2004-09-13 16:02:09

by Albert Cahalan

[permalink] [raw]
Subject: fake_ino fixes

This improves /proc inode numbering a bit.
If ino_t is 32-bit: just fix task 0 handling
If ino_t is 64-bit: fix large fd numbers too

Handling PID 0xffff on 32-bit was dropped in
favor of handling PID 0 correctly, since PID 0
can be seen with the default pid_max value.

Signed-off-by: Albert Cahalan <[email protected]>

diff -Naurd ol/fs/proc/base.c nl/fs/proc/base.c
--- ol/fs/proc/base.c 2004-09-13 11:13:54.000000000 -0400
+++ nl/fs/proc/base.c 2004-09-13 11:47:29.000000000 -0400
@@ -34,14 +34,16 @@
#include <linux/ptrace.h>

/*
- * For hysterical raisins we keep the same inumbers as in the old procfs.
- * Feel free to change the macro below - just keep the range distinct from
- * inumbers of the rest of procfs (currently those are in 0x0000--0xffff).
+ * This range should be distinct from inumbers of the rest of procfs.
+ * (currently those are in 0x0000--0xffff) Remember that PID 0 can
+ * be seen. Don't go above PID 0xfffe on any port with a 32-bit ino_t!
+ * (currently: Alpha, zSeries, and all 32-bit ports) Also, you'd best
+ * avoid using over 0x8000 file descriptors per task on such ports.
* As soon as we'll get a separate superblock we will be able to forget
* about magical ranges too.
*/

-#define fake_ino(pid,ino) (((pid)<<16)|(ino))
+#define fake_ino(pid,ino) ((((ino_t)pid+1)<<(sizeof(ino_t)*4))|(ino))

enum pid_directory_inos {
PROC_TGID_INO = 2,
@@ -779,7 +781,8 @@
{
struct inode *inode = filp->f_dentry->d_inode;
struct task_struct *p = proc_task(inode);
- unsigned int fd, tid, ino;
+ unsigned int fd, tid;
+ ino_t ino;
int retval;
char buf[NUMBUF];
struct files_struct * files;




2004-09-14 02:09:43

by William Lee Irwin III

[permalink] [raw]
Subject: Re: fake_ino fixes

On Mon, Sep 13, 2004 at 11:52:19AM -0400, Albert Cahalan wrote:
> This improves /proc inode numbering a bit.
> If ino_t is 32-bit: just fix task 0 handling
> If ino_t is 64-bit: fix large fd numbers too
> Handling PID 0xffff on 32-bit was dropped in
> favor of handling PID 0 correctly, since PID 0
> can be seen with the default pid_max value.

Sounds like a good stopgap measure unless someone's already got a more
comprehensive sweep that enables larger pid spaces on 32-bit.


-- wli