Commit ff7ec8dc1b64 ("proc: use the same treatment to check proc_lseek
as ones for proc_read_iter et.al") breaks lseek() for all /proc/net
entries, as shown for instance by pasta(1), a user-mode network
implementation using those entries to scan for bound ports:
$ strace -e openat,lseek -e s=none pasta -- true
[...]
openat(AT_FDCWD, "/proc/net/tcp", O_RDONLY|O_CLOEXEC) = 12
openat(AT_FDCWD, "/proc/net/tcp6", O_RDONLY|O_CLOEXEC) = 13
lseek(12, 0, SEEK_SET) = -1 ESPIPE (Illegal seek)
lseek() failed on /proc/net file: Illegal seek
lseek(13, 0, SEEK_SET) = -1 ESPIPE (Illegal seek)
lseek() failed on /proc/net file: Illegal seek
openat(AT_FDCWD, "/proc/net/udp", O_RDONLY|O_CLOEXEC) = 14
openat(AT_FDCWD, "/proc/net/udp6", O_RDONLY|O_CLOEXEC) = 15
lseek(14, 0, SEEK_SET) = -1 ESPIPE (Illegal seek)
lseek() failed on /proc/net file: Illegal seek
lseek(15, 0, SEEK_SET) = -1 ESPIPE (Illegal seek)
lseek() failed on /proc/net file: Illegal seek
[...]
That's because PROC_ENTRY_proc_lseek isn't set for /proc/net entries,
and it's now mandatory for lseek(). In fact, flags aren't set at all
for those entries because pde_set_flags() isn't called for them.
As commit d919b33dafb3 ("proc: faster open/read/close with "permanent"
files") introduced flags for procfs directory entries, along with the
pde_set_flags() helper, they weren't relevant for /proc/net entries,
so the lack of pde_set_flags() calls in proc_create_net_*() functions
was harmless.
Now that the calls are strictly needed for lseek() functionality,
add them.
Fixes: ff7ec8dc1b64 ("proc: use the same treatment to check proc_lseek as ones for proc_read_iter et.al")
Signed-off-by: Stefano Brivio
On Fri, 22 Aug 2025 19:23:35 +0200 Stefano Brivio
Commit ff7ec8dc1b64 ("proc: use the same treatment to check proc_lseek as ones for proc_read_iter et.al") breaks lseek() for all /proc/net entries, as shown for instance by pasta(1), a user-mode network implementation using those entries to scan for bound ports:
$ strace -e openat,lseek -e s=none pasta -- true [...] openat(AT_FDCWD, "/proc/net/tcp", O_RDONLY|O_CLOEXEC) = 12 openat(AT_FDCWD, "/proc/net/tcp6", O_RDONLY|O_CLOEXEC) = 13 lseek(12, 0, SEEK_SET) = -1 ESPIPE (Illegal seek) lseek() failed on /proc/net file: Illegal seek lseek(13, 0, SEEK_SET) = -1 ESPIPE (Illegal seek) lseek() failed on /proc/net file: Illegal seek openat(AT_FDCWD, "/proc/net/udp", O_RDONLY|O_CLOEXEC) = 14 openat(AT_FDCWD, "/proc/net/udp6", O_RDONLY|O_CLOEXEC) = 15 lseek(14, 0, SEEK_SET) = -1 ESPIPE (Illegal seek) lseek() failed on /proc/net file: Illegal seek lseek(15, 0, SEEK_SET) = -1 ESPIPE (Illegal seek) lseek() failed on /proc/net file: Illegal seek [...]
That's because PROC_ENTRY_proc_lseek isn't set for /proc/net entries, and it's now mandatory for lseek(). In fact, flags aren't set at all for those entries because pde_set_flags() isn't called for them.
As commit d919b33dafb3 ("proc: faster open/read/close with "permanent" files") introduced flags for procfs directory entries, along with the pde_set_flags() helper, they weren't relevant for /proc/net entries, so the lack of pde_set_flags() calls in proc_create_net_*() functions was harmless.
Now that the calls are strictly needed for lseek() functionality, add them.
Thanks. We already have https://lkml.kernel.org/r/20250821105806.1453833-1-wangzijie1@honor.com - does that look suitable?
On Fri, 22 Aug 2025 16:09:04 -0700
Andrew Morton
On Fri, 22 Aug 2025 19:23:35 +0200 Stefano Brivio
wrote: Commit ff7ec8dc1b64 ("proc: use the same treatment to check proc_lseek as ones for proc_read_iter et.al") breaks lseek() for all /proc/net entries, as shown for instance by pasta(1), a user-mode network implementation using those entries to scan for bound ports:
$ strace -e openat,lseek -e s=none pasta -- true [...] openat(AT_FDCWD, "/proc/net/tcp", O_RDONLY|O_CLOEXEC) = 12 openat(AT_FDCWD, "/proc/net/tcp6", O_RDONLY|O_CLOEXEC) = 13 lseek(12, 0, SEEK_SET) = -1 ESPIPE (Illegal seek) lseek() failed on /proc/net file: Illegal seek lseek(13, 0, SEEK_SET) = -1 ESPIPE (Illegal seek) lseek() failed on /proc/net file: Illegal seek openat(AT_FDCWD, "/proc/net/udp", O_RDONLY|O_CLOEXEC) = 14 openat(AT_FDCWD, "/proc/net/udp6", O_RDONLY|O_CLOEXEC) = 15 lseek(14, 0, SEEK_SET) = -1 ESPIPE (Illegal seek) lseek() failed on /proc/net file: Illegal seek lseek(15, 0, SEEK_SET) = -1 ESPIPE (Illegal seek) lseek() failed on /proc/net file: Illegal seek [...]
That's because PROC_ENTRY_proc_lseek isn't set for /proc/net entries, and it's now mandatory for lseek(). In fact, flags aren't set at all for those entries because pde_set_flags() isn't called for them.
As commit d919b33dafb3 ("proc: faster open/read/close with "permanent" files") introduced flags for procfs directory entries, along with the pde_set_flags() helper, they weren't relevant for /proc/net entries, so the lack of pde_set_flags() calls in proc_create_net_*() functions was harmless.
Now that the calls are strictly needed for lseek() functionality, add them.
Thanks. We already have https://lkml.kernel.org/r/20250821105806.1453833-1-wangzijie1@honor.com - does that look suitable?
Sorry, I didn't spot that one. It sure does! -- Stefano
Commit ff7ec8dc1b64 ("proc: use the same treatment to check proc_lseek as ones for proc_read_iter et.al") breaks lseek() for all /proc/net entries, as shown for instance by pasta(1), a user-mode network implementation using those entries to scan for bound ports:
$ strace -e openat,lseek -e s=none pasta -- true [...] openat(AT_FDCWD, "/proc/net/tcp", O_RDONLY|O_CLOEXEC) = 12 openat(AT_FDCWD, "/proc/net/tcp6", O_RDONLY|O_CLOEXEC) = 13 lseek(12, 0, SEEK_SET) = -1 ESPIPE (Illegal seek) lseek() failed on /proc/net file: Illegal seek lseek(13, 0, SEEK_SET) = -1 ESPIPE (Illegal seek) lseek() failed on /proc/net file: Illegal seek openat(AT_FDCWD, "/proc/net/udp", O_RDONLY|O_CLOEXEC) = 14 openat(AT_FDCWD, "/proc/net/udp6", O_RDONLY|O_CLOEXEC) = 15 lseek(14, 0, SEEK_SET) = -1 ESPIPE (Illegal seek) lseek() failed on /proc/net file: Illegal seek lseek(15, 0, SEEK_SET) = -1 ESPIPE (Illegal seek) lseek() failed on /proc/net file: Illegal seek [...]
That's because PROC_ENTRY_proc_lseek isn't set for /proc/net entries, and it's now mandatory for lseek(). In fact, flags aren't set at all for those entries because pde_set_flags() isn't called for them.
As commit d919b33dafb3 ("proc: faster open/read/close with "permanent" files") introduced flags for procfs directory entries, along with the pde_set_flags() helper, they weren't relevant for /proc/net entries, so the lack of pde_set_flags() calls in proc_create_net_*() functions was harmless.
Now that the calls are strictly needed for lseek() functionality, add them.
Fixes: ff7ec8dc1b64 ("proc: use the same treatment to check proc_lseek as ones for proc_read_iter et.al") Signed-off-by: Stefano Brivio
--- fs/proc/generic.c | 2 +- fs/proc/internal.h | 1 + fs/proc/proc_net.c | 4 ++++ 3 files changed, 6 insertions(+), 1 deletion(-) diff --git a/fs/proc/generic.c b/fs/proc/generic.c index 76e800e38c8f..57ec5e385d1b 100644 --- a/fs/proc/generic.c +++ b/fs/proc/generic.c @@ -561,7 +561,7 @@ 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) +void pde_set_flags(struct proc_dir_entry *pde) { if (pde->proc_ops->proc_flags & PROC_ENTRY_PERMANENT) pde->flags |= PROC_ENTRY_PERMANENT; diff --git a/fs/proc/internal.h b/fs/proc/internal.h index e737401d7383..a358974f14d2 100644 --- a/fs/proc/internal.h +++ b/fs/proc/internal.h @@ -284,6 +284,7 @@ extern struct dentry *proc_lookup(struct inode *, struct dentry *, unsigned int) struct dentry *proc_lookup_de(struct inode *, struct dentry *, struct proc_dir_entry *); extern int proc_readdir(struct file *, struct dir_context *); int proc_readdir_de(struct file *, struct dir_context *, struct proc_dir_entry *); +void pde_set_flags(struct proc_dir_entry *pde);
static inline void pde_get(struct proc_dir_entry *pde) { diff --git a/fs/proc/proc_net.c b/fs/proc/proc_net.c index 52f0b75cbce2..20bc7481b02c 100644 --- a/fs/proc/proc_net.c +++ b/fs/proc/proc_net.c @@ -124,6 +124,7 @@ struct proc_dir_entry *proc_create_net_data(const char *name, umode_t mode, p->proc_ops = &proc_net_seq_ops; p->seq_ops = ops; p->state_size = state_size; + pde_set_flags(p); return proc_register(parent, p); } EXPORT_SYMBOL_GPL(proc_create_net_data); @@ -170,6 +171,7 @@ struct proc_dir_entry *proc_create_net_data_write(const char *name, umode_t mode p->seq_ops = ops; p->state_size = state_size; p->write = write; + pde_set_flags(p); return proc_register(parent, p); } EXPORT_SYMBOL_GPL(proc_create_net_data_write); @@ -217,6 +219,7 @@ struct proc_dir_entry *proc_create_net_single(const char *name, umode_t mode, pde_force_lookup(p); p->proc_ops = &proc_net_single_ops; p->single_show = show; + pde_set_flags(p); return proc_register(parent, p); } EXPORT_SYMBOL_GPL(proc_create_net_single); @@ -261,6 +264,7 @@ struct proc_dir_entry *proc_create_net_single_write(const char *name, umode_t mo p->proc_ops = &proc_net_single_ops; p->single_show = show; p->write = write; + pde_set_flags(p); return proc_register(parent, p); } EXPORT_SYMBOL_GPL(proc_create_net_single_write); -- 2.43.0
Hi Stefano, Thanks for your patch, Lars reported this bug last week: https://lore.kernel.org/all/20250815195616.64497967@chagall.paradoxon.rec/ Jiri suggested to make pde_set_flags() part of proc_register(). I think it can help to avoid lack of pde_set_flags() calls in the future and make code clean. I have submitted a patch: https://lore.kernel.org/all/20250821105806.1453833-1-wangzijie1@honor.com
Hi wangzijie,
On Sat, 23 Aug 2025 09:53:49 +0800
wangzijie
Commit ff7ec8dc1b64 ("proc: use the same treatment to check proc_lseek as ones for proc_read_iter et.al") breaks lseek() for all /proc/net entries, as shown for instance by pasta(1), a user-mode network implementation using those entries to scan for bound ports:
$ strace -e openat,lseek -e s=none pasta -- true [...] openat(AT_FDCWD, "/proc/net/tcp", O_RDONLY|O_CLOEXEC) = 12 openat(AT_FDCWD, "/proc/net/tcp6", O_RDONLY|O_CLOEXEC) = 13 lseek(12, 0, SEEK_SET) = -1 ESPIPE (Illegal seek) lseek() failed on /proc/net file: Illegal seek lseek(13, 0, SEEK_SET) = -1 ESPIPE (Illegal seek) lseek() failed on /proc/net file: Illegal seek openat(AT_FDCWD, "/proc/net/udp", O_RDONLY|O_CLOEXEC) = 14 openat(AT_FDCWD, "/proc/net/udp6", O_RDONLY|O_CLOEXEC) = 15 lseek(14, 0, SEEK_SET) = -1 ESPIPE (Illegal seek) lseek() failed on /proc/net file: Illegal seek lseek(15, 0, SEEK_SET) = -1 ESPIPE (Illegal seek) lseek() failed on /proc/net file: Illegal seek [...]
That's because PROC_ENTRY_proc_lseek isn't set for /proc/net entries, and it's now mandatory for lseek(). In fact, flags aren't set at all for those entries because pde_set_flags() isn't called for them.
As commit d919b33dafb3 ("proc: faster open/read/close with "permanent" files") introduced flags for procfs directory entries, along with the pde_set_flags() helper, they weren't relevant for /proc/net entries, so the lack of pde_set_flags() calls in proc_create_net_*() functions was harmless.
Now that the calls are strictly needed for lseek() functionality, add them.
Fixes: ff7ec8dc1b64 ("proc: use the same treatment to check proc_lseek as ones for proc_read_iter et.al") Signed-off-by: Stefano Brivio
--- fs/proc/generic.c | 2 +- fs/proc/internal.h | 1 + fs/proc/proc_net.c | 4 ++++ 3 files changed, 6 insertions(+), 1 deletion(-) diff --git a/fs/proc/generic.c b/fs/proc/generic.c index 76e800e38c8f..57ec5e385d1b 100644 --- a/fs/proc/generic.c +++ b/fs/proc/generic.c @@ -561,7 +561,7 @@ 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) +void pde_set_flags(struct proc_dir_entry *pde) { if (pde->proc_ops->proc_flags & PROC_ENTRY_PERMANENT) pde->flags |= PROC_ENTRY_PERMANENT; diff --git a/fs/proc/internal.h b/fs/proc/internal.h index e737401d7383..a358974f14d2 100644 --- a/fs/proc/internal.h +++ b/fs/proc/internal.h @@ -284,6 +284,7 @@ extern struct dentry *proc_lookup(struct inode *, struct dentry *, unsigned int) struct dentry *proc_lookup_de(struct inode *, struct dentry *, struct proc_dir_entry *); extern int proc_readdir(struct file *, struct dir_context *); int proc_readdir_de(struct file *, struct dir_context *, struct proc_dir_entry *); +void pde_set_flags(struct proc_dir_entry *pde);
static inline void pde_get(struct proc_dir_entry *pde) { diff --git a/fs/proc/proc_net.c b/fs/proc/proc_net.c index 52f0b75cbce2..20bc7481b02c 100644 --- a/fs/proc/proc_net.c +++ b/fs/proc/proc_net.c @@ -124,6 +124,7 @@ struct proc_dir_entry *proc_create_net_data(const char *name, umode_t mode, p->proc_ops = &proc_net_seq_ops; p->seq_ops = ops; p->state_size = state_size; + pde_set_flags(p); return proc_register(parent, p); } EXPORT_SYMBOL_GPL(proc_create_net_data); @@ -170,6 +171,7 @@ struct proc_dir_entry *proc_create_net_data_write(const char *name, umode_t mode p->seq_ops = ops; p->state_size = state_size; p->write = write; + pde_set_flags(p); return proc_register(parent, p); } EXPORT_SYMBOL_GPL(proc_create_net_data_write); @@ -217,6 +219,7 @@ struct proc_dir_entry *proc_create_net_single(const char *name, umode_t mode, pde_force_lookup(p); p->proc_ops = &proc_net_single_ops; p->single_show = show; + pde_set_flags(p); return proc_register(parent, p); } EXPORT_SYMBOL_GPL(proc_create_net_single); @@ -261,6 +264,7 @@ struct proc_dir_entry *proc_create_net_single_write(const char *name, umode_t mo p->proc_ops = &proc_net_single_ops; p->single_show = show; p->write = write; + pde_set_flags(p); return proc_register(parent, p); } EXPORT_SYMBOL_GPL(proc_create_net_single_write); -- 2.43.0
Hi Stefano, Thanks for your patch, Lars reported this bug last week: https://lore.kernel.org/all/20250815195616.64497967@chagall.paradoxon.rec/
Apologies, I didn't see that. It's definitely the same issue.
Jiri suggested to make pde_set_flags() part of proc_register(). I think it can help to avoid lack of pde_set_flags() calls in the future and make code clean.
Right, I was pondering to try something like that, but I just wanted to submit a minimal fix for the moment being. In any case, you have it ready, and it's obviously cleaner than my solution, so, of course, let's discard my patch.
I have submitted a patch: https://lore.kernel.org/all/20250821105806.1453833-1-wangzijie1@honor.com
I just tested it, it works for me. Thanks for fixing that! -- Stefano
participants (3)
-
Andrew Morton
-
Stefano Brivio
-
wangzijie