-stable review patch. If anyone has any objections, please let us know.
------------------
From: Hugh Dickins <[email protected]>
Miklos Szeredi observes that during truncation of shmem page directories,
info->lock is released to improve latency (after lowering i_size and
next_index to exclude races); but this is quite wrong for holepunching,
which receives no such protection from i_size or next_index, and is left
vulnerable to races with shmem_unuse, shmem_getpage and shmem_writepage.
Hold info->lock throughout when holepunching? No, any user could prevent
rescheduling for far too long. Instead take info->lock just when needed:
in shmem_free_swp when removing the swap entries, and whenever removing
a directory page from the level above. But so long as we remove before
scanning, we can safely skip taking the lock at the lower levels, except
at misaligned start and end of the hole.
Signed-off-by: Hugh Dickins <[email protected]>
Signed-off-by: Greg Kroah-Hartman <[email protected]>
---
mm/shmem.c | 96 ++++++++++++++++++++++++++++++++++++++++++++++---------------
1 file changed, 73 insertions(+), 23 deletions(-)
--- a/mm/shmem.c
+++ b/mm/shmem.c
@@ -402,26 +402,38 @@ static swp_entry_t *shmem_swp_alloc(stru
/*
* shmem_free_swp - free some swap entries in a directory
*
- * @dir: pointer to the directory
- * @edir: pointer after last entry of the directory
+ * @dir: pointer to the directory
+ * @edir: pointer after last entry of the directory
+ * @punch_lock: pointer to spinlock when needed for the holepunch case
*/
-static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir)
+static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir,
+ spinlock_t *punch_lock)
{
+ spinlock_t *punch_unlock = NULL;
swp_entry_t *ptr;
int freed = 0;
for (ptr = dir; ptr < edir; ptr++) {
if (ptr->val) {
+ if (unlikely(punch_lock)) {
+ punch_unlock = punch_lock;
+ punch_lock = NULL;
+ spin_lock(punch_unlock);
+ if (!ptr->val)
+ continue;
+ }
free_swap_and_cache(*ptr);
*ptr = (swp_entry_t){0};
freed++;
}
}
+ if (punch_unlock)
+ spin_unlock(punch_unlock);
return freed;
}
-static int shmem_map_and_free_swp(struct page *subdir,
- int offset, int limit, struct page ***dir)
+static int shmem_map_and_free_swp(struct page *subdir, int offset,
+ int limit, struct page ***dir, spinlock_t *punch_lock)
{
swp_entry_t *ptr;
int freed = 0;
@@ -431,7 +443,8 @@ static int shmem_map_and_free_swp(struct
int size = limit - offset;
if (size > LATENCY_LIMIT)
size = LATENCY_LIMIT;
- freed += shmem_free_swp(ptr+offset, ptr+offset+size);
+ freed += shmem_free_swp(ptr+offset, ptr+offset+size,
+ punch_lock);
if (need_resched()) {
shmem_swp_unmap(ptr);
if (*dir) {
@@ -482,6 +495,8 @@ static void shmem_truncate_range(struct
int offset;
int freed;
int punch_hole;
+ spinlock_t *needs_lock;
+ spinlock_t *punch_lock;
unsigned long upper_limit;
inode->i_ctime = inode->i_mtime = CURRENT_TIME;
@@ -495,6 +510,7 @@ static void shmem_truncate_range(struct
limit = info->next_index;
upper_limit = SHMEM_MAX_INDEX;
info->next_index = idx;
+ needs_lock = NULL;
punch_hole = 0;
} else {
if (end + 1 >= inode->i_size) { /* we may free a little more */
@@ -505,6 +521,7 @@ static void shmem_truncate_range(struct
limit = (end + 1) >> PAGE_CACHE_SHIFT;
upper_limit = limit;
}
+ needs_lock = &info->lock;
punch_hole = 1;
}
@@ -521,7 +538,7 @@ static void shmem_truncate_range(struct
size = limit;
if (size > SHMEM_NR_DIRECT)
size = SHMEM_NR_DIRECT;
- nr_swaps_freed = shmem_free_swp(ptr+idx, ptr+size);
+ nr_swaps_freed = shmem_free_swp(ptr+idx, ptr+size, needs_lock);
}
/*
@@ -531,6 +548,19 @@ static void shmem_truncate_range(struct
if (!topdir || limit <= SHMEM_NR_DIRECT)
goto done2;
+ /*
+ * The truncation case has already dropped info->lock, and we're safe
+ * because i_size and next_index have already been lowered, preventing
+ * access beyond. But in the punch_hole case, we still need to take
+ * the lock when updating the swap directory, because there might be
+ * racing accesses by shmem_getpage(SGP_CACHE), shmem_unuse_inode or
+ * shmem_writepage. However, whenever we find we can remove a whole
+ * directory page (not at the misaligned start or end of the range),
+ * we first NULLify its pointer in the level above, and then have no
+ * need to take the lock when updating its contents: needs_lock and
+ * punch_lock (either pointing to info->lock or NULL) manage this.
+ */
+
upper_limit -= SHMEM_NR_DIRECT;
limit -= SHMEM_NR_DIRECT;
idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0;
@@ -552,7 +582,13 @@ static void shmem_truncate_range(struct
diroff = ((idx - ENTRIES_PER_PAGEPAGE/2) %
ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE;
if (!diroff && !offset && upper_limit >= stage) {
- *dir = NULL;
+ if (needs_lock) {
+ spin_lock(needs_lock);
+ *dir = NULL;
+ spin_unlock(needs_lock);
+ needs_lock = NULL;
+ } else
+ *dir = NULL;
nr_pages_to_free++;
list_add(&middir->lru, &pages_to_free);
}
@@ -578,8 +614,16 @@ static void shmem_truncate_range(struct
}
stage = idx + ENTRIES_PER_PAGEPAGE;
middir = *dir;
+ if (punch_hole)
+ needs_lock = &info->lock;
if (upper_limit >= stage) {
- *dir = NULL;
+ if (needs_lock) {
+ spin_lock(needs_lock);
+ *dir = NULL;
+ spin_unlock(needs_lock);
+ needs_lock = NULL;
+ } else
+ *dir = NULL;
nr_pages_to_free++;
list_add(&middir->lru, &pages_to_free);
}
@@ -588,31 +632,37 @@ static void shmem_truncate_range(struct
dir = shmem_dir_map(middir);
diroff = 0;
}
+ punch_lock = needs_lock;
subdir = dir[diroff];
- if (subdir && page_private(subdir)) {
+ if (subdir && !offset && upper_limit-idx >= ENTRIES_PER_PAGE) {
+ if (needs_lock) {
+ spin_lock(needs_lock);
+ dir[diroff] = NULL;
+ spin_unlock(needs_lock);
+ punch_lock = NULL;
+ } else
+ dir[diroff] = NULL;
+ nr_pages_to_free++;
+ list_add(&subdir->lru, &pages_to_free);
+ }
+ if (subdir && page_private(subdir) /* has swap entries */) {
size = limit - idx;
if (size > ENTRIES_PER_PAGE)
size = ENTRIES_PER_PAGE;
freed = shmem_map_and_free_swp(subdir,
- offset, size, &dir);
+ offset, size, &dir, punch_lock);
if (!dir)
dir = shmem_dir_map(middir);
nr_swaps_freed += freed;
- if (offset)
+ if (offset || punch_lock) {
spin_lock(&info->lock);
- set_page_private(subdir, page_private(subdir) - freed);
- if (offset)
+ set_page_private(subdir,
+ page_private(subdir) - freed);
spin_unlock(&info->lock);
- if (!punch_hole)
- BUG_ON(page_private(subdir) > offset);
- }
- if (offset)
- offset = 0;
- else if (subdir && upper_limit - idx >= ENTRIES_PER_PAGE) {
- dir[diroff] = NULL;
- nr_pages_to_free++;
- list_add(&subdir->lru, &pages_to_free);
+ } else
+ BUG_ON(page_private(subdir) != freed);
}
+ offset = 0;
}
done1:
shmem_dir_unmap(dir);
--
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
[Index of Archives]
[Kernel Newbies]
[Netfilter]
[Bugtraq]
[Photo]
[Stuff]
[Gimp]
[Yosemite News]
[MIPS Linux]
[ARM Linux]
[Linux Security]
[Linux RAID]
[Video 4 Linux]
[Linux for the blind]
[Linux Resources]
