SAFE public projects git trees. - safe/jmp/linux-2.6/blob - fs/reiserfs/journal.c

   1 /*
   2 ** Write ahead logging implementation copyright Chris Mason 2000
   3 **
   4 ** The background commits make this code very interelated, and
   5 ** overly complex.  I need to rethink things a bit....The major players:
   6 **
   7 ** journal_begin -- call with the number of blocks you expect to log.
   8 **                  If the current transaction is too
   9 **                  old, it will block until the current transaction is
  10 **                  finished, and then start a new one.
  11 **                  Usually, your transaction will get joined in with
  12 **                  previous ones for speed.
  13 **
  14 ** journal_join  -- same as journal_begin, but won't block on the current
  15 **                  transaction regardless of age.  Don't ever call
  16 **                  this.  Ever.  There are only two places it should be
  17 **                  called from, and they are both inside this file.
  18 **
  19 ** journal_mark_dirty -- adds blocks into this transaction.  clears any flags
  20 **                       that might make them get sent to disk
  21 **                       and then marks them BH_JDirty.  Puts the buffer head
  22 **                       into the current transaction hash.
  23 **
  24 ** journal_end -- if the current transaction is batchable, it does nothing
  25 **                   otherwise, it could do an async/synchronous commit, or
  26 **                   a full flush of all log and real blocks in the
  27 **                   transaction.
  28 **
  29 ** flush_old_commits -- if the current transaction is too old, it is ended and
  30 **                      commit blocks are sent to disk.  Forces commit blocks
  31 **                      to disk for all backgrounded commits that have been
  32 **                      around too long.
  33 **                   -- Note, if you call this as an immediate flush from
  34 **                      from within kupdate, it will ignore the immediate flag
  35 */
  36
  37 #include <linux/time.h>
  38 #include <linux/semaphore.h>
  39 #include <linux/vmalloc.h>
  40 #include <linux/reiserfs_fs.h>
  41 #include <linux/kernel.h>
  42 #include <linux/errno.h>
  43 #include <linux/fcntl.h>
  44 #include <linux/stat.h>
  45 #include <linux/string.h>
  46 #include <linux/smp_lock.h>
  47 #include <linux/buffer_head.h>
  48 #include <linux/workqueue.h>
  49 #include <linux/writeback.h>
  50 #include <linux/blkdev.h>
  51 #include <linux/backing-dev.h>
  52 #include <linux/uaccess.h>
  53
  54 #include <asm/system.h>
  55
  56 /* gets a struct reiserfs_journal_list * from a list head */
  57 #define JOURNAL_LIST_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
  58                                j_list))
  59 #define JOURNAL_WORK_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
  60                                j_working_list))
  61
  62 /* the number of mounted filesystems.  This is used to decide when to
  63 ** start and kill the commit workqueue
  64 */
  65 static int reiserfs_mounted_fs_count;
  66
  67 static struct workqueue_struct *commit_wq;
  68
  69 #define JOURNAL_TRANS_HALF 1018 /* must be correct to keep the desc and commit
  70                                    structs at 4k */
  71 #define BUFNR 64                /*read ahead */
  72
  73 /* cnode stat bits.  Move these into reiserfs_fs.h */
  74
  75 #define BLOCK_FREED 2           /* this block was freed, and can't be written.  */
  76 #define BLOCK_FREED_HOLDER 3    /* this block was freed during this transaction, and can't be written */
  77
  78 #define BLOCK_NEEDS_FLUSH 4     /* used in flush_journal_list */
  79 #define BLOCK_DIRTIED 5
  80
  81 /* journal list state bits */
  82 #define LIST_TOUCHED 1
  83 #define LIST_DIRTY   2
  84 #define LIST_COMMIT_PENDING  4  /* someone will commit this list */
  85
  86 /* flags for do_journal_end */
  87 #define FLUSH_ALL   1           /* flush commit and real blocks */
  88 #define COMMIT_NOW  2           /* end and commit this transaction */
  89 #define WAIT        4           /* wait for the log blocks to hit the disk */
  90
  91 static int do_journal_end(struct reiserfs_transaction_handle *,
  92                           struct super_block *, unsigned long nblocks,
  93                           int flags);
  94 static int flush_journal_list(struct super_block *s,
  95                               struct reiserfs_journal_list *jl, int flushall);
  96 static int flush_commit_list(struct super_block *s,
  97                              struct reiserfs_journal_list *jl, int flushall);
  98 static int can_dirty(struct reiserfs_journal_cnode *cn);
  99 static int journal_join(struct reiserfs_transaction_handle *th,
 100                         struct super_block *sb, unsigned long nblocks);
 101 static int release_journal_dev(struct super_block *super,
 102                                struct reiserfs_journal *journal);
 103 static int dirty_one_transaction(struct super_block *s,
 104                                  struct reiserfs_journal_list *jl);
 105 static void flush_async_commits(struct work_struct *work);
 106 static void queue_log_writer(struct super_block *s);
 107
 108 /* values for join in do_journal_begin_r */
 109 enum {
 110         JBEGIN_REG = 0,         /* regular journal begin */
 111         JBEGIN_JOIN = 1,        /* join the running transaction if at all possible */
 112         JBEGIN_ABORT = 2,       /* called from cleanup code, ignores aborted flag */
 113 };
 114
 115 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
 116                               struct super_block *sb,
 117                               unsigned long nblocks, int join);
 118
 119 static void init_journal_hash(struct super_block *sb)
 120 {
 121         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 122         memset(journal->j_hash_table, 0,
 123                JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
 124 }
 125
 126 /*
 127 ** clears BH_Dirty and sticks the buffer on the clean list.  Called because I can't allow refile_buffer to
 128 ** make schedule happen after I've freed a block.  Look at remove_from_transaction and journal_mark_freed for
 129 ** more details.
 130 */
 131 static int reiserfs_clean_and_file_buffer(struct buffer_head *bh)
 132 {
 133         if (bh) {
 134                 clear_buffer_dirty(bh);
 135                 clear_buffer_journal_test(bh);
 136         }
 137         return 0;
 138 }
 139
 140 static void disable_barrier(struct super_block *s)
 141 {
 142         REISERFS_SB(s)->s_mount_opt &= ~(1 << REISERFS_BARRIER_FLUSH);
 143         printk("reiserfs: disabling flush barriers on %s\n",
 144                reiserfs_bdevname(s));
 145 }
 146
 147 static struct reiserfs_bitmap_node *allocate_bitmap_node(struct super_block
 148                                                          *sb)
 149 {
 150         struct reiserfs_bitmap_node *bn;
 151         static int id;
 152
 153         bn = kmalloc(sizeof(struct reiserfs_bitmap_node), GFP_NOFS);
 154         if (!bn) {
 155                 return NULL;
 156         }
 157         bn->data = kzalloc(sb->s_blocksize, GFP_NOFS);
 158         if (!bn->data) {
 159                 kfree(bn);
 160                 return NULL;
 161         }
 162         bn->id = id++;
 163         INIT_LIST_HEAD(&bn->list);
 164         return bn;
 165 }
 166
 167 static struct reiserfs_bitmap_node *get_bitmap_node(struct super_block *sb)
 168 {
 169         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 170         struct reiserfs_bitmap_node *bn = NULL;
 171         struct list_head *entry = journal->j_bitmap_nodes.next;
 172
 173         journal->j_used_bitmap_nodes++;
 174       repeat:
 175
 176         if (entry != &journal->j_bitmap_nodes) {
 177                 bn = list_entry(entry, struct reiserfs_bitmap_node, list);
 178                 list_del(entry);
 179                 memset(bn->data, 0, sb->s_blocksize);
 180                 journal->j_free_bitmap_nodes--;
 181                 return bn;
 182         }
 183         bn = allocate_bitmap_node(sb);
 184         if (!bn) {
 185                 yield();
 186                 goto repeat;
 187         }
 188         return bn;
 189 }
 190 static inline void free_bitmap_node(struct super_block *sb,
 191                                     struct reiserfs_bitmap_node *bn)
 192 {
 193         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 194         journal->j_used_bitmap_nodes--;
 195         if (journal->j_free_bitmap_nodes > REISERFS_MAX_BITMAP_NODES) {
 196                 kfree(bn->data);
 197                 kfree(bn);
 198         } else {
 199                 list_add(&bn->list, &journal->j_bitmap_nodes);
 200                 journal->j_free_bitmap_nodes++;
 201         }
 202 }
 203
 204 static void allocate_bitmap_nodes(struct super_block *sb)
 205 {
 206         int i;
 207         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 208         struct reiserfs_bitmap_node *bn = NULL;
 209         for (i = 0; i < REISERFS_MIN_BITMAP_NODES; i++) {
 210                 bn = allocate_bitmap_node(sb);
 211                 if (bn) {
 212                         list_add(&bn->list, &journal->j_bitmap_nodes);
 213                         journal->j_free_bitmap_nodes++;
 214                 } else {
 215                         break;  /* this is ok, we'll try again when more are needed */
 216                 }
 217         }
 218 }
 219
 220 static int set_bit_in_list_bitmap(struct super_block *sb,
 221                                   b_blocknr_t block,
 222                                   struct reiserfs_list_bitmap *jb)
 223 {
 224         unsigned int bmap_nr = block / (sb->s_blocksize << 3);
 225         unsigned int bit_nr = block % (sb->s_blocksize << 3);
 226
 227         if (!jb->bitmaps[bmap_nr]) {
 228                 jb->bitmaps[bmap_nr] = get_bitmap_node(sb);
 229         }
 230         set_bit(bit_nr, (unsigned long *)jb->bitmaps[bmap_nr]->data);
 231         return 0;
 232 }
 233
 234 static void cleanup_bitmap_list(struct super_block *sb,
 235                                 struct reiserfs_list_bitmap *jb)
 236 {
 237         int i;
 238         if (jb->bitmaps == NULL)
 239                 return;
 240
 241         for (i = 0; i < reiserfs_bmap_count(sb); i++) {
 242                 if (jb->bitmaps[i]) {
 243                         free_bitmap_node(sb, jb->bitmaps[i]);
 244                         jb->bitmaps[i] = NULL;
 245                 }
 246         }
 247 }
 248
 249 /*
 250 ** only call this on FS unmount.
 251 */
 252 static int free_list_bitmaps(struct super_block *sb,
 253                              struct reiserfs_list_bitmap *jb_array)
 254 {
 255         int i;
 256         struct reiserfs_list_bitmap *jb;
 257         for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
 258                 jb = jb_array + i;
 259                 jb->journal_list = NULL;
 260                 cleanup_bitmap_list(sb, jb);
 261                 vfree(jb->bitmaps);
 262                 jb->bitmaps = NULL;
 263         }
 264         return 0;
 265 }
 266
 267 static int free_bitmap_nodes(struct super_block *sb)
 268 {
 269         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 270         struct list_head *next = journal->j_bitmap_nodes.next;
 271         struct reiserfs_bitmap_node *bn;
 272
 273         while (next != &journal->j_bitmap_nodes) {
 274                 bn = list_entry(next, struct reiserfs_bitmap_node, list);
 275                 list_del(next);
 276                 kfree(bn->data);
 277                 kfree(bn);
 278                 next = journal->j_bitmap_nodes.next;
 279                 journal->j_free_bitmap_nodes--;
 280         }
 281
 282         return 0;
 283 }
 284
 285 /*
 286 ** get memory for JOURNAL_NUM_BITMAPS worth of bitmaps.
 287 ** jb_array is the array to be filled in.
 288 */
 289 int reiserfs_allocate_list_bitmaps(struct super_block *sb,
 290                                    struct reiserfs_list_bitmap *jb_array,
 291                                    unsigned int bmap_nr)
 292 {
 293         int i;
 294         int failed = 0;
 295         struct reiserfs_list_bitmap *jb;
 296         int mem = bmap_nr * sizeof(struct reiserfs_bitmap_node *);
 297
 298         for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
 299                 jb = jb_array + i;
 300                 jb->journal_list = NULL;
 301                 jb->bitmaps = vmalloc(mem);
 302                 if (!jb->bitmaps) {
 303                         reiserfs_warning(sb, "clm-2000", "unable to "
 304                                          "allocate bitmaps for journal lists");
 305                         failed = 1;
 306                         break;
 307                 }
 308                 memset(jb->bitmaps, 0, mem);
 309         }
 310         if (failed) {
 311                 free_list_bitmaps(sb, jb_array);
 312                 return -1;
 313         }
 314         return 0;
 315 }
 316
 317 /*
 318 ** find an available list bitmap.  If you can't find one, flush a commit list
 319 ** and try again
 320 */
 321 static struct reiserfs_list_bitmap *get_list_bitmap(struct super_block *sb,
 322                                                     struct reiserfs_journal_list
 323                                                     *jl)
 324 {
 325         int i, j;
 326         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 327         struct reiserfs_list_bitmap *jb = NULL;
 328
 329         for (j = 0; j < (JOURNAL_NUM_BITMAPS * 3); j++) {
 330                 i = journal->j_list_bitmap_index;
 331                 journal->j_list_bitmap_index = (i + 1) % JOURNAL_NUM_BITMAPS;
 332                 jb = journal->j_list_bitmap + i;
 333                 if (journal->j_list_bitmap[i].journal_list) {
 334                         flush_commit_list(sb,
 335                                           journal->j_list_bitmap[i].
 336                                           journal_list, 1);
 337                         if (!journal->j_list_bitmap[i].journal_list) {
 338                                 break;
 339                         }
 340                 } else {
 341                         break;
 342                 }
 343         }
 344         if (jb->journal_list) { /* double check to make sure if flushed correctly */
 345                 return NULL;
 346         }
 347         jb->journal_list = jl;
 348         return jb;
 349 }
 350
 351 /*
 352 ** allocates a new chunk of X nodes, and links them all together as a list.
 353 ** Uses the cnode->next and cnode->prev pointers
 354 ** returns NULL on failure
 355 */
 356 static struct reiserfs_journal_cnode *allocate_cnodes(int num_cnodes)
 357 {
 358         struct reiserfs_journal_cnode *head;
 359         int i;
 360         if (num_cnodes <= 0) {
 361                 return NULL;
 362         }
 363         head = vmalloc(num_cnodes * sizeof(struct reiserfs_journal_cnode));
 364         if (!head) {
 365                 return NULL;
 366         }
 367         memset(head, 0, num_cnodes * sizeof(struct reiserfs_journal_cnode));
 368         head[0].prev = NULL;
 369         head[0].next = head + 1;
 370         for (i = 1; i < num_cnodes; i++) {
 371                 head[i].prev = head + (i - 1);
 372                 head[i].next = head + (i + 1);  /* if last one, overwrite it after the if */
 373         }
 374         head[num_cnodes - 1].next = NULL;
 375         return head;
 376 }
 377
 378 /*
 379 ** pulls a cnode off the free list, or returns NULL on failure
 380 */
 381 static struct reiserfs_journal_cnode *get_cnode(struct super_block *sb)
 382 {
 383         struct reiserfs_journal_cnode *cn;
 384         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 385
 386         reiserfs_check_lock_depth(sb, "get_cnode");
 387
 388         if (journal->j_cnode_free <= 0) {
 389                 return NULL;
 390         }
 391         journal->j_cnode_used++;
 392         journal->j_cnode_free--;
 393         cn = journal->j_cnode_free_list;
 394         if (!cn) {
 395                 return cn;
 396         }
 397         if (cn->next) {
 398                 cn->next->prev = NULL;
 399         }
 400         journal->j_cnode_free_list = cn->next;
 401         memset(cn, 0, sizeof(struct reiserfs_journal_cnode));
 402         return cn;
 403 }
 404
 405 /*
 406 ** returns a cnode to the free list
 407 */
 408 static void free_cnode(struct super_block *sb,
 409                        struct reiserfs_journal_cnode *cn)
 410 {
 411         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 412
 413         reiserfs_check_lock_depth(sb, "free_cnode");
 414
 415         journal->j_cnode_used--;
 416         journal->j_cnode_free++;
 417         /* memset(cn, 0, sizeof(struct reiserfs_journal_cnode)) ; */
 418         cn->next = journal->j_cnode_free_list;
 419         if (journal->j_cnode_free_list) {
 420                 journal->j_cnode_free_list->prev = cn;
 421         }
 422         cn->prev = NULL;        /* not needed with the memset, but I might kill the memset, and forget to do this */
 423         journal->j_cnode_free_list = cn;
 424 }
 425
 426 static void clear_prepared_bits(struct buffer_head *bh)
 427 {
 428         clear_buffer_journal_prepared(bh);
 429         clear_buffer_journal_restore_dirty(bh);
 430 }
 431
 432 /* return a cnode with same dev, block number and size in table, or null if not found */
 433 static inline struct reiserfs_journal_cnode *get_journal_hash_dev(struct
 434                                                                   super_block
 435                                                                   *sb,
 436                                                                   struct
 437                                                                   reiserfs_journal_cnode
 438                                                                   **table,
 439                                                                   long bl)
 440 {
 441         struct reiserfs_journal_cnode *cn;
 442         cn = journal_hash(table, sb, bl);
 443         while (cn) {
 444                 if (cn->blocknr == bl && cn->sb == sb)
 445                         return cn;
 446                 cn = cn->hnext;
 447         }
 448         return (struct reiserfs_journal_cnode *)0;
 449 }
 450
 451 /*
 452 ** this actually means 'can this block be reallocated yet?'.  If you set search_all, a block can only be allocated
 453 ** if it is not in the current transaction, was not freed by the current transaction, and has no chance of ever
 454 ** being overwritten by a replay after crashing.
 455 **
 456 ** If you don't set search_all, a block can only be allocated if it is not in the current transaction.  Since deleting
 457 ** a block removes it from the current transaction, this case should never happen.  If you don't set search_all, make
 458 ** sure you never write the block without logging it.
 459 **
 460 ** next_zero_bit is a suggestion about the next block to try for find_forward.
 461 ** when bl is rejected because it is set in a journal list bitmap, we search
 462 ** for the next zero bit in the bitmap that rejected bl.  Then, we return that
 463 ** through next_zero_bit for find_forward to try.
 464 **
 465 ** Just because we return something in next_zero_bit does not mean we won't
 466 ** reject it on the next call to reiserfs_in_journal
 467 **
 468 */
 469 int reiserfs_in_journal(struct super_block *sb,
 470                         unsigned int bmap_nr, int bit_nr, int search_all,
 471                         b_blocknr_t * next_zero_bit)
 472 {
 473         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 474         struct reiserfs_journal_cnode *cn;
 475         struct reiserfs_list_bitmap *jb;
 476         int i;
 477         unsigned long bl;
 478
 479         *next_zero_bit = 0;     /* always start this at zero. */
 480
 481         PROC_INFO_INC(sb, journal.in_journal);
 482         /* If we aren't doing a search_all, this is a metablock, and it will be logged before use.
 483          ** if we crash before the transaction that freed it commits,  this transaction won't
 484          ** have committed either, and the block will never be written
 485          */
 486         if (search_all) {
 487                 for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
 488                         PROC_INFO_INC(sb, journal.in_journal_bitmap);
 489                         jb = journal->j_list_bitmap + i;
 490                         if (jb->journal_list && jb->bitmaps[bmap_nr] &&
 491                             test_bit(bit_nr,
 492                                      (unsigned long *)jb->bitmaps[bmap_nr]->
 493                                      data)) {
 494                                 *next_zero_bit =
 495                                     find_next_zero_bit((unsigned long *)
 496                                                        (jb->bitmaps[bmap_nr]->
 497                                                         data),
 498                                                        sb->s_blocksize << 3,
 499                                                        bit_nr + 1);
 500                                 return 1;
 501                         }
 502                 }
 503         }
 504
 505         bl = bmap_nr * (sb->s_blocksize << 3) + bit_nr;
 506         /* is it in any old transactions? */
 507         if (search_all
 508             && (cn =
 509                 get_journal_hash_dev(sb, journal->j_list_hash_table, bl))) {
 510                 return 1;
 511         }
 512
 513         /* is it in the current transaction.  This should never happen */
 514         if ((cn = get_journal_hash_dev(sb, journal->j_hash_table, bl))) {
 515                 BUG();
 516                 return 1;
 517         }
 518
 519         PROC_INFO_INC(sb, journal.in_journal_reusable);
 520         /* safe for reuse */
 521         return 0;
 522 }
 523
 524 /* insert cn into table
 525 */
 526 static inline void insert_journal_hash(struct reiserfs_journal_cnode **table,
 527                                        struct reiserfs_journal_cnode *cn)
 528 {
 529         struct reiserfs_journal_cnode *cn_orig;
 530
 531         cn_orig = journal_hash(table, cn->sb, cn->blocknr);
 532         cn->hnext = cn_orig;
 533         cn->hprev = NULL;
 534         if (cn_orig) {
 535                 cn_orig->hprev = cn;
 536         }
 537         journal_hash(table, cn->sb, cn->blocknr) = cn;
 538 }
 539
 540 /*
 541  * Several mutexes depend on the write lock.
 542  * However sometimes we want to relax the write lock while we hold
 543  * these mutexes, according to the release/reacquire on schedule()
 544  * properties of the Bkl that were used.
 545  * Reiserfs performances and locking were based on this scheme.
 546  * Now that the write lock is a mutex and not the bkl anymore, doing so
 547  * may result in a deadlock:
 548  *
 549  * A acquire write_lock
 550  * A acquire j_commit_mutex
 551  * A release write_lock and wait for something
 552  * B acquire write_lock
 553  * B can't acquire j_commit_mutex and sleep
 554  * A can't acquire write lock anymore
 555  * deadlock
 556  *
 557  * What we do here is avoiding such deadlock by playing the same game
 558  * than the Bkl: if we can't acquire a mutex that depends on the write lock,
 559  * we release the write lock, wait a bit and then retry.
 560  *
 561  * The mutexes concerned by this hack are:
 562  * - The commit mutex of a journal list
 563  * - The flush mutex
 564  * - The journal lock
 565  */
 566 static inline void reiserfs_mutex_lock_safe(struct mutex *m,
 567                                struct super_block *s)
 568 {
 569         while (!mutex_trylock(m)) {
 570                 reiserfs_write_unlock(s);
 571                 schedule();
 572                 reiserfs_write_lock(s);
 573         }
 574 }
 575
 576 /* lock the current transaction */
 577 static inline void lock_journal(struct super_block *sb)
 578 {
 579         PROC_INFO_INC(sb, journal.lock_journal);
 580
 581         reiserfs_mutex_lock_safe(&SB_JOURNAL(sb)->j_mutex, sb);
 582 }
 583
 584 /* unlock the current transaction */
 585 static inline void unlock_journal(struct super_block *sb)
 586 {
 587         mutex_unlock(&SB_JOURNAL(sb)->j_mutex);
 588 }
 589
 590 static inline void get_journal_list(struct reiserfs_journal_list *jl)
 591 {
 592         jl->j_refcount++;
 593 }
 594
 595 static inline void put_journal_list(struct super_block *s,
 596                                     struct reiserfs_journal_list *jl)
 597 {
 598         if (jl->j_refcount < 1) {
 599                 reiserfs_panic(s, "journal-2", "trans id %u, refcount at %d",
 600                                jl->j_trans_id, jl->j_refcount);
 601         }
 602         if (--jl->j_refcount == 0)
 603                 kfree(jl);
 604 }
 605
 606 /*
 607 ** this used to be much more involved, and I'm keeping it just in case things get ugly again.
 608 ** it gets called by flush_commit_list, and cleans up any data stored about blocks freed during a
 609 ** transaction.
 610 */
 611 static void cleanup_freed_for_journal_list(struct super_block *sb,
 612                                            struct reiserfs_journal_list *jl)
 613 {
 614
 615         struct reiserfs_list_bitmap *jb = jl->j_list_bitmap;
 616         if (jb) {
 617                 cleanup_bitmap_list(sb, jb);
 618         }
 619         jl->j_list_bitmap->journal_list = NULL;
 620         jl->j_list_bitmap = NULL;
 621 }
 622
 623 static int journal_list_still_alive(struct super_block *s,
 624                                     unsigned int trans_id)
 625 {
 626         struct reiserfs_journal *journal = SB_JOURNAL(s);
 627         struct list_head *entry = &journal->j_journal_list;
 628         struct reiserfs_journal_list *jl;
 629
 630         if (!list_empty(entry)) {
 631                 jl = JOURNAL_LIST_ENTRY(entry->next);
 632                 if (jl->j_trans_id <= trans_id) {
 633                         return 1;
 634                 }
 635         }
 636         return 0;
 637 }
 638
 639 /*
 640  * If page->mapping was null, we failed to truncate this page for
 641  * some reason.  Most likely because it was truncated after being
 642  * logged via data=journal.
 643  *
 644  * This does a check to see if the buffer belongs to one of these
 645  * lost pages before doing the final put_bh.  If page->mapping was
 646  * null, it tries to free buffers on the page, which should make the
 647  * final page_cache_release drop the page from the lru.
 648  */
 649 static void release_buffer_page(struct buffer_head *bh)
 650 {
 651         struct page *page = bh->b_page;
 652         if (!page->mapping && trylock_page(page)) {
 653                 page_cache_get(page);
 654                 put_bh(bh);
 655                 if (!page->mapping)
 656                         try_to_free_buffers(page);
 657                 unlock_page(page);
 658                 page_cache_release(page);
 659         } else {
 660                 put_bh(bh);
 661         }
 662 }
 663
 664 static void reiserfs_end_buffer_io_sync(struct buffer_head *bh, int uptodate)
 665 {
 666         char b[BDEVNAME_SIZE];
 667
 668         if (buffer_journaled(bh)) {
 669                 reiserfs_warning(NULL, "clm-2084",
 670                                  "pinned buffer %lu:%s sent to disk",
 671                                  bh->b_blocknr, bdevname(bh->b_bdev, b));
 672         }
 673         if (uptodate)
 674                 set_buffer_uptodate(bh);
 675         else
 676                 clear_buffer_uptodate(bh);
 677
 678         unlock_buffer(bh);
 679         release_buffer_page(bh);
 680 }
 681
 682 static void reiserfs_end_ordered_io(struct buffer_head *bh, int uptodate)
 683 {
 684         if (uptodate)
 685                 set_buffer_uptodate(bh);
 686         else
 687                 clear_buffer_uptodate(bh);
 688         unlock_buffer(bh);
 689         put_bh(bh);
 690 }
 691
 692 static void submit_logged_buffer(struct buffer_head *bh)
 693 {
 694         get_bh(bh);
 695         bh->b_end_io = reiserfs_end_buffer_io_sync;
 696         clear_buffer_journal_new(bh);
 697         clear_buffer_dirty(bh);
 698         if (!test_clear_buffer_journal_test(bh))
 699                 BUG();
 700         if (!buffer_uptodate(bh))
 701                 BUG();
 702         submit_bh(WRITE, bh);
 703 }
 704
 705 static void submit_ordered_buffer(struct buffer_head *bh)
 706 {
 707         get_bh(bh);
 708         bh->b_end_io = reiserfs_end_ordered_io;
 709         clear_buffer_dirty(bh);
 710         if (!buffer_uptodate(bh))
 711                 BUG();
 712         submit_bh(WRITE, bh);
 713 }
 714
 715 static int submit_barrier_buffer(struct buffer_head *bh)
 716 {
 717         get_bh(bh);
 718         bh->b_end_io = reiserfs_end_ordered_io;
 719         clear_buffer_dirty(bh);
 720         if (!buffer_uptodate(bh))
 721                 BUG();
 722         return submit_bh(WRITE_BARRIER, bh);
 723 }
 724
 725 static void check_barrier_completion(struct super_block *s,
 726                                      struct buffer_head *bh)
 727 {
 728         if (buffer_eopnotsupp(bh)) {
 729                 clear_buffer_eopnotsupp(bh);
 730                 disable_barrier(s);
 731                 set_buffer_uptodate(bh);
 732                 set_buffer_dirty(bh);
 733                 reiserfs_write_unlock(s);
 734                 sync_dirty_buffer(bh);
 735                 reiserfs_write_lock(s);
 736         }
 737 }
 738
 739 #define CHUNK_SIZE 32
 740 struct buffer_chunk {
 741         struct buffer_head *bh[CHUNK_SIZE];
 742         int nr;
 743 };
 744
 745 static void write_chunk(struct buffer_chunk *chunk)
 746 {
 747         int i;
 748         get_fs_excl();
 749         for (i = 0; i < chunk->nr; i++) {
 750                 submit_logged_buffer(chunk->bh[i]);
 751         }
 752         chunk->nr = 0;
 753         put_fs_excl();
 754 }
 755
 756 static void write_ordered_chunk(struct buffer_chunk *chunk)
 757 {
 758         int i;
 759         get_fs_excl();
 760         for (i = 0; i < chunk->nr; i++) {
 761                 submit_ordered_buffer(chunk->bh[i]);
 762         }
 763         chunk->nr = 0;
 764         put_fs_excl();
 765 }
 766
 767 static int add_to_chunk(struct buffer_chunk *chunk, struct buffer_head *bh,
 768                         spinlock_t * lock, void (fn) (struct buffer_chunk *))
 769 {
 770         int ret = 0;
 771         BUG_ON(chunk->nr >= CHUNK_SIZE);
 772         chunk->bh[chunk->nr++] = bh;
 773         if (chunk->nr >= CHUNK_SIZE) {
 774                 ret = 1;
 775                 if (lock)
 776                         spin_unlock(lock);
 777                 fn(chunk);
 778                 if (lock)
 779                         spin_lock(lock);
 780         }
 781         return ret;
 782 }
 783
 784 static atomic_t nr_reiserfs_jh = ATOMIC_INIT(0);
 785 static struct reiserfs_jh *alloc_jh(void)
 786 {
 787         struct reiserfs_jh *jh;
 788         while (1) {
 789                 jh = kmalloc(sizeof(*jh), GFP_NOFS);
 790                 if (jh) {
 791                         atomic_inc(&nr_reiserfs_jh);
 792                         return jh;
 793                 }
 794                 yield();
 795         }
 796 }
 797
 798 /*
 799  * we want to free the jh when the buffer has been written
 800  * and waited on
 801  */
 802 void reiserfs_free_jh(struct buffer_head *bh)
 803 {
 804         struct reiserfs_jh *jh;
 805
 806         jh = bh->b_private;
 807         if (jh) {
 808                 bh->b_private = NULL;
 809                 jh->bh = NULL;
 810                 list_del_init(&jh->list);
 811                 kfree(jh);
 812                 if (atomic_read(&nr_reiserfs_jh) <= 0)
 813                         BUG();
 814                 atomic_dec(&nr_reiserfs_jh);
 815                 put_bh(bh);
 816         }
 817 }
 818
 819 static inline int __add_jh(struct reiserfs_journal *j, struct buffer_head *bh,
 820                            int tail)
 821 {
 822         struct reiserfs_jh *jh;
 823
 824         if (bh->b_private) {
 825                 spin_lock(&j->j_dirty_buffers_lock);
 826                 if (!bh->b_private) {
 827                         spin_unlock(&j->j_dirty_buffers_lock);
 828                         goto no_jh;
 829                 }
 830                 jh = bh->b_private;
 831                 list_del_init(&jh->list);
 832         } else {
 833               no_jh:
 834                 get_bh(bh);
 835                 jh = alloc_jh();
 836                 spin_lock(&j->j_dirty_buffers_lock);
 837                 /* buffer must be locked for __add_jh, should be able to have
 838                  * two adds at the same time
 839                  */
 840                 BUG_ON(bh->b_private);
 841                 jh->bh = bh;
 842                 bh->b_private = jh;
 843         }
 844         jh->jl = j->j_current_jl;
 845         if (tail)
 846                 list_add_tail(&jh->list, &jh->jl->j_tail_bh_list);
 847         else {
 848                 list_add_tail(&jh->list, &jh->jl->j_bh_list);
 849         }
 850         spin_unlock(&j->j_dirty_buffers_lock);
 851         return 0;
 852 }
 853
 854 int reiserfs_add_tail_list(struct inode *inode, struct buffer_head *bh)
 855 {
 856         return __add_jh(SB_JOURNAL(inode->i_sb), bh, 1);
 857 }
 858 int reiserfs_add_ordered_list(struct inode *inode, struct buffer_head *bh)
 859 {
 860         return __add_jh(SB_JOURNAL(inode->i_sb), bh, 0);
 861 }
 862
 863 #define JH_ENTRY(l) list_entry((l), struct reiserfs_jh, list)
 864 static int write_ordered_buffers(spinlock_t * lock,
 865                                  struct reiserfs_journal *j,
 866                                  struct reiserfs_journal_list *jl,
 867                                  struct list_head *list)
 868 {
 869         struct buffer_head *bh;
 870         struct reiserfs_jh *jh;
 871         int ret = j->j_errno;
 872         struct buffer_chunk chunk;
 873         struct list_head tmp;
 874         INIT_LIST_HEAD(&tmp);
 875
 876         chunk.nr = 0;
 877         spin_lock(lock);
 878         while (!list_empty(list)) {
 879                 jh = JH_ENTRY(list->next);
 880                 bh = jh->bh;
 881                 get_bh(bh);
 882                 if (!trylock_buffer(bh)) {
 883                         if (!buffer_dirty(bh)) {
 884                                 list_move(&jh->list, &tmp);
 885                                 goto loop_next;
 886                         }
 887                         spin_unlock(lock);
 888                         if (chunk.nr)
 889                                 write_ordered_chunk(&chunk);
 890                         wait_on_buffer(bh);
 891                         cond_resched();
 892                         spin_lock(lock);
 893                         goto loop_next;
 894                 }
 895                 /* in theory, dirty non-uptodate buffers should never get here,
 896                  * but the upper layer io error paths still have a few quirks.
 897                  * Handle them here as gracefully as we can
 898                  */
 899                 if (!buffer_uptodate(bh) && buffer_dirty(bh)) {
 900                         clear_buffer_dirty(bh);
 901                         ret = -EIO;
 902                 }
 903                 if (buffer_dirty(bh)) {
 904                         list_move(&jh->list, &tmp);
 905                         add_to_chunk(&chunk, bh, lock, write_ordered_chunk);
 906                 } else {
 907                         reiserfs_free_jh(bh);
 908                         unlock_buffer(bh);
 909                 }
 910               loop_next:
 911                 put_bh(bh);
 912                 cond_resched_lock(lock);
 913         }
 914         if (chunk.nr) {
 915                 spin_unlock(lock);
 916                 write_ordered_chunk(&chunk);
 917                 spin_lock(lock);
 918         }
 919         while (!list_empty(&tmp)) {
 920                 jh = JH_ENTRY(tmp.prev);
 921                 bh = jh->bh;
 922                 get_bh(bh);
 923                 reiserfs_free_jh(bh);
 924
 925                 if (buffer_locked(bh)) {
 926                         spin_unlock(lock);
 927                         wait_on_buffer(bh);
 928                         spin_lock(lock);
 929                 }
 930                 if (!buffer_uptodate(bh)) {
 931                         ret = -EIO;
 932                 }
 933                 /* ugly interaction with invalidatepage here.
 934                  * reiserfs_invalidate_page will pin any buffer that has a valid
 935                  * journal head from an older transaction.  If someone else sets
 936                  * our buffer dirty after we write it in the first loop, and
 937                  * then someone truncates the page away, nobody will ever write
 938                  * the buffer. We're safe if we write the page one last time
 939                  * after freeing the journal header.
 940                  */
 941                 if (buffer_dirty(bh) && unlikely(bh->b_page->mapping == NULL)) {
 942                         spin_unlock(lock);
 943                         ll_rw_block(WRITE, 1, &bh);
 944                         spin_lock(lock);
 945                 }
 946                 put_bh(bh);
 947                 cond_resched_lock(lock);
 948         }
 949         spin_unlock(lock);
 950         return ret;
 951 }
 952
 953 static int flush_older_commits(struct super_block *s,
 954                                struct reiserfs_journal_list *jl)
 955 {
 956         struct reiserfs_journal *journal = SB_JOURNAL(s);
 957         struct reiserfs_journal_list *other_jl;
 958         struct reiserfs_journal_list *first_jl;
 959         struct list_head *entry;
 960         unsigned int trans_id = jl->j_trans_id;
 961         unsigned int other_trans_id;
 962         unsigned int first_trans_id;
 963
 964       find_first:
 965         /*
 966          * first we walk backwards to find the oldest uncommitted transation
 967          */
 968         first_jl = jl;
 969         entry = jl->j_list.prev;
 970         while (1) {
 971                 other_jl = JOURNAL_LIST_ENTRY(entry);
 972                 if (entry == &journal->j_journal_list ||
 973                     atomic_read(&other_jl->j_older_commits_done))
 974                         break;
 975
 976                 first_jl = other_jl;
 977                 entry = other_jl->j_list.prev;
 978         }
 979
 980         /* if we didn't find any older uncommitted transactions, return now */
 981         if (first_jl == jl) {
 982                 return 0;
 983         }
 984
 985         first_trans_id = first_jl->j_trans_id;
 986
 987         entry = &first_jl->j_list;
 988         while (1) {
 989                 other_jl = JOURNAL_LIST_ENTRY(entry);
 990                 other_trans_id = other_jl->j_trans_id;
 991
 992                 if (other_trans_id < trans_id) {
 993                         if (atomic_read(&other_jl->j_commit_left) != 0) {
 994                                 flush_commit_list(s, other_jl, 0);
 995
 996                                 /* list we were called with is gone, return */
 997                                 if (!journal_list_still_alive(s, trans_id))
 998                                         return 1;
 999
1000                                 /* the one we just flushed is gone, this means all
1001                                  * older lists are also gone, so first_jl is no longer
1002                                  * valid either.  Go back to the beginning.
1003                                  */
1004                                 if (!journal_list_still_alive
1005                                     (s, other_trans_id)) {
1006                                         goto find_first;
1007                                 }
1008                         }
1009                         entry = entry->next;
1010                         if (entry == &journal->j_journal_list)
1011                                 return 0;
1012                 } else {
1013                         return 0;
1014                 }
1015         }
1016         return 0;
1017 }
1018
1019 static int reiserfs_async_progress_wait(struct super_block *s)
1020 {
1021         DEFINE_WAIT(wait);
1022         struct reiserfs_journal *j = SB_JOURNAL(s);
1023
1024         if (atomic_read(&j->j_async_throttle)) {
1025                 reiserfs_write_unlock(s);
1026                 congestion_wait(BLK_RW_ASYNC, HZ / 10);
1027                 reiserfs_write_lock(s);
1028         }
1029
1030         return 0;
1031 }
1032
1033 /*
1034 ** if this journal list still has commit blocks unflushed, send them to disk.
1035 **
1036 ** log areas must be flushed in order (transaction 2 can't commit before transaction 1)
1037 ** Before the commit block can by written, every other log block must be safely on disk
1038 **
1039 */
1040 static int flush_commit_list(struct super_block *s,
1041                              struct reiserfs_journal_list *jl, int flushall)
1042 {
1043         int i;
1044         b_blocknr_t bn;
1045         struct buffer_head *tbh = NULL;
1046         unsigned int trans_id = jl->j_trans_id;
1047         struct reiserfs_journal *journal = SB_JOURNAL(s);
1048         int barrier = 0;
1049         int retval = 0;
1050         int write_len;
1051
1052         reiserfs_check_lock_depth(s, "flush_commit_list");
1053
1054         if (atomic_read(&jl->j_older_commits_done)) {
1055                 return 0;
1056         }
1057
1058         get_fs_excl();
1059
1060         /* before we can put our commit blocks on disk, we have to make sure everyone older than
1061          ** us is on disk too
1062          */
1063         BUG_ON(jl->j_len <= 0);
1064         BUG_ON(trans_id == journal->j_trans_id);
1065
1066         get_journal_list(jl);
1067         if (flushall) {
1068                 if (flush_older_commits(s, jl) == 1) {
1069                         /* list disappeared during flush_older_commits.  return */
1070                         goto put_jl;
1071                 }
1072         }
1073
1074         /* make sure nobody is trying to flush this one at the same time */
1075         reiserfs_mutex_lock_safe(&jl->j_commit_mutex, s);
1076
1077         if (!journal_list_still_alive(s, trans_id)) {
1078                 mutex_unlock(&jl->j_commit_mutex);
1079                 goto put_jl;
1080         }
1081         BUG_ON(jl->j_trans_id == 0);
1082
1083         /* this commit is done, exit */
1084         if (atomic_read(&(jl->j_commit_left)) <= 0) {
1085                 if (flushall) {
1086                         atomic_set(&(jl->j_older_commits_done), 1);
1087                 }
1088                 mutex_unlock(&jl->j_commit_mutex);
1089                 goto put_jl;
1090         }
1091
1092         if (!list_empty(&jl->j_bh_list)) {
1093                 int ret;
1094
1095                 /*
1096                  * We might sleep in numerous places inside
1097                  * write_ordered_buffers. Relax the write lock.
1098                  */
1099                 reiserfs_write_unlock(s);
1100                 ret = write_ordered_buffers(&journal->j_dirty_buffers_lock,
1101                                             journal, jl, &jl->j_bh_list);
1102                 if (ret < 0 && retval == 0)
1103                         retval = ret;
1104                 reiserfs_write_lock(s);
1105         }
1106         BUG_ON(!list_empty(&jl->j_bh_list));
1107         /*
1108          * for the description block and all the log blocks, submit any buffers
1109          * that haven't already reached the disk.  Try to write at least 256
1110          * log blocks. later on, we will only wait on blocks that correspond
1111          * to this transaction, but while we're unplugging we might as well
1112          * get a chunk of data on there.
1113          */
1114         atomic_inc(&journal->j_async_throttle);
1115         write_len = jl->j_len + 1;
1116         if (write_len < 256)
1117                 write_len = 256;
1118         for (i = 0 ; i < write_len ; i++) {
1119                 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + (jl->j_start + i) %
1120                     SB_ONDISK_JOURNAL_SIZE(s);
1121                 tbh = journal_find_get_block(s, bn);
1122                 if (tbh) {
1123                         if (buffer_dirty(tbh)) {
1124                             reiserfs_write_unlock(s);
1125                             ll_rw_block(WRITE, 1, &tbh);
1126                             reiserfs_write_lock(s);
1127                         }
1128                         put_bh(tbh) ;
1129                 }
1130         }
1131         atomic_dec(&journal->j_async_throttle);
1132
1133         /* We're skipping the commit if there's an error */
1134         if (retval || reiserfs_is_journal_aborted(journal))
1135                 barrier = 0;
1136
1137         /* wait on everything written so far before writing the commit
1138          * if we are in barrier mode, send the commit down now
1139          */
1140         barrier = reiserfs_barrier_flush(s);
1141         if (barrier) {
1142                 int ret;
1143                 lock_buffer(jl->j_commit_bh);
1144                 ret = submit_barrier_buffer(jl->j_commit_bh);
1145                 if (ret == -EOPNOTSUPP) {
1146                         set_buffer_uptodate(jl->j_commit_bh);
1147                         disable_barrier(s);
1148                         barrier = 0;
1149                 }
1150         }
1151         for (i = 0; i < (jl->j_len + 1); i++) {
1152                 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) +
1153                     (jl->j_start + i) % SB_ONDISK_JOURNAL_SIZE(s);
1154                 tbh = journal_find_get_block(s, bn);
1155
1156                 reiserfs_write_unlock(s);
1157                 wait_on_buffer(tbh);
1158                 reiserfs_write_lock(s);
1159                 // since we're using ll_rw_blk above, it might have skipped over
1160                 // a locked buffer.  Double check here
1161                 //
1162                 /* redundant, sync_dirty_buffer() checks */
1163                 if (buffer_dirty(tbh)) {
1164                         reiserfs_write_unlock(s);
1165                         sync_dirty_buffer(tbh);
1166                         reiserfs_write_lock(s);
1167                 }
1168                 if (unlikely(!buffer_uptodate(tbh))) {
1169 #ifdef CONFIG_REISERFS_CHECK
1170                         reiserfs_warning(s, "journal-601",
1171                                          "buffer write failed");
1172 #endif
1173                         retval = -EIO;
1174                 }
1175                 put_bh(tbh);    /* once for journal_find_get_block */
1176                 put_bh(tbh);    /* once due to original getblk in do_journal_end */
1177                 atomic_dec(&(jl->j_commit_left));
1178         }
1179
1180         BUG_ON(atomic_read(&(jl->j_commit_left)) != 1);
1181
1182         if (!barrier) {
1183                 /* If there was a write error in the journal - we can't commit
1184                  * this transaction - it will be invalid and, if successful,
1185                  * will just end up propagating the write error out to
1186                  * the file system. */
1187                 if (likely(!retval && !reiserfs_is_journal_aborted (journal))) {
1188                         if (buffer_dirty(jl->j_commit_bh))
1189                                 BUG();
1190                         mark_buffer_dirty(jl->j_commit_bh) ;
1191                         reiserfs_write_unlock(s);
1192                         sync_dirty_buffer(jl->j_commit_bh) ;
1193                         reiserfs_write_lock(s);
1194                 }
1195         } else {
1196                 reiserfs_write_unlock(s);
1197                 wait_on_buffer(jl->j_commit_bh);
1198                 reiserfs_write_lock(s);
1199         }
1200
1201         check_barrier_completion(s, jl->j_commit_bh);
1202
1203         /* If there was a write error in the journal - we can't commit this
1204          * transaction - it will be invalid and, if successful, will just end
1205          * up propagating the write error out to the filesystem. */
1206         if (unlikely(!buffer_uptodate(jl->j_commit_bh))) {
1207 #ifdef CONFIG_REISERFS_CHECK
1208                 reiserfs_warning(s, "journal-615", "buffer write failed");
1209 #endif
1210                 retval = -EIO;
1211         }
1212         bforget(jl->j_commit_bh);
1213         if (journal->j_last_commit_id != 0 &&
1214             (jl->j_trans_id - journal->j_last_commit_id) != 1) {
1215                 reiserfs_warning(s, "clm-2200", "last commit %lu, current %lu",
1216                                  journal->j_last_commit_id, jl->j_trans_id);
1217         }
1218         journal->j_last_commit_id = jl->j_trans_id;
1219
1220         /* now, every commit block is on the disk.  It is safe to allow blocks freed during this transaction to be reallocated */
1221         cleanup_freed_for_journal_list(s, jl);
1222
1223         retval = retval ? retval : journal->j_errno;
1224
1225         /* mark the metadata dirty */
1226         if (!retval)
1227                 dirty_one_transaction(s, jl);
1228         atomic_dec(&(jl->j_commit_left));
1229
1230         if (flushall) {
1231                 atomic_set(&(jl->j_older_commits_done), 1);
1232         }
1233         mutex_unlock(&jl->j_commit_mutex);
1234       put_jl:
1235         put_journal_list(s, jl);
1236
1237         if (retval)
1238                 reiserfs_abort(s, retval, "Journal write error in %s",
1239                                __func__);
1240         put_fs_excl();
1241         return retval;
1242 }
1243
1244 /*
1245 ** flush_journal_list frequently needs to find a newer transaction for a given block.  This does that, or
1246 ** returns NULL if it can't find anything
1247 */
1248 static struct reiserfs_journal_list *find_newer_jl_for_cn(struct
1249                                                           reiserfs_journal_cnode
1250                                                           *cn)
1251 {
1252         struct super_block *sb = cn->sb;
1253         b_blocknr_t blocknr = cn->blocknr;
1254
1255         cn = cn->hprev;
1256         while (cn) {
1257                 if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist) {
1258                         return cn->jlist;
1259                 }
1260                 cn = cn->hprev;
1261         }
1262         return NULL;
1263 }
1264
1265 static int newer_jl_done(struct reiserfs_journal_cnode *cn)
1266 {
1267         struct super_block *sb = cn->sb;
1268         b_blocknr_t blocknr = cn->blocknr;
1269
1270         cn = cn->hprev;
1271         while (cn) {
1272                 if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist &&
1273                     atomic_read(&cn->jlist->j_commit_left) != 0)
1274                                     return 0;
1275                 cn = cn->hprev;
1276         }
1277         return 1;
1278 }
1279
1280 static void remove_journal_hash(struct super_block *,
1281                                 struct reiserfs_journal_cnode **,
1282                                 struct reiserfs_journal_list *, unsigned long,
1283                                 int);
1284
1285 /*
1286 ** once all the real blocks have been flushed, it is safe to remove them from the
1287 ** journal list for this transaction.  Aside from freeing the cnode, this also allows the
1288 ** block to be reallocated for data blocks if it had been deleted.
1289 */
1290 static void remove_all_from_journal_list(struct super_block *sb,
1291                                          struct reiserfs_journal_list *jl,
1292                                          int debug)
1293 {
1294         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1295         struct reiserfs_journal_cnode *cn, *last;
1296         cn = jl->j_realblock;
1297
1298         /* which is better, to lock once around the whole loop, or
1299          ** to lock for each call to remove_journal_hash?
1300          */
1301         while (cn) {
1302                 if (cn->blocknr != 0) {
1303                         if (debug) {
1304                                 reiserfs_warning(sb, "reiserfs-2201",
1305                                                  "block %u, bh is %d, state %ld",
1306                                                  cn->blocknr, cn->bh ? 1 : 0,
1307                                                  cn->state);
1308                         }
1309                         cn->state = 0;
1310                         remove_journal_hash(sb, journal->j_list_hash_table,
1311                                             jl, cn->blocknr, 1);
1312                 }
1313                 last = cn;
1314                 cn = cn->next;
1315                 free_cnode(sb, last);
1316         }
1317         jl->j_realblock = NULL;
1318 }
1319
1320 /*
1321 ** if this timestamp is greater than the timestamp we wrote last to the header block, write it to the header block.
1322 ** once this is done, I can safely say the log area for this transaction won't ever be replayed, and I can start
1323 ** releasing blocks in this transaction for reuse as data blocks.
1324 ** called by flush_journal_list, before it calls remove_all_from_journal_list
1325 **
1326 */
1327 static int _update_journal_header_block(struct super_block *sb,
1328                                         unsigned long offset,
1329                                         unsigned int trans_id)
1330 {
1331         struct reiserfs_journal_header *jh;
1332         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1333
1334         if (reiserfs_is_journal_aborted(journal))
1335                 return -EIO;
1336
1337         if (trans_id >= journal->j_last_flush_trans_id) {
1338                 if (buffer_locked((journal->j_header_bh))) {
1339                         reiserfs_write_unlock(sb);
1340                         wait_on_buffer((journal->j_header_bh));
1341                         reiserfs_write_lock(sb);
1342                         if (unlikely(!buffer_uptodate(journal->j_header_bh))) {
1343 #ifdef CONFIG_REISERFS_CHECK
1344                                 reiserfs_warning(sb, "journal-699",
1345                                                  "buffer write failed");
1346 #endif
1347                                 return -EIO;
1348                         }
1349                 }
1350                 journal->j_last_flush_trans_id = trans_id;
1351                 journal->j_first_unflushed_offset = offset;
1352                 jh = (struct reiserfs_journal_header *)(journal->j_header_bh->
1353                                                         b_data);
1354                 jh->j_last_flush_trans_id = cpu_to_le32(trans_id);
1355                 jh->j_first_unflushed_offset = cpu_to_le32(offset);
1356                 jh->j_mount_id = cpu_to_le32(journal->j_mount_id);
1357
1358                 if (reiserfs_barrier_flush(sb)) {
1359                         int ret;
1360                         lock_buffer(journal->j_header_bh);
1361                         ret = submit_barrier_buffer(journal->j_header_bh);
1362                         if (ret == -EOPNOTSUPP) {
1363                                 set_buffer_uptodate(journal->j_header_bh);
1364                                 disable_barrier(sb);
1365                                 goto sync;
1366                         }
1367                         reiserfs_write_unlock(sb);
1368                         wait_on_buffer(journal->j_header_bh);
1369                         reiserfs_write_lock(sb);
1370                         check_barrier_completion(sb, journal->j_header_bh);
1371                 } else {
1372                       sync:
1373                         set_buffer_dirty(journal->j_header_bh);
1374                         reiserfs_write_unlock(sb);
1375                         sync_dirty_buffer(journal->j_header_bh);
1376                         reiserfs_write_lock(sb);
1377                 }
1378                 if (!buffer_uptodate(journal->j_header_bh)) {
1379                         reiserfs_warning(sb, "journal-837",
1380                                          "IO error during journal replay");
1381                         return -EIO;
1382                 }
1383         }
1384         return 0;
1385 }
1386
1387 static int update_journal_header_block(struct super_block *sb,
1388                                        unsigned long offset,
1389                                        unsigned int trans_id)
1390 {
1391         return _update_journal_header_block(sb, offset, trans_id);
1392 }
1393
1394 /*
1395 ** flush any and all journal lists older than you are
1396 ** can only be called from flush_journal_list
1397 */
1398 static int flush_older_journal_lists(struct super_block *sb,
1399                                      struct reiserfs_journal_list *jl)
1400 {
1401         struct list_head *entry;
1402         struct reiserfs_journal_list *other_jl;
1403         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1404         unsigned int trans_id = jl->j_trans_id;
1405
1406         /* we know we are the only ones flushing things, no extra race
1407          * protection is required.
1408          */
1409       restart:
1410         entry = journal->j_journal_list.next;
1411         /* Did we wrap? */
1412         if (entry == &journal->j_journal_list)
1413                 return 0;
1414         other_jl = JOURNAL_LIST_ENTRY(entry);
1415         if (other_jl->j_trans_id < trans_id) {
1416                 BUG_ON(other_jl->j_refcount <= 0);
1417                 /* do not flush all */
1418                 flush_journal_list(sb, other_jl, 0);
1419
1420                 /* other_jl is now deleted from the list */
1421                 goto restart;
1422         }
1423         return 0;
1424 }
1425
1426 static void del_from_work_list(struct super_block *s,
1427                                struct reiserfs_journal_list *jl)
1428 {
1429         struct reiserfs_journal *journal = SB_JOURNAL(s);
1430         if (!list_empty(&jl->j_working_list)) {
1431                 list_del_init(&jl->j_working_list);
1432                 journal->j_num_work_lists--;
1433         }
1434 }
1435
1436 /* flush a journal list, both commit and real blocks
1437 **
1438 ** always set flushall to 1, unless you are calling from inside
1439 ** flush_journal_list
1440 **
1441 ** IMPORTANT.  This can only be called while there are no journal writers,
1442 ** and the journal is locked.  That means it can only be called from
1443 ** do_journal_end, or by journal_release
1444 */
1445 static int flush_journal_list(struct super_block *s,
1446                               struct reiserfs_journal_list *jl, int flushall)
1447 {
1448         struct reiserfs_journal_list *pjl;
1449         struct reiserfs_journal_cnode *cn, *last;
1450         int count;
1451         int was_jwait = 0;
1452         int was_dirty = 0;
1453         struct buffer_head *saved_bh;
1454         unsigned long j_len_saved = jl->j_len;
1455         struct reiserfs_journal *journal = SB_JOURNAL(s);
1456         int err = 0;
1457
1458         BUG_ON(j_len_saved <= 0);
1459
1460         if (atomic_read(&journal->j_wcount) != 0) {
1461                 reiserfs_warning(s, "clm-2048", "called with wcount %d",
1462                                  atomic_read(&journal->j_wcount));
1463         }
1464         BUG_ON(jl->j_trans_id == 0);
1465
1466         /* if flushall == 0, the lock is already held */
1467         if (flushall) {
1468                 reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s);
1469         } else if (mutex_trylock(&journal->j_flush_mutex)) {
1470                 BUG();
1471         }
1472
1473         count = 0;
1474         if (j_len_saved > journal->j_trans_max) {
1475                 reiserfs_panic(s, "journal-715", "length is %lu, trans id %lu",
1476                                j_len_saved, jl->j_trans_id);
1477                 return 0;
1478         }
1479
1480         get_fs_excl();
1481
1482         /* if all the work is already done, get out of here */
1483         if (atomic_read(&(jl->j_nonzerolen)) <= 0 &&
1484             atomic_read(&(jl->j_commit_left)) <= 0) {
1485                 goto flush_older_and_return;
1486         }
1487
1488         /* start by putting the commit list on disk.  This will also flush
1489          ** the commit lists of any olders transactions
1490          */
1491         flush_commit_list(s, jl, 1);
1492
1493         if (!(jl->j_state & LIST_DIRTY)
1494             && !reiserfs_is_journal_aborted(journal))
1495                 BUG();
1496
1497         /* are we done now? */
1498         if (atomic_read(&(jl->j_nonzerolen)) <= 0 &&
1499             atomic_read(&(jl->j_commit_left)) <= 0) {
1500                 goto flush_older_and_return;
1501         }
1502
1503         /* loop through each cnode, see if we need to write it,
1504          ** or wait on a more recent transaction, or just ignore it
1505          */
1506         if (atomic_read(&(journal->j_wcount)) != 0) {
1507                 reiserfs_panic(s, "journal-844", "journal list is flushing, "
1508                                "wcount is not 0");
1509         }
1510         cn = jl->j_realblock;
1511         while (cn) {
1512                 was_jwait = 0;
1513                 was_dirty = 0;
1514                 saved_bh = NULL;
1515                 /* blocknr of 0 is no longer in the hash, ignore it */
1516                 if (cn->blocknr == 0) {
1517                         goto free_cnode;
1518                 }
1519
1520                 /* This transaction failed commit. Don't write out to the disk */
1521                 if (!(jl->j_state & LIST_DIRTY))
1522                         goto free_cnode;
1523
1524                 pjl = find_newer_jl_for_cn(cn);
1525                 /* the order is important here.  We check pjl to make sure we
1526                  ** don't clear BH_JDirty_wait if we aren't the one writing this
1527                  ** block to disk
1528                  */
1529                 if (!pjl && cn->bh) {
1530                         saved_bh = cn->bh;
1531
1532                         /* we do this to make sure nobody releases the buffer while
1533                          ** we are working with it
1534                          */
1535                         get_bh(saved_bh);
1536
1537                         if (buffer_journal_dirty(saved_bh)) {
1538                                 BUG_ON(!can_dirty(cn));
1539                                 was_jwait = 1;
1540                                 was_dirty = 1;
1541                         } else if (can_dirty(cn)) {
1542                                 /* everything with !pjl && jwait should be writable */
1543                                 BUG();
1544                         }
1545                 }
1546
1547                 /* if someone has this block in a newer transaction, just make
1548                  ** sure they are committed, and don't try writing it to disk
1549                  */
1550                 if (pjl) {
1551                         if (atomic_read(&pjl->j_commit_left))
1552                                 flush_commit_list(s, pjl, 1);
1553                         goto free_cnode;
1554                 }
1555
1556                 /* bh == NULL when the block got to disk on its own, OR,
1557                  ** the block got freed in a future transaction
1558                  */
1559                 if (saved_bh == NULL) {
1560                         goto free_cnode;
1561                 }
1562
1563                 /* this should never happen.  kupdate_one_transaction has this list
1564                  ** locked while it works, so we should never see a buffer here that
1565                  ** is not marked JDirty_wait
1566                  */
1567                 if ((!was_jwait) && !buffer_locked(saved_bh)) {
1568                         reiserfs_warning(s, "journal-813",
1569                                          "BAD! buffer %llu %cdirty %cjwait, "
1570                                          "not in a newer tranasction",
1571                                          (unsigned long long)saved_bh->
1572                                          b_blocknr, was_dirty ? ' ' : '!',
1573                                          was_jwait ? ' ' : '!');
1574                 }
1575                 if (was_dirty) {
1576                         /* we inc again because saved_bh gets decremented at free_cnode */
1577                         get_bh(saved_bh);
1578                         set_bit(BLOCK_NEEDS_FLUSH, &cn->state);
1579                         lock_buffer(saved_bh);
1580                         BUG_ON(cn->blocknr != saved_bh->b_blocknr);
1581                         if (buffer_dirty(saved_bh))
1582                                 submit_logged_buffer(saved_bh);
1583                         else
1584                                 unlock_buffer(saved_bh);
1585                         count++;
1586                 } else {
1587                         reiserfs_warning(s, "clm-2082",
1588                                          "Unable to flush buffer %llu in %s",
1589                                          (unsigned long long)saved_bh->
1590                                          b_blocknr, __func__);
1591                 }
1592               free_cnode:
1593                 last = cn;
1594                 cn = cn->next;
1595                 if (saved_bh) {
1596                         /* we incremented this to keep others from taking the buffer head away */
1597                         put_bh(saved_bh);
1598                         if (atomic_read(&(saved_bh->b_count)) < 0) {
1599                                 reiserfs_warning(s, "journal-945",
1600                                                  "saved_bh->b_count < 0");
1601                         }
1602                 }
1603         }
1604         if (count > 0) {
1605                 cn = jl->j_realblock;
1606                 while (cn) {
1607                         if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) {
1608                                 if (!cn->bh) {
1609                                         reiserfs_panic(s, "journal-1011",
1610                                                        "cn->bh is NULL");
1611                                 }
1612
1613                                 reiserfs_write_unlock(s);
1614                                 wait_on_buffer(cn->bh);
1615                                 reiserfs_write_lock(s);
1616
1617                                 if (!cn->bh) {
1618                                         reiserfs_panic(s, "journal-1012",
1619                                                        "cn->bh is NULL");
1620                                 }
1621                                 if (unlikely(!buffer_uptodate(cn->bh))) {
1622 #ifdef CONFIG_REISERFS_CHECK
1623                                         reiserfs_warning(s, "journal-949",
1624                                                          "buffer write failed");
1625 #endif
1626                                         err = -EIO;
1627                                 }
1628                                 /* note, we must clear the JDirty_wait bit after the up to date
1629                                  ** check, otherwise we race against our flushpage routine
1630                                  */
1631                                 BUG_ON(!test_clear_buffer_journal_dirty
1632                                        (cn->bh));
1633
1634                                 /* drop one ref for us */
1635                                 put_bh(cn->bh);
1636                                 /* drop one ref for journal_mark_dirty */
1637                                 release_buffer_page(cn->bh);
1638                         }
1639                         cn = cn->next;
1640                 }
1641         }
1642
1643         if (err)
1644                 reiserfs_abort(s, -EIO,
1645                                "Write error while pushing transaction to disk in %s",
1646                                __func__);
1647       flush_older_and_return:
1648
1649         /* before we can update the journal header block, we _must_ flush all
1650          ** real blocks from all older transactions to disk.  This is because
1651          ** once the header block is updated, this transaction will not be
1652          ** replayed after a crash
1653          */
1654         if (flushall) {
1655                 flush_older_journal_lists(s, jl);
1656         }
1657
1658         err = journal->j_errno;
1659         /* before we can remove everything from the hash tables for this
1660          ** transaction, we must make sure it can never be replayed
1661          **
1662          ** since we are only called from do_journal_end, we know for sure there
1663          ** are no allocations going on while we are flushing journal lists.  So,
1664          ** we only need to update the journal header block for the last list
1665          ** being flushed
1666          */
1667         if (!err && flushall) {
1668                 err =
1669                     update_journal_header_block(s,
1670                                                 (jl->j_start + jl->j_len +
1671                                                  2) % SB_ONDISK_JOURNAL_SIZE(s),
1672                                                 jl->j_trans_id);
1673                 if (err)
1674                         reiserfs_abort(s, -EIO,
1675                                        "Write error while updating journal header in %s",
1676                                        __func__);
1677         }
1678         remove_all_from_journal_list(s, jl, 0);
1679         list_del_init(&jl->j_list);
1680         journal->j_num_lists--;
1681         del_from_work_list(s, jl);
1682
1683         if (journal->j_last_flush_id != 0 &&
1684             (jl->j_trans_id - journal->j_last_flush_id) != 1) {
1685                 reiserfs_warning(s, "clm-2201", "last flush %lu, current %lu",
1686                                  journal->j_last_flush_id, jl->j_trans_id);
1687         }
1688         journal->j_last_flush_id = jl->j_trans_id;
1689
1690         /* not strictly required since we are freeing the list, but it should
1691          * help find code using dead lists later on
1692          */
1693         jl->j_len = 0;
1694         atomic_set(&(jl->j_nonzerolen), 0);
1695         jl->j_start = 0;
1696         jl->j_realblock = NULL;
1697         jl->j_commit_bh = NULL;
1698         jl->j_trans_id = 0;
1699         jl->j_state = 0;
1700         put_journal_list(s, jl);
1701         if (flushall)
1702                 mutex_unlock(&journal->j_flush_mutex);
1703         put_fs_excl();
1704         return err;
1705 }
1706
1707 static int test_transaction(struct super_block *s,
1708                             struct reiserfs_journal_list *jl)
1709 {
1710         struct reiserfs_journal_cnode *cn;
1711
1712         if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0)
1713                 return 1;
1714
1715         cn = jl->j_realblock;
1716         while (cn) {
1717                 /* if the blocknr == 0, this has been cleared from the hash,
1718                  ** skip it
1719                  */
1720                 if (cn->blocknr == 0) {
1721                         goto next;
1722                 }
1723                 if (cn->bh && !newer_jl_done(cn))
1724                         return 0;
1725               next:
1726                 cn = cn->next;
1727                 cond_resched();
1728         }
1729         return 0;
1730 }
1731
1732 static int write_one_transaction(struct super_block *s,
1733                                  struct reiserfs_journal_list *jl,
1734                                  struct buffer_chunk *chunk)
1735 {
1736         struct reiserfs_journal_cnode *cn;
1737         int ret = 0;
1738
1739         jl->j_state |= LIST_TOUCHED;
1740         del_from_work_list(s, jl);
1741         if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0) {
1742                 return 0;
1743         }
1744
1745         cn = jl->j_realblock;
1746         while (cn) {
1747                 /* if the blocknr == 0, this has been cleared from the hash,
1748                  ** skip it
1749                  */
1750                 if (cn->blocknr == 0) {
1751                         goto next;
1752                 }
1753                 if (cn->bh && can_dirty(cn) && buffer_dirty(cn->bh)) {
1754                         struct buffer_head *tmp_bh;
1755                         /* we can race against journal_mark_freed when we try
1756                          * to lock_buffer(cn->bh), so we have to inc the buffer
1757                          * count, and recheck things after locking
1758                          */
1759                         tmp_bh = cn->bh;
1760                         get_bh(tmp_bh);
1761                         lock_buffer(tmp_bh);
1762                         if (cn->bh && can_dirty(cn) && buffer_dirty(tmp_bh)) {
1763                                 if (!buffer_journal_dirty(tmp_bh) ||
1764                                     buffer_journal_prepared(tmp_bh))
1765                                         BUG();
1766                                 add_to_chunk(chunk, tmp_bh, NULL, write_chunk);
1767                                 ret++;
1768                         } else {
1769                                 /* note, cn->bh might be null now */
1770                                 unlock_buffer(tmp_bh);
1771                         }
1772                         put_bh(tmp_bh);
1773                 }
1774               next:
1775                 cn = cn->next;
1776                 cond_resched();
1777         }
1778         return ret;
1779 }
1780
1781 /* used by flush_commit_list */
1782 static int dirty_one_transaction(struct super_block *s,
1783                                  struct reiserfs_journal_list *jl)
1784 {
1785         struct reiserfs_journal_cnode *cn;
1786         struct reiserfs_journal_list *pjl;
1787         int ret = 0;
1788
1789         jl->j_state |= LIST_DIRTY;
1790         cn = jl->j_realblock;
1791         while (cn) {
1792                 /* look for a more recent transaction that logged this
1793                  ** buffer.  Only the most recent transaction with a buffer in
1794                  ** it is allowed to send that buffer to disk
1795                  */
1796                 pjl = find_newer_jl_for_cn(cn);
1797                 if (!pjl && cn->blocknr && cn->bh
1798                     && buffer_journal_dirty(cn->bh)) {
1799                         BUG_ON(!can_dirty(cn));
1800                         /* if the buffer is prepared, it will either be logged
1801                          * or restored.  If restored, we need to make sure
1802                          * it actually gets marked dirty
1803                          */
1804                         clear_buffer_journal_new(cn->bh);
1805                         if (buffer_journal_prepared(cn->bh)) {
1806                                 set_buffer_journal_restore_dirty(cn->bh);
1807                         } else {
1808                                 set_buffer_journal_test(cn->bh);
1809                                 mark_buffer_dirty(cn->bh);
1810                         }
1811                 }
1812                 cn = cn->next;
1813         }
1814         return ret;
1815 }
1816
1817 static int kupdate_transactions(struct super_block *s,
1818                                 struct reiserfs_journal_list *jl,
1819                                 struct reiserfs_journal_list **next_jl,
1820                                 unsigned int *next_trans_id,
1821                                 int num_blocks, int num_trans)
1822 {
1823         int ret = 0;
1824         int written = 0;
1825         int transactions_flushed = 0;
1826         unsigned int orig_trans_id = jl->j_trans_id;
1827         struct buffer_chunk chunk;
1828         struct list_head *entry;
1829         struct reiserfs_journal *journal = SB_JOURNAL(s);
1830         chunk.nr = 0;
1831
1832         reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s);
1833         if (!journal_list_still_alive(s, orig_trans_id)) {
1834                 goto done;
1835         }
1836
1837         /* we've got j_flush_mutex held, nobody is going to delete any
1838          * of these lists out from underneath us
1839          */
1840         while ((num_trans && transactions_flushed < num_trans) ||
1841                (!num_trans && written < num_blocks)) {
1842
1843                 if (jl->j_len == 0 || (jl->j_state & LIST_TOUCHED) ||
1844                     atomic_read(&jl->j_commit_left)
1845                     || !(jl->j_state & LIST_DIRTY)) {
1846                         del_from_work_list(s, jl);
1847                         break;
1848                 }
1849                 ret = write_one_transaction(s, jl, &chunk);
1850
1851                 if (ret < 0)
1852                         goto done;
1853                 transactions_flushed++;
1854                 written += ret;
1855                 entry = jl->j_list.next;
1856
1857                 /* did we wrap? */
1858                 if (entry == &journal->j_journal_list) {
1859                         break;
1860                 }
1861                 jl = JOURNAL_LIST_ENTRY(entry);
1862
1863                 /* don't bother with older transactions */
1864                 if (jl->j_trans_id <= orig_trans_id)
1865                         break;
1866         }
1867         if (chunk.nr) {
1868                 write_chunk(&chunk);
1869         }
1870
1871       done:
1872         mutex_unlock(&journal->j_flush_mutex);
1873         return ret;
1874 }
1875
1876 /* for o_sync and fsync heavy applications, they tend to use
1877 ** all the journa list slots with tiny transactions.  These
1878 ** trigger lots and lots of calls to update the header block, which
1879 ** adds seeks and slows things down.
1880 **
1881 ** This function tries to clear out a large chunk of the journal lists
1882 ** at once, which makes everything faster since only the newest journal
1883 ** list updates the header block
1884 */
1885 static int flush_used_journal_lists(struct super_block *s,
1886                                     struct reiserfs_journal_list *jl)
1887 {
1888         unsigned long len = 0;
1889         unsigned long cur_len;
1890         int ret;
1891         int i;
1892         int limit = 256;
1893         struct reiserfs_journal_list *tjl;
1894         struct reiserfs_journal_list *flush_jl;
1895         unsigned int trans_id;
1896         struct reiserfs_journal *journal = SB_JOURNAL(s);
1897
1898         flush_jl = tjl = jl;
1899
1900         /* in data logging mode, try harder to flush a lot of blocks */
1901         if (reiserfs_data_log(s))
1902                 limit = 1024;
1903         /* flush for 256 transactions or limit blocks, whichever comes first */
1904         for (i = 0; i < 256 && len < limit; i++) {
1905                 if (atomic_read(&tjl->j_commit_left) ||
1906                     tjl->j_trans_id < jl->j_trans_id) {
1907                         break;
1908                 }
1909                 cur_len = atomic_read(&tjl->j_nonzerolen);
1910                 if (cur_len > 0) {
1911                         tjl->j_state &= ~LIST_TOUCHED;
1912                 }
1913                 len += cur_len;
1914                 flush_jl = tjl;
1915                 if (tjl->j_list.next == &journal->j_journal_list)
1916                         break;
1917                 tjl = JOURNAL_LIST_ENTRY(tjl->j_list.next);
1918         }
1919         /* try to find a group of blocks we can flush across all the
1920          ** transactions, but only bother if we've actually spanned
1921          ** across multiple lists
1922          */
1923         if (flush_jl != jl) {
1924                 ret = kupdate_transactions(s, jl, &tjl, &trans_id, len, i);
1925         }
1926         flush_journal_list(s, flush_jl, 1);
1927         return 0;
1928 }
1929
1930 /*
1931 ** removes any nodes in table with name block and dev as bh.
1932 ** only touchs the hnext and hprev pointers.
1933 */
1934 void remove_journal_hash(struct super_block *sb,
1935                          struct reiserfs_journal_cnode **table,
1936                          struct reiserfs_journal_list *jl,
1937                          unsigned long block, int remove_freed)
1938 {
1939         struct reiserfs_journal_cnode *cur;
1940         struct reiserfs_journal_cnode **head;
1941
1942         head = &(journal_hash(table, sb, block));
1943         if (!head) {
1944                 return;
1945         }
1946         cur = *head;
1947         while (cur) {
1948                 if (cur->blocknr == block && cur->sb == sb
1949                     && (jl == NULL || jl == cur->jlist)
1950                     && (!test_bit(BLOCK_FREED, &cur->state) || remove_freed)) {
1951                         if (cur->hnext) {
1952                                 cur->hnext->hprev = cur->hprev;
1953                         }
1954                         if (cur->hprev) {
1955                                 cur->hprev->hnext = cur->hnext;
1956                         } else {
1957                                 *head = cur->hnext;
1958                         }
1959                         cur->blocknr = 0;
1960                         cur->sb = NULL;
1961                         cur->state = 0;
1962                         if (cur->bh && cur->jlist)      /* anybody who clears the cur->bh will also dec the nonzerolen */
1963                                 atomic_dec(&(cur->jlist->j_nonzerolen));
1964                         cur->bh = NULL;
1965                         cur->jlist = NULL;
1966                 }
1967                 cur = cur->hnext;
1968         }
1969 }
1970
1971 static void free_journal_ram(struct super_block *sb)
1972 {
1973         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1974         kfree(journal->j_current_jl);
1975         journal->j_num_lists--;
1976
1977         vfree(journal->j_cnode_free_orig);
1978         free_list_bitmaps(sb, journal->j_list_bitmap);
1979         free_bitmap_nodes(sb);  /* must be after free_list_bitmaps */
1980         if (journal->j_header_bh) {
1981                 brelse(journal->j_header_bh);
1982         }
1983         /* j_header_bh is on the journal dev, make sure not to release the journal
1984          * dev until we brelse j_header_bh
1985          */
1986         release_journal_dev(sb, journal);
1987         vfree(journal);
1988 }
1989
1990 /*
1991 ** call on unmount.  Only set error to 1 if you haven't made your way out
1992 ** of read_super() yet.  Any other caller must keep error at 0.
1993 */
1994 static int do_journal_release(struct reiserfs_transaction_handle *th,
1995                               struct super_block *sb, int error)
1996 {
1997         struct reiserfs_transaction_handle myth;
1998         int flushed = 0;
1999         struct reiserfs_journal *journal = SB_JOURNAL(sb);
2000
2001         /* we only want to flush out transactions if we were called with error == 0
2002          */
2003         if (!error && !(sb->s_flags & MS_RDONLY)) {
2004                 /* end the current trans */
2005                 BUG_ON(!th->t_trans_id);
2006                 do_journal_end(th, sb, 10, FLUSH_ALL);
2007
2008                 /* make sure something gets logged to force our way into the flush code */
2009                 if (!journal_join(&myth, sb, 1)) {
2010                         reiserfs_prepare_for_journal(sb,
2011                                                      SB_BUFFER_WITH_SB(sb),
2012                                                      1);
2013                         journal_mark_dirty(&myth, sb,
2014                                            SB_BUFFER_WITH_SB(sb));
2015                         do_journal_end(&myth, sb, 1, FLUSH_ALL);
2016                         flushed = 1;
2017                 }
2018         }
2019
2020         /* this also catches errors during the do_journal_end above */
2021         if (!error && reiserfs_is_journal_aborted(journal)) {
2022                 memset(&myth, 0, sizeof(myth));
2023                 if (!journal_join_abort(&myth, sb, 1)) {
2024                         reiserfs_prepare_for_journal(sb,
2025                                                      SB_BUFFER_WITH_SB(sb),
2026                                                      1);
2027                         journal_mark_dirty(&myth, sb,
2028                                            SB_BUFFER_WITH_SB(sb));
2029                         do_journal_end(&myth, sb, 1, FLUSH_ALL);
2030                 }
2031         }
2032
2033         reiserfs_mounted_fs_count--;
2034         /* wait for all commits to finish */
2035         cancel_delayed_work(&SB_JOURNAL(sb)->j_work);
2036
2037         /*
2038          * We must release the write lock here because
2039          * the workqueue job (flush_async_commit) needs this lock
2040          */
2041         reiserfs_write_unlock(sb);
2042         flush_workqueue(commit_wq);
2043
2044         if (!reiserfs_mounted_fs_count) {
2045                 destroy_workqueue(commit_wq);
2046                 commit_wq = NULL;
2047         }
2048         reiserfs_write_lock(sb);
2049
2050         free_journal_ram(sb);
2051
2052         return 0;
2053 }
2054
2055 /*
2056 ** call on unmount.  flush all journal trans, release all alloc'd ram
2057 */
2058 int journal_release(struct reiserfs_transaction_handle *th,
2059                     struct super_block *sb)
2060 {
2061         return do_journal_release(th, sb, 0);
2062 }
2063
2064 /*
2065 ** only call from an error condition inside reiserfs_read_super!
2066 */
2067 int journal_release_error(struct reiserfs_transaction_handle *th,
2068                           struct super_block *sb)
2069 {
2070         return do_journal_release(th, sb, 1);
2071 }
2072
2073 /* compares description block with commit block.  returns 1 if they differ, 0 if they are the same */
2074 static int journal_compare_desc_commit(struct super_block *sb,
2075                                        struct reiserfs_journal_desc *desc,
2076                                        struct reiserfs_journal_commit *commit)
2077 {
2078         if (get_commit_trans_id(commit) != get_desc_trans_id(desc) ||
2079             get_commit_trans_len(commit) != get_desc_trans_len(desc) ||
2080             get_commit_trans_len(commit) > SB_JOURNAL(sb)->j_trans_max ||
2081             get_commit_trans_len(commit) <= 0) {
2082                 return 1;
2083         }
2084         return 0;
2085 }
2086
2087 /* returns 0 if it did not find a description block
2088 ** returns -1 if it found a corrupt commit block
2089 ** returns 1 if both desc and commit were valid
2090 */
2091 static int journal_transaction_is_valid(struct super_block *sb,
2092                                         struct buffer_head *d_bh,
2093                                         unsigned int *oldest_invalid_trans_id,
2094                                         unsigned long *newest_mount_id)
2095 {
2096         struct reiserfs_journal_desc *desc;
2097         struct reiserfs_journal_commit *commit;
2098         struct buffer_head *c_bh;
2099         unsigned long offset;
2100
2101         if (!d_bh)
2102                 return 0;
2103
2104         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2105         if (get_desc_trans_len(desc) > 0
2106             && !memcmp(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8)) {
2107                 if (oldest_invalid_trans_id && *oldest_invalid_trans_id
2108                     && get_desc_trans_id(desc) > *oldest_invalid_trans_id) {
2109                         reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2110                                        "journal-986: transaction "
2111                                        "is valid returning because trans_id %d is greater than "
2112                                        "oldest_invalid %lu",
2113                                        get_desc_trans_id(desc),
2114                                        *oldest_invalid_trans_id);
2115                         return 0;
2116                 }
2117                 if (newest_mount_id
2118                     && *newest_mount_id > get_desc_mount_id(desc)) {
2119                         reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2120                                        "journal-1087: transaction "
2121                                        "is valid returning because mount_id %d is less than "
2122                                        "newest_mount_id %lu",
2123                                        get_desc_mount_id(desc),
2124                                        *newest_mount_id);
2125                         return -1;
2126                 }
2127                 if (get_desc_trans_len(desc) > SB_JOURNAL(sb)->j_trans_max) {
2128                         reiserfs_warning(sb, "journal-2018",
2129                                          "Bad transaction length %d "
2130                                          "encountered, ignoring transaction",
2131                                          get_desc_trans_len(desc));
2132                         return -1;
2133                 }
2134                 offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2135
2136                 /* ok, we have a journal description block, lets see if the transaction was valid */
2137                 c_bh =
2138                     journal_bread(sb,
2139                                   SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2140                                   ((offset + get_desc_trans_len(desc) +
2141                                     1) % SB_ONDISK_JOURNAL_SIZE(sb)));
2142                 if (!c_bh)
2143                         return 0;
2144                 commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2145                 if (journal_compare_desc_commit(sb, desc, commit)) {
2146                         reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2147                                        "journal_transaction_is_valid, commit offset %ld had bad "
2148                                        "time %d or length %d",
2149                                        c_bh->b_blocknr -
2150                                        SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2151                                        get_commit_trans_id(commit),
2152                                        get_commit_trans_len(commit));
2153                         brelse(c_bh);
2154                         if (oldest_invalid_trans_id) {
2155                                 *oldest_invalid_trans_id =
2156                                     get_desc_trans_id(desc);
2157                                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2158                                                "journal-1004: "
2159                                                "transaction_is_valid setting oldest invalid trans_id "
2160                                                "to %d",
2161                                                get_desc_trans_id(desc));
2162                         }
2163                         return -1;
2164                 }
2165                 brelse(c_bh);
2166                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2167                                "journal-1006: found valid "
2168                                "transaction start offset %llu, len %d id %d",
2169                                d_bh->b_blocknr -
2170                                SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2171                                get_desc_trans_len(desc),
2172                                get_desc_trans_id(desc));
2173                 return 1;
2174         } else {
2175                 return 0;
2176         }
2177 }
2178
2179 static void brelse_array(struct buffer_head **heads, int num)
2180 {
2181         int i;
2182         for (i = 0; i < num; i++) {
2183                 brelse(heads[i]);
2184         }
2185 }
2186
2187 /*
2188 ** given the start, and values for the oldest acceptable transactions,
2189 ** this either reads in a replays a transaction, or returns because the transaction
2190 ** is invalid, or too old.
2191 */
2192 static int journal_read_transaction(struct super_block *sb,
2193                                     unsigned long cur_dblock,
2194                                     unsigned long oldest_start,
2195                                     unsigned int oldest_trans_id,
2196                                     unsigned long newest_mount_id)
2197 {
2198         struct reiserfs_journal *journal = SB_JOURNAL(sb);
2199         struct reiserfs_journal_desc *desc;
2200         struct reiserfs_journal_commit *commit;
2201         unsigned int trans_id = 0;
2202         struct buffer_head *c_bh;
2203         struct buffer_head *d_bh;
2204         struct buffer_head **log_blocks = NULL;
2205         struct buffer_head **real_blocks = NULL;
2206         unsigned int trans_offset;
2207         int i;
2208         int trans_half;
2209
2210         d_bh = journal_bread(sb, cur_dblock);
2211         if (!d_bh)
2212                 return 1;
2213         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2214         trans_offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2215         reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1037: "
2216                        "journal_read_transaction, offset %llu, len %d mount_id %d",
2217                        d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2218                        get_desc_trans_len(desc), get_desc_mount_id(desc));
2219         if (get_desc_trans_id(desc) < oldest_trans_id) {
2220                 reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1039: "
2221                                "journal_read_trans skipping because %lu is too old",
2222                                cur_dblock -
2223                                SB_ONDISK_JOURNAL_1st_BLOCK(sb));
2224                 brelse(d_bh);
2225                 return 1;
2226         }
2227         if (get_desc_mount_id(desc) != newest_mount_id) {
2228                 reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1146: "
2229                                "journal_read_trans skipping because %d is != "
2230                                "newest_mount_id %lu", get_desc_mount_id(desc),
2231                                newest_mount_id);
2232                 brelse(d_bh);
2233                 return 1;
2234         }
2235         c_bh = journal_bread(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2236                              ((trans_offset + get_desc_trans_len(desc) + 1) %
2237                               SB_ONDISK_JOURNAL_SIZE(sb)));
2238         if (!c_bh) {
2239                 brelse(d_bh);
2240                 return 1;
2241         }
2242         commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2243         if (journal_compare_desc_commit(sb, desc, commit)) {
2244                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2245                                "journal_read_transaction, "
2246                                "commit offset %llu had bad time %d or length %d",
2247                                c_bh->b_blocknr -
2248                                SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2249                                get_commit_trans_id(commit),
2250                                get_commit_trans_len(commit));
2251                 brelse(c_bh);
2252                 brelse(d_bh);
2253                 return 1;
2254         }
2255         trans_id = get_desc_trans_id(desc);
2256         /* now we know we've got a good transaction, and it was inside the valid time ranges */
2257         log_blocks = kmalloc(get_desc_trans_len(desc) *
2258                              sizeof(struct buffer_head *), GFP_NOFS);
2259         real_blocks = kmalloc(get_desc_trans_len(desc) *
2260                               sizeof(struct buffer_head *), GFP_NOFS);
2261         if (!log_blocks || !real_blocks) {
2262                 brelse(c_bh);
2263                 brelse(d_bh);
2264                 kfree(log_blocks);
2265                 kfree(real_blocks);
2266                 reiserfs_warning(sb, "journal-1169",
2267                                  "kmalloc failed, unable to mount FS");
2268                 return -1;
2269         }
2270         /* get all the buffer heads */
2271         trans_half = journal_trans_half(sb->s_blocksize);
2272         for (i = 0; i < get_desc_trans_len(desc); i++) {
2273                 log_blocks[i] =
2274                     journal_getblk(sb,
2275                                    SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2276                                    (trans_offset + 1 +
2277                                     i) % SB_ONDISK_JOURNAL_SIZE(sb));
2278                 if (i < trans_half) {
2279                         real_blocks[i] =
2280                             sb_getblk(sb,
2281                                       le32_to_cpu(desc->j_realblock[i]));
2282                 } else {
2283                         real_blocks[i] =
2284                             sb_getblk(sb,
2285                                       le32_to_cpu(commit->
2286                                                   j_realblock[i - trans_half]));
2287                 }
2288                 if (real_blocks[i]->b_blocknr > SB_BLOCK_COUNT(sb)) {
2289                         reiserfs_warning(sb, "journal-1207",
2290                                          "REPLAY FAILURE fsck required! "
2291                                          "Block to replay is outside of "
2292                                          "filesystem");
2293                         goto abort_replay;
2294                 }
2295                 /* make sure we don't try to replay onto log or reserved area */
2296                 if (is_block_in_log_or_reserved_area
2297                     (sb, real_blocks[i]->b_blocknr)) {
2298                         reiserfs_warning(sb, "journal-1204",
2299                                          "REPLAY FAILURE fsck required! "
2300                                          "Trying to replay onto a log block");
2301                       abort_replay:
2302                         brelse_array(log_blocks, i);
2303                         brelse_array(real_blocks, i);
2304                         brelse(c_bh);
2305                         brelse(d_bh);
2306                         kfree(log_blocks);
2307                         kfree(real_blocks);
2308                         return -1;
2309                 }
2310         }
2311         /* read in the log blocks, memcpy to the corresponding real block */
2312         ll_rw_block(READ, get_desc_trans_len(desc), log_blocks);
2313         for (i = 0; i < get_desc_trans_len(desc); i++) {
2314
2315                 reiserfs_write_unlock(sb);
2316                 wait_on_buffer(log_blocks[i]);
2317                 reiserfs_write_lock(sb);
2318
2319                 if (!buffer_uptodate(log_blocks[i])) {
2320                         reiserfs_warning(sb, "journal-1212",
2321                                          "REPLAY FAILURE fsck required! "
2322                                          "buffer write failed");
2323                         brelse_array(log_blocks + i,
2324                                      get_desc_trans_len(desc) - i);
2325                         brelse_array(real_blocks, get_desc_trans_len(desc));
2326                         brelse(c_bh);
2327                         brelse(d_bh);
2328                         kfree(log_blocks);
2329                         kfree(real_blocks);
2330                         return -1;
2331                 }
2332                 memcpy(real_blocks[i]->b_data, log_blocks[i]->b_data,
2333                        real_blocks[i]->b_size);
2334                 set_buffer_uptodate(real_blocks[i]);
2335                 brelse(log_blocks[i]);
2336         }
2337         /* flush out the real blocks */
2338         for (i = 0; i < get_desc_trans_len(desc); i++) {
2339                 set_buffer_dirty(real_blocks[i]);
2340                 ll_rw_block(SWRITE, 1, real_blocks + i);
2341         }
2342         for (i = 0; i < get_desc_trans_len(desc); i++) {
2343                 wait_on_buffer(real_blocks[i]);
2344                 if (!buffer_uptodate(real_blocks[i])) {
2345                         reiserfs_warning(sb, "journal-1226",
2346                                          "REPLAY FAILURE, fsck required! "
2347                                          "buffer write failed");
2348                         brelse_array(real_blocks + i,
2349                                      get_desc_trans_len(desc) - i);
2350                         brelse(c_bh);
2351                         brelse(d_bh);
2352                         kfree(log_blocks);
2353                         kfree(real_blocks);
2354                         return -1;
2355                 }
2356                 brelse(real_blocks[i]);
2357         }
2358         cur_dblock =
2359             SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2360             ((trans_offset + get_desc_trans_len(desc) +
2361               2) % SB_ONDISK_JOURNAL_SIZE(sb));
2362         reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2363                        "journal-1095: setting journal " "start to offset %ld",
2364                        cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb));
2365
2366         /* init starting values for the first transaction, in case this is the last transaction to be replayed. */
2367         journal->j_start = cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2368         journal->j_last_flush_trans_id = trans_id;
2369         journal->j_trans_id = trans_id + 1;
2370         /* check for trans_id overflow */
2371         if (journal->j_trans_id == 0)
2372                 journal->j_trans_id = 10;
2373         brelse(c_bh);
2374         brelse(d_bh);
2375         kfree(log_blocks);
2376         kfree(real_blocks);
2377         return 0;
2378 }
2379
2380 /* This function reads blocks starting from block and to max_block of bufsize
2381    size (but no more than BUFNR blocks at a time). This proved to improve
2382    mounting speed on self-rebuilding raid5 arrays at least.
2383    Right now it is only used from journal code. But later we might use it
2384    from other places.
2385    Note: Do not use journal_getblk/sb_getblk functions here! */
2386 static struct buffer_head *reiserfs_breada(struct block_device *dev,
2387                                            b_blocknr_t block, int bufsize,
2388                                            b_blocknr_t max_block)
2389 {
2390         struct buffer_head *bhlist[BUFNR];
2391         unsigned int blocks = BUFNR;
2392         struct buffer_head *bh;
2393         int i, j;
2394
2395         bh = __getblk(dev, block, bufsize);
2396         if (buffer_uptodate(bh))
2397                 return (bh);
2398
2399         if (block + BUFNR > max_block) {
2400                 blocks = max_block - block;
2401         }
2402         bhlist[0] = bh;
2403         j = 1;
2404         for (i = 1; i < blocks; i++) {
2405                 bh = __getblk(dev, block + i, bufsize);
2406                 if (buffer_uptodate(bh)) {
2407                         brelse(bh);
2408                         break;
2409                 } else
2410                         bhlist[j++] = bh;
2411         }
2412         ll_rw_block(READ, j, bhlist);
2413         for (i = 1; i < j; i++)
2414                 brelse(bhlist[i]);
2415         bh = bhlist[0];
2416         wait_on_buffer(bh);
2417         if (buffer_uptodate(bh))
2418                 return bh;
2419         brelse(bh);
2420         return NULL;
2421 }
2422
2423 /*
2424 ** read and replay the log
2425 ** on a clean unmount, the journal header's next unflushed pointer will be to an invalid
2426 ** transaction.  This tests that before finding all the transactions in the log, which makes normal mount times fast.
2427 **
2428 ** After a crash, this starts with the next unflushed transaction, and replays until it finds one too old, or invalid.
2429 **
2430 ** On exit, it sets things up so the first transaction will work correctly.
2431 */
2432 static int journal_read(struct super_block *sb)
2433 {
2434         struct reiserfs_journal *journal = SB_JOURNAL(sb);
2435         struct reiserfs_journal_desc *desc;
2436         unsigned int oldest_trans_id = 0;
2437         unsigned int oldest_invalid_trans_id = 0;
2438         time_t start;
2439         unsigned long oldest_start = 0;
2440         unsigned long cur_dblock = 0;
2441         unsigned long newest_mount_id = 9;
2442         struct buffer_head *d_bh;
2443         struct reiserfs_journal_header *jh;
2444         int valid_journal_header = 0;
2445         int replay_count = 0;
2446         int continue_replay = 1;
2447         int ret;
2448         char b[BDEVNAME_SIZE];
2449
2450         cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2451         reiserfs_info(sb, "checking transaction log (%s)\n",
2452                       bdevname(journal->j_dev_bd, b));
2453         start = get_seconds();
2454
2455         /* step 1, read in the journal header block.  Check the transaction it says
2456          ** is the first unflushed, and if that transaction is not valid,
2457          ** replay is done
2458          */
2459         journal->j_header_bh = journal_bread(sb,
2460                                              SB_ONDISK_JOURNAL_1st_BLOCK(sb)
2461                                              + SB_ONDISK_JOURNAL_SIZE(sb));
2462         if (!journal->j_header_bh) {
2463                 return 1;
2464         }
2465         jh = (struct reiserfs_journal_header *)(journal->j_header_bh->b_data);
2466         if (le32_to_cpu(jh->j_first_unflushed_offset) <
2467             SB_ONDISK_JOURNAL_SIZE(sb)
2468             && le32_to_cpu(jh->j_last_flush_trans_id) > 0) {
2469                 oldest_start =
2470                     SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2471                     le32_to_cpu(jh->j_first_unflushed_offset);
2472                 oldest_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2473                 newest_mount_id = le32_to_cpu(jh->j_mount_id);
2474                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2475                                "journal-1153: found in "
2476                                "header: first_unflushed_offset %d, last_flushed_trans_id "
2477                                "%lu", le32_to_cpu(jh->j_first_unflushed_offset),
2478                                le32_to_cpu(jh->j_last_flush_trans_id));
2479                 valid_journal_header = 1;
2480
2481                 /* now, we try to read the first unflushed offset.  If it is not valid,
2482                  ** there is nothing more we can do, and it makes no sense to read
2483                  ** through the whole log.
2484                  */
2485                 d_bh =
2486                     journal_bread(sb,
2487                                   SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2488                                   le32_to_cpu(jh->j_first_unflushed_offset));
2489                 ret = journal_transaction_is_valid(sb, d_bh, NULL, NULL);
2490                 if (!ret) {
2491                         continue_replay = 0;
2492                 }
2493                 brelse(d_bh);
2494                 goto start_log_replay;
2495         }
2496
2497         if (continue_replay && bdev_read_only(sb->s_bdev)) {
2498                 reiserfs_warning(sb, "clm-2076",
2499                                  "device is readonly, unable to replay log");
2500                 return -1;
2501         }
2502
2503         /* ok, there are transactions that need to be replayed.  start with the first log block, find
2504          ** all the valid transactions, and pick out the oldest.
2505          */
2506         while (continue_replay
2507                && cur_dblock <
2508                (SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2509                 SB_ONDISK_JOURNAL_SIZE(sb))) {
2510                 /* Note that it is required for blocksize of primary fs device and journal
2511                    device to be the same */
2512                 d_bh =
2513                     reiserfs_breada(journal->j_dev_bd, cur_dblock,
2514                                     sb->s_blocksize,
2515                                     SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2516                                     SB_ONDISK_JOURNAL_SIZE(sb));
2517                 ret =
2518                     journal_transaction_is_valid(sb, d_bh,
2519                                                  &oldest_invalid_trans_id,
2520                                                  &newest_mount_id);
2521                 if (ret == 1) {
2522                         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2523                         if (oldest_start == 0) {        /* init all oldest_ values */
2524                                 oldest_trans_id = get_desc_trans_id(desc);
2525                                 oldest_start = d_bh->b_blocknr;
2526                                 newest_mount_id = get_desc_mount_id(desc);
2527                                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2528                                                "journal-1179: Setting "
2529                                                "oldest_start to offset %llu, trans_id %lu",
2530                                                oldest_start -
2531                                                SB_ONDISK_JOURNAL_1st_BLOCK
2532                                                (sb), oldest_trans_id);
2533                         } else if (oldest_trans_id > get_desc_trans_id(desc)) {
2534                                 /* one we just read was older */
2535                                 oldest_trans_id = get_desc_trans_id(desc);
2536                                 oldest_start = d_bh->b_blocknr;
2537                                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2538                                                "journal-1180: Resetting "
2539                                                "oldest_start to offset %lu, trans_id %lu",
2540                                                oldest_start -
2541                                                SB_ONDISK_JOURNAL_1st_BLOCK
2542                                                (sb), oldest_trans_id);
2543                         }
2544                         if (newest_mount_id < get_desc_mount_id(desc)) {
2545                                 newest_mount_id = get_desc_mount_id(desc);
2546                                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2547                                                "journal-1299: Setting "
2548                                                "newest_mount_id to %d",
2549                                                get_desc_mount_id(desc));
2550                         }
2551                         cur_dblock += get_desc_trans_len(desc) + 2;
2552                 } else {
2553                         cur_dblock++;
2554                 }
2555                 brelse(d_bh);
2556         }
2557
2558       start_log_replay:
2559         cur_dblock = oldest_start;
2560         if (oldest_trans_id) {
2561                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2562                                "journal-1206: Starting replay "
2563                                "from offset %llu, trans_id %lu",
2564                                cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2565                                oldest_trans_id);
2566
2567         }
2568         replay_count = 0;
2569         while (continue_replay && oldest_trans_id > 0) {
2570                 ret =
2571                     journal_read_transaction(sb, cur_dblock, oldest_start,
2572                                              oldest_trans_id, newest_mount_id);
2573                 if (ret < 0) {
2574                         return ret;
2575                 } else if (ret != 0) {
2576                         break;
2577                 }
2578                 cur_dblock =
2579                     SB_ONDISK_JOURNAL_1st_BLOCK(sb) + journal->j_start;
2580                 replay_count++;
2581                 if (cur_dblock == oldest_start)
2582                         break;
2583         }
2584
2585         if (oldest_trans_id == 0) {
2586                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2587                                "journal-1225: No valid " "transactions found");
2588         }
2589         /* j_start does not get set correctly if we don't replay any transactions.
2590          ** if we had a valid journal_header, set j_start to the first unflushed transaction value,
2591          ** copy the trans_id from the header
2592          */
2593         if (valid_journal_header && replay_count == 0) {
2594                 journal->j_start = le32_to_cpu(jh->j_first_unflushed_offset);
2595                 journal->j_trans_id =
2596                     le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2597                 /* check for trans_id overflow */
2598                 if (journal->j_trans_id == 0)
2599                         journal->j_trans_id = 10;
2600                 journal->j_last_flush_trans_id =
2601                     le32_to_cpu(jh->j_last_flush_trans_id);
2602                 journal->j_mount_id = le32_to_cpu(jh->j_mount_id) + 1;
2603         } else {
2604                 journal->j_mount_id = newest_mount_id + 1;
2605         }
2606         reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1299: Setting "
2607                        "newest_mount_id to %lu", journal->j_mount_id);
2608         journal->j_first_unflushed_offset = journal->j_start;
2609         if (replay_count > 0) {
2610                 reiserfs_info(sb,
2611                               "replayed %d transactions in %lu seconds\n",
2612                               replay_count, get_seconds() - start);
2613         }
2614         if (!bdev_read_only(sb->s_bdev) &&
2615             _update_journal_header_block(sb, journal->j_start,
2616                                          journal->j_last_flush_trans_id)) {
2617                 /* replay failed, caller must call free_journal_ram and abort
2618                  ** the mount
2619                  */
2620                 return -1;
2621         }
2622         return 0;
2623 }
2624
2625 static struct reiserfs_journal_list *alloc_journal_list(struct super_block *s)
2626 {
2627         struct reiserfs_journal_list *jl;
2628         jl = kzalloc(sizeof(struct reiserfs_journal_list),
2629                      GFP_NOFS | __GFP_NOFAIL);
2630         INIT_LIST_HEAD(&jl->j_list);
2631         INIT_LIST_HEAD(&jl->j_working_list);
2632         INIT_LIST_HEAD(&jl->j_tail_bh_list);
2633         INIT_LIST_HEAD(&jl->j_bh_list);
2634         mutex_init(&jl->j_commit_mutex);
2635         SB_JOURNAL(s)->j_num_lists++;
2636         get_journal_list(jl);
2637         return jl;
2638 }
2639
2640 static void journal_list_init(struct super_block *sb)
2641 {
2642         SB_JOURNAL(sb)->j_current_jl = alloc_journal_list(sb);
2643 }
2644
2645 static int release_journal_dev(struct super_block *super,
2646                                struct reiserfs_journal *journal)
2647 {
2648         int result;
2649
2650         result = 0;
2651
2652         if (journal->j_dev_bd != NULL) {
2653                 if (journal->j_dev_bd->bd_dev != super->s_dev)
2654                         bd_release(journal->j_dev_bd);
2655                 result = blkdev_put(journal->j_dev_bd, journal->j_dev_mode);
2656                 journal->j_dev_bd = NULL;
2657         }
2658
2659         if (result != 0) {
2660                 reiserfs_warning(super, "sh-457",
2661                                  "Cannot release journal device: %i", result);
2662         }
2663         return result;
2664 }
2665
2666 static int journal_init_dev(struct super_block *super,
2667                             struct reiserfs_journal *journal,
2668                             const char *jdev_name)
2669 {
2670         int result;
2671         dev_t jdev;
2672         fmode_t blkdev_mode = FMODE_READ | FMODE_WRITE;
2673         char b[BDEVNAME_SIZE];
2674
2675         result = 0;
2676
2677         journal->j_dev_bd = NULL;
2678         jdev = SB_ONDISK_JOURNAL_DEVICE(super) ?
2679             new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super)) : super->s_dev;
2680
2681         if (bdev_read_only(super->s_bdev))
2682                 blkdev_mode = FMODE_READ;
2683
2684         /* there is no "jdev" option and journal is on separate device */
2685         if ((!jdev_name || !jdev_name[0])) {
2686                 journal->j_dev_bd = open_by_devnum(jdev, blkdev_mode);
2687                 journal->j_dev_mode = blkdev_mode;
2688                 if (IS_ERR(journal->j_dev_bd)) {
2689                         result = PTR_ERR(journal->j_dev_bd);
2690                         journal->j_dev_bd = NULL;
2691                         reiserfs_warning(super, "sh-458",
2692                                          "cannot init journal device '%s': %i",
2693                                          __bdevname(jdev, b), result);
2694                         return result;
2695                 } else if (jdev != super->s_dev) {
2696                         result = bd_claim(journal->j_dev_bd, journal);
2697                         if (result) {
2698                                 blkdev_put(journal->j_dev_bd, blkdev_mode);
2699                                 return result;
2700                         }
2701
2702                         set_blocksize(journal->j_dev_bd, super->s_blocksize);
2703                 }
2704
2705                 return 0;
2706         }
2707
2708         journal->j_dev_mode = blkdev_mode;
2709         journal->j_dev_bd = open_bdev_exclusive(jdev_name,
2710                                                 blkdev_mode, journal);
2711         if (IS_ERR(journal->j_dev_bd)) {
2712                 result = PTR_ERR(journal->j_dev_bd);
2713                 journal->j_dev_bd = NULL;
2714                 reiserfs_warning(super,
2715                                  "journal_init_dev: Cannot open '%s': %i",
2716                                  jdev_name, result);
2717                 return result;
2718         }
2719
2720         set_blocksize(journal->j_dev_bd, super->s_blocksize);
2721         reiserfs_info(super,
2722                       "journal_init_dev: journal device: %s\n",
2723                       bdevname(journal->j_dev_bd, b));
2724         return 0;
2725 }
2726
2727 /**
2728  * When creating/tuning a file system user can assign some
2729  * journal params within boundaries which depend on the ratio
2730  * blocksize/standard_blocksize.
2731  *
2732  * For blocks >= standard_blocksize transaction size should
2733  * be not less then JOURNAL_TRANS_MIN_DEFAULT, and not more
2734  * then JOURNAL_TRANS_MAX_DEFAULT.
2735  *
2736  * For blocks < standard_blocksize these boundaries should be
2737  * decreased proportionally.
2738  */
2739 #define REISERFS_STANDARD_BLKSIZE (4096)
2740
2741 static int check_advise_trans_params(struct super_block *sb,
2742                                      struct reiserfs_journal *journal)
2743 {
2744         if (journal->j_trans_max) {
2745                 /* Non-default journal params.
2746                    Do sanity check for them. */
2747                 int ratio = 1;
2748                 if (sb->s_blocksize < REISERFS_STANDARD_BLKSIZE)
2749                         ratio = REISERFS_STANDARD_BLKSIZE / sb->s_blocksize;
2750
2751                 if (journal->j_trans_max > JOURNAL_TRANS_MAX_DEFAULT / ratio ||
2752                     journal->j_trans_max < JOURNAL_TRANS_MIN_DEFAULT / ratio ||
2753                     SB_ONDISK_JOURNAL_SIZE(sb) / journal->j_trans_max <
2754                     JOURNAL_MIN_RATIO) {
2755                         reiserfs_warning(sb, "sh-462",
2756                                          "bad transaction max size (%u). "
2757                                          "FSCK?", journal->j_trans_max);
2758                         return 1;
2759                 }
2760                 if (journal->j_max_batch != (journal->j_trans_max) *
2761                         JOURNAL_MAX_BATCH_DEFAULT/JOURNAL_TRANS_MAX_DEFAULT) {
2762                         reiserfs_warning(sb, "sh-463",
2763                                          "bad transaction max batch (%u). "
2764                                          "FSCK?", journal->j_max_batch);
2765                         return 1;
2766                 }
2767         } else {
2768                 /* Default journal params.
2769                    The file system was created by old version
2770                    of mkreiserfs, so some fields contain zeros,
2771                    and we need to advise proper values for them */
2772                 if (sb->s_blocksize != REISERFS_STANDARD_BLKSIZE) {
2773                         reiserfs_warning(sb, "sh-464", "bad blocksize (%u)",
2774                                          sb->s_blocksize);
2775                         return 1;
2776                 }
2777                 journal->j_trans_max = JOURNAL_TRANS_MAX_DEFAULT;
2778                 journal->j_max_batch = JOURNAL_MAX_BATCH_DEFAULT;
2779                 journal->j_max_commit_age = JOURNAL_MAX_COMMIT_AGE;
2780         }
2781         return 0;
2782 }
2783
2784 /*
2785 ** must be called once on fs mount.  calls journal_read for you
2786 */
2787 int journal_init(struct super_block *sb, const char *j_dev_name,
2788                  int old_format, unsigned int commit_max_age)
2789 {
2790         int num_cnodes = SB_ONDISK_JOURNAL_SIZE(sb) * 2;
2791         struct buffer_head *bhjh;
2792         struct reiserfs_super_block *rs;
2793         struct reiserfs_journal_header *jh;
2794         struct reiserfs_journal *journal;
2795         struct reiserfs_journal_list *jl;
2796         char b[BDEVNAME_SIZE];
2797
2798         journal = SB_JOURNAL(sb) = vmalloc(sizeof(struct reiserfs_journal));
2799         if (!journal) {
2800                 reiserfs_warning(sb, "journal-1256",
2801                                  "unable to get memory for journal structure");
2802                 return 1;
2803         }
2804         memset(journal, 0, sizeof(struct reiserfs_journal));
2805         INIT_LIST_HEAD(&journal->j_bitmap_nodes);
2806         INIT_LIST_HEAD(&journal->j_prealloc_list);
2807         INIT_LIST_HEAD(&journal->j_working_list);
2808         INIT_LIST_HEAD(&journal->j_journal_list);
2809         journal->j_persistent_trans = 0;
2810         if (reiserfs_allocate_list_bitmaps(sb,
2811                                            journal->j_list_bitmap,
2812                                            reiserfs_bmap_count(sb)))
2813                 goto free_and_return;
2814         allocate_bitmap_nodes(sb);
2815
2816         /* reserved for journal area support */
2817         SB_JOURNAL_1st_RESERVED_BLOCK(sb) = (old_format ?
2818                                                  REISERFS_OLD_DISK_OFFSET_IN_BYTES
2819                                                  / sb->s_blocksize +
2820                                                  reiserfs_bmap_count(sb) +
2821                                                  1 :
2822                                                  REISERFS_DISK_OFFSET_IN_BYTES /
2823                                                  sb->s_blocksize + 2);
2824
2825         /* Sanity check to see is the standard journal fitting withing first bitmap
2826            (actual for small blocksizes) */
2827         if (!SB_ONDISK_JOURNAL_DEVICE(sb) &&
2828             (SB_JOURNAL_1st_RESERVED_BLOCK(sb) +
2829              SB_ONDISK_JOURNAL_SIZE(sb) > sb->s_blocksize * 8)) {
2830                 reiserfs_warning(sb, "journal-1393",
2831                                  "journal does not fit for area addressed "
2832                                  "by first of bitmap blocks. It starts at "
2833                                  "%u and its size is %u. Block size %ld",
2834                                  SB_JOURNAL_1st_RESERVED_BLOCK(sb),
2835                                  SB_ONDISK_JOURNAL_SIZE(sb),
2836                                  sb->s_blocksize);
2837                 goto free_and_return;
2838         }
2839
2840         if (journal_init_dev(sb, journal, j_dev_name) != 0) {
2841                 reiserfs_warning(sb, "sh-462",
2842                                  "unable to initialize jornal device");
2843                 goto free_and_return;
2844         }
2845
2846         rs = SB_DISK_SUPER_BLOCK(sb);
2847
2848         /* read journal header */
2849         bhjh = journal_bread(sb,
2850                              SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2851                              SB_ONDISK_JOURNAL_SIZE(sb));
2852         if (!bhjh) {
2853                 reiserfs_warning(sb, "sh-459",
2854                                  "unable to read journal header");
2855                 goto free_and_return;
2856         }
2857         jh = (struct reiserfs_journal_header *)(bhjh->b_data);
2858
2859         /* make sure that journal matches to the super block */
2860         if (is_reiserfs_jr(rs)
2861             && (le32_to_cpu(jh->jh_journal.jp_journal_magic) !=
2862                 sb_jp_journal_magic(rs))) {
2863                 reiserfs_warning(sb, "sh-460",
2864                                  "journal header magic %x (device %s) does "
2865                                  "not match to magic found in super block %x",
2866                                  jh->jh_journal.jp_journal_magic,
2867                                  bdevname(journal->j_dev_bd, b),
2868                                  sb_jp_journal_magic(rs));
2869                 brelse(bhjh);
2870                 goto free_and_return;
2871         }
2872
2873         journal->j_trans_max = le32_to_cpu(jh->jh_journal.jp_journal_trans_max);
2874         journal->j_max_batch = le32_to_cpu(jh->jh_journal.jp_journal_max_batch);
2875         journal->j_max_commit_age =
2876             le32_to_cpu(jh->jh_journal.jp_journal_max_commit_age);
2877         journal->j_max_trans_age = JOURNAL_MAX_TRANS_AGE;
2878
2879         if (check_advise_trans_params(sb, journal) != 0)
2880                 goto free_and_return;
2881         journal->j_default_max_commit_age = journal->j_max_commit_age;
2882
2883         if (commit_max_age != 0) {
2884                 journal->j_max_commit_age = commit_max_age;
2885                 journal->j_max_trans_age = commit_max_age;
2886         }
2887
2888         reiserfs_info(sb, "journal params: device %s, size %u, "
2889                       "journal first block %u, max trans len %u, max batch %u, "
2890                       "max commit age %u, max trans age %u\n",
2891                       bdevname(journal->j_dev_bd, b),
2892                       SB_ONDISK_JOURNAL_SIZE(sb),
2893                       SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2894                       journal->j_trans_max,
2895                       journal->j_max_batch,
2896                       journal->j_max_commit_age, journal->j_max_trans_age);
2897
2898         brelse(bhjh);
2899
2900         journal->j_list_bitmap_index = 0;
2901         journal_list_init(sb);
2902
2903         memset(journal->j_list_hash_table, 0,
2904                JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
2905
2906         INIT_LIST_HEAD(&journal->j_dirty_buffers);
2907         spin_lock_init(&journal->j_dirty_buffers_lock);
2908
2909         journal->j_start = 0;
2910         journal->j_len = 0;
2911         journal->j_len_alloc = 0;
2912         atomic_set(&(journal->j_wcount), 0);
2913         atomic_set(&(journal->j_async_throttle), 0);
2914         journal->j_bcount = 0;
2915         journal->j_trans_start_time = 0;
2916         journal->j_last = NULL;
2917         journal->j_first = NULL;
2918         init_waitqueue_head(&(journal->j_join_wait));
2919         mutex_init(&journal->j_mutex);
2920         mutex_init(&journal->j_flush_mutex);
2921
2922         journal->j_trans_id = 10;
2923         journal->j_mount_id = 10;
2924         journal->j_state = 0;
2925         atomic_set(&(journal->j_jlock), 0);
2926         journal->j_cnode_free_list = allocate_cnodes(num_cnodes);
2927         journal->j_cnode_free_orig = journal->j_cnode_free_list;
2928         journal->j_cnode_free = journal->j_cnode_free_list ? num_cnodes : 0;
2929         journal->j_cnode_used = 0;
2930         journal->j_must_wait = 0;
2931
2932         if (journal->j_cnode_free == 0) {
2933                 reiserfs_warning(sb, "journal-2004", "Journal cnode memory "
2934                                  "allocation failed (%ld bytes). Journal is "
2935                                  "too large for available memory. Usually "
2936                                  "this is due to a journal that is too large.",
2937                                  sizeof (struct reiserfs_journal_cnode) * num_cnodes);
2938                 goto free_and_return;
2939         }
2940
2941         init_journal_hash(sb);
2942         jl = journal->j_current_jl;
2943         jl->j_list_bitmap = get_list_bitmap(sb, jl);
2944         if (!jl->j_list_bitmap) {
2945                 reiserfs_warning(sb, "journal-2005",
2946                                  "get_list_bitmap failed for journal list 0");
2947                 goto free_and_return;
2948         }
2949         if (journal_read(sb) < 0) {
2950                 reiserfs_warning(sb, "reiserfs-2006",
2951                                  "Replay Failure, unable to mount");
2952                 goto free_and_return;
2953         }
2954
2955         reiserfs_mounted_fs_count++;
2956         if (reiserfs_mounted_fs_count <= 1)
2957                 commit_wq = create_workqueue("reiserfs");
2958
2959         INIT_DELAYED_WORK(&journal->j_work, flush_async_commits);
2960         journal->j_work_sb = sb;
2961         return 0;
2962       free_and_return:
2963         free_journal_ram(sb);
2964         return 1;
2965 }
2966
2967 /*
2968 ** test for a polite end of the current transaction.  Used by file_write, and should
2969 ** be used by delete to make sure they don't write more than can fit inside a single
2970 ** transaction
2971 */
2972 int journal_transaction_should_end(struct reiserfs_transaction_handle *th,
2973                                    int new_alloc)
2974 {
2975         struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2976         time_t now = get_seconds();
2977         /* cannot restart while nested */
2978         BUG_ON(!th->t_trans_id);
2979         if (th->t_refcount > 1)
2980                 return 0;
2981         if (journal->j_must_wait > 0 ||
2982             (journal->j_len_alloc + new_alloc) >= journal->j_max_batch ||
2983             atomic_read(&(journal->j_jlock)) ||
2984             (now - journal->j_trans_start_time) > journal->j_max_trans_age ||
2985             journal->j_cnode_free < (journal->j_trans_max * 3)) {
2986                 return 1;
2987         }
2988         /* protected by the BKL here */
2989         journal->j_len_alloc += new_alloc;
2990         th->t_blocks_allocated += new_alloc ;
2991         return 0;
2992 }
2993
2994 /* this must be called inside a transaction, and requires the
2995 ** kernel_lock to be held
2996 */
2997 void reiserfs_block_writes(struct reiserfs_transaction_handle *th)
2998 {
2999         struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
3000         BUG_ON(!th->t_trans_id);
3001         journal->j_must_wait = 1;
3002         set_bit(J_WRITERS_BLOCKED, &journal->j_state);
3003         return;
3004 }
3005
3006 /* this must be called without a transaction started, and does not
3007 ** require BKL
3008 */
3009 void reiserfs_allow_writes(struct super_block *s)
3010 {
3011         struct reiserfs_journal *journal = SB_JOURNAL(s);
3012         clear_bit(J_WRITERS_BLOCKED, &journal->j_state);
3013         wake_up(&journal->j_join_wait);
3014 }
3015
3016 /* this must be called without a transaction started, and does not
3017 ** require BKL
3018 */
3019 void reiserfs_wait_on_write_block(struct super_block *s)
3020 {
3021         struct reiserfs_journal *journal = SB_JOURNAL(s);
3022         wait_event(journal->j_join_wait,
3023                    !test_bit(J_WRITERS_BLOCKED, &journal->j_state));
3024 }
3025
3026 static void queue_log_writer(struct super_block *s)
3027 {
3028         wait_queue_t wait;
3029         struct reiserfs_journal *journal = SB_JOURNAL(s);
3030         set_bit(J_WRITERS_QUEUED, &journal->j_state);
3031
3032         /*
3033          * we don't want to use wait_event here because
3034          * we only want to wait once.
3035          */
3036         init_waitqueue_entry(&wait, current);
3037         add_wait_queue(&journal->j_join_wait, &wait);
3038         set_current_state(TASK_UNINTERRUPTIBLE);
3039         if (test_bit(J_WRITERS_QUEUED, &journal->j_state)) {
3040                 reiserfs_write_unlock(s);
3041                 schedule();
3042                 reiserfs_write_lock(s);
3043         }
3044         __set_current_state(TASK_RUNNING);
3045         remove_wait_queue(&journal->j_join_wait, &wait);
3046 }
3047
3048 static void wake_queued_writers(struct super_block *s)
3049 {
3050         struct reiserfs_journal *journal = SB_JOURNAL(s);
3051         if (test_and_clear_bit(J_WRITERS_QUEUED, &journal->j_state))
3052                 wake_up(&journal->j_join_wait);
3053 }
3054
3055 static void let_transaction_grow(struct super_block *sb, unsigned int trans_id)
3056 {
3057         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3058         unsigned long bcount = journal->j_bcount;
3059         while (1) {
3060                 reiserfs_write_unlock(sb);
3061                 schedule_timeout_uninterruptible(1);
3062                 reiserfs_write_lock(sb);
3063                 journal->j_current_jl->j_state |= LIST_COMMIT_PENDING;
3064                 while ((atomic_read(&journal->j_wcount) > 0 ||
3065                         atomic_read(&journal->j_jlock)) &&
3066                        journal->j_trans_id == trans_id) {
3067                         queue_log_writer(sb);
3068                 }
3069                 if (journal->j_trans_id != trans_id)
3070                         break;
3071                 if (bcount == journal->j_bcount)
3072                         break;
3073                 bcount = journal->j_bcount;
3074         }
3075 }
3076
3077 /* join == true if you must join an existing transaction.
3078 ** join == false if you can deal with waiting for others to finish
3079 **
3080 ** this will block until the transaction is joinable.  send the number of blocks you
3081 ** expect to use in nblocks.
3082 */
3083 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
3084                               struct super_block *sb, unsigned long nblocks,
3085                               int join)
3086 {
3087         time_t now = get_seconds();
3088         unsigned int old_trans_id;
3089         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3090         struct reiserfs_transaction_handle myth;
3091         int sched_count = 0;
3092         int retval;
3093
3094         reiserfs_check_lock_depth(sb, "journal_begin");
3095         BUG_ON(nblocks > journal->j_trans_max);
3096
3097         PROC_INFO_INC(sb, journal.journal_being);
3098         /* set here for journal_join */
3099         th->t_refcount = 1;
3100         th->t_super = sb;
3101
3102       relock:
3103         lock_journal(sb);
3104         if (join != JBEGIN_ABORT && reiserfs_is_journal_aborted(journal)) {
3105                 unlock_journal(sb);
3106                 retval = journal->j_errno;
3107                 goto out_fail;
3108         }
3109         journal->j_bcount++;
3110
3111         if (test_bit(J_WRITERS_BLOCKED, &journal->j_state)) {
3112                 unlock_journal(sb);
3113                 reiserfs_write_unlock(sb);
3114                 reiserfs_wait_on_write_block(sb);
3115                 reiserfs_write_lock(sb);
3116                 PROC_INFO_INC(sb, journal.journal_relock_writers);
3117                 goto relock;
3118         }
3119         now = get_seconds();
3120
3121         /* if there is no room in the journal OR
3122          ** if this transaction is too old, and we weren't called joinable, wait for it to finish before beginning
3123          ** we don't sleep if there aren't other writers
3124          */
3125
3126         if ((!join && journal->j_must_wait > 0) ||
3127             (!join
3128              && (journal->j_len_alloc + nblocks + 2) >= journal->j_max_batch)
3129             || (!join && atomic_read(&journal->j_wcount) > 0
3130                 && journal->j_trans_start_time > 0
3131                 && (now - journal->j_trans_start_time) >
3132                 journal->j_max_trans_age) || (!join
3133                                               && atomic_read(&journal->j_jlock))
3134             || (!join && journal->j_cnode_free < (journal->j_trans_max * 3))) {
3135
3136                 old_trans_id = journal->j_trans_id;
3137                 unlock_journal(sb);     /* allow others to finish this transaction */
3138
3139                 if (!join && (journal->j_len_alloc + nblocks + 2) >=
3140                     journal->j_max_batch &&
3141                     ((journal->j_len + nblocks + 2) * 100) <
3142                     (journal->j_len_alloc * 75)) {
3143                         if (atomic_read(&journal->j_wcount) > 10) {
3144                                 sched_count++;
3145                                 queue_log_writer(sb);
3146                                 goto relock;
3147                         }
3148                 }
3149                 /* don't mess with joining the transaction if all we have to do is
3150                  * wait for someone else to do a commit
3151                  */
3152                 if (atomic_read(&journal->j_jlock)) {
3153                         while (journal->j_trans_id == old_trans_id &&
3154                                atomic_read(&journal->j_jlock)) {
3155                                 queue_log_writer(sb);
3156                         }
3157                         goto relock;
3158                 }
3159                 retval = journal_join(&myth, sb, 1);
3160                 if (retval)
3161                         goto out_fail;
3162
3163                 /* someone might have ended the transaction while we joined */
3164                 if (old_trans_id != journal->j_trans_id) {
3165                         retval = do_journal_end(&myth, sb, 1, 0);
3166                 } else {
3167                         retval = do_journal_end(&myth, sb, 1, COMMIT_NOW);
3168                 }
3169
3170                 if (retval)
3171                         goto out_fail;
3172
3173                 PROC_INFO_INC(sb, journal.journal_relock_wcount);
3174                 goto relock;
3175         }
3176         /* we are the first writer, set trans_id */
3177         if (journal->j_trans_start_time == 0) {
3178                 journal->j_trans_start_time = get_seconds();
3179         }
3180         atomic_inc(&(journal->j_wcount));
3181         journal->j_len_alloc += nblocks;
3182         th->t_blocks_logged = 0;
3183         th->t_blocks_allocated = nblocks;
3184         th->t_trans_id = journal->j_trans_id;
3185         unlock_journal(sb);
3186         INIT_LIST_HEAD(&th->t_list);
3187         get_fs_excl();
3188         return 0;
3189
3190       out_fail:
3191         memset(th, 0, sizeof(*th));
3192         /* Re-set th->t_super, so we can properly keep track of how many
3193          * persistent transactions there are. We need to do this so if this
3194          * call is part of a failed restart_transaction, we can free it later */
3195         th->t_super = sb;
3196         return retval;
3197 }
3198
3199 struct reiserfs_transaction_handle *reiserfs_persistent_transaction(struct
3200                                                                     super_block
3201                                                                     *s,
3202                                                                     int nblocks)
3203 {
3204         int ret;
3205         struct reiserfs_transaction_handle *th;
3206
3207         /* if we're nesting into an existing transaction.  It will be
3208          ** persistent on its own
3209          */
3210         if (reiserfs_transaction_running(s)) {
3211                 th = current->journal_info;
3212                 th->t_refcount++;
3213                 BUG_ON(th->t_refcount < 2);
3214
3215                 return th;
3216         }
3217         th = kmalloc(sizeof(struct reiserfs_transaction_handle), GFP_NOFS);
3218         if (!th)
3219                 return NULL;
3220         ret = journal_begin(th, s, nblocks);
3221         if (ret) {
3222                 kfree(th);
3223                 return NULL;
3224         }
3225
3226         SB_JOURNAL(s)->j_persistent_trans++;
3227         return th;
3228 }
3229
3230 int reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle *th)
3231 {
3232         struct super_block *s = th->t_super;
3233         int ret = 0;
3234         if (th->t_trans_id)
3235                 ret = journal_end(th, th->t_super, th->t_blocks_allocated);
3236         else
3237                 ret = -EIO;
3238         if (th->t_refcount == 0) {
3239                 SB_JOURNAL(s)->j_persistent_trans--;
3240                 kfree(th);
3241         }
3242         return ret;
3243 }
3244
3245 static int journal_join(struct reiserfs_transaction_handle *th,
3246                         struct super_block *sb, unsigned long nblocks)
3247 {
3248         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3249
3250         /* this keeps do_journal_end from NULLing out the current->journal_info
3251          ** pointer
3252          */
3253         th->t_handle_save = cur_th;
3254         BUG_ON(cur_th && cur_th->t_refcount > 1);
3255         return do_journal_begin_r(th, sb, nblocks, JBEGIN_JOIN);
3256 }
3257
3258 int journal_join_abort(struct reiserfs_transaction_handle *th,
3259                        struct super_block *sb, unsigned long nblocks)
3260 {
3261         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3262
3263         /* this keeps do_journal_end from NULLing out the current->journal_info
3264          ** pointer
3265          */
3266         th->t_handle_save = cur_th;
3267         BUG_ON(cur_th && cur_th->t_refcount > 1);
3268         return do_journal_begin_r(th, sb, nblocks, JBEGIN_ABORT);
3269 }
3270
3271 int journal_begin(struct reiserfs_transaction_handle *th,
3272                   struct super_block *sb, unsigned long nblocks)
3273 {
3274         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3275         int ret;
3276
3277         th->t_handle_save = NULL;
3278         if (cur_th) {
3279                 /* we are nesting into the current transaction */
3280                 if (cur_th->t_super == sb) {
3281                         BUG_ON(!cur_th->t_refcount);
3282                         cur_th->t_refcount++;
3283                         memcpy(th, cur_th, sizeof(*th));
3284                         if (th->t_refcount <= 1)
3285                                 reiserfs_warning(sb, "reiserfs-2005",
3286                                                  "BAD: refcount <= 1, but "
3287                                                  "journal_info != 0");
3288                         return 0;
3289                 } else {
3290                         /* we've ended up with a handle from a different filesystem.
3291                          ** save it and restore on journal_end.  This should never
3292                          ** really happen...
3293                          */
3294                         reiserfs_warning(sb, "clm-2100",
3295                                          "nesting info a different FS");
3296                         th->t_handle_save = current->journal_info;
3297                         current->journal_info = th;
3298                 }
3299         } else {
3300                 current->journal_info = th;
3301         }
3302         ret = do_journal_begin_r(th, sb, nblocks, JBEGIN_REG);
3303         BUG_ON(current->journal_info != th);
3304
3305         /* I guess this boils down to being the reciprocal of clm-2100 above.
3306          * If do_journal_begin_r fails, we need to put it back, since journal_end
3307          * won't be called to do it. */
3308         if (ret)
3309                 current->journal_info = th->t_handle_save;
3310         else
3311                 BUG_ON(!th->t_refcount);
3312
3313         return ret;
3314 }
3315
3316 /*
3317 ** puts bh into the current transaction.  If it was already there, reorders removes the
3318 ** old pointers from the hash, and puts new ones in (to make sure replay happen in the right order).
3319 **
3320 ** if it was dirty, cleans and files onto the clean list.  I can't let it be dirty again until the
3321 ** transaction is committed.
3322 **
3323 ** if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len.
3324 */
3325 int journal_mark_dirty(struct reiserfs_transaction_handle *th,
3326                        struct super_block *sb, struct buffer_head *bh)
3327 {
3328         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3329         struct reiserfs_journal_cnode *cn = NULL;
3330         int count_already_incd = 0;
3331         int prepared = 0;
3332         BUG_ON(!th->t_trans_id);
3333
3334         PROC_INFO_INC(sb, journal.mark_dirty);
3335         if (th->t_trans_id != journal->j_trans_id) {
3336                 reiserfs_panic(th->t_super, "journal-1577",
3337                                "handle trans id %ld != current trans id %ld",
3338                                th->t_trans_id, journal->j_trans_id);
3339         }
3340
3341         sb->s_dirt = 1;
3342
3343         prepared = test_clear_buffer_journal_prepared(bh);
3344         clear_buffer_journal_restore_dirty(bh);
3345         /* already in this transaction, we are done */
3346         if (buffer_journaled(bh)) {
3347                 PROC_INFO_INC(sb, journal.mark_dirty_already);
3348                 return 0;
3349         }
3350
3351         /* this must be turned into a panic instead of a warning.  We can't allow
3352          ** a dirty or journal_dirty or locked buffer to be logged, as some changes
3353          ** could get to disk too early.  NOT GOOD.
3354          */
3355         if (!prepared || buffer_dirty(bh)) {
3356                 reiserfs_warning(sb, "journal-1777",
3357                                  "buffer %llu bad state "
3358                                  "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT",
3359                                  (unsigned long long)bh->b_blocknr,
3360                                  prepared ? ' ' : '!',
3361                                  buffer_locked(bh) ? ' ' : '!',
3362                                  buffer_dirty(bh) ? ' ' : '!',
3363                                  buffer_journal_dirty(bh) ? ' ' : '!');
3364         }
3365
3366         if (atomic_read(&(journal->j_wcount)) <= 0) {
3367                 reiserfs_warning(sb, "journal-1409",
3368                                  "returning because j_wcount was %d",
3369                                  atomic_read(&(journal->j_wcount)));
3370                 return 1;
3371         }
3372         /* this error means I've screwed up, and we've overflowed the transaction.
3373          ** Nothing can be done here, except make the FS readonly or panic.
3374          */
3375         if (journal->j_len >= journal->j_trans_max) {
3376                 reiserfs_panic(th->t_super, "journal-1413",
3377                                "j_len (%lu) is too big",
3378                                journal->j_len);
3379         }
3380
3381         if (buffer_journal_dirty(bh)) {
3382                 count_already_incd = 1;
3383                 PROC_INFO_INC(sb, journal.mark_dirty_notjournal);
3384                 clear_buffer_journal_dirty(bh);
3385         }
3386
3387         if (journal->j_len > journal->j_len_alloc) {
3388                 journal->j_len_alloc = journal->j_len + JOURNAL_PER_BALANCE_CNT;
3389         }
3390
3391         set_buffer_journaled(bh);
3392
3393         /* now put this guy on the end */
3394         if (!cn) {
3395                 cn = get_cnode(sb);
3396                 if (!cn) {
3397                         reiserfs_panic(sb, "journal-4", "get_cnode failed!");
3398                 }
3399
3400                 if (th->t_blocks_logged == th->t_blocks_allocated) {
3401                         th->t_blocks_allocated += JOURNAL_PER_BALANCE_CNT;
3402                         journal->j_len_alloc += JOURNAL_PER_BALANCE_CNT;
3403                 }
3404                 th->t_blocks_logged++;
3405                 journal->j_len++;
3406
3407                 cn->bh = bh;
3408                 cn->blocknr = bh->b_blocknr;
3409                 cn->sb = sb;
3410                 cn->jlist = NULL;
3411                 insert_journal_hash(journal->j_hash_table, cn);
3412                 if (!count_already_incd) {
3413                         get_bh(bh);
3414                 }
3415         }
3416         cn->next = NULL;
3417         cn->prev = journal->j_last;
3418         cn->bh = bh;
3419         if (journal->j_last) {
3420                 journal->j_last->next = cn;
3421                 journal->j_last = cn;
3422         } else {
3423                 journal->j_first = cn;
3424                 journal->j_last = cn;
3425         }
3426         return 0;
3427 }
3428
3429 int journal_end(struct reiserfs_transaction_handle *th,
3430                 struct super_block *sb, unsigned long nblocks)
3431 {
3432         if (!current->journal_info && th->t_refcount > 1)
3433                 reiserfs_warning(sb, "REISER-NESTING",
3434                                  "th NULL, refcount %d", th->t_refcount);
3435
3436         if (!th->t_trans_id) {
3437                 WARN_ON(1);
3438                 return -EIO;
3439         }
3440
3441         th->t_refcount--;
3442         if (th->t_refcount > 0) {
3443                 struct reiserfs_transaction_handle *cur_th =
3444                     current->journal_info;
3445
3446                 /* we aren't allowed to close a nested transaction on a different
3447                  ** filesystem from the one in the task struct
3448                  */
3449                 BUG_ON(cur_th->t_super != th->t_super);
3450
3451                 if (th != cur_th) {
3452                         memcpy(current->journal_info, th, sizeof(*th));
3453                         th->t_trans_id = 0;
3454                 }
3455                 return 0;
3456         } else {
3457                 return do_journal_end(th, sb, nblocks, 0);
3458         }
3459 }
3460
3461 /* removes from the current transaction, relsing and descrementing any counters.
3462 ** also files the removed buffer directly onto the clean list
3463 **
3464 ** called by journal_mark_freed when a block has been deleted
3465 **
3466 ** returns 1 if it cleaned and relsed the buffer. 0 otherwise
3467 */
3468 static int remove_from_transaction(struct super_block *sb,
3469                                    b_blocknr_t blocknr, int already_cleaned)
3470 {
3471         struct buffer_head *bh;
3472         struct reiserfs_journal_cnode *cn;
3473         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3474         int ret = 0;
3475
3476         cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr);
3477         if (!cn || !cn->bh) {
3478                 return ret;
3479         }
3480         bh = cn->bh;
3481         if (cn->prev) {
3482                 cn->prev->next = cn->next;
3483         }
3484         if (cn->next) {
3485                 cn->next->prev = cn->prev;
3486         }
3487         if (cn == journal->j_first) {
3488                 journal->j_first = cn->next;
3489         }
3490         if (cn == journal->j_last) {
3491                 journal->j_last = cn->prev;
3492         }
3493         if (bh)
3494                 remove_journal_hash(sb, journal->j_hash_table, NULL,
3495                                     bh->b_blocknr, 0);
3496         clear_buffer_journaled(bh);     /* don't log this one */
3497
3498         if (!already_cleaned) {
3499                 clear_buffer_journal_dirty(bh);
3500                 clear_buffer_dirty(bh);
3501                 clear_buffer_journal_test(bh);
3502                 put_bh(bh);
3503                 if (atomic_read(&(bh->b_count)) < 0) {
3504                         reiserfs_warning(sb, "journal-1752",
3505                                          "b_count < 0");
3506                 }
3507                 ret = 1;
3508         }
3509         journal->j_len--;
3510         journal->j_len_alloc--;
3511         free_cnode(sb, cn);
3512         return ret;
3513 }
3514
3515 /*
3516 ** for any cnode in a journal list, it can only be dirtied of all the
3517 ** transactions that include it are committed to disk.
3518 ** this checks through each transaction, and returns 1 if you are allowed to dirty,
3519 ** and 0 if you aren't
3520 **
3521 ** it is called by dirty_journal_list, which is called after flush_commit_list has gotten all the log
3522 ** blocks for a given transaction on disk
3523 **
3524 */
3525 static int can_dirty(struct reiserfs_journal_cnode *cn)
3526 {
3527         struct super_block *sb = cn->sb;
3528         b_blocknr_t blocknr = cn->blocknr;
3529         struct reiserfs_journal_cnode *cur = cn->hprev;
3530         int can_dirty = 1;
3531
3532         /* first test hprev.  These are all newer than cn, so any node here
3533          ** with the same block number and dev means this node can't be sent
3534          ** to disk right now.
3535          */
3536         while (cur && can_dirty) {
3537                 if (cur->jlist && cur->bh && cur->blocknr && cur->sb == sb &&
3538                     cur->blocknr == blocknr) {
3539                         can_dirty = 0;
3540                 }
3541                 cur = cur->hprev;
3542         }
3543         /* then test hnext.  These are all older than cn.  As long as they
3544          ** are committed to the log, it is safe to write cn to disk
3545          */
3546         cur = cn->hnext;
3547         while (cur && can_dirty) {
3548                 if (cur->jlist && cur->jlist->j_len > 0 &&
3549                     atomic_read(&(cur->jlist->j_commit_left)) > 0 && cur->bh &&
3550                     cur->blocknr && cur->sb == sb && cur->blocknr == blocknr) {
3551                         can_dirty = 0;
3552                 }
3553                 cur = cur->hnext;
3554         }
3555         return can_dirty;
3556 }
3557
3558 /* syncs the commit blocks, but does not force the real buffers to disk
3559 ** will wait until the current transaction is done/committed before returning
3560 */
3561 int journal_end_sync(struct reiserfs_transaction_handle *th,
3562                      struct super_block *sb, unsigned long nblocks)
3563 {
3564         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3565
3566         BUG_ON(!th->t_trans_id);
3567         /* you can sync while nested, very, very bad */
3568         BUG_ON(th->t_refcount > 1);
3569         if (journal->j_len == 0) {
3570                 reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3571                                              1);
3572                 journal_mark_dirty(th, sb, SB_BUFFER_WITH_SB(sb));
3573         }
3574         return do_journal_end(th, sb, nblocks, COMMIT_NOW | WAIT);
3575 }
3576
3577 /*
3578 ** writeback the pending async commits to disk
3579 */
3580 static void flush_async_commits(struct work_struct *work)
3581 {
3582         struct reiserfs_journal *journal =
3583                 container_of(work, struct reiserfs_journal, j_work.work);
3584         struct super_block *sb = journal->j_work_sb;
3585         struct reiserfs_journal_list *jl;
3586         struct list_head *entry;
3587
3588         reiserfs_write_lock(sb);
3589         if (!list_empty(&journal->j_journal_list)) {
3590                 /* last entry is the youngest, commit it and you get everything */
3591                 entry = journal->j_journal_list.prev;
3592                 jl = JOURNAL_LIST_ENTRY(entry);
3593                 flush_commit_list(sb, jl, 1);
3594         }
3595         reiserfs_write_unlock(sb);
3596 }
3597
3598 /*
3599 ** flushes any old transactions to disk
3600 ** ends the current transaction if it is too old
3601 */
3602 int reiserfs_flush_old_commits(struct super_block *sb)
3603 {
3604         time_t now;
3605         struct reiserfs_transaction_handle th;
3606         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3607
3608         now = get_seconds();
3609         /* safety check so we don't flush while we are replaying the log during
3610          * mount
3611          */
3612         if (list_empty(&journal->j_journal_list)) {
3613                 return 0;
3614         }
3615
3616         /* check the current transaction.  If there are no writers, and it is
3617          * too old, finish it, and force the commit blocks to disk
3618          */
3619         if (atomic_read(&journal->j_wcount) <= 0 &&
3620             journal->j_trans_start_time > 0 &&
3621             journal->j_len > 0 &&
3622             (now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3623                 if (!journal_join(&th, sb, 1)) {
3624                         reiserfs_prepare_for_journal(sb,
3625                                                      SB_BUFFER_WITH_SB(sb),
3626                                                      1);
3627                         journal_mark_dirty(&th, sb,
3628                                            SB_BUFFER_WITH_SB(sb));
3629
3630                         /* we're only being called from kreiserfsd, it makes no sense to do
3631                          ** an async commit so that kreiserfsd can do it later
3632                          */
3633                         do_journal_end(&th, sb, 1, COMMIT_NOW | WAIT);
3634                 }
3635         }
3636         return sb->s_dirt;
3637 }
3638
3639 /*
3640 ** returns 0 if do_journal_end should return right away, returns 1 if do_journal_end should finish the commit
3641 **
3642 ** if the current transaction is too old, but still has writers, this will wait on j_join_wait until all
3643 ** the writers are done.  By the time it wakes up, the transaction it was called has already ended, so it just
3644 ** flushes the commit list and returns 0.
3645 **
3646 ** Won't batch when flush or commit_now is set.  Also won't batch when others are waiting on j_join_wait.
3647 **
3648 ** Note, we can't allow the journal_end to proceed while there are still writers in the log.
3649 */
3650 static int check_journal_end(struct reiserfs_transaction_handle *th,
3651                              struct super_block *sb, unsigned long nblocks,
3652                              int flags)
3653 {
3654
3655         time_t now;
3656         int flush = flags & FLUSH_ALL;
3657         int commit_now = flags & COMMIT_NOW;
3658         int wait_on_commit = flags & WAIT;
3659         struct reiserfs_journal_list *jl;
3660         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3661
3662         BUG_ON(!th->t_trans_id);
3663
3664         if (th->t_trans_id != journal->j_trans_id) {
3665                 reiserfs_panic(th->t_super, "journal-1577",
3666                                "handle trans id %ld != current trans id %ld",
3667                                th->t_trans_id, journal->j_trans_id);
3668         }
3669
3670         journal->j_len_alloc -= (th->t_blocks_allocated - th->t_blocks_logged);
3671         if (atomic_read(&(journal->j_wcount)) > 0) {    /* <= 0 is allowed.  unmounting might not call begin */
3672                 atomic_dec(&(journal->j_wcount));
3673         }
3674
3675         /* BUG, deal with case where j_len is 0, but people previously freed blocks need to be released
3676          ** will be dealt with by next transaction that actually writes something, but should be taken
3677          ** care of in this trans
3678          */
3679         BUG_ON(journal->j_len == 0);
3680
3681         /* if wcount > 0, and we are called to with flush or commit_now,
3682          ** we wait on j_join_wait.  We will wake up when the last writer has
3683          ** finished the transaction, and started it on its way to the disk.
3684          ** Then, we flush the commit or journal list, and just return 0
3685          ** because the rest of journal end was already done for this transaction.
3686          */
3687         if (atomic_read(&(journal->j_wcount)) > 0) {
3688                 if (flush || commit_now) {
3689                         unsigned trans_id;
3690
3691                         jl = journal->j_current_jl;
3692                         trans_id = jl->j_trans_id;
3693                         if (wait_on_commit)
3694                                 jl->j_state |= LIST_COMMIT_PENDING;
3695                         atomic_set(&(journal->j_jlock), 1);
3696                         if (flush) {
3697                                 journal->j_next_full_flush = 1;
3698                         }
3699                         unlock_journal(sb);
3700
3701                         /* sleep while the current transaction is still j_jlocked */
3702                         while (journal->j_trans_id == trans_id) {
3703                                 if (atomic_read(&journal->j_jlock)) {
3704                                         queue_log_writer(sb);
3705                                 } else {
3706                                         lock_journal(sb);
3707                                         if (journal->j_trans_id == trans_id) {
3708                                                 atomic_set(&(journal->j_jlock),
3709                                                            1);
3710                                         }
3711                                         unlock_journal(sb);
3712                                 }
3713                         }
3714                         BUG_ON(journal->j_trans_id == trans_id);
3715
3716                         if (commit_now
3717                             && journal_list_still_alive(sb, trans_id)
3718                             && wait_on_commit) {
3719                                 flush_commit_list(sb, jl, 1);
3720                         }
3721                         return 0;
3722                 }
3723                 unlock_journal(sb);
3724                 return 0;
3725         }
3726
3727         /* deal with old transactions where we are the last writers */
3728         now = get_seconds();
3729         if ((now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3730                 commit_now = 1;
3731                 journal->j_next_async_flush = 1;
3732         }
3733         /* don't batch when someone is waiting on j_join_wait */
3734         /* don't batch when syncing the commit or flushing the whole trans */
3735         if (!(journal->j_must_wait > 0) && !(atomic_read(&(journal->j_jlock)))
3736             && !flush && !commit_now && (journal->j_len < journal->j_max_batch)
3737             && journal->j_len_alloc < journal->j_max_batch
3738             && journal->j_cnode_free > (journal->j_trans_max * 3)) {
3739                 journal->j_bcount++;
3740                 unlock_journal(sb);
3741                 return 0;
3742         }
3743
3744         if (journal->j_start > SB_ONDISK_JOURNAL_SIZE(sb)) {
3745                 reiserfs_panic(sb, "journal-003",
3746                                "j_start (%ld) is too high",
3747                                journal->j_start);
3748         }
3749         return 1;
3750 }
3751
3752 /*
3753 ** Does all the work that makes deleting blocks safe.
3754 ** when deleting a block mark BH_JNew, just remove it from the current transaction, clean it's buffer_head and move on.
3755 **
3756 ** otherwise:
3757 ** set a bit for the block in the journal bitmap.  That will prevent it from being allocated for unformatted nodes
3758 ** before this transaction has finished.
3759 **
3760 ** mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers.  That will prevent any old transactions with
3761 ** this block from trying to flush to the real location.  Since we aren't removing the cnode from the journal_list_hash,
3762 ** the block can't be reallocated yet.
3763 **
3764 ** Then remove it from the current transaction, decrementing any counters and filing it on the clean list.
3765 */
3766 int journal_mark_freed(struct reiserfs_transaction_handle *th,
3767                        struct super_block *sb, b_blocknr_t blocknr)
3768 {
3769         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3770         struct reiserfs_journal_cnode *cn = NULL;
3771         struct buffer_head *bh = NULL;
3772         struct reiserfs_list_bitmap *jb = NULL;
3773         int cleaned = 0;
3774         BUG_ON(!th->t_trans_id);
3775
3776         cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr);
3777         if (cn && cn->bh) {
3778                 bh = cn->bh;
3779                 get_bh(bh);
3780         }
3781         /* if it is journal new, we just remove it from this transaction */
3782         if (bh && buffer_journal_new(bh)) {
3783                 clear_buffer_journal_new(bh);
3784                 clear_prepared_bits(bh);
3785                 reiserfs_clean_and_file_buffer(bh);
3786                 cleaned = remove_from_transaction(sb, blocknr, cleaned);
3787         } else {
3788                 /* set the bit for this block in the journal bitmap for this transaction */
3789                 jb = journal->j_current_jl->j_list_bitmap;
3790                 if (!jb) {
3791                         reiserfs_panic(sb, "journal-1702",
3792                                        "journal_list_bitmap is NULL");
3793                 }
3794                 set_bit_in_list_bitmap(sb, blocknr, jb);
3795
3796                 /* Note, the entire while loop is not allowed to schedule.  */
3797
3798                 if (bh) {
3799                         clear_prepared_bits(bh);
3800                         reiserfs_clean_and_file_buffer(bh);
3801                 }
3802                 cleaned = remove_from_transaction(sb, blocknr, cleaned);
3803
3804                 /* find all older transactions with this block, make sure they don't try to write it out */
3805                 cn = get_journal_hash_dev(sb, journal->j_list_hash_table,
3806                                           blocknr);
3807                 while (cn) {
3808                         if (sb == cn->sb && blocknr == cn->blocknr) {
3809                                 set_bit(BLOCK_FREED, &cn->state);
3810                                 if (cn->bh) {
3811                                         if (!cleaned) {
3812                                                 /* remove_from_transaction will brelse the buffer if it was
3813                                                  ** in the current trans
3814                                                  */
3815                                                 clear_buffer_journal_dirty(cn->
3816                                                                            bh);
3817                                                 clear_buffer_dirty(cn->bh);
3818                                                 clear_buffer_journal_test(cn->
3819                                                                           bh);
3820                                                 cleaned = 1;
3821                                                 put_bh(cn->bh);
3822                                                 if (atomic_read
3823                                                     (&(cn->bh->b_count)) < 0) {
3824                                                         reiserfs_warning(sb,
3825                                                                  "journal-2138",
3826                                                                  "cn->bh->b_count < 0");
3827                                                 }
3828                                         }
3829                                         if (cn->jlist) {        /* since we are clearing the bh, we MUST dec nonzerolen */
3830                                                 atomic_dec(&
3831                                                            (cn->jlist->
3832                                                             j_nonzerolen));
3833                                         }
3834                                         cn->bh = NULL;
3835                                 }
3836                         }
3837                         cn = cn->hnext;
3838                 }
3839         }
3840
3841         if (bh)
3842                 release_buffer_page(bh); /* get_hash grabs the buffer */
3843         return 0;
3844 }
3845
3846 void reiserfs_update_inode_transaction(struct inode *inode)
3847 {
3848         struct reiserfs_journal *journal = SB_JOURNAL(inode->i_sb);
3849         REISERFS_I(inode)->i_jl = journal->j_current_jl;
3850         REISERFS_I(inode)->i_trans_id = journal->j_trans_id;
3851 }
3852
3853 /*
3854  * returns -1 on error, 0 if no commits/barriers were done and 1
3855  * if a transaction was actually committed and the barrier was done
3856  */
3857 static int __commit_trans_jl(struct inode *inode, unsigned long id,
3858                              struct reiserfs_journal_list *jl)
3859 {
3860         struct reiserfs_transaction_handle th;
3861         struct super_block *sb = inode->i_sb;
3862         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3863         int ret = 0;
3864
3865         /* is it from the current transaction, or from an unknown transaction? */
3866         if (id == journal->j_trans_id) {
3867                 jl = journal->j_current_jl;
3868                 /* try to let other writers come in and grow this transaction */
3869                 let_transaction_grow(sb, id);
3870                 if (journal->j_trans_id != id) {
3871                         goto flush_commit_only;
3872                 }
3873
3874                 ret = journal_begin(&th, sb, 1);
3875                 if (ret)
3876                         return ret;
3877
3878                 /* someone might have ended this transaction while we joined */
3879                 if (journal->j_trans_id != id) {
3880                         reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3881                                                      1);
3882                         journal_mark_dirty(&th, sb, SB_BUFFER_WITH_SB(sb));
3883                         ret = journal_end(&th, sb, 1);
3884                         goto flush_commit_only;
3885                 }
3886
3887                 ret = journal_end_sync(&th, sb, 1);
3888                 if (!ret)
3889                         ret = 1;
3890
3891         } else {
3892                 /* this gets tricky, we have to make sure the journal list in
3893                  * the inode still exists.  We know the list is still around
3894                  * if we've got a larger transaction id than the oldest list
3895                  */
3896               flush_commit_only:
3897                 if (journal_list_still_alive(inode->i_sb, id)) {
3898                         /*
3899                          * we only set ret to 1 when we know for sure
3900                          * the barrier hasn't been started yet on the commit
3901                          * block.
3902                          */
3903                         if (atomic_read(&jl->j_commit_left) > 1)
3904                                 ret = 1;
3905                         flush_commit_list(sb, jl, 1);
3906                         if (journal->j_errno)
3907                                 ret = journal->j_errno;
3908                 }
3909         }
3910         /* otherwise the list is gone, and long since committed */
3911         return ret;
3912 }
3913
3914 int reiserfs_commit_for_inode(struct inode *inode)
3915 {
3916         unsigned int id = REISERFS_I(inode)->i_trans_id;
3917         struct reiserfs_journal_list *jl = REISERFS_I(inode)->i_jl;
3918
3919         /* for the whole inode, assume unset id means it was
3920          * changed in the current transaction.  More conservative
3921          */
3922         if (!id || !jl) {
3923                 reiserfs_update_inode_transaction(inode);
3924                 id = REISERFS_I(inode)->i_trans_id;
3925                 /* jl will be updated in __commit_trans_jl */
3926         }
3927
3928         return __commit_trans_jl(inode, id, jl);
3929 }
3930
3931 void reiserfs_restore_prepared_buffer(struct super_block *sb,
3932                                       struct buffer_head *bh)
3933 {
3934         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3935         PROC_INFO_INC(sb, journal.restore_prepared);
3936         if (!bh) {
3937                 return;
3938         }
3939         if (test_clear_buffer_journal_restore_dirty(bh) &&
3940             buffer_journal_dirty(bh)) {
3941                 struct reiserfs_journal_cnode *cn;
3942                 cn = get_journal_hash_dev(sb,
3943                                           journal->j_list_hash_table,
3944                                           bh->b_blocknr);
3945                 if (cn && can_dirty(cn)) {
3946                         set_buffer_journal_test(bh);
3947                         mark_buffer_dirty(bh);
3948                 }
3949         }
3950         clear_buffer_journal_prepared(bh);
3951 }
3952
3953 extern struct tree_balance *cur_tb;
3954 /*
3955 ** before we can change a metadata block, we have to make sure it won't
3956 ** be written to disk while we are altering it.  So, we must:
3957 ** clean it
3958 ** wait on it.
3959 **
3960 */
3961 int reiserfs_prepare_for_journal(struct super_block *sb,
3962                                  struct buffer_head *bh, int wait)
3963 {
3964         PROC_INFO_INC(sb, journal.prepare);
3965
3966         if (!trylock_buffer(bh)) {
3967                 if (!wait)
3968                         return 0;
3969                 lock_buffer(bh);
3970         }
3971         set_buffer_journal_prepared(bh);
3972         if (test_clear_buffer_dirty(bh) && buffer_journal_dirty(bh)) {
3973                 clear_buffer_journal_test(bh);
3974                 set_buffer_journal_restore_dirty(bh);
3975         }
3976         unlock_buffer(bh);
3977         return 1;
3978 }
3979
3980 static void flush_old_journal_lists(struct super_block *s)
3981 {
3982         struct reiserfs_journal *journal = SB_JOURNAL(s);
3983         struct reiserfs_journal_list *jl;
3984         struct list_head *entry;
3985         time_t now = get_seconds();
3986
3987         while (!list_empty(&journal->j_journal_list)) {
3988                 entry = journal->j_journal_list.next;
3989                 jl = JOURNAL_LIST_ENTRY(entry);
3990                 /* this check should always be run, to send old lists to disk */
3991                 if (jl->j_timestamp < (now - (JOURNAL_MAX_TRANS_AGE * 4)) &&
3992                     atomic_read(&jl->j_commit_left) == 0 &&
3993                     test_transaction(s, jl)) {
3994                         flush_used_journal_lists(s, jl);
3995                 } else {
3996                         break;
3997                 }
3998         }
3999 }
4000
4001 /*
4002 ** long and ugly.  If flush, will not return until all commit
4003 ** blocks and all real buffers in the trans are on disk.
4004 ** If no_async, won't return until all commit blocks are on disk.
4005 **
4006 ** keep reading, there are comments as you go along
4007 **
4008 ** If the journal is aborted, we just clean up. Things like flushing
4009 ** journal lists, etc just won't happen.
4010 */
4011 static int do_journal_end(struct reiserfs_transaction_handle *th,
4012                           struct super_block *sb, unsigned long nblocks,
4013                           int flags)
4014 {
4015         struct reiserfs_journal *journal = SB_JOURNAL(sb);
4016         struct reiserfs_journal_cnode *cn, *next, *jl_cn;
4017         struct reiserfs_journal_cnode *last_cn = NULL;
4018         struct reiserfs_journal_desc *desc;
4019         struct reiserfs_journal_commit *commit;
4020         struct buffer_head *c_bh;       /* commit bh */
4021         struct buffer_head *d_bh;       /* desc bh */
4022         int cur_write_start = 0;        /* start index of current log write */
4023         int old_start;
4024         int i;
4025         int flush;
4026         int wait_on_commit;
4027         struct reiserfs_journal_list *jl, *temp_jl;
4028         struct list_head *entry, *safe;
4029         unsigned long jindex;
4030         unsigned int commit_trans_id;
4031         int trans_half;
4032
4033         BUG_ON(th->t_refcount > 1);
4034         BUG_ON(!th->t_trans_id);
4035
4036         /* protect flush_older_commits from doing mistakes if the
4037            transaction ID counter gets overflowed.  */
4038         if (th->t_trans_id == ~0U)
4039                 flags |= FLUSH_ALL | COMMIT_NOW | WAIT;
4040         flush = flags & FLUSH_ALL;
4041         wait_on_commit = flags & WAIT;
4042
4043         put_fs_excl();
4044         current->journal_info = th->t_handle_save;
4045         reiserfs_check_lock_depth(sb, "journal end");
4046         if (journal->j_len == 0) {
4047                 reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
4048                                              1);
4049                 journal_mark_dirty(th, sb, SB_BUFFER_WITH_SB(sb));
4050         }
4051
4052         lock_journal(sb);
4053         if (journal->j_next_full_flush) {
4054                 flags |= FLUSH_ALL;
4055                 flush = 1;
4056         }
4057         if (journal->j_next_async_flush) {
4058                 flags |= COMMIT_NOW | WAIT;
4059                 wait_on_commit = 1;
4060         }
4061
4062         /* check_journal_end locks the journal, and unlocks if it does not return 1
4063          ** it tells us if we should continue with the journal_end, or just return
4064          */
4065         if (!check_journal_end(th, sb, nblocks, flags)) {
4066                 sb->s_dirt = 1;
4067                 wake_queued_writers(sb);
4068                 reiserfs_async_progress_wait(sb);
4069                 goto out;
4070         }
4071
4072         /* check_journal_end might set these, check again */
4073         if (journal->j_next_full_flush) {
4074                 flush = 1;
4075         }
4076
4077         /*
4078          ** j must wait means we have to flush the log blocks, and the real blocks for
4079          ** this transaction
4080          */
4081         if (journal->j_must_wait > 0) {
4082                 flush = 1;
4083         }
4084 #ifdef REISERFS_PREALLOCATE
4085         /* quota ops might need to nest, setup the journal_info pointer for them
4086          * and raise the refcount so that it is > 0. */
4087         current->journal_info = th;
4088         th->t_refcount++;
4089         reiserfs_discard_all_prealloc(th);      /* it should not involve new blocks into
4090                                                  * the transaction */
4091         th->t_refcount--;
4092         current->journal_info = th->t_handle_save;
4093 #endif
4094
4095         /* setup description block */
4096         d_bh =
4097             journal_getblk(sb,
4098                            SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4099                            journal->j_start);
4100         set_buffer_uptodate(d_bh);
4101         desc = (struct reiserfs_journal_desc *)(d_bh)->b_data;
4102         memset(d_bh->b_data, 0, d_bh->b_size);
4103         memcpy(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8);
4104         set_desc_trans_id(desc, journal->j_trans_id);
4105
4106         /* setup commit block.  Don't write (keep it clean too) this one until after everyone else is written */
4107         c_bh = journal_getblk(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4108                               ((journal->j_start + journal->j_len +
4109                                 1) % SB_ONDISK_JOURNAL_SIZE(sb)));
4110         commit = (struct reiserfs_journal_commit *)c_bh->b_data;
4111         memset(c_bh->b_data, 0, c_bh->b_size);
4112         set_commit_trans_id(commit, journal->j_trans_id);
4113         set_buffer_uptodate(c_bh);
4114
4115         /* init this journal list */
4116         jl = journal->j_current_jl;
4117
4118         /* we lock the commit before doing anything because
4119          * we want to make sure nobody tries to run flush_commit_list until
4120          * the new transaction is fully setup, and we've already flushed the
4121          * ordered bh list
4122          */
4123         reiserfs_mutex_lock_safe(&jl->j_commit_mutex, sb);
4124
4125         /* save the transaction id in case we need to commit it later */
4126         commit_trans_id = jl->j_trans_id;
4127
4128         atomic_set(&jl->j_older_commits_done, 0);
4129         jl->j_trans_id = journal->j_trans_id;
4130         jl->j_timestamp = journal->j_trans_start_time;
4131         jl->j_commit_bh = c_bh;
4132         jl->j_start = journal->j_start;
4133         jl->j_len = journal->j_len;
4134         atomic_set(&jl->j_nonzerolen, journal->j_len);
4135         atomic_set(&jl->j_commit_left, journal->j_len + 2);
4136         jl->j_realblock = NULL;
4137
4138         /* The ENTIRE FOR LOOP MUST not cause schedule to occur.
4139          **  for each real block, add it to the journal list hash,
4140          ** copy into real block index array in the commit or desc block
4141          */
4142         trans_half = journal_trans_half(sb->s_blocksize);
4143         for (i = 0, cn = journal->j_first; cn; cn = cn->next, i++) {
4144                 if (buffer_journaled(cn->bh)) {
4145                         jl_cn = get_cnode(sb);
4146                         if (!jl_cn) {
4147                                 reiserfs_panic(sb, "journal-1676",
4148                                                "get_cnode returned NULL");
4149                         }
4150                         if (i == 0) {
4151                                 jl->j_realblock = jl_cn;
4152                         }
4153                         jl_cn->prev = last_cn;
4154                         jl_cn->next = NULL;
4155                         if (last_cn) {
4156                                 last_cn->next = jl_cn;
4157                         }
4158                         last_cn = jl_cn;
4159                         /* make sure the block we are trying to log is not a block
4160                            of journal or reserved area */
4161
4162                         if (is_block_in_log_or_reserved_area
4163                             (sb, cn->bh->b_blocknr)) {
4164                                 reiserfs_panic(sb, "journal-2332",
4165                                                "Trying to log block %lu, "
4166                                                "which is a log block",
4167                                                cn->bh->b_blocknr);
4168                         }
4169                         jl_cn->blocknr = cn->bh->b_blocknr;
4170                         jl_cn->state = 0;
4171                         jl_cn->sb = sb;
4172                         jl_cn->bh = cn->bh;
4173                         jl_cn->jlist = jl;
4174                         insert_journal_hash(journal->j_list_hash_table, jl_cn);
4175                         if (i < trans_half) {
4176                                 desc->j_realblock[i] =
4177                                     cpu_to_le32(cn->bh->b_blocknr);
4178                         } else {
4179                                 commit->j_realblock[i - trans_half] =
4180                                     cpu_to_le32(cn->bh->b_blocknr);
4181                         }
4182                 } else {
4183                         i--;
4184                 }
4185         }
4186         set_desc_trans_len(desc, journal->j_len);
4187         set_desc_mount_id(desc, journal->j_mount_id);
4188         set_desc_trans_id(desc, journal->j_trans_id);
4189         set_commit_trans_len(commit, journal->j_len);
4190
4191         /* special check in case all buffers in the journal were marked for not logging */
4192         BUG_ON(journal->j_len == 0);
4193
4194         /* we're about to dirty all the log blocks, mark the description block
4195          * dirty now too.  Don't mark the commit block dirty until all the
4196          * others are on disk
4197          */
4198         mark_buffer_dirty(d_bh);
4199
4200         /* first data block is j_start + 1, so add one to cur_write_start wherever you use it */
4201         cur_write_start = journal->j_start;
4202         cn = journal->j_first;
4203         jindex = 1;             /* start at one so we don't get the desc again */
4204         while (cn) {
4205                 clear_buffer_journal_new(cn->bh);
4206                 /* copy all the real blocks into log area.  dirty log blocks */
4207                 if (buffer_journaled(cn->bh)) {
4208                         struct buffer_head *tmp_bh;
4209                         char *addr;
4210                         struct page *page;
4211                         tmp_bh =
4212                             journal_getblk(sb,
4213                                            SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4214                                            ((cur_write_start +
4215                                              jindex) %
4216                                             SB_ONDISK_JOURNAL_SIZE(sb)));
4217                         set_buffer_uptodate(tmp_bh);
4218                         page = cn->bh->b_page;
4219                         addr = kmap(page);
4220                         memcpy(tmp_bh->b_data,
4221                                addr + offset_in_page(cn->bh->b_data),
4222                                cn->bh->b_size);
4223                         kunmap(page);
4224                         mark_buffer_dirty(tmp_bh);
4225                         jindex++;
4226                         set_buffer_journal_dirty(cn->bh);
4227                         clear_buffer_journaled(cn->bh);
4228                 } else {
4229                         /* JDirty cleared sometime during transaction.  don't log this one */
4230                         reiserfs_warning(sb, "journal-2048",
4231                                          "BAD, buffer in journal hash, "
4232                                          "but not JDirty!");
4233                         brelse(cn->bh);
4234                 }
4235                 next = cn->next;
4236                 free_cnode(sb, cn);
4237                 cn = next;
4238                 reiserfs_write_unlock(sb);
4239                 cond_resched();
4240                 reiserfs_write_lock(sb);
4241         }
4242
4243         /* we are done  with both the c_bh and d_bh, but
4244          ** c_bh must be written after all other commit blocks,
4245          ** so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1.
4246          */
4247
4248         journal->j_current_jl = alloc_journal_list(sb);
4249
4250         /* now it is safe to insert this transaction on the main list */
4251         list_add_tail(&jl->j_list, &journal->j_journal_list);
4252         list_add_tail(&jl->j_working_list, &journal->j_working_list);
4253         journal->j_num_work_lists++;
4254
4255         /* reset journal values for the next transaction */
4256         old_start = journal->j_start;
4257         journal->j_start =
4258             (journal->j_start + journal->j_len +
4259              2) % SB_ONDISK_JOURNAL_SIZE(sb);
4260         atomic_set(&(journal->j_wcount), 0);
4261         journal->j_bcount = 0;
4262         journal->j_last = NULL;
4263         journal->j_first = NULL;
4264         journal->j_len = 0;
4265         journal->j_trans_start_time = 0;
4266         /* check for trans_id overflow */
4267         if (++journal->j_trans_id == 0)
4268                 journal->j_trans_id = 10;
4269         journal->j_current_jl->j_trans_id = journal->j_trans_id;
4270         journal->j_must_wait = 0;
4271         journal->j_len_alloc = 0;
4272         journal->j_next_full_flush = 0;
4273         journal->j_next_async_flush = 0;
4274         init_journal_hash(sb);
4275
4276         // make sure reiserfs_add_jh sees the new current_jl before we
4277         // write out the tails
4278         smp_mb();
4279
4280         /* tail conversion targets have to hit the disk before we end the
4281          * transaction.  Otherwise a later transaction might repack the tail
4282          * before this transaction commits, leaving the data block unflushed and
4283          * clean, if we crash before the later transaction commits, the data block
4284          * is lost.
4285          */
4286         if (!list_empty(&jl->j_tail_bh_list)) {
4287                 reiserfs_write_unlock(sb);
4288                 write_ordered_buffers(&journal->j_dirty_buffers_lock,
4289                                       journal, jl, &jl->j_tail_bh_list);
4290                 reiserfs_write_lock(sb);
4291         }
4292         BUG_ON(!list_empty(&jl->j_tail_bh_list));
4293         mutex_unlock(&jl->j_commit_mutex);
4294
4295         /* honor the flush wishes from the caller, simple commits can
4296          ** be done outside the journal lock, they are done below
4297          **
4298          ** if we don't flush the commit list right now, we put it into
4299          ** the work queue so the people waiting on the async progress work
4300          ** queue don't wait for this proc to flush journal lists and such.
4301          */
4302         if (flush) {
4303                 flush_commit_list(sb, jl, 1);
4304                 flush_journal_list(sb, jl, 1);
4305         } else if (!(jl->j_state & LIST_COMMIT_PENDING))
4306                 queue_delayed_work(commit_wq, &journal->j_work, HZ / 10);
4307
4308         /* if the next transaction has any chance of wrapping, flush
4309          ** transactions that might get overwritten.  If any journal lists are very
4310          ** old flush them as well.
4311          */
4312       first_jl:
4313         list_for_each_safe(entry, safe, &journal->j_journal_list) {
4314                 temp_jl = JOURNAL_LIST_ENTRY(entry);
4315                 if (journal->j_start <= temp_jl->j_start) {
4316                         if ((journal->j_start + journal->j_trans_max + 1) >=
4317                             temp_jl->j_start) {
4318                                 flush_used_journal_lists(sb, temp_jl);
4319                                 goto first_jl;
4320                         } else if ((journal->j_start +
4321                                     journal->j_trans_max + 1) <
4322                                    SB_ONDISK_JOURNAL_SIZE(sb)) {
4323                                 /* if we don't cross into the next transaction and we don't
4324                                  * wrap, there is no way we can overlap any later transactions
4325                                  * break now
4326                                  */
4327                                 break;
4328                         }
4329                 } else if ((journal->j_start +
4330                             journal->j_trans_max + 1) >
4331                            SB_ONDISK_JOURNAL_SIZE(sb)) {
4332                         if (((journal->j_start + journal->j_trans_max + 1) %
4333                              SB_ONDISK_JOURNAL_SIZE(sb)) >=
4334                             temp_jl->j_start) {
4335                                 flush_used_journal_lists(sb, temp_jl);
4336                                 goto first_jl;
4337                         } else {
4338                                 /* we don't overlap anything from out start to the end of the
4339                                  * log, and our wrapped portion doesn't overlap anything at
4340                                  * the start of the log.  We can break
4341                                  */
4342                                 break;
4343                         }
4344                 }
4345         }
4346         flush_old_journal_lists(sb);
4347
4348         journal->j_current_jl->j_list_bitmap =
4349             get_list_bitmap(sb, journal->j_current_jl);
4350
4351         if (!(journal->j_current_jl->j_list_bitmap)) {
4352                 reiserfs_panic(sb, "journal-1996",
4353                                "could not get a list bitmap");
4354         }
4355
4356         atomic_set(&(journal->j_jlock), 0);
4357         unlock_journal(sb);
4358         /* wake up any body waiting to join. */
4359         clear_bit(J_WRITERS_QUEUED, &journal->j_state);
4360         wake_up(&(journal->j_join_wait));
4361
4362         if (!flush && wait_on_commit &&
4363             journal_list_still_alive(sb, commit_trans_id)) {
4364                 flush_commit_list(sb, jl, 1);
4365         }
4366       out:
4367         reiserfs_check_lock_depth(sb, "journal end2");
4368
4369         memset(th, 0, sizeof(*th));
4370         /* Re-set th->t_super, so we can properly keep track of how many
4371          * persistent transactions there are. We need to do this so if this
4372          * call is part of a failed restart_transaction, we can free it later */
4373         th->t_super = sb;
4374
4375         return journal->j_errno;
4376 }
4377
4378 /* Send the file system read only and refuse new transactions */
4379 void reiserfs_abort_journal(struct super_block *sb, int errno)
4380 {
4381         struct reiserfs_journal *journal = SB_JOURNAL(sb);
4382         if (test_bit(J_ABORTED, &journal->j_state))
4383                 return;
4384
4385         if (!journal->j_errno)
4386                 journal->j_errno = errno;
4387
4388         sb->s_flags |= MS_RDONLY;
4389         set_bit(J_ABORTED, &journal->j_state);
4390
4391 #ifdef CONFIG_REISERFS_CHECK
4392         dump_stack();
4393 #endif
4394 }
4395