]> err.no Git - linux-2.6/blobdiff - fs/ext4/inode.c
fs/ecryptfs/: possible cleanups
[linux-2.6] / fs / ext4 / inode.c
index e6634550cfc8caf8bb6b3e00be2d1a75e480259f..05c4145dd27d3e03b634d7525aea9fe472012a58 100644 (file)
@@ -243,13 +243,6 @@ static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
        p->bh = bh;
 }
 
-static int verify_chain(Indirect *from, Indirect *to)
-{
-       while (from <= to && from->key == *from->p)
-               from++;
-       return (from > to);
-}
-
 /**
  *     ext4_block_to_path - parse the block number into array of offsets
  *     @inode: inode in question (we are only interested in its superblock)
@@ -314,7 +307,10 @@ static int ext4_block_to_path(struct inode *inode,
                offsets[n++] = i_block & (ptrs - 1);
                final = ptrs;
        } else {
-               ext4_warning(inode->i_sb, "ext4_block_to_path", "block > big");
+               ext4_warning(inode->i_sb, "ext4_block_to_path",
+                               "block %lu > max",
+                               i_block + direct_blocks +
+                               indirect_blocks + double_blocks);
        }
        if (boundary)
                *boundary = final - 1 - (i_block & (ptrs - 1));
@@ -345,10 +341,11 @@ static int ext4_block_to_path(struct inode *inode,
  *             (pointer to last triple returned, *@err == 0)
  *     or when it gets an IO error reading an indirect block
  *             (ditto, *@err == -EIO)
- *     or when it notices that chain had been changed while it was reading
- *             (ditto, *@err == -EAGAIN)
  *     or when it reads all @depth-1 indirect blocks successfully and finds
  *     the whole chain, all way to the data (returns %NULL, *err == 0).
+ *
+ *      Need to be called with
+ *      down_read(&EXT4_I(inode)->i_data_sem)
  */
 static Indirect *ext4_get_branch(struct inode *inode, int depth,
                                 ext4_lblk_t  *offsets,
@@ -367,9 +364,6 @@ static Indirect *ext4_get_branch(struct inode *inode, int depth,
                bh = sb_bread(sb, le32_to_cpu(p->key));
                if (!bh)
                        goto failure;
-               /* Reader: pointers */
-               if (!verify_chain(chain, p))
-                       goto changed;
                add_chain(++p, bh, (__le32*)bh->b_data + *++offsets);
                /* Reader: end */
                if (!p->key)
@@ -377,10 +371,6 @@ static Indirect *ext4_get_branch(struct inode *inode, int depth,
        }
        return NULL;
 
-changed:
-       brelse(bh);
-       *err = -EAGAIN;
-       goto no_block;
 failure:
        *err = -EIO;
 no_block:
@@ -561,7 +551,7 @@ static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
        return ret;
 failed_out:
        for (i = 0; i <index; i++)
-               ext4_free_blocks(handle, inode, new_blocks[i], 1);
+               ext4_free_blocks(handle, inode, new_blocks[i], 1, 0);
        return ret;
 }
 
@@ -660,9 +650,9 @@ failed:
                ext4_journal_forget(handle, branch[i].bh);
        }
        for (i = 0; i <indirect_blks; i++)
-               ext4_free_blocks(handle, inode, new_blocks[i], 1);
+               ext4_free_blocks(handle, inode, new_blocks[i], 1, 0);
 
-       ext4_free_blocks(handle, inode, new_blocks[i], num);
+       ext4_free_blocks(handle, inode, new_blocks[i], num, 0);
 
        return err;
 }
@@ -759,9 +749,10 @@ err_out:
        for (i = 1; i <= num; i++) {
                BUFFER_TRACE(where[i].bh, "call jbd2_journal_forget");
                ext4_journal_forget(handle, where[i].bh);
-               ext4_free_blocks(handle,inode,le32_to_cpu(where[i-1].key),1);
+               ext4_free_blocks(handle, inode,
+                                       le32_to_cpu(where[i-1].key), 1, 0);
        }
-       ext4_free_blocks(handle, inode, le32_to_cpu(where[num].key), blks);
+       ext4_free_blocks(handle, inode, le32_to_cpu(where[num].key), blks, 0);
 
        return err;
 }
@@ -784,6 +775,11 @@ err_out:
  * return > 0, # of blocks mapped or allocated.
  * return = 0, if plain lookup failed.
  * return < 0, error case.
+ *
+ *
+ * Need to be called with
+ * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
+ * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
  */
 int ext4_get_blocks_handle(handle_t *handle, struct inode *inode,
                ext4_lblk_t iblock, unsigned long maxblocks,
@@ -822,18 +818,6 @@ int ext4_get_blocks_handle(handle_t *handle, struct inode *inode,
                while (count < maxblocks && count <= blocks_to_boundary) {
                        ext4_fsblk_t blk;
 
-                       if (!verify_chain(chain, partial)) {
-                               /*
-                                * Indirect block might be removed by
-                                * truncate while we were reading it.
-                                * Handling of that case: forget what we've
-                                * got now. Flag the err as EAGAIN, so it
-                                * will reread.
-                                */
-                               err = -EAGAIN;
-                               count = 0;
-                               break;
-                       }
                        blk = le32_to_cpu(*(chain[depth-1].p + count));
 
                        if (blk == first_block + count)
@@ -841,44 +825,13 @@ int ext4_get_blocks_handle(handle_t *handle, struct inode *inode,
                        else
                                break;
                }
-               if (err != -EAGAIN)
-                       goto got_it;
+               goto got_it;
        }
 
        /* Next simple case - plain lookup or failed read of indirect block */
        if (!create || err == -EIO)
                goto cleanup;
 
-       mutex_lock(&ei->truncate_mutex);
-
-       /*
-        * If the indirect block is missing while we are reading
-        * the chain(ext4_get_branch() returns -EAGAIN err), or
-        * if the chain has been changed after we grab the semaphore,
-        * (either because another process truncated this branch, or
-        * another get_block allocated this branch) re-grab the chain to see if
-        * the request block has been allocated or not.
-        *
-        * Since we already block the truncate/other get_block
-        * at this point, we will have the current copy of the chain when we
-        * splice the branch into the tree.
-        */
-       if (err == -EAGAIN || !verify_chain(chain, partial)) {
-               while (partial > chain) {
-                       brelse(partial->bh);
-                       partial--;
-               }
-               partial = ext4_get_branch(inode, depth, offsets, chain, &err);
-               if (!partial) {
-                       count++;
-                       mutex_unlock(&ei->truncate_mutex);
-                       if (err)
-                               goto cleanup;
-                       clear_buffer_new(bh_result);
-                       goto got_it;
-               }
-       }
-
        /*
         * Okay, we need to do block allocation.  Lazily initialize the block
         * allocation info here if necessary
@@ -914,13 +867,12 @@ int ext4_get_blocks_handle(handle_t *handle, struct inode *inode,
                err = ext4_splice_branch(handle, inode, iblock,
                                        partial, indirect_blks, count);
        /*
-        * i_disksize growing is protected by truncate_mutex.  Don't forget to
+        * i_disksize growing is protected by i_data_sem.  Don't forget to
         * protect it if you're about to implement concurrent
         * ext4_get_block() -bzzz
        */
        if (!err && extend_disksize && inode->i_size > ei->i_disksize)
                ei->i_disksize = inode->i_size;
-       mutex_unlock(&ei->truncate_mutex);
        if (err)
                goto cleanup;
 
@@ -945,6 +897,47 @@ out:
 
 #define DIO_CREDITS (EXT4_RESERVE_TRANS_BLOCKS + 32)
 
+int ext4_get_blocks_wrap(handle_t *handle, struct inode *inode, sector_t block,
+                       unsigned long max_blocks, struct buffer_head *bh,
+                       int create, int extend_disksize)
+{
+       int retval;
+       /*
+        * Try to see if we can get  the block without requesting
+        * for new file system block.
+        */
+       down_read((&EXT4_I(inode)->i_data_sem));
+       if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
+               retval =  ext4_ext_get_blocks(handle, inode, block, max_blocks,
+                               bh, 0, 0);
+       } else {
+               retval = ext4_get_blocks_handle(handle,
+                               inode, block, max_blocks, bh, 0, 0);
+       }
+       up_read((&EXT4_I(inode)->i_data_sem));
+       if (!create || (retval > 0))
+               return retval;
+
+       /*
+        * We need to allocate new blocks which will result
+        * in i_data update
+        */
+       down_write((&EXT4_I(inode)->i_data_sem));
+       /*
+        * We need to check for EXT4 here because migrate
+        * could have changed the inode type in between
+        */
+       if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
+               retval =  ext4_ext_get_blocks(handle, inode, block, max_blocks,
+                               bh, create, extend_disksize);
+       } else {
+               retval = ext4_get_blocks_handle(handle, inode, block,
+                               max_blocks, bh, create, extend_disksize);
+       }
+       up_write((&EXT4_I(inode)->i_data_sem));
+       return retval;
+}
+
 static int ext4_get_block(struct inode *inode, sector_t iblock,
                        struct buffer_head *bh_result, int create)
 {
@@ -1449,7 +1442,7 @@ static int jbd2_journal_dirty_data_fn(handle_t *handle, struct buffer_head *bh)
  *     ext4_file_write() -> generic_file_write() -> __alloc_pages() -> ...
  *
  * Same applies to ext4_get_block().  We will deadlock on various things like
- * lock_journal and i_truncate_mutex.
+ * lock_journal and i_data_sem
  *
  * Setting PF_MEMALLOC here doesn't work - too many internal memory
  * allocations fail.
@@ -1847,7 +1840,7 @@ int ext4_block_truncate_page(handle_t *handle, struct page *page,
         */
        if (!page_has_buffers(page) && test_opt(inode->i_sb, NOBH) &&
             ext4_should_writeback_data(inode) && PageUptodate(page)) {
-               zero_user_page(page, offset, length, KM_USER0);
+               zero_user(page, offset, length);
                set_page_dirty(page);
                goto unlock;
        }
@@ -1900,7 +1893,7 @@ int ext4_block_truncate_page(handle_t *handle, struct page *page,
                        goto unlock;
        }
 
-       zero_user_page(page, offset, length, KM_USER0);
+       zero_user(page, offset, length);
 
        BUFFER_TRACE(bh, "zeroed end of block");
 
@@ -2060,7 +2053,7 @@ static void ext4_clear_blocks(handle_t *handle, struct inode *inode,
                }
        }
 
-       ext4_free_blocks(handle, inode, block_to_free, count);
+       ext4_free_blocks(handle, inode, block_to_free, count, 0);
 }
 
 /**
@@ -2233,7 +2226,7 @@ static void ext4_free_branches(handle_t *handle, struct inode *inode,
                                ext4_journal_test_restart(handle, inode);
                        }
 
-                       ext4_free_blocks(handle, inode, nr, 1);
+                       ext4_free_blocks(handle, inode, nr, 1, 1);
 
                        if (parent_bh) {
                                /*
@@ -2375,7 +2368,7 @@ void ext4_truncate(struct inode *inode)
         * From here we block out all ext4_get_block() callers who want to
         * modify the block allocation tree.
         */
-       mutex_lock(&ei->truncate_mutex);
+       down_write(&ei->i_data_sem);
 
        if (n == 1) {           /* direct blocks */
                ext4_free_data(handle, inode, NULL, i_data+offsets[0],
@@ -2439,7 +2432,7 @@ do_indirects:
 
        ext4_discard_reservation(inode);
 
-       mutex_unlock(&ei->truncate_mutex);
+       up_write(&ei->i_data_sem);
        inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
        ext4_mark_inode_dirty(handle, inode);
 
@@ -2667,6 +2660,28 @@ void ext4_get_inode_flags(struct ext4_inode_info *ei)
        if (flags & S_DIRSYNC)
                ei->i_flags |= EXT4_DIRSYNC_FL;
 }
+static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
+                                       struct ext4_inode_info *ei)
+{
+       blkcnt_t i_blocks ;
+       struct inode *inode = &(ei->vfs_inode);
+       struct super_block *sb = inode->i_sb;
+
+       if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
+                               EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
+               /* we are using combined 48 bit field */
+               i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
+                                       le32_to_cpu(raw_inode->i_blocks_lo);
+               if (ei->i_flags & EXT4_HUGE_FILE_FL) {
+                       /* i_blocks represent file system block size */
+                       return i_blocks  << (inode->i_blkbits - 9);
+               } else {
+                       return i_blocks;
+               }
+       } else {
+               return le32_to_cpu(raw_inode->i_blocks_lo);
+       }
+}
 
 void ext4_read_inode(struct inode * inode)
 {
@@ -2715,8 +2730,8 @@ void ext4_read_inode(struct inode * inode)
                 * recovery code: that's fine, we're about to complete
                 * the process of deleting those. */
        }
-       inode->i_blocks = le32_to_cpu(raw_inode->i_blocks);
        ei->i_flags = le32_to_cpu(raw_inode->i_flags);
+       inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
        ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
        if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
            cpu_to_le32(EXT4_OS_HURD)) {
@@ -2767,6 +2782,13 @@ void ext4_read_inode(struct inode * inode)
        EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
        EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
 
+       inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
+       if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
+               if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
+                       inode->i_version |=
+                       (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
+       }
+
        if (S_ISREG(inode->i_mode)) {
                inode->i_op = &ext4_file_inode_operations;
                inode->i_fop = &ext4_file_operations;
@@ -2799,6 +2821,55 @@ bad_inode:
        return;
 }
 
+static int ext4_inode_blocks_set(handle_t *handle,
+                               struct ext4_inode *raw_inode,
+                               struct ext4_inode_info *ei)
+{
+       struct inode *inode = &(ei->vfs_inode);
+       u64 i_blocks = inode->i_blocks;
+       struct super_block *sb = inode->i_sb;
+       int err = 0;
+
+       if (i_blocks <= ~0U) {
+               /*
+                * i_blocks can be represnted in a 32 bit variable
+                * as multiple of 512 bytes
+                */
+               raw_inode->i_blocks_lo   = cpu_to_le32(i_blocks);
+               raw_inode->i_blocks_high = 0;
+               ei->i_flags &= ~EXT4_HUGE_FILE_FL;
+       } else if (i_blocks <= 0xffffffffffffULL) {
+               /*
+                * i_blocks can be represented in a 48 bit variable
+                * as multiple of 512 bytes
+                */
+               err = ext4_update_rocompat_feature(handle, sb,
+                                           EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
+               if (err)
+                       goto  err_out;
+               /* i_block is stored in the split  48 bit fields */
+               raw_inode->i_blocks_lo   = cpu_to_le32(i_blocks);
+               raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
+               ei->i_flags &= ~EXT4_HUGE_FILE_FL;
+       } else {
+               /*
+                * i_blocks should be represented in a 48 bit variable
+                * as multiple of  file system block size
+                */
+               err = ext4_update_rocompat_feature(handle, sb,
+                                           EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
+               if (err)
+                       goto  err_out;
+               ei->i_flags |= EXT4_HUGE_FILE_FL;
+               /* i_block is stored in file system block size */
+               i_blocks = i_blocks >> (inode->i_blkbits - 9);
+               raw_inode->i_blocks_lo   = cpu_to_le32(i_blocks);
+               raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
+       }
+err_out:
+       return err;
+}
+
 /*
  * Post the struct inode info into an on-disk inode location in the
  * buffer-cache.  This gobbles the caller's reference to the
@@ -2853,7 +2924,8 @@ static int ext4_do_update_inode(handle_t *handle,
        EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
        EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
 
-       raw_inode->i_blocks = cpu_to_le32(inode->i_blocks);
+       if (ext4_inode_blocks_set(handle, raw_inode, ei))
+               goto out_brelse;
        raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
        raw_inode->i_flags = cpu_to_le32(ei->i_flags);
        if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
@@ -2899,8 +2971,14 @@ static int ext4_do_update_inode(handle_t *handle,
        } else for (block = 0; block < EXT4_N_BLOCKS; block++)
                raw_inode->i_block[block] = ei->i_data[block];
 
-       if (ei->i_extra_isize)
+       raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
+       if (ei->i_extra_isize) {
+               if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
+                       raw_inode->i_version_hi =
+                       cpu_to_le32(inode->i_version >> 32);
                raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
+       }
+
 
        BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
        rc = ext4_journal_dirty_metadata(handle, bh);
@@ -3020,6 +3098,17 @@ int ext4_setattr(struct dentry *dentry, struct iattr *attr)
                ext4_journal_stop(handle);
        }
 
+       if (attr->ia_valid & ATTR_SIZE) {
+               if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)) {
+                       struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
+
+                       if (attr->ia_size > sbi->s_bitmap_maxbytes) {
+                               error = -EFBIG;
+                               goto err_out;
+                       }
+               }
+       }
+
        if (S_ISREG(inode->i_mode) &&
            attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) {
                handle_t *handle;
@@ -3116,6 +3205,9 @@ int ext4_mark_iloc_dirty(handle_t *handle,
 {
        int err = 0;
 
+       if (test_opt(inode->i_sb, I_VERSION))
+               inode_inc_iversion(inode);
+
        /* the do_update_inode consumes one bh->b_count */
        get_bh(iloc->bh);