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 <asm/uaccess.h>
  38 #include <asm/system.h>
  39
  40 #include <linux/time.h>
  41 #include <asm/semaphore.h>
  42
  43 #include <linux/vmalloc.h>
  44 #include <linux/reiserfs_fs.h>
  45
  46 #include <linux/kernel.h>
  47 #include <linux/errno.h>
  48 #include <linux/fcntl.h>
  49 #include <linux/stat.h>
  50 #include <linux/string.h>
  51 #include <linux/smp_lock.h>
  52 #include <linux/buffer_head.h>
  53 #include <linux/workqueue.h>
  54 #include <linux/writeback.h>
  55 #include <linux/blkdev.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 *p_s_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(void *p);
 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 *p_s_sb,
 118                               unsigned long nblocks, int join);
 119
 120 static void init_journal_hash(struct super_block *p_s_sb)
 121 {
 122         struct reiserfs_journal *journal = SB_JOURNAL(p_s_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 void disable_barrier(struct super_block *s)
 142 {
 143         REISERFS_SB(s)->s_mount_opt &= ~(1 << REISERFS_BARRIER_FLUSH);
 144         printk("reiserfs: disabling flush barriers on %s\n",
 145                reiserfs_bdevname(s));
 146 }
 147
 148 static struct reiserfs_bitmap_node *allocate_bitmap_node(struct super_block
 149                                                          *p_s_sb)
 150 {
 151         struct reiserfs_bitmap_node *bn;
 152         static int id;
 153
 154         bn = kmalloc(sizeof(struct reiserfs_bitmap_node), GFP_NOFS);
 155         if (!bn) {
 156                 return NULL;
 157         }
 158         bn->data = kzalloc(p_s_sb->s_blocksize, GFP_NOFS);
 159         if (!bn->data) {
 160                 kfree(bn);
 161                 return NULL;
 162         }
 163         bn->id = id++;
 164         INIT_LIST_HEAD(&bn->list);
 165         return bn;
 166 }
 167
 168 static struct reiserfs_bitmap_node *get_bitmap_node(struct super_block *p_s_sb)
 169 {
 170         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 171         struct reiserfs_bitmap_node *bn = NULL;
 172         struct list_head *entry = journal->j_bitmap_nodes.next;
 173
 174         journal->j_used_bitmap_nodes++;
 175       repeat:
 176
 177         if (entry != &journal->j_bitmap_nodes) {
 178                 bn = list_entry(entry, struct reiserfs_bitmap_node, list);
 179                 list_del(entry);
 180                 memset(bn->data, 0, p_s_sb->s_blocksize);
 181                 journal->j_free_bitmap_nodes--;
 182                 return bn;
 183         }
 184         bn = allocate_bitmap_node(p_s_sb);
 185         if (!bn) {
 186                 yield();
 187                 goto repeat;
 188         }
 189         return bn;
 190 }
 191 static inline void free_bitmap_node(struct super_block *p_s_sb,
 192                                     struct reiserfs_bitmap_node *bn)
 193 {
 194         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 195         journal->j_used_bitmap_nodes--;
 196         if (journal->j_free_bitmap_nodes > REISERFS_MAX_BITMAP_NODES) {
 197                 kfree(bn->data);
 198                 kfree(bn);
 199         } else {
 200                 list_add(&bn->list, &journal->j_bitmap_nodes);
 201                 journal->j_free_bitmap_nodes++;
 202         }
 203 }
 204
 205 static void allocate_bitmap_nodes(struct super_block *p_s_sb)
 206 {
 207         int i;
 208         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 209         struct reiserfs_bitmap_node *bn = NULL;
 210         for (i = 0; i < REISERFS_MIN_BITMAP_NODES; i++) {
 211                 bn = allocate_bitmap_node(p_s_sb);
 212                 if (bn) {
 213                         list_add(&bn->list, &journal->j_bitmap_nodes);
 214                         journal->j_free_bitmap_nodes++;
 215                 } else {
 216                         break;  // this is ok, we'll try again when more are needed
 217                 }
 218         }
 219 }
 220
 221 static int set_bit_in_list_bitmap(struct super_block *p_s_sb, int block,
 222                                   struct reiserfs_list_bitmap *jb)
 223 {
 224         int bmap_nr = block / (p_s_sb->s_blocksize << 3);
 225         int bit_nr = block % (p_s_sb->s_blocksize << 3);
 226
 227         if (!jb->bitmaps[bmap_nr]) {
 228                 jb->bitmaps[bmap_nr] = get_bitmap_node(p_s_sb);
 229         }
 230         set_bit(bit_nr, (unsigned long *)jb->bitmaps[bmap_nr]->data);
 231         return 0;
 232 }
 233
 234 static void cleanup_bitmap_list(struct super_block *p_s_sb,
 235                                 struct reiserfs_list_bitmap *jb)
 236 {
 237         int i;
 238         if (jb->bitmaps == NULL)
 239                 return;
 240
 241         for (i = 0; i < SB_BMAP_NR(p_s_sb); i++) {
 242                 if (jb->bitmaps[i]) {
 243                         free_bitmap_node(p_s_sb, jb->bitmaps[i]);
 244                         jb->bitmaps[i] = NULL;
 245                 }
 246         }
 247 }
 248
 249 /*
 250 ** only call this on FS unmount.
 251 */
 252 static int free_list_bitmaps(struct super_block *p_s_sb,
 253                              struct reiserfs_list_bitmap *jb_array)
 254 {
 255         int i;
 256         struct reiserfs_list_bitmap *jb;
 257         for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
 258                 jb = jb_array + i;
 259                 jb->journal_list = NULL;
 260                 cleanup_bitmap_list(p_s_sb, jb);
 261                 vfree(jb->bitmaps);
 262                 jb->bitmaps = NULL;
 263         }
 264         return 0;
 265 }
 266
 267 static int free_bitmap_nodes(struct super_block *p_s_sb)
 268 {
 269         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 270         struct list_head *next = journal->j_bitmap_nodes.next;
 271         struct reiserfs_bitmap_node *bn;
 272
 273         while (next != &journal->j_bitmap_nodes) {
 274                 bn = list_entry(next, struct reiserfs_bitmap_node, list);
 275                 list_del(next);
 276                 kfree(bn->data);
 277                 kfree(bn);
 278                 next = journal->j_bitmap_nodes.next;
 279                 journal->j_free_bitmap_nodes--;
 280         }
 281
 282         return 0;
 283 }
 284
 285 /*
 286 ** get memory for JOURNAL_NUM_BITMAPS worth of bitmaps.
 287 ** jb_array is the array to be filled in.
 288 */
 289 int reiserfs_allocate_list_bitmaps(struct super_block *p_s_sb,
 290                                    struct reiserfs_list_bitmap *jb_array,
 291                                    int bmap_nr)
 292 {
 293         int i;
 294         int failed = 0;
 295         struct reiserfs_list_bitmap *jb;
 296         int mem = bmap_nr * sizeof(struct reiserfs_bitmap_node *);
 297
 298         for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
 299                 jb = jb_array + i;
 300                 jb->journal_list = NULL;
 301                 jb->bitmaps = vmalloc(mem);
 302                 if (!jb->bitmaps) {
 303                         reiserfs_warning(p_s_sb,
 304                                          "clm-2000, unable to allocate bitmaps for journal lists");
 305                         failed = 1;
 306                         break;
 307                 }
 308                 memset(jb->bitmaps, 0, mem);
 309         }
 310         if (failed) {
 311                 free_list_bitmaps(p_s_sb, jb_array);
 312                 return -1;
 313         }
 314         return 0;
 315 }
 316
 317 /*
 318 ** find an available list bitmap.  If you can't find one, flush a commit list
 319 ** and try again
 320 */
 321 static struct reiserfs_list_bitmap *get_list_bitmap(struct super_block *p_s_sb,
 322                                                     struct reiserfs_journal_list
 323                                                     *jl)
 324 {
 325         int i, j;
 326         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 327         struct reiserfs_list_bitmap *jb = NULL;
 328
 329         for (j = 0; j < (JOURNAL_NUM_BITMAPS * 3); j++) {
 330                 i = journal->j_list_bitmap_index;
 331                 journal->j_list_bitmap_index = (i + 1) % JOURNAL_NUM_BITMAPS;
 332                 jb = journal->j_list_bitmap + i;
 333                 if (journal->j_list_bitmap[i].journal_list) {
 334                         flush_commit_list(p_s_sb,
 335                                           journal->j_list_bitmap[i].
 336                                           journal_list, 1);
 337                         if (!journal->j_list_bitmap[i].journal_list) {
 338                                 break;
 339                         }
 340                 } else {
 341                         break;
 342                 }
 343         }
 344         if (jb->journal_list) { /* double check to make sure if flushed correctly */
 345                 return NULL;
 346         }
 347         jb->journal_list = jl;
 348         return jb;
 349 }
 350
 351 /*
 352 ** allocates a new chunk of X nodes, and links them all together as a list.
 353 ** Uses the cnode->next and cnode->prev pointers
 354 ** returns NULL on failure
 355 */
 356 static struct reiserfs_journal_cnode *allocate_cnodes(int num_cnodes)
 357 {
 358         struct reiserfs_journal_cnode *head;
 359         int i;
 360         if (num_cnodes <= 0) {
 361                 return NULL;
 362         }
 363         head = vmalloc(num_cnodes * sizeof(struct reiserfs_journal_cnode));
 364         if (!head) {
 365                 return NULL;
 366         }
 367         memset(head, 0, num_cnodes * sizeof(struct reiserfs_journal_cnode));
 368         head[0].prev = NULL;
 369         head[0].next = head + 1;
 370         for (i = 1; i < num_cnodes; i++) {
 371                 head[i].prev = head + (i - 1);
 372                 head[i].next = head + (i + 1);  /* if last one, overwrite it after the if */
 373         }
 374         head[num_cnodes - 1].next = NULL;
 375         return head;
 376 }
 377
 378 /*
 379 ** pulls a cnode off the free list, or returns NULL on failure
 380 */
 381 static struct reiserfs_journal_cnode *get_cnode(struct super_block *p_s_sb)
 382 {
 383         struct reiserfs_journal_cnode *cn;
 384         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 385
 386         reiserfs_check_lock_depth(p_s_sb, "get_cnode");
 387
 388         if (journal->j_cnode_free <= 0) {
 389                 return NULL;
 390         }
 391         journal->j_cnode_used++;
 392         journal->j_cnode_free--;
 393         cn = journal->j_cnode_free_list;
 394         if (!cn) {
 395                 return cn;
 396         }
 397         if (cn->next) {
 398                 cn->next->prev = NULL;
 399         }
 400         journal->j_cnode_free_list = cn->next;
 401         memset(cn, 0, sizeof(struct reiserfs_journal_cnode));
 402         return cn;
 403 }
 404
 405 /*
 406 ** returns a cnode to the free list
 407 */
 408 static void free_cnode(struct super_block *p_s_sb,
 409                        struct reiserfs_journal_cnode *cn)
 410 {
 411         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 412
 413         reiserfs_check_lock_depth(p_s_sb, "free_cnode");
 414
 415         journal->j_cnode_used--;
 416         journal->j_cnode_free++;
 417         /* memset(cn, 0, sizeof(struct reiserfs_journal_cnode)) ; */
 418         cn->next = journal->j_cnode_free_list;
 419         if (journal->j_cnode_free_list) {
 420                 journal->j_cnode_free_list->prev = cn;
 421         }
 422         cn->prev = NULL;        /* not needed with the memset, but I might kill the memset, and forget to do this */
 423         journal->j_cnode_free_list = cn;
 424 }
 425
 426 static void clear_prepared_bits(struct buffer_head *bh)
 427 {
 428         clear_buffer_journal_prepared(bh);
 429         clear_buffer_journal_restore_dirty(bh);
 430 }
 431
 432 /* utility function to force a BUG if it is called without the big
 433 ** kernel lock held.  caller is the string printed just before calling BUG()
 434 */
 435 void reiserfs_check_lock_depth(struct super_block *sb, char *caller)
 436 {
 437 #ifdef CONFIG_SMP
 438         if (current->lock_depth < 0) {
 439                 reiserfs_panic(sb, "%s called without kernel lock held",
 440                                caller);
 441         }
 442 #else
 443         ;
 444 #endif
 445 }
 446
 447 /* return a cnode with same dev, block number and size in table, or null if not found */
 448 static inline struct reiserfs_journal_cnode *get_journal_hash_dev(struct
 449                                                                   super_block
 450                                                                   *sb,
 451                                                                   struct
 452                                                                   reiserfs_journal_cnode
 453                                                                   **table,
 454                                                                   long bl)
 455 {
 456         struct reiserfs_journal_cnode *cn;
 457         cn = journal_hash(table, sb, bl);
 458         while (cn) {
 459                 if (cn->blocknr == bl && cn->sb == sb)
 460                         return cn;
 461                 cn = cn->hnext;
 462         }
 463         return (struct reiserfs_journal_cnode *)0;
 464 }
 465
 466 /*
 467 ** this actually means 'can this block be reallocated yet?'.  If you set search_all, a block can only be allocated
 468 ** if it is not in the current transaction, was not freed by the current transaction, and has no chance of ever
 469 ** being overwritten by a replay after crashing.
 470 **
 471 ** If you don't set search_all, a block can only be allocated if it is not in the current transaction.  Since deleting
 472 ** a block removes it from the current transaction, this case should never happen.  If you don't set search_all, make
 473 ** sure you never write the block without logging it.
 474 **
 475 ** next_zero_bit is a suggestion about the next block to try for find_forward.
 476 ** when bl is rejected because it is set in a journal list bitmap, we search
 477 ** for the next zero bit in the bitmap that rejected bl.  Then, we return that
 478 ** through next_zero_bit for find_forward to try.
 479 **
 480 ** Just because we return something in next_zero_bit does not mean we won't
 481 ** reject it on the next call to reiserfs_in_journal
 482 **
 483 */
 484 int reiserfs_in_journal(struct super_block *p_s_sb,
 485                         int bmap_nr, int bit_nr, int search_all,
 486                         b_blocknr_t * next_zero_bit)
 487 {
 488         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 489         struct reiserfs_journal_cnode *cn;
 490         struct reiserfs_list_bitmap *jb;
 491         int i;
 492         unsigned long bl;
 493
 494         *next_zero_bit = 0;     /* always start this at zero. */
 495
 496         PROC_INFO_INC(p_s_sb, journal.in_journal);
 497         /* If we aren't doing a search_all, this is a metablock, and it will be logged before use.
 498          ** if we crash before the transaction that freed it commits,  this transaction won't
 499          ** have committed either, and the block will never be written
 500          */
 501         if (search_all) {
 502                 for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
 503                         PROC_INFO_INC(p_s_sb, journal.in_journal_bitmap);
 504                         jb = journal->j_list_bitmap + i;
 505                         if (jb->journal_list && jb->bitmaps[bmap_nr] &&
 506                             test_bit(bit_nr,
 507                                      (unsigned long *)jb->bitmaps[bmap_nr]->
 508                                      data)) {
 509                                 *next_zero_bit =
 510                                     find_next_zero_bit((unsigned long *)
 511                                                        (jb->bitmaps[bmap_nr]->
 512                                                         data),
 513                                                        p_s_sb->s_blocksize << 3,
 514                                                        bit_nr + 1);
 515                                 return 1;
 516                         }
 517                 }
 518         }
 519
 520         bl = bmap_nr * (p_s_sb->s_blocksize << 3) + bit_nr;
 521         /* is it in any old transactions? */
 522         if (search_all
 523             && (cn =
 524                 get_journal_hash_dev(p_s_sb, journal->j_list_hash_table, bl))) {
 525                 return 1;
 526         }
 527
 528         /* is it in the current transaction.  This should never happen */
 529         if ((cn = get_journal_hash_dev(p_s_sb, journal->j_hash_table, bl))) {
 530                 BUG();
 531                 return 1;
 532         }
 533
 534         PROC_INFO_INC(p_s_sb, journal.in_journal_reusable);
 535         /* safe for reuse */
 536         return 0;
 537 }
 538
 539 /* insert cn into table
 540 */
 541 static inline void insert_journal_hash(struct reiserfs_journal_cnode **table,
 542                                        struct reiserfs_journal_cnode *cn)
 543 {
 544         struct reiserfs_journal_cnode *cn_orig;
 545
 546         cn_orig = journal_hash(table, cn->sb, cn->blocknr);
 547         cn->hnext = cn_orig;
 548         cn->hprev = NULL;
 549         if (cn_orig) {
 550                 cn_orig->hprev = cn;
 551         }
 552         journal_hash(table, cn->sb, cn->blocknr) = cn;
 553 }
 554
 555 /* lock the current transaction */
 556 static inline void lock_journal(struct super_block *p_s_sb)
 557 {
 558         PROC_INFO_INC(p_s_sb, journal.lock_journal);
 559         down(&SB_JOURNAL(p_s_sb)->j_lock);
 560 }
 561
 562 /* unlock the current transaction */
 563 static inline void unlock_journal(struct super_block *p_s_sb)
 564 {
 565         up(&SB_JOURNAL(p_s_sb)->j_lock);
 566 }
 567
 568 static inline void get_journal_list(struct reiserfs_journal_list *jl)
 569 {
 570         jl->j_refcount++;
 571 }
 572
 573 static inline void put_journal_list(struct super_block *s,
 574                                     struct reiserfs_journal_list *jl)
 575 {
 576         if (jl->j_refcount < 1) {
 577                 reiserfs_panic(s, "trans id %lu, refcount at %d",
 578                                jl->j_trans_id, jl->j_refcount);
 579         }
 580         if (--jl->j_refcount == 0)
 581                 kfree(jl);
 582 }
 583
 584 /*
 585 ** this used to be much more involved, and I'm keeping it just in case things get ugly again.
 586 ** it gets called by flush_commit_list, and cleans up any data stored about blocks freed during a
 587 ** transaction.
 588 */
 589 static void cleanup_freed_for_journal_list(struct super_block *p_s_sb,
 590                                            struct reiserfs_journal_list *jl)
 591 {
 592
 593         struct reiserfs_list_bitmap *jb = jl->j_list_bitmap;
 594         if (jb) {
 595                 cleanup_bitmap_list(p_s_sb, jb);
 596         }
 597         jl->j_list_bitmap->journal_list = NULL;
 598         jl->j_list_bitmap = NULL;
 599 }
 600
 601 static int journal_list_still_alive(struct super_block *s,
 602                                     unsigned long trans_id)
 603 {
 604         struct reiserfs_journal *journal = SB_JOURNAL(s);
 605         struct list_head *entry = &journal->j_journal_list;
 606         struct reiserfs_journal_list *jl;
 607
 608         if (!list_empty(entry)) {
 609                 jl = JOURNAL_LIST_ENTRY(entry->next);
 610                 if (jl->j_trans_id <= trans_id) {
 611                         return 1;
 612                 }
 613         }
 614         return 0;
 615 }
 616
 617 static void reiserfs_end_buffer_io_sync(struct buffer_head *bh, int uptodate)
 618 {
 619         char b[BDEVNAME_SIZE];
 620
 621         if (buffer_journaled(bh)) {
 622                 reiserfs_warning(NULL,
 623                                  "clm-2084: pinned buffer %lu:%s sent to disk",
 624                                  bh->b_blocknr, bdevname(bh->b_bdev, b));
 625         }
 626         if (uptodate)
 627                 set_buffer_uptodate(bh);
 628         else
 629                 clear_buffer_uptodate(bh);
 630         unlock_buffer(bh);
 631         put_bh(bh);
 632 }
 633
 634 static void reiserfs_end_ordered_io(struct buffer_head *bh, int uptodate)
 635 {
 636         if (uptodate)
 637                 set_buffer_uptodate(bh);
 638         else
 639                 clear_buffer_uptodate(bh);
 640         unlock_buffer(bh);
 641         put_bh(bh);
 642 }
 643
 644 static void submit_logged_buffer(struct buffer_head *bh)
 645 {
 646         get_bh(bh);
 647         bh->b_end_io = reiserfs_end_buffer_io_sync;
 648         clear_buffer_journal_new(bh);
 649         clear_buffer_dirty(bh);
 650         if (!test_clear_buffer_journal_test(bh))
 651                 BUG();
 652         if (!buffer_uptodate(bh))
 653                 BUG();
 654         submit_bh(WRITE, bh);
 655 }
 656
 657 static void submit_ordered_buffer(struct buffer_head *bh)
 658 {
 659         get_bh(bh);
 660         bh->b_end_io = reiserfs_end_ordered_io;
 661         clear_buffer_dirty(bh);
 662         if (!buffer_uptodate(bh))
 663                 BUG();
 664         submit_bh(WRITE, bh);
 665 }
 666
 667 static int submit_barrier_buffer(struct buffer_head *bh)
 668 {
 669         get_bh(bh);
 670         bh->b_end_io = reiserfs_end_ordered_io;
 671         clear_buffer_dirty(bh);
 672         if (!buffer_uptodate(bh))
 673                 BUG();
 674         return submit_bh(WRITE_BARRIER, bh);
 675 }
 676
 677 static void check_barrier_completion(struct super_block *s,
 678                                      struct buffer_head *bh)
 679 {
 680         if (buffer_eopnotsupp(bh)) {
 681                 clear_buffer_eopnotsupp(bh);
 682                 disable_barrier(s);
 683                 set_buffer_uptodate(bh);
 684                 set_buffer_dirty(bh);
 685                 sync_dirty_buffer(bh);
 686         }
 687 }
 688
 689 #define CHUNK_SIZE 32
 690 struct buffer_chunk {
 691         struct buffer_head *bh[CHUNK_SIZE];
 692         int nr;
 693 };
 694
 695 static void write_chunk(struct buffer_chunk *chunk)
 696 {
 697         int i;
 698         get_fs_excl();
 699         for (i = 0; i < chunk->nr; i++) {
 700                 submit_logged_buffer(chunk->bh[i]);
 701         }
 702         chunk->nr = 0;
 703         put_fs_excl();
 704 }
 705
 706 static void write_ordered_chunk(struct buffer_chunk *chunk)
 707 {
 708         int i;
 709         get_fs_excl();
 710         for (i = 0; i < chunk->nr; i++) {
 711                 submit_ordered_buffer(chunk->bh[i]);
 712         }
 713         chunk->nr = 0;
 714         put_fs_excl();
 715 }
 716
 717 static int add_to_chunk(struct buffer_chunk *chunk, struct buffer_head *bh,
 718                         spinlock_t * lock, void (fn) (struct buffer_chunk *))
 719 {
 720         int ret = 0;
 721         if (chunk->nr >= CHUNK_SIZE)
 722                 BUG();
 723         chunk->bh[chunk->nr++] = bh;
 724         if (chunk->nr >= CHUNK_SIZE) {
 725                 ret = 1;
 726                 if (lock)
 727                         spin_unlock(lock);
 728                 fn(chunk);
 729                 if (lock)
 730                         spin_lock(lock);
 731         }
 732         return ret;
 733 }
 734
 735 static atomic_t nr_reiserfs_jh = ATOMIC_INIT(0);
 736 static struct reiserfs_jh *alloc_jh(void)
 737 {
 738         struct reiserfs_jh *jh;
 739         while (1) {
 740                 jh = kmalloc(sizeof(*jh), GFP_NOFS);
 741                 if (jh) {
 742                         atomic_inc(&nr_reiserfs_jh);
 743                         return jh;
 744                 }
 745                 yield();
 746         }
 747 }
 748
 749 /*
 750  * we want to free the jh when the buffer has been written
 751  * and waited on
 752  */
 753 void reiserfs_free_jh(struct buffer_head *bh)
 754 {
 755         struct reiserfs_jh *jh;
 756
 757         jh = bh->b_private;
 758         if (jh) {
 759                 bh->b_private = NULL;
 760                 jh->bh = NULL;
 761                 list_del_init(&jh->list);
 762                 kfree(jh);
 763                 if (atomic_read(&nr_reiserfs_jh) <= 0)
 764                         BUG();
 765                 atomic_dec(&nr_reiserfs_jh);
 766                 put_bh(bh);
 767         }
 768 }
 769
 770 static inline int __add_jh(struct reiserfs_journal *j, struct buffer_head *bh,
 771                            int tail)
 772 {
 773         struct reiserfs_jh *jh;
 774
 775         if (bh->b_private) {
 776                 spin_lock(&j->j_dirty_buffers_lock);
 777                 if (!bh->b_private) {
 778                         spin_unlock(&j->j_dirty_buffers_lock);
 779                         goto no_jh;
 780                 }
 781                 jh = bh->b_private;
 782                 list_del_init(&jh->list);
 783         } else {
 784               no_jh:
 785                 get_bh(bh);
 786                 jh = alloc_jh();
 787                 spin_lock(&j->j_dirty_buffers_lock);
 788                 /* buffer must be locked for __add_jh, should be able to have
 789                  * two adds at the same time
 790                  */
 791                 if (bh->b_private)
 792                         BUG();
 793                 jh->bh = bh;
 794                 bh->b_private = jh;
 795         }
 796         jh->jl = j->j_current_jl;
 797         if (tail)
 798                 list_add_tail(&jh->list, &jh->jl->j_tail_bh_list);
 799         else {
 800                 list_add_tail(&jh->list, &jh->jl->j_bh_list);
 801         }
 802         spin_unlock(&j->j_dirty_buffers_lock);
 803         return 0;
 804 }
 805
 806 int reiserfs_add_tail_list(struct inode *inode, struct buffer_head *bh)
 807 {
 808         return __add_jh(SB_JOURNAL(inode->i_sb), bh, 1);
 809 }
 810 int reiserfs_add_ordered_list(struct inode *inode, struct buffer_head *bh)
 811 {
 812         return __add_jh(SB_JOURNAL(inode->i_sb), bh, 0);
 813 }
 814
 815 #define JH_ENTRY(l) list_entry((l), struct reiserfs_jh, list)
 816 static int write_ordered_buffers(spinlock_t * lock,
 817                                  struct reiserfs_journal *j,
 818                                  struct reiserfs_journal_list *jl,
 819                                  struct list_head *list)
 820 {
 821         struct buffer_head *bh;
 822         struct reiserfs_jh *jh;
 823         int ret = j->j_errno;
 824         struct buffer_chunk chunk;
 825         struct list_head tmp;
 826         INIT_LIST_HEAD(&tmp);
 827
 828         chunk.nr = 0;
 829         spin_lock(lock);
 830         while (!list_empty(list)) {
 831                 jh = JH_ENTRY(list->next);
 832                 bh = jh->bh;
 833                 get_bh(bh);
 834                 if (test_set_buffer_locked(bh)) {
 835                         if (!buffer_dirty(bh)) {
 836                                 list_move(&jh->list, &tmp);
 837                                 goto loop_next;
 838                         }
 839                         spin_unlock(lock);
 840                         if (chunk.nr)
 841                                 write_ordered_chunk(&chunk);
 842                         wait_on_buffer(bh);
 843                         cond_resched();
 844                         spin_lock(lock);
 845                         goto loop_next;
 846                 }
 847                 /* in theory, dirty non-uptodate buffers should never get here,
 848                  * but the upper layer io error paths still have a few quirks.
 849                  * Handle them here as gracefully as we can
 850                  */
 851                 if (!buffer_uptodate(bh) && buffer_dirty(bh)) {
 852                         clear_buffer_dirty(bh);
 853                         ret = -EIO;
 854                 }
 855                 if (buffer_dirty(bh)) {
 856                         list_move(&jh->list, &tmp);
 857                         add_to_chunk(&chunk, bh, lock, write_ordered_chunk);
 858                 } else {
 859                         reiserfs_free_jh(bh);
 860                         unlock_buffer(bh);
 861                 }
 862               loop_next:
 863                 put_bh(bh);
 864                 cond_resched_lock(lock);
 865         }
 866         if (chunk.nr) {
 867                 spin_unlock(lock);
 868                 write_ordered_chunk(&chunk);
 869                 spin_lock(lock);
 870         }
 871         while (!list_empty(&tmp)) {
 872                 jh = JH_ENTRY(tmp.prev);
 873                 bh = jh->bh;
 874                 get_bh(bh);
 875                 reiserfs_free_jh(bh);
 876
 877                 if (buffer_locked(bh)) {
 878                         spin_unlock(lock);
 879                         wait_on_buffer(bh);
 880                         spin_lock(lock);
 881                 }
 882                 if (!buffer_uptodate(bh)) {
 883                         ret = -EIO;
 884                 }
 885                 /* ugly interaction with invalidatepage here.
 886                  * reiserfs_invalidate_page will pin any buffer that has a valid
 887                  * journal head from an older transaction.  If someone else sets
 888                  * our buffer dirty after we write it in the first loop, and
 889                  * then someone truncates the page away, nobody will ever write
 890                  * the buffer. We're safe if we write the page one last time
 891                  * after freeing the journal header.
 892                  */
 893                 if (buffer_dirty(bh) && unlikely(bh->b_page->mapping == NULL)) {
 894                         spin_unlock(lock);
 895                         ll_rw_block(WRITE, 1, &bh);
 896                         spin_lock(lock);
 897                 }
 898                 put_bh(bh);
 899                 cond_resched_lock(lock);
 900         }
 901         spin_unlock(lock);
 902         return ret;
 903 }
 904
 905 static int flush_older_commits(struct super_block *s,
 906                                struct reiserfs_journal_list *jl)
 907 {
 908         struct reiserfs_journal *journal = SB_JOURNAL(s);
 909         struct reiserfs_journal_list *other_jl;
 910         struct reiserfs_journal_list *first_jl;
 911         struct list_head *entry;
 912         unsigned long trans_id = jl->j_trans_id;
 913         unsigned long other_trans_id;
 914         unsigned long first_trans_id;
 915
 916       find_first:
 917         /*
 918          * first we walk backwards to find the oldest uncommitted transation
 919          */
 920         first_jl = jl;
 921         entry = jl->j_list.prev;
 922         while (1) {
 923                 other_jl = JOURNAL_LIST_ENTRY(entry);
 924                 if (entry == &journal->j_journal_list ||
 925                     atomic_read(&other_jl->j_older_commits_done))
 926                         break;
 927
 928                 first_jl = other_jl;
 929                 entry = other_jl->j_list.prev;
 930         }
 931
 932         /* if we didn't find any older uncommitted transactions, return now */
 933         if (first_jl == jl) {
 934                 return 0;
 935         }
 936
 937         first_trans_id = first_jl->j_trans_id;
 938
 939         entry = &first_jl->j_list;
 940         while (1) {
 941                 other_jl = JOURNAL_LIST_ENTRY(entry);
 942                 other_trans_id = other_jl->j_trans_id;
 943
 944                 if (other_trans_id < trans_id) {
 945                         if (atomic_read(&other_jl->j_commit_left) != 0) {
 946                                 flush_commit_list(s, other_jl, 0);
 947
 948                                 /* list we were called with is gone, return */
 949                                 if (!journal_list_still_alive(s, trans_id))
 950                                         return 1;
 951
 952                                 /* the one we just flushed is gone, this means all
 953                                  * older lists are also gone, so first_jl is no longer
 954                                  * valid either.  Go back to the beginning.
 955                                  */
 956                                 if (!journal_list_still_alive
 957                                     (s, other_trans_id)) {
 958                                         goto find_first;
 959                                 }
 960                         }
 961                         entry = entry->next;
 962                         if (entry == &journal->j_journal_list)
 963                                 return 0;
 964                 } else {
 965                         return 0;
 966                 }
 967         }
 968         return 0;
 969 }
 970 int reiserfs_async_progress_wait(struct super_block *s)
 971 {
 972         DEFINE_WAIT(wait);
 973         struct reiserfs_journal *j = SB_JOURNAL(s);
 974         if (atomic_read(&j->j_async_throttle))
 975                 blk_congestion_wait(WRITE, HZ / 10);
 976         return 0;
 977 }
 978
 979 /*
 980 ** if this journal list still has commit blocks unflushed, send them to disk.
 981 **
 982 ** log areas must be flushed in order (transaction 2 can't commit before transaction 1)
 983 ** Before the commit block can by written, every other log block must be safely on disk
 984 **
 985 */
 986 static int flush_commit_list(struct super_block *s,
 987                              struct reiserfs_journal_list *jl, int flushall)
 988 {
 989         int i;
 990         int bn;
 991         struct buffer_head *tbh = NULL;
 992         unsigned long trans_id = jl->j_trans_id;
 993         struct reiserfs_journal *journal = SB_JOURNAL(s);
 994         int barrier = 0;
 995         int retval = 0;
 996         int write_len;
 997
 998         reiserfs_check_lock_depth(s, "flush_commit_list");
 999
1000         if (atomic_read(&jl->j_older_commits_done)) {
1001                 return 0;
1002         }
1003
1004         get_fs_excl();
1005
1006         /* before we can put our commit blocks on disk, we have to make sure everyone older than
1007          ** us is on disk too
1008          */
1009         BUG_ON(jl->j_len <= 0);
1010         BUG_ON(trans_id == journal->j_trans_id);
1011
1012         get_journal_list(jl);
1013         if (flushall) {
1014                 if (flush_older_commits(s, jl) == 1) {
1015                         /* list disappeared during flush_older_commits.  return */
1016                         goto put_jl;
1017                 }
1018         }
1019
1020         /* make sure nobody is trying to flush this one at the same time */
1021         down(&jl->j_commit_lock);
1022         if (!journal_list_still_alive(s, trans_id)) {
1023                 up(&jl->j_commit_lock);
1024                 goto put_jl;
1025         }
1026         BUG_ON(jl->j_trans_id == 0);
1027
1028         /* this commit is done, exit */
1029         if (atomic_read(&(jl->j_commit_left)) <= 0) {
1030                 if (flushall) {
1031                         atomic_set(&(jl->j_older_commits_done), 1);
1032                 }
1033                 up(&jl->j_commit_lock);
1034                 goto put_jl;
1035         }
1036
1037         if (!list_empty(&jl->j_bh_list)) {
1038                 int ret;
1039                 unlock_kernel();
1040                 ret = write_ordered_buffers(&journal->j_dirty_buffers_lock,
1041                                             journal, jl, &jl->j_bh_list);
1042                 if (ret < 0 && retval == 0)
1043                         retval = ret;
1044                 lock_kernel();
1045         }
1046         BUG_ON(!list_empty(&jl->j_bh_list));
1047         /*
1048          * for the description block and all the log blocks, submit any buffers
1049          * that haven't already reached the disk.  Try to write at least 256
1050          * log blocks. later on, we will only wait on blocks that correspond
1051          * to this transaction, but while we're unplugging we might as well
1052          * get a chunk of data on there.
1053          */
1054         atomic_inc(&journal->j_async_throttle);
1055         write_len = jl->j_len + 1;
1056         if (write_len < 256)
1057                 write_len = 256;
1058         for (i = 0 ; i < write_len ; i++) {
1059                 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + (jl->j_start + i) %
1060                     SB_ONDISK_JOURNAL_SIZE(s);
1061                 tbh = journal_find_get_block(s, bn);
1062                 if (tbh) {
1063                         if (buffer_dirty(tbh))
1064                             ll_rw_block(WRITE, 1, &tbh) ;
1065                         put_bh(tbh) ;
1066                 }
1067         }
1068         atomic_dec(&journal->j_async_throttle);
1069
1070         /* We're skipping the commit if there's an error */
1071         if (retval || reiserfs_is_journal_aborted(journal))
1072                 barrier = 0;
1073
1074         /* wait on everything written so far before writing the commit
1075          * if we are in barrier mode, send the commit down now
1076          */
1077         barrier = reiserfs_barrier_flush(s);
1078         if (barrier) {
1079                 int ret;
1080                 lock_buffer(jl->j_commit_bh);
1081                 ret = submit_barrier_buffer(jl->j_commit_bh);
1082                 if (ret == -EOPNOTSUPP) {
1083                         set_buffer_uptodate(jl->j_commit_bh);
1084                         disable_barrier(s);
1085                         barrier = 0;
1086                 }
1087         }
1088         for (i = 0; i < (jl->j_len + 1); i++) {
1089                 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) +
1090                     (jl->j_start + i) % SB_ONDISK_JOURNAL_SIZE(s);
1091                 tbh = journal_find_get_block(s, bn);
1092                 wait_on_buffer(tbh);
1093                 // since we're using ll_rw_blk above, it might have skipped over
1094                 // a locked buffer.  Double check here
1095                 //
1096                 if (buffer_dirty(tbh))  /* redundant, sync_dirty_buffer() checks */
1097                         sync_dirty_buffer(tbh);
1098                 if (unlikely(!buffer_uptodate(tbh))) {
1099 #ifdef CONFIG_REISERFS_CHECK
1100                         reiserfs_warning(s, "journal-601, buffer write failed");
1101 #endif
1102                         retval = -EIO;
1103                 }
1104                 put_bh(tbh);    /* once for journal_find_get_block */
1105                 put_bh(tbh);    /* once due to original getblk in do_journal_end */
1106                 atomic_dec(&(jl->j_commit_left));
1107         }
1108
1109         BUG_ON(atomic_read(&(jl->j_commit_left)) != 1);
1110
1111         if (!barrier) {
1112                 /* If there was a write error in the journal - we can't commit
1113                  * this transaction - it will be invalid and, if successful,
1114                  * will just end up propogating the write error out to
1115                  * the file system. */
1116                 if (likely(!retval && !reiserfs_is_journal_aborted (journal))) {
1117                         if (buffer_dirty(jl->j_commit_bh))
1118                                 BUG();
1119                         mark_buffer_dirty(jl->j_commit_bh) ;
1120                         sync_dirty_buffer(jl->j_commit_bh) ;
1121                 }
1122         } else
1123                 wait_on_buffer(jl->j_commit_bh);
1124
1125         check_barrier_completion(s, jl->j_commit_bh);
1126
1127         /* If there was a write error in the journal - we can't commit this
1128          * transaction - it will be invalid and, if successful, will just end
1129          * up propogating the write error out to the filesystem. */
1130         if (unlikely(!buffer_uptodate(jl->j_commit_bh))) {
1131 #ifdef CONFIG_REISERFS_CHECK
1132                 reiserfs_warning(s, "journal-615: buffer write failed");
1133 #endif
1134                 retval = -EIO;
1135         }
1136         bforget(jl->j_commit_bh);
1137         if (journal->j_last_commit_id != 0 &&
1138             (jl->j_trans_id - journal->j_last_commit_id) != 1) {
1139                 reiserfs_warning(s, "clm-2200: last commit %lu, current %lu",
1140                                  journal->j_last_commit_id, jl->j_trans_id);
1141         }
1142         journal->j_last_commit_id = jl->j_trans_id;
1143
1144         /* now, every commit block is on the disk.  It is safe to allow blocks freed during this transaction to be reallocated */
1145         cleanup_freed_for_journal_list(s, jl);
1146
1147         retval = retval ? retval : journal->j_errno;
1148
1149         /* mark the metadata dirty */
1150         if (!retval)
1151                 dirty_one_transaction(s, jl);
1152         atomic_dec(&(jl->j_commit_left));
1153
1154         if (flushall) {
1155                 atomic_set(&(jl->j_older_commits_done), 1);
1156         }
1157         up(&jl->j_commit_lock);
1158       put_jl:
1159         put_journal_list(s, jl);
1160
1161         if (retval)
1162                 reiserfs_abort(s, retval, "Journal write error in %s",
1163                                __FUNCTION__);
1164         put_fs_excl();
1165         return retval;
1166 }
1167
1168 /*
1169 ** flush_journal_list frequently needs to find a newer transaction for a given block.  This does that, or
1170 ** returns NULL if it can't find anything
1171 */
1172 static struct reiserfs_journal_list *find_newer_jl_for_cn(struct
1173                                                           reiserfs_journal_cnode
1174                                                           *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                         return cn->jlist;
1183                 }
1184                 cn = cn->hprev;
1185         }
1186         return NULL;
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 *p_s_sb,
1200                                          struct reiserfs_journal_list *jl,
1201                                          int debug)
1202 {
1203         struct reiserfs_journal *journal = SB_JOURNAL(p_s_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(p_s_sb,
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(p_s_sb, journal->j_list_hash_table,
1220                                             jl, cn->blocknr, 1);
1221                 }
1222                 last = cn;
1223                 cn = cn->next;
1224                 free_cnode(p_s_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 *p_s_sb,
1237                                         unsigned long offset,
1238                                         unsigned long trans_id)
1239 {
1240         struct reiserfs_journal_header *jh;
1241         struct reiserfs_journal *journal = SB_JOURNAL(p_s_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                         wait_on_buffer((journal->j_header_bh));
1249                         if (unlikely(!buffer_uptodate(journal->j_header_bh))) {
1250 #ifdef CONFIG_REISERFS_CHECK
1251                                 reiserfs_warning(p_s_sb,
1252                                                  "journal-699: buffer write failed");
1253 #endif
1254                                 return -EIO;
1255                         }
1256                 }
1257                 journal->j_last_flush_trans_id = trans_id;
1258                 journal->j_first_unflushed_offset = offset;
1259                 jh = (struct reiserfs_journal_header *)(journal->j_header_bh->
1260                                                         b_data);
1261                 jh->j_last_flush_trans_id = cpu_to_le32(trans_id);
1262                 jh->j_first_unflushed_offset = cpu_to_le32(offset);
1263                 jh->j_mount_id = cpu_to_le32(journal->j_mount_id);
1264
1265                 if (reiserfs_barrier_flush(p_s_sb)) {
1266                         int ret;
1267                         lock_buffer(journal->j_header_bh);
1268                         ret = submit_barrier_buffer(journal->j_header_bh);
1269                         if (ret == -EOPNOTSUPP) {
1270                                 set_buffer_uptodate(journal->j_header_bh);
1271                                 disable_barrier(p_s_sb);
1272                                 goto sync;
1273                         }
1274                         wait_on_buffer(journal->j_header_bh);
1275                         check_barrier_completion(p_s_sb, journal->j_header_bh);
1276                 } else {
1277                       sync:
1278                         set_buffer_dirty(journal->j_header_bh);
1279                         sync_dirty_buffer(journal->j_header_bh);
1280                 }
1281                 if (!buffer_uptodate(journal->j_header_bh)) {
1282                         reiserfs_warning(p_s_sb,
1283                                          "journal-837: IO error during journal replay");
1284                         return -EIO;
1285                 }
1286         }
1287         return 0;
1288 }
1289
1290 static int update_journal_header_block(struct super_block *p_s_sb,
1291                                        unsigned long offset,
1292                                        unsigned long trans_id)
1293 {
1294         return _update_journal_header_block(p_s_sb, offset, trans_id);
1295 }
1296
1297 /*
1298 ** flush any and all journal lists older than you are
1299 ** can only be called from flush_journal_list
1300 */
1301 static int flush_older_journal_lists(struct super_block *p_s_sb,
1302                                      struct reiserfs_journal_list *jl)
1303 {
1304         struct list_head *entry;
1305         struct reiserfs_journal_list *other_jl;
1306         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1307         unsigned long trans_id = jl->j_trans_id;
1308
1309         /* we know we are the only ones flushing things, no extra race
1310          * protection is required.
1311          */
1312       restart:
1313         entry = journal->j_journal_list.next;
1314         /* Did we wrap? */
1315         if (entry == &journal->j_journal_list)
1316                 return 0;
1317         other_jl = JOURNAL_LIST_ENTRY(entry);
1318         if (other_jl->j_trans_id < trans_id) {
1319                 BUG_ON(other_jl->j_refcount <= 0);
1320                 /* do not flush all */
1321                 flush_journal_list(p_s_sb, other_jl, 0);
1322
1323                 /* other_jl is now deleted from the list */
1324                 goto restart;
1325         }
1326         return 0;
1327 }
1328
1329 static void del_from_work_list(struct super_block *s,
1330                                struct reiserfs_journal_list *jl)
1331 {
1332         struct reiserfs_journal *journal = SB_JOURNAL(s);
1333         if (!list_empty(&jl->j_working_list)) {
1334                 list_del_init(&jl->j_working_list);
1335                 journal->j_num_work_lists--;
1336         }
1337 }
1338
1339 /* flush a journal list, both commit and real blocks
1340 **
1341 ** always set flushall to 1, unless you are calling from inside
1342 ** flush_journal_list
1343 **
1344 ** IMPORTANT.  This can only be called while there are no journal writers,
1345 ** and the journal is locked.  That means it can only be called from
1346 ** do_journal_end, or by journal_release
1347 */
1348 static int flush_journal_list(struct super_block *s,
1349                               struct reiserfs_journal_list *jl, int flushall)
1350 {
1351         struct reiserfs_journal_list *pjl;
1352         struct reiserfs_journal_cnode *cn, *last;
1353         int count;
1354         int was_jwait = 0;
1355         int was_dirty = 0;
1356         struct buffer_head *saved_bh;
1357         unsigned long j_len_saved = jl->j_len;
1358         struct reiserfs_journal *journal = SB_JOURNAL(s);
1359         int err = 0;
1360
1361         BUG_ON(j_len_saved <= 0);
1362
1363         if (atomic_read(&journal->j_wcount) != 0) {
1364                 reiserfs_warning(s,
1365                                  "clm-2048: flush_journal_list called with wcount %d",
1366                                  atomic_read(&journal->j_wcount));
1367         }
1368         BUG_ON(jl->j_trans_id == 0);
1369
1370         /* if flushall == 0, the lock is already held */
1371         if (flushall) {
1372                 down(&journal->j_flush_sem);
1373         } else if (!down_trylock(&journal->j_flush_sem)) {
1374                 BUG();
1375         }
1376
1377         count = 0;
1378         if (j_len_saved > journal->j_trans_max) {
1379                 reiserfs_panic(s,
1380                                "journal-715: flush_journal_list, length is %lu, trans id %lu\n",
1381                                j_len_saved, jl->j_trans_id);
1382                 return 0;
1383         }
1384
1385         get_fs_excl();
1386
1387         /* if all the work is already done, get out of here */
1388         if (atomic_read(&(jl->j_nonzerolen)) <= 0 &&
1389             atomic_read(&(jl->j_commit_left)) <= 0) {
1390                 goto flush_older_and_return;
1391         }
1392
1393         /* start by putting the commit list on disk.  This will also flush
1394          ** the commit lists of any olders transactions
1395          */
1396         flush_commit_list(s, jl, 1);
1397
1398         if (!(jl->j_state & LIST_DIRTY)
1399             && !reiserfs_is_journal_aborted(journal))
1400                 BUG();
1401
1402         /* are we done now? */
1403         if (atomic_read(&(jl->j_nonzerolen)) <= 0 &&
1404             atomic_read(&(jl->j_commit_left)) <= 0) {
1405                 goto flush_older_and_return;
1406         }
1407
1408         /* loop through each cnode, see if we need to write it,
1409          ** or wait on a more recent transaction, or just ignore it
1410          */
1411         if (atomic_read(&(journal->j_wcount)) != 0) {
1412                 reiserfs_panic(s,
1413                                "journal-844: panic journal list is flushing, wcount is not 0\n");
1414         }
1415         cn = jl->j_realblock;
1416         while (cn) {
1417                 was_jwait = 0;
1418                 was_dirty = 0;
1419                 saved_bh = NULL;
1420                 /* blocknr of 0 is no longer in the hash, ignore it */
1421                 if (cn->blocknr == 0) {
1422                         goto free_cnode;
1423                 }
1424
1425                 /* This transaction failed commit. Don't write out to the disk */
1426                 if (!(jl->j_state & LIST_DIRTY))
1427                         goto free_cnode;
1428
1429                 pjl = find_newer_jl_for_cn(cn);
1430                 /* the order is important here.  We check pjl to make sure we
1431                  ** don't clear BH_JDirty_wait if we aren't the one writing this
1432                  ** block to disk
1433                  */
1434                 if (!pjl && cn->bh) {
1435                         saved_bh = cn->bh;
1436
1437                         /* we do this to make sure nobody releases the buffer while
1438                          ** we are working with it
1439                          */
1440                         get_bh(saved_bh);
1441
1442                         if (buffer_journal_dirty(saved_bh)) {
1443                                 BUG_ON(!can_dirty(cn));
1444                                 was_jwait = 1;
1445                                 was_dirty = 1;
1446                         } else if (can_dirty(cn)) {
1447                                 /* everything with !pjl && jwait should be writable */
1448                                 BUG();
1449                         }
1450                 }
1451
1452                 /* if someone has this block in a newer transaction, just make
1453                  ** sure they are commited, and don't try writing it to disk
1454                  */
1455                 if (pjl) {
1456                         if (atomic_read(&pjl->j_commit_left))
1457                                 flush_commit_list(s, pjl, 1);
1458                         goto free_cnode;
1459                 }
1460
1461                 /* bh == NULL when the block got to disk on its own, OR,
1462                  ** the block got freed in a future transaction
1463                  */
1464                 if (saved_bh == NULL) {
1465                         goto free_cnode;
1466                 }
1467
1468                 /* this should never happen.  kupdate_one_transaction has this list
1469                  ** locked while it works, so we should never see a buffer here that
1470                  ** is not marked JDirty_wait
1471                  */
1472                 if ((!was_jwait) && !buffer_locked(saved_bh)) {
1473                         reiserfs_warning(s,
1474                                          "journal-813: BAD! buffer %llu %cdirty %cjwait, "
1475                                          "not in a newer tranasction",
1476                                          (unsigned long long)saved_bh->
1477                                          b_blocknr, was_dirty ? ' ' : '!',
1478                                          was_jwait ? ' ' : '!');
1479                 }
1480                 if (was_dirty) {
1481                         /* we inc again because saved_bh gets decremented at free_cnode */
1482                         get_bh(saved_bh);
1483                         set_bit(BLOCK_NEEDS_FLUSH, &cn->state);
1484                         lock_buffer(saved_bh);
1485                         BUG_ON(cn->blocknr != saved_bh->b_blocknr);
1486                         if (buffer_dirty(saved_bh))
1487                                 submit_logged_buffer(saved_bh);
1488                         else
1489                                 unlock_buffer(saved_bh);
1490                         count++;
1491                 } else {
1492                         reiserfs_warning(s,
1493                                          "clm-2082: Unable to flush buffer %llu in %s",
1494                                          (unsigned long long)saved_bh->
1495                                          b_blocknr, __FUNCTION__);
1496                 }
1497               free_cnode:
1498                 last = cn;
1499                 cn = cn->next;
1500                 if (saved_bh) {
1501                         /* we incremented this to keep others from taking the buffer head away */
1502                         put_bh(saved_bh);
1503                         if (atomic_read(&(saved_bh->b_count)) < 0) {
1504                                 reiserfs_warning(s,
1505                                                  "journal-945: saved_bh->b_count < 0");
1506                         }
1507                 }
1508         }
1509         if (count > 0) {
1510                 cn = jl->j_realblock;
1511                 while (cn) {
1512                         if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) {
1513                                 if (!cn->bh) {
1514                                         reiserfs_panic(s,
1515                                                        "journal-1011: cn->bh is NULL\n");
1516                                 }
1517                                 wait_on_buffer(cn->bh);
1518                                 if (!cn->bh) {
1519                                         reiserfs_panic(s,
1520                                                        "journal-1012: cn->bh is NULL\n");
1521                                 }
1522                                 if (unlikely(!buffer_uptodate(cn->bh))) {
1523 #ifdef CONFIG_REISERFS_CHECK
1524                                         reiserfs_warning(s,
1525                                                          "journal-949: buffer write failed\n");
1526 #endif
1527                                         err = -EIO;
1528                                 }
1529                                 /* note, we must clear the JDirty_wait bit after the up to date
1530                                  ** check, otherwise we race against our flushpage routine
1531                                  */
1532                                 BUG_ON(!test_clear_buffer_journal_dirty
1533                                        (cn->bh));
1534
1535                                 /* undo the inc from journal_mark_dirty */
1536                                 put_bh(cn->bh);
1537                                 brelse(cn->bh);
1538                         }
1539                         cn = cn->next;
1540                 }
1541         }
1542
1543         if (err)
1544                 reiserfs_abort(s, -EIO,
1545                                "Write error while pushing transaction to disk in %s",
1546                                __FUNCTION__);
1547       flush_older_and_return:
1548
1549         /* before we can update the journal header block, we _must_ flush all
1550          ** real blocks from all older transactions to disk.  This is because
1551          ** once the header block is updated, this transaction will not be
1552          ** replayed after a crash
1553          */
1554         if (flushall) {
1555                 flush_older_journal_lists(s, jl);
1556         }
1557
1558         err = journal->j_errno;
1559         /* before we can remove everything from the hash tables for this
1560          ** transaction, we must make sure it can never be replayed
1561          **
1562          ** since we are only called from do_journal_end, we know for sure there
1563          ** are no allocations going on while we are flushing journal lists.  So,
1564          ** we only need to update the journal header block for the last list
1565          ** being flushed
1566          */
1567         if (!err && flushall) {
1568                 err =
1569                     update_journal_header_block(s,
1570                                                 (jl->j_start + jl->j_len +
1571                                                  2) % SB_ONDISK_JOURNAL_SIZE(s),
1572                                                 jl->j_trans_id);
1573                 if (err)
1574                         reiserfs_abort(s, -EIO,
1575                                        "Write error while updating journal header in %s",
1576                                        __FUNCTION__);
1577         }
1578         remove_all_from_journal_list(s, jl, 0);
1579         list_del_init(&jl->j_list);
1580         journal->j_num_lists--;
1581         del_from_work_list(s, jl);
1582
1583         if (journal->j_last_flush_id != 0 &&
1584             (jl->j_trans_id - journal->j_last_flush_id) != 1) {
1585                 reiserfs_warning(s, "clm-2201: last flush %lu, current %lu",
1586                                  journal->j_last_flush_id, jl->j_trans_id);
1587         }
1588         journal->j_last_flush_id = jl->j_trans_id;
1589
1590         /* not strictly required since we are freeing the list, but it should
1591          * help find code using dead lists later on
1592          */
1593         jl->j_len = 0;
1594         atomic_set(&(jl->j_nonzerolen), 0);
1595         jl->j_start = 0;
1596         jl->j_realblock = NULL;
1597         jl->j_commit_bh = NULL;
1598         jl->j_trans_id = 0;
1599         jl->j_state = 0;
1600         put_journal_list(s, jl);
1601         if (flushall)
1602                 up(&journal->j_flush_sem);
1603         put_fs_excl();
1604         return err;
1605 }
1606
1607 static int write_one_transaction(struct super_block *s,
1608                                  struct reiserfs_journal_list *jl,
1609                                  struct buffer_chunk *chunk)
1610 {
1611         struct reiserfs_journal_cnode *cn;
1612         int ret = 0;
1613
1614         jl->j_state |= LIST_TOUCHED;
1615         del_from_work_list(s, jl);
1616         if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0) {
1617                 return 0;
1618         }
1619
1620         cn = jl->j_realblock;
1621         while (cn) {
1622                 /* if the blocknr == 0, this has been cleared from the hash,
1623                  ** skip it
1624                  */
1625                 if (cn->blocknr == 0) {
1626                         goto next;
1627                 }
1628                 if (cn->bh && can_dirty(cn) && buffer_dirty(cn->bh)) {
1629                         struct buffer_head *tmp_bh;
1630                         /* we can race against journal_mark_freed when we try
1631                          * to lock_buffer(cn->bh), so we have to inc the buffer
1632                          * count, and recheck things after locking
1633                          */
1634                         tmp_bh = cn->bh;
1635                         get_bh(tmp_bh);
1636                         lock_buffer(tmp_bh);
1637                         if (cn->bh && can_dirty(cn) && buffer_dirty(tmp_bh)) {
1638                                 if (!buffer_journal_dirty(tmp_bh) ||
1639                                     buffer_journal_prepared(tmp_bh))
1640                                         BUG();
1641                                 add_to_chunk(chunk, tmp_bh, NULL, write_chunk);
1642                                 ret++;
1643                         } else {
1644                                 /* note, cn->bh might be null now */
1645                                 unlock_buffer(tmp_bh);
1646                         }
1647                         put_bh(tmp_bh);
1648                 }
1649               next:
1650                 cn = cn->next;
1651                 cond_resched();
1652         }
1653         return ret;
1654 }
1655
1656 /* used by flush_commit_list */
1657 static int dirty_one_transaction(struct super_block *s,
1658                                  struct reiserfs_journal_list *jl)
1659 {
1660         struct reiserfs_journal_cnode *cn;
1661         struct reiserfs_journal_list *pjl;
1662         int ret = 0;
1663
1664         jl->j_state |= LIST_DIRTY;
1665         cn = jl->j_realblock;
1666         while (cn) {
1667                 /* look for a more recent transaction that logged this
1668                  ** buffer.  Only the most recent transaction with a buffer in
1669                  ** it is allowed to send that buffer to disk
1670                  */
1671                 pjl = find_newer_jl_for_cn(cn);
1672                 if (!pjl && cn->blocknr && cn->bh
1673                     && buffer_journal_dirty(cn->bh)) {
1674                         BUG_ON(!can_dirty(cn));
1675                         /* if the buffer is prepared, it will either be logged
1676                          * or restored.  If restored, we need to make sure
1677                          * it actually gets marked dirty
1678                          */
1679                         clear_buffer_journal_new(cn->bh);
1680                         if (buffer_journal_prepared(cn->bh)) {
1681                                 set_buffer_journal_restore_dirty(cn->bh);
1682                         } else {
1683                                 set_buffer_journal_test(cn->bh);
1684                                 mark_buffer_dirty(cn->bh);
1685                         }
1686                 }
1687                 cn = cn->next;
1688         }
1689         return ret;
1690 }
1691
1692 static int kupdate_transactions(struct super_block *s,
1693                                 struct reiserfs_journal_list *jl,
1694                                 struct reiserfs_journal_list **next_jl,
1695                                 unsigned long *next_trans_id,
1696                                 int num_blocks, int num_trans)
1697 {
1698         int ret = 0;
1699         int written = 0;
1700         int transactions_flushed = 0;
1701         unsigned long orig_trans_id = jl->j_trans_id;
1702         struct buffer_chunk chunk;
1703         struct list_head *entry;
1704         struct reiserfs_journal *journal = SB_JOURNAL(s);
1705         chunk.nr = 0;
1706
1707         down(&journal->j_flush_sem);
1708         if (!journal_list_still_alive(s, orig_trans_id)) {
1709                 goto done;
1710         }
1711
1712         /* we've got j_flush_sem held, nobody is going to delete any
1713          * of these lists out from underneath us
1714          */
1715         while ((num_trans && transactions_flushed < num_trans) ||
1716                (!num_trans && written < num_blocks)) {
1717
1718                 if (jl->j_len == 0 || (jl->j_state & LIST_TOUCHED) ||
1719                     atomic_read(&jl->j_commit_left)
1720                     || !(jl->j_state & LIST_DIRTY)) {
1721                         del_from_work_list(s, jl);
1722                         break;
1723                 }
1724                 ret = write_one_transaction(s, jl, &chunk);
1725
1726                 if (ret < 0)
1727                         goto done;
1728                 transactions_flushed++;
1729                 written += ret;
1730                 entry = jl->j_list.next;
1731
1732                 /* did we wrap? */
1733                 if (entry == &journal->j_journal_list) {
1734                         break;
1735                 }
1736                 jl = JOURNAL_LIST_ENTRY(entry);
1737
1738                 /* don't bother with older transactions */
1739                 if (jl->j_trans_id <= orig_trans_id)
1740                         break;
1741         }
1742         if (chunk.nr) {
1743                 write_chunk(&chunk);
1744         }
1745
1746       done:
1747         up(&journal->j_flush_sem);
1748         return ret;
1749 }
1750
1751 /* for o_sync and fsync heavy applications, they tend to use
1752 ** all the journa list slots with tiny transactions.  These
1753 ** trigger lots and lots of calls to update the header block, which
1754 ** adds seeks and slows things down.
1755 **
1756 ** This function tries to clear out a large chunk of the journal lists
1757 ** at once, which makes everything faster since only the newest journal
1758 ** list updates the header block
1759 */
1760 static int flush_used_journal_lists(struct super_block *s,
1761                                     struct reiserfs_journal_list *jl)
1762 {
1763         unsigned long len = 0;
1764         unsigned long cur_len;
1765         int ret;
1766         int i;
1767         int limit = 256;
1768         struct reiserfs_journal_list *tjl;
1769         struct reiserfs_journal_list *flush_jl;
1770         unsigned long trans_id;
1771         struct reiserfs_journal *journal = SB_JOURNAL(s);
1772
1773         flush_jl = tjl = jl;
1774
1775         /* in data logging mode, try harder to flush a lot of blocks */
1776         if (reiserfs_data_log(s))
1777                 limit = 1024;
1778         /* flush for 256 transactions or limit blocks, whichever comes first */
1779         for (i = 0; i < 256 && len < limit; i++) {
1780                 if (atomic_read(&tjl->j_commit_left) ||
1781                     tjl->j_trans_id < jl->j_trans_id) {
1782                         break;
1783                 }
1784                 cur_len = atomic_read(&tjl->j_nonzerolen);
1785                 if (cur_len > 0) {
1786                         tjl->j_state &= ~LIST_TOUCHED;
1787                 }
1788                 len += cur_len;
1789                 flush_jl = tjl;
1790                 if (tjl->j_list.next == &journal->j_journal_list)
1791                         break;
1792                 tjl = JOURNAL_LIST_ENTRY(tjl->j_list.next);
1793         }
1794         /* try to find a group of blocks we can flush across all the
1795          ** transactions, but only bother if we've actually spanned
1796          ** across multiple lists
1797          */
1798         if (flush_jl != jl) {
1799                 ret = kupdate_transactions(s, jl, &tjl, &trans_id, len, i);
1800         }
1801         flush_journal_list(s, flush_jl, 1);
1802         return 0;
1803 }
1804
1805 /*
1806 ** removes any nodes in table with name block and dev as bh.
1807 ** only touchs the hnext and hprev pointers.
1808 */
1809 void remove_journal_hash(struct super_block *sb,
1810                          struct reiserfs_journal_cnode **table,
1811                          struct reiserfs_journal_list *jl,
1812                          unsigned long block, int remove_freed)
1813 {
1814         struct reiserfs_journal_cnode *cur;
1815         struct reiserfs_journal_cnode **head;
1816
1817         head = &(journal_hash(table, sb, block));
1818         if (!head) {
1819                 return;
1820         }
1821         cur = *head;
1822         while (cur) {
1823                 if (cur->blocknr == block && cur->sb == sb
1824                     && (jl == NULL || jl == cur->jlist)
1825                     && (!test_bit(BLOCK_FREED, &cur->state) || remove_freed)) {
1826                         if (cur->hnext) {
1827                                 cur->hnext->hprev = cur->hprev;
1828                         }
1829                         if (cur->hprev) {
1830                                 cur->hprev->hnext = cur->hnext;
1831                         } else {
1832                                 *head = cur->hnext;
1833                         }
1834                         cur->blocknr = 0;
1835                         cur->sb = NULL;
1836                         cur->state = 0;
1837                         if (cur->bh && cur->jlist)      /* anybody who clears the cur->bh will also dec the nonzerolen */
1838                                 atomic_dec(&(cur->jlist->j_nonzerolen));
1839                         cur->bh = NULL;
1840                         cur->jlist = NULL;
1841                 }
1842                 cur = cur->hnext;
1843         }
1844 }
1845
1846 static void free_journal_ram(struct super_block *p_s_sb)
1847 {
1848         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1849         kfree(journal->j_current_jl);
1850         journal->j_num_lists--;
1851
1852         vfree(journal->j_cnode_free_orig);
1853         free_list_bitmaps(p_s_sb, journal->j_list_bitmap);
1854         free_bitmap_nodes(p_s_sb);      /* must be after free_list_bitmaps */
1855         if (journal->j_header_bh) {
1856                 brelse(journal->j_header_bh);
1857         }
1858         /* j_header_bh is on the journal dev, make sure not to release the journal
1859          * dev until we brelse j_header_bh
1860          */
1861         release_journal_dev(p_s_sb, journal);
1862         vfree(journal);
1863 }
1864
1865 /*
1866 ** call on unmount.  Only set error to 1 if you haven't made your way out
1867 ** of read_super() yet.  Any other caller must keep error at 0.
1868 */
1869 static int do_journal_release(struct reiserfs_transaction_handle *th,
1870                               struct super_block *p_s_sb, int error)
1871 {
1872         struct reiserfs_transaction_handle myth;
1873         int flushed = 0;
1874         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1875
1876         /* we only want to flush out transactions if we were called with error == 0
1877          */
1878         if (!error && !(p_s_sb->s_flags & MS_RDONLY)) {
1879                 /* end the current trans */
1880                 BUG_ON(!th->t_trans_id);
1881                 do_journal_end(th, p_s_sb, 10, FLUSH_ALL);
1882
1883                 /* make sure something gets logged to force our way into the flush code */
1884                 if (!journal_join(&myth, p_s_sb, 1)) {
1885                         reiserfs_prepare_for_journal(p_s_sb,
1886                                                      SB_BUFFER_WITH_SB(p_s_sb),
1887                                                      1);
1888                         journal_mark_dirty(&myth, p_s_sb,
1889                                            SB_BUFFER_WITH_SB(p_s_sb));
1890                         do_journal_end(&myth, p_s_sb, 1, FLUSH_ALL);
1891                         flushed = 1;
1892                 }
1893         }
1894
1895         /* this also catches errors during the do_journal_end above */
1896         if (!error && reiserfs_is_journal_aborted(journal)) {
1897                 memset(&myth, 0, sizeof(myth));
1898                 if (!journal_join_abort(&myth, p_s_sb, 1)) {
1899                         reiserfs_prepare_for_journal(p_s_sb,
1900                                                      SB_BUFFER_WITH_SB(p_s_sb),
1901                                                      1);
1902                         journal_mark_dirty(&myth, p_s_sb,
1903                                            SB_BUFFER_WITH_SB(p_s_sb));
1904                         do_journal_end(&myth, p_s_sb, 1, FLUSH_ALL);
1905                 }
1906         }
1907
1908         reiserfs_mounted_fs_count--;
1909         /* wait for all commits to finish */
1910         cancel_delayed_work(&SB_JOURNAL(p_s_sb)->j_work);
1911         flush_workqueue(commit_wq);
1912         if (!reiserfs_mounted_fs_count) {
1913                 destroy_workqueue(commit_wq);
1914                 commit_wq = NULL;
1915         }
1916
1917         free_journal_ram(p_s_sb);
1918
1919         return 0;
1920 }
1921
1922 /*
1923 ** call on unmount.  flush all journal trans, release all alloc'd ram
1924 */
1925 int journal_release(struct reiserfs_transaction_handle *th,
1926                     struct super_block *p_s_sb)
1927 {
1928         return do_journal_release(th, p_s_sb, 0);
1929 }
1930
1931 /*
1932 ** only call from an error condition inside reiserfs_read_super!
1933 */
1934 int journal_release_error(struct reiserfs_transaction_handle *th,
1935                           struct super_block *p_s_sb)
1936 {
1937         return do_journal_release(th, p_s_sb, 1);
1938 }
1939
1940 /* compares description block with commit block.  returns 1 if they differ, 0 if they are the same */
1941 static int journal_compare_desc_commit(struct super_block *p_s_sb,
1942                                        struct reiserfs_journal_desc *desc,
1943                                        struct reiserfs_journal_commit *commit)
1944 {
1945         if (get_commit_trans_id(commit) != get_desc_trans_id(desc) ||
1946             get_commit_trans_len(commit) != get_desc_trans_len(desc) ||
1947             get_commit_trans_len(commit) > SB_JOURNAL(p_s_sb)->j_trans_max ||
1948             get_commit_trans_len(commit) <= 0) {
1949                 return 1;
1950         }
1951         return 0;
1952 }
1953
1954 /* returns 0 if it did not find a description block
1955 ** returns -1 if it found a corrupt commit block
1956 ** returns 1 if both desc and commit were valid
1957 */
1958 static int journal_transaction_is_valid(struct super_block *p_s_sb,
1959                                         struct buffer_head *d_bh,
1960                                         unsigned long *oldest_invalid_trans_id,
1961                                         unsigned long *newest_mount_id)
1962 {
1963         struct reiserfs_journal_desc *desc;
1964         struct reiserfs_journal_commit *commit;
1965         struct buffer_head *c_bh;
1966         unsigned long offset;
1967
1968         if (!d_bh)
1969                 return 0;
1970
1971         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
1972         if (get_desc_trans_len(desc) > 0
1973             && !memcmp(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8)) {
1974                 if (oldest_invalid_trans_id && *oldest_invalid_trans_id
1975                     && get_desc_trans_id(desc) > *oldest_invalid_trans_id) {
1976                         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
1977                                        "journal-986: transaction "
1978                                        "is valid returning because trans_id %d is greater than "
1979                                        "oldest_invalid %lu",
1980                                        get_desc_trans_id(desc),
1981                                        *oldest_invalid_trans_id);
1982                         return 0;
1983                 }
1984                 if (newest_mount_id
1985                     && *newest_mount_id > get_desc_mount_id(desc)) {
1986                         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
1987                                        "journal-1087: transaction "
1988                                        "is valid returning because mount_id %d is less than "
1989                                        "newest_mount_id %lu",
1990                                        get_desc_mount_id(desc),
1991                                        *newest_mount_id);
1992                         return -1;
1993                 }
1994                 if (get_desc_trans_len(desc) > SB_JOURNAL(p_s_sb)->j_trans_max) {
1995                         reiserfs_warning(p_s_sb,
1996                                          "journal-2018: Bad transaction length %d encountered, ignoring transaction",
1997                                          get_desc_trans_len(desc));
1998                         return -1;
1999                 }
2000                 offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb);
2001
2002                 /* ok, we have a journal description block, lets see if the transaction was valid */
2003                 c_bh =
2004                     journal_bread(p_s_sb,
2005                                   SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2006                                   ((offset + get_desc_trans_len(desc) +
2007                                     1) % SB_ONDISK_JOURNAL_SIZE(p_s_sb)));
2008                 if (!c_bh)
2009                         return 0;
2010                 commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2011                 if (journal_compare_desc_commit(p_s_sb, desc, commit)) {
2012                         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2013                                        "journal_transaction_is_valid, commit offset %ld had bad "
2014                                        "time %d or length %d",
2015                                        c_bh->b_blocknr -
2016                                        SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2017                                        get_commit_trans_id(commit),
2018                                        get_commit_trans_len(commit));
2019                         brelse(c_bh);
2020                         if (oldest_invalid_trans_id) {
2021                                 *oldest_invalid_trans_id =
2022                                     get_desc_trans_id(desc);
2023                                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2024                                                "journal-1004: "
2025                                                "transaction_is_valid setting oldest invalid trans_id "
2026                                                "to %d",
2027                                                get_desc_trans_id(desc));
2028                         }
2029                         return -1;
2030                 }
2031                 brelse(c_bh);
2032                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2033                                "journal-1006: found valid "
2034                                "transaction start offset %llu, len %d id %d",
2035                                d_bh->b_blocknr -
2036                                SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2037                                get_desc_trans_len(desc),
2038                                get_desc_trans_id(desc));
2039                 return 1;
2040         } else {
2041                 return 0;
2042         }
2043 }
2044
2045 static void brelse_array(struct buffer_head **heads, int num)
2046 {
2047         int i;
2048         for (i = 0; i < num; i++) {
2049                 brelse(heads[i]);
2050         }
2051 }
2052
2053 /*
2054 ** given the start, and values for the oldest acceptable transactions,
2055 ** this either reads in a replays a transaction, or returns because the transaction
2056 ** is invalid, or too old.
2057 */
2058 static int journal_read_transaction(struct super_block *p_s_sb,
2059                                     unsigned long cur_dblock,
2060                                     unsigned long oldest_start,
2061                                     unsigned long oldest_trans_id,
2062                                     unsigned long newest_mount_id)
2063 {
2064         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
2065         struct reiserfs_journal_desc *desc;
2066         struct reiserfs_journal_commit *commit;
2067         unsigned long trans_id = 0;
2068         struct buffer_head *c_bh;
2069         struct buffer_head *d_bh;
2070         struct buffer_head **log_blocks = NULL;
2071         struct buffer_head **real_blocks = NULL;
2072         unsigned long trans_offset;
2073         int i;
2074         int trans_half;
2075
2076         d_bh = journal_bread(p_s_sb, cur_dblock);
2077         if (!d_bh)
2078                 return 1;
2079         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2080         trans_offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb);
2081         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1037: "
2082                        "journal_read_transaction, offset %llu, len %d mount_id %d",
2083                        d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2084                        get_desc_trans_len(desc), get_desc_mount_id(desc));
2085         if (get_desc_trans_id(desc) < oldest_trans_id) {
2086                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1039: "
2087                                "journal_read_trans skipping because %lu is too old",
2088                                cur_dblock -
2089                                SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb));
2090                 brelse(d_bh);
2091                 return 1;
2092         }
2093         if (get_desc_mount_id(desc) != newest_mount_id) {
2094                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1146: "
2095                                "journal_read_trans skipping because %d is != "
2096                                "newest_mount_id %lu", get_desc_mount_id(desc),
2097                                newest_mount_id);
2098                 brelse(d_bh);
2099                 return 1;
2100         }
2101         c_bh = journal_bread(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2102                              ((trans_offset + get_desc_trans_len(desc) + 1) %
2103                               SB_ONDISK_JOURNAL_SIZE(p_s_sb)));
2104         if (!c_bh) {
2105                 brelse(d_bh);
2106                 return 1;
2107         }
2108         commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2109         if (journal_compare_desc_commit(p_s_sb, desc, commit)) {
2110                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2111                                "journal_read_transaction, "
2112                                "commit offset %llu had bad time %d or length %d",
2113                                c_bh->b_blocknr -
2114                                SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2115                                get_commit_trans_id(commit),
2116                                get_commit_trans_len(commit));
2117                 brelse(c_bh);
2118                 brelse(d_bh);
2119                 return 1;
2120         }
2121         trans_id = get_desc_trans_id(desc);
2122         /* now we know we've got a good transaction, and it was inside the valid time ranges */
2123         log_blocks = kmalloc(get_desc_trans_len(desc) *
2124                              sizeof(struct buffer_head *), GFP_NOFS);
2125         real_blocks = kmalloc(get_desc_trans_len(desc) *
2126                               sizeof(struct buffer_head *), GFP_NOFS);
2127         if (!log_blocks || !real_blocks) {
2128                 brelse(c_bh);
2129                 brelse(d_bh);
2130                 kfree(log_blocks);
2131                 kfree(real_blocks);
2132                 reiserfs_warning(p_s_sb,
2133                                  "journal-1169: kmalloc failed, unable to mount FS");
2134                 return -1;
2135         }
2136         /* get all the buffer heads */
2137         trans_half = journal_trans_half(p_s_sb->s_blocksize);
2138         for (i = 0; i < get_desc_trans_len(desc); i++) {
2139                 log_blocks[i] =
2140                     journal_getblk(p_s_sb,
2141                                    SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2142                                    (trans_offset + 1 +
2143                                     i) % SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2144                 if (i < trans_half) {
2145                         real_blocks[i] =
2146                             sb_getblk(p_s_sb,
2147                                       le32_to_cpu(desc->j_realblock[i]));
2148                 } else {
2149                         real_blocks[i] =
2150                             sb_getblk(p_s_sb,
2151                                       le32_to_cpu(commit->
2152                                                   j_realblock[i - trans_half]));
2153                 }
2154                 if (real_blocks[i]->b_blocknr > SB_BLOCK_COUNT(p_s_sb)) {
2155                         reiserfs_warning(p_s_sb,
2156                                          "journal-1207: REPLAY FAILURE fsck required! Block to replay is outside of filesystem");
2157                         goto abort_replay;
2158                 }
2159                 /* make sure we don't try to replay onto log or reserved area */
2160                 if (is_block_in_log_or_reserved_area
2161                     (p_s_sb, real_blocks[i]->b_blocknr)) {
2162                         reiserfs_warning(p_s_sb,
2163                                          "journal-1204: REPLAY FAILURE fsck required! Trying to replay onto a log block");
2164                       abort_replay:
2165                         brelse_array(log_blocks, i);
2166                         brelse_array(real_blocks, i);
2167                         brelse(c_bh);
2168                         brelse(d_bh);
2169                         kfree(log_blocks);
2170                         kfree(real_blocks);
2171                         return -1;
2172                 }
2173         }
2174         /* read in the log blocks, memcpy to the corresponding real block */
2175         ll_rw_block(READ, get_desc_trans_len(desc), log_blocks);
2176         for (i = 0; i < get_desc_trans_len(desc); i++) {
2177                 wait_on_buffer(log_blocks[i]);
2178                 if (!buffer_uptodate(log_blocks[i])) {
2179                         reiserfs_warning(p_s_sb,
2180                                          "journal-1212: REPLAY FAILURE fsck required! buffer write failed");
2181                         brelse_array(log_blocks + i,
2182                                      get_desc_trans_len(desc) - i);
2183                         brelse_array(real_blocks, get_desc_trans_len(desc));
2184                         brelse(c_bh);
2185                         brelse(d_bh);
2186                         kfree(log_blocks);
2187                         kfree(real_blocks);
2188                         return -1;
2189                 }
2190                 memcpy(real_blocks[i]->b_data, log_blocks[i]->b_data,
2191                        real_blocks[i]->b_size);
2192                 set_buffer_uptodate(real_blocks[i]);
2193                 brelse(log_blocks[i]);
2194         }
2195         /* flush out the real blocks */
2196         for (i = 0; i < get_desc_trans_len(desc); i++) {
2197                 set_buffer_dirty(real_blocks[i]);
2198                 ll_rw_block(SWRITE, 1, real_blocks + i);
2199         }
2200         for (i = 0; i < get_desc_trans_len(desc); i++) {
2201                 wait_on_buffer(real_blocks[i]);
2202                 if (!buffer_uptodate(real_blocks[i])) {
2203                         reiserfs_warning(p_s_sb,
2204                                          "journal-1226: REPLAY FAILURE, fsck required! buffer write failed");
2205                         brelse_array(real_blocks + i,
2206                                      get_desc_trans_len(desc) - i);
2207                         brelse(c_bh);
2208                         brelse(d_bh);
2209                         kfree(log_blocks);
2210                         kfree(real_blocks);
2211                         return -1;
2212                 }
2213                 brelse(real_blocks[i]);
2214         }
2215         cur_dblock =
2216             SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2217             ((trans_offset + get_desc_trans_len(desc) +
2218               2) % SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2219         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2220                        "journal-1095: setting journal " "start to offset %ld",
2221                        cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb));
2222
2223         /* init starting values for the first transaction, in case this is the last transaction to be replayed. */
2224         journal->j_start = cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb);
2225         journal->j_last_flush_trans_id = trans_id;
2226         journal->j_trans_id = trans_id + 1;
2227         /* check for trans_id overflow */
2228         if (journal->j_trans_id == 0)
2229                 journal->j_trans_id = 10;
2230         brelse(c_bh);
2231         brelse(d_bh);
2232         kfree(log_blocks);
2233         kfree(real_blocks);
2234         return 0;
2235 }
2236
2237 /* This function reads blocks starting from block and to max_block of bufsize
2238    size (but no more than BUFNR blocks at a time). This proved to improve
2239    mounting speed on self-rebuilding raid5 arrays at least.
2240    Right now it is only used from journal code. But later we might use it
2241    from other places.
2242    Note: Do not use journal_getblk/sb_getblk functions here! */
2243 static struct buffer_head *reiserfs_breada(struct block_device *dev, int block,
2244                                            int bufsize, unsigned int max_block)
2245 {
2246         struct buffer_head *bhlist[BUFNR];
2247         unsigned int blocks = BUFNR;
2248         struct buffer_head *bh;
2249         int i, j;
2250
2251         bh = __getblk(dev, block, bufsize);
2252         if (buffer_uptodate(bh))
2253                 return (bh);
2254
2255         if (block + BUFNR > max_block) {
2256                 blocks = max_block - block;
2257         }
2258         bhlist[0] = bh;
2259         j = 1;
2260         for (i = 1; i < blocks; i++) {
2261                 bh = __getblk(dev, block + i, bufsize);
2262                 if (buffer_uptodate(bh)) {
2263                         brelse(bh);
2264                         break;
2265                 } else
2266                         bhlist[j++] = bh;
2267         }
2268         ll_rw_block(READ, j, bhlist);
2269         for (i = 1; i < j; i++)
2270                 brelse(bhlist[i]);
2271         bh = bhlist[0];
2272         wait_on_buffer(bh);
2273         if (buffer_uptodate(bh))
2274                 return bh;
2275         brelse(bh);
2276         return NULL;
2277 }
2278
2279 /*
2280 ** read and replay the log
2281 ** on a clean unmount, the journal header's next unflushed pointer will be to an invalid
2282 ** transaction.  This tests that before finding all the transactions in the log, which makes normal mount times fast.
2283 **
2284 ** After a crash, this starts with the next unflushed transaction, and replays until it finds one too old, or invalid.
2285 **
2286 ** On exit, it sets things up so the first transaction will work correctly.
2287 */
2288 static int journal_read(struct super_block *p_s_sb)
2289 {
2290         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
2291         struct reiserfs_journal_desc *desc;
2292         unsigned long oldest_trans_id = 0;
2293         unsigned long oldest_invalid_trans_id = 0;
2294         time_t start;
2295         unsigned long oldest_start = 0;
2296         unsigned long cur_dblock = 0;
2297         unsigned long newest_mount_id = 9;
2298         struct buffer_head *d_bh;
2299         struct reiserfs_journal_header *jh;
2300         int valid_journal_header = 0;
2301         int replay_count = 0;
2302         int continue_replay = 1;
2303         int ret;
2304         char b[BDEVNAME_SIZE];
2305
2306         cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb);
2307         reiserfs_info(p_s_sb, "checking transaction log (%s)\n",
2308                       bdevname(journal->j_dev_bd, b));
2309         start = get_seconds();
2310
2311         /* step 1, read in the journal header block.  Check the transaction it says
2312          ** is the first unflushed, and if that transaction is not valid,
2313          ** replay is done
2314          */
2315         journal->j_header_bh = journal_bread(p_s_sb,
2316                                              SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb)
2317                                              + SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2318         if (!journal->j_header_bh) {
2319                 return 1;
2320         }
2321         jh = (struct reiserfs_journal_header *)(journal->j_header_bh->b_data);
2322         if (le32_to_cpu(jh->j_first_unflushed_offset) <
2323             SB_ONDISK_JOURNAL_SIZE(p_s_sb)
2324             && le32_to_cpu(jh->j_last_flush_trans_id) > 0) {
2325                 oldest_start =
2326                     SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2327                     le32_to_cpu(jh->j_first_unflushed_offset);
2328                 oldest_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2329                 newest_mount_id = le32_to_cpu(jh->j_mount_id);
2330                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2331                                "journal-1153: found in "
2332                                "header: first_unflushed_offset %d, last_flushed_trans_id "
2333                                "%lu", le32_to_cpu(jh->j_first_unflushed_offset),
2334                                le32_to_cpu(jh->j_last_flush_trans_id));
2335                 valid_journal_header = 1;
2336
2337                 /* now, we try to read the first unflushed offset.  If it is not valid,
2338                  ** there is nothing more we can do, and it makes no sense to read
2339                  ** through the whole log.
2340                  */
2341                 d_bh =
2342                     journal_bread(p_s_sb,
2343                                   SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2344                                   le32_to_cpu(jh->j_first_unflushed_offset));
2345                 ret = journal_transaction_is_valid(p_s_sb, d_bh, NULL, NULL);
2346                 if (!ret) {
2347                         continue_replay = 0;
2348                 }
2349                 brelse(d_bh);
2350                 goto start_log_replay;
2351         }
2352
2353         if (continue_replay && bdev_read_only(p_s_sb->s_bdev)) {
2354                 reiserfs_warning(p_s_sb,
2355                                  "clm-2076: device is readonly, unable to replay log");
2356                 return -1;
2357         }
2358
2359         /* ok, there are transactions that need to be replayed.  start with the first log block, find
2360          ** all the valid transactions, and pick out the oldest.
2361          */
2362         while (continue_replay
2363                && cur_dblock <
2364                (SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2365                 SB_ONDISK_JOURNAL_SIZE(p_s_sb))) {
2366                 /* Note that it is required for blocksize of primary fs device and journal
2367                    device to be the same */
2368                 d_bh =
2369                     reiserfs_breada(journal->j_dev_bd, cur_dblock,
2370                                     p_s_sb->s_blocksize,
2371                                     SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2372                                     SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2373                 ret =
2374                     journal_transaction_is_valid(p_s_sb, d_bh,
2375                                                  &oldest_invalid_trans_id,
2376                                                  &newest_mount_id);
2377                 if (ret == 1) {
2378                         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2379                         if (oldest_start == 0) {        /* init all oldest_ values */
2380                                 oldest_trans_id = get_desc_trans_id(desc);
2381                                 oldest_start = d_bh->b_blocknr;
2382                                 newest_mount_id = get_desc_mount_id(desc);
2383                                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2384                                                "journal-1179: Setting "
2385                                                "oldest_start to offset %llu, trans_id %lu",
2386                                                oldest_start -
2387                                                SB_ONDISK_JOURNAL_1st_BLOCK
2388                                                (p_s_sb), oldest_trans_id);
2389                         } else if (oldest_trans_id > get_desc_trans_id(desc)) {
2390                                 /* one we just read was older */
2391                                 oldest_trans_id = get_desc_trans_id(desc);
2392                                 oldest_start = d_bh->b_blocknr;
2393                                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2394                                                "journal-1180: Resetting "
2395                                                "oldest_start to offset %lu, trans_id %lu",
2396                                                oldest_start -
2397                                                SB_ONDISK_JOURNAL_1st_BLOCK
2398                                                (p_s_sb), oldest_trans_id);
2399                         }
2400                         if (newest_mount_id < get_desc_mount_id(desc)) {
2401                                 newest_mount_id = get_desc_mount_id(desc);
2402                                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2403                                                "journal-1299: Setting "
2404                                                "newest_mount_id to %d",
2405                                                get_desc_mount_id(desc));
2406                         }
2407                         cur_dblock += get_desc_trans_len(desc) + 2;
2408                 } else {
2409                         cur_dblock++;
2410                 }
2411                 brelse(d_bh);
2412         }
2413
2414       start_log_replay:
2415         cur_dblock = oldest_start;
2416         if (oldest_trans_id) {
2417                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2418                                "journal-1206: Starting replay "
2419                                "from offset %llu, trans_id %lu",
2420                                cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2421                                oldest_trans_id);
2422
2423         }
2424         replay_count = 0;
2425         while (continue_replay && oldest_trans_id > 0) {
2426                 ret =
2427                     journal_read_transaction(p_s_sb, cur_dblock, oldest_start,
2428                                              oldest_trans_id, newest_mount_id);
2429                 if (ret < 0) {
2430                         return ret;
2431                 } else if (ret != 0) {
2432                         break;
2433                 }
2434                 cur_dblock =
2435                     SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + journal->j_start;
2436                 replay_count++;
2437                 if (cur_dblock == oldest_start)
2438                         break;
2439         }
2440
2441         if (oldest_trans_id == 0) {
2442                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2443                                "journal-1225: No valid " "transactions found");
2444         }
2445         /* j_start does not get set correctly if we don't replay any transactions.
2446          ** if we had a valid journal_header, set j_start to the first unflushed transaction value,
2447          ** copy the trans_id from the header
2448          */
2449         if (valid_journal_header && replay_count == 0) {
2450                 journal->j_start = le32_to_cpu(jh->j_first_unflushed_offset);
2451                 journal->j_trans_id =
2452                     le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2453                 /* check for trans_id overflow */
2454                 if (journal->j_trans_id == 0)
2455                         journal->j_trans_id = 10;
2456                 journal->j_last_flush_trans_id =
2457                     le32_to_cpu(jh->j_last_flush_trans_id);
2458                 journal->j_mount_id = le32_to_cpu(jh->j_mount_id) + 1;
2459         } else {
2460                 journal->j_mount_id = newest_mount_id + 1;
2461         }
2462         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1299: Setting "
2463                        "newest_mount_id to %lu", journal->j_mount_id);
2464         journal->j_first_unflushed_offset = journal->j_start;
2465         if (replay_count > 0) {
2466                 reiserfs_info(p_s_sb,
2467                               "replayed %d transactions in %lu seconds\n",
2468                               replay_count, get_seconds() - start);
2469         }
2470         if (!bdev_read_only(p_s_sb->s_bdev) &&
2471             _update_journal_header_block(p_s_sb, journal->j_start,
2472                                          journal->j_last_flush_trans_id)) {
2473                 /* replay failed, caller must call free_journal_ram and abort
2474                  ** the mount
2475                  */
2476                 return -1;
2477         }
2478         return 0;
2479 }
2480
2481 static struct reiserfs_journal_list *alloc_journal_list(struct super_block *s)
2482 {
2483         struct reiserfs_journal_list *jl;
2484         jl = kzalloc(sizeof(struct reiserfs_journal_list),
2485                      GFP_NOFS | __GFP_NOFAIL);
2486         INIT_LIST_HEAD(&jl->j_list);
2487         INIT_LIST_HEAD(&jl->j_working_list);
2488         INIT_LIST_HEAD(&jl->j_tail_bh_list);
2489         INIT_LIST_HEAD(&jl->j_bh_list);
2490         sema_init(&jl->j_commit_lock, 1);
2491         SB_JOURNAL(s)->j_num_lists++;
2492         get_journal_list(jl);
2493         return jl;
2494 }
2495
2496 static void journal_list_init(struct super_block *p_s_sb)
2497 {
2498         SB_JOURNAL(p_s_sb)->j_current_jl = alloc_journal_list(p_s_sb);
2499 }
2500
2501 static int release_journal_dev(struct super_block *super,
2502                                struct reiserfs_journal *journal)
2503 {
2504         int result;
2505
2506         result = 0;
2507
2508         if (journal->j_dev_file != NULL) {
2509                 result = filp_close(journal->j_dev_file, NULL);
2510                 journal->j_dev_file = NULL;
2511                 journal->j_dev_bd = NULL;
2512         } else if (journal->j_dev_bd != NULL) {
2513                 result = blkdev_put(journal->j_dev_bd);
2514                 journal->j_dev_bd = NULL;
2515         }
2516
2517         if (result != 0) {
2518                 reiserfs_warning(super,
2519                                  "sh-457: release_journal_dev: Cannot release journal device: %i",
2520                                  result);
2521         }
2522         return result;
2523 }
2524
2525 static int journal_init_dev(struct super_block *super,
2526                             struct reiserfs_journal *journal,
2527                             const char *jdev_name)
2528 {
2529         int result;
2530         dev_t jdev;
2531         int blkdev_mode = FMODE_READ | FMODE_WRITE;
2532         char b[BDEVNAME_SIZE];
2533
2534         result = 0;
2535
2536         journal->j_dev_bd = NULL;
2537         journal->j_dev_file = NULL;
2538         jdev = SB_ONDISK_JOURNAL_DEVICE(super) ?
2539             new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super)) : super->s_dev;
2540
2541         if (bdev_read_only(super->s_bdev))
2542                 blkdev_mode = FMODE_READ;
2543
2544         /* there is no "jdev" option and journal is on separate device */
2545         if ((!jdev_name || !jdev_name[0])) {
2546                 journal->j_dev_bd = open_by_devnum(jdev, blkdev_mode);
2547                 if (IS_ERR(journal->j_dev_bd)) {
2548                         result = PTR_ERR(journal->j_dev_bd);
2549                         journal->j_dev_bd = NULL;
2550                         reiserfs_warning(super, "sh-458: journal_init_dev: "
2551                                          "cannot init journal device '%s': %i",
2552                                          __bdevname(jdev, b), result);
2553                         return result;
2554                 } else if (jdev != super->s_dev)
2555                         set_blocksize(journal->j_dev_bd, super->s_blocksize);
2556                 return 0;
2557         }
2558
2559         journal->j_dev_file = filp_open(jdev_name, 0, 0);
2560         if (!IS_ERR(journal->j_dev_file)) {
2561                 struct inode *jdev_inode = journal->j_dev_file->f_mapping->host;
2562                 if (!S_ISBLK(jdev_inode->i_mode)) {
2563                         reiserfs_warning(super, "journal_init_dev: '%s' is "
2564                                          "not a block device", jdev_name);
2565                         result = -ENOTBLK;
2566                         release_journal_dev(super, journal);
2567                 } else {
2568                         /* ok */
2569                         journal->j_dev_bd = I_BDEV(jdev_inode);
2570                         set_blocksize(journal->j_dev_bd, super->s_blocksize);
2571                         reiserfs_info(super,
2572                                       "journal_init_dev: journal device: %s\n",
2573                                       bdevname(journal->j_dev_bd, b));
2574                 }
2575         } else {
2576                 result = PTR_ERR(journal->j_dev_file);
2577                 journal->j_dev_file = NULL;
2578                 reiserfs_warning(super,
2579                                  "journal_init_dev: Cannot open '%s': %i",
2580                                  jdev_name, result);
2581         }
2582         return result;
2583 }
2584
2585 /*
2586 ** must be called once on fs mount.  calls journal_read for you
2587 */
2588 int journal_init(struct super_block *p_s_sb, const char *j_dev_name,
2589                  int old_format, unsigned int commit_max_age)
2590 {
2591         int num_cnodes = SB_ONDISK_JOURNAL_SIZE(p_s_sb) * 2;
2592         struct buffer_head *bhjh;
2593         struct reiserfs_super_block *rs;
2594         struct reiserfs_journal_header *jh;
2595         struct reiserfs_journal *journal;
2596         struct reiserfs_journal_list *jl;
2597         char b[BDEVNAME_SIZE];
2598
2599         journal = SB_JOURNAL(p_s_sb) = vmalloc(sizeof(struct reiserfs_journal));
2600         if (!journal) {
2601                 reiserfs_warning(p_s_sb,
2602                                  "journal-1256: unable to get memory for journal structure");
2603                 return 1;
2604         }
2605         memset(journal, 0, sizeof(struct reiserfs_journal));
2606         INIT_LIST_HEAD(&journal->j_bitmap_nodes);
2607         INIT_LIST_HEAD(&journal->j_prealloc_list);
2608         INIT_LIST_HEAD(&journal->j_working_list);
2609         INIT_LIST_HEAD(&journal->j_journal_list);
2610         journal->j_persistent_trans = 0;
2611         if (reiserfs_allocate_list_bitmaps(p_s_sb,
2612                                            journal->j_list_bitmap,
2613                                            SB_BMAP_NR(p_s_sb)))
2614                 goto free_and_return;
2615         allocate_bitmap_nodes(p_s_sb);
2616
2617         /* reserved for journal area support */
2618         SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb) = (old_format ?
2619                                                  REISERFS_OLD_DISK_OFFSET_IN_BYTES
2620                                                  / p_s_sb->s_blocksize +
2621                                                  SB_BMAP_NR(p_s_sb) +
2622                                                  1 :
2623                                                  REISERFS_DISK_OFFSET_IN_BYTES /
2624                                                  p_s_sb->s_blocksize + 2);
2625
2626         /* Sanity check to see is the standard journal fitting withing first bitmap
2627            (actual for small blocksizes) */
2628         if (!SB_ONDISK_JOURNAL_DEVICE(p_s_sb) &&
2629             (SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb) +
2630              SB_ONDISK_JOURNAL_SIZE(p_s_sb) > p_s_sb->s_blocksize * 8)) {
2631                 reiserfs_warning(p_s_sb,
2632                                  "journal-1393: journal does not fit for area "
2633                                  "addressed by first of bitmap blocks. It starts at "
2634                                  "%u and its size is %u. Block size %ld",
2635                                  SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb),
2636                                  SB_ONDISK_JOURNAL_SIZE(p_s_sb),
2637                                  p_s_sb->s_blocksize);
2638                 goto free_and_return;
2639         }
2640
2641         if (journal_init_dev(p_s_sb, journal, j_dev_name) != 0) {
2642                 reiserfs_warning(p_s_sb,
2643                                  "sh-462: unable to initialize jornal device");
2644                 goto free_and_return;
2645         }
2646
2647         rs = SB_DISK_SUPER_BLOCK(p_s_sb);
2648
2649         /* read journal header */
2650         bhjh = journal_bread(p_s_sb,
2651                              SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2652                              SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2653         if (!bhjh) {
2654                 reiserfs_warning(p_s_sb,
2655                                  "sh-459: unable to read journal header");
2656                 goto free_and_return;
2657         }
2658         jh = (struct reiserfs_journal_header *)(bhjh->b_data);
2659
2660         /* make sure that journal matches to the super block */
2661         if (is_reiserfs_jr(rs)
2662             && (le32_to_cpu(jh->jh_journal.jp_journal_magic) !=
2663                 sb_jp_journal_magic(rs))) {
2664                 reiserfs_warning(p_s_sb,
2665                                  "sh-460: journal header magic %x "
2666                                  "(device %s) does not match to magic found in super "
2667                                  "block %x", jh->jh_journal.jp_journal_magic,
2668                                  bdevname(journal->j_dev_bd, b),
2669                                  sb_jp_journal_magic(rs));
2670                 brelse(bhjh);
2671                 goto free_and_return;
2672         }
2673
2674         journal->j_trans_max = le32_to_cpu(jh->jh_journal.jp_journal_trans_max);
2675         journal->j_max_batch = le32_to_cpu(jh->jh_journal.jp_journal_max_batch);
2676         journal->j_max_commit_age =
2677             le32_to_cpu(jh->jh_journal.jp_journal_max_commit_age);
2678         journal->j_max_trans_age = JOURNAL_MAX_TRANS_AGE;
2679
2680         if (journal->j_trans_max) {
2681                 /* make sure these parameters are available, assign it if they are not */
2682                 __u32 initial = journal->j_trans_max;
2683                 __u32 ratio = 1;
2684
2685                 if (p_s_sb->s_blocksize < 4096)
2686                         ratio = 4096 / p_s_sb->s_blocksize;
2687
2688                 if (SB_ONDISK_JOURNAL_SIZE(p_s_sb) / journal->j_trans_max <
2689                     JOURNAL_MIN_RATIO)
2690                         journal->j_trans_max =
2691                             SB_ONDISK_JOURNAL_SIZE(p_s_sb) / JOURNAL_MIN_RATIO;
2692                 if (journal->j_trans_max > JOURNAL_TRANS_MAX_DEFAULT / ratio)
2693                         journal->j_trans_max =
2694                             JOURNAL_TRANS_MAX_DEFAULT / ratio;
2695                 if (journal->j_trans_max < JOURNAL_TRANS_MIN_DEFAULT / ratio)
2696                         journal->j_trans_max =
2697                             JOURNAL_TRANS_MIN_DEFAULT / ratio;
2698
2699                 if (journal->j_trans_max != initial)
2700                         reiserfs_warning(p_s_sb,
2701                                          "sh-461: journal_init: wrong transaction max size (%u). Changed to %u",
2702                                          initial, journal->j_trans_max);
2703
2704                 journal->j_max_batch = journal->j_trans_max *
2705                     JOURNAL_MAX_BATCH_DEFAULT / JOURNAL_TRANS_MAX_DEFAULT;
2706         }
2707
2708         if (!journal->j_trans_max) {
2709                 /*we have the file system was created by old version of mkreiserfs
2710                    so this field contains zero value */
2711                 journal->j_trans_max = JOURNAL_TRANS_MAX_DEFAULT;
2712                 journal->j_max_batch = JOURNAL_MAX_BATCH_DEFAULT;
2713                 journal->j_max_commit_age = JOURNAL_MAX_COMMIT_AGE;
2714
2715                 /* for blocksize >= 4096 - max transaction size is 1024. For block size < 4096
2716                    trans max size is decreased proportionally */
2717                 if (p_s_sb->s_blocksize < 4096) {
2718                         journal->j_trans_max /= (4096 / p_s_sb->s_blocksize);
2719                         journal->j_max_batch = (journal->j_trans_max) * 9 / 10;
2720                 }
2721         }
2722
2723         journal->j_default_max_commit_age = journal->j_max_commit_age;
2724
2725         if (commit_max_age != 0) {
2726                 journal->j_max_commit_age = commit_max_age;
2727                 journal->j_max_trans_age = commit_max_age;
2728         }
2729
2730         reiserfs_info(p_s_sb, "journal params: device %s, size %u, "
2731                       "journal first block %u, max trans len %u, max batch %u, "
2732                       "max commit age %u, max trans age %u\n",
2733                       bdevname(journal->j_dev_bd, b),
2734                       SB_ONDISK_JOURNAL_SIZE(p_s_sb),
2735                       SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2736                       journal->j_trans_max,
2737                       journal->j_max_batch,
2738                       journal->j_max_commit_age, journal->j_max_trans_age);
2739
2740         brelse(bhjh);
2741
2742         journal->j_list_bitmap_index = 0;
2743         journal_list_init(p_s_sb);
2744
2745         memset(journal->j_list_hash_table, 0,
2746                JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
2747
2748         INIT_LIST_HEAD(&journal->j_dirty_buffers);
2749         spin_lock_init(&journal->j_dirty_buffers_lock);
2750
2751         journal->j_start = 0;
2752         journal->j_len = 0;
2753         journal->j_len_alloc = 0;
2754         atomic_set(&(journal->j_wcount), 0);
2755         atomic_set(&(journal->j_async_throttle), 0);
2756         journal->j_bcount = 0;
2757         journal->j_trans_start_time = 0;
2758         journal->j_last = NULL;
2759         journal->j_first = NULL;
2760         init_waitqueue_head(&(journal->j_join_wait));
2761         sema_init(&journal->j_lock, 1);
2762         sema_init(&journal->j_flush_sem, 1);
2763
2764         journal->j_trans_id = 10;
2765         journal->j_mount_id = 10;
2766         journal->j_state = 0;
2767         atomic_set(&(journal->j_jlock), 0);
2768         journal->j_cnode_free_list = allocate_cnodes(num_cnodes);
2769         journal->j_cnode_free_orig = journal->j_cnode_free_list;
2770         journal->j_cnode_free = journal->j_cnode_free_list ? num_cnodes : 0;
2771         journal->j_cnode_used = 0;
2772         journal->j_must_wait = 0;
2773
2774         if (journal->j_cnode_free == 0) {
2775                 reiserfs_warning(p_s_sb, "journal-2004: Journal cnode memory "
2776                                  "allocation failed (%ld bytes). Journal is "
2777                                  "too large for available memory. Usually "
2778                                  "this is due to a journal that is too large.",
2779                                  sizeof (struct reiserfs_journal_cnode) * num_cnodes);
2780                 goto free_and_return;
2781         }
2782
2783         init_journal_hash(p_s_sb);
2784         jl = journal->j_current_jl;
2785         jl->j_list_bitmap = get_list_bitmap(p_s_sb, jl);
2786         if (!jl->j_list_bitmap) {
2787                 reiserfs_warning(p_s_sb,
2788                                  "journal-2005, get_list_bitmap failed for journal list 0");
2789                 goto free_and_return;
2790         }
2791         if (journal_read(p_s_sb) < 0) {
2792                 reiserfs_warning(p_s_sb, "Replay Failure, unable to mount");
2793                 goto free_and_return;
2794         }
2795
2796         reiserfs_mounted_fs_count++;
2797         if (reiserfs_mounted_fs_count <= 1)
2798                 commit_wq = create_workqueue("reiserfs");
2799
2800         INIT_WORK(&journal->j_work, flush_async_commits, p_s_sb);
2801         return 0;
2802       free_and_return:
2803         free_journal_ram(p_s_sb);
2804         return 1;
2805 }
2806
2807 /*
2808 ** test for a polite end of the current transaction.  Used by file_write, and should
2809 ** be used by delete to make sure they don't write more than can fit inside a single
2810 ** transaction
2811 */
2812 int journal_transaction_should_end(struct reiserfs_transaction_handle *th,
2813                                    int new_alloc)
2814 {
2815         struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2816         time_t now = get_seconds();
2817         /* cannot restart while nested */
2818         BUG_ON(!th->t_trans_id);
2819         if (th->t_refcount > 1)
2820                 return 0;
2821         if (journal->j_must_wait > 0 ||
2822             (journal->j_len_alloc + new_alloc) >= journal->j_max_batch ||
2823             atomic_read(&(journal->j_jlock)) ||
2824             (now - journal->j_trans_start_time) > journal->j_max_trans_age ||
2825             journal->j_cnode_free < (journal->j_trans_max * 3)) {
2826                 return 1;
2827         }
2828         /* protected by the BKL here */
2829         journal->j_len_alloc += new_alloc;
2830         th->t_blocks_allocated += new_alloc ;
2831         return 0;
2832 }
2833
2834 /* this must be called inside a transaction, and requires the
2835 ** kernel_lock to be held
2836 */
2837 void reiserfs_block_writes(struct reiserfs_transaction_handle *th)
2838 {
2839         struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2840         BUG_ON(!th->t_trans_id);
2841         journal->j_must_wait = 1;
2842         set_bit(J_WRITERS_BLOCKED, &journal->j_state);
2843         return;
2844 }
2845
2846 /* this must be called without a transaction started, and does not
2847 ** require BKL
2848 */
2849 void reiserfs_allow_writes(struct super_block *s)
2850 {
2851         struct reiserfs_journal *journal = SB_JOURNAL(s);
2852         clear_bit(J_WRITERS_BLOCKED, &journal->j_state);
2853         wake_up(&journal->j_join_wait);
2854 }
2855
2856 /* this must be called without a transaction started, and does not
2857 ** require BKL
2858 */
2859 void reiserfs_wait_on_write_block(struct super_block *s)
2860 {
2861         struct reiserfs_journal *journal = SB_JOURNAL(s);
2862         wait_event(journal->j_join_wait,
2863                    !test_bit(J_WRITERS_BLOCKED, &journal->j_state));
2864 }
2865
2866 static void queue_log_writer(struct super_block *s)
2867 {
2868         wait_queue_t wait;
2869         struct reiserfs_journal *journal = SB_JOURNAL(s);
2870         set_bit(J_WRITERS_QUEUED, &journal->j_state);
2871
2872         /*
2873          * we don't want to use wait_event here because
2874          * we only want to wait once.
2875          */
2876         init_waitqueue_entry(&wait, current);
2877         add_wait_queue(&journal->j_join_wait, &wait);
2878         set_current_state(TASK_UNINTERRUPTIBLE);
2879         if (test_bit(J_WRITERS_QUEUED, &journal->j_state))
2880                 schedule();
2881         current->state = TASK_RUNNING;
2882         remove_wait_queue(&journal->j_join_wait, &wait);
2883 }
2884
2885 static void wake_queued_writers(struct super_block *s)
2886 {
2887         struct reiserfs_journal *journal = SB_JOURNAL(s);
2888         if (test_and_clear_bit(J_WRITERS_QUEUED, &journal->j_state))
2889                 wake_up(&journal->j_join_wait);
2890 }
2891
2892 static void let_transaction_grow(struct super_block *sb, unsigned long trans_id)
2893 {
2894         struct reiserfs_journal *journal = SB_JOURNAL(sb);
2895         unsigned long bcount = journal->j_bcount;
2896         while (1) {
2897                 schedule_timeout_uninterruptible(1);
2898                 journal->j_current_jl->j_state |= LIST_COMMIT_PENDING;
2899                 while ((atomic_read(&journal->j_wcount) > 0 ||
2900                         atomic_read(&journal->j_jlock)) &&
2901                        journal->j_trans_id == trans_id) {
2902                         queue_log_writer(sb);
2903                 }
2904                 if (journal->j_trans_id != trans_id)
2905                         break;
2906                 if (bcount == journal->j_bcount)
2907                         break;
2908                 bcount = journal->j_bcount;
2909         }
2910 }
2911
2912 /* join == true if you must join an existing transaction.
2913 ** join == false if you can deal with waiting for others to finish
2914 **
2915 ** this will block until the transaction is joinable.  send the number of blocks you
2916 ** expect to use in nblocks.
2917 */
2918 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
2919                               struct super_block *p_s_sb, unsigned long nblocks,
2920                               int join)
2921 {
2922         time_t now = get_seconds();
2923         int old_trans_id;
2924         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
2925         struct reiserfs_transaction_handle myth;
2926         int sched_count = 0;
2927         int retval;
2928
2929         reiserfs_check_lock_depth(p_s_sb, "journal_begin");
2930         if (nblocks > journal->j_trans_max)
2931                 BUG();
2932
2933         PROC_INFO_INC(p_s_sb, journal.journal_being);
2934         /* set here for journal_join */
2935         th->t_refcount = 1;
2936         th->t_super = p_s_sb;
2937
2938       relock:
2939         lock_journal(p_s_sb);
2940         if (join != JBEGIN_ABORT && reiserfs_is_journal_aborted(journal)) {
2941                 unlock_journal(p_s_sb);
2942                 retval = journal->j_errno;
2943                 goto out_fail;
2944         }
2945         journal->j_bcount++;
2946
2947         if (test_bit(J_WRITERS_BLOCKED, &journal->j_state)) {
2948                 unlock_journal(p_s_sb);
2949                 reiserfs_wait_on_write_block(p_s_sb);
2950                 PROC_INFO_INC(p_s_sb, journal.journal_relock_writers);
2951                 goto relock;
2952         }
2953         now = get_seconds();
2954
2955         /* if there is no room in the journal OR
2956          ** if this transaction is too old, and we weren't called joinable, wait for it to finish before beginning
2957          ** we don't sleep if there aren't other writers
2958          */
2959
2960         if ((!join && journal->j_must_wait > 0) ||
2961             (!join
2962              && (journal->j_len_alloc + nblocks + 2) >= journal->j_max_batch)
2963             || (!join && atomic_read(&journal->j_wcount) > 0
2964                 && journal->j_trans_start_time > 0
2965                 && (now - journal->j_trans_start_time) >
2966                 journal->j_max_trans_age) || (!join
2967                                               && atomic_read(&journal->j_jlock))
2968             || (!join && journal->j_cnode_free < (journal->j_trans_max * 3))) {
2969
2970                 old_trans_id = journal->j_trans_id;
2971                 unlock_journal(p_s_sb); /* allow others to finish this transaction */
2972
2973                 if (!join && (journal->j_len_alloc + nblocks + 2) >=
2974                     journal->j_max_batch &&
2975                     ((journal->j_len + nblocks + 2) * 100) <
2976                     (journal->j_len_alloc * 75)) {
2977                         if (atomic_read(&journal->j_wcount) > 10) {
2978                                 sched_count++;
2979                                 queue_log_writer(p_s_sb);
2980                                 goto relock;
2981                         }
2982                 }
2983                 /* don't mess with joining the transaction if all we have to do is
2984                  * wait for someone else to do a commit
2985                  */
2986                 if (atomic_read(&journal->j_jlock)) {
2987                         while (journal->j_trans_id == old_trans_id &&
2988                                atomic_read(&journal->j_jlock)) {
2989                                 queue_log_writer(p_s_sb);
2990                         }
2991                         goto relock;
2992                 }
2993                 retval = journal_join(&myth, p_s_sb, 1);
2994                 if (retval)
2995                         goto out_fail;
2996
2997                 /* someone might have ended the transaction while we joined */
2998                 if (old_trans_id != journal->j_trans_id) {
2999                         retval = do_journal_end(&myth, p_s_sb, 1, 0);
3000                 } else {
3001                         retval = do_journal_end(&myth, p_s_sb, 1, COMMIT_NOW);
3002                 }
3003
3004                 if (retval)
3005                         goto out_fail;
3006
3007                 PROC_INFO_INC(p_s_sb, journal.journal_relock_wcount);
3008                 goto relock;
3009         }
3010         /* we are the first writer, set trans_id */
3011         if (journal->j_trans_start_time == 0) {
3012                 journal->j_trans_start_time = get_seconds();
3013         }
3014         atomic_inc(&(journal->j_wcount));
3015         journal->j_len_alloc += nblocks;
3016         th->t_blocks_logged = 0;
3017         th->t_blocks_allocated = nblocks;
3018         th->t_trans_id = journal->j_trans_id;
3019         unlock_journal(p_s_sb);
3020         INIT_LIST_HEAD(&th->t_list);
3021         get_fs_excl();
3022         return 0;
3023
3024       out_fail:
3025         memset(th, 0, sizeof(*th));
3026         /* Re-set th->t_super, so we can properly keep track of how many
3027          * persistent transactions there are. We need to do this so if this
3028          * call is part of a failed restart_transaction, we can free it later */
3029         th->t_super = p_s_sb;
3030         return retval;
3031 }
3032
3033 struct reiserfs_transaction_handle *reiserfs_persistent_transaction(struct
3034                                                                     super_block
3035                                                                     *s,
3036                                                                     int nblocks)
3037 {
3038         int ret;
3039         struct reiserfs_transaction_handle *th;
3040
3041         /* if we're nesting into an existing transaction.  It will be
3042          ** persistent on its own
3043          */
3044         if (reiserfs_transaction_running(s)) {
3045                 th = current->journal_info;
3046                 th->t_refcount++;
3047                 if (th->t_refcount < 2) {
3048                         BUG();
3049                 }
3050                 return th;
3051         }
3052         th = kmalloc(sizeof(struct reiserfs_transaction_handle), GFP_NOFS);
3053         if (!th)
3054                 return NULL;
3055         ret = journal_begin(th, s, nblocks);
3056         if (ret) {
3057                 kfree(th);
3058                 return NULL;
3059         }
3060
3061         SB_JOURNAL(s)->j_persistent_trans++;
3062         return th;
3063 }
3064
3065 int reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle *th)
3066 {
3067         struct super_block *s = th->t_super;
3068         int ret = 0;
3069         if (th->t_trans_id)
3070                 ret = journal_end(th, th->t_super, th->t_blocks_allocated);
3071         else
3072                 ret = -EIO;
3073         if (th->t_refcount == 0) {
3074                 SB_JOURNAL(s)->j_persistent_trans--;
3075                 kfree(th);
3076         }
3077         return ret;
3078 }
3079
3080 static int journal_join(struct reiserfs_transaction_handle *th,
3081                         struct super_block *p_s_sb, unsigned long nblocks)
3082 {
3083         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3084
3085         /* this keeps do_journal_end from NULLing out the current->journal_info
3086          ** pointer
3087          */
3088         th->t_handle_save = cur_th;
3089         if (cur_th && cur_th->t_refcount > 1) {
3090                 BUG();
3091         }
3092         return do_journal_begin_r(th, p_s_sb, nblocks, JBEGIN_JOIN);
3093 }
3094
3095 int journal_join_abort(struct reiserfs_transaction_handle *th,
3096                        struct super_block *p_s_sb, unsigned long nblocks)
3097 {
3098         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3099
3100         /* this keeps do_journal_end from NULLing out the current->journal_info
3101          ** pointer
3102          */
3103         th->t_handle_save = cur_th;
3104         if (cur_th && cur_th->t_refcount > 1) {
3105                 BUG();
3106         }
3107         return do_journal_begin_r(th, p_s_sb, nblocks, JBEGIN_ABORT);
3108 }
3109
3110 int journal_begin(struct reiserfs_transaction_handle *th,
3111                   struct super_block *p_s_sb, unsigned long nblocks)
3112 {
3113         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3114         int ret;
3115
3116         th->t_handle_save = NULL;
3117         if (cur_th) {
3118                 /* we are nesting into the current transaction */
3119                 if (cur_th->t_super == p_s_sb) {
3120                         BUG_ON(!cur_th->t_refcount);
3121                         cur_th->t_refcount++;
3122                         memcpy(th, cur_th, sizeof(*th));
3123                         if (th->t_refcount <= 1)
3124                                 reiserfs_warning(p_s_sb,
3125                                                  "BAD: refcount <= 1, but journal_info != 0");
3126                         return 0;
3127                 } else {
3128                         /* we've ended up with a handle from a different filesystem.
3129                          ** save it and restore on journal_end.  This should never
3130                          ** really happen...
3131                          */
3132                         reiserfs_warning(p_s_sb,
3133                                          "clm-2100: nesting info a different FS");
3134                         th->t_handle_save = current->journal_info;
3135                         current->journal_info = th;
3136                 }
3137         } else {
3138                 current->journal_info = th;
3139         }
3140         ret = do_journal_begin_r(th, p_s_sb, nblocks, JBEGIN_REG);
3141         if (current->journal_info != th)
3142                 BUG();
3143
3144         /* I guess this boils down to being the reciprocal of clm-2100 above.
3145          * If do_journal_begin_r fails, we need to put it back, since journal_end
3146          * won't be called to do it. */
3147         if (ret)
3148                 current->journal_info = th->t_handle_save;
3149         else
3150                 BUG_ON(!th->t_refcount);
3151
3152         return ret;
3153 }
3154
3155 /*
3156 ** puts bh into the current transaction.  If it was already there, reorders removes the
3157 ** old pointers from the hash, and puts new ones in (to make sure replay happen in the right order).
3158 **
3159 ** if it was dirty, cleans and files onto the clean list.  I can't let it be dirty again until the
3160 ** transaction is committed.
3161 **
3162 ** if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len.
3163 */
3164 int journal_mark_dirty(struct reiserfs_transaction_handle *th,
3165                        struct super_block *p_s_sb, struct buffer_head *bh)
3166 {
3167         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3168         struct reiserfs_journal_cnode *cn = NULL;
3169         int count_already_incd = 0;
3170         int prepared = 0;
3171         BUG_ON(!th->t_trans_id);
3172
3173         PROC_INFO_INC(p_s_sb, journal.mark_dirty);
3174         if (th->t_trans_id != journal->j_trans_id) {
3175                 reiserfs_panic(th->t_super,
3176                                "journal-1577: handle trans id %ld != current trans id %ld\n",
3177                                th->t_trans_id, journal->j_trans_id);
3178         }
3179
3180         p_s_sb->s_dirt = 1;
3181
3182         prepared = test_clear_buffer_journal_prepared(bh);
3183         clear_buffer_journal_restore_dirty(bh);
3184         /* already in this transaction, we are done */
3185         if (buffer_journaled(bh)) {
3186                 PROC_INFO_INC(p_s_sb, journal.mark_dirty_already);
3187                 return 0;
3188         }
3189
3190         /* this must be turned into a panic instead of a warning.  We can't allow
3191          ** a dirty or journal_dirty or locked buffer to be logged, as some changes
3192          ** could get to disk too early.  NOT GOOD.
3193          */
3194         if (!prepared || buffer_dirty(bh)) {
3195                 reiserfs_warning(p_s_sb, "journal-1777: buffer %llu bad state "
3196                                  "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT",
3197                                  (unsigned long long)bh->b_blocknr,
3198                                  prepared ? ' ' : '!',
3199                                  buffer_locked(bh) ? ' ' : '!',
3200                                  buffer_dirty(bh) ? ' ' : '!',
3201                                  buffer_journal_dirty(bh) ? ' ' : '!');
3202         }
3203
3204         if (atomic_read(&(journal->j_wcount)) <= 0) {
3205                 reiserfs_warning(p_s_sb,
3206                                  "journal-1409: journal_mark_dirty returning because j_wcount was %d",
3207                                  atomic_read(&(journal->j_wcount)));
3208                 return 1;
3209         }
3210         /* this error means I've screwed up, and we've overflowed the transaction.
3211          ** Nothing can be done here, except make the FS readonly or panic.
3212          */
3213         if (journal->j_len >= journal->j_trans_max) {
3214                 reiserfs_panic(th->t_super,
3215                                "journal-1413: journal_mark_dirty: j_len (%lu) is too big\n",
3216                                journal->j_len);
3217         }
3218
3219         if (buffer_journal_dirty(bh)) {
3220                 count_already_incd = 1;
3221                 PROC_INFO_INC(p_s_sb, journal.mark_dirty_notjournal);
3222                 clear_buffer_journal_dirty(bh);
3223         }
3224
3225         if (journal->j_len > journal->j_len_alloc) {
3226                 journal->j_len_alloc = journal->j_len + JOURNAL_PER_BALANCE_CNT;
3227         }
3228
3229         set_buffer_journaled(bh);
3230
3231         /* now put this guy on the end */
3232         if (!cn) {
3233                 cn = get_cnode(p_s_sb);
3234                 if (!cn) {
3235                         reiserfs_panic(p_s_sb, "get_cnode failed!\n");
3236                 }
3237
3238                 if (th->t_blocks_logged == th->t_blocks_allocated) {
3239                         th->t_blocks_allocated += JOURNAL_PER_BALANCE_CNT;
3240                         journal->j_len_alloc += JOURNAL_PER_BALANCE_CNT;
3241                 }
3242                 th->t_blocks_logged++;
3243                 journal->j_len++;
3244
3245                 cn->bh = bh;
3246                 cn->blocknr = bh->b_blocknr;
3247                 cn->sb = p_s_sb;
3248                 cn->jlist = NULL;
3249                 insert_journal_hash(journal->j_hash_table, cn);
3250                 if (!count_already_incd) {
3251                         get_bh(bh);
3252                 }
3253         }
3254         cn->next = NULL;
3255         cn->prev = journal->j_last;
3256         cn->bh = bh;
3257         if (journal->j_last) {
3258                 journal->j_last->next = cn;
3259                 journal->j_last = cn;
3260         } else {
3261                 journal->j_first = cn;
3262                 journal->j_last = cn;
3263         }
3264         return 0;
3265 }
3266
3267 int journal_end(struct reiserfs_transaction_handle *th,
3268                 struct super_block *p_s_sb, unsigned long nblocks)
3269 {
3270         if (!current->journal_info && th->t_refcount > 1)
3271                 reiserfs_warning(p_s_sb, "REISER-NESTING: th NULL, refcount %d",
3272                                  th->t_refcount);
3273
3274         if (!th->t_trans_id) {
3275                 WARN_ON(1);
3276                 return -EIO;
3277         }
3278
3279         th->t_refcount--;
3280         if (th->t_refcount > 0) {
3281                 struct reiserfs_transaction_handle *cur_th =
3282                     current->journal_info;
3283
3284                 /* we aren't allowed to close a nested transaction on a different
3285                  ** filesystem from the one in the task struct
3286                  */
3287                 if (cur_th->t_super != th->t_super)
3288                         BUG();
3289
3290                 if (th != cur_th) {
3291                         memcpy(current->journal_info, th, sizeof(*th));
3292                         th->t_trans_id = 0;
3293                 }
3294                 return 0;
3295         } else {
3296                 return do_journal_end(th, p_s_sb, nblocks, 0);
3297         }
3298 }
3299
3300 /* removes from the current transaction, relsing and descrementing any counters.
3301 ** also files the removed buffer directly onto the clean list
3302 **
3303 ** called by journal_mark_freed when a block has been deleted
3304 **
3305 ** returns 1 if it cleaned and relsed the buffer. 0 otherwise
3306 */
3307 static int remove_from_transaction(struct super_block *p_s_sb,
3308                                    b_blocknr_t blocknr, int already_cleaned)
3309 {
3310         struct buffer_head *bh;
3311         struct reiserfs_journal_cnode *cn;
3312         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3313         int ret = 0;
3314
3315         cn = get_journal_hash_dev(p_s_sb, journal->j_hash_table, blocknr);
3316         if (!cn || !cn->bh) {
3317                 return ret;
3318         }
3319         bh = cn->bh;
3320         if (cn->prev) {
3321                 cn->prev->next = cn->next;
3322         }
3323         if (cn->next) {
3324                 cn->next->prev = cn->prev;
3325         }
3326         if (cn == journal->j_first) {
3327                 journal->j_first = cn->next;
3328         }
3329         if (cn == journal->j_last) {
3330                 journal->j_last = cn->prev;
3331         }
3332         if (bh)
3333                 remove_journal_hash(p_s_sb, journal->j_hash_table, NULL,
3334                                     bh->b_blocknr, 0);
3335         clear_buffer_journaled(bh);     /* don't log this one */
3336
3337         if (!already_cleaned) {
3338                 clear_buffer_journal_dirty(bh);
3339                 clear_buffer_dirty(bh);
3340                 clear_buffer_journal_test(bh);
3341                 put_bh(bh);
3342                 if (atomic_read(&(bh->b_count)) < 0) {
3343                         reiserfs_warning(p_s_sb,
3344                                          "journal-1752: remove from trans, b_count < 0");
3345                 }
3346                 ret = 1;
3347         }
3348         journal->j_len--;
3349         journal->j_len_alloc--;
3350         free_cnode(p_s_sb, cn);
3351         return ret;
3352 }
3353
3354 /*
3355 ** for any cnode in a journal list, it can only be dirtied of all the
3356 ** transactions that include it are commited to disk.
3357 ** this checks through each transaction, and returns 1 if you are allowed to dirty,
3358 ** and 0 if you aren't
3359 **
3360 ** it is called by dirty_journal_list, which is called after flush_commit_list has gotten all the log
3361 ** blocks for a given transaction on disk
3362 **
3363 */
3364 static int can_dirty(struct reiserfs_journal_cnode *cn)
3365 {
3366         struct super_block *sb = cn->sb;
3367         b_blocknr_t blocknr = cn->blocknr;
3368         struct reiserfs_journal_cnode *cur = cn->hprev;
3369         int can_dirty = 1;
3370
3371         /* first test hprev.  These are all newer than cn, so any node here
3372          ** with the same block number and dev means this node can't be sent
3373          ** to disk right now.
3374          */
3375         while (cur && can_dirty) {
3376                 if (cur->jlist && cur->bh && cur->blocknr && cur->sb == sb &&
3377                     cur->blocknr == blocknr) {
3378                         can_dirty = 0;
3379                 }
3380                 cur = cur->hprev;
3381         }
3382         /* then test hnext.  These are all older than cn.  As long as they
3383          ** are committed to the log, it is safe to write cn to disk
3384          */
3385         cur = cn->hnext;
3386         while (cur && can_dirty) {
3387                 if (cur->jlist && cur->jlist->j_len > 0 &&
3388                     atomic_read(&(cur->jlist->j_commit_left)) > 0 && cur->bh &&
3389                     cur->blocknr && cur->sb == sb && cur->blocknr == blocknr) {
3390                         can_dirty = 0;
3391                 }
3392                 cur = cur->hnext;
3393         }
3394         return can_dirty;
3395 }
3396
3397 /* syncs the commit blocks, but does not force the real buffers to disk
3398 ** will wait until the current transaction is done/commited before returning
3399 */
3400 int journal_end_sync(struct reiserfs_transaction_handle *th,
3401                      struct super_block *p_s_sb, unsigned long nblocks)
3402 {
3403         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3404
3405         BUG_ON(!th->t_trans_id);
3406         /* you can sync while nested, very, very bad */
3407         if (th->t_refcount > 1) {
3408                 BUG();
3409         }
3410         if (journal->j_len == 0) {
3411                 reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb),
3412                                              1);
3413                 journal_mark_dirty(th, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb));
3414         }
3415         return do_journal_end(th, p_s_sb, nblocks, COMMIT_NOW | WAIT);
3416 }
3417
3418 /*
3419 ** writeback the pending async commits to disk
3420 */
3421 static void flush_async_commits(void *p)
3422 {
3423         struct super_block *p_s_sb = p;
3424         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3425         struct reiserfs_journal_list *jl;
3426         struct list_head *entry;
3427
3428         lock_kernel();
3429         if (!list_empty(&journal->j_journal_list)) {
3430                 /* last entry is the youngest, commit it and you get everything */
3431                 entry = journal->j_journal_list.prev;
3432                 jl = JOURNAL_LIST_ENTRY(entry);
3433                 flush_commit_list(p_s_sb, jl, 1);
3434         }
3435         unlock_kernel();
3436         /*
3437          * this is a little racey, but there's no harm in missing
3438          * the filemap_fdata_write
3439          */
3440         if (!atomic_read(&journal->j_async_throttle)
3441             && !reiserfs_is_journal_aborted(journal)) {
3442                 atomic_inc(&journal->j_async_throttle);
3443                 filemap_fdatawrite(p_s_sb->s_bdev->bd_inode->i_mapping);
3444                 atomic_dec(&journal->j_async_throttle);
3445         }
3446 }
3447
3448 /*
3449 ** flushes any old transactions to disk
3450 ** ends the current transaction if it is too old
3451 */
3452 int reiserfs_flush_old_commits(struct super_block *p_s_sb)
3453 {
3454         time_t now;
3455         struct reiserfs_transaction_handle th;
3456         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3457
3458         now = get_seconds();
3459         /* safety check so we don't flush while we are replaying the log during
3460          * mount
3461          */
3462         if (list_empty(&journal->j_journal_list)) {
3463                 return 0;
3464         }
3465
3466         /* check the current transaction.  If there are no writers, and it is
3467          * too old, finish it, and force the commit blocks to disk
3468          */
3469         if (atomic_read(&journal->j_wcount) <= 0 &&
3470             journal->j_trans_start_time > 0 &&
3471             journal->j_len > 0 &&
3472             (now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3473                 if (!journal_join(&th, p_s_sb, 1)) {
3474                         reiserfs_prepare_for_journal(p_s_sb,
3475                                                      SB_BUFFER_WITH_SB(p_s_sb),
3476                                                      1);
3477                         journal_mark_dirty(&th, p_s_sb,
3478                                            SB_BUFFER_WITH_SB(p_s_sb));
3479
3480                         /* we're only being called from kreiserfsd, it makes no sense to do
3481                          ** an async commit so that kreiserfsd can do it later
3482                          */
3483                         do_journal_end(&th, p_s_sb, 1, COMMIT_NOW | WAIT);
3484                 }
3485         }
3486         return p_s_sb->s_dirt;
3487 }
3488
3489 /*
3490 ** returns 0 if do_journal_end should return right away, returns 1 if do_journal_end should finish the commit
3491 **
3492 ** if the current transaction is too old, but still has writers, this will wait on j_join_wait until all
3493 ** the writers are done.  By the time it wakes up, the transaction it was called has already ended, so it just
3494 ** flushes the commit list and returns 0.
3495 **
3496 ** Won't batch when flush or commit_now is set.  Also won't batch when others are waiting on j_join_wait.
3497 **
3498 ** Note, we can't allow the journal_end to proceed while there are still writers in the log.
3499 */
3500 static int check_journal_end(struct reiserfs_transaction_handle *th,
3501                              struct super_block *p_s_sb, unsigned long nblocks,
3502                              int flags)
3503 {
3504
3505         time_t now;
3506         int flush = flags & FLUSH_ALL;
3507         int commit_now = flags & COMMIT_NOW;
3508         int wait_on_commit = flags & WAIT;
3509         struct reiserfs_journal_list *jl;
3510         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3511
3512         BUG_ON(!th->t_trans_id);
3513
3514         if (th->t_trans_id != journal->j_trans_id) {
3515                 reiserfs_panic(th->t_super,
3516                                "journal-1577: handle trans id %ld != current trans id %ld\n",
3517                                th->t_trans_id, journal->j_trans_id);
3518         }
3519
3520         journal->j_len_alloc -= (th->t_blocks_allocated - th->t_blocks_logged);
3521         if (atomic_read(&(journal->j_wcount)) > 0) {    /* <= 0 is allowed.  unmounting might not call begin */
3522                 atomic_dec(&(journal->j_wcount));
3523         }
3524
3525         /* BUG, deal with case where j_len is 0, but people previously freed blocks need to be released
3526          ** will be dealt with by next transaction that actually writes something, but should be taken
3527          ** care of in this trans
3528          */
3529         if (journal->j_len == 0) {
3530                 BUG();
3531         }
3532         /* if wcount > 0, and we are called to with flush or commit_now,
3533          ** we wait on j_join_wait.  We will wake up when the last writer has
3534          ** finished the transaction, and started it on its way to the disk.
3535          ** Then, we flush the commit or journal list, and just return 0
3536          ** because the rest of journal end was already done for this transaction.
3537          */
3538         if (atomic_read(&(journal->j_wcount)) > 0) {
3539                 if (flush || commit_now) {
3540                         unsigned trans_id;
3541
3542                         jl = journal->j_current_jl;
3543                         trans_id = jl->j_trans_id;
3544                         if (wait_on_commit)
3545                                 jl->j_state |= LIST_COMMIT_PENDING;
3546                         atomic_set(&(journal->j_jlock), 1);
3547                         if (flush) {
3548                                 journal->j_next_full_flush = 1;
3549                         }
3550                         unlock_journal(p_s_sb);
3551
3552                         /* sleep while the current transaction is still j_jlocked */
3553                         while (journal->j_trans_id == trans_id) {
3554                                 if (atomic_read(&journal->j_jlock)) {
3555                                         queue_log_writer(p_s_sb);
3556                                 } else {
3557                                         lock_journal(p_s_sb);
3558                                         if (journal->j_trans_id == trans_id) {
3559                                                 atomic_set(&(journal->j_jlock),
3560                                                            1);
3561                                         }
3562                                         unlock_journal(p_s_sb);
3563                                 }
3564                         }
3565                         if (journal->j_trans_id == trans_id) {
3566                                 BUG();
3567                         }
3568                         if (commit_now
3569                             && journal_list_still_alive(p_s_sb, trans_id)
3570                             && wait_on_commit) {
3571                                 flush_commit_list(p_s_sb, jl, 1);
3572                         }
3573                         return 0;
3574                 }
3575                 unlock_journal(p_s_sb);
3576                 return 0;
3577         }
3578
3579         /* deal with old transactions where we are the last writers */
3580         now = get_seconds();
3581         if ((now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3582                 commit_now = 1;
3583                 journal->j_next_async_flush = 1;
3584         }
3585         /* don't batch when someone is waiting on j_join_wait */
3586         /* don't batch when syncing the commit or flushing the whole trans */
3587         if (!(journal->j_must_wait > 0) && !(atomic_read(&(journal->j_jlock)))
3588             && !flush && !commit_now && (journal->j_len < journal->j_max_batch)
3589             && journal->j_len_alloc < journal->j_max_batch
3590             && journal->j_cnode_free > (journal->j_trans_max * 3)) {
3591                 journal->j_bcount++;
3592                 unlock_journal(p_s_sb);
3593                 return 0;
3594         }
3595
3596         if (journal->j_start > SB_ONDISK_JOURNAL_SIZE(p_s_sb)) {
3597                 reiserfs_panic(p_s_sb,
3598                                "journal-003: journal_end: j_start (%ld) is too high\n",
3599                                journal->j_start);
3600         }
3601         return 1;
3602 }
3603
3604 /*
3605 ** Does all the work that makes deleting blocks safe.
3606 ** when deleting a block mark BH_JNew, just remove it from the current transaction, clean it's buffer_head and move on.
3607 **
3608 ** otherwise:
3609 ** set a bit for the block in the journal bitmap.  That will prevent it from being allocated for unformatted nodes
3610 ** before this transaction has finished.
3611 **
3612 ** mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers.  That will prevent any old transactions with
3613 ** this block from trying to flush to the real location.  Since we aren't removing the cnode from the journal_list_hash,
3614 ** the block can't be reallocated yet.
3615 **
3616 ** Then remove it from the current transaction, decrementing any counters and filing it on the clean list.
3617 */
3618 int journal_mark_freed(struct reiserfs_transaction_handle *th,
3619                        struct super_block *p_s_sb, b_blocknr_t blocknr)
3620 {
3621         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3622         struct reiserfs_journal_cnode *cn = NULL;
3623         struct buffer_head *bh = NULL;
3624         struct reiserfs_list_bitmap *jb = NULL;
3625         int cleaned = 0;
3626         BUG_ON(!th->t_trans_id);
3627
3628         cn = get_journal_hash_dev(p_s_sb, journal->j_hash_table, blocknr);
3629         if (cn && cn->bh) {
3630                 bh = cn->bh;
3631                 get_bh(bh);
3632         }
3633         /* if it is journal new, we just remove it from this transaction */
3634         if (bh && buffer_journal_new(bh)) {
3635                 clear_buffer_journal_new(bh);
3636                 clear_prepared_bits(bh);
3637                 reiserfs_clean_and_file_buffer(bh);
3638                 cleaned = remove_from_transaction(p_s_sb, blocknr, cleaned);
3639         } else {
3640                 /* set the bit for this block in the journal bitmap for this transaction */
3641                 jb = journal->j_current_jl->j_list_bitmap;
3642                 if (!jb) {
3643                         reiserfs_panic(p_s_sb,
3644                                        "journal-1702: journal_mark_freed, journal_list_bitmap is NULL\n");
3645                 }
3646                 set_bit_in_list_bitmap(p_s_sb, blocknr, jb);
3647
3648                 /* Note, the entire while loop is not allowed to schedule.  */
3649
3650                 if (bh) {
3651                         clear_prepared_bits(bh);
3652                         reiserfs_clean_and_file_buffer(bh);
3653                 }
3654                 cleaned = remove_from_transaction(p_s_sb, blocknr, cleaned);
3655
3656                 /* find all older transactions with this block, make sure they don't try to write it out */
3657                 cn = get_journal_hash_dev(p_s_sb, journal->j_list_hash_table,
3658                                           blocknr);
3659                 while (cn) {
3660                         if (p_s_sb == cn->sb && blocknr == cn->blocknr) {
3661                                 set_bit(BLOCK_FREED, &cn->state);
3662                                 if (cn->bh) {
3663                                         if (!cleaned) {
3664                                                 /* remove_from_transaction will brelse the buffer if it was
3665                                                  ** in the current trans
3666                                                  */
3667                                                 clear_buffer_journal_dirty(cn->
3668                                                                            bh);
3669                                                 clear_buffer_dirty(cn->bh);
3670                                                 clear_buffer_journal_test(cn->
3671                                                                           bh);
3672                                                 cleaned = 1;
3673                                                 put_bh(cn->bh);
3674                                                 if (atomic_read
3675                                                     (&(cn->bh->b_count)) < 0) {
3676                                                         reiserfs_warning(p_s_sb,
3677                                                                          "journal-2138: cn->bh->b_count < 0");
3678                                                 }
3679                                         }
3680                                         if (cn->jlist) {        /* since we are clearing the bh, we MUST dec nonzerolen */
3681                                                 atomic_dec(&
3682                                                            (cn->jlist->
3683                                                             j_nonzerolen));
3684                                         }
3685                                         cn->bh = NULL;
3686                                 }
3687                         }
3688                         cn = cn->hnext;
3689                 }
3690         }
3691
3692         if (bh) {
3693                 put_bh(bh);     /* get_hash grabs the buffer */
3694                 if (atomic_read(&(bh->b_count)) < 0) {
3695                         reiserfs_warning(p_s_sb,
3696                                          "journal-2165: bh->b_count < 0");
3697                 }
3698         }
3699         return 0;
3700 }
3701
3702 void reiserfs_update_inode_transaction(struct inode *inode)
3703 {
3704         struct reiserfs_journal *journal = SB_JOURNAL(inode->i_sb);
3705         REISERFS_I(inode)->i_jl = journal->j_current_jl;
3706         REISERFS_I(inode)->i_trans_id = journal->j_trans_id;
3707 }
3708
3709 /*
3710  * returns -1 on error, 0 if no commits/barriers were done and 1
3711  * if a transaction was actually committed and the barrier was done
3712  */
3713 static int __commit_trans_jl(struct inode *inode, unsigned long id,
3714                              struct reiserfs_journal_list *jl)
3715 {
3716         struct reiserfs_transaction_handle th;
3717         struct super_block *sb = inode->i_sb;
3718         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3719         int ret = 0;
3720
3721         /* is it from the current transaction, or from an unknown transaction? */
3722         if (id == journal->j_trans_id) {
3723                 jl = journal->j_current_jl;
3724                 /* try to let other writers come in and grow this transaction */
3725                 let_transaction_grow(sb, id);
3726                 if (journal->j_trans_id != id) {
3727                         goto flush_commit_only;
3728                 }
3729
3730                 ret = journal_begin(&th, sb, 1);
3731                 if (ret)
3732                         return ret;
3733
3734                 /* someone might have ended this transaction while we joined */
3735                 if (journal->j_trans_id != id) {
3736                         reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3737                                                      1);
3738                         journal_mark_dirty(&th, sb, SB_BUFFER_WITH_SB(sb));
3739                         ret = journal_end(&th, sb, 1);
3740                         goto flush_commit_only;
3741                 }
3742
3743                 ret = journal_end_sync(&th, sb, 1);
3744                 if (!ret)
3745                         ret = 1;
3746
3747         } else {
3748                 /* this gets tricky, we have to make sure the journal list in
3749                  * the inode still exists.  We know the list is still around
3750                  * if we've got a larger transaction id than the oldest list
3751                  */
3752               flush_commit_only:
3753                 if (journal_list_still_alive(inode->i_sb, id)) {
3754                         /*
3755                          * we only set ret to 1 when we know for sure
3756                          * the barrier hasn't been started yet on the commit
3757                          * block.
3758                          */
3759                         if (atomic_read(&jl->j_commit_left) > 1)
3760                                 ret = 1;
3761                         flush_commit_list(sb, jl, 1);
3762                         if (journal->j_errno)
3763                                 ret = journal->j_errno;
3764                 }
3765         }
3766         /* otherwise the list is gone, and long since committed */
3767         return ret;
3768 }
3769
3770 int reiserfs_commit_for_inode(struct inode *inode)
3771 {
3772         unsigned long id = REISERFS_I(inode)->i_trans_id;
3773         struct reiserfs_journal_list *jl = REISERFS_I(inode)->i_jl;
3774
3775         /* for the whole inode, assume unset id means it was
3776          * changed in the current transaction.  More conservative
3777          */
3778         if (!id || !jl) {
3779                 reiserfs_update_inode_transaction(inode);
3780                 id = REISERFS_I(inode)->i_trans_id;
3781                 /* jl will be updated in __commit_trans_jl */
3782         }
3783
3784         return __commit_trans_jl(inode, id, jl);
3785 }
3786
3787 void reiserfs_restore_prepared_buffer(struct super_block *p_s_sb,
3788                                       struct buffer_head *bh)
3789 {
3790         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3791         PROC_INFO_INC(p_s_sb, journal.restore_prepared);
3792         if (!bh) {
3793                 return;
3794         }
3795         if (test_clear_buffer_journal_restore_dirty(bh) &&
3796             buffer_journal_dirty(bh)) {
3797                 struct reiserfs_journal_cnode *cn;
3798                 cn = get_journal_hash_dev(p_s_sb,
3799                                           journal->j_list_hash_table,
3800                                           bh->b_blocknr);
3801                 if (cn && can_dirty(cn)) {
3802                         set_buffer_journal_test(bh);
3803                         mark_buffer_dirty(bh);
3804                 }
3805         }
3806         clear_buffer_journal_prepared(bh);
3807 }
3808
3809 extern struct tree_balance *cur_tb;
3810 /*
3811 ** before we can change a metadata block, we have to make sure it won't
3812 ** be written to disk while we are altering it.  So, we must:
3813 ** clean it
3814 ** wait on it.
3815 **
3816 */
3817 int reiserfs_prepare_for_journal(struct super_block *p_s_sb,
3818                                  struct buffer_head *bh, int wait)
3819 {
3820         PROC_INFO_INC(p_s_sb, journal.prepare);
3821
3822         if (test_set_buffer_locked(bh)) {
3823                 if (!wait)
3824                         return 0;
3825                 lock_buffer(bh);
3826         }
3827         set_buffer_journal_prepared(bh);
3828         if (test_clear_buffer_dirty(bh) && buffer_journal_dirty(bh)) {
3829                 clear_buffer_journal_test(bh);
3830                 set_buffer_journal_restore_dirty(bh);
3831         }
3832         unlock_buffer(bh);
3833         return 1;
3834 }
3835
3836 static void flush_old_journal_lists(struct super_block *s)
3837 {
3838         struct reiserfs_journal *journal = SB_JOURNAL(s);
3839         struct reiserfs_journal_list *jl;
3840         struct list_head *entry;
3841         time_t now = get_seconds();
3842
3843         while (!list_empty(&journal->j_journal_list)) {
3844                 entry = journal->j_journal_list.next;
3845                 jl = JOURNAL_LIST_ENTRY(entry);
3846                 /* this check should always be run, to send old lists to disk */
3847                 if (jl->j_timestamp < (now - (JOURNAL_MAX_TRANS_AGE * 4))) {
3848                         flush_used_journal_lists(s, jl);
3849                 } else {
3850                         break;
3851                 }
3852         }
3853 }
3854
3855 /*
3856 ** long and ugly.  If flush, will not return until all commit
3857 ** blocks and all real buffers in the trans are on disk.
3858 ** If no_async, won't return until all commit blocks are on disk.
3859 **
3860 ** keep reading, there are comments as you go along
3861 **
3862 ** If the journal is aborted, we just clean up. Things like flushing
3863 ** journal lists, etc just won't happen.
3864 */
3865 static int do_journal_end(struct reiserfs_transaction_handle *th,
3866                           struct super_block *p_s_sb, unsigned long nblocks,
3867                           int flags)
3868 {
3869         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3870         struct reiserfs_journal_cnode *cn, *next, *jl_cn;
3871         struct reiserfs_journal_cnode *last_cn = NULL;
3872         struct reiserfs_journal_desc *desc;
3873         struct reiserfs_journal_commit *commit;
3874         struct buffer_head *c_bh;       /* commit bh */
3875         struct buffer_head *d_bh;       /* desc bh */
3876         int cur_write_start = 0;        /* start index of current log write */
3877         int old_start;
3878         int i;
3879         int flush;
3880         int wait_on_commit;
3881         struct reiserfs_journal_list *jl, *temp_jl;
3882         struct list_head *entry, *safe;
3883         unsigned long jindex;
3884         unsigned long commit_trans_id;
3885         int trans_half;
3886
3887         BUG_ON(th->t_refcount > 1);
3888         BUG_ON(!th->t_trans_id);
3889
3890         /* protect flush_older_commits from doing mistakes if the
3891            transaction ID counter gets overflowed.  */
3892         if (th->t_trans_id == ~0UL)
3893                 flags |= FLUSH_ALL | COMMIT_NOW | WAIT;
3894         flush = flags & FLUSH_ALL;
3895         wait_on_commit = flags & WAIT;
3896
3897         put_fs_excl();
3898         current->journal_info = th->t_handle_save;
3899         reiserfs_check_lock_depth(p_s_sb, "journal end");
3900         if (journal->j_len == 0) {
3901                 reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb),
3902                                              1);
3903                 journal_mark_dirty(th, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb));
3904         }
3905
3906         lock_journal(p_s_sb);
3907         if (journal->j_next_full_flush) {
3908                 flags |= FLUSH_ALL;
3909                 flush = 1;
3910         }
3911         if (journal->j_next_async_flush) {
3912                 flags |= COMMIT_NOW | WAIT;
3913                 wait_on_commit = 1;
3914         }
3915
3916         /* check_journal_end locks the journal, and unlocks if it does not return 1
3917          ** it tells us if we should continue with the journal_end, or just return
3918          */
3919         if (!check_journal_end(th, p_s_sb, nblocks, flags)) {
3920                 p_s_sb->s_dirt = 1;
3921                 wake_queued_writers(p_s_sb);
3922                 reiserfs_async_progress_wait(p_s_sb);
3923                 goto out;
3924         }
3925
3926         /* check_journal_end might set these, check again */
3927         if (journal->j_next_full_flush) {
3928                 flush = 1;
3929         }
3930
3931         /*
3932          ** j must wait means we have to flush the log blocks, and the real blocks for
3933          ** this transaction
3934          */
3935         if (journal->j_must_wait > 0) {
3936                 flush = 1;
3937         }
3938 #ifdef REISERFS_PREALLOCATE
3939         /* quota ops might need to nest, setup the journal_info pointer for them
3940          * and raise the refcount so that it is > 0. */
3941         current->journal_info = th;
3942         th->t_refcount++;
3943         reiserfs_discard_all_prealloc(th);      /* it should not involve new blocks into
3944                                                  * the transaction */
3945         th->t_refcount--;
3946         current->journal_info = th->t_handle_save;
3947 #endif
3948
3949         /* setup description block */
3950         d_bh =
3951             journal_getblk(p_s_sb,
3952                            SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
3953                            journal->j_start);
3954         set_buffer_uptodate(d_bh);
3955         desc = (struct reiserfs_journal_desc *)(d_bh)->b_data;
3956         memset(d_bh->b_data, 0, d_bh->b_size);
3957         memcpy(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8);
3958         set_desc_trans_id(desc, journal->j_trans_id);
3959
3960         /* setup commit block.  Don't write (keep it clean too) this one until after everyone else is written */
3961         c_bh = journal_getblk(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
3962                               ((journal->j_start + journal->j_len +
3963                                 1) % SB_ONDISK_JOURNAL_SIZE(p_s_sb)));
3964         commit = (struct reiserfs_journal_commit *)c_bh->b_data;
3965         memset(c_bh->b_data, 0, c_bh->b_size);
3966         set_commit_trans_id(commit, journal->j_trans_id);
3967         set_buffer_uptodate(c_bh);
3968
3969         /* init this journal list */
3970         jl = journal->j_current_jl;
3971
3972         /* we lock the commit before doing anything because
3973          * we want to make sure nobody tries to run flush_commit_list until
3974          * the new transaction is fully setup, and we've already flushed the
3975          * ordered bh list
3976          */
3977         down(&jl->j_commit_lock);
3978
3979         /* save the transaction id in case we need to commit it later */
3980         commit_trans_id = jl->j_trans_id;
3981
3982         atomic_set(&jl->j_older_commits_done, 0);
3983         jl->j_trans_id = journal->j_trans_id;
3984         jl->j_timestamp = journal->j_trans_start_time;
3985         jl->j_commit_bh = c_bh;
3986         jl->j_start = journal->j_start;
3987         jl->j_len = journal->j_len;
3988         atomic_set(&jl->j_nonzerolen, journal->j_len);
3989         atomic_set(&jl->j_commit_left, journal->j_len + 2);
3990         jl->j_realblock = NULL;
3991
3992         /* The ENTIRE FOR LOOP MUST not cause schedule to occur.
3993          **  for each real block, add it to the journal list hash,
3994          ** copy into real block index array in the commit or desc block
3995          */
3996         trans_half = journal_trans_half(p_s_sb->s_blocksize);
3997         for (i = 0, cn = journal->j_first; cn; cn = cn->next, i++) {
3998                 if (buffer_journaled(cn->bh)) {
3999                         jl_cn = get_cnode(p_s_sb);
4000                         if (!jl_cn) {
4001                                 reiserfs_panic(p_s_sb,
4002                                                "journal-1676, get_cnode returned NULL\n");
4003                         }
4004                         if (i == 0) {
4005                                 jl->j_realblock = jl_cn;
4006                         }
4007                         jl_cn->prev = last_cn;
4008                         jl_cn->next = NULL;
4009                         if (last_cn) {
4010                                 last_cn->next = jl_cn;
4011                         }
4012                         last_cn = jl_cn;
4013                         /* make sure the block we are trying to log is not a block
4014                            of journal or reserved area */
4015
4016                         if (is_block_in_log_or_reserved_area
4017                             (p_s_sb, cn->bh->b_blocknr)) {
4018                                 reiserfs_panic(p_s_sb,
4019                                                "journal-2332: Trying to log block %lu, which is a log block\n",
4020                                                cn->bh->b_blocknr);
4021                         }
4022                         jl_cn->blocknr = cn->bh->b_blocknr;
4023                         jl_cn->state = 0;
4024                         jl_cn->sb = p_s_sb;
4025                         jl_cn->bh = cn->bh;
4026                         jl_cn->jlist = jl;
4027                         insert_journal_hash(journal->j_list_hash_table, jl_cn);
4028                         if (i < trans_half) {
4029                                 desc->j_realblock[i] =
4030                                     cpu_to_le32(cn->bh->b_blocknr);
4031                         } else {
4032                                 commit->j_realblock[i - trans_half] =
4033                                     cpu_to_le32(cn->bh->b_blocknr);
4034                         }
4035                 } else {
4036                         i--;
4037                 }
4038         }
4039         set_desc_trans_len(desc, journal->j_len);
4040         set_desc_mount_id(desc, journal->j_mount_id);
4041         set_desc_trans_id(desc, journal->j_trans_id);
4042         set_commit_trans_len(commit, journal->j_len);
4043
4044         /* special check in case all buffers in the journal were marked for not logging */
4045         if (journal->j_len == 0) {
4046                 BUG();
4047         }
4048
4049         /* we're about to dirty all the log blocks, mark the description block
4050          * dirty now too.  Don't mark the commit block dirty until all the
4051          * others are on disk
4052          */
4053         mark_buffer_dirty(d_bh);
4054
4055         /* first data block is j_start + 1, so add one to cur_write_start wherever you use it */
4056         cur_write_start = journal->j_start;
4057         cn = journal->j_first;
4058         jindex = 1;             /* start at one so we don't get the desc again */
4059         while (cn) {
4060                 clear_buffer_journal_new(cn->bh);
4061                 /* copy all the real blocks into log area.  dirty log blocks */
4062                 if (buffer_journaled(cn->bh)) {
4063                         struct buffer_head *tmp_bh;
4064                         char *addr;
4065                         struct page *page;
4066                         tmp_bh =
4067                             journal_getblk(p_s_sb,
4068                                            SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
4069                                            ((cur_write_start +
4070                                              jindex) %
4071                                             SB_ONDISK_JOURNAL_SIZE(p_s_sb)));
4072                         set_buffer_uptodate(tmp_bh);
4073                         page = cn->bh->b_page;
4074                         addr = kmap(page);
4075                         memcpy(tmp_bh->b_data,
4076                                addr + offset_in_page(cn->bh->b_data),
4077                                cn->bh->b_size);
4078                         kunmap(page);
4079                         mark_buffer_dirty(tmp_bh);
4080                         jindex++;
4081                         set_buffer_journal_dirty(cn->bh);
4082                         clear_buffer_journaled(cn->bh);
4083                 } else {
4084                         /* JDirty cleared sometime during transaction.  don't log this one */
4085                         reiserfs_warning(p_s_sb,
4086                                          "journal-2048: do_journal_end: BAD, buffer in journal hash, but not JDirty!");
4087                         brelse(cn->bh);
4088                 }
4089                 next = cn->next;
4090                 free_cnode(p_s_sb, cn);
4091                 cn = next;
4092                 cond_resched();
4093         }
4094
4095         /* we are done  with both the c_bh and d_bh, but
4096          ** c_bh must be written after all other commit blocks,
4097          ** so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1.
4098          */
4099
4100         journal->j_current_jl = alloc_journal_list(p_s_sb);
4101
4102         /* now it is safe to insert this transaction on the main list */
4103         list_add_tail(&jl->j_list, &journal->j_journal_list);
4104         list_add_tail(&jl->j_working_list, &journal->j_working_list);
4105         journal->j_num_work_lists++;
4106
4107         /* reset journal values for the next transaction */
4108         old_start = journal->j_start;
4109         journal->j_start =
4110             (journal->j_start + journal->j_len +
4111              2) % SB_ONDISK_JOURNAL_SIZE(p_s_sb);
4112         atomic_set(&(journal->j_wcount), 0);
4113         journal->j_bcount = 0;
4114         journal->j_last = NULL;
4115         journal->j_first = NULL;
4116         journal->j_len = 0;
4117         journal->j_trans_start_time = 0;
4118         /* check for trans_id overflow */
4119         if (++journal->j_trans_id == 0)
4120                 journal->j_trans_id = 10;
4121         journal->j_current_jl->j_trans_id = journal->j_trans_id;
4122         journal->j_must_wait = 0;
4123         journal->j_len_alloc = 0;
4124         journal->j_next_full_flush = 0;
4125         journal->j_next_async_flush = 0;
4126         init_journal_hash(p_s_sb);
4127
4128         // make sure reiserfs_add_jh sees the new current_jl before we
4129         // write out the tails
4130         smp_mb();
4131
4132         /* tail conversion targets have to hit the disk before we end the
4133          * transaction.  Otherwise a later transaction might repack the tail
4134          * before this transaction commits, leaving the data block unflushed and
4135          * clean, if we crash before the later transaction commits, the data block
4136          * is lost.
4137          */
4138         if (!list_empty(&jl->j_tail_bh_list)) {
4139                 unlock_kernel();
4140                 write_ordered_buffers(&journal->j_dirty_buffers_lock,
4141                                       journal, jl, &jl->j_tail_bh_list);
4142                 lock_kernel();
4143         }
4144         if (!list_empty(&jl->j_tail_bh_list))
4145                 BUG();
4146         up(&jl->j_commit_lock);
4147
4148         /* honor the flush wishes from the caller, simple commits can
4149          ** be done outside the journal lock, they are done below
4150          **
4151          ** if we don't flush the commit list right now, we put it into
4152          ** the work queue so the people waiting on the async progress work
4153          ** queue don't wait for this proc to flush journal lists and such.
4154          */
4155         if (flush) {
4156                 flush_commit_list(p_s_sb, jl, 1);
4157                 flush_journal_list(p_s_sb, jl, 1);
4158         } else if (!(jl->j_state & LIST_COMMIT_PENDING))
4159                 queue_delayed_work(commit_wq, &journal->j_work, HZ / 10);
4160
4161         /* if the next transaction has any chance of wrapping, flush
4162          ** transactions that might get overwritten.  If any journal lists are very
4163          ** old flush them as well.
4164          */
4165       first_jl:
4166         list_for_each_safe(entry, safe, &journal->j_journal_list) {
4167                 temp_jl = JOURNAL_LIST_ENTRY(entry);
4168                 if (journal->j_start <= temp_jl->j_start) {
4169                         if ((journal->j_start + journal->j_trans_max + 1) >=
4170                             temp_jl->j_start) {
4171                                 flush_used_journal_lists(p_s_sb, temp_jl);
4172                                 goto first_jl;
4173                         } else if ((journal->j_start +
4174                                     journal->j_trans_max + 1) <
4175                                    SB_ONDISK_JOURNAL_SIZE(p_s_sb)) {
4176                                 /* if we don't cross into the next transaction and we don't
4177                                  * wrap, there is no way we can overlap any later transactions
4178                                  * break now
4179                                  */
4180                                 break;
4181                         }
4182                 } else if ((journal->j_start +
4183                             journal->j_trans_max + 1) >
4184                            SB_ONDISK_JOURNAL_SIZE(p_s_sb)) {
4185                         if (((journal->j_start + journal->j_trans_max + 1) %
4186                              SB_ONDISK_JOURNAL_SIZE(p_s_sb)) >=
4187                             temp_jl->j_start) {
4188                                 flush_used_journal_lists(p_s_sb, temp_jl);
4189                                 goto first_jl;
4190                         } else {
4191                                 /* we don't overlap anything from out start to the end of the
4192                                  * log, and our wrapped portion doesn't overlap anything at
4193                                  * the start of the log.  We can break
4194                                  */
4195                                 break;
4196                         }
4197                 }
4198         }
4199         flush_old_journal_lists(p_s_sb);
4200
4201         journal->j_current_jl->j_list_bitmap =
4202             get_list_bitmap(p_s_sb, journal->j_current_jl);
4203
4204         if (!(journal->j_current_jl->j_list_bitmap)) {
4205                 reiserfs_panic(p_s_sb,
4206                                "journal-1996: do_journal_end, could not get a list bitmap\n");
4207         }
4208
4209         atomic_set(&(journal->j_jlock), 0);
4210         unlock_journal(p_s_sb);
4211         /* wake up any body waiting to join. */
4212         clear_bit(J_WRITERS_QUEUED, &journal->j_state);
4213         wake_up(&(journal->j_join_wait));
4214
4215         if (!flush && wait_on_commit &&
4216             journal_list_still_alive(p_s_sb, commit_trans_id)) {
4217                 flush_commit_list(p_s_sb, jl, 1);
4218         }
4219       out:
4220         reiserfs_check_lock_depth(p_s_sb, "journal end2");
4221
4222         memset(th, 0, sizeof(*th));
4223         /* Re-set th->t_super, so we can properly keep track of how many
4224          * persistent transactions there are. We need to do this so if this
4225          * call is part of a failed restart_transaction, we can free it later */
4226         th->t_super = p_s_sb;
4227
4228         return journal->j_errno;
4229 }
4230
4231 static void __reiserfs_journal_abort_hard(struct super_block *sb)
4232 {
4233         struct reiserfs_journal *journal = SB_JOURNAL(sb);
4234         if (test_bit(J_ABORTED, &journal->j_state))
4235                 return;
4236
4237         printk(KERN_CRIT "REISERFS: Aborting journal for filesystem on %s\n",
4238                reiserfs_bdevname(sb));
4239
4240         sb->s_flags |= MS_RDONLY;
4241         set_bit(J_ABORTED, &journal->j_state);
4242
4243 #ifdef CONFIG_REISERFS_CHECK
4244         dump_stack();
4245 #endif
4246 }
4247
4248 static void __reiserfs_journal_abort_soft(struct super_block *sb, int errno)
4249 {
4250         struct reiserfs_journal *journal = SB_JOURNAL(sb);
4251         if (test_bit(J_ABORTED, &journal->j_state))
4252                 return;
4253
4254         if (!journal->j_errno)
4255                 journal->j_errno = errno;
4256
4257         __reiserfs_journal_abort_hard(sb);
4258 }
4259
4260 void reiserfs_journal_abort(struct super_block *sb, int errno)
4261 {
4262         return __reiserfs_journal_abort_soft(sb, errno);
4263 }