mm, memcg: prevent memory.high load/store tearing
authorChris Down <chris@chrisdown.name>
Thu, 2 Apr 2020 04:07:17 +0000 (21:07 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Thu, 2 Apr 2020 16:35:28 +0000 (09:35 -0700)
A mem_cgroup's high attribute can be concurrently set at the same time as
we are trying to read it -- for example, if we are in memory_high_write at
the same time as we are trying to do high reclaim.

Signed-off-by: Chris Down <chris@chrisdown.name>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Roman Gushchin <guro@fb.com>
Cc: Tejun Heo <tj@kernel.org>
Link: http://lkml.kernel.org/r/2f66f7038ed1d4688e59de72b627ae0ea52efa83.1584034301.git.chris@chrisdown.name
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/memcontrol.c

index badd043..b1bb624 100644 (file)
@@ -2242,7 +2242,7 @@ static void reclaim_high(struct mem_cgroup *memcg,
                         gfp_t gfp_mask)
 {
        do {
-               if (page_counter_read(&memcg->memory) <= memcg->high)
+               if (page_counter_read(&memcg->memory) <= READ_ONCE(memcg->high))
                        continue;
                memcg_memory_event(memcg, MEMCG_HIGH);
                try_to_free_mem_cgroup_pages(memcg, nr_pages, gfp_mask, true);
@@ -2582,7 +2582,7 @@ done_restock:
         * reclaim, the cost of mismatch is negligible.
         */
        do {
-               if (page_counter_read(&memcg->memory) > memcg->high) {
+               if (page_counter_read(&memcg->memory) > READ_ONCE(memcg->high)) {
                        /* Don't bother a random interrupted task */
                        if (in_interrupt()) {
                                schedule_work(&memcg->high_work);
@@ -4325,7 +4325,8 @@ void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages,
        *pheadroom = PAGE_COUNTER_MAX;
 
        while ((parent = parent_mem_cgroup(memcg))) {
-               unsigned long ceiling = min(memcg->memory.max, memcg->high);
+               unsigned long ceiling = min(memcg->memory.max,
+                                           READ_ONCE(memcg->high));
                unsigned long used = page_counter_read(&memcg->memory);
 
                *pheadroom = min(*pheadroom, ceiling - min(ceiling, used));
@@ -5047,7 +5048,7 @@ mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
        if (!memcg)
                return ERR_PTR(error);
 
-       memcg->high = PAGE_COUNTER_MAX;
+       WRITE_ONCE(memcg->high, PAGE_COUNTER_MAX);
        memcg->soft_limit = PAGE_COUNTER_MAX;
        if (parent) {
                memcg->swappiness = mem_cgroup_swappiness(parent);
@@ -5200,7 +5201,7 @@ static void mem_cgroup_css_reset(struct cgroup_subsys_state *css)
        page_counter_set_max(&memcg->tcpmem, PAGE_COUNTER_MAX);
        page_counter_set_min(&memcg->memory, 0);
        page_counter_set_low(&memcg->memory, 0);
-       memcg->high = PAGE_COUNTER_MAX;
+       WRITE_ONCE(memcg->high, PAGE_COUNTER_MAX);
        memcg->soft_limit = PAGE_COUNTER_MAX;
        memcg_wb_domain_size_changed(memcg);
 }
@@ -6016,7 +6017,7 @@ static ssize_t memory_high_write(struct kernfs_open_file *of,
        if (err)
                return err;
 
-       memcg->high = high;
+       WRITE_ONCE(memcg->high, high);
 
        for (;;) {
                unsigned long nr_pages = page_counter_read(&memcg->memory);