diff options
-rw-r--r-- | Documentation/cgroups/cgroups.txt | 27 | ||||
-rw-r--r-- | include/linux/cgroup.h | 17 | ||||
-rw-r--r-- | kernel/cgroup.c | 47 |
3 files changed, 10 insertions, 81 deletions
diff --git a/Documentation/cgroups/cgroups.txt b/Documentation/cgroups/cgroups.txt index 8e74980ab385..4a0b64c605fc 100644 --- a/Documentation/cgroups/cgroups.txt +++ b/Documentation/cgroups/cgroups.txt | |||
@@ -370,15 +370,12 @@ To mount a cgroup hierarchy with just the cpuset and memory | |||
370 | subsystems, type: | 370 | subsystems, type: |
371 | # mount -t cgroup -o cpuset,memory hier1 /sys/fs/cgroup/rg1 | 371 | # mount -t cgroup -o cpuset,memory hier1 /sys/fs/cgroup/rg1 |
372 | 372 | ||
373 | To change the set of subsystems bound to a mounted hierarchy, just | 373 | While remounting cgroups is currently supported, it is not recommend |
374 | remount with different options: | 374 | to use it. Remounting allows changing bound subsystems and |
375 | # mount -o remount,cpuset,blkio hier1 /sys/fs/cgroup/rg1 | 375 | release_agent. Rebinding is hardly useful as it only works when the |
376 | 376 | hierarchy is empty and release_agent itself should be replaced with | |
377 | Now memory is removed from the hierarchy and blkio is added. | 377 | conventional fsnotify. The support for remounting will be removed in |
378 | 378 | the future. | |
379 | Note this will add blkio to the hierarchy but won't remove memory or | ||
380 | cpuset, because the new options are appended to the old ones: | ||
381 | # mount -o remount,blkio /sys/fs/cgroup/rg1 | ||
382 | 379 | ||
383 | To Specify a hierarchy's release_agent: | 380 | To Specify a hierarchy's release_agent: |
384 | # mount -t cgroup -o cpuset,release_agent="/sbin/cpuset_release_agent" \ | 381 | # mount -t cgroup -o cpuset,release_agent="/sbin/cpuset_release_agent" \ |
@@ -637,16 +634,6 @@ void exit(struct task_struct *task) | |||
637 | 634 | ||
638 | Called during task exit. | 635 | Called during task exit. |
639 | 636 | ||
640 | int populate(struct cgroup *cgrp) | ||
641 | (cgroup_mutex held by caller) | ||
642 | |||
643 | Called after creation of a cgroup to allow a subsystem to populate | ||
644 | the cgroup directory with file entries. The subsystem should make | ||
645 | calls to cgroup_add_file() with objects of type cftype (see | ||
646 | include/linux/cgroup.h for details). Note that although this | ||
647 | method can return an error code, the error code is currently not | ||
648 | always handled well. | ||
649 | |||
650 | void post_clone(struct cgroup *cgrp) | 637 | void post_clone(struct cgroup *cgrp) |
651 | (cgroup_mutex held by caller) | 638 | (cgroup_mutex held by caller) |
652 | 639 | ||
@@ -656,7 +643,7 @@ example in cpusets, no task may attach before 'cpus' and 'mems' are set | |||
656 | up. | 643 | up. |
657 | 644 | ||
658 | void bind(struct cgroup *root) | 645 | void bind(struct cgroup *root) |
659 | (cgroup_mutex and ss->hierarchy_mutex held by caller) | 646 | (cgroup_mutex held by caller) |
660 | 647 | ||
661 | Called when a cgroup subsystem is rebound to a different hierarchy | 648 | Called when a cgroup subsystem is rebound to a different hierarchy |
662 | and root cgroup. Currently this will only involve movement between | 649 | and root cgroup. Currently this will only involve movement between |
diff --git a/include/linux/cgroup.h b/include/linux/cgroup.h index d3f5fba2c159..c90eaa803440 100644 --- a/include/linux/cgroup.h +++ b/include/linux/cgroup.h | |||
@@ -500,21 +500,8 @@ struct cgroup_subsys { | |||
500 | const char *name; | 500 | const char *name; |
501 | 501 | ||
502 | /* | 502 | /* |
503 | * Protects sibling/children links of cgroups in this | ||
504 | * hierarchy, plus protects which hierarchy (or none) the | ||
505 | * subsystem is a part of (i.e. root/sibling). To avoid | ||
506 | * potential deadlocks, the following operations should not be | ||
507 | * undertaken while holding any hierarchy_mutex: | ||
508 | * | ||
509 | * - allocating memory | ||
510 | * - initiating hotplug events | ||
511 | */ | ||
512 | struct mutex hierarchy_mutex; | ||
513 | struct lock_class_key subsys_key; | ||
514 | |||
515 | /* | ||
516 | * Link to parent, and list entry in parent's children. | 503 | * Link to parent, and list entry in parent's children. |
517 | * Protected by this->hierarchy_mutex and cgroup_lock() | 504 | * Protected by cgroup_lock() |
518 | */ | 505 | */ |
519 | struct cgroupfs_root *root; | 506 | struct cgroupfs_root *root; |
520 | struct list_head sibling; | 507 | struct list_head sibling; |
@@ -602,7 +589,7 @@ int cgroup_attach_task_all(struct task_struct *from, struct task_struct *); | |||
602 | * the lifetime of cgroup_subsys_state is subsys's matter. | 589 | * the lifetime of cgroup_subsys_state is subsys's matter. |
603 | * | 590 | * |
604 | * Looking up and scanning function should be called under rcu_read_lock(). | 591 | * Looking up and scanning function should be called under rcu_read_lock(). |
605 | * Taking cgroup_mutex()/hierarchy_mutex() is not necessary for following calls. | 592 | * Taking cgroup_mutex is not necessary for following calls. |
606 | * But the css returned by this routine can be "not populated yet" or "being | 593 | * But the css returned by this routine can be "not populated yet" or "being |
607 | * destroyed". The caller should check css and cgroup's status. | 594 | * destroyed". The caller should check css and cgroup's status. |
608 | */ | 595 | */ |
diff --git a/kernel/cgroup.c b/kernel/cgroup.c index af2b5641fc8b..79818507e444 100644 --- a/kernel/cgroup.c +++ b/kernel/cgroup.c | |||
@@ -954,7 +954,7 @@ static int cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft) | |||
954 | 954 | ||
955 | dget(d); | 955 | dget(d); |
956 | d_delete(d); | 956 | d_delete(d); |
957 | simple_unlink(d->d_inode, d); | 957 | simple_unlink(cgrp->dentry->d_inode, d); |
958 | list_del_init(&cfe->node); | 958 | list_del_init(&cfe->node); |
959 | dput(d); | 959 | dput(d); |
960 | 960 | ||
@@ -1068,28 +1068,24 @@ static int rebind_subsystems(struct cgroupfs_root *root, | |||
1068 | BUG_ON(cgrp->subsys[i]); | 1068 | BUG_ON(cgrp->subsys[i]); |
1069 | BUG_ON(!dummytop->subsys[i]); | 1069 | BUG_ON(!dummytop->subsys[i]); |
1070 | BUG_ON(dummytop->subsys[i]->cgroup != dummytop); | 1070 | BUG_ON(dummytop->subsys[i]->cgroup != dummytop); |
1071 | mutex_lock(&ss->hierarchy_mutex); | ||
1072 | cgrp->subsys[i] = dummytop->subsys[i]; | 1071 | cgrp->subsys[i] = dummytop->subsys[i]; |
1073 | cgrp->subsys[i]->cgroup = cgrp; | 1072 | cgrp->subsys[i]->cgroup = cgrp; |
1074 | list_move(&ss->sibling, &root->subsys_list); | 1073 | list_move(&ss->sibling, &root->subsys_list); |
1075 | ss->root = root; | 1074 | ss->root = root; |
1076 | if (ss->bind) | 1075 | if (ss->bind) |
1077 | ss->bind(cgrp); | 1076 | ss->bind(cgrp); |
1078 | mutex_unlock(&ss->hierarchy_mutex); | ||
1079 | /* refcount was already taken, and we're keeping it */ | 1077 | /* refcount was already taken, and we're keeping it */ |
1080 | } else if (bit & removed_bits) { | 1078 | } else if (bit & removed_bits) { |
1081 | /* We're removing this subsystem */ | 1079 | /* We're removing this subsystem */ |
1082 | BUG_ON(ss == NULL); | 1080 | BUG_ON(ss == NULL); |
1083 | BUG_ON(cgrp->subsys[i] != dummytop->subsys[i]); | 1081 | BUG_ON(cgrp->subsys[i] != dummytop->subsys[i]); |
1084 | BUG_ON(cgrp->subsys[i]->cgroup != cgrp); | 1082 | BUG_ON(cgrp->subsys[i]->cgroup != cgrp); |
1085 | mutex_lock(&ss->hierarchy_mutex); | ||
1086 | if (ss->bind) | 1083 | if (ss->bind) |
1087 | ss->bind(dummytop); | 1084 | ss->bind(dummytop); |
1088 | dummytop->subsys[i]->cgroup = dummytop; | 1085 | dummytop->subsys[i]->cgroup = dummytop; |
1089 | cgrp->subsys[i] = NULL; | 1086 | cgrp->subsys[i] = NULL; |
1090 | subsys[i]->root = &rootnode; | 1087 | subsys[i]->root = &rootnode; |
1091 | list_move(&ss->sibling, &rootnode.subsys_list); | 1088 | list_move(&ss->sibling, &rootnode.subsys_list); |
1092 | mutex_unlock(&ss->hierarchy_mutex); | ||
1093 | /* subsystem is now free - drop reference on module */ | 1089 | /* subsystem is now free - drop reference on module */ |
1094 | module_put(ss->module); | 1090 | module_put(ss->module); |
1095 | } else if (bit & final_bits) { | 1091 | } else if (bit & final_bits) { |
@@ -3915,37 +3911,6 @@ static void init_cgroup_css(struct cgroup_subsys_state *css, | |||
3915 | set_bit(CSS_CLEAR_CSS_REFS, &css->flags); | 3911 | set_bit(CSS_CLEAR_CSS_REFS, &css->flags); |
3916 | } | 3912 | } |
3917 | 3913 | ||
3918 | static void cgroup_lock_hierarchy(struct cgroupfs_root *root) | ||
3919 | { | ||
3920 | /* We need to take each hierarchy_mutex in a consistent order */ | ||
3921 | int i; | ||
3922 | |||
3923 | /* | ||
3924 | * No worry about a race with rebind_subsystems that might mess up the | ||
3925 | * locking order, since both parties are under cgroup_mutex. | ||
3926 | */ | ||
3927 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | ||
3928 | struct cgroup_subsys *ss = subsys[i]; | ||
3929 | if (ss == NULL) | ||
3930 | continue; | ||
3931 | if (ss->root == root) | ||
3932 | mutex_lock(&ss->hierarchy_mutex); | ||
3933 | } | ||
3934 | } | ||
3935 | |||
3936 | static void cgroup_unlock_hierarchy(struct cgroupfs_root *root) | ||
3937 | { | ||
3938 | int i; | ||
3939 | |||
3940 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | ||
3941 | struct cgroup_subsys *ss = subsys[i]; | ||
3942 | if (ss == NULL) | ||
3943 | continue; | ||
3944 | if (ss->root == root) | ||
3945 | mutex_unlock(&ss->hierarchy_mutex); | ||
3946 | } | ||
3947 | } | ||
3948 | |||
3949 | /* | 3914 | /* |
3950 | * cgroup_create - create a cgroup | 3915 | * cgroup_create - create a cgroup |
3951 | * @parent: cgroup that will be parent of the new cgroup | 3916 | * @parent: cgroup that will be parent of the new cgroup |
@@ -4006,9 +3971,7 @@ static long cgroup_create(struct cgroup *parent, struct dentry *dentry, | |||
4006 | ss->post_clone(cgrp); | 3971 | ss->post_clone(cgrp); |
4007 | } | 3972 | } |
4008 | 3973 | ||
4009 | cgroup_lock_hierarchy(root); | ||
4010 | list_add(&cgrp->sibling, &cgrp->parent->children); | 3974 | list_add(&cgrp->sibling, &cgrp->parent->children); |
4011 | cgroup_unlock_hierarchy(root); | ||
4012 | root->number_of_cgroups++; | 3975 | root->number_of_cgroups++; |
4013 | 3976 | ||
4014 | err = cgroup_create_dir(cgrp, dentry, mode); | 3977 | err = cgroup_create_dir(cgrp, dentry, mode); |
@@ -4035,9 +3998,7 @@ static long cgroup_create(struct cgroup *parent, struct dentry *dentry, | |||
4035 | 3998 | ||
4036 | err_remove: | 3999 | err_remove: |
4037 | 4000 | ||
4038 | cgroup_lock_hierarchy(root); | ||
4039 | list_del(&cgrp->sibling); | 4001 | list_del(&cgrp->sibling); |
4040 | cgroup_unlock_hierarchy(root); | ||
4041 | root->number_of_cgroups--; | 4002 | root->number_of_cgroups--; |
4042 | 4003 | ||
4043 | err_destroy: | 4004 | err_destroy: |
@@ -4245,10 +4206,8 @@ again: | |||
4245 | list_del_init(&cgrp->release_list); | 4206 | list_del_init(&cgrp->release_list); |
4246 | raw_spin_unlock(&release_list_lock); | 4207 | raw_spin_unlock(&release_list_lock); |
4247 | 4208 | ||
4248 | cgroup_lock_hierarchy(cgrp->root); | ||
4249 | /* delete this cgroup from parent->children */ | 4209 | /* delete this cgroup from parent->children */ |
4250 | list_del_init(&cgrp->sibling); | 4210 | list_del_init(&cgrp->sibling); |
4251 | cgroup_unlock_hierarchy(cgrp->root); | ||
4252 | 4211 | ||
4253 | list_del_init(&cgrp->allcg_node); | 4212 | list_del_init(&cgrp->allcg_node); |
4254 | 4213 | ||
@@ -4322,8 +4281,6 @@ static void __init cgroup_init_subsys(struct cgroup_subsys *ss) | |||
4322 | * need to invoke fork callbacks here. */ | 4281 | * need to invoke fork callbacks here. */ |
4323 | BUG_ON(!list_empty(&init_task.tasks)); | 4282 | BUG_ON(!list_empty(&init_task.tasks)); |
4324 | 4283 | ||
4325 | mutex_init(&ss->hierarchy_mutex); | ||
4326 | lockdep_set_class(&ss->hierarchy_mutex, &ss->subsys_key); | ||
4327 | ss->active = 1; | 4284 | ss->active = 1; |
4328 | 4285 | ||
4329 | /* this function shouldn't be used with modular subsystems, since they | 4286 | /* this function shouldn't be used with modular subsystems, since they |
@@ -4450,8 +4407,6 @@ int __init_or_module cgroup_load_subsys(struct cgroup_subsys *ss) | |||
4450 | } | 4407 | } |
4451 | write_unlock(&css_set_lock); | 4408 | write_unlock(&css_set_lock); |
4452 | 4409 | ||
4453 | mutex_init(&ss->hierarchy_mutex); | ||
4454 | lockdep_set_class(&ss->hierarchy_mutex, &ss->subsys_key); | ||
4455 | ss->active = 1; | 4410 | ss->active = 1; |
4456 | 4411 | ||
4457 | /* success! */ | 4412 | /* success! */ |