fs: symlink write_begin allocation context fix
[safe/jmp/linux-2.6] / fs / ext4 / extents.c
index 5c5dd3a..ea2ce3c 100644 (file)
@@ -40,6 +40,7 @@
 #include <linux/slab.h>
 #include <linux/falloc.h>
 #include <asm/uaccess.h>
+#include <linux/fiemap.h>
 #include "ext4_jbd2.h"
 #include "ext4_extents.h"
 
@@ -383,8 +384,8 @@ static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
        ext_debug("\n");
 }
 #else
-#define ext4_ext_show_path(inode,path)
-#define ext4_ext_show_leaf(inode,path)
+#define ext4_ext_show_path(inode, path)
+#define ext4_ext_show_leaf(inode, path)
 #endif
 
 void ext4_ext_drop_refs(struct ext4_ext_path *path)
@@ -440,9 +441,10 @@ ext4_ext_binsearch_idx(struct inode *inode,
                for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
                  if (k != 0 &&
                      le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
-                               printk("k=%d, ix=0x%p, first=0x%p\n", k,
-                                       ix, EXT_FIRST_INDEX(eh));
-                               printk("%u <= %u\n",
+                               printk(KERN_DEBUG "k=%d, ix=0x%p, "
+                                      "first=0x%p\n", k,
+                                      ix, EXT_FIRST_INDEX(eh));
+                               printk(KERN_DEBUG "%u <= %u\n",
                                       le32_to_cpu(ix->ei_block),
                                       le32_to_cpu(ix[-1].ei_block));
                        }
@@ -1475,7 +1477,7 @@ int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
                                struct ext4_ext_path *path,
                                struct ext4_extent *newext)
 {
-       struct ext4_extent_header * eh;
+       struct ext4_extent_header *eh;
        struct ext4_extent *ex, *fex;
        struct ext4_extent *nearex; /* nearest extent */
        struct ext4_ext_path *npath = NULL;
@@ -1625,6 +1627,113 @@ cleanup:
        return err;
 }
 
+int ext4_ext_walk_space(struct inode *inode, ext4_lblk_t block,
+                       ext4_lblk_t num, ext_prepare_callback func,
+                       void *cbdata)
+{
+       struct ext4_ext_path *path = NULL;
+       struct ext4_ext_cache cbex;
+       struct ext4_extent *ex;
+       ext4_lblk_t next, start = 0, end = 0;
+       ext4_lblk_t last = block + num;
+       int depth, exists, err = 0;
+
+       BUG_ON(func == NULL);
+       BUG_ON(inode == NULL);
+
+       while (block < last && block != EXT_MAX_BLOCK) {
+               num = last - block;
+               /* find extent for this block */
+               path = ext4_ext_find_extent(inode, block, path);
+               if (IS_ERR(path)) {
+                       err = PTR_ERR(path);
+                       path = NULL;
+                       break;
+               }
+
+               depth = ext_depth(inode);
+               BUG_ON(path[depth].p_hdr == NULL);
+               ex = path[depth].p_ext;
+               next = ext4_ext_next_allocated_block(path);
+
+               exists = 0;
+               if (!ex) {
+                       /* there is no extent yet, so try to allocate
+                        * all requested space */
+                       start = block;
+                       end = block + num;
+               } else if (le32_to_cpu(ex->ee_block) > block) {
+                       /* need to allocate space before found extent */
+                       start = block;
+                       end = le32_to_cpu(ex->ee_block);
+                       if (block + num < end)
+                               end = block + num;
+               } else if (block >= le32_to_cpu(ex->ee_block)
+                                       + ext4_ext_get_actual_len(ex)) {
+                       /* need to allocate space after found extent */
+                       start = block;
+                       end = block + num;
+                       if (end >= next)
+                               end = next;
+               } else if (block >= le32_to_cpu(ex->ee_block)) {
+                       /*
+                        * some part of requested space is covered
+                        * by found extent
+                        */
+                       start = block;
+                       end = le32_to_cpu(ex->ee_block)
+                               + ext4_ext_get_actual_len(ex);
+                       if (block + num < end)
+                               end = block + num;
+                       exists = 1;
+               } else {
+                       BUG();
+               }
+               BUG_ON(end <= start);
+
+               if (!exists) {
+                       cbex.ec_block = start;
+                       cbex.ec_len = end - start;
+                       cbex.ec_start = 0;
+                       cbex.ec_type = EXT4_EXT_CACHE_GAP;
+               } else {
+                       cbex.ec_block = le32_to_cpu(ex->ee_block);
+                       cbex.ec_len = ext4_ext_get_actual_len(ex);
+                       cbex.ec_start = ext_pblock(ex);
+                       cbex.ec_type = EXT4_EXT_CACHE_EXTENT;
+               }
+
+               BUG_ON(cbex.ec_len == 0);
+               err = func(inode, path, &cbex, ex, cbdata);
+               ext4_ext_drop_refs(path);
+
+               if (err < 0)
+                       break;
+
+               if (err == EXT_REPEAT)
+                       continue;
+               else if (err == EXT_BREAK) {
+                       err = 0;
+                       break;
+               }
+
+               if (ext_depth(inode) != depth) {
+                       /* depth was changed. we have to realloc path */
+                       kfree(path);
+                       path = NULL;
+               }
+
+               block = cbex.ec_block + cbex.ec_len;
+       }
+
+       if (path) {
+               ext4_ext_drop_refs(path);
+               kfree(path);
+       }
+
+       return err;
+}
+
 static void
 ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block,
                        __u32 len, ext4_fsblk_t start, int type)
@@ -1753,12 +1862,12 @@ static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
  * When pass the actual path, the caller should calculate credits
  * under i_data_sem.
  */
-int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int num,
+int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
                                                struct ext4_ext_path *path)
 {
        if (path) {
                int depth = ext_depth(inode);
-               int ret;
+               int ret = 0;
 
                /* probably there is space in leaf? */
                if (le16_to_cpu(path[depth].p_hdr->eh_entries)
@@ -1772,12 +1881,12 @@ int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int num,
                         *  and other metadat blocks still need to be
                         *  accounted.
                         */
-                       /* 1 one bitmap, 1 block group descriptor */
+                       /* 1 bitmap, 1 block group descriptor */
                        ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
                }
        }
 
-       return ext4_meta_trans_blocks(inode, num, 1);
+       return ext4_chunk_trans_blocks(inode, nrblocks);
 }
 
 /*
@@ -1791,7 +1900,7 @@ int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int num,
  * If the nrblocks are discontiguous, they could cause
  * the whole tree split more than once, but this is really rare.
  */
-int ext4_ext_index_trans_blocks(struct inode *inode, int num, int chunk)
+int ext4_ext_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
 {
        int index;
        int depth = ext_depth(inode);
@@ -2142,7 +2251,7 @@ void ext4_ext_init(struct super_block *sb)
         */
 
        if (test_opt(sb, EXTENTS)) {
-               printk("EXT4-fs: file extents enabled");
+               printk(KERN_INFO "EXT4-fs: file extents enabled");
 #ifdef AGGRESSIVE_TEST
                printk(", aggressive tests");
 #endif
@@ -2696,11 +2805,8 @@ int ext4_ext_get_blocks(handle_t *handle, struct inode *inode,
                goto out2;
        }
        /*
-        * Okay, we need to do block allocation.  Lazily initialize the block
-        * allocation info here if necessary.
+        * Okay, we need to do block allocation.
         */
-       if (S_ISREG(inode->i_mode) && (!EXT4_I(inode)->i_block_alloc_info))
-               ext4_init_block_alloc_info(inode);
 
        /* find neighbour allocated blocks */
        ar.lleft = iblock;
@@ -2760,7 +2866,7 @@ int ext4_ext_get_blocks(handle_t *handle, struct inode *inode,
                /* free data blocks we just allocated */
                /* not a good idea to call discard here directly,
                 * but otherwise we'd need to call it every free() */
-               ext4_mb_discard_inode_preallocations(inode);
+               ext4_discard_preallocations(inode);
                ext4_free_blocks(handle, inode, ext_pblock(&newex),
                                        ext4_ext_get_actual_len(&newex), 0);
                goto out2;
@@ -2810,7 +2916,7 @@ void ext4_ext_truncate(struct inode *inode)
        /*
         * probably first extent we're gonna free will be last in block
         */
-       err = ext4_writepage_trans_blocks(inode) + 3;
+       err = ext4_writepage_trans_blocks(inode);
        handle = ext4_journal_start(inode, err);
        if (IS_ERR(handle))
                return;
@@ -2824,7 +2930,7 @@ void ext4_ext_truncate(struct inode *inode)
        down_write(&EXT4_I(inode)->i_data_sem);
        ext4_ext_invalidate_cache(inode);
 
-       ext4_discard_reservation(inode);
+       ext4_discard_preallocations(inode);
 
        /*
         * TODO: optimization is possible here.
@@ -2877,10 +2983,11 @@ static void ext4_falloc_update_inode(struct inode *inode,
         * Update only when preallocation was requested beyond
         * the file size.
         */
-       if (!(mode & FALLOC_FL_KEEP_SIZE) &&
-                               new_size > i_size_read(inode)) {
-               i_size_write(inode, new_size);
-               EXT4_I(inode)->i_disksize = new_size;
+       if (!(mode & FALLOC_FL_KEEP_SIZE)) {
+               if (new_size > i_size_read(inode))
+                       i_size_write(inode, new_size);
+               if (new_size > EXT4_I(inode)->i_disksize)
+                       ext4_update_i_disksize(inode, new_size);
        }
 
 }
@@ -2923,10 +3030,9 @@ long ext4_fallocate(struct inode *inode, int mode, loff_t offset, loff_t len)
        max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
                                                        - block;
        /*
-        * credits to insert 1 extent into extent tree + buffers to be able to
-        * modify 1 super block, 1 block bitmap and 1 group descriptor.
+        * credits to insert 1 extent into extent tree
         */
-       credits = EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + 3;
+       credits = ext4_chunk_trans_blocks(inode, max_blocks);
        mutex_lock(&inode->i_mutex);
 retry:
        while (ret >= 0 && ret < max_blocks) {
@@ -2973,3 +3079,143 @@ retry:
        mutex_unlock(&inode->i_mutex);
        return ret > 0 ? ret2 : ret;
 }
+
+/*
+ * Callback function called for each extent to gather FIEMAP information.
+ */
+int ext4_ext_fiemap_cb(struct inode *inode, struct ext4_ext_path *path,
+                      struct ext4_ext_cache *newex, struct ext4_extent *ex,
+                      void *data)
+{
+       struct fiemap_extent_info *fieinfo = data;
+       unsigned long blksize_bits = inode->i_sb->s_blocksize_bits;
+       __u64   logical;
+       __u64   physical;
+       __u64   length;
+       __u32   flags = 0;
+       int     error;
+
+       logical =  (__u64)newex->ec_block << blksize_bits;
+
+       if (newex->ec_type == EXT4_EXT_CACHE_GAP) {
+               pgoff_t offset;
+               struct page *page;
+               struct buffer_head *bh = NULL;
+
+               offset = logical >> PAGE_SHIFT;
+               page = find_get_page(inode->i_mapping, offset);
+               if (!page || !page_has_buffers(page))
+                       return EXT_CONTINUE;
+
+               bh = page_buffers(page);
+
+               if (!bh)
+                       return EXT_CONTINUE;
+
+               if (buffer_delay(bh)) {
+                       flags |= FIEMAP_EXTENT_DELALLOC;
+                       page_cache_release(page);
+               } else {
+                       page_cache_release(page);
+                       return EXT_CONTINUE;
+               }
+       }
+
+       physical = (__u64)newex->ec_start << blksize_bits;
+       length =   (__u64)newex->ec_len << blksize_bits;
+
+       if (ex && ext4_ext_is_uninitialized(ex))
+               flags |= FIEMAP_EXTENT_UNWRITTEN;
+
+       /*
+        * If this extent reaches EXT_MAX_BLOCK, it must be last.
+        *
+        * Or if ext4_ext_next_allocated_block is EXT_MAX_BLOCK,
+        * this also indicates no more allocated blocks.
+        *
+        * XXX this might miss a single-block extent at EXT_MAX_BLOCK
+        */
+       if (logical + length - 1 == EXT_MAX_BLOCK ||
+           ext4_ext_next_allocated_block(path) == EXT_MAX_BLOCK)
+               flags |= FIEMAP_EXTENT_LAST;
+
+       error = fiemap_fill_next_extent(fieinfo, logical, physical,
+                                       length, flags);
+       if (error < 0)
+               return error;
+       if (error == 1)
+               return EXT_BREAK;
+
+       return EXT_CONTINUE;
+}
+
+/* fiemap flags we can handle specified here */
+#define EXT4_FIEMAP_FLAGS      (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
+
+int ext4_xattr_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo)
+{
+       __u64 physical = 0;
+       __u64 length;
+       __u32 flags = FIEMAP_EXTENT_LAST;
+       int blockbits = inode->i_sb->s_blocksize_bits;
+       int error = 0;
+
+       /* in-inode? */
+       if (EXT4_I(inode)->i_state & EXT4_STATE_XATTR) {
+               struct ext4_iloc iloc;
+               int offset;     /* offset of xattr in inode */
+
+               error = ext4_get_inode_loc(inode, &iloc);
+               if (error)
+                       return error;
+               physical = iloc.bh->b_blocknr << blockbits;
+               offset = EXT4_GOOD_OLD_INODE_SIZE +
+                               EXT4_I(inode)->i_extra_isize;
+               physical += offset;
+               length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
+               flags |= FIEMAP_EXTENT_DATA_INLINE;
+       } else { /* external block */
+               physical = EXT4_I(inode)->i_file_acl << blockbits;
+               length = inode->i_sb->s_blocksize;
+       }
+
+       if (physical)
+               error = fiemap_fill_next_extent(fieinfo, 0, physical,
+                                               length, flags);
+       return (error < 0 ? error : 0);
+}
+
+int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
+               __u64 start, __u64 len)
+{
+       ext4_lblk_t start_blk;
+       ext4_lblk_t len_blks;
+       int error = 0;
+
+       /* fallback to generic here if not in extents fmt */
+       if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
+               return generic_block_fiemap(inode, fieinfo, start, len,
+                       ext4_get_block);
+
+       if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
+               return -EBADR;
+
+       if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
+               error = ext4_xattr_fiemap(inode, fieinfo);
+       } else {
+               start_blk = start >> inode->i_sb->s_blocksize_bits;
+               len_blks = len >> inode->i_sb->s_blocksize_bits;
+
+               /*
+                * Walk the extent tree gathering extent information.
+                * ext4_ext_fiemap_cb will push extents back to user.
+                */
+               down_write(&EXT4_I(inode)->i_data_sem);
+               error = ext4_ext_walk_space(inode, start_blk, len_blks,
+                                         ext4_ext_fiemap_cb, fieinfo);
+               up_write(&EXT4_I(inode)->i_data_sem);
+       }
+
+       return error;
+}
+