staging: android: ashmem: Avoid deadlock with mmap/shrink
authorLaura Abbott <lauraa@codeaurora.org>
Fri, 24 Jan 2014 23:19:49 +0000 (15:19 -0800)
committerAmit Pundir <amit.pundir@linaro.org>
Mon, 11 Jan 2016 04:18:49 +0000 (09:48 +0530)
Both ashmem_mmap and ashmem_shrink take the ashmem_lock. It may
be possible for ashmem_mmap to invoke ashmem_shrink:

-000|mutex_lock(lock = 0x0)
-001|ashmem_shrink(?, sc = 0x0) <--- try to take ashmem_mutex again
-002|shrink_slab(shrink = 0xDA5F1CC0, nr_pages_scanned = 0, lru_pages
-002|=
-002|124)
-003|try_to_free_pages(zonelist = 0x0, ?, ?, ?)
-004|__alloc_pages_nodemask(gfp_mask = 21200, order = 1, zonelist =
-004|0xC11D0940,
-005|new_slab(s = 0xE4841E80, ?, node = -1)
-006|__slab_alloc.isra.43.constprop.50(s = 0xE4841E80, gfpflags =
-006|2148925462, ad
-007|kmem_cache_alloc(s = 0xE4841E80, gfpflags = 208)
-008|shmem_alloc_inode(?)
-009|alloc_inode(sb = 0xE480E800)
-010|new_inode_pseudo(?)
-011|new_inode(?)
-012|shmem_get_inode(sb = 0xE480E800, dir = 0x0, ?, dev = 0, flags =
-012|187)
-013|shmem_file_setup(?, ?, flags = 187)
-014|ashmem_mmap(?, vma = 0xC5D64210) <---- Acquire ashmem_mutex
-015|mmap_region(file = 0xDF8E2C00, addr = 1772974080, len = 233472,
-015|flags = 57,
-016|sys_mmap_pgoff(addr = 0, len = 230400, prot = 3, flags = 1, fd =
-016|157, pgoff
-017|ret_fast_syscall(asm)
-->|exception
-018|NUR:0x40097508(asm)
---|end of frame

Avoid this deadlock by using mutex_trylock in ashmem_shrink; if the mutex
is already held, do not attempt to shrink.

Change-Id: I222bbf55856d5849da813b730de0636c80966c8e
Reported-by: Matt Wagantall <mattw@codeaurora.org>
Reported-by: Syed Rameez Mustafa <rameezmustafa@codeaurora.org>
Reported-by: Osvaldo Banuelos <osvaldob@codeaurora.org>
Reported-by: Subbaraman Narayanamurthy <subbaram@codeaurora.org>
Signed-off-by: Laura Abbott <lauraa@codeaurora.org>
drivers/staging/android/ashmem.c

index ae0166ff667fd60e8cfa7b4773e0817b53af41a8..1830548e29a2d6f05f989cf12d24232ff3c8cd26 100644 (file)
@@ -433,7 +433,9 @@ ashmem_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
        if (!(sc->gfp_mask & __GFP_FS))
                return SHRINK_STOP;
 
-       mutex_lock(&ashmem_mutex);
+       if (!mutex_trylock(&ashmem_mutex))
+               return -1;
+
        list_for_each_entry_safe(range, next, &ashmem_lru_list, lru) {
                loff_t start = range->pgstart * PAGE_SIZE;
                loff_t end = (range->pgend + 1) * PAGE_SIZE;