[PATCH] proc: restore seekdir("/proc", 256) semantics

[PATCH] proc: restore seekdir("/proc", 256) semantics

[Alexey Dobriyan]

Long time ago "/proc/self" was an honest symlink and all not-PID entries
were output before /proc/$PID. To not lose /proc/self in readdir output
after it became permanently positive dentry it was stuck before /proc/1.

One side effect of the change was that the code

	d = opendir("/proc");
	seekdir(d, 256);

stopped pointing to the first PID for applications that want to skip all
the crap.

Later "/proc/thread-self" was added in the same way.

It looks like ps and top aren't seeking over /proc but are simply
skipping over so nobody noticed.

Restore old behaviour, make seekdir(254) point to /proc/self and
seekdir(255) point to /proc/thread-self.

Commit in question:

	commit 021ada7dff22d0d9540ff596cb0f8bb866755ee1
	procfs: switch /proc/self away from proc_dir_entry

Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
---

 fs/proc/base.c |   14 ++++++--------
 fs/proc/root.c |    4 ++--
 2 files changed, 8 insertions(+), 10 deletions(-)

--- a/fs/proc/base.c
+++ b/fs/proc/base.c
@@ -3230,8 +3230,6 @@ static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter ite
 	return iter;
 }
 
-#define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2)
-
 /* for the /proc/ directory itself, after non-process stuff has been done */
 int proc_pid_readdir(struct file *file, struct dir_context *ctx)
 {
@@ -3239,22 +3237,22 @@ int proc_pid_readdir(struct file *file, struct dir_context *ctx)
 	struct pid_namespace *ns = file_inode(file)->i_sb->s_fs_info;
 	loff_t pos = ctx->pos;
 
-	if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
+	if (pos >= PID_MAX_LIMIT + FIRST_PROCESS_ENTRY)
 		return 0;
 
-	if (pos == TGID_OFFSET - 2) {
+	if (pos == FIRST_PROCESS_ENTRY - 2) {
 		struct inode *inode = d_inode(ns->proc_self);
 		if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
 			return 0;
 		ctx->pos = pos = pos + 1;
 	}
-	if (pos == TGID_OFFSET - 1) {
+	if (pos == FIRST_PROCESS_ENTRY - 1) {
 		struct inode *inode = d_inode(ns->proc_thread_self);
 		if (!dir_emit(ctx, "thread-self", 11, inode->i_ino, DT_LNK))
 			return 0;
 		ctx->pos = pos = pos + 1;
 	}
-	iter.tgid = pos - TGID_OFFSET;
+	iter.tgid = pos - FIRST_PROCESS_ENTRY;
 	iter.task = NULL;
 	for (iter = next_tgid(ns, iter);
 	     iter.task;
@@ -3267,14 +3265,14 @@ int proc_pid_readdir(struct file *file, struct dir_context *ctx)
 			continue;
 
 		len = snprintf(name, sizeof(name), "%u", iter.tgid);
-		ctx->pos = iter.tgid + TGID_OFFSET;
+		ctx->pos = iter.tgid + FIRST_PROCESS_ENTRY;
 		if (!proc_fill_cache(file, ctx, name, len,
 				     proc_pid_instantiate, iter.task, NULL)) {
 			put_task_struct(iter.task);
 			return 0;
 		}
 	}
-	ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
+	ctx->pos = PID_MAX_LIMIT + FIRST_PROCESS_ENTRY;
 	return 0;
 }
 
--- a/fs/proc/root.c
+++ b/fs/proc/root.c
@@ -162,11 +162,11 @@ static struct dentry *proc_root_lookup(struct inode * dir, struct dentry * dentr
 
 static int proc_root_readdir(struct file *file, struct dir_context *ctx)
 {
-	if (ctx->pos < FIRST_PROCESS_ENTRY) {
+	if (ctx->pos < FIRST_PROCESS_ENTRY - 2) {
 		int error = proc_readdir(file, ctx);
 		if (unlikely(error <= 0))
 			return error;
-		ctx->pos = FIRST_PROCESS_ENTRY;
+		ctx->pos = FIRST_PROCESS_ENTRY - 2;
 	}
 
 	return proc_pid_readdir(file, ctx);

Re: [PATCH] proc: restore seekdir("/proc", 256) semantics

[Andrew Morton]

On Tue, 24 Apr 2018 00:50:09 +0300 Alexey Dobriyan <adobriyan@gmail.com> wrote:

> Long time ago "/proc/self" was an honest symlink and all not-PID entries
> were output before /proc/$PID. To not lose /proc/self in readdir output
> after it became permanently positive dentry it was stuck before /proc/1.
> 
> One side effect of the change was that the code
> 
> 	d = opendir("/proc");
> 	seekdir(d, 256);
> 
> stopped pointing to the first PID for applications that want to skip all
> the crap.
> 
> Later "/proc/thread-self" was added in the same way.
> 
> It looks like ps and top aren't seeking over /proc but are simply
> skipping over so nobody noticed.
> 
> Restore old behaviour, make seekdir(254) point to /proc/self and
> seekdir(255) point to /proc/thread-self.
> 

Gee.  Why?  That's a pretty weird artifact in the userspace API and if
we were able to withdraw it without damage then good, let's leave it
withdrawn?

I mean, if we're to make this a permanent part of the userspace API
then it should be documented in the proc(5) manpage and we should have
something in tools/testing/selftests to detect regressions in the
interface.  Good luck with all that!

Re: [PATCH] proc: restore seekdir("/proc", 256) semantics

[Pavel Machek]

Hi!

> Long time ago "/proc/self" was an honest symlink and all not-PID entries
> were output before /proc/$PID. To not lose /proc/self in readdir output
> after it became permanently positive dentry it was stuck before /proc/1.
> 
> One side effect of the change was that the code
> 
> 	d = opendir("/proc");
> 	seekdir(d, 256);
> 
> stopped pointing to the first PID for applications that want to skip all
> the crap.
> 
> Later "/proc/thread-self" was added in the same way.
> 
> It looks like ps and top aren't seeking over /proc but are simply
> skipping over so nobody noticed.
> 
> Restore old behaviour, make seekdir(254) point to /proc/self and
> seekdir(255) point to /proc/thread-self.

Is there anyone relying on the old behaviour?

									Pavel
-- 
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html

Re: [PATCH] proc: restore seekdir("/proc", 256) semantics

[Alexey Dobriyan]

On Tue, May 01, 2018 at 03:23:07PM -0700, Andrew Morton wrote:
> On Tue, 24 Apr 2018 00:50:09 +0300 Alexey Dobriyan <adobriyan@gmail.com> wrote:
> 
> > Long time ago "/proc/self" was an honest symlink and all not-PID entries
> > were output before /proc/$PID. To not lose /proc/self in readdir output
> > after it became permanently positive dentry it was stuck before /proc/1.
> > 
> > One side effect of the change was that the code
> > 
> > 	d = opendir("/proc");
> > 	seekdir(d, 256);
> > 
> > stopped pointing to the first PID for applications that want to skip all
> > the crap.
> > 
> > Later "/proc/thread-self" was added in the same way.
> > 
> > It looks like ps and top aren't seeking over /proc but are simply
> > skipping over so nobody noticed.
> > 
> > Restore old behaviour, make seekdir(254) point to /proc/self and
> > seekdir(255) point to /proc/thread-self.
> > 
> 
> Gee.  Why?  That's a pretty weird artifact in the userspace API and if
> we were able to withdraw it without damage then good, let's leave it
> withdrawn?

I checked ps, top and htop, they are not doing seekdir(256).
Apparently userspace devs were unaware of this trick.
I'll remove this before anyone notices.

> I mean, if we're to make this a permanent part of the userspace API
> then it should be documented in the proc(5) manpage and we should have
> something in tools/testing/selftests to detect regressions in the
> interface.  Good luck with all that!