A cgroup containing only dying tasks will be seen as empty when a userspace process reads its cgroup.procs or cgroup.tasks files. It should be safe to delete such a cgroup as it is considered empty. However if one of the dying tasks did not reach cgroup_exit then an attempt to delete the cgroup will fail with EBUSY because cgroup_is_populated() will not consider it empty until all tasks reach cgroup_exit. Such a condition can be triggered when a task consumes large amounts of memory and spends enough time in exit_mm to create delay between the moment it is flagged as PF_EXITING and the moment it reaches cgroup_exit. Fix this by detecting cgroups containing only dying tasks during cgroup destruction and proceeding with it while postponing the final step of releasing the last reference until the last task reaches cgroup_exit.
Signed-off-by: Suren Baghdasaryan surenb@google.com Reported-by: JeiFeng Lee linger.lee@mediatek.com Fixes: c03cd7738a83 ("cgroup: Include dying leaders with live threads in PROCS iterations") --- include/linux/cgroup-defs.h | 3 ++ kernel/cgroup/cgroup.c | 65 +++++++++++++++++++++++++++++++++---- 2 files changed, 61 insertions(+), 7 deletions(-)
diff --git a/include/linux/cgroup-defs.h b/include/linux/cgroup-defs.h index 63097cb243cb..f9bcccbac8dd 100644 --- a/include/linux/cgroup-defs.h +++ b/include/linux/cgroup-defs.h @@ -71,6 +71,9 @@ enum {
/* Cgroup is frozen. */ CGRP_FROZEN, + + /* Cgroup is dead. */ + CGRP_DEAD, };
/* cgroup_root->flags */ diff --git a/kernel/cgroup/cgroup.c b/kernel/cgroup/cgroup.c index 735af8f15f95..a99ebddd37d9 100644 --- a/kernel/cgroup/cgroup.c +++ b/kernel/cgroup/cgroup.c @@ -795,10 +795,11 @@ static bool css_set_populated(struct css_set *cset) * that the content of the interface file has changed. This can be used to * detect when @cgrp and its descendants become populated or empty. */ -static void cgroup_update_populated(struct cgroup *cgrp, bool populated) +static bool cgroup_update_populated(struct cgroup *cgrp, bool populated) { struct cgroup *child = NULL; int adj = populated ? 1 : -1; + bool state_change = false;
lockdep_assert_held(&css_set_lock);
@@ -817,6 +818,7 @@ static void cgroup_update_populated(struct cgroup *cgrp, bool populated) if (was_populated == cgroup_is_populated(cgrp)) break;
+ state_change = true; cgroup1_check_for_release(cgrp); TRACE_CGROUP_PATH(notify_populated, cgrp, cgroup_is_populated(cgrp)); @@ -825,6 +827,21 @@ static void cgroup_update_populated(struct cgroup *cgrp, bool populated) child = cgrp; cgrp = cgroup_parent(cgrp); } while (cgrp); + + return state_change; +} + +static void cgroup_prune_dead(struct cgroup *cgrp) +{ + lockdep_assert_held(&css_set_lock); + + do { + /* put the base reference if cgroup was already destroyed */ + if (!cgroup_is_populated(cgrp) && + test_bit(CGRP_DEAD, &cgrp->flags)) + percpu_ref_kill(&cgrp->self.refcnt); + cgrp = cgroup_parent(cgrp); + } while (cgrp); }
/** @@ -838,11 +855,15 @@ static void cgroup_update_populated(struct cgroup *cgrp, bool populated) static void css_set_update_populated(struct css_set *cset, bool populated) { struct cgrp_cset_link *link; + bool state_change;
lockdep_assert_held(&css_set_lock);
- list_for_each_entry(link, &cset->cgrp_links, cgrp_link) - cgroup_update_populated(link->cgrp, populated); + list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { + state_change = cgroup_update_populated(link->cgrp, populated); + if (state_change && !populated) + cgroup_prune_dead(link->cgrp); + } }
/* @@ -5458,8 +5479,26 @@ static int cgroup_destroy_locked(struct cgroup *cgrp) * Only migration can raise populated from zero and we're already * holding cgroup_mutex. */ - if (cgroup_is_populated(cgrp)) - return -EBUSY; + if (cgroup_is_populated(cgrp)) { + struct css_task_iter it; + struct task_struct *task; + + /* + * cgroup_is_populated does not account for exiting tasks + * that did not reach cgroup_exit yet. Check if all the tasks + * in this cgroup are exiting. + */ + css_task_iter_start(&cgrp->self, 0, &it); + do { + task = css_task_iter_next(&it); + } while (task && (task->flags & PF_EXITING)); + css_task_iter_end(&it); + + if (task) { + /* cgroup is indeed populated */ + return -EBUSY; + } + }
/* * Make sure there's no live children. We can't test emptiness of @@ -5510,8 +5549,20 @@ static int cgroup_destroy_locked(struct cgroup *cgrp)
cgroup_bpf_offline(cgrp);
- /* put the base reference */ - percpu_ref_kill(&cgrp->self.refcnt); + /* + * Take css_set_lock because of the possible race with + * cgroup_update_populated. + */ + spin_lock_irq(&css_set_lock); + /* The last task might have died since we last checked */ + if (cgroup_is_populated(cgrp)) { + /* mark cgroup for future destruction */ + set_bit(CGRP_DEAD, &cgrp->flags); + } else { + /* put the base reference */ + percpu_ref_kill(&cgrp->self.refcnt); + } + spin_unlock_irq(&css_set_lock);
return 0; };
Add new test to verify that a cgroup with dead processes can be destroyed. The test spawns a child process which allocates and touches 100MB of RAM to ensure prolonged exit. Subsequently it kills the child, waits until the cgroup containing the child is empty and destroys the cgroup.
Signed-off-by: Suren Baghdasaryan surenb@google.com --- tools/testing/selftests/cgroup/test_core.c | 113 +++++++++++++++++++++ 1 file changed, 113 insertions(+)
diff --git a/tools/testing/selftests/cgroup/test_core.c b/tools/testing/selftests/cgroup/test_core.c index c5ca669feb2b..2a5242ec1a49 100644 --- a/tools/testing/selftests/cgroup/test_core.c +++ b/tools/testing/selftests/cgroup/test_core.c @@ -2,7 +2,10 @@
#include <linux/limits.h> #include <sys/types.h> +#include <sys/mman.h> +#include <sys/wait.h> #include <unistd.h> +#include <fcntl.h> #include <stdio.h> #include <errno.h> #include <signal.h> @@ -12,6 +15,115 @@ #include "../kselftest.h" #include "cgroup_util.h"
+static int touch_anon(char *buf, size_t size) +{ + int fd; + char *pos = buf; + + fd = open("/dev/urandom", O_RDONLY); + if (fd < 0) + return -1; + + while (size > 0) { + ssize_t ret = read(fd, pos, size); + + if (ret < 0) { + if (errno != EINTR) { + close(fd); + return -1; + } + } else { + pos += ret; + size -= ret; + } + } + close(fd); + + return 0; +} + +static int alloc_and_touch_anon_noexit(const char *cgroup, void *arg) +{ + int ppid = getppid(); + size_t size = (size_t)arg; + void *buf; + + buf = mmap(NULL, size, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANON, + 0, 0); + if (buf == MAP_FAILED) + return -1; + + if (touch_anon((char *)buf, size)) { + munmap(buf, size); + return -1; + } + + while (getppid() == ppid) + sleep(1); + + munmap(buf, size); + return 0; +} + +/* + * Create a child process that allocates and touches 100MB, then waits to be + * killed. Wait until the child is attached to the cgroup, kill all processes + * in that cgroup and wait until "cgroup.events" is empty. At this point try to + * destroy the empty cgroup. The test helps detect race conditions between + * dying processes leaving the cgroup and cgroup destruction path. + */ +static int test_cgcore_destroy(const char *root) +{ + int ret = KSFT_FAIL; + char *cg_test = NULL; + int child_pid; + char buf[PAGE_SIZE]; + + cg_test = cg_name(root, "cg_test"); + + if (!cg_test) + goto cleanup; + + for (int i = 0; i < 10; i++) { + if (cg_create(cg_test)) + goto cleanup; + + child_pid = cg_run_nowait(cg_test, alloc_and_touch_anon_noexit, + (void *) MB(100)); + + if (child_pid < 0) + goto cleanup; + + /* wait for the child to enter cgroup */ + if (cg_wait_for_proc_count(cg_test, 1)) + goto cleanup; + + if (cg_killall(cg_test)) + goto cleanup; + + /* wait for cgroup to be empty */ + while (1) { + if (cg_read(cg_test, "cgroup.procs", buf, sizeof(buf))) + goto cleanup; + if (buf[0] == '\0') + break; + usleep(1000); + } + + if (rmdir(cg_test)) + goto cleanup; + + if (waitpid(child_pid, NULL, 0) < 0) + goto cleanup; + } + ret = KSFT_PASS; +cleanup: + if (cg_test) + cg_destroy(cg_test); + free(cg_test); + return ret; +} + /* * A(0) - B(0) - C(1) * \ D(0) @@ -512,6 +624,7 @@ struct corecg_test { T(test_cgcore_populated), T(test_cgcore_proc_migration), T(test_cgcore_thread_migration), + T(test_cgcore_destroy), }; #undef T
Hi, I was looking into the issue and came up with an alternative solution that changes task iteration to be consistent with cgroup_is_populated() check and moving the responsibility to check PF_EXITING on the consumers of iterator API.
I haven't check your approach thoroughly, however, it appears to me it complicates (already non-trivial) cgroup destruction path. I ran your selftest on the iterators approach and it proved working.
Michal Koutný (2): cgroup: Unify css_set task lists cgroup: Iterate tasks that did not finish do_exit()
Suren Baghdasaryan (1): kselftest/cgroup: add cgroup destruction test
include/linux/cgroup-defs.h | 15 ++- include/linux/cgroup.h | 4 +- kernel/cgroup/cgroup.c | 86 ++++++++-------- kernel/cgroup/debug.c | 16 ++- tools/testing/selftests/cgroup/test_core.c | 113 +++++++++++++++++++++ 5 files changed, 176 insertions(+), 58 deletions(-)
We track tasks of css_set in three different lists. The iterators unnecessarily process each list specially. Use an array of lists and simplify the iterator code. This is refactoring with no intended functional change.
Signed-off-by: Michal Koutný mkoutny@suse.com --- include/linux/cgroup-defs.h | 15 ++++--- include/linux/cgroup.h | 4 +- kernel/cgroup/cgroup.c | 81 +++++++++++++++++++------------------ kernel/cgroup/debug.c | 16 ++++---- 4 files changed, 60 insertions(+), 56 deletions(-)
diff --git a/include/linux/cgroup-defs.h b/include/linux/cgroup-defs.h index 63097cb243cb..e474b5a67438 100644 --- a/include/linux/cgroup-defs.h +++ b/include/linux/cgroup-defs.h @@ -180,6 +180,13 @@ struct cgroup_subsys_state { struct cgroup_subsys_state *parent; };
+enum css_set_task_list { + CSS_SET_TASKS = 0, + CSS_SET_TASKS_MG, + CSS_SET_TASKS_DYING, + CSS_SET_TASKS_COUNT, +}; + /* * A css_set is a structure holding pointers to a set of * cgroup_subsys_state objects. This saves space in the task struct @@ -214,14 +221,12 @@ struct css_set {
/* * Lists running through all tasks using this cgroup group. - * mg_tasks lists tasks which belong to this cset but are in the + * CSS_SET_TASKS_MG lists tasks which belong to this cset but are in the * process of being migrated out or in. Protected by * css_set_rwsem, but, during migration, once tasks are moved to - * mg_tasks, it can be read safely while holding cgroup_mutex. + * CSS_SET_TASKS_MG, it can be read safely while holding cgroup_mutex. */ - struct list_head tasks; - struct list_head mg_tasks; - struct list_head dying_tasks; + struct list_head tasks[CSS_SET_TASKS_COUNT];
/* all css_task_iters currently walking this cset */ struct list_head task_iters; diff --git a/include/linux/cgroup.h b/include/linux/cgroup.h index d7ddebd0cdec..284e6b4b43cc 100644 --- a/include/linux/cgroup.h +++ b/include/linux/cgroup.h @@ -58,10 +58,8 @@ struct css_task_iter { struct list_head *tcset_head;
struct list_head *task_pos; - struct list_head *tasks_head; - struct list_head *mg_tasks_head; - struct list_head *dying_tasks_head;
+ enum css_set_task_list cur_list; struct css_set *cur_cset; struct css_set *cur_dcset; struct task_struct *cur_task; diff --git a/kernel/cgroup/cgroup.c b/kernel/cgroup/cgroup.c index 735af8f15f95..b56283e13491 100644 --- a/kernel/cgroup/cgroup.c +++ b/kernel/cgroup/cgroup.c @@ -737,9 +737,9 @@ EXPORT_SYMBOL_GPL(of_css); struct css_set init_css_set = { .refcount = REFCOUNT_INIT(1), .dom_cset = &init_css_set, - .tasks = LIST_HEAD_INIT(init_css_set.tasks), - .mg_tasks = LIST_HEAD_INIT(init_css_set.mg_tasks), - .dying_tasks = LIST_HEAD_INIT(init_css_set.dying_tasks), + .tasks[CSS_SET_TASKS] = LIST_HEAD_INIT(init_css_set.tasks[CSS_SET_TASKS]), + .tasks[CSS_SET_TASKS_MG] = LIST_HEAD_INIT(init_css_set.tasks[CSS_SET_TASKS_MG]), + .tasks[CSS_SET_TASKS_DYING] = LIST_HEAD_INIT(init_css_set.tasks[CSS_SET_TASKS_DYING]), .task_iters = LIST_HEAD_INIT(init_css_set.task_iters), .threaded_csets = LIST_HEAD_INIT(init_css_set.threaded_csets), .cgrp_links = LIST_HEAD_INIT(init_css_set.cgrp_links), @@ -775,7 +775,8 @@ static bool css_set_populated(struct css_set *cset) { lockdep_assert_held(&css_set_lock);
- return !list_empty(&cset->tasks) || !list_empty(&cset->mg_tasks); + return !list_empty(&cset->tasks[CSS_SET_TASKS]) || + !list_empty(&cset->tasks[CSS_SET_TASKS_MG]); }
/** @@ -865,7 +866,7 @@ static void css_set_skip_task_iters(struct css_set *cset, * @task: task being moved * @from_cset: css_set @task currently belongs to (may be NULL) * @to_cset: new css_set @task is being moved to (may be NULL) - * @use_mg_tasks: move to @to_cset->mg_tasks instead of ->tasks + * @use_mg_tasks: move to @to_cset->tasks[CSS_SET_TASKS_MG] instead of ->tasks[CSS_SET_TASKS] * * Move @task from @from_cset to @to_cset. If @task didn't belong to any * css_set, @from_cset can be NULL. If @task is being disassociated @@ -896,6 +897,8 @@ static void css_set_move_task(struct task_struct *task, }
if (to_cset) { + enum css_set_task_list ls = use_mg_tasks ? CSS_SET_TASKS_MG : CSS_SET_TASKS; + /* * We are synchronized through cgroup_threadgroup_rwsem * against PF_EXITING setting such that we can't race @@ -904,8 +907,7 @@ static void css_set_move_task(struct task_struct *task, WARN_ON_ONCE(task->flags & PF_EXITING);
cgroup_move_task(task, to_cset); - list_add_tail(&task->cg_list, use_mg_tasks ? &to_cset->mg_tasks : - &to_cset->tasks); + list_add_tail(&task->cg_list, &to_cset->tasks[ls]); } }
@@ -1211,9 +1213,9 @@ static struct css_set *find_css_set(struct css_set *old_cset,
refcount_set(&cset->refcount, 1); cset->dom_cset = cset; - INIT_LIST_HEAD(&cset->tasks); - INIT_LIST_HEAD(&cset->mg_tasks); - INIT_LIST_HEAD(&cset->dying_tasks); + INIT_LIST_HEAD(&cset->tasks[CSS_SET_TASKS]); + INIT_LIST_HEAD(&cset->tasks[CSS_SET_TASKS_MG]); + INIT_LIST_HEAD(&cset->tasks[CSS_SET_TASKS_DYING]); INIT_LIST_HEAD(&cset->task_iters); INIT_LIST_HEAD(&cset->threaded_csets); INIT_HLIST_NODE(&cset->hlist); @@ -2285,7 +2287,7 @@ EXPORT_SYMBOL_GPL(task_cgroup_path); * Add @task, which is a migration target, to @mgctx->tset. This function * becomes noop if @task doesn't need to be migrated. @task's css_set * should have been added as a migration source and @task->cg_list will be - * moved from the css_set's tasks list to mg_tasks one. + * moved from the css_set's tasks list to tasks[CSS_SET_TASKS_MG] one. */ static void cgroup_migrate_add_task(struct task_struct *task, struct cgroup_mgctx *mgctx) @@ -2307,7 +2309,7 @@ static void cgroup_migrate_add_task(struct task_struct *task,
mgctx->tset.nr_tasks++;
- list_move_tail(&task->cg_list, &cset->mg_tasks); + list_move_tail(&task->cg_list, &cset->tasks[CSS_SET_TASKS_MG]); if (list_empty(&cset->mg_node)) list_add_tail(&cset->mg_node, &mgctx->tset.src_csets); @@ -2348,12 +2350,12 @@ struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset,
while (&cset->mg_node != tset->csets) { if (!task) - task = list_first_entry(&cset->mg_tasks, + task = list_first_entry(&cset->tasks[CSS_SET_TASKS_MG], struct task_struct, cg_list); else task = list_next_entry(task, cg_list);
- if (&task->cg_list != &cset->mg_tasks) { + if (&task->cg_list != &cset->tasks[CSS_SET_TASKS_MG]) { tset->cur_cset = cset; tset->cur_task = task;
@@ -2416,7 +2418,7 @@ static int cgroup_migrate_execute(struct cgroup_mgctx *mgctx) */ spin_lock_irq(&css_set_lock); list_for_each_entry(cset, &tset->src_csets, mg_node) { - list_for_each_entry_safe(task, tmp_task, &cset->mg_tasks, cg_list) { + list_for_each_entry_safe(task, tmp_task, &cset->tasks[CSS_SET_TASKS_MG], cg_list) { struct css_set *from_cset = task_css_set(task); struct css_set *to_cset = cset->mg_dst_cset;
@@ -2470,7 +2472,7 @@ static int cgroup_migrate_execute(struct cgroup_mgctx *mgctx) spin_lock_irq(&css_set_lock); list_splice_init(&tset->dst_csets, &tset->src_csets); list_for_each_entry_safe(cset, tmp_cset, &tset->src_csets, mg_node) { - list_splice_tail_init(&cset->mg_tasks, &cset->tasks); + list_splice_tail_init(&cset->tasks[CSS_SET_TASKS_MG], &cset->tasks[CSS_SET_TASKS]); list_del_init(&cset->mg_node); } spin_unlock_irq(&css_set_lock); @@ -2592,7 +2594,7 @@ void cgroup_migrate_add_src(struct css_set *src_cset,
WARN_ON(src_cset->mg_src_cgrp); WARN_ON(src_cset->mg_dst_cgrp); - WARN_ON(!list_empty(&src_cset->mg_tasks)); + WARN_ON(!list_empty(&src_cset->tasks[CSS_SET_TASKS_MG])); WARN_ON(!list_empty(&src_cset->mg_node));
src_cset->mg_src_cgrp = src_cgrp; @@ -2905,7 +2907,7 @@ static int cgroup_update_dfl_csses(struct cgroup *cgrp) struct task_struct *task, *ntask;
/* all tasks in src_csets need to be migrated */ - list_for_each_entry_safe(task, ntask, &src_cset->tasks, cg_list) + list_for_each_entry_safe(task, ntask, &src_cset->tasks[CSS_SET_TASKS], cg_list) cgroup_migrate_add_task(task, &mgctx); } spin_unlock_irq(&css_set_lock); @@ -4402,18 +4404,18 @@ static void css_task_iter_advance_css_set(struct css_task_iter *it) it->task_pos = NULL; return; } - } while (!css_set_populated(cset) && list_empty(&cset->dying_tasks)); - - if (!list_empty(&cset->tasks)) - it->task_pos = cset->tasks.next; - else if (!list_empty(&cset->mg_tasks)) - it->task_pos = cset->mg_tasks.next; - else - it->task_pos = cset->dying_tasks.next; + } while (!css_set_populated(cset) && list_empty(&cset->tasks[CSS_SET_TASKS_DYING]));
- it->tasks_head = &cset->tasks; - it->mg_tasks_head = &cset->mg_tasks; - it->dying_tasks_head = &cset->dying_tasks; + it->cur_list = CSS_SET_TASKS; + while (list_empty(&cset->tasks[it->cur_list])) { + /* At least one list should be non-empty */ + if (WARN_ON(++it->cur_list == CSS_SET_TASKS_COUNT)) { + it->cur_list = CSS_SET_TASKS; + it->task_pos = NULL; + return; + } + } + it->task_pos = cset->tasks[it->cur_list].next;
/* * We don't keep css_sets locked across iteration steps and thus @@ -4458,21 +4460,22 @@ static void css_task_iter_advance(struct css_task_iter *it) repeat: if (it->task_pos) { /* - * Advance iterator to find next entry. cset->tasks is - * consumed first and then ->mg_tasks. After ->mg_tasks, - * we move onto the next cset. + * Advance iterator to find next entry. All lists in + * cset->tasks[] are consumed, then we move onto the next cset. */ if (it->flags & CSS_TASK_ITER_SKIPPED) it->flags &= ~CSS_TASK_ITER_SKIPPED; else it->task_pos = it->task_pos->next;
- if (it->task_pos == it->tasks_head) - it->task_pos = it->mg_tasks_head->next; - if (it->task_pos == it->mg_tasks_head) - it->task_pos = it->dying_tasks_head->next; - if (it->task_pos == it->dying_tasks_head) - css_task_iter_advance_css_set(it); + while (it->task_pos == &it->cur_cset->tasks[it->cur_list]) { + if (++it->cur_list == CSS_SET_TASKS_COUNT) { + css_task_iter_advance_css_set(it); + break; + } + + it->task_pos = it->cur_cset->tasks[it->cur_list].next; + } } else { /* called from start, proceed to the first cset */ css_task_iter_advance_css_set(it); @@ -5986,7 +5989,7 @@ void cgroup_exit(struct task_struct *tsk) WARN_ON_ONCE(list_empty(&tsk->cg_list)); cset = task_css_set(tsk); css_set_move_task(tsk, cset, NULL, false); - list_add_tail(&tsk->cg_list, &cset->dying_tasks); + list_add_tail(&tsk->cg_list, &cset->tasks[CSS_SET_TASKS_DYING]); cset->nr_tasks--;
WARN_ON_ONCE(cgroup_task_frozen(tsk)); diff --git a/kernel/cgroup/debug.c b/kernel/cgroup/debug.c index 80aa3f027ac3..26b4908600f7 100644 --- a/kernel/cgroup/debug.c +++ b/kernel/cgroup/debug.c @@ -124,6 +124,7 @@ static int cgroup_css_links_read(struct seq_file *seq, void *v) struct task_struct *task; int count = 0; int refcnt = refcount_read(&cset->refcount); + enum css_set_task_list ls;
/* * Print out the proc_cset and threaded_cset relationship @@ -161,17 +162,14 @@ static int cgroup_css_links_read(struct seq_file *seq, void *v) } seq_puts(seq, "\n");
- list_for_each_entry(task, &cset->tasks, cg_list) { - if (count++ <= MAX_TASKS_SHOWN_PER_CSS) - seq_printf(seq, " task %d\n", - task_pid_vnr(task)); + for (ls = CSS_SET_TASKS; ls < CSS_SET_TASKS_DYING; ++ls) { + list_for_each_entry(task, &cset->tasks[ls], cg_list) { + if (count++ <= MAX_TASKS_SHOWN_PER_CSS) + seq_printf(seq, " task %d\n", + task_pid_vnr(task)); + } }
- list_for_each_entry(task, &cset->mg_tasks, cg_list) { - if (count++ <= MAX_TASKS_SHOWN_PER_CSS) - seq_printf(seq, " task %d\n", - task_pid_vnr(task)); - } /* show # of overflowed tasks */ if (count > MAX_TASKS_SHOWN_PER_CSS) seq_printf(seq, " ... (%d)\n",
PF_EXITING is set earlier than actual removal from css_set when a task is exitting. This can confuse cgroup.procs readers who see no PF_EXITING tasks, however, rmdir is checking against css_set membership so it can transitionally fail with EBUSY.
Fix this by listing tasks that weren't unlinked from css_set active lists. It may happen that other users of the task iterator (without CSS_TASK_ITER_PROCS) spot a PF_EXITING task before cgroup_exit(). This is equal to the state before commit c03cd7738a83 ("cgroup: Include dying leaders with live threads in PROCS iterations") but it may be reviewed later.
Reported-by: Suren Baghdasaryan surenb@google.com Fixes: c03cd7738a83 ("cgroup: Include dying leaders with live threads in PROCS iterations") Signed-off-by: Michal Koutný mkoutny@suse.com --- kernel/cgroup/cgroup.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)
diff --git a/kernel/cgroup/cgroup.c b/kernel/cgroup/cgroup.c index b56283e13491..132d258e7172 100644 --- a/kernel/cgroup/cgroup.c +++ b/kernel/cgroup/cgroup.c @@ -4492,11 +4492,12 @@ static void css_task_iter_advance(struct css_task_iter *it) goto repeat;
/* and dying leaders w/o live member threads */ - if (!atomic_read(&task->signal->live)) + if (it->cur_list == CSS_SET_TASKS_DYING && + !atomic_read(&task->signal->live)) goto repeat; } else { /* skip all dying ones */ - if (task->flags & PF_EXITING) + if (it->cur_list == CSS_SET_TASKS_DYING) goto repeat; } }
From: Suren Baghdasaryan surenb@google.com
Add new test to verify that a cgroup with dead processes can be destroyed. The test spawns a child process which allocates and touches 100MB of RAM to ensure prolonged exit. Subsequently it kills the child, waits until the cgroup containing the child is empty and destroys the cgroup.
Signed-off-by: Suren Baghdasaryan surenb@google.com Signed-off-by: Michal Koutný mkoutny@suse.com --- tools/testing/selftests/cgroup/test_core.c | 113 +++++++++++++++++++++ 1 file changed, 113 insertions(+)
diff --git a/tools/testing/selftests/cgroup/test_core.c b/tools/testing/selftests/cgroup/test_core.c index c5ca669feb2b..2a5242ec1a49 100644 --- a/tools/testing/selftests/cgroup/test_core.c +++ b/tools/testing/selftests/cgroup/test_core.c @@ -2,7 +2,10 @@
#include <linux/limits.h> #include <sys/types.h> +#include <sys/mman.h> +#include <sys/wait.h> #include <unistd.h> +#include <fcntl.h> #include <stdio.h> #include <errno.h> #include <signal.h> @@ -12,6 +15,115 @@ #include "../kselftest.h" #include "cgroup_util.h"
+static int touch_anon(char *buf, size_t size) +{ + int fd; + char *pos = buf; + + fd = open("/dev/urandom", O_RDONLY); + if (fd < 0) + return -1; + + while (size > 0) { + ssize_t ret = read(fd, pos, size); + + if (ret < 0) { + if (errno != EINTR) { + close(fd); + return -1; + } + } else { + pos += ret; + size -= ret; + } + } + close(fd); + + return 0; +} + +static int alloc_and_touch_anon_noexit(const char *cgroup, void *arg) +{ + int ppid = getppid(); + size_t size = (size_t)arg; + void *buf; + + buf = mmap(NULL, size, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANON, + 0, 0); + if (buf == MAP_FAILED) + return -1; + + if (touch_anon((char *)buf, size)) { + munmap(buf, size); + return -1; + } + + while (getppid() == ppid) + sleep(1); + + munmap(buf, size); + return 0; +} + +/* + * Create a child process that allocates and touches 100MB, then waits to be + * killed. Wait until the child is attached to the cgroup, kill all processes + * in that cgroup and wait until "cgroup.events" is empty. At this point try to + * destroy the empty cgroup. The test helps detect race conditions between + * dying processes leaving the cgroup and cgroup destruction path. + */ +static int test_cgcore_destroy(const char *root) +{ + int ret = KSFT_FAIL; + char *cg_test = NULL; + int child_pid; + char buf[PAGE_SIZE]; + + cg_test = cg_name(root, "cg_test"); + + if (!cg_test) + goto cleanup; + + for (int i = 0; i < 10; i++) { + if (cg_create(cg_test)) + goto cleanup; + + child_pid = cg_run_nowait(cg_test, alloc_and_touch_anon_noexit, + (void *) MB(100)); + + if (child_pid < 0) + goto cleanup; + + /* wait for the child to enter cgroup */ + if (cg_wait_for_proc_count(cg_test, 1)) + goto cleanup; + + if (cg_killall(cg_test)) + goto cleanup; + + /* wait for cgroup to be empty */ + while (1) { + if (cg_read(cg_test, "cgroup.procs", buf, sizeof(buf))) + goto cleanup; + if (buf[0] == '\0') + break; + usleep(1000); + } + + if (rmdir(cg_test)) + goto cleanup; + + if (waitpid(child_pid, NULL, 0) < 0) + goto cleanup; + } + ret = KSFT_PASS; +cleanup: + if (cg_test) + cg_destroy(cg_test); + free(cg_test); + return ret; +} + /* * A(0) - B(0) - C(1) * \ D(0) @@ -512,6 +624,7 @@ struct corecg_test { T(test_cgcore_populated), T(test_cgcore_proc_migration), T(test_cgcore_thread_migration), + T(test_cgcore_destroy), }; #undef T
On Fri, Jan 17, 2020 at 04:15:31PM +0100, Michal Koutný wrote:
We track tasks of css_set in three different lists. The iterators unnecessarily process each list specially. Use an array of lists and simplify the iterator code. This is refactoring with no intended functional change.
Signed-off-by: Michal Koutný mkoutny@suse.com
include/linux/cgroup-defs.h | 15 ++++--- include/linux/cgroup.h | 4 +- kernel/cgroup/cgroup.c | 81 +++++++++++++++++++------------------ kernel/cgroup/debug.c | 16 ++++---- 4 files changed, 60 insertions(+), 56 deletions(-)
So, I get the urge to move the lists into an array and thought about doing that while adding the third list for sure. However, it does make code paths which don't walk all lists wordier and the code path is already tricky like hell. Given that there aren't that many places which loop over the lists, if you really wanna clean it up, maybe add an interator over the lists so that both parts of code can look lean?
Thanks.
On Fri, Jan 17, 2020 at 04:15:32PM +0100, Michal Koutný wrote:
PF_EXITING is set earlier than actual removal from css_set when a task is exitting. This can confuse cgroup.procs readers who see no PF_EXITING tasks, however, rmdir is checking against css_set membership so it can transitionally fail with EBUSY.
Fix this by listing tasks that weren't unlinked from css_set active lists. It may happen that other users of the task iterator (without CSS_TASK_ITER_PROCS) spot a PF_EXITING task before cgroup_exit(). This is equal to the state before commit c03cd7738a83 ("cgroup: Include dying leaders with live threads in PROCS iterations") but it may be reviewed later.
Yeah, this looks fine to me. Any chance you can order this before the clean up so that we can mark it for -stable.
Thanks.
Hi Michal,
On Fri, Jan 17, 2020 at 7:15 AM Michal Koutný mkoutny@suse.com wrote:
Hi, I was looking into the issue and came up with an alternative solution that changes task iteration to be consistent with cgroup_is_populated() check and moving the responsibility to check PF_EXITING on the consumers of iterator API.
Yeah, that was my first thought which basically reverts a part of c03cd7738a83. When I first brought up this issue in the other email thread, Tejun's comment was "the right thing to do is allowing destruction of cgroups w/ only dead processes in it". I assumed, maybe incorrectly, that the desire here is not to include dying processes into cgroup.procs but to allow cgroups with dying processes to be deleted.
To be clear, either way is fine with me since both ways solve the issue and this way the code is definitely simpler. I'll rerun the tests with your patches just to confirm the issue is gone. Thanks!
I haven't check your approach thoroughly, however, it appears to me it complicates (already non-trivial) cgroup destruction path. I ran your selftest on the iterators approach and it proved working.
Michal Koutný (2): cgroup: Unify css_set task lists cgroup: Iterate tasks that did not finish do_exit()
Suren Baghdasaryan (1): kselftest/cgroup: add cgroup destruction test
include/linux/cgroup-defs.h | 15 ++- include/linux/cgroup.h | 4 +- kernel/cgroup/cgroup.c | 86 ++++++++-------- kernel/cgroup/debug.c | 16 ++- tools/testing/selftests/cgroup/test_core.c | 113 +++++++++++++++++++++ 5 files changed, 176 insertions(+), 58 deletions(-)
-- 2.24.1
-- To unsubscribe from this group and stop receiving emails from it, send an email to kernel-team+unsubscribe@android.com.
On Fri, Jan 17, 2020 at 9:28 AM Tejun Heo tj@kernel.org wrote:
On Fri, Jan 17, 2020 at 04:15:32PM +0100, Michal Koutný wrote:
PF_EXITING is set earlier than actual removal from css_set when a task is exitting. This can confuse cgroup.procs readers who see no PF_EXITING tasks, however, rmdir is checking against css_set membership so it can transitionally fail with EBUSY.
Fix this by listing tasks that weren't unlinked from css_set active lists. It may happen that other users of the task iterator (without CSS_TASK_ITER_PROCS) spot a PF_EXITING task before cgroup_exit(). This is equal to the state before commit c03cd7738a83 ("cgroup: Include dying leaders with live threads in PROCS iterations") but it may be reviewed later.
Tested-by: Suren Baghdasaryan surenb@google.com
Yeah, this looks fine to me. Any chance you can order this before the clean up so that we can mark it for -stable.
+1 for reordering. Makes it easier to backport. Thanks, Suren.
Thanks.
-- tejun
-- To unsubscribe from this group and stop receiving emails from it, send an email to kernel-team+unsubscribe@android.com.
On Fri, Jan 17, 2020 at 10:41:29AM -0800, Suren Baghdasaryan surenb@google.com wrote:
Tested-by: Suren Baghdasaryan surenb@google.com
Thanks.
Yeah, this looks fine to me. Any chance you can order this before the clean up so that we can mark it for -stable.
+1 for reordering. Makes it easier to backport.
The grounds still need to be prepared for css_task_iter to store additional information. Let me see how the preceding changes can be minimized.
Michal
Hi, thanks for your feedback. I'm sending updated solution that changes task iteration to be consistent with cgroup_is_populated() check and moving the responsibility to check PF_EXITING on the consumers of iterator API.
Changes from v1 - v1: https://lore.kernel.org/lkml/20200117151533.12381-1-mkoutny@suse.com/ - Swap the fixing and refactoring patch (in order to make the fix easier for backport) - Don't introduce tasks lists array because of unnecessarily long access code - Fix comment in selftest - I leave the CC:stable on discretion of the maintainer
Michal Koutný (2): cgroup: Iterate tasks that did not finish do_exit() cgroup: Clean up css_set task traversal
Suren Baghdasaryan (1): kselftest/cgroup: add cgroup destruction test
include/linux/cgroup.h | 4 +- kernel/cgroup/cgroup.c | 60 ++++++----- tools/testing/selftests/cgroup/test_core.c | 113 +++++++++++++++++++++ 3 files changed, 146 insertions(+), 31 deletions(-)
PF_EXITING is set earlier than actual removal from css_set when a task is exitting. This can confuse cgroup.procs readers who see no PF_EXITING tasks, however, rmdir is checking against css_set membership so it can transitionally fail with EBUSY.
Fix this by listing tasks that weren't unlinked from css_set active lists. It may happen that other users of the task iterator (without CSS_TASK_ITER_PROCS) spot a PF_EXITING task before cgroup_exit(). This is equal to the state before commit c03cd7738a83 ("cgroup: Include dying leaders with live threads in PROCS iterations") but it may be reviewed later.
Reported-by: Suren Baghdasaryan surenb@google.com Fixes: c03cd7738a83 ("cgroup: Include dying leaders with live threads in PROCS iterations") Signed-off-by: Michal Koutný mkoutny@suse.com --- include/linux/cgroup.h | 1 + kernel/cgroup/cgroup.c | 23 ++++++++++++++++------- 2 files changed, 17 insertions(+), 7 deletions(-)
diff --git a/include/linux/cgroup.h b/include/linux/cgroup.h index d7ddebd0cdec..e75d2191226b 100644 --- a/include/linux/cgroup.h +++ b/include/linux/cgroup.h @@ -62,6 +62,7 @@ struct css_task_iter { struct list_head *mg_tasks_head; struct list_head *dying_tasks_head;
+ struct list_head *cur_tasks_head; struct css_set *cur_cset; struct css_set *cur_dcset; struct task_struct *cur_task; diff --git a/kernel/cgroup/cgroup.c b/kernel/cgroup/cgroup.c index 735af8f15f95..a6e3619e013b 100644 --- a/kernel/cgroup/cgroup.c +++ b/kernel/cgroup/cgroup.c @@ -4404,12 +4404,16 @@ static void css_task_iter_advance_css_set(struct css_task_iter *it) } } while (!css_set_populated(cset) && list_empty(&cset->dying_tasks));
- if (!list_empty(&cset->tasks)) + if (!list_empty(&cset->tasks)) { it->task_pos = cset->tasks.next; - else if (!list_empty(&cset->mg_tasks)) + it->cur_tasks_head = &cset->tasks; + } else if (!list_empty(&cset->mg_tasks)) { it->task_pos = cset->mg_tasks.next; - else + it->cur_tasks_head = &cset->mg_tasks; + } else { it->task_pos = cset->dying_tasks.next; + it->cur_tasks_head = &cset->dying_tasks; + }
it->tasks_head = &cset->tasks; it->mg_tasks_head = &cset->mg_tasks; @@ -4467,10 +4471,14 @@ static void css_task_iter_advance(struct css_task_iter *it) else it->task_pos = it->task_pos->next;
- if (it->task_pos == it->tasks_head) + if (it->task_pos == it->tasks_head) { it->task_pos = it->mg_tasks_head->next; - if (it->task_pos == it->mg_tasks_head) + it->cur_tasks_head = it->mg_tasks_head; + } + if (it->task_pos == it->mg_tasks_head) { it->task_pos = it->dying_tasks_head->next; + it->cur_tasks_head = it->dying_tasks_head; + } if (it->task_pos == it->dying_tasks_head) css_task_iter_advance_css_set(it); } else { @@ -4489,11 +4497,12 @@ static void css_task_iter_advance(struct css_task_iter *it) goto repeat;
/* and dying leaders w/o live member threads */ - if (!atomic_read(&task->signal->live)) + if (it->cur_tasks_head == it->dying_tasks_head && + !atomic_read(&task->signal->live)) goto repeat; } else { /* skip all dying ones */ - if (task->flags & PF_EXITING) + if (it->cur_tasks_head == it->dying_tasks_head) goto repeat; } }
css_task_iter stores pointer to head of each iterable list, this dates back to commit 0f0a2b4fa621 ("cgroup: reorganize css_task_iter") when we did not store cur_cset. Let us utilize list heads directly in cur_cset and streamline css_task_iter_advance_css_set a bit. This is no intentional function change.
Signed-off-by: Michal Koutný mkoutny@suse.com --- include/linux/cgroup.h | 3 --- kernel/cgroup/cgroup.c | 61 +++++++++++++++++++----------------------- 2 files changed, 28 insertions(+), 36 deletions(-)
diff --git a/include/linux/cgroup.h b/include/linux/cgroup.h index e75d2191226b..f1219b927817 100644 --- a/include/linux/cgroup.h +++ b/include/linux/cgroup.h @@ -58,9 +58,6 @@ struct css_task_iter { struct list_head *tcset_head;
struct list_head *task_pos; - struct list_head *tasks_head; - struct list_head *mg_tasks_head; - struct list_head *dying_tasks_head;
struct list_head *cur_tasks_head; struct css_set *cur_cset; diff --git a/kernel/cgroup/cgroup.c b/kernel/cgroup/cgroup.c index a6e3619e013b..14e0e360a2b4 100644 --- a/kernel/cgroup/cgroup.c +++ b/kernel/cgroup/cgroup.c @@ -4395,29 +4395,24 @@ static void css_task_iter_advance_css_set(struct css_task_iter *it)
lockdep_assert_held(&css_set_lock);
- /* Advance to the next non-empty css_set */ - do { - cset = css_task_iter_next_css_set(it); - if (!cset) { - it->task_pos = NULL; - return; + /* Advance to the next non-empty css_set and find first non-empty tasks list*/ + while ((cset = css_task_iter_next_css_set(it))) { + if (!list_empty(&cset->tasks)) { + it->cur_tasks_head = &cset->tasks; + break; + } else if (!list_empty(&cset->mg_tasks)) { + it->cur_tasks_head = &cset->mg_tasks; + break; + } else if (!list_empty(&cset->dying_tasks)) { + it->cur_tasks_head = &cset->dying_tasks; + break; } - } while (!css_set_populated(cset) && list_empty(&cset->dying_tasks)); - - if (!list_empty(&cset->tasks)) { - it->task_pos = cset->tasks.next; - it->cur_tasks_head = &cset->tasks; - } else if (!list_empty(&cset->mg_tasks)) { - it->task_pos = cset->mg_tasks.next; - it->cur_tasks_head = &cset->mg_tasks; - } else { - it->task_pos = cset->dying_tasks.next; - it->cur_tasks_head = &cset->dying_tasks; } - - it->tasks_head = &cset->tasks; - it->mg_tasks_head = &cset->mg_tasks; - it->dying_tasks_head = &cset->dying_tasks; + if (!cset) { + it->task_pos = NULL; + return; + } + it->task_pos = it->cur_tasks_head->next;
/* * We don't keep css_sets locked across iteration steps and thus @@ -4462,24 +4457,24 @@ static void css_task_iter_advance(struct css_task_iter *it) repeat: if (it->task_pos) { /* - * Advance iterator to find next entry. cset->tasks is - * consumed first and then ->mg_tasks. After ->mg_tasks, - * we move onto the next cset. + * Advance iterator to find next entry. We go through cset + * tasks, mg_tasks and dying_tasks, when consumed we move onto + * the next cset. */ if (it->flags & CSS_TASK_ITER_SKIPPED) it->flags &= ~CSS_TASK_ITER_SKIPPED; else it->task_pos = it->task_pos->next;
- if (it->task_pos == it->tasks_head) { - it->task_pos = it->mg_tasks_head->next; - it->cur_tasks_head = it->mg_tasks_head; + if (it->task_pos == &it->cur_cset->tasks) { + it->cur_tasks_head = &it->cur_cset->mg_tasks; + it->task_pos = it->cur_tasks_head->next; } - if (it->task_pos == it->mg_tasks_head) { - it->task_pos = it->dying_tasks_head->next; - it->cur_tasks_head = it->dying_tasks_head; + if (it->task_pos == &it->cur_cset->mg_tasks) { + it->cur_tasks_head = &it->cur_cset->dying_tasks; + it->task_pos = it->cur_tasks_head->next; } - if (it->task_pos == it->dying_tasks_head) + if (it->task_pos == &it->cur_cset->dying_tasks) css_task_iter_advance_css_set(it); } else { /* called from start, proceed to the first cset */ @@ -4497,12 +4492,12 @@ static void css_task_iter_advance(struct css_task_iter *it) goto repeat;
/* and dying leaders w/o live member threads */ - if (it->cur_tasks_head == it->dying_tasks_head && + if (it->cur_tasks_head == &it->cur_cset->dying_tasks && !atomic_read(&task->signal->live)) goto repeat; } else { /* skip all dying ones */ - if (it->cur_tasks_head == it->dying_tasks_head) + if (it->cur_tasks_head == &it->cur_cset->dying_tasks) goto repeat; } }
From: Suren Baghdasaryan surenb@google.com
Add new test to verify that a cgroup with dead processes can be destroyed. The test spawns a child process which allocates and touches 100MB of RAM to ensure prolonged exit. Subsequently it kills the child, waits until the cgroup containing the child is empty and destroys the cgroup.
Signed-off-by: Suren Baghdasaryan surenb@google.com [mkoutny@suse.com: Fix typo in test_cgcore_destroy comment] Acked-by: Michal Koutný mkoutny@suse.com Signed-off-by: Michal Koutný mkoutny@suse.com --- tools/testing/selftests/cgroup/test_core.c | 113 +++++++++++++++++++++ 1 file changed, 113 insertions(+)
diff --git a/tools/testing/selftests/cgroup/test_core.c b/tools/testing/selftests/cgroup/test_core.c index c5ca669feb2b..76c9dd578ba5 100644 --- a/tools/testing/selftests/cgroup/test_core.c +++ b/tools/testing/selftests/cgroup/test_core.c @@ -2,7 +2,10 @@
#include <linux/limits.h> #include <sys/types.h> +#include <sys/mman.h> +#include <sys/wait.h> #include <unistd.h> +#include <fcntl.h> #include <stdio.h> #include <errno.h> #include <signal.h> @@ -12,6 +15,115 @@ #include "../kselftest.h" #include "cgroup_util.h"
+static int touch_anon(char *buf, size_t size) +{ + int fd; + char *pos = buf; + + fd = open("/dev/urandom", O_RDONLY); + if (fd < 0) + return -1; + + while (size > 0) { + ssize_t ret = read(fd, pos, size); + + if (ret < 0) { + if (errno != EINTR) { + close(fd); + return -1; + } + } else { + pos += ret; + size -= ret; + } + } + close(fd); + + return 0; +} + +static int alloc_and_touch_anon_noexit(const char *cgroup, void *arg) +{ + int ppid = getppid(); + size_t size = (size_t)arg; + void *buf; + + buf = mmap(NULL, size, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANON, + 0, 0); + if (buf == MAP_FAILED) + return -1; + + if (touch_anon((char *)buf, size)) { + munmap(buf, size); + return -1; + } + + while (getppid() == ppid) + sleep(1); + + munmap(buf, size); + return 0; +} + +/* + * Create a child process that allocates and touches 100MB, then waits to be + * killed. Wait until the child is attached to the cgroup, kill all processes + * in that cgroup and wait until "cgroup.procs" is empty. At this point try to + * destroy the empty cgroup. The test helps detect race conditions between + * dying processes leaving the cgroup and cgroup destruction path. + */ +static int test_cgcore_destroy(const char *root) +{ + int ret = KSFT_FAIL; + char *cg_test = NULL; + int child_pid; + char buf[PAGE_SIZE]; + + cg_test = cg_name(root, "cg_test"); + + if (!cg_test) + goto cleanup; + + for (int i = 0; i < 10; i++) { + if (cg_create(cg_test)) + goto cleanup; + + child_pid = cg_run_nowait(cg_test, alloc_and_touch_anon_noexit, + (void *) MB(100)); + + if (child_pid < 0) + goto cleanup; + + /* wait for the child to enter cgroup */ + if (cg_wait_for_proc_count(cg_test, 1)) + goto cleanup; + + if (cg_killall(cg_test)) + goto cleanup; + + /* wait for cgroup to be empty */ + while (1) { + if (cg_read(cg_test, "cgroup.procs", buf, sizeof(buf))) + goto cleanup; + if (buf[0] == '\0') + break; + usleep(1000); + } + + if (rmdir(cg_test)) + goto cleanup; + + if (waitpid(child_pid, NULL, 0) < 0) + goto cleanup; + } + ret = KSFT_PASS; +cleanup: + if (cg_test) + cg_destroy(cg_test); + free(cg_test); + return ret; +} + /* * A(0) - B(0) - C(1) * \ D(0) @@ -512,6 +624,7 @@ struct corecg_test { T(test_cgcore_populated), T(test_cgcore_proc_migration), T(test_cgcore_thread_migration), + T(test_cgcore_destroy), }; #undef T
On Fri, Jan 24, 2020 at 3:40 AM Michal Koutný mkoutny@suse.com wrote:
PF_EXITING is set earlier than actual removal from css_set when a task is exitting. This can confuse cgroup.procs readers who see no PF_EXITING tasks, however, rmdir is checking against css_set membership so it can transitionally fail with EBUSY.
Fix this by listing tasks that weren't unlinked from css_set active lists. It may happen that other users of the task iterator (without CSS_TASK_ITER_PROCS) spot a PF_EXITING task before cgroup_exit(). This is equal to the state before commit c03cd7738a83 ("cgroup: Include dying leaders with live threads in PROCS iterations") but it may be reviewed later.
Reported-by: Suren Baghdasaryan surenb@google.com Fixes: c03cd7738a83 ("cgroup: Include dying leaders with live threads in PROCS iterations") Signed-off-by: Michal Koutný mkoutny@suse.com
include/linux/cgroup.h | 1 + kernel/cgroup/cgroup.c | 23 ++++++++++++++++------- 2 files changed, 17 insertions(+), 7 deletions(-)
diff --git a/include/linux/cgroup.h b/include/linux/cgroup.h index d7ddebd0cdec..e75d2191226b 100644 --- a/include/linux/cgroup.h +++ b/include/linux/cgroup.h @@ -62,6 +62,7 @@ struct css_task_iter { struct list_head *mg_tasks_head; struct list_head *dying_tasks_head;
struct list_head *cur_tasks_head; struct css_set *cur_cset; struct css_set *cur_dcset; struct task_struct *cur_task;diff --git a/kernel/cgroup/cgroup.c b/kernel/cgroup/cgroup.c index 735af8f15f95..a6e3619e013b 100644 --- a/kernel/cgroup/cgroup.c +++ b/kernel/cgroup/cgroup.c @@ -4404,12 +4404,16 @@ static void css_task_iter_advance_css_set(struct css_task_iter *it) } } while (!css_set_populated(cset) && list_empty(&cset->dying_tasks));
if (!list_empty(&cset->tasks))
if (!list_empty(&cset->tasks)) { it->task_pos = cset->tasks.next;
else if (!list_empty(&cset->mg_tasks))
it->cur_tasks_head = &cset->tasks;} else if (!list_empty(&cset->mg_tasks)) { it->task_pos = cset->mg_tasks.next;
else
it->cur_tasks_head = &cset->mg_tasks;} else { it->task_pos = cset->dying_tasks.next;it->cur_tasks_head = &cset->dying_tasks;} it->tasks_head = &cset->tasks; it->mg_tasks_head = &cset->mg_tasks;@@ -4467,10 +4471,14 @@ static void css_task_iter_advance(struct css_task_iter *it) else it->task_pos = it->task_pos->next;
if (it->task_pos == it->tasks_head)
if (it->task_pos == it->tasks_head) { it->task_pos = it->mg_tasks_head->next;
if (it->task_pos == it->mg_tasks_head)
it->cur_tasks_head = it->mg_tasks_head;}if (it->task_pos == it->mg_tasks_head) { it->task_pos = it->dying_tasks_head->next;it->cur_tasks_head = it->dying_tasks_head;} if (it->task_pos == it->dying_tasks_head) css_task_iter_advance_css_set(it); } else {@@ -4489,11 +4497,12 @@ static void css_task_iter_advance(struct css_task_iter *it) goto repeat;
/* and dying leaders w/o live member threads */
if (!atomic_read(&task->signal->live))
if (it->cur_tasks_head == it->dying_tasks_head &&!atomic_read(&task->signal->live)) goto repeat; } else { /* skip all dying ones */
if (task->flags & PF_EXITING)
if (it->cur_tasks_head == it->dying_tasks_head) goto repeat; }}
2.24.1
Tested-by: Suren Baghdasaryan surenb@google.com
Thanks!
On Fri, Jan 24, 2020 at 2:56 PM Suren Baghdasaryan surenb@google.com wrote:
On Fri, Jan 24, 2020 at 3:40 AM Michal Koutný mkoutny@suse.com wrote:
PF_EXITING is set earlier than actual removal from css_set when a task is exitting. This can confuse cgroup.procs readers who see no PF_EXITING tasks, however, rmdir is checking against css_set membership so it can transitionally fail with EBUSY.
Fix this by listing tasks that weren't unlinked from css_set active lists. It may happen that other users of the task iterator (without CSS_TASK_ITER_PROCS) spot a PF_EXITING task before cgroup_exit(). This is equal to the state before commit c03cd7738a83 ("cgroup: Include dying leaders with live threads in PROCS iterations") but it may be reviewed later.
Reported-by: Suren Baghdasaryan surenb@google.com Fixes: c03cd7738a83 ("cgroup: Include dying leaders with live threads in PROCS iterations") Signed-off-by: Michal Koutný mkoutny@suse.com
include/linux/cgroup.h | 1 + kernel/cgroup/cgroup.c | 23 ++++++++++++++++------- 2 files changed, 17 insertions(+), 7 deletions(-)
diff --git a/include/linux/cgroup.h b/include/linux/cgroup.h index d7ddebd0cdec..e75d2191226b 100644 --- a/include/linux/cgroup.h +++ b/include/linux/cgroup.h @@ -62,6 +62,7 @@ struct css_task_iter { struct list_head *mg_tasks_head; struct list_head *dying_tasks_head;
struct list_head *cur_tasks_head; struct css_set *cur_cset; struct css_set *cur_dcset; struct task_struct *cur_task;diff --git a/kernel/cgroup/cgroup.c b/kernel/cgroup/cgroup.c index 735af8f15f95..a6e3619e013b 100644 --- a/kernel/cgroup/cgroup.c +++ b/kernel/cgroup/cgroup.c @@ -4404,12 +4404,16 @@ static void css_task_iter_advance_css_set(struct css_task_iter *it) } } while (!css_set_populated(cset) && list_empty(&cset->dying_tasks));
if (!list_empty(&cset->tasks))
if (!list_empty(&cset->tasks)) { it->task_pos = cset->tasks.next;
else if (!list_empty(&cset->mg_tasks))
it->cur_tasks_head = &cset->tasks;} else if (!list_empty(&cset->mg_tasks)) { it->task_pos = cset->mg_tasks.next;
else
it->cur_tasks_head = &cset->mg_tasks;} else { it->task_pos = cset->dying_tasks.next;it->cur_tasks_head = &cset->dying_tasks;} it->tasks_head = &cset->tasks; it->mg_tasks_head = &cset->mg_tasks;@@ -4467,10 +4471,14 @@ static void css_task_iter_advance(struct css_task_iter *it) else it->task_pos = it->task_pos->next;
if (it->task_pos == it->tasks_head)
if (it->task_pos == it->tasks_head) { it->task_pos = it->mg_tasks_head->next;
if (it->task_pos == it->mg_tasks_head)
it->cur_tasks_head = it->mg_tasks_head;}if (it->task_pos == it->mg_tasks_head) { it->task_pos = it->dying_tasks_head->next;it->cur_tasks_head = it->dying_tasks_head;} if (it->task_pos == it->dying_tasks_head) css_task_iter_advance_css_set(it); } else {@@ -4489,11 +4497,12 @@ static void css_task_iter_advance(struct css_task_iter *it) goto repeat;
/* and dying leaders w/o live member threads */
if (!atomic_read(&task->signal->live))
if (it->cur_tasks_head == it->dying_tasks_head &&!atomic_read(&task->signal->live)) goto repeat; } else { /* skip all dying ones */
if (task->flags & PF_EXITING)
if (it->cur_tasks_head == it->dying_tasks_head) goto repeat; }}
2.24.1
Tested-by: Suren Baghdasaryan surenb@google.com
Thanks!
Hi Folks, If this new version looks good could we get an Ack please? I need to start backporting this fix to Android and would like a confirmation before doing so. Thanks!
On Fri, Jan 24, 2020 at 12:40:15PM +0100, Michal Koutný wrote:
PF_EXITING is set earlier than actual removal from css_set when a task is exitting. This can confuse cgroup.procs readers who see no PF_EXITING tasks, however, rmdir is checking against css_set membership so it can transitionally fail with EBUSY.
Fix this by listing tasks that weren't unlinked from css_set active lists. It may happen that other users of the task iterator (without CSS_TASK_ITER_PROCS) spot a PF_EXITING task before cgroup_exit(). This is equal to the state before commit c03cd7738a83 ("cgroup: Include dying leaders with live threads in PROCS iterations") but it may be reviewed later.
Reported-by: Suren Baghdasaryan surenb@google.com Fixes: c03cd7738a83 ("cgroup: Include dying leaders with live threads in PROCS iterations") Signed-off-by: Michal Koutný mkoutny@suse.com
Applied to cgroup/for-5.6-fixes.
Thanks.
On Fri, Jan 24, 2020 at 12:40:17PM +0100, Michal Koutný wrote:
From: Suren Baghdasaryan surenb@google.com
Add new test to verify that a cgroup with dead processes can be destroyed. The test spawns a child process which allocates and touches 100MB of RAM to ensure prolonged exit. Subsequently it kills the child, waits until the cgroup containing the child is empty and destroys the cgroup.
Signed-off-by: Suren Baghdasaryan surenb@google.com [mkoutny@suse.com: Fix typo in test_cgcore_destroy comment] Acked-by: Michal Koutný mkoutny@suse.com Signed-off-by: Michal Koutný mkoutny@suse.com
Applied 2-3 to cgroup/for-5.7.
Thanks a lot for working on this!
linux-kselftest-mirror@lists.linaro.org
-
Michal Koutný -
Suren Baghdasaryan -
Tejun Heo