007fff973723c0 R08: 0000000000000000 R09: 0000000000000000
[    2.697434] R10: 0000000000000000 R11: 0000000000000202 R12: 0000000000000000
[    2.697434] R13: 00005ea9a2c027c0 R14: 00005ea9a2be5608 R15: 00005ea9a2be55f0
[    2.697436]  </TASK>
[    2.697436] ---[ end trace ]---

This situation can happen for block devices because when
CONFIG_TRANSPARENT_HUGEPAGE is enabled, the maximum logical_block_size
is 64 KiB. set_init_blocksize() then sets the block device inode->i_blkbits
to 8 KiB, which is within this limit.

File I/O does not trigger this problem because for filesystems that do not
support the FS_LBS feature, sb_set_blocksize() prevents sb->s_blocksize_bits
from being larger than PAGE_SHIFT. During inode allocation,
alloc_inode()->inode_init_always() assigns inode->i_blkbits from
sb->s_blocksize_bits. Currently, only xfs_fs_type has the FS_LBS flag, and
since xfs I/O paths do not reach submit_bh_wbc(), it does not hit the
left-shift underflow issue.

Signed-off-by: Yongpeng Yang <yangyongpeng@xiaomi.com>
---
v2:
- Added more explanations about the issue in the commit message.
---
 fs/crypto/inline_crypt.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/fs/crypto/inline_crypt.c b/fs/crypto/inline_crypt.c
index 5dee7c498bc8..6beb5f490612 100644
--- a/fs/crypto/inline_crypt.c
+++ b/fs/crypto/inline_crypt.c
@@ -333,7 +333,7 @@ static bool bh_get_inode_and_lblk_num(const struct buffer_head *bh,
 	inode = mapping->host;
 
 	*inode_ret = inode;
-	*lblk_num_ret = ((u64)folio->index << (PAGE_SHIFT - inode->i_blkbits)) +
+	*lblk_num_ret = (((u64)folio->index << PAGE_SHIFT) >> inode->i_blkbits) +
 			(bh_offset(bh) >> inode->i_blkbits);
 	return true;
 }
-- 
2.43.0

Re: [PATCH v2] fscrypt: fix left shift underflow when inode->i_blkbits > PAGE_SHIFT

[Eric Biggers]

[+linux-fsdevel, linux-block, and Luis]

On Thu, Oct 30, 2025 at 03:29:56PM +0800, Yongpeng Yang wrote:
> From: Yongpeng Yang <yangyongpeng@xiaomi.com>
> 
> When simulating an nvme device on qemu with both logical_block_size and
> physical_block_size set to 8 KiB, a error trace appears during partition
> table reading at boot time. The issue is caused by inode->i_blkbits being
> larger than PAGE_SHIFT, which leads to a left shift of -1 and triggering a
> UBSAN warning.
> 
> [    2.697306] ------------[ cut here ]------------
> [    2.697309] UBSAN: shift-out-of-bounds in fs/crypto/inline_crypt.c:336:37
> [    2.697311] shift exponent -1 is negative
> [    2.697315] CPU: 3 UID: 0 PID: 274 Comm: (udev-worker) Not tainted 6.18.0-rc2+ #34 PREEMPT(voluntary)
> [    2.697317] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.16.3-0-ga6ed6b701f0a-prebuilt.qemu.org 04/01/2014
> [    2.697320] Call Trace:
> [    2.697324]  <TASK>
> [    2.697325]  dump_stack_lvl+0x76/0xa0
> [    2.697340]  dump_stack+0x10/0x20
> [    2.697342]  __ubsan_handle_shift_out_of_bounds+0x1e3/0x390
> [    2.697351]  bh_get_inode_and_lblk_num.cold+0x12/0x94
> [    2.697359]  fscrypt_set_bio_crypt_ctx_bh+0x44/0x90
> [    2.697365]  submit_bh_wbc+0xb6/0x190
> [    2.697370]  block_read_full_folio+0x194/0x270
> [    2.697371]  ? __pfx_blkdev_get_block+0x10/0x10
> [    2.697375]  ? __pfx_blkdev_read_folio+0x10/0x10
> [    2.697377]  blkdev_read_folio+0x18/0x30
> [    2.697379]  filemap_read_folio+0x40/0xe0
> [    2.697382]  filemap_get_pages+0x5ef/0x7a0
> [    2.697385]  ? mmap_region+0x63/0xd0
> [    2.697389]  filemap_read+0x11d/0x520
> [    2.697392]  blkdev_read_iter+0x7c/0x180
> [    2.697393]  vfs_read+0x261/0x390
> [    2.697397]  ksys_read+0x71/0xf0
> [    2.697398]  __x64_sys_read+0x19/0x30
> [    2.697399]  x64_sys_call+0x1e88/0x26a0
> [    2.697405]  do_syscall_64+0x80/0x670
> [    2.697410]  ? __x64_sys_newfstat+0x15/0x20
> [    2.697414]  ? x64_sys_call+0x204a/0x26a0
> [    2.697415]  ? do_syscall_64+0xb8/0x670
> [    2.697417]  ? irqentry_exit_to_user_mode+0x2e/0x2a0
> [    2.697420]  ? irqentry_exit+0x43/0x50
> [    2.697421]  ? exc_page_fault+0x90/0x1b0
> [    2.697422]  entry_SYSCALL_64_after_hwframe+0x76/0x7e
> [    2.697425] RIP: 0033:0x75054cba4a06
> [    2.697426] Code: 5d e8 41 8b 93 08 03 00 00 59 5e 48 83 f8 fc 75 19 83 e2 39 83 fa 08 75 11 e8 26 ff ff ff 66 0f 1f 44 00 00 48 8b 45 10 0f 05 <48> 8b 5d f8 c9 c3 0f 1f 40 00 f3 0f 1e fa 55 48 89 e5 48 83 ec 08
> [    2.697427] RSP: 002b:00007fff973723a0 EFLAGS: 00000202 ORIG_RAX: 0000000000000000
> [    2.697430] RAX: ffffffffffffffda RBX: 00005ea9a2c02760 RCX: 000075054cba4a06
> [    2.697432] RDX: 0000000000002000 RSI: 000075054c190000 RDI: 000000000000001b
> [    2.697433] RBP: 00007fff973723c0 R08: 0000000000000000 R09: 0000000000000000
> [    2.697434] R10: 0000000000000000 R11: 0000000000000202 R12: 0000000000000000
> [    2.697434] R13: 00005ea9a2c027c0 R14: 00005ea9a2be5608 R15: 00005ea9a2be55f0
> [    2.697436]  </TASK>
> [    2.697436] ---[ end trace ]---
> 
> This situation can happen for block devices because when
> CONFIG_TRANSPARENT_HUGEPAGE is enabled, the maximum logical_block_size
> is 64 KiB. set_init_blocksize() then sets the block device inode->i_blkbits
> to 8 KiB, which is within this limit.
> 
> File I/O does not trigger this problem because for filesystems that do not
> support the FS_LBS feature, sb_set_blocksize() prevents sb->s_blocksize_bits
> from being larger than PAGE_SHIFT. During inode allocation,
> alloc_inode()->inode_init_always() assigns inode->i_blkbits from
> sb->s_blocksize_bits. Currently, only xfs_fs_type has the FS_LBS flag, and
> since xfs I/O paths do not reach submit_bh_wbc(), it does not hit the
> left-shift underflow issue.
> 
> Signed-off-by: Yongpeng Yang <yangyongpeng@xiaomi.com>
> ---
> v2:
> - Added more explanations about the issue in the commit message.
> ---
>  fs/crypto/inline_crypt.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/fs/crypto/inline_crypt.c b/fs/crypto/inline_crypt.c
> index 5dee7c498bc8..6beb5f490612 100644
> --- a/fs/crypto/inline_crypt.c
> +++ b/fs/crypto/inline_crypt.c
> @@ -333,7 +333,7 @@ static bool bh_get_inode_and_lblk_num(const struct buffer_head *bh,
>  	inode = mapping->host;
>  
>  	*inode_ret = inode;
> -	*lblk_num_ret = ((u64)folio->index << (PAGE_SHIFT - inode->i_blkbits)) +
> +	*lblk_num_ret = (((u64)folio->index << PAGE_SHIFT) >> inode->i_blkbits) +
>  			(bh_offset(bh) >> inode->i_blkbits);
>  	return true;
>  }

Applied to https://git.kernel.org/pub/scm/fs/fscrypt/linux.git/log/?h=for-current

I also added:

    Fixes: 47dd67532303 ("block/bdev: lift block size restrictions to 64k")
    Cc: stable@vger.kernel.org

- Eric

Re: [PATCH v2] fscrypt: fix left shift underflow when inode->i_blkbits > PAGE_SHIFT

[Christoph Hellwig]

On Mon, Nov 03, 2025 at 08:48:29AM -0800, Eric Biggers wrote:
> >  	*inode_ret = inode;
> > -	*lblk_num_ret = ((u64)folio->index << (PAGE_SHIFT - inode->i_blkbits)) +
> > +	*lblk_num_ret = (((u64)folio->index << PAGE_SHIFT) >> inode->i_blkbits) +

This should be using folio_pos() instead of open coding the arithmetics.

Re: [PATCH v2] fscrypt: fix left shift underflow when inode->i_blkbits > PAGE_SHIFT

[Yongpeng Yang]

On 11/4/2025 7:12 PM, Christoph Hellwig wrote:
> On Mon, Nov 03, 2025 at 08:48:29AM -0800, Eric Biggers wrote:
>>>   	*inode_ret = inode;
>>> -	*lblk_num_ret = ((u64)folio->index << (PAGE_SHIFT - inode->i_blkbits)) +
>>> +	*lblk_num_ret = (((u64)folio->index << PAGE_SHIFT) >> inode->i_blkbits) +
> 
> This should be using folio_pos() instead of open coding the arithmetics.
> 

How about this modification: using "<< PAGE_SHIFT" instead of "* 
PAGE_SIZE" for page_offset and folio_pos?

--- a/fs/crypto/inline_crypt.c
+++ b/fs/crypto/inline_crypt.c
@@ -333,7 +333,7 @@ static bool bh_get_inode_and_lblk_num(const struct 
buffer_head *bh,
         inode = mapping->host;

         *inode_ret = inode;
-       *lblk_num_ret = ((u64)folio->index << (PAGE_SHIFT - 
inode->i_blkbits)) +
+       *lblk_num_ret = ((u64)folio_pos(folio) >> inode->i_blkbits) +
                         (bh_offset(bh) >> inode->i_blkbits);
         return true;
  }
diff --git a/include/linux/pagemap.h b/include/linux/pagemap.h
index 09b581c1d878..72eeb1841bc3 100644
--- a/include/linux/pagemap.h
+++ b/include/linux/pagemap.h
@@ -1026,7 +1026,7 @@ static inline pgoff_t page_pgoff(const struct 
folio *folio,
   */
  static inline loff_t folio_pos(const struct folio *folio)
  {
-       return ((loff_t)folio->index) * PAGE_SIZE;
+       return ((loff_t)folio->index) << PAGE_SHIFT;
  }

  /*
@@ -1036,7 +1036,7 @@ static inline loff_t page_offset(struct page *page)
  {
         struct folio *folio = page_folio(page);

-       return folio_pos(folio) + folio_page_idx(folio, page) * PAGE_SIZE;
+       return folio_pos(folio) + (folio_page_idx(folio, page) << 
PAGE_SHIFT);
  }

Yongpeng,

Re: [PATCH v2] fscrypt: fix left shift underflow when inode->i_blkbits > PAGE_SHIFT

[Eric Biggers]

On Tue, Nov 04, 2025 at 03:12:53AM -0800, Christoph Hellwig wrote:
> On Mon, Nov 03, 2025 at 08:48:29AM -0800, Eric Biggers wrote:
> > >  	*inode_ret = inode;
> > > -	*lblk_num_ret = ((u64)folio->index << (PAGE_SHIFT - inode->i_blkbits)) +
> > > +	*lblk_num_ret = (((u64)folio->index << PAGE_SHIFT) >> inode->i_blkbits) +
> 
> This should be using folio_pos() instead of open coding the arithmetics.

Well, folio_pos() doesn't work with sizes greater than S64_MAX, and it
uses multiplication rather than a shift.

Probably doesn't matter, but I always feel like I have to actually check
that.

It looks like the size of block device can come from several different
places, including set_capacity(), bdev_resize_partition(), and
add_partition().  The first has a size check.  I don't immediately see a
size check in the other two.  Maybe it's there and I need to look
closer.  Also can the size of a block device be set in other ways?

Then I have to remember whether a multiplication of a signed value gets
reliably optimized to a shift on all architectures or not.  I think so.

Anyway, the trivial version avoids having to consider any of this...

- Eric

Re: [PATCH v2] fscrypt: fix left shift underflow when inode->i_blkbits > PAGE_SHIFT

[Christoph Hellwig]

On Tue, Nov 04, 2025 at 11:05:49PM +0800, Yongpeng Yang wrote:
> On 11/4/2025 7:12 PM, Christoph Hellwig wrote:
> > On Mon, Nov 03, 2025 at 08:48:29AM -0800, Eric Biggers wrote:
> > > >   	*inode_ret = inode;
> > > > -	*lblk_num_ret = ((u64)folio->index << (PAGE_SHIFT - inode->i_blkbits)) +
> > > > +	*lblk_num_ret = (((u64)folio->index << PAGE_SHIFT) >> inode->i_blkbits) +
> > 
> > This should be using folio_pos() instead of open coding the arithmetics.
> > 
> 
> How about this modification: using "<< PAGE_SHIFT" instead of "* PAGE_SIZE"
> for page_offset and folio_pos?

Any decent compiler turns a multiplication by a compіle time fixed
power of two constant in shifts, so why bother?

(and yes, I just double checked this happens here)

Re: [PATCH v2] fscrypt: fix left shift underflow when inode->i_blkbits > PAGE_SHIFT

[Christoph Hellwig]

On Tue, Nov 04, 2025 at 10:10:06AM -0800, Eric Biggers wrote:
> On Tue, Nov 04, 2025 at 03:12:53AM -0800, Christoph Hellwig wrote:
> > On Mon, Nov 03, 2025 at 08:48:29AM -0800, Eric Biggers wrote:
> > > >  	*inode_ret = inode;
> > > > -	*lblk_num_ret = ((u64)folio->index << (PAGE_SHIFT - inode->i_blkbits)) +
> > > > +	*lblk_num_ret = (((u64)folio->index << PAGE_SHIFT) >> inode->i_blkbits) +
> > 
> > This should be using folio_pos() instead of open coding the arithmetics.
> 
> Well, folio_pos() doesn't work with sizes greater than S64_MAX, and it
> uses multiplication rather than a shift.

What do you mean with "sizes greater than S64_MAX"?  folio_pos works
on a folio and is the MM designated helper to get the file offset
from a folio, where a file offset is a loff_t, aka s64.

And as answered to the previous mail, the compiler turns that
multiplication into a shift.

> Anyway, the trivial version avoids having to consider any of this...

I see it the other way around - folio_pos is the defined way to get
the index into the inode (block device inode here) in the abstract
way.