add-indirection-to-metadata-block-read-paths

   1 Add indirection to metadata read paths
   2
   3 From: Abutalib Aghayev <agayev@cs.cmu.edu>
   4
   5 Change all metadata block reads to use jmap-aware function that first looks
   6 up the metadata block in the jmap.  If lookup is successful, the function
   7 reads the corresponding log block from the journal and copies it to the
   8 metadata block buffer head.  Otherwise, it reads the metadata block from
   9 the file system, just like standard jmap-unaware function.
  10
  11 Signed-off-by: Abutalib Aghayev <agayev@cs.cmu.edu>
  12
  13 ---
  14  fs/ext4/extents.c     |  3 ++-
  15  fs/ext4/ialloc.c      |  5 ++++-
  16  fs/ext4/indirect.c    |  3 ++-
  17  fs/ext4/inode.c       |  8 ++++++--
  18  fs/ext4/move_extent.c |  3 ++-
  19  fs/ext4/namei.c       |  8 +++++---
  20  fs/ext4/resize.c      |  4 +++-
  21  fs/jbd2/jmap.c        | 17 ++++++++++-------
  22  8 files changed, 34 insertions(+), 17 deletions(-)
  23
  24 diff --git a/fs/ext4/extents.c b/fs/ext4/extents.c
  25 index c930a0110fb4..aea6e67d9037 100644
  26 --- a/fs/ext4/extents.c
  27 +++ b/fs/ext4/extents.c
  28 @@ -517,6 +517,7 @@ __read_extent_tree_block(const char *function, unsigned int line,
  29  {
  30         struct buffer_head              *bh;
  31         int                             err;
  32 +       journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
  33
  34         bh = sb_getblk_gfp(inode->i_sb, pblk, __GFP_MOVABLE | GFP_NOFS);
  35         if (unlikely(!bh))
  36 @@ -524,7 +525,7 @@ __read_extent_tree_block(const char *function, unsigned int line,
  37
  38         if (!bh_uptodate_or_lock(bh)) {
  39                 trace_ext4_ext_load_extent(inode, pblk, _RET_IP_);
  40 -               err = bh_submit_read(bh);
  41 +               err = jbd2_bh_submit_read(journal, bh, __func__);
  42                 if (err < 0)
  43                         goto errout;
  44         }
  45 diff --git a/fs/ext4/ialloc.c b/fs/ext4/ialloc.c
  46 index 170421edfdfe..919c2d114fb5 100644
  47 --- a/fs/ext4/ialloc.c
  48 +++ b/fs/ext4/ialloc.c
  49 @@ -14,6 +14,7 @@
  50
  51  #include <linux/time.h>
  52  #include <linux/fs.h>
  53 +#include <linux/jbd2.h>
  54  #include <linux/stat.h>
  55  #include <linux/string.h>
  56  #include <linux/quotaops.h>
  57 @@ -160,6 +161,7 @@ ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
  58         struct buffer_head *bh = NULL;
  59         ext4_fsblk_t bitmap_blk;
  60         int err;
  61 +       journal_t *journal = EXT4_SB(sb)->s_journal;
  62
  63         desc = ext4_get_group_desc(sb, block_group, NULL);
  64         if (!desc)
  65 @@ -214,7 +216,8 @@ ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
  66         trace_ext4_load_inode_bitmap(sb, block_group);
  67         bh->b_end_io = ext4_end_bitmap_read;
  68         get_bh(bh);
  69 -       submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh);
  70 +       jbd2_submit_bh(journal, REQ_OP_READ, REQ_META | REQ_PRIO, bh, __func__);
  71 +
  72         wait_on_buffer(bh);
  73         if (!buffer_uptodate(bh)) {
  74                 put_bh(bh);
  75 diff --git a/fs/ext4/indirect.c b/fs/ext4/indirect.c
  76 index bc15c2c17633..21531ef4a182 100644
  77 --- a/fs/ext4/indirect.c
  78 +++ b/fs/ext4/indirect.c
  79 @@ -145,6 +145,7 @@ static Indirect *ext4_get_branch(struct inode *inode, int depth,
  80                                  Indirect chain[4], int *err)
  81  {
  82         struct super_block *sb = inode->i_sb;
  83 +       journal_t *journal = EXT4_SB(sb)->s_journal;
  84         Indirect *p = chain;
  85         struct buffer_head *bh;
  86         int ret = -EIO;
  87 @@ -162,7 +163,7 @@ static Indirect *ext4_get_branch(struct inode *inode, int depth,
  88                 }
  89
  90                 if (!bh_uptodate_or_lock(bh)) {
  91 -                       if (bh_submit_read(bh) < 0) {
  92 +                       if (jbd2_bh_submit_read(journal, bh, __func__) < 0) {
  93                                 put_bh(bh);
  94                                 goto failure;
  95                         }
  96 diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c
  97 index 9c064727ed62..0e2f7c3b499e 100644
  98 --- a/fs/ext4/inode.c
  99 +++ b/fs/ext4/inode.c
 100 @@ -989,13 +989,15 @@ struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
 101                                ext4_lblk_t block, int map_flags)
 102  {
 103         struct buffer_head *bh;
 104 +       journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
 105
 106         bh = ext4_getblk(handle, inode, block, map_flags);
 107         if (IS_ERR(bh))
 108                 return bh;
 109         if (!bh || buffer_uptodate(bh))
 110                 return bh;
 111 -       ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &bh);
 112 +       jbd2_ll_rw_block(journal, REQ_OP_READ, REQ_META | REQ_PRIO, 1, &bh,
 113 +                        __func__);
 114         wait_on_buffer(bh);
 115         if (buffer_uptodate(bh))
 116                 return bh;
 117 @@ -4201,6 +4203,7 @@ static int __ext4_get_inode_loc(struct inode *inode,
 118         struct super_block      *sb = inode->i_sb;
 119         ext4_fsblk_t            block;
 120         int                     inodes_per_block, inode_offset;
 121 +       journal_t               *journal = EXT4_SB(sb)->s_journal;
 122
 123         iloc->bh = NULL;
 124         if (!ext4_valid_inum(sb, inode->i_ino))
 125 @@ -4316,7 +4319,8 @@ static int __ext4_get_inode_loc(struct inode *inode,
 126                 trace_ext4_load_inode(inode);
 127                 get_bh(bh);
 128                 bh->b_end_io = end_buffer_read_sync;
 129 -               submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh);
 130 +               jbd2_submit_bh(journal, REQ_OP_READ, REQ_META | REQ_PRIO, bh,
 131 +                              __func__);
 132                 wait_on_buffer(bh);
 133                 if (!buffer_uptodate(bh)) {
 134                         EXT4_ERROR_INODE_BLOCK(inode, block,
 135 diff --git a/fs/ext4/move_extent.c b/fs/ext4/move_extent.c
 136 index 6fc14def0c70..b6c25638f5f4 100644
 137 --- a/fs/ext4/move_extent.c
 138 +++ b/fs/ext4/move_extent.c
 139 @@ -177,6 +177,7 @@ static int
 140  mext_page_mkuptodate(struct page *page, unsigned from, unsigned to)
 141  {
 142         struct inode *inode = page->mapping->host;
 143 +       journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
 144         sector_t block;
 145         struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE];
 146         unsigned int blocksize, block_start, block_end;
 147 @@ -225,7 +226,7 @@ mext_page_mkuptodate(struct page *page, unsigned from, unsigned to)
 148         for (i = 0; i < nr; i++) {
 149                 bh = arr[i];
 150                 if (!bh_uptodate_or_lock(bh)) {
 151 -                       err = bh_submit_read(bh);
 152 +                       err = jbd2_bh_submit_read(journal, bh, __func__);
 153                         if (err)
 154                                 return err;
 155                 }
 156 diff --git a/fs/ext4/namei.c b/fs/ext4/namei.c
 157 index 104f8bfba718..9c23616e7702 100644
 158 --- a/fs/ext4/namei.c
 159 +++ b/fs/ext4/namei.c
 160 @@ -1361,6 +1361,7 @@ static struct buffer_head * ext4_find_entry (struct inode *dir,
 161         struct buffer_head *bh_use[NAMEI_RA_SIZE];
 162         struct buffer_head *bh, *ret = NULL;
 163         ext4_lblk_t start, block, b;
 164 +       journal_t *journal;
 165         const u8 *name = d_name->name;
 166         int ra_max = 0;         /* Number of bh's in the readahead
 167                                    buffer, bh_use[] */
 168 @@ -1373,6 +1374,7 @@ static struct buffer_head * ext4_find_entry (struct inode *dir,
 169
 170         *res_dir = NULL;
 171         sb = dir->i_sb;
 172 +       journal = EXT4_SB(sb)->s_journal;
 173         namelen = d_name->len;
 174         if (namelen > EXT4_NAME_LEN)
 175                 return NULL;
 176 @@ -1449,9 +1451,9 @@ static struct buffer_head * ext4_find_entry (struct inode *dir,
 177                                 }
 178                                 bh_use[ra_max] = bh;
 179                                 if (bh)
 180 -                                       ll_rw_block(REQ_OP_READ,
 181 -                                                   REQ_META | REQ_PRIO,
 182 -                                                   1, &bh);
 183 +                                       jbd2_ll_rw_block(journal, REQ_OP_READ,
 184 +                                                        REQ_META | REQ_PRIO,
 185 +                                                        1, &bh, __func__);
 186                         }
 187                 }
 188                 if ((bh = bh_use[ra_ptr++]) == NULL)
 189 diff --git a/fs/ext4/resize.c b/fs/ext4/resize.c
 190 index cf681004b196..4e8711abc333 100644
 191 --- a/fs/ext4/resize.c
 192 +++ b/fs/ext4/resize.c
 193 @@ -1192,10 +1192,12 @@ static int ext4_add_new_descs(handle_t *handle, struct super_block *sb,
 194  static struct buffer_head *ext4_get_bitmap(struct super_block *sb, __u64 block)
 195  {
 196         struct buffer_head *bh = sb_getblk(sb, block);
 197 +       journal_t *journal = EXT4_SB(sb)->s_journal;
 198 +
 199         if (unlikely(!bh))
 200                 return NULL;
 201         if (!bh_uptodate_or_lock(bh)) {
 202 -               if (bh_submit_read(bh) < 0) {
 203 +               if (jbd2_bh_submit_read(journal, bh, __func__) < 0) {
 204                         brelse(bh);
 205                         return NULL;
 206                 }
 207 diff --git a/fs/jbd2/jmap.c b/fs/jbd2/jmap.c
 208 index 7d7b4eb389ed..8c844f65eeaa 100644
 209 --- a/fs/jbd2/jmap.c
 210 +++ b/fs/jbd2/jmap.c
 211 @@ -87,17 +87,19 @@ static int process_existing_mappings(journal_t *journal,
 212                         mappings[nr_new++] = mappings[i];
 213                         continue;
 214                 }
 215 +               /*
 216 +                * We are either deleting the entry because it was revoked, or
 217 +                * we are moving it to the live blocks list of this transaction.
 218 +                * In either case, we remove it from its existing list.
 219 +                */
 220 +               list_del(&je->list);
 221 +
 222                 if (je->revoked) {
 223                         rb_erase(&je->rb_node, &journal->j_jmap);
 224                         kmem_cache_free(jbd2_jmap_cache, je);
 225                 } else {
 226 -                       /*
 227 -                        * Delete jmap entry from the old transaction's list
 228 -                        * before adding it to the new transaction's list.
 229 -                        */
 230 -                       list_del(&je->list);
 231 -                       fill_entry(je, &mappings[i], t_idx, &ti->live_logblks);
 232                         trace_jbd2_jmap_replace(je, &mappings[i], t_idx);
 233 +                       fill_entry(je, &mappings[i], t_idx, &ti->live_logblks);
 234                 }
 235         }
 236         return nr_new;
 237 @@ -141,12 +143,13 @@ static void add_new_mappings(journal_t *journal, struct transaction_info *ti,
 238                         int t_idx, struct blk_mapping *mappings,
 239                         struct jmap_entry **new_entries, int nr_new)
 240  {
 241 -       struct rb_node **p = &journal->j_jmap.rb_node;
 242 +       struct rb_node **p;
 243         struct rb_node *parent = NULL;
 244         struct jmap_entry *je;
 245         int i;
 246
 247         for (i = 0; i < nr_new; ++i) {
 248 +               p = &journal->j_jmap.rb_node;
 249                 while (*p) {
 250                         parent = *p;
 251                         je = rb_entry(parent, struct jmap_entry, rb_node);