[PATCH] holepunch: fix shmem_truncate_range punching too far
authorHugh Dickins <hugh@veritas.com>
Thu, 29 Mar 2007 08:20:35 +0000 (01:20 -0700)
committerLinus Torvalds <torvalds@woody.linux-foundation.org>
Thu, 29 Mar 2007 15:22:25 +0000 (08:22 -0700)
Miklos Szeredi observes BUG_ON(!entry) in shmem_writepage() triggered in rare
circumstances, because shmem_truncate_range() erroneously removes partially
truncated directory pages at the end of the range: later reclaim on pages
pointing to these removed directories triggers the BUG.  Indeed, and it can
also cause data loss beyond the hole.

Fix this as in the patch proposed by Miklos, but distinguish between "limit"
(how far we need to search: ignore truncation's next_index optimization in the
holepunch case - if there are races it's more consistent to act on the whole
range specified) and "upper_limit" (how far we can free directory pages:
generally we must be careful to keep partially punched pages, but can relax at
end of file - i_size being held stable by i_mutex).

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Cc: Miklos Szeredi <mszeredi@suse.cs>
Cc: Badari Pulavarty <pbadari@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/shmem.c

index b8c429a2d271b676d21840f095ec5449000424db..1077b1d903d26c877431465721452331d18a080e 100644 (file)
@@ -481,7 +481,8 @@ static void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end)
        long nr_swaps_freed = 0;
        int offset;
        int freed;
-       int punch_hole = 0;
+       int punch_hole;
+       unsigned long upper_limit;
 
        inode->i_ctime = inode->i_mtime = CURRENT_TIME;
        idx = (start + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
@@ -492,11 +493,18 @@ static void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end)
        info->flags |= SHMEM_TRUNCATE;
        if (likely(end == (loff_t) -1)) {
                limit = info->next_index;
+               upper_limit = SHMEM_MAX_INDEX;
                info->next_index = idx;
+               punch_hole = 0;
        } else {
-               limit = (end + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
-               if (limit > info->next_index)
-                       limit = info->next_index;
+               if (end + 1 >= inode->i_size) { /* we may free a little more */
+                       limit = (inode->i_size + PAGE_CACHE_SIZE - 1) >>
+                                                       PAGE_CACHE_SHIFT;
+                       upper_limit = SHMEM_MAX_INDEX;
+               } else {
+                       limit = (end + 1) >> PAGE_CACHE_SHIFT;
+                       upper_limit = limit;
+               }
                punch_hole = 1;
        }
 
@@ -520,10 +528,10 @@ static void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end)
         * If there are no indirect blocks or we are punching a hole
         * below indirect blocks, nothing to be done.
         */
-       if (!topdir || (punch_hole && (limit <= SHMEM_NR_DIRECT)))
+       if (!topdir || limit <= SHMEM_NR_DIRECT)
                goto done2;
 
-       BUG_ON(limit <= SHMEM_NR_DIRECT);
+       upper_limit -= SHMEM_NR_DIRECT;
        limit -= SHMEM_NR_DIRECT;
        idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0;
        offset = idx % ENTRIES_PER_PAGE;
@@ -543,7 +551,7 @@ static void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end)
                if (*dir) {
                        diroff = ((idx - ENTRIES_PER_PAGEPAGE/2) %
                                ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE;
-                       if (!diroff && !offset) {
+                       if (!diroff && !offset && upper_limit >= stage) {
                                *dir = NULL;
                                nr_pages_to_free++;
                                list_add(&middir->lru, &pages_to_free);
@@ -570,9 +578,11 @@ static void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end)
                        }
                        stage = idx + ENTRIES_PER_PAGEPAGE;
                        middir = *dir;
-                       *dir = NULL;
-                       nr_pages_to_free++;
-                       list_add(&middir->lru, &pages_to_free);
+                       if (upper_limit >= stage) {
+                               *dir = NULL;
+                               nr_pages_to_free++;
+                               list_add(&middir->lru, &pages_to_free);
+                       }
                        shmem_dir_unmap(dir);
                        cond_resched();
                        dir = shmem_dir_map(middir);
@@ -598,7 +608,7 @@ static void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end)
                }
                if (offset)
                        offset = 0;
-               else if (subdir && !page_private(subdir)) {
+               else if (subdir && upper_limit - idx >= ENTRIES_PER_PAGE) {
                        dir[diroff] = NULL;
                        nr_pages_to_free++;
                        list_add(&subdir->lru, &pages_to_free);