To avoid potential UAF issues during module removal races, we use pde_set_flags() to save proc_ops flags in PDE itself before proc_register(), and then use pde_has_proc_*() helpers instead of directly dereferencing pde->proc_ops->*.
However, the pde_set_flags() call was missing when creating net related proc files. This omission caused incorrect behavior which FMODE_LSEEK was being cleared inappropriately in proc_reg_open() for net proc files. Lars reported it in this link[1].
Fix this by ensuring pde_set_flags() is called when register proc entry, and add NULL check for proc_ops in pde_set_flags().
Fixes: ff7ec8dc1b64 ("proc: use the same treatment to check proc_lseek as ones for proc_read_iter et.al") Cc: stable@vger.kernel.org Reported-by: Lars Wendler polynomial-c@gmx.de Signed-off-by: wangzijie wangzijie1@honor.com
v3:
- followed by Christian's suggestion to stash pde->proc_ops in a local const variable
v2:
- followed by Jiri's suggestion to refractor code and reformat commit message
fs/proc/generic.c | 38 +++++++++++++++++++++----------------- 1 file changed, 21 insertions(+), 17 deletions(-)
diff --git a/fs/proc/generic.c b/fs/proc/generic.c index 76e800e38..bd0c099cf 100644 --- a/fs/proc/generic.c +++ b/fs/proc/generic.c @@ -367,6 +367,25 @@ static const struct inode_operations proc_dir_inode_operations = { .setattr = proc_notify_change, }; +static void pde_set_flags(struct proc_dir_entry *pde) +{
- const struct proc_ops *proc_ops = pde->proc_ops;
- if (!proc_ops)
return;- if (proc_ops->proc_flags & PROC_ENTRY_PERMANENT)
pde->flags |= PROC_ENTRY_PERMANENT;- if (proc_ops->proc_read_iter)
pde->flags |= PROC_ENTRY_proc_read_iter;+#ifdef CONFIG_COMPAT
- if (proc_ops->proc_compat_ioctl)
pde->flags |= PROC_ENTRY_proc_compat_ioctl;+#endif
- if (proc_ops->proc_lseek)
pde->flags |= PROC_ENTRY_proc_lseek;+}
/* returns the registered entry, or frees dp and returns NULL on failure */ struct proc_dir_entry *proc_register(struct proc_dir_entry *dir, struct proc_dir_entry *dp) @@ -374,6 +393,8 @@ struct proc_dir_entry *proc_register(struct proc_dir_entry *dir, if (proc_alloc_inum(&dp->low_ino)) goto out_free_entry;
- pde_set_flags(dp);
- write_lock(&proc_subdir_lock); dp->parent = dir; if (pde_subdir_insert(dir, dp) == false) {
@@ -561,20 +582,6 @@ struct proc_dir_entry *proc_create_reg(const char *name, umode_t mode, return p; } -static void pde_set_flags(struct proc_dir_entry *pde) -{
- if (pde->proc_ops->proc_flags & PROC_ENTRY_PERMANENT)
pde->flags |= PROC_ENTRY_PERMANENT;- if (pde->proc_ops->proc_read_iter)
pde->flags |= PROC_ENTRY_proc_read_iter;-#ifdef CONFIG_COMPAT
- if (pde->proc_ops->proc_compat_ioctl)
pde->flags |= PROC_ENTRY_proc_compat_ioctl;-#endif
- if (pde->proc_ops->proc_lseek)
pde->flags |= PROC_ENTRY_proc_lseek;-}
struct proc_dir_entry *proc_create_data(const char *name, umode_t mode, struct proc_dir_entry *parent, const struct proc_ops *proc_ops, void *data) @@ -585,7 +592,6 @@ struct proc_dir_entry *proc_create_data(const char *name, umode_t mode, if (!p) return NULL; p->proc_ops = proc_ops;
- pde_set_flags(p); return proc_register(parent, p);
} EXPORT_SYMBOL(proc_create_data); @@ -636,7 +642,6 @@ struct proc_dir_entry *proc_create_seq_private(const char *name, umode_t mode, p->proc_ops = &proc_seq_ops; p->seq_ops = ops; p->state_size = state_size;
- pde_set_flags(p); return proc_register(parent, p);
} EXPORT_SYMBOL(proc_create_seq_private); @@ -667,7 +672,6 @@ struct proc_dir_entry *proc_create_single_data(const char *name, umode_t mode, return NULL; p->proc_ops = &proc_single_ops; p->single_show = show;
- pde_set_flags(p); return proc_register(parent, p);
} EXPORT_SYMBOL(proc_create_single_data); -- 2.25.1
Hi all,
Sorry for disturbing, Brad reported type confusion introduced by this patch to me, we missed a part in the definition of proc_dir_entry: union { const struct proc_ops *proc_ops; const struct file_operations *proc_dir_ops; };
Quoting the email he sent to me: " any dereference of ->proc_ops assuming it is a proc_ops structure results in type confusion, it probably won't be (easily) noticed without CONFIG_RANDSTRUCT because of the way in which proc_flags overlaps with the module owner pointer in the file_operations structure(and the alignment of the struct module pointed to).
With RANDSTRUCT enabled, users are likely to observe warnings on proc dir removal like: [ 19.093369] removing permanent /proc entry '12/i8042' [ 19.093396] WARNING: CPU: 0 PID: 1 at fs/proc/generic.c:886 remove_proc_subtree+0xcc/0x218 "
I examined where proc_register() is called, we can do !S_ISDIR(dp->mode) test before calling pde_set_flags() to fix it(Brad also suggested to do this), or we can drop this patch and apply the initial version[1]. I think the former is better.
Dose anyone have any suggestions?
[1] https://lore.kernel.org/all/20250818040535.564611-1-wangzijie1@honor.com/.