Skip to contentSkip to conversation

tmpfs: ensure O_LARGEFILE with generic_file_open()

PATCH Applied Thomas Weißschuh 2 participants 3 messages Sep 28, 2022

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 tmpfs: ensure O_LARGEFILE with generic_file_open()

From: Thomas Weißschuh <[email protected]>

Without this check open() will open large files on tmpfs although O_LARGEFILE was not specified. This is inconsistent with other filesystems.
Also it will later result in EOVERFLOW on stat() or EFBIG on write().

---
 mm/shmem.c | 1 +
 1 file changed, 1 insertion(+)
mm/shmem.c +1 -0
index 42e5888bf84d..902c5550fabc 100644
@@ -3876,6 +3876,7 @@ EXPORT_SYMBOL(shmem_aops);
 
 static const struct file_operations shmem_file_operations = {
 	.mmap		= shmem_mmap,
+	.open		= generic_file_open,
 	.get_unmapped_area = shmem_get_unmapped_area,
 #ifdef CONFIG_TMPFS
 	.llseek		= shmem_file_llseek,

base-commit: f76349cf41451c5c42a99f18a9163377e4b364ff

Signature · 1 line
2.37.3
maintainer Acked-by on v1 1/1
#1 lore raw
Quoted 9 lines · Thomas Weißschuh

From: Thomas Weißschuh <[email protected]>

Without this check open() will open large files on tmpfs although O_LARGEFILE was not specified. This is inconsistent with other filesystems.
Also it will later result in EOVERFLOW on stat() or EFBIG on write().

Thanks! I think you deserve some special award for finding and fixing such an ancient bug/inconsistency - dating back to early v2.4 it seems.

But only affecting 32-bit; and since we've happily lived with it so long, and the fix does change behaviour for userspace, better not to Cc stable.

There are some other filesystems still behaving as tmpfs was (ramfs, hugetlbfsi, and I didn't look further); but we do make more of an effort to keep tmpfs in line with the major filesystems - thank you.

Quoted 20 lines
---
 mm/shmem.c | 1 +
 1 file changed, 1 insertion(+)
mm/shmem.c +1 -0
index 42e5888bf84d..902c5550fabc 100644
@@ -3876,6 +3876,7 @@ EXPORT_SYMBOL(shmem_aops);
 
 static const struct file_operations shmem_file_operations = {
 	.mmap		= shmem_mmap,
+	.open		= generic_file_open,
 	.get_unmapped_area = shmem_get_unmapped_area,
 #ifdef CONFIG_TMPFS
 	.llseek		= shmem_file_llseek,

base-commit: f76349cf41451c5c42a99f18a9163377e4b364ff

Signature · 1 line
2.37.3
#2 lore raw
Quoted 10 lines · Hugh Dickins

From: Hugh Dickins <[email protected]>

Quoted 7 lines · Thomas Weißschuh

From: Thomas Weißschuh <[email protected]>

Without this check open() will open large files on tmpfs although O_LARGEFILE was not specified. This is inconsistent with other filesystems.
Also it will later result in EOVERFLOW on stat() or EFBIG on write().

Quoted 7 lines
Quoted 2 lines

Thanks! I think you deserve some special award for finding and fixing such an ancient bug/inconsistency - dating back to early v2.4 it seems.

Seems to be my special talent.

Quoted 2 lines

But only affecting 32-bit; and since we've happily lived with it so long, and the fix does change behaviour for userspace, better not to Cc stable.

That was my feeling, too.

Quoted 3 lines

There are some other filesystems still behaving as tmpfs was (ramfs, hugetlbfsi, and I didn't look further); but we do make more of an effort to keep tmpfs in line with the major filesystems - thank you.

NFS seems also to be affected, which I think warrants a fix.
My plan was to wait for the feedback to this patch and if it works out to then submit patches for NFS as validating that will be a bit more involved.

Quoted 20 lines
Quoted 20 lines
---
 mm/shmem.c | 1 +
 1 file changed, 1 insertion(+)
mm/shmem.c +1 -0
index 42e5888bf84d..902c5550fabc 100644
@@ -3876,6 +3876,7 @@ EXPORT_SYMBOL(shmem_aops);
 
 static const struct file_operations shmem_file_operations = {
 	.mmap		= shmem_mmap,
+	.open		= generic_file_open,
 	.get_unmapped_area = shmem_get_unmapped_area,
 #ifdef CONFIG_TMPFS
 	.llseek		= shmem_file_llseek,

base-commit: f76349cf41451c5c42a99f18a9163377e4b364ff

Signature · 1 line
2.37.3