git.oblomov.eu Git - linux-2.6/blob - fs/reiserfs/journal.c

   1 /*
   2 ** Write ahead logging implementation copyright Chris Mason 2000
   3 **
   4 ** The background commits make this code very interelated, and
   5 ** overly complex.  I need to rethink things a bit....The major players:
   6 **
   7 ** journal_begin -- call with the number of blocks you expect to log.
   8 **                  If the current transaction is too
   9 **                  old, it will block until the current transaction is
  10 **                  finished, and then start a new one.
  11 **                  Usually, your transaction will get joined in with
  12 **                  previous ones for speed.
  13 **
  14 ** journal_join  -- same as journal_begin, but won't block on the current
  15 **                  transaction regardless of age.  Don't ever call
  16 **                  this.  Ever.  There are only two places it should be
  17 **                  called from, and they are both inside this file.
  18 **
  19 ** journal_mark_dirty -- adds blocks into this transaction.  clears any flags
  20 **                       that might make them get sent to disk
  21 **                       and then marks them BH_JDirty.  Puts the buffer head
  22 **                       into the current transaction hash.
  23 **
  24 ** journal_end -- if the current transaction is batchable, it does nothing
  25 **                   otherwise, it could do an async/synchronous commit, or
  26 **                   a full flush of all log and real blocks in the
  27 **                   transaction.
  28 **
  29 ** flush_old_commits -- if the current transaction is too old, it is ended and
  30 **                      commit blocks are sent to disk.  Forces commit blocks
  31 **                      to disk for all backgrounded commits that have been
  32 **                      around too long.
  33 **                   -- Note, if you call this as an immediate flush from
  34 **                      from within kupdate, it will ignore the immediate flag
  35 */
  36
  37 #include <linux/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 = reiserfs_kmalloc(sizeof(struct reiserfs_bitmap_node), GFP_NOFS,
 156                               p_s_sb);
 157         if (!bn) {
 158                 return NULL;
 159         }
 160         bn->data = reiserfs_kmalloc(p_s_sb->s_blocksize, GFP_NOFS, p_s_sb);
 161         if (!bn->data) {
 162                 reiserfs_kfree(bn, sizeof(struct reiserfs_bitmap_node), p_s_sb);
 163                 return NULL;
 164         }
 165         bn->id = id++;
 166         memset(bn->data, 0, p_s_sb->s_blocksize);
 167         INIT_LIST_HEAD(&bn->list);
 168         return bn;
 169 }
 170
 171 static struct reiserfs_bitmap_node *get_bitmap_node(struct super_block *p_s_sb)
 172 {
 173         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 174         struct reiserfs_bitmap_node *bn = NULL;
 175         struct list_head *entry = journal->j_bitmap_nodes.next;
 176
 177         journal->j_used_bitmap_nodes++;
 178       repeat:
 179
 180         if (entry != &journal->j_bitmap_nodes) {
 181                 bn = list_entry(entry, struct reiserfs_bitmap_node, list);
 182                 list_del(entry);
 183                 memset(bn->data, 0, p_s_sb->s_blocksize);
 184                 journal->j_free_bitmap_nodes--;
 185                 return bn;
 186         }
 187         bn = allocate_bitmap_node(p_s_sb);
 188         if (!bn) {
 189                 yield();
 190                 goto repeat;
 191         }
 192         return bn;
 193 }
 194 static inline void free_bitmap_node(struct super_block *p_s_sb,
 195                                     struct reiserfs_bitmap_node *bn)
 196 {
 197         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 198         journal->j_used_bitmap_nodes--;
 199         if (journal->j_free_bitmap_nodes > REISERFS_MAX_BITMAP_NODES) {
 200                 reiserfs_kfree(bn->data, p_s_sb->s_blocksize, p_s_sb);
 201                 reiserfs_kfree(bn, sizeof(struct reiserfs_bitmap_node), p_s_sb);
 202         } else {
 203                 list_add(&bn->list, &journal->j_bitmap_nodes);
 204                 journal->j_free_bitmap_nodes++;
 205         }
 206 }
 207
 208 static void allocate_bitmap_nodes(struct super_block *p_s_sb)
 209 {
 210         int i;
 211         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 212         struct reiserfs_bitmap_node *bn = NULL;
 213         for (i = 0; i < REISERFS_MIN_BITMAP_NODES; i++) {
 214                 bn = allocate_bitmap_node(p_s_sb);
 215                 if (bn) {
 216                         list_add(&bn->list, &journal->j_bitmap_nodes);
 217                         journal->j_free_bitmap_nodes++;
 218                 } else {
 219                         break;  // this is ok, we'll try again when more are needed
 220                 }
 221         }
 222 }
 223
 224 static int set_bit_in_list_bitmap(struct super_block *p_s_sb, int block,
 225                                   struct reiserfs_list_bitmap *jb)
 226 {
 227         int bmap_nr = block / (p_s_sb->s_blocksize << 3);
 228         int bit_nr = block % (p_s_sb->s_blocksize << 3);
 229
 230         if (!jb->bitmaps[bmap_nr]) {
 231                 jb->bitmaps[bmap_nr] = get_bitmap_node(p_s_sb);
 232         }
 233         set_bit(bit_nr, (unsigned long *)jb->bitmaps[bmap_nr]->data);
 234         return 0;
 235 }
 236
 237 static void cleanup_bitmap_list(struct super_block *p_s_sb,
 238                                 struct reiserfs_list_bitmap *jb)
 239 {
 240         int i;
 241         if (jb->bitmaps == NULL)
 242                 return;
 243
 244         for (i = 0; i < SB_BMAP_NR(p_s_sb); i++) {
 245                 if (jb->bitmaps[i]) {
 246                         free_bitmap_node(p_s_sb, jb->bitmaps[i]);
 247                         jb->bitmaps[i] = NULL;
 248                 }
 249         }
 250 }
 251
 252 /*
 253 ** only call this on FS unmount.
 254 */
 255 static int free_list_bitmaps(struct super_block *p_s_sb,
 256                              struct reiserfs_list_bitmap *jb_array)
 257 {
 258         int i;
 259         struct reiserfs_list_bitmap *jb;
 260         for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
 261                 jb = jb_array + i;
 262                 jb->journal_list = NULL;
 263                 cleanup_bitmap_list(p_s_sb, jb);
 264                 vfree(jb->bitmaps);
 265                 jb->bitmaps = NULL;
 266         }
 267         return 0;
 268 }
 269
 270 static int free_bitmap_nodes(struct super_block *p_s_sb)
 271 {
 272         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 273         struct list_head *next = journal->j_bitmap_nodes.next;
 274         struct reiserfs_bitmap_node *bn;
 275
 276         while (next != &journal->j_bitmap_nodes) {
 277                 bn = list_entry(next, struct reiserfs_bitmap_node, list);
 278                 list_del(next);
 279                 reiserfs_kfree(bn->data, p_s_sb->s_blocksize, p_s_sb);
 280                 reiserfs_kfree(bn, sizeof(struct reiserfs_bitmap_node), p_s_sb);
 281                 next = journal->j_bitmap_nodes.next;
 282                 journal->j_free_bitmap_nodes--;
 283         }
 284
 285         return 0;
 286 }
 287
 288 /*
 289 ** get memory for JOURNAL_NUM_BITMAPS worth of bitmaps.
 290 ** jb_array is the array to be filled in.
 291 */
 292 int reiserfs_allocate_list_bitmaps(struct super_block *p_s_sb,
 293                                    struct reiserfs_list_bitmap *jb_array,
 294                                    int bmap_nr)
 295 {
 296         int i;
 297         int failed = 0;
 298         struct reiserfs_list_bitmap *jb;
 299         int mem = bmap_nr * sizeof(struct reiserfs_bitmap_node *);
 300
 301         for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
 302                 jb = jb_array + i;
 303                 jb->journal_list = NULL;
 304                 jb->bitmaps = vmalloc(mem);
 305                 if (!jb->bitmaps) {
 306                         reiserfs_warning(p_s_sb,
 307                                          "clm-2000, unable to allocate bitmaps for journal lists");
 308                         failed = 1;
 309                         break;
 310                 }
 311                 memset(jb->bitmaps, 0, mem);
 312         }
 313         if (failed) {
 314                 free_list_bitmaps(p_s_sb, jb_array);
 315                 return -1;
 316         }
 317         return 0;
 318 }
 319
 320 /*
 321 ** find an available list bitmap.  If you can't find one, flush a commit list
 322 ** and try again
 323 */
 324 static struct reiserfs_list_bitmap *get_list_bitmap(struct super_block *p_s_sb,
 325                                                     struct reiserfs_journal_list
 326                                                     *jl)
 327 {
 328         int i, j;
 329         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 330         struct reiserfs_list_bitmap *jb = NULL;
 331
 332         for (j = 0; j < (JOURNAL_NUM_BITMAPS * 3); j++) {
 333                 i = journal->j_list_bitmap_index;
 334                 journal->j_list_bitmap_index = (i + 1) % JOURNAL_NUM_BITMAPS;
 335                 jb = journal->j_list_bitmap + i;
 336                 if (journal->j_list_bitmap[i].journal_list) {
 337                         flush_commit_list(p_s_sb,
 338                                           journal->j_list_bitmap[i].
 339                                           journal_list, 1);
 340                         if (!journal->j_list_bitmap[i].journal_list) {
 341                                 break;
 342                         }
 343                 } else {
 344                         break;
 345                 }
 346         }
 347         if (jb->journal_list) { /* double check to make sure if flushed correctly */
 348                 return NULL;
 349         }
 350         jb->journal_list = jl;
 351         return jb;
 352 }
 353
 354 /*
 355 ** allocates a new chunk of X nodes, and links them all together as a list.
 356 ** Uses the cnode->next and cnode->prev pointers
 357 ** returns NULL on failure
 358 */
 359 static struct reiserfs_journal_cnode *allocate_cnodes(int num_cnodes)
 360 {
 361         struct reiserfs_journal_cnode *head;
 362         int i;
 363         if (num_cnodes <= 0) {
 364                 return NULL;
 365         }
 366         head = vmalloc(num_cnodes * sizeof(struct reiserfs_journal_cnode));
 367         if (!head) {
 368                 return NULL;
 369         }
 370         memset(head, 0, num_cnodes * sizeof(struct reiserfs_journal_cnode));
 371         head[0].prev = NULL;
 372         head[0].next = head + 1;
 373         for (i = 1; i < num_cnodes; i++) {
 374                 head[i].prev = head + (i - 1);
 375                 head[i].next = head + (i + 1);  /* if last one, overwrite it after the if */
 376         }
 377         head[num_cnodes - 1].next = NULL;
 378         return head;
 379 }
 380
 381 /*
 382 ** pulls a cnode off the free list, or returns NULL on failure
 383 */
 384 static struct reiserfs_journal_cnode *get_cnode(struct super_block *p_s_sb)
 385 {
 386         struct reiserfs_journal_cnode *cn;
 387         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 388
 389         reiserfs_check_lock_depth(p_s_sb, "get_cnode");
 390
 391         if (journal->j_cnode_free <= 0) {
 392                 return NULL;
 393         }
 394         journal->j_cnode_used++;
 395         journal->j_cnode_free--;
 396         cn = journal->j_cnode_free_list;
 397         if (!cn) {
 398                 return cn;
 399         }
 400         if (cn->next) {
 401                 cn->next->prev = NULL;
 402         }
 403         journal->j_cnode_free_list = cn->next;
 404         memset(cn, 0, sizeof(struct reiserfs_journal_cnode));
 405         return cn;
 406 }
 407
 408 /*
 409 ** returns a cnode to the free list
 410 */
 411 static void free_cnode(struct super_block *p_s_sb,
 412                        struct reiserfs_journal_cnode *cn)
 413 {
 414         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 415
 416         reiserfs_check_lock_depth(p_s_sb, "free_cnode");
 417
 418         journal->j_cnode_used--;
 419         journal->j_cnode_free++;
 420         /* memset(cn, 0, sizeof(struct reiserfs_journal_cnode)) ; */
 421         cn->next = journal->j_cnode_free_list;
 422         if (journal->j_cnode_free_list) {
 423                 journal->j_cnode_free_list->prev = cn;
 424         }
 425         cn->prev = NULL;        /* not needed with the memset, but I might kill the memset, and forget to do this */
 426         journal->j_cnode_free_list = cn;
 427 }
 428
 429 static void clear_prepared_bits(struct buffer_head *bh)
 430 {
 431         clear_buffer_journal_prepared(bh);
 432         clear_buffer_journal_restore_dirty(bh);
 433 }
 434
 435 /* utility function to force a BUG if it is called without the big
 436 ** kernel lock held.  caller is the string printed just before calling BUG()
 437 */
 438 void reiserfs_check_lock_depth(struct super_block *sb, char *caller)
 439 {
 440 #ifdef CONFIG_SMP
 441         if (current->lock_depth < 0) {
 442                 reiserfs_panic(sb, "%s called without kernel lock held",
 443                                caller);
 444         }
 445 #else
 446         ;
 447 #endif
 448 }
 449
 450 /* return a cnode with same dev, block number and size in table, or null if not found */
 451 static inline struct reiserfs_journal_cnode *get_journal_hash_dev(struct
 452                                                                   super_block
 453                                                                   *sb,
 454                                                                   struct
 455                                                                   reiserfs_journal_cnode
 456                                                                   **table,
 457                                                                   long bl)
 458 {
 459         struct reiserfs_journal_cnode *cn;
 460         cn = journal_hash(table, sb, bl);
 461         while (cn) {
 462                 if (cn->blocknr == bl && cn->sb == sb)
 463                         return cn;
 464                 cn = cn->hnext;
 465         }
 466         return (struct reiserfs_journal_cnode *)0;
 467 }
 468
 469 /*
 470 ** this actually means 'can this block be reallocated yet?'.  If you set search_all, a block can only be allocated
 471 ** if it is not in the current transaction, was not freed by the current transaction, and has no chance of ever
 472 ** being overwritten by a replay after crashing.
 473 **
 474 ** If you don't set search_all, a block can only be allocated if it is not in the current transaction.  Since deleting
 475 ** a block removes it from the current transaction, this case should never happen.  If you don't set search_all, make
 476 ** sure you never write the block without logging it.
 477 **
 478 ** next_zero_bit is a suggestion about the next block to try for find_forward.
 479 ** when bl is rejected because it is set in a journal list bitmap, we search
 480 ** for the next zero bit in the bitmap that rejected bl.  Then, we return that
 481 ** through next_zero_bit for find_forward to try.
 482 **
 483 ** Just because we return something in next_zero_bit does not mean we won't
 484 ** reject it on the next call to reiserfs_in_journal
 485 **
 486 */
 487 int reiserfs_in_journal(struct super_block *p_s_sb,
 488                         int bmap_nr, int bit_nr, int search_all,
 489                         b_blocknr_t * next_zero_bit)
 490 {
 491         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 492         struct reiserfs_journal_cnode *cn;
 493         struct reiserfs_list_bitmap *jb;
 494         int i;
 495         unsigned long bl;
 496
 497         *next_zero_bit = 0;     /* always start this at zero. */
 498
 499         PROC_INFO_INC(p_s_sb, journal.in_journal);
 500         /* If we aren't doing a search_all, this is a metablock, and it will be logged before use.
 501          ** if we crash before the transaction that freed it commits,  this transaction won't
 502          ** have committed either, and the block will never be written
 503          */
 504         if (search_all) {
 505                 for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
 506                         PROC_INFO_INC(p_s_sb, journal.in_journal_bitmap);
 507                         jb = journal->j_list_bitmap + i;
 508                         if (jb->journal_list && jb->bitmaps[bmap_nr] &&
 509                             test_bit(bit_nr,
 510                                      (unsigned long *)jb->bitmaps[bmap_nr]->
 511                                      data)) {
 512                                 *next_zero_bit =
 513                                     find_next_zero_bit((unsigned long *)
 514                                                        (jb->bitmaps[bmap_nr]->
 515                                                         data),
 516                                                        p_s_sb->s_blocksize << 3,
 517                                                        bit_nr + 1);
 518                                 return 1;
 519                         }
 520                 }
 521         }
 522
 523         bl = bmap_nr * (p_s_sb->s_blocksize << 3) + bit_nr;
 524         /* is it in any old transactions? */
 525         if (search_all
 526             && (cn =
 527                 get_journal_hash_dev(p_s_sb, journal->j_list_hash_table, bl))) {
 528                 return 1;
 529         }
 530
 531         /* is it in the current transaction.  This should never happen */
 532         if ((cn = get_journal_hash_dev(p_s_sb, journal->j_hash_table, bl))) {
 533                 BUG();
 534                 return 1;
 535         }
 536
 537         PROC_INFO_INC(p_s_sb, journal.in_journal_reusable);
 538         /* safe for reuse */
 539         return 0;
 540 }
 541
 542 /* insert cn into table
 543 */
 544 static inline void insert_journal_hash(struct reiserfs_journal_cnode **table,
 545                                        struct reiserfs_journal_cnode *cn)
 546 {
 547         struct reiserfs_journal_cnode *cn_orig;
 548
 549         cn_orig = journal_hash(table, cn->sb, cn->blocknr);
 550         cn->hnext = cn_orig;
 551         cn->hprev = NULL;
 552         if (cn_orig) {
 553                 cn_orig->hprev = cn;
 554         }
 555         journal_hash(table, cn->sb, cn->blocknr) = cn;
 556 }
 557
 558 /* lock the current transaction */
 559 static inline void lock_journal(struct super_block *p_s_sb)
 560 {
 561         PROC_INFO_INC(p_s_sb, journal.lock_journal);
 562         down(&SB_JOURNAL(p_s_sb)->j_lock);
 563 }
 564
 565 /* unlock the current transaction */
 566 static inline void unlock_journal(struct super_block *p_s_sb)
 567 {
 568         up(&SB_JOURNAL(p_s_sb)->j_lock);
 569 }
 570
 571 static inline void get_journal_list(struct reiserfs_journal_list *jl)
 572 {
 573         jl->j_refcount++;
 574 }
 575
 576 static inline void put_journal_list(struct super_block *s,
 577                                     struct reiserfs_journal_list *jl)
 578 {
 579         if (jl->j_refcount < 1) {
 580                 reiserfs_panic(s, "trans id %lu, refcount at %d",
 581                                jl->j_trans_id, jl->j_refcount);
 582         }
 583         if (--jl->j_refcount == 0)
 584                 reiserfs_kfree(jl, sizeof(struct reiserfs_journal_list), s);
 585 }
 586
 587 /*
 588 ** this used to be much more involved, and I'm keeping it just in case things get ugly again.
 589 ** it gets called by flush_commit_list, and cleans up any data stored about blocks freed during a
 590 ** transaction.
 591 */
 592 static void cleanup_freed_for_journal_list(struct super_block *p_s_sb,
 593                                            struct reiserfs_journal_list *jl)
 594 {
 595
 596         struct reiserfs_list_bitmap *jb = jl->j_list_bitmap;
 597         if (jb) {
 598                 cleanup_bitmap_list(p_s_sb, jb);
 599         }
 600         jl->j_list_bitmap->journal_list = NULL;
 601         jl->j_list_bitmap = NULL;
 602 }
 603
 604 static int journal_list_still_alive(struct super_block *s,
 605                                     unsigned long trans_id)
 606 {
 607         struct reiserfs_journal *journal = SB_JOURNAL(s);
 608         struct list_head *entry = &journal->j_journal_list;
 609         struct reiserfs_journal_list *jl;
 610
 611         if (!list_empty(entry)) {
 612                 jl = JOURNAL_LIST_ENTRY(entry->next);
 613                 if (jl->j_trans_id <= trans_id) {
 614                         return 1;
 615                 }
 616         }
 617         return 0;
 618 }
 619
 620 static void reiserfs_end_buffer_io_sync(struct buffer_head *bh, int uptodate)
 621 {
 622         char b[BDEVNAME_SIZE];
 623
 624         if (buffer_journaled(bh)) {
 625                 reiserfs_warning(NULL,
 626                                  "clm-2084: pinned buffer %lu:%s sent to disk",
 627                                  bh->b_blocknr, bdevname(bh->b_bdev, b));
 628         }
 629         if (uptodate)
 630                 set_buffer_uptodate(bh);
 631         else
 632                 clear_buffer_uptodate(bh);
 633         unlock_buffer(bh);
 634         put_bh(bh);
 635 }
 636
 637 static void reiserfs_end_ordered_io(struct buffer_head *bh, int uptodate)
 638 {
 639         if (uptodate)
 640                 set_buffer_uptodate(bh);
 641         else
 642                 clear_buffer_uptodate(bh);
 643         unlock_buffer(bh);
 644         put_bh(bh);
 645 }
 646
 647 static void submit_logged_buffer(struct buffer_head *bh)
 648 {
 649         get_bh(bh);
 650         bh->b_end_io = reiserfs_end_buffer_io_sync;
 651         clear_buffer_journal_new(bh);
 652         clear_buffer_dirty(bh);
 653         if (!test_clear_buffer_journal_test(bh))
 654                 BUG();
 655         if (!buffer_uptodate(bh))
 656                 BUG();
 657         submit_bh(WRITE, bh);
 658 }
 659
 660 static void submit_ordered_buffer(struct buffer_head *bh)
 661 {
 662         get_bh(bh);
 663         bh->b_end_io = reiserfs_end_ordered_io;
 664         clear_buffer_dirty(bh);
 665         if (!buffer_uptodate(bh))
 666                 BUG();
 667         submit_bh(WRITE, bh);
 668 }
 669
 670 static int submit_barrier_buffer(struct buffer_head *bh)
 671 {
 672         get_bh(bh);
 673         bh->b_end_io = reiserfs_end_ordered_io;
 674         clear_buffer_dirty(bh);
 675         if (!buffer_uptodate(bh))
 676                 BUG();
 677         return submit_bh(WRITE_BARRIER, bh);
 678 }
 679
 680 static void check_barrier_completion(struct super_block *s,
 681                                      struct buffer_head *bh)
 682 {
 683         if (buffer_eopnotsupp(bh)) {
 684                 clear_buffer_eopnotsupp(bh);
 685                 disable_barrier(s);
 686                 set_buffer_uptodate(bh);
 687                 set_buffer_dirty(bh);
 688                 sync_dirty_buffer(bh);
 689         }
 690 }
 691
 692 #define CHUNK_SIZE 32
 693 struct buffer_chunk {
 694         struct buffer_head *bh[CHUNK_SIZE];
 695         int nr;
 696 };
 697
 698 static void write_chunk(struct buffer_chunk *chunk)
 699 {
 700         int i;
 701         get_fs_excl();
 702         for (i = 0; i < chunk->nr; i++) {
 703                 submit_logged_buffer(chunk->bh[i]);
 704         }
 705         chunk->nr = 0;
 706         put_fs_excl();
 707 }
 708
 709 static void write_ordered_chunk(struct buffer_chunk *chunk)
 710 {
 711         int i;
 712         get_fs_excl();
 713         for (i = 0; i < chunk->nr; i++) {
 714                 submit_ordered_buffer(chunk->bh[i]);
 715         }
 716         chunk->nr = 0;
 717         put_fs_excl();
 718 }
 719
 720 static int add_to_chunk(struct buffer_chunk *chunk, struct buffer_head *bh,
 721                         spinlock_t * lock, void (fn) (struct buffer_chunk *))
 722 {
 723         int ret = 0;
 724         if (chunk->nr >= CHUNK_SIZE)
 725                 BUG();
 726         chunk->bh[chunk->nr++] = bh;
 727         if (chunk->nr >= CHUNK_SIZE) {
 728                 ret = 1;
 729                 if (lock)
 730                         spin_unlock(lock);
 731                 fn(chunk);
 732                 if (lock)
 733                         spin_lock(lock);
 734         }
 735         return ret;
 736 }
 737
 738 static atomic_t nr_reiserfs_jh = ATOMIC_INIT(0);
 739 static struct reiserfs_jh *alloc_jh(void)
 740 {
 741         struct reiserfs_jh *jh;
 742         while (1) {
 743                 jh = kmalloc(sizeof(*jh), GFP_NOFS);
 744                 if (jh) {
 745                         atomic_inc(&nr_reiserfs_jh);
 746                         return jh;
 747                 }
 748                 yield();
 749         }
 750 }
 751
 752 /*
 753  * we want to free the jh when the buffer has been written
 754  * and waited on
 755  */
 756 void reiserfs_free_jh(struct buffer_head *bh)
 757 {
 758         struct reiserfs_jh *jh;
 759
 760         jh = bh->b_private;
 761         if (jh) {
 762                 bh->b_private = NULL;
 763                 jh->bh = NULL;
 764                 list_del_init(&jh->list);
 765                 kfree(jh);
 766                 if (atomic_read(&nr_reiserfs_jh) <= 0)
 767                         BUG();
 768                 atomic_dec(&nr_reiserfs_jh);
 769                 put_bh(bh);
 770         }
 771 }
 772
 773 static inline int __add_jh(struct reiserfs_journal *j, struct buffer_head *bh,
 774                            int tail)
 775 {
 776         struct reiserfs_jh *jh;
 777
 778         if (bh->b_private) {
 779                 spin_lock(&j->j_dirty_buffers_lock);
 780                 if (!bh->b_private) {
 781                         spin_unlock(&j->j_dirty_buffers_lock);
 782                         goto no_jh;
 783                 }
 784                 jh = bh->b_private;
 785                 list_del_init(&jh->list);
 786         } else {
 787               no_jh:
 788                 get_bh(bh);
 789                 jh = alloc_jh();
 790                 spin_lock(&j->j_dirty_buffers_lock);
 791                 /* buffer must be locked for __add_jh, should be able to have
 792                  * two adds at the same time
 793                  */
 794                 if (bh->b_private)
 795                         BUG();
 796                 jh->bh = bh;
 797                 bh->b_private = jh;
 798         }
 799         jh->jl = j->j_current_jl;
 800         if (tail)
 801                 list_add_tail(&jh->list, &jh->jl->j_tail_bh_list);
 802         else {
 803                 list_add_tail(&jh->list, &jh->jl->j_bh_list);
 804         }
 805         spin_unlock(&j->j_dirty_buffers_lock);
 806         return 0;
 807 }
 808
 809 int reiserfs_add_tail_list(struct inode *inode, struct buffer_head *bh)
 810 {
 811         return __add_jh(SB_JOURNAL(inode->i_sb), bh, 1);
 812 }
 813 int reiserfs_add_ordered_list(struct inode *inode, struct buffer_head *bh)
 814 {
 815         return __add_jh(SB_JOURNAL(inode->i_sb), bh, 0);
 816 }
 817
 818 #define JH_ENTRY(l) list_entry((l), struct reiserfs_jh, list)
 819 static int write_ordered_buffers(spinlock_t * lock,
 820                                  struct reiserfs_journal *j,
 821                                  struct reiserfs_journal_list *jl,
 822                                  struct list_head *list)
 823 {
 824         struct buffer_head *bh;
 825         struct reiserfs_jh *jh;
 826         int ret = j->j_errno;
 827         struct buffer_chunk chunk;
 828         struct list_head tmp;
 829         INIT_LIST_HEAD(&tmp);
 830
 831         chunk.nr = 0;
 832         spin_lock(lock);
 833         while (!list_empty(list)) {
 834                 jh = JH_ENTRY(list->next);
 835                 bh = jh->bh;
 836                 get_bh(bh);
 837                 if (test_set_buffer_locked(bh)) {
 838                         if (!buffer_dirty(bh)) {
 839                                 list_del_init(&jh->list);
 840                                 list_add(&jh->list, &tmp);
 841                                 goto loop_next;
 842                         }
 843                         spin_unlock(lock);
 844                         if (chunk.nr)
 845                                 write_ordered_chunk(&chunk);
 846                         wait_on_buffer(bh);
 847                         cond_resched();
 848                         spin_lock(lock);
 849                         goto loop_next;
 850                 }
 851                 if (buffer_dirty(bh)) {
 852                         list_del_init(&jh->list);
 853                         list_add(&jh->list, &tmp);
 854                         add_to_chunk(&chunk, bh, lock, write_ordered_chunk);
 855                 } else {
 856                         reiserfs_free_jh(bh);
 857                         unlock_buffer(bh);
 858                 }
 859               loop_next:
 860                 put_bh(bh);
 861                 cond_resched_lock(lock);
 862         }
 863         if (chunk.nr) {
 864                 spin_unlock(lock);
 865                 write_ordered_chunk(&chunk);
 866                 spin_lock(lock);
 867         }
 868         while (!list_empty(&tmp)) {
 869                 jh = JH_ENTRY(tmp.prev);
 870                 bh = jh->bh;
 871                 get_bh(bh);
 872                 reiserfs_free_jh(bh);
 873
 874                 if (buffer_locked(bh)) {
 875                         spin_unlock(lock);
 876                         wait_on_buffer(bh);
 877                         spin_lock(lock);
 878                 }
 879                 if (!buffer_uptodate(bh)) {
 880                         ret = -EIO;
 881                 }
 882                 put_bh(bh);
 883                 cond_resched_lock(lock);
 884         }
 885         spin_unlock(lock);
 886         return ret;
 887 }
 888
 889 static int flush_older_commits(struct super_block *s,
 890                                struct reiserfs_journal_list *jl)
 891 {
 892         struct reiserfs_journal *journal = SB_JOURNAL(s);
 893         struct reiserfs_journal_list *other_jl;
 894         struct reiserfs_journal_list *first_jl;
 895         struct list_head *entry;
 896         unsigned long trans_id = jl->j_trans_id;
 897         unsigned long other_trans_id;
 898         unsigned long first_trans_id;
 899
 900       find_first:
 901         /*
 902          * first we walk backwards to find the oldest uncommitted transation
 903          */
 904         first_jl = jl;
 905         entry = jl->j_list.prev;
 906         while (1) {
 907                 other_jl = JOURNAL_LIST_ENTRY(entry);
 908                 if (entry == &journal->j_journal_list ||
 909                     atomic_read(&other_jl->j_older_commits_done))
 910                         break;
 911
 912                 first_jl = other_jl;
 913                 entry = other_jl->j_list.prev;
 914         }
 915
 916         /* if we didn't find any older uncommitted transactions, return now */
 917         if (first_jl == jl) {
 918                 return 0;
 919         }
 920
 921         first_trans_id = first_jl->j_trans_id;
 922
 923         entry = &first_jl->j_list;
 924         while (1) {
 925                 other_jl = JOURNAL_LIST_ENTRY(entry);
 926                 other_trans_id = other_jl->j_trans_id;
 927
 928                 if (other_trans_id < trans_id) {
 929                         if (atomic_read(&other_jl->j_commit_left) != 0) {
 930                                 flush_commit_list(s, other_jl, 0);
 931
 932                                 /* list we were called with is gone, return */
 933                                 if (!journal_list_still_alive(s, trans_id))
 934                                         return 1;
 935
 936                                 /* the one we just flushed is gone, this means all
 937                                  * older lists are also gone, so first_jl is no longer
 938                                  * valid either.  Go back to the beginning.
 939                                  */
 940                                 if (!journal_list_still_alive
 941                                     (s, other_trans_id)) {
 942                                         goto find_first;
 943                                 }
 944                         }
 945                         entry = entry->next;
 946                         if (entry == &journal->j_journal_list)
 947                                 return 0;
 948                 } else {
 949                         return 0;
 950                 }
 951         }
 952         return 0;
 953 }
 954 int reiserfs_async_progress_wait(struct super_block *s)
 955 {
 956         DEFINE_WAIT(wait);
 957         struct reiserfs_journal *j = SB_JOURNAL(s);
 958         if (atomic_read(&j->j_async_throttle))
 959                 blk_congestion_wait(WRITE, HZ / 10);
 960         return 0;
 961 }
 962
 963 /*
 964 ** if this journal list still has commit blocks unflushed, send them to disk.
 965 **
 966 ** log areas must be flushed in order (transaction 2 can't commit before transaction 1)
 967 ** Before the commit block can by written, every other log block must be safely on disk
 968 **
 969 */
 970 static int flush_commit_list(struct super_block *s,
 971                              struct reiserfs_journal_list *jl, int flushall)
 972 {
 973         int i;
 974         int bn;
 975         struct buffer_head *tbh = NULL;
 976         unsigned long trans_id = jl->j_trans_id;
 977         struct reiserfs_journal *journal = SB_JOURNAL(s);
 978         int barrier = 0;
 979         int retval = 0;
 980
 981         reiserfs_check_lock_depth(s, "flush_commit_list");
 982
 983         if (atomic_read(&jl->j_older_commits_done)) {
 984                 return 0;
 985         }
 986
 987         get_fs_excl();
 988
 989         /* before we can put our commit blocks on disk, we have to make sure everyone older than
 990          ** us is on disk too
 991          */
 992         BUG_ON(jl->j_len <= 0);
 993         BUG_ON(trans_id == journal->j_trans_id);
 994
 995         get_journal_list(jl);
 996         if (flushall) {
 997                 if (flush_older_commits(s, jl) == 1) {
 998                         /* list disappeared during flush_older_commits.  return */
 999                         goto put_jl;
1000                 }
1001         }
1002
1003         /* make sure nobody is trying to flush this one at the same time */
1004         down(&jl->j_commit_lock);
1005         if (!journal_list_still_alive(s, trans_id)) {
1006                 up(&jl->j_commit_lock);
1007                 goto put_jl;
1008         }
1009         BUG_ON(jl->j_trans_id == 0);
1010
1011         /* this commit is done, exit */
1012         if (atomic_read(&(jl->j_commit_left)) <= 0) {
1013                 if (flushall) {
1014                         atomic_set(&(jl->j_older_commits_done), 1);
1015                 }
1016                 up(&jl->j_commit_lock);
1017                 goto put_jl;
1018         }
1019
1020         if (!list_empty(&jl->j_bh_list)) {
1021                 unlock_kernel();
1022                 write_ordered_buffers(&journal->j_dirty_buffers_lock,
1023                                       journal, jl, &jl->j_bh_list);
1024                 lock_kernel();
1025         }
1026         BUG_ON(!list_empty(&jl->j_bh_list));
1027         /*
1028          * for the description block and all the log blocks, submit any buffers
1029          * that haven't already reached the disk
1030          */
1031         atomic_inc(&journal->j_async_throttle);
1032         for (i = 0; i < (jl->j_len + 1); i++) {
1033                 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + (jl->j_start + i) %
1034                     SB_ONDISK_JOURNAL_SIZE(s);
1035                 tbh = journal_find_get_block(s, bn);
1036                 if (buffer_dirty(tbh))  /* redundant, ll_rw_block() checks */
1037                         ll_rw_block(SWRITE, 1, &tbh);
1038                 put_bh(tbh);
1039         }
1040         atomic_dec(&journal->j_async_throttle);
1041
1042         /* wait on everything written so far before writing the commit
1043          * if we are in barrier mode, send the commit down now
1044          */
1045         barrier = reiserfs_barrier_flush(s);
1046         if (barrier) {
1047                 int ret;
1048                 lock_buffer(jl->j_commit_bh);
1049                 ret = submit_barrier_buffer(jl->j_commit_bh);
1050                 if (ret == -EOPNOTSUPP) {
1051                         set_buffer_uptodate(jl->j_commit_bh);
1052                         disable_barrier(s);
1053                         barrier = 0;
1054                 }
1055         }
1056         for (i = 0; i < (jl->j_len + 1); i++) {
1057                 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) +
1058                     (jl->j_start + i) % SB_ONDISK_JOURNAL_SIZE(s);
1059                 tbh = journal_find_get_block(s, bn);
1060                 wait_on_buffer(tbh);
1061                 // since we're using ll_rw_blk above, it might have skipped over
1062                 // a locked buffer.  Double check here
1063                 //
1064                 if (buffer_dirty(tbh))  /* redundant, sync_dirty_buffer() checks */
1065                         sync_dirty_buffer(tbh);
1066                 if (unlikely(!buffer_uptodate(tbh))) {
1067 #ifdef CONFIG_REISERFS_CHECK
1068                         reiserfs_warning(s, "journal-601, buffer write failed");
1069 #endif
1070                         retval = -EIO;
1071                 }
1072                 put_bh(tbh);    /* once for journal_find_get_block */
1073                 put_bh(tbh);    /* once due to original getblk in do_journal_end */
1074                 atomic_dec(&(jl->j_commit_left));
1075         }
1076
1077         BUG_ON(atomic_read(&(jl->j_commit_left)) != 1);
1078
1079         if (!barrier) {
1080                 if (buffer_dirty(jl->j_commit_bh))
1081                         BUG();
1082                 mark_buffer_dirty(jl->j_commit_bh);
1083                 sync_dirty_buffer(jl->j_commit_bh);
1084         } else
1085                 wait_on_buffer(jl->j_commit_bh);
1086
1087         check_barrier_completion(s, jl->j_commit_bh);
1088
1089         /* If there was a write error in the journal - we can't commit this
1090          * transaction - it will be invalid and, if successful, will just end
1091          * up propogating the write error out to the filesystem. */
1092         if (unlikely(!buffer_uptodate(jl->j_commit_bh))) {
1093 #ifdef CONFIG_REISERFS_CHECK
1094                 reiserfs_warning(s, "journal-615: buffer write failed");
1095 #endif
1096                 retval = -EIO;
1097         }
1098         bforget(jl->j_commit_bh);
1099         if (journal->j_last_commit_id != 0 &&
1100             (jl->j_trans_id - journal->j_last_commit_id) != 1) {
1101                 reiserfs_warning(s, "clm-2200: last commit %lu, current %lu",
1102                                  journal->j_last_commit_id, jl->j_trans_id);
1103         }
1104         journal->j_last_commit_id = jl->j_trans_id;
1105
1106         /* now, every commit block is on the disk.  It is safe to allow blocks freed during this transaction to be reallocated */
1107         cleanup_freed_for_journal_list(s, jl);
1108
1109         retval = retval ? retval : journal->j_errno;
1110
1111         /* mark the metadata dirty */
1112         if (!retval)
1113                 dirty_one_transaction(s, jl);
1114         atomic_dec(&(jl->j_commit_left));
1115
1116         if (flushall) {
1117                 atomic_set(&(jl->j_older_commits_done), 1);
1118         }
1119         up(&jl->j_commit_lock);
1120       put_jl:
1121         put_journal_list(s, jl);
1122
1123         if (retval)
1124                 reiserfs_abort(s, retval, "Journal write error in %s",
1125                                __FUNCTION__);
1126         put_fs_excl();
1127         return retval;
1128 }
1129
1130 /*
1131 ** flush_journal_list frequently needs to find a newer transaction for a given block.  This does that, or
1132 ** returns NULL if it can't find anything
1133 */
1134 static struct reiserfs_journal_list *find_newer_jl_for_cn(struct
1135                                                           reiserfs_journal_cnode
1136                                                           *cn)
1137 {
1138         struct super_block *sb = cn->sb;
1139         b_blocknr_t blocknr = cn->blocknr;
1140
1141         cn = cn->hprev;
1142         while (cn) {
1143                 if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist) {
1144                         return cn->jlist;
1145                 }
1146                 cn = cn->hprev;
1147         }
1148         return NULL;
1149 }
1150
1151 static void remove_journal_hash(struct super_block *,
1152                                 struct reiserfs_journal_cnode **,
1153                                 struct reiserfs_journal_list *, unsigned long,
1154                                 int);
1155
1156 /*
1157 ** once all the real blocks have been flushed, it is safe to remove them from the
1158 ** journal list for this transaction.  Aside from freeing the cnode, this also allows the
1159 ** block to be reallocated for data blocks if it had been deleted.
1160 */
1161 static void remove_all_from_journal_list(struct super_block *p_s_sb,
1162                                          struct reiserfs_journal_list *jl,
1163                                          int debug)
1164 {
1165         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1166         struct reiserfs_journal_cnode *cn, *last;
1167         cn = jl->j_realblock;
1168
1169         /* which is better, to lock once around the whole loop, or
1170          ** to lock for each call to remove_journal_hash?
1171          */
1172         while (cn) {
1173                 if (cn->blocknr != 0) {
1174                         if (debug) {
1175                                 reiserfs_warning(p_s_sb,
1176                                                  "block %u, bh is %d, state %ld",
1177                                                  cn->blocknr, cn->bh ? 1 : 0,
1178                                                  cn->state);
1179                         }
1180                         cn->state = 0;
1181                         remove_journal_hash(p_s_sb, journal->j_list_hash_table,
1182                                             jl, cn->blocknr, 1);
1183                 }
1184                 last = cn;
1185                 cn = cn->next;
1186                 free_cnode(p_s_sb, last);
1187         }
1188         jl->j_realblock = NULL;
1189 }
1190
1191 /*
1192 ** if this timestamp is greater than the timestamp we wrote last to the header block, write it to the header block.
1193 ** once this is done, I can safely say the log area for this transaction won't ever be replayed, and I can start
1194 ** releasing blocks in this transaction for reuse as data blocks.
1195 ** called by flush_journal_list, before it calls remove_all_from_journal_list
1196 **
1197 */
1198 static int _update_journal_header_block(struct super_block *p_s_sb,
1199                                         unsigned long offset,
1200                                         unsigned long trans_id)
1201 {
1202         struct reiserfs_journal_header *jh;
1203         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1204
1205         if (reiserfs_is_journal_aborted(journal))
1206                 return -EIO;
1207
1208         if (trans_id >= journal->j_last_flush_trans_id) {
1209                 if (buffer_locked((journal->j_header_bh))) {
1210                         wait_on_buffer((journal->j_header_bh));
1211                         if (unlikely(!buffer_uptodate(journal->j_header_bh))) {
1212 #ifdef CONFIG_REISERFS_CHECK
1213                                 reiserfs_warning(p_s_sb,
1214                                                  "journal-699: buffer write failed");
1215 #endif
1216                                 return -EIO;
1217                         }
1218                 }
1219                 journal->j_last_flush_trans_id = trans_id;
1220                 journal->j_first_unflushed_offset = offset;
1221                 jh = (struct reiserfs_journal_header *)(journal->j_header_bh->
1222                                                         b_data);
1223                 jh->j_last_flush_trans_id = cpu_to_le32(trans_id);
1224                 jh->j_first_unflushed_offset = cpu_to_le32(offset);
1225                 jh->j_mount_id = cpu_to_le32(journal->j_mount_id);
1226
1227                 if (reiserfs_barrier_flush(p_s_sb)) {
1228                         int ret;
1229                         lock_buffer(journal->j_header_bh);
1230                         ret = submit_barrier_buffer(journal->j_header_bh);
1231                         if (ret == -EOPNOTSUPP) {
1232                                 set_buffer_uptodate(journal->j_header_bh);
1233                                 disable_barrier(p_s_sb);
1234                                 goto sync;
1235                         }
1236                         wait_on_buffer(journal->j_header_bh);
1237                         check_barrier_completion(p_s_sb, journal->j_header_bh);
1238                 } else {
1239                       sync:
1240                         set_buffer_dirty(journal->j_header_bh);
1241                         sync_dirty_buffer(journal->j_header_bh);
1242                 }
1243                 if (!buffer_uptodate(journal->j_header_bh)) {
1244                         reiserfs_warning(p_s_sb,
1245                                          "journal-837: IO error during journal replay");
1246                         return -EIO;
1247                 }
1248         }
1249         return 0;
1250 }
1251
1252 static int update_journal_header_block(struct super_block *p_s_sb,
1253                                        unsigned long offset,
1254                                        unsigned long trans_id)
1255 {
1256         return _update_journal_header_block(p_s_sb, offset, trans_id);
1257 }
1258
1259 /*
1260 ** flush any and all journal lists older than you are
1261 ** can only be called from flush_journal_list
1262 */
1263 static int flush_older_journal_lists(struct super_block *p_s_sb,
1264                                      struct reiserfs_journal_list *jl)
1265 {
1266         struct list_head *entry;
1267         struct reiserfs_journal_list *other_jl;
1268         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1269         unsigned long trans_id = jl->j_trans_id;
1270
1271         /* we know we are the only ones flushing things, no extra race
1272          * protection is required.
1273          */
1274       restart:
1275         entry = journal->j_journal_list.next;
1276         /* Did we wrap? */
1277         if (entry == &journal->j_journal_list)
1278                 return 0;
1279         other_jl = JOURNAL_LIST_ENTRY(entry);
1280         if (other_jl->j_trans_id < trans_id) {
1281                 BUG_ON(other_jl->j_refcount <= 0);
1282                 /* do not flush all */
1283                 flush_journal_list(p_s_sb, other_jl, 0);
1284
1285                 /* other_jl is now deleted from the list */
1286                 goto restart;
1287         }
1288         return 0;
1289 }
1290
1291 static void del_from_work_list(struct super_block *s,
1292                                struct reiserfs_journal_list *jl)
1293 {
1294         struct reiserfs_journal *journal = SB_JOURNAL(s);
1295         if (!list_empty(&jl->j_working_list)) {
1296                 list_del_init(&jl->j_working_list);
1297                 journal->j_num_work_lists--;
1298         }
1299 }
1300
1301 /* flush a journal list, both commit and real blocks
1302 **
1303 ** always set flushall to 1, unless you are calling from inside
1304 ** flush_journal_list
1305 **
1306 ** IMPORTANT.  This can only be called while there are no journal writers,
1307 ** and the journal is locked.  That means it can only be called from
1308 ** do_journal_end, or by journal_release
1309 */
1310 static int flush_journal_list(struct super_block *s,
1311                               struct reiserfs_journal_list *jl, int flushall)
1312 {
1313         struct reiserfs_journal_list *pjl;
1314         struct reiserfs_journal_cnode *cn, *last;
1315         int count;
1316         int was_jwait = 0;
1317         int was_dirty = 0;
1318         struct buffer_head *saved_bh;
1319         unsigned long j_len_saved = jl->j_len;
1320         struct reiserfs_journal *journal = SB_JOURNAL(s);
1321         int err = 0;
1322
1323         BUG_ON(j_len_saved <= 0);
1324
1325         if (atomic_read(&journal->j_wcount) != 0) {
1326                 reiserfs_warning(s,
1327                                  "clm-2048: flush_journal_list called with wcount %d",
1328                                  atomic_read(&journal->j_wcount));
1329         }
1330         BUG_ON(jl->j_trans_id == 0);
1331
1332         /* if flushall == 0, the lock is already held */
1333         if (flushall) {
1334                 down(&journal->j_flush_sem);
1335         } else if (!down_trylock(&journal->j_flush_sem)) {
1336                 BUG();
1337         }
1338
1339         count = 0;
1340         if (j_len_saved > journal->j_trans_max) {
1341                 reiserfs_panic(s,
1342                                "journal-715: flush_journal_list, length is %lu, trans id %lu\n",
1343                                j_len_saved, jl->j_trans_id);
1344                 return 0;
1345         }
1346
1347         get_fs_excl();
1348
1349         /* if all the work is already done, get out of here */
1350         if (atomic_read(&(jl->j_nonzerolen)) <= 0 &&
1351             atomic_read(&(jl->j_commit_left)) <= 0) {
1352                 goto flush_older_and_return;
1353         }
1354
1355         /* start by putting the commit list on disk.  This will also flush
1356          ** the commit lists of any olders transactions
1357          */
1358         flush_commit_list(s, jl, 1);
1359
1360         if (!(jl->j_state & LIST_DIRTY)
1361             && !reiserfs_is_journal_aborted(journal))
1362                 BUG();
1363
1364         /* are we done now? */
1365         if (atomic_read(&(jl->j_nonzerolen)) <= 0 &&
1366             atomic_read(&(jl->j_commit_left)) <= 0) {
1367                 goto flush_older_and_return;
1368         }
1369
1370         /* loop through each cnode, see if we need to write it,
1371          ** or wait on a more recent transaction, or just ignore it
1372          */
1373         if (atomic_read(&(journal->j_wcount)) != 0) {
1374                 reiserfs_panic(s,
1375                                "journal-844: panic journal list is flushing, wcount is not 0\n");
1376         }
1377         cn = jl->j_realblock;
1378         while (cn) {
1379                 was_jwait = 0;
1380                 was_dirty = 0;
1381                 saved_bh = NULL;
1382                 /* blocknr of 0 is no longer in the hash, ignore it */
1383                 if (cn->blocknr == 0) {
1384                         goto free_cnode;
1385                 }
1386
1387                 /* This transaction failed commit. Don't write out to the disk */
1388                 if (!(jl->j_state & LIST_DIRTY))
1389                         goto free_cnode;
1390
1391                 pjl = find_newer_jl_for_cn(cn);
1392                 /* the order is important here.  We check pjl to make sure we
1393                  ** don't clear BH_JDirty_wait if we aren't the one writing this
1394                  ** block to disk
1395                  */
1396                 if (!pjl && cn->bh) {
1397                         saved_bh = cn->bh;
1398
1399                         /* we do this to make sure nobody releases the buffer while
1400                          ** we are working with it
1401                          */
1402                         get_bh(saved_bh);
1403
1404                         if (buffer_journal_dirty(saved_bh)) {
1405                                 BUG_ON(!can_dirty(cn));
1406                                 was_jwait = 1;
1407                                 was_dirty = 1;
1408                         } else if (can_dirty(cn)) {
1409                                 /* everything with !pjl && jwait should be writable */
1410                                 BUG();
1411                         }
1412                 }
1413
1414                 /* if someone has this block in a newer transaction, just make
1415                  ** sure they are commited, and don't try writing it to disk
1416                  */
1417                 if (pjl) {
1418                         if (atomic_read(&pjl->j_commit_left))
1419                                 flush_commit_list(s, pjl, 1);
1420                         goto free_cnode;
1421                 }
1422
1423                 /* bh == NULL when the block got to disk on its own, OR,
1424                  ** the block got freed in a future transaction
1425                  */
1426                 if (saved_bh == NULL) {
1427                         goto free_cnode;
1428                 }
1429
1430                 /* this should never happen.  kupdate_one_transaction has this list
1431                  ** locked while it works, so we should never see a buffer here that
1432                  ** is not marked JDirty_wait
1433                  */
1434                 if ((!was_jwait) && !buffer_locked(saved_bh)) {
1435                         reiserfs_warning(s,
1436                                          "journal-813: BAD! buffer %llu %cdirty %cjwait, "
1437                                          "not in a newer tranasction",
1438                                          (unsigned long long)saved_bh->
1439                                          b_blocknr, was_dirty ? ' ' : '!',
1440                                          was_jwait ? ' ' : '!');
1441                 }
1442                 if (was_dirty) {
1443                         /* we inc again because saved_bh gets decremented at free_cnode */
1444                         get_bh(saved_bh);
1445                         set_bit(BLOCK_NEEDS_FLUSH, &cn->state);
1446                         lock_buffer(saved_bh);
1447                         BUG_ON(cn->blocknr != saved_bh->b_blocknr);
1448                         if (buffer_dirty(saved_bh))
1449                                 submit_logged_buffer(saved_bh);
1450                         else
1451                                 unlock_buffer(saved_bh);
1452                         count++;
1453                 } else {
1454                         reiserfs_warning(s,
1455                                          "clm-2082: Unable to flush buffer %llu in %s",
1456                                          (unsigned long long)saved_bh->
1457                                          b_blocknr, __FUNCTION__);
1458                 }
1459               free_cnode:
1460                 last = cn;
1461                 cn = cn->next;
1462                 if (saved_bh) {
1463                         /* we incremented this to keep others from taking the buffer head away */
1464                         put_bh(saved_bh);
1465                         if (atomic_read(&(saved_bh->b_count)) < 0) {
1466                                 reiserfs_warning(s,
1467                                                  "journal-945: saved_bh->b_count < 0");
1468                         }
1469                 }
1470         }
1471         if (count > 0) {
1472                 cn = jl->j_realblock;
1473                 while (cn) {
1474                         if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) {
1475                                 if (!cn->bh) {
1476                                         reiserfs_panic(s,
1477                                                        "journal-1011: cn->bh is NULL\n");
1478                                 }
1479                                 wait_on_buffer(cn->bh);
1480                                 if (!cn->bh) {
1481                                         reiserfs_panic(s,
1482                                                        "journal-1012: cn->bh is NULL\n");
1483                                 }
1484                                 if (unlikely(!buffer_uptodate(cn->bh))) {
1485 #ifdef CONFIG_REISERFS_CHECK
1486                                         reiserfs_warning(s,
1487                                                          "journal-949: buffer write failed\n");
1488 #endif
1489                                         err = -EIO;
1490                                 }
1491                                 /* note, we must clear the JDirty_wait bit after the up to date
1492                                  ** check, otherwise we race against our flushpage routine
1493                                  */
1494                                 BUG_ON(!test_clear_buffer_journal_dirty
1495                                        (cn->bh));
1496
1497                                 /* undo the inc from journal_mark_dirty */
1498                                 put_bh(cn->bh);
1499                                 brelse(cn->bh);
1500                         }
1501                         cn = cn->next;
1502                 }
1503         }
1504
1505         if (err)
1506                 reiserfs_abort(s, -EIO,
1507                                "Write error while pushing transaction to disk in %s",
1508                                __FUNCTION__);
1509       flush_older_and_return:
1510
1511         /* before we can update the journal header block, we _must_ flush all
1512          ** real blocks from all older transactions to disk.  This is because
1513          ** once the header block is updated, this transaction will not be
1514          ** replayed after a crash
1515          */
1516         if (flushall) {
1517                 flush_older_journal_lists(s, jl);
1518         }
1519
1520         err = journal->j_errno;
1521         /* before we can remove everything from the hash tables for this
1522          ** transaction, we must make sure it can never be replayed
1523          **
1524          ** since we are only called from do_journal_end, we know for sure there
1525          ** are no allocations going on while we are flushing journal lists.  So,
1526          ** we only need to update the journal header block for the last list
1527          ** being flushed
1528          */
1529         if (!err && flushall) {
1530                 err =
1531                     update_journal_header_block(s,
1532                                                 (jl->j_start + jl->j_len +
1533                                                  2) % SB_ONDISK_JOURNAL_SIZE(s),
1534                                                 jl->j_trans_id);
1535                 if (err)
1536                         reiserfs_abort(s, -EIO,
1537                                        "Write error while updating journal header in %s",
1538                                        __FUNCTION__);
1539         }
1540         remove_all_from_journal_list(s, jl, 0);
1541         list_del_init(&jl->j_list);
1542         journal->j_num_lists--;
1543         del_from_work_list(s, jl);
1544
1545         if (journal->j_last_flush_id != 0 &&
1546             (jl->j_trans_id - journal->j_last_flush_id) != 1) {
1547                 reiserfs_warning(s, "clm-2201: last flush %lu, current %lu",
1548                                  journal->j_last_flush_id, jl->j_trans_id);
1549         }
1550         journal->j_last_flush_id = jl->j_trans_id;
1551
1552         /* not strictly required since we are freeing the list, but it should
1553          * help find code using dead lists later on
1554          */
1555         jl->j_len = 0;
1556         atomic_set(&(jl->j_nonzerolen), 0);
1557         jl->j_start = 0;
1558         jl->j_realblock = NULL;
1559         jl->j_commit_bh = NULL;
1560         jl->j_trans_id = 0;
1561         jl->j_state = 0;
1562         put_journal_list(s, jl);
1563         if (flushall)
1564                 up(&journal->j_flush_sem);
1565         put_fs_excl();
1566         return err;
1567 }
1568
1569 static int write_one_transaction(struct super_block *s,
1570                                  struct reiserfs_journal_list *jl,
1571                                  struct buffer_chunk *chunk)
1572 {
1573         struct reiserfs_journal_cnode *cn;
1574         int ret = 0;
1575
1576         jl->j_state |= LIST_TOUCHED;
1577         del_from_work_list(s, jl);
1578         if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0) {
1579                 return 0;
1580         }
1581
1582         cn = jl->j_realblock;
1583         while (cn) {
1584                 /* if the blocknr == 0, this has been cleared from the hash,
1585                  ** skip it
1586                  */
1587                 if (cn->blocknr == 0) {
1588                         goto next;
1589                 }
1590                 if (cn->bh && can_dirty(cn) && buffer_dirty(cn->bh)) {
1591                         struct buffer_head *tmp_bh;
1592                         /* we can race against journal_mark_freed when we try
1593                          * to lock_buffer(cn->bh), so we have to inc the buffer
1594                          * count, and recheck things after locking
1595                          */
1596                         tmp_bh = cn->bh;
1597                         get_bh(tmp_bh);
1598                         lock_buffer(tmp_bh);
1599                         if (cn->bh && can_dirty(cn) && buffer_dirty(tmp_bh)) {
1600                                 if (!buffer_journal_dirty(tmp_bh) ||
1601                                     buffer_journal_prepared(tmp_bh))
1602                                         BUG();
1603                                 add_to_chunk(chunk, tmp_bh, NULL, write_chunk);
1604                                 ret++;
1605                         } else {
1606                                 /* note, cn->bh might be null now */
1607                                 unlock_buffer(tmp_bh);
1608                         }
1609                         put_bh(tmp_bh);
1610                 }
1611               next:
1612                 cn = cn->next;
1613                 cond_resched();
1614         }
1615         return ret;
1616 }
1617
1618 /* used by flush_commit_list */
1619 static int dirty_one_transaction(struct super_block *s,
1620                                  struct reiserfs_journal_list *jl)
1621 {
1622         struct reiserfs_journal_cnode *cn;
1623         struct reiserfs_journal_list *pjl;
1624         int ret = 0;
1625
1626         jl->j_state |= LIST_DIRTY;
1627         cn = jl->j_realblock;
1628         while (cn) {
1629                 /* look for a more recent transaction that logged this
1630                  ** buffer.  Only the most recent transaction with a buffer in
1631                  ** it is allowed to send that buffer to disk
1632                  */
1633                 pjl = find_newer_jl_for_cn(cn);
1634                 if (!pjl && cn->blocknr && cn->bh
1635                     && buffer_journal_dirty(cn->bh)) {
1636                         BUG_ON(!can_dirty(cn));
1637                         /* if the buffer is prepared, it will either be logged
1638                          * or restored.  If restored, we need to make sure
1639                          * it actually gets marked dirty
1640                          */
1641                         clear_buffer_journal_new(cn->bh);
1642                         if (buffer_journal_prepared(cn->bh)) {
1643                                 set_buffer_journal_restore_dirty(cn->bh);
1644                         } else {
1645                                 set_buffer_journal_test(cn->bh);
1646                                 mark_buffer_dirty(cn->bh);
1647                         }
1648                 }
1649                 cn = cn->next;
1650         }
1651         return ret;
1652 }
1653
1654 static int kupdate_transactions(struct super_block *s,
1655                                 struct reiserfs_journal_list *jl,
1656                                 struct reiserfs_journal_list **next_jl,
1657                                 unsigned long *next_trans_id,
1658                                 int num_blocks, int num_trans)
1659 {
1660         int ret = 0;
1661         int written = 0;
1662         int transactions_flushed = 0;
1663         unsigned long orig_trans_id = jl->j_trans_id;
1664         struct buffer_chunk chunk;
1665         struct list_head *entry;
1666         struct reiserfs_journal *journal = SB_JOURNAL(s);
1667         chunk.nr = 0;
1668
1669         down(&journal->j_flush_sem);
1670         if (!journal_list_still_alive(s, orig_trans_id)) {
1671                 goto done;
1672         }
1673
1674         /* we've got j_flush_sem held, nobody is going to delete any
1675          * of these lists out from underneath us
1676          */
1677         while ((num_trans && transactions_flushed < num_trans) ||
1678                (!num_trans && written < num_blocks)) {
1679
1680                 if (jl->j_len == 0 || (jl->j_state & LIST_TOUCHED) ||
1681                     atomic_read(&jl->j_commit_left)
1682                     || !(jl->j_state & LIST_DIRTY)) {
1683                         del_from_work_list(s, jl);
1684                         break;
1685                 }
1686                 ret = write_one_transaction(s, jl, &chunk);
1687
1688                 if (ret < 0)
1689                         goto done;
1690                 transactions_flushed++;
1691                 written += ret;
1692                 entry = jl->j_list.next;
1693
1694                 /* did we wrap? */
1695                 if (entry == &journal->j_journal_list) {
1696                         break;
1697                 }
1698                 jl = JOURNAL_LIST_ENTRY(entry);
1699
1700                 /* don't bother with older transactions */
1701                 if (jl->j_trans_id <= orig_trans_id)
1702                         break;
1703         }
1704         if (chunk.nr) {
1705                 write_chunk(&chunk);
1706         }
1707
1708       done:
1709         up(&journal->j_flush_sem);
1710         return ret;
1711 }
1712
1713 /* for o_sync and fsync heavy applications, they tend to use
1714 ** all the journa list slots with tiny transactions.  These
1715 ** trigger lots and lots of calls to update the header block, which
1716 ** adds seeks and slows things down.
1717 **
1718 ** This function tries to clear out a large chunk of the journal lists
1719 ** at once, which makes everything faster since only the newest journal
1720 ** list updates the header block
1721 */
1722 static int flush_used_journal_lists(struct super_block *s,
1723                                     struct reiserfs_journal_list *jl)
1724 {
1725         unsigned long len = 0;
1726         unsigned long cur_len;
1727         int ret;
1728         int i;
1729         int limit = 256;
1730         struct reiserfs_journal_list *tjl;
1731         struct reiserfs_journal_list *flush_jl;
1732         unsigned long trans_id;
1733         struct reiserfs_journal *journal = SB_JOURNAL(s);
1734
1735         flush_jl = tjl = jl;
1736
1737         /* in data logging mode, try harder to flush a lot of blocks */
1738         if (reiserfs_data_log(s))
1739                 limit = 1024;
1740         /* flush for 256 transactions or limit blocks, whichever comes first */
1741         for (i = 0; i < 256 && len < limit; i++) {
1742                 if (atomic_read(&tjl->j_commit_left) ||
1743                     tjl->j_trans_id < jl->j_trans_id) {
1744                         break;
1745                 }
1746                 cur_len = atomic_read(&tjl->j_nonzerolen);
1747                 if (cur_len > 0) {
1748                         tjl->j_state &= ~LIST_TOUCHED;
1749                 }
1750                 len += cur_len;
1751                 flush_jl = tjl;
1752                 if (tjl->j_list.next == &journal->j_journal_list)
1753                         break;
1754                 tjl = JOURNAL_LIST_ENTRY(tjl->j_list.next);
1755         }
1756         /* try to find a group of blocks we can flush across all the
1757          ** transactions, but only bother if we've actually spanned
1758          ** across multiple lists
1759          */
1760         if (flush_jl != jl) {
1761                 ret = kupdate_transactions(s, jl, &tjl, &trans_id, len, i);
1762         }
1763         flush_journal_list(s, flush_jl, 1);
1764         return 0;
1765 }
1766
1767 /*
1768 ** removes any nodes in table with name block and dev as bh.
1769 ** only touchs the hnext and hprev pointers.
1770 */
1771 void remove_journal_hash(struct super_block *sb,
1772                          struct reiserfs_journal_cnode **table,
1773                          struct reiserfs_journal_list *jl,
1774                          unsigned long block, int remove_freed)
1775 {
1776         struct reiserfs_journal_cnode *cur;
1777         struct reiserfs_journal_cnode **head;
1778
1779         head = &(journal_hash(table, sb, block));
1780         if (!head) {
1781                 return;
1782         }
1783         cur = *head;
1784         while (cur) {
1785                 if (cur->blocknr == block && cur->sb == sb
1786                     && (jl == NULL || jl == cur->jlist)
1787                     && (!test_bit(BLOCK_FREED, &cur->state) || remove_freed)) {
1788                         if (cur->hnext) {
1789                                 cur->hnext->hprev = cur->hprev;
1790                         }
1791                         if (cur->hprev) {
1792                                 cur->hprev->hnext = cur->hnext;
1793                         } else {
1794                                 *head = cur->hnext;
1795                         }
1796                         cur->blocknr = 0;
1797                         cur->sb = NULL;
1798                         cur->state = 0;
1799                         if (cur->bh && cur->jlist)      /* anybody who clears the cur->bh will also dec the nonzerolen */
1800                                 atomic_dec(&(cur->jlist->j_nonzerolen));
1801                         cur->bh = NULL;
1802                         cur->jlist = NULL;
1803                 }
1804                 cur = cur->hnext;
1805         }
1806 }
1807
1808 static void free_journal_ram(struct super_block *p_s_sb)
1809 {
1810         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1811         reiserfs_kfree(journal->j_current_jl,
1812                        sizeof(struct reiserfs_journal_list), p_s_sb);
1813         journal->j_num_lists--;
1814
1815         vfree(journal->j_cnode_free_orig);
1816         free_list_bitmaps(p_s_sb, journal->j_list_bitmap);
1817         free_bitmap_nodes(p_s_sb);      /* must be after free_list_bitmaps */
1818         if (journal->j_header_bh) {
1819                 brelse(journal->j_header_bh);
1820         }
1821         /* j_header_bh is on the journal dev, make sure not to release the journal
1822          * dev until we brelse j_header_bh
1823          */
1824         release_journal_dev(p_s_sb, journal);
1825         vfree(journal);
1826 }
1827
1828 /*
1829 ** call on unmount.  Only set error to 1 if you haven't made your way out
1830 ** of read_super() yet.  Any other caller must keep error at 0.
1831 */
1832 static int do_journal_release(struct reiserfs_transaction_handle *th,
1833                               struct super_block *p_s_sb, int error)
1834 {
1835         struct reiserfs_transaction_handle myth;
1836         int flushed = 0;
1837         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1838
1839         /* we only want to flush out transactions if we were called with error == 0
1840          */
1841         if (!error && !(p_s_sb->s_flags & MS_RDONLY)) {
1842                 /* end the current trans */
1843                 BUG_ON(!th->t_trans_id);
1844                 do_journal_end(th, p_s_sb, 10, FLUSH_ALL);
1845
1846                 /* make sure something gets logged to force our way into the flush code */
1847                 if (!journal_join(&myth, p_s_sb, 1)) {
1848                         reiserfs_prepare_for_journal(p_s_sb,
1849                                                      SB_BUFFER_WITH_SB(p_s_sb),
1850                                                      1);
1851                         journal_mark_dirty(&myth, p_s_sb,
1852                                            SB_BUFFER_WITH_SB(p_s_sb));
1853                         do_journal_end(&myth, p_s_sb, 1, FLUSH_ALL);
1854                         flushed = 1;
1855                 }
1856         }
1857
1858         /* this also catches errors during the do_journal_end above */
1859         if (!error && reiserfs_is_journal_aborted(journal)) {
1860                 memset(&myth, 0, sizeof(myth));
1861                 if (!journal_join_abort(&myth, p_s_sb, 1)) {
1862                         reiserfs_prepare_for_journal(p_s_sb,
1863                                                      SB_BUFFER_WITH_SB(p_s_sb),
1864                                                      1);
1865                         journal_mark_dirty(&myth, p_s_sb,
1866                                            SB_BUFFER_WITH_SB(p_s_sb));
1867                         do_journal_end(&myth, p_s_sb, 1, FLUSH_ALL);
1868                 }
1869         }
1870
1871         reiserfs_mounted_fs_count--;
1872         /* wait for all commits to finish */
1873         cancel_delayed_work(&SB_JOURNAL(p_s_sb)->j_work);
1874         flush_workqueue(commit_wq);
1875         if (!reiserfs_mounted_fs_count) {
1876                 destroy_workqueue(commit_wq);
1877                 commit_wq = NULL;
1878         }
1879
1880         free_journal_ram(p_s_sb);
1881
1882         return 0;
1883 }
1884
1885 /*
1886 ** call on unmount.  flush all journal trans, release all alloc'd ram
1887 */
1888 int journal_release(struct reiserfs_transaction_handle *th,
1889                     struct super_block *p_s_sb)
1890 {
1891         return do_journal_release(th, p_s_sb, 0);
1892 }
1893
1894 /*
1895 ** only call from an error condition inside reiserfs_read_super!
1896 */
1897 int journal_release_error(struct reiserfs_transaction_handle *th,
1898                           struct super_block *p_s_sb)
1899 {
1900         return do_journal_release(th, p_s_sb, 1);
1901 }
1902
1903 /* compares description block with commit block.  returns 1 if they differ, 0 if they are the same */
1904 static int journal_compare_desc_commit(struct super_block *p_s_sb,
1905                                        struct reiserfs_journal_desc *desc,
1906                                        struct reiserfs_journal_commit *commit)
1907 {
1908         if (get_commit_trans_id(commit) != get_desc_trans_id(desc) ||
1909             get_commit_trans_len(commit) != get_desc_trans_len(desc) ||
1910             get_commit_trans_len(commit) > SB_JOURNAL(p_s_sb)->j_trans_max ||
1911             get_commit_trans_len(commit) <= 0) {
1912                 return 1;
1913         }
1914         return 0;
1915 }
1916
1917 /* returns 0 if it did not find a description block
1918 ** returns -1 if it found a corrupt commit block
1919 ** returns 1 if both desc and commit were valid
1920 */
1921 static int journal_transaction_is_valid(struct super_block *p_s_sb,
1922                                         struct buffer_head *d_bh,
1923                                         unsigned long *oldest_invalid_trans_id,
1924                                         unsigned long *newest_mount_id)
1925 {
1926         struct reiserfs_journal_desc *desc;
1927         struct reiserfs_journal_commit *commit;
1928         struct buffer_head *c_bh;
1929         unsigned long offset;
1930
1931         if (!d_bh)
1932                 return 0;
1933
1934         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
1935         if (get_desc_trans_len(desc) > 0
1936             && !memcmp(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8)) {
1937                 if (oldest_invalid_trans_id && *oldest_invalid_trans_id
1938                     && get_desc_trans_id(desc) > *oldest_invalid_trans_id) {
1939                         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
1940                                        "journal-986: transaction "
1941                                        "is valid returning because trans_id %d is greater than "
1942                                        "oldest_invalid %lu",
1943                                        get_desc_trans_id(desc),
1944                                        *oldest_invalid_trans_id);
1945                         return 0;
1946                 }
1947                 if (newest_mount_id
1948                     && *newest_mount_id > get_desc_mount_id(desc)) {
1949                         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
1950                                        "journal-1087: transaction "
1951                                        "is valid returning because mount_id %d is less than "
1952                                        "newest_mount_id %lu",
1953                                        get_desc_mount_id(desc),
1954                                        *newest_mount_id);
1955                         return -1;
1956                 }
1957                 if (get_desc_trans_len(desc) > SB_JOURNAL(p_s_sb)->j_trans_max) {
1958                         reiserfs_warning(p_s_sb,
1959                                          "journal-2018: Bad transaction length %d encountered, ignoring transaction",
1960                                          get_desc_trans_len(desc));
1961                         return -1;
1962                 }
1963                 offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb);
1964
1965                 /* ok, we have a journal description block, lets see if the transaction was valid */
1966                 c_bh =
1967                     journal_bread(p_s_sb,
1968                                   SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
1969                                   ((offset + get_desc_trans_len(desc) +
1970                                     1) % SB_ONDISK_JOURNAL_SIZE(p_s_sb)));
1971                 if (!c_bh)
1972                         return 0;
1973                 commit = (struct reiserfs_journal_commit *)c_bh->b_data;
1974                 if (journal_compare_desc_commit(p_s_sb, desc, commit)) {
1975                         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
1976                                        "journal_transaction_is_valid, commit offset %ld had bad "
1977                                        "time %d or length %d",
1978                                        c_bh->b_blocknr -
1979                                        SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
1980                                        get_commit_trans_id(commit),
1981                                        get_commit_trans_len(commit));
1982                         brelse(c_bh);
1983                         if (oldest_invalid_trans_id) {
1984                                 *oldest_invalid_trans_id =
1985                                     get_desc_trans_id(desc);
1986                                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
1987                                                "journal-1004: "
1988                                                "transaction_is_valid setting oldest invalid trans_id "
1989                                                "to %d",
1990                                                get_desc_trans_id(desc));
1991                         }
1992                         return -1;
1993                 }
1994                 brelse(c_bh);
1995                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
1996                                "journal-1006: found valid "
1997                                "transaction start offset %llu, len %d id %d",
1998                                d_bh->b_blocknr -
1999                                SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2000                                get_desc_trans_len(desc),
2001                                get_desc_trans_id(desc));
2002                 return 1;
2003         } else {
2004                 return 0;
2005         }
2006 }
2007
2008 static void brelse_array(struct buffer_head **heads, int num)
2009 {
2010         int i;
2011         for (i = 0; i < num; i++) {
2012                 brelse(heads[i]);
2013         }
2014 }
2015
2016 /*
2017 ** given the start, and values for the oldest acceptable transactions,
2018 ** this either reads in a replays a transaction, or returns because the transaction
2019 ** is invalid, or too old.
2020 */
2021 static int journal_read_transaction(struct super_block *p_s_sb,
2022                                     unsigned long cur_dblock,
2023                                     unsigned long oldest_start,
2024                                     unsigned long oldest_trans_id,
2025                                     unsigned long newest_mount_id)
2026 {
2027         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
2028         struct reiserfs_journal_desc *desc;
2029         struct reiserfs_journal_commit *commit;
2030         unsigned long trans_id = 0;
2031         struct buffer_head *c_bh;
2032         struct buffer_head *d_bh;
2033         struct buffer_head **log_blocks = NULL;
2034         struct buffer_head **real_blocks = NULL;
2035         unsigned long trans_offset;
2036         int i;
2037         int trans_half;
2038
2039         d_bh = journal_bread(p_s_sb, cur_dblock);
2040         if (!d_bh)
2041                 return 1;
2042         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2043         trans_offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb);
2044         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1037: "
2045                        "journal_read_transaction, offset %llu, len %d mount_id %d",
2046                        d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2047                        get_desc_trans_len(desc), get_desc_mount_id(desc));
2048         if (get_desc_trans_id(desc) < oldest_trans_id) {
2049                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1039: "
2050                                "journal_read_trans skipping because %lu is too old",
2051                                cur_dblock -
2052                                SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb));
2053                 brelse(d_bh);
2054                 return 1;
2055         }
2056         if (get_desc_mount_id(desc) != newest_mount_id) {
2057                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1146: "
2058                                "journal_read_trans skipping because %d is != "
2059                                "newest_mount_id %lu", get_desc_mount_id(desc),
2060                                newest_mount_id);
2061                 brelse(d_bh);
2062                 return 1;
2063         }
2064         c_bh = journal_bread(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2065                              ((trans_offset + get_desc_trans_len(desc) + 1) %
2066                               SB_ONDISK_JOURNAL_SIZE(p_s_sb)));
2067         if (!c_bh) {
2068                 brelse(d_bh);
2069                 return 1;
2070         }
2071         commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2072         if (journal_compare_desc_commit(p_s_sb, desc, commit)) {
2073                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2074                                "journal_read_transaction, "
2075                                "commit offset %llu had bad time %d or length %d",
2076                                c_bh->b_blocknr -
2077                                SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2078                                get_commit_trans_id(commit),
2079                                get_commit_trans_len(commit));
2080                 brelse(c_bh);
2081                 brelse(d_bh);
2082                 return 1;
2083         }
2084         trans_id = get_desc_trans_id(desc);
2085         /* now we know we've got a good transaction, and it was inside the valid time ranges */
2086         log_blocks =
2087             reiserfs_kmalloc(get_desc_trans_len(desc) *
2088                              sizeof(struct buffer_head *), GFP_NOFS, p_s_sb);
2089         real_blocks =
2090             reiserfs_kmalloc(get_desc_trans_len(desc) *
2091                              sizeof(struct buffer_head *), GFP_NOFS, p_s_sb);
2092         if (!log_blocks || !real_blocks) {
2093                 brelse(c_bh);
2094                 brelse(d_bh);
2095                 reiserfs_kfree(log_blocks,
2096                                get_desc_trans_len(desc) *
2097                                sizeof(struct buffer_head *), p_s_sb);
2098                 reiserfs_kfree(real_blocks,
2099                                get_desc_trans_len(desc) *
2100                                sizeof(struct buffer_head *), p_s_sb);
2101                 reiserfs_warning(p_s_sb,
2102                                  "journal-1169: kmalloc failed, unable to mount FS");
2103                 return -1;
2104         }
2105         /* get all the buffer heads */
2106         trans_half = journal_trans_half(p_s_sb->s_blocksize);
2107         for (i = 0; i < get_desc_trans_len(desc); i++) {
2108                 log_blocks[i] =
2109                     journal_getblk(p_s_sb,
2110                                    SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2111                                    (trans_offset + 1 +
2112                                     i) % SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2113                 if (i < trans_half) {
2114                         real_blocks[i] =
2115                             sb_getblk(p_s_sb,
2116                                       le32_to_cpu(desc->j_realblock[i]));
2117                 } else {
2118                         real_blocks[i] =
2119                             sb_getblk(p_s_sb,
2120                                       le32_to_cpu(commit->
2121                                                   j_realblock[i - trans_half]));
2122                 }
2123                 if (real_blocks[i]->b_blocknr > SB_BLOCK_COUNT(p_s_sb)) {
2124                         reiserfs_warning(p_s_sb,
2125                                          "journal-1207: REPLAY FAILURE fsck required! Block to replay is outside of filesystem");
2126                         goto abort_replay;
2127                 }
2128                 /* make sure we don't try to replay onto log or reserved area */
2129                 if (is_block_in_log_or_reserved_area
2130                     (p_s_sb, real_blocks[i]->b_blocknr)) {
2131                         reiserfs_warning(p_s_sb,
2132                                          "journal-1204: REPLAY FAILURE fsck required! Trying to replay onto a log block");
2133                       abort_replay:
2134                         brelse_array(log_blocks, i);
2135                         brelse_array(real_blocks, i);
2136                         brelse(c_bh);
2137                         brelse(d_bh);
2138                         reiserfs_kfree(log_blocks,
2139                                        get_desc_trans_len(desc) *
2140                                        sizeof(struct buffer_head *), p_s_sb);
2141                         reiserfs_kfree(real_blocks,
2142                                        get_desc_trans_len(desc) *
2143                                        sizeof(struct buffer_head *), p_s_sb);
2144                         return -1;
2145                 }
2146         }
2147         /* read in the log blocks, memcpy to the corresponding real block */
2148         ll_rw_block(READ, get_desc_trans_len(desc), log_blocks);
2149         for (i = 0; i < get_desc_trans_len(desc); i++) {
2150                 wait_on_buffer(log_blocks[i]);
2151                 if (!buffer_uptodate(log_blocks[i])) {
2152                         reiserfs_warning(p_s_sb,
2153                                          "journal-1212: REPLAY FAILURE fsck required! buffer write failed");
2154                         brelse_array(log_blocks + i,
2155                                      get_desc_trans_len(desc) - i);
2156                         brelse_array(real_blocks, get_desc_trans_len(desc));
2157                         brelse(c_bh);
2158                         brelse(d_bh);
2159                         reiserfs_kfree(log_blocks,
2160                                        get_desc_trans_len(desc) *
2161                                        sizeof(struct buffer_head *), p_s_sb);
2162                         reiserfs_kfree(real_blocks,
2163                                        get_desc_trans_len(desc) *
2164                                        sizeof(struct buffer_head *), p_s_sb);
2165                         return -1;
2166                 }
2167                 memcpy(real_blocks[i]->b_data, log_blocks[i]->b_data,
2168                        real_blocks[i]->b_size);
2169                 set_buffer_uptodate(real_blocks[i]);
2170                 brelse(log_blocks[i]);
2171         }
2172         /* flush out the real blocks */
2173         for (i = 0; i < get_desc_trans_len(desc); i++) {
2174                 set_buffer_dirty(real_blocks[i]);
2175                 ll_rw_block(SWRITE, 1, real_blocks + i);
2176         }
2177         for (i = 0; i < get_desc_trans_len(desc); i++) {
2178                 wait_on_buffer(real_blocks[i]);
2179                 if (!buffer_uptodate(real_blocks[i])) {
2180                         reiserfs_warning(p_s_sb,
2181                                          "journal-1226: REPLAY FAILURE, fsck required! buffer write failed");
2182                         brelse_array(real_blocks + i,
2183                                      get_desc_trans_len(desc) - i);
2184                         brelse(c_bh);
2185                         brelse(d_bh);
2186                         reiserfs_kfree(log_blocks,
2187                                        get_desc_trans_len(desc) *
2188                                        sizeof(struct buffer_head *), p_s_sb);
2189                         reiserfs_kfree(real_blocks,
2190                                        get_desc_trans_len(desc) *
2191                                        sizeof(struct buffer_head *), p_s_sb);
2192                         return -1;
2193                 }
2194                 brelse(real_blocks[i]);
2195         }
2196         cur_dblock =
2197             SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2198             ((trans_offset + get_desc_trans_len(desc) +
2199               2) % SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2200         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2201                        "journal-1095: setting journal " "start to offset %ld",
2202                        cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb));
2203
2204         /* init starting values for the first transaction, in case this is the last transaction to be replayed. */
2205         journal->j_start = cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb);
2206         journal->j_last_flush_trans_id = trans_id;
2207         journal->j_trans_id = trans_id + 1;
2208         brelse(c_bh);
2209         brelse(d_bh);
2210         reiserfs_kfree(log_blocks,
2211                        le32_to_cpu(desc->j_len) * sizeof(struct buffer_head *),
2212                        p_s_sb);
2213         reiserfs_kfree(real_blocks,
2214                        le32_to_cpu(desc->j_len) * sizeof(struct buffer_head *),
2215                        p_s_sb);
2216         return 0;
2217 }
2218
2219 /* This function reads blocks starting from block and to max_block of bufsize
2220    size (but no more than BUFNR blocks at a time). This proved to improve
2221    mounting speed on self-rebuilding raid5 arrays at least.
2222    Right now it is only used from journal code. But later we might use it
2223    from other places.
2224    Note: Do not use journal_getblk/sb_getblk functions here! */
2225 static struct buffer_head *reiserfs_breada(struct block_device *dev, int block,
2226                                            int bufsize, unsigned int max_block)
2227 {
2228         struct buffer_head *bhlist[BUFNR];
2229         unsigned int blocks = BUFNR;
2230         struct buffer_head *bh;
2231         int i, j;
2232
2233         bh = __getblk(dev, block, bufsize);
2234         if (buffer_uptodate(bh))
2235                 return (bh);
2236
2237         if (block + BUFNR > max_block) {
2238                 blocks = max_block - block;
2239         }
2240         bhlist[0] = bh;
2241         j = 1;
2242         for (i = 1; i < blocks; i++) {
2243                 bh = __getblk(dev, block + i, bufsize);
2244                 if (buffer_uptodate(bh)) {
2245                         brelse(bh);
2246                         break;
2247                 } else
2248                         bhlist[j++] = bh;
2249         }
2250         ll_rw_block(READ, j, bhlist);
2251         for (i = 1; i < j; i++)
2252                 brelse(bhlist[i]);
2253         bh = bhlist[0];
2254         wait_on_buffer(bh);
2255         if (buffer_uptodate(bh))
2256                 return bh;
2257         brelse(bh);
2258         return NULL;
2259 }
2260
2261 /*
2262 ** read and replay the log
2263 ** on a clean unmount, the journal header's next unflushed pointer will be to an invalid
2264 ** transaction.  This tests that before finding all the transactions in the log, which makes normal mount times fast.
2265 **
2266 ** After a crash, this starts with the next unflushed transaction, and replays until it finds one too old, or invalid.
2267 **
2268 ** On exit, it sets things up so the first transaction will work correctly.
2269 */
2270 static int journal_read(struct super_block *p_s_sb)
2271 {
2272         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
2273         struct reiserfs_journal_desc *desc;
2274         unsigned long oldest_trans_id = 0;
2275         unsigned long oldest_invalid_trans_id = 0;
2276         time_t start;
2277         unsigned long oldest_start = 0;
2278         unsigned long cur_dblock = 0;
2279         unsigned long newest_mount_id = 9;
2280         struct buffer_head *d_bh;
2281         struct reiserfs_journal_header *jh;
2282         int valid_journal_header = 0;
2283         int replay_count = 0;
2284         int continue_replay = 1;
2285         int ret;
2286         char b[BDEVNAME_SIZE];
2287
2288         cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb);
2289         reiserfs_info(p_s_sb, "checking transaction log (%s)\n",
2290                       bdevname(journal->j_dev_bd, b));
2291         start = get_seconds();
2292
2293         /* step 1, read in the journal header block.  Check the transaction it says
2294          ** is the first unflushed, and if that transaction is not valid,
2295          ** replay is done
2296          */
2297         journal->j_header_bh = journal_bread(p_s_sb,
2298                                              SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb)
2299                                              + SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2300         if (!journal->j_header_bh) {
2301                 return 1;
2302         }
2303         jh = (struct reiserfs_journal_header *)(journal->j_header_bh->b_data);
2304         if (le32_to_cpu(jh->j_first_unflushed_offset) >= 0 &&
2305             le32_to_cpu(jh->j_first_unflushed_offset) <
2306             SB_ONDISK_JOURNAL_SIZE(p_s_sb)
2307             && le32_to_cpu(jh->j_last_flush_trans_id) > 0) {
2308                 oldest_start =
2309                     SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2310                     le32_to_cpu(jh->j_first_unflushed_offset);
2311                 oldest_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2312                 newest_mount_id = le32_to_cpu(jh->j_mount_id);
2313                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2314                                "journal-1153: found in "
2315                                "header: first_unflushed_offset %d, last_flushed_trans_id "
2316                                "%lu", le32_to_cpu(jh->j_first_unflushed_offset),
2317                                le32_to_cpu(jh->j_last_flush_trans_id));
2318                 valid_journal_header = 1;
2319
2320                 /* now, we try to read the first unflushed offset.  If it is not valid,
2321                  ** there is nothing more we can do, and it makes no sense to read
2322                  ** through the whole log.
2323                  */
2324                 d_bh =
2325                     journal_bread(p_s_sb,
2326                                   SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2327                                   le32_to_cpu(jh->j_first_unflushed_offset));
2328                 ret = journal_transaction_is_valid(p_s_sb, d_bh, NULL, NULL);
2329                 if (!ret) {
2330                         continue_replay = 0;
2331                 }
2332                 brelse(d_bh);
2333                 goto start_log_replay;
2334         }
2335
2336         if (continue_replay && bdev_read_only(p_s_sb->s_bdev)) {
2337                 reiserfs_warning(p_s_sb,
2338                                  "clm-2076: device is readonly, unable to replay log");
2339                 return -1;
2340         }
2341
2342         /* ok, there are transactions that need to be replayed.  start with the first log block, find
2343          ** all the valid transactions, and pick out the oldest.
2344          */
2345         while (continue_replay
2346                && cur_dblock <
2347                (SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2348                 SB_ONDISK_JOURNAL_SIZE(p_s_sb))) {
2349                 /* Note that it is required for blocksize of primary fs device and journal
2350                    device to be the same */
2351                 d_bh =
2352                     reiserfs_breada(journal->j_dev_bd, cur_dblock,
2353                                     p_s_sb->s_blocksize,
2354                                     SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2355                                     SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2356                 ret =
2357                     journal_transaction_is_valid(p_s_sb, d_bh,
2358                                                  &oldest_invalid_trans_id,
2359                                                  &newest_mount_id);
2360                 if (ret == 1) {
2361                         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2362                         if (oldest_start == 0) {        /* init all oldest_ values */
2363                                 oldest_trans_id = get_desc_trans_id(desc);
2364                                 oldest_start = d_bh->b_blocknr;
2365                                 newest_mount_id = get_desc_mount_id(desc);
2366                                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2367                                                "journal-1179: Setting "
2368                                                "oldest_start to offset %llu, trans_id %lu",
2369                                                oldest_start -
2370                                                SB_ONDISK_JOURNAL_1st_BLOCK
2371                                                (p_s_sb), oldest_trans_id);
2372                         } else if (oldest_trans_id > get_desc_trans_id(desc)) {
2373                                 /* one we just read was older */
2374                                 oldest_trans_id = get_desc_trans_id(desc);
2375                                 oldest_start = d_bh->b_blocknr;
2376                                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2377                                                "journal-1180: Resetting "
2378                                                "oldest_start to offset %lu, trans_id %lu",
2379                                                oldest_start -
2380                                                SB_ONDISK_JOURNAL_1st_BLOCK
2381                                                (p_s_sb), oldest_trans_id);
2382                         }
2383                         if (newest_mount_id < get_desc_mount_id(desc)) {
2384                                 newest_mount_id = get_desc_mount_id(desc);
2385                                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2386                                                "journal-1299: Setting "
2387                                                "newest_mount_id to %d",
2388                                                get_desc_mount_id(desc));
2389                         }
2390                         cur_dblock += get_desc_trans_len(desc) + 2;
2391                 } else {
2392                         cur_dblock++;
2393                 }
2394                 brelse(d_bh);
2395         }
2396
2397       start_log_replay:
2398         cur_dblock = oldest_start;
2399         if (oldest_trans_id) {
2400                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2401                                "journal-1206: Starting replay "
2402                                "from offset %llu, trans_id %lu",
2403                                cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2404                                oldest_trans_id);
2405
2406         }
2407         replay_count = 0;
2408         while (continue_replay && oldest_trans_id > 0) {
2409                 ret =
2410                     journal_read_transaction(p_s_sb, cur_dblock, oldest_start,
2411                                              oldest_trans_id, newest_mount_id);
2412                 if (ret < 0) {
2413                         return ret;
2414                 } else if (ret != 0) {
2415                         break;
2416                 }
2417                 cur_dblock =
2418                     SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + journal->j_start;
2419                 replay_count++;
2420                 if (cur_dblock == oldest_start)
2421                         break;
2422         }
2423
2424         if (oldest_trans_id == 0) {
2425                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2426                                "journal-1225: No valid " "transactions found");
2427         }
2428         /* j_start does not get set correctly if we don't replay any transactions.
2429          ** if we had a valid journal_header, set j_start to the first unflushed transaction value,
2430          ** copy the trans_id from the header
2431          */
2432         if (valid_journal_header && replay_count == 0) {
2433                 journal->j_start = le32_to_cpu(jh->j_first_unflushed_offset);
2434                 journal->j_trans_id =
2435                     le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2436                 journal->j_last_flush_trans_id =
2437                     le32_to_cpu(jh->j_last_flush_trans_id);
2438                 journal->j_mount_id = le32_to_cpu(jh->j_mount_id) + 1;
2439         } else {
2440                 journal->j_mount_id = newest_mount_id + 1;
2441         }
2442         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1299: Setting "
2443                        "newest_mount_id to %lu", journal->j_mount_id);
2444         journal->j_first_unflushed_offset = journal->j_start;
2445         if (replay_count > 0) {
2446                 reiserfs_info(p_s_sb,
2447                               "replayed %d transactions in %lu seconds\n",
2448                               replay_count, get_seconds() - start);
2449         }
2450         if (!bdev_read_only(p_s_sb->s_bdev) &&
2451             _update_journal_header_block(p_s_sb, journal->j_start,
2452                                          journal->j_last_flush_trans_id)) {
2453                 /* replay failed, caller must call free_journal_ram and abort
2454                  ** the mount
2455                  */
2456                 return -1;
2457         }
2458         return 0;
2459 }
2460
2461 static struct reiserfs_journal_list *alloc_journal_list(struct super_block *s)
2462 {
2463         struct reiserfs_journal_list *jl;
2464       retry:
2465         jl = reiserfs_kmalloc(sizeof(struct reiserfs_journal_list), GFP_NOFS,
2466                               s);
2467         if (!jl) {
2468                 yield();
2469                 goto retry;
2470         }
2471         memset(jl, 0, sizeof(*jl));
2472         INIT_LIST_HEAD(&jl->j_list);
2473         INIT_LIST_HEAD(&jl->j_working_list);
2474         INIT_LIST_HEAD(&jl->j_tail_bh_list);
2475         INIT_LIST_HEAD(&jl->j_bh_list);
2476         sema_init(&jl->j_commit_lock, 1);
2477         SB_JOURNAL(s)->j_num_lists++;
2478         get_journal_list(jl);
2479         return jl;
2480 }
2481
2482 static void journal_list_init(struct super_block *p_s_sb)
2483 {
2484         SB_JOURNAL(p_s_sb)->j_current_jl = alloc_journal_list(p_s_sb);
2485 }
2486
2487 static int release_journal_dev(struct super_block *super,
2488                                struct reiserfs_journal *journal)
2489 {
2490         int result;
2491
2492         result = 0;
2493
2494         if (journal->j_dev_file != NULL) {
2495                 result = filp_close(journal->j_dev_file, NULL);
2496                 journal->j_dev_file = NULL;
2497                 journal->j_dev_bd = NULL;
2498         } else if (journal->j_dev_bd != NULL) {
2499                 result = blkdev_put(journal->j_dev_bd);
2500                 journal->j_dev_bd = NULL;
2501         }
2502
2503         if (result != 0) {
2504                 reiserfs_warning(super,
2505                                  "sh-457: release_journal_dev: Cannot release journal device: %i",
2506                                  result);
2507         }
2508         return result;
2509 }
2510
2511 static int journal_init_dev(struct super_block *super,
2512                             struct reiserfs_journal *journal,
2513                             const char *jdev_name)
2514 {
2515         int result;
2516         dev_t jdev;
2517         int blkdev_mode = FMODE_READ | FMODE_WRITE;
2518         char b[BDEVNAME_SIZE];
2519
2520         result = 0;
2521
2522         journal->j_dev_bd = NULL;
2523         journal->j_dev_file = NULL;
2524         jdev = SB_ONDISK_JOURNAL_DEVICE(super) ?
2525             new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super)) : super->s_dev;
2526
2527         if (bdev_read_only(super->s_bdev))
2528                 blkdev_mode = FMODE_READ;
2529
2530         /* there is no "jdev" option and journal is on separate device */
2531         if ((!jdev_name || !jdev_name[0])) {
2532                 journal->j_dev_bd = open_by_devnum(jdev, blkdev_mode);
2533                 if (IS_ERR(journal->j_dev_bd)) {
2534                         result = PTR_ERR(journal->j_dev_bd);
2535                         journal->j_dev_bd = NULL;
2536                         reiserfs_warning(super, "sh-458: journal_init_dev: "
2537                                          "cannot init journal device '%s': %i",
2538                                          __bdevname(jdev, b), result);
2539                         return result;
2540                 } else if (jdev != super->s_dev)
2541                         set_blocksize(journal->j_dev_bd, super->s_blocksize);
2542                 return 0;
2543         }
2544
2545         journal->j_dev_file = filp_open(jdev_name, 0, 0);
2546         if (!IS_ERR(journal->j_dev_file)) {
2547                 struct inode *jdev_inode = journal->j_dev_file->f_mapping->host;
2548                 if (!S_ISBLK(jdev_inode->i_mode)) {
2549                         reiserfs_warning(super, "journal_init_dev: '%s' is "
2550                                          "not a block device", jdev_name);
2551                         result = -ENOTBLK;
2552                         release_journal_dev(super, journal);
2553                 } else {
2554                         /* ok */
2555                         journal->j_dev_bd = I_BDEV(jdev_inode);
2556                         set_blocksize(journal->j_dev_bd, super->s_blocksize);
2557                         reiserfs_info(super,
2558                                       "journal_init_dev: journal device: %s\n",
2559                                       bdevname(journal->j_dev_bd, b));
2560                 }
2561         } else {
2562                 result = PTR_ERR(journal->j_dev_file);
2563                 journal->j_dev_file = NULL;
2564                 reiserfs_warning(super,
2565                                  "journal_init_dev: Cannot open '%s': %i",
2566                                  jdev_name, result);
2567         }
2568         return result;
2569 }
2570
2571 /*
2572 ** must be called once on fs mount.  calls journal_read for you
2573 */
2574 int journal_init(struct super_block *p_s_sb, const char *j_dev_name,
2575                  int old_format, unsigned int commit_max_age)
2576 {
2577         int num_cnodes = SB_ONDISK_JOURNAL_SIZE(p_s_sb) * 2;
2578         struct buffer_head *bhjh;
2579         struct reiserfs_super_block *rs;
2580         struct reiserfs_journal_header *jh;
2581         struct reiserfs_journal *journal;
2582         struct reiserfs_journal_list *jl;
2583         char b[BDEVNAME_SIZE];
2584
2585         journal = SB_JOURNAL(p_s_sb) = vmalloc(sizeof(struct reiserfs_journal));
2586         if (!journal) {
2587                 reiserfs_warning(p_s_sb,
2588                                  "journal-1256: unable to get memory for journal structure");
2589                 return 1;
2590         }
2591         memset(journal, 0, sizeof(struct reiserfs_journal));
2592         INIT_LIST_HEAD(&journal->j_bitmap_nodes);
2593         INIT_LIST_HEAD(&journal->j_prealloc_list);
2594         INIT_LIST_HEAD(&journal->j_working_list);
2595         INIT_LIST_HEAD(&journal->j_journal_list);
2596         journal->j_persistent_trans = 0;
2597         if (reiserfs_allocate_list_bitmaps(p_s_sb,
2598                                            journal->j_list_bitmap,
2599                                            SB_BMAP_NR(p_s_sb)))
2600                 goto free_and_return;
2601         allocate_bitmap_nodes(p_s_sb);
2602
2603         /* reserved for journal area support */
2604         SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb) = (old_format ?
2605                                                  REISERFS_OLD_DISK_OFFSET_IN_BYTES
2606                                                  / p_s_sb->s_blocksize +
2607                                                  SB_BMAP_NR(p_s_sb) +
2608                                                  1 :
2609                                                  REISERFS_DISK_OFFSET_IN_BYTES /
2610                                                  p_s_sb->s_blocksize + 2);
2611
2612         /* Sanity check to see is the standard journal fitting withing first bitmap
2613            (actual for small blocksizes) */
2614         if (!SB_ONDISK_JOURNAL_DEVICE(p_s_sb) &&
2615             (SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb) +
2616              SB_ONDISK_JOURNAL_SIZE(p_s_sb) > p_s_sb->s_blocksize * 8)) {
2617                 reiserfs_warning(p_s_sb,
2618                                  "journal-1393: journal does not fit for area "
2619                                  "addressed by first of bitmap blocks. It starts at "
2620                                  "%u and its size is %u. Block size %ld",
2621                                  SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb),
2622                                  SB_ONDISK_JOURNAL_SIZE(p_s_sb),
2623                                  p_s_sb->s_blocksize);
2624                 goto free_and_return;
2625         }
2626
2627         if (journal_init_dev(p_s_sb, journal, j_dev_name) != 0) {
2628                 reiserfs_warning(p_s_sb,
2629                                  "sh-462: unable to initialize jornal device");
2630                 goto free_and_return;
2631         }
2632
2633         rs = SB_DISK_SUPER_BLOCK(p_s_sb);
2634
2635         /* read journal header */
2636         bhjh = journal_bread(p_s_sb,
2637                              SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2638                              SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2639         if (!bhjh) {
2640                 reiserfs_warning(p_s_sb,
2641                                  "sh-459: unable to read journal header");
2642                 goto free_and_return;
2643         }
2644         jh = (struct reiserfs_journal_header *)(bhjh->b_data);
2645
2646         /* make sure that journal matches to the super block */
2647         if (is_reiserfs_jr(rs)
2648             && (le32_to_cpu(jh->jh_journal.jp_journal_magic) !=
2649                 sb_jp_journal_magic(rs))) {
2650                 reiserfs_warning(p_s_sb,
2651                                  "sh-460: journal header magic %x "
2652                                  "(device %s) does not match to magic found in super "
2653                                  "block %x", jh->jh_journal.jp_journal_magic,
2654                                  bdevname(journal->j_dev_bd, b),
2655                                  sb_jp_journal_magic(rs));
2656                 brelse(bhjh);
2657                 goto free_and_return;
2658         }
2659
2660         journal->j_trans_max = le32_to_cpu(jh->jh_journal.jp_journal_trans_max);
2661         journal->j_max_batch = le32_to_cpu(jh->jh_journal.jp_journal_max_batch);
2662         journal->j_max_commit_age =
2663             le32_to_cpu(jh->jh_journal.jp_journal_max_commit_age);
2664         journal->j_max_trans_age = JOURNAL_MAX_TRANS_AGE;
2665
2666         if (journal->j_trans_max) {
2667                 /* make sure these parameters are available, assign it if they are not */
2668                 __u32 initial = journal->j_trans_max;
2669                 __u32 ratio = 1;
2670
2671                 if (p_s_sb->s_blocksize < 4096)
2672                         ratio = 4096 / p_s_sb->s_blocksize;
2673
2674                 if (SB_ONDISK_JOURNAL_SIZE(p_s_sb) / journal->j_trans_max <
2675                     JOURNAL_MIN_RATIO)
2676                         journal->j_trans_max =
2677                             SB_ONDISK_JOURNAL_SIZE(p_s_sb) / JOURNAL_MIN_RATIO;
2678                 if (journal->j_trans_max > JOURNAL_TRANS_MAX_DEFAULT / ratio)
2679                         journal->j_trans_max =
2680                             JOURNAL_TRANS_MAX_DEFAULT / ratio;
2681                 if (journal->j_trans_max < JOURNAL_TRANS_MIN_DEFAULT / ratio)
2682                         journal->j_trans_max =
2683                             JOURNAL_TRANS_MIN_DEFAULT / ratio;
2684
2685                 if (journal->j_trans_max != initial)
2686                         reiserfs_warning(p_s_sb,
2687                                          "sh-461: journal_init: wrong transaction max size (%u). Changed to %u",
2688                                          initial, journal->j_trans_max);
2689
2690                 journal->j_max_batch = journal->j_trans_max *
2691                     JOURNAL_MAX_BATCH_DEFAULT / JOURNAL_TRANS_MAX_DEFAULT;
2692         }
2693
2694         if (!journal->j_trans_max) {
2695                 /*we have the file system was created by old version of mkreiserfs
2696                    so this field contains zero value */
2697                 journal->j_trans_max = JOURNAL_TRANS_MAX_DEFAULT;
2698                 journal->j_max_batch = JOURNAL_MAX_BATCH_DEFAULT;
2699                 journal->j_max_commit_age = JOURNAL_MAX_COMMIT_AGE;
2700
2701                 /* for blocksize >= 4096 - max transaction size is 1024. For block size < 4096
2702                    trans max size is decreased proportionally */
2703                 if (p_s_sb->s_blocksize < 4096) {
2704                         journal->j_trans_max /= (4096 / p_s_sb->s_blocksize);
2705                         journal->j_max_batch = (journal->j_trans_max) * 9 / 10;
2706                 }
2707         }
2708
2709         journal->j_default_max_commit_age = journal->j_max_commit_age;
2710
2711         if (commit_max_age != 0) {
2712                 journal->j_max_commit_age = commit_max_age;
2713                 journal->j_max_trans_age = commit_max_age;
2714         }
2715
2716         reiserfs_info(p_s_sb, "journal params: device %s, size %u, "
2717                       "journal first block %u, max trans len %u, max batch %u, "
2718                       "max commit age %u, max trans age %u\n",
2719                       bdevname(journal->j_dev_bd, b),
2720                       SB_ONDISK_JOURNAL_SIZE(p_s_sb),
2721                       SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2722                       journal->j_trans_max,
2723                       journal->j_max_batch,
2724                       journal->j_max_commit_age, journal->j_max_trans_age);
2725
2726         brelse(bhjh);
2727
2728         journal->j_list_bitmap_index = 0;
2729         journal_list_init(p_s_sb);
2730
2731         memset(journal->j_list_hash_table, 0,
2732                JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
2733
2734         INIT_LIST_HEAD(&journal->j_dirty_buffers);
2735         spin_lock_init(&journal->j_dirty_buffers_lock);
2736
2737         journal->j_start = 0;
2738         journal->j_len = 0;
2739         journal->j_len_alloc = 0;
2740         atomic_set(&(journal->j_wcount), 0);
2741         atomic_set(&(journal->j_async_throttle), 0);
2742         journal->j_bcount = 0;
2743         journal->j_trans_start_time = 0;
2744         journal->j_last = NULL;
2745         journal->j_first = NULL;
2746         init_waitqueue_head(&(journal->j_join_wait));
2747         sema_init(&journal->j_lock, 1);
2748         sema_init(&journal->j_flush_sem, 1);
2749
2750         journal->j_trans_id = 10;
2751         journal->j_mount_id = 10;
2752         journal->j_state = 0;
2753         atomic_set(&(journal->j_jlock), 0);
2754         journal->j_cnode_free_list = allocate_cnodes(num_cnodes);
2755         journal->j_cnode_free_orig = journal->j_cnode_free_list;
2756         journal->j_cnode_free = journal->j_cnode_free_list ? num_cnodes : 0;
2757         journal->j_cnode_used = 0;
2758         journal->j_must_wait = 0;
2759
2760         if (journal->j_cnode_free == 0) {
2761                 reiserfs_warning(p_s_sb, "journal-2004: Journal cnode memory "
2762                                  "allocation failed (%ld bytes). Journal is "
2763                                  "too large for available memory. Usually "
2764                                  "this is due to a journal that is too large.",
2765                                  sizeof (struct reiserfs_journal_cnode) * num_cnodes);
2766                 goto free_and_return;
2767         }
2768
2769         init_journal_hash(p_s_sb);
2770         jl = journal->j_current_jl;
2771         jl->j_list_bitmap = get_list_bitmap(p_s_sb, jl);
2772         if (!jl->j_list_bitmap) {
2773                 reiserfs_warning(p_s_sb,
2774                                  "journal-2005, get_list_bitmap failed for journal list 0");
2775                 goto free_and_return;
2776         }
2777         if (journal_read(p_s_sb) < 0) {
2778                 reiserfs_warning(p_s_sb, "Replay Failure, unable to mount");
2779                 goto free_and_return;
2780         }
2781
2782         reiserfs_mounted_fs_count++;
2783         if (reiserfs_mounted_fs_count <= 1)
2784                 commit_wq = create_workqueue("reiserfs");
2785
2786         INIT_WORK(&journal->j_work, flush_async_commits, p_s_sb);
2787         return 0;
2788       free_and_return:
2789         free_journal_ram(p_s_sb);
2790         return 1;
2791 }
2792
2793 /*
2794 ** test for a polite end of the current transaction.  Used by file_write, and should
2795 ** be used by delete to make sure they don't write more than can fit inside a single
2796 ** transaction
2797 */
2798 int journal_transaction_should_end(struct reiserfs_transaction_handle *th,
2799                                    int new_alloc)
2800 {
2801         struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2802         time_t now = get_seconds();
2803         /* cannot restart while nested */
2804         BUG_ON(!th->t_trans_id);
2805         if (th->t_refcount > 1)
2806                 return 0;
2807         if (journal->j_must_wait > 0 ||
2808             (journal->j_len_alloc + new_alloc) >= journal->j_max_batch ||
2809             atomic_read(&(journal->j_jlock)) ||
2810             (now - journal->j_trans_start_time) > journal->j_max_trans_age ||
2811             journal->j_cnode_free < (journal->j_trans_max * 3)) {
2812                 return 1;
2813         }
2814         return 0;
2815 }
2816
2817 /* this must be called inside a transaction, and requires the
2818 ** kernel_lock to be held
2819 */
2820 void reiserfs_block_writes(struct reiserfs_transaction_handle *th)
2821 {
2822         struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2823         BUG_ON(!th->t_trans_id);
2824         journal->j_must_wait = 1;
2825         set_bit(J_WRITERS_BLOCKED, &journal->j_state);
2826         return;
2827 }
2828
2829 /* this must be called without a transaction started, and does not
2830 ** require BKL
2831 */
2832 void reiserfs_allow_writes(struct super_block *s)
2833 {
2834         struct reiserfs_journal *journal = SB_JOURNAL(s);
2835         clear_bit(J_WRITERS_BLOCKED, &journal->j_state);
2836         wake_up(&journal->j_join_wait);
2837 }
2838
2839 /* this must be called without a transaction started, and does not
2840 ** require BKL
2841 */
2842 void reiserfs_wait_on_write_block(struct super_block *s)
2843 {
2844         struct reiserfs_journal *journal = SB_JOURNAL(s);
2845         wait_event(journal->j_join_wait,
2846                    !test_bit(J_WRITERS_BLOCKED, &journal->j_state));
2847 }
2848
2849 static void queue_log_writer(struct super_block *s)
2850 {
2851         wait_queue_t wait;
2852         struct reiserfs_journal *journal = SB_JOURNAL(s);
2853         set_bit(J_WRITERS_QUEUED, &journal->j_state);
2854
2855         /*
2856          * we don't want to use wait_event here because
2857          * we only want to wait once.
2858          */
2859         init_waitqueue_entry(&wait, current);
2860         add_wait_queue(&journal->j_join_wait, &wait);
2861         set_current_state(TASK_UNINTERRUPTIBLE);
2862         if (test_bit(J_WRITERS_QUEUED, &journal->j_state))
2863                 schedule();
2864         current->state = TASK_RUNNING;
2865         remove_wait_queue(&journal->j_join_wait, &wait);
2866 }
2867
2868 static void wake_queued_writers(struct super_block *s)
2869 {
2870         struct reiserfs_journal *journal = SB_JOURNAL(s);
2871         if (test_and_clear_bit(J_WRITERS_QUEUED, &journal->j_state))
2872                 wake_up(&journal->j_join_wait);
2873 }
2874
2875 static void let_transaction_grow(struct super_block *sb, unsigned long trans_id)
2876 {
2877         struct reiserfs_journal *journal = SB_JOURNAL(sb);
2878         unsigned long bcount = journal->j_bcount;
2879         while (1) {
2880                 schedule_timeout_uninterruptible(1);
2881                 journal->j_current_jl->j_state |= LIST_COMMIT_PENDING;
2882                 while ((atomic_read(&journal->j_wcount) > 0 ||
2883                         atomic_read(&journal->j_jlock)) &&
2884                        journal->j_trans_id == trans_id) {
2885                         queue_log_writer(sb);
2886                 }
2887                 if (journal->j_trans_id != trans_id)
2888                         break;
2889                 if (bcount == journal->j_bcount)
2890                         break;
2891                 bcount = journal->j_bcount;
2892         }
2893 }
2894
2895 /* join == true if you must join an existing transaction.
2896 ** join == false if you can deal with waiting for others to finish
2897 **
2898 ** this will block until the transaction is joinable.  send the number of blocks you
2899 ** expect to use in nblocks.
2900 */
2901 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
2902                               struct super_block *p_s_sb, unsigned long nblocks,
2903                               int join)
2904 {
2905         time_t now = get_seconds();
2906         int old_trans_id;
2907         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
2908         struct reiserfs_transaction_handle myth;
2909         int sched_count = 0;
2910         int retval;
2911
2912         reiserfs_check_lock_depth(p_s_sb, "journal_begin");
2913         if (nblocks > journal->j_trans_max)
2914                 BUG();
2915
2916         PROC_INFO_INC(p_s_sb, journal.journal_being);
2917         /* set here for journal_join */
2918         th->t_refcount = 1;
2919         th->t_super = p_s_sb;
2920
2921       relock:
2922         lock_journal(p_s_sb);
2923         if (join != JBEGIN_ABORT && reiserfs_is_journal_aborted(journal)) {
2924                 unlock_journal(p_s_sb);
2925                 retval = journal->j_errno;
2926                 goto out_fail;
2927         }
2928         journal->j_bcount++;
2929
2930         if (test_bit(J_WRITERS_BLOCKED, &journal->j_state)) {
2931                 unlock_journal(p_s_sb);
2932                 reiserfs_wait_on_write_block(p_s_sb);
2933                 PROC_INFO_INC(p_s_sb, journal.journal_relock_writers);
2934                 goto relock;
2935         }
2936         now = get_seconds();
2937
2938         /* if there is no room in the journal OR
2939          ** if this transaction is too old, and we weren't called joinable, wait for it to finish before beginning
2940          ** we don't sleep if there aren't other writers
2941          */
2942
2943         if ((!join && journal->j_must_wait > 0) ||
2944             (!join
2945              && (journal->j_len_alloc + nblocks + 2) >= journal->j_max_batch)
2946             || (!join && atomic_read(&journal->j_wcount) > 0
2947                 && journal->j_trans_start_time > 0
2948                 && (now - journal->j_trans_start_time) >
2949                 journal->j_max_trans_age) || (!join
2950                                               && atomic_read(&journal->j_jlock))
2951             || (!join && journal->j_cnode_free < (journal->j_trans_max * 3))) {
2952
2953                 old_trans_id = journal->j_trans_id;
2954                 unlock_journal(p_s_sb); /* allow others to finish this transaction */
2955
2956                 if (!join && (journal->j_len_alloc + nblocks + 2) >=
2957                     journal->j_max_batch &&
2958                     ((journal->j_len + nblocks + 2) * 100) <
2959                     (journal->j_len_alloc * 75)) {
2960                         if (atomic_read(&journal->j_wcount) > 10) {
2961                                 sched_count++;
2962                                 queue_log_writer(p_s_sb);
2963                                 goto relock;
2964                         }
2965                 }
2966                 /* don't mess with joining the transaction if all we have to do is
2967                  * wait for someone else to do a commit
2968                  */
2969                 if (atomic_read(&journal->j_jlock)) {
2970                         while (journal->j_trans_id == old_trans_id &&
2971                                atomic_read(&journal->j_jlock)) {
2972                                 queue_log_writer(p_s_sb);
2973                         }
2974                         goto relock;
2975                 }
2976                 retval = journal_join(&myth, p_s_sb, 1);
2977                 if (retval)
2978                         goto out_fail;
2979
2980                 /* someone might have ended the transaction while we joined */
2981                 if (old_trans_id != journal->j_trans_id) {
2982                         retval = do_journal_end(&myth, p_s_sb, 1, 0);
2983                 } else {
2984                         retval = do_journal_end(&myth, p_s_sb, 1, COMMIT_NOW);
2985                 }
2986
2987                 if (retval)
2988                         goto out_fail;
2989
2990                 PROC_INFO_INC(p_s_sb, journal.journal_relock_wcount);
2991                 goto relock;
2992         }
2993         /* we are the first writer, set trans_id */
2994         if (journal->j_trans_start_time == 0) {
2995                 journal->j_trans_start_time = get_seconds();
2996         }
2997         atomic_inc(&(journal->j_wcount));
2998         journal->j_len_alloc += nblocks;
2999         th->t_blocks_logged = 0;
3000         th->t_blocks_allocated = nblocks;
3001         th->t_trans_id = journal->j_trans_id;
3002         unlock_journal(p_s_sb);
3003         INIT_LIST_HEAD(&th->t_list);
3004         get_fs_excl();
3005         return 0;
3006
3007       out_fail:
3008         memset(th, 0, sizeof(*th));
3009         /* Re-set th->t_super, so we can properly keep track of how many
3010          * persistent transactions there are. We need to do this so if this
3011          * call is part of a failed restart_transaction, we can free it later */
3012         th->t_super = p_s_sb;
3013         return retval;
3014 }
3015
3016 struct reiserfs_transaction_handle *reiserfs_persistent_transaction(struct
3017                                                                     super_block
3018                                                                     *s,
3019                                                                     int nblocks)
3020 {
3021         int ret;
3022         struct reiserfs_transaction_handle *th;
3023
3024         /* if we're nesting into an existing transaction.  It will be
3025          ** persistent on its own
3026          */
3027         if (reiserfs_transaction_running(s)) {
3028                 th = current->journal_info;
3029                 th->t_refcount++;
3030                 if (th->t_refcount < 2) {
3031                         BUG();
3032                 }
3033                 return th;
3034         }
3035         th = reiserfs_kmalloc(sizeof(struct reiserfs_transaction_handle),
3036                               GFP_NOFS, s);
3037         if (!th)
3038                 return NULL;
3039         ret = journal_begin(th, s, nblocks);
3040         if (ret) {
3041                 reiserfs_kfree(th, sizeof(struct reiserfs_transaction_handle),
3042                                s);
3043                 return NULL;
3044         }
3045
3046         SB_JOURNAL(s)->j_persistent_trans++;
3047         return th;
3048 }
3049
3050 int reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle *th)
3051 {
3052         struct super_block *s = th->t_super;
3053         int ret = 0;
3054         if (th->t_trans_id)
3055                 ret = journal_end(th, th->t_super, th->t_blocks_allocated);
3056         else
3057                 ret = -EIO;
3058         if (th->t_refcount == 0) {
3059                 SB_JOURNAL(s)->j_persistent_trans--;
3060                 reiserfs_kfree(th, sizeof(struct reiserfs_transaction_handle),
3061                                s);
3062         }
3063         return ret;
3064 }
3065
3066 static int journal_join(struct reiserfs_transaction_handle *th,
3067                         struct super_block *p_s_sb, unsigned long nblocks)
3068 {
3069         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3070
3071         /* this keeps do_journal_end from NULLing out the current->journal_info
3072          ** pointer
3073          */
3074         th->t_handle_save = cur_th;
3075         if (cur_th && cur_th->t_refcount > 1) {
3076                 BUG();
3077         }
3078         return do_journal_begin_r(th, p_s_sb, nblocks, JBEGIN_JOIN);
3079 }
3080
3081 int journal_join_abort(struct reiserfs_transaction_handle *th,
3082                        struct super_block *p_s_sb, unsigned long nblocks)
3083 {
3084         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3085
3086         /* this keeps do_journal_end from NULLing out the current->journal_info
3087          ** pointer
3088          */
3089         th->t_handle_save = cur_th;
3090         if (cur_th && cur_th->t_refcount > 1) {
3091                 BUG();
3092         }
3093         return do_journal_begin_r(th, p_s_sb, nblocks, JBEGIN_ABORT);
3094 }
3095
3096 int journal_begin(struct reiserfs_transaction_handle *th,
3097                   struct super_block *p_s_sb, unsigned long nblocks)
3098 {
3099         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3100         int ret;
3101
3102         th->t_handle_save = NULL;
3103         if (cur_th) {
3104                 /* we are nesting into the current transaction */
3105                 if (cur_th->t_super == p_s_sb) {
3106                         BUG_ON(!cur_th->t_refcount);
3107                         cur_th->t_refcount++;
3108                         memcpy(th, cur_th, sizeof(*th));
3109                         if (th->t_refcount <= 1)
3110                                 reiserfs_warning(p_s_sb,
3111                                                  "BAD: refcount <= 1, but journal_info != 0");
3112                         return 0;
3113                 } else {
3114                         /* we've ended up with a handle from a different filesystem.
3115                          ** save it and restore on journal_end.  This should never
3116                          ** really happen...
3117                          */
3118                         reiserfs_warning(p_s_sb,
3119                                          "clm-2100: nesting info a different FS");
3120                         th->t_handle_save = current->journal_info;
3121                         current->journal_info = th;
3122                 }
3123         } else {
3124                 current->journal_info = th;
3125         }
3126         ret = do_journal_begin_r(th, p_s_sb, nblocks, JBEGIN_REG);
3127         if (current->journal_info != th)
3128                 BUG();
3129
3130         /* I guess this boils down to being the reciprocal of clm-2100 above.
3131          * If do_journal_begin_r fails, we need to put it back, since journal_end
3132          * won't be called to do it. */
3133         if (ret)
3134                 current->journal_info = th->t_handle_save;
3135         else
3136                 BUG_ON(!th->t_refcount);
3137
3138         return ret;
3139 }
3140
3141 /*
3142 ** puts bh into the current transaction.  If it was already there, reorders removes the
3143 ** old pointers from the hash, and puts new ones in (to make sure replay happen in the right order).
3144 **
3145 ** if it was dirty, cleans and files onto the clean list.  I can't let it be dirty again until the
3146 ** transaction is committed.
3147 **
3148 ** if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len.
3149 */
3150 int journal_mark_dirty(struct reiserfs_transaction_handle *th,
3151                        struct super_block *p_s_sb, struct buffer_head *bh)
3152 {
3153         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3154         struct reiserfs_journal_cnode *cn = NULL;
3155         int count_already_incd = 0;
3156         int prepared = 0;
3157         BUG_ON(!th->t_trans_id);
3158
3159         PROC_INFO_INC(p_s_sb, journal.mark_dirty);
3160         if (th->t_trans_id != journal->j_trans_id) {
3161                 reiserfs_panic(th->t_super,
3162                                "journal-1577: handle trans id %ld != current trans id %ld\n",
3163                                th->t_trans_id, journal->j_trans_id);
3164         }
3165
3166         p_s_sb->s_dirt = 1;
3167
3168         prepared = test_clear_buffer_journal_prepared(bh);
3169         clear_buffer_journal_restore_dirty(bh);
3170         /* already in this transaction, we are done */
3171         if (buffer_journaled(bh)) {
3172                 PROC_INFO_INC(p_s_sb, journal.mark_dirty_already);
3173                 return 0;
3174         }
3175
3176         /* this must be turned into a panic instead of a warning.  We can't allow
3177          ** a dirty or journal_dirty or locked buffer to be logged, as some changes
3178          ** could get to disk too early.  NOT GOOD.
3179          */
3180         if (!prepared || buffer_dirty(bh)) {
3181                 reiserfs_warning(p_s_sb, "journal-1777: buffer %llu bad state "
3182                                  "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT",
3183                                  (unsigned long long)bh->b_blocknr,
3184                                  prepared ? ' ' : '!',
3185                                  buffer_locked(bh) ? ' ' : '!',
3186                                  buffer_dirty(bh) ? ' ' : '!',
3187                                  buffer_journal_dirty(bh) ? ' ' : '!');
3188         }
3189
3190         if (atomic_read(&(journal->j_wcount)) <= 0) {
3191                 reiserfs_warning(p_s_sb,
3192                                  "journal-1409: journal_mark_dirty returning because j_wcount was %d",
3193                                  atomic_read(&(journal->j_wcount)));
3194                 return 1;
3195         }
3196         /* this error means I've screwed up, and we've overflowed the transaction.
3197          ** Nothing can be done here, except make the FS readonly or panic.
3198          */
3199         if (journal->j_len >= journal->j_trans_max) {
3200                 reiserfs_panic(th->t_super,
3201                                "journal-1413: journal_mark_dirty: j_len (%lu) is too big\n",
3202                                journal->j_len);
3203         }
3204
3205         if (buffer_journal_dirty(bh)) {
3206                 count_already_incd = 1;
3207                 PROC_INFO_INC(p_s_sb, journal.mark_dirty_notjournal);
3208                 clear_buffer_journal_dirty(bh);
3209         }
3210
3211         if (journal->j_len > journal->j_len_alloc) {
3212                 journal->j_len_alloc = journal->j_len + JOURNAL_PER_BALANCE_CNT;
3213         }
3214
3215         set_buffer_journaled(bh);
3216
3217         /* now put this guy on the end */
3218         if (!cn) {
3219                 cn = get_cnode(p_s_sb);
3220                 if (!cn) {
3221                         reiserfs_panic(p_s_sb, "get_cnode failed!\n");
3222                 }
3223
3224                 if (th->t_blocks_logged == th->t_blocks_allocated) {
3225                         th->t_blocks_allocated += JOURNAL_PER_BALANCE_CNT;
3226                         journal->j_len_alloc += JOURNAL_PER_BALANCE_CNT;
3227                 }
3228                 th->t_blocks_logged++;
3229                 journal->j_len++;
3230
3231                 cn->bh = bh;
3232                 cn->blocknr = bh->b_blocknr;
3233                 cn->sb = p_s_sb;
3234                 cn->jlist = NULL;
3235                 insert_journal_hash(journal->j_hash_table, cn);
3236                 if (!count_already_incd) {
3237                         get_bh(bh);
3238                 }
3239         }
3240         cn->next = NULL;
3241         cn->prev = journal->j_last;
3242         cn->bh = bh;
3243         if (journal->j_last) {
3244                 journal->j_last->next = cn;
3245                 journal->j_last = cn;
3246         } else {
3247                 journal->j_first = cn;
3248                 journal->j_last = cn;
3249         }
3250         return 0;
3251 }
3252
3253 int journal_end(struct reiserfs_transaction_handle *th,
3254                 struct super_block *p_s_sb, unsigned long nblocks)
3255 {
3256         if (!current->journal_info && th->t_refcount > 1)
3257                 reiserfs_warning(p_s_sb, "REISER-NESTING: th NULL, refcount %d",
3258                                  th->t_refcount);
3259
3260         if (!th->t_trans_id) {
3261                 WARN_ON(1);
3262                 return -EIO;
3263         }
3264
3265         th->t_refcount--;
3266         if (th->t_refcount > 0) {
3267                 struct reiserfs_transaction_handle *cur_th =
3268                     current->journal_info;
3269
3270                 /* we aren't allowed to close a nested transaction on a different
3271                  ** filesystem from the one in the task struct
3272                  */
3273                 if (cur_th->t_super != th->t_super)
3274                         BUG();
3275
3276                 if (th != cur_th) {
3277                         memcpy(current->journal_info, th, sizeof(*th));
3278                         th->t_trans_id = 0;
3279                 }
3280                 return 0;
3281         } else {
3282                 return do_journal_end(th, p_s_sb, nblocks, 0);
3283         }
3284 }
3285
3286 /* removes from the current transaction, relsing and descrementing any counters.
3287 ** also files the removed buffer directly onto the clean list
3288 **
3289 ** called by journal_mark_freed when a block has been deleted
3290 **
3291 ** returns 1 if it cleaned and relsed the buffer. 0 otherwise
3292 */
3293 static int remove_from_transaction(struct super_block *p_s_sb,
3294                                    b_blocknr_t blocknr, int already_cleaned)
3295 {
3296         struct buffer_head *bh;
3297         struct reiserfs_journal_cnode *cn;
3298         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3299         int ret = 0;
3300
3301         cn = get_journal_hash_dev(p_s_sb, journal->j_hash_table, blocknr);
3302         if (!cn || !cn->bh) {
3303                 return ret;
3304         }
3305         bh = cn->bh;
3306         if (cn->prev) {
3307                 cn->prev->next = cn->next;
3308         }
3309         if (cn->next) {
3310                 cn->next->prev = cn->prev;
3311         }
3312         if (cn == journal->j_first) {
3313                 journal->j_first = cn->next;
3314         }
3315         if (cn == journal->j_last) {
3316                 journal->j_last = cn->prev;
3317         }
3318         if (bh)
3319                 remove_journal_hash(p_s_sb, journal->j_hash_table, NULL,
3320                                     bh->b_blocknr, 0);
3321         clear_buffer_journaled(bh);     /* don't log this one */
3322
3323         if (!already_cleaned) {
3324                 clear_buffer_journal_dirty(bh);
3325                 clear_buffer_dirty(bh);
3326                 clear_buffer_journal_test(bh);
3327                 put_bh(bh);
3328                 if (atomic_read(&(bh->b_count)) < 0) {
3329                         reiserfs_warning(p_s_sb,
3330                                          "journal-1752: remove from trans, b_count < 0");
3331                 }
3332                 ret = 1;
3333         }
3334         journal->j_len--;
3335         journal->j_len_alloc--;
3336         free_cnode(p_s_sb, cn);
3337         return ret;
3338 }
3339
3340 /*
3341 ** for any cnode in a journal list, it can only be dirtied of all the
3342 ** transactions that include it are commited to disk.
3343 ** this checks through each transaction, and returns 1 if you are allowed to dirty,
3344 ** and 0 if you aren't
3345 **
3346 ** it is called by dirty_journal_list, which is called after flush_commit_list has gotten all the log
3347 ** blocks for a given transaction on disk
3348 **
3349 */
3350 static int can_dirty(struct reiserfs_journal_cnode *cn)
3351 {
3352         struct super_block *sb = cn->sb;
3353         b_blocknr_t blocknr = cn->blocknr;
3354         struct reiserfs_journal_cnode *cur = cn->hprev;
3355         int can_dirty = 1;
3356
3357         /* first test hprev.  These are all newer than cn, so any node here
3358          ** with the same block number and dev means this node can't be sent
3359          ** to disk right now.
3360          */
3361         while (cur && can_dirty) {
3362                 if (cur->jlist && cur->bh && cur->blocknr && cur->sb == sb &&
3363                     cur->blocknr == blocknr) {
3364                         can_dirty = 0;
3365                 }
3366                 cur = cur->hprev;
3367         }
3368         /* then test hnext.  These are all older than cn.  As long as they
3369          ** are committed to the log, it is safe to write cn to disk
3370          */
3371         cur = cn->hnext;
3372         while (cur && can_dirty) {
3373                 if (cur->jlist && cur->jlist->j_len > 0 &&
3374                     atomic_read(&(cur->jlist->j_commit_left)) > 0 && cur->bh &&
3375                     cur->blocknr && cur->sb == sb && cur->blocknr == blocknr) {
3376                         can_dirty = 0;
3377                 }
3378                 cur = cur->hnext;
3379         }
3380         return can_dirty;
3381 }
3382
3383 /* syncs the commit blocks, but does not force the real buffers to disk
3384 ** will wait until the current transaction is done/commited before returning
3385 */
3386 int journal_end_sync(struct reiserfs_transaction_handle *th,
3387                      struct super_block *p_s_sb, unsigned long nblocks)
3388 {
3389         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3390
3391         BUG_ON(!th->t_trans_id);
3392         /* you can sync while nested, very, very bad */
3393         if (th->t_refcount > 1) {
3394                 BUG();
3395         }
3396         if (journal->j_len == 0) {
3397                 reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb),
3398                                              1);
3399                 journal_mark_dirty(th, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb));
3400         }
3401         return do_journal_end(th, p_s_sb, nblocks, COMMIT_NOW | WAIT);
3402 }
3403
3404 /*
3405 ** writeback the pending async commits to disk
3406 */
3407 static void flush_async_commits(void *p)
3408 {
3409         struct super_block *p_s_sb = p;
3410         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3411         struct reiserfs_journal_list *jl;
3412         struct list_head *entry;
3413
3414         lock_kernel();
3415         if (!list_empty(&journal->j_journal_list)) {
3416                 /* last entry is the youngest, commit it and you get everything */
3417                 entry = journal->j_journal_list.prev;
3418                 jl = JOURNAL_LIST_ENTRY(entry);
3419                 flush_commit_list(p_s_sb, jl, 1);
3420         }
3421         unlock_kernel();
3422         /*
3423          * this is a little racey, but there's no harm in missing
3424          * the filemap_fdata_write
3425          */
3426         if (!atomic_read(&journal->j_async_throttle)
3427             && !reiserfs_is_journal_aborted(journal)) {
3428                 atomic_inc(&journal->j_async_throttle);
3429                 filemap_fdatawrite(p_s_sb->s_bdev->bd_inode->i_mapping);
3430                 atomic_dec(&journal->j_async_throttle);
3431         }
3432 }
3433
3434 /*
3435 ** flushes any old transactions to disk
3436 ** ends the current transaction if it is too old
3437 */
3438 int reiserfs_flush_old_commits(struct super_block *p_s_sb)
3439 {
3440         time_t now;
3441         struct reiserfs_transaction_handle th;
3442         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3443
3444         now = get_seconds();
3445         /* safety check so we don't flush while we are replaying the log during
3446          * mount
3447          */
3448         if (list_empty(&journal->j_journal_list)) {
3449                 return 0;
3450         }
3451
3452         /* check the current transaction.  If there are no writers, and it is
3453          * too old, finish it, and force the commit blocks to disk
3454          */
3455         if (atomic_read(&journal->j_wcount) <= 0 &&
3456             journal->j_trans_start_time > 0 &&
3457             journal->j_len > 0 &&
3458             (now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3459                 if (!journal_join(&th, p_s_sb, 1)) {
3460                         reiserfs_prepare_for_journal(p_s_sb,
3461                                                      SB_BUFFER_WITH_SB(p_s_sb),
3462                                                      1);
3463                         journal_mark_dirty(&th, p_s_sb,
3464                                            SB_BUFFER_WITH_SB(p_s_sb));
3465
3466                         /* we're only being called from kreiserfsd, it makes no sense to do
3467                          ** an async commit so that kreiserfsd can do it later
3468                          */
3469                         do_journal_end(&th, p_s_sb, 1, COMMIT_NOW | WAIT);
3470                 }
3471         }
3472         return p_s_sb->s_dirt;
3473 }
3474
3475 /*
3476 ** returns 0 if do_journal_end should return right away, returns 1 if do_journal_end should finish the commit
3477 **
3478 ** if the current transaction is too old, but still has writers, this will wait on j_join_wait until all
3479 ** the writers are done.  By the time it wakes up, the transaction it was called has already ended, so it just
3480 ** flushes the commit list and returns 0.
3481 **
3482 ** Won't batch when flush or commit_now is set.  Also won't batch when others are waiting on j_join_wait.
3483 **
3484 ** Note, we can't allow the journal_end to proceed while there are still writers in the log.
3485 */
3486 static int check_journal_end(struct reiserfs_transaction_handle *th,
3487                              struct super_block *p_s_sb, unsigned long nblocks,
3488                              int flags)
3489 {
3490
3491         time_t now;
3492         int flush = flags & FLUSH_ALL;
3493         int commit_now = flags & COMMIT_NOW;
3494         int wait_on_commit = flags & WAIT;
3495         struct reiserfs_journal_list *jl;
3496         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3497
3498         BUG_ON(!th->t_trans_id);
3499
3500         if (th->t_trans_id != journal->j_trans_id) {
3501                 reiserfs_panic(th->t_super,
3502                                "journal-1577: handle trans id %ld != current trans id %ld\n",
3503                                th->t_trans_id, journal->j_trans_id);
3504         }
3505
3506         journal->j_len_alloc -= (th->t_blocks_allocated - th->t_blocks_logged);
3507         if (atomic_read(&(journal->j_wcount)) > 0) {    /* <= 0 is allowed.  unmounting might not call begin */
3508                 atomic_dec(&(journal->j_wcount));
3509         }
3510
3511         /* BUG, deal with case where j_len is 0, but people previously freed blocks need to be released
3512          ** will be dealt with by next transaction that actually writes something, but should be taken
3513          ** care of in this trans
3514          */
3515         if (journal->j_len == 0) {
3516                 BUG();
3517         }
3518         /* if wcount > 0, and we are called to with flush or commit_now,
3519          ** we wait on j_join_wait.  We will wake up when the last writer has
3520          ** finished the transaction, and started it on its way to the disk.
3521          ** Then, we flush the commit or journal list, and just return 0
3522          ** because the rest of journal end was already done for this transaction.
3523          */
3524         if (atomic_read(&(journal->j_wcount)) > 0) {
3525                 if (flush || commit_now) {
3526                         unsigned trans_id;
3527
3528                         jl = journal->j_current_jl;
3529                         trans_id = jl->j_trans_id;
3530                         if (wait_on_commit)
3531                                 jl->j_state |= LIST_COMMIT_PENDING;
3532                         atomic_set(&(journal->j_jlock), 1);
3533                         if (flush) {
3534                                 journal->j_next_full_flush = 1;
3535                         }
3536                         unlock_journal(p_s_sb);
3537
3538                         /* sleep while the current transaction is still j_jlocked */
3539                         while (journal->j_trans_id == trans_id) {
3540                                 if (atomic_read(&journal->j_jlock)) {
3541                                         queue_log_writer(p_s_sb);
3542                                 } else {
3543                                         lock_journal(p_s_sb);
3544                                         if (journal->j_trans_id == trans_id) {
3545                                                 atomic_set(&(journal->j_jlock),
3546                                                            1);
3547                                         }
3548                                         unlock_journal(p_s_sb);
3549                                 }
3550                         }
3551                         if (journal->j_trans_id == trans_id) {
3552                                 BUG();
3553                         }
3554                         if (commit_now
3555                             && journal_list_still_alive(p_s_sb, trans_id)
3556                             && wait_on_commit) {
3557                                 flush_commit_list(p_s_sb, jl, 1);
3558                         }
3559                         return 0;
3560                 }
3561                 unlock_journal(p_s_sb);
3562                 return 0;
3563         }
3564
3565         /* deal with old transactions where we are the last writers */
3566         now = get_seconds();
3567         if ((now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3568                 commit_now = 1;
3569                 journal->j_next_async_flush = 1;
3570         }
3571         /* don't batch when someone is waiting on j_join_wait */
3572         /* don't batch when syncing the commit or flushing the whole trans */
3573         if (!(journal->j_must_wait > 0) && !(atomic_read(&(journal->j_jlock)))
3574             && !flush && !commit_now && (journal->j_len < journal->j_max_batch)
3575             && journal->j_len_alloc < journal->j_max_batch
3576             && journal->j_cnode_free > (journal->j_trans_max * 3)) {
3577                 journal->j_bcount++;
3578                 unlock_journal(p_s_sb);
3579                 return 0;
3580         }
3581
3582         if (journal->j_start > SB_ONDISK_JOURNAL_SIZE(p_s_sb)) {
3583                 reiserfs_panic(p_s_sb,
3584                                "journal-003: journal_end: j_start (%ld) is too high\n",
3585                                journal->j_start);
3586         }
3587         return 1;
3588 }
3589
3590 /*
3591 ** Does all the work that makes deleting blocks safe.
3592 ** when deleting a block mark BH_JNew, just remove it from the current transaction, clean it's buffer_head and move on.
3593 **
3594 ** otherwise:
3595 ** set a bit for the block in the journal bitmap.  That will prevent it from being allocated for unformatted nodes
3596 ** before this transaction has finished.
3597 **
3598 ** mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers.  That will prevent any old transactions with
3599 ** this block from trying to flush to the real location.  Since we aren't removing the cnode from the journal_list_hash,
3600 ** the block can't be reallocated yet.
3601 **
3602 ** Then remove it from the current transaction, decrementing any counters and filing it on the clean list.
3603 */
3604 int journal_mark_freed(struct reiserfs_transaction_handle *th,
3605                        struct super_block *p_s_sb, b_blocknr_t blocknr)
3606 {
3607         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3608         struct reiserfs_journal_cnode *cn = NULL;
3609         struct buffer_head *bh = NULL;
3610         struct reiserfs_list_bitmap *jb = NULL;
3611         int cleaned = 0;
3612         BUG_ON(!th->t_trans_id);
3613
3614         cn = get_journal_hash_dev(p_s_sb, journal->j_hash_table, blocknr);
3615         if (cn && cn->bh) {
3616                 bh = cn->bh;
3617                 get_bh(bh);
3618         }
3619         /* if it is journal new, we just remove it from this transaction */
3620         if (bh && buffer_journal_new(bh)) {
3621                 clear_buffer_journal_new(bh);
3622                 clear_prepared_bits(bh);
3623                 reiserfs_clean_and_file_buffer(bh);
3624                 cleaned = remove_from_transaction(p_s_sb, blocknr, cleaned);
3625         } else {
3626                 /* set the bit for this block in the journal bitmap for this transaction */
3627                 jb = journal->j_current_jl->j_list_bitmap;
3628                 if (!jb) {
3629                         reiserfs_panic(p_s_sb,
3630                                        "journal-1702: journal_mark_freed, journal_list_bitmap is NULL\n");
3631                 }
3632                 set_bit_in_list_bitmap(p_s_sb, blocknr, jb);
3633
3634                 /* Note, the entire while loop is not allowed to schedule.  */
3635
3636                 if (bh) {
3637                         clear_prepared_bits(bh);
3638                         reiserfs_clean_and_file_buffer(bh);
3639                 }
3640                 cleaned = remove_from_transaction(p_s_sb, blocknr, cleaned);
3641
3642                 /* find all older transactions with this block, make sure they don't try to write it out */
3643                 cn = get_journal_hash_dev(p_s_sb, journal->j_list_hash_table,
3644                                           blocknr);
3645                 while (cn) {
3646                         if (p_s_sb == cn->sb && blocknr == cn->blocknr) {
3647                                 set_bit(BLOCK_FREED, &cn->state);
3648                                 if (cn->bh) {
3649                                         if (!cleaned) {
3650                                                 /* remove_from_transaction will brelse the buffer if it was
3651                                                  ** in the current trans
3652                                                  */
3653                                                 clear_buffer_journal_dirty(cn->
3654                                                                            bh);
3655                                                 clear_buffer_dirty(cn->bh);
3656                                                 clear_buffer_journal_test(cn->
3657                                                                           bh);
3658                                                 cleaned = 1;
3659                                                 put_bh(cn->bh);
3660                                                 if (atomic_read
3661                                                     (&(cn->bh->b_count)) < 0) {
3662                                                         reiserfs_warning(p_s_sb,
3663                                                                          "journal-2138: cn->bh->b_count < 0");
3664                                                 }
3665                                         }
3666                                         if (cn->jlist) {        /* since we are clearing the bh, we MUST dec nonzerolen */
3667                                                 atomic_dec(&
3668                                                            (cn->jlist->
3669                                                             j_nonzerolen));
3670                                         }
3671                                         cn->bh = NULL;
3672                                 }
3673                         }
3674                         cn = cn->hnext;
3675                 }
3676         }
3677
3678         if (bh) {
3679                 put_bh(bh);     /* get_hash grabs the buffer */
3680                 if (atomic_read(&(bh->b_count)) < 0) {
3681                         reiserfs_warning(p_s_sb,
3682                                          "journal-2165: bh->b_count < 0");
3683                 }
3684         }
3685         return 0;
3686 }
3687
3688 void reiserfs_update_inode_transaction(struct inode *inode)
3689 {
3690         struct reiserfs_journal *journal = SB_JOURNAL(inode->i_sb);
3691         REISERFS_I(inode)->i_jl = journal->j_current_jl;
3692         REISERFS_I(inode)->i_trans_id = journal->j_trans_id;
3693 }
3694
3695 /*
3696  * returns -1 on error, 0 if no commits/barriers were done and 1
3697  * if a transaction was actually committed and the barrier was done
3698  */
3699 static int __commit_trans_jl(struct inode *inode, unsigned long id,
3700                              struct reiserfs_journal_list *jl)
3701 {
3702         struct reiserfs_transaction_handle th;
3703         struct super_block *sb = inode->i_sb;
3704         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3705         int ret = 0;
3706
3707         /* is it from the current transaction, or from an unknown transaction? */
3708         if (id == journal->j_trans_id) {
3709                 jl = journal->j_current_jl;
3710                 /* try to let other writers come in and grow this transaction */
3711                 let_transaction_grow(sb, id);
3712                 if (journal->j_trans_id != id) {
3713                         goto flush_commit_only;
3714                 }
3715
3716                 ret = journal_begin(&th, sb, 1);
3717                 if (ret)
3718                         return ret;
3719
3720                 /* someone might have ended this transaction while we joined */
3721                 if (journal->j_trans_id != id) {
3722                         reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3723                                                      1);
3724                         journal_mark_dirty(&th, sb, SB_BUFFER_WITH_SB(sb));
3725                         ret = journal_end(&th, sb, 1);
3726                         goto flush_commit_only;
3727                 }
3728
3729                 ret = journal_end_sync(&th, sb, 1);
3730                 if (!ret)
3731                         ret = 1;
3732
3733         } else {
3734                 /* this gets tricky, we have to make sure the journal list in
3735                  * the inode still exists.  We know the list is still around
3736                  * if we've got a larger transaction id than the oldest list
3737                  */
3738               flush_commit_only:
3739                 if (journal_list_still_alive(inode->i_sb, id)) {
3740                         /*
3741                          * we only set ret to 1 when we know for sure
3742                          * the barrier hasn't been started yet on the commit
3743                          * block.
3744                          */
3745                         if (atomic_read(&jl->j_commit_left) > 1)
3746                                 ret = 1;
3747                         flush_commit_list(sb, jl, 1);
3748                         if (journal->j_errno)
3749                                 ret = journal->j_errno;
3750                 }
3751         }
3752         /* otherwise the list is gone, and long since committed */
3753         return ret;
3754 }
3755
3756 int reiserfs_commit_for_inode(struct inode *inode)
3757 {
3758         unsigned long id = REISERFS_I(inode)->i_trans_id;
3759         struct reiserfs_journal_list *jl = REISERFS_I(inode)->i_jl;
3760
3761         /* for the whole inode, assume unset id means it was
3762          * changed in the current transaction.  More conservative
3763          */
3764         if (!id || !jl) {
3765                 reiserfs_update_inode_transaction(inode);
3766                 id = REISERFS_I(inode)->i_trans_id;
3767                 /* jl will be updated in __commit_trans_jl */
3768         }
3769
3770         return __commit_trans_jl(inode, id, jl);
3771 }
3772
3773 void reiserfs_restore_prepared_buffer(struct super_block *p_s_sb,
3774                                       struct buffer_head *bh)
3775 {
3776         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3777         PROC_INFO_INC(p_s_sb, journal.restore_prepared);
3778         if (!bh) {
3779                 return;
3780         }
3781         if (test_clear_buffer_journal_restore_dirty(bh) &&
3782             buffer_journal_dirty(bh)) {
3783                 struct reiserfs_journal_cnode *cn;
3784                 cn = get_journal_hash_dev(p_s_sb,
3785                                           journal->j_list_hash_table,
3786                                           bh->b_blocknr);
3787                 if (cn && can_dirty(cn)) {
3788                         set_buffer_journal_test(bh);
3789                         mark_buffer_dirty(bh);
3790                 }
3791         }
3792         clear_buffer_journal_prepared(bh);
3793 }
3794
3795 extern struct tree_balance *cur_tb;
3796 /*
3797 ** before we can change a metadata block, we have to make sure it won't
3798 ** be written to disk while we are altering it.  So, we must:
3799 ** clean it
3800 ** wait on it.
3801 **
3802 */
3803 int reiserfs_prepare_for_journal(struct super_block *p_s_sb,
3804                                  struct buffer_head *bh, int wait)
3805 {
3806         PROC_INFO_INC(p_s_sb, journal.prepare);
3807
3808         if (test_set_buffer_locked(bh)) {
3809                 if (!wait)
3810                         return 0;
3811                 lock_buffer(bh);
3812         }
3813         set_buffer_journal_prepared(bh);
3814         if (test_clear_buffer_dirty(bh) && buffer_journal_dirty(bh)) {
3815                 clear_buffer_journal_test(bh);
3816                 set_buffer_journal_restore_dirty(bh);
3817         }
3818         unlock_buffer(bh);
3819         return 1;
3820 }
3821
3822 static void flush_old_journal_lists(struct super_block *s)
3823 {
3824         struct reiserfs_journal *journal = SB_JOURNAL(s);
3825         struct reiserfs_journal_list *jl;
3826         struct list_head *entry;
3827         time_t now = get_seconds();
3828
3829         while (!list_empty(&journal->j_journal_list)) {
3830                 entry = journal->j_journal_list.next;
3831                 jl = JOURNAL_LIST_ENTRY(entry);
3832                 /* this check should always be run, to send old lists to disk */
3833                 if (jl->j_timestamp < (now - (JOURNAL_MAX_TRANS_AGE * 4))) {
3834                         flush_used_journal_lists(s, jl);
3835                 } else {
3836                         break;
3837                 }
3838         }
3839 }
3840
3841 /*
3842 ** long and ugly.  If flush, will not return until all commit
3843 ** blocks and all real buffers in the trans are on disk.
3844 ** If no_async, won't return until all commit blocks are on disk.
3845 **
3846 ** keep reading, there are comments as you go along
3847 **
3848 ** If the journal is aborted, we just clean up. Things like flushing
3849 ** journal lists, etc just won't happen.
3850 */
3851 static int do_journal_end(struct reiserfs_transaction_handle *th,
3852                           struct super_block *p_s_sb, unsigned long nblocks,
3853                           int flags)
3854 {
3855         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3856         struct reiserfs_journal_cnode *cn, *next, *jl_cn;
3857         struct reiserfs_journal_cnode *last_cn = NULL;
3858         struct reiserfs_journal_desc *desc;
3859         struct reiserfs_journal_commit *commit;
3860         struct buffer_head *c_bh;       /* commit bh */
3861         struct buffer_head *d_bh;       /* desc bh */
3862         int cur_write_start = 0;        /* start index of current log write */
3863         int old_start;
3864         int i;
3865         int flush = flags & FLUSH_ALL;
3866         int wait_on_commit = flags & WAIT;
3867         struct reiserfs_journal_list *jl, *temp_jl;
3868         struct list_head *entry, *safe;
3869         unsigned long jindex;
3870         unsigned long commit_trans_id;
3871         int trans_half;
3872
3873         BUG_ON(th->t_refcount > 1);
3874         BUG_ON(!th->t_trans_id);
3875
3876         put_fs_excl();
3877         current->journal_info = th->t_handle_save;
3878         reiserfs_check_lock_depth(p_s_sb, "journal end");
3879         if (journal->j_len == 0) {
3880                 reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb),
3881                                              1);
3882                 journal_mark_dirty(th, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb));
3883         }
3884
3885         lock_journal(p_s_sb);
3886         if (journal->j_next_full_flush) {
3887                 flags |= FLUSH_ALL;
3888                 flush = 1;
3889         }
3890         if (journal->j_next_async_flush) {
3891                 flags |= COMMIT_NOW | WAIT;
3892                 wait_on_commit = 1;
3893         }
3894
3895         /* check_journal_end locks the journal, and unlocks if it does not return 1
3896          ** it tells us if we should continue with the journal_end, or just return
3897          */
3898         if (!check_journal_end(th, p_s_sb, nblocks, flags)) {
3899                 p_s_sb->s_dirt = 1;
3900                 wake_queued_writers(p_s_sb);
3901                 reiserfs_async_progress_wait(p_s_sb);
3902                 goto out;
3903         }
3904
3905         /* check_journal_end might set these, check again */
3906         if (journal->j_next_full_flush) {
3907                 flush = 1;
3908         }
3909
3910         /*
3911          ** j must wait means we have to flush the log blocks, and the real blocks for
3912          ** this transaction
3913          */
3914         if (journal->j_must_wait > 0) {
3915                 flush = 1;
3916         }
3917 #ifdef REISERFS_PREALLOCATE
3918         /* quota ops might need to nest, setup the journal_info pointer for them */
3919         current->journal_info = th;
3920         reiserfs_discard_all_prealloc(th);      /* it should not involve new blocks into
3921                                                  * the transaction */
3922         current->journal_info = th->t_handle_save;
3923 #endif
3924
3925         /* setup description block */
3926         d_bh =
3927             journal_getblk(p_s_sb,
3928                            SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
3929                            journal->j_start);
3930         set_buffer_uptodate(d_bh);
3931         desc = (struct reiserfs_journal_desc *)(d_bh)->b_data;
3932         memset(d_bh->b_data, 0, d_bh->b_size);
3933         memcpy(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8);
3934         set_desc_trans_id(desc, journal->j_trans_id);
3935
3936         /* setup commit block.  Don't write (keep it clean too) this one until after everyone else is written */
3937         c_bh = journal_getblk(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
3938                               ((journal->j_start + journal->j_len +
3939                                 1) % SB_ONDISK_JOURNAL_SIZE(p_s_sb)));
3940         commit = (struct reiserfs_journal_commit *)c_bh->b_data;
3941         memset(c_bh->b_data, 0, c_bh->b_size);
3942         set_commit_trans_id(commit, journal->j_trans_id);
3943         set_buffer_uptodate(c_bh);
3944
3945         /* init this journal list */
3946         jl = journal->j_current_jl;
3947
3948         /* we lock the commit before doing anything because
3949          * we want to make sure nobody tries to run flush_commit_list until
3950          * the new transaction is fully setup, and we've already flushed the
3951          * ordered bh list
3952          */
3953         down(&jl->j_commit_lock);
3954
3955         /* save the transaction id in case we need to commit it later */
3956         commit_trans_id = jl->j_trans_id;
3957
3958         atomic_set(&jl->j_older_commits_done, 0);
3959         jl->j_trans_id = journal->j_trans_id;
3960         jl->j_timestamp = journal->j_trans_start_time;
3961         jl->j_commit_bh = c_bh;
3962         jl->j_start = journal->j_start;
3963         jl->j_len = journal->j_len;
3964         atomic_set(&jl->j_nonzerolen, journal->j_len);
3965         atomic_set(&jl->j_commit_left, journal->j_len + 2);
3966         jl->j_realblock = NULL;
3967
3968         /* The ENTIRE FOR LOOP MUST not cause schedule to occur.
3969          **  for each real block, add it to the journal list hash,
3970          ** copy into real block index array in the commit or desc block
3971          */
3972         trans_half = journal_trans_half(p_s_sb->s_blocksize);
3973         for (i = 0, cn = journal->j_first; cn; cn = cn->next, i++) {
3974                 if (buffer_journaled(cn->bh)) {
3975                         jl_cn = get_cnode(p_s_sb);
3976                         if (!jl_cn) {
3977                                 reiserfs_panic(p_s_sb,
3978                                                "journal-1676, get_cnode returned NULL\n");
3979                         }
3980                         if (i == 0) {
3981                                 jl->j_realblock = jl_cn;
3982                         }
3983                         jl_cn->prev = last_cn;
3984                         jl_cn->next = NULL;
3985                         if (last_cn) {
3986                                 last_cn->next = jl_cn;
3987                         }
3988                         last_cn = jl_cn;
3989                         /* make sure the block we are trying to log is not a block
3990                            of journal or reserved area */
3991
3992                         if (is_block_in_log_or_reserved_area
3993                             (p_s_sb, cn->bh->b_blocknr)) {
3994                                 reiserfs_panic(p_s_sb,
3995                                                "journal-2332: Trying to log block %lu, which is a log block\n",
3996                                                cn->bh->b_blocknr);
3997                         }
3998                         jl_cn->blocknr = cn->bh->b_blocknr;
3999                         jl_cn->state = 0;
4000                         jl_cn->sb = p_s_sb;
4001                         jl_cn->bh = cn->bh;
4002                         jl_cn->jlist = jl;
4003                         insert_journal_hash(journal->j_list_hash_table, jl_cn);
4004                         if (i < trans_half) {
4005                                 desc->j_realblock[i] =
4006                                     cpu_to_le32(cn->bh->b_blocknr);
4007                         } else {
4008                                 commit->j_realblock[i - trans_half] =
4009                                     cpu_to_le32(cn->bh->b_blocknr);
4010                         }
4011                 } else {
4012                         i--;
4013                 }
4014         }
4015         set_desc_trans_len(desc, journal->j_len);
4016         set_desc_mount_id(desc, journal->j_mount_id);
4017         set_desc_trans_id(desc, journal->j_trans_id);
4018         set_commit_trans_len(commit, journal->j_len);
4019
4020         /* special check in case all buffers in the journal were marked for not logging */
4021         if (journal->j_len == 0) {
4022                 BUG();
4023         }
4024
4025         /* we're about to dirty all the log blocks, mark the description block
4026          * dirty now too.  Don't mark the commit block dirty until all the
4027          * others are on disk
4028          */
4029         mark_buffer_dirty(d_bh);
4030
4031         /* first data block is j_start + 1, so add one to cur_write_start wherever you use it */
4032         cur_write_start = journal->j_start;
4033         cn = journal->j_first;
4034         jindex = 1;             /* start at one so we don't get the desc again */
4035         while (cn) {
4036                 clear_buffer_journal_new(cn->bh);
4037                 /* copy all the real blocks into log area.  dirty log blocks */
4038                 if (buffer_journaled(cn->bh)) {
4039                         struct buffer_head *tmp_bh;
4040                         char *addr;
4041                         struct page *page;
4042                         tmp_bh =
4043                             journal_getblk(p_s_sb,
4044                                            SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
4045                                            ((cur_write_start +
4046                                              jindex) %
4047                                             SB_ONDISK_JOURNAL_SIZE(p_s_sb)));
4048                         set_buffer_uptodate(tmp_bh);
4049                         page = cn->bh->b_page;
4050                         addr = kmap(page);
4051                         memcpy(tmp_bh->b_data,
4052                                addr + offset_in_page(cn->bh->b_data),
4053                                cn->bh->b_size);
4054                         kunmap(page);
4055                         mark_buffer_dirty(tmp_bh);
4056                         jindex++;
4057                         set_buffer_journal_dirty(cn->bh);
4058                         clear_buffer_journaled(cn->bh);
4059                 } else {
4060                         /* JDirty cleared sometime during transaction.  don't log this one */
4061                         reiserfs_warning(p_s_sb,
4062                                          "journal-2048: do_journal_end: BAD, buffer in journal hash, but not JDirty!");
4063                         brelse(cn->bh);
4064                 }
4065                 next = cn->next;
4066                 free_cnode(p_s_sb, cn);
4067                 cn = next;
4068                 cond_resched();
4069         }
4070
4071         /* we are done  with both the c_bh and d_bh, but
4072          ** c_bh must be written after all other commit blocks,
4073          ** so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1.
4074          */
4075
4076         journal->j_current_jl = alloc_journal_list(p_s_sb);
4077
4078         /* now it is safe to insert this transaction on the main list */
4079         list_add_tail(&jl->j_list, &journal->j_journal_list);
4080         list_add_tail(&jl->j_working_list, &journal->j_working_list);
4081         journal->j_num_work_lists++;
4082
4083         /* reset journal values for the next transaction */
4084         old_start = journal->j_start;
4085         journal->j_start =
4086             (journal->j_start + journal->j_len +
4087              2) % SB_ONDISK_JOURNAL_SIZE(p_s_sb);
4088         atomic_set(&(journal->j_wcount), 0);
4089         journal->j_bcount = 0;
4090         journal->j_last = NULL;
4091         journal->j_first = NULL;
4092         journal->j_len = 0;
4093         journal->j_trans_start_time = 0;
4094         journal->j_trans_id++;
4095         journal->j_current_jl->j_trans_id = journal->j_trans_id;
4096         journal->j_must_wait = 0;
4097         journal->j_len_alloc = 0;
4098         journal->j_next_full_flush = 0;
4099         journal->j_next_async_flush = 0;
4100         init_journal_hash(p_s_sb);
4101
4102         // make sure reiserfs_add_jh sees the new current_jl before we
4103         // write out the tails
4104         smp_mb();
4105
4106         /* tail conversion targets have to hit the disk before we end the
4107          * transaction.  Otherwise a later transaction might repack the tail
4108          * before this transaction commits, leaving the data block unflushed and
4109          * clean, if we crash before the later transaction commits, the data block
4110          * is lost.
4111          */
4112         if (!list_empty(&jl->j_tail_bh_list)) {
4113                 unlock_kernel();
4114                 write_ordered_buffers(&journal->j_dirty_buffers_lock,
4115                                       journal, jl, &jl->j_tail_bh_list);
4116                 lock_kernel();
4117         }
4118         if (!list_empty(&jl->j_tail_bh_list))
4119                 BUG();
4120         up(&jl->j_commit_lock);
4121
4122         /* honor the flush wishes from the caller, simple commits can
4123          ** be done outside the journal lock, they are done below
4124          **
4125          ** if we don't flush the commit list right now, we put it into
4126          ** the work queue so the people waiting on the async progress work
4127          ** queue don't wait for this proc to flush journal lists and such.
4128          */
4129         if (flush) {
4130                 flush_commit_list(p_s_sb, jl, 1);
4131                 flush_journal_list(p_s_sb, jl, 1);
4132         } else if (!(jl->j_state & LIST_COMMIT_PENDING))
4133                 queue_delayed_work(commit_wq, &journal->j_work, HZ / 10);
4134
4135         /* if the next transaction has any chance of wrapping, flush
4136          ** transactions that might get overwritten.  If any journal lists are very
4137          ** old flush them as well.
4138          */
4139       first_jl:
4140         list_for_each_safe(entry, safe, &journal->j_journal_list) {
4141                 temp_jl = JOURNAL_LIST_ENTRY(entry);
4142                 if (journal->j_start <= temp_jl->j_start) {
4143                         if ((journal->j_start + journal->j_trans_max + 1) >=
4144                             temp_jl->j_start) {
4145                                 flush_used_journal_lists(p_s_sb, temp_jl);
4146                                 goto first_jl;
4147                         } else if ((journal->j_start +
4148                                     journal->j_trans_max + 1) <
4149                                    SB_ONDISK_JOURNAL_SIZE(p_s_sb)) {
4150                                 /* if we don't cross into the next transaction and we don't
4151                                  * wrap, there is no way we can overlap any later transactions
4152                                  * break now
4153                                  */
4154                                 break;
4155                         }
4156                 } else if ((journal->j_start +
4157                             journal->j_trans_max + 1) >
4158                            SB_ONDISK_JOURNAL_SIZE(p_s_sb)) {
4159                         if (((journal->j_start + journal->j_trans_max + 1) %
4160                              SB_ONDISK_JOURNAL_SIZE(p_s_sb)) >=
4161                             temp_jl->j_start) {
4162                                 flush_used_journal_lists(p_s_sb, temp_jl);
4163                                 goto first_jl;
4164                         } else {
4165                                 /* we don't overlap anything from out start to the end of the
4166                                  * log, and our wrapped portion doesn't overlap anything at
4167                                  * the start of the log.  We can break
4168                                  */
4169                                 break;
4170                         }
4171                 }
4172         }
4173         flush_old_journal_lists(p_s_sb);
4174
4175         journal->j_current_jl->j_list_bitmap =
4176             get_list_bitmap(p_s_sb, journal->j_current_jl);
4177
4178         if (!(journal->j_current_jl->j_list_bitmap)) {
4179                 reiserfs_panic(p_s_sb,
4180                                "journal-1996: do_journal_end, could not get a list bitmap\n");
4181         }
4182
4183         atomic_set(&(journal->j_jlock), 0);
4184         unlock_journal(p_s_sb);
4185         /* wake up any body waiting to join. */
4186         clear_bit(J_WRITERS_QUEUED, &journal->j_state);
4187         wake_up(&(journal->j_join_wait));
4188
4189         if (!flush && wait_on_commit &&
4190             journal_list_still_alive(p_s_sb, commit_trans_id)) {
4191                 flush_commit_list(p_s_sb, jl, 1);
4192         }
4193       out:
4194         reiserfs_check_lock_depth(p_s_sb, "journal end2");
4195
4196         memset(th, 0, sizeof(*th));
4197         /* Re-set th->t_super, so we can properly keep track of how many
4198          * persistent transactions there are. We need to do this so if this
4199          * call is part of a failed restart_transaction, we can free it later */
4200         th->t_super = p_s_sb;
4201
4202         return journal->j_errno;
4203 }
4204
4205 static void __reiserfs_journal_abort_hard(struct super_block *sb)
4206 {
4207         struct reiserfs_journal *journal = SB_JOURNAL(sb);
4208         if (test_bit(J_ABORTED, &journal->j_state))
4209                 return;
4210
4211         printk(KERN_CRIT "REISERFS: Aborting journal for filesystem on %s\n",
4212                reiserfs_bdevname(sb));
4213
4214         sb->s_flags |= MS_RDONLY;
4215         set_bit(J_ABORTED, &journal->j_state);
4216
4217 #ifdef CONFIG_REISERFS_CHECK
4218         dump_stack();
4219 #endif
4220 }
4221
4222 static void __reiserfs_journal_abort_soft(struct super_block *sb, int errno)
4223 {
4224         struct reiserfs_journal *journal = SB_JOURNAL(sb);
4225         if (test_bit(J_ABORTED, &journal->j_state))
4226                 return;
4227
4228         if (!journal->j_errno)
4229                 journal->j_errno = errno;
4230
4231         __reiserfs_journal_abort_hard(sb);
4232 }
4233
4234 void reiserfs_journal_abort(struct super_block *sb, int errno)
4235 {
4236         return __reiserfs_journal_abort_soft(sb, errno);
4237 }