mm/vmscan.c: fix potential deadlock in reclaim_pages()
Subsystems (3)
PATCH 1/1 mm/vmscan.c: fix potential deadlock in reclaim_pages()
Use memalloc_noreclaim_save()/memalloc_noreclaim_restore() in reclaim_pages() to prevent the page reclaim from going into the block I/O layer recursively and deadlock.
--- mm/vmscan.c | 15 +++++++++++++++ 1 file changed, 15 insertions(+)
mm/vmscan.c +15 -0
@@ -1701,6 +1701,7 @@ unsigned int reclaim_clean_pages_from_list(struct zone *zone,unsigned int nr_reclaimed;struct page *page, *next;LIST_HEAD (clean_pages); +unsigned int noreclaim_flag;list_for_each_entry_safe (page, next, page_list, lru) { if (!PageHuge(page) && page_is_file_lru(page) &&
@@ -1711,8 +1712,17 @@ unsigned int reclaim_clean_pages_from_list(struct zone *zone, } } +/* +* We should be safe here since we are only dealing with file pages and +* we are not kswapd and therefore cannot write dirty file pages. But +* call memalloc_noreclaim_save() anyway, just in case these conditions +* change in the future. +*/ + noreclaim_flag =memalloc_noreclaim_save (); nr_reclaimed =shrink_page_list (&clean_pages, zone->zone_pgdat, &sc, &stat, true); +memalloc_noreclaim_restore (noreclaim_flag); +list_splice (&clean_pages, page_list);mod_node_page_state (zone->zone_pgdat,NR_ISOLATED_FILE , -(long )nr_reclaimed);
@@ -2306,6 +2316,7 @@ unsigned long reclaim_pages(struct list_head *page_list)LIST_HEAD (node_page_list);struct reclaim_stat dummy_stat;struct page *page; +unsigned int noreclaim_flag;struct scan_control sc = { .gfp_mask =GFP_KERNEL , .priority =DEF_PRIORITY ,
@@ -2314,6 +2325,8 @@ unsigned long reclaim_pages(struct list_head *page_list) .may_swap = 1, }; + noreclaim_flag =memalloc_noreclaim_save (); +while (!list_empty (page_list)) { page =lru_to_page (page_list);if (nid ==NUMA_NO_NODE ) {
@@ -2350,6 +2363,8 @@ unsigned long reclaim_pages(struct list_head *page_list) } } +memalloc_noreclaim_restore (noreclaim_flag); +return nr_reclaimed; }
Signature · 1 line
2.32.0.272.g935e593368-goog
Andrew Morton
maintainerQuoted 3 lines · On Mon, 14 Jun 2021 13:47:27 -0600 Yu Zhao <[email protected]> wrote:
Use memalloc_noreclaim_save()/memalloc_noreclaim_restore() in
reclaim_pages()to prevent the page reclaim from going into the block I/O layer recursively and deadlock.
Well. Deadlocking the kernel is considered a bad thing ;)
From the lack of a cc:stable I'm assuming that this is a theoretical from-code-inspection thing and that such a deadlock has not been observed?
If not, why do we think that is the case? What is saving us?
(In other words, more detailed changelogging, please!)
PATCH v2 1/1 mm/vmscan.c: fix potential deadlock in reclaim_pages()
Theoretically without the protect from memalloc_noreclaim_save() and memalloc_noreclaim_restore(), reclaim_pages() can go into the block I/O layer recursively and deadlock.
Querying 'reclaim_pages' in our kernel crash databases didn't yield any results. So the deadlock seems unlikely to happen. A possible explanation is that the only user of reclaim_pages(), i.e., MADV_PAGEOUT, is usually called before memory pressure builds up, e.g., on Android and Chrome OS. Under such a condition, allocations in the block I/O layer can be fulfilled without diverting to direct reclaim and therefore the recursion is avoided.
---
v1 -> v2: update the commit message pre request
mm/vmscan.c | 15 +++++++++++++++ 1 file changed, 15 insertions(+)
mm/vmscan.c +15 -0
@@ -1701,6 +1701,7 @@ unsigned int reclaim_clean_pages_from_list(struct zone *zone,unsigned int nr_reclaimed;struct page *page, *next;LIST_HEAD (clean_pages); +unsigned int noreclaim_flag;list_for_each_entry_safe (page, next, page_list, lru) { if (!PageHuge(page) && page_is_file_lru(page) &&
@@ -1711,8 +1712,17 @@ unsigned int reclaim_clean_pages_from_list(struct zone *zone, } } +/* +* We should be safe here since we are only dealing with file pages and +* we are not kswapd and therefore cannot write dirty file pages. But +* call memalloc_noreclaim_save() anyway, just in case these conditions +* change in the future. +*/ + noreclaim_flag =memalloc_noreclaim_save (); nr_reclaimed =shrink_page_list (&clean_pages, zone->zone_pgdat, &sc, &stat, true); +memalloc_noreclaim_restore (noreclaim_flag); +list_splice (&clean_pages, page_list);mod_node_page_state (zone->zone_pgdat,NR_ISOLATED_FILE , -(long )nr_reclaimed);
@@ -2306,6 +2316,7 @@ unsigned long reclaim_pages(struct list_head *page_list)LIST_HEAD (node_page_list);struct reclaim_stat dummy_stat;struct page *page; +unsigned int noreclaim_flag;struct scan_control sc = { .gfp_mask =GFP_KERNEL , .priority =DEF_PRIORITY ,
@@ -2314,6 +2325,8 @@ unsigned long reclaim_pages(struct list_head *page_list) .may_swap = 1, }; + noreclaim_flag =memalloc_noreclaim_save (); +while (!list_empty (page_list)) { page =lru_to_page (page_list);if (nid ==NUMA_NO_NODE ) {
@@ -2350,6 +2363,8 @@ unsigned long reclaim_pages(struct list_head *page_list) } } +memalloc_noreclaim_restore (noreclaim_flag); +return nr_reclaimed; }
Signature · 1 line
2.32.0.288.g62a8d224e6-goog