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

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