NTFS: - In fs/ntfs/compress.c, use i_size_read() at the start and then use the
Anton Altaparmakov [Thu, 18 Nov 2004 13:46:45 +0000 (13:46 +0000)]
cached value everywhere.  Cache the initialized_size in the same way
and protect the critical region where the two sizes are read using the
new size_lock of the ntfs inode.
      - Add the new size_lock to the ntfs_inode structure (fs/ntfs/inode.h)
and initialize it (fs/ntfs/inode.c).

Signed-off-by: Anton Altaparmakov <aia21@cantab.net>

fs/ntfs/ChangeLog
fs/ntfs/compress.c
fs/ntfs/inode.c
fs/ntfs/inode.h

index 6d2a99c..97c4fe9 100644 (file)
@@ -34,6 +34,13 @@ ToDo/Notes:
        - Use i_size_read() once and then use the cached value in
          fs/ntfs/lcnalloc.c::ntfs_cluster_alloc().
        - Use i_size_read() in fs/ntfs/file.c::ntfs_file_open().
+       - Add size_lock to the ntfs_inode structure.  This is an rw spinlock
+         and it locks against access to the inode sizes.  Note, ->size_lock
+         is also accessed from irq context so you must use the _irqsave and
+         _irqrestore lock and unlock functions, respectively.
+       - Use i_size_read() in fs/ntfs/compress.c at the start of the read and
+         use the cached value afterwards.  Cache the initialized_size in the
+         same way and protect access to the two sizes using the size_lock.
 
 2.1.22 - Many bug and race fixes and error handling improvements.
 
index ee5ae70..6d265cf 100644 (file)
@@ -96,13 +96,14 @@ void free_compression_buffers(void)
 /**
  * zero_partial_compressed_page - zero out of bounds compressed page region
  */
-static void zero_partial_compressed_page(ntfs_inode *ni, struct page *page)
+static void zero_partial_compressed_page(struct page *page,
+               const s64 initialized_size)
 {
        u8 *kp = page_address(page);
        unsigned int kp_ofs;
 
        ntfs_debug("Zeroing page region outside initialized size.");
-       if (((s64)page->index << PAGE_CACHE_SHIFT) >= ni->initialized_size) {
+       if (((s64)page->index << PAGE_CACHE_SHIFT) >= initialized_size) {
                /*
                 * FIXME: Using clear_page() will become wrong when we get
                 * PAGE_CACHE_SIZE != PAGE_SIZE but for now there is no problem.
@@ -110,7 +111,7 @@ static void zero_partial_compressed_page(ntfs_inode *ni, struct page *page)
                clear_page(kp);
                return;
        }
-       kp_ofs = ni->initialized_size & ~PAGE_CACHE_MASK;
+       kp_ofs = initialized_size & ~PAGE_CACHE_MASK;
        memset(kp + kp_ofs, 0, PAGE_CACHE_SIZE - kp_ofs);
        return;
 }
@@ -118,12 +119,12 @@ static void zero_partial_compressed_page(ntfs_inode *ni, struct page *page)
 /**
  * handle_bounds_compressed_page - test for&handle out of bounds compressed page
  */
-static inline void handle_bounds_compressed_page(ntfs_inode *ni,
-               struct page *page)
+static inline void handle_bounds_compressed_page(struct page *page,
+               const loff_t i_size, const s64 initialized_size)
 {
-       if ((page->index >= (ni->initialized_size >> PAGE_CACHE_SHIFT)) &&
-                       (ni->initialized_size < VFS_I(ni)->i_size))
-               zero_partial_compressed_page(ni, page);
+       if ((page->index >= (initialized_size >> PAGE_CACHE_SHIFT)) &&
+                       (initialized_size < i_size))
+               zero_partial_compressed_page(page, initialized_size);
        return;
 }
 
@@ -138,6 +139,8 @@ static inline void handle_bounds_compressed_page(ntfs_inode *ni,
  * @xpage_done:                set to 1 if xpage was completed successfully (IN/OUT)
  * @cb_start:          compression block to decompress (IN)
  * @cb_size:           size of compression block @cb_start in bytes (IN)
+ * @i_size:            file size when we started the read (IN)
+ * @initialized_size:  initialized file size when we started the read (IN)
  *
  * The caller must have disabled preemption. ntfs_decompress() reenables it when
  * the critical section is finished.
@@ -165,7 +168,8 @@ static inline void handle_bounds_compressed_page(ntfs_inode *ni,
 static int ntfs_decompress(struct page *dest_pages[], int *dest_index,
                int *dest_ofs, const int dest_max_index, const int dest_max_ofs,
                const int xpage, char *xpage_done, u8 *const cb_start,
-               const u32 cb_size)
+               const u32 cb_size, const loff_t i_size,
+               const s64 initialized_size)
 {
        /*
         * Pointers into the compressed data, i.e. the compression block (cb),
@@ -219,9 +223,6 @@ return_error:
                spin_unlock(&ntfs_cb_lock);
                /* Second stage: finalize completed pages. */
                if (nr_completed_pages > 0) {
-                       struct page *page = dest_pages[completed_pages[0]];
-                       ntfs_inode *ni = NTFS_I(page->mapping->host);
-
                        for (i = 0; i < nr_completed_pages; i++) {
                                int di = completed_pages[i];
 
@@ -230,7 +231,8 @@ return_error:
                                 * If we are outside the initialized size, zero
                                 * the out of bounds page range.
                                 */
-                               handle_bounds_compressed_page(ni, dp);
+                               handle_bounds_compressed_page(dp, i_size,
+                                               initialized_size);
                                flush_dcache_page(dp);
                                kunmap(dp);
                                SetPageUptodate(dp);
@@ -478,12 +480,14 @@ return_overflow:
  */
 int ntfs_read_compressed_block(struct page *page)
 {
+       loff_t i_size;
+       s64 initialized_size;
        struct address_space *mapping = page->mapping;
        ntfs_inode *ni = NTFS_I(mapping->host);
        ntfs_volume *vol = ni->vol;
        struct super_block *sb = vol->sb;
        runlist_element *rl;
-       unsigned long block_size = sb->s_blocksize;
+       unsigned long flags, block_size = sb->s_blocksize;
        unsigned char block_size_bits = sb->s_blocksize_bits;
        u8 *cb, *cb_pos, *cb_end;
        struct buffer_head **bhs;
@@ -552,8 +556,12 @@ int ntfs_read_compressed_block(struct page *page)
         * The remaining pages need to be allocated and inserted into the page
         * cache, alignment guarantees keep all the below much simpler. (-8
         */
-       max_page = ((VFS_I(ni)->i_size + PAGE_CACHE_SIZE - 1) >>
-                       PAGE_CACHE_SHIFT) - offset;
+       read_lock_irqsave(&ni->size_lock, flags);
+       i_size = i_size_read(VFS_I(ni));
+       initialized_size = ni->initialized_size;
+       read_unlock_irqrestore(&ni->size_lock, flags);
+       max_page = ((i_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) -
+                       offset;
        if (nr_pages < max_page)
                max_page = nr_pages;
        for (i = 0; i < max_page; i++, offset++) {
@@ -824,7 +832,8 @@ lock_retry_remap:
                                 * If we are outside the initialized size, zero
                                 * the out of bounds page range.
                                 */
-                               handle_bounds_compressed_page(ni, page);
+                               handle_bounds_compressed_page(page, i_size,
+                                               initialized_size);
                                flush_dcache_page(page);
                                kunmap(page);
                                SetPageUptodate(page);
@@ -847,7 +856,8 @@ lock_retry_remap:
                ntfs_debug("Found compressed compression block.");
                err = ntfs_decompress(pages, &cur_page, &cur_ofs,
                                cb_max_page, cb_max_ofs, xpage, &xpage_done,
-                               cb_pos, cb_size - (cb_pos - cb));
+                               cb_pos, cb_size - (cb_pos - cb), i_size,
+                               initialized_size);
                /*
                 * We can sleep from now on, lock already dropped by
                 * ntfs_decompress().
index 31840ba..a02d8d9 100644 (file)
@@ -376,6 +376,7 @@ static void ntfs_destroy_extent_inode(ntfs_inode *ni)
 void __ntfs_init_inode(struct super_block *sb, ntfs_inode *ni)
 {
        ntfs_debug("Entering.");
+       rwlock_init(&ni->size_lock);
        ni->initialized_size = ni->allocated_size = 0;
        ni->seq_no = 0;
        atomic_set(&ni->count, 1);
index 9958045..56b2d6e 100644 (file)
@@ -44,6 +44,7 @@ typedef struct _ntfs_inode ntfs_inode;
  * fields already provided in the VFS inode.
  */
 struct _ntfs_inode {
+       rwlock_t size_lock;     /* Lock serializing access to inode sizes. */
        s64 initialized_size;   /* Copy from the attribute record. */
        s64 allocated_size;     /* Copy from the attribute record. */
        unsigned long state;    /* NTFS specific flags describing this inode.