ext4: Retry block allocation if we have free blocks left
authorAneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Tue, 9 Sep 2008 03:05:34 +0000 (23:05 -0400)
committerTheodore Ts'o <tytso@mit.edu>
Tue, 9 Sep 2008 03:05:34 +0000 (23:05 -0400)
When we truncate files, the meta-data blocks released are not reused
untill we commit the truncate transaction.  That means delayed get_block
request will return ENOSPC even if we have free blocks left.  Force a
journal commit and retry block allocation if we get ENOSPC with free
blocks left.

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Mingming Cao <cmm@us.ibm.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
fs/ext4/inode.c

index b567e71f5be99d91cf6881de40d2c0dace4f9938..f97b3478eb894c4fcdc075e9bf4308e27e5f31dd 100644 (file)
@@ -1634,6 +1634,7 @@ struct mpage_da_data {
        struct writeback_control *wbc;
        int io_done;
        long pages_written;
+       int retval;
 };
 
 /*
@@ -1820,6 +1821,24 @@ static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd,
        return;
 }
 
+static void ext4_print_free_blocks(struct inode *inode)
+{
+       struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
+       printk(KERN_EMERG "Total free blocks count %lld\n",
+                       ext4_count_free_blocks(inode->i_sb));
+       printk(KERN_EMERG "Free/Dirty block details\n");
+       printk(KERN_EMERG "free_blocks=%lld\n",
+                       percpu_counter_sum(&sbi->s_freeblocks_counter));
+       printk(KERN_EMERG "dirty_blocks=%lld\n",
+                       percpu_counter_sum(&sbi->s_dirtyblocks_counter));
+       printk(KERN_EMERG "Block reservation details\n");
+       printk(KERN_EMERG "i_reserved_data_blocks=%lu\n",
+                       EXT4_I(inode)->i_reserved_data_blocks);
+       printk(KERN_EMERG "i_reserved_meta_blocks=%lu\n",
+                       EXT4_I(inode)->i_reserved_meta_blocks);
+       return;
+}
+
 /*
  * mpage_da_map_blocks - go through given space
  *
@@ -1834,7 +1853,7 @@ static int  mpage_da_map_blocks(struct mpage_da_data *mpd)
        int err = 0;
        struct buffer_head new;
        struct buffer_head *lbh = &mpd->lbh;
-       sector_t next = lbh->b_blocknr;
+       sector_t next;
 
        /*
         * We consider only non-mapped and non-allocated blocks
@@ -1844,6 +1863,7 @@ static int  mpage_da_map_blocks(struct mpage_da_data *mpd)
        new.b_state = lbh->b_state;
        new.b_blocknr = 0;
        new.b_size = lbh->b_size;
+       next = lbh->b_blocknr;
        /*
         * If we didn't accumulate anything
         * to write simply return
@@ -1860,6 +1880,13 @@ static int  mpage_da_map_blocks(struct mpage_da_data *mpd)
                 */
                if (err == -EAGAIN)
                        return 0;
+
+               if (err == -ENOSPC &&
+                               ext4_count_free_blocks(mpd->inode->i_sb)) {
+                       mpd->retval = err;
+                       return 0;
+               }
+
                /*
                 * get block failure will cause us
                 * to loop in writepages. Because
@@ -1877,8 +1904,7 @@ static int  mpage_da_map_blocks(struct mpage_da_data *mpd)
                printk(KERN_EMERG "This should not happen.!! "
                                        "Data will be lost\n");
                if (err == -ENOSPC) {
-                       printk(KERN_CRIT "Total free blocks count %lld\n",
-                               ext4_count_free_blocks(mpd->inode->i_sb));
+                       ext4_print_free_blocks(mpd->inode);
                }
                /* invlaidate all the pages */
                ext4_da_block_invalidatepages(mpd, next,
@@ -2085,39 +2111,36 @@ static int __mpage_da_writepage(struct page *page,
  */
 static int mpage_da_writepages(struct address_space *mapping,
                               struct writeback_control *wbc,
-                              get_block_t get_block)
+                              struct mpage_da_data *mpd)
 {
-       struct mpage_da_data mpd;
        long to_write;
        int ret;
 
-       if (!get_block)
+       if (!mpd->get_block)
                return generic_writepages(mapping, wbc);
 
-       mpd.wbc = wbc;
-       mpd.inode = mapping->host;
-       mpd.lbh.b_size = 0;
-       mpd.lbh.b_state = 0;
-       mpd.lbh.b_blocknr = 0;
-       mpd.first_page = 0;
-       mpd.next_page = 0;
-       mpd.get_block = get_block;
-       mpd.io_done = 0;
-       mpd.pages_written = 0;
+       mpd->lbh.b_size = 0;
+       mpd->lbh.b_state = 0;
+       mpd->lbh.b_blocknr = 0;
+       mpd->first_page = 0;
+       mpd->next_page = 0;
+       mpd->io_done = 0;
+       mpd->pages_written = 0;
+       mpd->retval = 0;
 
        to_write = wbc->nr_to_write;
 
-       ret = write_cache_pages(mapping, wbc, __mpage_da_writepage, &mpd);
+       ret = write_cache_pages(mapping, wbc, __mpage_da_writepage, mpd);
 
        /*
         * Handle last extent of pages
         */
-       if (!mpd.io_done && mpd.next_page != mpd.first_page) {
-               if (mpage_da_map_blocks(&mpd) == 0)
-                       mpage_da_submit_io(&mpd);
+       if (!mpd->io_done && mpd->next_page != mpd->first_page) {
+               if (mpage_da_map_blocks(mpd) == 0)
+                       mpage_da_submit_io(mpd);
        }
 
-       wbc->nr_to_write = to_write - mpd.pages_written;
+       wbc->nr_to_write = to_write - mpd->pages_written;
        return ret;
 }
 
@@ -2357,6 +2380,7 @@ static int ext4_da_writepages(struct address_space *mapping,
 {
        handle_t *handle = NULL;
        loff_t range_start = 0;
+       struct mpage_da_data mpd;
        struct inode *inode = mapping->host;
        int needed_blocks, ret = 0, nr_to_writebump = 0;
        long to_write, pages_skipped = 0;
@@ -2390,6 +2414,9 @@ static int ext4_da_writepages(struct address_space *mapping,
        range_start =  wbc->range_start;
        pages_skipped = wbc->pages_skipped;
 
+       mpd.wbc = wbc;
+       mpd.inode = mapping->host;
+
 restart_loop:
        to_write = wbc->nr_to_write;
        while (!ret && to_write > 0) {
@@ -2413,11 +2440,17 @@ restart_loop:
                        dump_stack();
                        goto out_writepages;
                }
-
                to_write -= wbc->nr_to_write;
-               ret = mpage_da_writepages(mapping, wbc,
-                                         ext4_da_get_block_write);
+
+               mpd.get_block = ext4_da_get_block_write;
+               ret = mpage_da_writepages(mapping, wbc, &mpd);
+
                ext4_journal_stop(handle);
+
+               if (mpd.retval == -ENOSPC)
+                       jbd2_journal_force_commit_nested(sbi->s_journal);
+
+               /* reset the retry count */
                if (ret == MPAGE_DA_EXTENT_TAIL) {
                        /*
                         * got one extent now try with