ext4-Retry-block-reservation.patch

   1 ext4: Retry block reservation
   2
   3 From: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
   4
   5 During block reservation if we don't have enough blocks left, retry
   6 block reservation with smaller block counts.  This makes sure we try
   7 fallocate and DIO with smaller request size and don't fail early.  The
   8 delayed allocation reservation cannot try with smaller block count. So
   9 retry block reservation to handle temporary disk full conditions.  Also
  10 print free blocks details if we fail block allocation during writepages.
  11
  12 Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
  13 Signed-off-by: Mingming Cao <cmm@us.ibm.com>
  14 Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
  15 ---
  16  fs/ext4/balloc.c  |    8 +++++++-
  17  fs/ext4/inode.c   |   14 +++++++++++---
  18  fs/ext4/mballoc.c |    7 ++++++-
  19  3 files changed, 24 insertions(+), 5 deletions(-)
  20
  21 Index: linux-2.6.27-rc3/fs/ext4/balloc.c
  22 ===================================================================
  23 --- linux-2.6.27-rc3.orig/fs/ext4/balloc.c      2008-08-27 09:45:05.000000000 -0700
  24 +++ linux-2.6.27-rc3/fs/ext4/balloc.c   2008-08-27 13:54:26.000000000 -0700
  25 @@ -1736,10 +1736,16 @@
  26                 /*
  27                  * With delalloc we already reserved the blocks
  28                  */
  29 -               if (ext4_claim_free_blocks(sbi, *count)) {
  30 +               while (*count && ext4_claim_free_blocks(sbi, *count)) {
  31 +                       /* let others to free the space */
  32 +                       yield();
  33 +                       *count = *count >> 1;
  34 +               }
  35 +               if (!*count) {
  36                         *errp = -ENOSPC;
  37                         return 0;       /*return with ENOSPC error */
  38                 }
  39 +               num = *count;
  40         }
  41         /*
  42          * Check quota for allocation of this block.
  43 Index: linux-2.6.27-rc3/fs/ext4/inode.c
  44 ===================================================================
  45 --- linux-2.6.27-rc3.orig/fs/ext4/inode.c       2008-08-27 09:45:05.000000000 -0700
  46 +++ linux-2.6.27-rc3/fs/ext4/inode.c    2008-08-27 13:54:26.000000000 -0700
  47 @@ -1521,6 +1521,7 @@
  48
  49  static int ext4_da_reserve_space(struct inode *inode, int nrblocks)
  50  {
  51 +       int retries = 0;
  52         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
  53         unsigned long md_needed, mdblocks, total = 0;
  54
  55 @@ -1529,6 +1530,7 @@
  56          * in order to allocate nrblocks
  57          * worse case is one extent per block
  58          */
  59 +repeat:
  60         spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
  61         total = EXT4_I(inode)->i_reserved_data_blocks + nrblocks;
  62         mdblocks = ext4_calc_metadata_amount(inode, total);
  63 @@ -1539,6 +1541,10 @@
  64
  65         if (ext4_claim_free_blocks(sbi, total)) {
  66                 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
  67 +               if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
  68 +                       yield();
  69 +                       goto repeat;
  70 +               }
  71                 return -ENOSPC;
  72         }
  73         EXT4_I(inode)->i_reserved_data_blocks += nrblocks;
  74 @@ -1825,20 +1831,18 @@
  75  static int  mpage_da_map_blocks(struct mpage_da_data *mpd)
  76  {
  77         int err = 0;
  78 +       struct buffer_head new;
  79         struct buffer_head *lbh = &mpd->lbh;
  80         sector_t next = lbh->b_blocknr;
  81 -       struct buffer_head new;
  82
  83         /*
  84          * We consider only non-mapped and non-allocated blocks
  85          */
  86         if (buffer_mapped(lbh) && !buffer_delay(lbh))
  87                 return 0;
  88 -
  89         new.b_state = lbh->b_state;
  90         new.b_blocknr = 0;
  91         new.b_size = lbh->b_size;
  92 -
  93         /*
  94          * If we didn't accumulate anything
  95          * to write simply return
  96 @@ -1871,6 +1875,10 @@
  97                                   lbh->b_size >> mpd->inode->i_blkbits, err);
  98                 printk(KERN_EMERG "This should not happen.!! "
  99                                         "Data will be lost\n");
 100 +               if (err == -ENOSPC) {
 101 +                       printk(KERN_CRIT "Total free blocks count %lld\n",
 102 +                               ext4_count_free_blocks(mpd->inode->i_sb));
 103 +               }
 104                 /* invlaidate all the pages */
 105                 ext4_da_block_invalidatepages(mpd, next,
 106                                 lbh->b_size >> mpd->inode->i_blkbits);
 107 Index: linux-2.6.27-rc3/fs/ext4/mballoc.c
 108 ===================================================================
 109 --- linux-2.6.27-rc3.orig/fs/ext4/mballoc.c     2008-08-27 09:45:05.000000000 -0700
 110 +++ linux-2.6.27-rc3/fs/ext4/mballoc.c  2008-08-27 13:54:26.000000000 -0700
 111 @@ -4394,7 +4394,12 @@
 112                 /*
 113                  * With delalloc we already reserved the blocks
 114                  */
 115 -               if (ext4_claim_free_blocks(sbi, ar->len)) {
 116 +               while (ar->len && ext4_claim_free_blocks(sbi, ar->len)) {
 117 +                       /* let others to free the space */
 118 +                       yield();
 119 +                       ar->len = ar->len >> 1;
 120 +               }
 121 +               if (!ar->len) {
 122                         *errp = -ENOSPC;
 123                         return 0;
 124                 }