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 = kmalloc(sizeof(struct reiserfs_bitmap_node), GFP_NOFS);
 156         if (!bn) {
 157                 return NULL;
 158         }
 159         bn->data = kzalloc(p_s_sb->s_blocksize, GFP_NOFS);
 160         if (!bn->data) {
 161                 kfree(bn);
 162                 return NULL;
 163         }
 164         bn->id = id++;
 165         INIT_LIST_HEAD(&bn->list);
 166         return bn;
 167 }
 168
 169 static struct reiserfs_bitmap_node *get_bitmap_node(struct super_block *p_s_sb)
 170 {
 171         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 172         struct reiserfs_bitmap_node *bn = NULL;
 173         struct list_head *entry = journal->j_bitmap_nodes.next;
 174
 175         journal->j_used_bitmap_nodes++;
 176       repeat:
 177
 178         if (entry != &journal->j_bitmap_nodes) {
 179                 bn = list_entry(entry, struct reiserfs_bitmap_node, list);
 180                 list_del(entry);
 181                 memset(bn->data, 0, p_s_sb->s_blocksize);
 182                 journal->j_free_bitmap_nodes--;
 183                 return bn;
 184         }
 185         bn = allocate_bitmap_node(p_s_sb);
 186         if (!bn) {
 187                 yield();
 188                 goto repeat;
 189         }
 190         return bn;
 191 }
 192 static inline void free_bitmap_node(struct super_block *p_s_sb,
 193                                     struct reiserfs_bitmap_node *bn)
 194 {
 195         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 196         journal->j_used_bitmap_nodes--;
 197         if (journal->j_free_bitmap_nodes > REISERFS_MAX_BITMAP_NODES) {
 198                 kfree(bn->data);
 199                 kfree(bn);
 200         } else {
 201                 list_add(&bn->list, &journal->j_bitmap_nodes);
 202                 journal->j_free_bitmap_nodes++;
 203         }
 204 }
 205
 206 static void allocate_bitmap_nodes(struct super_block *p_s_sb)
 207 {
 208         int i;
 209         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 210         struct reiserfs_bitmap_node *bn = NULL;
 211         for (i = 0; i < REISERFS_MIN_BITMAP_NODES; i++) {
 212                 bn = allocate_bitmap_node(p_s_sb);
 213                 if (bn) {
 214                         list_add(&bn->list, &journal->j_bitmap_nodes);
 215                         journal->j_free_bitmap_nodes++;
 216                 } else {
 217                         break;  // this is ok, we'll try again when more are needed
 218                 }
 219         }
 220 }
 221
 222 static int set_bit_in_list_bitmap(struct super_block *p_s_sb, int block,
 223                                   struct reiserfs_list_bitmap *jb)
 224 {
 225         int bmap_nr = block / (p_s_sb->s_blocksize << 3);
 226         int bit_nr = block % (p_s_sb->s_blocksize << 3);
 227
 228         if (!jb->bitmaps[bmap_nr]) {
 229                 jb->bitmaps[bmap_nr] = get_bitmap_node(p_s_sb);
 230         }
 231         set_bit(bit_nr, (unsigned long *)jb->bitmaps[bmap_nr]->data);
 232         return 0;
 233 }
 234
 235 static void cleanup_bitmap_list(struct super_block *p_s_sb,
 236                                 struct reiserfs_list_bitmap *jb)
 237 {
 238         int i;
 239         if (jb->bitmaps == NULL)
 240                 return;
 241
 242         for (i = 0; i < SB_BMAP_NR(p_s_sb); i++) {
 243                 if (jb->bitmaps[i]) {
 244                         free_bitmap_node(p_s_sb, jb->bitmaps[i]);
 245                         jb->bitmaps[i] = NULL;
 246                 }
 247         }
 248 }
 249
 250 /*
 251 ** only call this on FS unmount.
 252 */
 253 static int free_list_bitmaps(struct super_block *p_s_sb,
 254                              struct reiserfs_list_bitmap *jb_array)
 255 {
 256         int i;
 257         struct reiserfs_list_bitmap *jb;
 258         for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
 259                 jb = jb_array + i;
 260                 jb->journal_list = NULL;
 261                 cleanup_bitmap_list(p_s_sb, jb);
 262                 vfree(jb->bitmaps);
 263                 jb->bitmaps = NULL;
 264         }
 265         return 0;
 266 }
 267
 268 static int free_bitmap_nodes(struct super_block *p_s_sb)
 269 {
 270         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 271         struct list_head *next = journal->j_bitmap_nodes.next;
 272         struct reiserfs_bitmap_node *bn;
 273
 274         while (next != &journal->j_bitmap_nodes) {
 275                 bn = list_entry(next, struct reiserfs_bitmap_node, list);
 276                 list_del(next);
 277                 kfree(bn->data);
 278                 kfree(bn);
 279                 next = journal->j_bitmap_nodes.next;
 280                 journal->j_free_bitmap_nodes--;
 281         }
 282
 283         return 0;
 284 }
 285
 286 /*
 287 ** get memory for JOURNAL_NUM_BITMAPS worth of bitmaps.
 288 ** jb_array is the array to be filled in.
 289 */
 290 int reiserfs_allocate_list_bitmaps(struct super_block *p_s_sb,
 291                                    struct reiserfs_list_bitmap *jb_array,
 292                                    int bmap_nr)
 293 {
 294         int i;
 295         int failed = 0;
 296         struct reiserfs_list_bitmap *jb;
 297         int mem = bmap_nr * sizeof(struct reiserfs_bitmap_node *);
 298
 299         for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
 300                 jb = jb_array + i;
 301                 jb->journal_list = NULL;
 302                 jb->bitmaps = vmalloc(mem);
 303                 if (!jb->bitmaps) {
 304                         reiserfs_warning(p_s_sb,
 305                                          "clm-2000, unable to allocate bitmaps for journal lists");
 306                         failed = 1;
 307                         break;
 308                 }
 309                 memset(jb->bitmaps, 0, mem);
 310         }
 311         if (failed) {
 312                 free_list_bitmaps(p_s_sb, jb_array);
 313                 return -1;
 314         }
 315         return 0;
 316 }
 317
 318 /*
 319 ** find an available list bitmap.  If you can't find one, flush a commit list
 320 ** and try again
 321 */
 322 static struct reiserfs_list_bitmap *get_list_bitmap(struct super_block *p_s_sb,
 323                                                     struct reiserfs_journal_list
 324                                                     *jl)
 325 {
 326         int i, j;
 327         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 328         struct reiserfs_list_bitmap *jb = NULL;
 329
 330         for (j = 0; j < (JOURNAL_NUM_BITMAPS * 3); j++) {
 331                 i = journal->j_list_bitmap_index;
 332                 journal->j_list_bitmap_index = (i + 1) % JOURNAL_NUM_BITMAPS;
 333                 jb = journal->j_list_bitmap + i;
 334                 if (journal->j_list_bitmap[i].journal_list) {
 335                         flush_commit_list(p_s_sb,
 336                                           journal->j_list_bitmap[i].
 337                                           journal_list, 1);
 338                         if (!journal->j_list_bitmap[i].journal_list) {
 339                                 break;
 340                         }
 341                 } else {
 342                         break;
 343                 }
 344         }
 345         if (jb->journal_list) { /* double check to make sure if flushed correctly */
 346                 return NULL;
 347         }
 348         jb->journal_list = jl;
 349         return jb;
 350 }
 351
 352 /*
 353 ** allocates a new chunk of X nodes, and links them all together as a list.
 354 ** Uses the cnode->next and cnode->prev pointers
 355 ** returns NULL on failure
 356 */
 357 static struct reiserfs_journal_cnode *allocate_cnodes(int num_cnodes)
 358 {
 359         struct reiserfs_journal_cnode *head;
 360         int i;
 361         if (num_cnodes <= 0) {
 362                 return NULL;
 363         }
 364         head = vmalloc(num_cnodes * sizeof(struct reiserfs_journal_cnode));
 365         if (!head) {
 366                 return NULL;
 367         }
 368         memset(head, 0, num_cnodes * sizeof(struct reiserfs_journal_cnode));
 369         head[0].prev = NULL;
 370         head[0].next = head + 1;
 371         for (i = 1; i < num_cnodes; i++) {
 372                 head[i].prev = head + (i - 1);
 373                 head[i].next = head + (i + 1);  /* if last one, overwrite it after the if */
 374         }
 375         head[num_cnodes - 1].next = NULL;
 376         return head;
 377 }
 378
 379 /*
 380 ** pulls a cnode off the free list, or returns NULL on failure
 381 */
 382 static struct reiserfs_journal_cnode *get_cnode(struct super_block *p_s_sb)
 383 {
 384         struct reiserfs_journal_cnode *cn;
 385         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 386
 387         reiserfs_check_lock_depth(p_s_sb, "get_cnode");
 388
 389         if (journal->j_cnode_free <= 0) {
 390                 return NULL;
 391         }
 392         journal->j_cnode_used++;
 393         journal->j_cnode_free--;
 394         cn = journal->j_cnode_free_list;
 395         if (!cn) {
 396                 return cn;
 397         }
 398         if (cn->next) {
 399                 cn->next->prev = NULL;
 400         }
 401         journal->j_cnode_free_list = cn->next;
 402         memset(cn, 0, sizeof(struct reiserfs_journal_cnode));
 403         return cn;
 404 }
 405
 406 /*
 407 ** returns a cnode to the free list
 408 */
 409 static void free_cnode(struct super_block *p_s_sb,
 410                        struct reiserfs_journal_cnode *cn)
 411 {
 412         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 413
 414         reiserfs_check_lock_depth(p_s_sb, "free_cnode");
 415
 416         journal->j_cnode_used--;
 417         journal->j_cnode_free++;
 418         /* memset(cn, 0, sizeof(struct reiserfs_journal_cnode)) ; */
 419         cn->next = journal->j_cnode_free_list;
 420         if (journal->j_cnode_free_list) {
 421                 journal->j_cnode_free_list->prev = cn;
 422         }
 423         cn->prev = NULL;        /* not needed with the memset, but I might kill the memset, and forget to do this */
 424         journal->j_cnode_free_list = cn;
 425 }
 426
 427 static void clear_prepared_bits(struct buffer_head *bh)
 428 {
 429         clear_buffer_journal_prepared(bh);
 430         clear_buffer_journal_restore_dirty(bh);
 431 }
 432
 433 /* utility function to force a BUG if it is called without the big
 434 ** kernel lock held.  caller is the string printed just before calling BUG()
 435 */
 436 void reiserfs_check_lock_depth(struct super_block *sb, char *caller)
 437 {
 438 #ifdef CONFIG_SMP
 439         if (current->lock_depth < 0) {
 440                 reiserfs_panic(sb, "%s called without kernel lock held",
 441                                caller);
 442         }
 443 #else
 444         ;
 445 #endif
 446 }
 447
 448 /* return a cnode with same dev, block number and size in table, or null if not found */
 449 static inline struct reiserfs_journal_cnode *get_journal_hash_dev(struct
 450                                                                   super_block
 451                                                                   *sb,
 452                                                                   struct
 453                                                                   reiserfs_journal_cnode
 454                                                                   **table,
 455                                                                   long bl)
 456 {
 457         struct reiserfs_journal_cnode *cn;
 458         cn = journal_hash(table, sb, bl);
 459         while (cn) {
 460                 if (cn->blocknr == bl && cn->sb == sb)
 461                         return cn;
 462                 cn = cn->hnext;
 463         }
 464         return (struct reiserfs_journal_cnode *)0;
 465 }
 466
 467 /*
 468 ** this actually means 'can this block be reallocated yet?'.  If you set search_all, a block can only be allocated
 469 ** if it is not in the current transaction, was not freed by the current transaction, and has no chance of ever
 470 ** being overwritten by a replay after crashing.
 471 **
 472 ** If you don't set search_all, a block can only be allocated if it is not in the current transaction.  Since deleting
 473 ** a block removes it from the current transaction, this case should never happen.  If you don't set search_all, make
 474 ** sure you never write the block without logging it.
 475 **
 476 ** next_zero_bit is a suggestion about the next block to try for find_forward.
 477 ** when bl is rejected because it is set in a journal list bitmap, we search
 478 ** for the next zero bit in the bitmap that rejected bl.  Then, we return that
 479 ** through next_zero_bit for find_forward to try.
 480 **
 481 ** Just because we return something in next_zero_bit does not mean we won't
 482 ** reject it on the next call to reiserfs_in_journal
 483 **
 484 */
 485 int reiserfs_in_journal(struct super_block *p_s_sb,
 486                         int bmap_nr, int bit_nr, int search_all,
 487                         b_blocknr_t * next_zero_bit)
 488 {
 489         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 490         struct reiserfs_journal_cnode *cn;
 491         struct reiserfs_list_bitmap *jb;
 492         int i;
 493         unsigned long bl;
 494
 495         *next_zero_bit = 0;     /* always start this at zero. */
 496
 497         PROC_INFO_INC(p_s_sb, journal.in_journal);
 498         /* If we aren't doing a search_all, this is a metablock, and it will be logged before use.
 499          ** if we crash before the transaction that freed it commits,  this transaction won't
 500          ** have committed either, and the block will never be written
 501          */
 502         if (search_all) {
 503                 for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
 504                         PROC_INFO_INC(p_s_sb, journal.in_journal_bitmap);
 505                         jb = journal->j_list_bitmap + i;
 506                         if (jb->journal_list && jb->bitmaps[bmap_nr] &&
 507                             test_bit(bit_nr,
 508                                      (unsigned long *)jb->bitmaps[bmap_nr]->
 509                                      data)) {
 510                                 *next_zero_bit =
 511                                     find_next_zero_bit((unsigned long *)
 512                                                        (jb->bitmaps[bmap_nr]->
 513                                                         data),
 514                                                        p_s_sb->s_blocksize << 3,
 515                                                        bit_nr + 1);
 516                                 return 1;
 517                         }
 518                 }
 519         }
 520
 521         bl = bmap_nr * (p_s_sb->s_blocksize << 3) + bit_nr;
 522         /* is it in any old transactions? */
 523         if (search_all
 524             && (cn =
 525                 get_journal_hash_dev(p_s_sb, journal->j_list_hash_table, bl))) {
 526                 return 1;
 527         }
 528
 529         /* is it in the current transaction.  This should never happen */
 530         if ((cn = get_journal_hash_dev(p_s_sb, journal->j_hash_table, bl))) {
 531                 BUG();
 532                 return 1;
 533         }
 534
 535         PROC_INFO_INC(p_s_sb, journal.in_journal_reusable);
 536         /* safe for reuse */
 537         return 0;
 538 }
 539
 540 /* insert cn into table
 541 */
 542 static inline void insert_journal_hash(struct reiserfs_journal_cnode **table,
 543                                        struct reiserfs_journal_cnode *cn)
 544 {
 545         struct reiserfs_journal_cnode *cn_orig;
 546
 547         cn_orig = journal_hash(table, cn->sb, cn->blocknr);
 548         cn->hnext = cn_orig;
 549         cn->hprev = NULL;
 550         if (cn_orig) {
 551                 cn_orig->hprev = cn;
 552         }
 553         journal_hash(table, cn->sb, cn->blocknr) = cn;
 554 }
 555
 556 /* lock the current transaction */
 557 static inline void lock_journal(struct super_block *p_s_sb)
 558 {
 559         PROC_INFO_INC(p_s_sb, journal.lock_journal);
 560         down(&SB_JOURNAL(p_s_sb)->j_lock);
 561 }
 562
 563 /* unlock the current transaction */
 564 static inline void unlock_journal(struct super_block *p_s_sb)
 565 {
 566         up(&SB_JOURNAL(p_s_sb)->j_lock);
 567 }
 568
 569 static inline void get_journal_list(struct reiserfs_journal_list *jl)
 570 {
 571         jl->j_refcount++;
 572 }
 573
 574 static inline void put_journal_list(struct super_block *s,
 575                                     struct reiserfs_journal_list *jl)
 576 {
 577         if (jl->j_refcount < 1) {
 578                 reiserfs_panic(s, "trans id %lu, refcount at %d",
 579                                jl->j_trans_id, jl->j_refcount);
 580         }
 581         if (--jl->j_refcount == 0)
 582                 kfree(jl);
 583 }
 584
 585 /*
 586 ** this used to be much more involved, and I'm keeping it just in case things get ugly again.
 587 ** it gets called by flush_commit_list, and cleans up any data stored about blocks freed during a
 588 ** transaction.
 589 */
 590 static void cleanup_freed_for_journal_list(struct super_block *p_s_sb,
 591                                            struct reiserfs_journal_list *jl)
 592 {
 593
 594         struct reiserfs_list_bitmap *jb = jl->j_list_bitmap;
 595         if (jb) {
 596                 cleanup_bitmap_list(p_s_sb, jb);
 597         }
 598         jl->j_list_bitmap->journal_list = NULL;
 599         jl->j_list_bitmap = NULL;
 600 }
 601
 602 static int journal_list_still_alive(struct super_block *s,
 603                                     unsigned long trans_id)
 604 {
 605         struct reiserfs_journal *journal = SB_JOURNAL(s);
 606         struct list_head *entry = &journal->j_journal_list;
 607         struct reiserfs_journal_list *jl;
 608
 609         if (!list_empty(entry)) {
 610                 jl = JOURNAL_LIST_ENTRY(entry->next);
 611                 if (jl->j_trans_id <= trans_id) {
 612                         return 1;
 613                 }
 614         }
 615         return 0;
 616 }
 617
 618 static void reiserfs_end_buffer_io_sync(struct buffer_head *bh, int uptodate)
 619 {
 620         char b[BDEVNAME_SIZE];
 621
 622         if (buffer_journaled(bh)) {
 623                 reiserfs_warning(NULL,
 624                                  "clm-2084: pinned buffer %lu:%s sent to disk",
 625                                  bh->b_blocknr, bdevname(bh->b_bdev, b));
 626         }
 627         if (uptodate)
 628                 set_buffer_uptodate(bh);
 629         else
 630                 clear_buffer_uptodate(bh);
 631         unlock_buffer(bh);
 632         put_bh(bh);
 633 }
 634
 635 static void reiserfs_end_ordered_io(struct buffer_head *bh, int uptodate)
 636 {
 637         if (uptodate)
 638                 set_buffer_uptodate(bh);
 639         else
 640                 clear_buffer_uptodate(bh);
 641         unlock_buffer(bh);
 642         put_bh(bh);
 643 }
 644
 645 static void submit_logged_buffer(struct buffer_head *bh)
 646 {
 647         get_bh(bh);
 648         bh->b_end_io = reiserfs_end_buffer_io_sync;
 649         clear_buffer_journal_new(bh);
 650         clear_buffer_dirty(bh);
 651         if (!test_clear_buffer_journal_test(bh))
 652                 BUG();
 653         if (!buffer_uptodate(bh))
 654                 BUG();
 655         submit_bh(WRITE, bh);
 656 }
 657
 658 static void submit_ordered_buffer(struct buffer_head *bh)
 659 {
 660         get_bh(bh);
 661         bh->b_end_io = reiserfs_end_ordered_io;
 662         clear_buffer_dirty(bh);
 663         if (!buffer_uptodate(bh))
 664                 BUG();
 665         submit_bh(WRITE, bh);
 666 }
 667
 668 static int submit_barrier_buffer(struct buffer_head *bh)
 669 {
 670         get_bh(bh);
 671         bh->b_end_io = reiserfs_end_ordered_io;
 672         clear_buffer_dirty(bh);
 673         if (!buffer_uptodate(bh))
 674                 BUG();
 675         return submit_bh(WRITE_BARRIER, bh);
 676 }
 677
 678 static void check_barrier_completion(struct super_block *s,
 679                                      struct buffer_head *bh)
 680 {
 681         if (buffer_eopnotsupp(bh)) {
 682                 clear_buffer_eopnotsupp(bh);
 683                 disable_barrier(s);
 684                 set_buffer_uptodate(bh);
 685                 set_buffer_dirty(bh);
 686                 sync_dirty_buffer(bh);
 687         }
 688 }
 689
 690 #define CHUNK_SIZE 32
 691 struct buffer_chunk {
 692         struct buffer_head *bh[CHUNK_SIZE];
 693         int nr;
 694 };
 695
 696 static void write_chunk(struct buffer_chunk *chunk)
 697 {
 698         int i;
 699         get_fs_excl();
 700         for (i = 0; i < chunk->nr; i++) {
 701                 submit_logged_buffer(chunk->bh[i]);
 702         }
 703         chunk->nr = 0;
 704         put_fs_excl();
 705 }
 706
 707 static void write_ordered_chunk(struct buffer_chunk *chunk)
 708 {
 709         int i;
 710         get_fs_excl();
 711         for (i = 0; i < chunk->nr; i++) {
 712                 submit_ordered_buffer(chunk->bh[i]);
 713         }
 714         chunk->nr = 0;
 715         put_fs_excl();
 716 }
 717
 718 static int add_to_chunk(struct buffer_chunk *chunk, struct buffer_head *bh,
 719                         spinlock_t * lock, void (fn) (struct buffer_chunk *))
 720 {
 721         int ret = 0;
 722         if (chunk->nr >= CHUNK_SIZE)
 723                 BUG();
 724         chunk->bh[chunk->nr++] = bh;
 725         if (chunk->nr >= CHUNK_SIZE) {
 726                 ret = 1;
 727                 if (lock)
 728                         spin_unlock(lock);
 729                 fn(chunk);
 730                 if (lock)
 731                         spin_lock(lock);
 732         }
 733         return ret;
 734 }
 735
 736 static atomic_t nr_reiserfs_jh = ATOMIC_INIT(0);
 737 static struct reiserfs_jh *alloc_jh(void)
 738 {
 739         struct reiserfs_jh *jh;
 740         while (1) {
 741                 jh = kmalloc(sizeof(*jh), GFP_NOFS);
 742                 if (jh) {
 743                         atomic_inc(&nr_reiserfs_jh);
 744                         return jh;
 745                 }
 746                 yield();
 747         }
 748 }
 749
 750 /*
 751  * we want to free the jh when the buffer has been written
 752  * and waited on
 753  */
 754 void reiserfs_free_jh(struct buffer_head *bh)
 755 {
 756         struct reiserfs_jh *jh;
 757
 758         jh = bh->b_private;
 759         if (jh) {
 760                 bh->b_private = NULL;
 761                 jh->bh = NULL;
 762                 list_del_init(&jh->list);
 763                 kfree(jh);
 764                 if (atomic_read(&nr_reiserfs_jh) <= 0)
 765                         BUG();
 766                 atomic_dec(&nr_reiserfs_jh);
 767                 put_bh(bh);
 768         }
 769 }
 770
 771 static inline int __add_jh(struct reiserfs_journal *j, struct buffer_head *bh,
 772                            int tail)
 773 {
 774         struct reiserfs_jh *jh;
 775
 776         if (bh->b_private) {
 777                 spin_lock(&j->j_dirty_buffers_lock);
 778                 if (!bh->b_private) {
 779                         spin_unlock(&j->j_dirty_buffers_lock);
 780                         goto no_jh;
 781                 }
 782                 jh = bh->b_private;
 783                 list_del_init(&jh->list);
 784         } else {
 785               no_jh:
 786                 get_bh(bh);
 787                 jh = alloc_jh();
 788                 spin_lock(&j->j_dirty_buffers_lock);
 789                 /* buffer must be locked for __add_jh, should be able to have
 790                  * two adds at the same time
 791                  */
 792                 if (bh->b_private)
 793                         BUG();
 794                 jh->bh = bh;
 795                 bh->b_private = jh;
 796         }
 797         jh->jl = j->j_current_jl;
 798         if (tail)
 799                 list_add_tail(&jh->list, &jh->jl->j_tail_bh_list);
 800         else {
 801                 list_add_tail(&jh->list, &jh->jl->j_bh_list);
 802         }
 803         spin_unlock(&j->j_dirty_buffers_lock);
 804         return 0;
 805 }
 806
 807 int reiserfs_add_tail_list(struct inode *inode, struct buffer_head *bh)
 808 {
 809         return __add_jh(SB_JOURNAL(inode->i_sb), bh, 1);
 810 }
 811 int reiserfs_add_ordered_list(struct inode *inode, struct buffer_head *bh)
 812 {
 813         return __add_jh(SB_JOURNAL(inode->i_sb), bh, 0);
 814 }
 815
 816 #define JH_ENTRY(l) list_entry((l), struct reiserfs_jh, list)
 817 static int write_ordered_buffers(spinlock_t * lock,
 818                                  struct reiserfs_journal *j,
 819                                  struct reiserfs_journal_list *jl,
 820                                  struct list_head *list)
 821 {
 822         struct buffer_head *bh;
 823         struct reiserfs_jh *jh;
 824         int ret = j->j_errno;
 825         struct buffer_chunk chunk;
 826         struct list_head tmp;
 827         INIT_LIST_HEAD(&tmp);
 828
 829         chunk.nr = 0;
 830         spin_lock(lock);
 831         while (!list_empty(list)) {
 832                 jh = JH_ENTRY(list->next);
 833                 bh = jh->bh;
 834                 get_bh(bh);
 835                 if (test_set_buffer_locked(bh)) {
 836                         if (!buffer_dirty(bh)) {
 837                                 list_del_init(&jh->list);
 838                                 list_add(&jh->list, &tmp);
 839                                 goto loop_next;
 840                         }
 841                         spin_unlock(lock);
 842                         if (chunk.nr)
 843                                 write_ordered_chunk(&chunk);
 844                         wait_on_buffer(bh);
 845                         cond_resched();
 846                         spin_lock(lock);
 847                         goto loop_next;
 848                 }
 849                 /* in theory, dirty non-uptodate buffers should never get here,
 850                  * but the upper layer io error paths still have a few quirks.
 851                  * Handle them here as gracefully as we can
 852                  */
 853                 if (!buffer_uptodate(bh) && buffer_dirty(bh)) {
 854                         clear_buffer_dirty(bh);
 855                         ret = -EIO;
 856                 }
 857                 if (buffer_dirty(bh)) {
 858                         list_del_init(&jh->list);
 859                         list_add(&jh->list, &tmp);
 860                         add_to_chunk(&chunk, bh, lock, write_ordered_chunk);
 861                 } else {
 862                         reiserfs_free_jh(bh);
 863                         unlock_buffer(bh);
 864                 }
 865               loop_next:
 866                 put_bh(bh);
 867                 cond_resched_lock(lock);
 868         }
 869         if (chunk.nr) {
 870                 spin_unlock(lock);
 871                 write_ordered_chunk(&chunk);
 872                 spin_lock(lock);
 873         }
 874         while (!list_empty(&tmp)) {
 875                 jh = JH_ENTRY(tmp.prev);
 876                 bh = jh->bh;
 877                 get_bh(bh);
 878                 reiserfs_free_jh(bh);
 879
 880                 if (buffer_locked(bh)) {
 881                         spin_unlock(lock);
 882                         wait_on_buffer(bh);
 883                         spin_lock(lock);
 884                 }
 885                 if (!buffer_uptodate(bh)) {
 886                         ret = -EIO;
 887                 }
 888                 /* ugly interaction with invalidatepage here.
 889                  * reiserfs_invalidate_page will pin any buffer that has a valid
 890                  * journal head from an older transaction.  If someone else sets
 891                  * our buffer dirty after we write it in the first loop, and
 892                  * then someone truncates the page away, nobody will ever write
 893                  * the buffer. We're safe if we write the page one last time
 894                  * after freeing the journal header.
 895                  */
 896                 if (buffer_dirty(bh) && unlikely(bh->b_page->mapping == NULL)) {
 897                         spin_unlock(lock);
 898                         ll_rw_block(WRITE, 1, &bh);
 899                         spin_lock(lock);
 900                 }
 901                 put_bh(bh);
 902                 cond_resched_lock(lock);
 903         }
 904         spin_unlock(lock);
 905         return ret;
 906 }
 907
 908 static int flush_older_commits(struct super_block *s,
 909                                struct reiserfs_journal_list *jl)
 910 {
 911         struct reiserfs_journal *journal = SB_JOURNAL(s);
 912         struct reiserfs_journal_list *other_jl;
 913         struct reiserfs_journal_list *first_jl;
 914         struct list_head *entry;
 915         unsigned long trans_id = jl->j_trans_id;
 916         unsigned long other_trans_id;
 917         unsigned long first_trans_id;
 918
 919       find_first:
 920         /*
 921          * first we walk backwards to find the oldest uncommitted transation
 922          */
 923         first_jl = jl;
 924         entry = jl->j_list.prev;
 925         while (1) {
 926                 other_jl = JOURNAL_LIST_ENTRY(entry);
 927                 if (entry == &journal->j_journal_list ||
 928                     atomic_read(&other_jl->j_older_commits_done))
 929                         break;
 930
 931                 first_jl = other_jl;
 932                 entry = other_jl->j_list.prev;
 933         }
 934
 935         /* if we didn't find any older uncommitted transactions, return now */
 936         if (first_jl == jl) {
 937                 return 0;
 938         }
 939
 940         first_trans_id = first_jl->j_trans_id;
 941
 942         entry = &first_jl->j_list;
 943         while (1) {
 944                 other_jl = JOURNAL_LIST_ENTRY(entry);
 945                 other_trans_id = other_jl->j_trans_id;
 946
 947                 if (other_trans_id < trans_id) {
 948                         if (atomic_read(&other_jl->j_commit_left) != 0) {
 949                                 flush_commit_list(s, other_jl, 0);
 950
 951                                 /* list we were called with is gone, return */
 952                                 if (!journal_list_still_alive(s, trans_id))
 953                                         return 1;
 954
 955                                 /* the one we just flushed is gone, this means all
 956                                  * older lists are also gone, so first_jl is no longer
 957                                  * valid either.  Go back to the beginning.
 958                                  */
 959                                 if (!journal_list_still_alive
 960                                     (s, other_trans_id)) {
 961                                         goto find_first;
 962                                 }
 963                         }
 964                         entry = entry->next;
 965                         if (entry == &journal->j_journal_list)
 966                                 return 0;
 967                 } else {
 968                         return 0;
 969                 }
 970         }
 971         return 0;
 972 }
 973 int reiserfs_async_progress_wait(struct super_block *s)
 974 {
 975         DEFINE_WAIT(wait);
 976         struct reiserfs_journal *j = SB_JOURNAL(s);
 977         if (atomic_read(&j->j_async_throttle))
 978                 blk_congestion_wait(WRITE, HZ / 10);
 979         return 0;
 980 }
 981
 982 /*
 983 ** if this journal list still has commit blocks unflushed, send them to disk.
 984 **
 985 ** log areas must be flushed in order (transaction 2 can't commit before transaction 1)
 986 ** Before the commit block can by written, every other log block must be safely on disk
 987 **
 988 */
 989 static int flush_commit_list(struct super_block *s,
 990                              struct reiserfs_journal_list *jl, int flushall)
 991 {
 992         int i;
 993         int bn;
 994         struct buffer_head *tbh = NULL;
 995         unsigned long trans_id = jl->j_trans_id;
 996         struct reiserfs_journal *journal = SB_JOURNAL(s);
 997         int barrier = 0;
 998         int retval = 0;
 999         int write_len;
1000
1001         reiserfs_check_lock_depth(s, "flush_commit_list");
1002
1003         if (atomic_read(&jl->j_older_commits_done)) {
1004                 return 0;
1005         }
1006
1007         get_fs_excl();
1008
1009         /* before we can put our commit blocks on disk, we have to make sure everyone older than
1010          ** us is on disk too
1011          */
1012         BUG_ON(jl->j_len <= 0);
1013         BUG_ON(trans_id == journal->j_trans_id);
1014
1015         get_journal_list(jl);
1016         if (flushall) {
1017                 if (flush_older_commits(s, jl) == 1) {
1018                         /* list disappeared during flush_older_commits.  return */
1019                         goto put_jl;
1020                 }
1021         }
1022
1023         /* make sure nobody is trying to flush this one at the same time */
1024         down(&jl->j_commit_lock);
1025         if (!journal_list_still_alive(s, trans_id)) {
1026                 up(&jl->j_commit_lock);
1027                 goto put_jl;
1028         }
1029         BUG_ON(jl->j_trans_id == 0);
1030
1031         /* this commit is done, exit */
1032         if (atomic_read(&(jl->j_commit_left)) <= 0) {
1033                 if (flushall) {
1034                         atomic_set(&(jl->j_older_commits_done), 1);
1035                 }
1036                 up(&jl->j_commit_lock);
1037                 goto put_jl;
1038         }
1039
1040         if (!list_empty(&jl->j_bh_list)) {
1041                 int ret;
1042                 unlock_kernel();
1043                 ret = write_ordered_buffers(&journal->j_dirty_buffers_lock,
1044                                             journal, jl, &jl->j_bh_list);
1045                 if (ret < 0 && retval == 0)
1046                         retval = ret;
1047                 lock_kernel();
1048         }
1049         BUG_ON(!list_empty(&jl->j_bh_list));
1050         /*
1051          * for the description block and all the log blocks, submit any buffers
1052          * that haven't already reached the disk.  Try to write at least 256
1053          * log blocks. later on, we will only wait on blocks that correspond
1054          * to this transaction, but while we're unplugging we might as well
1055          * get a chunk of data on there.
1056          */
1057         atomic_inc(&journal->j_async_throttle);
1058         write_len = jl->j_len + 1;
1059         if (write_len < 256)
1060                 write_len = 256;
1061         for (i = 0 ; i < write_len ; i++) {
1062                 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + (jl->j_start + i) %
1063                     SB_ONDISK_JOURNAL_SIZE(s);
1064                 tbh = journal_find_get_block(s, bn);
1065                 if (tbh) {
1066                         if (buffer_dirty(tbh))
1067                             ll_rw_block(WRITE, 1, &tbh) ;
1068                         put_bh(tbh) ;
1069                 }
1070         }
1071         atomic_dec(&journal->j_async_throttle);
1072
1073         /* We're skipping the commit if there's an error */
1074         if (retval || reiserfs_is_journal_aborted(journal))
1075                 barrier = 0;
1076
1077         /* wait on everything written so far before writing the commit
1078          * if we are in barrier mode, send the commit down now
1079          */
1080         barrier = reiserfs_barrier_flush(s);
1081         if (barrier) {
1082                 int ret;
1083                 lock_buffer(jl->j_commit_bh);
1084                 ret = submit_barrier_buffer(jl->j_commit_bh);
1085                 if (ret == -EOPNOTSUPP) {
1086                         set_buffer_uptodate(jl->j_commit_bh);
1087                         disable_barrier(s);
1088                         barrier = 0;
1089                 }
1090         }
1091         for (i = 0; i < (jl->j_len + 1); i++) {
1092                 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) +
1093                     (jl->j_start + i) % SB_ONDISK_JOURNAL_SIZE(s);
1094                 tbh = journal_find_get_block(s, bn);
1095                 wait_on_buffer(tbh);
1096                 // since we're using ll_rw_blk above, it might have skipped over
1097                 // a locked buffer.  Double check here
1098                 //
1099                 if (buffer_dirty(tbh))  /* redundant, sync_dirty_buffer() checks */
1100                         sync_dirty_buffer(tbh);
1101                 if (unlikely(!buffer_uptodate(tbh))) {
1102 #ifdef CONFIG_REISERFS_CHECK
1103                         reiserfs_warning(s, "journal-601, buffer write failed");
1104 #endif
1105                         retval = -EIO;
1106                 }
1107                 put_bh(tbh);    /* once for journal_find_get_block */
1108                 put_bh(tbh);    /* once due to original getblk in do_journal_end */
1109                 atomic_dec(&(jl->j_commit_left));
1110         }
1111
1112         BUG_ON(atomic_read(&(jl->j_commit_left)) != 1);
1113
1114         if (!barrier) {
1115                 /* If there was a write error in the journal - we can't commit
1116                  * this transaction - it will be invalid and, if successful,
1117                  * will just end up propogating the write error out to
1118                  * the file system. */
1119                 if (likely(!retval && !reiserfs_is_journal_aborted (journal))) {
1120                         if (buffer_dirty(jl->j_commit_bh))
1121                                 BUG();
1122                         mark_buffer_dirty(jl->j_commit_bh) ;
1123                         sync_dirty_buffer(jl->j_commit_bh) ;
1124                 }
1125         } else
1126                 wait_on_buffer(jl->j_commit_bh);
1127
1128         check_barrier_completion(s, jl->j_commit_bh);
1129
1130         /* If there was a write error in the journal - we can't commit this
1131          * transaction - it will be invalid and, if successful, will just end
1132          * up propogating the write error out to the filesystem. */
1133         if (unlikely(!buffer_uptodate(jl->j_commit_bh))) {
1134 #ifdef CONFIG_REISERFS_CHECK
1135                 reiserfs_warning(s, "journal-615: buffer write failed");
1136 #endif
1137                 retval = -EIO;
1138         }
1139         bforget(jl->j_commit_bh);
1140         if (journal->j_last_commit_id != 0 &&
1141             (jl->j_trans_id - journal->j_last_commit_id) != 1) {
1142                 reiserfs_warning(s, "clm-2200: last commit %lu, current %lu",
1143                                  journal->j_last_commit_id, jl->j_trans_id);
1144         }
1145         journal->j_last_commit_id = jl->j_trans_id;
1146
1147         /* now, every commit block is on the disk.  It is safe to allow blocks freed during this transaction to be reallocated */
1148         cleanup_freed_for_journal_list(s, jl);
1149
1150         retval = retval ? retval : journal->j_errno;
1151
1152         /* mark the metadata dirty */
1153         if (!retval)
1154                 dirty_one_transaction(s, jl);
1155         atomic_dec(&(jl->j_commit_left));
1156
1157         if (flushall) {
1158                 atomic_set(&(jl->j_older_commits_done), 1);
1159         }
1160         up(&jl->j_commit_lock);
1161       put_jl:
1162         put_journal_list(s, jl);
1163
1164         if (retval)
1165                 reiserfs_abort(s, retval, "Journal write error in %s",
1166                                __FUNCTION__);
1167         put_fs_excl();
1168         return retval;
1169 }
1170
1171 /*
1172 ** flush_journal_list frequently needs to find a newer transaction for a given block.  This does that, or
1173 ** returns NULL if it can't find anything
1174 */
1175 static struct reiserfs_journal_list *find_newer_jl_for_cn(struct
1176                                                           reiserfs_journal_cnode
1177                                                           *cn)
1178 {
1179         struct super_block *sb = cn->sb;
1180         b_blocknr_t blocknr = cn->blocknr;
1181
1182         cn = cn->hprev;
1183         while (cn) {
1184                 if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist) {
1185                         return cn->jlist;
1186                 }
1187                 cn = cn->hprev;
1188         }
1189         return NULL;
1190 }
1191
1192 static void remove_journal_hash(struct super_block *,
1193                                 struct reiserfs_journal_cnode **,
1194                                 struct reiserfs_journal_list *, unsigned long,
1195                                 int);
1196
1197 /*
1198 ** once all the real blocks have been flushed, it is safe to remove them from the
1199 ** journal list for this transaction.  Aside from freeing the cnode, this also allows the
1200 ** block to be reallocated for data blocks if it had been deleted.
1201 */
1202 static void remove_all_from_journal_list(struct super_block *p_s_sb,
1203                                          struct reiserfs_journal_list *jl,
1204                                          int debug)
1205 {
1206         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1207         struct reiserfs_journal_cnode *cn, *last;
1208         cn = jl->j_realblock;
1209
1210         /* which is better, to lock once around the whole loop, or
1211          ** to lock for each call to remove_journal_hash?
1212          */
1213         while (cn) {
1214                 if (cn->blocknr != 0) {
1215                         if (debug) {
1216                                 reiserfs_warning(p_s_sb,
1217                                                  "block %u, bh is %d, state %ld",
1218                                                  cn->blocknr, cn->bh ? 1 : 0,
1219                                                  cn->state);
1220                         }
1221                         cn->state = 0;
1222                         remove_journal_hash(p_s_sb, journal->j_list_hash_table,
1223                                             jl, cn->blocknr, 1);
1224                 }
1225                 last = cn;
1226                 cn = cn->next;
1227                 free_cnode(p_s_sb, last);
1228         }
1229         jl->j_realblock = NULL;
1230 }
1231
1232 /*
1233 ** if this timestamp is greater than the timestamp we wrote last to the header block, write it to the header block.
1234 ** once this is done, I can safely say the log area for this transaction won't ever be replayed, and I can start
1235 ** releasing blocks in this transaction for reuse as data blocks.
1236 ** called by flush_journal_list, before it calls remove_all_from_journal_list
1237 **
1238 */
1239 static int _update_journal_header_block(struct super_block *p_s_sb,
1240                                         unsigned long offset,
1241                                         unsigned long trans_id)
1242 {
1243         struct reiserfs_journal_header *jh;
1244         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1245
1246         if (reiserfs_is_journal_aborted(journal))
1247                 return -EIO;
1248
1249         if (trans_id >= journal->j_last_flush_trans_id) {
1250                 if (buffer_locked((journal->j_header_bh))) {
1251                         wait_on_buffer((journal->j_header_bh));
1252                         if (unlikely(!buffer_uptodate(journal->j_header_bh))) {
1253 #ifdef CONFIG_REISERFS_CHECK
1254                                 reiserfs_warning(p_s_sb,
1255                                                  "journal-699: buffer write failed");
1256 #endif
1257                                 return -EIO;
1258                         }
1259                 }
1260                 journal->j_last_flush_trans_id = trans_id;
1261                 journal->j_first_unflushed_offset = offset;
1262                 jh = (struct reiserfs_journal_header *)(journal->j_header_bh->
1263                                                         b_data);
1264                 jh->j_last_flush_trans_id = cpu_to_le32(trans_id);
1265                 jh->j_first_unflushed_offset = cpu_to_le32(offset);
1266                 jh->j_mount_id = cpu_to_le32(journal->j_mount_id);
1267
1268                 if (reiserfs_barrier_flush(p_s_sb)) {
1269                         int ret;
1270                         lock_buffer(journal->j_header_bh);
1271                         ret = submit_barrier_buffer(journal->j_header_bh);
1272                         if (ret == -EOPNOTSUPP) {
1273                                 set_buffer_uptodate(journal->j_header_bh);
1274                                 disable_barrier(p_s_sb);
1275                                 goto sync;
1276                         }
1277                         wait_on_buffer(journal->j_header_bh);
1278                         check_barrier_completion(p_s_sb, journal->j_header_bh);
1279                 } else {
1280                       sync:
1281                         set_buffer_dirty(journal->j_header_bh);
1282                         sync_dirty_buffer(journal->j_header_bh);
1283                 }
1284                 if (!buffer_uptodate(journal->j_header_bh)) {
1285                         reiserfs_warning(p_s_sb,
1286                                          "journal-837: IO error during journal replay");
1287                         return -EIO;
1288                 }
1289         }
1290         return 0;
1291 }
1292
1293 static int update_journal_header_block(struct super_block *p_s_sb,
1294                                        unsigned long offset,
1295                                        unsigned long trans_id)
1296 {
1297         return _update_journal_header_block(p_s_sb, offset, trans_id);
1298 }
1299
1300 /*
1301 ** flush any and all journal lists older than you are
1302 ** can only be called from flush_journal_list
1303 */
1304 static int flush_older_journal_lists(struct super_block *p_s_sb,
1305                                      struct reiserfs_journal_list *jl)
1306 {
1307         struct list_head *entry;
1308         struct reiserfs_journal_list *other_jl;
1309         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1310         unsigned long trans_id = jl->j_trans_id;
1311
1312         /* we know we are the only ones flushing things, no extra race
1313          * protection is required.
1314          */
1315       restart:
1316         entry = journal->j_journal_list.next;
1317         /* Did we wrap? */
1318         if (entry == &journal->j_journal_list)
1319                 return 0;
1320         other_jl = JOURNAL_LIST_ENTRY(entry);
1321         if (other_jl->j_trans_id < trans_id) {
1322                 BUG_ON(other_jl->j_refcount <= 0);
1323                 /* do not flush all */
1324                 flush_journal_list(p_s_sb, other_jl, 0);
1325
1326                 /* other_jl is now deleted from the list */
1327                 goto restart;
1328         }
1329         return 0;
1330 }
1331
1332 static void del_from_work_list(struct super_block *s,
1333                                struct reiserfs_journal_list *jl)
1334 {
1335         struct reiserfs_journal *journal = SB_JOURNAL(s);
1336         if (!list_empty(&jl->j_working_list)) {
1337                 list_del_init(&jl->j_working_list);
1338                 journal->j_num_work_lists--;
1339         }
1340 }
1341
1342 /* flush a journal list, both commit and real blocks
1343 **
1344 ** always set flushall to 1, unless you are calling from inside
1345 ** flush_journal_list
1346 **
1347 ** IMPORTANT.  This can only be called while there are no journal writers,
1348 ** and the journal is locked.  That means it can only be called from
1349 ** do_journal_end, or by journal_release
1350 */
1351 static int flush_journal_list(struct super_block *s,
1352                               struct reiserfs_journal_list *jl, int flushall)
1353 {
1354         struct reiserfs_journal_list *pjl;
1355         struct reiserfs_journal_cnode *cn, *last;
1356         int count;
1357         int was_jwait = 0;
1358         int was_dirty = 0;
1359         struct buffer_head *saved_bh;
1360         unsigned long j_len_saved = jl->j_len;
1361         struct reiserfs_journal *journal = SB_JOURNAL(s);
1362         int err = 0;
1363
1364         BUG_ON(j_len_saved <= 0);
1365
1366         if (atomic_read(&journal->j_wcount) != 0) {
1367                 reiserfs_warning(s,
1368                                  "clm-2048: flush_journal_list called with wcount %d",
1369                                  atomic_read(&journal->j_wcount));
1370         }
1371         BUG_ON(jl->j_trans_id == 0);
1372
1373         /* if flushall == 0, the lock is already held */
1374         if (flushall) {
1375                 down(&journal->j_flush_sem);
1376         } else if (!down_trylock(&journal->j_flush_sem)) {
1377                 BUG();
1378         }
1379
1380         count = 0;
1381         if (j_len_saved > journal->j_trans_max) {
1382                 reiserfs_panic(s,
1383                                "journal-715: flush_journal_list, length is %lu, trans id %lu\n",
1384                                j_len_saved, jl->j_trans_id);
1385                 return 0;
1386         }
1387
1388         get_fs_excl();
1389
1390         /* if all the work is already done, get out of here */
1391         if (atomic_read(&(jl->j_nonzerolen)) <= 0 &&
1392             atomic_read(&(jl->j_commit_left)) <= 0) {
1393                 goto flush_older_and_return;
1394         }
1395
1396         /* start by putting the commit list on disk.  This will also flush
1397          ** the commit lists of any olders transactions
1398          */
1399         flush_commit_list(s, jl, 1);
1400
1401         if (!(jl->j_state & LIST_DIRTY)
1402             && !reiserfs_is_journal_aborted(journal))
1403                 BUG();
1404
1405         /* are we done now? */
1406         if (atomic_read(&(jl->j_nonzerolen)) <= 0 &&
1407             atomic_read(&(jl->j_commit_left)) <= 0) {
1408                 goto flush_older_and_return;
1409         }
1410
1411         /* loop through each cnode, see if we need to write it,
1412          ** or wait on a more recent transaction, or just ignore it
1413          */
1414         if (atomic_read(&(journal->j_wcount)) != 0) {
1415                 reiserfs_panic(s,
1416                                "journal-844: panic journal list is flushing, wcount is not 0\n");
1417         }
1418         cn = jl->j_realblock;
1419         while (cn) {
1420                 was_jwait = 0;
1421                 was_dirty = 0;
1422                 saved_bh = NULL;
1423                 /* blocknr of 0 is no longer in the hash, ignore it */
1424                 if (cn->blocknr == 0) {
1425                         goto free_cnode;
1426                 }
1427
1428                 /* This transaction failed commit. Don't write out to the disk */
1429                 if (!(jl->j_state & LIST_DIRTY))
1430                         goto free_cnode;
1431
1432                 pjl = find_newer_jl_for_cn(cn);
1433                 /* the order is important here.  We check pjl to make sure we
1434                  ** don't clear BH_JDirty_wait if we aren't the one writing this
1435                  ** block to disk
1436                  */
1437                 if (!pjl && cn->bh) {
1438                         saved_bh = cn->bh;
1439
1440                         /* we do this to make sure nobody releases the buffer while
1441                          ** we are working with it
1442                          */
1443                         get_bh(saved_bh);
1444
1445                         if (buffer_journal_dirty(saved_bh)) {
1446                                 BUG_ON(!can_dirty(cn));
1447                                 was_jwait = 1;
1448                                 was_dirty = 1;
1449                         } else if (can_dirty(cn)) {
1450                                 /* everything with !pjl && jwait should be writable */
1451                                 BUG();
1452                         }
1453                 }
1454
1455                 /* if someone has this block in a newer transaction, just make
1456                  ** sure they are commited, and don't try writing it to disk
1457                  */
1458                 if (pjl) {
1459                         if (atomic_read(&pjl->j_commit_left))
1460                                 flush_commit_list(s, pjl, 1);
1461                         goto free_cnode;
1462                 }
1463
1464                 /* bh == NULL when the block got to disk on its own, OR,
1465                  ** the block got freed in a future transaction
1466                  */
1467                 if (saved_bh == NULL) {
1468                         goto free_cnode;
1469                 }
1470
1471                 /* this should never happen.  kupdate_one_transaction has this list
1472                  ** locked while it works, so we should never see a buffer here that
1473                  ** is not marked JDirty_wait
1474                  */
1475                 if ((!was_jwait) && !buffer_locked(saved_bh)) {
1476                         reiserfs_warning(s,
1477                                          "journal-813: BAD! buffer %llu %cdirty %cjwait, "
1478                                          "not in a newer tranasction",
1479                                          (unsigned long long)saved_bh->
1480                                          b_blocknr, was_dirty ? ' ' : '!',
1481                                          was_jwait ? ' ' : '!');
1482                 }
1483                 if (was_dirty) {
1484                         /* we inc again because saved_bh gets decremented at free_cnode */
1485                         get_bh(saved_bh);
1486                         set_bit(BLOCK_NEEDS_FLUSH, &cn->state);
1487                         lock_buffer(saved_bh);
1488                         BUG_ON(cn->blocknr != saved_bh->b_blocknr);
1489                         if (buffer_dirty(saved_bh))
1490                                 submit_logged_buffer(saved_bh);
1491                         else
1492                                 unlock_buffer(saved_bh);
1493                         count++;
1494                 } else {
1495                         reiserfs_warning(s,
1496                                          "clm-2082: Unable to flush buffer %llu in %s",
1497                                          (unsigned long long)saved_bh->
1498                                          b_blocknr, __FUNCTION__);
1499                 }
1500               free_cnode:
1501                 last = cn;
1502                 cn = cn->next;
1503                 if (saved_bh) {
1504                         /* we incremented this to keep others from taking the buffer head away */
1505                         put_bh(saved_bh);
1506                         if (atomic_read(&(saved_bh->b_count)) < 0) {
1507                                 reiserfs_warning(s,
1508                                                  "journal-945: saved_bh->b_count < 0");
1509                         }
1510                 }
1511         }
1512         if (count > 0) {
1513                 cn = jl->j_realblock;
1514                 while (cn) {
1515                         if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) {
1516                                 if (!cn->bh) {
1517                                         reiserfs_panic(s,
1518                                                        "journal-1011: cn->bh is NULL\n");
1519                                 }
1520                                 wait_on_buffer(cn->bh);
1521                                 if (!cn->bh) {
1522                                         reiserfs_panic(s,
1523                                                        "journal-1012: cn->bh is NULL\n");
1524                                 }
1525                                 if (unlikely(!buffer_uptodate(cn->bh))) {
1526 #ifdef CONFIG_REISERFS_CHECK
1527                                         reiserfs_warning(s,
1528                                                          "journal-949: buffer write failed\n");
1529 #endif
1530                                         err = -EIO;
1531                                 }
1532                                 /* note, we must clear the JDirty_wait bit after the up to date
1533                                  ** check, otherwise we race against our flushpage routine
1534                                  */
1535                                 BUG_ON(!test_clear_buffer_journal_dirty
1536                                        (cn->bh));
1537
1538                                 /* undo the inc from journal_mark_dirty */
1539                                 put_bh(cn->bh);
1540                                 brelse(cn->bh);
1541                         }
1542                         cn = cn->next;
1543                 }
1544         }
1545
1546         if (err)
1547                 reiserfs_abort(s, -EIO,
1548                                "Write error while pushing transaction to disk in %s",
1549                                __FUNCTION__);
1550       flush_older_and_return:
1551
1552         /* before we can update the journal header block, we _must_ flush all
1553          ** real blocks from all older transactions to disk.  This is because
1554          ** once the header block is updated, this transaction will not be
1555          ** replayed after a crash
1556          */
1557         if (flushall) {
1558                 flush_older_journal_lists(s, jl);
1559         }
1560
1561         err = journal->j_errno;
1562         /* before we can remove everything from the hash tables for this
1563          ** transaction, we must make sure it can never be replayed
1564          **
1565          ** since we are only called from do_journal_end, we know for sure there
1566          ** are no allocations going on while we are flushing journal lists.  So,
1567          ** we only need to update the journal header block for the last list
1568          ** being flushed
1569          */
1570         if (!err && flushall) {
1571                 err =
1572                     update_journal_header_block(s,
1573                                                 (jl->j_start + jl->j_len +
1574                                                  2) % SB_ONDISK_JOURNAL_SIZE(s),
1575                                                 jl->j_trans_id);
1576                 if (err)
1577                         reiserfs_abort(s, -EIO,
1578                                        "Write error while updating journal header in %s",
1579                                        __FUNCTION__);
1580         }
1581         remove_all_from_journal_list(s, jl, 0);
1582         list_del_init(&jl->j_list);
1583         journal->j_num_lists--;
1584         del_from_work_list(s, jl);
1585
1586         if (journal->j_last_flush_id != 0 &&
1587             (jl->j_trans_id - journal->j_last_flush_id) != 1) {
1588                 reiserfs_warning(s, "clm-2201: last flush %lu, current %lu",
1589                                  journal->j_last_flush_id, jl->j_trans_id);
1590         }
1591         journal->j_last_flush_id = jl->j_trans_id;
1592
1593         /* not strictly required since we are freeing the list, but it should
1594          * help find code using dead lists later on
1595          */
1596         jl->j_len = 0;
1597         atomic_set(&(jl->j_nonzerolen), 0);
1598         jl->j_start = 0;
1599         jl->j_realblock = NULL;
1600         jl->j_commit_bh = NULL;
1601         jl->j_trans_id = 0;
1602         jl->j_state = 0;
1603         put_journal_list(s, jl);
1604         if (flushall)
1605                 up(&journal->j_flush_sem);
1606         put_fs_excl();
1607         return err;
1608 }
1609
1610 static int write_one_transaction(struct super_block *s,
1611                                  struct reiserfs_journal_list *jl,
1612                                  struct buffer_chunk *chunk)
1613 {
1614         struct reiserfs_journal_cnode *cn;
1615         int ret = 0;
1616
1617         jl->j_state |= LIST_TOUCHED;
1618         del_from_work_list(s, jl);
1619         if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0) {
1620                 return 0;
1621         }
1622
1623         cn = jl->j_realblock;
1624         while (cn) {
1625                 /* if the blocknr == 0, this has been cleared from the hash,
1626                  ** skip it
1627                  */
1628                 if (cn->blocknr == 0) {
1629                         goto next;
1630                 }
1631                 if (cn->bh && can_dirty(cn) && buffer_dirty(cn->bh)) {
1632                         struct buffer_head *tmp_bh;
1633                         /* we can race against journal_mark_freed when we try
1634                          * to lock_buffer(cn->bh), so we have to inc the buffer
1635                          * count, and recheck things after locking
1636                          */
1637                         tmp_bh = cn->bh;
1638                         get_bh(tmp_bh);
1639                         lock_buffer(tmp_bh);
1640                         if (cn->bh && can_dirty(cn) && buffer_dirty(tmp_bh)) {
1641                                 if (!buffer_journal_dirty(tmp_bh) ||
1642                                     buffer_journal_prepared(tmp_bh))
1643                                         BUG();
1644                                 add_to_chunk(chunk, tmp_bh, NULL, write_chunk);
1645                                 ret++;
1646                         } else {
1647                                 /* note, cn->bh might be null now */
1648                                 unlock_buffer(tmp_bh);
1649                         }
1650                         put_bh(tmp_bh);
1651                 }
1652               next:
1653                 cn = cn->next;
1654                 cond_resched();
1655         }
1656         return ret;
1657 }
1658
1659 /* used by flush_commit_list */
1660 static int dirty_one_transaction(struct super_block *s,
1661                                  struct reiserfs_journal_list *jl)
1662 {
1663         struct reiserfs_journal_cnode *cn;
1664         struct reiserfs_journal_list *pjl;
1665         int ret = 0;
1666
1667         jl->j_state |= LIST_DIRTY;
1668         cn = jl->j_realblock;
1669         while (cn) {
1670                 /* look for a more recent transaction that logged this
1671                  ** buffer.  Only the most recent transaction with a buffer in
1672                  ** it is allowed to send that buffer to disk
1673                  */
1674                 pjl = find_newer_jl_for_cn(cn);
1675                 if (!pjl && cn->blocknr && cn->bh
1676                     && buffer_journal_dirty(cn->bh)) {
1677                         BUG_ON(!can_dirty(cn));
1678                         /* if the buffer is prepared, it will either be logged
1679                          * or restored.  If restored, we need to make sure
1680                          * it actually gets marked dirty
1681                          */
1682                         clear_buffer_journal_new(cn->bh);
1683                         if (buffer_journal_prepared(cn->bh)) {
1684                                 set_buffer_journal_restore_dirty(cn->bh);
1685                         } else {
1686                                 set_buffer_journal_test(cn->bh);
1687                                 mark_buffer_dirty(cn->bh);
1688                         }
1689                 }
1690                 cn = cn->next;
1691         }
1692         return ret;
1693 }
1694
1695 static int kupdate_transactions(struct super_block *s,
1696                                 struct reiserfs_journal_list *jl,
1697                                 struct reiserfs_journal_list **next_jl,
1698                                 unsigned long *next_trans_id,
1699                                 int num_blocks, int num_trans)
1700 {
1701         int ret = 0;
1702         int written = 0;
1703         int transactions_flushed = 0;
1704         unsigned long orig_trans_id = jl->j_trans_id;
1705         struct buffer_chunk chunk;
1706         struct list_head *entry;
1707         struct reiserfs_journal *journal = SB_JOURNAL(s);
1708         chunk.nr = 0;
1709
1710         down(&journal->j_flush_sem);
1711         if (!journal_list_still_alive(s, orig_trans_id)) {
1712                 goto done;
1713         }
1714
1715         /* we've got j_flush_sem held, nobody is going to delete any
1716          * of these lists out from underneath us
1717          */
1718         while ((num_trans && transactions_flushed < num_trans) ||
1719                (!num_trans && written < num_blocks)) {
1720
1721                 if (jl->j_len == 0 || (jl->j_state & LIST_TOUCHED) ||
1722                     atomic_read(&jl->j_commit_left)
1723                     || !(jl->j_state & LIST_DIRTY)) {
1724                         del_from_work_list(s, jl);
1725                         break;
1726                 }
1727                 ret = write_one_transaction(s, jl, &chunk);
1728
1729                 if (ret < 0)
1730                         goto done;
1731                 transactions_flushed++;
1732                 written += ret;
1733                 entry = jl->j_list.next;
1734
1735                 /* did we wrap? */
1736                 if (entry == &journal->j_journal_list) {
1737                         break;
1738                 }
1739                 jl = JOURNAL_LIST_ENTRY(entry);
1740
1741                 /* don't bother with older transactions */
1742                 if (jl->j_trans_id <= orig_trans_id)
1743                         break;
1744         }
1745         if (chunk.nr) {
1746                 write_chunk(&chunk);
1747         }
1748
1749       done:
1750         up(&journal->j_flush_sem);
1751         return ret;
1752 }
1753
1754 /* for o_sync and fsync heavy applications, they tend to use
1755 ** all the journa list slots with tiny transactions.  These
1756 ** trigger lots and lots of calls to update the header block, which
1757 ** adds seeks and slows things down.
1758 **
1759 ** This function tries to clear out a large chunk of the journal lists
1760 ** at once, which makes everything faster since only the newest journal
1761 ** list updates the header block
1762 */
1763 static int flush_used_journal_lists(struct super_block *s,
1764                                     struct reiserfs_journal_list *jl)
1765 {
1766         unsigned long len = 0;
1767         unsigned long cur_len;
1768         int ret;
1769         int i;
1770         int limit = 256;
1771         struct reiserfs_journal_list *tjl;
1772         struct reiserfs_journal_list *flush_jl;
1773         unsigned long trans_id;
1774         struct reiserfs_journal *journal = SB_JOURNAL(s);
1775
1776         flush_jl = tjl = jl;
1777
1778         /* in data logging mode, try harder to flush a lot of blocks */
1779         if (reiserfs_data_log(s))
1780                 limit = 1024;
1781         /* flush for 256 transactions or limit blocks, whichever comes first */
1782         for (i = 0; i < 256 && len < limit; i++) {
1783                 if (atomic_read(&tjl->j_commit_left) ||
1784                     tjl->j_trans_id < jl->j_trans_id) {
1785                         break;
1786                 }
1787                 cur_len = atomic_read(&tjl->j_nonzerolen);
1788                 if (cur_len > 0) {
1789                         tjl->j_state &= ~LIST_TOUCHED;
1790                 }
1791                 len += cur_len;
1792                 flush_jl = tjl;
1793                 if (tjl->j_list.next == &journal->j_journal_list)
1794                         break;
1795                 tjl = JOURNAL_LIST_ENTRY(tjl->j_list.next);
1796         }
1797         /* try to find a group of blocks we can flush across all the
1798          ** transactions, but only bother if we've actually spanned
1799          ** across multiple lists
1800          */
1801         if (flush_jl != jl) {
1802                 ret = kupdate_transactions(s, jl, &tjl, &trans_id, len, i);
1803         }
1804         flush_journal_list(s, flush_jl, 1);
1805         return 0;
1806 }
1807
1808 /*
1809 ** removes any nodes in table with name block and dev as bh.
1810 ** only touchs the hnext and hprev pointers.
1811 */
1812 void remove_journal_hash(struct super_block *sb,
1813                          struct reiserfs_journal_cnode **table,
1814                          struct reiserfs_journal_list *jl,
1815                          unsigned long block, int remove_freed)
1816 {
1817         struct reiserfs_journal_cnode *cur;
1818         struct reiserfs_journal_cnode **head;
1819
1820         head = &(journal_hash(table, sb, block));
1821         if (!head) {
1822                 return;
1823         }
1824         cur = *head;
1825         while (cur) {
1826                 if (cur->blocknr == block && cur->sb == sb
1827                     && (jl == NULL || jl == cur->jlist)
1828                     && (!test_bit(BLOCK_FREED, &cur->state) || remove_freed)) {
1829                         if (cur->hnext) {
1830                                 cur->hnext->hprev = cur->hprev;
1831                         }
1832                         if (cur->hprev) {
1833                                 cur->hprev->hnext = cur->hnext;
1834                         } else {
1835                                 *head = cur->hnext;
1836                         }
1837                         cur->blocknr = 0;
1838                         cur->sb = NULL;
1839                         cur->state = 0;
1840                         if (cur->bh && cur->jlist)      /* anybody who clears the cur->bh will also dec the nonzerolen */
1841                                 atomic_dec(&(cur->jlist->j_nonzerolen));
1842                         cur->bh = NULL;
1843                         cur->jlist = NULL;
1844                 }
1845                 cur = cur->hnext;
1846         }
1847 }
1848
1849 static void free_journal_ram(struct super_block *p_s_sb)
1850 {
1851         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1852         kfree(journal->j_current_jl);
1853         journal->j_num_lists--;
1854
1855         vfree(journal->j_cnode_free_orig);
1856         free_list_bitmaps(p_s_sb, journal->j_list_bitmap);
1857         free_bitmap_nodes(p_s_sb);      /* must be after free_list_bitmaps */
1858         if (journal->j_header_bh) {
1859                 brelse(journal->j_header_bh);
1860         }
1861         /* j_header_bh is on the journal dev, make sure not to release the journal
1862          * dev until we brelse j_header_bh
1863          */
1864         release_journal_dev(p_s_sb, journal);
1865         vfree(journal);
1866 }
1867
1868 /*
1869 ** call on unmount.  Only set error to 1 if you haven't made your way out
1870 ** of read_super() yet.  Any other caller must keep error at 0.
1871 */
1872 static int do_journal_release(struct reiserfs_transaction_handle *th,
1873                               struct super_block *p_s_sb, int error)
1874 {
1875         struct reiserfs_transaction_handle myth;
1876         int flushed = 0;
1877         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1878
1879         /* we only want to flush out transactions if we were called with error == 0
1880          */
1881         if (!error && !(p_s_sb->s_flags & MS_RDONLY)) {
1882                 /* end the current trans */
1883                 BUG_ON(!th->t_trans_id);
1884                 do_journal_end(th, p_s_sb, 10, FLUSH_ALL);
1885
1886                 /* make sure something gets logged to force our way into the flush code */
1887                 if (!journal_join(&myth, p_s_sb, 1)) {
1888                         reiserfs_prepare_for_journal(p_s_sb,
1889                                                      SB_BUFFER_WITH_SB(p_s_sb),
1890                                                      1);
1891                         journal_mark_dirty(&myth, p_s_sb,
1892                                            SB_BUFFER_WITH_SB(p_s_sb));
1893                         do_journal_end(&myth, p_s_sb, 1, FLUSH_ALL);
1894                         flushed = 1;
1895                 }
1896         }
1897
1898         /* this also catches errors during the do_journal_end above */
1899         if (!error && reiserfs_is_journal_aborted(journal)) {
1900                 memset(&myth, 0, sizeof(myth));
1901                 if (!journal_join_abort(&myth, p_s_sb, 1)) {
1902                         reiserfs_prepare_for_journal(p_s_sb,
1903                                                      SB_BUFFER_WITH_SB(p_s_sb),
1904                                                      1);
1905                         journal_mark_dirty(&myth, p_s_sb,
1906                                            SB_BUFFER_WITH_SB(p_s_sb));
1907                         do_journal_end(&myth, p_s_sb, 1, FLUSH_ALL);
1908                 }
1909         }
1910
1911         reiserfs_mounted_fs_count--;
1912         /* wait for all commits to finish */
1913         cancel_delayed_work(&SB_JOURNAL(p_s_sb)->j_work);
1914         flush_workqueue(commit_wq);
1915         if (!reiserfs_mounted_fs_count) {
1916                 destroy_workqueue(commit_wq);
1917                 commit_wq = NULL;
1918         }
1919
1920         free_journal_ram(p_s_sb);
1921
1922         return 0;
1923 }
1924
1925 /*
1926 ** call on unmount.  flush all journal trans, release all alloc'd ram
1927 */
1928 int journal_release(struct reiserfs_transaction_handle *th,
1929                     struct super_block *p_s_sb)
1930 {
1931         return do_journal_release(th, p_s_sb, 0);
1932 }
1933
1934 /*
1935 ** only call from an error condition inside reiserfs_read_super!
1936 */
1937 int journal_release_error(struct reiserfs_transaction_handle *th,
1938                           struct super_block *p_s_sb)
1939 {
1940         return do_journal_release(th, p_s_sb, 1);
1941 }
1942
1943 /* compares description block with commit block.  returns 1 if they differ, 0 if they are the same */
1944 static int journal_compare_desc_commit(struct super_block *p_s_sb,
1945                                        struct reiserfs_journal_desc *desc,
1946                                        struct reiserfs_journal_commit *commit)
1947 {
1948         if (get_commit_trans_id(commit) != get_desc_trans_id(desc) ||
1949             get_commit_trans_len(commit) != get_desc_trans_len(desc) ||
1950             get_commit_trans_len(commit) > SB_JOURNAL(p_s_sb)->j_trans_max ||
1951             get_commit_trans_len(commit) <= 0) {
1952                 return 1;
1953         }
1954         return 0;
1955 }
1956
1957 /* returns 0 if it did not find a description block
1958 ** returns -1 if it found a corrupt commit block
1959 ** returns 1 if both desc and commit were valid
1960 */
1961 static int journal_transaction_is_valid(struct super_block *p_s_sb,
1962                                         struct buffer_head *d_bh,
1963                                         unsigned long *oldest_invalid_trans_id,
1964                                         unsigned long *newest_mount_id)
1965 {
1966         struct reiserfs_journal_desc *desc;
1967         struct reiserfs_journal_commit *commit;
1968         struct buffer_head *c_bh;
1969         unsigned long offset;
1970
1971         if (!d_bh)
1972                 return 0;
1973
1974         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
1975         if (get_desc_trans_len(desc) > 0
1976             && !memcmp(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8)) {
1977                 if (oldest_invalid_trans_id && *oldest_invalid_trans_id
1978                     && get_desc_trans_id(desc) > *oldest_invalid_trans_id) {
1979                         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
1980                                        "journal-986: transaction "
1981                                        "is valid returning because trans_id %d is greater than "
1982                                        "oldest_invalid %lu",
1983                                        get_desc_trans_id(desc),
1984                                        *oldest_invalid_trans_id);
1985                         return 0;
1986                 }
1987                 if (newest_mount_id
1988                     && *newest_mount_id > get_desc_mount_id(desc)) {
1989                         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
1990                                        "journal-1087: transaction "
1991                                        "is valid returning because mount_id %d is less than "
1992                                        "newest_mount_id %lu",
1993                                        get_desc_mount_id(desc),
1994                                        *newest_mount_id);
1995                         return -1;
1996                 }
1997                 if (get_desc_trans_len(desc) > SB_JOURNAL(p_s_sb)->j_trans_max) {
1998                         reiserfs_warning(p_s_sb,
1999                                          "journal-2018: Bad transaction length %d encountered, ignoring transaction",
2000                                          get_desc_trans_len(desc));
2001                         return -1;
2002                 }
2003                 offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb);
2004
2005                 /* ok, we have a journal description block, lets see if the transaction was valid */
2006                 c_bh =
2007                     journal_bread(p_s_sb,
2008                                   SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2009                                   ((offset + get_desc_trans_len(desc) +
2010                                     1) % SB_ONDISK_JOURNAL_SIZE(p_s_sb)));
2011                 if (!c_bh)
2012                         return 0;
2013                 commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2014                 if (journal_compare_desc_commit(p_s_sb, desc, commit)) {
2015                         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2016                                        "journal_transaction_is_valid, commit offset %ld had bad "
2017                                        "time %d or length %d",
2018                                        c_bh->b_blocknr -
2019                                        SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2020                                        get_commit_trans_id(commit),
2021                                        get_commit_trans_len(commit));
2022                         brelse(c_bh);
2023                         if (oldest_invalid_trans_id) {
2024                                 *oldest_invalid_trans_id =
2025                                     get_desc_trans_id(desc);
2026                                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2027                                                "journal-1004: "
2028                                                "transaction_is_valid setting oldest invalid trans_id "
2029                                                "to %d",
2030                                                get_desc_trans_id(desc));
2031                         }
2032                         return -1;
2033                 }
2034                 brelse(c_bh);
2035                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2036                                "journal-1006: found valid "
2037                                "transaction start offset %llu, len %d id %d",
2038                                d_bh->b_blocknr -
2039                                SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2040                                get_desc_trans_len(desc),
2041                                get_desc_trans_id(desc));
2042                 return 1;
2043         } else {
2044                 return 0;
2045         }
2046 }
2047
2048 static void brelse_array(struct buffer_head **heads, int num)
2049 {
2050         int i;
2051         for (i = 0; i < num; i++) {
2052                 brelse(heads[i]);
2053         }
2054 }
2055
2056 /*
2057 ** given the start, and values for the oldest acceptable transactions,
2058 ** this either reads in a replays a transaction, or returns because the transaction
2059 ** is invalid, or too old.
2060 */
2061 static int journal_read_transaction(struct super_block *p_s_sb,
2062                                     unsigned long cur_dblock,
2063                                     unsigned long oldest_start,
2064                                     unsigned long oldest_trans_id,
2065                                     unsigned long newest_mount_id)
2066 {
2067         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
2068         struct reiserfs_journal_desc *desc;
2069         struct reiserfs_journal_commit *commit;
2070         unsigned long trans_id = 0;
2071         struct buffer_head *c_bh;
2072         struct buffer_head *d_bh;
2073         struct buffer_head **log_blocks = NULL;
2074         struct buffer_head **real_blocks = NULL;
2075         unsigned long trans_offset;
2076         int i;
2077         int trans_half;
2078
2079         d_bh = journal_bread(p_s_sb, cur_dblock);
2080         if (!d_bh)
2081                 return 1;
2082         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2083         trans_offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb);
2084         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1037: "
2085                        "journal_read_transaction, offset %llu, len %d mount_id %d",
2086                        d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2087                        get_desc_trans_len(desc), get_desc_mount_id(desc));
2088         if (get_desc_trans_id(desc) < oldest_trans_id) {
2089                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1039: "
2090                                "journal_read_trans skipping because %lu is too old",
2091                                cur_dblock -
2092                                SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb));
2093                 brelse(d_bh);
2094                 return 1;
2095         }
2096         if (get_desc_mount_id(desc) != newest_mount_id) {
2097                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1146: "
2098                                "journal_read_trans skipping because %d is != "
2099                                "newest_mount_id %lu", get_desc_mount_id(desc),
2100                                newest_mount_id);
2101                 brelse(d_bh);
2102                 return 1;
2103         }
2104         c_bh = journal_bread(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2105                              ((trans_offset + get_desc_trans_len(desc) + 1) %
2106                               SB_ONDISK_JOURNAL_SIZE(p_s_sb)));
2107         if (!c_bh) {
2108                 brelse(d_bh);
2109                 return 1;
2110         }
2111         commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2112         if (journal_compare_desc_commit(p_s_sb, desc, commit)) {
2113                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2114                                "journal_read_transaction, "
2115                                "commit offset %llu had bad time %d or length %d",
2116                                c_bh->b_blocknr -
2117                                SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2118                                get_commit_trans_id(commit),
2119                                get_commit_trans_len(commit));
2120                 brelse(c_bh);
2121                 brelse(d_bh);
2122                 return 1;
2123         }
2124         trans_id = get_desc_trans_id(desc);
2125         /* now we know we've got a good transaction, and it was inside the valid time ranges */
2126         log_blocks = kmalloc(get_desc_trans_len(desc) *
2127                              sizeof(struct buffer_head *), GFP_NOFS);
2128         real_blocks = kmalloc(get_desc_trans_len(desc) *
2129                               sizeof(struct buffer_head *), GFP_NOFS);
2130         if (!log_blocks || !real_blocks) {
2131                 brelse(c_bh);
2132                 brelse(d_bh);
2133                 kfree(log_blocks);
2134                 kfree(real_blocks);
2135                 reiserfs_warning(p_s_sb,
2136                                  "journal-1169: kmalloc failed, unable to mount FS");
2137                 return -1;
2138         }
2139         /* get all the buffer heads */
2140         trans_half = journal_trans_half(p_s_sb->s_blocksize);
2141         for (i = 0; i < get_desc_trans_len(desc); i++) {
2142                 log_blocks[i] =
2143                     journal_getblk(p_s_sb,
2144                                    SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2145                                    (trans_offset + 1 +
2146                                     i) % SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2147                 if (i < trans_half) {
2148                         real_blocks[i] =
2149                             sb_getblk(p_s_sb,
2150                                       le32_to_cpu(desc->j_realblock[i]));
2151                 } else {
2152                         real_blocks[i] =
2153                             sb_getblk(p_s_sb,
2154                                       le32_to_cpu(commit->
2155                                                   j_realblock[i - trans_half]));
2156                 }
2157                 if (real_blocks[i]->b_blocknr > SB_BLOCK_COUNT(p_s_sb)) {
2158                         reiserfs_warning(p_s_sb,
2159                                          "journal-1207: REPLAY FAILURE fsck required! Block to replay is outside of filesystem");
2160                         goto abort_replay;
2161                 }
2162                 /* make sure we don't try to replay onto log or reserved area */
2163                 if (is_block_in_log_or_reserved_area
2164                     (p_s_sb, real_blocks[i]->b_blocknr)) {
2165                         reiserfs_warning(p_s_sb,
2166                                          "journal-1204: REPLAY FAILURE fsck required! Trying to replay onto a log block");
2167                       abort_replay:
2168                         brelse_array(log_blocks, i);
2169                         brelse_array(real_blocks, i);
2170                         brelse(c_bh);
2171                         brelse(d_bh);
2172                         kfree(log_blocks);
2173                         kfree(real_blocks);
2174                         return -1;
2175                 }
2176         }
2177         /* read in the log blocks, memcpy to the corresponding real block */
2178         ll_rw_block(READ, get_desc_trans_len(desc), log_blocks);
2179         for (i = 0; i < get_desc_trans_len(desc); i++) {
2180                 wait_on_buffer(log_blocks[i]);
2181                 if (!buffer_uptodate(log_blocks[i])) {
2182                         reiserfs_warning(p_s_sb,
2183                                          "journal-1212: REPLAY FAILURE fsck required! buffer write failed");
2184                         brelse_array(log_blocks + i,
2185                                      get_desc_trans_len(desc) - i);
2186                         brelse_array(real_blocks, get_desc_trans_len(desc));
2187                         brelse(c_bh);
2188                         brelse(d_bh);
2189                         kfree(log_blocks);
2190                         kfree(real_blocks);
2191                         return -1;
2192                 }
2193                 memcpy(real_blocks[i]->b_data, log_blocks[i]->b_data,
2194                        real_blocks[i]->b_size);
2195                 set_buffer_uptodate(real_blocks[i]);
2196                 brelse(log_blocks[i]);
2197         }
2198         /* flush out the real blocks */
2199         for (i = 0; i < get_desc_trans_len(desc); i++) {
2200                 set_buffer_dirty(real_blocks[i]);
2201                 ll_rw_block(SWRITE, 1, real_blocks + i);
2202         }
2203         for (i = 0; i < get_desc_trans_len(desc); i++) {
2204                 wait_on_buffer(real_blocks[i]);
2205                 if (!buffer_uptodate(real_blocks[i])) {
2206                         reiserfs_warning(p_s_sb,
2207                                          "journal-1226: REPLAY FAILURE, fsck required! buffer write failed");
2208                         brelse_array(real_blocks + i,
2209                                      get_desc_trans_len(desc) - i);
2210                         brelse(c_bh);
2211                         brelse(d_bh);
2212                         kfree(log_blocks);
2213                         kfree(real_blocks);
2214                         return -1;
2215                 }
2216                 brelse(real_blocks[i]);
2217         }
2218         cur_dblock =
2219             SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2220             ((trans_offset + get_desc_trans_len(desc) +
2221               2) % SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2222         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2223                        "journal-1095: setting journal " "start to offset %ld",
2224                        cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb));
2225
2226         /* init starting values for the first transaction, in case this is the last transaction to be replayed. */
2227         journal->j_start = cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb);
2228         journal->j_last_flush_trans_id = trans_id;
2229         journal->j_trans_id = trans_id + 1;
2230         /* check for trans_id overflow */
2231         if (journal->j_trans_id == 0)
2232                 journal->j_trans_id = 10;
2233         brelse(c_bh);
2234         brelse(d_bh);
2235         kfree(log_blocks);
2236         kfree(real_blocks);
2237         return 0;
2238 }
2239
2240 /* This function reads blocks starting from block and to max_block of bufsize
2241    size (but no more than BUFNR blocks at a time). This proved to improve
2242    mounting speed on self-rebuilding raid5 arrays at least.
2243    Right now it is only used from journal code. But later we might use it
2244    from other places.
2245    Note: Do not use journal_getblk/sb_getblk functions here! */
2246 static struct buffer_head *reiserfs_breada(struct block_device *dev, int block,
2247                                            int bufsize, unsigned int max_block)
2248 {
2249         struct buffer_head *bhlist[BUFNR];
2250         unsigned int blocks = BUFNR;
2251         struct buffer_head *bh;
2252         int i, j;
2253
2254         bh = __getblk(dev, block, bufsize);
2255         if (buffer_uptodate(bh))
2256                 return (bh);
2257
2258         if (block + BUFNR > max_block) {
2259                 blocks = max_block - block;
2260         }
2261         bhlist[0] = bh;
2262         j = 1;
2263         for (i = 1; i < blocks; i++) {
2264                 bh = __getblk(dev, block + i, bufsize);
2265                 if (buffer_uptodate(bh)) {
2266                         brelse(bh);
2267                         break;
2268                 } else
2269                         bhlist[j++] = bh;
2270         }
2271         ll_rw_block(READ, j, bhlist);
2272         for (i = 1; i < j; i++)
2273                 brelse(bhlist[i]);
2274         bh = bhlist[0];
2275         wait_on_buffer(bh);
2276         if (buffer_uptodate(bh))
2277                 return bh;
2278         brelse(bh);
2279         return NULL;
2280 }
2281
2282 /*
2283 ** read and replay the log
2284 ** on a clean unmount, the journal header's next unflushed pointer will be to an invalid
2285 ** transaction.  This tests that before finding all the transactions in the log, which makes normal mount times fast.
2286 **
2287 ** After a crash, this starts with the next unflushed transaction, and replays until it finds one too old, or invalid.
2288 **
2289 ** On exit, it sets things up so the first transaction will work correctly.
2290 */
2291 static int journal_read(struct super_block *p_s_sb)
2292 {
2293         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
2294         struct reiserfs_journal_desc *desc;
2295         unsigned long oldest_trans_id = 0;
2296         unsigned long oldest_invalid_trans_id = 0;
2297         time_t start;
2298         unsigned long oldest_start = 0;
2299         unsigned long cur_dblock = 0;
2300         unsigned long newest_mount_id = 9;
2301         struct buffer_head *d_bh;
2302         struct reiserfs_journal_header *jh;
2303         int valid_journal_header = 0;
2304         int replay_count = 0;
2305         int continue_replay = 1;
2306         int ret;
2307         char b[BDEVNAME_SIZE];
2308
2309         cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb);
2310         reiserfs_info(p_s_sb, "checking transaction log (%s)\n",
2311                       bdevname(journal->j_dev_bd, b));
2312         start = get_seconds();
2313
2314         /* step 1, read in the journal header block.  Check the transaction it says
2315          ** is the first unflushed, and if that transaction is not valid,
2316          ** replay is done
2317          */
2318         journal->j_header_bh = journal_bread(p_s_sb,
2319                                              SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb)
2320                                              + SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2321         if (!journal->j_header_bh) {
2322                 return 1;
2323         }
2324         jh = (struct reiserfs_journal_header *)(journal->j_header_bh->b_data);
2325         if (le32_to_cpu(jh->j_first_unflushed_offset) <
2326             SB_ONDISK_JOURNAL_SIZE(p_s_sb)
2327             && le32_to_cpu(jh->j_last_flush_trans_id) > 0) {
2328                 oldest_start =
2329                     SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2330                     le32_to_cpu(jh->j_first_unflushed_offset);
2331                 oldest_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2332                 newest_mount_id = le32_to_cpu(jh->j_mount_id);
2333                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2334                                "journal-1153: found in "
2335                                "header: first_unflushed_offset %d, last_flushed_trans_id "
2336                                "%lu", le32_to_cpu(jh->j_first_unflushed_offset),
2337                                le32_to_cpu(jh->j_last_flush_trans_id));
2338                 valid_journal_header = 1;
2339
2340                 /* now, we try to read the first unflushed offset.  If it is not valid,
2341                  ** there is nothing more we can do, and it makes no sense to read
2342                  ** through the whole log.
2343                  */
2344                 d_bh =
2345                     journal_bread(p_s_sb,
2346                                   SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2347                                   le32_to_cpu(jh->j_first_unflushed_offset));
2348                 ret = journal_transaction_is_valid(p_s_sb, d_bh, NULL, NULL);
2349                 if (!ret) {
2350                         continue_replay = 0;
2351                 }
2352                 brelse(d_bh);
2353                 goto start_log_replay;
2354         }
2355
2356         if (continue_replay && bdev_read_only(p_s_sb->s_bdev)) {
2357                 reiserfs_warning(p_s_sb,
2358                                  "clm-2076: device is readonly, unable to replay log");
2359                 return -1;
2360         }
2361
2362         /* ok, there are transactions that need to be replayed.  start with the first log block, find
2363          ** all the valid transactions, and pick out the oldest.
2364          */
2365         while (continue_replay
2366                && cur_dblock <
2367                (SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2368                 SB_ONDISK_JOURNAL_SIZE(p_s_sb))) {
2369                 /* Note that it is required for blocksize of primary fs device and journal
2370                    device to be the same */
2371                 d_bh =
2372                     reiserfs_breada(journal->j_dev_bd, cur_dblock,
2373                                     p_s_sb->s_blocksize,
2374                                     SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2375                                     SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2376                 ret =
2377                     journal_transaction_is_valid(p_s_sb, d_bh,
2378                                                  &oldest_invalid_trans_id,
2379                                                  &newest_mount_id);
2380                 if (ret == 1) {
2381                         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2382                         if (oldest_start == 0) {        /* init all oldest_ values */
2383                                 oldest_trans_id = get_desc_trans_id(desc);
2384                                 oldest_start = d_bh->b_blocknr;
2385                                 newest_mount_id = get_desc_mount_id(desc);
2386                                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2387                                                "journal-1179: Setting "
2388                                                "oldest_start to offset %llu, trans_id %lu",
2389                                                oldest_start -
2390                                                SB_ONDISK_JOURNAL_1st_BLOCK
2391                                                (p_s_sb), oldest_trans_id);
2392                         } else if (oldest_trans_id > get_desc_trans_id(desc)) {
2393                                 /* one we just read was older */
2394                                 oldest_trans_id = get_desc_trans_id(desc);
2395                                 oldest_start = d_bh->b_blocknr;
2396                                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2397                                                "journal-1180: Resetting "
2398                                                "oldest_start to offset %lu, trans_id %lu",
2399                                                oldest_start -
2400                                                SB_ONDISK_JOURNAL_1st_BLOCK
2401                                                (p_s_sb), oldest_trans_id);
2402                         }
2403                         if (newest_mount_id < get_desc_mount_id(desc)) {
2404                                 newest_mount_id = get_desc_mount_id(desc);
2405                                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2406                                                "journal-1299: Setting "
2407                                                "newest_mount_id to %d",
2408                                                get_desc_mount_id(desc));
2409                         }
2410                         cur_dblock += get_desc_trans_len(desc) + 2;
2411                 } else {
2412                         cur_dblock++;
2413                 }
2414                 brelse(d_bh);
2415         }
2416
2417       start_log_replay:
2418         cur_dblock = oldest_start;
2419         if (oldest_trans_id) {
2420                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2421                                "journal-1206: Starting replay "
2422                                "from offset %llu, trans_id %lu",
2423                                cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2424                                oldest_trans_id);
2425
2426         }
2427         replay_count = 0;
2428         while (continue_replay && oldest_trans_id > 0) {
2429                 ret =
2430                     journal_read_transaction(p_s_sb, cur_dblock, oldest_start,
2431                                              oldest_trans_id, newest_mount_id);
2432                 if (ret < 0) {
2433                         return ret;
2434                 } else if (ret != 0) {
2435                         break;
2436                 }
2437                 cur_dblock =
2438                     SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + journal->j_start;
2439                 replay_count++;
2440                 if (cur_dblock == oldest_start)
2441                         break;
2442         }
2443
2444         if (oldest_trans_id == 0) {
2445                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2446                                "journal-1225: No valid " "transactions found");
2447         }
2448         /* j_start does not get set correctly if we don't replay any transactions.
2449          ** if we had a valid journal_header, set j_start to the first unflushed transaction value,
2450          ** copy the trans_id from the header
2451          */
2452         if (valid_journal_header && replay_count == 0) {
2453                 journal->j_start = le32_to_cpu(jh->j_first_unflushed_offset);
2454                 journal->j_trans_id =
2455                     le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2456                 /* check for trans_id overflow */
2457                 if (journal->j_trans_id == 0)
2458                         journal->j_trans_id = 10;
2459                 journal->j_last_flush_trans_id =
2460                     le32_to_cpu(jh->j_last_flush_trans_id);
2461                 journal->j_mount_id = le32_to_cpu(jh->j_mount_id) + 1;
2462         } else {
2463                 journal->j_mount_id = newest_mount_id + 1;
2464         }
2465         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1299: Setting "
2466                        "newest_mount_id to %lu", journal->j_mount_id);
2467         journal->j_first_unflushed_offset = journal->j_start;
2468         if (replay_count > 0) {
2469                 reiserfs_info(p_s_sb,
2470                               "replayed %d transactions in %lu seconds\n",
2471                               replay_count, get_seconds() - start);
2472         }
2473         if (!bdev_read_only(p_s_sb->s_bdev) &&
2474             _update_journal_header_block(p_s_sb, journal->j_start,
2475                                          journal->j_last_flush_trans_id)) {
2476                 /* replay failed, caller must call free_journal_ram and abort
2477                  ** the mount
2478                  */
2479                 return -1;
2480         }
2481         return 0;
2482 }
2483
2484 static struct reiserfs_journal_list *alloc_journal_list(struct super_block *s)
2485 {
2486         struct reiserfs_journal_list *jl;
2487         jl = kzalloc(sizeof(struct reiserfs_journal_list),
2488                      GFP_NOFS | __GFP_NOFAIL);
2489         INIT_LIST_HEAD(&jl->j_list);
2490         INIT_LIST_HEAD(&jl->j_working_list);
2491         INIT_LIST_HEAD(&jl->j_tail_bh_list);
2492         INIT_LIST_HEAD(&jl->j_bh_list);
2493         sema_init(&jl->j_commit_lock, 1);
2494         SB_JOURNAL(s)->j_num_lists++;
2495         get_journal_list(jl);
2496         return jl;
2497 }
2498
2499 static void journal_list_init(struct super_block *p_s_sb)
2500 {
2501         SB_JOURNAL(p_s_sb)->j_current_jl = alloc_journal_list(p_s_sb);
2502 }
2503
2504 static int release_journal_dev(struct super_block *super,
2505                                struct reiserfs_journal *journal)
2506 {
2507         int result;
2508
2509         result = 0;
2510
2511         if (journal->j_dev_file != NULL) {
2512                 result = filp_close(journal->j_dev_file, NULL);
2513                 journal->j_dev_file = NULL;
2514                 journal->j_dev_bd = NULL;
2515         } else if (journal->j_dev_bd != NULL) {
2516                 result = blkdev_put(journal->j_dev_bd);
2517                 journal->j_dev_bd = NULL;
2518         }
2519
2520         if (result != 0) {
2521                 reiserfs_warning(super,
2522                                  "sh-457: release_journal_dev: Cannot release journal device: %i",
2523                                  result);
2524         }
2525         return result;
2526 }
2527
2528 static int journal_init_dev(struct super_block *super,
2529                             struct reiserfs_journal *journal,
2530                             const char *jdev_name)
2531 {
2532         int result;
2533         dev_t jdev;
2534         int blkdev_mode = FMODE_READ | FMODE_WRITE;
2535         char b[BDEVNAME_SIZE];
2536
2537         result = 0;
2538
2539         journal->j_dev_bd = NULL;
2540         journal->j_dev_file = NULL;
2541         jdev = SB_ONDISK_JOURNAL_DEVICE(super) ?
2542             new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super)) : super->s_dev;
2543
2544         if (bdev_read_only(super->s_bdev))
2545                 blkdev_mode = FMODE_READ;
2546
2547         /* there is no "jdev" option and journal is on separate device */
2548         if ((!jdev_name || !jdev_name[0])) {
2549                 journal->j_dev_bd = open_by_devnum(jdev, blkdev_mode);
2550                 if (IS_ERR(journal->j_dev_bd)) {
2551                         result = PTR_ERR(journal->j_dev_bd);
2552                         journal->j_dev_bd = NULL;
2553                         reiserfs_warning(super, "sh-458: journal_init_dev: "
2554                                          "cannot init journal device '%s': %i",
2555                                          __bdevname(jdev, b), result);
2556                         return result;
2557                 } else if (jdev != super->s_dev)
2558                         set_blocksize(journal->j_dev_bd, super->s_blocksize);
2559                 return 0;
2560         }
2561
2562         journal->j_dev_file = filp_open(jdev_name, 0, 0);
2563         if (!IS_ERR(journal->j_dev_file)) {
2564                 struct inode *jdev_inode = journal->j_dev_file->f_mapping->host;
2565                 if (!S_ISBLK(jdev_inode->i_mode)) {
2566                         reiserfs_warning(super, "journal_init_dev: '%s' is "
2567                                          "not a block device", jdev_name);
2568                         result = -ENOTBLK;
2569                         release_journal_dev(super, journal);
2570                 } else {
2571                         /* ok */
2572                         journal->j_dev_bd = I_BDEV(jdev_inode);
2573                         set_blocksize(journal->j_dev_bd, super->s_blocksize);
2574                         reiserfs_info(super,
2575                                       "journal_init_dev: journal device: %s\n",
2576                                       bdevname(journal->j_dev_bd, b));
2577                 }
2578         } else {
2579                 result = PTR_ERR(journal->j_dev_file);
2580                 journal->j_dev_file = NULL;
2581                 reiserfs_warning(super,
2582                                  "journal_init_dev: Cannot open '%s': %i",
2583                                  jdev_name, result);
2584         }
2585         return result;
2586 }
2587
2588 /*
2589 ** must be called once on fs mount.  calls journal_read for you
2590 */
2591 int journal_init(struct super_block *p_s_sb, const char *j_dev_name,
2592                  int old_format, unsigned int commit_max_age)
2593 {
2594         int num_cnodes = SB_ONDISK_JOURNAL_SIZE(p_s_sb) * 2;
2595         struct buffer_head *bhjh;
2596         struct reiserfs_super_block *rs;
2597         struct reiserfs_journal_header *jh;
2598         struct reiserfs_journal *journal;
2599         struct reiserfs_journal_list *jl;
2600         char b[BDEVNAME_SIZE];
2601
2602         journal = SB_JOURNAL(p_s_sb) = vmalloc(sizeof(struct reiserfs_journal));
2603         if (!journal) {
2604                 reiserfs_warning(p_s_sb,
2605                                  "journal-1256: unable to get memory for journal structure");
2606                 return 1;
2607         }
2608         memset(journal, 0, sizeof(struct reiserfs_journal));
2609         INIT_LIST_HEAD(&journal->j_bitmap_nodes);
2610         INIT_LIST_HEAD(&journal->j_prealloc_list);
2611         INIT_LIST_HEAD(&journal->j_working_list);
2612         INIT_LIST_HEAD(&journal->j_journal_list);
2613         journal->j_persistent_trans = 0;
2614         if (reiserfs_allocate_list_bitmaps(p_s_sb,
2615                                            journal->j_list_bitmap,
2616                                            SB_BMAP_NR(p_s_sb)))
2617                 goto free_and_return;
2618         allocate_bitmap_nodes(p_s_sb);
2619
2620         /* reserved for journal area support */
2621         SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb) = (old_format ?
2622                                                  REISERFS_OLD_DISK_OFFSET_IN_BYTES
2623                                                  / p_s_sb->s_blocksize +
2624                                                  SB_BMAP_NR(p_s_sb) +
2625                                                  1 :
2626                                                  REISERFS_DISK_OFFSET_IN_BYTES /
2627                                                  p_s_sb->s_blocksize + 2);
2628
2629         /* Sanity check to see is the standard journal fitting withing first bitmap
2630            (actual for small blocksizes) */
2631         if (!SB_ONDISK_JOURNAL_DEVICE(p_s_sb) &&
2632             (SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb) +
2633              SB_ONDISK_JOURNAL_SIZE(p_s_sb) > p_s_sb->s_blocksize * 8)) {
2634                 reiserfs_warning(p_s_sb,
2635                                  "journal-1393: journal does not fit for area "
2636                                  "addressed by first of bitmap blocks. It starts at "
2637                                  "%u and its size is %u. Block size %ld",
2638                                  SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb),
2639                                  SB_ONDISK_JOURNAL_SIZE(p_s_sb),
2640                                  p_s_sb->s_blocksize);
2641                 goto free_and_return;
2642         }
2643
2644         if (journal_init_dev(p_s_sb, journal, j_dev_name) != 0) {
2645                 reiserfs_warning(p_s_sb,
2646                                  "sh-462: unable to initialize jornal device");
2647                 goto free_and_return;
2648         }
2649
2650         rs = SB_DISK_SUPER_BLOCK(p_s_sb);
2651
2652         /* read journal header */
2653         bhjh = journal_bread(p_s_sb,
2654                              SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2655                              SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2656         if (!bhjh) {
2657                 reiserfs_warning(p_s_sb,
2658                                  "sh-459: unable to read journal header");
2659                 goto free_and_return;
2660         }
2661         jh = (struct reiserfs_journal_header *)(bhjh->b_data);
2662
2663         /* make sure that journal matches to the super block */
2664         if (is_reiserfs_jr(rs)
2665             && (le32_to_cpu(jh->jh_journal.jp_journal_magic) !=
2666                 sb_jp_journal_magic(rs))) {
2667                 reiserfs_warning(p_s_sb,
2668                                  "sh-460: journal header magic %x "
2669                                  "(device %s) does not match to magic found in super "
2670                                  "block %x", jh->jh_journal.jp_journal_magic,
2671                                  bdevname(journal->j_dev_bd, b),
2672                                  sb_jp_journal_magic(rs));
2673                 brelse(bhjh);
2674                 goto free_and_return;
2675         }
2676
2677         journal->j_trans_max = le32_to_cpu(jh->jh_journal.jp_journal_trans_max);
2678         journal->j_max_batch = le32_to_cpu(jh->jh_journal.jp_journal_max_batch);
2679         journal->j_max_commit_age =
2680             le32_to_cpu(jh->jh_journal.jp_journal_max_commit_age);
2681         journal->j_max_trans_age = JOURNAL_MAX_TRANS_AGE;
2682
2683         if (journal->j_trans_max) {
2684                 /* make sure these parameters are available, assign it if they are not */
2685                 __u32 initial = journal->j_trans_max;
2686                 __u32 ratio = 1;
2687
2688                 if (p_s_sb->s_blocksize < 4096)
2689                         ratio = 4096 / p_s_sb->s_blocksize;
2690
2691                 if (SB_ONDISK_JOURNAL_SIZE(p_s_sb) / journal->j_trans_max <
2692                     JOURNAL_MIN_RATIO)
2693                         journal->j_trans_max =
2694                             SB_ONDISK_JOURNAL_SIZE(p_s_sb) / JOURNAL_MIN_RATIO;
2695                 if (journal->j_trans_max > JOURNAL_TRANS_MAX_DEFAULT / ratio)
2696                         journal->j_trans_max =
2697                             JOURNAL_TRANS_MAX_DEFAULT / ratio;
2698                 if (journal->j_trans_max < JOURNAL_TRANS_MIN_DEFAULT / ratio)
2699                         journal->j_trans_max =
2700                             JOURNAL_TRANS_MIN_DEFAULT / ratio;
2701
2702                 if (journal->j_trans_max != initial)
2703                         reiserfs_warning(p_s_sb,
2704                                          "sh-461: journal_init: wrong transaction max size (%u). Changed to %u",
2705                                          initial, journal->j_trans_max);
2706
2707                 journal->j_max_batch = journal->j_trans_max *
2708                     JOURNAL_MAX_BATCH_DEFAULT / JOURNAL_TRANS_MAX_DEFAULT;
2709         }
2710
2711         if (!journal->j_trans_max) {
2712                 /*we have the file system was created by old version of mkreiserfs
2713                    so this field contains zero value */
2714                 journal->j_trans_max = JOURNAL_TRANS_MAX_DEFAULT;
2715                 journal->j_max_batch = JOURNAL_MAX_BATCH_DEFAULT;
2716                 journal->j_max_commit_age = JOURNAL_MAX_COMMIT_AGE;
2717
2718                 /* for blocksize >= 4096 - max transaction size is 1024. For block size < 4096
2719                    trans max size is decreased proportionally */
2720                 if (p_s_sb->s_blocksize < 4096) {
2721                         journal->j_trans_max /= (4096 / p_s_sb->s_blocksize);
2722                         journal->j_max_batch = (journal->j_trans_max) * 9 / 10;
2723                 }
2724         }
2725
2726         journal->j_default_max_commit_age = journal->j_max_commit_age;
2727
2728         if (commit_max_age != 0) {
2729                 journal->j_max_commit_age = commit_max_age;
2730                 journal->j_max_trans_age = commit_max_age;
2731         }
2732
2733         reiserfs_info(p_s_sb, "journal params: device %s, size %u, "
2734                       "journal first block %u, max trans len %u, max batch %u, "
2735                       "max commit age %u, max trans age %u\n",
2736                       bdevname(journal->j_dev_bd, b),
2737                       SB_ONDISK_JOURNAL_SIZE(p_s_sb),
2738                       SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2739                       journal->j_trans_max,
2740                       journal->j_max_batch,
2741                       journal->j_max_commit_age, journal->j_max_trans_age);
2742
2743         brelse(bhjh);
2744
2745         journal->j_list_bitmap_index = 0;
2746         journal_list_init(p_s_sb);
2747
2748         memset(journal->j_list_hash_table, 0,
2749                JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
2750
2751         INIT_LIST_HEAD(&journal->j_dirty_buffers);
2752         spin_lock_init(&journal->j_dirty_buffers_lock);
2753
2754         journal->j_start = 0;
2755         journal->j_len = 0;
2756         journal->j_len_alloc = 0;
2757         atomic_set(&(journal->j_wcount), 0);
2758         atomic_set(&(journal->j_async_throttle), 0);
2759         journal->j_bcount = 0;
2760         journal->j_trans_start_time = 0;
2761         journal->j_last = NULL;
2762         journal->j_first = NULL;
2763         init_waitqueue_head(&(journal->j_join_wait));
2764         sema_init(&journal->j_lock, 1);
2765         sema_init(&journal->j_flush_sem, 1);
2766
2767         journal->j_trans_id = 10;
2768         journal->j_mount_id = 10;
2769         journal->j_state = 0;
2770         atomic_set(&(journal->j_jlock), 0);
2771         journal->j_cnode_free_list = allocate_cnodes(num_cnodes);
2772         journal->j_cnode_free_orig = journal->j_cnode_free_list;
2773         journal->j_cnode_free = journal->j_cnode_free_list ? num_cnodes : 0;
2774         journal->j_cnode_used = 0;
2775         journal->j_must_wait = 0;
2776
2777         if (journal->j_cnode_free == 0) {
2778                 reiserfs_warning(p_s_sb, "journal-2004: Journal cnode memory "
2779                                  "allocation failed (%ld bytes). Journal is "
2780                                  "too large for available memory. Usually "
2781                                  "this is due to a journal that is too large.",
2782                                  sizeof (struct reiserfs_journal_cnode) * num_cnodes);
2783                 goto free_and_return;
2784         }
2785
2786         init_journal_hash(p_s_sb);
2787         jl = journal->j_current_jl;
2788         jl->j_list_bitmap = get_list_bitmap(p_s_sb, jl);
2789         if (!jl->j_list_bitmap) {
2790                 reiserfs_warning(p_s_sb,
2791                                  "journal-2005, get_list_bitmap failed for journal list 0");
2792                 goto free_and_return;
2793         }
2794         if (journal_read(p_s_sb) < 0) {
2795                 reiserfs_warning(p_s_sb, "Replay Failure, unable to mount");
2796                 goto free_and_return;
2797         }
2798
2799         reiserfs_mounted_fs_count++;
2800         if (reiserfs_mounted_fs_count <= 1)
2801                 commit_wq = create_workqueue("reiserfs");
2802
2803         INIT_WORK(&journal->j_work, flush_async_commits, p_s_sb);
2804         return 0;
2805       free_and_return:
2806         free_journal_ram(p_s_sb);
2807         return 1;
2808 }
2809
2810 /*
2811 ** test for a polite end of the current transaction.  Used by file_write, and should
2812 ** be used by delete to make sure they don't write more than can fit inside a single
2813 ** transaction
2814 */
2815 int journal_transaction_should_end(struct reiserfs_transaction_handle *th,
2816                                    int new_alloc)
2817 {
2818         struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2819         time_t now = get_seconds();
2820         /* cannot restart while nested */
2821         BUG_ON(!th->t_trans_id);
2822         if (th->t_refcount > 1)
2823                 return 0;
2824         if (journal->j_must_wait > 0 ||
2825             (journal->j_len_alloc + new_alloc) >= journal->j_max_batch ||
2826             atomic_read(&(journal->j_jlock)) ||
2827             (now - journal->j_trans_start_time) > journal->j_max_trans_age ||
2828             journal->j_cnode_free < (journal->j_trans_max * 3)) {
2829                 return 1;
2830         }
2831         /* protected by the BKL here */
2832         journal->j_len_alloc += new_alloc;
2833         th->t_blocks_allocated += new_alloc ;
2834         return 0;
2835 }
2836
2837 /* this must be called inside a transaction, and requires the
2838 ** kernel_lock to be held
2839 */
2840 void reiserfs_block_writes(struct reiserfs_transaction_handle *th)
2841 {
2842         struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2843         BUG_ON(!th->t_trans_id);
2844         journal->j_must_wait = 1;
2845         set_bit(J_WRITERS_BLOCKED, &journal->j_state);
2846         return;
2847 }
2848
2849 /* this must be called without a transaction started, and does not
2850 ** require BKL
2851 */
2852 void reiserfs_allow_writes(struct super_block *s)
2853 {
2854         struct reiserfs_journal *journal = SB_JOURNAL(s);
2855         clear_bit(J_WRITERS_BLOCKED, &journal->j_state);
2856         wake_up(&journal->j_join_wait);
2857 }
2858
2859 /* this must be called without a transaction started, and does not
2860 ** require BKL
2861 */
2862 void reiserfs_wait_on_write_block(struct super_block *s)
2863 {
2864         struct reiserfs_journal *journal = SB_JOURNAL(s);
2865         wait_event(journal->j_join_wait,
2866                    !test_bit(J_WRITERS_BLOCKED, &journal->j_state));
2867 }
2868
2869 static void queue_log_writer(struct super_block *s)
2870 {
2871         wait_queue_t wait;
2872         struct reiserfs_journal *journal = SB_JOURNAL(s);
2873         set_bit(J_WRITERS_QUEUED, &journal->j_state);
2874
2875         /*
2876          * we don't want to use wait_event here because
2877          * we only want to wait once.
2878          */
2879         init_waitqueue_entry(&wait, current);
2880         add_wait_queue(&journal->j_join_wait, &wait);
2881         set_current_state(TASK_UNINTERRUPTIBLE);
2882         if (test_bit(J_WRITERS_QUEUED, &journal->j_state))
2883                 schedule();
2884         current->state = TASK_RUNNING;
2885         remove_wait_queue(&journal->j_join_wait, &wait);
2886 }
2887
2888 static void wake_queued_writers(struct super_block *s)
2889 {
2890         struct reiserfs_journal *journal = SB_JOURNAL(s);
2891         if (test_and_clear_bit(J_WRITERS_QUEUED, &journal->j_state))
2892                 wake_up(&journal->j_join_wait);
2893 }
2894
2895 static void let_transaction_grow(struct super_block *sb, unsigned long trans_id)
2896 {
2897         struct reiserfs_journal *journal = SB_JOURNAL(sb);
2898         unsigned long bcount = journal->j_bcount;
2899         while (1) {
2900                 schedule_timeout_uninterruptible(1);
2901                 journal->j_current_jl->j_state |= LIST_COMMIT_PENDING;
2902                 while ((atomic_read(&journal->j_wcount) > 0 ||
2903                         atomic_read(&journal->j_jlock)) &&
2904                        journal->j_trans_id == trans_id) {
2905                         queue_log_writer(sb);
2906                 }
2907                 if (journal->j_trans_id != trans_id)
2908                         break;
2909                 if (bcount == journal->j_bcount)
2910                         break;
2911                 bcount = journal->j_bcount;
2912         }
2913 }
2914
2915 /* join == true if you must join an existing transaction.
2916 ** join == false if you can deal with waiting for others to finish
2917 **
2918 ** this will block until the transaction is joinable.  send the number of blocks you
2919 ** expect to use in nblocks.
2920 */
2921 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
2922                               struct super_block *p_s_sb, unsigned long nblocks,
2923                               int join)
2924 {
2925         time_t now = get_seconds();
2926         int old_trans_id;
2927         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
2928         struct reiserfs_transaction_handle myth;
2929         int sched_count = 0;
2930         int retval;
2931
2932         reiserfs_check_lock_depth(p_s_sb, "journal_begin");
2933         if (nblocks > journal->j_trans_max)
2934                 BUG();
2935
2936         PROC_INFO_INC(p_s_sb, journal.journal_being);
2937         /* set here for journal_join */
2938         th->t_refcount = 1;
2939         th->t_super = p_s_sb;
2940
2941       relock:
2942         lock_journal(p_s_sb);
2943         if (join != JBEGIN_ABORT && reiserfs_is_journal_aborted(journal)) {
2944                 unlock_journal(p_s_sb);
2945                 retval = journal->j_errno;
2946                 goto out_fail;
2947         }
2948         journal->j_bcount++;
2949
2950         if (test_bit(J_WRITERS_BLOCKED, &journal->j_state)) {
2951                 unlock_journal(p_s_sb);
2952                 reiserfs_wait_on_write_block(p_s_sb);
2953                 PROC_INFO_INC(p_s_sb, journal.journal_relock_writers);
2954                 goto relock;
2955         }
2956         now = get_seconds();
2957
2958         /* if there is no room in the journal OR
2959          ** if this transaction is too old, and we weren't called joinable, wait for it to finish before beginning
2960          ** we don't sleep if there aren't other writers
2961          */
2962
2963         if ((!join && journal->j_must_wait > 0) ||
2964             (!join
2965              && (journal->j_len_alloc + nblocks + 2) >= journal->j_max_batch)
2966             || (!join && atomic_read(&journal->j_wcount) > 0
2967                 && journal->j_trans_start_time > 0
2968                 && (now - journal->j_trans_start_time) >
2969                 journal->j_max_trans_age) || (!join
2970                                               && atomic_read(&journal->j_jlock))
2971             || (!join && journal->j_cnode_free < (journal->j_trans_max * 3))) {
2972
2973                 old_trans_id = journal->j_trans_id;
2974                 unlock_journal(p_s_sb); /* allow others to finish this transaction */
2975
2976                 if (!join && (journal->j_len_alloc + nblocks + 2) >=
2977                     journal->j_max_batch &&
2978                     ((journal->j_len + nblocks + 2) * 100) <
2979                     (journal->j_len_alloc * 75)) {
2980                         if (atomic_read(&journal->j_wcount) > 10) {
2981                                 sched_count++;
2982                                 queue_log_writer(p_s_sb);
2983                                 goto relock;
2984                         }
2985                 }
2986                 /* don't mess with joining the transaction if all we have to do is
2987                  * wait for someone else to do a commit
2988                  */
2989                 if (atomic_read(&journal->j_jlock)) {
2990                         while (journal->j_trans_id == old_trans_id &&
2991                                atomic_read(&journal->j_jlock)) {
2992                                 queue_log_writer(p_s_sb);
2993                         }
2994                         goto relock;
2995                 }
2996                 retval = journal_join(&myth, p_s_sb, 1);
2997                 if (retval)
2998                         goto out_fail;
2999
3000                 /* someone might have ended the transaction while we joined */
3001                 if (old_trans_id != journal->j_trans_id) {
3002                         retval = do_journal_end(&myth, p_s_sb, 1, 0);
3003                 } else {
3004                         retval = do_journal_end(&myth, p_s_sb, 1, COMMIT_NOW);
3005                 }
3006
3007                 if (retval)
3008                         goto out_fail;
3009
3010                 PROC_INFO_INC(p_s_sb, journal.journal_relock_wcount);
3011                 goto relock;
3012         }
3013         /* we are the first writer, set trans_id */
3014         if (journal->j_trans_start_time == 0) {
3015                 journal->j_trans_start_time = get_seconds();
3016         }
3017         atomic_inc(&(journal->j_wcount));
3018         journal->j_len_alloc += nblocks;
3019         th->t_blocks_logged = 0;
3020         th->t_blocks_allocated = nblocks;
3021         th->t_trans_id = journal->j_trans_id;
3022         unlock_journal(p_s_sb);
3023         INIT_LIST_HEAD(&th->t_list);
3024         get_fs_excl();
3025         return 0;
3026
3027       out_fail:
3028         memset(th, 0, sizeof(*th));
3029         /* Re-set th->t_super, so we can properly keep track of how many
3030          * persistent transactions there are. We need to do this so if this
3031          * call is part of a failed restart_transaction, we can free it later */
3032         th->t_super = p_s_sb;
3033         return retval;
3034 }
3035
3036 struct reiserfs_transaction_handle *reiserfs_persistent_transaction(struct
3037                                                                     super_block
3038                                                                     *s,
3039                                                                     int nblocks)
3040 {
3041         int ret;
3042         struct reiserfs_transaction_handle *th;
3043
3044         /* if we're nesting into an existing transaction.  It will be
3045          ** persistent on its own
3046          */
3047         if (reiserfs_transaction_running(s)) {
3048                 th = current->journal_info;
3049                 th->t_refcount++;
3050                 if (th->t_refcount < 2) {
3051                         BUG();
3052                 }
3053                 return th;
3054         }
3055         th = kmalloc(sizeof(struct reiserfs_transaction_handle), GFP_NOFS);
3056         if (!th)
3057                 return NULL;
3058         ret = journal_begin(th, s, nblocks);
3059         if (ret) {
3060                 kfree(th);
3061                 return NULL;
3062         }
3063
3064         SB_JOURNAL(s)->j_persistent_trans++;
3065         return th;
3066 }
3067
3068 int reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle *th)
3069 {
3070         struct super_block *s = th->t_super;
3071         int ret = 0;
3072         if (th->t_trans_id)
3073                 ret = journal_end(th, th->t_super, th->t_blocks_allocated);
3074         else
3075                 ret = -EIO;
3076         if (th->t_refcount == 0) {
3077                 SB_JOURNAL(s)->j_persistent_trans--;
3078                 kfree(th);
3079         }
3080         return ret;
3081 }
3082
3083 static int journal_join(struct reiserfs_transaction_handle *th,
3084                         struct super_block *p_s_sb, unsigned long nblocks)
3085 {
3086         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3087
3088         /* this keeps do_journal_end from NULLing out the current->journal_info
3089          ** pointer
3090          */
3091         th->t_handle_save = cur_th;
3092         if (cur_th && cur_th->t_refcount > 1) {
3093                 BUG();
3094         }
3095         return do_journal_begin_r(th, p_s_sb, nblocks, JBEGIN_JOIN);
3096 }
3097
3098 int journal_join_abort(struct reiserfs_transaction_handle *th,
3099                        struct super_block *p_s_sb, unsigned long nblocks)
3100 {
3101         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3102
3103         /* this keeps do_journal_end from NULLing out the current->journal_info
3104          ** pointer
3105          */
3106         th->t_handle_save = cur_th;
3107         if (cur_th && cur_th->t_refcount > 1) {
3108                 BUG();
3109         }
3110         return do_journal_begin_r(th, p_s_sb, nblocks, JBEGIN_ABORT);
3111 }
3112
3113 int journal_begin(struct reiserfs_transaction_handle *th,
3114                   struct super_block *p_s_sb, unsigned long nblocks)
3115 {
3116         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3117         int ret;
3118
3119         th->t_handle_save = NULL;
3120         if (cur_th) {
3121                 /* we are nesting into the current transaction */
3122                 if (cur_th->t_super == p_s_sb) {
3123                         BUG_ON(!cur_th->t_refcount);
3124                         cur_th->t_refcount++;
3125                         memcpy(th, cur_th, sizeof(*th));
3126                         if (th->t_refcount <= 1)
3127                                 reiserfs_warning(p_s_sb,
3128                                                  "BAD: refcount <= 1, but journal_info != 0");
3129                         return 0;
3130                 } else {
3131                         /* we've ended up with a handle from a different filesystem.
3132                          ** save it and restore on journal_end.  This should never
3133                          ** really happen...
3134                          */
3135                         reiserfs_warning(p_s_sb,
3136                                          "clm-2100: nesting info a different FS");
3137                         th->t_handle_save = current->journal_info;
3138                         current->journal_info = th;
3139                 }
3140         } else {
3141                 current->journal_info = th;
3142         }
3143         ret = do_journal_begin_r(th, p_s_sb, nblocks, JBEGIN_REG);
3144         if (current->journal_info != th)
3145                 BUG();
3146
3147         /* I guess this boils down to being the reciprocal of clm-2100 above.
3148          * If do_journal_begin_r fails, we need to put it back, since journal_end
3149          * won't be called to do it. */
3150         if (ret)
3151                 current->journal_info = th->t_handle_save;
3152         else
3153                 BUG_ON(!th->t_refcount);
3154
3155         return ret;
3156 }
3157
3158 /*
3159 ** puts bh into the current transaction.  If it was already there, reorders removes the
3160 ** old pointers from the hash, and puts new ones in (to make sure replay happen in the right order).
3161 **
3162 ** if it was dirty, cleans and files onto the clean list.  I can't let it be dirty again until the
3163 ** transaction is committed.
3164 **
3165 ** if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len.
3166 */
3167 int journal_mark_dirty(struct reiserfs_transaction_handle *th,
3168                        struct super_block *p_s_sb, struct buffer_head *bh)
3169 {
3170         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3171         struct reiserfs_journal_cnode *cn = NULL;
3172         int count_already_incd = 0;
3173         int prepared = 0;
3174         BUG_ON(!th->t_trans_id);
3175
3176         PROC_INFO_INC(p_s_sb, journal.mark_dirty);
3177         if (th->t_trans_id != journal->j_trans_id) {
3178                 reiserfs_panic(th->t_super,
3179                                "journal-1577: handle trans id %ld != current trans id %ld\n",
3180                                th->t_trans_id, journal->j_trans_id);
3181         }
3182
3183         p_s_sb->s_dirt = 1;
3184
3185         prepared = test_clear_buffer_journal_prepared(bh);
3186         clear_buffer_journal_restore_dirty(bh);
3187         /* already in this transaction, we are done */
3188         if (buffer_journaled(bh)) {
3189                 PROC_INFO_INC(p_s_sb, journal.mark_dirty_already);
3190                 return 0;
3191         }
3192
3193         /* this must be turned into a panic instead of a warning.  We can't allow
3194          ** a dirty or journal_dirty or locked buffer to be logged, as some changes
3195          ** could get to disk too early.  NOT GOOD.
3196          */
3197         if (!prepared || buffer_dirty(bh)) {
3198                 reiserfs_warning(p_s_sb, "journal-1777: buffer %llu bad state "
3199                                  "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT",
3200                                  (unsigned long long)bh->b_blocknr,
3201                                  prepared ? ' ' : '!',
3202                                  buffer_locked(bh) ? ' ' : '!',
3203                                  buffer_dirty(bh) ? ' ' : '!',
3204                                  buffer_journal_dirty(bh) ? ' ' : '!');
3205         }
3206
3207         if (atomic_read(&(journal->j_wcount)) <= 0) {
3208                 reiserfs_warning(p_s_sb,
3209                                  "journal-1409: journal_mark_dirty returning because j_wcount was %d",
3210                                  atomic_read(&(journal->j_wcount)));
3211                 return 1;
3212         }
3213         /* this error means I've screwed up, and we've overflowed the transaction.
3214          ** Nothing can be done here, except make the FS readonly or panic.
3215          */
3216         if (journal->j_len >= journal->j_trans_max) {
3217                 reiserfs_panic(th->t_super,
3218                                "journal-1413: journal_mark_dirty: j_len (%lu) is too big\n",
3219                                journal->j_len);
3220         }
3221
3222         if (buffer_journal_dirty(bh)) {
3223                 count_already_incd = 1;
3224                 PROC_INFO_INC(p_s_sb, journal.mark_dirty_notjournal);
3225                 clear_buffer_journal_dirty(bh);
3226         }
3227
3228         if (journal->j_len > journal->j_len_alloc) {
3229                 journal->j_len_alloc = journal->j_len + JOURNAL_PER_BALANCE_CNT;
3230         }
3231
3232         set_buffer_journaled(bh);
3233
3234         /* now put this guy on the end */
3235         if (!cn) {
3236                 cn = get_cnode(p_s_sb);
3237                 if (!cn) {
3238                         reiserfs_panic(p_s_sb, "get_cnode failed!\n");
3239                 }
3240
3241                 if (th->t_blocks_logged == th->t_blocks_allocated) {
3242                         th->t_blocks_allocated += JOURNAL_PER_BALANCE_CNT;
3243                         journal->j_len_alloc += JOURNAL_PER_BALANCE_CNT;
3244                 }
3245                 th->t_blocks_logged++;
3246                 journal->j_len++;
3247
3248                 cn->bh = bh;
3249                 cn->blocknr = bh->b_blocknr;
3250                 cn->sb = p_s_sb;
3251                 cn->jlist = NULL;
3252                 insert_journal_hash(journal->j_hash_table, cn);
3253                 if (!count_already_incd) {
3254                         get_bh(bh);
3255                 }
3256         }
3257         cn->next = NULL;
3258         cn->prev = journal->j_last;
3259         cn->bh = bh;
3260         if (journal->j_last) {
3261                 journal->j_last->next = cn;
3262                 journal->j_last = cn;
3263         } else {
3264                 journal->j_first = cn;
3265                 journal->j_last = cn;
3266         }
3267         return 0;
3268 }
3269
3270 int journal_end(struct reiserfs_transaction_handle *th,
3271                 struct super_block *p_s_sb, unsigned long nblocks)
3272 {
3273         if (!current->journal_info && th->t_refcount > 1)
3274                 reiserfs_warning(p_s_sb, "REISER-NESTING: th NULL, refcount %d",
3275                                  th->t_refcount);
3276
3277         if (!th->t_trans_id) {
3278                 WARN_ON(1);
3279                 return -EIO;
3280         }
3281
3282         th->t_refcount--;
3283         if (th->t_refcount > 0) {
3284                 struct reiserfs_transaction_handle *cur_th =
3285                     current->journal_info;
3286
3287                 /* we aren't allowed to close a nested transaction on a different
3288                  ** filesystem from the one in the task struct
3289                  */
3290                 if (cur_th->t_super != th->t_super)
3291                         BUG();
3292
3293                 if (th != cur_th) {
3294                         memcpy(current->journal_info, th, sizeof(*th));
3295                         th->t_trans_id = 0;
3296                 }
3297                 return 0;
3298         } else {
3299                 return do_journal_end(th, p_s_sb, nblocks, 0);
3300         }
3301 }
3302
3303 /* removes from the current transaction, relsing and descrementing any counters.
3304 ** also files the removed buffer directly onto the clean list
3305 **
3306 ** called by journal_mark_freed when a block has been deleted
3307 **
3308 ** returns 1 if it cleaned and relsed the buffer. 0 otherwise
3309 */
3310 static int remove_from_transaction(struct super_block *p_s_sb,
3311                                    b_blocknr_t blocknr, int already_cleaned)
3312 {
3313         struct buffer_head *bh;
3314         struct reiserfs_journal_cnode *cn;
3315         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3316         int ret = 0;
3317
3318         cn = get_journal_hash_dev(p_s_sb, journal->j_hash_table, blocknr);
3319         if (!cn || !cn->bh) {
3320                 return ret;
3321         }
3322         bh = cn->bh;
3323         if (cn->prev) {
3324                 cn->prev->next = cn->next;
3325         }
3326         if (cn->next) {
3327                 cn->next->prev = cn->prev;
3328         }
3329         if (cn == journal->j_first) {
3330                 journal->j_first = cn->next;
3331         }
3332         if (cn == journal->j_last) {
3333                 journal->j_last = cn->prev;
3334         }
3335         if (bh)
3336                 remove_journal_hash(p_s_sb, journal->j_hash_table, NULL,
3337                                     bh->b_blocknr, 0);
3338         clear_buffer_journaled(bh);     /* don't log this one */
3339
3340         if (!already_cleaned) {
3341                 clear_buffer_journal_dirty(bh);
3342                 clear_buffer_dirty(bh);
3343                 clear_buffer_journal_test(bh);
3344                 put_bh(bh);
3345                 if (atomic_read(&(bh->b_count)) < 0) {
3346                         reiserfs_warning(p_s_sb,
3347                                          "journal-1752: remove from trans, b_count < 0");
3348                 }
3349                 ret = 1;
3350         }
3351         journal->j_len--;
3352         journal->j_len_alloc--;
3353         free_cnode(p_s_sb, cn);
3354         return ret;
3355 }
3356
3357 /*
3358 ** for any cnode in a journal list, it can only be dirtied of all the
3359 ** transactions that include it are commited to disk.
3360 ** this checks through each transaction, and returns 1 if you are allowed to dirty,
3361 ** and 0 if you aren't
3362 **
3363 ** it is called by dirty_journal_list, which is called after flush_commit_list has gotten all the log
3364 ** blocks for a given transaction on disk
3365 **
3366 */
3367 static int can_dirty(struct reiserfs_journal_cnode *cn)
3368 {
3369         struct super_block *sb = cn->sb;
3370         b_blocknr_t blocknr = cn->blocknr;
3371         struct reiserfs_journal_cnode *cur = cn->hprev;
3372         int can_dirty = 1;
3373
3374         /* first test hprev.  These are all newer than cn, so any node here
3375          ** with the same block number and dev means this node can't be sent
3376          ** to disk right now.
3377          */
3378         while (cur && can_dirty) {
3379                 if (cur->jlist && cur->bh && cur->blocknr && cur->sb == sb &&
3380                     cur->blocknr == blocknr) {
3381                         can_dirty = 0;
3382                 }
3383                 cur = cur->hprev;
3384         }
3385         /* then test hnext.  These are all older than cn.  As long as they
3386          ** are committed to the log, it is safe to write cn to disk
3387          */
3388         cur = cn->hnext;
3389         while (cur && can_dirty) {
3390                 if (cur->jlist && cur->jlist->j_len > 0 &&
3391                     atomic_read(&(cur->jlist->j_commit_left)) > 0 && cur->bh &&
3392                     cur->blocknr && cur->sb == sb && cur->blocknr == blocknr) {
3393                         can_dirty = 0;
3394                 }
3395                 cur = cur->hnext;
3396         }
3397         return can_dirty;
3398 }
3399
3400 /* syncs the commit blocks, but does not force the real buffers to disk
3401 ** will wait until the current transaction is done/commited before returning
3402 */
3403 int journal_end_sync(struct reiserfs_transaction_handle *th,
3404                      struct super_block *p_s_sb, unsigned long nblocks)
3405 {
3406         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3407
3408         BUG_ON(!th->t_trans_id);
3409         /* you can sync while nested, very, very bad */
3410         if (th->t_refcount > 1) {
3411                 BUG();
3412         }
3413         if (journal->j_len == 0) {
3414                 reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb),
3415                                              1);
3416                 journal_mark_dirty(th, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb));
3417         }
3418         return do_journal_end(th, p_s_sb, nblocks, COMMIT_NOW | WAIT);
3419 }
3420
3421 /*
3422 ** writeback the pending async commits to disk
3423 */
3424 static void flush_async_commits(void *p)
3425 {
3426         struct super_block *p_s_sb = p;
3427         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3428         struct reiserfs_journal_list *jl;
3429         struct list_head *entry;
3430
3431         lock_kernel();
3432         if (!list_empty(&journal->j_journal_list)) {
3433                 /* last entry is the youngest, commit it and you get everything */
3434                 entry = journal->j_journal_list.prev;
3435                 jl = JOURNAL_LIST_ENTRY(entry);
3436                 flush_commit_list(p_s_sb, jl, 1);
3437         }
3438         unlock_kernel();
3439         /*
3440          * this is a little racey, but there's no harm in missing
3441          * the filemap_fdata_write
3442          */
3443         if (!atomic_read(&journal->j_async_throttle)
3444             && !reiserfs_is_journal_aborted(journal)) {
3445                 atomic_inc(&journal->j_async_throttle);
3446                 filemap_fdatawrite(p_s_sb->s_bdev->bd_inode->i_mapping);
3447                 atomic_dec(&journal->j_async_throttle);
3448         }
3449 }
3450
3451 /*
3452 ** flushes any old transactions to disk
3453 ** ends the current transaction if it is too old
3454 */
3455 int reiserfs_flush_old_commits(struct super_block *p_s_sb)
3456 {
3457         time_t now;
3458         struct reiserfs_transaction_handle th;
3459         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3460
3461         now = get_seconds();
3462         /* safety check so we don't flush while we are replaying the log during
3463          * mount
3464          */
3465         if (list_empty(&journal->j_journal_list)) {
3466                 return 0;
3467         }
3468
3469         /* check the current transaction.  If there are no writers, and it is
3470          * too old, finish it, and force the commit blocks to disk
3471          */
3472         if (atomic_read(&journal->j_wcount) <= 0 &&
3473             journal->j_trans_start_time > 0 &&
3474             journal->j_len > 0 &&
3475             (now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3476                 if (!journal_join(&th, p_s_sb, 1)) {
3477                         reiserfs_prepare_for_journal(p_s_sb,
3478                                                      SB_BUFFER_WITH_SB(p_s_sb),
3479                                                      1);
3480                         journal_mark_dirty(&th, p_s_sb,
3481                                            SB_BUFFER_WITH_SB(p_s_sb));
3482
3483                         /* we're only being called from kreiserfsd, it makes no sense to do
3484                          ** an async commit so that kreiserfsd can do it later
3485                          */
3486                         do_journal_end(&th, p_s_sb, 1, COMMIT_NOW | WAIT);
3487                 }
3488         }
3489         return p_s_sb->s_dirt;
3490 }
3491
3492 /*
3493 ** returns 0 if do_journal_end should return right away, returns 1 if do_journal_end should finish the commit
3494 **
3495 ** if the current transaction is too old, but still has writers, this will wait on j_join_wait until all
3496 ** the writers are done.  By the time it wakes up, the transaction it was called has already ended, so it just
3497 ** flushes the commit list and returns 0.
3498 **
3499 ** Won't batch when flush or commit_now is set.  Also won't batch when others are waiting on j_join_wait.
3500 **
3501 ** Note, we can't allow the journal_end to proceed while there are still writers in the log.
3502 */
3503 static int check_journal_end(struct reiserfs_transaction_handle *th,
3504                              struct super_block *p_s_sb, unsigned long nblocks,
3505                              int flags)
3506 {
3507
3508         time_t now;
3509         int flush = flags & FLUSH_ALL;
3510         int commit_now = flags & COMMIT_NOW;
3511         int wait_on_commit = flags & WAIT;
3512         struct reiserfs_journal_list *jl;
3513         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3514
3515         BUG_ON(!th->t_trans_id);
3516
3517         if (th->t_trans_id != journal->j_trans_id) {
3518                 reiserfs_panic(th->t_super,
3519                                "journal-1577: handle trans id %ld != current trans id %ld\n",
3520                                th->t_trans_id, journal->j_trans_id);
3521         }
3522
3523         journal->j_len_alloc -= (th->t_blocks_allocated - th->t_blocks_logged);
3524         if (atomic_read(&(journal->j_wcount)) > 0) {    /* <= 0 is allowed.  unmounting might not call begin */
3525                 atomic_dec(&(journal->j_wcount));
3526         }
3527
3528         /* BUG, deal with case where j_len is 0, but people previously freed blocks need to be released
3529          ** will be dealt with by next transaction that actually writes something, but should be taken
3530          ** care of in this trans
3531          */
3532         if (journal->j_len == 0) {
3533                 BUG();
3534         }
3535         /* if wcount > 0, and we are called to with flush or commit_now,
3536          ** we wait on j_join_wait.  We will wake up when the last writer has
3537          ** finished the transaction, and started it on its way to the disk.
3538          ** Then, we flush the commit or journal list, and just return 0
3539          ** because the rest of journal end was already done for this transaction.
3540          */
3541         if (atomic_read(&(journal->j_wcount)) > 0) {
3542                 if (flush || commit_now) {
3543                         unsigned trans_id;
3544
3545                         jl = journal->j_current_jl;
3546                         trans_id = jl->j_trans_id;
3547                         if (wait_on_commit)
3548                                 jl->j_state |= LIST_COMMIT_PENDING;
3549                         atomic_set(&(journal->j_jlock), 1);
3550                         if (flush) {
3551                                 journal->j_next_full_flush = 1;
3552                         }
3553                         unlock_journal(p_s_sb);
3554
3555                         /* sleep while the current transaction is still j_jlocked */
3556                         while (journal->j_trans_id == trans_id) {
3557                                 if (atomic_read(&journal->j_jlock)) {
3558                                         queue_log_writer(p_s_sb);
3559                                 } else {
3560                                         lock_journal(p_s_sb);
3561                                         if (journal->j_trans_id == trans_id) {
3562                                                 atomic_set(&(journal->j_jlock),
3563                                                            1);
3564                                         }
3565                                         unlock_journal(p_s_sb);
3566                                 }
3567                         }
3568                         if (journal->j_trans_id == trans_id) {
3569                                 BUG();
3570                         }
3571                         if (commit_now
3572                             && journal_list_still_alive(p_s_sb, trans_id)
3573                             && wait_on_commit) {
3574                                 flush_commit_list(p_s_sb, jl, 1);
3575                         }
3576                         return 0;
3577                 }
3578                 unlock_journal(p_s_sb);
3579                 return 0;
3580         }
3581
3582         /* deal with old transactions where we are the last writers */
3583         now = get_seconds();
3584         if ((now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3585                 commit_now = 1;
3586                 journal->j_next_async_flush = 1;
3587         }
3588         /* don't batch when someone is waiting on j_join_wait */
3589         /* don't batch when syncing the commit or flushing the whole trans */
3590         if (!(journal->j_must_wait > 0) && !(atomic_read(&(journal->j_jlock)))
3591             && !flush && !commit_now && (journal->j_len < journal->j_max_batch)
3592             && journal->j_len_alloc < journal->j_max_batch
3593             && journal->j_cnode_free > (journal->j_trans_max * 3)) {
3594                 journal->j_bcount++;
3595                 unlock_journal(p_s_sb);
3596                 return 0;
3597         }
3598
3599         if (journal->j_start > SB_ONDISK_JOURNAL_SIZE(p_s_sb)) {
3600                 reiserfs_panic(p_s_sb,
3601                                "journal-003: journal_end: j_start (%ld) is too high\n",
3602                                journal->j_start);
3603         }
3604         return 1;
3605 }
3606
3607 /*
3608 ** Does all the work that makes deleting blocks safe.
3609 ** when deleting a block mark BH_JNew, just remove it from the current transaction, clean it's buffer_head and move on.
3610 **
3611 ** otherwise:
3612 ** set a bit for the block in the journal bitmap.  That will prevent it from being allocated for unformatted nodes
3613 ** before this transaction has finished.
3614 **
3615 ** mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers.  That will prevent any old transactions with
3616 ** this block from trying to flush to the real location.  Since we aren't removing the cnode from the journal_list_hash,
3617 ** the block can't be reallocated yet.
3618 **
3619 ** Then remove it from the current transaction, decrementing any counters and filing it on the clean list.
3620 */
3621 int journal_mark_freed(struct reiserfs_transaction_handle *th,
3622                        struct super_block *p_s_sb, b_blocknr_t blocknr)
3623 {
3624         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3625         struct reiserfs_journal_cnode *cn = NULL;
3626         struct buffer_head *bh = NULL;
3627         struct reiserfs_list_bitmap *jb = NULL;
3628         int cleaned = 0;
3629         BUG_ON(!th->t_trans_id);
3630
3631         cn = get_journal_hash_dev(p_s_sb, journal->j_hash_table, blocknr);
3632         if (cn && cn->bh) {
3633                 bh = cn->bh;
3634                 get_bh(bh);
3635         }
3636         /* if it is journal new, we just remove it from this transaction */
3637         if (bh && buffer_journal_new(bh)) {
3638                 clear_buffer_journal_new(bh);
3639                 clear_prepared_bits(bh);
3640                 reiserfs_clean_and_file_buffer(bh);
3641                 cleaned = remove_from_transaction(p_s_sb, blocknr, cleaned);
3642         } else {
3643                 /* set the bit for this block in the journal bitmap for this transaction */
3644                 jb = journal->j_current_jl->j_list_bitmap;
3645                 if (!jb) {
3646                         reiserfs_panic(p_s_sb,
3647                                        "journal-1702: journal_mark_freed, journal_list_bitmap is NULL\n");
3648                 }
3649                 set_bit_in_list_bitmap(p_s_sb, blocknr, jb);
3650
3651                 /* Note, the entire while loop is not allowed to schedule.  */
3652
3653                 if (bh) {
3654                         clear_prepared_bits(bh);
3655                         reiserfs_clean_and_file_buffer(bh);
3656                 }
3657                 cleaned = remove_from_transaction(p_s_sb, blocknr, cleaned);
3658
3659                 /* find all older transactions with this block, make sure they don't try to write it out */
3660                 cn = get_journal_hash_dev(p_s_sb, journal->j_list_hash_table,
3661                                           blocknr);
3662                 while (cn) {
3663                         if (p_s_sb == cn->sb && blocknr == cn->blocknr) {
3664                                 set_bit(BLOCK_FREED, &cn->state);
3665                                 if (cn->bh) {
3666                                         if (!cleaned) {
3667                                                 /* remove_from_transaction will brelse the buffer if it was
3668                                                  ** in the current trans
3669                                                  */
3670                                                 clear_buffer_journal_dirty(cn->
3671                                                                            bh);
3672                                                 clear_buffer_dirty(cn->bh);
3673                                                 clear_buffer_journal_test(cn->
3674                                                                           bh);
3675                                                 cleaned = 1;
3676                                                 put_bh(cn->bh);
3677                                                 if (atomic_read
3678                                                     (&(cn->bh->b_count)) < 0) {
3679                                                         reiserfs_warning(p_s_sb,
3680                                                                          "journal-2138: cn->bh->b_count < 0");
3681                                                 }
3682                                         }
3683                                         if (cn->jlist) {        /* since we are clearing the bh, we MUST dec nonzerolen */
3684                                                 atomic_dec(&
3685                                                            (cn->jlist->
3686                                                             j_nonzerolen));
3687                                         }
3688                                         cn->bh = NULL;
3689                                 }
3690                         }
3691                         cn = cn->hnext;
3692                 }
3693         }
3694
3695         if (bh) {
3696                 put_bh(bh);     /* get_hash grabs the buffer */
3697                 if (atomic_read(&(bh->b_count)) < 0) {
3698                         reiserfs_warning(p_s_sb,
3699                                          "journal-2165: bh->b_count < 0");
3700                 }
3701         }
3702         return 0;
3703 }
3704
3705 void reiserfs_update_inode_transaction(struct inode *inode)
3706 {
3707         struct reiserfs_journal *journal = SB_JOURNAL(inode->i_sb);
3708         REISERFS_I(inode)->i_jl = journal->j_current_jl;
3709         REISERFS_I(inode)->i_trans_id = journal->j_trans_id;
3710 }
3711
3712 /*
3713  * returns -1 on error, 0 if no commits/barriers were done and 1
3714  * if a transaction was actually committed and the barrier was done
3715  */
3716 static int __commit_trans_jl(struct inode *inode, unsigned long id,
3717                              struct reiserfs_journal_list *jl)
3718 {
3719         struct reiserfs_transaction_handle th;
3720         struct super_block *sb = inode->i_sb;
3721         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3722         int ret = 0;
3723
3724         /* is it from the current transaction, or from an unknown transaction? */
3725         if (id == journal->j_trans_id) {
3726                 jl = journal->j_current_jl;
3727                 /* try to let other writers come in and grow this transaction */
3728                 let_transaction_grow(sb, id);
3729                 if (journal->j_trans_id != id) {
3730                         goto flush_commit_only;
3731                 }
3732
3733                 ret = journal_begin(&th, sb, 1);
3734                 if (ret)
3735                         return ret;
3736
3737                 /* someone might have ended this transaction while we joined */
3738                 if (journal->j_trans_id != id) {
3739                         reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3740                                                      1);
3741                         journal_mark_dirty(&th, sb, SB_BUFFER_WITH_SB(sb));
3742                         ret = journal_end(&th, sb, 1);
3743                         goto flush_commit_only;
3744                 }
3745
3746                 ret = journal_end_sync(&th, sb, 1);
3747                 if (!ret)
3748                         ret = 1;
3749
3750         } else {
3751                 /* this gets tricky, we have to make sure the journal list in
3752                  * the inode still exists.  We know the list is still around
3753                  * if we've got a larger transaction id than the oldest list
3754                  */
3755               flush_commit_only:
3756                 if (journal_list_still_alive(inode->i_sb, id)) {
3757                         /*
3758                          * we only set ret to 1 when we know for sure
3759                          * the barrier hasn't been started yet on the commit
3760                          * block.
3761                          */
3762                         if (atomic_read(&jl->j_commit_left) > 1)
3763                                 ret = 1;
3764                         flush_commit_list(sb, jl, 1);
3765                         if (journal->j_errno)
3766                                 ret = journal->j_errno;
3767                 }
3768         }
3769         /* otherwise the list is gone, and long since committed */
3770         return ret;
3771 }
3772
3773 int reiserfs_commit_for_inode(struct inode *inode)
3774 {
3775         unsigned long id = REISERFS_I(inode)->i_trans_id;
3776         struct reiserfs_journal_list *jl = REISERFS_I(inode)->i_jl;
3777
3778         /* for the whole inode, assume unset id means it was
3779          * changed in the current transaction.  More conservative
3780          */
3781         if (!id || !jl) {
3782                 reiserfs_update_inode_transaction(inode);
3783                 id = REISERFS_I(inode)->i_trans_id;
3784                 /* jl will be updated in __commit_trans_jl */
3785         }
3786
3787         return __commit_trans_jl(inode, id, jl);
3788 }
3789
3790 void reiserfs_restore_prepared_buffer(struct super_block *p_s_sb,
3791                                       struct buffer_head *bh)
3792 {
3793         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3794         PROC_INFO_INC(p_s_sb, journal.restore_prepared);
3795         if (!bh) {
3796                 return;
3797         }
3798         if (test_clear_buffer_journal_restore_dirty(bh) &&
3799             buffer_journal_dirty(bh)) {
3800                 struct reiserfs_journal_cnode *cn;
3801                 cn = get_journal_hash_dev(p_s_sb,
3802                                           journal->j_list_hash_table,
3803                                           bh->b_blocknr);
3804                 if (cn && can_dirty(cn)) {
3805                         set_buffer_journal_test(bh);
3806                         mark_buffer_dirty(bh);
3807                 }
3808         }
3809         clear_buffer_journal_prepared(bh);
3810 }
3811
3812 extern struct tree_balance *cur_tb;
3813 /*
3814 ** before we can change a metadata block, we have to make sure it won't
3815 ** be written to disk while we are altering it.  So, we must:
3816 ** clean it
3817 ** wait on it.
3818 **
3819 */
3820 int reiserfs_prepare_for_journal(struct super_block *p_s_sb,
3821                                  struct buffer_head *bh, int wait)
3822 {
3823         PROC_INFO_INC(p_s_sb, journal.prepare);
3824
3825         if (test_set_buffer_locked(bh)) {
3826                 if (!wait)
3827                         return 0;
3828                 lock_buffer(bh);
3829         }
3830         set_buffer_journal_prepared(bh);
3831         if (test_clear_buffer_dirty(bh) && buffer_journal_dirty(bh)) {
3832                 clear_buffer_journal_test(bh);
3833                 set_buffer_journal_restore_dirty(bh);
3834         }
3835         unlock_buffer(bh);
3836         return 1;
3837 }
3838
3839 static void flush_old_journal_lists(struct super_block *s)
3840 {
3841         struct reiserfs_journal *journal = SB_JOURNAL(s);
3842         struct reiserfs_journal_list *jl;
3843         struct list_head *entry;
3844         time_t now = get_seconds();
3845
3846         while (!list_empty(&journal->j_journal_list)) {
3847                 entry = journal->j_journal_list.next;
3848                 jl = JOURNAL_LIST_ENTRY(entry);
3849                 /* this check should always be run, to send old lists to disk */
3850                 if (jl->j_timestamp < (now - (JOURNAL_MAX_TRANS_AGE * 4))) {
3851                         flush_used_journal_lists(s, jl);
3852                 } else {
3853                         break;
3854                 }
3855         }
3856 }
3857
3858 /*
3859 ** long and ugly.  If flush, will not return until all commit
3860 ** blocks and all real buffers in the trans are on disk.
3861 ** If no_async, won't return until all commit blocks are on disk.
3862 **
3863 ** keep reading, there are comments as you go along
3864 **
3865 ** If the journal is aborted, we just clean up. Things like flushing
3866 ** journal lists, etc just won't happen.
3867 */
3868 static int do_journal_end(struct reiserfs_transaction_handle *th,
3869                           struct super_block *p_s_sb, unsigned long nblocks,
3870                           int flags)
3871 {
3872         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3873         struct reiserfs_journal_cnode *cn, *next, *jl_cn;
3874         struct reiserfs_journal_cnode *last_cn = NULL;
3875         struct reiserfs_journal_desc *desc;
3876         struct reiserfs_journal_commit *commit;
3877         struct buffer_head *c_bh;       /* commit bh */
3878         struct buffer_head *d_bh;       /* desc bh */
3879         int cur_write_start = 0;        /* start index of current log write */
3880         int old_start;
3881         int i;
3882         int flush;
3883         int wait_on_commit;
3884         struct reiserfs_journal_list *jl, *temp_jl;
3885         struct list_head *entry, *safe;
3886         unsigned long jindex;
3887         unsigned long commit_trans_id;
3888         int trans_half;
3889
3890         BUG_ON(th->t_refcount > 1);
3891         BUG_ON(!th->t_trans_id);
3892
3893         /* protect flush_older_commits from doing mistakes if the
3894            transaction ID counter gets overflowed.  */
3895         if (th->t_trans_id == ~0UL)
3896                 flags |= FLUSH_ALL | COMMIT_NOW | WAIT;
3897         flush = flags & FLUSH_ALL;
3898         wait_on_commit = flags & WAIT;
3899
3900         put_fs_excl();
3901         current->journal_info = th->t_handle_save;
3902         reiserfs_check_lock_depth(p_s_sb, "journal end");
3903         if (journal->j_len == 0) {
3904                 reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb),
3905                                              1);
3906                 journal_mark_dirty(th, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb));
3907         }
3908
3909         lock_journal(p_s_sb);
3910         if (journal->j_next_full_flush) {
3911                 flags |= FLUSH_ALL;
3912                 flush = 1;
3913         }
3914         if (journal->j_next_async_flush) {
3915                 flags |= COMMIT_NOW | WAIT;
3916                 wait_on_commit = 1;
3917         }
3918
3919         /* check_journal_end locks the journal, and unlocks if it does not return 1
3920          ** it tells us if we should continue with the journal_end, or just return
3921          */
3922         if (!check_journal_end(th, p_s_sb, nblocks, flags)) {
3923                 p_s_sb->s_dirt = 1;
3924                 wake_queued_writers(p_s_sb);
3925                 reiserfs_async_progress_wait(p_s_sb);
3926                 goto out;
3927         }
3928
3929         /* check_journal_end might set these, check again */
3930         if (journal->j_next_full_flush) {
3931                 flush = 1;
3932         }
3933
3934         /*
3935          ** j must wait means we have to flush the log blocks, and the real blocks for
3936          ** this transaction
3937          */
3938         if (journal->j_must_wait > 0) {
3939                 flush = 1;
3940         }
3941 #ifdef REISERFS_PREALLOCATE
3942         /* quota ops might need to nest, setup the journal_info pointer for them
3943          * and raise the refcount so that it is > 0. */
3944         current->journal_info = th;
3945         th->t_refcount++;
3946         reiserfs_discard_all_prealloc(th);      /* it should not involve new blocks into
3947                                                  * the transaction */
3948         th->t_refcount--;
3949         current->journal_info = th->t_handle_save;
3950 #endif
3951
3952         /* setup description block */
3953         d_bh =
3954             journal_getblk(p_s_sb,
3955                            SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
3956                            journal->j_start);
3957         set_buffer_uptodate(d_bh);
3958         desc = (struct reiserfs_journal_desc *)(d_bh)->b_data;
3959         memset(d_bh->b_data, 0, d_bh->b_size);
3960         memcpy(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8);
3961         set_desc_trans_id(desc, journal->j_trans_id);
3962
3963         /* setup commit block.  Don't write (keep it clean too) this one until after everyone else is written */
3964         c_bh = journal_getblk(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
3965                               ((journal->j_start + journal->j_len +
3966                                 1) % SB_ONDISK_JOURNAL_SIZE(p_s_sb)));
3967         commit = (struct reiserfs_journal_commit *)c_bh->b_data;
3968         memset(c_bh->b_data, 0, c_bh->b_size);
3969         set_commit_trans_id(commit, journal->j_trans_id);
3970         set_buffer_uptodate(c_bh);
3971
3972         /* init this journal list */
3973         jl = journal->j_current_jl;
3974
3975         /* we lock the commit before doing anything because
3976          * we want to make sure nobody tries to run flush_commit_list until
3977          * the new transaction is fully setup, and we've already flushed the
3978          * ordered bh list
3979          */
3980         down(&jl->j_commit_lock);
3981
3982         /* save the transaction id in case we need to commit it later */
3983         commit_trans_id = jl->j_trans_id;
3984
3985         atomic_set(&jl->j_older_commits_done, 0);
3986         jl->j_trans_id = journal->j_trans_id;
3987         jl->j_timestamp = journal->j_trans_start_time;
3988         jl->j_commit_bh = c_bh;
3989         jl->j_start = journal->j_start;
3990         jl->j_len = journal->j_len;
3991         atomic_set(&jl->j_nonzerolen, journal->j_len);
3992         atomic_set(&jl->j_commit_left, journal->j_len + 2);
3993         jl->j_realblock = NULL;
3994
3995         /* The ENTIRE FOR LOOP MUST not cause schedule to occur.
3996          **  for each real block, add it to the journal list hash,
3997          ** copy into real block index array in the commit or desc block
3998          */
3999         trans_half = journal_trans_half(p_s_sb->s_blocksize);
4000         for (i = 0, cn = journal->j_first; cn; cn = cn->next, i++) {
4001                 if (buffer_journaled(cn->bh)) {
4002                         jl_cn = get_cnode(p_s_sb);
4003                         if (!jl_cn) {
4004                                 reiserfs_panic(p_s_sb,
4005                                                "journal-1676, get_cnode returned NULL\n");
4006                         }
4007                         if (i == 0) {
4008                                 jl->j_realblock = jl_cn;
4009                         }
4010                         jl_cn->prev = last_cn;
4011                         jl_cn->next = NULL;
4012                         if (last_cn) {
4013                                 last_cn->next = jl_cn;
4014                         }
4015                         last_cn = jl_cn;
4016                         /* make sure the block we are trying to log is not a block
4017                            of journal or reserved area */
4018
4019                         if (is_block_in_log_or_reserved_area
4020                             (p_s_sb, cn->bh->b_blocknr)) {
4021                                 reiserfs_panic(p_s_sb,
4022                                                "journal-2332: Trying to log block %lu, which is a log block\n",
4023                                                cn->bh->b_blocknr);
4024                         }
4025                         jl_cn->blocknr = cn->bh->b_blocknr;
4026                         jl_cn->state = 0;
4027                         jl_cn->sb = p_s_sb;
4028                         jl_cn->bh = cn->bh;
4029                         jl_cn->jlist = jl;
4030                         insert_journal_hash(journal->j_list_hash_table, jl_cn);
4031                         if (i < trans_half) {
4032                                 desc->j_realblock[i] =
4033                                     cpu_to_le32(cn->bh->b_blocknr);
4034                         } else {
4035                                 commit->j_realblock[i - trans_half] =
4036                                     cpu_to_le32(cn->bh->b_blocknr);
4037                         }
4038                 } else {
4039                         i--;
4040                 }
4041         }
4042         set_desc_trans_len(desc, journal->j_len);
4043         set_desc_mount_id(desc, journal->j_mount_id);
4044         set_desc_trans_id(desc, journal->j_trans_id);
4045         set_commit_trans_len(commit, journal->j_len);
4046
4047         /* special check in case all buffers in the journal were marked for not logging */
4048         if (journal->j_len == 0) {
4049                 BUG();
4050         }
4051
4052         /* we're about to dirty all the log blocks, mark the description block
4053          * dirty now too.  Don't mark the commit block dirty until all the
4054          * others are on disk
4055          */
4056         mark_buffer_dirty(d_bh);
4057
4058         /* first data block is j_start + 1, so add one to cur_write_start wherever you use it */
4059         cur_write_start = journal->j_start;
4060         cn = journal->j_first;
4061         jindex = 1;             /* start at one so we don't get the desc again */
4062         while (cn) {
4063                 clear_buffer_journal_new(cn->bh);
4064                 /* copy all the real blocks into log area.  dirty log blocks */
4065                 if (buffer_journaled(cn->bh)) {
4066                         struct buffer_head *tmp_bh;
4067                         char *addr;
4068                         struct page *page;
4069                         tmp_bh =
4070                             journal_getblk(p_s_sb,
4071                                            SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
4072                                            ((cur_write_start +
4073                                              jindex) %
4074                                             SB_ONDISK_JOURNAL_SIZE(p_s_sb)));
4075                         set_buffer_uptodate(tmp_bh);
4076                         page = cn->bh->b_page;
4077                         addr = kmap(page);
4078                         memcpy(tmp_bh->b_data,
4079                                addr + offset_in_page(cn->bh->b_data),
4080                                cn->bh->b_size);
4081                         kunmap(page);
4082                         mark_buffer_dirty(tmp_bh);
4083                         jindex++;
4084                         set_buffer_journal_dirty(cn->bh);
4085                         clear_buffer_journaled(cn->bh);
4086                 } else {
4087                         /* JDirty cleared sometime during transaction.  don't log this one */
4088                         reiserfs_warning(p_s_sb,
4089                                          "journal-2048: do_journal_end: BAD, buffer in journal hash, but not JDirty!");
4090                         brelse(cn->bh);
4091                 }
4092                 next = cn->next;
4093                 free_cnode(p_s_sb, cn);
4094                 cn = next;
4095                 cond_resched();
4096         }
4097
4098         /* we are done  with both the c_bh and d_bh, but
4099          ** c_bh must be written after all other commit blocks,
4100          ** so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1.
4101          */
4102
4103         journal->j_current_jl = alloc_journal_list(p_s_sb);
4104
4105         /* now it is safe to insert this transaction on the main list */
4106         list_add_tail(&jl->j_list, &journal->j_journal_list);
4107         list_add_tail(&jl->j_working_list, &journal->j_working_list);
4108         journal->j_num_work_lists++;
4109
4110         /* reset journal values for the next transaction */
4111         old_start = journal->j_start;
4112         journal->j_start =
4113             (journal->j_start + journal->j_len +
4114              2) % SB_ONDISK_JOURNAL_SIZE(p_s_sb);
4115         atomic_set(&(journal->j_wcount), 0);
4116         journal->j_bcount = 0;
4117         journal->j_last = NULL;
4118         journal->j_first = NULL;
4119         journal->j_len = 0;
4120         journal->j_trans_start_time = 0;
4121         /* check for trans_id overflow */
4122         if (++journal->j_trans_id == 0)
4123                 journal->j_trans_id = 10;
4124         journal->j_current_jl->j_trans_id = journal->j_trans_id;
4125         journal->j_must_wait = 0;
4126         journal->j_len_alloc = 0;
4127         journal->j_next_full_flush = 0;
4128         journal->j_next_async_flush = 0;
4129         init_journal_hash(p_s_sb);
4130
4131         // make sure reiserfs_add_jh sees the new current_jl before we
4132         // write out the tails
4133         smp_mb();
4134
4135         /* tail conversion targets have to hit the disk before we end the
4136          * transaction.  Otherwise a later transaction might repack the tail
4137          * before this transaction commits, leaving the data block unflushed and
4138          * clean, if we crash before the later transaction commits, the data block
4139          * is lost.
4140          */
4141         if (!list_empty(&jl->j_tail_bh_list)) {
4142                 unlock_kernel();
4143                 write_ordered_buffers(&journal->j_dirty_buffers_lock,
4144                                       journal, jl, &jl->j_tail_bh_list);
4145                 lock_kernel();
4146         }
4147         if (!list_empty(&jl->j_tail_bh_list))
4148                 BUG();
4149         up(&jl->j_commit_lock);
4150
4151         /* honor the flush wishes from the caller, simple commits can
4152          ** be done outside the journal lock, they are done below
4153          **
4154          ** if we don't flush the commit list right now, we put it into
4155          ** the work queue so the people waiting on the async progress work
4156          ** queue don't wait for this proc to flush journal lists and such.
4157          */
4158         if (flush) {
4159                 flush_commit_list(p_s_sb, jl, 1);
4160                 flush_journal_list(p_s_sb, jl, 1);
4161         } else if (!(jl->j_state & LIST_COMMIT_PENDING))
4162                 queue_delayed_work(commit_wq, &journal->j_work, HZ / 10);
4163
4164         /* if the next transaction has any chance of wrapping, flush
4165          ** transactions that might get overwritten.  If any journal lists are very
4166          ** old flush them as well.
4167          */
4168       first_jl:
4169         list_for_each_safe(entry, safe, &journal->j_journal_list) {
4170                 temp_jl = JOURNAL_LIST_ENTRY(entry);
4171                 if (journal->j_start <= temp_jl->j_start) {
4172                         if ((journal->j_start + journal->j_trans_max + 1) >=
4173                             temp_jl->j_start) {
4174                                 flush_used_journal_lists(p_s_sb, temp_jl);
4175                                 goto first_jl;
4176                         } else if ((journal->j_start +
4177                                     journal->j_trans_max + 1) <
4178                                    SB_ONDISK_JOURNAL_SIZE(p_s_sb)) {
4179                                 /* if we don't cross into the next transaction and we don't
4180                                  * wrap, there is no way we can overlap any later transactions
4181                                  * break now
4182                                  */
4183                                 break;
4184                         }
4185                 } else if ((journal->j_start +
4186                             journal->j_trans_max + 1) >
4187                            SB_ONDISK_JOURNAL_SIZE(p_s_sb)) {
4188                         if (((journal->j_start + journal->j_trans_max + 1) %
4189                              SB_ONDISK_JOURNAL_SIZE(p_s_sb)) >=
4190                             temp_jl->j_start) {
4191                                 flush_used_journal_lists(p_s_sb, temp_jl);
4192                                 goto first_jl;
4193                         } else {
4194                                 /* we don't overlap anything from out start to the end of the
4195                                  * log, and our wrapped portion doesn't overlap anything at
4196                                  * the start of the log.  We can break
4197                                  */
4198                                 break;
4199                         }
4200                 }
4201         }
4202         flush_old_journal_lists(p_s_sb);
4203
4204         journal->j_current_jl->j_list_bitmap =
4205             get_list_bitmap(p_s_sb, journal->j_current_jl);
4206
4207         if (!(journal->j_current_jl->j_list_bitmap)) {
4208                 reiserfs_panic(p_s_sb,
4209                                "journal-1996: do_journal_end, could not get a list bitmap\n");
4210         }
4211
4212         atomic_set(&(journal->j_jlock), 0);
4213         unlock_journal(p_s_sb);
4214         /* wake up any body waiting to join. */
4215         clear_bit(J_WRITERS_QUEUED, &journal->j_state);
4216         wake_up(&(journal->j_join_wait));
4217
4218         if (!flush && wait_on_commit &&
4219             journal_list_still_alive(p_s_sb, commit_trans_id)) {
4220                 flush_commit_list(p_s_sb, jl, 1);
4221         }
4222       out:
4223         reiserfs_check_lock_depth(p_s_sb, "journal end2");
4224
4225         memset(th, 0, sizeof(*th));
4226         /* Re-set th->t_super, so we can properly keep track of how many
4227          * persistent transactions there are. We need to do this so if this
4228          * call is part of a failed restart_transaction, we can free it later */
4229         th->t_super = p_s_sb;
4230
4231         return journal->j_errno;
4232 }
4233
4234 static void __reiserfs_journal_abort_hard(struct super_block *sb)
4235 {
4236         struct reiserfs_journal *journal = SB_JOURNAL(sb);
4237         if (test_bit(J_ABORTED, &journal->j_state))
4238                 return;
4239
4240         printk(KERN_CRIT "REISERFS: Aborting journal for filesystem on %s\n",
4241                reiserfs_bdevname(sb));
4242
4243         sb->s_flags |= MS_RDONLY;
4244         set_bit(J_ABORTED, &journal->j_state);
4245
4246 #ifdef CONFIG_REISERFS_CHECK
4247         dump_stack();
4248 #endif
4249 }
4250
4251 static void __reiserfs_journal_abort_soft(struct super_block *sb, int errno)
4252 {
4253         struct reiserfs_journal *journal = SB_JOURNAL(sb);
4254         if (test_bit(J_ABORTED, &journal->j_state))
4255                 return;
4256
4257         if (!journal->j_errno)
4258                 journal->j_errno = errno;
4259
4260         __reiserfs_journal_abort_hard(sb);
4261 }
4262
4263 void reiserfs_journal_abort(struct super_block *sb, int errno)
4264 {
4265         return __reiserfs_journal_abort_soft(sb, errno);
4266 }