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

   1 /*
   2 ** Write ahead logging implementation copyright Chris Mason 2000
   3 **
   4 ** The background commits make this code very interelated, and
   5 ** overly complex.  I need to rethink things a bit....The major players:
   6 **
   7 ** journal_begin -- call with the number of blocks you expect to log.
   8 **                  If the current transaction is too
   9 **                  old, it will block until the current transaction is
  10 **                  finished, and then start a new one.
  11 **                  Usually, your transaction will get joined in with
  12 **                  previous ones for speed.
  13 **
  14 ** journal_join  -- same as journal_begin, but won't block on the current
  15 **                  transaction regardless of age.  Don't ever call
  16 **                  this.  Ever.  There are only two places it should be
  17 **                  called from, and they are both inside this file.
  18 **
  19 ** journal_mark_dirty -- adds blocks into this transaction.  clears any flags
  20 **                       that might make them get sent to disk
  21 **                       and then marks them BH_JDirty.  Puts the buffer head
  22 **                       into the current transaction hash.
  23 **
  24 ** journal_end -- if the current transaction is batchable, it does nothing
  25 **                   otherwise, it could do an async/synchronous commit, or
  26 **                   a full flush of all log and real blocks in the
  27 **                   transaction.
  28 **
  29 ** flush_old_commits -- if the current transaction is too old, it is ended and
  30 **                      commit blocks are sent to disk.  Forces commit blocks
  31 **                      to disk for all backgrounded commits that have been
  32 **                      around too long.
  33 **                   -- Note, if you call this as an immediate flush from
  34 **                      from within kupdate, it will ignore the immediate flag
  35 */
  36
  37 #include <linux/config.h>
  38 #include <asm/uaccess.h>
  39 #include <asm/system.h>
  40
  41 #include <linux/time.h>
  42 #include <asm/semaphore.h>
  43
  44 #include <linux/vmalloc.h>
  45 #include <linux/reiserfs_fs.h>
  46
  47 #include <linux/kernel.h>
  48 #include <linux/errno.h>
  49 #include <linux/fcntl.h>
  50 #include <linux/stat.h>
  51 #include <linux/string.h>
  52 #include <linux/smp_lock.h>
  53 #include <linux/buffer_head.h>
  54 #include <linux/workqueue.h>
  55 #include <linux/writeback.h>
  56 #include <linux/blkdev.h>
  57
  58 /* gets a struct reiserfs_journal_list * from a list head */
  59 #define JOURNAL_LIST_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
  60                                j_list))
  61 #define JOURNAL_WORK_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
  62                                j_working_list))
  63
  64 /* the number of mounted filesystems.  This is used to decide when to
  65 ** start and kill the commit workqueue
  66 */
  67 static int reiserfs_mounted_fs_count;
  68
  69 static struct workqueue_struct *commit_wq;
  70
  71 #define JOURNAL_TRANS_HALF 1018 /* must be correct to keep the desc and commit
  72                                    structs at 4k */
  73 #define BUFNR 64                /*read ahead */
  74
  75 /* cnode stat bits.  Move these into reiserfs_fs.h */
  76
  77 #define BLOCK_FREED 2           /* this block was freed, and can't be written.  */
  78 #define BLOCK_FREED_HOLDER 3    /* this block was freed during this transaction, and can't be written */
  79
  80 #define BLOCK_NEEDS_FLUSH 4     /* used in flush_journal_list */
  81 #define BLOCK_DIRTIED 5
  82
  83 /* journal list state bits */
  84 #define LIST_TOUCHED 1
  85 #define LIST_DIRTY   2
  86 #define LIST_COMMIT_PENDING  4  /* someone will commit this list */
  87
  88 /* flags for do_journal_end */
  89 #define FLUSH_ALL   1           /* flush commit and real blocks */
  90 #define COMMIT_NOW  2           /* end and commit this transaction */
  91 #define WAIT        4           /* wait for the log blocks to hit the disk */
  92
  93 static int do_journal_end(struct reiserfs_transaction_handle *,
  94                           struct super_block *, unsigned long nblocks,
  95                           int flags);
  96 static int flush_journal_list(struct super_block *s,
  97                               struct reiserfs_journal_list *jl, int flushall);
  98 static int flush_commit_list(struct super_block *s,
  99                              struct reiserfs_journal_list *jl, int flushall);
 100 static int can_dirty(struct reiserfs_journal_cnode *cn);
 101 static int journal_join(struct reiserfs_transaction_handle *th,
 102                         struct super_block *p_s_sb, unsigned long nblocks);
 103 static int release_journal_dev(struct super_block *super,
 104                                struct reiserfs_journal *journal);
 105 static int dirty_one_transaction(struct super_block *s,
 106                                  struct reiserfs_journal_list *jl);
 107 static void flush_async_commits(void *p);
 108 static void queue_log_writer(struct super_block *s);
 109
 110 /* values for join in do_journal_begin_r */
 111 enum {
 112         JBEGIN_REG = 0,         /* regular journal begin */
 113         JBEGIN_JOIN = 1,        /* join the running transaction if at all possible */
 114         JBEGIN_ABORT = 2,       /* called from cleanup code, ignores aborted flag */
 115 };
 116
 117 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
 118                               struct super_block *p_s_sb,
 119                               unsigned long nblocks, int join);
 120
 121 static void init_journal_hash(struct super_block *p_s_sb)
 122 {
 123         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 124         memset(journal->j_hash_table, 0,
 125                JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
 126 }
 127
 128 /*
 129 ** clears BH_Dirty and sticks the buffer on the clean list.  Called because I can't allow refile_buffer to
 130 ** make schedule happen after I've freed a block.  Look at remove_from_transaction and journal_mark_freed for
 131 ** more details.
 132 */
 133 static int reiserfs_clean_and_file_buffer(struct buffer_head *bh)
 134 {
 135         if (bh) {
 136                 clear_buffer_dirty(bh);
 137                 clear_buffer_journal_test(bh);
 138         }
 139         return 0;
 140 }
 141
 142 static void disable_barrier(struct super_block *s)
 143 {
 144         REISERFS_SB(s)->s_mount_opt &= ~(1 << REISERFS_BARRIER_FLUSH);
 145         printk("reiserfs: disabling flush barriers on %s\n",
 146                reiserfs_bdevname(s));
 147 }
 148
 149 static struct reiserfs_bitmap_node *allocate_bitmap_node(struct super_block
 150                                                          *p_s_sb)
 151 {
 152         struct reiserfs_bitmap_node *bn;
 153         static int id;
 154
 155         bn = reiserfs_kmalloc(sizeof(struct reiserfs_bitmap_node), GFP_NOFS,
 156                               p_s_sb);
 157         if (!bn) {
 158                 return NULL;
 159         }
 160         bn->data = reiserfs_kmalloc(p_s_sb->s_blocksize, GFP_NOFS, p_s_sb);
 161         if (!bn->data) {
 162                 reiserfs_kfree(bn, sizeof(struct reiserfs_bitmap_node), p_s_sb);
 163                 return NULL;
 164         }
 165         bn->id = id++;
 166         memset(bn->data, 0, p_s_sb->s_blocksize);
 167         INIT_LIST_HEAD(&bn->list);
 168         return bn;
 169 }
 170
 171 static struct reiserfs_bitmap_node *get_bitmap_node(struct super_block *p_s_sb)
 172 {
 173         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 174         struct reiserfs_bitmap_node *bn = NULL;
 175         struct list_head *entry = journal->j_bitmap_nodes.next;
 176
 177         journal->j_used_bitmap_nodes++;
 178       repeat:
 179
 180         if (entry != &journal->j_bitmap_nodes) {
 181                 bn = list_entry(entry, struct reiserfs_bitmap_node, list);
 182                 list_del(entry);
 183                 memset(bn->data, 0, p_s_sb->s_blocksize);
 184                 journal->j_free_bitmap_nodes--;
 185                 return bn;
 186         }
 187         bn = allocate_bitmap_node(p_s_sb);
 188         if (!bn) {
 189                 yield();
 190                 goto repeat;
 191         }
 192         return bn;
 193 }
 194 static inline void free_bitmap_node(struct super_block *p_s_sb,
 195                                     struct reiserfs_bitmap_node *bn)
 196 {
 197         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 198         journal->j_used_bitmap_nodes--;
 199         if (journal->j_free_bitmap_nodes > REISERFS_MAX_BITMAP_NODES) {
 200                 reiserfs_kfree(bn->data, p_s_sb->s_blocksize, p_s_sb);
 201                 reiserfs_kfree(bn, sizeof(struct reiserfs_bitmap_node), p_s_sb);
 202         } else {
 203                 list_add(&bn->list, &journal->j_bitmap_nodes);
 204                 journal->j_free_bitmap_nodes++;
 205         }
 206 }
 207
 208 static void allocate_bitmap_nodes(struct super_block *p_s_sb)
 209 {
 210         int i;
 211         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 212         struct reiserfs_bitmap_node *bn = NULL;
 213         for (i = 0; i < REISERFS_MIN_BITMAP_NODES; i++) {
 214                 bn = allocate_bitmap_node(p_s_sb);
 215                 if (bn) {
 216                         list_add(&bn->list, &journal->j_bitmap_nodes);
 217                         journal->j_free_bitmap_nodes++;
 218                 } else {
 219                         break;  // this is ok, we'll try again when more are needed
 220                 }
 221         }
 222 }
 223
 224 static int set_bit_in_list_bitmap(struct super_block *p_s_sb, int block,
 225                                   struct reiserfs_list_bitmap *jb)
 226 {
 227         int bmap_nr = block / (p_s_sb->s_blocksize << 3);
 228         int bit_nr = block % (p_s_sb->s_blocksize << 3);
 229
 230         if (!jb->bitmaps[bmap_nr]) {
 231                 jb->bitmaps[bmap_nr] = get_bitmap_node(p_s_sb);
 232         }
 233         set_bit(bit_nr, (unsigned long *)jb->bitmaps[bmap_nr]->data);
 234         return 0;
 235 }
 236
 237 static void cleanup_bitmap_list(struct super_block *p_s_sb,
 238                                 struct reiserfs_list_bitmap *jb)
 239 {
 240         int i;
 241         if (jb->bitmaps == NULL)
 242                 return;
 243
 244         for (i = 0; i < SB_BMAP_NR(p_s_sb); i++) {
 245                 if (jb->bitmaps[i]) {
 246                         free_bitmap_node(p_s_sb, jb->bitmaps[i]);
 247                         jb->bitmaps[i] = NULL;
 248                 }
 249         }
 250 }
 251
 252 /*
 253 ** only call this on FS unmount.
 254 */
 255 static int free_list_bitmaps(struct super_block *p_s_sb,
 256                              struct reiserfs_list_bitmap *jb_array)
 257 {
 258         int i;
 259         struct reiserfs_list_bitmap *jb;
 260         for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
 261                 jb = jb_array + i;
 262                 jb->journal_list = NULL;
 263                 cleanup_bitmap_list(p_s_sb, jb);
 264                 vfree(jb->bitmaps);
 265                 jb->bitmaps = NULL;
 266         }
 267         return 0;
 268 }
 269
 270 static int free_bitmap_nodes(struct super_block *p_s_sb)
 271 {
 272         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 273         struct list_head *next = journal->j_bitmap_nodes.next;
 274         struct reiserfs_bitmap_node *bn;
 275
 276         while (next != &journal->j_bitmap_nodes) {
 277                 bn = list_entry(next, struct reiserfs_bitmap_node, list);
 278                 list_del(next);
 279                 reiserfs_kfree(bn->data, p_s_sb->s_blocksize, p_s_sb);
 280                 reiserfs_kfree(bn, sizeof(struct reiserfs_bitmap_node), p_s_sb);
 281                 next = journal->j_bitmap_nodes.next;
 282                 journal->j_free_bitmap_nodes--;
 283         }
 284
 285         return 0;
 286 }
 287
 288 /*
 289 ** get memory for JOURNAL_NUM_BITMAPS worth of bitmaps.
 290 ** jb_array is the array to be filled in.
 291 */
 292 int reiserfs_allocate_list_bitmaps(struct super_block *p_s_sb,
 293                                    struct reiserfs_list_bitmap *jb_array,
 294                                    int bmap_nr)
 295 {
 296         int i;
 297         int failed = 0;
 298         struct reiserfs_list_bitmap *jb;
 299         int mem = bmap_nr * sizeof(struct reiserfs_bitmap_node *);
 300
 301         for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
 302                 jb = jb_array + i;
 303                 jb->journal_list = NULL;
 304                 jb->bitmaps = vmalloc(mem);
 305                 if (!jb->bitmaps) {
 306                         reiserfs_warning(p_s_sb,
 307                                          "clm-2000, unable to allocate bitmaps for journal lists");
 308                         failed = 1;
 309                         break;
 310                 }
 311                 memset(jb->bitmaps, 0, mem);
 312         }
 313         if (failed) {
 314                 free_list_bitmaps(p_s_sb, jb_array);
 315                 return -1;
 316         }
 317         return 0;
 318 }
 319
 320 /*
 321 ** find an available list bitmap.  If you can't find one, flush a commit list
 322 ** and try again
 323 */
 324 static struct reiserfs_list_bitmap *get_list_bitmap(struct super_block *p_s_sb,
 325                                                     struct reiserfs_journal_list
 326                                                     *jl)
 327 {
 328         int i, j;
 329         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 330         struct reiserfs_list_bitmap *jb = NULL;
 331
 332         for (j = 0; j < (JOURNAL_NUM_BITMAPS * 3); j++) {
 333                 i = journal->j_list_bitmap_index;
 334                 journal->j_list_bitmap_index = (i + 1) % JOURNAL_NUM_BITMAPS;
 335                 jb = journal->j_list_bitmap + i;
 336                 if (journal->j_list_bitmap[i].journal_list) {
 337                         flush_commit_list(p_s_sb,
 338                                           journal->j_list_bitmap[i].
 339                                           journal_list, 1);
 340                         if (!journal->j_list_bitmap[i].journal_list) {
 341                                 break;
 342                         }
 343                 } else {
 344                         break;
 345                 }
 346         }
 347         if (jb->journal_list) { /* double check to make sure if flushed correctly */
 348                 return NULL;
 349         }
 350         jb->journal_list = jl;
 351         return jb;
 352 }
 353
 354 /*
 355 ** allocates a new chunk of X nodes, and links them all together as a list.
 356 ** Uses the cnode->next and cnode->prev pointers
 357 ** returns NULL on failure
 358 */
 359 static struct reiserfs_journal_cnode *allocate_cnodes(int num_cnodes)
 360 {
 361         struct reiserfs_journal_cnode *head;
 362         int i;
 363         if (num_cnodes <= 0) {
 364                 return NULL;
 365         }
 366         head = vmalloc(num_cnodes * sizeof(struct reiserfs_journal_cnode));
 367         if (!head) {
 368                 return NULL;
 369         }
 370         memset(head, 0, num_cnodes * sizeof(struct reiserfs_journal_cnode));
 371         head[0].prev = NULL;
 372         head[0].next = head + 1;
 373         for (i = 1; i < num_cnodes; i++) {
 374                 head[i].prev = head + (i - 1);
 375                 head[i].next = head + (i + 1);  /* if last one, overwrite it after the if */
 376         }
 377         head[num_cnodes - 1].next = NULL;
 378         return head;
 379 }
 380
 381 /*
 382 ** pulls a cnode off the free list, or returns NULL on failure
 383 */
 384 static struct reiserfs_journal_cnode *get_cnode(struct super_block *p_s_sb)
 385 {
 386         struct reiserfs_journal_cnode *cn;
 387         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 388
 389         reiserfs_check_lock_depth(p_s_sb, "get_cnode");
 390
 391         if (journal->j_cnode_free <= 0) {
 392                 return NULL;
 393         }
 394         journal->j_cnode_used++;
 395         journal->j_cnode_free--;
 396         cn = journal->j_cnode_free_list;
 397         if (!cn) {
 398                 return cn;
 399         }
 400         if (cn->next) {
 401                 cn->next->prev = NULL;
 402         }
 403         journal->j_cnode_free_list = cn->next;
 404         memset(cn, 0, sizeof(struct reiserfs_journal_cnode));
 405         return cn;
 406 }
 407
 408 /*
 409 ** returns a cnode to the free list
 410 */
 411 static void free_cnode(struct super_block *p_s_sb,
 412                        struct reiserfs_journal_cnode *cn)
 413 {
 414         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 415
 416         reiserfs_check_lock_depth(p_s_sb, "free_cnode");
 417
 418         journal->j_cnode_used--;
 419         journal->j_cnode_free++;
 420         /* memset(cn, 0, sizeof(struct reiserfs_journal_cnode)) ; */
 421         cn->next = journal->j_cnode_free_list;
 422         if (journal->j_cnode_free_list) {
 423                 journal->j_cnode_free_list->prev = cn;
 424         }
 425         cn->prev = NULL;        /* not needed with the memset, but I might kill the memset, and forget to do this */
 426         journal->j_cnode_free_list = cn;
 427 }
 428
 429 static void clear_prepared_bits(struct buffer_head *bh)
 430 {
 431         clear_buffer_journal_prepared(bh);
 432         clear_buffer_journal_restore_dirty(bh);
 433 }
 434
 435 /* utility function to force a BUG if it is called without the big
 436 ** kernel lock held.  caller is the string printed just before calling BUG()
 437 */
 438 void reiserfs_check_lock_depth(struct super_block *sb, char *caller)
 439 {
 440 #ifdef CONFIG_SMP
 441         if (current->lock_depth < 0) {
 442                 reiserfs_panic(sb, "%s called without kernel lock held",
 443                                caller);
 444         }
 445 #else
 446         ;
 447 #endif
 448 }
 449
 450 /* return a cnode with same dev, block number and size in table, or null if not found */
 451 static inline struct reiserfs_journal_cnode *get_journal_hash_dev(struct
 452                                                                   super_block
 453                                                                   *sb,
 454                                                                   struct
 455                                                                   reiserfs_journal_cnode
 456                                                                   **table,
 457                                                                   long bl)
 458 {
 459         struct reiserfs_journal_cnode *cn;
 460         cn = journal_hash(table, sb, bl);
 461         while (cn) {
 462                 if (cn->blocknr == bl && cn->sb == sb)
 463                         return cn;
 464                 cn = cn->hnext;
 465         }
 466         return (struct reiserfs_journal_cnode *)0;
 467 }
 468
 469 /*
 470 ** this actually means 'can this block be reallocated yet?'.  If you set search_all, a block can only be allocated
 471 ** if it is not in the current transaction, was not freed by the current transaction, and has no chance of ever
 472 ** being overwritten by a replay after crashing.
 473 **
 474 ** If you don't set search_all, a block can only be allocated if it is not in the current transaction.  Since deleting
 475 ** a block removes it from the current transaction, this case should never happen.  If you don't set search_all, make
 476 ** sure you never write the block without logging it.
 477 **
 478 ** next_zero_bit is a suggestion about the next block to try for find_forward.
 479 ** when bl is rejected because it is set in a journal list bitmap, we search
 480 ** for the next zero bit in the bitmap that rejected bl.  Then, we return that
 481 ** through next_zero_bit for find_forward to try.
 482 **
 483 ** Just because we return something in next_zero_bit does not mean we won't
 484 ** reject it on the next call to reiserfs_in_journal
 485 **
 486 */
 487 int reiserfs_in_journal(struct super_block *p_s_sb,
 488                         int bmap_nr, int bit_nr, int search_all,
 489                         b_blocknr_t * next_zero_bit)
 490 {
 491         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 492         struct reiserfs_journal_cnode *cn;
 493         struct reiserfs_list_bitmap *jb;
 494         int i;
 495         unsigned long bl;
 496
 497         *next_zero_bit = 0;     /* always start this at zero. */
 498
 499         PROC_INFO_INC(p_s_sb, journal.in_journal);
 500         /* If we aren't doing a search_all, this is a metablock, and it will be logged before use.
 501          ** if we crash before the transaction that freed it commits,  this transaction won't
 502          ** have committed either, and the block will never be written
 503          */
 504         if (search_all) {
 505                 for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
 506                         PROC_INFO_INC(p_s_sb, journal.in_journal_bitmap);
 507                         jb = journal->j_list_bitmap + i;
 508                         if (jb->journal_list && jb->bitmaps[bmap_nr] &&
 509                             test_bit(bit_nr,
 510                                      (unsigned long *)jb->bitmaps[bmap_nr]->
 511                                      data)) {
 512                                 *next_zero_bit =
 513                                     find_next_zero_bit((unsigned long *)
 514                                                        (jb->bitmaps[bmap_nr]->
 515                                                         data),
 516                                                        p_s_sb->s_blocksize << 3,
 517                                                        bit_nr + 1);
 518                                 return 1;
 519                         }
 520                 }
 521         }
 522
 523         bl = bmap_nr * (p_s_sb->s_blocksize << 3) + bit_nr;
 524         /* is it in any old transactions? */
 525         if (search_all
 526             && (cn =
 527                 get_journal_hash_dev(p_s_sb, journal->j_list_hash_table, bl))) {
 528                 return 1;
 529         }
 530
 531         /* is it in the current transaction.  This should never happen */
 532         if ((cn = get_journal_hash_dev(p_s_sb, journal->j_hash_table, bl))) {
 533                 BUG();
 534                 return 1;
 535         }
 536
 537         PROC_INFO_INC(p_s_sb, journal.in_journal_reusable);
 538         /* safe for reuse */
 539         return 0;
 540 }
 541
 542 /* insert cn into table
 543 */
 544 static inline void insert_journal_hash(struct reiserfs_journal_cnode **table,
 545                                        struct reiserfs_journal_cnode *cn)
 546 {
 547         struct reiserfs_journal_cnode *cn_orig;
 548
 549         cn_orig = journal_hash(table, cn->sb, cn->blocknr);
 550         cn->hnext = cn_orig;
 551         cn->hprev = NULL;
 552         if (cn_orig) {
 553                 cn_orig->hprev = cn;
 554         }
 555         journal_hash(table, cn->sb, cn->blocknr) = cn;
 556 }
 557
 558 /* lock the current transaction */
 559 static inline void lock_journal(struct super_block *p_s_sb)
 560 {
 561         PROC_INFO_INC(p_s_sb, journal.lock_journal);
 562         down(&SB_JOURNAL(p_s_sb)->j_lock);
 563 }
 564
 565 /* unlock the current transaction */
 566 static inline void unlock_journal(struct super_block *p_s_sb)
 567 {
 568         up(&SB_JOURNAL(p_s_sb)->j_lock);
 569 }
 570
 571 static inline void get_journal_list(struct reiserfs_journal_list *jl)
 572 {
 573         jl->j_refcount++;
 574 }
 575
 576 static inline void put_journal_list(struct super_block *s,
 577                                     struct reiserfs_journal_list *jl)
 578 {
 579         if (jl->j_refcount < 1) {
 580                 reiserfs_panic(s, "trans id %lu, refcount at %d",
 581                                jl->j_trans_id, jl->j_refcount);
 582         }
 583         if (--jl->j_refcount == 0)
 584                 reiserfs_kfree(jl, sizeof(struct reiserfs_journal_list), s);
 585 }
 586
 587 /*
 588 ** this used to be much more involved, and I'm keeping it just in case things get ugly again.
 589 ** it gets called by flush_commit_list, and cleans up any data stored about blocks freed during a
 590 ** transaction.
 591 */
 592 static void cleanup_freed_for_journal_list(struct super_block *p_s_sb,
 593                                            struct reiserfs_journal_list *jl)
 594 {
 595
 596         struct reiserfs_list_bitmap *jb = jl->j_list_bitmap;
 597         if (jb) {
 598                 cleanup_bitmap_list(p_s_sb, jb);
 599         }
 600         jl->j_list_bitmap->journal_list = NULL;
 601         jl->j_list_bitmap = NULL;
 602 }
 603
 604 static int journal_list_still_alive(struct super_block *s,
 605                                     unsigned long trans_id)
 606 {
 607         struct reiserfs_journal *journal = SB_JOURNAL(s);
 608         struct list_head *entry = &journal->j_journal_list;
 609         struct reiserfs_journal_list *jl;
 610
 611         if (!list_empty(entry)) {
 612                 jl = JOURNAL_LIST_ENTRY(entry->next);
 613                 if (jl->j_trans_id <= trans_id) {
 614                         return 1;
 615                 }
 616         }
 617         return 0;
 618 }
 619
 620 static void reiserfs_end_buffer_io_sync(struct buffer_head *bh, int uptodate)
 621 {
 622         char b[BDEVNAME_SIZE];
 623
 624         if (buffer_journaled(bh)) {
 625                 reiserfs_warning(NULL,
 626                                  "clm-2084: pinned buffer %lu:%s sent to disk",
 627                                  bh->b_blocknr, bdevname(bh->b_bdev, b));
 628         }
 629         if (uptodate)
 630                 set_buffer_uptodate(bh);
 631         else
 632                 clear_buffer_uptodate(bh);
 633         unlock_buffer(bh);
 634         put_bh(bh);
 635 }
 636
 637 static void reiserfs_end_ordered_io(struct buffer_head *bh, int uptodate)
 638 {
 639         if (uptodate)
 640                 set_buffer_uptodate(bh);
 641         else
 642                 clear_buffer_uptodate(bh);
 643         unlock_buffer(bh);
 644         put_bh(bh);
 645 }
 646
 647 static void submit_logged_buffer(struct buffer_head *bh)
 648 {
 649         get_bh(bh);
 650         bh->b_end_io = reiserfs_end_buffer_io_sync;
 651         clear_buffer_journal_new(bh);
 652         clear_buffer_dirty(bh);
 653         if (!test_clear_buffer_journal_test(bh))
 654                 BUG();
 655         if (!buffer_uptodate(bh))
 656                 BUG();
 657         submit_bh(WRITE, bh);
 658 }
 659
 660 static void submit_ordered_buffer(struct buffer_head *bh)
 661 {
 662         get_bh(bh);
 663         bh->b_end_io = reiserfs_end_ordered_io;
 664         clear_buffer_dirty(bh);
 665         if (!buffer_uptodate(bh))
 666                 BUG();
 667         submit_bh(WRITE, bh);
 668 }
 669
 670 static int submit_barrier_buffer(struct buffer_head *bh)
 671 {
 672         get_bh(bh);
 673         bh->b_end_io = reiserfs_end_ordered_io;
 674         clear_buffer_dirty(bh);
 675         if (!buffer_uptodate(bh))
 676                 BUG();
 677         return submit_bh(WRITE_BARRIER, bh);
 678 }
 679
 680 static void check_barrier_completion(struct super_block *s,
 681                                      struct buffer_head *bh)
 682 {
 683         if (buffer_eopnotsupp(bh)) {
 684                 clear_buffer_eopnotsupp(bh);
 685                 disable_barrier(s);
 686                 set_buffer_uptodate(bh);
 687                 set_buffer_dirty(bh);
 688                 sync_dirty_buffer(bh);
 689         }
 690 }
 691
 692 #define CHUNK_SIZE 32
 693 struct buffer_chunk {
 694         struct buffer_head *bh[CHUNK_SIZE];
 695         int nr;
 696 };
 697
 698 static void write_chunk(struct buffer_chunk *chunk)
 699 {
 700         int i;
 701         get_fs_excl();
 702         for (i = 0; i < chunk->nr; i++) {
 703                 submit_logged_buffer(chunk->bh[i]);
 704         }
 705         chunk->nr = 0;
 706         put_fs_excl();
 707 }
 708
 709 static void write_ordered_chunk(struct buffer_chunk *chunk)
 710 {
 711         int i;
 712         get_fs_excl();
 713         for (i = 0; i < chunk->nr; i++) {
 714                 submit_ordered_buffer(chunk->bh[i]);
 715         }
 716         chunk->nr = 0;
 717         put_fs_excl();
 718 }
 719
 720 static int add_to_chunk(struct buffer_chunk *chunk, struct buffer_head *bh,
 721                         spinlock_t * lock, void (fn) (struct buffer_chunk *))
 722 {
 723         int ret = 0;
 724         if (chunk->nr >= CHUNK_SIZE)
 725                 BUG();
 726         chunk->bh[chunk->nr++] = bh;
 727         if (chunk->nr >= CHUNK_SIZE) {
 728                 ret = 1;
 729                 if (lock)
 730                         spin_unlock(lock);
 731                 fn(chunk);
 732                 if (lock)
 733                         spin_lock(lock);
 734         }
 735         return ret;
 736 }
 737
 738 static atomic_t nr_reiserfs_jh = ATOMIC_INIT(0);
 739 static struct reiserfs_jh *alloc_jh(void)
 740 {
 741         struct reiserfs_jh *jh;
 742         while (1) {
 743                 jh = kmalloc(sizeof(*jh), GFP_NOFS);
 744                 if (jh) {
 745                         atomic_inc(&nr_reiserfs_jh);
 746                         return jh;
 747                 }
 748                 yield();
 749         }
 750 }
 751
 752 /*
 753  * we want to free the jh when the buffer has been written
 754  * and waited on
 755  */
 756 void reiserfs_free_jh(struct buffer_head *bh)
 757 {
 758         struct reiserfs_jh *jh;
 759
 760         jh = bh->b_private;
 761         if (jh) {
 762                 bh->b_private = NULL;
 763                 jh->bh = NULL;
 764                 list_del_init(&jh->list);
 765                 kfree(jh);
 766                 if (atomic_read(&nr_reiserfs_jh) <= 0)
 767                         BUG();
 768                 atomic_dec(&nr_reiserfs_jh);
 769                 put_bh(bh);
 770         }
 771 }
 772
 773 static inline int __add_jh(struct reiserfs_journal *j, struct buffer_head *bh,
 774                            int tail)
 775 {
 776         struct reiserfs_jh *jh;
 777
 778         if (bh->b_private) {
 779                 spin_lock(&j->j_dirty_buffers_lock);
 780                 if (!bh->b_private) {
 781                         spin_unlock(&j->j_dirty_buffers_lock);
 782                         goto no_jh;
 783                 }
 784                 jh = bh->b_private;
 785                 list_del_init(&jh->list);
 786         } else {
 787               no_jh:
 788                 get_bh(bh);
 789                 jh = alloc_jh();
 790                 spin_lock(&j->j_dirty_buffers_lock);
 791                 /* buffer must be locked for __add_jh, should be able to have
 792                  * two adds at the same time
 793                  */
 794                 if (bh->b_private)
 795                         BUG();
 796                 jh->bh = bh;
 797                 bh->b_private = jh;
 798         }
 799         jh->jl = j->j_current_jl;
 800         if (tail)
 801                 list_add_tail(&jh->list, &jh->jl->j_tail_bh_list);
 802         else {
 803                 list_add_tail(&jh->list, &jh->jl->j_bh_list);
 804         }
 805         spin_unlock(&j->j_dirty_buffers_lock);
 806         return 0;
 807 }
 808
 809 int reiserfs_add_tail_list(struct inode *inode, struct buffer_head *bh)
 810 {
 811         return __add_jh(SB_JOURNAL(inode->i_sb), bh, 1);
 812 }
 813 int reiserfs_add_ordered_list(struct inode *inode, struct buffer_head *bh)
 814 {
 815         return __add_jh(SB_JOURNAL(inode->i_sb), bh, 0);
 816 }
 817
 818 #define JH_ENTRY(l) list_entry((l), struct reiserfs_jh, list)
 819 static int write_ordered_buffers(spinlock_t * lock,
 820                                  struct reiserfs_journal *j,
 821                                  struct reiserfs_journal_list *jl,
 822                                  struct list_head *list)
 823 {
 824         struct buffer_head *bh;
 825         struct reiserfs_jh *jh;
 826         int ret = j->j_errno;
 827         struct buffer_chunk chunk;
 828         struct list_head tmp;
 829         INIT_LIST_HEAD(&tmp);
 830
 831         chunk.nr = 0;
 832         spin_lock(lock);
 833         while (!list_empty(list)) {
 834                 jh = JH_ENTRY(list->next);
 835                 bh = jh->bh;
 836                 get_bh(bh);
 837                 if (test_set_buffer_locked(bh)) {
 838                         if (!buffer_dirty(bh)) {
 839                                 list_del_init(&jh->list);
 840                                 list_add(&jh->list, &tmp);
 841                                 goto loop_next;
 842                         }
 843                         spin_unlock(lock);
 844                         if (chunk.nr)
 845                                 write_ordered_chunk(&chunk);
 846                         wait_on_buffer(bh);
 847                         cond_resched();
 848                         spin_lock(lock);
 849                         goto loop_next;
 850                 }
 851                 if (buffer_dirty(bh)) {
 852                         list_del_init(&jh->list);
 853                         list_add(&jh->list, &tmp);
 854                         add_to_chunk(&chunk, bh, lock, write_ordered_chunk);
 855                 } else {
 856                         reiserfs_free_jh(bh);
 857                         unlock_buffer(bh);
 858                 }
 859               loop_next:
 860                 put_bh(bh);
 861                 cond_resched_lock(lock);
 862         }
 863         if (chunk.nr) {
 864                 spin_unlock(lock);
 865                 write_ordered_chunk(&chunk);
 866                 spin_lock(lock);
 867         }
 868         while (!list_empty(&tmp)) {
 869                 jh = JH_ENTRY(tmp.prev);
 870                 bh = jh->bh;
 871                 get_bh(bh);
 872                 reiserfs_free_jh(bh);
 873
 874                 if (buffer_locked(bh)) {
 875                         spin_unlock(lock);
 876                         wait_on_buffer(bh);
 877                         spin_lock(lock);
 878                 }
 879                 if (!buffer_uptodate(bh)) {
 880                         ret = -EIO;
 881                 }
 882                 put_bh(bh);
 883                 cond_resched_lock(lock);
 884         }
 885         spin_unlock(lock);
 886         return ret;
 887 }
 888
 889 static int flush_older_commits(struct super_block *s,
 890                                struct reiserfs_journal_list *jl)
 891 {
 892         struct reiserfs_journal *journal = SB_JOURNAL(s);
 893         struct reiserfs_journal_list *other_jl;
 894         struct reiserfs_journal_list *first_jl;
 895         struct list_head *entry;
 896         unsigned long trans_id = jl->j_trans_id;
 897         unsigned long other_trans_id;
 898         unsigned long first_trans_id;
 899
 900       find_first:
 901         /*
 902          * first we walk backwards to find the oldest uncommitted transation
 903          */
 904         first_jl = jl;
 905         entry = jl->j_list.prev;
 906         while (1) {
 907                 other_jl = JOURNAL_LIST_ENTRY(entry);
 908                 if (entry == &journal->j_journal_list ||
 909                     atomic_read(&other_jl->j_older_commits_done))
 910                         break;
 911
 912                 first_jl = other_jl;
 913                 entry = other_jl->j_list.prev;
 914         }
 915
 916         /* if we didn't find any older uncommitted transactions, return now */
 917         if (first_jl == jl) {
 918                 return 0;
 919         }
 920
 921         first_trans_id = first_jl->j_trans_id;
 922
 923         entry = &first_jl->j_list;
 924         while (1) {
 925                 other_jl = JOURNAL_LIST_ENTRY(entry);
 926                 other_trans_id = other_jl->j_trans_id;
 927
 928                 if (other_trans_id < trans_id) {
 929                         if (atomic_read(&other_jl->j_commit_left) != 0) {
 930                                 flush_commit_list(s, other_jl, 0);
 931
 932                                 /* list we were called with is gone, return */
 933                                 if (!journal_list_still_alive(s, trans_id))
 934                                         return 1;
 935
 936                                 /* the one we just flushed is gone, this means all
 937                                  * older lists are also gone, so first_jl is no longer
 938                                  * valid either.  Go back to the beginning.
 939                                  */
 940                                 if (!journal_list_still_alive
 941                                     (s, other_trans_id)) {
 942                                         goto find_first;
 943                                 }
 944                         }
 945                         entry = entry->next;
 946                         if (entry == &journal->j_journal_list)
 947                                 return 0;
 948                 } else {
 949                         return 0;
 950                 }
 951         }
 952         return 0;
 953 }
 954 int reiserfs_async_progress_wait(struct super_block *s)
 955 {
 956         DEFINE_WAIT(wait);
 957         struct reiserfs_journal *j = SB_JOURNAL(s);
 958         if (atomic_read(&j->j_async_throttle))
 959                 blk_congestion_wait(WRITE, HZ / 10);
 960         return 0;
 961 }
 962
 963 /*
 964 ** if this journal list still has commit blocks unflushed, send them to disk.
 965 **
 966 ** log areas must be flushed in order (transaction 2 can't commit before transaction 1)
 967 ** Before the commit block can by written, every other log block must be safely on disk
 968 **
 969 */
 970 static int flush_commit_list(struct super_block *s,
 971                              struct reiserfs_journal_list *jl, int flushall)
 972 {
 973         int i;
 974         int bn;
 975         struct buffer_head *tbh = NULL;
 976         unsigned long trans_id = jl->j_trans_id;
 977         struct reiserfs_journal *journal = SB_JOURNAL(s);
 978         int barrier = 0;
 979         int retval = 0;
 980
 981         reiserfs_check_lock_depth(s, "flush_commit_list");
 982
 983         if (atomic_read(&jl->j_older_commits_done)) {
 984                 return 0;
 985         }
 986
 987         get_fs_excl();
 988
 989         /* before we can put our commit blocks on disk, we have to make sure everyone older than
 990          ** us is on disk too
 991          */
 992         BUG_ON(jl->j_len <= 0);
 993         BUG_ON(trans_id == journal->j_trans_id);
 994
 995         get_journal_list(jl);
 996         if (flushall) {
 997                 if (flush_older_commits(s, jl) == 1) {
 998                         /* list disappeared during flush_older_commits.  return */
 999                         goto put_jl;
1000                 }
1001         }
1002
1003         /* make sure nobody is trying to flush this one at the same time */
1004         down(&jl->j_commit_lock);
1005         if (!journal_list_still_alive(s, trans_id)) {
1006                 up(&jl->j_commit_lock);
1007                 goto put_jl;
1008         }
1009         BUG_ON(jl->j_trans_id == 0);
1010
1011         /* this commit is done, exit */
1012         if (atomic_read(&(jl->j_commit_left)) <= 0) {
1013                 if (flushall) {
1014                         atomic_set(&(jl->j_older_commits_done), 1);
1015                 }
1016                 up(&jl->j_commit_lock);
1017                 goto put_jl;
1018         }
1019
1020         if (!list_empty(&jl->j_bh_list)) {
1021                 unlock_kernel();
1022                 write_ordered_buffers(&journal->j_dirty_buffers_lock,
1023                                       journal, jl, &jl->j_bh_list);
1024                 lock_kernel();
1025         }
1026         BUG_ON(!list_empty(&jl->j_bh_list));
1027         /*
1028          * for the description block and all the log blocks, submit any buffers
1029          * that haven't already reached the disk
1030          */
1031         atomic_inc(&journal->j_async_throttle);
1032         for (i = 0; i < (jl->j_len + 1); i++) {
1033                 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + (jl->j_start + i) %
1034                     SB_ONDISK_JOURNAL_SIZE(s);
1035                 tbh = journal_find_get_block(s, bn);
1036                 if (buffer_dirty(tbh))  /* redundant, ll_rw_block() checks */
1037                         ll_rw_block(SWRITE, 1, &tbh);
1038                 put_bh(tbh);
1039         }
1040         atomic_dec(&journal->j_async_throttle);
1041
1042         /* We're skipping the commit if there's an error */
1043         if (retval || reiserfs_is_journal_aborted(journal))
1044                 barrier = 0;
1045
1046         /* wait on everything written so far before writing the commit
1047          * if we are in barrier mode, send the commit down now
1048          */
1049         barrier = reiserfs_barrier_flush(s);
1050         if (barrier) {
1051                 int ret;
1052                 lock_buffer(jl->j_commit_bh);
1053                 ret = submit_barrier_buffer(jl->j_commit_bh);
1054                 if (ret == -EOPNOTSUPP) {
1055                         set_buffer_uptodate(jl->j_commit_bh);
1056                         disable_barrier(s);
1057                         barrier = 0;
1058                 }
1059         }
1060         for (i = 0; i < (jl->j_len + 1); i++) {
1061                 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) +
1062                     (jl->j_start + i) % SB_ONDISK_JOURNAL_SIZE(s);
1063                 tbh = journal_find_get_block(s, bn);
1064                 wait_on_buffer(tbh);
1065                 // since we're using ll_rw_blk above, it might have skipped over
1066                 // a locked buffer.  Double check here
1067                 //
1068                 if (buffer_dirty(tbh))  /* redundant, sync_dirty_buffer() checks */
1069                         sync_dirty_buffer(tbh);
1070                 if (unlikely(!buffer_uptodate(tbh))) {
1071 #ifdef CONFIG_REISERFS_CHECK
1072                         reiserfs_warning(s, "journal-601, buffer write failed");
1073 #endif
1074                         retval = -EIO;
1075                 }
1076                 put_bh(tbh);    /* once for journal_find_get_block */
1077                 put_bh(tbh);    /* once due to original getblk in do_journal_end */
1078                 atomic_dec(&(jl->j_commit_left));
1079         }
1080
1081         BUG_ON(atomic_read(&(jl->j_commit_left)) != 1);
1082
1083         if (!barrier) {
1084                 /* If there was a write error in the journal - we can't commit
1085                  * this transaction - it will be invalid and, if successful,
1086                  * will just end up propogating the write error out to
1087                  * the file system. */
1088                 if (likely(!retval && !reiserfs_is_journal_aborted (journal))) {
1089                         if (buffer_dirty(jl->j_commit_bh))
1090                                 BUG();
1091                         mark_buffer_dirty(jl->j_commit_bh) ;
1092                         sync_dirty_buffer(jl->j_commit_bh) ;
1093                 }
1094         } else
1095                 wait_on_buffer(jl->j_commit_bh);
1096
1097         check_barrier_completion(s, jl->j_commit_bh);
1098
1099         /* If there was a write error in the journal - we can't commit this
1100          * transaction - it will be invalid and, if successful, will just end
1101          * up propogating the write error out to the filesystem. */
1102         if (unlikely(!buffer_uptodate(jl->j_commit_bh))) {
1103 #ifdef CONFIG_REISERFS_CHECK
1104                 reiserfs_warning(s, "journal-615: buffer write failed");
1105 #endif
1106                 retval = -EIO;
1107         }
1108         bforget(jl->j_commit_bh);
1109         if (journal->j_last_commit_id != 0 &&
1110             (jl->j_trans_id - journal->j_last_commit_id) != 1) {
1111                 reiserfs_warning(s, "clm-2200: last commit %lu, current %lu",
1112                                  journal->j_last_commit_id, jl->j_trans_id);
1113         }
1114         journal->j_last_commit_id = jl->j_trans_id;
1115
1116         /* now, every commit block is on the disk.  It is safe to allow blocks freed during this transaction to be reallocated */
1117         cleanup_freed_for_journal_list(s, jl);
1118
1119         retval = retval ? retval : journal->j_errno;
1120
1121         /* mark the metadata dirty */
1122         if (!retval)
1123                 dirty_one_transaction(s, jl);
1124         atomic_dec(&(jl->j_commit_left));
1125
1126         if (flushall) {
1127                 atomic_set(&(jl->j_older_commits_done), 1);
1128         }
1129         up(&jl->j_commit_lock);
1130       put_jl:
1131         put_journal_list(s, jl);
1132
1133         if (retval)
1134                 reiserfs_abort(s, retval, "Journal write error in %s",
1135                                __FUNCTION__);
1136         put_fs_excl();
1137         return retval;
1138 }
1139
1140 /*
1141 ** flush_journal_list frequently needs to find a newer transaction for a given block.  This does that, or
1142 ** returns NULL if it can't find anything
1143 */
1144 static struct reiserfs_journal_list *find_newer_jl_for_cn(struct
1145                                                           reiserfs_journal_cnode
1146                                                           *cn)
1147 {
1148         struct super_block *sb = cn->sb;
1149         b_blocknr_t blocknr = cn->blocknr;
1150
1151         cn = cn->hprev;
1152         while (cn) {
1153                 if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist) {
1154                         return cn->jlist;
1155                 }
1156                 cn = cn->hprev;
1157         }
1158         return NULL;
1159 }
1160
1161 static void remove_journal_hash(struct super_block *,
1162                                 struct reiserfs_journal_cnode **,
1163                                 struct reiserfs_journal_list *, unsigned long,
1164                                 int);
1165
1166 /*
1167 ** once all the real blocks have been flushed, it is safe to remove them from the
1168 ** journal list for this transaction.  Aside from freeing the cnode, this also allows the
1169 ** block to be reallocated for data blocks if it had been deleted.
1170 */
1171 static void remove_all_from_journal_list(struct super_block *p_s_sb,
1172                                          struct reiserfs_journal_list *jl,
1173                                          int debug)
1174 {
1175         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1176         struct reiserfs_journal_cnode *cn, *last;
1177         cn = jl->j_realblock;
1178
1179         /* which is better, to lock once around the whole loop, or
1180          ** to lock for each call to remove_journal_hash?
1181          */
1182         while (cn) {
1183                 if (cn->blocknr != 0) {
1184                         if (debug) {
1185                                 reiserfs_warning(p_s_sb,
1186                                                  "block %u, bh is %d, state %ld",
1187                                                  cn->blocknr, cn->bh ? 1 : 0,
1188                                                  cn->state);
1189                         }
1190                         cn->state = 0;
1191                         remove_journal_hash(p_s_sb, journal->j_list_hash_table,
1192                                             jl, cn->blocknr, 1);
1193                 }
1194                 last = cn;
1195                 cn = cn->next;
1196                 free_cnode(p_s_sb, last);
1197         }
1198         jl->j_realblock = NULL;
1199 }
1200
1201 /*
1202 ** if this timestamp is greater than the timestamp we wrote last to the header block, write it to the header block.
1203 ** once this is done, I can safely say the log area for this transaction won't ever be replayed, and I can start
1204 ** releasing blocks in this transaction for reuse as data blocks.
1205 ** called by flush_journal_list, before it calls remove_all_from_journal_list
1206 **
1207 */
1208 static int _update_journal_header_block(struct super_block *p_s_sb,
1209                                         unsigned long offset,
1210                                         unsigned long trans_id)
1211 {
1212         struct reiserfs_journal_header *jh;
1213         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1214
1215         if (reiserfs_is_journal_aborted(journal))
1216                 return -EIO;
1217
1218         if (trans_id >= journal->j_last_flush_trans_id) {
1219                 if (buffer_locked((journal->j_header_bh))) {
1220                         wait_on_buffer((journal->j_header_bh));
1221                         if (unlikely(!buffer_uptodate(journal->j_header_bh))) {
1222 #ifdef CONFIG_REISERFS_CHECK
1223                                 reiserfs_warning(p_s_sb,
1224                                                  "journal-699: buffer write failed");
1225 #endif
1226                                 return -EIO;
1227                         }
1228                 }
1229                 journal->j_last_flush_trans_id = trans_id;
1230                 journal->j_first_unflushed_offset = offset;
1231                 jh = (struct reiserfs_journal_header *)(journal->j_header_bh->
1232                                                         b_data);
1233                 jh->j_last_flush_trans_id = cpu_to_le32(trans_id);
1234                 jh->j_first_unflushed_offset = cpu_to_le32(offset);
1235                 jh->j_mount_id = cpu_to_le32(journal->j_mount_id);
1236
1237                 if (reiserfs_barrier_flush(p_s_sb)) {
1238                         int ret;
1239                         lock_buffer(journal->j_header_bh);
1240                         ret = submit_barrier_buffer(journal->j_header_bh);
1241                         if (ret == -EOPNOTSUPP) {
1242                                 set_buffer_uptodate(journal->j_header_bh);
1243                                 disable_barrier(p_s_sb);
1244                                 goto sync;
1245                         }
1246                         wait_on_buffer(journal->j_header_bh);
1247                         check_barrier_completion(p_s_sb, journal->j_header_bh);
1248                 } else {
1249                       sync:
1250                         set_buffer_dirty(journal->j_header_bh);
1251                         sync_dirty_buffer(journal->j_header_bh);
1252                 }
1253                 if (!buffer_uptodate(journal->j_header_bh)) {
1254                         reiserfs_warning(p_s_sb,
1255                                          "journal-837: IO error during journal replay");
1256                         return -EIO;
1257                 }
1258         }
1259         return 0;
1260 }
1261
1262 static int update_journal_header_block(struct super_block *p_s_sb,
1263                                        unsigned long offset,
1264                                        unsigned long trans_id)
1265 {
1266         return _update_journal_header_block(p_s_sb, offset, trans_id);
1267 }
1268
1269 /*
1270 ** flush any and all journal lists older than you are
1271 ** can only be called from flush_journal_list
1272 */
1273 static int flush_older_journal_lists(struct super_block *p_s_sb,
1274                                      struct reiserfs_journal_list *jl)
1275 {
1276         struct list_head *entry;
1277         struct reiserfs_journal_list *other_jl;
1278         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1279         unsigned long trans_id = jl->j_trans_id;
1280
1281         /* we know we are the only ones flushing things, no extra race
1282          * protection is required.
1283          */
1284       restart:
1285         entry = journal->j_journal_list.next;
1286         /* Did we wrap? */
1287         if (entry == &journal->j_journal_list)
1288                 return 0;
1289         other_jl = JOURNAL_LIST_ENTRY(entry);
1290         if (other_jl->j_trans_id < trans_id) {
1291                 BUG_ON(other_jl->j_refcount <= 0);
1292                 /* do not flush all */
1293                 flush_journal_list(p_s_sb, other_jl, 0);
1294
1295                 /* other_jl is now deleted from the list */
1296                 goto restart;
1297         }
1298         return 0;
1299 }
1300
1301 static void del_from_work_list(struct super_block *s,
1302                                struct reiserfs_journal_list *jl)
1303 {
1304         struct reiserfs_journal *journal = SB_JOURNAL(s);
1305         if (!list_empty(&jl->j_working_list)) {
1306                 list_del_init(&jl->j_working_list);
1307                 journal->j_num_work_lists--;
1308         }
1309 }
1310
1311 /* flush a journal list, both commit and real blocks
1312 **
1313 ** always set flushall to 1, unless you are calling from inside
1314 ** flush_journal_list
1315 **
1316 ** IMPORTANT.  This can only be called while there are no journal writers,
1317 ** and the journal is locked.  That means it can only be called from
1318 ** do_journal_end, or by journal_release
1319 */
1320 static int flush_journal_list(struct super_block *s,
1321                               struct reiserfs_journal_list *jl, int flushall)
1322 {
1323         struct reiserfs_journal_list *pjl;
1324         struct reiserfs_journal_cnode *cn, *last;
1325         int count;
1326         int was_jwait = 0;
1327         int was_dirty = 0;
1328         struct buffer_head *saved_bh;
1329         unsigned long j_len_saved = jl->j_len;
1330         struct reiserfs_journal *journal = SB_JOURNAL(s);
1331         int err = 0;
1332
1333         BUG_ON(j_len_saved <= 0);
1334
1335         if (atomic_read(&journal->j_wcount) != 0) {
1336                 reiserfs_warning(s,
1337                                  "clm-2048: flush_journal_list called with wcount %d",
1338                                  atomic_read(&journal->j_wcount));
1339         }
1340         BUG_ON(jl->j_trans_id == 0);
1341
1342         /* if flushall == 0, the lock is already held */
1343         if (flushall) {
1344                 down(&journal->j_flush_sem);
1345         } else if (!down_trylock(&journal->j_flush_sem)) {
1346                 BUG();
1347         }
1348
1349         count = 0;
1350         if (j_len_saved > journal->j_trans_max) {
1351                 reiserfs_panic(s,
1352                                "journal-715: flush_journal_list, length is %lu, trans id %lu\n",
1353                                j_len_saved, jl->j_trans_id);
1354                 return 0;
1355         }
1356
1357         get_fs_excl();
1358
1359         /* if all the work is already done, get out of here */
1360         if (atomic_read(&(jl->j_nonzerolen)) <= 0 &&
1361             atomic_read(&(jl->j_commit_left)) <= 0) {
1362                 goto flush_older_and_return;
1363         }
1364
1365         /* start by putting the commit list on disk.  This will also flush
1366          ** the commit lists of any olders transactions
1367          */
1368         flush_commit_list(s, jl, 1);
1369
1370         if (!(jl->j_state & LIST_DIRTY)
1371             && !reiserfs_is_journal_aborted(journal))
1372                 BUG();
1373
1374         /* are we done now? */
1375         if (atomic_read(&(jl->j_nonzerolen)) <= 0 &&
1376             atomic_read(&(jl->j_commit_left)) <= 0) {
1377                 goto flush_older_and_return;
1378         }
1379
1380         /* loop through each cnode, see if we need to write it,
1381          ** or wait on a more recent transaction, or just ignore it
1382          */
1383         if (atomic_read(&(journal->j_wcount)) != 0) {
1384                 reiserfs_panic(s,
1385                                "journal-844: panic journal list is flushing, wcount is not 0\n");
1386         }
1387         cn = jl->j_realblock;
1388         while (cn) {
1389                 was_jwait = 0;
1390                 was_dirty = 0;
1391                 saved_bh = NULL;
1392                 /* blocknr of 0 is no longer in the hash, ignore it */
1393                 if (cn->blocknr == 0) {
1394                         goto free_cnode;
1395                 }
1396
1397                 /* This transaction failed commit. Don't write out to the disk */
1398                 if (!(jl->j_state & LIST_DIRTY))
1399                         goto free_cnode;
1400
1401                 pjl = find_newer_jl_for_cn(cn);
1402                 /* the order is important here.  We check pjl to make sure we
1403                  ** don't clear BH_JDirty_wait if we aren't the one writing this
1404                  ** block to disk
1405                  */
1406                 if (!pjl && cn->bh) {
1407                         saved_bh = cn->bh;
1408
1409                         /* we do this to make sure nobody releases the buffer while
1410                          ** we are working with it
1411                          */
1412                         get_bh(saved_bh);
1413
1414                         if (buffer_journal_dirty(saved_bh)) {
1415                                 BUG_ON(!can_dirty(cn));
1416                                 was_jwait = 1;
1417                                 was_dirty = 1;
1418                         } else if (can_dirty(cn)) {
1419                                 /* everything with !pjl && jwait should be writable */
1420                                 BUG();
1421                         }
1422                 }
1423
1424                 /* if someone has this block in a newer transaction, just make
1425                  ** sure they are commited, and don't try writing it to disk
1426                  */
1427                 if (pjl) {
1428                         if (atomic_read(&pjl->j_commit_left))
1429                                 flush_commit_list(s, pjl, 1);
1430                         goto free_cnode;
1431                 }
1432
1433                 /* bh == NULL when the block got to disk on its own, OR,
1434                  ** the block got freed in a future transaction
1435                  */
1436                 if (saved_bh == NULL) {
1437                         goto free_cnode;
1438                 }
1439
1440                 /* this should never happen.  kupdate_one_transaction has this list
1441                  ** locked while it works, so we should never see a buffer here that
1442                  ** is not marked JDirty_wait
1443                  */
1444                 if ((!was_jwait) && !buffer_locked(saved_bh)) {
1445                         reiserfs_warning(s,
1446                                          "journal-813: BAD! buffer %llu %cdirty %cjwait, "
1447                                          "not in a newer tranasction",
1448                                          (unsigned long long)saved_bh->
1449                                          b_blocknr, was_dirty ? ' ' : '!',
1450                                          was_jwait ? ' ' : '!');
1451                 }
1452                 if (was_dirty) {
1453                         /* we inc again because saved_bh gets decremented at free_cnode */
1454                         get_bh(saved_bh);
1455                         set_bit(BLOCK_NEEDS_FLUSH, &cn->state);
1456                         lock_buffer(saved_bh);
1457                         BUG_ON(cn->blocknr != saved_bh->b_blocknr);
1458                         if (buffer_dirty(saved_bh))
1459                                 submit_logged_buffer(saved_bh);
1460                         else
1461                                 unlock_buffer(saved_bh);
1462                         count++;
1463                 } else {
1464                         reiserfs_warning(s,
1465                                          "clm-2082: Unable to flush buffer %llu in %s",
1466                                          (unsigned long long)saved_bh->
1467                                          b_blocknr, __FUNCTION__);
1468                 }
1469               free_cnode:
1470                 last = cn;
1471                 cn = cn->next;
1472                 if (saved_bh) {
1473                         /* we incremented this to keep others from taking the buffer head away */
1474                         put_bh(saved_bh);
1475                         if (atomic_read(&(saved_bh->b_count)) < 0) {
1476                                 reiserfs_warning(s,
1477                                                  "journal-945: saved_bh->b_count < 0");
1478                         }
1479                 }
1480         }
1481         if (count > 0) {
1482                 cn = jl->j_realblock;
1483                 while (cn) {
1484                         if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) {
1485                                 if (!cn->bh) {
1486                                         reiserfs_panic(s,
1487                                                        "journal-1011: cn->bh is NULL\n");
1488                                 }
1489                                 wait_on_buffer(cn->bh);
1490                                 if (!cn->bh) {
1491                                         reiserfs_panic(s,
1492                                                        "journal-1012: cn->bh is NULL\n");
1493                                 }
1494                                 if (unlikely(!buffer_uptodate(cn->bh))) {
1495 #ifdef CONFIG_REISERFS_CHECK
1496                                         reiserfs_warning(s,
1497                                                          "journal-949: buffer write failed\n");
1498 #endif
1499                                         err = -EIO;
1500                                 }
1501                                 /* note, we must clear the JDirty_wait bit after the up to date
1502                                  ** check, otherwise we race against our flushpage routine
1503                                  */
1504                                 BUG_ON(!test_clear_buffer_journal_dirty
1505                                        (cn->bh));
1506
1507                                 /* undo the inc from journal_mark_dirty */
1508                                 put_bh(cn->bh);
1509                                 brelse(cn->bh);
1510                         }
1511                         cn = cn->next;
1512                 }
1513         }
1514
1515         if (err)
1516                 reiserfs_abort(s, -EIO,
1517                                "Write error while pushing transaction to disk in %s",
1518                                __FUNCTION__);
1519       flush_older_and_return:
1520
1521         /* before we can update the journal header block, we _must_ flush all
1522          ** real blocks from all older transactions to disk.  This is because
1523          ** once the header block is updated, this transaction will not be
1524          ** replayed after a crash
1525          */
1526         if (flushall) {
1527                 flush_older_journal_lists(s, jl);
1528         }
1529
1530         err = journal->j_errno;
1531         /* before we can remove everything from the hash tables for this
1532          ** transaction, we must make sure it can never be replayed
1533          **
1534          ** since we are only called from do_journal_end, we know for sure there
1535          ** are no allocations going on while we are flushing journal lists.  So,
1536          ** we only need to update the journal header block for the last list
1537          ** being flushed
1538          */
1539         if (!err && flushall) {
1540                 err =
1541                     update_journal_header_block(s,
1542                                                 (jl->j_start + jl->j_len +
1543                                                  2) % SB_ONDISK_JOURNAL_SIZE(s),
1544                                                 jl->j_trans_id);
1545                 if (err)
1546                         reiserfs_abort(s, -EIO,
1547                                        "Write error while updating journal header in %s",
1548                                        __FUNCTION__);
1549         }
1550         remove_all_from_journal_list(s, jl, 0);
1551         list_del_init(&jl->j_list);
1552         journal->j_num_lists--;
1553         del_from_work_list(s, jl);
1554
1555         if (journal->j_last_flush_id != 0 &&
1556             (jl->j_trans_id - journal->j_last_flush_id) != 1) {
1557                 reiserfs_warning(s, "clm-2201: last flush %lu, current %lu",
1558                                  journal->j_last_flush_id, jl->j_trans_id);
1559         }
1560         journal->j_last_flush_id = jl->j_trans_id;
1561
1562         /* not strictly required since we are freeing the list, but it should
1563          * help find code using dead lists later on
1564          */
1565         jl->j_len = 0;
1566         atomic_set(&(jl->j_nonzerolen), 0);
1567         jl->j_start = 0;
1568         jl->j_realblock = NULL;
1569         jl->j_commit_bh = NULL;
1570         jl->j_trans_id = 0;
1571         jl->j_state = 0;
1572         put_journal_list(s, jl);
1573         if (flushall)
1574                 up(&journal->j_flush_sem);
1575         put_fs_excl();
1576         return err;
1577 }
1578
1579 static int write_one_transaction(struct super_block *s,
1580                                  struct reiserfs_journal_list *jl,
1581                                  struct buffer_chunk *chunk)
1582 {
1583         struct reiserfs_journal_cnode *cn;
1584         int ret = 0;
1585
1586         jl->j_state |= LIST_TOUCHED;
1587         del_from_work_list(s, jl);
1588         if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0) {
1589                 return 0;
1590         }
1591
1592         cn = jl->j_realblock;
1593         while (cn) {
1594                 /* if the blocknr == 0, this has been cleared from the hash,
1595                  ** skip it
1596                  */
1597                 if (cn->blocknr == 0) {
1598                         goto next;
1599                 }
1600                 if (cn->bh && can_dirty(cn) && buffer_dirty(cn->bh)) {
1601                         struct buffer_head *tmp_bh;
1602                         /* we can race against journal_mark_freed when we try
1603                          * to lock_buffer(cn->bh), so we have to inc the buffer
1604                          * count, and recheck things after locking
1605                          */
1606                         tmp_bh = cn->bh;
1607                         get_bh(tmp_bh);
1608                         lock_buffer(tmp_bh);
1609                         if (cn->bh && can_dirty(cn) && buffer_dirty(tmp_bh)) {
1610                                 if (!buffer_journal_dirty(tmp_bh) ||
1611                                     buffer_journal_prepared(tmp_bh))
1612                                         BUG();
1613                                 add_to_chunk(chunk, tmp_bh, NULL, write_chunk);
1614                                 ret++;
1615                         } else {
1616                                 /* note, cn->bh might be null now */
1617                                 unlock_buffer(tmp_bh);
1618                         }
1619                         put_bh(tmp_bh);
1620                 }
1621               next:
1622                 cn = cn->next;
1623                 cond_resched();
1624         }
1625         return ret;
1626 }
1627
1628 /* used by flush_commit_list */
1629 static int dirty_one_transaction(struct super_block *s,
1630                                  struct reiserfs_journal_list *jl)
1631 {
1632         struct reiserfs_journal_cnode *cn;
1633         struct reiserfs_journal_list *pjl;
1634         int ret = 0;
1635
1636         jl->j_state |= LIST_DIRTY;
1637         cn = jl->j_realblock;
1638         while (cn) {
1639                 /* look for a more recent transaction that logged this
1640                  ** buffer.  Only the most recent transaction with a buffer in
1641                  ** it is allowed to send that buffer to disk
1642                  */
1643                 pjl = find_newer_jl_for_cn(cn);
1644                 if (!pjl && cn->blocknr && cn->bh
1645                     && buffer_journal_dirty(cn->bh)) {
1646                         BUG_ON(!can_dirty(cn));
1647                         /* if the buffer is prepared, it will either be logged
1648                          * or restored.  If restored, we need to make sure
1649                          * it actually gets marked dirty
1650                          */
1651                         clear_buffer_journal_new(cn->bh);
1652                         if (buffer_journal_prepared(cn->bh)) {
1653                                 set_buffer_journal_restore_dirty(cn->bh);
1654                         } else {
1655                                 set_buffer_journal_test(cn->bh);
1656                                 mark_buffer_dirty(cn->bh);
1657                         }
1658                 }
1659                 cn = cn->next;
1660         }
1661         return ret;
1662 }
1663
1664 static int kupdate_transactions(struct super_block *s,
1665                                 struct reiserfs_journal_list *jl,
1666                                 struct reiserfs_journal_list **next_jl,
1667                                 unsigned long *next_trans_id,
1668                                 int num_blocks, int num_trans)
1669 {
1670         int ret = 0;
1671         int written = 0;
1672         int transactions_flushed = 0;
1673         unsigned long orig_trans_id = jl->j_trans_id;
1674         struct buffer_chunk chunk;
1675         struct list_head *entry;
1676         struct reiserfs_journal *journal = SB_JOURNAL(s);
1677         chunk.nr = 0;
1678
1679         down(&journal->j_flush_sem);
1680         if (!journal_list_still_alive(s, orig_trans_id)) {
1681                 goto done;
1682         }
1683
1684         /* we've got j_flush_sem held, nobody is going to delete any
1685          * of these lists out from underneath us
1686          */
1687         while ((num_trans && transactions_flushed < num_trans) ||
1688                (!num_trans && written < num_blocks)) {
1689
1690                 if (jl->j_len == 0 || (jl->j_state & LIST_TOUCHED) ||
1691                     atomic_read(&jl->j_commit_left)
1692                     || !(jl->j_state & LIST_DIRTY)) {
1693                         del_from_work_list(s, jl);
1694                         break;
1695                 }
1696                 ret = write_one_transaction(s, jl, &chunk);
1697
1698                 if (ret < 0)
1699                         goto done;
1700                 transactions_flushed++;
1701                 written += ret;
1702                 entry = jl->j_list.next;
1703
1704                 /* did we wrap? */
1705                 if (entry == &journal->j_journal_list) {
1706                         break;
1707                 }
1708                 jl = JOURNAL_LIST_ENTRY(entry);
1709
1710                 /* don't bother with older transactions */
1711                 if (jl->j_trans_id <= orig_trans_id)
1712                         break;
1713         }
1714         if (chunk.nr) {
1715                 write_chunk(&chunk);
1716         }
1717
1718       done:
1719         up(&journal->j_flush_sem);
1720         return ret;
1721 }
1722
1723 /* for o_sync and fsync heavy applications, they tend to use
1724 ** all the journa list slots with tiny transactions.  These
1725 ** trigger lots and lots of calls to update the header block, which
1726 ** adds seeks and slows things down.
1727 **
1728 ** This function tries to clear out a large chunk of the journal lists
1729 ** at once, which makes everything faster since only the newest journal
1730 ** list updates the header block
1731 */
1732 static int flush_used_journal_lists(struct super_block *s,
1733                                     struct reiserfs_journal_list *jl)
1734 {
1735         unsigned long len = 0;
1736         unsigned long cur_len;
1737         int ret;
1738         int i;
1739         int limit = 256;
1740         struct reiserfs_journal_list *tjl;
1741         struct reiserfs_journal_list *flush_jl;
1742         unsigned long trans_id;
1743         struct reiserfs_journal *journal = SB_JOURNAL(s);
1744
1745         flush_jl = tjl = jl;
1746
1747         /* in data logging mode, try harder to flush a lot of blocks */
1748         if (reiserfs_data_log(s))
1749                 limit = 1024;
1750         /* flush for 256 transactions or limit blocks, whichever comes first */
1751         for (i = 0; i < 256 && len < limit; i++) {
1752                 if (atomic_read(&tjl->j_commit_left) ||
1753                     tjl->j_trans_id < jl->j_trans_id) {
1754                         break;
1755                 }
1756                 cur_len = atomic_read(&tjl->j_nonzerolen);
1757                 if (cur_len > 0) {
1758                         tjl->j_state &= ~LIST_TOUCHED;
1759                 }
1760                 len += cur_len;
1761                 flush_jl = tjl;
1762                 if (tjl->j_list.next == &journal->j_journal_list)
1763                         break;
1764                 tjl = JOURNAL_LIST_ENTRY(tjl->j_list.next);
1765         }
1766         /* try to find a group of blocks we can flush across all the
1767          ** transactions, but only bother if we've actually spanned
1768          ** across multiple lists
1769          */
1770         if (flush_jl != jl) {
1771                 ret = kupdate_transactions(s, jl, &tjl, &trans_id, len, i);
1772         }
1773         flush_journal_list(s, flush_jl, 1);
1774         return 0;
1775 }
1776
1777 /*
1778 ** removes any nodes in table with name block and dev as bh.
1779 ** only touchs the hnext and hprev pointers.
1780 */
1781 void remove_journal_hash(struct super_block *sb,
1782                          struct reiserfs_journal_cnode **table,
1783                          struct reiserfs_journal_list *jl,
1784                          unsigned long block, int remove_freed)
1785 {
1786         struct reiserfs_journal_cnode *cur;
1787         struct reiserfs_journal_cnode **head;
1788
1789         head = &(journal_hash(table, sb, block));
1790         if (!head) {
1791                 return;
1792         }
1793         cur = *head;
1794         while (cur) {
1795                 if (cur->blocknr == block && cur->sb == sb
1796                     && (jl == NULL || jl == cur->jlist)
1797                     && (!test_bit(BLOCK_FREED, &cur->state) || remove_freed)) {
1798                         if (cur->hnext) {
1799                                 cur->hnext->hprev = cur->hprev;
1800                         }
1801                         if (cur->hprev) {
1802                                 cur->hprev->hnext = cur->hnext;
1803                         } else {
1804                                 *head = cur->hnext;
1805                         }
1806                         cur->blocknr = 0;
1807                         cur->sb = NULL;
1808                         cur->state = 0;
1809                         if (cur->bh && cur->jlist)      /* anybody who clears the cur->bh will also dec the nonzerolen */
1810                                 atomic_dec(&(cur->jlist->j_nonzerolen));
1811                         cur->bh = NULL;
1812                         cur->jlist = NULL;
1813                 }
1814                 cur = cur->hnext;
1815         }
1816 }
1817
1818 static void free_journal_ram(struct super_block *p_s_sb)
1819 {
1820         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1821         reiserfs_kfree(journal->j_current_jl,
1822                        sizeof(struct reiserfs_journal_list), p_s_sb);
1823         journal->j_num_lists--;
1824
1825         vfree(journal->j_cnode_free_orig);
1826         free_list_bitmaps(p_s_sb, journal->j_list_bitmap);
1827         free_bitmap_nodes(p_s_sb);      /* must be after free_list_bitmaps */
1828         if (journal->j_header_bh) {
1829                 brelse(journal->j_header_bh);
1830         }
1831         /* j_header_bh is on the journal dev, make sure not to release the journal
1832          * dev until we brelse j_header_bh
1833          */
1834         release_journal_dev(p_s_sb, journal);
1835         vfree(journal);
1836 }
1837
1838 /*
1839 ** call on unmount.  Only set error to 1 if you haven't made your way out
1840 ** of read_super() yet.  Any other caller must keep error at 0.
1841 */
1842 static int do_journal_release(struct reiserfs_transaction_handle *th,
1843                               struct super_block *p_s_sb, int error)
1844 {
1845         struct reiserfs_transaction_handle myth;
1846         int flushed = 0;
1847         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1848
1849         /* we only want to flush out transactions if we were called with error == 0
1850          */
1851         if (!error && !(p_s_sb->s_flags & MS_RDONLY)) {
1852                 /* end the current trans */
1853                 BUG_ON(!th->t_trans_id);
1854                 do_journal_end(th, p_s_sb, 10, FLUSH_ALL);
1855
1856                 /* make sure something gets logged to force our way into the flush code */
1857                 if (!journal_join(&myth, p_s_sb, 1)) {
1858                         reiserfs_prepare_for_journal(p_s_sb,
1859                                                      SB_BUFFER_WITH_SB(p_s_sb),
1860                                                      1);
1861                         journal_mark_dirty(&myth, p_s_sb,
1862                                            SB_BUFFER_WITH_SB(p_s_sb));
1863                         do_journal_end(&myth, p_s_sb, 1, FLUSH_ALL);
1864                         flushed = 1;
1865                 }
1866         }
1867
1868         /* this also catches errors during the do_journal_end above */
1869         if (!error && reiserfs_is_journal_aborted(journal)) {
1870                 memset(&myth, 0, sizeof(myth));
1871                 if (!journal_join_abort(&myth, p_s_sb, 1)) {
1872                         reiserfs_prepare_for_journal(p_s_sb,
1873                                                      SB_BUFFER_WITH_SB(p_s_sb),
1874                                                      1);
1875                         journal_mark_dirty(&myth, p_s_sb,
1876                                            SB_BUFFER_WITH_SB(p_s_sb));
1877                         do_journal_end(&myth, p_s_sb, 1, FLUSH_ALL);
1878                 }
1879         }
1880
1881         reiserfs_mounted_fs_count--;
1882         /* wait for all commits to finish */
1883         cancel_delayed_work(&SB_JOURNAL(p_s_sb)->j_work);
1884         flush_workqueue(commit_wq);
1885         if (!reiserfs_mounted_fs_count) {
1886                 destroy_workqueue(commit_wq);
1887                 commit_wq = NULL;
1888         }
1889
1890         free_journal_ram(p_s_sb);
1891
1892         return 0;
1893 }
1894
1895 /*
1896 ** call on unmount.  flush all journal trans, release all alloc'd ram
1897 */
1898 int journal_release(struct reiserfs_transaction_handle *th,
1899                     struct super_block *p_s_sb)
1900 {
1901         return do_journal_release(th, p_s_sb, 0);
1902 }
1903
1904 /*
1905 ** only call from an error condition inside reiserfs_read_super!
1906 */
1907 int journal_release_error(struct reiserfs_transaction_handle *th,
1908                           struct super_block *p_s_sb)
1909 {
1910         return do_journal_release(th, p_s_sb, 1);
1911 }
1912
1913 /* compares description block with commit block.  returns 1 if they differ, 0 if they are the same */
1914 static int journal_compare_desc_commit(struct super_block *p_s_sb,
1915                                        struct reiserfs_journal_desc *desc,
1916                                        struct reiserfs_journal_commit *commit)
1917 {
1918         if (get_commit_trans_id(commit) != get_desc_trans_id(desc) ||
1919             get_commit_trans_len(commit) != get_desc_trans_len(desc) ||
1920             get_commit_trans_len(commit) > SB_JOURNAL(p_s_sb)->j_trans_max ||
1921             get_commit_trans_len(commit) <= 0) {
1922                 return 1;
1923         }
1924         return 0;
1925 }
1926
1927 /* returns 0 if it did not find a description block
1928 ** returns -1 if it found a corrupt commit block
1929 ** returns 1 if both desc and commit were valid
1930 */
1931 static int journal_transaction_is_valid(struct super_block *p_s_sb,
1932                                         struct buffer_head *d_bh,
1933                                         unsigned long *oldest_invalid_trans_id,
1934                                         unsigned long *newest_mount_id)
1935 {
1936         struct reiserfs_journal_desc *desc;
1937         struct reiserfs_journal_commit *commit;
1938         struct buffer_head *c_bh;
1939         unsigned long offset;
1940
1941         if (!d_bh)
1942                 return 0;
1943
1944         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
1945         if (get_desc_trans_len(desc) > 0
1946             && !memcmp(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8)) {
1947                 if (oldest_invalid_trans_id && *oldest_invalid_trans_id
1948                     && get_desc_trans_id(desc) > *oldest_invalid_trans_id) {
1949                         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
1950                                        "journal-986: transaction "
1951                                        "is valid returning because trans_id %d is greater than "
1952                                        "oldest_invalid %lu",
1953                                        get_desc_trans_id(desc),
1954                                        *oldest_invalid_trans_id);
1955                         return 0;
1956                 }
1957                 if (newest_mount_id
1958                     && *newest_mount_id > get_desc_mount_id(desc)) {
1959                         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
1960                                        "journal-1087: transaction "
1961                                        "is valid returning because mount_id %d is less than "
1962                                        "newest_mount_id %lu",
1963                                        get_desc_mount_id(desc),
1964                                        *newest_mount_id);
1965                         return -1;
1966                 }
1967                 if (get_desc_trans_len(desc) > SB_JOURNAL(p_s_sb)->j_trans_max) {
1968                         reiserfs_warning(p_s_sb,
1969                                          "journal-2018: Bad transaction length %d encountered, ignoring transaction",
1970                                          get_desc_trans_len(desc));
1971                         return -1;
1972                 }
1973                 offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb);
1974
1975                 /* ok, we have a journal description block, lets see if the transaction was valid */
1976                 c_bh =
1977                     journal_bread(p_s_sb,
1978                                   SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
1979                                   ((offset + get_desc_trans_len(desc) +
1980                                     1) % SB_ONDISK_JOURNAL_SIZE(p_s_sb)));
1981                 if (!c_bh)
1982                         return 0;
1983                 commit = (struct reiserfs_journal_commit *)c_bh->b_data;
1984                 if (journal_compare_desc_commit(p_s_sb, desc, commit)) {
1985                         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
1986                                        "journal_transaction_is_valid, commit offset %ld had bad "
1987                                        "time %d or length %d",
1988                                        c_bh->b_blocknr -
1989                                        SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
1990                                        get_commit_trans_id(commit),
1991                                        get_commit_trans_len(commit));
1992                         brelse(c_bh);
1993                         if (oldest_invalid_trans_id) {
1994                                 *oldest_invalid_trans_id =
1995                                     get_desc_trans_id(desc);
1996                                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
1997                                                "journal-1004: "
1998                                                "transaction_is_valid setting oldest invalid trans_id "
1999                                                "to %d",
2000                                                get_desc_trans_id(desc));
2001                         }
2002                         return -1;
2003                 }
2004                 brelse(c_bh);
2005                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2006                                "journal-1006: found valid "
2007                                "transaction start offset %llu, len %d id %d",
2008                                d_bh->b_blocknr -
2009                                SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2010                                get_desc_trans_len(desc),
2011                                get_desc_trans_id(desc));
2012                 return 1;
2013         } else {
2014                 return 0;
2015         }
2016 }
2017
2018 static void brelse_array(struct buffer_head **heads, int num)
2019 {
2020         int i;
2021         for (i = 0; i < num; i++) {
2022                 brelse(heads[i]);
2023         }
2024 }
2025
2026 /*
2027 ** given the start, and values for the oldest acceptable transactions,
2028 ** this either reads in a replays a transaction, or returns because the transaction
2029 ** is invalid, or too old.
2030 */
2031 static int journal_read_transaction(struct super_block *p_s_sb,
2032                                     unsigned long cur_dblock,
2033                                     unsigned long oldest_start,
2034                                     unsigned long oldest_trans_id,
2035                                     unsigned long newest_mount_id)
2036 {
2037         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
2038         struct reiserfs_journal_desc *desc;
2039         struct reiserfs_journal_commit *commit;
2040         unsigned long trans_id = 0;
2041         struct buffer_head *c_bh;
2042         struct buffer_head *d_bh;
2043         struct buffer_head **log_blocks = NULL;
2044         struct buffer_head **real_blocks = NULL;
2045         unsigned long trans_offset;
2046         int i;
2047         int trans_half;
2048
2049         d_bh = journal_bread(p_s_sb, cur_dblock);
2050         if (!d_bh)
2051                 return 1;
2052         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2053         trans_offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb);
2054         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1037: "
2055                        "journal_read_transaction, offset %llu, len %d mount_id %d",
2056                        d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2057                        get_desc_trans_len(desc), get_desc_mount_id(desc));
2058         if (get_desc_trans_id(desc) < oldest_trans_id) {
2059                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1039: "
2060                                "journal_read_trans skipping because %lu is too old",
2061                                cur_dblock -
2062                                SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb));
2063                 brelse(d_bh);
2064                 return 1;
2065         }
2066         if (get_desc_mount_id(desc) != newest_mount_id) {
2067                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1146: "
2068                                "journal_read_trans skipping because %d is != "
2069                                "newest_mount_id %lu", get_desc_mount_id(desc),
2070                                newest_mount_id);
2071                 brelse(d_bh);
2072                 return 1;
2073         }
2074         c_bh = journal_bread(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2075                              ((trans_offset + get_desc_trans_len(desc) + 1) %
2076                               SB_ONDISK_JOURNAL_SIZE(p_s_sb)));
2077         if (!c_bh) {
2078                 brelse(d_bh);
2079                 return 1;
2080         }
2081         commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2082         if (journal_compare_desc_commit(p_s_sb, desc, commit)) {
2083                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2084                                "journal_read_transaction, "
2085                                "commit offset %llu had bad time %d or length %d",
2086                                c_bh->b_blocknr -
2087                                SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2088                                get_commit_trans_id(commit),
2089                                get_commit_trans_len(commit));
2090                 brelse(c_bh);
2091                 brelse(d_bh);
2092                 return 1;
2093         }
2094         trans_id = get_desc_trans_id(desc);
2095         /* now we know we've got a good transaction, and it was inside the valid time ranges */
2096         log_blocks =
2097             reiserfs_kmalloc(get_desc_trans_len(desc) *
2098                              sizeof(struct buffer_head *), GFP_NOFS, p_s_sb);
2099         real_blocks =
2100             reiserfs_kmalloc(get_desc_trans_len(desc) *
2101                              sizeof(struct buffer_head *), GFP_NOFS, p_s_sb);
2102         if (!log_blocks || !real_blocks) {
2103                 brelse(c_bh);
2104                 brelse(d_bh);
2105                 reiserfs_kfree(log_blocks,
2106                                get_desc_trans_len(desc) *
2107                                sizeof(struct buffer_head *), p_s_sb);
2108                 reiserfs_kfree(real_blocks,
2109                                get_desc_trans_len(desc) *
2110                                sizeof(struct buffer_head *), p_s_sb);
2111                 reiserfs_warning(p_s_sb,
2112                                  "journal-1169: kmalloc failed, unable to mount FS");
2113                 return -1;
2114         }
2115         /* get all the buffer heads */
2116         trans_half = journal_trans_half(p_s_sb->s_blocksize);
2117         for (i = 0; i < get_desc_trans_len(desc); i++) {
2118                 log_blocks[i] =
2119                     journal_getblk(p_s_sb,
2120                                    SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2121                                    (trans_offset + 1 +
2122                                     i) % SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2123                 if (i < trans_half) {
2124                         real_blocks[i] =
2125                             sb_getblk(p_s_sb,
2126                                       le32_to_cpu(desc->j_realblock[i]));
2127                 } else {
2128                         real_blocks[i] =
2129                             sb_getblk(p_s_sb,
2130                                       le32_to_cpu(commit->
2131                                                   j_realblock[i - trans_half]));
2132                 }
2133                 if (real_blocks[i]->b_blocknr > SB_BLOCK_COUNT(p_s_sb)) {
2134                         reiserfs_warning(p_s_sb,
2135                                          "journal-1207: REPLAY FAILURE fsck required! Block to replay is outside of filesystem");
2136                         goto abort_replay;
2137                 }
2138                 /* make sure we don't try to replay onto log or reserved area */
2139                 if (is_block_in_log_or_reserved_area
2140                     (p_s_sb, real_blocks[i]->b_blocknr)) {
2141                         reiserfs_warning(p_s_sb,
2142                                          "journal-1204: REPLAY FAILURE fsck required! Trying to replay onto a log block");
2143                       abort_replay:
2144                         brelse_array(log_blocks, i);
2145                         brelse_array(real_blocks, i);
2146                         brelse(c_bh);
2147                         brelse(d_bh);
2148                         reiserfs_kfree(log_blocks,
2149                                        get_desc_trans_len(desc) *
2150                                        sizeof(struct buffer_head *), p_s_sb);
2151                         reiserfs_kfree(real_blocks,
2152                                        get_desc_trans_len(desc) *
2153                                        sizeof(struct buffer_head *), p_s_sb);
2154                         return -1;
2155                 }
2156         }
2157         /* read in the log blocks, memcpy to the corresponding real block */
2158         ll_rw_block(READ, get_desc_trans_len(desc), log_blocks);
2159         for (i = 0; i < get_desc_trans_len(desc); i++) {
2160                 wait_on_buffer(log_blocks[i]);
2161                 if (!buffer_uptodate(log_blocks[i])) {
2162                         reiserfs_warning(p_s_sb,
2163                                          "journal-1212: REPLAY FAILURE fsck required! buffer write failed");
2164                         brelse_array(log_blocks + i,
2165                                      get_desc_trans_len(desc) - i);
2166                         brelse_array(real_blocks, get_desc_trans_len(desc));
2167                         brelse(c_bh);
2168                         brelse(d_bh);
2169                         reiserfs_kfree(log_blocks,
2170                                        get_desc_trans_len(desc) *
2171                                        sizeof(struct buffer_head *), p_s_sb);
2172                         reiserfs_kfree(real_blocks,
2173                                        get_desc_trans_len(desc) *
2174                                        sizeof(struct buffer_head *), p_s_sb);
2175                         return -1;
2176                 }
2177                 memcpy(real_blocks[i]->b_data, log_blocks[i]->b_data,
2178                        real_blocks[i]->b_size);
2179                 set_buffer_uptodate(real_blocks[i]);
2180                 brelse(log_blocks[i]);
2181         }
2182         /* flush out the real blocks */
2183         for (i = 0; i < get_desc_trans_len(desc); i++) {
2184                 set_buffer_dirty(real_blocks[i]);
2185                 ll_rw_block(SWRITE, 1, real_blocks + i);
2186         }
2187         for (i = 0; i < get_desc_trans_len(desc); i++) {
2188                 wait_on_buffer(real_blocks[i]);
2189                 if (!buffer_uptodate(real_blocks[i])) {
2190                         reiserfs_warning(p_s_sb,
2191                                          "journal-1226: REPLAY FAILURE, fsck required! buffer write failed");
2192                         brelse_array(real_blocks + i,
2193                                      get_desc_trans_len(desc) - i);
2194                         brelse(c_bh);
2195                         brelse(d_bh);
2196                         reiserfs_kfree(log_blocks,
2197                                        get_desc_trans_len(desc) *
2198                                        sizeof(struct buffer_head *), p_s_sb);
2199                         reiserfs_kfree(real_blocks,
2200                                        get_desc_trans_len(desc) *
2201                                        sizeof(struct buffer_head *), p_s_sb);
2202                         return -1;
2203                 }
2204                 brelse(real_blocks[i]);
2205         }
2206         cur_dblock =
2207             SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2208             ((trans_offset + get_desc_trans_len(desc) +
2209               2) % SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2210         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2211                        "journal-1095: setting journal " "start to offset %ld",
2212                        cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb));
2213
2214         /* init starting values for the first transaction, in case this is the last transaction to be replayed. */
2215         journal->j_start = cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb);
2216         journal->j_last_flush_trans_id = trans_id;
2217         journal->j_trans_id = trans_id + 1;
2218         brelse(c_bh);
2219         brelse(d_bh);
2220         reiserfs_kfree(log_blocks,
2221                        le32_to_cpu(desc->j_len) * sizeof(struct buffer_head *),
2222                        p_s_sb);
2223         reiserfs_kfree(real_blocks,
2224                        le32_to_cpu(desc->j_len) * sizeof(struct buffer_head *),
2225                        p_s_sb);
2226         return 0;
2227 }
2228
2229 /* This function reads blocks starting from block and to max_block of bufsize
2230    size (but no more than BUFNR blocks at a time). This proved to improve
2231    mounting speed on self-rebuilding raid5 arrays at least.
2232    Right now it is only used from journal code. But later we might use it
2233    from other places.
2234    Note: Do not use journal_getblk/sb_getblk functions here! */
2235 static struct buffer_head *reiserfs_breada(struct block_device *dev, int block,
2236                                            int bufsize, unsigned int max_block)
2237 {
2238         struct buffer_head *bhlist[BUFNR];
2239         unsigned int blocks = BUFNR;
2240         struct buffer_head *bh;
2241         int i, j;
2242
2243         bh = __getblk(dev, block, bufsize);
2244         if (buffer_uptodate(bh))
2245                 return (bh);
2246
2247         if (block + BUFNR > max_block) {
2248                 blocks = max_block - block;
2249         }
2250         bhlist[0] = bh;
2251         j = 1;
2252         for (i = 1; i < blocks; i++) {
2253                 bh = __getblk(dev, block + i, bufsize);
2254                 if (buffer_uptodate(bh)) {
2255                         brelse(bh);
2256                         break;
2257                 } else
2258                         bhlist[j++] = bh;
2259         }
2260         ll_rw_block(READ, j, bhlist);
2261         for (i = 1; i < j; i++)
2262                 brelse(bhlist[i]);
2263         bh = bhlist[0];
2264         wait_on_buffer(bh);
2265         if (buffer_uptodate(bh))
2266                 return bh;
2267         brelse(bh);
2268         return NULL;
2269 }
2270
2271 /*
2272 ** read and replay the log
2273 ** on a clean unmount, the journal header's next unflushed pointer will be to an invalid
2274 ** transaction.  This tests that before finding all the transactions in the log, which makes normal mount times fast.
2275 **
2276 ** After a crash, this starts with the next unflushed transaction, and replays until it finds one too old, or invalid.
2277 **
2278 ** On exit, it sets things up so the first transaction will work correctly.
2279 */
2280 static int journal_read(struct super_block *p_s_sb)
2281 {
2282         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
2283         struct reiserfs_journal_desc *desc;
2284         unsigned long oldest_trans_id = 0;
2285         unsigned long oldest_invalid_trans_id = 0;
2286         time_t start;
2287         unsigned long oldest_start = 0;
2288         unsigned long cur_dblock = 0;
2289         unsigned long newest_mount_id = 9;
2290         struct buffer_head *d_bh;
2291         struct reiserfs_journal_header *jh;
2292         int valid_journal_header = 0;
2293         int replay_count = 0;
2294         int continue_replay = 1;
2295         int ret;
2296         char b[BDEVNAME_SIZE];
2297
2298         cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb);
2299         reiserfs_info(p_s_sb, "checking transaction log (%s)\n",
2300                       bdevname(journal->j_dev_bd, b));
2301         start = get_seconds();
2302
2303         /* step 1, read in the journal header block.  Check the transaction it says
2304          ** is the first unflushed, and if that transaction is not valid,
2305          ** replay is done
2306          */
2307         journal->j_header_bh = journal_bread(p_s_sb,
2308                                              SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb)
2309                                              + SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2310         if (!journal->j_header_bh) {
2311                 return 1;
2312         }
2313         jh = (struct reiserfs_journal_header *)(journal->j_header_bh->b_data);
2314         if (le32_to_cpu(jh->j_first_unflushed_offset) >= 0 &&
2315             le32_to_cpu(jh->j_first_unflushed_offset) <
2316             SB_ONDISK_JOURNAL_SIZE(p_s_sb)
2317             && le32_to_cpu(jh->j_last_flush_trans_id) > 0) {
2318                 oldest_start =
2319                     SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2320                     le32_to_cpu(jh->j_first_unflushed_offset);
2321                 oldest_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2322                 newest_mount_id = le32_to_cpu(jh->j_mount_id);
2323                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2324                                "journal-1153: found in "
2325                                "header: first_unflushed_offset %d, last_flushed_trans_id "
2326                                "%lu", le32_to_cpu(jh->j_first_unflushed_offset),
2327                                le32_to_cpu(jh->j_last_flush_trans_id));
2328                 valid_journal_header = 1;
2329
2330                 /* now, we try to read the first unflushed offset.  If it is not valid,
2331                  ** there is nothing more we can do, and it makes no sense to read
2332                  ** through the whole log.
2333                  */
2334                 d_bh =
2335                     journal_bread(p_s_sb,
2336                                   SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2337                                   le32_to_cpu(jh->j_first_unflushed_offset));
2338                 ret = journal_transaction_is_valid(p_s_sb, d_bh, NULL, NULL);
2339                 if (!ret) {
2340                         continue_replay = 0;
2341                 }
2342                 brelse(d_bh);
2343                 goto start_log_replay;
2344         }
2345
2346         if (continue_replay && bdev_read_only(p_s_sb->s_bdev)) {
2347                 reiserfs_warning(p_s_sb,
2348                                  "clm-2076: device is readonly, unable to replay log");
2349                 return -1;
2350         }
2351
2352         /* ok, there are transactions that need to be replayed.  start with the first log block, find
2353          ** all the valid transactions, and pick out the oldest.
2354          */
2355         while (continue_replay
2356                && cur_dblock <
2357                (SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2358                 SB_ONDISK_JOURNAL_SIZE(p_s_sb))) {
2359                 /* Note that it is required for blocksize of primary fs device and journal
2360                    device to be the same */
2361                 d_bh =
2362                     reiserfs_breada(journal->j_dev_bd, cur_dblock,
2363                                     p_s_sb->s_blocksize,
2364                                     SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2365                                     SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2366                 ret =
2367                     journal_transaction_is_valid(p_s_sb, d_bh,
2368                                                  &oldest_invalid_trans_id,
2369                                                  &newest_mount_id);
2370                 if (ret == 1) {
2371                         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2372                         if (oldest_start == 0) {        /* init all oldest_ values */
2373                                 oldest_trans_id = get_desc_trans_id(desc);
2374                                 oldest_start = d_bh->b_blocknr;
2375                                 newest_mount_id = get_desc_mount_id(desc);
2376                                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2377                                                "journal-1179: Setting "
2378                                                "oldest_start to offset %llu, trans_id %lu",
2379                                                oldest_start -
2380                                                SB_ONDISK_JOURNAL_1st_BLOCK
2381                                                (p_s_sb), oldest_trans_id);
2382                         } else if (oldest_trans_id > get_desc_trans_id(desc)) {
2383                                 /* one we just read was older */
2384                                 oldest_trans_id = get_desc_trans_id(desc);
2385                                 oldest_start = d_bh->b_blocknr;
2386                                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2387                                                "journal-1180: Resetting "
2388                                                "oldest_start to offset %lu, trans_id %lu",
2389                                                oldest_start -
2390                                                SB_ONDISK_JOURNAL_1st_BLOCK
2391                                                (p_s_sb), oldest_trans_id);
2392                         }
2393                         if (newest_mount_id < get_desc_mount_id(desc)) {
2394                                 newest_mount_id = get_desc_mount_id(desc);
2395                                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2396                                                "journal-1299: Setting "
2397                                                "newest_mount_id to %d",
2398                                                get_desc_mount_id(desc));
2399                         }
2400                         cur_dblock += get_desc_trans_len(desc) + 2;
2401                 } else {
2402                         cur_dblock++;
2403                 }
2404                 brelse(d_bh);
2405         }
2406
2407       start_log_replay:
2408         cur_dblock = oldest_start;
2409         if (oldest_trans_id) {
2410                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2411                                "journal-1206: Starting replay "
2412                                "from offset %llu, trans_id %lu",
2413                                cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2414                                oldest_trans_id);
2415
2416         }
2417         replay_count = 0;
2418         while (continue_replay && oldest_trans_id > 0) {
2419                 ret =
2420                     journal_read_transaction(p_s_sb, cur_dblock, oldest_start,
2421                                              oldest_trans_id, newest_mount_id);
2422                 if (ret < 0) {
2423                         return ret;
2424                 } else if (ret != 0) {
2425                         break;
2426                 }
2427                 cur_dblock =
2428                     SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + journal->j_start;
2429                 replay_count++;
2430                 if (cur_dblock == oldest_start)
2431                         break;
2432         }
2433
2434         if (oldest_trans_id == 0) {
2435                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2436                                "journal-1225: No valid " "transactions found");
2437         }
2438         /* j_start does not get set correctly if we don't replay any transactions.
2439          ** if we had a valid journal_header, set j_start to the first unflushed transaction value,
2440          ** copy the trans_id from the header
2441          */
2442         if (valid_journal_header && replay_count == 0) {
2443                 journal->j_start = le32_to_cpu(jh->j_first_unflushed_offset);
2444                 journal->j_trans_id =
2445                     le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2446                 journal->j_last_flush_trans_id =
2447                     le32_to_cpu(jh->j_last_flush_trans_id);
2448                 journal->j_mount_id = le32_to_cpu(jh->j_mount_id) + 1;
2449         } else {
2450                 journal->j_mount_id = newest_mount_id + 1;
2451         }
2452         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1299: Setting "
2453                        "newest_mount_id to %lu", journal->j_mount_id);
2454         journal->j_first_unflushed_offset = journal->j_start;
2455         if (replay_count > 0) {
2456                 reiserfs_info(p_s_sb,
2457                               "replayed %d transactions in %lu seconds\n",
2458                               replay_count, get_seconds() - start);
2459         }
2460         if (!bdev_read_only(p_s_sb->s_bdev) &&
2461             _update_journal_header_block(p_s_sb, journal->j_start,
2462                                          journal->j_last_flush_trans_id)) {
2463                 /* replay failed, caller must call free_journal_ram and abort
2464                  ** the mount
2465                  */
2466                 return -1;
2467         }
2468         return 0;
2469 }
2470
2471 static struct reiserfs_journal_list *alloc_journal_list(struct super_block *s)
2472 {
2473         struct reiserfs_journal_list *jl;
2474       retry:
2475         jl = reiserfs_kmalloc(sizeof(struct reiserfs_journal_list), GFP_NOFS,
2476                               s);
2477         if (!jl) {
2478                 yield();
2479                 goto retry;
2480         }
2481         memset(jl, 0, sizeof(*jl));
2482         INIT_LIST_HEAD(&jl->j_list);
2483         INIT_LIST_HEAD(&jl->j_working_list);
2484         INIT_LIST_HEAD(&jl->j_tail_bh_list);
2485         INIT_LIST_HEAD(&jl->j_bh_list);
2486         sema_init(&jl->j_commit_lock, 1);
2487         SB_JOURNAL(s)->j_num_lists++;
2488         get_journal_list(jl);
2489         return jl;
2490 }
2491
2492 static void journal_list_init(struct super_block *p_s_sb)
2493 {
2494         SB_JOURNAL(p_s_sb)->j_current_jl = alloc_journal_list(p_s_sb);
2495 }
2496
2497 static int release_journal_dev(struct super_block *super,
2498                                struct reiserfs_journal *journal)
2499 {
2500         int result;
2501
2502         result = 0;
2503
2504         if (journal->j_dev_file != NULL) {
2505                 result = filp_close(journal->j_dev_file, NULL);
2506                 journal->j_dev_file = NULL;
2507                 journal->j_dev_bd = NULL;
2508         } else if (journal->j_dev_bd != NULL) {
2509                 result = blkdev_put(journal->j_dev_bd);
2510                 journal->j_dev_bd = NULL;
2511         }
2512
2513         if (result != 0) {
2514                 reiserfs_warning(super,
2515                                  "sh-457: release_journal_dev: Cannot release journal device: %i",
2516                                  result);
2517         }
2518         return result;
2519 }
2520
2521 static int journal_init_dev(struct super_block *super,
2522                             struct reiserfs_journal *journal,
2523                             const char *jdev_name)
2524 {
2525         int result;
2526         dev_t jdev;
2527         int blkdev_mode = FMODE_READ | FMODE_WRITE;
2528         char b[BDEVNAME_SIZE];
2529
2530         result = 0;
2531
2532         journal->j_dev_bd = NULL;
2533         journal->j_dev_file = NULL;
2534         jdev = SB_ONDISK_JOURNAL_DEVICE(super) ?
2535             new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super)) : super->s_dev;
2536
2537         if (bdev_read_only(super->s_bdev))
2538                 blkdev_mode = FMODE_READ;
2539
2540         /* there is no "jdev" option and journal is on separate device */
2541         if ((!jdev_name || !jdev_name[0])) {
2542                 journal->j_dev_bd = open_by_devnum(jdev, blkdev_mode);
2543                 if (IS_ERR(journal->j_dev_bd)) {
2544                         result = PTR_ERR(journal->j_dev_bd);
2545                         journal->j_dev_bd = NULL;
2546                         reiserfs_warning(super, "sh-458: journal_init_dev: "
2547                                          "cannot init journal device '%s': %i",
2548                                          __bdevname(jdev, b), result);
2549                         return result;
2550                 } else if (jdev != super->s_dev)
2551                         set_blocksize(journal->j_dev_bd, super->s_blocksize);
2552                 return 0;
2553         }
2554
2555         journal->j_dev_file = filp_open(jdev_name, 0, 0);
2556         if (!IS_ERR(journal->j_dev_file)) {
2557                 struct inode *jdev_inode = journal->j_dev_file->f_mapping->host;
2558                 if (!S_ISBLK(jdev_inode->i_mode)) {
2559                         reiserfs_warning(super, "journal_init_dev: '%s' is "
2560                                          "not a block device", jdev_name);
2561                         result = -ENOTBLK;
2562                         release_journal_dev(super, journal);
2563                 } else {
2564                         /* ok */
2565                         journal->j_dev_bd = I_BDEV(jdev_inode);
2566                         set_blocksize(journal->j_dev_bd, super->s_blocksize);
2567                         reiserfs_info(super,
2568                                       "journal_init_dev: journal device: %s\n",
2569                                       bdevname(journal->j_dev_bd, b));
2570                 }
2571         } else {
2572                 result = PTR_ERR(journal->j_dev_file);
2573                 journal->j_dev_file = NULL;
2574                 reiserfs_warning(super,
2575                                  "journal_init_dev: Cannot open '%s': %i",
2576                                  jdev_name, result);
2577         }
2578         return result;
2579 }
2580
2581 /*
2582 ** must be called once on fs mount.  calls journal_read for you
2583 */
2584 int journal_init(struct super_block *p_s_sb, const char *j_dev_name,
2585                  int old_format, unsigned int commit_max_age)
2586 {
2587         int num_cnodes = SB_ONDISK_JOURNAL_SIZE(p_s_sb) * 2;
2588         struct buffer_head *bhjh;
2589         struct reiserfs_super_block *rs;
2590         struct reiserfs_journal_header *jh;
2591         struct reiserfs_journal *journal;
2592         struct reiserfs_journal_list *jl;
2593         char b[BDEVNAME_SIZE];
2594
2595         journal = SB_JOURNAL(p_s_sb) = vmalloc(sizeof(struct reiserfs_journal));
2596         if (!journal) {
2597                 reiserfs_warning(p_s_sb,
2598                                  "journal-1256: unable to get memory for journal structure");
2599                 return 1;
2600         }
2601         memset(journal, 0, sizeof(struct reiserfs_journal));
2602         INIT_LIST_HEAD(&journal->j_bitmap_nodes);
2603         INIT_LIST_HEAD(&journal->j_prealloc_list);
2604         INIT_LIST_HEAD(&journal->j_working_list);
2605         INIT_LIST_HEAD(&journal->j_journal_list);
2606         journal->j_persistent_trans = 0;
2607         if (reiserfs_allocate_list_bitmaps(p_s_sb,
2608                                            journal->j_list_bitmap,
2609                                            SB_BMAP_NR(p_s_sb)))
2610                 goto free_and_return;
2611         allocate_bitmap_nodes(p_s_sb);
2612
2613         /* reserved for journal area support */
2614         SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb) = (old_format ?
2615                                                  REISERFS_OLD_DISK_OFFSET_IN_BYTES
2616                                                  / p_s_sb->s_blocksize +
2617                                                  SB_BMAP_NR(p_s_sb) +
2618                                                  1 :
2619                                                  REISERFS_DISK_OFFSET_IN_BYTES /
2620                                                  p_s_sb->s_blocksize + 2);
2621
2622         /* Sanity check to see is the standard journal fitting withing first bitmap
2623            (actual for small blocksizes) */
2624         if (!SB_ONDISK_JOURNAL_DEVICE(p_s_sb) &&
2625             (SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb) +
2626              SB_ONDISK_JOURNAL_SIZE(p_s_sb) > p_s_sb->s_blocksize * 8)) {
2627                 reiserfs_warning(p_s_sb,
2628                                  "journal-1393: journal does not fit for area "
2629                                  "addressed by first of bitmap blocks. It starts at "
2630                                  "%u and its size is %u. Block size %ld",
2631                                  SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb),
2632                                  SB_ONDISK_JOURNAL_SIZE(p_s_sb),
2633                                  p_s_sb->s_blocksize);
2634                 goto free_and_return;
2635         }
2636
2637         if (journal_init_dev(p_s_sb, journal, j_dev_name) != 0) {
2638                 reiserfs_warning(p_s_sb,
2639                                  "sh-462: unable to initialize jornal device");
2640                 goto free_and_return;
2641         }
2642
2643         rs = SB_DISK_SUPER_BLOCK(p_s_sb);
2644
2645         /* read journal header */
2646         bhjh = journal_bread(p_s_sb,
2647                              SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2648                              SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2649         if (!bhjh) {
2650                 reiserfs_warning(p_s_sb,
2651                                  "sh-459: unable to read journal header");
2652                 goto free_and_return;
2653         }
2654         jh = (struct reiserfs_journal_header *)(bhjh->b_data);
2655
2656         /* make sure that journal matches to the super block */
2657         if (is_reiserfs_jr(rs)
2658             && (le32_to_cpu(jh->jh_journal.jp_journal_magic) !=
2659                 sb_jp_journal_magic(rs))) {
2660                 reiserfs_warning(p_s_sb,
2661                                  "sh-460: journal header magic %x "
2662                                  "(device %s) does not match to magic found in super "
2663                                  "block %x", jh->jh_journal.jp_journal_magic,
2664                                  bdevname(journal->j_dev_bd, b),
2665                                  sb_jp_journal_magic(rs));
2666                 brelse(bhjh);
2667                 goto free_and_return;
2668         }
2669
2670         journal->j_trans_max = le32_to_cpu(jh->jh_journal.jp_journal_trans_max);
2671         journal->j_max_batch = le32_to_cpu(jh->jh_journal.jp_journal_max_batch);
2672         journal->j_max_commit_age =
2673             le32_to_cpu(jh->jh_journal.jp_journal_max_commit_age);
2674         journal->j_max_trans_age = JOURNAL_MAX_TRANS_AGE;
2675
2676         if (journal->j_trans_max) {
2677                 /* make sure these parameters are available, assign it if they are not */
2678                 __u32 initial = journal->j_trans_max;
2679                 __u32 ratio = 1;
2680
2681                 if (p_s_sb->s_blocksize < 4096)
2682                         ratio = 4096 / p_s_sb->s_blocksize;
2683
2684                 if (SB_ONDISK_JOURNAL_SIZE(p_s_sb) / journal->j_trans_max <
2685                     JOURNAL_MIN_RATIO)
2686                         journal->j_trans_max =
2687                             SB_ONDISK_JOURNAL_SIZE(p_s_sb) / JOURNAL_MIN_RATIO;
2688                 if (journal->j_trans_max > JOURNAL_TRANS_MAX_DEFAULT / ratio)
2689                         journal->j_trans_max =
2690                             JOURNAL_TRANS_MAX_DEFAULT / ratio;
2691                 if (journal->j_trans_max < JOURNAL_TRANS_MIN_DEFAULT / ratio)
2692                         journal->j_trans_max =
2693                             JOURNAL_TRANS_MIN_DEFAULT / ratio;
2694
2695                 if (journal->j_trans_max != initial)
2696                         reiserfs_warning(p_s_sb,
2697                                          "sh-461: journal_init: wrong transaction max size (%u). Changed to %u",
2698                                          initial, journal->j_trans_max);
2699
2700                 journal->j_max_batch = journal->j_trans_max *
2701                     JOURNAL_MAX_BATCH_DEFAULT / JOURNAL_TRANS_MAX_DEFAULT;
2702         }
2703
2704         if (!journal->j_trans_max) {
2705                 /*we have the file system was created by old version of mkreiserfs
2706                    so this field contains zero value */
2707                 journal->j_trans_max = JOURNAL_TRANS_MAX_DEFAULT;
2708                 journal->j_max_batch = JOURNAL_MAX_BATCH_DEFAULT;
2709                 journal->j_max_commit_age = JOURNAL_MAX_COMMIT_AGE;
2710
2711                 /* for blocksize >= 4096 - max transaction size is 1024. For block size < 4096
2712                    trans max size is decreased proportionally */
2713                 if (p_s_sb->s_blocksize < 4096) {
2714                         journal->j_trans_max /= (4096 / p_s_sb->s_blocksize);
2715                         journal->j_max_batch = (journal->j_trans_max) * 9 / 10;
2716                 }
2717         }
2718
2719         journal->j_default_max_commit_age = journal->j_max_commit_age;
2720
2721         if (commit_max_age != 0) {
2722                 journal->j_max_commit_age = commit_max_age;
2723                 journal->j_max_trans_age = commit_max_age;
2724         }
2725
2726         reiserfs_info(p_s_sb, "journal params: device %s, size %u, "
2727                       "journal first block %u, max trans len %u, max batch %u, "
2728                       "max commit age %u, max trans age %u\n",
2729                       bdevname(journal->j_dev_bd, b),
2730                       SB_ONDISK_JOURNAL_SIZE(p_s_sb),
2731                       SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2732                       journal->j_trans_max,
2733                       journal->j_max_batch,
2734                       journal->j_max_commit_age, journal->j_max_trans_age);
2735
2736         brelse(bhjh);
2737
2738         journal->j_list_bitmap_index = 0;
2739         journal_list_init(p_s_sb);
2740
2741         memset(journal->j_list_hash_table, 0,
2742                JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
2743
2744         INIT_LIST_HEAD(&journal->j_dirty_buffers);
2745         spin_lock_init(&journal->j_dirty_buffers_lock);
2746
2747         journal->j_start = 0;
2748         journal->j_len = 0;
2749         journal->j_len_alloc = 0;
2750         atomic_set(&(journal->j_wcount), 0);
2751         atomic_set(&(journal->j_async_throttle), 0);
2752         journal->j_bcount = 0;
2753         journal->j_trans_start_time = 0;
2754         journal->j_last = NULL;
2755         journal->j_first = NULL;
2756         init_waitqueue_head(&(journal->j_join_wait));
2757         sema_init(&journal->j_lock, 1);
2758         sema_init(&journal->j_flush_sem, 1);
2759
2760         journal->j_trans_id = 10;
2761         journal->j_mount_id = 10;
2762         journal->j_state = 0;
2763         atomic_set(&(journal->j_jlock), 0);
2764         journal->j_cnode_free_list = allocate_cnodes(num_cnodes);
2765         journal->j_cnode_free_orig = journal->j_cnode_free_list;
2766         journal->j_cnode_free = journal->j_cnode_free_list ? num_cnodes : 0;
2767         journal->j_cnode_used = 0;
2768         journal->j_must_wait = 0;
2769
2770         if (journal->j_cnode_free == 0) {
2771                 reiserfs_warning(p_s_sb, "journal-2004: Journal cnode memory "
2772                                  "allocation failed (%ld bytes). Journal is "
2773                                  "too large for available memory. Usually "
2774                                  "this is due to a journal that is too large.",
2775                                  sizeof (struct reiserfs_journal_cnode) * num_cnodes);
2776                 goto free_and_return;
2777         }
2778
2779         init_journal_hash(p_s_sb);
2780         jl = journal->j_current_jl;
2781         jl->j_list_bitmap = get_list_bitmap(p_s_sb, jl);
2782         if (!jl->j_list_bitmap) {
2783                 reiserfs_warning(p_s_sb,
2784                                  "journal-2005, get_list_bitmap failed for journal list 0");
2785                 goto free_and_return;
2786         }
2787         if (journal_read(p_s_sb) < 0) {
2788                 reiserfs_warning(p_s_sb, "Replay Failure, unable to mount");
2789                 goto free_and_return;
2790         }
2791
2792         reiserfs_mounted_fs_count++;
2793         if (reiserfs_mounted_fs_count <= 1)
2794                 commit_wq = create_workqueue("reiserfs");
2795
2796         INIT_WORK(&journal->j_work, flush_async_commits, p_s_sb);
2797         return 0;
2798       free_and_return:
2799         free_journal_ram(p_s_sb);
2800         return 1;
2801 }
2802
2803 /*
2804 ** test for a polite end of the current transaction.  Used by file_write, and should
2805 ** be used by delete to make sure they don't write more than can fit inside a single
2806 ** transaction
2807 */
2808 int journal_transaction_should_end(struct reiserfs_transaction_handle *th,
2809                                    int new_alloc)
2810 {
2811         struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2812         time_t now = get_seconds();
2813         /* cannot restart while nested */
2814         BUG_ON(!th->t_trans_id);
2815         if (th->t_refcount > 1)
2816                 return 0;
2817         if (journal->j_must_wait > 0 ||
2818             (journal->j_len_alloc + new_alloc) >= journal->j_max_batch ||
2819             atomic_read(&(journal->j_jlock)) ||
2820             (now - journal->j_trans_start_time) > journal->j_max_trans_age ||
2821             journal->j_cnode_free < (journal->j_trans_max * 3)) {
2822                 return 1;
2823         }
2824         return 0;
2825 }
2826
2827 /* this must be called inside a transaction, and requires the
2828 ** kernel_lock to be held
2829 */
2830 void reiserfs_block_writes(struct reiserfs_transaction_handle *th)
2831 {
2832         struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2833         BUG_ON(!th->t_trans_id);
2834         journal->j_must_wait = 1;
2835         set_bit(J_WRITERS_BLOCKED, &journal->j_state);
2836         return;
2837 }
2838
2839 /* this must be called without a transaction started, and does not
2840 ** require BKL
2841 */
2842 void reiserfs_allow_writes(struct super_block *s)
2843 {
2844         struct reiserfs_journal *journal = SB_JOURNAL(s);
2845         clear_bit(J_WRITERS_BLOCKED, &journal->j_state);
2846         wake_up(&journal->j_join_wait);
2847 }
2848
2849 /* this must be called without a transaction started, and does not
2850 ** require BKL
2851 */
2852 void reiserfs_wait_on_write_block(struct super_block *s)
2853 {
2854         struct reiserfs_journal *journal = SB_JOURNAL(s);
2855         wait_event(journal->j_join_wait,
2856                    !test_bit(J_WRITERS_BLOCKED, &journal->j_state));
2857 }
2858
2859 static void queue_log_writer(struct super_block *s)
2860 {
2861         wait_queue_t wait;
2862         struct reiserfs_journal *journal = SB_JOURNAL(s);
2863         set_bit(J_WRITERS_QUEUED, &journal->j_state);
2864
2865         /*
2866          * we don't want to use wait_event here because
2867          * we only want to wait once.
2868          */
2869         init_waitqueue_entry(&wait, current);
2870         add_wait_queue(&journal->j_join_wait, &wait);
2871         set_current_state(TASK_UNINTERRUPTIBLE);
2872         if (test_bit(J_WRITERS_QUEUED, &journal->j_state))
2873                 schedule();
2874         current->state = TASK_RUNNING;
2875         remove_wait_queue(&journal->j_join_wait, &wait);
2876 }
2877
2878 static void wake_queued_writers(struct super_block *s)
2879 {
2880         struct reiserfs_journal *journal = SB_JOURNAL(s);
2881         if (test_and_clear_bit(J_WRITERS_QUEUED, &journal->j_state))
2882                 wake_up(&journal->j_join_wait);
2883 }
2884
2885 static void let_transaction_grow(struct super_block *sb, unsigned long trans_id)
2886 {
2887         struct reiserfs_journal *journal = SB_JOURNAL(sb);
2888         unsigned long bcount = journal->j_bcount;
2889         while (1) {
2890                 schedule_timeout_uninterruptible(1);
2891                 journal->j_current_jl->j_state |= LIST_COMMIT_PENDING;
2892                 while ((atomic_read(&journal->j_wcount) > 0 ||
2893                         atomic_read(&journal->j_jlock)) &&
2894                        journal->j_trans_id == trans_id) {
2895                         queue_log_writer(sb);
2896                 }
2897                 if (journal->j_trans_id != trans_id)
2898                         break;
2899                 if (bcount == journal->j_bcount)
2900                         break;
2901                 bcount = journal->j_bcount;
2902         }
2903 }
2904
2905 /* join == true if you must join an existing transaction.
2906 ** join == false if you can deal with waiting for others to finish
2907 **
2908 ** this will block until the transaction is joinable.  send the number of blocks you
2909 ** expect to use in nblocks.
2910 */
2911 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
2912                               struct super_block *p_s_sb, unsigned long nblocks,
2913                               int join)
2914 {
2915         time_t now = get_seconds();
2916         int old_trans_id;
2917         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
2918         struct reiserfs_transaction_handle myth;
2919         int sched_count = 0;
2920         int retval;
2921
2922         reiserfs_check_lock_depth(p_s_sb, "journal_begin");
2923         if (nblocks > journal->j_trans_max)
2924                 BUG();
2925
2926         PROC_INFO_INC(p_s_sb, journal.journal_being);
2927         /* set here for journal_join */
2928         th->t_refcount = 1;
2929         th->t_super = p_s_sb;
2930
2931       relock:
2932         lock_journal(p_s_sb);
2933         if (join != JBEGIN_ABORT && reiserfs_is_journal_aborted(journal)) {
2934                 unlock_journal(p_s_sb);
2935                 retval = journal->j_errno;
2936                 goto out_fail;
2937         }
2938         journal->j_bcount++;
2939
2940         if (test_bit(J_WRITERS_BLOCKED, &journal->j_state)) {
2941                 unlock_journal(p_s_sb);
2942                 reiserfs_wait_on_write_block(p_s_sb);
2943                 PROC_INFO_INC(p_s_sb, journal.journal_relock_writers);
2944                 goto relock;
2945         }
2946         now = get_seconds();
2947
2948         /* if there is no room in the journal OR
2949          ** if this transaction is too old, and we weren't called joinable, wait for it to finish before beginning
2950          ** we don't sleep if there aren't other writers
2951          */
2952
2953         if ((!join && journal->j_must_wait > 0) ||
2954             (!join
2955              && (journal->j_len_alloc + nblocks + 2) >= journal->j_max_batch)
2956             || (!join && atomic_read(&journal->j_wcount) > 0
2957                 && journal->j_trans_start_time > 0
2958                 && (now - journal->j_trans_start_time) >
2959                 journal->j_max_trans_age) || (!join
2960                                               && atomic_read(&journal->j_jlock))
2961             || (!join && journal->j_cnode_free < (journal->j_trans_max * 3))) {
2962
2963                 old_trans_id = journal->j_trans_id;
2964                 unlock_journal(p_s_sb); /* allow others to finish this transaction */
2965
2966                 if (!join && (journal->j_len_alloc + nblocks + 2) >=
2967                     journal->j_max_batch &&
2968                     ((journal->j_len + nblocks + 2) * 100) <
2969                     (journal->j_len_alloc * 75)) {
2970                         if (atomic_read(&journal->j_wcount) > 10) {
2971                                 sched_count++;
2972                                 queue_log_writer(p_s_sb);
2973                                 goto relock;
2974                         }
2975                 }
2976                 /* don't mess with joining the transaction if all we have to do is
2977                  * wait for someone else to do a commit
2978                  */
2979                 if (atomic_read(&journal->j_jlock)) {
2980                         while (journal->j_trans_id == old_trans_id &&
2981                                atomic_read(&journal->j_jlock)) {
2982                                 queue_log_writer(p_s_sb);
2983                         }
2984                         goto relock;
2985                 }
2986                 retval = journal_join(&myth, p_s_sb, 1);
2987                 if (retval)
2988                         goto out_fail;
2989
2990                 /* someone might have ended the transaction while we joined */
2991                 if (old_trans_id != journal->j_trans_id) {
2992                         retval = do_journal_end(&myth, p_s_sb, 1, 0);
2993                 } else {
2994                         retval = do_journal_end(&myth, p_s_sb, 1, COMMIT_NOW);
2995                 }
2996
2997                 if (retval)
2998                         goto out_fail;
2999
3000                 PROC_INFO_INC(p_s_sb, journal.journal_relock_wcount);
3001                 goto relock;
3002         }
3003         /* we are the first writer, set trans_id */
3004         if (journal->j_trans_start_time == 0) {
3005                 journal->j_trans_start_time = get_seconds();
3006         }
3007         atomic_inc(&(journal->j_wcount));
3008         journal->j_len_alloc += nblocks;
3009         th->t_blocks_logged = 0;
3010         th->t_blocks_allocated = nblocks;
3011         th->t_trans_id = journal->j_trans_id;
3012         unlock_journal(p_s_sb);
3013         INIT_LIST_HEAD(&th->t_list);
3014         get_fs_excl();
3015         return 0;
3016
3017       out_fail:
3018         memset(th, 0, sizeof(*th));
3019         /* Re-set th->t_super, so we can properly keep track of how many
3020          * persistent transactions there are. We need to do this so if this
3021          * call is part of a failed restart_transaction, we can free it later */
3022         th->t_super = p_s_sb;
3023         return retval;
3024 }
3025
3026 struct reiserfs_transaction_handle *reiserfs_persistent_transaction(struct
3027                                                                     super_block
3028                                                                     *s,
3029                                                                     int nblocks)
3030 {
3031         int ret;
3032         struct reiserfs_transaction_handle *th;
3033
3034         /* if we're nesting into an existing transaction.  It will be
3035          ** persistent on its own
3036          */
3037         if (reiserfs_transaction_running(s)) {
3038                 th = current->journal_info;
3039                 th->t_refcount++;
3040                 if (th->t_refcount < 2) {
3041                         BUG();
3042                 }
3043                 return th;
3044         }
3045         th = reiserfs_kmalloc(sizeof(struct reiserfs_transaction_handle),
3046                               GFP_NOFS, s);
3047         if (!th)
3048                 return NULL;
3049         ret = journal_begin(th, s, nblocks);
3050         if (ret) {
3051                 reiserfs_kfree(th, sizeof(struct reiserfs_transaction_handle),
3052                                s);
3053                 return NULL;
3054         }
3055
3056         SB_JOURNAL(s)->j_persistent_trans++;
3057         return th;
3058 }
3059
3060 int reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle *th)
3061 {
3062         struct super_block *s = th->t_super;
3063         int ret = 0;
3064         if (th->t_trans_id)
3065                 ret = journal_end(th, th->t_super, th->t_blocks_allocated);
3066         else
3067                 ret = -EIO;
3068         if (th->t_refcount == 0) {
3069                 SB_JOURNAL(s)->j_persistent_trans--;
3070                 reiserfs_kfree(th, sizeof(struct reiserfs_transaction_handle),
3071                                s);
3072         }
3073         return ret;
3074 }
3075
3076 static int journal_join(struct reiserfs_transaction_handle *th,
3077                         struct super_block *p_s_sb, unsigned long nblocks)
3078 {
3079         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3080
3081         /* this keeps do_journal_end from NULLing out the current->journal_info
3082          ** pointer
3083          */
3084         th->t_handle_save = cur_th;
3085         if (cur_th && cur_th->t_refcount > 1) {
3086                 BUG();
3087         }
3088         return do_journal_begin_r(th, p_s_sb, nblocks, JBEGIN_JOIN);
3089 }
3090
3091 int journal_join_abort(struct reiserfs_transaction_handle *th,
3092                        struct super_block *p_s_sb, unsigned long nblocks)
3093 {
3094         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3095
3096         /* this keeps do_journal_end from NULLing out the current->journal_info
3097          ** pointer
3098          */
3099         th->t_handle_save = cur_th;
3100         if (cur_th && cur_th->t_refcount > 1) {
3101                 BUG();
3102         }
3103         return do_journal_begin_r(th, p_s_sb, nblocks, JBEGIN_ABORT);
3104 }
3105
3106 int journal_begin(struct reiserfs_transaction_handle *th,
3107                   struct super_block *p_s_sb, unsigned long nblocks)
3108 {
3109         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3110         int ret;
3111
3112         th->t_handle_save = NULL;
3113         if (cur_th) {
3114                 /* we are nesting into the current transaction */
3115                 if (cur_th->t_super == p_s_sb) {
3116                         BUG_ON(!cur_th->t_refcount);
3117                         cur_th->t_refcount++;
3118                         memcpy(th, cur_th, sizeof(*th));
3119                         if (th->t_refcount <= 1)
3120                                 reiserfs_warning(p_s_sb,
3121                                                  "BAD: refcount <= 1, but journal_info != 0");
3122                         return 0;
3123                 } else {
3124                         /* we've ended up with a handle from a different filesystem.
3125                          ** save it and restore on journal_end.  This should never
3126                          ** really happen...
3127                          */
3128                         reiserfs_warning(p_s_sb,
3129                                          "clm-2100: nesting info a different FS");
3130                         th->t_handle_save = current->journal_info;
3131                         current->journal_info = th;
3132                 }
3133         } else {
3134                 current->journal_info = th;
3135         }
3136         ret = do_journal_begin_r(th, p_s_sb, nblocks, JBEGIN_REG);
3137         if (current->journal_info != th)
3138                 BUG();
3139
3140         /* I guess this boils down to being the reciprocal of clm-2100 above.
3141          * If do_journal_begin_r fails, we need to put it back, since journal_end
3142          * won't be called to do it. */
3143         if (ret)
3144                 current->journal_info = th->t_handle_save;
3145         else
3146                 BUG_ON(!th->t_refcount);
3147
3148         return ret;
3149 }
3150
3151 /*
3152 ** puts bh into the current transaction.  If it was already there, reorders removes the
3153 ** old pointers from the hash, and puts new ones in (to make sure replay happen in the right order).
3154 **
3155 ** if it was dirty, cleans and files onto the clean list.  I can't let it be dirty again until the
3156 ** transaction is committed.
3157 **
3158 ** if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len.
3159 */
3160 int journal_mark_dirty(struct reiserfs_transaction_handle *th,
3161                        struct super_block *p_s_sb, struct buffer_head *bh)
3162 {
3163         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3164         struct reiserfs_journal_cnode *cn = NULL;
3165         int count_already_incd = 0;
3166         int prepared = 0;
3167         BUG_ON(!th->t_trans_id);
3168
3169         PROC_INFO_INC(p_s_sb, journal.mark_dirty);
3170         if (th->t_trans_id != journal->j_trans_id) {
3171                 reiserfs_panic(th->t_super,
3172                                "journal-1577: handle trans id %ld != current trans id %ld\n",
3173                                th->t_trans_id, journal->j_trans_id);
3174         }
3175
3176         p_s_sb->s_dirt = 1;
3177
3178         prepared = test_clear_buffer_journal_prepared(bh);
3179         clear_buffer_journal_restore_dirty(bh);
3180         /* already in this transaction, we are done */
3181         if (buffer_journaled(bh)) {
3182                 PROC_INFO_INC(p_s_sb, journal.mark_dirty_already);
3183                 return 0;
3184         }
3185
3186         /* this must be turned into a panic instead of a warning.  We can't allow
3187          ** a dirty or journal_dirty or locked buffer to be logged, as some changes
3188          ** could get to disk too early.  NOT GOOD.
3189          */
3190         if (!prepared || buffer_dirty(bh)) {
3191                 reiserfs_warning(p_s_sb, "journal-1777: buffer %llu bad state "
3192                                  "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT",
3193                                  (unsigned long long)bh->b_blocknr,
3194                                  prepared ? ' ' : '!',
3195                                  buffer_locked(bh) ? ' ' : '!',
3196                                  buffer_dirty(bh) ? ' ' : '!',
3197                                  buffer_journal_dirty(bh) ? ' ' : '!');
3198         }
3199
3200         if (atomic_read(&(journal->j_wcount)) <= 0) {
3201                 reiserfs_warning(p_s_sb,
3202                                  "journal-1409: journal_mark_dirty returning because j_wcount was %d",
3203                                  atomic_read(&(journal->j_wcount)));
3204                 return 1;
3205         }
3206         /* this error means I've screwed up, and we've overflowed the transaction.
3207          ** Nothing can be done here, except make the FS readonly or panic.
3208          */
3209         if (journal->j_len >= journal->j_trans_max) {
3210                 reiserfs_panic(th->t_super,
3211                                "journal-1413: journal_mark_dirty: j_len (%lu) is too big\n",
3212                                journal->j_len);
3213         }
3214
3215         if (buffer_journal_dirty(bh)) {
3216                 count_already_incd = 1;
3217                 PROC_INFO_INC(p_s_sb, journal.mark_dirty_notjournal);
3218                 clear_buffer_journal_dirty(bh);
3219         }
3220
3221         if (journal->j_len > journal->j_len_alloc) {
3222                 journal->j_len_alloc = journal->j_len + JOURNAL_PER_BALANCE_CNT;
3223         }
3224
3225         set_buffer_journaled(bh);
3226
3227         /* now put this guy on the end */
3228         if (!cn) {
3229                 cn = get_cnode(p_s_sb);
3230                 if (!cn) {
3231                         reiserfs_panic(p_s_sb, "get_cnode failed!\n");
3232                 }
3233
3234                 if (th->t_blocks_logged == th->t_blocks_allocated) {
3235                         th->t_blocks_allocated += JOURNAL_PER_BALANCE_CNT;
3236                         journal->j_len_alloc += JOURNAL_PER_BALANCE_CNT;
3237                 }
3238                 th->t_blocks_logged++;
3239                 journal->j_len++;
3240
3241                 cn->bh = bh;
3242                 cn->blocknr = bh->b_blocknr;
3243                 cn->sb = p_s_sb;
3244                 cn->jlist = NULL;
3245                 insert_journal_hash(journal->j_hash_table, cn);
3246                 if (!count_already_incd) {
3247                         get_bh(bh);
3248                 }
3249         }
3250         cn->next = NULL;
3251         cn->prev = journal->j_last;
3252         cn->bh = bh;
3253         if (journal->j_last) {
3254                 journal->j_last->next = cn;
3255                 journal->j_last = cn;
3256         } else {
3257                 journal->j_first = cn;
3258                 journal->j_last = cn;
3259         }
3260         return 0;
3261 }
3262
3263 int journal_end(struct reiserfs_transaction_handle *th,
3264                 struct super_block *p_s_sb, unsigned long nblocks)
3265 {
3266         if (!current->journal_info && th->t_refcount > 1)
3267                 reiserfs_warning(p_s_sb, "REISER-NESTING: th NULL, refcount %d",
3268                                  th->t_refcount);
3269
3270         if (!th->t_trans_id) {
3271                 WARN_ON(1);
3272                 return -EIO;
3273         }
3274
3275         th->t_refcount--;
3276         if (th->t_refcount > 0) {
3277                 struct reiserfs_transaction_handle *cur_th =
3278                     current->journal_info;
3279
3280                 /* we aren't allowed to close a nested transaction on a different
3281                  ** filesystem from the one in the task struct
3282                  */
3283                 if (cur_th->t_super != th->t_super)
3284                         BUG();
3285
3286                 if (th != cur_th) {
3287                         memcpy(current->journal_info, th, sizeof(*th));
3288                         th->t_trans_id = 0;
3289                 }
3290                 return 0;
3291         } else {
3292                 return do_journal_end(th, p_s_sb, nblocks, 0);
3293         }
3294 }
3295
3296 /* removes from the current transaction, relsing and descrementing any counters.
3297 ** also files the removed buffer directly onto the clean list
3298 **
3299 ** called by journal_mark_freed when a block has been deleted
3300 **
3301 ** returns 1 if it cleaned and relsed the buffer. 0 otherwise
3302 */
3303 static int remove_from_transaction(struct super_block *p_s_sb,
3304                                    b_blocknr_t blocknr, int already_cleaned)
3305 {
3306         struct buffer_head *bh;
3307         struct reiserfs_journal_cnode *cn;
3308         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3309         int ret = 0;
3310
3311         cn = get_journal_hash_dev(p_s_sb, journal->j_hash_table, blocknr);
3312         if (!cn || !cn->bh) {
3313                 return ret;
3314         }
3315         bh = cn->bh;
3316         if (cn->prev) {
3317                 cn->prev->next = cn->next;
3318         }
3319         if (cn->next) {
3320                 cn->next->prev = cn->prev;
3321         }
3322         if (cn == journal->j_first) {
3323                 journal->j_first = cn->next;
3324         }
3325         if (cn == journal->j_last) {
3326                 journal->j_last = cn->prev;
3327         }
3328         if (bh)
3329                 remove_journal_hash(p_s_sb, journal->j_hash_table, NULL,
3330                                     bh->b_blocknr, 0);
3331         clear_buffer_journaled(bh);     /* don't log this one */
3332
3333         if (!already_cleaned) {
3334                 clear_buffer_journal_dirty(bh);
3335                 clear_buffer_dirty(bh);
3336                 clear_buffer_journal_test(bh);
3337                 put_bh(bh);
3338                 if (atomic_read(&(bh->b_count)) < 0) {
3339                         reiserfs_warning(p_s_sb,
3340                                          "journal-1752: remove from trans, b_count < 0");
3341                 }
3342                 ret = 1;
3343         }
3344         journal->j_len--;
3345         journal->j_len_alloc--;
3346         free_cnode(p_s_sb, cn);
3347         return ret;
3348 }
3349
3350 /*
3351 ** for any cnode in a journal list, it can only be dirtied of all the
3352 ** transactions that include it are commited to disk.
3353 ** this checks through each transaction, and returns 1 if you are allowed to dirty,
3354 ** and 0 if you aren't
3355 **
3356 ** it is called by dirty_journal_list, which is called after flush_commit_list has gotten all the log
3357 ** blocks for a given transaction on disk
3358 **
3359 */
3360 static int can_dirty(struct reiserfs_journal_cnode *cn)
3361 {
3362         struct super_block *sb = cn->sb;
3363         b_blocknr_t blocknr = cn->blocknr;
3364         struct reiserfs_journal_cnode *cur = cn->hprev;
3365         int can_dirty = 1;
3366
3367         /* first test hprev.  These are all newer than cn, so any node here
3368          ** with the same block number and dev means this node can't be sent
3369          ** to disk right now.
3370          */
3371         while (cur && can_dirty) {
3372                 if (cur->jlist && cur->bh && cur->blocknr && cur->sb == sb &&
3373                     cur->blocknr == blocknr) {
3374                         can_dirty = 0;
3375                 }
3376                 cur = cur->hprev;
3377         }
3378         /* then test hnext.  These are all older than cn.  As long as they
3379          ** are committed to the log, it is safe to write cn to disk
3380          */
3381         cur = cn->hnext;
3382         while (cur && can_dirty) {
3383                 if (cur->jlist && cur->jlist->j_len > 0 &&
3384                     atomic_read(&(cur->jlist->j_commit_left)) > 0 && cur->bh &&
3385                     cur->blocknr && cur->sb == sb && cur->blocknr == blocknr) {
3386                         can_dirty = 0;
3387                 }
3388                 cur = cur->hnext;
3389         }
3390         return can_dirty;
3391 }
3392
3393 /* syncs the commit blocks, but does not force the real buffers to disk
3394 ** will wait until the current transaction is done/commited before returning
3395 */
3396 int journal_end_sync(struct reiserfs_transaction_handle *th,
3397                      struct super_block *p_s_sb, unsigned long nblocks)
3398 {
3399         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3400
3401         BUG_ON(!th->t_trans_id);
3402         /* you can sync while nested, very, very bad */
3403         if (th->t_refcount > 1) {
3404                 BUG();
3405         }
3406         if (journal->j_len == 0) {
3407                 reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb),
3408                                              1);
3409                 journal_mark_dirty(th, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb));
3410         }
3411         return do_journal_end(th, p_s_sb, nblocks, COMMIT_NOW | WAIT);
3412 }
3413
3414 /*
3415 ** writeback the pending async commits to disk
3416 */
3417 static void flush_async_commits(void *p)
3418 {
3419         struct super_block *p_s_sb = p;
3420         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3421         struct reiserfs_journal_list *jl;
3422         struct list_head *entry;
3423
3424         lock_kernel();
3425         if (!list_empty(&journal->j_journal_list)) {
3426                 /* last entry is the youngest, commit it and you get everything */
3427                 entry = journal->j_journal_list.prev;
3428                 jl = JOURNAL_LIST_ENTRY(entry);
3429                 flush_commit_list(p_s_sb, jl, 1);
3430         }
3431         unlock_kernel();
3432         /*
3433          * this is a little racey, but there's no harm in missing
3434          * the filemap_fdata_write
3435          */
3436         if (!atomic_read(&journal->j_async_throttle)
3437             && !reiserfs_is_journal_aborted(journal)) {
3438                 atomic_inc(&journal->j_async_throttle);
3439                 filemap_fdatawrite(p_s_sb->s_bdev->bd_inode->i_mapping);
3440                 atomic_dec(&journal->j_async_throttle);
3441         }
3442 }
3443
3444 /*
3445 ** flushes any old transactions to disk
3446 ** ends the current transaction if it is too old
3447 */
3448 int reiserfs_flush_old_commits(struct super_block *p_s_sb)
3449 {
3450         time_t now;
3451         struct reiserfs_transaction_handle th;
3452         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3453
3454         now = get_seconds();
3455         /* safety check so we don't flush while we are replaying the log during
3456          * mount
3457          */
3458         if (list_empty(&journal->j_journal_list)) {
3459                 return 0;
3460         }
3461
3462         /* check the current transaction.  If there are no writers, and it is
3463          * too old, finish it, and force the commit blocks to disk
3464          */
3465         if (atomic_read(&journal->j_wcount) <= 0 &&
3466             journal->j_trans_start_time > 0 &&
3467             journal->j_len > 0 &&
3468             (now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3469                 if (!journal_join(&th, p_s_sb, 1)) {
3470                         reiserfs_prepare_for_journal(p_s_sb,
3471                                                      SB_BUFFER_WITH_SB(p_s_sb),
3472                                                      1);
3473                         journal_mark_dirty(&th, p_s_sb,
3474                                            SB_BUFFER_WITH_SB(p_s_sb));
3475
3476                         /* we're only being called from kreiserfsd, it makes no sense to do
3477                          ** an async commit so that kreiserfsd can do it later
3478                          */
3479                         do_journal_end(&th, p_s_sb, 1, COMMIT_NOW | WAIT);
3480                 }
3481         }
3482         return p_s_sb->s_dirt;
3483 }
3484
3485 /*
3486 ** returns 0 if do_journal_end should return right away, returns 1 if do_journal_end should finish the commit
3487 **
3488 ** if the current transaction is too old, but still has writers, this will wait on j_join_wait until all
3489 ** the writers are done.  By the time it wakes up, the transaction it was called has already ended, so it just
3490 ** flushes the commit list and returns 0.
3491 **
3492 ** Won't batch when flush or commit_now is set.  Also won't batch when others are waiting on j_join_wait.
3493 **
3494 ** Note, we can't allow the journal_end to proceed while there are still writers in the log.
3495 */
3496 static int check_journal_end(struct reiserfs_transaction_handle *th,
3497                              struct super_block *p_s_sb, unsigned long nblocks,
3498                              int flags)
3499 {
3500
3501         time_t now;
3502         int flush = flags & FLUSH_ALL;
3503         int commit_now = flags & COMMIT_NOW;
3504         int wait_on_commit = flags & WAIT;
3505         struct reiserfs_journal_list *jl;
3506         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3507
3508         BUG_ON(!th->t_trans_id);
3509
3510         if (th->t_trans_id != journal->j_trans_id) {
3511                 reiserfs_panic(th->t_super,
3512                                "journal-1577: handle trans id %ld != current trans id %ld\n",
3513                                th->t_trans_id, journal->j_trans_id);
3514         }
3515
3516         journal->j_len_alloc -= (th->t_blocks_allocated - th->t_blocks_logged);
3517         if (atomic_read(&(journal->j_wcount)) > 0) {    /* <= 0 is allowed.  unmounting might not call begin */
3518                 atomic_dec(&(journal->j_wcount));
3519         }
3520
3521         /* BUG, deal with case where j_len is 0, but people previously freed blocks need to be released
3522          ** will be dealt with by next transaction that actually writes something, but should be taken
3523          ** care of in this trans
3524          */
3525         if (journal->j_len == 0) {
3526                 BUG();
3527         }
3528         /* if wcount > 0, and we are called to with flush or commit_now,
3529          ** we wait on j_join_wait.  We will wake up when the last writer has
3530          ** finished the transaction, and started it on its way to the disk.
3531          ** Then, we flush the commit or journal list, and just return 0
3532          ** because the rest of journal end was already done for this transaction.
3533          */
3534         if (atomic_read(&(journal->j_wcount)) > 0) {
3535                 if (flush || commit_now) {
3536                         unsigned trans_id;
3537
3538                         jl = journal->j_current_jl;
3539                         trans_id = jl->j_trans_id;
3540                         if (wait_on_commit)
3541                                 jl->j_state |= LIST_COMMIT_PENDING;
3542                         atomic_set(&(journal->j_jlock), 1);
3543                         if (flush) {
3544                                 journal->j_next_full_flush = 1;
3545                         }
3546                         unlock_journal(p_s_sb);
3547
3548                         /* sleep while the current transaction is still j_jlocked */
3549                         while (journal->j_trans_id == trans_id) {
3550                                 if (atomic_read(&journal->j_jlock)) {
3551                                         queue_log_writer(p_s_sb);
3552                                 } else {
3553                                         lock_journal(p_s_sb);
3554                                         if (journal->j_trans_id == trans_id) {
3555                                                 atomic_set(&(journal->j_jlock),
3556                                                            1);
3557                                         }
3558                                         unlock_journal(p_s_sb);
3559                                 }
3560                         }
3561                         if (journal->j_trans_id == trans_id) {
3562                                 BUG();
3563                         }
3564                         if (commit_now
3565                             && journal_list_still_alive(p_s_sb, trans_id)
3566                             && wait_on_commit) {
3567                                 flush_commit_list(p_s_sb, jl, 1);
3568                         }
3569                         return 0;
3570                 }
3571                 unlock_journal(p_s_sb);
3572                 return 0;
3573         }
3574
3575         /* deal with old transactions where we are the last writers */
3576         now = get_seconds();
3577         if ((now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3578                 commit_now = 1;
3579                 journal->j_next_async_flush = 1;
3580         }
3581         /* don't batch when someone is waiting on j_join_wait */
3582         /* don't batch when syncing the commit or flushing the whole trans */
3583         if (!(journal->j_must_wait > 0) && !(atomic_read(&(journal->j_jlock)))
3584             && !flush && !commit_now && (journal->j_len < journal->j_max_batch)
3585             && journal->j_len_alloc < journal->j_max_batch
3586             && journal->j_cnode_free > (journal->j_trans_max * 3)) {
3587                 journal->j_bcount++;
3588                 unlock_journal(p_s_sb);
3589                 return 0;
3590         }
3591
3592         if (journal->j_start > SB_ONDISK_JOURNAL_SIZE(p_s_sb)) {
3593                 reiserfs_panic(p_s_sb,
3594                                "journal-003: journal_end: j_start (%ld) is too high\n",
3595                                journal->j_start);
3596         }
3597         return 1;
3598 }
3599
3600 /*
3601 ** Does all the work that makes deleting blocks safe.
3602 ** when deleting a block mark BH_JNew, just remove it from the current transaction, clean it's buffer_head and move on.
3603 **
3604 ** otherwise:
3605 ** set a bit for the block in the journal bitmap.  That will prevent it from being allocated for unformatted nodes
3606 ** before this transaction has finished.
3607 **
3608 ** mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers.  That will prevent any old transactions with
3609 ** this block from trying to flush to the real location.  Since we aren't removing the cnode from the journal_list_hash,
3610 ** the block can't be reallocated yet.
3611 **
3612 ** Then remove it from the current transaction, decrementing any counters and filing it on the clean list.
3613 */
3614 int journal_mark_freed(struct reiserfs_transaction_handle *th,
3615                        struct super_block *p_s_sb, b_blocknr_t blocknr)
3616 {
3617         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3618         struct reiserfs_journal_cnode *cn = NULL;
3619         struct buffer_head *bh = NULL;
3620         struct reiserfs_list_bitmap *jb = NULL;
3621         int cleaned = 0;
3622         BUG_ON(!th->t_trans_id);
3623
3624         cn = get_journal_hash_dev(p_s_sb, journal->j_hash_table, blocknr);
3625         if (cn && cn->bh) {
3626                 bh = cn->bh;
3627                 get_bh(bh);
3628         }
3629         /* if it is journal new, we just remove it from this transaction */
3630         if (bh && buffer_journal_new(bh)) {
3631                 clear_buffer_journal_new(bh);
3632                 clear_prepared_bits(bh);
3633                 reiserfs_clean_and_file_buffer(bh);
3634                 cleaned = remove_from_transaction(p_s_sb, blocknr, cleaned);
3635         } else {
3636                 /* set the bit for this block in the journal bitmap for this transaction */
3637                 jb = journal->j_current_jl->j_list_bitmap;
3638                 if (!jb) {
3639                         reiserfs_panic(p_s_sb,
3640                                        "journal-1702: journal_mark_freed, journal_list_bitmap is NULL\n");
3641                 }
3642                 set_bit_in_list_bitmap(p_s_sb, blocknr, jb);
3643
3644                 /* Note, the entire while loop is not allowed to schedule.  */
3645
3646                 if (bh) {
3647                         clear_prepared_bits(bh);
3648                         reiserfs_clean_and_file_buffer(bh);
3649                 }
3650                 cleaned = remove_from_transaction(p_s_sb, blocknr, cleaned);
3651
3652                 /* find all older transactions with this block, make sure they don't try to write it out */
3653                 cn = get_journal_hash_dev(p_s_sb, journal->j_list_hash_table,
3654                                           blocknr);
3655                 while (cn) {
3656                         if (p_s_sb == cn->sb && blocknr == cn->blocknr) {
3657                                 set_bit(BLOCK_FREED, &cn->state);
3658                                 if (cn->bh) {
3659                                         if (!cleaned) {
3660                                                 /* remove_from_transaction will brelse the buffer if it was
3661                                                  ** in the current trans
3662                                                  */
3663                                                 clear_buffer_journal_dirty(cn->
3664                                                                            bh);
3665                                                 clear_buffer_dirty(cn->bh);
3666                                                 clear_buffer_journal_test(cn->
3667                                                                           bh);
3668                                                 cleaned = 1;
3669                                                 put_bh(cn->bh);
3670                                                 if (atomic_read
3671                                                     (&(cn->bh->b_count)) < 0) {
3672                                                         reiserfs_warning(p_s_sb,
3673                                                                          "journal-2138: cn->bh->b_count < 0");
3674                                                 }
3675                                         }
3676                                         if (cn->jlist) {        /* since we are clearing the bh, we MUST dec nonzerolen */
3677                                                 atomic_dec(&
3678                                                            (cn->jlist->
3679                                                             j_nonzerolen));
3680                                         }
3681                                         cn->bh = NULL;
3682                                 }
3683                         }
3684                         cn = cn->hnext;
3685                 }
3686         }
3687
3688         if (bh) {
3689                 put_bh(bh);     /* get_hash grabs the buffer */
3690                 if (atomic_read(&(bh->b_count)) < 0) {
3691                         reiserfs_warning(p_s_sb,
3692                                          "journal-2165: bh->b_count < 0");
3693                 }
3694         }
3695         return 0;
3696 }
3697
3698 void reiserfs_update_inode_transaction(struct inode *inode)
3699 {
3700         struct reiserfs_journal *journal = SB_JOURNAL(inode->i_sb);
3701         REISERFS_I(inode)->i_jl = journal->j_current_jl;
3702         REISERFS_I(inode)->i_trans_id = journal->j_trans_id;
3703 }
3704
3705 /*
3706  * returns -1 on error, 0 if no commits/barriers were done and 1
3707  * if a transaction was actually committed and the barrier was done
3708  */
3709 static int __commit_trans_jl(struct inode *inode, unsigned long id,
3710                              struct reiserfs_journal_list *jl)
3711 {
3712         struct reiserfs_transaction_handle th;
3713         struct super_block *sb = inode->i_sb;
3714         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3715         int ret = 0;
3716
3717         /* is it from the current transaction, or from an unknown transaction? */
3718         if (id == journal->j_trans_id) {
3719                 jl = journal->j_current_jl;
3720                 /* try to let other writers come in and grow this transaction */
3721                 let_transaction_grow(sb, id);
3722                 if (journal->j_trans_id != id) {
3723                         goto flush_commit_only;
3724                 }
3725
3726                 ret = journal_begin(&th, sb, 1);
3727                 if (ret)
3728                         return ret;
3729
3730                 /* someone might have ended this transaction while we joined */
3731                 if (journal->j_trans_id != id) {
3732                         reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3733                                                      1);
3734                         journal_mark_dirty(&th, sb, SB_BUFFER_WITH_SB(sb));
3735                         ret = journal_end(&th, sb, 1);
3736                         goto flush_commit_only;
3737                 }
3738
3739                 ret = journal_end_sync(&th, sb, 1);
3740                 if (!ret)
3741                         ret = 1;
3742
3743         } else {
3744                 /* this gets tricky, we have to make sure the journal list in
3745                  * the inode still exists.  We know the list is still around
3746                  * if we've got a larger transaction id than the oldest list
3747                  */
3748               flush_commit_only:
3749                 if (journal_list_still_alive(inode->i_sb, id)) {
3750                         /*
3751                          * we only set ret to 1 when we know for sure
3752                          * the barrier hasn't been started yet on the commit
3753                          * block.
3754                          */
3755                         if (atomic_read(&jl->j_commit_left) > 1)
3756                                 ret = 1;
3757                         flush_commit_list(sb, jl, 1);
3758                         if (journal->j_errno)
3759                                 ret = journal->j_errno;
3760                 }
3761         }
3762         /* otherwise the list is gone, and long since committed */
3763         return ret;
3764 }
3765
3766 int reiserfs_commit_for_inode(struct inode *inode)
3767 {
3768         unsigned long id = REISERFS_I(inode)->i_trans_id;
3769         struct reiserfs_journal_list *jl = REISERFS_I(inode)->i_jl;
3770
3771         /* for the whole inode, assume unset id means it was
3772          * changed in the current transaction.  More conservative
3773          */
3774         if (!id || !jl) {
3775                 reiserfs_update_inode_transaction(inode);
3776                 id = REISERFS_I(inode)->i_trans_id;
3777                 /* jl will be updated in __commit_trans_jl */
3778         }
3779
3780         return __commit_trans_jl(inode, id, jl);
3781 }
3782
3783 void reiserfs_restore_prepared_buffer(struct super_block *p_s_sb,
3784                                       struct buffer_head *bh)
3785 {
3786         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3787         PROC_INFO_INC(p_s_sb, journal.restore_prepared);
3788         if (!bh) {
3789                 return;
3790         }
3791         if (test_clear_buffer_journal_restore_dirty(bh) &&
3792             buffer_journal_dirty(bh)) {
3793                 struct reiserfs_journal_cnode *cn;
3794                 cn = get_journal_hash_dev(p_s_sb,
3795                                           journal->j_list_hash_table,
3796                                           bh->b_blocknr);
3797                 if (cn && can_dirty(cn)) {
3798                         set_buffer_journal_test(bh);
3799                         mark_buffer_dirty(bh);
3800                 }
3801         }
3802         clear_buffer_journal_prepared(bh);
3803 }
3804
3805 extern struct tree_balance *cur_tb;
3806 /*
3807 ** before we can change a metadata block, we have to make sure it won't
3808 ** be written to disk while we are altering it.  So, we must:
3809 ** clean it
3810 ** wait on it.
3811 **
3812 */
3813 int reiserfs_prepare_for_journal(struct super_block *p_s_sb,
3814                                  struct buffer_head *bh, int wait)
3815 {
3816         PROC_INFO_INC(p_s_sb, journal.prepare);
3817
3818         if (test_set_buffer_locked(bh)) {
3819                 if (!wait)
3820                         return 0;
3821                 lock_buffer(bh);
3822         }
3823         set_buffer_journal_prepared(bh);
3824         if (test_clear_buffer_dirty(bh) && buffer_journal_dirty(bh)) {
3825                 clear_buffer_journal_test(bh);
3826                 set_buffer_journal_restore_dirty(bh);
3827         }
3828         unlock_buffer(bh);
3829         return 1;
3830 }
3831
3832 static void flush_old_journal_lists(struct super_block *s)
3833 {
3834         struct reiserfs_journal *journal = SB_JOURNAL(s);
3835         struct reiserfs_journal_list *jl;
3836         struct list_head *entry;
3837         time_t now = get_seconds();
3838
3839         while (!list_empty(&journal->j_journal_list)) {
3840                 entry = journal->j_journal_list.next;
3841                 jl = JOURNAL_LIST_ENTRY(entry);
3842                 /* this check should always be run, to send old lists to disk */
3843                 if (jl->j_timestamp < (now - (JOURNAL_MAX_TRANS_AGE * 4))) {
3844                         flush_used_journal_lists(s, jl);
3845                 } else {
3846                         break;
3847                 }
3848         }
3849 }
3850
3851 /*
3852 ** long and ugly.  If flush, will not return until all commit
3853 ** blocks and all real buffers in the trans are on disk.
3854 ** If no_async, won't return until all commit blocks are on disk.
3855 **
3856 ** keep reading, there are comments as you go along
3857 **
3858 ** If the journal is aborted, we just clean up. Things like flushing
3859 ** journal lists, etc just won't happen.
3860 */
3861 static int do_journal_end(struct reiserfs_transaction_handle *th,
3862                           struct super_block *p_s_sb, unsigned long nblocks,
3863                           int flags)
3864 {
3865         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3866         struct reiserfs_journal_cnode *cn, *next, *jl_cn;
3867         struct reiserfs_journal_cnode *last_cn = NULL;
3868         struct reiserfs_journal_desc *desc;
3869         struct reiserfs_journal_commit *commit;
3870         struct buffer_head *c_bh;       /* commit bh */
3871         struct buffer_head *d_bh;       /* desc bh */
3872         int cur_write_start = 0;        /* start index of current log write */
3873         int old_start;
3874         int i;
3875         int flush = flags & FLUSH_ALL;
3876         int wait_on_commit = flags & WAIT;
3877         struct reiserfs_journal_list *jl, *temp_jl;
3878         struct list_head *entry, *safe;
3879         unsigned long jindex;
3880         unsigned long commit_trans_id;
3881         int trans_half;
3882
3883         BUG_ON(th->t_refcount > 1);
3884         BUG_ON(!th->t_trans_id);
3885
3886         put_fs_excl();
3887         current->journal_info = th->t_handle_save;
3888         reiserfs_check_lock_depth(p_s_sb, "journal end");
3889         if (journal->j_len == 0) {
3890                 reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb),
3891                                              1);
3892                 journal_mark_dirty(th, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb));
3893         }
3894
3895         lock_journal(p_s_sb);
3896         if (journal->j_next_full_flush) {
3897                 flags |= FLUSH_ALL;
3898                 flush = 1;
3899         }
3900         if (journal->j_next_async_flush) {
3901                 flags |= COMMIT_NOW | WAIT;
3902                 wait_on_commit = 1;
3903         }
3904
3905         /* check_journal_end locks the journal, and unlocks if it does not return 1
3906          ** it tells us if we should continue with the journal_end, or just return
3907          */
3908         if (!check_journal_end(th, p_s_sb, nblocks, flags)) {
3909                 p_s_sb->s_dirt = 1;
3910                 wake_queued_writers(p_s_sb);
3911                 reiserfs_async_progress_wait(p_s_sb);
3912                 goto out;
3913         }
3914
3915         /* check_journal_end might set these, check again */
3916         if (journal->j_next_full_flush) {
3917                 flush = 1;
3918         }
3919
3920         /*
3921          ** j must wait means we have to flush the log blocks, and the real blocks for
3922          ** this transaction
3923          */
3924         if (journal->j_must_wait > 0) {
3925                 flush = 1;
3926         }
3927 #ifdef REISERFS_PREALLOCATE
3928         /* quota ops might need to nest, setup the journal_info pointer for them
3929          * and raise the refcount so that it is > 0. */
3930         current->journal_info = th;
3931         th->t_refcount++;
3932         reiserfs_discard_all_prealloc(th);      /* it should not involve new blocks into
3933                                                  * the transaction */
3934         th->t_refcount--;
3935         current->journal_info = th->t_handle_save;
3936 #endif
3937
3938         /* setup description block */
3939         d_bh =
3940             journal_getblk(p_s_sb,
3941                            SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
3942                            journal->j_start);
3943         set_buffer_uptodate(d_bh);
3944         desc = (struct reiserfs_journal_desc *)(d_bh)->b_data;
3945         memset(d_bh->b_data, 0, d_bh->b_size);
3946         memcpy(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8);
3947         set_desc_trans_id(desc, journal->j_trans_id);
3948
3949         /* setup commit block.  Don't write (keep it clean too) this one until after everyone else is written */
3950         c_bh = journal_getblk(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
3951                               ((journal->j_start + journal->j_len +
3952                                 1) % SB_ONDISK_JOURNAL_SIZE(p_s_sb)));
3953         commit = (struct reiserfs_journal_commit *)c_bh->b_data;
3954         memset(c_bh->b_data, 0, c_bh->b_size);
3955         set_commit_trans_id(commit, journal->j_trans_id);
3956         set_buffer_uptodate(c_bh);
3957
3958         /* init this journal list */
3959         jl = journal->j_current_jl;
3960
3961         /* we lock the commit before doing anything because
3962          * we want to make sure nobody tries to run flush_commit_list until
3963          * the new transaction is fully setup, and we've already flushed the
3964          * ordered bh list
3965          */
3966         down(&jl->j_commit_lock);
3967
3968         /* save the transaction id in case we need to commit it later */
3969         commit_trans_id = jl->j_trans_id;
3970
3971         atomic_set(&jl->j_older_commits_done, 0);
3972         jl->j_trans_id = journal->j_trans_id;
3973         jl->j_timestamp = journal->j_trans_start_time;
3974         jl->j_commit_bh = c_bh;
3975         jl->j_start = journal->j_start;
3976         jl->j_len = journal->j_len;
3977         atomic_set(&jl->j_nonzerolen, journal->j_len);
3978         atomic_set(&jl->j_commit_left, journal->j_len + 2);
3979         jl->j_realblock = NULL;
3980
3981         /* The ENTIRE FOR LOOP MUST not cause schedule to occur.
3982          **  for each real block, add it to the journal list hash,
3983          ** copy into real block index array in the commit or desc block
3984          */
3985         trans_half = journal_trans_half(p_s_sb->s_blocksize);
3986         for (i = 0, cn = journal->j_first; cn; cn = cn->next, i++) {
3987                 if (buffer_journaled(cn->bh)) {
3988                         jl_cn = get_cnode(p_s_sb);
3989                         if (!jl_cn) {
3990                                 reiserfs_panic(p_s_sb,
3991                                                "journal-1676, get_cnode returned NULL\n");
3992                         }
3993                         if (i == 0) {
3994                                 jl->j_realblock = jl_cn;
3995                         }
3996                         jl_cn->prev = last_cn;
3997                         jl_cn->next = NULL;
3998                         if (last_cn) {
3999                                 last_cn->next = jl_cn;
4000                         }
4001                         last_cn = jl_cn;
4002                         /* make sure the block we are trying to log is not a block
4003                            of journal or reserved area */
4004
4005                         if (is_block_in_log_or_reserved_area
4006                             (p_s_sb, cn->bh->b_blocknr)) {
4007                                 reiserfs_panic(p_s_sb,
4008                                                "journal-2332: Trying to log block %lu, which is a log block\n",
4009                                                cn->bh->b_blocknr);
4010                         }
4011                         jl_cn->blocknr = cn->bh->b_blocknr;
4012                         jl_cn->state = 0;
4013                         jl_cn->sb = p_s_sb;
4014                         jl_cn->bh = cn->bh;
4015                         jl_cn->jlist = jl;
4016                         insert_journal_hash(journal->j_list_hash_table, jl_cn);
4017                         if (i < trans_half) {
4018                                 desc->j_realblock[i] =
4019                                     cpu_to_le32(cn->bh->b_blocknr);
4020                         } else {
4021                                 commit->j_realblock[i - trans_half] =
4022                                     cpu_to_le32(cn->bh->b_blocknr);
4023                         }
4024                 } else {
4025                         i--;
4026                 }
4027         }
4028         set_desc_trans_len(desc, journal->j_len);
4029         set_desc_mount_id(desc, journal->j_mount_id);
4030         set_desc_trans_id(desc, journal->j_trans_id);
4031         set_commit_trans_len(commit, journal->j_len);
4032
4033         /* special check in case all buffers in the journal were marked for not logging */
4034         if (journal->j_len == 0) {
4035                 BUG();
4036         }
4037
4038         /* we're about to dirty all the log blocks, mark the description block
4039          * dirty now too.  Don't mark the commit block dirty until all the
4040          * others are on disk
4041          */
4042         mark_buffer_dirty(d_bh);
4043
4044         /* first data block is j_start + 1, so add one to cur_write_start wherever you use it */
4045         cur_write_start = journal->j_start;
4046         cn = journal->j_first;
4047         jindex = 1;             /* start at one so we don't get the desc again */
4048         while (cn) {
4049                 clear_buffer_journal_new(cn->bh);
4050                 /* copy all the real blocks into log area.  dirty log blocks */
4051                 if (buffer_journaled(cn->bh)) {
4052                         struct buffer_head *tmp_bh;
4053                         char *addr;
4054                         struct page *page;
4055                         tmp_bh =
4056                             journal_getblk(p_s_sb,
4057                                            SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
4058                                            ((cur_write_start +
4059                                              jindex) %
4060                                             SB_ONDISK_JOURNAL_SIZE(p_s_sb)));
4061                         set_buffer_uptodate(tmp_bh);
4062                         page = cn->bh->b_page;
4063                         addr = kmap(page);
4064                         memcpy(tmp_bh->b_data,
4065                                addr + offset_in_page(cn->bh->b_data),
4066                                cn->bh->b_size);
4067                         kunmap(page);
4068                         mark_buffer_dirty(tmp_bh);
4069                         jindex++;
4070                         set_buffer_journal_dirty(cn->bh);
4071                         clear_buffer_journaled(cn->bh);
4072                 } else {
4073                         /* JDirty cleared sometime during transaction.  don't log this one */
4074                         reiserfs_warning(p_s_sb,
4075                                          "journal-2048: do_journal_end: BAD, buffer in journal hash, but not JDirty!");
4076                         brelse(cn->bh);
4077                 }
4078                 next = cn->next;
4079                 free_cnode(p_s_sb, cn);
4080                 cn = next;
4081                 cond_resched();
4082         }
4083
4084         /* we are done  with both the c_bh and d_bh, but
4085          ** c_bh must be written after all other commit blocks,
4086          ** so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1.
4087          */
4088
4089         journal->j_current_jl = alloc_journal_list(p_s_sb);
4090
4091         /* now it is safe to insert this transaction on the main list */
4092         list_add_tail(&jl->j_list, &journal->j_journal_list);
4093         list_add_tail(&jl->j_working_list, &journal->j_working_list);
4094         journal->j_num_work_lists++;
4095
4096         /* reset journal values for the next transaction */
4097         old_start = journal->j_start;
4098         journal->j_start =
4099             (journal->j_start + journal->j_len +
4100              2) % SB_ONDISK_JOURNAL_SIZE(p_s_sb);
4101         atomic_set(&(journal->j_wcount), 0);
4102         journal->j_bcount = 0;
4103         journal->j_last = NULL;
4104         journal->j_first = NULL;
4105         journal->j_len = 0;
4106         journal->j_trans_start_time = 0;
4107         journal->j_trans_id++;
4108         journal->j_current_jl->j_trans_id = journal->j_trans_id;
4109         journal->j_must_wait = 0;
4110         journal->j_len_alloc = 0;
4111         journal->j_next_full_flush = 0;
4112         journal->j_next_async_flush = 0;
4113         init_journal_hash(p_s_sb);
4114
4115         // make sure reiserfs_add_jh sees the new current_jl before we
4116         // write out the tails
4117         smp_mb();
4118
4119         /* tail conversion targets have to hit the disk before we end the
4120          * transaction.  Otherwise a later transaction might repack the tail
4121          * before this transaction commits, leaving the data block unflushed and
4122          * clean, if we crash before the later transaction commits, the data block
4123          * is lost.
4124          */
4125         if (!list_empty(&jl->j_tail_bh_list)) {
4126                 unlock_kernel();
4127                 write_ordered_buffers(&journal->j_dirty_buffers_lock,
4128                                       journal, jl, &jl->j_tail_bh_list);
4129                 lock_kernel();
4130         }
4131         if (!list_empty(&jl->j_tail_bh_list))
4132                 BUG();
4133         up(&jl->j_commit_lock);
4134
4135         /* honor the flush wishes from the caller, simple commits can
4136          ** be done outside the journal lock, they are done below
4137          **
4138          ** if we don't flush the commit list right now, we put it into
4139          ** the work queue so the people waiting on the async progress work
4140          ** queue don't wait for this proc to flush journal lists and such.
4141          */
4142         if (flush) {
4143                 flush_commit_list(p_s_sb, jl, 1);
4144                 flush_journal_list(p_s_sb, jl, 1);
4145         } else if (!(jl->j_state & LIST_COMMIT_PENDING))
4146                 queue_delayed_work(commit_wq, &journal->j_work, HZ / 10);
4147
4148         /* if the next transaction has any chance of wrapping, flush
4149          ** transactions that might get overwritten.  If any journal lists are very
4150          ** old flush them as well.
4151          */
4152       first_jl:
4153         list_for_each_safe(entry, safe, &journal->j_journal_list) {
4154                 temp_jl = JOURNAL_LIST_ENTRY(entry);
4155                 if (journal->j_start <= temp_jl->j_start) {
4156                         if ((journal->j_start + journal->j_trans_max + 1) >=
4157                             temp_jl->j_start) {
4158                                 flush_used_journal_lists(p_s_sb, temp_jl);
4159                                 goto first_jl;
4160                         } else if ((journal->j_start +
4161                                     journal->j_trans_max + 1) <
4162                                    SB_ONDISK_JOURNAL_SIZE(p_s_sb)) {
4163                                 /* if we don't cross into the next transaction and we don't
4164                                  * wrap, there is no way we can overlap any later transactions
4165                                  * break now
4166                                  */
4167                                 break;
4168                         }
4169                 } else if ((journal->j_start +
4170                             journal->j_trans_max + 1) >
4171                            SB_ONDISK_JOURNAL_SIZE(p_s_sb)) {
4172                         if (((journal->j_start + journal->j_trans_max + 1) %
4173                              SB_ONDISK_JOURNAL_SIZE(p_s_sb)) >=
4174                             temp_jl->j_start) {
4175                                 flush_used_journal_lists(p_s_sb, temp_jl);
4176                                 goto first_jl;
4177                         } else {
4178                                 /* we don't overlap anything from out start to the end of the
4179                                  * log, and our wrapped portion doesn't overlap anything at
4180                                  * the start of the log.  We can break
4181                                  */
4182                                 break;
4183                         }
4184                 }
4185         }
4186         flush_old_journal_lists(p_s_sb);
4187
4188         journal->j_current_jl->j_list_bitmap =
4189             get_list_bitmap(p_s_sb, journal->j_current_jl);
4190
4191         if (!(journal->j_current_jl->j_list_bitmap)) {
4192                 reiserfs_panic(p_s_sb,
4193                                "journal-1996: do_journal_end, could not get a list bitmap\n");
4194         }
4195
4196         atomic_set(&(journal->j_jlock), 0);
4197         unlock_journal(p_s_sb);
4198         /* wake up any body waiting to join. */
4199         clear_bit(J_WRITERS_QUEUED, &journal->j_state);
4200         wake_up(&(journal->j_join_wait));
4201
4202         if (!flush && wait_on_commit &&
4203             journal_list_still_alive(p_s_sb, commit_trans_id)) {
4204                 flush_commit_list(p_s_sb, jl, 1);
4205         }
4206       out:
4207         reiserfs_check_lock_depth(p_s_sb, "journal end2");
4208
4209         memset(th, 0, sizeof(*th));
4210         /* Re-set th->t_super, so we can properly keep track of how many
4211          * persistent transactions there are. We need to do this so if this
4212          * call is part of a failed restart_transaction, we can free it later */
4213         th->t_super = p_s_sb;
4214
4215         return journal->j_errno;
4216 }
4217
4218 static void __reiserfs_journal_abort_hard(struct super_block *sb)
4219 {
4220         struct reiserfs_journal *journal = SB_JOURNAL(sb);
4221         if (test_bit(J_ABORTED, &journal->j_state))
4222                 return;
4223
4224         printk(KERN_CRIT "REISERFS: Aborting journal for filesystem on %s\n",
4225                reiserfs_bdevname(sb));
4226
4227         sb->s_flags |= MS_RDONLY;
4228         set_bit(J_ABORTED, &journal->j_state);
4229
4230 #ifdef CONFIG_REISERFS_CHECK
4231         dump_stack();
4232 #endif
4233 }
4234
4235 static void __reiserfs_journal_abort_soft(struct super_block *sb, int errno)
4236 {
4237         struct reiserfs_journal *journal = SB_JOURNAL(sb);
4238         if (test_bit(J_ABORTED, &journal->j_state))
4239                 return;
4240
4241         if (!journal->j_errno)
4242                 journal->j_errno = errno;
4243
4244         __reiserfs_journal_abort_hard(sb);
4245 }
4246
4247 void reiserfs_journal_abort(struct super_block *sb, int errno)
4248 {
4249         return __reiserfs_journal_abort_soft(sb, errno);
4250 }