* [PATCH] btrfs: reject encoded reads from write-only file descriptors
@ 2026-11-17 4:29 Sieun Yang
2026-11-17 4:41 ` Sieun Yang
0 siblings, 1 reply; 2+ messages in thread
From: Sieun Yang @ 2026-11-17 4:29 UTC (permalink / raw)
To: dsterba; +Cc: mason, linux-btrfs, axboe, io-uring, linux-kernel
BTRFS_IOC_ENCODED_READ and the matching io_uring command require
CAP_SYS_ADMIN, but neither checks FMODE_READ. A caller can therefore
read file contents through an O_WRONLY descriptor, whereas read() and
IORING_OP_READ on that descriptor return EBADF.
This is observable under three read restrictions while CAP_SYS_ADMIN
is retained:
- Landlock denies READ_FILE. A new O_RDONLY open fails with EACCES, but
an O_WRONLY open succeeds after the restriction is installed.
- DAC mode 0200 denies O_RDONLY after CAP_DAC_OVERRIDE and
CAP_DAC_READ_SEARCH are dropped, but allows O_WRONLY.
- With CONFIG_BTRFS_FS_POSIX_ACL=y, a POSIX ACL grants only write access
to a named user. With fsuid set to that user and the DAC bypass
capabilities dropped, O_RDONLY fails and O_WRONLY succeeds.
On Linux 7.4.0-rc3-g666539eecdf4, both encoded-read paths returned
the same 8192 bytes from an uncompressed Btrfs extent in each case.
Normal read() on the O_WRONLY descriptor returned EBADF.
Check FMODE_READ in both entry points, matching the FMODE_WRITE checks
already present in the encoded-write paths. Both encoded reads then
return EBADF for O_WRONLY; O_RDONLY controls with read permission still
succeed. The Landlock and DAC cases were also verified on Linux 7.3.4.
Fixes: 1881fba89bd5 ("btrfs: add BTRFS_IOC_ENCODED_READ ioctl")
Fixes: 34310c442e17 ("btrfs: add io_uring command for encoded reads (ENCODED_READ ioctl)")
Cc: stable@vger.kernel.org
Signed-off-by: Sieun Yang <didtldms2525@jnu.ac.kr>
---
fs/btrfs/ioctl.c | 10 ++++++++++
1 file changed, 10 insertions(+)
diff --git a/fs/btrfs/ioctl.c b/fs/btrfs/ioctl.c
index 6ba883db1..f1e46b4e6 100644
--- a/fs/btrfs/ioctl.c
+++ b/fs/btrfs/ioctl.c
@@ -4392,6 +4392,11 @@ static int btrfs_ioctl_encoded_read(struct file *file, void __user *argp,
goto out_acct;
}
+ if (!(file->f_mode & FMODE_READ)) {
+ ret = -EBADF;
+ goto out_acct;
+ }
+
if (compat) {
#if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
struct btrfs_ioctl_encoded_io_args_32 args32;
@@ -4779,6 +4784,11 @@ static int btrfs_uring_encoded_read(struct io_uring_cmd *cmd, unsigned int issue
}
sqe_addr = u64_to_user_ptr(READ_ONCE(cmd->sqe->addr));
+ if (!(file->f_mode & FMODE_READ)) {
+ ret = -EBADF;
+ goto out_acct;
+ }
+
if (issue_flags & IO_URING_F_COMPAT) {
#if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
copy_end = offsetofend(struct btrfs_ioctl_encoded_io_args_32, flags);
base-commit: 666539eecdf4db623f5e3c3e55cf9f2dfad24aa4
--
2.39.5
^ permalink raw reply related [flat|nested] 2+ messages in thread* [PATCH] btrfs: reject encoded reads from write-only file descriptors
2026-11-17 4:29 [PATCH] btrfs: reject encoded reads from write-only file descriptors Sieun Yang
@ 2026-11-17 4:41 ` Sieun Yang
0 siblings, 0 replies; 2+ messages in thread
From: Sieun Yang @ 2026-11-17 4:41 UTC (permalink / raw)
To: dsterba; +Cc: mason, linux-btrfs, axboe, io-uring, linux-kernel
BTRFS_IOC_ENCODED_READ and the matching io_uring command require
CAP_SYS_ADMIN, but neither checks FMODE_READ. A caller can therefore
read file contents through an O_WRONLY descriptor, whereas read() and
IORING_OP_READ on that descriptor return EBADF.
This is observable under three read restrictions while CAP_SYS_ADMIN
is retained:
- Landlock denies READ_FILE. A new O_RDONLY open fails with EACCES, but
an O_WRONLY open succeeds after the restriction is installed.
- DAC mode 0200 denies O_RDONLY after CAP_DAC_OVERRIDE and
CAP_DAC_READ_SEARCH are dropped, but allows O_WRONLY.
- With CONFIG_BTRFS_FS_POSIX_ACL=y, a POSIX ACL grants only write access
to a named user. With fsuid set to that user and the DAC bypass
capabilities dropped, O_RDONLY fails and O_WRONLY succeeds.
On Linux 7.4.0-rc3-g666539eecdf4, both encoded-read paths returned
the same 8192 bytes from an uncompressed Btrfs extent in each case.
Normal read() on the O_WRONLY descriptor returned EBADF.
Check FMODE_READ in both entry points, matching the FMODE_WRITE checks
already present in the encoded-write paths. Both encoded reads then
return EBADF for O_WRONLY; O_RDONLY controls with read permission still
succeed. The Landlock and DAC cases were also verified on Linux 7.3.4.
Fixes: 1881fba89bd5 ("btrfs: add BTRFS_IOC_ENCODED_READ ioctl")
Fixes: 34310c442e17 ("btrfs: add io_uring command for encoded reads (ENCODED_READ ioctl)")
Cc: stable@vger.kernel.org
Signed-off-by: Sieun Yang <didtldms2525@jnu.ac.kr>
---
v2: Resend with ASCII spaces in the diff context. The v1 email had
U+00A0 at the start of context lines. No code changes.
fs/btrfs/ioctl.c | 10 ++++++++++
1 file changed, 10 insertions(+)
diff --git a/fs/btrfs/ioctl.c b/fs/btrfs/ioctl.c
index 6ba883db1..f1e46b4e6 100644
--- a/fs/btrfs/ioctl.c
+++ b/fs/btrfs/ioctl.c
@@ -4392,6 +4392,11 @@ static int btrfs_ioctl_encoded_read(struct file *file, void __user *argp,
goto out_acct;
}
+ if (!(file->f_mode & FMODE_READ)) {
+ ret = -EBADF;
+ goto out_acct;
+ }
+
if (compat) {
#if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
struct btrfs_ioctl_encoded_io_args_32 args32;
@@ -4779,6 +4784,11 @@ static int btrfs_uring_encoded_read(struct io_uring_cmd *cmd, unsigned int issue
}
sqe_addr = u64_to_user_ptr(READ_ONCE(cmd->sqe->addr));
+ if (!(file->f_mode & FMODE_READ)) {
+ ret = -EBADF;
+ goto out_acct;
+ }
+
if (issue_flags & IO_URING_F_COMPAT) {
#if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
copy_end = offsetofend(struct btrfs_ioctl_encoded_io_args_32, flags);
base-commit: 666539eecdf4db623f5e3c3e55cf9f2dfad24aa4
--
2.39.5
^ permalink raw reply related [flat|nested] 2+ messages in thread
end of thread, other threads:[~2026-11-17 4:41 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2026-11-17 4:29 [PATCH] btrfs: reject encoded reads from write-only file descriptors Sieun Yang
2026-11-17 4:41 ` Sieun Yang
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox