[PATCH] epoll: handle timeout overflow
[safe/jmp/linux-2.6] / fs / ext3 / inode.c
index 040eb28..b5177c9 100644 (file)
@@ -128,7 +128,7 @@ static unsigned long blocks_for_truncate(struct inode *inode)
        if (needed > EXT3_MAX_TRANS_DATA) 
                needed = EXT3_MAX_TRANS_DATA;
 
-       return EXT3_DATA_TRANS_BLOCKS + needed;
+       return EXT3_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
 }
 
 /* 
@@ -187,6 +187,8 @@ void ext3_delete_inode (struct inode * inode)
 {
        handle_t *handle;
 
+       truncate_inode_pages(&inode->i_data, 0);
+
        if (is_bad_inode(inode))
                goto no_delete;
 
@@ -455,12 +457,11 @@ static unsigned long ext3_find_near(struct inode *inode, Indirect *ind)
  *     @goal:  place to store the result.
  *
  *     Normally this function find the prefered place for block allocation,
- *     stores it in *@goal and returns zero. If the branch had been changed
- *     under us we return -EAGAIN.
+ *     stores it in *@goal and returns zero.
  */
 
-static int ext3_find_goal(struct inode *inode, long block, Indirect chain[4],
-                         Indirect *partial, unsigned long *goal)
+static unsigned long ext3_find_goal(struct inode *inode, long block,
+               Indirect chain[4], Indirect *partial)
 {
        struct ext3_block_alloc_info *block_i =  EXT3_I(inode)->i_block_alloc_info;
 
@@ -470,15 +471,10 @@ static int ext3_find_goal(struct inode *inode, long block, Indirect chain[4],
         */
        if (block_i && (block == block_i->last_alloc_logical_block + 1)
                && (block_i->last_alloc_physical_block != 0)) {
-               *goal = block_i->last_alloc_physical_block + 1;
-               return 0;
+               return block_i->last_alloc_physical_block + 1;
        }
 
-       if (verify_chain(chain, partial)) {
-               *goal = ext3_find_near(inode, partial);
-               return 0;
-       }
-       return -EAGAIN;
+       return ext3_find_near(inode, partial);
 }
 
 /**
@@ -582,12 +578,9 @@ static int ext3_alloc_branch(handle_t *handle, struct inode *inode,
  *     @where: location of missing link
  *     @num:   number of blocks we are adding
  *
- *     This function verifies that chain (up to the missing link) had not
- *     changed, fills the missing link and does all housekeeping needed in
+ *     This function fills the missing link and does all housekeeping needed in
  *     inode (->i_blocks, etc.). In case of success we end up with the full
- *     chain to new block and return 0. Otherwise (== chain had been changed)
- *     we free the new blocks (forgetting their buffer_heads, indeed) and
- *     return -EAGAIN.
+ *     chain to new block and return 0.
  */
 
 static int ext3_splice_branch(handle_t *handle, struct inode *inode, long block,
@@ -608,12 +601,6 @@ static int ext3_splice_branch(handle_t *handle, struct inode *inode, long block,
                if (err)
                        goto err_out;
        }
-       /* Verify that place we are splicing to is still there and vacant */
-
-       if (!verify_chain(chain, where-1) || *where->p)
-               /* Writer: end */
-               goto changed;
-
        /* That's it */
 
        *where->p = where->key;
@@ -657,26 +644,11 @@ static int ext3_splice_branch(handle_t *handle, struct inode *inode, long block,
        }
        return err;
 
-changed:
-       /*
-        * AKPM: if where[i].bh isn't part of the current updating
-        * transaction then we explode nastily.  Test this code path.
-        */
-       jbd_debug(1, "the chain changed: try again\n");
-       err = -EAGAIN;
-
 err_out:
        for (i = 1; i < num; i++) {
                BUFFER_TRACE(where[i].bh, "call journal_forget");
                ext3_journal_forget(handle, where[i].bh);
        }
-       /* For the normal collision cleanup case, we free up the blocks.
-        * On genuine filesystem errors we don't even think about doing
-        * that. */
-       if (err == -EAGAIN)
-               for (i = 0; i < num; i++)
-                       ext3_free_blocks(handle, inode, 
-                                        le32_to_cpu(where[i].key), 1);
        return err;
 }
 
@@ -708,7 +680,7 @@ ext3_get_block_handle(handle_t *handle, struct inode *inode, sector_t iblock,
        unsigned long goal;
        int left;
        int boundary = 0;
-       int depth = ext3_block_to_path(inode, iblock, offsets, &boundary);
+       const int depth = ext3_block_to_path(inode, iblock, offsets, &boundary);
        struct ext3_inode_info *ei = EXT3_I(inode);
 
        J_ASSERT(handle != NULL || create == 0);
@@ -716,54 +688,55 @@ ext3_get_block_handle(handle_t *handle, struct inode *inode, sector_t iblock,
        if (depth == 0)
                goto out;
 
-reread:
        partial = ext3_get_branch(inode, depth, offsets, chain, &err);
 
        /* Simplest case - block found, no allocation needed */
        if (!partial) {
                clear_buffer_new(bh_result);
-got_it:
-               map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
-               if (boundary)
-                       set_buffer_boundary(bh_result);
-               /* Clean up and exit */
-               partial = chain+depth-1; /* the whole chain */
-               goto cleanup;
+               goto got_it;
        }
 
        /* Next simple case - plain lookup or failed read of indirect block */
-       if (!create || err == -EIO) {
-cleanup:
+       if (!create || err == -EIO)
+               goto cleanup;
+
+       down(&ei->truncate_sem);
+
+       /*
+        * If the indirect block is missing while we are reading
+        * the chain(ext3_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) {
-                       BUFFER_TRACE(partial->bh, "call brelse");
                        brelse(partial->bh);
                        partial--;
                }
-               BUFFER_TRACE(bh_result, "returned");
-out:
-               return err;
+               partial = ext3_get_branch(inode, depth, offsets, chain, &err);
+               if (!partial) {
+                       up(&ei->truncate_sem);
+                       if (err)
+                               goto cleanup;
+                       clear_buffer_new(bh_result);
+                       goto got_it;
+               }
        }
 
        /*
-        * Indirect block might be removed by truncate while we were
-        * reading it. Handling of that case (forget what we've got and
-        * reread) is taken out of the main path.
-        */
-       if (err == -EAGAIN)
-               goto changed;
-
-       goal = 0;
-       down(&ei->truncate_sem);
-
-       /* lazy initialize the block allocation info here if necessary */
-       if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info)) {
+        * Okay, we need to do block allocation.  Lazily initialize the block
+        * allocation info here if necessary
+       */
+       if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info))
                ext3_init_block_alloc_info(inode);
-       }
 
-       if (ext3_find_goal(inode, iblock, chain, partial, &goal) < 0) {
-               up(&ei->truncate_sem);
-               goto changed;
-       }
+       goal = ext3_find_goal(inode, iblock, chain, partial);
 
        left = (chain + depth) - partial;
 
@@ -771,38 +744,45 @@ out:
         * Block out ext3_truncate while we alter the tree
         */
        err = ext3_alloc_branch(handle, inode, left, goal,
-                                       offsets+(partial-chain), partial);
+                               offsets + (partial - chain), partial);
 
-       /* The ext3_splice_branch call will free and forget any buffers
+       /*
+        * The ext3_splice_branch call will free and forget any buffers
         * on the new chain if there is a failure, but that risks using
         * up transaction credits, especially for bitmaps where the
         * credits cannot be returned.  Can we handle this somehow?  We
-        * may need to return -EAGAIN upwards in the worst case.  --sct */
+        * may need to return -EAGAIN upwards in the worst case.  --sct
+        */
        if (!err)
                err = ext3_splice_branch(handle, inode, iblock, chain,
                                         partial, left);
-       /* i_disksize growing is protected by truncate_sem
-        * don't forget to protect it if you're about to implement
-        * concurrent ext3_get_block() -bzzz */
+       /*
+        * i_disksize growing is protected by truncate_sem.  Don't forget to
+        * protect it if you're about to implement concurrent
+        * ext3_get_block() -bzzz
+       */
        if (!err && extend_disksize && inode->i_size > ei->i_disksize)
                ei->i_disksize = inode->i_size;
        up(&ei->truncate_sem);
-       if (err == -EAGAIN)
-               goto changed;
        if (err)
                goto cleanup;
 
        set_buffer_new(bh_result);
-       goto got_it;
-
-changed:
+got_it:
+       map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
+       if (boundary)
+               set_buffer_boundary(bh_result);
+       /* Clean up and exit */
+       partial = chain + depth - 1;    /* the whole chain */
+cleanup:
        while (partial > chain) {
-               jbd_debug(1, "buffer chain changed, retrying\n");
-               BUFFER_TRACE(partial->bh, "brelsing");
+               BUFFER_TRACE(partial->bh, "call brelse");
                brelse(partial->bh);
                partial--;
        }
-       goto reread;
+       BUFFER_TRACE(bh_result, "returned");
+out:
+       return err;
 }
 
 static int ext3_get_block(struct inode *inode, sector_t iblock,
@@ -866,12 +846,6 @@ get_block:
        return ret;
 }
 
-static int ext3_writepages_get_block(struct inode *inode, sector_t iblock,
-                       struct buffer_head *bh, int create)
-{
-       return ext3_direct_io_get_blocks(inode, iblock, 1, bh, create);
-}
-
 /*
  * `handle' can be NULL if create is zero
  */
@@ -1345,45 +1319,6 @@ out_fail:
        return ret;
 }
 
-static int
-ext3_writeback_writepage_helper(struct page *page,
-                               struct writeback_control *wbc)
-{
-       return block_write_full_page(page, ext3_get_block, wbc);
-}
-
-static int
-ext3_writeback_writepages(struct address_space *mapping,
-                               struct writeback_control *wbc)
-{
-       struct inode *inode = mapping->host;
-       handle_t *handle = NULL;
-       int err, ret = 0;
-
-       if (!mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
-               return ret;
-
-       handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
-       if (IS_ERR(handle)) {
-               ret = PTR_ERR(handle);
-               return ret;
-       }
-
-        ret = __mpage_writepages(mapping, wbc, ext3_writepages_get_block,
-                                       ext3_writeback_writepage_helper);
-
-       /*
-        * Need to reaquire the handle since ext3_writepages_get_block()
-        * can restart the handle
-        */
-       handle = journal_current_handle();
-
-       err = ext3_journal_stop(handle);
-       if (!ret)
-               ret = err;
-       return ret;
-}
-
 static int ext3_writeback_writepage(struct page *page,
                                struct writeback_control *wbc)
 {
@@ -1621,7 +1556,6 @@ static struct address_space_operations ext3_writeback_aops = {
        .readpage       = ext3_readpage,
        .readpages      = ext3_readpages,
        .writepage      = ext3_writeback_writepage,
-       .writepages     = ext3_writeback_writepages,
        .sync_page      = block_sync_page,
        .prepare_write  = ext3_prepare_write,
        .commit_write   = ext3_writeback_commit_write,
@@ -2731,7 +2665,7 @@ static int ext3_do_update_inode(handle_t *handle,
        } else for (block = 0; block < EXT3_N_BLOCKS; block++)
                raw_inode->i_block[block] = ei->i_data[block];
 
-       if (EXT3_INODE_SIZE(inode->i_sb) > EXT3_GOOD_OLD_INODE_SIZE)
+       if (ei->i_extra_isize)
                raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
 
        BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
@@ -2831,7 +2765,8 @@ int ext3_setattr(struct dentry *dentry, struct iattr *attr)
 
                /* (user+group)*(old+new) structure, inode write (sb,
                 * inode block, ? - but truncate inode update has it) */
-               handle = ext3_journal_start(inode, 4*EXT3_QUOTA_INIT_BLOCKS+3);
+               handle = ext3_journal_start(inode, 2*(EXT3_QUOTA_INIT_BLOCKS(inode->i_sb)+
+                                       EXT3_QUOTA_DEL_BLOCKS(inode->i_sb))+3);
                if (IS_ERR(handle)) {
                        error = PTR_ERR(handle);
                        goto err_out;
@@ -2929,7 +2864,7 @@ static int ext3_writepage_trans_blocks(struct inode *inode)
 #ifdef CONFIG_QUOTA
        /* We know that structure was already allocated during DQUOT_INIT so
         * we will be updating only the data blocks + inodes */
-       ret += 2*EXT3_QUOTA_TRANS_BLOCKS;
+       ret += 2*EXT3_QUOTA_TRANS_BLOCKS(inode->i_sb);
 #endif
 
        return ret;