ldiskfs/kernel_patches/patches/ext4-journal-callback-rhel5.patch

   1 Index: linux-stage/fs/ext4/ext4_jbd2.h
   2 ===================================================================
   3 --- linux-stage.orig/fs/ext4/ext4_jbd2.h
   4 +++ linux-stage/fs/ext4/ext4_jbd2.h
   5 @@ -106,6 +106,80 @@
   6  #define EXT4_MAXQUOTAS_INIT_BLOCKS(sb) (MAXQUOTAS*EXT4_QUOTA_INIT_BLOCKS(sb))
   7  #define EXT4_MAXQUOTAS_DEL_BLOCKS(sb) (MAXQUOTAS*EXT4_QUOTA_DEL_BLOCKS(sb))
   8
   9 +/**
  10 + *   struct ext4_journal_cb_entry - Base structure for callback information.
  11 + *
  12 + *   This struct is a 'seed' structure for a using with your own callback
  13 + *   structs. If you are using callbacks you must allocate one of these
  14 + *   or another struct of your own definition which has this struct
  15 + *   as it's first element and pass it to ext4_journal_callback_add().
  16 + */
  17 +struct ext4_journal_cb_entry {
  18 +       /* list information for other callbacks attached to the same handle */
  19 +       struct list_head jce_list;
  20 +
  21 +       /*  Function to call with this callback structure */
  22 +       void (*jce_func)(struct super_block *sb,
  23 +                        struct ext4_journal_cb_entry *jce, int error);
  24 +
  25 +       /* user data goes here */
  26 +};
  27 +
  28 +/**
  29 + * ext4_journal_callback_add: add a function to call after transaction commit
  30 + * @handle: active journal transaction handle to register callback on
  31 + * @func: callback function to call after the transaction has committed:
  32 + *        @sb: superblock of current filesystem for transaction
  33 + *        @jce: returned journal callback data
  34 + *        @rc: journal state at commit (0 = transaction committed properly)
  35 + * @jce: journal callback data (internal and function private data struct)
  36 + *
  37 + * The registered function will be called in the context of the journal thread
  38 + * after the transaction for which the handle was created has completed.
  39 + *
  40 + * No locks are held when the callback function is called, so it is safe to
  41 + * call blocking functions from within the callback, but the callback should
  42 + * not block or run for too long, or the filesystem will be blocked waiting for
  43 + * the next transaction to commit. No journaling functions can be used, or
  44 + * there is a risk of deadlock.
  45 + *
  46 + * There is no guaranteed calling order of multiple registered callbacks on
  47 + * the same transaction.
  48 + */
  49 +static inline void ext4_journal_callback_add(handle_t *handle,
  50 +                       void (*func)(struct super_block *sb,
  51 +                                    struct ext4_journal_cb_entry *jce,
  52 +                                    int rc),
  53 +                       struct ext4_journal_cb_entry *jce)
  54 +{
  55 +       struct ext4_sb_info *sbi =
  56 +                       EXT4_SB(handle->h_transaction->t_journal->j_private);
  57 +
  58 +       /* Add the jce to transaction's private list */
  59 +       jce->jce_func = func;
  60 +       spin_lock(&sbi->s_md_lock);
  61 +       list_add_tail(&jce->jce_list, &handle->h_transaction->t_private_list);
  62 +       spin_unlock(&sbi->s_md_lock);
  63 +}
  64 +
  65 +/**
  66 + * ext4_journal_callback_del: delete a registered callback
  67 + * @handle: active journal transaction handle on which callback was registered
  68 + * @jce: registered journal callback entry to unregister
  69 + */
  70 +static inline void ext4_journal_callback_del(handle_t *handle,
  71 +                                            struct ext4_journal_cb_entry *jce)
  72 +{
  73 +       struct ext4_sb_info *sbi =
  74 +                       EXT4_SB(handle->h_transaction->t_journal->j_private);
  75 +
  76 +       spin_lock(&sbi->s_md_lock);
  77 +       list_del_init(&jce->jce_list);
  78 +       spin_unlock(&sbi->s_md_lock);
  79 +}
  80 +
  81 +#define HAVE_EXT4_JOURNAL_CALLBACK_ADD
  82 +
  83  int
  84  ext4_mark_iloc_dirty(handle_t *handle,
  85                      struct inode *inode,
  86 Index: linux-stage/fs/ext4/mballoc.c
  87 ===================================================================
  88 --- linux-stage.orig/fs/ext4/mballoc.c
  89 +++ linux-stage/fs/ext4/mballoc.c
  90 @@ -21,6 +21,7 @@
  91   * mballoc.c contains the multiblocks allocation routines
  92   */
  93
  94 +#include "ext4_jbd2.h"
  95  #include "mballoc.h"
  96  #include <linux/debugfs.h>
  97
  98 @@ -335,14 +336,12 @@
  99   */
 100  static struct kmem_cache *ext4_pspace_cachep;
 101  static struct kmem_cache *ext4_ac_cachep;
 102 -static struct kmem_cache *ext4_free_ext_cachep;
 103 +static struct kmem_cache *ext4_free_data_cachep;
 104  static int ext4_mb_generate_from_pa(struct super_block *sb, void *bitmap,
 105                                         ext4_group_t group);
 106  static void ext4_mb_generate_from_freelist(struct super_block *sb, void *bitmap,
 107                                                 ext4_group_t group);
 108 -static void release_blocks_on_commit(journal_t *journal, transaction_t *txn);
 109 -
 110 -
 111 +static void ext4_free_data_callback(struct super_block *sb, struct ext4_journal_cb_entry *jce, int error);
 112
 113  static inline void *mb_correct_addr_and_bit(int *bit, void *addr)
 114  {
 115 @@ -2942,8 +2941,6 @@ int ext4_mb_init(struct super_block *sb,
 116
 117         ext4_mb_history_init(sb);
 118
 119 -       if (sbi->s_journal)
 120 -               sbi->s_journal->j_commit_callback = release_blocks_on_commit;
 121         return 0;
 122  }
 123
 124 @@ -3032,46 +3029,42 @@ int ext4_mb_release(struct super_block *
 125   * This function is called by the jbd2 layer once the commit has finished,
 126   * so we know we can free the blocks that were released with that commit.
 127   */
 128 -static void release_blocks_on_commit(journal_t *journal, transaction_t *txn)
 129 +static void ext4_free_data_callback(struct super_block *sb,
 130 +                                   struct ext4_journal_cb_entry *jce,
 131 +                                   int rc)
 132  {
 133 -       struct super_block *sb = journal->j_private;
 134 +       struct ext4_free_data *entry = (struct ext4_free_data *)jce;
 135         struct ext4_buddy e4b;
 136         struct ext4_group_info *db;
 137         int err, count = 0, count2 = 0;
 138 -       struct ext4_free_data *entry;
 139 -       struct list_head *l, *ltmp;
 140
 141 -       list_for_each_safe(l, ltmp, &txn->t_private_list) {
 142 -               entry = list_entry(l, struct ext4_free_data, list);
 143 -
 144 -               mb_debug(1, "gonna free %u blocks in group %u (0x%p):",
 145 -                        entry->count, entry->group, entry);
 146 -
 147 -               err = ext4_mb_load_buddy(sb, entry->group, &e4b);
 148 -               /* we expect to find existing buddy because it's pinned */
 149 -               BUG_ON(err != 0);
 150 -
 151 -               db = e4b.bd_info;
 152 -               /* there are blocks to put in buddy to make them really free */
 153 -               count += entry->count;
 154 -               count2++;
 155 -               ext4_lock_group(sb, entry->group);
 156 -               /* Take it out of per group rb tree */
 157 -               rb_erase(&entry->node, &(db->bb_free_root));
 158 -               mb_free_blocks(NULL, &e4b, entry->start_blk, entry->count);
 159 -
 160 -               if (!db->bb_free_root.rb_node) {
 161 -                       /* No more items in the per group rb tree
 162 -                        * balance refcounts from ext4_mb_free_metadata()
 163 -                        */
 164 -                       page_cache_release(e4b.bd_buddy_page);
 165 -                       page_cache_release(e4b.bd_bitmap_page);
 166 -               }
 167 -               ext4_unlock_group(sb, entry->group);
 168 +       mb_debug(1, "gonna free %u blocks in group %u (0x%p):",
 169 +                entry->efd_count, entry->efd_group, entry);
 170
 171 -               kmem_cache_free(ext4_free_ext_cachep, entry);
 172 -               ext4_mb_unload_buddy(&e4b);
 173 +       err = ext4_mb_load_buddy(sb, entry->efd_group, &e4b);
 174 +       /* we expect to find existing buddy because it's pinned */
 175 +       BUG_ON(err != 0);
 176 +
 177 +       db = e4b.bd_info;
 178 +       /* there are blocks to put in buddy to make them really free */
 179 +       count += entry->efd_count;
 180 +       count2++;
 181 +       ext4_lock_group(sb, entry->efd_group);
 182 +       /* Take it out of per group rb tree */
 183 +       rb_erase(&entry->efd_node, &(db->bb_free_root));
 184 +       mb_free_blocks(NULL, &e4b, entry->efd_start_blk, entry->efd_count);
 185 +
 186 +       if (!db->bb_free_root.rb_node) {
 187 +               /* No more items in the per group rb tree
 188 +                * balance refcounts from ext4_mb_free_metadata()
 189 +                */
 190 +               page_cache_release(e4b.bd_buddy_page);
 191 +               page_cache_release(e4b.bd_bitmap_page);
 192         }
 193 +       ext4_unlock_group(sb, entry->efd_group);
 194 +
 195 +       kmem_cache_free(ext4_free_data_cachep, entry);
 196 +       ext4_mb_unload_buddy(&e4b);
 197
 198         mb_debug(1, "freed %u blocks in %u structures\n", count, count2);
 199  }
 200 @@ -3123,22 +3116,24 @@ int __init init_ext4_mballoc(void)
 201                 kmem_cache_create("ext4_alloc_context",
 202                                      sizeof(struct ext4_allocation_context),
 203                                      0, SLAB_RECLAIM_ACCOUNT, NULL, NULL);
 204 -       if (ext4_ac_cachep == NULL) {
 205 -               kmem_cache_destroy(ext4_pspace_cachep);
 206 -               return -ENOMEM;
 207 -       }
 208 +       if (ext4_ac_cachep == NULL)
 209 +               goto out_err;
 210
 211 -       ext4_free_ext_cachep =
 212 -               kmem_cache_create("ext4_free_block_extents",
 213 +       ext4_free_data_cachep =
 214 +               kmem_cache_create("ext4_free_data",
 215                                      sizeof(struct ext4_free_data),
 216                                      0, SLAB_RECLAIM_ACCOUNT, NULL, NULL);
 217 -       if (ext4_free_ext_cachep == NULL) {
 218 -               kmem_cache_destroy(ext4_pspace_cachep);
 219 -               kmem_cache_destroy(ext4_ac_cachep);
 220 -               return -ENOMEM;
 221 -       }
 222 +       if (ext4_free_data_cachep == NULL)
 223 +               goto out1_err;
 224 +
 225         ext4_create_debugfs_entry();
 226         return 0;
 227 +
 228 +out1_err:
 229 +       kmem_cache_destroy(ext4_ac_cachep);
 230 +out_err:
 231 +       kmem_cache_destroy(ext4_pspace_cachep);
 232 +       return -ENOMEM;
 233  }
 234
 235  void exit_ext4_mballoc(void)
 236 @@ -3150,7 +3145,7 @@ void exit_ext4_mballoc(void)
 237         rcu_barrier();
 238         kmem_cache_destroy(ext4_pspace_cachep);
 239         kmem_cache_destroy(ext4_ac_cachep);
 240 -       kmem_cache_destroy(ext4_free_ext_cachep);
 241 +       kmem_cache_destroy(ext4_free_data_cachep);
 242         ext4_remove_debugfs_entry();
 243  }
 244
 245 @@ -3688,8 +3683,8 @@ static void ext4_mb_generate_from_freeli
 246         n = rb_first(&(grp->bb_free_root));
 247
 248         while (n) {
 249 -               entry = rb_entry(n, struct ext4_free_data, node);
 250 -               mb_set_bits(bitmap, entry->start_blk, entry->count);
 251 +               entry = rb_entry(n, struct ext4_free_data, efd_node);
 252 +               mb_set_bits(bitmap, entry->efd_start_blk, entry->efd_count);
 253                 n = rb_next(n);
 254         }
 255         return;
 256 @@ -4974,11 +4969,11 @@ out3:
 257   * AND the blocks are associated with the same group.
 258   */
 259  static int can_merge(struct ext4_free_data *entry1,
 260 -                       struct ext4_free_data *entry2)
 261 +                    struct ext4_free_data *entry2)
 262  {
 263 -       if ((entry1->t_tid == entry2->t_tid) &&
 264 -           (entry1->group == entry2->group) &&
 265 -           ((entry1->start_blk + entry1->count) == entry2->start_blk))
 266 +       if ((entry1->efd_tid == entry2->efd_tid) &&
 267 +           (entry1->efd_group == entry2->efd_group) &&
 268 +           ((entry1->efd_start_blk + entry1->efd_count) == entry2->efd_start_blk))
 269                 return 1;
 270         return 0;
 271  }
 272 @@ -4991,7 +4986,6 @@ ext4_mb_free_metadata(handle_t *handle,
 273         struct ext4_free_data *entry;
 274         struct ext4_group_info *db = e4b->bd_info;
 275         struct super_block *sb = e4b->bd_sb;
 276 -       struct ext4_sb_info *sbi = EXT4_SB(sb);
 277         struct rb_node **n = &db->bb_free_root.rb_node, *node;
 278         struct rb_node *parent = NULL, *new_node;
 279
 280 @@ -4999,8 +4993,8 @@ ext4_mb_free_metadata(handle_t *handle,
 281         BUG_ON(e4b->bd_bitmap_page == NULL);
 282         BUG_ON(e4b->bd_buddy_page == NULL);
 283
 284 -       new_node = &new_entry->node;
 285 -       block = new_entry->start_blk;
 286 +       new_node = &new_entry->efd_node;
 287 +       block = new_entry->efd_start_blk;
 288
 289         if (!*n) {
 290                 /* first free block exent. We need to
 291 @@ -5013,15 +5007,15 @@ ext4_mb_free_metadata(handle_t *handle,
 292         }
 293         while (*n) {
 294                 parent = *n;
 295 -               entry = rb_entry(parent, struct ext4_free_data, node);
 296 -               if (block < entry->start_blk)
 297 +               entry = rb_entry(parent, struct ext4_free_data, efd_node);
 298 +               if (block < entry->efd_start_blk)
 299                         n = &(*n)->rb_left;
 300 -               else if (block >= (entry->start_blk + entry->count))
 301 +               else if (block >= (entry->efd_start_blk + entry->efd_count))
 302                         n = &(*n)->rb_right;
 303                 else {
 304                         ext4_grp_locked_error(sb, e4b->bd_group, __func__,
 305                                         "Double free of blocks %d (%d %d)",
 306 -                                       block, entry->start_blk, entry->count);
 307 +                                       block, entry->efd_start_blk, entry->efd_count);
 308                         return 0;
 309                 }
 310         }
 311 @@ -5032,34 +5026,29 @@ ext4_mb_free_metadata(handle_t *handle,
 312         /* Now try to see the extent can be merged to left and right */
 313         node = rb_prev(new_node);
 314         if (node) {
 315 -               entry = rb_entry(node, struct ext4_free_data, node);
 316 +               entry = rb_entry(node, struct ext4_free_data, efd_node);
 317                 if (can_merge(entry, new_entry)) {
 318 -                       new_entry->start_blk = entry->start_blk;
 319 -                       new_entry->count += entry->count;
 320 +                       new_entry->efd_start_blk = entry->efd_start_blk;
 321 +                       new_entry->efd_count += entry->efd_count;
 322                         rb_erase(node, &(db->bb_free_root));
 323 -                       spin_lock(&sbi->s_md_lock);
 324 -                       list_del(&entry->list);
 325 -                       spin_unlock(&sbi->s_md_lock);
 326 -                       kmem_cache_free(ext4_free_ext_cachep, entry);
 327 +                       ext4_journal_callback_del(handle, &entry->efd_jce);
 328 +                       kmem_cache_free(ext4_free_data_cachep, entry);
 329                 }
 330         }
 331
 332         node = rb_next(new_node);
 333         if (node) {
 334 -               entry = rb_entry(node, struct ext4_free_data, node);
 335 +               entry = rb_entry(node, struct ext4_free_data, efd_node);
 336                 if (can_merge(new_entry, entry)) {
 337 -                       new_entry->count += entry->count;
 338 +                       new_entry->efd_count += entry->efd_count;
 339                         rb_erase(node, &(db->bb_free_root));
 340 -                       spin_lock(&sbi->s_md_lock);
 341 -                       list_del(&entry->list);
 342 -                       spin_unlock(&sbi->s_md_lock);
 343 -                       kmem_cache_free(ext4_free_ext_cachep, entry);
 344 +                       ext4_journal_callback_del(handle, &entry->efd_jce);
 345 +                       kmem_cache_free(ext4_free_data_cachep, entry);
 346                 }
 347         }
 348         /* Add the extent to transaction's private list */
 349 -       spin_lock(&sbi->s_md_lock);
 350 -       list_add(&new_entry->list, &handle->h_transaction->t_private_list);
 351 -       spin_unlock(&sbi->s_md_lock);
 352 +       ext4_journal_callback_add(handle, ext4_free_data_callback,
 353 +                                 &new_entry->efd_jce);
 354         return 0;
 355  }
 356
 357 @@ -5191,11 +5180,11 @@ do_more:
 358                  * blocks being freed are metadata. these blocks shouldn't
 359                  * be used until this transaction is committed
 360                  */
 361 -               new_entry  = kmem_cache_alloc(ext4_free_ext_cachep, GFP_NOFS);
 362 -               new_entry->start_blk = bit;
 363 -               new_entry->group  = block_group;
 364 -               new_entry->count = count;
 365 -               new_entry->t_tid = handle->h_transaction->t_tid;
 366 +               new_entry = kmem_cache_alloc(ext4_free_data_cachep, GFP_NOFS);
 367 +               new_entry->efd_start_blk = bit;
 368 +               new_entry->efd_group  = block_group;
 369 +               new_entry->efd_count = count;
 370 +               new_entry->efd_tid = handle->h_transaction->t_tid;
 371
 372                 ext4_lock_group(sb, block_group);
 373                 mb_clear_bits(bitmap_bh->b_data, bit, count);
 374 Index: linux-stage/fs/ext4/mballoc.h
 375 ===================================================================
 376 --- linux-stage.orig/fs/ext4/mballoc.h
 377 +++ linux-stage/fs/ext4/mballoc.h
 378 @@ -107,23 +107,24 @@ extern u8 mb_enable_debug;
 379   */
 380  #define MB_DEFAULT_GROUP_PREALLOC      512
 381
 382 -
 383  struct ext4_free_data {
 384 -       /* this links the free block information from group_info */
 385 -       struct rb_node node;
 386 +       /* MUST be the first member */
 387 +       struct ext4_journal_cb_entry    efd_jce;
 388
 389 -       /* this links the free block information from ext4_sb_info */
 390 -       struct list_head list;
 391 +       /* ext4_free_data private data starts from here */
 392 +
 393 +       /* this links the free block information from group_info */
 394 +       struct rb_node          efd_node;
 395
 396         /* group which free block extent belongs */
 397 -       ext4_group_t group;
 398 +       ext4_group_t            efd_group;
 399
 400         /* free block extent */
 401 -       ext4_grpblk_t start_blk;
 402 -       ext4_grpblk_t count;
 403 +       ext4_grpblk_t           efd_start_blk;
 404 +       ext4_grpblk_t           efd_count;
 405
 406         /* transaction which freed this extent */
 407 -       tid_t   t_tid;
 408 +       tid_t                   efd_tid;
 409  };
 410
 411  struct ext4_prealloc_space {
 412 Index: linux-stage/fs/ext4/super.c
 413 ===================================================================
 414 --- linux-stage.orig/fs/ext4/super.c
 415 +++ linux-stage/fs/ext4/super.c
 416 @@ -304,6 +304,23 @@ void ext4_journal_abort_handle(const cha
 417
 418  EXPORT_SYMBOL(ext4_journal_abort_handle);
 419
 420 +static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 421 +{
 422 +       struct super_block              *sb = journal->j_private;
 423 +       struct ext4_sb_info             *sbi = EXT4_SB(sb);
 424 +       int                             error = is_journal_aborted(journal);
 425 +       struct ext4_journal_cb_entry    *jce, *tmp;
 426 +
 427 +       spin_lock(&sbi->s_md_lock);
 428 +       list_for_each_entry_safe(jce, tmp, &txn->t_private_list, jce_list) {
 429 +               list_del_init(&jce->jce_list);
 430 +               spin_unlock(&sbi->s_md_lock);
 431 +               jce->jce_func(sb, jce, error);
 432 +               spin_lock(&sbi->s_md_lock);
 433 +       }
 434 +       spin_unlock(&sbi->s_md_lock);
 435 +}
 436 +
 437  /* Deal with the reporting of failure conditions on a filesystem such as
 438   * inconsistencies detected or read IO failures.
 439   *
 440 @@ -2997,6 +3014,8 @@ static int ext4_fill_super(struct super_
 441         }
 442         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
 443
 444 +       sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
 445 +
 446  no_journal:
 447
 448         if (test_opt(sb, NOBH)) {