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