aboutsummaryrefslogtreecommitdiffstats
path: root/kernel
diff options
context:
space:
mode:
authorPaul Menage <menage@google.com>2009-01-07 21:08:38 -0500
committerLinus Torvalds <torvalds@linux-foundation.org>2009-01-08 11:31:10 -0500
commite7c5ec9193d32b9559a3bb8893ceedbda85201ff (patch)
treefd7505c03e2c7525a110a702907c0604c233e2d9 /kernel
parent2cb378c862777d050c20db903b119a029845fdcb (diff)
cgroups: add css_tryget()
Add css_tryget(), that obtains a counted reference on a CSS. It is used in situations where the caller has a "weak" reference to the CSS, i.e. one that does not protect the cgroup from removal via a reference count, but would instead be cleaned up by a destroy() callback. css_tryget() will return true on success, or false if the cgroup is being removed. This is similar to Kamezawa Hiroyuki's patch from a week or two ago, but with the difference that in the event of css_tryget() racing with a cgroup_rmdir(), css_tryget() will only return false if the cgroup really does get removed. This implementation is done by biasing css->refcnt, so that a refcnt of 1 means "releasable" and 0 means "released or releasing". In the event of a race, css_tryget() distinguishes between "released" and "releasing" by checking for the CSS_REMOVED flag in css->flags. Signed-off-by: Paul Menage <menage@google.com> Tested-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Li Zefan <lizf@cn.fujitsu.com> Cc: Balbir Singh <balbir@in.ibm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Diffstat (limited to 'kernel')
-rw-r--r--kernel/cgroup.c61
1 files changed, 56 insertions, 5 deletions
diff --git a/kernel/cgroup.c b/kernel/cgroup.c
index 8b6379cdf637..c29831076e7a 100644
--- a/kernel/cgroup.c
+++ b/kernel/cgroup.c
@@ -2333,7 +2333,7 @@ static void init_cgroup_css(struct cgroup_subsys_state *css,
2333 struct cgroup *cgrp) 2333 struct cgroup *cgrp)
2334{ 2334{
2335 css->cgroup = cgrp; 2335 css->cgroup = cgrp;
2336 atomic_set(&css->refcnt, 0); 2336 atomic_set(&css->refcnt, 1);
2337 css->flags = 0; 2337 css->flags = 0;
2338 if (cgrp == dummytop) 2338 if (cgrp == dummytop)
2339 set_bit(CSS_ROOT, &css->flags); 2339 set_bit(CSS_ROOT, &css->flags);
@@ -2465,7 +2465,7 @@ static int cgroup_has_css_refs(struct cgroup *cgrp)
2465{ 2465{
2466 /* Check the reference count on each subsystem. Since we 2466 /* Check the reference count on each subsystem. Since we
2467 * already established that there are no tasks in the 2467 * already established that there are no tasks in the
2468 * cgroup, if the css refcount is also 0, then there should 2468 * cgroup, if the css refcount is also 1, then there should
2469 * be no outstanding references, so the subsystem is safe to 2469 * be no outstanding references, so the subsystem is safe to
2470 * destroy. We scan across all subsystems rather than using 2470 * destroy. We scan across all subsystems rather than using
2471 * the per-hierarchy linked list of mounted subsystems since 2471 * the per-hierarchy linked list of mounted subsystems since
@@ -2486,12 +2486,62 @@ static int cgroup_has_css_refs(struct cgroup *cgrp)
2486 * matter, since it can only happen if the cgroup 2486 * matter, since it can only happen if the cgroup
2487 * has been deleted and hence no longer needs the 2487 * has been deleted and hence no longer needs the
2488 * release agent to be called anyway. */ 2488 * release agent to be called anyway. */
2489 if (css && atomic_read(&css->refcnt)) 2489 if (css && (atomic_read(&css->refcnt) > 1))
2490 return 1; 2490 return 1;
2491 } 2491 }
2492 return 0; 2492 return 0;
2493} 2493}
2494 2494
2495/*
2496 * Atomically mark all (or else none) of the cgroup's CSS objects as
2497 * CSS_REMOVED. Return true on success, or false if the cgroup has
2498 * busy subsystems. Call with cgroup_mutex held
2499 */
2500
2501static int cgroup_clear_css_refs(struct cgroup *cgrp)
2502{
2503 struct cgroup_subsys *ss;
2504 unsigned long flags;
2505 bool failed = false;
2506 local_irq_save(flags);
2507 for_each_subsys(cgrp->root, ss) {
2508 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
2509 int refcnt;
2510 do {
2511 /* We can only remove a CSS with a refcnt==1 */
2512 refcnt = atomic_read(&css->refcnt);
2513 if (refcnt > 1) {
2514 failed = true;
2515 goto done;
2516 }
2517 BUG_ON(!refcnt);
2518 /*
2519 * Drop the refcnt to 0 while we check other
2520 * subsystems. This will cause any racing
2521 * css_tryget() to spin until we set the
2522 * CSS_REMOVED bits or abort
2523 */
2524 } while (atomic_cmpxchg(&css->refcnt, refcnt, 0) != refcnt);
2525 }
2526 done:
2527 for_each_subsys(cgrp->root, ss) {
2528 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
2529 if (failed) {
2530 /*
2531 * Restore old refcnt if we previously managed
2532 * to clear it from 1 to 0
2533 */
2534 if (!atomic_read(&css->refcnt))
2535 atomic_set(&css->refcnt, 1);
2536 } else {
2537 /* Commit the fact that the CSS is removed */
2538 set_bit(CSS_REMOVED, &css->flags);
2539 }
2540 }
2541 local_irq_restore(flags);
2542 return !failed;
2543}
2544
2495static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry) 2545static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry)
2496{ 2546{
2497 struct cgroup *cgrp = dentry->d_fsdata; 2547 struct cgroup *cgrp = dentry->d_fsdata;
@@ -2522,7 +2572,7 @@ static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry)
2522 2572
2523 if (atomic_read(&cgrp->count) 2573 if (atomic_read(&cgrp->count)
2524 || !list_empty(&cgrp->children) 2574 || !list_empty(&cgrp->children)
2525 || cgroup_has_css_refs(cgrp)) { 2575 || !cgroup_clear_css_refs(cgrp)) {
2526 mutex_unlock(&cgroup_mutex); 2576 mutex_unlock(&cgroup_mutex);
2527 return -EBUSY; 2577 return -EBUSY;
2528 } 2578 }
@@ -3078,7 +3128,8 @@ void __css_put(struct cgroup_subsys_state *css)
3078{ 3128{
3079 struct cgroup *cgrp = css->cgroup; 3129 struct cgroup *cgrp = css->cgroup;
3080 rcu_read_lock(); 3130 rcu_read_lock();
3081 if (atomic_dec_and_test(&css->refcnt) && notify_on_release(cgrp)) { 3131 if ((atomic_dec_return(&css->refcnt) == 1) &&
3132 notify_on_release(cgrp)) {
3082 set_bit(CGRP_RELEASABLE, &cgrp->flags); 3133 set_bit(CGRP_RELEASABLE, &cgrp->flags);
3083 check_for_release(cgrp); 3134 check_for_release(cgrp);
3084 } 3135 }