aboutsummaryrefslogtreecommitdiffstats
path: root/kernel/cpuset.c
diff options
context:
space:
mode:
authorPaul Jackson <pj@sgi.com>2005-08-09 13:07:59 -0400
committerLinus Torvalds <torvalds@g5.osdl.org>2005-08-09 15:08:22 -0400
commit3077a260e9f316b611436b1506eec9cc5c4f8aa6 (patch)
tree43b7d5faa5f204904c713c463015792d9ff56b01 /kernel/cpuset.c
parenta242b44da6feb604c4c659b78f63dedb69b2d4a3 (diff)
[PATCH] cpuset release ABBA deadlock fix
Fix possible cpuset_sem ABBA deadlock if 'notify_on_release' set. For a particular usage pattern, creating and destroying cpusets fairly frequently using notify_on_release, on a very large system, this deadlock can be seen every few days. If you are not using the cpuset notify_on_release feature, you will never see this deadlock. The existing code, on task exit (or cpuset deletion) did: get cpuset_sem if cpuset marked notify_on_release and is ready to release: compute cpuset path relative to /dev/cpuset mount point call_usermodehelper() forks /sbin/cpuset_release_agent with path drop cpuset_sem Unfortunately, the fork in call_usermodehelper can allocate memory, and allocating memory can require cpuset_sem, if the mems_generation values changed in the interim. This results in an ABBA deadlock, trying to obtain cpuset_sem when it is already held by the current task. To fix this, I put the cpuset path (which must be computed while holding cpuset_sem) in a temporary buffer, to be used in the call_usermodehelper call of /sbin/cpuset_release_agent only _after_ dropping cpuset_sem. So the new logic is: get cpuset_sem if cpuset marked notify_on_release and is ready to release: compute cpuset path relative to /dev/cpuset mount point stash path in kmalloc'd buffer drop cpuset_sem call_usermodehelper() forks /sbin/cpuset_release_agent with path free path The sharp eyed reader might notice that this patch does not contain any calls to kmalloc. The existing code in the check_for_release() routine was already kmalloc'ing a buffer to hold the cpuset path. In the old code, it just held the buffer for a few lines, over the cpuset_release_agent() call that in turn invoked call_usermodehelper(). In the new code, with the application of this patch, it returns that buffer via the new char **ppathbuf parameter, for later use and freeing in cpuset_release_agent(), which is called after cpuset_sem is dropped. Whereas the old code has just one call to cpuset_release_agent(), right in the check_for_release() routine, the new code has three calls to cpuset_release_agent(), from the various places that a cpuset can be released. This patch has been build and booted on SN2, and passed a stress test that previously hit the deadlock within a few seconds. Signed-off-by: Paul Jackson <pj@sgi.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Diffstat (limited to 'kernel/cpuset.c')
-rw-r--r--kernel/cpuset.c68
1 files changed, 48 insertions, 20 deletions
diff --git a/kernel/cpuset.c b/kernel/cpuset.c
index 805fb9097318..21a4e3b2cbda 100644
--- a/kernel/cpuset.c
+++ b/kernel/cpuset.c
@@ -398,21 +398,31 @@ static int cpuset_path(const struct cpuset *cs, char *buf, int buflen)
398 * to continue to serve a useful existence. Next time it's released, 398 * to continue to serve a useful existence. Next time it's released,
399 * we will get notified again, if it still has 'notify_on_release' set. 399 * we will get notified again, if it still has 'notify_on_release' set.
400 * 400 *
401 * Note final arg to call_usermodehelper() is 0 - that means 401 * The final arg to call_usermodehelper() is 0, which means don't
402 * don't wait. Since we are holding the global cpuset_sem here, 402 * wait. The separate /sbin/cpuset_release_agent task is forked by
403 * and we are asking another thread (started from keventd) to rmdir a 403 * call_usermodehelper(), then control in this thread returns here,
404 * cpuset, we can't wait - or we'd deadlock with the removing thread 404 * without waiting for the release agent task. We don't bother to
405 * on cpuset_sem. 405 * wait because the caller of this routine has no use for the exit
406 * status of the /sbin/cpuset_release_agent task, so no sense holding
407 * our caller up for that.
408 *
409 * The simple act of forking that task might require more memory,
410 * which might need cpuset_sem. So this routine must be called while
411 * cpuset_sem is not held, to avoid a possible deadlock. See also
412 * comments for check_for_release(), below.
406 */ 413 */
407 414
408static int cpuset_release_agent(char *cpuset_str) 415static void cpuset_release_agent(const char *pathbuf)
409{ 416{
410 char *argv[3], *envp[3]; 417 char *argv[3], *envp[3];
411 int i; 418 int i;
412 419
420 if (!pathbuf)
421 return;
422
413 i = 0; 423 i = 0;
414 argv[i++] = "/sbin/cpuset_release_agent"; 424 argv[i++] = "/sbin/cpuset_release_agent";
415 argv[i++] = cpuset_str; 425 argv[i++] = (char *)pathbuf;
416 argv[i] = NULL; 426 argv[i] = NULL;
417 427
418 i = 0; 428 i = 0;
@@ -421,17 +431,29 @@ static int cpuset_release_agent(char *cpuset_str)
421 envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin"; 431 envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
422 envp[i] = NULL; 432 envp[i] = NULL;
423 433
424 return call_usermodehelper(argv[0], argv, envp, 0); 434 call_usermodehelper(argv[0], argv, envp, 0);
435 kfree(pathbuf);
425} 436}
426 437
427/* 438/*
428 * Either cs->count of using tasks transitioned to zero, or the 439 * Either cs->count of using tasks transitioned to zero, or the
429 * cs->children list of child cpusets just became empty. If this 440 * cs->children list of child cpusets just became empty. If this
430 * cs is notify_on_release() and now both the user count is zero and 441 * cs is notify_on_release() and now both the user count is zero and
431 * the list of children is empty, send notice to user land. 442 * the list of children is empty, prepare cpuset path in a kmalloc'd
443 * buffer, to be returned via ppathbuf, so that the caller can invoke
444 * cpuset_release_agent() with it later on, once cpuset_sem is dropped.
445 * Call here with cpuset_sem held.
446 *
447 * This check_for_release() routine is responsible for kmalloc'ing
448 * pathbuf. The above cpuset_release_agent() is responsible for
449 * kfree'ing pathbuf. The caller of these routines is responsible
450 * for providing a pathbuf pointer, initialized to NULL, then
451 * calling check_for_release() with cpuset_sem held and the address
452 * of the pathbuf pointer, then dropping cpuset_sem, then calling
453 * cpuset_release_agent() with pathbuf, as set by check_for_release().
432 */ 454 */
433 455
434static void check_for_release(struct cpuset *cs) 456static void check_for_release(struct cpuset *cs, char **ppathbuf)
435{ 457{
436 if (notify_on_release(cs) && atomic_read(&cs->count) == 0 && 458 if (notify_on_release(cs) && atomic_read(&cs->count) == 0 &&
437 list_empty(&cs->children)) { 459 list_empty(&cs->children)) {
@@ -441,10 +463,9 @@ static void check_for_release(struct cpuset *cs)
441 if (!buf) 463 if (!buf)
442 return; 464 return;
443 if (cpuset_path(cs, buf, PAGE_SIZE) < 0) 465 if (cpuset_path(cs, buf, PAGE_SIZE) < 0)
444 goto out; 466 kfree(buf);
445 cpuset_release_agent(buf); 467 else
446out: 468 *ppathbuf = buf;
447 kfree(buf);
448 } 469 }
449} 470}
450 471
@@ -727,14 +748,14 @@ static int update_flag(cpuset_flagbits_t bit, struct cpuset *cs, char *buf)
727 return 0; 748 return 0;
728} 749}
729 750
730static int attach_task(struct cpuset *cs, char *buf) 751static int attach_task(struct cpuset *cs, char *pidbuf, char **ppathbuf)
731{ 752{
732 pid_t pid; 753 pid_t pid;
733 struct task_struct *tsk; 754 struct task_struct *tsk;
734 struct cpuset *oldcs; 755 struct cpuset *oldcs;
735 cpumask_t cpus; 756 cpumask_t cpus;
736 757
737 if (sscanf(buf, "%d", &pid) != 1) 758 if (sscanf(pidbuf, "%d", &pid) != 1)
738 return -EIO; 759 return -EIO;
739 if (cpus_empty(cs->cpus_allowed) || nodes_empty(cs->mems_allowed)) 760 if (cpus_empty(cs->cpus_allowed) || nodes_empty(cs->mems_allowed))
740 return -ENOSPC; 761 return -ENOSPC;
@@ -777,7 +798,7 @@ static int attach_task(struct cpuset *cs, char *buf)
777 798
778 put_task_struct(tsk); 799 put_task_struct(tsk);
779 if (atomic_dec_and_test(&oldcs->count)) 800 if (atomic_dec_and_test(&oldcs->count))
780 check_for_release(oldcs); 801 check_for_release(oldcs, ppathbuf);
781 return 0; 802 return 0;
782} 803}
783 804
@@ -801,6 +822,7 @@ static ssize_t cpuset_common_file_write(struct file *file, const char __user *us
801 struct cftype *cft = __d_cft(file->f_dentry); 822 struct cftype *cft = __d_cft(file->f_dentry);
802 cpuset_filetype_t type = cft->private; 823 cpuset_filetype_t type = cft->private;
803 char *buffer; 824 char *buffer;
825 char *pathbuf = NULL;
804 int retval = 0; 826 int retval = 0;
805 827
806 /* Crude upper limit on largest legitimate cpulist user might write. */ 828 /* Crude upper limit on largest legitimate cpulist user might write. */
@@ -841,7 +863,7 @@ static ssize_t cpuset_common_file_write(struct file *file, const char __user *us
841 retval = update_flag(CS_NOTIFY_ON_RELEASE, cs, buffer); 863 retval = update_flag(CS_NOTIFY_ON_RELEASE, cs, buffer);
842 break; 864 break;
843 case FILE_TASKLIST: 865 case FILE_TASKLIST:
844 retval = attach_task(cs, buffer); 866 retval = attach_task(cs, buffer, &pathbuf);
845 break; 867 break;
846 default: 868 default:
847 retval = -EINVAL; 869 retval = -EINVAL;
@@ -852,6 +874,7 @@ static ssize_t cpuset_common_file_write(struct file *file, const char __user *us
852 retval = nbytes; 874 retval = nbytes;
853out2: 875out2:
854 up(&cpuset_sem); 876 up(&cpuset_sem);
877 cpuset_release_agent(pathbuf);
855out1: 878out1:
856 kfree(buffer); 879 kfree(buffer);
857 return retval; 880 return retval;
@@ -1357,6 +1380,7 @@ static int cpuset_rmdir(struct inode *unused_dir, struct dentry *dentry)
1357 struct cpuset *cs = dentry->d_fsdata; 1380 struct cpuset *cs = dentry->d_fsdata;
1358 struct dentry *d; 1381 struct dentry *d;
1359 struct cpuset *parent; 1382 struct cpuset *parent;
1383 char *pathbuf = NULL;
1360 1384
1361 /* the vfs holds both inode->i_sem already */ 1385 /* the vfs holds both inode->i_sem already */
1362 1386
@@ -1376,7 +1400,7 @@ static int cpuset_rmdir(struct inode *unused_dir, struct dentry *dentry)
1376 update_cpu_domains(cs); 1400 update_cpu_domains(cs);
1377 list_del(&cs->sibling); /* delete my sibling from parent->children */ 1401 list_del(&cs->sibling); /* delete my sibling from parent->children */
1378 if (list_empty(&parent->children)) 1402 if (list_empty(&parent->children))
1379 check_for_release(parent); 1403 check_for_release(parent, &pathbuf);
1380 spin_lock(&cs->dentry->d_lock); 1404 spin_lock(&cs->dentry->d_lock);
1381 d = dget(cs->dentry); 1405 d = dget(cs->dentry);
1382 cs->dentry = NULL; 1406 cs->dentry = NULL;
@@ -1384,6 +1408,7 @@ static int cpuset_rmdir(struct inode *unused_dir, struct dentry *dentry)
1384 cpuset_d_remove_dir(d); 1408 cpuset_d_remove_dir(d);
1385 dput(d); 1409 dput(d);
1386 up(&cpuset_sem); 1410 up(&cpuset_sem);
1411 cpuset_release_agent(pathbuf);
1387 return 0; 1412 return 0;
1388} 1413}
1389 1414
@@ -1483,10 +1508,13 @@ void cpuset_exit(struct task_struct *tsk)
1483 task_unlock(tsk); 1508 task_unlock(tsk);
1484 1509
1485 if (notify_on_release(cs)) { 1510 if (notify_on_release(cs)) {
1511 char *pathbuf = NULL;
1512
1486 down(&cpuset_sem); 1513 down(&cpuset_sem);
1487 if (atomic_dec_and_test(&cs->count)) 1514 if (atomic_dec_and_test(&cs->count))
1488 check_for_release(cs); 1515 check_for_release(cs, &pathbuf);
1489 up(&cpuset_sem); 1516 up(&cpuset_sem);
1517 cpuset_release_agent(pathbuf);
1490 } else { 1518 } else {
1491 atomic_dec(&cs->count); 1519 atomic_dec(&cs->count);
1492 } 1520 }