fs/reiserfs/journal.c

   1 /*
   2 ** Write ahead logging implementation copyright Chris Mason 2000
   3 **
   4 ** The background commits make this code very interelated, and
   5 ** overly complex.  I need to rethink things a bit....The major players:
   6 **
   7 ** journal_begin -- call with the number of blocks you expect to log.
   8 **                  If the current transaction is too
   9 **                  old, it will block until the current transaction is
  10 **                  finished, and then start a new one.
  11 **                  Usually, your transaction will get joined in with
  12 **                  previous ones for speed.
  13 **
  14 ** journal_join  -- same as journal_begin, but won't block on the current
  15 **                  transaction regardless of age.  Don't ever call
  16 **                  this.  Ever.  There are only two places it should be
  17 **                  called from, and they are both inside this file.
  18 **
  19 ** journal_mark_dirty -- adds blocks into this transaction.  clears any flags
  20 **                       that might make them get sent to disk
  21 **                       and then marks them BH_JDirty.  Puts the buffer head
  22 **                       into the current transaction hash.
  23 **
  24 ** journal_end -- if the current transaction is batchable, it does nothing
  25 **                   otherwise, it could do an async/synchronous commit, or
  26 **                   a full flush of all log and real blocks in the
  27 **                   transaction.
  28 **
  29 ** flush_old_commits -- if the current transaction is too old, it is ended and
  30 **                      commit blocks are sent to disk.  Forces commit blocks
  31 **                      to disk for all backgrounded commits that have been
  32 **                      around too long.
  33 **                   -- Note, if you call this as an immediate flush from
  34 **                      from within kupdate, it will ignore the immediate flag
  35 */
  36
  37 #include <linux/config.h>
  38 #include <asm/uaccess.h>
  39 #include <asm/system.h>
  40
  41 #include <linux/time.h>
  42 #include <asm/semaphore.h>
  43
  44 #include <linux/vmalloc.h>
  45 #include <linux/reiserfs_fs.h>
  46
  47 #include <linux/kernel.h>
  48 #include <linux/errno.h>
  49 #include <linux/fcntl.h>
  50 #include <linux/stat.h>
  51 #include <linux/string.h>
  52 #include <linux/smp_lock.h>
  53 #include <linux/buffer_head.h>
  54 #include <linux/workqueue.h>
  55 #include <linux/writeback.h>
  56 #include <linux/blkdev.h>
  57
  58 /* gets a struct reiserfs_journal_list * from a list head */
  59 #define JOURNAL_LIST_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
  60                                j_list))
  61 #define JOURNAL_WORK_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
  62                                j_working_list))
  63
  64 /* the number of mounted filesystems.  This is used to decide when to
  65 ** start and kill the commit workqueue
  66 */
  67 static int reiserfs_mounted_fs_count;
  68
  69 static struct workqueue_struct *commit_wq;
  70
  71 #define JOURNAL_TRANS_HALF 1018 /* must be correct to keep the desc and commit
  72                                    structs at 4k */
  73 #define BUFNR 64                /*read ahead */
  74
  75 /* cnode stat bits.  Move these into reiserfs_fs.h */
  76
  77 #define BLOCK_FREED 2           /* this block was freed, and can't be written.  */
  78 #define BLOCK_FREED_HOLDER 3    /* this block was freed during this transaction, and can't be written */
  79
  80 #define BLOCK_NEEDS_FLUSH 4     /* used in flush_journal_list */
  81 #define BLOCK_DIRTIED 5
  82
  83 /* journal list state bits */
  84 #define LIST_TOUCHED 1
  85 #define LIST_DIRTY   2
  86 #define LIST_COMMIT_PENDING  4  /* someone will commit this list */
  87
  88 /* flags for do_journal_end */
  89 #define FLUSH_ALL   1           /* flush commit and real blocks */
  90 #define COMMIT_NOW  2           /* end and commit this transaction */
  91 #define WAIT        4           /* wait for the log blocks to hit the disk */
  92
  93 static int do_journal_end(struct reiserfs_transaction_handle *,
  94                           struct super_block *, unsigned long nblocks,
  95                           int flags);
  96 static int flush_journal_list(struct super_block *s,
  97                               struct reiserfs_journal_list *jl, int flushall);
  98 static int flush_commit_list(struct super_block *s,
  99                              struct reiserfs_journal_list *jl, int flushall);
 100 static int can_dirty(struct reiserfs_journal_cnode *cn);
 101 static int journal_join(struct reiserfs_transaction_handle *th,
 102                         struct super_block *p_s_sb, unsigned long nblocks);
 103 static int release_journal_dev(struct super_block *super,
 104                                struct reiserfs_journal *journal);
 105 static int dirty_one_transaction(struct super_block *s,
 106                                  struct reiserfs_journal_list *jl);
 107 static void flush_async_commits(void *p);
 108 static void queue_log_writer(struct super_block *s);
 109
 110 /* values for join in do_journal_begin_r */
 111 enum {
 112         JBEGIN_REG = 0,         /* regular journal begin */
 113         JBEGIN_JOIN = 1,        /* join the running transaction if at all possible */
 114         JBEGIN_ABORT = 2,       /* called from cleanup code, ignores aborted flag */
 115 };
 116
 117 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
 118                               struct super_block *p_s_sb,
 119                               unsigned long nblocks, int join);
 120
 121 static void init_journal_hash(struct super_block *p_s_sb)
 122 {
 123         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 124         memset(journal->j_hash_table, 0,
 125                JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
 126 }
 127
 128 /*
 129 ** clears BH_Dirty and sticks the buffer on the clean list.  Called because I can't allow refile_buffer to
 130 ** make schedule happen after I've freed a block.  Look at remove_from_transaction and journal_mark_freed for
 131 ** more details.
 132 */
 133 static int reiserfs_clean_and_file_buffer(struct buffer_head *bh)
 134 {
 135         if (bh) {
 136                 clear_buffer_dirty(bh);
 137                 clear_buffer_journal_test(bh);
 138         }
 139         return 0;
 140 }
 141
 142 static void disable_barrier(struct super_block *s)
 143 {
 144         REISERFS_SB(s)->s_mount_opt &= ~(1 << REISERFS_BARRIER_FLUSH);
 145         printk("reiserfs: disabling flush barriers on %s\n",
 146                reiserfs_bdevname(s));
 147 }
 148
 149 static struct reiserfs_bitmap_node *allocate_bitmap_node(struct super_block
 150                                                          *p_s_sb)
 151 {
 152         struct reiserfs_bitmap_node *bn;
 153         static int id;
 154
 155         bn = kmalloc(sizeof(struct reiserfs_bitmap_node), GFP_NOFS);
 156         if (!bn) {
 157                 return NULL;
 158         }
 159         bn->data = kzalloc(p_s_sb->s_blocksize, GFP_NOFS);
 160         if (!bn->data) {
 161                 kfree(bn);
 162                 return NULL;
 163         }
 164         bn->id = id++;
 165         INIT_LIST_HEAD(&bn->list);
 166         return bn;
 167 }
 168
 169 static struct reiserfs_bitmap_node *get_bitmap_node(struct super_block *p_s_sb)
 170 {
 171         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 172         struct reiserfs_bitmap_node *bn = NULL;
 173         struct list_head *entry = journal->j_bitmap_nodes.next;
 174
 175         journal->j_used_bitmap_nodes++;
 176       repeat:
 177
 178         if (entry != &journal->j_bitmap_nodes) {
 179                 bn = list_entry(entry, struct reiserfs_bitmap_node, list);
 180                 list_del(entry);
 181                 memset(bn->data, 0, p_s_sb->s_blocksize);
 182                 journal->j_free_bitmap_nodes--;
 183                 return bn;
 184         }
 185         bn = allocate_bitmap_node(p_s_sb);
 186         if (!bn) {
 187                 yield();
 188                 goto repeat;
 189         }
 190         return bn;
 191 }
 192 static inline void free_bitmap_node(struct super_block *p_s_sb,
 193                                     struct reiserfs_bitmap_node *bn)
 194 {
 195         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 196         journal->j_used_bitmap_nodes--;
 197         if (journal->j_free_bitmap_nodes > REISERFS_MAX_BITMAP_NODES) {
 198                 kfree(bn->data);
 199                 kfree(bn);
 200         } else {
 201                 list_add(&bn->list, &journal->j_bitmap_nodes);
 202                 journal->j_free_bitmap_nodes++;
 203         }
 204 }
 205
 206 static void allocate_bitmap_nodes(struct super_block *p_s_sb)
 207 {
 208         int i;
 209         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 210         struct reiserfs_bitmap_node *bn = NULL;
 211         for (i = 0; i < REISERFS_MIN_BITMAP_NODES; i++) {
 212                 bn = allocate_bitmap_node(p_s_sb);
 213                 if (bn) {
 214                         list_add(&bn->list, &journal->j_bitmap_nodes);
 215                         journal->j_free_bitmap_nodes++;
 216                 } else {
 217                         break;  // this is ok, we'll try again when more are needed
 218                 }
 219         }
 220 }
 221
 222 static int set_bit_in_list_bitmap(struct super_block *p_s_sb, int block,
 223                                   struct reiserfs_list_bitmap *jb)
 224 {
 225         int bmap_nr = block / (p_s_sb->s_blocksize << 3);
 226         int bit_nr = block % (p_s_sb->s_blocksize << 3);
 227
 228         if (!jb->bitmaps[bmap_nr]) {
 229                 jb->bitmaps[bmap_nr] = get_bitmap_node(p_s_sb);
 230         }
 231         set_bit(bit_nr, (unsigned long *)jb->bitmaps[bmap_nr]->data);
 232         return 0;
 233 }
 234
 235 static void cleanup_bitmap_list(struct super_block *p_s_sb,
 236                                 struct reiserfs_list_bitmap *jb)
 237 {
 238         int i;
 239         if (jb->bitmaps == NULL)
 240                 return;
 241
 242         for (i = 0; i < SB_BMAP_NR(p_s_sb); i++) {
 243                 if (jb->bitmaps[i]) {
 244                         free_bitmap_node(p_s_sb, jb->bitmaps[i]);
 245                         jb->bitmaps[i] = NULL;
 246                 }
 247         }
 248 }
 249
 250 /*
 251 ** only call this on FS unmount.
 252 */
 253 static int free_list_bitmaps(struct super_block *p_s_sb,
 254                              struct reiserfs_list_bitmap *jb_array)
 255 {
 256         int i;
 257         struct reiserfs_list_bitmap *jb;
 258         for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
 259                 jb = jb_array + i;
 260                 jb->journal_list = NULL;
 261                 cleanup_bitmap_list(p_s_sb, jb);
 262                 vfree(jb->bitmaps);
 263                 jb->bitmaps = NULL;
 264         }
 265         return 0;
 266 }
 267
 268 static int free_bitmap_nodes(struct super_block *p_s_sb)
 269 {
 270         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 271         struct list_head *next = journal->j_bitmap_nodes.next;
 272         struct reiserfs_bitmap_node *bn;
 273
 274         while (next != &journal->j_bitmap_nodes) {
 275                 bn = list_entry(next, struct reiserfs_bitmap_node, list);
 276                 list_del(next);
 277                 kfree(bn->data);
 278                 kfree(bn);
 279                 next = journal->j_bitmap_nodes.next;
 280                 journal->j_free_bitmap_nodes--;
 281         }
 282
 283         return 0;
 284 }
 285
 286 /*
 287 ** get memory for JOURNAL_NUM_BITMAPS worth of bitmaps.
 288 ** jb_array is the array to be filled in.
 289 */
 290 int reiserfs_allocate_list_bitmaps(struct super_block *p_s_sb,
 291                                    struct reiserfs_list_bitmap *jb_array,
 292                                    int bmap_nr)
 293 {
 294         int i;
 295         int failed = 0;
 296         struct reiserfs_list_bitmap *jb;
 297         int mem = bmap_nr * sizeof(struct reiserfs_bitmap_node *);
 298
 299         for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
 300                 jb = jb_array + i;
 301                 jb->journal_list = NULL;
 302                 jb->bitmaps = vmalloc(mem);
 303                 if (!jb->bitmaps) {
 304                         reiserfs_warning(p_s_sb,
 305                                          "clm-2000, unable to allocate bitmaps for journal lists");
 306                         failed = 1;
 307                         break;
 308                 }
 309                 memset(jb->bitmaps, 0, mem);
 310         }
 311         if (failed) {
 312                 free_list_bitmaps(p_s_sb, jb_array);
 313                 return -1;
 314         }
 315         return 0;
 316 }
 317
 318 /*
 319 ** find an available list bitmap.  If you can't find one, flush a commit list
 320 ** and try again
 321 */
 322 static struct reiserfs_list_bitmap *get_list_bitmap(struct super_block *p_s_sb,
 323                                                     struct reiserfs_journal_list
 324                                                     *jl)
 325 {
 326         int i, j;
 327         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 328         struct reiserfs_list_bitmap *jb = NULL;
 329
 330         for (j = 0; j < (JOURNAL_NUM_BITMAPS * 3); j++) {
 331                 i = journal->j_list_bitmap_index;
 332                 journal->j_list_bitmap_index = (i + 1) % JOURNAL_NUM_BITMAPS;
 333                 jb = journal->j_list_bitmap + i;
 334                 if (journal->j_list_bitmap[i].journal_list) {
 335                         flush_commit_list(p_s_sb,
 336                                           journal->j_list_bitmap[i].
 337                                           journal_list, 1);
 338                         if (!journal->j_list_bitmap[i].journal_list) {
 339                                 break;
 340                         }
 341                 } else {
 342                         break;
 343                 }
 344         }
 345         if (jb->journal_list) { /* double check to make sure if flushed correctly */
 346                 return NULL;
 347         }
 348         jb->journal_list = jl;
 349         return jb;
 350 }
 351
 352 /*
 353 ** allocates a new chunk of X nodes, and links them all together as a list.
 354 ** Uses the cnode->next and cnode->prev pointers
 355 ** returns NULL on failure
 356 */
 357 static struct reiserfs_journal_cnode *allocate_cnodes(int num_cnodes)
 358 {
 359         struct reiserfs_journal_cnode *head;
 360         int i;
 361         if (num_cnodes <= 0) {
 362                 return NULL;
 363         }
 364         head = vmalloc(num_cnodes * sizeof(struct reiserfs_journal_cnode));
 365         if (!head) {
 366                 return NULL;
 367         }
 368         memset(head, 0, num_cnodes * sizeof(struct reiserfs_journal_cnode));
 369         head[0].prev = NULL;
 370         head[0].next = head + 1;
 371         for (i = 1; i < num_cnodes; i++) {
 372                 head[i].prev = head + (i - 1);
 373                 head[i].next = head + (i + 1);  /* if last one, overwrite it after the if */
 374         }
 375         head[num_cnodes - 1].next = NULL;
 376         return head;
 377 }
 378
 379 /*
 380 ** pulls a cnode off the free list, or returns NULL on failure
 381 */
 382 static struct reiserfs_journal_cnode *get_cnode(struct super_block *p_s_sb)
 383 {
 384         struct reiserfs_journal_cnode *cn;
 385         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 386
 387         reiserfs_check_lock_depth(p_s_sb, "get_cnode");
 388
 389         if (journal->j_cnode_free <= 0) {
 390                 return NULL;
 391         }
 392         journal->j_cnode_used++;
 393         journal->j_cnode_free--;
 394         cn = journal->j_cnode_free_list;
 395         if (!cn) {
 396                 return cn;
 397         }
 398         if (cn->next) {
 399                 cn->next->prev = NULL;
 400         }
 401         journal->j_cnode_free_list = cn->next;
 402         memset(cn, 0, sizeof(struct reiserfs_journal_cnode));
 403         return cn;
 404 }
 405
 406 /*
 407 ** returns a cnode to the free list
 408 */
 409 static void free_cnode(struct super_block *p_s_sb,
 410                        struct reiserfs_journal_cnode *cn)
 411 {
 412         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 413
 414         reiserfs_check_lock_depth(p_s_sb, "free_cnode");
 415
 416         journal->j_cnode_used--;
 417         journal->j_cnode_free++;
 418         /* memset(cn, 0, sizeof(struct reiserfs_journal_cnode)) ; */
 419         cn->next = journal->j_cnode_free_list;
 420         if (journal->j_cnode_free_list) {
 421                 journal->j_cnode_free_list->prev = cn;
 422         }
 423         cn->prev = NULL;        /* not needed with the memset, but I might kill the memset, and forget to do this */
 424         journal->j_cnode_free_list = cn;
 425 }
 426
 427 static void clear_prepared_bits(struct buffer_head *bh)
 428 {
 429         clear_buffer_journal_prepared(bh);
 430         clear_buffer_journal_restore_dirty(bh);
 431 }
 432
 433 /* utility function to force a BUG if it is called without the big
 434 ** kernel lock held.  caller is the string printed just before calling BUG()
 435 */
 436 void reiserfs_check_lock_depth(struct super_block *sb, char *caller)
 437 {
 438 #ifdef CONFIG_SMP
 439         if (current->lock_depth < 0) {
 440                 reiserfs_panic(sb, "%s called without kernel lock held",
 441                                caller);
 442         }
 443 #else
 444         ;
 445 #endif
 446 }
 447
 448 /* return a cnode with same dev, block number and size in table, or null if not found */
 449 static inline struct reiserfs_journal_cnode *get_journal_hash_dev(struct
 450                                                                   super_block
 451                                                                   *sb,
 452                                                                   struct
 453                                                                   reiserfs_journal_cnode
 454                                                                   **table,
 455                                                                   long bl)
 456 {
 457         struct reiserfs_journal_cnode *cn;
 458         cn = journal_hash(table, sb, bl);
 459         while (cn) {
 460                 if (cn->blocknr == bl && cn->sb == sb)
 461                         return cn;
 462                 cn = cn->hnext;
 463         }
 464         return (struct reiserfs_journal_cnode *)0;
 465 }
 466
 467 /*
 468 ** this actually means 'can this block be reallocated yet?'.  If you set search_all, a block can only be allocated
 469 ** if it is not in the current transaction, was not freed by the current transaction, and has no chance of ever
 470 ** being overwritten by a replay after crashing.
 471 **
 472 ** If you don't set search_all, a block can only be allocated if it is not in the current transaction.  Since deleting
 473 ** a block removes it from the current transaction, this case should never happen.  If you don't set search_all, make
 474 ** sure you never write the block without logging it.
 475 **
 476 ** next_zero_bit is a suggestion about the next block to try for find_forward.
 477 ** when bl is rejected because it is set in a journal list bitmap, we search
 478 ** for the next zero bit in the bitmap that rejected bl.  Then, we return that
 479 ** through next_zero_bit for find_forward to try.
 480 **
 481 ** Just because we return something in next_zero_bit does not mean we won't
 482 ** reject it on the next call to reiserfs_in_journal
 483 **
 484 */
 485 int reiserfs_in_journal(struct super_block *p_s_sb,
 486                         int bmap_nr, int bit_nr, int search_all,
 487                         b_blocknr_t * next_zero_bit)
 488 {
 489         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 490         struct reiserfs_journal_cnode *cn;
 491         struct reiserfs_list_bitmap *jb;
 492         int i;
 493         unsigned long bl;
 494
 495         *next_zero_bit = 0;     /* always start this at zero. */
 496
 497         PROC_INFO_INC(p_s_sb, journal.in_journal);
 498         /* If we aren't doing a search_all, this is a metablock, and it will be logged before use.
 499          ** if we crash before the transaction that freed it commits,  this transaction won't
 500          ** have committed either, and the block will never be written
 501          */
 502         if (search_all) {
 503                 for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
 504                         PROC_INFO_INC(p_s_sb, journal.in_journal_bitmap);
 505                         jb = journal->j_list_bitmap + i;
 506                         if (jb->journal_list && jb->bitmaps[bmap_nr] &&
 507                             test_bit(bit_nr,
 508                                      (unsigned long *)jb->bitmaps[bmap_nr]->
 509                                      data)) {
 510                                 *next_zero_bit =
 511                                     find_next_zero_bit((unsigned long *)
 512                                                        (jb->bitmaps[bmap_nr]->
 513                                                         data),
 514                                                        p_s_sb->s_blocksize << 3,
 515                                                        bit_nr + 1);
 516                                 return 1;
 517                         }
 518                 }
 519         }
 520
 521         bl = bmap_nr * (p_s_sb->s_blocksize << 3) + bit_nr;
 522         /* is it in any old transactions? */
 523         if (search_all
 524             && (cn =
 525                 get_journal_hash_dev(p_s_sb, journal->j_list_hash_table, bl))) {
 526                 return 1;
 527         }
 528
 529         /* is it in the current transaction.  This should never happen */
 530         if ((cn = get_journal_hash_dev(p_s_sb, journal->j_hash_table, bl))) {
 531                 BUG();
 532                 return 1;
 533         }
 534
 535         PROC_INFO_INC(p_s_sb, journal.in_journal_reusable);
 536         /* safe for reuse */
 537         return 0;
 538 }
 539
 540 /* insert cn into table
 541 */
 542 static inline void insert_journal_hash(struct reiserfs_journal_cnode **table,
 543                                        struct reiserfs_journal_cnode *cn)
 544 {
 545         struct reiserfs_journal_cnode *cn_orig;
 546
 547         cn_orig = journal_hash(table, cn->sb, cn->blocknr);
 548         cn->hnext = cn_orig;
 549         cn->hprev = NULL;
 550         if (cn_orig) {
 551                 cn_orig->hprev = cn;
 552         }
 553         journal_hash(table, cn->sb, cn->blocknr) = cn;
 554 }
 555
 556 /* lock the current transaction */
 557 static inline void lock_journal(struct super_block *p_s_sb)
 558 {
 559         PROC_INFO_INC(p_s_sb, journal.lock_journal);
 560         down(&SB_JOURNAL(p_s_sb)->j_lock);
 561 }
 562
 563 /* unlock the current transaction */
 564 static inline void unlock_journal(struct super_block *p_s_sb)
 565 {
 566         up(&SB_JOURNAL(p_s_sb)->j_lock);
 567 }
 568
 569 static inline void get_journal_list(struct reiserfs_journal_list *jl)
 570 {
 571         jl->j_refcount++;
 572 }
 573
 574 static inline void put_journal_list(struct super_block *s,
 575                                     struct reiserfs_journal_list *jl)
 576 {
 577         if (jl->j_refcount < 1) {
 578                 reiserfs_panic(s, "trans id %lu, refcount at %d",
 579                                jl->j_trans_id, jl->j_refcount);
 580         }
 581         if (--jl->j_refcount == 0)
 582                 kfree(jl);
 583 }
 584
 585 /*
 586 ** this used to be much more involved, and I'm keeping it just in case things get ugly again.
 587 ** it gets called by flush_commit_list, and cleans up any data stored about blocks freed during a
 588 ** transaction.
 589 */
 590 static void cleanup_freed_for_journal_list(struct super_block *p_s_sb,
 591                                            struct reiserfs_journal_list *jl)
 592 {
 593
 594         struct reiserfs_list_bitmap *jb = jl->j_list_bitmap;
 595         if (jb) {
 596                 cleanup_bitmap_list(p_s_sb, jb);
 597         }
 598         jl->j_list_bitmap->journal_list = NULL;
 599         jl->j_list_bitmap = NULL;
 600 }
 601
 602 static int journal_list_still_alive(struct super_block *s,
 603                                     unsigned long trans_id)
 604 {
 605         struct reiserfs_journal *journal = SB_JOURNAL(s);
 606         struct list_head *entry = &journal->j_journal_list;
 607         struct reiserfs_journal_list *jl;
 608
 609         if (!list_empty(entry)) {
 610                 jl = JOURNAL_LIST_ENTRY(entry->next);
 611                 if (jl->j_trans_id <= trans_id) {
 612                         return 1;
 613                 }
 614         }
 615         return 0;
 616 }
 617
 618 static void reiserfs_end_buffer_io_sync(struct buffer_head *bh, int uptodate)
 619 {
 620         char b[BDEVNAME_SIZE];
 621
 622         if (buffer_journaled(bh)) {
 623                 reiserfs_warning(NULL,
 624                                  "clm-2084: pinned buffer %lu:%s sent to disk",
 625                                  bh->b_blocknr, bdevname(bh->b_bdev, b));
 626         }
 627         if (uptodate)
 628                 set_buffer_uptodate(bh);
 629         else
 630                 clear_buffer_uptodate(bh);
 631         unlock_buffer(bh);
 632         put_bh(bh);
 633 }
 634
 635 static void reiserfs_end_ordered_io(struct buffer_head *bh, int uptodate)
 636 {
 637         if (uptodate)
 638                 set_buffer_uptodate(bh);
 639         else
 640                 clear_buffer_uptodate(bh);
 641         unlock_buffer(bh);
 642         put_bh(bh);
 643 }
 644
 645 static void submit_logged_buffer(struct buffer_head *bh)
 646 {
 647         get_bh(bh);
 648         bh->b_end_io = reiserfs_end_buffer_io_sync;
 649         clear_buffer_journal_new(bh);
 650         clear_buffer_dirty(bh);
 651         if (!test_clear_buffer_journal_test(bh))
 652                 BUG();
 653         if (!buffer_uptodate(bh))
 654                 BUG();
 655         submit_bh(WRITE, bh);
 656 }
 657
 658 static void submit_ordered_buffer(struct buffer_head *bh)
 659 {
 660         get_bh(bh);
 661         bh->b_end_io = reiserfs_end_ordered_io;
 662         clear_buffer_dirty(bh);
 663         if (!buffer_uptodate(bh))
 664                 BUG();
 665         submit_bh(WRITE, bh);
 666 }
 667
 668 static int submit_barrier_buffer(struct buffer_head *bh)
 669 {
 670         get_bh(bh);
 671         bh->b_end_io = reiserfs_end_ordered_io;
 672         clear_buffer_dirty(bh);
 673         if (!buffer_uptodate(bh))
 674                 BUG();
 675         return submit_bh(WRITE_BARRIER, bh);
 676 }
 677
 678 static void check_barrier_completion(struct super_block *s,
 679                                      struct buffer_head *bh)
 680 {
 681         if (buffer_eopnotsupp(bh)) {
 682                 clear_buffer_eopnotsupp(bh);
 683                 disable_barrier(s);
 684                 set_buffer_uptodate(bh);
 685                 set_buffer_dirty(bh);
 686                 sync_dirty_buffer(bh);
 687         }
 688 }
 689
 690 #define CHUNK_SIZE 32
 691 struct buffer_chunk {
 692         struct buffer_head *bh[CHUNK_SIZE];
 693         int nr;
 694 };
 695
 696 static void write_chunk(struct buffer_chunk *chunk)
 697 {
 698         int i;
 699         get_fs_excl();
 700         for (i = 0; i < chunk->nr; i++) {
 701                 submit_logged_buffer(chunk->bh[i]);
 702         }
 703         chunk->nr = 0;
 704         put_fs_excl();
 705 }
 706
 707 static void write_ordered_chunk(struct buffer_chunk *chunk)
 708 {
 709         int i;
 710         get_fs_excl();
 711         for (i = 0; i < chunk->nr; i++) {
 712                 submit_ordered_buffer(chunk->bh[i]);
 713         }
 714         chunk->nr = 0;
 715         put_fs_excl();
 716 }
 717
 718 static int add_to_chunk(struct buffer_chunk *chunk, struct buffer_head *bh,
 719                         spinlock_t * lock, void (fn) (struct buffer_chunk *))
 720 {
 721         int ret = 0;
 722         if (chunk->nr >= CHUNK_SIZE)
 723                 BUG();
 724         chunk->bh[chunk->nr++] = bh;
 725         if (chunk->nr >= CHUNK_SIZE) {
 726                 ret = 1;
 727                 if (lock)
 728                         spin_unlock(lock);
 729                 fn(chunk);
 730                 if (lock)
 731                         spin_lock(lock);
 732         }
 733         return ret;
 734 }
 735
 736 static atomic_t nr_reiserfs_jh = ATOMIC_INIT(0);
 737 static struct reiserfs_jh *alloc_jh(void)
 738 {
 739         struct reiserfs_jh *jh;
 740         while (1) {
 741                 jh = kmalloc(sizeof(*jh), GFP_NOFS);
 742                 if (jh) {
 743                         atomic_inc(&nr_reiserfs_jh);
 744                         return jh;
 745                 }
 746                 yield();
 747         }
 748 }
 749
 750 /*
 751  * we want to free the jh when the buffer has been written
 752  * and waited on
 753  */
 754 void reiserfs_free_jh(struct buffer_head *bh)
 755 {
 756         struct reiserfs_jh *jh;
 757
 758         jh = bh->b_private;
 759         if (jh) {
 760                 bh->b_private = NULL;
 761                 jh->bh = NULL;
 762                 list_del_init(&jh->list);
 763                 kfree(jh);
 764                 if (atomic_read(&nr_reiserfs_jh) <= 0)
 765                         BUG();
 766                 atomic_dec(&nr_reiserfs_jh);
 767                 put_bh(bh);
 768         }
 769 }
 770
 771 static inline int __add_jh(struct reiserfs_journal *j, struct buffer_head *bh,
 772                            int tail)
 773 {
 774         struct reiserfs_jh *jh;
 775
 776         if (bh->b_private) {
 777                 spin_lock(&j->j_dirty_buffers_lock);
 778                 if (!bh->b_private) {
 779                         spin_unlock(&j->j_dirty_buffers_lock);
 780                         goto no_jh;
 781                 }
 782                 jh = bh->b_private;
 783                 list_del_init(&jh->list);
 784         } else {
 785               no_jh:
 786                 get_bh(bh);
 787                 jh = alloc_jh();
 788                 spin_lock(&j->j_dirty_buffers_lock);
 789                 /* buffer must be locked for __add_jh, should be able to have
 790                  * two adds at the same time
 791                  */
 792                 if (bh->b_private)
 793                         BUG();
 794                 jh->bh = bh;
 795                 bh->b_private = jh;
 796         }
 797         jh->jl = j->j_current_jl;
 798         if (tail)
 799                 list_add_tail(&jh->list, &jh->jl->j_tail_bh_list);
 800         else {
 801                 list_add_tail(&jh->list, &jh->jl->j_bh_list);
 802         }
 803         spin_unlock(&j->j_dirty_buffers_lock);
 804         return 0;
 805 }
 806
 807 int reiserfs_add_tail_list(struct inode *inode, struct buffer_head *bh)
 808 {
 809         return __add_jh(SB_JOURNAL(inode->i_sb), bh, 1);
 810 }
 811 int reiserfs_add_ordered_list(struct inode *inode, struct buffer_head *bh)
 812 {
 813         return __add_jh(SB_JOURNAL(inode->i_sb), bh, 0);
 814 }
 815
 816 #define JH_ENTRY(l) list_entry((l), struct reiserfs_jh, list)
 817 static int write_ordered_buffers(spinlock_t * lock,
 818                                  struct reiserfs_journal *j,
 819                                  struct reiserfs_journal_list *jl,
 820                                  struct list_head *list)
 821 {
 822         struct buffer_head *bh;
 823         struct reiserfs_jh *jh;
 824         int ret = j->j_errno;
 825         struct buffer_chunk chunk;
 826         struct list_head tmp;
 827         INIT_LIST_HEAD(&tmp);
 828
 829         chunk.nr = 0;
 830         spin_lock(lock);
 831         while (!list_empty(list)) {
 832                 jh = JH_ENTRY(list->next);
 833                 bh = jh->bh;
 834                 get_bh(bh);
 835                 if (test_set_buffer_locked(bh)) {
 836                         if (!buffer_dirty(bh)) {
 837                                 list_del_init(&jh->list);
 838                                 list_add(&jh->list, &tmp);
 839                                 goto loop_next;
 840                         }
 841                         spin_unlock(lock);
 842                         if (chunk.nr)
 843                                 write_ordered_chunk(&chunk);
 844                         wait_on_buffer(bh);
 845                         cond_resched();
 846                         spin_lock(lock);
 847                         goto loop_next;
 848                 }
 849                 if (buffer_dirty(bh)) {
 850                         list_del_init(&jh->list);
 851                         list_add(&jh->list, &tmp);
 852                         add_to_chunk(&chunk, bh, lock, write_ordered_chunk);
 853                 } else {
 854                         reiserfs_free_jh(bh);
 855                         unlock_buffer(bh);
 856                 }
 857               loop_next:
 858                 put_bh(bh);
 859                 cond_resched_lock(lock);
 860         }
 861         if (chunk.nr) {
 862                 spin_unlock(lock);
 863                 write_ordered_chunk(&chunk);
 864                 spin_lock(lock);
 865         }
 866         while (!list_empty(&tmp)) {
 867                 jh = JH_ENTRY(tmp.prev);
 868                 bh = jh->bh;
 869                 get_bh(bh);
 870                 reiserfs_free_jh(bh);
 871
 872                 if (buffer_locked(bh)) {
 873                         spin_unlock(lock);
 874                         wait_on_buffer(bh);
 875                         spin_lock(lock);
 876                 }
 877                 if (!buffer_uptodate(bh)) {
 878                         ret = -EIO;
 879                 }
 880                 /* ugly interaction with invalidatepage here.
 881                  * reiserfs_invalidate_page will pin any buffer that has a valid
 882                  * journal head from an older transaction.  If someone else sets
 883                  * our buffer dirty after we write it in the first loop, and
 884                  * then someone truncates the page away, nobody will ever write
 885                  * the buffer. We're safe if we write the page one last time
 886                  * after freeing the journal header.
 887                  */
 888                 if (buffer_dirty(bh) && unlikely(bh->b_page->mapping == NULL)) {
 889                         spin_unlock(lock);
 890                         ll_rw_block(WRITE, 1, &bh);
 891                         spin_lock(lock);
 892                 }
 893                 put_bh(bh);
 894                 cond_resched_lock(lock);
 895         }
 896         spin_unlock(lock);
 897         return ret;
 898 }
 899
 900 static int flush_older_commits(struct super_block *s,
 901                                struct reiserfs_journal_list *jl)
 902 {
 903         struct reiserfs_journal *journal = SB_JOURNAL(s);
 904         struct reiserfs_journal_list *other_jl;
 905         struct reiserfs_journal_list *first_jl;
 906         struct list_head *entry;
 907         unsigned long trans_id = jl->j_trans_id;
 908         unsigned long other_trans_id;
 909         unsigned long first_trans_id;
 910
 911       find_first:
 912         /*
 913          * first we walk backwards to find the oldest uncommitted transation
 914          */
 915         first_jl = jl;
 916         entry = jl->j_list.prev;
 917         while (1) {
 918                 other_jl = JOURNAL_LIST_ENTRY(entry);
 919                 if (entry == &journal->j_journal_list ||
 920                     atomic_read(&other_jl->j_older_commits_done))
 921                         break;
 922
 923                 first_jl = other_jl;
 924                 entry = other_jl->j_list.prev;
 925         }
 926
 927         /* if we didn't find any older uncommitted transactions, return now */
 928         if (first_jl == jl) {
 929                 return 0;
 930         }
 931
 932         first_trans_id = first_jl->j_trans_id;
 933
 934         entry = &first_jl->j_list;
 935         while (1) {
 936                 other_jl = JOURNAL_LIST_ENTRY(entry);
 937                 other_trans_id = other_jl->j_trans_id;
 938
 939                 if (other_trans_id < trans_id) {
 940                         if (atomic_read(&other_jl->j_commit_left) != 0) {
 941                                 flush_commit_list(s, other_jl, 0);
 942
 943                                 /* list we were called with is gone, return */
 944                                 if (!journal_list_still_alive(s, trans_id))
 945                                         return 1;
 946
 947                                 /* the one we just flushed is gone, this means all
 948                                  * older lists are also gone, so first_jl is no longer
 949                                  * valid either.  Go back to the beginning.
 950                                  */
 951                                 if (!journal_list_still_alive
 952                                     (s, other_trans_id)) {
 953                                         goto find_first;
 954                                 }
 955                         }
 956                         entry = entry->next;
 957                         if (entry == &journal->j_journal_list)
 958                                 return 0;
 959                 } else {
 960                         return 0;
 961                 }
 962         }
 963         return 0;
 964 }
 965 int reiserfs_async_progress_wait(struct super_block *s)
 966 {
 967         DEFINE_WAIT(wait);
 968         struct reiserfs_journal *j = SB_JOURNAL(s);
 969         if (atomic_read(&j->j_async_throttle))
 970                 blk_congestion_wait(WRITE, HZ / 10);
 971         return 0;
 972 }
 973
 974 /*
 975 ** if this journal list still has commit blocks unflushed, send them to disk.
 976 **
 977 ** log areas must be flushed in order (transaction 2 can't commit before transaction 1)
 978 ** Before the commit block can by written, every other log block must be safely on disk
 979 **
 980 */
 981 static int flush_commit_list(struct super_block *s,
 982                              struct reiserfs_journal_list *jl, int flushall)
 983 {
 984         int i;
 985         int bn;
 986         struct buffer_head *tbh = NULL;
 987         unsigned long trans_id = jl->j_trans_id;
 988         struct reiserfs_journal *journal = SB_JOURNAL(s);
 989         int barrier = 0;
 990         int retval = 0;
 991
 992         reiserfs_check_lock_depth(s, "flush_commit_list");
 993
 994         if (atomic_read(&jl->j_older_commits_done)) {
 995                 return 0;
 996         }
 997
 998         get_fs_excl();
 999
1000         /* before we can put our commit blocks on disk, we have to make sure everyone older than
1001          ** us is on disk too
1002          */
1003         BUG_ON(jl->j_len <= 0);
1004         BUG_ON(trans_id == journal->j_trans_id);
1005
1006         get_journal_list(jl);
1007         if (flushall) {
1008                 if (flush_older_commits(s, jl) == 1) {
1009                         /* list disappeared during flush_older_commits.  return */
1010                         goto put_jl;
1011                 }
1012         }
1013
1014         /* make sure nobody is trying to flush this one at the same time */
1015         down(&jl->j_commit_lock);
1016         if (!journal_list_still_alive(s, trans_id)) {
1017                 up(&jl->j_commit_lock);
1018                 goto put_jl;
1019         }
1020         BUG_ON(jl->j_trans_id == 0);
1021
1022         /* this commit is done, exit */
1023         if (atomic_read(&(jl->j_commit_left)) <= 0) {
1024                 if (flushall) {
1025                         atomic_set(&(jl->j_older_commits_done), 1);
1026                 }
1027                 up(&jl->j_commit_lock);
1028                 goto put_jl;
1029         }
1030
1031         if (!list_empty(&jl->j_bh_list)) {
1032                 unlock_kernel();
1033                 write_ordered_buffers(&journal->j_dirty_buffers_lock,
1034                                       journal, jl, &jl->j_bh_list);
1035                 lock_kernel();
1036         }
1037         BUG_ON(!list_empty(&jl->j_bh_list));
1038         /*
1039          * for the description block and all the log blocks, submit any buffers
1040          * that haven't already reached the disk
1041          */
1042         atomic_inc(&journal->j_async_throttle);
1043         for (i = 0; i < (jl->j_len + 1); i++) {
1044                 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + (jl->j_start + i) %
1045                     SB_ONDISK_JOURNAL_SIZE(s);
1046                 tbh = journal_find_get_block(s, bn);
1047                 if (buffer_dirty(tbh))  /* redundant, ll_rw_block() checks */
1048                         ll_rw_block(SWRITE, 1, &tbh);
1049                 put_bh(tbh);
1050         }
1051         atomic_dec(&journal->j_async_throttle);
1052
1053         /* We're skipping the commit if there's an error */
1054         if (retval || reiserfs_is_journal_aborted(journal))
1055                 barrier = 0;
1056
1057         /* wait on everything written so far before writing the commit
1058          * if we are in barrier mode, send the commit down now
1059          */
1060         barrier = reiserfs_barrier_flush(s);
1061         if (barrier) {
1062                 int ret;
1063                 lock_buffer(jl->j_commit_bh);
1064                 ret = submit_barrier_buffer(jl->j_commit_bh);
1065                 if (ret == -EOPNOTSUPP) {
1066                         set_buffer_uptodate(jl->j_commit_bh);
1067                         disable_barrier(s);
1068                         barrier = 0;
1069                 }
1070         }
1071         for (i = 0; i < (jl->j_len + 1); i++) {
1072                 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) +
1073                     (jl->j_start + i) % SB_ONDISK_JOURNAL_SIZE(s);
1074                 tbh = journal_find_get_block(s, bn);
1075                 wait_on_buffer(tbh);
1076                 // since we're using ll_rw_blk above, it might have skipped over
1077                 // a locked buffer.  Double check here
1078                 //
1079                 if (buffer_dirty(tbh))  /* redundant, sync_dirty_buffer() checks */
1080                         sync_dirty_buffer(tbh);
1081                 if (unlikely(!buffer_uptodate(tbh))) {
1082 #ifdef CONFIG_REISERFS_CHECK
1083                         reiserfs_warning(s, "journal-601, buffer write failed");
1084 #endif
1085                         retval = -EIO;
1086                 }
1087                 put_bh(tbh);    /* once for journal_find_get_block */
1088                 put_bh(tbh);    /* once due to original getblk in do_journal_end */
1089                 atomic_dec(&(jl->j_commit_left));
1090         }
1091
1092         BUG_ON(atomic_read(&(jl->j_commit_left)) != 1);
1093
1094         if (!barrier) {
1095                 /* If there was a write error in the journal - we can't commit
1096                  * this transaction - it will be invalid and, if successful,
1097                  * will just end up propogating the write error out to
1098                  * the file system. */
1099                 if (likely(!retval && !reiserfs_is_journal_aborted (journal))) {
1100                         if (buffer_dirty(jl->j_commit_bh))
1101                                 BUG();
1102                         mark_buffer_dirty(jl->j_commit_bh) ;
1103                         sync_dirty_buffer(jl->j_commit_bh) ;
1104                 }
1105         } else
1106                 wait_on_buffer(jl->j_commit_bh);
1107
1108         check_barrier_completion(s, jl->j_commit_bh);
1109
1110         /* If there was a write error in the journal - we can't commit this
1111          * transaction - it will be invalid and, if successful, will just end
1112          * up propogating the write error out to the filesystem. */
1113         if (unlikely(!buffer_uptodate(jl->j_commit_bh))) {
1114 #ifdef CONFIG_REISERFS_CHECK
1115                 reiserfs_warning(s, "journal-615: buffer write failed");
1116 #endif
1117                 retval = -EIO;
1118         }
1119         bforget(jl->j_commit_bh);
1120         if (journal->j_last_commit_id != 0 &&
1121             (jl->j_trans_id - journal->j_last_commit_id) != 1) {
1122                 reiserfs_warning(s, "clm-2200: last commit %lu, current %lu",
1123                                  journal->j_last_commit_id, jl->j_trans_id);
1124         }
1125         journal->j_last_commit_id = jl->j_trans_id;
1126
1127         /* now, every commit block is on the disk.  It is safe to allow blocks freed during this transaction to be reallocated */
1128         cleanup_freed_for_journal_list(s, jl);
1129
1130         retval = retval ? retval : journal->j_errno;
1131
1132         /* mark the metadata dirty */
1133         if (!retval)
1134                 dirty_one_transaction(s, jl);
1135         atomic_dec(&(jl->j_commit_left));
1136
1137         if (flushall) {
1138                 atomic_set(&(jl->j_older_commits_done), 1);
1139         }
1140         up(&jl->j_commit_lock);
1141       put_jl:
1142         put_journal_list(s, jl);
1143
1144         if (retval)
1145                 reiserfs_abort(s, retval, "Journal write error in %s",
1146                                __FUNCTION__);
1147         put_fs_excl();
1148         return retval;
1149 }
1150
1151 /*
1152 ** flush_journal_list frequently needs to find a newer transaction for a given block.  This does that, or
1153 ** returns NULL if it can't find anything
1154 */
1155 static struct reiserfs_journal_list *find_newer_jl_for_cn(struct
1156                                                           reiserfs_journal_cnode
1157                                                           *cn)
1158 {
1159         struct super_block *sb = cn->sb;
1160         b_blocknr_t blocknr = cn->blocknr;
1161
1162         cn = cn->hprev;
1163         while (cn) {
1164                 if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist) {
1165                         return cn->jlist;
1166                 }
1167                 cn = cn->hprev;
1168         }
1169         return NULL;
1170 }
1171
1172 static void remove_journal_hash(struct super_block *,
1173                                 struct reiserfs_journal_cnode **,
1174                                 struct reiserfs_journal_list *, unsigned long,
1175                                 int);
1176
1177 /*
1178 ** once all the real blocks have been flushed, it is safe to remove them from the
1179 ** journal list for this transaction.  Aside from freeing the cnode, this also allows the
1180 ** block to be reallocated for data blocks if it had been deleted.
1181 */
1182 static void remove_all_from_journal_list(struct super_block *p_s_sb,
1183                                          struct reiserfs_journal_list *jl,
1184                                          int debug)
1185 {
1186         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1187         struct reiserfs_journal_cnode *cn, *last;
1188         cn = jl->j_realblock;
1189
1190         /* which is better, to lock once around the whole loop, or
1191          ** to lock for each call to remove_journal_hash?
1192          */
1193         while (cn) {
1194                 if (cn->blocknr != 0) {
1195                         if (debug) {
1196                                 reiserfs_warning(p_s_sb,
1197                                                  "block %u, bh is %d, state %ld",
1198                                                  cn->blocknr, cn->bh ? 1 : 0,
1199                                                  cn->state);
1200                         }
1201                         cn->state = 0;
1202                         remove_journal_hash(p_s_sb, journal->j_list_hash_table,
1203                                             jl, cn->blocknr, 1);
1204                 }
1205                 last = cn;
1206                 cn = cn->next;
1207                 free_cnode(p_s_sb, last);
1208         }
1209         jl->j_realblock = NULL;
1210 }
1211
1212 /*
1213 ** if this timestamp is greater than the timestamp we wrote last to the header block, write it to the header block.
1214 ** once this is done, I can safely say the log area for this transaction won't ever be replayed, and I can start
1215 ** releasing blocks in this transaction for reuse as data blocks.
1216 ** called by flush_journal_list, before it calls remove_all_from_journal_list
1217 **
1218 */
1219 static int _update_journal_header_block(struct super_block *p_s_sb,
1220                                         unsigned long offset,
1221                                         unsigned long trans_id)
1222 {
1223         struct reiserfs_journal_header *jh;
1224         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1225
1226         if (reiserfs_is_journal_aborted(journal))
1227                 return -EIO;
1228
1229         if (trans_id >= journal->j_last_flush_trans_id) {
1230                 if (buffer_locked((journal->j_header_bh))) {
1231                         wait_on_buffer((journal->j_header_bh));
1232                         if (unlikely(!buffer_uptodate(journal->j_header_bh))) {
1233 #ifdef CONFIG_REISERFS_CHECK
1234                                 reiserfs_warning(p_s_sb,
1235                                                  "journal-699: buffer write failed");
1236 #endif
1237                                 return -EIO;
1238                         }
1239                 }
1240                 journal->j_last_flush_trans_id = trans_id;
1241                 journal->j_first_unflushed_offset = offset;
1242                 jh = (struct reiserfs_journal_header *)(journal->j_header_bh->
1243                                                         b_data);
1244                 jh->j_last_flush_trans_id = cpu_to_le32(trans_id);
1245                 jh->j_first_unflushed_offset = cpu_to_le32(offset);
1246                 jh->j_mount_id = cpu_to_le32(journal->j_mount_id);
1247
1248                 if (reiserfs_barrier_flush(p_s_sb)) {
1249                         int ret;
1250                         lock_buffer(journal->j_header_bh);
1251                         ret = submit_barrier_buffer(journal->j_header_bh);
1252                         if (ret == -EOPNOTSUPP) {
1253                                 set_buffer_uptodate(journal->j_header_bh);
1254                                 disable_barrier(p_s_sb);
1255                                 goto sync;
1256                         }
1257                         wait_on_buffer(journal->j_header_bh);
1258                         check_barrier_completion(p_s_sb, journal->j_header_bh);
1259                 } else {
1260                       sync:
1261                         set_buffer_dirty(journal->j_header_bh);
1262                         sync_dirty_buffer(journal->j_header_bh);
1263                 }
1264                 if (!buffer_uptodate(journal->j_header_bh)) {
1265                         reiserfs_warning(p_s_sb,
1266                                          "journal-837: IO error during journal replay");
1267                         return -EIO;
1268                 }
1269         }
1270         return 0;
1271 }
1272
1273 static int update_journal_header_block(struct super_block *p_s_sb,
1274                                        unsigned long offset,
1275                                        unsigned long trans_id)
1276 {
1277         return _update_journal_header_block(p_s_sb, offset, trans_id);
1278 }
1279
1280 /*
1281 ** flush any and all journal lists older than you are
1282 ** can only be called from flush_journal_list
1283 */
1284 static int flush_older_journal_lists(struct super_block *p_s_sb,
1285                                      struct reiserfs_journal_list *jl)
1286 {
1287         struct list_head *entry;
1288         struct reiserfs_journal_list *other_jl;
1289         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1290         unsigned long trans_id = jl->j_trans_id;
1291
1292         /* we know we are the only ones flushing things, no extra race
1293          * protection is required.
1294          */
1295       restart:
1296         entry = journal->j_journal_list.next;
1297         /* Did we wrap? */
1298         if (entry == &journal->j_journal_list)
1299                 return 0;
1300         other_jl = JOURNAL_LIST_ENTRY(entry);
1301         if (other_jl->j_trans_id < trans_id) {
1302                 BUG_ON(other_jl->j_refcount <= 0);
1303                 /* do not flush all */
1304                 flush_journal_list(p_s_sb, other_jl, 0);
1305
1306                 /* other_jl is now deleted from the list */
1307                 goto restart;
1308         }
1309         return 0;
1310 }
1311
1312 static void del_from_work_list(struct super_block *s,
1313                                struct reiserfs_journal_list *jl)
1314 {
1315         struct reiserfs_journal *journal = SB_JOURNAL(s);
1316         if (!list_empty(&jl->j_working_list)) {
1317                 list_del_init(&jl->j_working_list);
1318                 journal->j_num_work_lists--;
1319         }
1320 }
1321
1322 /* flush a journal list, both commit and real blocks
1323 **
1324 ** always set flushall to 1, unless you are calling from inside
1325 ** flush_journal_list
1326 **
1327 ** IMPORTANT.  This can only be called while there are no journal writers,
1328 ** and the journal is locked.  That means it can only be called from
1329 ** do_journal_end, or by journal_release
1330 */
1331 static int flush_journal_list(struct super_block *s,
1332                               struct reiserfs_journal_list *jl, int flushall)
1333 {
1334         struct reiserfs_journal_list *pjl;
1335         struct reiserfs_journal_cnode *cn, *last;
1336         int count;
1337         int was_jwait = 0;
1338         int was_dirty = 0;
1339         struct buffer_head *saved_bh;
1340         unsigned long j_len_saved = jl->j_len;
1341         struct reiserfs_journal *journal = SB_JOURNAL(s);
1342         int err = 0;
1343
1344         BUG_ON(j_len_saved <= 0);
1345
1346         if (atomic_read(&journal->j_wcount) != 0) {
1347                 reiserfs_warning(s,
1348                                  "clm-2048: flush_journal_list called with wcount %d",
1349                                  atomic_read(&journal->j_wcount));
1350         }
1351         BUG_ON(jl->j_trans_id == 0);
1352
1353         /* if flushall == 0, the lock is already held */
1354         if (flushall) {
1355                 down(&journal->j_flush_sem);
1356         } else if (!down_trylock(&journal->j_flush_sem)) {
1357                 BUG();
1358         }
1359
1360         count = 0;
1361         if (j_len_saved > journal->j_trans_max) {
1362                 reiserfs_panic(s,
1363                                "journal-715: flush_journal_list, length is %lu, trans id %lu\n",
1364                                j_len_saved, jl->j_trans_id);
1365                 return 0;
1366         }
1367
1368         get_fs_excl();
1369
1370         /* if all the work is already done, get out of here */
1371         if (atomic_read(&(jl->j_nonzerolen)) <= 0 &&
1372             atomic_read(&(jl->j_commit_left)) <= 0) {
1373                 goto flush_older_and_return;
1374         }
1375
1376         /* start by putting the commit list on disk.  This will also flush
1377          ** the commit lists of any olders transactions
1378          */
1379         flush_commit_list(s, jl, 1);
1380
1381         if (!(jl->j_state & LIST_DIRTY)
1382             && !reiserfs_is_journal_aborted(journal))
1383                 BUG();
1384
1385         /* are we done now? */
1386         if (atomic_read(&(jl->j_nonzerolen)) <= 0 &&
1387             atomic_read(&(jl->j_commit_left)) <= 0) {
1388                 goto flush_older_and_return;
1389         }
1390
1391         /* loop through each cnode, see if we need to write it,
1392          ** or wait on a more recent transaction, or just ignore it
1393          */
1394         if (atomic_read(&(journal->j_wcount)) != 0) {
1395                 reiserfs_panic(s,
1396                                "journal-844: panic journal list is flushing, wcount is not 0\n");
1397         }
1398         cn = jl->j_realblock;
1399         while (cn) {
1400                 was_jwait = 0;
1401                 was_dirty = 0;
1402                 saved_bh = NULL;
1403                 /* blocknr of 0 is no longer in the hash, ignore it */
1404                 if (cn->blocknr == 0) {
1405                         goto free_cnode;
1406                 }
1407
1408                 /* This transaction failed commit. Don't write out to the disk */
1409                 if (!(jl->j_state & LIST_DIRTY))
1410                         goto free_cnode;
1411
1412                 pjl = find_newer_jl_for_cn(cn);
1413                 /* the order is important here.  We check pjl to make sure we
1414                  ** don't clear BH_JDirty_wait if we aren't the one writing this
1415                  ** block to disk
1416                  */
1417                 if (!pjl && cn->bh) {
1418                         saved_bh = cn->bh;
1419
1420                         /* we do this to make sure nobody releases the buffer while
1421                          ** we are working with it
1422                          */
1423                         get_bh(saved_bh);
1424
1425                         if (buffer_journal_dirty(saved_bh)) {
1426                                 BUG_ON(!can_dirty(cn));
1427                                 was_jwait = 1;
1428                                 was_dirty = 1;
1429                         } else if (can_dirty(cn)) {
1430                                 /* everything with !pjl && jwait should be writable */
1431                                 BUG();
1432                         }
1433                 }
1434
1435                 /* if someone has this block in a newer transaction, just make
1436                  ** sure they are commited, and don't try writing it to disk
1437                  */
1438                 if (pjl) {
1439                         if (atomic_read(&pjl->j_commit_left))
1440                                 flush_commit_list(s, pjl, 1);
1441                         goto free_cnode;
1442                 }
1443
1444                 /* bh == NULL when the block got to disk on its own, OR,
1445                  ** the block got freed in a future transaction
1446                  */
1447                 if (saved_bh == NULL) {
1448                         goto free_cnode;
1449                 }
1450
1451                 /* this should never happen.  kupdate_one_transaction has this list
1452                  ** locked while it works, so we should never see a buffer here that
1453                  ** is not marked JDirty_wait
1454                  */
1455                 if ((!was_jwait) && !buffer_locked(saved_bh)) {
1456                         reiserfs_warning(s,
1457                                          "journal-813: BAD! buffer %llu %cdirty %cjwait, "
1458                                          "not in a newer tranasction",
1459                                          (unsigned long long)saved_bh->
1460                                          b_blocknr, was_dirty ? ' ' : '!',
1461                                          was_jwait ? ' ' : '!');
1462                 }
1463                 if (was_dirty) {
1464                         /* we inc again because saved_bh gets decremented at free_cnode */
1465                         get_bh(saved_bh);
1466                         set_bit(BLOCK_NEEDS_FLUSH, &cn->state);
1467                         lock_buffer(saved_bh);
1468                         BUG_ON(cn->blocknr != saved_bh->b_blocknr);
1469                         if (buffer_dirty(saved_bh))
1470                                 submit_logged_buffer(saved_bh);
1471                         else
1472                                 unlock_buffer(saved_bh);
1473                         count++;
1474                 } else {
1475                         reiserfs_warning(s,
1476                                          "clm-2082: Unable to flush buffer %llu in %s",
1477                                          (unsigned long long)saved_bh->
1478                                          b_blocknr, __FUNCTION__);
1479                 }
1480               free_cnode:
1481                 last = cn;
1482                 cn = cn->next;
1483                 if (saved_bh) {
1484                         /* we incremented this to keep others from taking the buffer head away */
1485                         put_bh(saved_bh);
1486                         if (atomic_read(&(saved_bh->b_count)) < 0) {
1487                                 reiserfs_warning(s,
1488                                                  "journal-945: saved_bh->b_count < 0");
1489                         }
1490                 }
1491         }
1492         if (count > 0) {
1493                 cn = jl->j_realblock;
1494                 while (cn) {
1495                         if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) {
1496                                 if (!cn->bh) {
1497                                         reiserfs_panic(s,
1498                                                        "journal-1011: cn->bh is NULL\n");
1499                                 }
1500                                 wait_on_buffer(cn->bh);
1501                                 if (!cn->bh) {
1502                                         reiserfs_panic(s,
1503                                                        "journal-1012: cn->bh is NULL\n");
1504                                 }
1505                                 if (unlikely(!buffer_uptodate(cn->bh))) {
1506 #ifdef CONFIG_REISERFS_CHECK
1507                                         reiserfs_warning(s,
1508                                                          "journal-949: buffer write failed\n");
1509 #endif
1510                                         err = -EIO;
1511                                 }
1512                                 /* note, we must clear the JDirty_wait bit after the up to date
1513                                  ** check, otherwise we race against our flushpage routine
1514                                  */
1515                                 BUG_ON(!test_clear_buffer_journal_dirty
1516                                        (cn->bh));
1517
1518                                 /* undo the inc from journal_mark_dirty */
1519                                 put_bh(cn->bh);
1520                                 brelse(cn->bh);
1521                         }
1522                         cn = cn->next;
1523                 }
1524         }
1525
1526         if (err)
1527                 reiserfs_abort(s, -EIO,
1528                                "Write error while pushing transaction to disk in %s",
1529                                __FUNCTION__);
1530       flush_older_and_return:
1531
1532         /* before we can update the journal header block, we _must_ flush all
1533          ** real blocks from all older transactions to disk.  This is because
1534          ** once the header block is updated, this transaction will not be
1535          ** replayed after a crash
1536          */
1537         if (flushall) {
1538                 flush_older_journal_lists(s, jl);
1539         }
1540
1541         err = journal->j_errno;
1542         /* before we can remove everything from the hash tables for this
1543          ** transaction, we must make sure it can never be replayed
1544          **
1545          ** since we are only called from do_journal_end, we know for sure there
1546          ** are no allocations going on while we are flushing journal lists.  So,
1547          ** we only need to update the journal header block for the last list
1548          ** being flushed
1549          */
1550         if (!err && flushall) {
1551                 err =
1552                     update_journal_header_block(s,
1553                                                 (jl->j_start + jl->j_len +
1554                                                  2) % SB_ONDISK_JOURNAL_SIZE(s),
1555                                                 jl->j_trans_id);
1556                 if (err)
1557                         reiserfs_abort(s, -EIO,
1558                                        "Write error while updating journal header in %s",
1559                                        __FUNCTION__);
1560         }
1561         remove_all_from_journal_list(s, jl, 0);
1562         list_del_init(&jl->j_list);
1563         journal->j_num_lists--;
1564         del_from_work_list(s, jl);
1565
1566         if (journal->j_last_flush_id != 0 &&
1567             (jl->j_trans_id - journal->j_last_flush_id) != 1) {
1568                 reiserfs_warning(s, "clm-2201: last flush %lu, current %lu",
1569                                  journal->j_last_flush_id, jl->j_trans_id);
1570         }
1571         journal->j_last_flush_id = jl->j_trans_id;
1572
1573         /* not strictly required since we are freeing the list, but it should
1574          * help find code using dead lists later on
1575          */
1576         jl->j_len = 0;
1577         atomic_set(&(jl->j_nonzerolen), 0);
1578         jl->j_start = 0;
1579         jl->j_realblock = NULL;
1580         jl->j_commit_bh = NULL;
1581         jl->j_trans_id = 0;
1582         jl->j_state = 0;
1583         put_journal_list(s, jl);
1584         if (flushall)
1585                 up(&journal->j_flush_sem);
1586         put_fs_excl();
1587         return err;
1588 }
1589
1590 static int write_one_transaction(struct super_block *s,
1591                                  struct reiserfs_journal_list *jl,
1592                                  struct buffer_chunk *chunk)
1593 {
1594         struct reiserfs_journal_cnode *cn;
1595         int ret = 0;
1596
1597         jl->j_state |= LIST_TOUCHED;
1598         del_from_work_list(s, jl);
1599         if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0) {
1600                 return 0;
1601         }
1602
1603         cn = jl->j_realblock;
1604         while (cn) {
1605                 /* if the blocknr == 0, this has been cleared from the hash,
1606                  ** skip it
1607                  */
1608                 if (cn->blocknr == 0) {
1609                         goto next;
1610                 }
1611                 if (cn->bh && can_dirty(cn) && buffer_dirty(cn->bh)) {
1612                         struct buffer_head *tmp_bh;
1613                         /* we can race against journal_mark_freed when we try
1614                          * to lock_buffer(cn->bh), so we have to inc the buffer
1615                          * count, and recheck things after locking
1616                          */
1617                         tmp_bh = cn->bh;
1618                         get_bh(tmp_bh);
1619                         lock_buffer(tmp_bh);
1620                         if (cn->bh && can_dirty(cn) && buffer_dirty(tmp_bh)) {
1621                                 if (!buffer_journal_dirty(tmp_bh) ||
1622                                     buffer_journal_prepared(tmp_bh))
1623                                         BUG();
1624                                 add_to_chunk(chunk, tmp_bh, NULL, write_chunk);
1625                                 ret++;
1626                         } else {
1627                                 /* note, cn->bh might be null now */
1628                                 unlock_buffer(tmp_bh);
1629                         }
1630                         put_bh(tmp_bh);
1631                 }
1632               next:
1633                 cn = cn->next;
1634                 cond_resched();
1635         }
1636         return ret;
1637 }
1638
1639 /* used by flush_commit_list */
1640 static int dirty_one_transaction(struct super_block *s,
1641                                  struct reiserfs_journal_list *jl)
1642 {
1643         struct reiserfs_journal_cnode *cn;
1644         struct reiserfs_journal_list *pjl;
1645         int ret = 0;
1646
1647         jl->j_state |= LIST_DIRTY;
1648         cn = jl->j_realblock;
1649         while (cn) {
1650                 /* look for a more recent transaction that logged this
1651                  ** buffer.  Only the most recent transaction with a buffer in
1652                  ** it is allowed to send that buffer to disk
1653                  */
1654                 pjl = find_newer_jl_for_cn(cn);
1655                 if (!pjl && cn->blocknr && cn->bh
1656                     && buffer_journal_dirty(cn->bh)) {
1657                         BUG_ON(!can_dirty(cn));
1658                         /* if the buffer is prepared, it will either be logged
1659                          * or restored.  If restored, we need to make sure
1660                          * it actually gets marked dirty
1661                          */
1662                         clear_buffer_journal_new(cn->bh);
1663                         if (buffer_journal_prepared(cn->bh)) {
1664                                 set_buffer_journal_restore_dirty(cn->bh);
1665                         } else {
1666                                 set_buffer_journal_test(cn->bh);
1667                                 mark_buffer_dirty(cn->bh);
1668                         }
1669                 }
1670                 cn = cn->next;
1671         }
1672         return ret;
1673 }
1674
1675 static int kupdate_transactions(struct super_block *s,
1676                                 struct reiserfs_journal_list *jl,
1677                                 struct reiserfs_journal_list **next_jl,
1678                                 unsigned long *next_trans_id,
1679                                 int num_blocks, int num_trans)
1680 {
1681         int ret = 0;
1682         int written = 0;
1683         int transactions_flushed = 0;
1684         unsigned long orig_trans_id = jl->j_trans_id;
1685         struct buffer_chunk chunk;
1686         struct list_head *entry;
1687         struct reiserfs_journal *journal = SB_JOURNAL(s);
1688         chunk.nr = 0;
1689
1690         down(&journal->j_flush_sem);
1691         if (!journal_list_still_alive(s, orig_trans_id)) {
1692                 goto done;
1693         }
1694
1695         /* we've got j_flush_sem held, nobody is going to delete any
1696          * of these lists out from underneath us
1697          */
1698         while ((num_trans && transactions_flushed < num_trans) ||
1699                (!num_trans && written < num_blocks)) {
1700
1701                 if (jl->j_len == 0 || (jl->j_state & LIST_TOUCHED) ||
1702                     atomic_read(&jl->j_commit_left)
1703                     || !(jl->j_state & LIST_DIRTY)) {
1704                         del_from_work_list(s, jl);
1705                         break;
1706                 }
1707                 ret = write_one_transaction(s, jl, &chunk);
1708
1709                 if (ret < 0)
1710                         goto done;
1711                 transactions_flushed++;
1712                 written += ret;
1713                 entry = jl->j_list.next;
1714
1715                 /* did we wrap? */
1716                 if (entry == &journal->j_journal_list) {
1717                         break;
1718                 }
1719                 jl = JOURNAL_LIST_ENTRY(entry);
1720
1721                 /* don't bother with older transactions */
1722                 if (jl->j_trans_id <= orig_trans_id)
1723                         break;
1724         }
1725         if (chunk.nr) {
1726                 write_chunk(&chunk);
1727         }
1728
1729       done:
1730         up(&journal->j_flush_sem);
1731         return ret;
1732 }
1733
1734 /* for o_sync and fsync heavy applications, they tend to use
1735 ** all the journa list slots with tiny transactions.  These
1736 ** trigger lots and lots of calls to update the header block, which
1737 ** adds seeks and slows things down.
1738 **
1739 ** This function tries to clear out a large chunk of the journal lists
1740 ** at once, which makes everything faster since only the newest journal
1741 ** list updates the header block
1742 */
1743 static int flush_used_journal_lists(struct super_block *s,
1744                                     struct reiserfs_journal_list *jl)
1745 {
1746         unsigned long len = 0;
1747         unsigned long cur_len;
1748         int ret;
1749         int i;
1750         int limit = 256;
1751         struct reiserfs_journal_list *tjl;
1752         struct reiserfs_journal_list *flush_jl;
1753         unsigned long trans_id;
1754         struct reiserfs_journal *journal = SB_JOURNAL(s);
1755
1756         flush_jl = tjl = jl;
1757
1758         /* in data logging mode, try harder to flush a lot of blocks */
1759         if (reiserfs_data_log(s))
1760                 limit = 1024;
1761         /* flush for 256 transactions or limit blocks, whichever comes first */
1762         for (i = 0; i < 256 && len < limit; i++) {
1763                 if (atomic_read(&tjl->j_commit_left) ||
1764                     tjl->j_trans_id < jl->j_trans_id) {
1765                         break;
1766                 }
1767                 cur_len = atomic_read(&tjl->j_nonzerolen);
1768                 if (cur_len > 0) {
1769                         tjl->j_state &= ~LIST_TOUCHED;
1770                 }
1771                 len += cur_len;
1772                 flush_jl = tjl;
1773                 if (tjl->j_list.next == &journal->j_journal_list)
1774                         break;
1775                 tjl = JOURNAL_LIST_ENTRY(tjl->j_list.next);
1776         }
1777         /* try to find a group of blocks we can flush across all the
1778          ** transactions, but only bother if we've actually spanned
1779          ** across multiple lists
1780          */
1781         if (flush_jl != jl) {
1782                 ret = kupdate_transactions(s, jl, &tjl, &trans_id, len, i);
1783         }
1784         flush_journal_list(s, flush_jl, 1);
1785         return 0;
1786 }
1787
1788 /*
1789 ** removes any nodes in table with name block and dev as bh.
1790 ** only touchs the hnext and hprev pointers.
1791 */
1792 void remove_journal_hash(struct super_block *sb,
1793                          struct reiserfs_journal_cnode **table,
1794                          struct reiserfs_journal_list *jl,
1795                          unsigned long block, int remove_freed)
1796 {
1797         struct reiserfs_journal_cnode *cur;
1798         struct reiserfs_journal_cnode **head;
1799
1800         head = &(journal_hash(table, sb, block));
1801         if (!head) {
1802                 return;
1803         }
1804         cur = *head;
1805         while (cur) {
1806                 if (cur->blocknr == block && cur->sb == sb
1807                     && (jl == NULL || jl == cur->jlist)
1808                     && (!test_bit(BLOCK_FREED, &cur->state) || remove_freed)) {
1809                         if (cur->hnext) {
1810                                 cur->hnext->hprev = cur->hprev;
1811                         }
1812                         if (cur->hprev) {
1813                                 cur->hprev->hnext = cur->hnext;
1814                         } else {
1815                                 *head = cur->hnext;
1816                         }
1817                         cur->blocknr = 0;
1818                         cur->sb = NULL;
1819                         cur->state = 0;
1820                         if (cur->bh && cur->jlist)      /* anybody who clears the cur->bh will also dec the nonzerolen */
1821                                 atomic_dec(&(cur->jlist->j_nonzerolen));
1822                         cur->bh = NULL;
1823                         cur->jlist = NULL;
1824                 }
1825                 cur = cur->hnext;
1826         }
1827 }
1828
1829 static void free_journal_ram(struct super_block *p_s_sb)
1830 {
1831         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1832         kfree(journal->j_current_jl);
1833         journal->j_num_lists--;
1834
1835         vfree(journal->j_cnode_free_orig);
1836         free_list_bitmaps(p_s_sb, journal->j_list_bitmap);
1837         free_bitmap_nodes(p_s_sb);      /* must be after free_list_bitmaps */
1838         if (journal->j_header_bh) {
1839                 brelse(journal->j_header_bh);
1840         }
1841         /* j_header_bh is on the journal dev, make sure not to release the journal
1842          * dev until we brelse j_header_bh
1843          */
1844         release_journal_dev(p_s_sb, journal);
1845         vfree(journal);
1846 }
1847
1848 /*
1849 ** call on unmount.  Only set error to 1 if you haven't made your way out
1850 ** of read_super() yet.  Any other caller must keep error at 0.
1851 */
1852 static int do_journal_release(struct reiserfs_transaction_handle *th,
1853                               struct super_block *p_s_sb, int error)
1854 {
1855         struct reiserfs_transaction_handle myth;
1856         int flushed = 0;
1857         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1858
1859         /* we only want to flush out transactions if we were called with error == 0
1860          */
1861         if (!error && !(p_s_sb->s_flags & MS_RDONLY)) {
1862                 /* end the current trans */
1863                 BUG_ON(!th->t_trans_id);
1864                 do_journal_end(th, p_s_sb, 10, FLUSH_ALL);
1865
1866                 /* make sure something gets logged to force our way into the flush code */
1867                 if (!journal_join(&myth, p_s_sb, 1)) {
1868                         reiserfs_prepare_for_journal(p_s_sb,
1869                                                      SB_BUFFER_WITH_SB(p_s_sb),
1870                                                      1);
1871                         journal_mark_dirty(&myth, p_s_sb,
1872                                            SB_BUFFER_WITH_SB(p_s_sb));
1873                         do_journal_end(&myth, p_s_sb, 1, FLUSH_ALL);
1874                         flushed = 1;
1875                 }
1876         }
1877
1878         /* this also catches errors during the do_journal_end above */
1879         if (!error && reiserfs_is_journal_aborted(journal)) {
1880                 memset(&myth, 0, sizeof(myth));
1881                 if (!journal_join_abort(&myth, p_s_sb, 1)) {
1882                         reiserfs_prepare_for_journal(p_s_sb,
1883                                                      SB_BUFFER_WITH_SB(p_s_sb),
1884                                                      1);
1885                         journal_mark_dirty(&myth, p_s_sb,
1886                                            SB_BUFFER_WITH_SB(p_s_sb));
1887                         do_journal_end(&myth, p_s_sb, 1, FLUSH_ALL);
1888                 }
1889         }
1890
1891         reiserfs_mounted_fs_count--;
1892         /* wait for all commits to finish */
1893         cancel_delayed_work(&SB_JOURNAL(p_s_sb)->j_work);
1894         flush_workqueue(commit_wq);
1895         if (!reiserfs_mounted_fs_count) {
1896                 destroy_workqueue(commit_wq);
1897                 commit_wq = NULL;
1898         }
1899
1900         free_journal_ram(p_s_sb);
1901
1902         return 0;
1903 }
1904
1905 /*
1906 ** call on unmount.  flush all journal trans, release all alloc'd ram
1907 */
1908 int journal_release(struct reiserfs_transaction_handle *th,
1909                     struct super_block *p_s_sb)
1910 {
1911         return do_journal_release(th, p_s_sb, 0);
1912 }
1913
1914 /*
1915 ** only call from an error condition inside reiserfs_read_super!
1916 */
1917 int journal_release_error(struct reiserfs_transaction_handle *th,
1918                           struct super_block *p_s_sb)
1919 {
1920         return do_journal_release(th, p_s_sb, 1);
1921 }
1922
1923 /* compares description block with commit block.  returns 1 if they differ, 0 if they are the same */
1924 static int journal_compare_desc_commit(struct super_block *p_s_sb,
1925                                        struct reiserfs_journal_desc *desc,
1926                                        struct reiserfs_journal_commit *commit)
1927 {
1928         if (get_commit_trans_id(commit) != get_desc_trans_id(desc) ||
1929             get_commit_trans_len(commit) != get_desc_trans_len(desc) ||
1930             get_commit_trans_len(commit) > SB_JOURNAL(p_s_sb)->j_trans_max ||
1931             get_commit_trans_len(commit) <= 0) {
1932                 return 1;
1933         }
1934         return 0;
1935 }
1936
1937 /* returns 0 if it did not find a description block
1938 ** returns -1 if it found a corrupt commit block
1939 ** returns 1 if both desc and commit were valid
1940 */
1941 static int journal_transaction_is_valid(struct super_block *p_s_sb,
1942                                         struct buffer_head *d_bh,
1943                                         unsigned long *oldest_invalid_trans_id,
1944                                         unsigned long *newest_mount_id)
1945 {
1946         struct reiserfs_journal_desc *desc;
1947         struct reiserfs_journal_commit *commit;
1948         struct buffer_head *c_bh;
1949         unsigned long offset;
1950
1951         if (!d_bh)
1952                 return 0;
1953
1954         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
1955         if (get_desc_trans_len(desc) > 0
1956             && !memcmp(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8)) {
1957                 if (oldest_invalid_trans_id && *oldest_invalid_trans_id
1958                     && get_desc_trans_id(desc) > *oldest_invalid_trans_id) {
1959                         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
1960                                        "journal-986: transaction "
1961                                        "is valid returning because trans_id %d is greater than "
1962                                        "oldest_invalid %lu",
1963                                        get_desc_trans_id(desc),
1964                                        *oldest_invalid_trans_id);
1965                         return 0;
1966                 }
1967                 if (newest_mount_id
1968                     && *newest_mount_id > get_desc_mount_id(desc)) {
1969                         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
1970                                        "journal-1087: transaction "
1971                                        "is valid returning because mount_id %d is less than "
1972                                        "newest_mount_id %lu",
1973                                        get_desc_mount_id(desc),
1974                                        *newest_mount_id);
1975                         return -1;
1976                 }
1977                 if (get_desc_trans_len(desc) > SB_JOURNAL(p_s_sb)->j_trans_max) {
1978                         reiserfs_warning(p_s_sb,
1979                                          "journal-2018: Bad transaction length %d encountered, ignoring transaction",
1980                                          get_desc_trans_len(desc));
1981                         return -1;
1982                 }
1983                 offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb);
1984
1985                 /* ok, we have a journal description block, lets see if the transaction was valid */
1986                 c_bh =
1987                     journal_bread(p_s_sb,
1988                                   SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
1989                                   ((offset + get_desc_trans_len(desc) +
1990                                     1) % SB_ONDISK_JOURNAL_SIZE(p_s_sb)));
1991                 if (!c_bh)
1992                         return 0;
1993                 commit = (struct reiserfs_journal_commit *)c_bh->b_data;
1994                 if (journal_compare_desc_commit(p_s_sb, desc, commit)) {
1995                         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
1996                                        "journal_transaction_is_valid, commit offset %ld had bad "
1997                                        "time %d or length %d",
1998                                        c_bh->b_blocknr -
1999                                        SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2000                                        get_commit_trans_id(commit),
2001                                        get_commit_trans_len(commit));
2002                         brelse(c_bh);
2003                         if (oldest_invalid_trans_id) {
2004                                 *oldest_invalid_trans_id =
2005                                     get_desc_trans_id(desc);
2006                                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2007                                                "journal-1004: "
2008                                                "transaction_is_valid setting oldest invalid trans_id "
2009                                                "to %d",
2010                                                get_desc_trans_id(desc));
2011                         }
2012                         return -1;
2013                 }
2014                 brelse(c_bh);
2015                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2016                                "journal-1006: found valid "
2017                                "transaction start offset %llu, len %d id %d",
2018                                d_bh->b_blocknr -
2019                                SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2020                                get_desc_trans_len(desc),
2021                                get_desc_trans_id(desc));
2022                 return 1;
2023         } else {
2024                 return 0;
2025         }
2026 }
2027
2028 static void brelse_array(struct buffer_head **heads, int num)
2029 {
2030         int i;
2031         for (i = 0; i < num; i++) {
2032                 brelse(heads[i]);
2033         }
2034 }
2035
2036 /*
2037 ** given the start, and values for the oldest acceptable transactions,
2038 ** this either reads in a replays a transaction, or returns because the transaction
2039 ** is invalid, or too old.
2040 */
2041 static int journal_read_transaction(struct super_block *p_s_sb,
2042                                     unsigned long cur_dblock,
2043                                     unsigned long oldest_start,
2044                                     unsigned long oldest_trans_id,
2045                                     unsigned long newest_mount_id)
2046 {
2047         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
2048         struct reiserfs_journal_desc *desc;
2049         struct reiserfs_journal_commit *commit;
2050         unsigned long trans_id = 0;
2051         struct buffer_head *c_bh;
2052         struct buffer_head *d_bh;
2053         struct buffer_head **log_blocks = NULL;
2054         struct buffer_head **real_blocks = NULL;
2055         unsigned long trans_offset;
2056         int i;
2057         int trans_half;
2058
2059         d_bh = journal_bread(p_s_sb, cur_dblock);
2060         if (!d_bh)
2061                 return 1;
2062         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2063         trans_offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb);
2064         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1037: "
2065                        "journal_read_transaction, offset %llu, len %d mount_id %d",
2066                        d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2067                        get_desc_trans_len(desc), get_desc_mount_id(desc));
2068         if (get_desc_trans_id(desc) < oldest_trans_id) {
2069                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1039: "
2070                                "journal_read_trans skipping because %lu is too old",
2071                                cur_dblock -
2072                                SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb));
2073                 brelse(d_bh);
2074                 return 1;
2075         }
2076         if (get_desc_mount_id(desc) != newest_mount_id) {
2077                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1146: "
2078                                "journal_read_trans skipping because %d is != "
2079                                "newest_mount_id %lu", get_desc_mount_id(desc),
2080                                newest_mount_id);
2081                 brelse(d_bh);
2082                 return 1;
2083         }
2084         c_bh = journal_bread(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2085                              ((trans_offset + get_desc_trans_len(desc) + 1) %
2086                               SB_ONDISK_JOURNAL_SIZE(p_s_sb)));
2087         if (!c_bh) {
2088                 brelse(d_bh);
2089                 return 1;
2090         }
2091         commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2092         if (journal_compare_desc_commit(p_s_sb, desc, commit)) {
2093                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2094                                "journal_read_transaction, "
2095                                "commit offset %llu had bad time %d or length %d",
2096                                c_bh->b_blocknr -
2097                                SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2098                                get_commit_trans_id(commit),
2099                                get_commit_trans_len(commit));
2100                 brelse(c_bh);
2101                 brelse(d_bh);
2102                 return 1;
2103         }
2104         trans_id = get_desc_trans_id(desc);
2105         /* now we know we've got a good transaction, and it was inside the valid time ranges */
2106         log_blocks = kmalloc(get_desc_trans_len(desc) *
2107                              sizeof(struct buffer_head *), GFP_NOFS);
2108         real_blocks = kmalloc(get_desc_trans_len(desc) *
2109                               sizeof(struct buffer_head *), GFP_NOFS);
2110         if (!log_blocks || !real_blocks) {
2111                 brelse(c_bh);
2112                 brelse(d_bh);
2113                 kfree(log_blocks);
2114                 kfree(real_blocks);
2115                 reiserfs_warning(p_s_sb,
2116                                  "journal-1169: kmalloc failed, unable to mount FS");
2117                 return -1;
2118         }
2119         /* get all the buffer heads */
2120         trans_half = journal_trans_half(p_s_sb->s_blocksize);
2121         for (i = 0; i < get_desc_trans_len(desc); i++) {
2122                 log_blocks[i] =
2123                     journal_getblk(p_s_sb,
2124                                    SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2125                                    (trans_offset + 1 +
2126                                     i) % SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2127                 if (i < trans_half) {
2128                         real_blocks[i] =
2129                             sb_getblk(p_s_sb,
2130                                       le32_to_cpu(desc->j_realblock[i]));
2131                 } else {
2132                         real_blocks[i] =
2133                             sb_getblk(p_s_sb,
2134                                       le32_to_cpu(commit->
2135                                                   j_realblock[i - trans_half]));
2136                 }
2137                 if (real_blocks[i]->b_blocknr > SB_BLOCK_COUNT(p_s_sb)) {
2138                         reiserfs_warning(p_s_sb,
2139                                          "journal-1207: REPLAY FAILURE fsck required! Block to replay is outside of filesystem");
2140                         goto abort_replay;
2141                 }
2142                 /* make sure we don't try to replay onto log or reserved area */
2143                 if (is_block_in_log_or_reserved_area
2144                     (p_s_sb, real_blocks[i]->b_blocknr)) {
2145                         reiserfs_warning(p_s_sb,
2146                                          "journal-1204: REPLAY FAILURE fsck required! Trying to replay onto a log block");
2147                       abort_replay:
2148                         brelse_array(log_blocks, i);
2149                         brelse_array(real_blocks, i);
2150                         brelse(c_bh);
2151                         brelse(d_bh);
2152                         kfree(log_blocks);
2153                         kfree(real_blocks);
2154                         return -1;
2155                 }
2156         }
2157         /* read in the log blocks, memcpy to the corresponding real block */
2158         ll_rw_block(READ, get_desc_trans_len(desc), log_blocks);
2159         for (i = 0; i < get_desc_trans_len(desc); i++) {
2160                 wait_on_buffer(log_blocks[i]);
2161                 if (!buffer_uptodate(log_blocks[i])) {
2162                         reiserfs_warning(p_s_sb,
2163                                          "journal-1212: REPLAY FAILURE fsck required! buffer write failed");
2164                         brelse_array(log_blocks + i,
2165                                      get_desc_trans_len(desc) - i);
2166                         brelse_array(real_blocks, get_desc_trans_len(desc));
2167                         brelse(c_bh);
2168                         brelse(d_bh);
2169                         kfree(log_blocks);
2170                         kfree(real_blocks);
2171                         return -1;
2172                 }
2173                 memcpy(real_blocks[i]->b_data, log_blocks[i]->b_data,
2174                        real_blocks[i]->b_size);
2175                 set_buffer_uptodate(real_blocks[i]);
2176                 brelse(log_blocks[i]);
2177         }
2178         /* flush out the real blocks */
2179         for (i = 0; i < get_desc_trans_len(desc); i++) {
2180                 set_buffer_dirty(real_blocks[i]);
2181                 ll_rw_block(SWRITE, 1, real_blocks + i);
2182         }
2183         for (i = 0; i < get_desc_trans_len(desc); i++) {
2184                 wait_on_buffer(real_blocks[i]);
2185                 if (!buffer_uptodate(real_blocks[i])) {
2186                         reiserfs_warning(p_s_sb,
2187                                          "journal-1226: REPLAY FAILURE, fsck required! buffer write failed");
2188                         brelse_array(real_blocks + i,
2189                                      get_desc_trans_len(desc) - i);
2190                         brelse(c_bh);
2191                         brelse(d_bh);
2192                         kfree(log_blocks);
2193                         kfree(real_blocks);
2194                         return -1;
2195                 }
2196                 brelse(real_blocks[i]);
2197         }
2198         cur_dblock =
2199             SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2200             ((trans_offset + get_desc_trans_len(desc) +
2201               2) % SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2202         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2203                        "journal-1095: setting journal " "start to offset %ld",
2204                        cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb));
2205
2206         /* init starting values for the first transaction, in case this is the last transaction to be replayed. */
2207         journal->j_start = cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb);
2208         journal->j_last_flush_trans_id = trans_id;
2209         journal->j_trans_id = trans_id + 1;
2210         brelse(c_bh);
2211         brelse(d_bh);
2212         kfree(log_blocks);
2213         kfree(real_blocks);
2214         return 0;
2215 }
2216
2217 /* This function reads blocks starting from block and to max_block of bufsize
2218    size (but no more than BUFNR blocks at a time). This proved to improve
2219    mounting speed on self-rebuilding raid5 arrays at least.
2220    Right now it is only used from journal code. But later we might use it
2221    from other places.
2222    Note: Do not use journal_getblk/sb_getblk functions here! */
2223 static struct buffer_head *reiserfs_breada(struct block_device *dev, int block,
2224                                            int bufsize, unsigned int max_block)
2225 {
2226         struct buffer_head *bhlist[BUFNR];
2227         unsigned int blocks = BUFNR;
2228         struct buffer_head *bh;
2229         int i, j;
2230
2231         bh = __getblk(dev, block, bufsize);
2232         if (buffer_uptodate(bh))
2233                 return (bh);
2234
2235         if (block + BUFNR > max_block) {
2236                 blocks = max_block - block;
2237         }
2238         bhlist[0] = bh;
2239         j = 1;
2240         for (i = 1; i < blocks; i++) {
2241                 bh = __getblk(dev, block + i, bufsize);
2242                 if (buffer_uptodate(bh)) {
2243                         brelse(bh);
2244                         break;
2245                 } else
2246                         bhlist[j++] = bh;
2247         }
2248         ll_rw_block(READ, j, bhlist);
2249         for (i = 1; i < j; i++)
2250                 brelse(bhlist[i]);
2251         bh = bhlist[0];
2252         wait_on_buffer(bh);
2253         if (buffer_uptodate(bh))
2254                 return bh;
2255         brelse(bh);
2256         return NULL;
2257 }
2258
2259 /*
2260 ** read and replay the log
2261 ** on a clean unmount, the journal header's next unflushed pointer will be to an invalid
2262 ** transaction.  This tests that before finding all the transactions in the log, which makes normal mount times fast.
2263 **
2264 ** After a crash, this starts with the next unflushed transaction, and replays until it finds one too old, or invalid.
2265 **
2266 ** On exit, it sets things up so the first transaction will work correctly.
2267 */
2268 static int journal_read(struct super_block *p_s_sb)
2269 {
2270         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
2271         struct reiserfs_journal_desc *desc;
2272         unsigned long oldest_trans_id = 0;
2273         unsigned long oldest_invalid_trans_id = 0;
2274         time_t start;
2275         unsigned long oldest_start = 0;
2276         unsigned long cur_dblock = 0;
2277         unsigned long newest_mount_id = 9;
2278         struct buffer_head *d_bh;
2279         struct reiserfs_journal_header *jh;
2280         int valid_journal_header = 0;
2281         int replay_count = 0;
2282         int continue_replay = 1;
2283         int ret;
2284         char b[BDEVNAME_SIZE];
2285
2286         cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb);
2287         reiserfs_info(p_s_sb, "checking transaction log (%s)\n",
2288                       bdevname(journal->j_dev_bd, b));
2289         start = get_seconds();
2290
2291         /* step 1, read in the journal header block.  Check the transaction it says
2292          ** is the first unflushed, and if that transaction is not valid,
2293          ** replay is done
2294          */
2295         journal->j_header_bh = journal_bread(p_s_sb,
2296                                              SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb)
2297                                              + SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2298         if (!journal->j_header_bh) {
2299                 return 1;
2300         }
2301         jh = (struct reiserfs_journal_header *)(journal->j_header_bh->b_data);
2302         if (le32_to_cpu(jh->j_first_unflushed_offset) >= 0 &&
2303             le32_to_cpu(jh->j_first_unflushed_offset) <
2304             SB_ONDISK_JOURNAL_SIZE(p_s_sb)
2305             && le32_to_cpu(jh->j_last_flush_trans_id) > 0) {
2306                 oldest_start =
2307                     SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2308                     le32_to_cpu(jh->j_first_unflushed_offset);
2309                 oldest_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2310                 newest_mount_id = le32_to_cpu(jh->j_mount_id);
2311                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2312                                "journal-1153: found in "
2313                                "header: first_unflushed_offset %d, last_flushed_trans_id "
2314                                "%lu", le32_to_cpu(jh->j_first_unflushed_offset),
2315                                le32_to_cpu(jh->j_last_flush_trans_id));
2316                 valid_journal_header = 1;
2317
2318                 /* now, we try to read the first unflushed offset.  If it is not valid,
2319                  ** there is nothing more we can do, and it makes no sense to read
2320                  ** through the whole log.
2321                  */
2322                 d_bh =
2323                     journal_bread(p_s_sb,
2324                                   SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2325                                   le32_to_cpu(jh->j_first_unflushed_offset));
2326                 ret = journal_transaction_is_valid(p_s_sb, d_bh, NULL, NULL);
2327                 if (!ret) {
2328                         continue_replay = 0;
2329                 }
2330                 brelse(d_bh);
2331                 goto start_log_replay;
2332         }
2333
2334         if (continue_replay && bdev_read_only(p_s_sb->s_bdev)) {
2335                 reiserfs_warning(p_s_sb,
2336                                  "clm-2076: device is readonly, unable to replay log");
2337                 return -1;
2338         }
2339
2340         /* ok, there are transactions that need to be replayed.  start with the first log block, find
2341          ** all the valid transactions, and pick out the oldest.
2342          */
2343         while (continue_replay
2344                && cur_dblock <
2345                (SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2346                 SB_ONDISK_JOURNAL_SIZE(p_s_sb))) {
2347                 /* Note that it is required for blocksize of primary fs device and journal
2348                    device to be the same */
2349                 d_bh =
2350                     reiserfs_breada(journal->j_dev_bd, cur_dblock,
2351                                     p_s_sb->s_blocksize,
2352                                     SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2353                                     SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2354                 ret =
2355                     journal_transaction_is_valid(p_s_sb, d_bh,
2356                                                  &oldest_invalid_trans_id,
2357                                                  &newest_mount_id);
2358                 if (ret == 1) {
2359                         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2360                         if (oldest_start == 0) {        /* init all oldest_ values */
2361                                 oldest_trans_id = get_desc_trans_id(desc);
2362                                 oldest_start = d_bh->b_blocknr;
2363                                 newest_mount_id = get_desc_mount_id(desc);
2364                                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2365                                                "journal-1179: Setting "
2366                                                "oldest_start to offset %llu, trans_id %lu",
2367                                                oldest_start -
2368                                                SB_ONDISK_JOURNAL_1st_BLOCK
2369                                                (p_s_sb), oldest_trans_id);
2370                         } else if (oldest_trans_id > get_desc_trans_id(desc)) {
2371                                 /* one we just read was older */
2372                                 oldest_trans_id = get_desc_trans_id(desc);
2373                                 oldest_start = d_bh->b_blocknr;
2374                                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2375                                                "journal-1180: Resetting "
2376                                                "oldest_start to offset %lu, trans_id %lu",
2377                                                oldest_start -
2378                                                SB_ONDISK_JOURNAL_1st_BLOCK
2379                                                (p_s_sb), oldest_trans_id);
2380                         }
2381                         if (newest_mount_id < get_desc_mount_id(desc)) {
2382                                 newest_mount_id = get_desc_mount_id(desc);
2383                                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2384                                                "journal-1299: Setting "
2385                                                "newest_mount_id to %d",
2386                                                get_desc_mount_id(desc));
2387                         }
2388                         cur_dblock += get_desc_trans_len(desc) + 2;
2389                 } else {
2390                         cur_dblock++;
2391                 }
2392                 brelse(d_bh);
2393         }
2394
2395       start_log_replay:
2396         cur_dblock = oldest_start;
2397         if (oldest_trans_id) {
2398                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2399                                "journal-1206: Starting replay "
2400                                "from offset %llu, trans_id %lu",
2401                                cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2402                                oldest_trans_id);
2403
2404         }
2405         replay_count = 0;
2406         while (continue_replay && oldest_trans_id > 0) {
2407                 ret =
2408                     journal_read_transaction(p_s_sb, cur_dblock, oldest_start,
2409                                              oldest_trans_id, newest_mount_id);
2410                 if (ret < 0) {
2411                         return ret;
2412                 } else if (ret != 0) {
2413                         break;
2414                 }
2415                 cur_dblock =
2416                     SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + journal->j_start;
2417                 replay_count++;
2418                 if (cur_dblock == oldest_start)
2419                         break;
2420         }
2421
2422         if (oldest_trans_id == 0) {
2423                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2424                                "journal-1225: No valid " "transactions found");
2425         }
2426         /* j_start does not get set correctly if we don't replay any transactions.
2427          ** if we had a valid journal_header, set j_start to the first unflushed transaction value,
2428          ** copy the trans_id from the header
2429          */
2430         if (valid_journal_header && replay_count == 0) {
2431                 journal->j_start = le32_to_cpu(jh->j_first_unflushed_offset);
2432                 journal->j_trans_id =
2433                     le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2434                 journal->j_last_flush_trans_id =
2435                     le32_to_cpu(jh->j_last_flush_trans_id);
2436                 journal->j_mount_id = le32_to_cpu(jh->j_mount_id) + 1;
2437         } else {
2438                 journal->j_mount_id = newest_mount_id + 1;
2439         }
2440         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1299: Setting "
2441                        "newest_mount_id to %lu", journal->j_mount_id);
2442         journal->j_first_unflushed_offset = journal->j_start;
2443         if (replay_count > 0) {
2444                 reiserfs_info(p_s_sb,
2445                               "replayed %d transactions in %lu seconds\n",
2446                               replay_count, get_seconds() - start);
2447         }
2448         if (!bdev_read_only(p_s_sb->s_bdev) &&
2449             _update_journal_header_block(p_s_sb, journal->j_start,
2450                                          journal->j_last_flush_trans_id)) {
2451                 /* replay failed, caller must call free_journal_ram and abort
2452                  ** the mount
2453                  */
2454                 return -1;
2455         }
2456         return 0;
2457 }
2458
2459 static struct reiserfs_journal_list *alloc_journal_list(struct super_block *s)
2460 {
2461         struct reiserfs_journal_list *jl;
2462         jl = kzalloc(sizeof(struct reiserfs_journal_list),
2463                      GFP_NOFS | __GFP_NOFAIL);
2464         INIT_LIST_HEAD(&jl->j_list);
2465         INIT_LIST_HEAD(&jl->j_working_list);
2466         INIT_LIST_HEAD(&jl->j_tail_bh_list);
2467         INIT_LIST_HEAD(&jl->j_bh_list);
2468         sema_init(&jl->j_commit_lock, 1);
2469         SB_JOURNAL(s)->j_num_lists++;
2470         get_journal_list(jl);
2471         return jl;
2472 }
2473
2474 static void journal_list_init(struct super_block *p_s_sb)
2475 {
2476         SB_JOURNAL(p_s_sb)->j_current_jl = alloc_journal_list(p_s_sb);
2477 }
2478
2479 static int release_journal_dev(struct super_block *super,
2480                                struct reiserfs_journal *journal)
2481 {
2482         int result;
2483
2484         result = 0;
2485
2486         if (journal->j_dev_file != NULL) {
2487                 result = filp_close(journal->j_dev_file, NULL);
2488                 journal->j_dev_file = NULL;
2489                 journal->j_dev_bd = NULL;
2490         } else if (journal->j_dev_bd != NULL) {
2491                 result = blkdev_put(journal->j_dev_bd);
2492                 journal->j_dev_bd = NULL;
2493         }
2494
2495         if (result != 0) {
2496                 reiserfs_warning(super,
2497                                  "sh-457: release_journal_dev: Cannot release journal device: %i",
2498                                  result);
2499         }
2500         return result;
2501 }
2502
2503 static int journal_init_dev(struct super_block *super,
2504                             struct reiserfs_journal *journal,
2505                             const char *jdev_name)
2506 {
2507         int result;
2508         dev_t jdev;
2509         int blkdev_mode = FMODE_READ | FMODE_WRITE;
2510         char b[BDEVNAME_SIZE];
2511
2512         result = 0;
2513
2514         journal->j_dev_bd = NULL;
2515         journal->j_dev_file = NULL;
2516         jdev = SB_ONDISK_JOURNAL_DEVICE(super) ?
2517             new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super)) : super->s_dev;
2518
2519         if (bdev_read_only(super->s_bdev))
2520                 blkdev_mode = FMODE_READ;
2521
2522         /* there is no "jdev" option and journal is on separate device */
2523         if ((!jdev_name || !jdev_name[0])) {
2524                 journal->j_dev_bd = open_by_devnum(jdev, blkdev_mode);
2525                 if (IS_ERR(journal->j_dev_bd)) {
2526                         result = PTR_ERR(journal->j_dev_bd);
2527                         journal->j_dev_bd = NULL;
2528                         reiserfs_warning(super, "sh-458: journal_init_dev: "
2529                                          "cannot init journal device '%s': %i",
2530                                          __bdevname(jdev, b), result);
2531                         return result;
2532                 } else if (jdev != super->s_dev)
2533                         set_blocksize(journal->j_dev_bd, super->s_blocksize);
2534                 return 0;
2535         }
2536
2537         journal->j_dev_file = filp_open(jdev_name, 0, 0);
2538         if (!IS_ERR(journal->j_dev_file)) {
2539                 struct inode *jdev_inode = journal->j_dev_file->f_mapping->host;
2540                 if (!S_ISBLK(jdev_inode->i_mode)) {
2541                         reiserfs_warning(super, "journal_init_dev: '%s' is "
2542                                          "not a block device", jdev_name);
2543                         result = -ENOTBLK;
2544                         release_journal_dev(super, journal);
2545                 } else {
2546                         /* ok */
2547                         journal->j_dev_bd = I_BDEV(jdev_inode);
2548                         set_blocksize(journal->j_dev_bd, super->s_blocksize);
2549                         reiserfs_info(super,
2550                                       "journal_init_dev: journal device: %s\n",
2551                                       bdevname(journal->j_dev_bd, b));
2552                 }
2553         } else {
2554                 result = PTR_ERR(journal->j_dev_file);
2555                 journal->j_dev_file = NULL;
2556                 reiserfs_warning(super,
2557                                  "journal_init_dev: Cannot open '%s': %i",
2558                                  jdev_name, result);
2559         }
2560         return result;
2561 }
2562
2563 /*
2564 ** must be called once on fs mount.  calls journal_read for you
2565 */
2566 int journal_init(struct super_block *p_s_sb, const char *j_dev_name,
2567                  int old_format, unsigned int commit_max_age)
2568 {
2569         int num_cnodes = SB_ONDISK_JOURNAL_SIZE(p_s_sb) * 2;
2570         struct buffer_head *bhjh;
2571         struct reiserfs_super_block *rs;
2572         struct reiserfs_journal_header *jh;
2573         struct reiserfs_journal *journal;
2574         struct reiserfs_journal_list *jl;
2575         char b[BDEVNAME_SIZE];
2576
2577         journal = SB_JOURNAL(p_s_sb) = vmalloc(sizeof(struct reiserfs_journal));
2578         if (!journal) {
2579                 reiserfs_warning(p_s_sb,
2580                                  "journal-1256: unable to get memory for journal structure");
2581                 return 1;
2582         }
2583         memset(journal, 0, sizeof(struct reiserfs_journal));
2584         INIT_LIST_HEAD(&journal->j_bitmap_nodes);
2585         INIT_LIST_HEAD(&journal->j_prealloc_list);
2586         INIT_LIST_HEAD(&journal->j_working_list);
2587         INIT_LIST_HEAD(&journal->j_journal_list);
2588         journal->j_persistent_trans = 0;
2589         if (reiserfs_allocate_list_bitmaps(p_s_sb,
2590                                            journal->j_list_bitmap,
2591                                            SB_BMAP_NR(p_s_sb)))
2592                 goto free_and_return;
2593         allocate_bitmap_nodes(p_s_sb);
2594
2595         /* reserved for journal area support */
2596         SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb) = (old_format ?
2597                                                  REISERFS_OLD_DISK_OFFSET_IN_BYTES
2598                                                  / p_s_sb->s_blocksize +
2599                                                  SB_BMAP_NR(p_s_sb) +
2600                                                  1 :
2601                                                  REISERFS_DISK_OFFSET_IN_BYTES /
2602                                                  p_s_sb->s_blocksize + 2);
2603
2604         /* Sanity check to see is the standard journal fitting withing first bitmap
2605            (actual for small blocksizes) */
2606         if (!SB_ONDISK_JOURNAL_DEVICE(p_s_sb) &&
2607             (SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb) +
2608              SB_ONDISK_JOURNAL_SIZE(p_s_sb) > p_s_sb->s_blocksize * 8)) {
2609                 reiserfs_warning(p_s_sb,
2610                                  "journal-1393: journal does not fit for area "
2611                                  "addressed by first of bitmap blocks. It starts at "
2612                                  "%u and its size is %u. Block size %ld",
2613                                  SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb),
2614                                  SB_ONDISK_JOURNAL_SIZE(p_s_sb),
2615                                  p_s_sb->s_blocksize);
2616                 goto free_and_return;
2617         }
2618
2619         if (journal_init_dev(p_s_sb, journal, j_dev_name) != 0) {
2620                 reiserfs_warning(p_s_sb,
2621                                  "sh-462: unable to initialize jornal device");
2622                 goto free_and_return;
2623         }
2624
2625         rs = SB_DISK_SUPER_BLOCK(p_s_sb);
2626
2627         /* read journal header */
2628         bhjh = journal_bread(p_s_sb,
2629                              SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2630                              SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2631         if (!bhjh) {
2632                 reiserfs_warning(p_s_sb,
2633                                  "sh-459: unable to read journal header");
2634                 goto free_and_return;
2635         }
2636         jh = (struct reiserfs_journal_header *)(bhjh->b_data);
2637
2638         /* make sure that journal matches to the super block */
2639         if (is_reiserfs_jr(rs)
2640             && (le32_to_cpu(jh->jh_journal.jp_journal_magic) !=
2641                 sb_jp_journal_magic(rs))) {
2642                 reiserfs_warning(p_s_sb,
2643                                  "sh-460: journal header magic %x "
2644                                  "(device %s) does not match to magic found in super "
2645                                  "block %x", jh->jh_journal.jp_journal_magic,
2646                                  bdevname(journal->j_dev_bd, b),
2647                                  sb_jp_journal_magic(rs));
2648                 brelse(bhjh);
2649                 goto free_and_return;
2650         }
2651
2652         journal->j_trans_max = le32_to_cpu(jh->jh_journal.jp_journal_trans_max);
2653         journal->j_max_batch = le32_to_cpu(jh->jh_journal.jp_journal_max_batch);
2654         journal->j_max_commit_age =
2655             le32_to_cpu(jh->jh_journal.jp_journal_max_commit_age);
2656         journal->j_max_trans_age = JOURNAL_MAX_TRANS_AGE;
2657
2658         if (journal->j_trans_max) {
2659                 /* make sure these parameters are available, assign it if they are not */
2660                 __u32 initial = journal->j_trans_max;
2661                 __u32 ratio = 1;
2662
2663                 if (p_s_sb->s_blocksize < 4096)
2664                         ratio = 4096 / p_s_sb->s_blocksize;
2665
2666                 if (SB_ONDISK_JOURNAL_SIZE(p_s_sb) / journal->j_trans_max <
2667                     JOURNAL_MIN_RATIO)
2668                         journal->j_trans_max =
2669                             SB_ONDISK_JOURNAL_SIZE(p_s_sb) / JOURNAL_MIN_RATIO;
2670                 if (journal->j_trans_max > JOURNAL_TRANS_MAX_DEFAULT / ratio)
2671                         journal->j_trans_max =
2672                             JOURNAL_TRANS_MAX_DEFAULT / ratio;
2673                 if (journal->j_trans_max < JOURNAL_TRANS_MIN_DEFAULT / ratio)
2674                         journal->j_trans_max =
2675                             JOURNAL_TRANS_MIN_DEFAULT / ratio;
2676
2677                 if (journal->j_trans_max != initial)
2678                         reiserfs_warning(p_s_sb,
2679                                          "sh-461: journal_init: wrong transaction max size (%u). Changed to %u",
2680                                          initial, journal->j_trans_max);
2681
2682                 journal->j_max_batch = journal->j_trans_max *
2683                     JOURNAL_MAX_BATCH_DEFAULT / JOURNAL_TRANS_MAX_DEFAULT;
2684         }
2685
2686         if (!journal->j_trans_max) {
2687                 /*we have the file system was created by old version of mkreiserfs
2688                    so this field contains zero value */
2689                 journal->j_trans_max = JOURNAL_TRANS_MAX_DEFAULT;
2690                 journal->j_max_batch = JOURNAL_MAX_BATCH_DEFAULT;
2691                 journal->j_max_commit_age = JOURNAL_MAX_COMMIT_AGE;
2692
2693                 /* for blocksize >= 4096 - max transaction size is 1024. For block size < 4096
2694                    trans max size is decreased proportionally */
2695                 if (p_s_sb->s_blocksize < 4096) {
2696                         journal->j_trans_max /= (4096 / p_s_sb->s_blocksize);
2697                         journal->j_max_batch = (journal->j_trans_max) * 9 / 10;
2698                 }
2699         }
2700
2701         journal->j_default_max_commit_age = journal->j_max_commit_age;
2702
2703         if (commit_max_age != 0) {
2704                 journal->j_max_commit_age = commit_max_age;
2705                 journal->j_max_trans_age = commit_max_age;
2706         }
2707
2708         reiserfs_info(p_s_sb, "journal params: device %s, size %u, "
2709                       "journal first block %u, max trans len %u, max batch %u, "
2710                       "max commit age %u, max trans age %u\n",
2711                       bdevname(journal->j_dev_bd, b),
2712                       SB_ONDISK_JOURNAL_SIZE(p_s_sb),
2713                       SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2714                       journal->j_trans_max,
2715                       journal->j_max_batch,
2716                       journal->j_max_commit_age, journal->j_max_trans_age);
2717
2718         brelse(bhjh);
2719
2720         journal->j_list_bitmap_index = 0;
2721         journal_list_init(p_s_sb);
2722
2723         memset(journal->j_list_hash_table, 0,
2724                JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
2725
2726         INIT_LIST_HEAD(&journal->j_dirty_buffers);
2727         spin_lock_init(&journal->j_dirty_buffers_lock);
2728
2729         journal->j_start = 0;
2730         journal->j_len = 0;
2731         journal->j_len_alloc = 0;
2732         atomic_set(&(journal->j_wcount), 0);
2733         atomic_set(&(journal->j_async_throttle), 0);
2734         journal->j_bcount = 0;
2735         journal->j_trans_start_time = 0;
2736         journal->j_last = NULL;
2737         journal->j_first = NULL;
2738         init_waitqueue_head(&(journal->j_join_wait));
2739         sema_init(&journal->j_lock, 1);
2740         sema_init(&journal->j_flush_sem, 1);
2741
2742         journal->j_trans_id = 10;
2743         journal->j_mount_id = 10;
2744         journal->j_state = 0;
2745         atomic_set(&(journal->j_jlock), 0);
2746         journal->j_cnode_free_list = allocate_cnodes(num_cnodes);
2747         journal->j_cnode_free_orig = journal->j_cnode_free_list;
2748         journal->j_cnode_free = journal->j_cnode_free_list ? num_cnodes : 0;
2749         journal->j_cnode_used = 0;
2750         journal->j_must_wait = 0;
2751
2752         if (journal->j_cnode_free == 0) {
2753                 reiserfs_warning(p_s_sb, "journal-2004: Journal cnode memory "
2754                                  "allocation failed (%ld bytes). Journal is "
2755                                  "too large for available memory. Usually "
2756                                  "this is due to a journal that is too large.",
2757                                  sizeof (struct reiserfs_journal_cnode) * num_cnodes);
2758                 goto free_and_return;
2759         }
2760
2761         init_journal_hash(p_s_sb);
2762         jl = journal->j_current_jl;
2763         jl->j_list_bitmap = get_list_bitmap(p_s_sb, jl);
2764         if (!jl->j_list_bitmap) {
2765                 reiserfs_warning(p_s_sb,
2766                                  "journal-2005, get_list_bitmap failed for journal list 0");
2767                 goto free_and_return;
2768         }
2769         if (journal_read(p_s_sb) < 0) {
2770                 reiserfs_warning(p_s_sb, "Replay Failure, unable to mount");
2771                 goto free_and_return;
2772         }
2773
2774         reiserfs_mounted_fs_count++;
2775         if (reiserfs_mounted_fs_count <= 1)
2776                 commit_wq = create_workqueue("reiserfs");
2777
2778         INIT_WORK(&journal->j_work, flush_async_commits, p_s_sb);
2779         return 0;
2780       free_and_return:
2781         free_journal_ram(p_s_sb);
2782         return 1;
2783 }
2784
2785 /*
2786 ** test for a polite end of the current transaction.  Used by file_write, and should
2787 ** be used by delete to make sure they don't write more than can fit inside a single
2788 ** transaction
2789 */
2790 int journal_transaction_should_end(struct reiserfs_transaction_handle *th,
2791                                    int new_alloc)
2792 {
2793         struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2794         time_t now = get_seconds();
2795         /* cannot restart while nested */
2796         BUG_ON(!th->t_trans_id);
2797         if (th->t_refcount > 1)
2798                 return 0;
2799         if (journal->j_must_wait > 0 ||
2800             (journal->j_len_alloc + new_alloc) >= journal->j_max_batch ||
2801             atomic_read(&(journal->j_jlock)) ||
2802             (now - journal->j_trans_start_time) > journal->j_max_trans_age ||
2803             journal->j_cnode_free < (journal->j_trans_max * 3)) {
2804                 return 1;
2805         }
2806         return 0;
2807 }
2808
2809 /* this must be called inside a transaction, and requires the
2810 ** kernel_lock to be held
2811 */
2812 void reiserfs_block_writes(struct reiserfs_transaction_handle *th)
2813 {
2814         struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2815         BUG_ON(!th->t_trans_id);
2816         journal->j_must_wait = 1;
2817         set_bit(J_WRITERS_BLOCKED, &journal->j_state);
2818         return;
2819 }
2820
2821 /* this must be called without a transaction started, and does not
2822 ** require BKL
2823 */
2824 void reiserfs_allow_writes(struct super_block *s)
2825 {
2826         struct reiserfs_journal *journal = SB_JOURNAL(s);
2827         clear_bit(J_WRITERS_BLOCKED, &journal->j_state);
2828         wake_up(&journal->j_join_wait);
2829 }
2830
2831 /* this must be called without a transaction started, and does not
2832 ** require BKL
2833 */
2834 void reiserfs_wait_on_write_block(struct super_block *s)
2835 {
2836         struct reiserfs_journal *journal = SB_JOURNAL(s);
2837         wait_event(journal->j_join_wait,
2838                    !test_bit(J_WRITERS_BLOCKED, &journal->j_state));
2839 }
2840
2841 static void queue_log_writer(struct super_block *s)
2842 {
2843         wait_queue_t wait;
2844         struct reiserfs_journal *journal = SB_JOURNAL(s);
2845         set_bit(J_WRITERS_QUEUED, &journal->j_state);
2846
2847         /*
2848          * we don't want to use wait_event here because
2849          * we only want to wait once.
2850          */
2851         init_waitqueue_entry(&wait, current);
2852         add_wait_queue(&journal->j_join_wait, &wait);
2853         set_current_state(TASK_UNINTERRUPTIBLE);
2854         if (test_bit(J_WRITERS_QUEUED, &journal->j_state))
2855                 schedule();
2856         current->state = TASK_RUNNING;
2857         remove_wait_queue(&journal->j_join_wait, &wait);
2858 }
2859
2860 static void wake_queued_writers(struct super_block *s)
2861 {
2862         struct reiserfs_journal *journal = SB_JOURNAL(s);
2863         if (test_and_clear_bit(J_WRITERS_QUEUED, &journal->j_state))
2864                 wake_up(&journal->j_join_wait);
2865 }
2866
2867 static void let_transaction_grow(struct super_block *sb, unsigned long trans_id)
2868 {
2869         struct reiserfs_journal *journal = SB_JOURNAL(sb);
2870         unsigned long bcount = journal->j_bcount;
2871         while (1) {
2872                 schedule_timeout_uninterruptible(1);
2873                 journal->j_current_jl->j_state |= LIST_COMMIT_PENDING;
2874                 while ((atomic_read(&journal->j_wcount) > 0 ||
2875                         atomic_read(&journal->j_jlock)) &&
2876                        journal->j_trans_id == trans_id) {
2877                         queue_log_writer(sb);
2878                 }
2879                 if (journal->j_trans_id != trans_id)
2880                         break;
2881                 if (bcount == journal->j_bcount)
2882                         break;
2883                 bcount = journal->j_bcount;
2884         }
2885 }
2886
2887 /* join == true if you must join an existing transaction.
2888 ** join == false if you can deal with waiting for others to finish
2889 **
2890 ** this will block until the transaction is joinable.  send the number of blocks you
2891 ** expect to use in nblocks.
2892 */
2893 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
2894                               struct super_block *p_s_sb, unsigned long nblocks,
2895                               int join)
2896 {
2897         time_t now = get_seconds();
2898         int old_trans_id;
2899         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
2900         struct reiserfs_transaction_handle myth;
2901         int sched_count = 0;
2902         int retval;
2903
2904         reiserfs_check_lock_depth(p_s_sb, "journal_begin");
2905         if (nblocks > journal->j_trans_max)
2906                 BUG();
2907
2908         PROC_INFO_INC(p_s_sb, journal.journal_being);
2909         /* set here for journal_join */
2910         th->t_refcount = 1;
2911         th->t_super = p_s_sb;
2912
2913       relock:
2914         lock_journal(p_s_sb);
2915         if (join != JBEGIN_ABORT && reiserfs_is_journal_aborted(journal)) {
2916                 unlock_journal(p_s_sb);
2917                 retval = journal->j_errno;
2918                 goto out_fail;
2919         }
2920         journal->j_bcount++;
2921
2922         if (test_bit(J_WRITERS_BLOCKED, &journal->j_state)) {
2923                 unlock_journal(p_s_sb);
2924                 reiserfs_wait_on_write_block(p_s_sb);
2925                 PROC_INFO_INC(p_s_sb, journal.journal_relock_writers);
2926                 goto relock;
2927         }
2928         now = get_seconds();
2929
2930         /* if there is no room in the journal OR
2931          ** if this transaction is too old, and we weren't called joinable, wait for it to finish before beginning
2932          ** we don't sleep if there aren't other writers
2933          */
2934
2935         if ((!join && journal->j_must_wait > 0) ||
2936             (!join
2937              && (journal->j_len_alloc + nblocks + 2) >= journal->j_max_batch)
2938             || (!join && atomic_read(&journal->j_wcount) > 0
2939                 && journal->j_trans_start_time > 0
2940                 && (now - journal->j_trans_start_time) >
2941                 journal->j_max_trans_age) || (!join
2942                                               && atomic_read(&journal->j_jlock))
2943             || (!join && journal->j_cnode_free < (journal->j_trans_max * 3))) {
2944
2945                 old_trans_id = journal->j_trans_id;
2946                 unlock_journal(p_s_sb); /* allow others to finish this transaction */
2947
2948                 if (!join && (journal->j_len_alloc + nblocks + 2) >=
2949                     journal->j_max_batch &&
2950                     ((journal->j_len + nblocks + 2) * 100) <
2951                     (journal->j_len_alloc * 75)) {
2952                         if (atomic_read(&journal->j_wcount) > 10) {
2953                                 sched_count++;
2954                                 queue_log_writer(p_s_sb);
2955                                 goto relock;
2956                         }
2957                 }
2958                 /* don't mess with joining the transaction if all we have to do is
2959                  * wait for someone else to do a commit
2960                  */
2961                 if (atomic_read(&journal->j_jlock)) {
2962                         while (journal->j_trans_id == old_trans_id &&
2963                                atomic_read(&journal->j_jlock)) {
2964                                 queue_log_writer(p_s_sb);
2965                         }
2966                         goto relock;
2967                 }
2968                 retval = journal_join(&myth, p_s_sb, 1);
2969                 if (retval)
2970                         goto out_fail;
2971
2972                 /* someone might have ended the transaction while we joined */
2973                 if (old_trans_id != journal->j_trans_id) {
2974                         retval = do_journal_end(&myth, p_s_sb, 1, 0);
2975                 } else {
2976                         retval = do_journal_end(&myth, p_s_sb, 1, COMMIT_NOW);
2977                 }
2978
2979                 if (retval)
2980                         goto out_fail;
2981
2982                 PROC_INFO_INC(p_s_sb, journal.journal_relock_wcount);
2983                 goto relock;
2984         }
2985         /* we are the first writer, set trans_id */
2986         if (journal->j_trans_start_time == 0) {
2987                 journal->j_trans_start_time = get_seconds();
2988         }
2989         atomic_inc(&(journal->j_wcount));
2990         journal->j_len_alloc += nblocks;
2991         th->t_blocks_logged = 0;
2992         th->t_blocks_allocated = nblocks;
2993         th->t_trans_id = journal->j_trans_id;
2994         unlock_journal(p_s_sb);
2995         INIT_LIST_HEAD(&th->t_list);
2996         get_fs_excl();
2997         return 0;
2998
2999       out_fail:
3000         memset(th, 0, sizeof(*th));
3001         /* Re-set th->t_super, so we can properly keep track of how many
3002          * persistent transactions there are. We need to do this so if this
3003          * call is part of a failed restart_transaction, we can free it later */
3004         th->t_super = p_s_sb;
3005         return retval;
3006 }
3007
3008 struct reiserfs_transaction_handle *reiserfs_persistent_transaction(struct
3009                                                                     super_block
3010                                                                     *s,
3011                                                                     int nblocks)
3012 {
3013         int ret;
3014         struct reiserfs_transaction_handle *th;
3015
3016         /* if we're nesting into an existing transaction.  It will be
3017          ** persistent on its own
3018          */
3019         if (reiserfs_transaction_running(s)) {
3020                 th = current->journal_info;
3021                 th->t_refcount++;
3022                 if (th->t_refcount < 2) {
3023                         BUG();
3024                 }
3025                 return th;
3026         }
3027         th = kmalloc(sizeof(struct reiserfs_transaction_handle), GFP_NOFS);
3028         if (!th)
3029                 return NULL;
3030         ret = journal_begin(th, s, nblocks);
3031         if (ret) {
3032                 kfree(th);
3033                 return NULL;
3034         }
3035
3036         SB_JOURNAL(s)->j_persistent_trans++;
3037         return th;
3038 }
3039
3040 int reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle *th)
3041 {
3042         struct super_block *s = th->t_super;
3043         int ret = 0;
3044         if (th->t_trans_id)
3045                 ret = journal_end(th, th->t_super, th->t_blocks_allocated);
3046         else
3047                 ret = -EIO;
3048         if (th->t_refcount == 0) {
3049                 SB_JOURNAL(s)->j_persistent_trans--;
3050                 kfree(th);
3051         }
3052         return ret;
3053 }
3054
3055 static int journal_join(struct reiserfs_transaction_handle *th,
3056                         struct super_block *p_s_sb, unsigned long nblocks)
3057 {
3058         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3059
3060         /* this keeps do_journal_end from NULLing out the current->journal_info
3061          ** pointer
3062          */
3063         th->t_handle_save = cur_th;
3064         if (cur_th && cur_th->t_refcount > 1) {
3065                 BUG();
3066         }
3067         return do_journal_begin_r(th, p_s_sb, nblocks, JBEGIN_JOIN);
3068 }
3069
3070 int journal_join_abort(struct reiserfs_transaction_handle *th,
3071                        struct super_block *p_s_sb, unsigned long nblocks)
3072 {
3073         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3074
3075         /* this keeps do_journal_end from NULLing out the current->journal_info
3076          ** pointer
3077          */
3078         th->t_handle_save = cur_th;
3079         if (cur_th && cur_th->t_refcount > 1) {
3080                 BUG();
3081         }
3082         return do_journal_begin_r(th, p_s_sb, nblocks, JBEGIN_ABORT);
3083 }
3084
3085 int journal_begin(struct reiserfs_transaction_handle *th,
3086                   struct super_block *p_s_sb, unsigned long nblocks)
3087 {
3088         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3089         int ret;
3090
3091         th->t_handle_save = NULL;
3092         if (cur_th) {
3093                 /* we are nesting into the current transaction */
3094                 if (cur_th->t_super == p_s_sb) {
3095                         BUG_ON(!cur_th->t_refcount);
3096                         cur_th->t_refcount++;
3097                         memcpy(th, cur_th, sizeof(*th));
3098                         if (th->t_refcount <= 1)
3099                                 reiserfs_warning(p_s_sb,
3100                                                  "BAD: refcount <= 1, but journal_info != 0");
3101                         return 0;
3102                 } else {
3103                         /* we've ended up with a handle from a different filesystem.
3104                          ** save it and restore on journal_end.  This should never
3105                          ** really happen...
3106                          */
3107                         reiserfs_warning(p_s_sb,
3108                                          "clm-2100: nesting info a different FS");
3109                         th->t_handle_save = current->journal_info;
3110                         current->journal_info = th;
3111                 }
3112         } else {
3113                 current->journal_info = th;
3114         }
3115         ret = do_journal_begin_r(th, p_s_sb, nblocks, JBEGIN_REG);
3116         if (current->journal_info != th)
3117                 BUG();
3118
3119         /* I guess this boils down to being the reciprocal of clm-2100 above.
3120          * If do_journal_begin_r fails, we need to put it back, since journal_end
3121          * won't be called to do it. */
3122         if (ret)
3123                 current->journal_info = th->t_handle_save;
3124         else
3125                 BUG_ON(!th->t_refcount);
3126
3127         return ret;
3128 }
3129
3130 /*
3131 ** puts bh into the current transaction.  If it was already there, reorders removes the
3132 ** old pointers from the hash, and puts new ones in (to make sure replay happen in the right order).
3133 **
3134 ** if it was dirty, cleans and files onto the clean list.  I can't let it be dirty again until the
3135 ** transaction is committed.
3136 **
3137 ** if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len.
3138 */
3139 int journal_mark_dirty(struct reiserfs_transaction_handle *th,
3140                        struct super_block *p_s_sb, struct buffer_head *bh)
3141 {
3142         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3143         struct reiserfs_journal_cnode *cn = NULL;
3144         int count_already_incd = 0;
3145         int prepared = 0;
3146         BUG_ON(!th->t_trans_id);
3147
3148         PROC_INFO_INC(p_s_sb, journal.mark_dirty);
3149         if (th->t_trans_id != journal->j_trans_id) {
3150                 reiserfs_panic(th->t_super,
3151                                "journal-1577: handle trans id %ld != current trans id %ld\n",
3152                                th->t_trans_id, journal->j_trans_id);
3153         }
3154
3155         p_s_sb->s_dirt = 1;
3156
3157         prepared = test_clear_buffer_journal_prepared(bh);
3158         clear_buffer_journal_restore_dirty(bh);
3159         /* already in this transaction, we are done */
3160         if (buffer_journaled(bh)) {
3161                 PROC_INFO_INC(p_s_sb, journal.mark_dirty_already);
3162                 return 0;
3163         }
3164
3165         /* this must be turned into a panic instead of a warning.  We can't allow
3166          ** a dirty or journal_dirty or locked buffer to be logged, as some changes
3167          ** could get to disk too early.  NOT GOOD.
3168          */
3169         if (!prepared || buffer_dirty(bh)) {
3170                 reiserfs_warning(p_s_sb, "journal-1777: buffer %llu bad state "
3171                                  "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT",
3172                                  (unsigned long long)bh->b_blocknr,
3173                                  prepared ? ' ' : '!',
3174                                  buffer_locked(bh) ? ' ' : '!',
3175                                  buffer_dirty(bh) ? ' ' : '!',
3176                                  buffer_journal_dirty(bh) ? ' ' : '!');
3177         }
3178
3179         if (atomic_read(&(journal->j_wcount)) <= 0) {
3180                 reiserfs_warning(p_s_sb,
3181                                  "journal-1409: journal_mark_dirty returning because j_wcount was %d",
3182                                  atomic_read(&(journal->j_wcount)));
3183                 return 1;
3184         }
3185         /* this error means I've screwed up, and we've overflowed the transaction.
3186          ** Nothing can be done here, except make the FS readonly or panic.
3187          */
3188         if (journal->j_len >= journal->j_trans_max) {
3189                 reiserfs_panic(th->t_super,
3190                                "journal-1413: journal_mark_dirty: j_len (%lu) is too big\n",
3191                                journal->j_len);
3192         }
3193
3194         if (buffer_journal_dirty(bh)) {
3195                 count_already_incd = 1;
3196                 PROC_INFO_INC(p_s_sb, journal.mark_dirty_notjournal);
3197                 clear_buffer_journal_dirty(bh);
3198         }
3199
3200         if (journal->j_len > journal->j_len_alloc) {
3201                 journal->j_len_alloc = journal->j_len + JOURNAL_PER_BALANCE_CNT;
3202         }
3203
3204         set_buffer_journaled(bh);
3205
3206         /* now put this guy on the end */
3207         if (!cn) {
3208                 cn = get_cnode(p_s_sb);
3209                 if (!cn) {
3210                         reiserfs_panic(p_s_sb, "get_cnode failed!\n");
3211                 }
3212
3213                 if (th->t_blocks_logged == th->t_blocks_allocated) {
3214                         th->t_blocks_allocated += JOURNAL_PER_BALANCE_CNT;
3215                         journal->j_len_alloc += JOURNAL_PER_BALANCE_CNT;
3216                 }
3217                 th->t_blocks_logged++;
3218                 journal->j_len++;
3219
3220                 cn->bh = bh;
3221                 cn->blocknr = bh->b_blocknr;
3222                 cn->sb = p_s_sb;
3223                 cn->jlist = NULL;
3224                 insert_journal_hash(journal->j_hash_table, cn);
3225                 if (!count_already_incd) {
3226                         get_bh(bh);
3227                 }
3228         }
3229         cn->next = NULL;
3230         cn->prev = journal->j_last;
3231         cn->bh = bh;
3232         if (journal->j_last) {
3233                 journal->j_last->next = cn;
3234                 journal->j_last = cn;
3235         } else {
3236                 journal->j_first = cn;
3237                 journal->j_last = cn;
3238         }
3239         return 0;
3240 }
3241
3242 int journal_end(struct reiserfs_transaction_handle *th,
3243                 struct super_block *p_s_sb, unsigned long nblocks)
3244 {
3245         if (!current->journal_info && th->t_refcount > 1)
3246                 reiserfs_warning(p_s_sb, "REISER-NESTING: th NULL, refcount %d",
3247                                  th->t_refcount);
3248
3249         if (!th->t_trans_id) {
3250                 WARN_ON(1);
3251                 return -EIO;
3252         }
3253
3254         th->t_refcount--;
3255         if (th->t_refcount > 0) {
3256                 struct reiserfs_transaction_handle *cur_th =
3257                     current->journal_info;
3258
3259                 /* we aren't allowed to close a nested transaction on a different
3260                  ** filesystem from the one in the task struct
3261                  */
3262                 if (cur_th->t_super != th->t_super)
3263                         BUG();
3264
3265                 if (th != cur_th) {
3266                         memcpy(current->journal_info, th, sizeof(*th));
3267                         th->t_trans_id = 0;
3268                 }
3269                 return 0;
3270         } else {
3271                 return do_journal_end(th, p_s_sb, nblocks, 0);
3272         }
3273 }
3274
3275 /* removes from the current transaction, relsing and descrementing any counters.
3276 ** also files the removed buffer directly onto the clean list
3277 **
3278 ** called by journal_mark_freed when a block has been deleted
3279 **
3280 ** returns 1 if it cleaned and relsed the buffer. 0 otherwise
3281 */
3282 static int remove_from_transaction(struct super_block *p_s_sb,
3283                                    b_blocknr_t blocknr, int already_cleaned)
3284 {
3285         struct buffer_head *bh;
3286         struct reiserfs_journal_cnode *cn;
3287         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3288         int ret = 0;
3289
3290         cn = get_journal_hash_dev(p_s_sb, journal->j_hash_table, blocknr);
3291         if (!cn || !cn->bh) {
3292                 return ret;
3293         }
3294         bh = cn->bh;
3295         if (cn->prev) {
3296                 cn->prev->next = cn->next;
3297         }
3298         if (cn->next) {
3299                 cn->next->prev = cn->prev;
3300         }
3301         if (cn == journal->j_first) {
3302                 journal->j_first = cn->next;
3303         }
3304         if (cn == journal->j_last) {
3305                 journal->j_last = cn->prev;
3306         }
3307         if (bh)
3308                 remove_journal_hash(p_s_sb, journal->j_hash_table, NULL,
3309                                     bh->b_blocknr, 0);
3310         clear_buffer_journaled(bh);     /* don't log this one */
3311
3312         if (!already_cleaned) {
3313                 clear_buffer_journal_dirty(bh);
3314                 clear_buffer_dirty(bh);
3315                 clear_buffer_journal_test(bh);
3316                 put_bh(bh);
3317                 if (atomic_read(&(bh->b_count)) < 0) {
3318                         reiserfs_warning(p_s_sb,
3319                                          "journal-1752: remove from trans, b_count < 0");
3320                 }
3321                 ret = 1;
3322         }
3323         journal->j_len--;
3324         journal->j_len_alloc--;
3325         free_cnode(p_s_sb, cn);
3326         return ret;
3327 }
3328
3329 /*
3330 ** for any cnode in a journal list, it can only be dirtied of all the
3331 ** transactions that include it are commited to disk.
3332 ** this checks through each transaction, and returns 1 if you are allowed to dirty,
3333 ** and 0 if you aren't
3334 **
3335 ** it is called by dirty_journal_list, which is called after flush_commit_list has gotten all the log
3336 ** blocks for a given transaction on disk
3337 **
3338 */
3339 static int can_dirty(struct reiserfs_journal_cnode *cn)
3340 {
3341         struct super_block *sb = cn->sb;
3342         b_blocknr_t blocknr = cn->blocknr;
3343         struct reiserfs_journal_cnode *cur = cn->hprev;
3344         int can_dirty = 1;
3345
3346         /* first test hprev.  These are all newer than cn, so any node here
3347          ** with the same block number and dev means this node can't be sent
3348          ** to disk right now.
3349          */
3350         while (cur && can_dirty) {
3351                 if (cur->jlist && cur->bh && cur->blocknr && cur->sb == sb &&
3352                     cur->blocknr == blocknr) {
3353                         can_dirty = 0;
3354                 }
3355                 cur = cur->hprev;
3356         }
3357         /* then test hnext.  These are all older than cn.  As long as they
3358          ** are committed to the log, it is safe to write cn to disk
3359          */
3360         cur = cn->hnext;
3361         while (cur && can_dirty) {
3362                 if (cur->jlist && cur->jlist->j_len > 0 &&
3363                     atomic_read(&(cur->jlist->j_commit_left)) > 0 && cur->bh &&
3364                     cur->blocknr && cur->sb == sb && cur->blocknr == blocknr) {
3365                         can_dirty = 0;
3366                 }
3367                 cur = cur->hnext;
3368         }
3369         return can_dirty;
3370 }
3371
3372 /* syncs the commit blocks, but does not force the real buffers to disk
3373 ** will wait until the current transaction is done/commited before returning
3374 */
3375 int journal_end_sync(struct reiserfs_transaction_handle *th,
3376                      struct super_block *p_s_sb, unsigned long nblocks)
3377 {
3378         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3379
3380         BUG_ON(!th->t_trans_id);
3381         /* you can sync while nested, very, very bad */
3382         if (th->t_refcount > 1) {
3383                 BUG();
3384         }
3385         if (journal->j_len == 0) {
3386                 reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb),
3387                                              1);
3388                 journal_mark_dirty(th, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb));
3389         }
3390         return do_journal_end(th, p_s_sb, nblocks, COMMIT_NOW | WAIT);
3391 }
3392
3393 /*
3394 ** writeback the pending async commits to disk
3395 */
3396 static void flush_async_commits(void *p)
3397 {
3398         struct super_block *p_s_sb = p;
3399         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3400         struct reiserfs_journal_list *jl;
3401         struct list_head *entry;
3402
3403         lock_kernel();
3404         if (!list_empty(&journal->j_journal_list)) {
3405                 /* last entry is the youngest, commit it and you get everything */
3406                 entry = journal->j_journal_list.prev;
3407                 jl = JOURNAL_LIST_ENTRY(entry);
3408                 flush_commit_list(p_s_sb, jl, 1);
3409         }
3410         unlock_kernel();
3411         /*
3412          * this is a little racey, but there's no harm in missing
3413          * the filemap_fdata_write
3414          */
3415         if (!atomic_read(&journal->j_async_throttle)
3416             && !reiserfs_is_journal_aborted(journal)) {
3417                 atomic_inc(&journal->j_async_throttle);
3418                 filemap_fdatawrite(p_s_sb->s_bdev->bd_inode->i_mapping);
3419                 atomic_dec(&journal->j_async_throttle);
3420         }
3421 }
3422
3423 /*
3424 ** flushes any old transactions to disk
3425 ** ends the current transaction if it is too old
3426 */
3427 int reiserfs_flush_old_commits(struct super_block *p_s_sb)
3428 {
3429         time_t now;
3430         struct reiserfs_transaction_handle th;
3431         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3432
3433         now = get_seconds();
3434         /* safety check so we don't flush while we are replaying the log during
3435          * mount
3436          */
3437         if (list_empty(&journal->j_journal_list)) {
3438                 return 0;
3439         }
3440
3441         /* check the current transaction.  If there are no writers, and it is
3442          * too old, finish it, and force the commit blocks to disk
3443          */
3444         if (atomic_read(&journal->j_wcount) <= 0 &&
3445             journal->j_trans_start_time > 0 &&
3446             journal->j_len > 0 &&
3447             (now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3448                 if (!journal_join(&th, p_s_sb, 1)) {
3449                         reiserfs_prepare_for_journal(p_s_sb,
3450                                                      SB_BUFFER_WITH_SB(p_s_sb),
3451                                                      1);
3452                         journal_mark_dirty(&th, p_s_sb,
3453                                            SB_BUFFER_WITH_SB(p_s_sb));
3454
3455                         /* we're only being called from kreiserfsd, it makes no sense to do
3456                          ** an async commit so that kreiserfsd can do it later
3457                          */
3458                         do_journal_end(&th, p_s_sb, 1, COMMIT_NOW | WAIT);
3459                 }
3460         }
3461         return p_s_sb->s_dirt;
3462 }
3463
3464 /*
3465 ** returns 0 if do_journal_end should return right away, returns 1 if do_journal_end should finish the commit
3466 **
3467 ** if the current transaction is too old, but still has writers, this will wait on j_join_wait until all
3468 ** the writers are done.  By the time it wakes up, the transaction it was called has already ended, so it just
3469 ** flushes the commit list and returns 0.
3470 **
3471 ** Won't batch when flush or commit_now is set.  Also won't batch when others are waiting on j_join_wait.
3472 **
3473 ** Note, we can't allow the journal_end to proceed while there are still writers in the log.
3474 */
3475 static int check_journal_end(struct reiserfs_transaction_handle *th,
3476                              struct super_block *p_s_sb, unsigned long nblocks,
3477                              int flags)
3478 {
3479
3480         time_t now;
3481         int flush = flags & FLUSH_ALL;
3482         int commit_now = flags & COMMIT_NOW;
3483         int wait_on_commit = flags & WAIT;
3484         struct reiserfs_journal_list *jl;
3485         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3486
3487         BUG_ON(!th->t_trans_id);
3488
3489         if (th->t_trans_id != journal->j_trans_id) {
3490                 reiserfs_panic(th->t_super,
3491                                "journal-1577: handle trans id %ld != current trans id %ld\n",
3492                                th->t_trans_id, journal->j_trans_id);
3493         }
3494
3495         journal->j_len_alloc -= (th->t_blocks_allocated - th->t_blocks_logged);
3496         if (atomic_read(&(journal->j_wcount)) > 0) {    /* <= 0 is allowed.  unmounting might not call begin */
3497                 atomic_dec(&(journal->j_wcount));
3498         }
3499
3500         /* BUG, deal with case where j_len is 0, but people previously freed blocks need to be released
3501          ** will be dealt with by next transaction that actually writes something, but should be taken
3502          ** care of in this trans
3503          */
3504         if (journal->j_len == 0) {
3505                 BUG();
3506         }
3507         /* if wcount > 0, and we are called to with flush or commit_now,
3508          ** we wait on j_join_wait.  We will wake up when the last writer has
3509          ** finished the transaction, and started it on its way to the disk.
3510          ** Then, we flush the commit or journal list, and just return 0
3511          ** because the rest of journal end was already done for this transaction.
3512          */
3513         if (atomic_read(&(journal->j_wcount)) > 0) {
3514                 if (flush || commit_now) {
3515                         unsigned trans_id;
3516
3517                         jl = journal->j_current_jl;
3518                         trans_id = jl->j_trans_id;
3519                         if (wait_on_commit)
3520                                 jl->j_state |= LIST_COMMIT_PENDING;
3521                         atomic_set(&(journal->j_jlock), 1);
3522                         if (flush) {
3523                                 journal->j_next_full_flush = 1;
3524                         }
3525                         unlock_journal(p_s_sb);
3526
3527                         /* sleep while the current transaction is still j_jlocked */
3528                         while (journal->j_trans_id == trans_id) {
3529                                 if (atomic_read(&journal->j_jlock)) {
3530                                         queue_log_writer(p_s_sb);
3531                                 } else {
3532                                         lock_journal(p_s_sb);
3533                                         if (journal->j_trans_id == trans_id) {
3534                                                 atomic_set(&(journal->j_jlock),
3535                                                            1);
3536                                         }
3537                                         unlock_journal(p_s_sb);
3538                                 }
3539                         }
3540                         if (journal->j_trans_id == trans_id) {
3541                                 BUG();
3542                         }
3543                         if (commit_now
3544                             && journal_list_still_alive(p_s_sb, trans_id)
3545                             && wait_on_commit) {
3546                                 flush_commit_list(p_s_sb, jl, 1);
3547                         }
3548                         return 0;
3549                 }
3550                 unlock_journal(p_s_sb);
3551                 return 0;
3552         }
3553
3554         /* deal with old transactions where we are the last writers */
3555         now = get_seconds();
3556         if ((now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3557                 commit_now = 1;
3558                 journal->j_next_async_flush = 1;
3559         }
3560         /* don't batch when someone is waiting on j_join_wait */
3561         /* don't batch when syncing the commit or flushing the whole trans */
3562         if (!(journal->j_must_wait > 0) && !(atomic_read(&(journal->j_jlock)))
3563             && !flush && !commit_now && (journal->j_len < journal->j_max_batch)
3564             && journal->j_len_alloc < journal->j_max_batch
3565             && journal->j_cnode_free > (journal->j_trans_max * 3)) {
3566                 journal->j_bcount++;
3567                 unlock_journal(p_s_sb);
3568                 return 0;
3569         }
3570
3571         if (journal->j_start > SB_ONDISK_JOURNAL_SIZE(p_s_sb)) {
3572                 reiserfs_panic(p_s_sb,
3573                                "journal-003: journal_end: j_start (%ld) is too high\n",
3574                                journal->j_start);
3575         }
3576         return 1;
3577 }
3578
3579 /*
3580 ** Does all the work that makes deleting blocks safe.
3581 ** when deleting a block mark BH_JNew, just remove it from the current transaction, clean it's buffer_head and move on.
3582 **
3583 ** otherwise:
3584 ** set a bit for the block in the journal bitmap.  That will prevent it from being allocated for unformatted nodes
3585 ** before this transaction has finished.
3586 **
3587 ** mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers.  That will prevent any old transactions with
3588 ** this block from trying to flush to the real location.  Since we aren't removing the cnode from the journal_list_hash,
3589 ** the block can't be reallocated yet.
3590 **
3591 ** Then remove it from the current transaction, decrementing any counters and filing it on the clean list.
3592 */
3593 int journal_mark_freed(struct reiserfs_transaction_handle *th,
3594                        struct super_block *p_s_sb, b_blocknr_t blocknr)
3595 {
3596         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3597         struct reiserfs_journal_cnode *cn = NULL;
3598         struct buffer_head *bh = NULL;
3599         struct reiserfs_list_bitmap *jb = NULL;
3600         int cleaned = 0;
3601         BUG_ON(!th->t_trans_id);
3602
3603         cn = get_journal_hash_dev(p_s_sb, journal->j_hash_table, blocknr);
3604         if (cn && cn->bh) {
3605                 bh = cn->bh;
3606                 get_bh(bh);
3607         }
3608         /* if it is journal new, we just remove it from this transaction */
3609         if (bh && buffer_journal_new(bh)) {
3610                 clear_buffer_journal_new(bh);
3611                 clear_prepared_bits(bh);
3612                 reiserfs_clean_and_file_buffer(bh);
3613                 cleaned = remove_from_transaction(p_s_sb, blocknr, cleaned);
3614         } else {
3615                 /* set the bit for this block in the journal bitmap for this transaction */
3616                 jb = journal->j_current_jl->j_list_bitmap;
3617                 if (!jb) {
3618                         reiserfs_panic(p_s_sb,
3619                                        "journal-1702: journal_mark_freed, journal_list_bitmap is NULL\n");
3620                 }
3621                 set_bit_in_list_bitmap(p_s_sb, blocknr, jb);
3622
3623                 /* Note, the entire while loop is not allowed to schedule.  */
3624
3625                 if (bh) {
3626                         clear_prepared_bits(bh);
3627                         reiserfs_clean_and_file_buffer(bh);
3628                 }
3629                 cleaned = remove_from_transaction(p_s_sb, blocknr, cleaned);
3630
3631                 /* find all older transactions with this block, make sure they don't try to write it out */
3632                 cn = get_journal_hash_dev(p_s_sb, journal->j_list_hash_table,
3633                                           blocknr);
3634                 while (cn) {
3635                         if (p_s_sb == cn->sb && blocknr == cn->blocknr) {
3636                                 set_bit(BLOCK_FREED, &cn->state);
3637                                 if (cn->bh) {
3638                                         if (!cleaned) {
3639                                                 /* remove_from_transaction will brelse the buffer if it was
3640                                                  ** in the current trans
3641                                                  */
3642                                                 clear_buffer_journal_dirty(cn->
3643                                                                            bh);
3644                                                 clear_buffer_dirty(cn->bh);
3645                                                 clear_buffer_journal_test(cn->
3646                                                                           bh);
3647                                                 cleaned = 1;
3648                                                 put_bh(cn->bh);
3649                                                 if (atomic_read
3650                                                     (&(cn->bh->b_count)) < 0) {
3651                                                         reiserfs_warning(p_s_sb,
3652                                                                          "journal-2138: cn->bh->b_count < 0");
3653                                                 }
3654                                         }
3655                                         if (cn->jlist) {        /* since we are clearing the bh, we MUST dec nonzerolen */
3656                                                 atomic_dec(&
3657                                                            (cn->jlist->
3658                                                             j_nonzerolen));
3659                                         }
3660                                         cn->bh = NULL;
3661                                 }
3662                         }
3663                         cn = cn->hnext;
3664                 }
3665         }
3666
3667         if (bh) {
3668                 put_bh(bh);     /* get_hash grabs the buffer */
3669                 if (atomic_read(&(bh->b_count)) < 0) {
3670                         reiserfs_warning(p_s_sb,
3671                                          "journal-2165: bh->b_count < 0");
3672                 }
3673         }
3674         return 0;
3675 }
3676
3677 void reiserfs_update_inode_transaction(struct inode *inode)
3678 {
3679         struct reiserfs_journal *journal = SB_JOURNAL(inode->i_sb);
3680         REISERFS_I(inode)->i_jl = journal->j_current_jl;
3681         REISERFS_I(inode)->i_trans_id = journal->j_trans_id;
3682 }
3683
3684 /*
3685  * returns -1 on error, 0 if no commits/barriers were done and 1
3686  * if a transaction was actually committed and the barrier was done
3687  */
3688 static int __commit_trans_jl(struct inode *inode, unsigned long id,
3689                              struct reiserfs_journal_list *jl)
3690 {
3691         struct reiserfs_transaction_handle th;
3692         struct super_block *sb = inode->i_sb;
3693         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3694         int ret = 0;
3695
3696         /* is it from the current transaction, or from an unknown transaction? */
3697         if (id == journal->j_trans_id) {
3698                 jl = journal->j_current_jl;
3699                 /* try to let other writers come in and grow this transaction */
3700                 let_transaction_grow(sb, id);
3701                 if (journal->j_trans_id != id) {
3702                         goto flush_commit_only;
3703                 }
3704
3705                 ret = journal_begin(&th, sb, 1);
3706                 if (ret)
3707                         return ret;
3708
3709                 /* someone might have ended this transaction while we joined */
3710                 if (journal->j_trans_id != id) {
3711                         reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3712                                                      1);
3713                         journal_mark_dirty(&th, sb, SB_BUFFER_WITH_SB(sb));
3714                         ret = journal_end(&th, sb, 1);
3715                         goto flush_commit_only;
3716                 }
3717
3718                 ret = journal_end_sync(&th, sb, 1);
3719                 if (!ret)
3720                         ret = 1;
3721
3722         } else {
3723                 /* this gets tricky, we have to make sure the journal list in
3724                  * the inode still exists.  We know the list is still around
3725                  * if we've got a larger transaction id than the oldest list
3726                  */
3727               flush_commit_only:
3728                 if (journal_list_still_alive(inode->i_sb, id)) {
3729                         /*
3730                          * we only set ret to 1 when we know for sure
3731                          * the barrier hasn't been started yet on the commit
3732                          * block.
3733                          */
3734                         if (atomic_read(&jl->j_commit_left) > 1)
3735                                 ret = 1;
3736                         flush_commit_list(sb, jl, 1);
3737                         if (journal->j_errno)
3738                                 ret = journal->j_errno;
3739                 }
3740         }
3741         /* otherwise the list is gone, and long since committed */
3742         return ret;
3743 }
3744
3745 int reiserfs_commit_for_inode(struct inode *inode)
3746 {
3747         unsigned long id = REISERFS_I(inode)->i_trans_id;
3748         struct reiserfs_journal_list *jl = REISERFS_I(inode)->i_jl;
3749
3750         /* for the whole inode, assume unset id means it was
3751          * changed in the current transaction.  More conservative
3752          */
3753         if (!id || !jl) {
3754                 reiserfs_update_inode_transaction(inode);
3755                 id = REISERFS_I(inode)->i_trans_id;
3756                 /* jl will be updated in __commit_trans_jl */
3757         }
3758
3759         return __commit_trans_jl(inode, id, jl);
3760 }
3761
3762 void reiserfs_restore_prepared_buffer(struct super_block *p_s_sb,
3763                                       struct buffer_head *bh)
3764 {
3765         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3766         PROC_INFO_INC(p_s_sb, journal.restore_prepared);
3767         if (!bh) {
3768                 return;
3769         }
3770         if (test_clear_buffer_journal_restore_dirty(bh) &&
3771             buffer_journal_dirty(bh)) {
3772                 struct reiserfs_journal_cnode *cn;
3773                 cn = get_journal_hash_dev(p_s_sb,
3774                                           journal->j_list_hash_table,
3775                                           bh->b_blocknr);
3776                 if (cn && can_dirty(cn)) {
3777                         set_buffer_journal_test(bh);
3778                         mark_buffer_dirty(bh);
3779                 }
3780         }
3781         clear_buffer_journal_prepared(bh);
3782 }
3783
3784 extern struct tree_balance *cur_tb;
3785 /*
3786 ** before we can change a metadata block, we have to make sure it won't
3787 ** be written to disk while we are altering it.  So, we must:
3788 ** clean it
3789 ** wait on it.
3790 **
3791 */
3792 int reiserfs_prepare_for_journal(struct super_block *p_s_sb,
3793                                  struct buffer_head *bh, int wait)
3794 {
3795         PROC_INFO_INC(p_s_sb, journal.prepare);
3796
3797         if (test_set_buffer_locked(bh)) {
3798                 if (!wait)
3799                         return 0;
3800                 lock_buffer(bh);
3801         }
3802         set_buffer_journal_prepared(bh);
3803         if (test_clear_buffer_dirty(bh) && buffer_journal_dirty(bh)) {
3804                 clear_buffer_journal_test(bh);
3805                 set_buffer_journal_restore_dirty(bh);
3806         }
3807         unlock_buffer(bh);
3808         return 1;
3809 }
3810
3811 static void flush_old_journal_lists(struct super_block *s)
3812 {
3813         struct reiserfs_journal *journal = SB_JOURNAL(s);
3814         struct reiserfs_journal_list *jl;
3815         struct list_head *entry;
3816         time_t now = get_seconds();
3817
3818         while (!list_empty(&journal->j_journal_list)) {
3819                 entry = journal->j_journal_list.next;
3820                 jl = JOURNAL_LIST_ENTRY(entry);
3821                 /* this check should always be run, to send old lists to disk */
3822                 if (jl->j_timestamp < (now - (JOURNAL_MAX_TRANS_AGE * 4))) {
3823                         flush_used_journal_lists(s, jl);
3824                 } else {
3825                         break;
3826                 }
3827         }
3828 }
3829
3830 /*
3831 ** long and ugly.  If flush, will not return until all commit
3832 ** blocks and all real buffers in the trans are on disk.
3833 ** If no_async, won't return until all commit blocks are on disk.
3834 **
3835 ** keep reading, there are comments as you go along
3836 **
3837 ** If the journal is aborted, we just clean up. Things like flushing
3838 ** journal lists, etc just won't happen.
3839 */
3840 static int do_journal_end(struct reiserfs_transaction_handle *th,
3841                           struct super_block *p_s_sb, unsigned long nblocks,
3842                           int flags)
3843 {
3844         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3845         struct reiserfs_journal_cnode *cn, *next, *jl_cn;
3846         struct reiserfs_journal_cnode *last_cn = NULL;
3847         struct reiserfs_journal_desc *desc;
3848         struct reiserfs_journal_commit *commit;
3849         struct buffer_head *c_bh;       /* commit bh */
3850         struct buffer_head *d_bh;       /* desc bh */
3851         int cur_write_start = 0;        /* start index of current log write */
3852         int old_start;
3853         int i;
3854         int flush = flags & FLUSH_ALL;
3855         int wait_on_commit = flags & WAIT;
3856         struct reiserfs_journal_list *jl, *temp_jl;
3857         struct list_head *entry, *safe;
3858         unsigned long jindex;
3859         unsigned long commit_trans_id;
3860         int trans_half;
3861
3862         BUG_ON(th->t_refcount > 1);
3863         BUG_ON(!th->t_trans_id);
3864
3865         put_fs_excl();
3866         current->journal_info = th->t_handle_save;
3867         reiserfs_check_lock_depth(p_s_sb, "journal end");
3868         if (journal->j_len == 0) {
3869                 reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb),
3870                                              1);
3871                 journal_mark_dirty(th, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb));
3872         }
3873
3874         lock_journal(p_s_sb);
3875         if (journal->j_next_full_flush) {
3876                 flags |= FLUSH_ALL;
3877                 flush = 1;
3878         }
3879         if (journal->j_next_async_flush) {
3880                 flags |= COMMIT_NOW | WAIT;
3881                 wait_on_commit = 1;
3882         }
3883
3884         /* check_journal_end locks the journal, and unlocks if it does not return 1
3885          ** it tells us if we should continue with the journal_end, or just return
3886          */
3887         if (!check_journal_end(th, p_s_sb, nblocks, flags)) {
3888                 p_s_sb->s_dirt = 1;
3889                 wake_queued_writers(p_s_sb);
3890                 reiserfs_async_progress_wait(p_s_sb);
3891                 goto out;
3892         }
3893
3894         /* check_journal_end might set these, check again */
3895         if (journal->j_next_full_flush) {
3896                 flush = 1;
3897         }
3898
3899         /*
3900          ** j must wait means we have to flush the log blocks, and the real blocks for
3901          ** this transaction
3902          */
3903         if (journal->j_must_wait > 0) {
3904                 flush = 1;
3905         }
3906 #ifdef REISERFS_PREALLOCATE
3907         /* quota ops might need to nest, setup the journal_info pointer for them
3908          * and raise the refcount so that it is > 0. */
3909         current->journal_info = th;
3910         th->t_refcount++;
3911         reiserfs_discard_all_prealloc(th);      /* it should not involve new blocks into
3912                                                  * the transaction */
3913         th->t_refcount--;
3914         current->journal_info = th->t_handle_save;
3915 #endif
3916
3917         /* setup description block */
3918         d_bh =
3919             journal_getblk(p_s_sb,
3920                            SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
3921                            journal->j_start);
3922         set_buffer_uptodate(d_bh);
3923         desc = (struct reiserfs_journal_desc *)(d_bh)->b_data;
3924         memset(d_bh->b_data, 0, d_bh->b_size);
3925         memcpy(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8);
3926         set_desc_trans_id(desc, journal->j_trans_id);
3927
3928         /* setup commit block.  Don't write (keep it clean too) this one until after everyone else is written */
3929         c_bh = journal_getblk(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
3930                               ((journal->j_start + journal->j_len +
3931                                 1) % SB_ONDISK_JOURNAL_SIZE(p_s_sb)));
3932         commit = (struct reiserfs_journal_commit *)c_bh->b_data;
3933         memset(c_bh->b_data, 0, c_bh->b_size);
3934         set_commit_trans_id(commit, journal->j_trans_id);
3935         set_buffer_uptodate(c_bh);
3936
3937         /* init this journal list */
3938         jl = journal->j_current_jl;
3939
3940         /* we lock the commit before doing anything because
3941          * we want to make sure nobody tries to run flush_commit_list until
3942          * the new transaction is fully setup, and we've already flushed the
3943          * ordered bh list
3944          */
3945         down(&jl->j_commit_lock);
3946
3947         /* save the transaction id in case we need to commit it later */
3948         commit_trans_id = jl->j_trans_id;
3949
3950         atomic_set(&jl->j_older_commits_done, 0);
3951         jl->j_trans_id = journal->j_trans_id;
3952         jl->j_timestamp = journal->j_trans_start_time;
3953         jl->j_commit_bh = c_bh;
3954         jl->j_start = journal->j_start;
3955         jl->j_len = journal->j_len;
3956         atomic_set(&jl->j_nonzerolen, journal->j_len);
3957         atomic_set(&jl->j_commit_left, journal->j_len + 2);
3958         jl->j_realblock = NULL;
3959
3960         /* The ENTIRE FOR LOOP MUST not cause schedule to occur.
3961          **  for each real block, add it to the journal list hash,
3962          ** copy into real block index array in the commit or desc block
3963          */
3964         trans_half = journal_trans_half(p_s_sb->s_blocksize);
3965         for (i = 0, cn = journal->j_first; cn; cn = cn->next, i++) {
3966                 if (buffer_journaled(cn->bh)) {
3967                         jl_cn = get_cnode(p_s_sb);
3968                         if (!jl_cn) {
3969                                 reiserfs_panic(p_s_sb,
3970                                                "journal-1676, get_cnode returned NULL\n");
3971                         }
3972                         if (i == 0) {
3973                                 jl->j_realblock = jl_cn;
3974                         }
3975                         jl_cn->prev = last_cn;
3976                         jl_cn->next = NULL;
3977                         if (last_cn) {
3978                                 last_cn->next = jl_cn;
3979                         }
3980                         last_cn = jl_cn;
3981                         /* make sure the block we are trying to log is not a block
3982                            of journal or reserved area */
3983
3984                         if (is_block_in_log_or_reserved_area
3985                             (p_s_sb, cn->bh->b_blocknr)) {
3986                                 reiserfs_panic(p_s_sb,
3987                                                "journal-2332: Trying to log block %lu, which is a log block\n",
3988                                                cn->bh->b_blocknr);
3989                         }
3990                         jl_cn->blocknr = cn->bh->b_blocknr;
3991                         jl_cn->state = 0;
3992                         jl_cn->sb = p_s_sb;
3993                         jl_cn->bh = cn->bh;
3994                         jl_cn->jlist = jl;
3995                         insert_journal_hash(journal->j_list_hash_table, jl_cn);
3996                         if (i < trans_half) {
3997                                 desc->j_realblock[i] =
3998                                     cpu_to_le32(cn->bh->b_blocknr);
3999                         } else {
4000                                 commit->j_realblock[i - trans_half] =
4001                                     cpu_to_le32(cn->bh->b_blocknr);
4002                         }
4003                 } else {
4004                         i--;
4005                 }
4006         }
4007         set_desc_trans_len(desc, journal->j_len);
4008         set_desc_mount_id(desc, journal->j_mount_id);
4009         set_desc_trans_id(desc, journal->j_trans_id);
4010         set_commit_trans_len(commit, journal->j_len);
4011
4012         /* special check in case all buffers in the journal were marked for not logging */
4013         if (journal->j_len == 0) {
4014                 BUG();
4015         }
4016
4017         /* we're about to dirty all the log blocks, mark the description block
4018          * dirty now too.  Don't mark the commit block dirty until all the
4019          * others are on disk
4020          */
4021         mark_buffer_dirty(d_bh);
4022
4023         /* first data block is j_start + 1, so add one to cur_write_start wherever you use it */
4024         cur_write_start = journal->j_start;
4025         cn = journal->j_first;
4026         jindex = 1;             /* start at one so we don't get the desc again */
4027         while (cn) {
4028                 clear_buffer_journal_new(cn->bh);
4029                 /* copy all the real blocks into log area.  dirty log blocks */
4030                 if (buffer_journaled(cn->bh)) {
4031                         struct buffer_head *tmp_bh;
4032                         char *addr;
4033                         struct page *page;
4034                         tmp_bh =
4035                             journal_getblk(p_s_sb,
4036                                            SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
4037                                            ((cur_write_start +
4038                                              jindex) %
4039                                             SB_ONDISK_JOURNAL_SIZE(p_s_sb)));
4040                         set_buffer_uptodate(tmp_bh);
4041                         page = cn->bh->b_page;
4042                         addr = kmap(page);
4043                         memcpy(tmp_bh->b_data,
4044                                addr + offset_in_page(cn->bh->b_data),
4045                                cn->bh->b_size);
4046                         kunmap(page);
4047                         mark_buffer_dirty(tmp_bh);
4048                         jindex++;
4049                         set_buffer_journal_dirty(cn->bh);
4050                         clear_buffer_journaled(cn->bh);
4051                 } else {
4052                         /* JDirty cleared sometime during transaction.  don't log this one */
4053                         reiserfs_warning(p_s_sb,
4054                                          "journal-2048: do_journal_end: BAD, buffer in journal hash, but not JDirty!");
4055                         brelse(cn->bh);
4056                 }
4057                 next = cn->next;
4058                 free_cnode(p_s_sb, cn);
4059                 cn = next;
4060                 cond_resched();
4061         }
4062
4063         /* we are done  with both the c_bh and d_bh, but
4064          ** c_bh must be written after all other commit blocks,
4065          ** so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1.
4066          */
4067
4068         journal->j_current_jl = alloc_journal_list(p_s_sb);
4069
4070         /* now it is safe to insert this transaction on the main list */
4071         list_add_tail(&jl->j_list, &journal->j_journal_list);
4072         list_add_tail(&jl->j_working_list, &journal->j_working_list);
4073         journal->j_num_work_lists++;
4074
4075         /* reset journal values for the next transaction */
4076         old_start = journal->j_start;
4077         journal->j_start =
4078             (journal->j_start + journal->j_len +
4079              2) % SB_ONDISK_JOURNAL_SIZE(p_s_sb);
4080         atomic_set(&(journal->j_wcount), 0);
4081         journal->j_bcount = 0;
4082         journal->j_last = NULL;
4083         journal->j_first = NULL;
4084         journal->j_len = 0;
4085         journal->j_trans_start_time = 0;
4086         journal->j_trans_id++;
4087         journal->j_current_jl->j_trans_id = journal->j_trans_id;
4088         journal->j_must_wait = 0;
4089         journal->j_len_alloc = 0;
4090         journal->j_next_full_flush = 0;
4091         journal->j_next_async_flush = 0;
4092         init_journal_hash(p_s_sb);
4093
4094         // make sure reiserfs_add_jh sees the new current_jl before we
4095         // write out the tails
4096         smp_mb();
4097
4098         /* tail conversion targets have to hit the disk before we end the
4099          * transaction.  Otherwise a later transaction might repack the tail
4100          * before this transaction commits, leaving the data block unflushed and
4101          * clean, if we crash before the later transaction commits, the data block
4102          * is lost.
4103          */
4104         if (!list_empty(&jl->j_tail_bh_list)) {
4105                 unlock_kernel();
4106                 write_ordered_buffers(&journal->j_dirty_buffers_lock,
4107                                       journal, jl, &jl->j_tail_bh_list);
4108                 lock_kernel();
4109         }
4110         if (!list_empty(&jl->j_tail_bh_list))
4111                 BUG();
4112         up(&jl->j_commit_lock);
4113
4114         /* honor the flush wishes from the caller, simple commits can
4115          ** be done outside the journal lock, they are done below
4116          **
4117          ** if we don't flush the commit list right now, we put it into
4118          ** the work queue so the people waiting on the async progress work
4119          ** queue don't wait for this proc to flush journal lists and such.
4120          */
4121         if (flush) {
4122                 flush_commit_list(p_s_sb, jl, 1);
4123                 flush_journal_list(p_s_sb, jl, 1);
4124         } else if (!(jl->j_state & LIST_COMMIT_PENDING))
4125                 queue_delayed_work(commit_wq, &journal->j_work, HZ / 10);
4126
4127         /* if the next transaction has any chance of wrapping, flush
4128          ** transactions that might get overwritten.  If any journal lists are very
4129          ** old flush them as well.
4130          */
4131       first_jl:
4132         list_for_each_safe(entry, safe, &journal->j_journal_list) {
4133                 temp_jl = JOURNAL_LIST_ENTRY(entry);
4134                 if (journal->j_start <= temp_jl->j_start) {
4135                         if ((journal->j_start + journal->j_trans_max + 1) >=
4136                             temp_jl->j_start) {
4137                                 flush_used_journal_lists(p_s_sb, temp_jl);
4138                                 goto first_jl;
4139                         } else if ((journal->j_start +
4140                                     journal->j_trans_max + 1) <
4141                                    SB_ONDISK_JOURNAL_SIZE(p_s_sb)) {
4142                                 /* if we don't cross into the next transaction and we don't
4143                                  * wrap, there is no way we can overlap any later transactions
4144                                  * break now
4145                                  */
4146                                 break;
4147                         }
4148                 } else if ((journal->j_start +
4149                             journal->j_trans_max + 1) >
4150                            SB_ONDISK_JOURNAL_SIZE(p_s_sb)) {
4151                         if (((journal->j_start + journal->j_trans_max + 1) %
4152                              SB_ONDISK_JOURNAL_SIZE(p_s_sb)) >=
4153                             temp_jl->j_start) {
4154                                 flush_used_journal_lists(p_s_sb, temp_jl);
4155                                 goto first_jl;
4156                         } else {
4157                                 /* we don't overlap anything from out start to the end of the
4158                                  * log, and our wrapped portion doesn't overlap anything at
4159                                  * the start of the log.  We can break
4160                                  */
4161                                 break;
4162                         }
4163                 }
4164         }
4165         flush_old_journal_lists(p_s_sb);
4166
4167         journal->j_current_jl->j_list_bitmap =
4168             get_list_bitmap(p_s_sb, journal->j_current_jl);
4169
4170         if (!(journal->j_current_jl->j_list_bitmap)) {
4171                 reiserfs_panic(p_s_sb,
4172                                "journal-1996: do_journal_end, could not get a list bitmap\n");
4173         }
4174
4175         atomic_set(&(journal->j_jlock), 0);
4176         unlock_journal(p_s_sb);
4177         /* wake up any body waiting to join. */
4178         clear_bit(J_WRITERS_QUEUED, &journal->j_state);
4179         wake_up(&(journal->j_join_wait));
4180
4181         if (!flush && wait_on_commit &&
4182             journal_list_still_alive(p_s_sb, commit_trans_id)) {
4183                 flush_commit_list(p_s_sb, jl, 1);
4184         }
4185       out:
4186         reiserfs_check_lock_depth(p_s_sb, "journal end2");
4187
4188         memset(th, 0, sizeof(*th));
4189         /* Re-set th->t_super, so we can properly keep track of how many
4190          * persistent transactions there are. We need to do this so if this
4191          * call is part of a failed restart_transaction, we can free it later */
4192         th->t_super = p_s_sb;
4193
4194         return journal->j_errno;
4195 }
4196
4197 static void __reiserfs_journal_abort_hard(struct super_block *sb)
4198 {
4199         struct reiserfs_journal *journal = SB_JOURNAL(sb);
4200         if (test_bit(J_ABORTED, &journal->j_state))
4201                 return;
4202
4203         printk(KERN_CRIT "REISERFS: Aborting journal for filesystem on %s\n",
4204                reiserfs_bdevname(sb));
4205
4206         sb->s_flags |= MS_RDONLY;
4207         set_bit(J_ABORTED, &journal->j_state);
4208
4209 #ifdef CONFIG_REISERFS_CHECK
4210         dump_stack();
4211 #endif
4212 }
4213
4214 static void __reiserfs_journal_abort_soft(struct super_block *sb, int errno)
4215 {
4216         struct reiserfs_journal *journal = SB_JOURNAL(sb);
4217         if (test_bit(J_ABORTED, &journal->j_state))
4218                 return;
4219
4220         if (!journal->j_errno)
4221                 journal->j_errno = errno;
4222
4223         __reiserfs_journal_abort_hard(sb);
4224 }
4225
4226 void reiserfs_journal_abort(struct super_block *sb, int errno)
4227 {
4228         return __reiserfs_journal_abort_soft(sb, errno);
4229 }