]> err.no Git - linux-2.6/commitdiff
[GFS2] Red Hat bz 228540: owner references
authorRobert Peterson <rpeterso@redhat.com>
Fri, 23 Mar 2007 22:05:15 +0000 (17:05 -0500)
committerSteven Whitehouse <swhiteho@redhat.com>
Tue, 1 May 2007 08:10:55 +0000 (09:10 +0100)
In Testing the previously posted and accepted patch for
https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=228540
I uncovered some gfs2 badness.  It turns out that the current
gfs2 code saves off a process pointer when glocks is taken
in both the glock and glock holder structures.  Those
structures will persist in memory long after the process has
ended; pointers to poisoned memory.

This problem isn't caused by the 228540 fix; the new capability
introduced by the fix just uncovered the problem.

I wrote this patch that avoids saving process pointers
and instead saves off the process pid.  Rather than
referencing the bad pointers, it now does process lookups.
There is special code that makes the output nicer for
printing holder information for processes that have ended.

This patch also adds a stub for the new "sprint_symbol"
function that exists in Andrew Morton's -mm patch set, but
won't go into the base kernel until 2.6.22, since it adds
functionality but doesn't fix a bug.

Signed-off-by: Bob Peterson <rpeterso@redhat.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
fs/gfs2/glock.c
fs/gfs2/glock.h
fs/gfs2/incore.h

index b8aa816bb6eb278c77474e0c3cdf0910e13d2502..d2e3094c40f8614123e7aa1f2d19500efa0b17c7 100644 (file)
@@ -25,6 +25,8 @@
 #include <asm/uaccess.h>
 #include <linux/seq_file.h>
 #include <linux/debugfs.h>
+#include <linux/module.h>
+#include <linux/kallsyms.h>
 
 #include "gfs2.h"
 #include "incore.h"
@@ -54,6 +56,7 @@ struct glock_iter {
 typedef void (*glock_examiner) (struct gfs2_glock * gl);
 
 static int gfs2_dump_lockstate(struct gfs2_sbd *sdp);
+static int dump_glock(struct glock_iter *gi, struct gfs2_glock *gl);
 static void gfs2_glock_xmote_th(struct gfs2_glock *gl, struct gfs2_holder *gh);
 static void gfs2_glock_drop_th(struct gfs2_glock *gl);
 static DECLARE_RWSEM(gfs2_umount_flush_sem);
@@ -64,6 +67,7 @@ static struct dentry *gfs2_root;
 #define GFS2_GL_HASH_MASK       (GFS2_GL_HASH_SIZE - 1)
 
 static struct gfs2_gl_hash_bucket gl_hash_table[GFS2_GL_HASH_SIZE];
+static struct dentry *gfs2_root;
 
 /*
  * Despite what you might think, the numbers below are not arbitrary :-)
@@ -312,7 +316,7 @@ int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
        atomic_set(&gl->gl_ref, 1);
        gl->gl_state = LM_ST_UNLOCKED;
        gl->gl_hash = hash;
-       gl->gl_owner = NULL;
+       gl->gl_owner_pid = 0;
        gl->gl_ip = 0;
        gl->gl_ops = glops;
        gl->gl_req_gh = NULL;
@@ -376,7 +380,7 @@ void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, unsigned flags,
        INIT_LIST_HEAD(&gh->gh_list);
        gh->gh_gl = gl;
        gh->gh_ip = (unsigned long)__builtin_return_address(0);
-       gh->gh_owner = current;
+       gh->gh_owner_pid = current->pid;
        gh->gh_state = state;
        gh->gh_flags = flags;
        gh->gh_error = 0;
@@ -601,7 +605,7 @@ static void gfs2_glmutex_lock(struct gfs2_glock *gl)
        if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
                list_add_tail(&gh.gh_list, &gl->gl_waiters1);
        } else {
-               gl->gl_owner = current;
+               gl->gl_owner_pid = current->pid;
                gl->gl_ip = (unsigned long)__builtin_return_address(0);
                clear_bit(HIF_WAIT, &gh.gh_iflags);
                smp_mb();
@@ -628,7 +632,7 @@ static int gfs2_glmutex_trylock(struct gfs2_glock *gl)
        if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
                acquired = 0;
        } else {
-               gl->gl_owner = current;
+               gl->gl_owner_pid = current->pid;
                gl->gl_ip = (unsigned long)__builtin_return_address(0);
        }
        spin_unlock(&gl->gl_spin);
@@ -646,7 +650,7 @@ static void gfs2_glmutex_unlock(struct gfs2_glock *gl)
 {
        spin_lock(&gl->gl_spin);
        clear_bit(GLF_LOCK, &gl->gl_flags);
-       gl->gl_owner = NULL;
+       gl->gl_owner_pid = 0;
        gl->gl_ip = 0;
        run_queue(gl);
        BUG_ON(!spin_is_locked(&gl->gl_spin));
@@ -999,12 +1003,12 @@ static int glock_wait_internal(struct gfs2_holder *gh)
 }
 
 static inline struct gfs2_holder *
-find_holder_by_owner(struct list_head *head, struct task_struct *owner)
+find_holder_by_owner(struct list_head *head, pid_t pid)
 {
        struct gfs2_holder *gh;
 
        list_for_each_entry(gh, head, gh_list) {
-               if (gh->gh_owner == owner)
+               if (gh->gh_owner_pid == pid)
                        return gh;
        }
 
@@ -1036,24 +1040,24 @@ static void add_to_queue(struct gfs2_holder *gh)
        struct gfs2_glock *gl = gh->gh_gl;
        struct gfs2_holder *existing;
 
-       BUG_ON(!gh->gh_owner);
+       BUG_ON(!gh->gh_owner_pid);
        if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
                BUG();
 
-       existing = find_holder_by_owner(&gl->gl_holders, gh->gh_owner);
+       existing = find_holder_by_owner(&gl->gl_holders, gh->gh_owner_pid);
        if (existing) {
                print_symbol(KERN_WARNING "original: %s\n", existing->gh_ip);
-               printk(KERN_INFO "pid : %d\n", existing->gh_owner->pid);
+               printk(KERN_INFO "pid : %d\n", existing->gh_owner_pid);
                printk(KERN_INFO "lock type : %d lock state : %d\n",
                                existing->gh_gl->gl_name.ln_type, existing->gh_gl->gl_state);
                print_symbol(KERN_WARNING "new: %s\n", gh->gh_ip);
-               printk(KERN_INFO "pid : %d\n", gh->gh_owner->pid);
+               printk(KERN_INFO "pid : %d\n", gh->gh_owner_pid);
                printk(KERN_INFO "lock type : %d lock state : %d\n",
                                gl->gl_name.ln_type, gl->gl_state);
                BUG();
        }
 
-       existing = find_holder_by_owner(&gl->gl_waiters3, gh->gh_owner);
+       existing = find_holder_by_owner(&gl->gl_waiters3, gh->gh_owner_pid);
        if (existing) {
                print_symbol(KERN_WARNING "original: %s\n", existing->gh_ip);
                print_symbol(KERN_WARNING "new: %s\n", gh->gh_ip);
@@ -1756,6 +1760,22 @@ void gfs2_gl_hash_clear(struct gfs2_sbd *sdp, int wait)
  *  Diagnostic routines to help debug distributed deadlock
  */
 
+static void gfs2_print_symbol(struct glock_iter *gi, const char *fmt,
+                              unsigned long address)
+{
+/* when sprint_symbol becomes available in the new kernel, replace this */
+/* function with:
+        char buffer[KSYM_SYMBOL_LEN];
+
+        sprint_symbol(buffer, address);
+        print_dbg(gi, fmt, buffer);
+*/
+        if (gi)
+                print_dbg(gi, fmt, address);
+        else
+                print_symbol(fmt, address);
+}
+
 /**
  * dump_holder - print information about a glock holder
  * @str: a string naming the type of holder
@@ -1768,10 +1788,18 @@ static int dump_holder(struct glock_iter *gi, char *str,
                       struct gfs2_holder *gh)
 {
        unsigned int x;
+       struct task_struct *gh_owner;
 
        print_dbg(gi, "  %s\n", str);
-       print_dbg(gi, "    owner = %ld\n",
-                  (gh->gh_owner) ? (long)gh->gh_owner->pid : -1);
+       if (gh->gh_owner_pid) {
+               print_dbg(gi, "    owner = %ld ", (long)gh->gh_owner_pid);
+               gh_owner = find_task_by_pid(gh->gh_owner_pid);
+               if (gh_owner)
+                       print_dbg(gi, "(%s)\n", gh_owner->comm);
+               else
+                       print_dbg(gi, "(ended)\n");
+       } else
+               print_dbg(gi, "    owner = -1\n");
        print_dbg(gi, "    gh_state = %u\n", gh->gh_state);
        print_dbg(gi, "    gh_flags =");
        for (x = 0; x < 32; x++)
@@ -1784,10 +1812,7 @@ static int dump_holder(struct glock_iter *gi, char *str,
                if (test_bit(x, &gh->gh_iflags))
                        print_dbg(gi, " %u", x);
        print_dbg(gi, " \n");
-       if (gi)
-               print_dbg(gi, "    initialized at: 0x%x\n", gh->gh_ip);
-       else
-               print_symbol(KERN_INFO "    initialized at: %s\n", gh->gh_ip);
+        gfs2_print_symbol(gi, "    initialized at: %s\n", gh->gh_ip);
 
        return 0;
 }
@@ -1828,6 +1853,7 @@ static int dump_glock(struct glock_iter *gi, struct gfs2_glock *gl)
        struct gfs2_holder *gh;
        unsigned int x;
        int error = -ENOBUFS;
+       struct task_struct *gl_owner;
 
        spin_lock(&gl->gl_spin);
 
@@ -1838,10 +1864,21 @@ static int dump_glock(struct glock_iter *gi, struct gfs2_glock *gl)
                if (test_bit(x, &gl->gl_flags))
                        print_dbg(gi, " %u", x);
        }
+       if (!test_bit(GLF_LOCK, &gl->gl_flags))
+               print_dbg(gi, " (unlocked)");
        print_dbg(gi, " \n");
        print_dbg(gi, "  gl_ref = %d\n", atomic_read(&gl->gl_ref));
        print_dbg(gi, "  gl_state = %u\n", gl->gl_state);
-       print_dbg(gi, "  gl_owner = %s\n", gl->gl_owner->comm);
+       if (gl->gl_owner_pid) {
+               gl_owner = find_task_by_pid(gl->gl_owner_pid);
+               if (gl_owner)
+                       print_dbg(gi, "  gl_owner = pid %d (%s)\n",
+                                 gl->gl_owner_pid, gl_owner->comm);
+               else
+                       print_dbg(gi, "  gl_owner = %d (ended)\n",
+                                 gl->gl_owner_pid);
+       } else
+               print_dbg(gi, "  gl_owner = -1\n");
        print_dbg(gi, "  gl_ip = %lu\n", gl->gl_ip);
        print_dbg(gi, "  req_gh = %s\n", (gl->gl_req_gh) ? "yes" : "no");
        print_dbg(gi, "  req_bh = %s\n", (gl->gl_req_bh) ? "yes" : "no");
index 5e662eadc6f24e9fab1fc8bed04e6c3397b679d8..11477ca3a3c00ccf918d8df3dc577c85045cbe75 100644 (file)
@@ -38,7 +38,7 @@ static inline int gfs2_glock_is_locked_by_me(struct gfs2_glock *gl)
        /* Look in glock's list of holders for one with current task as owner */
        spin_lock(&gl->gl_spin);
        list_for_each_entry(gh, &gl->gl_holders, gh_list) {
-               if (gh->gh_owner == current) {
+               if (gh->gh_owner_pid == current->pid) {
                        locked = 1;
                        break;
                }
index 9c125823d760c48f5167829bbb3d927b6ba1fbfe..fdf04705906f056595edcd2c814e79fee8655035 100644 (file)
@@ -127,7 +127,7 @@ struct gfs2_holder {
        struct list_head gh_list;
 
        struct gfs2_glock *gh_gl;
-       struct task_struct *gh_owner;
+       pid_t gh_owner_pid;
        unsigned int gh_state;
        unsigned gh_flags;
 
@@ -155,7 +155,7 @@ struct gfs2_glock {
        unsigned int gl_hash;
        unsigned int gl_demote_state; /* state requested by remote node */
        unsigned long gl_demote_time; /* time of first demote request */
-       struct task_struct *gl_owner;
+       pid_t gl_owner_pid;
        unsigned long gl_ip;
        struct list_head gl_holders;
        struct list_head gl_waiters1;   /* HIF_MUTEX */