ldiskfs/kernel_patches/patches/rhel7.3/ext4-disable-mb-cache.patch

   1 mbcache provides absolutely no value for Lustre xattrs (because
   2 they are unique and cannot be shared between files) and as we can
   3 see it has a noticable overhead in some cases. In the past there
   4 was a CONFIG_MBCACHE option that would allow it to be disabled,
   5 but this was removed in newer kernels, so we will need to patch
   6 ldiskfs to fix this.
   7
   8 Index: linux-stage/fs/ext4/ext4.h
   9 ===================================================================
  10 --- linux-stage.orig/fs/ext4/ext4.h
  11 +++ linux-stage/fs/ext4/ext4.h
  12 @@ -963,6 +963,7 @@ struct ext4_inode_info {
  13  /*
  14   * Mount flags set via mount options or defaults
  15   */
  16 +#define EXT4_MOUNT_NO_MBCACHE          0x00001 /* Disable mbcache */
  17  #define EXT4_MOUNT_DIRDATA             0x00002 /* Data in directory entries*/
  18  #define EXT4_MOUNT_GRPID               0x00004 /* Create files with directory's group */
  19  #define EXT4_MOUNT_DEBUG               0x00008 /* Some debugging messages */
  20 Index: linux-stage/fs/ext4/super.c
  21 ===================================================================
  22 --- linux-stage.orig/fs/ext4/super.c
  23 +++ linux-stage/fs/ext4/super.c
  24 @@ -1161,6 +1161,7 @@ enum {
  25         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
  26         Opt_inode_readahead_blks, Opt_journal_ioprio,
  27         Opt_dioread_nolock, Opt_dioread_lock,
  28 +       Opt_no_mbcache,
  29         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
  30         Opt_max_dir_size_kb, Opt_nojournal_checksum,
  31  };
  32 @@ -1238,6 +1239,7 @@ static const match_table_t tokens = {
  33         {Opt_discard, "discard"},
  34         {Opt_nodiscard, "nodiscard"},
  35         {Opt_init_itable, "init_itable=%u"},
  36 +       {Opt_no_mbcache, "no_mbcache"},
  37         {Opt_init_itable, "init_itable"},
  38         {Opt_noinit_itable, "noinit_itable"},
  39         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
  40 @@ -1400,6 +1402,7 @@ static const struct mount_opts {
  41         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
  42         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
  43         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
  44 +       {Opt_no_mbcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET},
  45         {Opt_commit, 0, MOPT_GTE0},
  46         {Opt_max_batch_time, 0, MOPT_GTE0},
  47         {Opt_min_batch_time, 0, MOPT_GTE0},
  48 Index: linux-stage/fs/ext4/xattr.c
  49 ===================================================================
  50 --- linux-stage.orig/fs/ext4/xattr.c
  51 +++ linux-stage/fs/ext4/xattr.c
  52 @@ -81,7 +81,8 @@
  53  # define ea_bdebug(bh, fmt, ...)       no_printk(fmt, ##__VA_ARGS__)
  54  #endif
  55
  56 -static void ext4_xattr_cache_insert(struct buffer_head *);
  57 +static void ext4_xattr_cache_insert(struct super_block *,
  58 +                                   struct buffer_head *);
  59  static struct buffer_head *ext4_xattr_cache_find(struct inode *,
  60                                                  struct ext4_xattr_header *,
  61                                                  struct mb_cache_entry **);
  62 @@ -405,7 +406,7 @@ bad_block:
  63                 error = -EIO;
  64                 goto cleanup;
  65         }
  66 -       ext4_xattr_cache_insert(bh);
  67 +       ext4_xattr_cache_insert(inode->i_sb, bh);
  68         entry = BFIRST(bh);
  69         error = ext4_xattr_find_entry(&entry, name_index, name, bh->b_size, 1,
  70                                       inode);
  71 @@ -569,7 +570,7 @@ ext4_xattr_block_list(struct dentry *den
  72                 error = -EIO;
  73                 goto cleanup;
  74         }
  75 -       ext4_xattr_cache_insert(bh);
  76 +       ext4_xattr_cache_insert(inode->i_sb, bh);
  77         error = ext4_xattr_list_entries(dentry, BFIRST(bh), buffer, buffer_size);
  78
  79  cleanup:
  80 @@ -667,7 +668,9 @@ ext4_xattr_release_block(handle_t *handl
  81         struct mb_cache_entry *ce = NULL;
  82         int error = 0;
  83
  84 -       ce = mb_cache_entry_get(ext4_xattr_cache, bh->b_bdev, bh->b_blocknr);
  85 +       if (!test_opt(inode->i_sb, NO_MBCACHE))
  86 +               ce = mb_cache_entry_get(ext4_xattr_cache, bh->b_bdev,
  87 +                                       bh->b_blocknr);
  88         BUFFER_TRACE(bh, "get_write_access");
  89         error = ext4_journal_get_write_access(handle, bh);
  90         if (error)
  91 @@ -1082,8 +1085,10 @@ ext4_xattr_block_set(handle_t *handle, s
  92  #define header(x) ((struct ext4_xattr_header *)(x))
  93
  94         if (s->base) {
  95 -               ce = mb_cache_entry_get(ext4_xattr_cache, bs->bh->b_bdev,
  96 -                                       bs->bh->b_blocknr);
  97 +               if (!test_opt(inode->i_sb, NO_MBCACHE))
  98 +                       ce = mb_cache_entry_get(ext4_xattr_cache,
  99 +                                               bs->bh->b_bdev,
 100 +                                               bs->bh->b_blocknr);
 101                 BUFFER_TRACE(bs->bh, "get_write_access");
 102                 error = ext4_journal_get_write_access(handle, bs->bh);
 103                 if (error)
 104 @@ -1101,7 +1106,7 @@ ext4_xattr_block_set(handle_t *handle, s
 105                                 if (!IS_LAST_ENTRY(s->first))
 106                                         ext4_xattr_rehash(header(s->base),
 107                                                           s->here);
 108 -                               ext4_xattr_cache_insert(bs->bh);
 109 +                               ext4_xattr_cache_insert(sb, bs->bh);
 110                         }
 111                         unlock_buffer(bs->bh);
 112                         if (error == -EIO)
 113 @@ -1185,7 +1190,8 @@ inserted:
 114                                 if (error)
 115                                         goto cleanup_dquot;
 116                         }
 117 -                       mb_cache_entry_release(ce);
 118 +                       if (ce)
 119 +                               mb_cache_entry_release(ce);
 120                         ce = NULL;
 121                 } else if (bs->bh && s->base == bs->bh->b_data) {
 122                         /* We were modifying this block in-place. */
 123 @@ -1238,7 +1244,7 @@ getblk_failed:
 124                         memcpy(new_bh->b_data, s->base, new_bh->b_size);
 125                         set_buffer_uptodate(new_bh);
 126                         unlock_buffer(new_bh);
 127 -                       ext4_xattr_cache_insert(new_bh);
 128 +                       ext4_xattr_cache_insert(sb, new_bh);
 129                         error = ext4_handle_dirty_xattr_block(handle,
 130                                                               inode, new_bh);
 131                         if (error)
 132 @@ -2022,12 +2028,15 @@ ext4_xattr_put_super(struct super_block
 133   * Returns 0, or a negative error number on failure.
 134   */
 135  static void
 136 -ext4_xattr_cache_insert(struct buffer_head *bh)
 137 +ext4_xattr_cache_insert(struct super_block *sb, struct buffer_head *bh)
 138  {
 139         __u32 hash = le32_to_cpu(BHDR(bh)->h_hash);
 140         struct mb_cache_entry *ce;
 141         int error;
 142
 143 +       if (test_opt(sb, NO_MBCACHE))
 144 +               return;
 145 +
 146         ce = mb_cache_entry_alloc(ext4_xattr_cache, GFP_NOFS);
 147         if (!ce) {
 148                 ea_bdebug(bh, "out of memory");
 149 @@ -2100,6 +2109,8 @@ ext4_xattr_cache_find(struct inode *inod
 150         __u32 hash = le32_to_cpu(header->h_hash);
 151         struct mb_cache_entry *ce;
 152
 153 +       if (test_opt(inode->i_sb, NO_MBCACHE))
 154 +               return NULL;
 155         if (!header->h_hash)
 156                 return NULL;  /* never share */
 157         ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);