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/time.h>
  38 #include <linux/semaphore.h>
  39 #include <linux/vmalloc.h>
  40 #include <linux/reiserfs_fs.h>
  41 #include <linux/kernel.h>
  42 #include <linux/errno.h>
  43 #include <linux/fcntl.h>
  44 #include <linux/stat.h>
  45 #include <linux/string.h>
  46 #include <linux/smp_lock.h>
  47 #include <linux/buffer_head.h>
  48 #include <linux/workqueue.h>
  49 #include <linux/writeback.h>
  50 #include <linux/blkdev.h>
  51 #include <linux/backing-dev.h>
  52 #include <linux/uaccess.h>
  53
  54 #include <asm/system.h>
  55
  56 /* gets a struct reiserfs_journal_list * from a list head */
  57 #define JOURNAL_LIST_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
  58                                j_list))
  59 #define JOURNAL_WORK_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
  60                                j_working_list))
  61
  62 /* the number of mounted filesystems.  This is used to decide when to
  63 ** start and kill the commit workqueue
  64 */
  65 static int reiserfs_mounted_fs_count;
  66
  67 static struct workqueue_struct *commit_wq;
  68
  69 #define JOURNAL_TRANS_HALF 1018 /* must be correct to keep the desc and commit
  70                                    structs at 4k */
  71 #define BUFNR 64                /*read ahead */
  72
  73 /* cnode stat bits.  Move these into reiserfs_fs.h */
  74
  75 #define BLOCK_FREED 2           /* this block was freed, and can't be written.  */
  76 #define BLOCK_FREED_HOLDER 3    /* this block was freed during this transaction, and can't be written */
  77
  78 #define BLOCK_NEEDS_FLUSH 4     /* used in flush_journal_list */
  79 #define BLOCK_DIRTIED 5
  80
  81 /* journal list state bits */
  82 #define LIST_TOUCHED 1
  83 #define LIST_DIRTY   2
  84 #define LIST_COMMIT_PENDING  4  /* someone will commit this list */
  85
  86 /* flags for do_journal_end */
  87 #define FLUSH_ALL   1           /* flush commit and real blocks */
  88 #define COMMIT_NOW  2           /* end and commit this transaction */
  89 #define WAIT        4           /* wait for the log blocks to hit the disk */
  90
  91 static int do_journal_end(struct reiserfs_transaction_handle *,
  92                           struct super_block *, unsigned long nblocks,
  93                           int flags);
  94 static int flush_journal_list(struct super_block *s,
  95                               struct reiserfs_journal_list *jl, int flushall);
  96 static int flush_commit_list(struct super_block *s,
  97                              struct reiserfs_journal_list *jl, int flushall);
  98 static int can_dirty(struct reiserfs_journal_cnode *cn);
  99 static int journal_join(struct reiserfs_transaction_handle *th,
 100                         struct super_block *p_s_sb, unsigned long nblocks);
 101 static int release_journal_dev(struct super_block *super,
 102                                struct reiserfs_journal *journal);
 103 static int dirty_one_transaction(struct super_block *s,
 104                                  struct reiserfs_journal_list *jl);
 105 static void flush_async_commits(struct work_struct *work);
 106 static void queue_log_writer(struct super_block *s);
 107
 108 /* values for join in do_journal_begin_r */
 109 enum {
 110         JBEGIN_REG = 0,         /* regular journal begin */
 111         JBEGIN_JOIN = 1,        /* join the running transaction if at all possible */
 112         JBEGIN_ABORT = 2,       /* called from cleanup code, ignores aborted flag */
 113 };
 114
 115 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
 116                               struct super_block *p_s_sb,
 117                               unsigned long nblocks, int join);
 118
 119 static void init_journal_hash(struct super_block *p_s_sb)
 120 {
 121         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 122         memset(journal->j_hash_table, 0,
 123                JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
 124 }
 125
 126 /*
 127 ** clears BH_Dirty and sticks the buffer on the clean list.  Called because I can't allow refile_buffer to
 128 ** make schedule happen after I've freed a block.  Look at remove_from_transaction and journal_mark_freed for
 129 ** more details.
 130 */
 131 static int reiserfs_clean_and_file_buffer(struct buffer_head *bh)
 132 {
 133         if (bh) {
 134                 clear_buffer_dirty(bh);
 135                 clear_buffer_journal_test(bh);
 136         }
 137         return 0;
 138 }
 139
 140 static void disable_barrier(struct super_block *s)
 141 {
 142         REISERFS_SB(s)->s_mount_opt &= ~(1 << REISERFS_BARRIER_FLUSH);
 143         printk("reiserfs: disabling flush barriers on %s\n",
 144                reiserfs_bdevname(s));
 145 }
 146
 147 static struct reiserfs_bitmap_node *allocate_bitmap_node(struct super_block
 148                                                          *p_s_sb)
 149 {
 150         struct reiserfs_bitmap_node *bn;
 151         static int id;
 152
 153         bn = kmalloc(sizeof(struct reiserfs_bitmap_node), GFP_NOFS);
 154         if (!bn) {
 155                 return NULL;
 156         }
 157         bn->data = kzalloc(p_s_sb->s_blocksize, GFP_NOFS);
 158         if (!bn->data) {
 159                 kfree(bn);
 160                 return NULL;
 161         }
 162         bn->id = id++;
 163         INIT_LIST_HEAD(&bn->list);
 164         return bn;
 165 }
 166
 167 static struct reiserfs_bitmap_node *get_bitmap_node(struct super_block *p_s_sb)
 168 {
 169         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 170         struct reiserfs_bitmap_node *bn = NULL;
 171         struct list_head *entry = journal->j_bitmap_nodes.next;
 172
 173         journal->j_used_bitmap_nodes++;
 174       repeat:
 175
 176         if (entry != &journal->j_bitmap_nodes) {
 177                 bn = list_entry(entry, struct reiserfs_bitmap_node, list);
 178                 list_del(entry);
 179                 memset(bn->data, 0, p_s_sb->s_blocksize);
 180                 journal->j_free_bitmap_nodes--;
 181                 return bn;
 182         }
 183         bn = allocate_bitmap_node(p_s_sb);
 184         if (!bn) {
 185                 yield();
 186                 goto repeat;
 187         }
 188         return bn;
 189 }
 190 static inline void free_bitmap_node(struct super_block *p_s_sb,
 191                                     struct reiserfs_bitmap_node *bn)
 192 {
 193         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 194         journal->j_used_bitmap_nodes--;
 195         if (journal->j_free_bitmap_nodes > REISERFS_MAX_BITMAP_NODES) {
 196                 kfree(bn->data);
 197                 kfree(bn);
 198         } else {
 199                 list_add(&bn->list, &journal->j_bitmap_nodes);
 200                 journal->j_free_bitmap_nodes++;
 201         }
 202 }
 203
 204 static void allocate_bitmap_nodes(struct super_block *p_s_sb)
 205 {
 206         int i;
 207         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 208         struct reiserfs_bitmap_node *bn = NULL;
 209         for (i = 0; i < REISERFS_MIN_BITMAP_NODES; i++) {
 210                 bn = allocate_bitmap_node(p_s_sb);
 211                 if (bn) {
 212                         list_add(&bn->list, &journal->j_bitmap_nodes);
 213                         journal->j_free_bitmap_nodes++;
 214                 } else {
 215                         break;  // this is ok, we'll try again when more are needed
 216                 }
 217         }
 218 }
 219
 220 static int set_bit_in_list_bitmap(struct super_block *p_s_sb,
 221                                   b_blocknr_t block,
 222                                   struct reiserfs_list_bitmap *jb)
 223 {
 224         unsigned int bmap_nr = block / (p_s_sb->s_blocksize << 3);
 225         unsigned int bit_nr = block % (p_s_sb->s_blocksize << 3);
 226
 227         if (!jb->bitmaps[bmap_nr]) {
 228                 jb->bitmaps[bmap_nr] = get_bitmap_node(p_s_sb);
 229         }
 230         set_bit(bit_nr, (unsigned long *)jb->bitmaps[bmap_nr]->data);
 231         return 0;
 232 }
 233
 234 static void cleanup_bitmap_list(struct super_block *p_s_sb,
 235                                 struct reiserfs_list_bitmap *jb)
 236 {
 237         int i;
 238         if (jb->bitmaps == NULL)
 239                 return;
 240
 241         for (i = 0; i < reiserfs_bmap_count(p_s_sb); i++) {
 242                 if (jb->bitmaps[i]) {
 243                         free_bitmap_node(p_s_sb, jb->bitmaps[i]);
 244                         jb->bitmaps[i] = NULL;
 245                 }
 246         }
 247 }
 248
 249 /*
 250 ** only call this on FS unmount.
 251 */
 252 static int free_list_bitmaps(struct super_block *p_s_sb,
 253                              struct reiserfs_list_bitmap *jb_array)
 254 {
 255         int i;
 256         struct reiserfs_list_bitmap *jb;
 257         for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
 258                 jb = jb_array + i;
 259                 jb->journal_list = NULL;
 260                 cleanup_bitmap_list(p_s_sb, jb);
 261                 vfree(jb->bitmaps);
 262                 jb->bitmaps = NULL;
 263         }
 264         return 0;
 265 }
 266
 267 static int free_bitmap_nodes(struct super_block *p_s_sb)
 268 {
 269         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 270         struct list_head *next = journal->j_bitmap_nodes.next;
 271         struct reiserfs_bitmap_node *bn;
 272
 273         while (next != &journal->j_bitmap_nodes) {
 274                 bn = list_entry(next, struct reiserfs_bitmap_node, list);
 275                 list_del(next);
 276                 kfree(bn->data);
 277                 kfree(bn);
 278                 next = journal->j_bitmap_nodes.next;
 279                 journal->j_free_bitmap_nodes--;
 280         }
 281
 282         return 0;
 283 }
 284
 285 /*
 286 ** get memory for JOURNAL_NUM_BITMAPS worth of bitmaps.
 287 ** jb_array is the array to be filled in.
 288 */
 289 int reiserfs_allocate_list_bitmaps(struct super_block *p_s_sb,
 290                                    struct reiserfs_list_bitmap *jb_array,
 291                                    unsigned int bmap_nr)
 292 {
 293         int i;
 294         int failed = 0;
 295         struct reiserfs_list_bitmap *jb;
 296         int mem = bmap_nr * sizeof(struct reiserfs_bitmap_node *);
 297
 298         for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
 299                 jb = jb_array + i;
 300                 jb->journal_list = NULL;
 301                 jb->bitmaps = vmalloc(mem);
 302                 if (!jb->bitmaps) {
 303                         reiserfs_warning(p_s_sb,
 304                                          "clm-2000, unable to allocate bitmaps for journal lists");
 305                         failed = 1;
 306                         break;
 307                 }
 308                 memset(jb->bitmaps, 0, mem);
 309         }
 310         if (failed) {
 311                 free_list_bitmaps(p_s_sb, jb_array);
 312                 return -1;
 313         }
 314         return 0;
 315 }
 316
 317 /*
 318 ** find an available list bitmap.  If you can't find one, flush a commit list
 319 ** and try again
 320 */
 321 static struct reiserfs_list_bitmap *get_list_bitmap(struct super_block *p_s_sb,
 322                                                     struct reiserfs_journal_list
 323                                                     *jl)
 324 {
 325         int i, j;
 326         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 327         struct reiserfs_list_bitmap *jb = NULL;
 328
 329         for (j = 0; j < (JOURNAL_NUM_BITMAPS * 3); j++) {
 330                 i = journal->j_list_bitmap_index;
 331                 journal->j_list_bitmap_index = (i + 1) % JOURNAL_NUM_BITMAPS;
 332                 jb = journal->j_list_bitmap + i;
 333                 if (journal->j_list_bitmap[i].journal_list) {
 334                         flush_commit_list(p_s_sb,
 335                                           journal->j_list_bitmap[i].
 336                                           journal_list, 1);
 337                         if (!journal->j_list_bitmap[i].journal_list) {
 338                                 break;
 339                         }
 340                 } else {
 341                         break;
 342                 }
 343         }
 344         if (jb->journal_list) { /* double check to make sure if flushed correctly */
 345                 return NULL;
 346         }
 347         jb->journal_list = jl;
 348         return jb;
 349 }
 350
 351 /*
 352 ** allocates a new chunk of X nodes, and links them all together as a list.
 353 ** Uses the cnode->next and cnode->prev pointers
 354 ** returns NULL on failure
 355 */
 356 static struct reiserfs_journal_cnode *allocate_cnodes(int num_cnodes)
 357 {
 358         struct reiserfs_journal_cnode *head;
 359         int i;
 360         if (num_cnodes <= 0) {
 361                 return NULL;
 362         }
 363         head = vmalloc(num_cnodes * sizeof(struct reiserfs_journal_cnode));
 364         if (!head) {
 365                 return NULL;
 366         }
 367         memset(head, 0, num_cnodes * sizeof(struct reiserfs_journal_cnode));
 368         head[0].prev = NULL;
 369         head[0].next = head + 1;
 370         for (i = 1; i < num_cnodes; i++) {
 371                 head[i].prev = head + (i - 1);
 372                 head[i].next = head + (i + 1);  /* if last one, overwrite it after the if */
 373         }
 374         head[num_cnodes - 1].next = NULL;
 375         return head;
 376 }
 377
 378 /*
 379 ** pulls a cnode off the free list, or returns NULL on failure
 380 */
 381 static struct reiserfs_journal_cnode *get_cnode(struct super_block *p_s_sb)
 382 {
 383         struct reiserfs_journal_cnode *cn;
 384         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 385
 386         reiserfs_check_lock_depth(p_s_sb, "get_cnode");
 387
 388         if (journal->j_cnode_free <= 0) {
 389                 return NULL;
 390         }
 391         journal->j_cnode_used++;
 392         journal->j_cnode_free--;
 393         cn = journal->j_cnode_free_list;
 394         if (!cn) {
 395                 return cn;
 396         }
 397         if (cn->next) {
 398                 cn->next->prev = NULL;
 399         }
 400         journal->j_cnode_free_list = cn->next;
 401         memset(cn, 0, sizeof(struct reiserfs_journal_cnode));
 402         return cn;
 403 }
 404
 405 /*
 406 ** returns a cnode to the free list
 407 */
 408 static void free_cnode(struct super_block *p_s_sb,
 409                        struct reiserfs_journal_cnode *cn)
 410 {
 411         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 412
 413         reiserfs_check_lock_depth(p_s_sb, "free_cnode");
 414
 415         journal->j_cnode_used--;
 416         journal->j_cnode_free++;
 417         /* memset(cn, 0, sizeof(struct reiserfs_journal_cnode)) ; */
 418         cn->next = journal->j_cnode_free_list;
 419         if (journal->j_cnode_free_list) {
 420                 journal->j_cnode_free_list->prev = cn;
 421         }
 422         cn->prev = NULL;        /* not needed with the memset, but I might kill the memset, and forget to do this */
 423         journal->j_cnode_free_list = cn;
 424 }
 425
 426 static void clear_prepared_bits(struct buffer_head *bh)
 427 {
 428         clear_buffer_journal_prepared(bh);
 429         clear_buffer_journal_restore_dirty(bh);
 430 }
 431
 432 /* utility function to force a BUG if it is called without the big
 433 ** kernel lock held.  caller is the string printed just before calling BUG()
 434 */
 435 void reiserfs_check_lock_depth(struct super_block *sb, char *caller)
 436 {
 437 #ifdef CONFIG_SMP
 438         if (current->lock_depth < 0) {
 439                 reiserfs_panic(sb, "%s called without kernel lock held",
 440                                caller);
 441         }
 442 #else
 443         ;
 444 #endif
 445 }
 446
 447 /* return a cnode with same dev, block number and size in table, or null if not found */
 448 static inline struct reiserfs_journal_cnode *get_journal_hash_dev(struct
 449                                                                   super_block
 450                                                                   *sb,
 451                                                                   struct
 452                                                                   reiserfs_journal_cnode
 453                                                                   **table,
 454                                                                   long bl)
 455 {
 456         struct reiserfs_journal_cnode *cn;
 457         cn = journal_hash(table, sb, bl);
 458         while (cn) {
 459                 if (cn->blocknr == bl && cn->sb == sb)
 460                         return cn;
 461                 cn = cn->hnext;
 462         }
 463         return (struct reiserfs_journal_cnode *)0;
 464 }
 465
 466 /*
 467 ** this actually means 'can this block be reallocated yet?'.  If you set search_all, a block can only be allocated
 468 ** if it is not in the current transaction, was not freed by the current transaction, and has no chance of ever
 469 ** being overwritten by a replay after crashing.
 470 **
 471 ** If you don't set search_all, a block can only be allocated if it is not in the current transaction.  Since deleting
 472 ** a block removes it from the current transaction, this case should never happen.  If you don't set search_all, make
 473 ** sure you never write the block without logging it.
 474 **
 475 ** next_zero_bit is a suggestion about the next block to try for find_forward.
 476 ** when bl is rejected because it is set in a journal list bitmap, we search
 477 ** for the next zero bit in the bitmap that rejected bl.  Then, we return that
 478 ** through next_zero_bit for find_forward to try.
 479 **
 480 ** Just because we return something in next_zero_bit does not mean we won't
 481 ** reject it on the next call to reiserfs_in_journal
 482 **
 483 */
 484 int reiserfs_in_journal(struct super_block *p_s_sb,
 485                         unsigned int bmap_nr, int bit_nr, int search_all,
 486                         b_blocknr_t * next_zero_bit)
 487 {
 488         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 489         struct reiserfs_journal_cnode *cn;
 490         struct reiserfs_list_bitmap *jb;
 491         int i;
 492         unsigned long bl;
 493
 494         *next_zero_bit = 0;     /* always start this at zero. */
 495
 496         PROC_INFO_INC(p_s_sb, journal.in_journal);
 497         /* If we aren't doing a search_all, this is a metablock, and it will be logged before use.
 498          ** if we crash before the transaction that freed it commits,  this transaction won't
 499          ** have committed either, and the block will never be written
 500          */
 501         if (search_all) {
 502                 for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
 503                         PROC_INFO_INC(p_s_sb, journal.in_journal_bitmap);
 504                         jb = journal->j_list_bitmap + i;
 505                         if (jb->journal_list && jb->bitmaps[bmap_nr] &&
 506                             test_bit(bit_nr,
 507                                      (unsigned long *)jb->bitmaps[bmap_nr]->
 508                                      data)) {
 509                                 *next_zero_bit =
 510                                     find_next_zero_bit((unsigned long *)
 511                                                        (jb->bitmaps[bmap_nr]->
 512                                                         data),
 513                                                        p_s_sb->s_blocksize << 3,
 514                                                        bit_nr + 1);
 515                                 return 1;
 516                         }
 517                 }
 518         }
 519
 520         bl = bmap_nr * (p_s_sb->s_blocksize << 3) + bit_nr;
 521         /* is it in any old transactions? */
 522         if (search_all
 523             && (cn =
 524                 get_journal_hash_dev(p_s_sb, journal->j_list_hash_table, bl))) {
 525                 return 1;
 526         }
 527
 528         /* is it in the current transaction.  This should never happen */
 529         if ((cn = get_journal_hash_dev(p_s_sb, journal->j_hash_table, bl))) {
 530                 BUG();
 531                 return 1;
 532         }
 533
 534         PROC_INFO_INC(p_s_sb, journal.in_journal_reusable);
 535         /* safe for reuse */
 536         return 0;
 537 }
 538
 539 /* insert cn into table
 540 */
 541 static inline void insert_journal_hash(struct reiserfs_journal_cnode **table,
 542                                        struct reiserfs_journal_cnode *cn)
 543 {
 544         struct reiserfs_journal_cnode *cn_orig;
 545
 546         cn_orig = journal_hash(table, cn->sb, cn->blocknr);
 547         cn->hnext = cn_orig;
 548         cn->hprev = NULL;
 549         if (cn_orig) {
 550                 cn_orig->hprev = cn;
 551         }
 552         journal_hash(table, cn->sb, cn->blocknr) = cn;
 553 }
 554
 555 /* lock the current transaction */
 556 static inline void lock_journal(struct super_block *p_s_sb)
 557 {
 558         PROC_INFO_INC(p_s_sb, journal.lock_journal);
 559         mutex_lock(&SB_JOURNAL(p_s_sb)->j_mutex);
 560 }
 561
 562 /* unlock the current transaction */
 563 static inline void unlock_journal(struct super_block *p_s_sb)
 564 {
 565         mutex_unlock(&SB_JOURNAL(p_s_sb)->j_mutex);
 566 }
 567
 568 static inline void get_journal_list(struct reiserfs_journal_list *jl)
 569 {
 570         jl->j_refcount++;
 571 }
 572
 573 static inline void put_journal_list(struct super_block *s,
 574                                     struct reiserfs_journal_list *jl)
 575 {
 576         if (jl->j_refcount < 1) {
 577                 reiserfs_panic(s, "trans id %u, refcount at %d",
 578                                jl->j_trans_id, jl->j_refcount);
 579         }
 580         if (--jl->j_refcount == 0)
 581                 kfree(jl);
 582 }
 583
 584 /*
 585 ** this used to be much more involved, and I'm keeping it just in case things get ugly again.
 586 ** it gets called by flush_commit_list, and cleans up any data stored about blocks freed during a
 587 ** transaction.
 588 */
 589 static void cleanup_freed_for_journal_list(struct super_block *p_s_sb,
 590                                            struct reiserfs_journal_list *jl)
 591 {
 592
 593         struct reiserfs_list_bitmap *jb = jl->j_list_bitmap;
 594         if (jb) {
 595                 cleanup_bitmap_list(p_s_sb, jb);
 596         }
 597         jl->j_list_bitmap->journal_list = NULL;
 598         jl->j_list_bitmap = NULL;
 599 }
 600
 601 static int journal_list_still_alive(struct super_block *s,
 602                                     unsigned int trans_id)
 603 {
 604         struct reiserfs_journal *journal = SB_JOURNAL(s);
 605         struct list_head *entry = &journal->j_journal_list;
 606         struct reiserfs_journal_list *jl;
 607
 608         if (!list_empty(entry)) {
 609                 jl = JOURNAL_LIST_ENTRY(entry->next);
 610                 if (jl->j_trans_id <= trans_id) {
 611                         return 1;
 612                 }
 613         }
 614         return 0;
 615 }
 616
 617 /*
 618  * If page->mapping was null, we failed to truncate this page for
 619  * some reason.  Most likely because it was truncated after being
 620  * logged via data=journal.
 621  *
 622  * This does a check to see if the buffer belongs to one of these
 623  * lost pages before doing the final put_bh.  If page->mapping was
 624  * null, it tries to free buffers on the page, which should make the
 625  * final page_cache_release drop the page from the lru.
 626  */
 627 static void release_buffer_page(struct buffer_head *bh)
 628 {
 629         struct page *page = bh->b_page;
 630         if (!page->mapping && trylock_page(page)) {
 631                 page_cache_get(page);
 632                 put_bh(bh);
 633                 if (!page->mapping)
 634                         try_to_free_buffers(page);
 635                 unlock_page(page);
 636                 page_cache_release(page);
 637         } else {
 638                 put_bh(bh);
 639         }
 640 }
 641
 642 static void reiserfs_end_buffer_io_sync(struct buffer_head *bh, int uptodate)
 643 {
 644         char b[BDEVNAME_SIZE];
 645
 646         if (buffer_journaled(bh)) {
 647                 reiserfs_warning(NULL,
 648                                  "clm-2084: pinned buffer %lu:%s sent to disk",
 649                                  bh->b_blocknr, bdevname(bh->b_bdev, b));
 650         }
 651         if (uptodate)
 652                 set_buffer_uptodate(bh);
 653         else
 654                 clear_buffer_uptodate(bh);
 655
 656         unlock_buffer(bh);
 657         release_buffer_page(bh);
 658 }
 659
 660 static void reiserfs_end_ordered_io(struct buffer_head *bh, int uptodate)
 661 {
 662         if (uptodate)
 663                 set_buffer_uptodate(bh);
 664         else
 665                 clear_buffer_uptodate(bh);
 666         unlock_buffer(bh);
 667         put_bh(bh);
 668 }
 669
 670 static void submit_logged_buffer(struct buffer_head *bh)
 671 {
 672         get_bh(bh);
 673         bh->b_end_io = reiserfs_end_buffer_io_sync;
 674         clear_buffer_journal_new(bh);
 675         clear_buffer_dirty(bh);
 676         if (!test_clear_buffer_journal_test(bh))
 677                 BUG();
 678         if (!buffer_uptodate(bh))
 679                 BUG();
 680         submit_bh(WRITE, bh);
 681 }
 682
 683 static void submit_ordered_buffer(struct buffer_head *bh)
 684 {
 685         get_bh(bh);
 686         bh->b_end_io = reiserfs_end_ordered_io;
 687         clear_buffer_dirty(bh);
 688         if (!buffer_uptodate(bh))
 689                 BUG();
 690         submit_bh(WRITE, bh);
 691 }
 692
 693 static int submit_barrier_buffer(struct buffer_head *bh)
 694 {
 695         get_bh(bh);
 696         bh->b_end_io = reiserfs_end_ordered_io;
 697         clear_buffer_dirty(bh);
 698         if (!buffer_uptodate(bh))
 699                 BUG();
 700         return submit_bh(WRITE_BARRIER, bh);
 701 }
 702
 703 static void check_barrier_completion(struct super_block *s,
 704                                      struct buffer_head *bh)
 705 {
 706         if (buffer_eopnotsupp(bh)) {
 707                 clear_buffer_eopnotsupp(bh);
 708                 disable_barrier(s);
 709                 set_buffer_uptodate(bh);
 710                 set_buffer_dirty(bh);
 711                 sync_dirty_buffer(bh);
 712         }
 713 }
 714
 715 #define CHUNK_SIZE 32
 716 struct buffer_chunk {
 717         struct buffer_head *bh[CHUNK_SIZE];
 718         int nr;
 719 };
 720
 721 static void write_chunk(struct buffer_chunk *chunk)
 722 {
 723         int i;
 724         get_fs_excl();
 725         for (i = 0; i < chunk->nr; i++) {
 726                 submit_logged_buffer(chunk->bh[i]);
 727         }
 728         chunk->nr = 0;
 729         put_fs_excl();
 730 }
 731
 732 static void write_ordered_chunk(struct buffer_chunk *chunk)
 733 {
 734         int i;
 735         get_fs_excl();
 736         for (i = 0; i < chunk->nr; i++) {
 737                 submit_ordered_buffer(chunk->bh[i]);
 738         }
 739         chunk->nr = 0;
 740         put_fs_excl();
 741 }
 742
 743 static int add_to_chunk(struct buffer_chunk *chunk, struct buffer_head *bh,
 744                         spinlock_t * lock, void (fn) (struct buffer_chunk *))
 745 {
 746         int ret = 0;
 747         BUG_ON(chunk->nr >= CHUNK_SIZE);
 748         chunk->bh[chunk->nr++] = bh;
 749         if (chunk->nr >= CHUNK_SIZE) {
 750                 ret = 1;
 751                 if (lock)
 752                         spin_unlock(lock);
 753                 fn(chunk);
 754                 if (lock)
 755                         spin_lock(lock);
 756         }
 757         return ret;
 758 }
 759
 760 static atomic_t nr_reiserfs_jh = ATOMIC_INIT(0);
 761 static struct reiserfs_jh *alloc_jh(void)
 762 {
 763         struct reiserfs_jh *jh;
 764         while (1) {
 765                 jh = kmalloc(sizeof(*jh), GFP_NOFS);
 766                 if (jh) {
 767                         atomic_inc(&nr_reiserfs_jh);
 768                         return jh;
 769                 }
 770                 yield();
 771         }
 772 }
 773
 774 /*
 775  * we want to free the jh when the buffer has been written
 776  * and waited on
 777  */
 778 void reiserfs_free_jh(struct buffer_head *bh)
 779 {
 780         struct reiserfs_jh *jh;
 781
 782         jh = bh->b_private;
 783         if (jh) {
 784                 bh->b_private = NULL;
 785                 jh->bh = NULL;
 786                 list_del_init(&jh->list);
 787                 kfree(jh);
 788                 if (atomic_read(&nr_reiserfs_jh) <= 0)
 789                         BUG();
 790                 atomic_dec(&nr_reiserfs_jh);
 791                 put_bh(bh);
 792         }
 793 }
 794
 795 static inline int __add_jh(struct reiserfs_journal *j, struct buffer_head *bh,
 796                            int tail)
 797 {
 798         struct reiserfs_jh *jh;
 799
 800         if (bh->b_private) {
 801                 spin_lock(&j->j_dirty_buffers_lock);
 802                 if (!bh->b_private) {
 803                         spin_unlock(&j->j_dirty_buffers_lock);
 804                         goto no_jh;
 805                 }
 806                 jh = bh->b_private;
 807                 list_del_init(&jh->list);
 808         } else {
 809               no_jh:
 810                 get_bh(bh);
 811                 jh = alloc_jh();
 812                 spin_lock(&j->j_dirty_buffers_lock);
 813                 /* buffer must be locked for __add_jh, should be able to have
 814                  * two adds at the same time
 815                  */
 816                 BUG_ON(bh->b_private);
 817                 jh->bh = bh;
 818                 bh->b_private = jh;
 819         }
 820         jh->jl = j->j_current_jl;
 821         if (tail)
 822                 list_add_tail(&jh->list, &jh->jl->j_tail_bh_list);
 823         else {
 824                 list_add_tail(&jh->list, &jh->jl->j_bh_list);
 825         }
 826         spin_unlock(&j->j_dirty_buffers_lock);
 827         return 0;
 828 }
 829
 830 int reiserfs_add_tail_list(struct inode *inode, struct buffer_head *bh)
 831 {
 832         return __add_jh(SB_JOURNAL(inode->i_sb), bh, 1);
 833 }
 834 int reiserfs_add_ordered_list(struct inode *inode, struct buffer_head *bh)
 835 {
 836         return __add_jh(SB_JOURNAL(inode->i_sb), bh, 0);
 837 }
 838
 839 #define JH_ENTRY(l) list_entry((l), struct reiserfs_jh, list)
 840 static int write_ordered_buffers(spinlock_t * lock,
 841                                  struct reiserfs_journal *j,
 842                                  struct reiserfs_journal_list *jl,
 843                                  struct list_head *list)
 844 {
 845         struct buffer_head *bh;
 846         struct reiserfs_jh *jh;
 847         int ret = j->j_errno;
 848         struct buffer_chunk chunk;
 849         struct list_head tmp;
 850         INIT_LIST_HEAD(&tmp);
 851
 852         chunk.nr = 0;
 853         spin_lock(lock);
 854         while (!list_empty(list)) {
 855                 jh = JH_ENTRY(list->next);
 856                 bh = jh->bh;
 857                 get_bh(bh);
 858                 if (!trylock_buffer(bh)) {
 859                         if (!buffer_dirty(bh)) {
 860                                 list_move(&jh->list, &tmp);
 861                                 goto loop_next;
 862                         }
 863                         spin_unlock(lock);
 864                         if (chunk.nr)
 865                                 write_ordered_chunk(&chunk);
 866                         wait_on_buffer(bh);
 867                         cond_resched();
 868                         spin_lock(lock);
 869                         goto loop_next;
 870                 }
 871                 /* in theory, dirty non-uptodate buffers should never get here,
 872                  * but the upper layer io error paths still have a few quirks.
 873                  * Handle them here as gracefully as we can
 874                  */
 875                 if (!buffer_uptodate(bh) && buffer_dirty(bh)) {
 876                         clear_buffer_dirty(bh);
 877                         ret = -EIO;
 878                 }
 879                 if (buffer_dirty(bh)) {
 880                         list_move(&jh->list, &tmp);
 881                         add_to_chunk(&chunk, bh, lock, write_ordered_chunk);
 882                 } else {
 883                         reiserfs_free_jh(bh);
 884                         unlock_buffer(bh);
 885                 }
 886               loop_next:
 887                 put_bh(bh);
 888                 cond_resched_lock(lock);
 889         }
 890         if (chunk.nr) {
 891                 spin_unlock(lock);
 892                 write_ordered_chunk(&chunk);
 893                 spin_lock(lock);
 894         }
 895         while (!list_empty(&tmp)) {
 896                 jh = JH_ENTRY(tmp.prev);
 897                 bh = jh->bh;
 898                 get_bh(bh);
 899                 reiserfs_free_jh(bh);
 900
 901                 if (buffer_locked(bh)) {
 902                         spin_unlock(lock);
 903                         wait_on_buffer(bh);
 904                         spin_lock(lock);
 905                 }
 906                 if (!buffer_uptodate(bh)) {
 907                         ret = -EIO;
 908                 }
 909                 /* ugly interaction with invalidatepage here.
 910                  * reiserfs_invalidate_page will pin any buffer that has a valid
 911                  * journal head from an older transaction.  If someone else sets
 912                  * our buffer dirty after we write it in the first loop, and
 913                  * then someone truncates the page away, nobody will ever write
 914                  * the buffer. We're safe if we write the page one last time
 915                  * after freeing the journal header.
 916                  */
 917                 if (buffer_dirty(bh) && unlikely(bh->b_page->mapping == NULL)) {
 918                         spin_unlock(lock);
 919                         ll_rw_block(WRITE, 1, &bh);
 920                         spin_lock(lock);
 921                 }
 922                 put_bh(bh);
 923                 cond_resched_lock(lock);
 924         }
 925         spin_unlock(lock);
 926         return ret;
 927 }
 928
 929 static int flush_older_commits(struct super_block *s,
 930                                struct reiserfs_journal_list *jl)
 931 {
 932         struct reiserfs_journal *journal = SB_JOURNAL(s);
 933         struct reiserfs_journal_list *other_jl;
 934         struct reiserfs_journal_list *first_jl;
 935         struct list_head *entry;
 936         unsigned int trans_id = jl->j_trans_id;
 937         unsigned int other_trans_id;
 938         unsigned int first_trans_id;
 939
 940       find_first:
 941         /*
 942          * first we walk backwards to find the oldest uncommitted transation
 943          */
 944         first_jl = jl;
 945         entry = jl->j_list.prev;
 946         while (1) {
 947                 other_jl = JOURNAL_LIST_ENTRY(entry);
 948                 if (entry == &journal->j_journal_list ||
 949                     atomic_read(&other_jl->j_older_commits_done))
 950                         break;
 951
 952                 first_jl = other_jl;
 953                 entry = other_jl->j_list.prev;
 954         }
 955
 956         /* if we didn't find any older uncommitted transactions, return now */
 957         if (first_jl == jl) {
 958                 return 0;
 959         }
 960
 961         first_trans_id = first_jl->j_trans_id;
 962
 963         entry = &first_jl->j_list;
 964         while (1) {
 965                 other_jl = JOURNAL_LIST_ENTRY(entry);
 966                 other_trans_id = other_jl->j_trans_id;
 967
 968                 if (other_trans_id < trans_id) {
 969                         if (atomic_read(&other_jl->j_commit_left) != 0) {
 970                                 flush_commit_list(s, other_jl, 0);
 971
 972                                 /* list we were called with is gone, return */
 973                                 if (!journal_list_still_alive(s, trans_id))
 974                                         return 1;
 975
 976                                 /* the one we just flushed is gone, this means all
 977                                  * older lists are also gone, so first_jl is no longer
 978                                  * valid either.  Go back to the beginning.
 979                                  */
 980                                 if (!journal_list_still_alive
 981                                     (s, other_trans_id)) {
 982                                         goto find_first;
 983                                 }
 984                         }
 985                         entry = entry->next;
 986                         if (entry == &journal->j_journal_list)
 987                                 return 0;
 988                 } else {
 989                         return 0;
 990                 }
 991         }
 992         return 0;
 993 }
 994
 995 static int reiserfs_async_progress_wait(struct super_block *s)
 996 {
 997         DEFINE_WAIT(wait);
 998         struct reiserfs_journal *j = SB_JOURNAL(s);
 999         if (atomic_read(&j->j_async_throttle))
1000                 congestion_wait(WRITE, HZ / 10);
1001         return 0;
1002 }
1003
1004 /*
1005 ** if this journal list still has commit blocks unflushed, send them to disk.
1006 **
1007 ** log areas must be flushed in order (transaction 2 can't commit before transaction 1)
1008 ** Before the commit block can by written, every other log block must be safely on disk
1009 **
1010 */
1011 static int flush_commit_list(struct super_block *s,
1012                              struct reiserfs_journal_list *jl, int flushall)
1013 {
1014         int i;
1015         b_blocknr_t bn;
1016         struct buffer_head *tbh = NULL;
1017         unsigned int trans_id = jl->j_trans_id;
1018         struct reiserfs_journal *journal = SB_JOURNAL(s);
1019         int barrier = 0;
1020         int retval = 0;
1021         int write_len;
1022
1023         reiserfs_check_lock_depth(s, "flush_commit_list");
1024
1025         if (atomic_read(&jl->j_older_commits_done)) {
1026                 return 0;
1027         }
1028
1029         get_fs_excl();
1030
1031         /* before we can put our commit blocks on disk, we have to make sure everyone older than
1032          ** us is on disk too
1033          */
1034         BUG_ON(jl->j_len <= 0);
1035         BUG_ON(trans_id == journal->j_trans_id);
1036
1037         get_journal_list(jl);
1038         if (flushall) {
1039                 if (flush_older_commits(s, jl) == 1) {
1040                         /* list disappeared during flush_older_commits.  return */
1041                         goto put_jl;
1042                 }
1043         }
1044
1045         /* make sure nobody is trying to flush this one at the same time */
1046         mutex_lock(&jl->j_commit_mutex);
1047         if (!journal_list_still_alive(s, trans_id)) {
1048                 mutex_unlock(&jl->j_commit_mutex);
1049                 goto put_jl;
1050         }
1051         BUG_ON(jl->j_trans_id == 0);
1052
1053         /* this commit is done, exit */
1054         if (atomic_read(&(jl->j_commit_left)) <= 0) {
1055                 if (flushall) {
1056                         atomic_set(&(jl->j_older_commits_done), 1);
1057                 }
1058                 mutex_unlock(&jl->j_commit_mutex);
1059                 goto put_jl;
1060         }
1061
1062         if (!list_empty(&jl->j_bh_list)) {
1063                 int ret;
1064                 unlock_kernel();
1065                 ret = write_ordered_buffers(&journal->j_dirty_buffers_lock,
1066                                             journal, jl, &jl->j_bh_list);
1067                 if (ret < 0 && retval == 0)
1068                         retval = ret;
1069                 lock_kernel();
1070         }
1071         BUG_ON(!list_empty(&jl->j_bh_list));
1072         /*
1073          * for the description block and all the log blocks, submit any buffers
1074          * that haven't already reached the disk.  Try to write at least 256
1075          * log blocks. later on, we will only wait on blocks that correspond
1076          * to this transaction, but while we're unplugging we might as well
1077          * get a chunk of data on there.
1078          */
1079         atomic_inc(&journal->j_async_throttle);
1080         write_len = jl->j_len + 1;
1081         if (write_len < 256)
1082                 write_len = 256;
1083         for (i = 0 ; i < write_len ; i++) {
1084                 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + (jl->j_start + i) %
1085                     SB_ONDISK_JOURNAL_SIZE(s);
1086                 tbh = journal_find_get_block(s, bn);
1087                 if (tbh) {
1088                         if (buffer_dirty(tbh))
1089                             ll_rw_block(WRITE, 1, &tbh) ;
1090                         put_bh(tbh) ;
1091                 }
1092         }
1093         atomic_dec(&journal->j_async_throttle);
1094
1095         /* We're skipping the commit if there's an error */
1096         if (retval || reiserfs_is_journal_aborted(journal))
1097                 barrier = 0;
1098
1099         /* wait on everything written so far before writing the commit
1100          * if we are in barrier mode, send the commit down now
1101          */
1102         barrier = reiserfs_barrier_flush(s);
1103         if (barrier) {
1104                 int ret;
1105                 lock_buffer(jl->j_commit_bh);
1106                 ret = submit_barrier_buffer(jl->j_commit_bh);
1107                 if (ret == -EOPNOTSUPP) {
1108                         set_buffer_uptodate(jl->j_commit_bh);
1109                         disable_barrier(s);
1110                         barrier = 0;
1111                 }
1112         }
1113         for (i = 0; i < (jl->j_len + 1); i++) {
1114                 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) +
1115                     (jl->j_start + i) % SB_ONDISK_JOURNAL_SIZE(s);
1116                 tbh = journal_find_get_block(s, bn);
1117                 wait_on_buffer(tbh);
1118                 // since we're using ll_rw_blk above, it might have skipped over
1119                 // a locked buffer.  Double check here
1120                 //
1121                 if (buffer_dirty(tbh))  /* redundant, sync_dirty_buffer() checks */
1122                         sync_dirty_buffer(tbh);
1123                 if (unlikely(!buffer_uptodate(tbh))) {
1124 #ifdef CONFIG_REISERFS_CHECK
1125                         reiserfs_warning(s, "journal-601, buffer write failed");
1126 #endif
1127                         retval = -EIO;
1128                 }
1129                 put_bh(tbh);    /* once for journal_find_get_block */
1130                 put_bh(tbh);    /* once due to original getblk in do_journal_end */
1131                 atomic_dec(&(jl->j_commit_left));
1132         }
1133
1134         BUG_ON(atomic_read(&(jl->j_commit_left)) != 1);
1135
1136         if (!barrier) {
1137                 /* If there was a write error in the journal - we can't commit
1138                  * this transaction - it will be invalid and, if successful,
1139                  * will just end up propagating the write error out to
1140                  * the file system. */
1141                 if (likely(!retval && !reiserfs_is_journal_aborted (journal))) {
1142                         if (buffer_dirty(jl->j_commit_bh))
1143                                 BUG();
1144                         mark_buffer_dirty(jl->j_commit_bh) ;
1145                         sync_dirty_buffer(jl->j_commit_bh) ;
1146                 }
1147         } else
1148                 wait_on_buffer(jl->j_commit_bh);
1149
1150         check_barrier_completion(s, jl->j_commit_bh);
1151
1152         /* If there was a write error in the journal - we can't commit this
1153          * transaction - it will be invalid and, if successful, will just end
1154          * up propagating the write error out to the filesystem. */
1155         if (unlikely(!buffer_uptodate(jl->j_commit_bh))) {
1156 #ifdef CONFIG_REISERFS_CHECK
1157                 reiserfs_warning(s, "journal-615: buffer write failed");
1158 #endif
1159                 retval = -EIO;
1160         }
1161         bforget(jl->j_commit_bh);
1162         if (journal->j_last_commit_id != 0 &&
1163             (jl->j_trans_id - journal->j_last_commit_id) != 1) {
1164                 reiserfs_warning(s, "clm-2200: last commit %lu, current %lu",
1165                                  journal->j_last_commit_id, jl->j_trans_id);
1166         }
1167         journal->j_last_commit_id = jl->j_trans_id;
1168
1169         /* now, every commit block is on the disk.  It is safe to allow blocks freed during this transaction to be reallocated */
1170         cleanup_freed_for_journal_list(s, jl);
1171
1172         retval = retval ? retval : journal->j_errno;
1173
1174         /* mark the metadata dirty */
1175         if (!retval)
1176                 dirty_one_transaction(s, jl);
1177         atomic_dec(&(jl->j_commit_left));
1178
1179         if (flushall) {
1180                 atomic_set(&(jl->j_older_commits_done), 1);
1181         }
1182         mutex_unlock(&jl->j_commit_mutex);
1183       put_jl:
1184         put_journal_list(s, jl);
1185
1186         if (retval)
1187                 reiserfs_abort(s, retval, "Journal write error in %s",
1188                                __func__);
1189         put_fs_excl();
1190         return retval;
1191 }
1192
1193 /*
1194 ** flush_journal_list frequently needs to find a newer transaction for a given block.  This does that, or
1195 ** returns NULL if it can't find anything
1196 */
1197 static struct reiserfs_journal_list *find_newer_jl_for_cn(struct
1198                                                           reiserfs_journal_cnode
1199                                                           *cn)
1200 {
1201         struct super_block *sb = cn->sb;
1202         b_blocknr_t blocknr = cn->blocknr;
1203
1204         cn = cn->hprev;
1205         while (cn) {
1206                 if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist) {
1207                         return cn->jlist;
1208                 }
1209                 cn = cn->hprev;
1210         }
1211         return NULL;
1212 }
1213
1214 static int newer_jl_done(struct reiserfs_journal_cnode *cn)
1215 {
1216         struct super_block *sb = cn->sb;
1217         b_blocknr_t blocknr = cn->blocknr;
1218
1219         cn = cn->hprev;
1220         while (cn) {
1221                 if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist &&
1222                     atomic_read(&cn->jlist->j_commit_left) != 0)
1223                                     return 0;
1224                 cn = cn->hprev;
1225         }
1226         return 1;
1227 }
1228
1229 static void remove_journal_hash(struct super_block *,
1230                                 struct reiserfs_journal_cnode **,
1231                                 struct reiserfs_journal_list *, unsigned long,
1232                                 int);
1233
1234 /*
1235 ** once all the real blocks have been flushed, it is safe to remove them from the
1236 ** journal list for this transaction.  Aside from freeing the cnode, this also allows the
1237 ** block to be reallocated for data blocks if it had been deleted.
1238 */
1239 static void remove_all_from_journal_list(struct super_block *p_s_sb,
1240                                          struct reiserfs_journal_list *jl,
1241                                          int debug)
1242 {
1243         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1244         struct reiserfs_journal_cnode *cn, *last;
1245         cn = jl->j_realblock;
1246
1247         /* which is better, to lock once around the whole loop, or
1248          ** to lock for each call to remove_journal_hash?
1249          */
1250         while (cn) {
1251                 if (cn->blocknr != 0) {
1252                         if (debug) {
1253                                 reiserfs_warning(p_s_sb,
1254                                                  "block %u, bh is %d, state %ld",
1255                                                  cn->blocknr, cn->bh ? 1 : 0,
1256                                                  cn->state);
1257                         }
1258                         cn->state = 0;
1259                         remove_journal_hash(p_s_sb, journal->j_list_hash_table,
1260                                             jl, cn->blocknr, 1);
1261                 }
1262                 last = cn;
1263                 cn = cn->next;
1264                 free_cnode(p_s_sb, last);
1265         }
1266         jl->j_realblock = NULL;
1267 }
1268
1269 /*
1270 ** if this timestamp is greater than the timestamp we wrote last to the header block, write it to the header block.
1271 ** once this is done, I can safely say the log area for this transaction won't ever be replayed, and I can start
1272 ** releasing blocks in this transaction for reuse as data blocks.
1273 ** called by flush_journal_list, before it calls remove_all_from_journal_list
1274 **
1275 */
1276 static int _update_journal_header_block(struct super_block *p_s_sb,
1277                                         unsigned long offset,
1278                                         unsigned int trans_id)
1279 {
1280         struct reiserfs_journal_header *jh;
1281         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1282
1283         if (reiserfs_is_journal_aborted(journal))
1284                 return -EIO;
1285
1286         if (trans_id >= journal->j_last_flush_trans_id) {
1287                 if (buffer_locked((journal->j_header_bh))) {
1288                         wait_on_buffer((journal->j_header_bh));
1289                         if (unlikely(!buffer_uptodate(journal->j_header_bh))) {
1290 #ifdef CONFIG_REISERFS_CHECK
1291                                 reiserfs_warning(p_s_sb,
1292                                                  "journal-699: buffer write failed");
1293 #endif
1294                                 return -EIO;
1295                         }
1296                 }
1297                 journal->j_last_flush_trans_id = trans_id;
1298                 journal->j_first_unflushed_offset = offset;
1299                 jh = (struct reiserfs_journal_header *)(journal->j_header_bh->
1300                                                         b_data);
1301                 jh->j_last_flush_trans_id = cpu_to_le32(trans_id);
1302                 jh->j_first_unflushed_offset = cpu_to_le32(offset);
1303                 jh->j_mount_id = cpu_to_le32(journal->j_mount_id);
1304
1305                 if (reiserfs_barrier_flush(p_s_sb)) {
1306                         int ret;
1307                         lock_buffer(journal->j_header_bh);
1308                         ret = submit_barrier_buffer(journal->j_header_bh);
1309                         if (ret == -EOPNOTSUPP) {
1310                                 set_buffer_uptodate(journal->j_header_bh);
1311                                 disable_barrier(p_s_sb);
1312                                 goto sync;
1313                         }
1314                         wait_on_buffer(journal->j_header_bh);
1315                         check_barrier_completion(p_s_sb, journal->j_header_bh);
1316                 } else {
1317                       sync:
1318                         set_buffer_dirty(journal->j_header_bh);
1319                         sync_dirty_buffer(journal->j_header_bh);
1320                 }
1321                 if (!buffer_uptodate(journal->j_header_bh)) {
1322                         reiserfs_warning(p_s_sb,
1323                                          "journal-837: IO error during journal replay");
1324                         return -EIO;
1325                 }
1326         }
1327         return 0;
1328 }
1329
1330 static int update_journal_header_block(struct super_block *p_s_sb,
1331                                        unsigned long offset,
1332                                        unsigned int trans_id)
1333 {
1334         return _update_journal_header_block(p_s_sb, offset, trans_id);
1335 }
1336
1337 /*
1338 ** flush any and all journal lists older than you are
1339 ** can only be called from flush_journal_list
1340 */
1341 static int flush_older_journal_lists(struct super_block *p_s_sb,
1342                                      struct reiserfs_journal_list *jl)
1343 {
1344         struct list_head *entry;
1345         struct reiserfs_journal_list *other_jl;
1346         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1347         unsigned int trans_id = jl->j_trans_id;
1348
1349         /* we know we are the only ones flushing things, no extra race
1350          * protection is required.
1351          */
1352       restart:
1353         entry = journal->j_journal_list.next;
1354         /* Did we wrap? */
1355         if (entry == &journal->j_journal_list)
1356                 return 0;
1357         other_jl = JOURNAL_LIST_ENTRY(entry);
1358         if (other_jl->j_trans_id < trans_id) {
1359                 BUG_ON(other_jl->j_refcount <= 0);
1360                 /* do not flush all */
1361                 flush_journal_list(p_s_sb, other_jl, 0);
1362
1363                 /* other_jl is now deleted from the list */
1364                 goto restart;
1365         }
1366         return 0;
1367 }
1368
1369 static void del_from_work_list(struct super_block *s,
1370                                struct reiserfs_journal_list *jl)
1371 {
1372         struct reiserfs_journal *journal = SB_JOURNAL(s);
1373         if (!list_empty(&jl->j_working_list)) {
1374                 list_del_init(&jl->j_working_list);
1375                 journal->j_num_work_lists--;
1376         }
1377 }
1378
1379 /* flush a journal list, both commit and real blocks
1380 **
1381 ** always set flushall to 1, unless you are calling from inside
1382 ** flush_journal_list
1383 **
1384 ** IMPORTANT.  This can only be called while there are no journal writers,
1385 ** and the journal is locked.  That means it can only be called from
1386 ** do_journal_end, or by journal_release
1387 */
1388 static int flush_journal_list(struct super_block *s,
1389                               struct reiserfs_journal_list *jl, int flushall)
1390 {
1391         struct reiserfs_journal_list *pjl;
1392         struct reiserfs_journal_cnode *cn, *last;
1393         int count;
1394         int was_jwait = 0;
1395         int was_dirty = 0;
1396         struct buffer_head *saved_bh;
1397         unsigned long j_len_saved = jl->j_len;
1398         struct reiserfs_journal *journal = SB_JOURNAL(s);
1399         int err = 0;
1400
1401         BUG_ON(j_len_saved <= 0);
1402
1403         if (atomic_read(&journal->j_wcount) != 0) {
1404                 reiserfs_warning(s,
1405                                  "clm-2048: flush_journal_list called with wcount %d",
1406                                  atomic_read(&journal->j_wcount));
1407         }
1408         BUG_ON(jl->j_trans_id == 0);
1409
1410         /* if flushall == 0, the lock is already held */
1411         if (flushall) {
1412                 mutex_lock(&journal->j_flush_mutex);
1413         } else if (mutex_trylock(&journal->j_flush_mutex)) {
1414                 BUG();
1415         }
1416
1417         count = 0;
1418         if (j_len_saved > journal->j_trans_max) {
1419                 reiserfs_panic(s,
1420                                "journal-715: flush_journal_list, length is %lu, trans id %lu\n",
1421                                j_len_saved, jl->j_trans_id);
1422                 return 0;
1423         }
1424
1425         get_fs_excl();
1426
1427         /* if all the work is already done, get out of here */
1428         if (atomic_read(&(jl->j_nonzerolen)) <= 0 &&
1429             atomic_read(&(jl->j_commit_left)) <= 0) {
1430                 goto flush_older_and_return;
1431         }
1432
1433         /* start by putting the commit list on disk.  This will also flush
1434          ** the commit lists of any olders transactions
1435          */
1436         flush_commit_list(s, jl, 1);
1437
1438         if (!(jl->j_state & LIST_DIRTY)
1439             && !reiserfs_is_journal_aborted(journal))
1440                 BUG();
1441
1442         /* are we done now? */
1443         if (atomic_read(&(jl->j_nonzerolen)) <= 0 &&
1444             atomic_read(&(jl->j_commit_left)) <= 0) {
1445                 goto flush_older_and_return;
1446         }
1447
1448         /* loop through each cnode, see if we need to write it,
1449          ** or wait on a more recent transaction, or just ignore it
1450          */
1451         if (atomic_read(&(journal->j_wcount)) != 0) {
1452                 reiserfs_panic(s,
1453                                "journal-844: panic journal list is flushing, wcount is not 0\n");
1454         }
1455         cn = jl->j_realblock;
1456         while (cn) {
1457                 was_jwait = 0;
1458                 was_dirty = 0;
1459                 saved_bh = NULL;
1460                 /* blocknr of 0 is no longer in the hash, ignore it */
1461                 if (cn->blocknr == 0) {
1462                         goto free_cnode;
1463                 }
1464
1465                 /* This transaction failed commit. Don't write out to the disk */
1466                 if (!(jl->j_state & LIST_DIRTY))
1467                         goto free_cnode;
1468
1469                 pjl = find_newer_jl_for_cn(cn);
1470                 /* the order is important here.  We check pjl to make sure we
1471                  ** don't clear BH_JDirty_wait if we aren't the one writing this
1472                  ** block to disk
1473                  */
1474                 if (!pjl && cn->bh) {
1475                         saved_bh = cn->bh;
1476
1477                         /* we do this to make sure nobody releases the buffer while
1478                          ** we are working with it
1479                          */
1480                         get_bh(saved_bh);
1481
1482                         if (buffer_journal_dirty(saved_bh)) {
1483                                 BUG_ON(!can_dirty(cn));
1484                                 was_jwait = 1;
1485                                 was_dirty = 1;
1486                         } else if (can_dirty(cn)) {
1487                                 /* everything with !pjl && jwait should be writable */
1488                                 BUG();
1489                         }
1490                 }
1491
1492                 /* if someone has this block in a newer transaction, just make
1493                  ** sure they are committed, and don't try writing it to disk
1494                  */
1495                 if (pjl) {
1496                         if (atomic_read(&pjl->j_commit_left))
1497                                 flush_commit_list(s, pjl, 1);
1498                         goto free_cnode;
1499                 }
1500
1501                 /* bh == NULL when the block got to disk on its own, OR,
1502                  ** the block got freed in a future transaction
1503                  */
1504                 if (saved_bh == NULL) {
1505                         goto free_cnode;
1506                 }
1507
1508                 /* this should never happen.  kupdate_one_transaction has this list
1509                  ** locked while it works, so we should never see a buffer here that
1510                  ** is not marked JDirty_wait
1511                  */
1512                 if ((!was_jwait) && !buffer_locked(saved_bh)) {
1513                         reiserfs_warning(s,
1514                                          "journal-813: BAD! buffer %llu %cdirty %cjwait, "
1515                                          "not in a newer tranasction",
1516                                          (unsigned long long)saved_bh->
1517                                          b_blocknr, was_dirty ? ' ' : '!',
1518                                          was_jwait ? ' ' : '!');
1519                 }
1520                 if (was_dirty) {
1521                         /* we inc again because saved_bh gets decremented at free_cnode */
1522                         get_bh(saved_bh);
1523                         set_bit(BLOCK_NEEDS_FLUSH, &cn->state);
1524                         lock_buffer(saved_bh);
1525                         BUG_ON(cn->blocknr != saved_bh->b_blocknr);
1526                         if (buffer_dirty(saved_bh))
1527                                 submit_logged_buffer(saved_bh);
1528                         else
1529                                 unlock_buffer(saved_bh);
1530                         count++;
1531                 } else {
1532                         reiserfs_warning(s,
1533                                          "clm-2082: Unable to flush buffer %llu in %s",
1534                                          (unsigned long long)saved_bh->
1535                                          b_blocknr, __func__);
1536                 }
1537               free_cnode:
1538                 last = cn;
1539                 cn = cn->next;
1540                 if (saved_bh) {
1541                         /* we incremented this to keep others from taking the buffer head away */
1542                         put_bh(saved_bh);
1543                         if (atomic_read(&(saved_bh->b_count)) < 0) {
1544                                 reiserfs_warning(s,
1545                                                  "journal-945: saved_bh->b_count < 0");
1546                         }
1547                 }
1548         }
1549         if (count > 0) {
1550                 cn = jl->j_realblock;
1551                 while (cn) {
1552                         if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) {
1553                                 if (!cn->bh) {
1554                                         reiserfs_panic(s,
1555                                                        "journal-1011: cn->bh is NULL\n");
1556                                 }
1557                                 wait_on_buffer(cn->bh);
1558                                 if (!cn->bh) {
1559                                         reiserfs_panic(s,
1560                                                        "journal-1012: cn->bh is NULL\n");
1561                                 }
1562                                 if (unlikely(!buffer_uptodate(cn->bh))) {
1563 #ifdef CONFIG_REISERFS_CHECK
1564                                         reiserfs_warning(s,
1565                                                          "journal-949: buffer write failed\n");
1566 #endif
1567                                         err = -EIO;
1568                                 }
1569                                 /* note, we must clear the JDirty_wait bit after the up to date
1570                                  ** check, otherwise we race against our flushpage routine
1571                                  */
1572                                 BUG_ON(!test_clear_buffer_journal_dirty
1573                                        (cn->bh));
1574
1575                                 /* drop one ref for us */
1576                                 put_bh(cn->bh);
1577                                 /* drop one ref for journal_mark_dirty */
1578                                 release_buffer_page(cn->bh);
1579                         }
1580                         cn = cn->next;
1581                 }
1582         }
1583
1584         if (err)
1585                 reiserfs_abort(s, -EIO,
1586                                "Write error while pushing transaction to disk in %s",
1587                                __func__);
1588       flush_older_and_return:
1589
1590         /* before we can update the journal header block, we _must_ flush all
1591          ** real blocks from all older transactions to disk.  This is because
1592          ** once the header block is updated, this transaction will not be
1593          ** replayed after a crash
1594          */
1595         if (flushall) {
1596                 flush_older_journal_lists(s, jl);
1597         }
1598
1599         err = journal->j_errno;
1600         /* before we can remove everything from the hash tables for this
1601          ** transaction, we must make sure it can never be replayed
1602          **
1603          ** since we are only called from do_journal_end, we know for sure there
1604          ** are no allocations going on while we are flushing journal lists.  So,
1605          ** we only need to update the journal header block for the last list
1606          ** being flushed
1607          */
1608         if (!err && flushall) {
1609                 err =
1610                     update_journal_header_block(s,
1611                                                 (jl->j_start + jl->j_len +
1612                                                  2) % SB_ONDISK_JOURNAL_SIZE(s),
1613                                                 jl->j_trans_id);
1614                 if (err)
1615                         reiserfs_abort(s, -EIO,
1616                                        "Write error while updating journal header in %s",
1617                                        __func__);
1618         }
1619         remove_all_from_journal_list(s, jl, 0);
1620         list_del_init(&jl->j_list);
1621         journal->j_num_lists--;
1622         del_from_work_list(s, jl);
1623
1624         if (journal->j_last_flush_id != 0 &&
1625             (jl->j_trans_id - journal->j_last_flush_id) != 1) {
1626                 reiserfs_warning(s, "clm-2201: last flush %lu, current %lu",
1627                                  journal->j_last_flush_id, jl->j_trans_id);
1628         }
1629         journal->j_last_flush_id = jl->j_trans_id;
1630
1631         /* not strictly required since we are freeing the list, but it should
1632          * help find code using dead lists later on
1633          */
1634         jl->j_len = 0;
1635         atomic_set(&(jl->j_nonzerolen), 0);
1636         jl->j_start = 0;
1637         jl->j_realblock = NULL;
1638         jl->j_commit_bh = NULL;
1639         jl->j_trans_id = 0;
1640         jl->j_state = 0;
1641         put_journal_list(s, jl);
1642         if (flushall)
1643                 mutex_unlock(&journal->j_flush_mutex);
1644         put_fs_excl();
1645         return err;
1646 }
1647
1648 static int test_transaction(struct super_block *s,
1649                             struct reiserfs_journal_list *jl)
1650 {
1651         struct reiserfs_journal_cnode *cn;
1652
1653         if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0)
1654                 return 1;
1655
1656         cn = jl->j_realblock;
1657         while (cn) {
1658                 /* if the blocknr == 0, this has been cleared from the hash,
1659                  ** skip it
1660                  */
1661                 if (cn->blocknr == 0) {
1662                         goto next;
1663                 }
1664                 if (cn->bh && !newer_jl_done(cn))
1665                         return 0;
1666               next:
1667                 cn = cn->next;
1668                 cond_resched();
1669         }
1670         return 0;
1671 }
1672
1673 static int write_one_transaction(struct super_block *s,
1674                                  struct reiserfs_journal_list *jl,
1675                                  struct buffer_chunk *chunk)
1676 {
1677         struct reiserfs_journal_cnode *cn;
1678         int ret = 0;
1679
1680         jl->j_state |= LIST_TOUCHED;
1681         del_from_work_list(s, jl);
1682         if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0) {
1683                 return 0;
1684         }
1685
1686         cn = jl->j_realblock;
1687         while (cn) {
1688                 /* if the blocknr == 0, this has been cleared from the hash,
1689                  ** skip it
1690                  */
1691                 if (cn->blocknr == 0) {
1692                         goto next;
1693                 }
1694                 if (cn->bh && can_dirty(cn) && buffer_dirty(cn->bh)) {
1695                         struct buffer_head *tmp_bh;
1696                         /* we can race against journal_mark_freed when we try
1697                          * to lock_buffer(cn->bh), so we have to inc the buffer
1698                          * count, and recheck things after locking
1699                          */
1700                         tmp_bh = cn->bh;
1701                         get_bh(tmp_bh);
1702                         lock_buffer(tmp_bh);
1703                         if (cn->bh && can_dirty(cn) && buffer_dirty(tmp_bh)) {
1704                                 if (!buffer_journal_dirty(tmp_bh) ||
1705                                     buffer_journal_prepared(tmp_bh))
1706                                         BUG();
1707                                 add_to_chunk(chunk, tmp_bh, NULL, write_chunk);
1708                                 ret++;
1709                         } else {
1710                                 /* note, cn->bh might be null now */
1711                                 unlock_buffer(tmp_bh);
1712                         }
1713                         put_bh(tmp_bh);
1714                 }
1715               next:
1716                 cn = cn->next;
1717                 cond_resched();
1718         }
1719         return ret;
1720 }
1721
1722 /* used by flush_commit_list */
1723 static int dirty_one_transaction(struct super_block *s,
1724                                  struct reiserfs_journal_list *jl)
1725 {
1726         struct reiserfs_journal_cnode *cn;
1727         struct reiserfs_journal_list *pjl;
1728         int ret = 0;
1729
1730         jl->j_state |= LIST_DIRTY;
1731         cn = jl->j_realblock;
1732         while (cn) {
1733                 /* look for a more recent transaction that logged this
1734                  ** buffer.  Only the most recent transaction with a buffer in
1735                  ** it is allowed to send that buffer to disk
1736                  */
1737                 pjl = find_newer_jl_for_cn(cn);
1738                 if (!pjl && cn->blocknr && cn->bh
1739                     && buffer_journal_dirty(cn->bh)) {
1740                         BUG_ON(!can_dirty(cn));
1741                         /* if the buffer is prepared, it will either be logged
1742                          * or restored.  If restored, we need to make sure
1743                          * it actually gets marked dirty
1744                          */
1745                         clear_buffer_journal_new(cn->bh);
1746                         if (buffer_journal_prepared(cn->bh)) {
1747                                 set_buffer_journal_restore_dirty(cn->bh);
1748                         } else {
1749                                 set_buffer_journal_test(cn->bh);
1750                                 mark_buffer_dirty(cn->bh);
1751                         }
1752                 }
1753                 cn = cn->next;
1754         }
1755         return ret;
1756 }
1757
1758 static int kupdate_transactions(struct super_block *s,
1759                                 struct reiserfs_journal_list *jl,
1760                                 struct reiserfs_journal_list **next_jl,
1761                                 unsigned int *next_trans_id,
1762                                 int num_blocks, int num_trans)
1763 {
1764         int ret = 0;
1765         int written = 0;
1766         int transactions_flushed = 0;
1767         unsigned int orig_trans_id = jl->j_trans_id;
1768         struct buffer_chunk chunk;
1769         struct list_head *entry;
1770         struct reiserfs_journal *journal = SB_JOURNAL(s);
1771         chunk.nr = 0;
1772
1773         mutex_lock(&journal->j_flush_mutex);
1774         if (!journal_list_still_alive(s, orig_trans_id)) {
1775                 goto done;
1776         }
1777
1778         /* we've got j_flush_mutex held, nobody is going to delete any
1779          * of these lists out from underneath us
1780          */
1781         while ((num_trans && transactions_flushed < num_trans) ||
1782                (!num_trans && written < num_blocks)) {
1783
1784                 if (jl->j_len == 0 || (jl->j_state & LIST_TOUCHED) ||
1785                     atomic_read(&jl->j_commit_left)
1786                     || !(jl->j_state & LIST_DIRTY)) {
1787                         del_from_work_list(s, jl);
1788                         break;
1789                 }
1790                 ret = write_one_transaction(s, jl, &chunk);
1791
1792                 if (ret < 0)
1793                         goto done;
1794                 transactions_flushed++;
1795                 written += ret;
1796                 entry = jl->j_list.next;
1797
1798                 /* did we wrap? */
1799                 if (entry == &journal->j_journal_list) {
1800                         break;
1801                 }
1802                 jl = JOURNAL_LIST_ENTRY(entry);
1803
1804                 /* don't bother with older transactions */
1805                 if (jl->j_trans_id <= orig_trans_id)
1806                         break;
1807         }
1808         if (chunk.nr) {
1809                 write_chunk(&chunk);
1810         }
1811
1812       done:
1813         mutex_unlock(&journal->j_flush_mutex);
1814         return ret;
1815 }
1816
1817 /* for o_sync and fsync heavy applications, they tend to use
1818 ** all the journa list slots with tiny transactions.  These
1819 ** trigger lots and lots of calls to update the header block, which
1820 ** adds seeks and slows things down.
1821 **
1822 ** This function tries to clear out a large chunk of the journal lists
1823 ** at once, which makes everything faster since only the newest journal
1824 ** list updates the header block
1825 */
1826 static int flush_used_journal_lists(struct super_block *s,
1827                                     struct reiserfs_journal_list *jl)
1828 {
1829         unsigned long len = 0;
1830         unsigned long cur_len;
1831         int ret;
1832         int i;
1833         int limit = 256;
1834         struct reiserfs_journal_list *tjl;
1835         struct reiserfs_journal_list *flush_jl;
1836         unsigned int trans_id;
1837         struct reiserfs_journal *journal = SB_JOURNAL(s);
1838
1839         flush_jl = tjl = jl;
1840
1841         /* in data logging mode, try harder to flush a lot of blocks */
1842         if (reiserfs_data_log(s))
1843                 limit = 1024;
1844         /* flush for 256 transactions or limit blocks, whichever comes first */
1845         for (i = 0; i < 256 && len < limit; i++) {
1846                 if (atomic_read(&tjl->j_commit_left) ||
1847                     tjl->j_trans_id < jl->j_trans_id) {
1848                         break;
1849                 }
1850                 cur_len = atomic_read(&tjl->j_nonzerolen);
1851                 if (cur_len > 0) {
1852                         tjl->j_state &= ~LIST_TOUCHED;
1853                 }
1854                 len += cur_len;
1855                 flush_jl = tjl;
1856                 if (tjl->j_list.next == &journal->j_journal_list)
1857                         break;
1858                 tjl = JOURNAL_LIST_ENTRY(tjl->j_list.next);
1859         }
1860         /* try to find a group of blocks we can flush across all the
1861          ** transactions, but only bother if we've actually spanned
1862          ** across multiple lists
1863          */
1864         if (flush_jl != jl) {
1865                 ret = kupdate_transactions(s, jl, &tjl, &trans_id, len, i);
1866         }
1867         flush_journal_list(s, flush_jl, 1);
1868         return 0;
1869 }
1870
1871 /*
1872 ** removes any nodes in table with name block and dev as bh.
1873 ** only touchs the hnext and hprev pointers.
1874 */
1875 void remove_journal_hash(struct super_block *sb,
1876                          struct reiserfs_journal_cnode **table,
1877                          struct reiserfs_journal_list *jl,
1878                          unsigned long block, int remove_freed)
1879 {
1880         struct reiserfs_journal_cnode *cur;
1881         struct reiserfs_journal_cnode **head;
1882
1883         head = &(journal_hash(table, sb, block));
1884         if (!head) {
1885                 return;
1886         }
1887         cur = *head;
1888         while (cur) {
1889                 if (cur->blocknr == block && cur->sb == sb
1890                     && (jl == NULL || jl == cur->jlist)
1891                     && (!test_bit(BLOCK_FREED, &cur->state) || remove_freed)) {
1892                         if (cur->hnext) {
1893                                 cur->hnext->hprev = cur->hprev;
1894                         }
1895                         if (cur->hprev) {
1896                                 cur->hprev->hnext = cur->hnext;
1897                         } else {
1898                                 *head = cur->hnext;
1899                         }
1900                         cur->blocknr = 0;
1901                         cur->sb = NULL;
1902                         cur->state = 0;
1903                         if (cur->bh && cur->jlist)      /* anybody who clears the cur->bh will also dec the nonzerolen */
1904                                 atomic_dec(&(cur->jlist->j_nonzerolen));
1905                         cur->bh = NULL;
1906                         cur->jlist = NULL;
1907                 }
1908                 cur = cur->hnext;
1909         }
1910 }
1911
1912 static void free_journal_ram(struct super_block *p_s_sb)
1913 {
1914         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1915         kfree(journal->j_current_jl);
1916         journal->j_num_lists--;
1917
1918         vfree(journal->j_cnode_free_orig);
1919         free_list_bitmaps(p_s_sb, journal->j_list_bitmap);
1920         free_bitmap_nodes(p_s_sb);      /* must be after free_list_bitmaps */
1921         if (journal->j_header_bh) {
1922                 brelse(journal->j_header_bh);
1923         }
1924         /* j_header_bh is on the journal dev, make sure not to release the journal
1925          * dev until we brelse j_header_bh
1926          */
1927         release_journal_dev(p_s_sb, journal);
1928         vfree(journal);
1929 }
1930
1931 /*
1932 ** call on unmount.  Only set error to 1 if you haven't made your way out
1933 ** of read_super() yet.  Any other caller must keep error at 0.
1934 */
1935 static int do_journal_release(struct reiserfs_transaction_handle *th,
1936                               struct super_block *p_s_sb, int error)
1937 {
1938         struct reiserfs_transaction_handle myth;
1939         int flushed = 0;
1940         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1941
1942         /* we only want to flush out transactions if we were called with error == 0
1943          */
1944         if (!error && !(p_s_sb->s_flags & MS_RDONLY)) {
1945                 /* end the current trans */
1946                 BUG_ON(!th->t_trans_id);
1947                 do_journal_end(th, p_s_sb, 10, FLUSH_ALL);
1948
1949                 /* make sure something gets logged to force our way into the flush code */
1950                 if (!journal_join(&myth, p_s_sb, 1)) {
1951                         reiserfs_prepare_for_journal(p_s_sb,
1952                                                      SB_BUFFER_WITH_SB(p_s_sb),
1953                                                      1);
1954                         journal_mark_dirty(&myth, p_s_sb,
1955                                            SB_BUFFER_WITH_SB(p_s_sb));
1956                         do_journal_end(&myth, p_s_sb, 1, FLUSH_ALL);
1957                         flushed = 1;
1958                 }
1959         }
1960
1961         /* this also catches errors during the do_journal_end above */
1962         if (!error && reiserfs_is_journal_aborted(journal)) {
1963                 memset(&myth, 0, sizeof(myth));
1964                 if (!journal_join_abort(&myth, p_s_sb, 1)) {
1965                         reiserfs_prepare_for_journal(p_s_sb,
1966                                                      SB_BUFFER_WITH_SB(p_s_sb),
1967                                                      1);
1968                         journal_mark_dirty(&myth, p_s_sb,
1969                                            SB_BUFFER_WITH_SB(p_s_sb));
1970                         do_journal_end(&myth, p_s_sb, 1, FLUSH_ALL);
1971                 }
1972         }
1973
1974         reiserfs_mounted_fs_count--;
1975         /* wait for all commits to finish */
1976         cancel_delayed_work(&SB_JOURNAL(p_s_sb)->j_work);
1977         flush_workqueue(commit_wq);
1978         if (!reiserfs_mounted_fs_count) {
1979                 destroy_workqueue(commit_wq);
1980                 commit_wq = NULL;
1981         }
1982
1983         free_journal_ram(p_s_sb);
1984
1985         return 0;
1986 }
1987
1988 /*
1989 ** call on unmount.  flush all journal trans, release all alloc'd ram
1990 */
1991 int journal_release(struct reiserfs_transaction_handle *th,
1992                     struct super_block *p_s_sb)
1993 {
1994         return do_journal_release(th, p_s_sb, 0);
1995 }
1996
1997 /*
1998 ** only call from an error condition inside reiserfs_read_super!
1999 */
2000 int journal_release_error(struct reiserfs_transaction_handle *th,
2001                           struct super_block *p_s_sb)
2002 {
2003         return do_journal_release(th, p_s_sb, 1);
2004 }
2005
2006 /* compares description block with commit block.  returns 1 if they differ, 0 if they are the same */
2007 static int journal_compare_desc_commit(struct super_block *p_s_sb,
2008                                        struct reiserfs_journal_desc *desc,
2009                                        struct reiserfs_journal_commit *commit)
2010 {
2011         if (get_commit_trans_id(commit) != get_desc_trans_id(desc) ||
2012             get_commit_trans_len(commit) != get_desc_trans_len(desc) ||
2013             get_commit_trans_len(commit) > SB_JOURNAL(p_s_sb)->j_trans_max ||
2014             get_commit_trans_len(commit) <= 0) {
2015                 return 1;
2016         }
2017         return 0;
2018 }
2019
2020 /* returns 0 if it did not find a description block
2021 ** returns -1 if it found a corrupt commit block
2022 ** returns 1 if both desc and commit were valid
2023 */
2024 static int journal_transaction_is_valid(struct super_block *p_s_sb,
2025                                         struct buffer_head *d_bh,
2026                                         unsigned int *oldest_invalid_trans_id,
2027                                         unsigned long *newest_mount_id)
2028 {
2029         struct reiserfs_journal_desc *desc;
2030         struct reiserfs_journal_commit *commit;
2031         struct buffer_head *c_bh;
2032         unsigned long offset;
2033
2034         if (!d_bh)
2035                 return 0;
2036
2037         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2038         if (get_desc_trans_len(desc) > 0
2039             && !memcmp(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8)) {
2040                 if (oldest_invalid_trans_id && *oldest_invalid_trans_id
2041                     && get_desc_trans_id(desc) > *oldest_invalid_trans_id) {
2042                         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2043                                        "journal-986: transaction "
2044                                        "is valid returning because trans_id %d is greater than "
2045                                        "oldest_invalid %lu",
2046                                        get_desc_trans_id(desc),
2047                                        *oldest_invalid_trans_id);
2048                         return 0;
2049                 }
2050                 if (newest_mount_id
2051                     && *newest_mount_id > get_desc_mount_id(desc)) {
2052                         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2053                                        "journal-1087: transaction "
2054                                        "is valid returning because mount_id %d is less than "
2055                                        "newest_mount_id %lu",
2056                                        get_desc_mount_id(desc),
2057                                        *newest_mount_id);
2058                         return -1;
2059                 }
2060                 if (get_desc_trans_len(desc) > SB_JOURNAL(p_s_sb)->j_trans_max) {
2061                         reiserfs_warning(p_s_sb,
2062                                          "journal-2018: Bad transaction length %d encountered, ignoring transaction",
2063                                          get_desc_trans_len(desc));
2064                         return -1;
2065                 }
2066                 offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb);
2067
2068                 /* ok, we have a journal description block, lets see if the transaction was valid */
2069                 c_bh =
2070                     journal_bread(p_s_sb,
2071                                   SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2072                                   ((offset + get_desc_trans_len(desc) +
2073                                     1) % SB_ONDISK_JOURNAL_SIZE(p_s_sb)));
2074                 if (!c_bh)
2075                         return 0;
2076                 commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2077                 if (journal_compare_desc_commit(p_s_sb, desc, commit)) {
2078                         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2079                                        "journal_transaction_is_valid, commit offset %ld had bad "
2080                                        "time %d or length %d",
2081                                        c_bh->b_blocknr -
2082                                        SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2083                                        get_commit_trans_id(commit),
2084                                        get_commit_trans_len(commit));
2085                         brelse(c_bh);
2086                         if (oldest_invalid_trans_id) {
2087                                 *oldest_invalid_trans_id =
2088                                     get_desc_trans_id(desc);
2089                                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2090                                                "journal-1004: "
2091                                                "transaction_is_valid setting oldest invalid trans_id "
2092                                                "to %d",
2093                                                get_desc_trans_id(desc));
2094                         }
2095                         return -1;
2096                 }
2097                 brelse(c_bh);
2098                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2099                                "journal-1006: found valid "
2100                                "transaction start offset %llu, len %d id %d",
2101                                d_bh->b_blocknr -
2102                                SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2103                                get_desc_trans_len(desc),
2104                                get_desc_trans_id(desc));
2105                 return 1;
2106         } else {
2107                 return 0;
2108         }
2109 }
2110
2111 static void brelse_array(struct buffer_head **heads, int num)
2112 {
2113         int i;
2114         for (i = 0; i < num; i++) {
2115                 brelse(heads[i]);
2116         }
2117 }
2118
2119 /*
2120 ** given the start, and values for the oldest acceptable transactions,
2121 ** this either reads in a replays a transaction, or returns because the transaction
2122 ** is invalid, or too old.
2123 */
2124 static int journal_read_transaction(struct super_block *p_s_sb,
2125                                     unsigned long cur_dblock,
2126                                     unsigned long oldest_start,
2127                                     unsigned int oldest_trans_id,
2128                                     unsigned long newest_mount_id)
2129 {
2130         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
2131         struct reiserfs_journal_desc *desc;
2132         struct reiserfs_journal_commit *commit;
2133         unsigned int trans_id = 0;
2134         struct buffer_head *c_bh;
2135         struct buffer_head *d_bh;
2136         struct buffer_head **log_blocks = NULL;
2137         struct buffer_head **real_blocks = NULL;
2138         unsigned int trans_offset;
2139         int i;
2140         int trans_half;
2141
2142         d_bh = journal_bread(p_s_sb, cur_dblock);
2143         if (!d_bh)
2144                 return 1;
2145         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2146         trans_offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb);
2147         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1037: "
2148                        "journal_read_transaction, offset %llu, len %d mount_id %d",
2149                        d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2150                        get_desc_trans_len(desc), get_desc_mount_id(desc));
2151         if (get_desc_trans_id(desc) < oldest_trans_id) {
2152                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1039: "
2153                                "journal_read_trans skipping because %lu is too old",
2154                                cur_dblock -
2155                                SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb));
2156                 brelse(d_bh);
2157                 return 1;
2158         }
2159         if (get_desc_mount_id(desc) != newest_mount_id) {
2160                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1146: "
2161                                "journal_read_trans skipping because %d is != "
2162                                "newest_mount_id %lu", get_desc_mount_id(desc),
2163                                newest_mount_id);
2164                 brelse(d_bh);
2165                 return 1;
2166         }
2167         c_bh = journal_bread(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2168                              ((trans_offset + get_desc_trans_len(desc) + 1) %
2169                               SB_ONDISK_JOURNAL_SIZE(p_s_sb)));
2170         if (!c_bh) {
2171                 brelse(d_bh);
2172                 return 1;
2173         }
2174         commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2175         if (journal_compare_desc_commit(p_s_sb, desc, commit)) {
2176                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2177                                "journal_read_transaction, "
2178                                "commit offset %llu had bad time %d or length %d",
2179                                c_bh->b_blocknr -
2180                                SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2181                                get_commit_trans_id(commit),
2182                                get_commit_trans_len(commit));
2183                 brelse(c_bh);
2184                 brelse(d_bh);
2185                 return 1;
2186         }
2187         trans_id = get_desc_trans_id(desc);
2188         /* now we know we've got a good transaction, and it was inside the valid time ranges */
2189         log_blocks = kmalloc(get_desc_trans_len(desc) *
2190                              sizeof(struct buffer_head *), GFP_NOFS);
2191         real_blocks = kmalloc(get_desc_trans_len(desc) *
2192                               sizeof(struct buffer_head *), GFP_NOFS);
2193         if (!log_blocks || !real_blocks) {
2194                 brelse(c_bh);
2195                 brelse(d_bh);
2196                 kfree(log_blocks);
2197                 kfree(real_blocks);
2198                 reiserfs_warning(p_s_sb,
2199                                  "journal-1169: kmalloc failed, unable to mount FS");
2200                 return -1;
2201         }
2202         /* get all the buffer heads */
2203         trans_half = journal_trans_half(p_s_sb->s_blocksize);
2204         for (i = 0; i < get_desc_trans_len(desc); i++) {
2205                 log_blocks[i] =
2206                     journal_getblk(p_s_sb,
2207                                    SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2208                                    (trans_offset + 1 +
2209                                     i) % SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2210                 if (i < trans_half) {
2211                         real_blocks[i] =
2212                             sb_getblk(p_s_sb,
2213                                       le32_to_cpu(desc->j_realblock[i]));
2214                 } else {
2215                         real_blocks[i] =
2216                             sb_getblk(p_s_sb,
2217                                       le32_to_cpu(commit->
2218                                                   j_realblock[i - trans_half]));
2219                 }
2220                 if (real_blocks[i]->b_blocknr > SB_BLOCK_COUNT(p_s_sb)) {
2221                         reiserfs_warning(p_s_sb,
2222                                          "journal-1207: REPLAY FAILURE fsck required! Block to replay is outside of filesystem");
2223                         goto abort_replay;
2224                 }
2225                 /* make sure we don't try to replay onto log or reserved area */
2226                 if (is_block_in_log_or_reserved_area
2227                     (p_s_sb, real_blocks[i]->b_blocknr)) {
2228                         reiserfs_warning(p_s_sb,
2229                                          "journal-1204: REPLAY FAILURE fsck required! Trying to replay onto a log block");
2230                       abort_replay:
2231                         brelse_array(log_blocks, i);
2232                         brelse_array(real_blocks, i);
2233                         brelse(c_bh);
2234                         brelse(d_bh);
2235                         kfree(log_blocks);
2236                         kfree(real_blocks);
2237                         return -1;
2238                 }
2239         }
2240         /* read in the log blocks, memcpy to the corresponding real block */
2241         ll_rw_block(READ, get_desc_trans_len(desc), log_blocks);
2242         for (i = 0; i < get_desc_trans_len(desc); i++) {
2243                 wait_on_buffer(log_blocks[i]);
2244                 if (!buffer_uptodate(log_blocks[i])) {
2245                         reiserfs_warning(p_s_sb,
2246                                          "journal-1212: REPLAY FAILURE fsck required! buffer write failed");
2247                         brelse_array(log_blocks + i,
2248                                      get_desc_trans_len(desc) - i);
2249                         brelse_array(real_blocks, get_desc_trans_len(desc));
2250                         brelse(c_bh);
2251                         brelse(d_bh);
2252                         kfree(log_blocks);
2253                         kfree(real_blocks);
2254                         return -1;
2255                 }
2256                 memcpy(real_blocks[i]->b_data, log_blocks[i]->b_data,
2257                        real_blocks[i]->b_size);
2258                 set_buffer_uptodate(real_blocks[i]);
2259                 brelse(log_blocks[i]);
2260         }
2261         /* flush out the real blocks */
2262         for (i = 0; i < get_desc_trans_len(desc); i++) {
2263                 set_buffer_dirty(real_blocks[i]);
2264                 ll_rw_block(SWRITE, 1, real_blocks + i);
2265         }
2266         for (i = 0; i < get_desc_trans_len(desc); i++) {
2267                 wait_on_buffer(real_blocks[i]);
2268                 if (!buffer_uptodate(real_blocks[i])) {
2269                         reiserfs_warning(p_s_sb,
2270                                          "journal-1226: REPLAY FAILURE, fsck required! buffer write failed");
2271                         brelse_array(real_blocks + i,
2272                                      get_desc_trans_len(desc) - i);
2273                         brelse(c_bh);
2274                         brelse(d_bh);
2275                         kfree(log_blocks);
2276                         kfree(real_blocks);
2277                         return -1;
2278                 }
2279                 brelse(real_blocks[i]);
2280         }
2281         cur_dblock =
2282             SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2283             ((trans_offset + get_desc_trans_len(desc) +
2284               2) % SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2285         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2286                        "journal-1095: setting journal " "start to offset %ld",
2287                        cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb));
2288
2289         /* init starting values for the first transaction, in case this is the last transaction to be replayed. */
2290         journal->j_start = cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb);
2291         journal->j_last_flush_trans_id = trans_id;
2292         journal->j_trans_id = trans_id + 1;
2293         /* check for trans_id overflow */
2294         if (journal->j_trans_id == 0)
2295                 journal->j_trans_id = 10;
2296         brelse(c_bh);
2297         brelse(d_bh);
2298         kfree(log_blocks);
2299         kfree(real_blocks);
2300         return 0;
2301 }
2302
2303 /* This function reads blocks starting from block and to max_block of bufsize
2304    size (but no more than BUFNR blocks at a time). This proved to improve
2305    mounting speed on self-rebuilding raid5 arrays at least.
2306    Right now it is only used from journal code. But later we might use it
2307    from other places.
2308    Note: Do not use journal_getblk/sb_getblk functions here! */
2309 static struct buffer_head *reiserfs_breada(struct block_device *dev,
2310                                            b_blocknr_t block, int bufsize,
2311                                            b_blocknr_t max_block)
2312 {
2313         struct buffer_head *bhlist[BUFNR];
2314         unsigned int blocks = BUFNR;
2315         struct buffer_head *bh;
2316         int i, j;
2317
2318         bh = __getblk(dev, block, bufsize);
2319         if (buffer_uptodate(bh))
2320                 return (bh);
2321
2322         if (block + BUFNR > max_block) {
2323                 blocks = max_block - block;
2324         }
2325         bhlist[0] = bh;
2326         j = 1;
2327         for (i = 1; i < blocks; i++) {
2328                 bh = __getblk(dev, block + i, bufsize);
2329                 if (buffer_uptodate(bh)) {
2330                         brelse(bh);
2331                         break;
2332                 } else
2333                         bhlist[j++] = bh;
2334         }
2335         ll_rw_block(READ, j, bhlist);
2336         for (i = 1; i < j; i++)
2337                 brelse(bhlist[i]);
2338         bh = bhlist[0];
2339         wait_on_buffer(bh);
2340         if (buffer_uptodate(bh))
2341                 return bh;
2342         brelse(bh);
2343         return NULL;
2344 }
2345
2346 /*
2347 ** read and replay the log
2348 ** on a clean unmount, the journal header's next unflushed pointer will be to an invalid
2349 ** transaction.  This tests that before finding all the transactions in the log, which makes normal mount times fast.
2350 **
2351 ** After a crash, this starts with the next unflushed transaction, and replays until it finds one too old, or invalid.
2352 **
2353 ** On exit, it sets things up so the first transaction will work correctly.
2354 */
2355 static int journal_read(struct super_block *p_s_sb)
2356 {
2357         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
2358         struct reiserfs_journal_desc *desc;
2359         unsigned int oldest_trans_id = 0;
2360         unsigned int oldest_invalid_trans_id = 0;
2361         time_t start;
2362         unsigned long oldest_start = 0;
2363         unsigned long cur_dblock = 0;
2364         unsigned long newest_mount_id = 9;
2365         struct buffer_head *d_bh;
2366         struct reiserfs_journal_header *jh;
2367         int valid_journal_header = 0;
2368         int replay_count = 0;
2369         int continue_replay = 1;
2370         int ret;
2371         char b[BDEVNAME_SIZE];
2372
2373         cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb);
2374         reiserfs_info(p_s_sb, "checking transaction log (%s)\n",
2375                       bdevname(journal->j_dev_bd, b));
2376         start = get_seconds();
2377
2378         /* step 1, read in the journal header block.  Check the transaction it says
2379          ** is the first unflushed, and if that transaction is not valid,
2380          ** replay is done
2381          */
2382         journal->j_header_bh = journal_bread(p_s_sb,
2383                                              SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb)
2384                                              + SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2385         if (!journal->j_header_bh) {
2386                 return 1;
2387         }
2388         jh = (struct reiserfs_journal_header *)(journal->j_header_bh->b_data);
2389         if (le32_to_cpu(jh->j_first_unflushed_offset) <
2390             SB_ONDISK_JOURNAL_SIZE(p_s_sb)
2391             && le32_to_cpu(jh->j_last_flush_trans_id) > 0) {
2392                 oldest_start =
2393                     SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2394                     le32_to_cpu(jh->j_first_unflushed_offset);
2395                 oldest_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2396                 newest_mount_id = le32_to_cpu(jh->j_mount_id);
2397                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2398                                "journal-1153: found in "
2399                                "header: first_unflushed_offset %d, last_flushed_trans_id "
2400                                "%lu", le32_to_cpu(jh->j_first_unflushed_offset),
2401                                le32_to_cpu(jh->j_last_flush_trans_id));
2402                 valid_journal_header = 1;
2403
2404                 /* now, we try to read the first unflushed offset.  If it is not valid,
2405                  ** there is nothing more we can do, and it makes no sense to read
2406                  ** through the whole log.
2407                  */
2408                 d_bh =
2409                     journal_bread(p_s_sb,
2410                                   SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2411                                   le32_to_cpu(jh->j_first_unflushed_offset));
2412                 ret = journal_transaction_is_valid(p_s_sb, d_bh, NULL, NULL);
2413                 if (!ret) {
2414                         continue_replay = 0;
2415                 }
2416                 brelse(d_bh);
2417                 goto start_log_replay;
2418         }
2419
2420         if (continue_replay && bdev_read_only(p_s_sb->s_bdev)) {
2421                 reiserfs_warning(p_s_sb,
2422                                  "clm-2076: device is readonly, unable to replay log");
2423                 return -1;
2424         }
2425
2426         /* ok, there are transactions that need to be replayed.  start with the first log block, find
2427          ** all the valid transactions, and pick out the oldest.
2428          */
2429         while (continue_replay
2430                && cur_dblock <
2431                (SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2432                 SB_ONDISK_JOURNAL_SIZE(p_s_sb))) {
2433                 /* Note that it is required for blocksize of primary fs device and journal
2434                    device to be the same */
2435                 d_bh =
2436                     reiserfs_breada(journal->j_dev_bd, cur_dblock,
2437                                     p_s_sb->s_blocksize,
2438                                     SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2439                                     SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2440                 ret =
2441                     journal_transaction_is_valid(p_s_sb, d_bh,
2442                                                  &oldest_invalid_trans_id,
2443                                                  &newest_mount_id);
2444                 if (ret == 1) {
2445                         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2446                         if (oldest_start == 0) {        /* init all oldest_ values */
2447                                 oldest_trans_id = get_desc_trans_id(desc);
2448                                 oldest_start = d_bh->b_blocknr;
2449                                 newest_mount_id = get_desc_mount_id(desc);
2450                                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2451                                                "journal-1179: Setting "
2452                                                "oldest_start to offset %llu, trans_id %lu",
2453                                                oldest_start -
2454                                                SB_ONDISK_JOURNAL_1st_BLOCK
2455                                                (p_s_sb), oldest_trans_id);
2456                         } else if (oldest_trans_id > get_desc_trans_id(desc)) {
2457                                 /* one we just read was older */
2458                                 oldest_trans_id = get_desc_trans_id(desc);
2459                                 oldest_start = d_bh->b_blocknr;
2460                                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2461                                                "journal-1180: Resetting "
2462                                                "oldest_start to offset %lu, trans_id %lu",
2463                                                oldest_start -
2464                                                SB_ONDISK_JOURNAL_1st_BLOCK
2465                                                (p_s_sb), oldest_trans_id);
2466                         }
2467                         if (newest_mount_id < get_desc_mount_id(desc)) {
2468                                 newest_mount_id = get_desc_mount_id(desc);
2469                                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2470                                                "journal-1299: Setting "
2471                                                "newest_mount_id to %d",
2472                                                get_desc_mount_id(desc));
2473                         }
2474                         cur_dblock += get_desc_trans_len(desc) + 2;
2475                 } else {
2476                         cur_dblock++;
2477                 }
2478                 brelse(d_bh);
2479         }
2480
2481       start_log_replay:
2482         cur_dblock = oldest_start;
2483         if (oldest_trans_id) {
2484                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2485                                "journal-1206: Starting replay "
2486                                "from offset %llu, trans_id %lu",
2487                                cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2488                                oldest_trans_id);
2489
2490         }
2491         replay_count = 0;
2492         while (continue_replay && oldest_trans_id > 0) {
2493                 ret =
2494                     journal_read_transaction(p_s_sb, cur_dblock, oldest_start,
2495                                              oldest_trans_id, newest_mount_id);
2496                 if (ret < 0) {
2497                         return ret;
2498                 } else if (ret != 0) {
2499                         break;
2500                 }
2501                 cur_dblock =
2502                     SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + journal->j_start;
2503                 replay_count++;
2504                 if (cur_dblock == oldest_start)
2505                         break;
2506         }
2507
2508         if (oldest_trans_id == 0) {
2509                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2510                                "journal-1225: No valid " "transactions found");
2511         }
2512         /* j_start does not get set correctly if we don't replay any transactions.
2513          ** if we had a valid journal_header, set j_start to the first unflushed transaction value,
2514          ** copy the trans_id from the header
2515          */
2516         if (valid_journal_header && replay_count == 0) {
2517                 journal->j_start = le32_to_cpu(jh->j_first_unflushed_offset);
2518                 journal->j_trans_id =
2519                     le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2520                 /* check for trans_id overflow */
2521                 if (journal->j_trans_id == 0)
2522                         journal->j_trans_id = 10;
2523                 journal->j_last_flush_trans_id =
2524                     le32_to_cpu(jh->j_last_flush_trans_id);
2525                 journal->j_mount_id = le32_to_cpu(jh->j_mount_id) + 1;
2526         } else {
2527                 journal->j_mount_id = newest_mount_id + 1;
2528         }
2529         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1299: Setting "
2530                        "newest_mount_id to %lu", journal->j_mount_id);
2531         journal->j_first_unflushed_offset = journal->j_start;
2532         if (replay_count > 0) {
2533                 reiserfs_info(p_s_sb,
2534                               "replayed %d transactions in %lu seconds\n",
2535                               replay_count, get_seconds() - start);
2536         }
2537         if (!bdev_read_only(p_s_sb->s_bdev) &&
2538             _update_journal_header_block(p_s_sb, journal->j_start,
2539                                          journal->j_last_flush_trans_id)) {
2540                 /* replay failed, caller must call free_journal_ram and abort
2541                  ** the mount
2542                  */
2543                 return -1;
2544         }
2545         return 0;
2546 }
2547
2548 static struct reiserfs_journal_list *alloc_journal_list(struct super_block *s)
2549 {
2550         struct reiserfs_journal_list *jl;
2551         jl = kzalloc(sizeof(struct reiserfs_journal_list),
2552                      GFP_NOFS | __GFP_NOFAIL);
2553         INIT_LIST_HEAD(&jl->j_list);
2554         INIT_LIST_HEAD(&jl->j_working_list);
2555         INIT_LIST_HEAD(&jl->j_tail_bh_list);
2556         INIT_LIST_HEAD(&jl->j_bh_list);
2557         mutex_init(&jl->j_commit_mutex);
2558         SB_JOURNAL(s)->j_num_lists++;
2559         get_journal_list(jl);
2560         return jl;
2561 }
2562
2563 static void journal_list_init(struct super_block *p_s_sb)
2564 {
2565         SB_JOURNAL(p_s_sb)->j_current_jl = alloc_journal_list(p_s_sb);
2566 }
2567
2568 static int release_journal_dev(struct super_block *super,
2569                                struct reiserfs_journal *journal)
2570 {
2571         int result;
2572
2573         result = 0;
2574
2575         if (journal->j_dev_bd != NULL) {
2576                 if (journal->j_dev_bd->bd_dev != super->s_dev)
2577                         bd_release(journal->j_dev_bd);
2578                 result = blkdev_put(journal->j_dev_bd, journal->j_dev_mode);
2579                 journal->j_dev_bd = NULL;
2580         }
2581
2582         if (result != 0) {
2583                 reiserfs_warning(super,
2584                                  "sh-457: release_journal_dev: Cannot release journal device: %i",
2585                                  result);
2586         }
2587         return result;
2588 }
2589
2590 static int journal_init_dev(struct super_block *super,
2591                             struct reiserfs_journal *journal,
2592                             const char *jdev_name)
2593 {
2594         int result;
2595         dev_t jdev;
2596         fmode_t blkdev_mode = FMODE_READ | FMODE_WRITE;
2597         char b[BDEVNAME_SIZE];
2598
2599         result = 0;
2600
2601         journal->j_dev_bd = NULL;
2602         jdev = SB_ONDISK_JOURNAL_DEVICE(super) ?
2603             new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super)) : super->s_dev;
2604
2605         if (bdev_read_only(super->s_bdev))
2606                 blkdev_mode = FMODE_READ;
2607
2608         /* there is no "jdev" option and journal is on separate device */
2609         if ((!jdev_name || !jdev_name[0])) {
2610                 journal->j_dev_bd = open_by_devnum(jdev, blkdev_mode);
2611                 journal->j_dev_mode = blkdev_mode;
2612                 if (IS_ERR(journal->j_dev_bd)) {
2613                         result = PTR_ERR(journal->j_dev_bd);
2614                         journal->j_dev_bd = NULL;
2615                         reiserfs_warning(super, "sh-458: journal_init_dev: "
2616                                          "cannot init journal device '%s': %i",
2617                                          __bdevname(jdev, b), result);
2618                         return result;
2619                 } else if (jdev != super->s_dev) {
2620                         result = bd_claim(journal->j_dev_bd, journal);
2621                         if (result) {
2622                                 blkdev_put(journal->j_dev_bd, blkdev_mode);
2623                                 return result;
2624                         }
2625
2626                         set_blocksize(journal->j_dev_bd, super->s_blocksize);
2627                 }
2628
2629                 return 0;
2630         }
2631
2632         journal->j_dev_mode = blkdev_mode;
2633         journal->j_dev_bd = open_bdev_exclusive(jdev_name,
2634                                                 blkdev_mode, journal);
2635         if (IS_ERR(journal->j_dev_bd)) {
2636                 result = PTR_ERR(journal->j_dev_bd);
2637                 journal->j_dev_bd = NULL;
2638                 reiserfs_warning(super,
2639                                  "journal_init_dev: Cannot open '%s': %i",
2640                                  jdev_name, result);
2641                 return result;
2642         }
2643
2644         set_blocksize(journal->j_dev_bd, super->s_blocksize);
2645         reiserfs_info(super,
2646                       "journal_init_dev: journal device: %s\n",
2647                       bdevname(journal->j_dev_bd, b));
2648         return 0;
2649 }
2650
2651 /**
2652  * When creating/tuning a file system user can assign some
2653  * journal params within boundaries which depend on the ratio
2654  * blocksize/standard_blocksize.
2655  *
2656  * For blocks >= standard_blocksize transaction size should
2657  * be not less then JOURNAL_TRANS_MIN_DEFAULT, and not more
2658  * then JOURNAL_TRANS_MAX_DEFAULT.
2659  *
2660  * For blocks < standard_blocksize these boundaries should be
2661  * decreased proportionally.
2662  */
2663 #define REISERFS_STANDARD_BLKSIZE (4096)
2664
2665 static int check_advise_trans_params(struct super_block *p_s_sb,
2666                                      struct reiserfs_journal *journal)
2667 {
2668         if (journal->j_trans_max) {
2669                 /* Non-default journal params.
2670                    Do sanity check for them. */
2671                 int ratio = 1;
2672                 if (p_s_sb->s_blocksize < REISERFS_STANDARD_BLKSIZE)
2673                         ratio = REISERFS_STANDARD_BLKSIZE / p_s_sb->s_blocksize;
2674
2675                 if (journal->j_trans_max > JOURNAL_TRANS_MAX_DEFAULT / ratio ||
2676                     journal->j_trans_max < JOURNAL_TRANS_MIN_DEFAULT / ratio ||
2677                     SB_ONDISK_JOURNAL_SIZE(p_s_sb) / journal->j_trans_max <
2678                     JOURNAL_MIN_RATIO) {
2679                         reiserfs_warning(p_s_sb,
2680                                  "sh-462: bad transaction max size (%u). FSCK?",
2681                                  journal->j_trans_max);
2682                         return 1;
2683                 }
2684                 if (journal->j_max_batch != (journal->j_trans_max) *
2685                         JOURNAL_MAX_BATCH_DEFAULT/JOURNAL_TRANS_MAX_DEFAULT) {
2686                         reiserfs_warning(p_s_sb,
2687                                 "sh-463: bad transaction max batch (%u). FSCK?",
2688                                 journal->j_max_batch);
2689                         return 1;
2690                 }
2691         } else {
2692                 /* Default journal params.
2693                    The file system was created by old version
2694                    of mkreiserfs, so some fields contain zeros,
2695                    and we need to advise proper values for them */
2696                 if (p_s_sb->s_blocksize != REISERFS_STANDARD_BLKSIZE)
2697                         reiserfs_panic(p_s_sb, "sh-464: bad blocksize (%u)",
2698                                        p_s_sb->s_blocksize);
2699                 journal->j_trans_max = JOURNAL_TRANS_MAX_DEFAULT;
2700                 journal->j_max_batch = JOURNAL_MAX_BATCH_DEFAULT;
2701                 journal->j_max_commit_age = JOURNAL_MAX_COMMIT_AGE;
2702         }
2703         return 0;
2704 }
2705
2706 /*
2707 ** must be called once on fs mount.  calls journal_read for you
2708 */
2709 int journal_init(struct super_block *p_s_sb, const char *j_dev_name,
2710                  int old_format, unsigned int commit_max_age)
2711 {
2712         int num_cnodes = SB_ONDISK_JOURNAL_SIZE(p_s_sb) * 2;
2713         struct buffer_head *bhjh;
2714         struct reiserfs_super_block *rs;
2715         struct reiserfs_journal_header *jh;
2716         struct reiserfs_journal *journal;
2717         struct reiserfs_journal_list *jl;
2718         char b[BDEVNAME_SIZE];
2719
2720         journal = SB_JOURNAL(p_s_sb) = vmalloc(sizeof(struct reiserfs_journal));
2721         if (!journal) {
2722                 reiserfs_warning(p_s_sb,
2723                                  "journal-1256: unable to get memory for journal structure");
2724                 return 1;
2725         }
2726         memset(journal, 0, sizeof(struct reiserfs_journal));
2727         INIT_LIST_HEAD(&journal->j_bitmap_nodes);
2728         INIT_LIST_HEAD(&journal->j_prealloc_list);
2729         INIT_LIST_HEAD(&journal->j_working_list);
2730         INIT_LIST_HEAD(&journal->j_journal_list);
2731         journal->j_persistent_trans = 0;
2732         if (reiserfs_allocate_list_bitmaps(p_s_sb,
2733                                            journal->j_list_bitmap,
2734                                            reiserfs_bmap_count(p_s_sb)))
2735                 goto free_and_return;
2736         allocate_bitmap_nodes(p_s_sb);
2737
2738         /* reserved for journal area support */
2739         SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb) = (old_format ?
2740                                                  REISERFS_OLD_DISK_OFFSET_IN_BYTES
2741                                                  / p_s_sb->s_blocksize +
2742                                                  reiserfs_bmap_count(p_s_sb) +
2743                                                  1 :
2744                                                  REISERFS_DISK_OFFSET_IN_BYTES /
2745                                                  p_s_sb->s_blocksize + 2);
2746
2747         /* Sanity check to see is the standard journal fitting withing first bitmap
2748            (actual for small blocksizes) */
2749         if (!SB_ONDISK_JOURNAL_DEVICE(p_s_sb) &&
2750             (SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb) +
2751              SB_ONDISK_JOURNAL_SIZE(p_s_sb) > p_s_sb->s_blocksize * 8)) {
2752                 reiserfs_warning(p_s_sb,
2753                                  "journal-1393: journal does not fit for area "
2754                                  "addressed by first of bitmap blocks. It starts at "
2755                                  "%u and its size is %u. Block size %ld",
2756                                  SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb),
2757                                  SB_ONDISK_JOURNAL_SIZE(p_s_sb),
2758                                  p_s_sb->s_blocksize);
2759                 goto free_and_return;
2760         }
2761
2762         if (journal_init_dev(p_s_sb, journal, j_dev_name) != 0) {
2763                 reiserfs_warning(p_s_sb,
2764                                  "sh-462: unable to initialize jornal device");
2765                 goto free_and_return;
2766         }
2767
2768         rs = SB_DISK_SUPER_BLOCK(p_s_sb);
2769
2770         /* read journal header */
2771         bhjh = journal_bread(p_s_sb,
2772                              SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2773                              SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2774         if (!bhjh) {
2775                 reiserfs_warning(p_s_sb,
2776                                  "sh-459: unable to read journal header");
2777                 goto free_and_return;
2778         }
2779         jh = (struct reiserfs_journal_header *)(bhjh->b_data);
2780
2781         /* make sure that journal matches to the super block */
2782         if (is_reiserfs_jr(rs)
2783             && (le32_to_cpu(jh->jh_journal.jp_journal_magic) !=
2784                 sb_jp_journal_magic(rs))) {
2785                 reiserfs_warning(p_s_sb,
2786                                  "sh-460: journal header magic %x "
2787                                  "(device %s) does not match to magic found in super "
2788                                  "block %x", jh->jh_journal.jp_journal_magic,
2789                                  bdevname(journal->j_dev_bd, b),
2790                                  sb_jp_journal_magic(rs));
2791                 brelse(bhjh);
2792                 goto free_and_return;
2793         }
2794
2795         journal->j_trans_max = le32_to_cpu(jh->jh_journal.jp_journal_trans_max);
2796         journal->j_max_batch = le32_to_cpu(jh->jh_journal.jp_journal_max_batch);
2797         journal->j_max_commit_age =
2798             le32_to_cpu(jh->jh_journal.jp_journal_max_commit_age);
2799         journal->j_max_trans_age = JOURNAL_MAX_TRANS_AGE;
2800
2801         if (check_advise_trans_params(p_s_sb, journal) != 0)
2802                 goto free_and_return;
2803         journal->j_default_max_commit_age = journal->j_max_commit_age;
2804
2805         if (commit_max_age != 0) {
2806                 journal->j_max_commit_age = commit_max_age;
2807                 journal->j_max_trans_age = commit_max_age;
2808         }
2809
2810         reiserfs_info(p_s_sb, "journal params: device %s, size %u, "
2811                       "journal first block %u, max trans len %u, max batch %u, "
2812                       "max commit age %u, max trans age %u\n",
2813                       bdevname(journal->j_dev_bd, b),
2814                       SB_ONDISK_JOURNAL_SIZE(p_s_sb),
2815                       SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2816                       journal->j_trans_max,
2817                       journal->j_max_batch,
2818                       journal->j_max_commit_age, journal->j_max_trans_age);
2819
2820         brelse(bhjh);
2821
2822         journal->j_list_bitmap_index = 0;
2823         journal_list_init(p_s_sb);
2824
2825         memset(journal->j_list_hash_table, 0,
2826                JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
2827
2828         INIT_LIST_HEAD(&journal->j_dirty_buffers);
2829         spin_lock_init(&journal->j_dirty_buffers_lock);
2830
2831         journal->j_start = 0;
2832         journal->j_len = 0;
2833         journal->j_len_alloc = 0;
2834         atomic_set(&(journal->j_wcount), 0);
2835         atomic_set(&(journal->j_async_throttle), 0);
2836         journal->j_bcount = 0;
2837         journal->j_trans_start_time = 0;
2838         journal->j_last = NULL;
2839         journal->j_first = NULL;
2840         init_waitqueue_head(&(journal->j_join_wait));
2841         mutex_init(&journal->j_mutex);
2842         mutex_init(&journal->j_flush_mutex);
2843
2844         journal->j_trans_id = 10;
2845         journal->j_mount_id = 10;
2846         journal->j_state = 0;
2847         atomic_set(&(journal->j_jlock), 0);
2848         journal->j_cnode_free_list = allocate_cnodes(num_cnodes);
2849         journal->j_cnode_free_orig = journal->j_cnode_free_list;
2850         journal->j_cnode_free = journal->j_cnode_free_list ? num_cnodes : 0;
2851         journal->j_cnode_used = 0;
2852         journal->j_must_wait = 0;
2853
2854         if (journal->j_cnode_free == 0) {
2855                 reiserfs_warning(p_s_sb, "journal-2004: Journal cnode memory "
2856                                  "allocation failed (%ld bytes). Journal is "
2857                                  "too large for available memory. Usually "
2858                                  "this is due to a journal that is too large.",
2859                                  sizeof (struct reiserfs_journal_cnode) * num_cnodes);
2860                 goto free_and_return;
2861         }
2862
2863         init_journal_hash(p_s_sb);
2864         jl = journal->j_current_jl;
2865         jl->j_list_bitmap = get_list_bitmap(p_s_sb, jl);
2866         if (!jl->j_list_bitmap) {
2867                 reiserfs_warning(p_s_sb,
2868                                  "journal-2005, get_list_bitmap failed for journal list 0");
2869                 goto free_and_return;
2870         }
2871         if (journal_read(p_s_sb) < 0) {
2872                 reiserfs_warning(p_s_sb, "Replay Failure, unable to mount");
2873                 goto free_and_return;
2874         }
2875
2876         reiserfs_mounted_fs_count++;
2877         if (reiserfs_mounted_fs_count <= 1)
2878                 commit_wq = create_workqueue("reiserfs");
2879
2880         INIT_DELAYED_WORK(&journal->j_work, flush_async_commits);
2881         journal->j_work_sb = p_s_sb;
2882         return 0;
2883       free_and_return:
2884         free_journal_ram(p_s_sb);
2885         return 1;
2886 }
2887
2888 /*
2889 ** test for a polite end of the current transaction.  Used by file_write, and should
2890 ** be used by delete to make sure they don't write more than can fit inside a single
2891 ** transaction
2892 */
2893 int journal_transaction_should_end(struct reiserfs_transaction_handle *th,
2894                                    int new_alloc)
2895 {
2896         struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2897         time_t now = get_seconds();
2898         /* cannot restart while nested */
2899         BUG_ON(!th->t_trans_id);
2900         if (th->t_refcount > 1)
2901                 return 0;
2902         if (journal->j_must_wait > 0 ||
2903             (journal->j_len_alloc + new_alloc) >= journal->j_max_batch ||
2904             atomic_read(&(journal->j_jlock)) ||
2905             (now - journal->j_trans_start_time) > journal->j_max_trans_age ||
2906             journal->j_cnode_free < (journal->j_trans_max * 3)) {
2907                 return 1;
2908         }
2909         /* protected by the BKL here */
2910         journal->j_len_alloc += new_alloc;
2911         th->t_blocks_allocated += new_alloc ;
2912         return 0;
2913 }
2914
2915 /* this must be called inside a transaction, and requires the
2916 ** kernel_lock to be held
2917 */
2918 void reiserfs_block_writes(struct reiserfs_transaction_handle *th)
2919 {
2920         struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2921         BUG_ON(!th->t_trans_id);
2922         journal->j_must_wait = 1;
2923         set_bit(J_WRITERS_BLOCKED, &journal->j_state);
2924         return;
2925 }
2926
2927 /* this must be called without a transaction started, and does not
2928 ** require BKL
2929 */
2930 void reiserfs_allow_writes(struct super_block *s)
2931 {
2932         struct reiserfs_journal *journal = SB_JOURNAL(s);
2933         clear_bit(J_WRITERS_BLOCKED, &journal->j_state);
2934         wake_up(&journal->j_join_wait);
2935 }
2936
2937 /* this must be called without a transaction started, and does not
2938 ** require BKL
2939 */
2940 void reiserfs_wait_on_write_block(struct super_block *s)
2941 {
2942         struct reiserfs_journal *journal = SB_JOURNAL(s);
2943         wait_event(journal->j_join_wait,
2944                    !test_bit(J_WRITERS_BLOCKED, &journal->j_state));
2945 }
2946
2947 static void queue_log_writer(struct super_block *s)
2948 {
2949         wait_queue_t wait;
2950         struct reiserfs_journal *journal = SB_JOURNAL(s);
2951         set_bit(J_WRITERS_QUEUED, &journal->j_state);
2952
2953         /*
2954          * we don't want to use wait_event here because
2955          * we only want to wait once.
2956          */
2957         init_waitqueue_entry(&wait, current);
2958         add_wait_queue(&journal->j_join_wait, &wait);
2959         set_current_state(TASK_UNINTERRUPTIBLE);
2960         if (test_bit(J_WRITERS_QUEUED, &journal->j_state))
2961                 schedule();
2962         __set_current_state(TASK_RUNNING);
2963         remove_wait_queue(&journal->j_join_wait, &wait);
2964 }
2965
2966 static void wake_queued_writers(struct super_block *s)
2967 {
2968         struct reiserfs_journal *journal = SB_JOURNAL(s);
2969         if (test_and_clear_bit(J_WRITERS_QUEUED, &journal->j_state))
2970                 wake_up(&journal->j_join_wait);
2971 }
2972
2973 static void let_transaction_grow(struct super_block *sb, unsigned int trans_id)
2974 {
2975         struct reiserfs_journal *journal = SB_JOURNAL(sb);
2976         unsigned long bcount = journal->j_bcount;
2977         while (1) {
2978                 schedule_timeout_uninterruptible(1);
2979                 journal->j_current_jl->j_state |= LIST_COMMIT_PENDING;
2980                 while ((atomic_read(&journal->j_wcount) > 0 ||
2981                         atomic_read(&journal->j_jlock)) &&
2982                        journal->j_trans_id == trans_id) {
2983                         queue_log_writer(sb);
2984                 }
2985                 if (journal->j_trans_id != trans_id)
2986                         break;
2987                 if (bcount == journal->j_bcount)
2988                         break;
2989                 bcount = journal->j_bcount;
2990         }
2991 }
2992
2993 /* join == true if you must join an existing transaction.
2994 ** join == false if you can deal with waiting for others to finish
2995 **
2996 ** this will block until the transaction is joinable.  send the number of blocks you
2997 ** expect to use in nblocks.
2998 */
2999 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
3000                               struct super_block *p_s_sb, unsigned long nblocks,
3001                               int join)
3002 {
3003         time_t now = get_seconds();
3004         unsigned int old_trans_id;
3005         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3006         struct reiserfs_transaction_handle myth;
3007         int sched_count = 0;
3008         int retval;
3009
3010         reiserfs_check_lock_depth(p_s_sb, "journal_begin");
3011         BUG_ON(nblocks > journal->j_trans_max);
3012
3013         PROC_INFO_INC(p_s_sb, journal.journal_being);
3014         /* set here for journal_join */
3015         th->t_refcount = 1;
3016         th->t_super = p_s_sb;
3017
3018       relock:
3019         lock_journal(p_s_sb);
3020         if (join != JBEGIN_ABORT && reiserfs_is_journal_aborted(journal)) {
3021                 unlock_journal(p_s_sb);
3022                 retval = journal->j_errno;
3023                 goto out_fail;
3024         }
3025         journal->j_bcount++;
3026
3027         if (test_bit(J_WRITERS_BLOCKED, &journal->j_state)) {
3028                 unlock_journal(p_s_sb);
3029                 reiserfs_wait_on_write_block(p_s_sb);
3030                 PROC_INFO_INC(p_s_sb, journal.journal_relock_writers);
3031                 goto relock;
3032         }
3033         now = get_seconds();
3034
3035         /* if there is no room in the journal OR
3036          ** if this transaction is too old, and we weren't called joinable, wait for it to finish before beginning
3037          ** we don't sleep if there aren't other writers
3038          */
3039
3040         if ((!join && journal->j_must_wait > 0) ||
3041             (!join
3042              && (journal->j_len_alloc + nblocks + 2) >= journal->j_max_batch)
3043             || (!join && atomic_read(&journal->j_wcount) > 0
3044                 && journal->j_trans_start_time > 0
3045                 && (now - journal->j_trans_start_time) >
3046                 journal->j_max_trans_age) || (!join
3047                                               && atomic_read(&journal->j_jlock))
3048             || (!join && journal->j_cnode_free < (journal->j_trans_max * 3))) {
3049
3050                 old_trans_id = journal->j_trans_id;
3051                 unlock_journal(p_s_sb); /* allow others to finish this transaction */
3052
3053                 if (!join && (journal->j_len_alloc + nblocks + 2) >=
3054                     journal->j_max_batch &&
3055                     ((journal->j_len + nblocks + 2) * 100) <
3056                     (journal->j_len_alloc * 75)) {
3057                         if (atomic_read(&journal->j_wcount) > 10) {
3058                                 sched_count++;
3059                                 queue_log_writer(p_s_sb);
3060                                 goto relock;
3061                         }
3062                 }
3063                 /* don't mess with joining the transaction if all we have to do is
3064                  * wait for someone else to do a commit
3065                  */
3066                 if (atomic_read(&journal->j_jlock)) {
3067                         while (journal->j_trans_id == old_trans_id &&
3068                                atomic_read(&journal->j_jlock)) {
3069                                 queue_log_writer(p_s_sb);
3070                         }
3071                         goto relock;
3072                 }
3073                 retval = journal_join(&myth, p_s_sb, 1);
3074                 if (retval)
3075                         goto out_fail;
3076
3077                 /* someone might have ended the transaction while we joined */
3078                 if (old_trans_id != journal->j_trans_id) {
3079                         retval = do_journal_end(&myth, p_s_sb, 1, 0);
3080                 } else {
3081                         retval = do_journal_end(&myth, p_s_sb, 1, COMMIT_NOW);
3082                 }
3083
3084                 if (retval)
3085                         goto out_fail;
3086
3087                 PROC_INFO_INC(p_s_sb, journal.journal_relock_wcount);
3088                 goto relock;
3089         }
3090         /* we are the first writer, set trans_id */
3091         if (journal->j_trans_start_time == 0) {
3092                 journal->j_trans_start_time = get_seconds();
3093         }
3094         atomic_inc(&(journal->j_wcount));
3095         journal->j_len_alloc += nblocks;
3096         th->t_blocks_logged = 0;
3097         th->t_blocks_allocated = nblocks;
3098         th->t_trans_id = journal->j_trans_id;
3099         unlock_journal(p_s_sb);
3100         INIT_LIST_HEAD(&th->t_list);
3101         get_fs_excl();
3102         return 0;
3103
3104       out_fail:
3105         memset(th, 0, sizeof(*th));
3106         /* Re-set th->t_super, so we can properly keep track of how many
3107          * persistent transactions there are. We need to do this so if this
3108          * call is part of a failed restart_transaction, we can free it later */
3109         th->t_super = p_s_sb;
3110         return retval;
3111 }
3112
3113 struct reiserfs_transaction_handle *reiserfs_persistent_transaction(struct
3114                                                                     super_block
3115                                                                     *s,
3116                                                                     int nblocks)
3117 {
3118         int ret;
3119         struct reiserfs_transaction_handle *th;
3120
3121         /* if we're nesting into an existing transaction.  It will be
3122          ** persistent on its own
3123          */
3124         if (reiserfs_transaction_running(s)) {
3125                 th = current->journal_info;
3126                 th->t_refcount++;
3127                 BUG_ON(th->t_refcount < 2);
3128
3129                 return th;
3130         }
3131         th = kmalloc(sizeof(struct reiserfs_transaction_handle), GFP_NOFS);
3132         if (!th)
3133                 return NULL;
3134         ret = journal_begin(th, s, nblocks);
3135         if (ret) {
3136                 kfree(th);
3137                 return NULL;
3138         }
3139
3140         SB_JOURNAL(s)->j_persistent_trans++;
3141         return th;
3142 }
3143
3144 int reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle *th)
3145 {
3146         struct super_block *s = th->t_super;
3147         int ret = 0;
3148         if (th->t_trans_id)
3149                 ret = journal_end(th, th->t_super, th->t_blocks_allocated);
3150         else
3151                 ret = -EIO;
3152         if (th->t_refcount == 0) {
3153                 SB_JOURNAL(s)->j_persistent_trans--;
3154                 kfree(th);
3155         }
3156         return ret;
3157 }
3158
3159 static int journal_join(struct reiserfs_transaction_handle *th,
3160                         struct super_block *p_s_sb, unsigned long nblocks)
3161 {
3162         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3163
3164         /* this keeps do_journal_end from NULLing out the current->journal_info
3165          ** pointer
3166          */
3167         th->t_handle_save = cur_th;
3168         BUG_ON(cur_th && cur_th->t_refcount > 1);
3169         return do_journal_begin_r(th, p_s_sb, nblocks, JBEGIN_JOIN);
3170 }
3171
3172 int journal_join_abort(struct reiserfs_transaction_handle *th,
3173                        struct super_block *p_s_sb, unsigned long nblocks)
3174 {
3175         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3176
3177         /* this keeps do_journal_end from NULLing out the current->journal_info
3178          ** pointer
3179          */
3180         th->t_handle_save = cur_th;
3181         BUG_ON(cur_th && cur_th->t_refcount > 1);
3182         return do_journal_begin_r(th, p_s_sb, nblocks, JBEGIN_ABORT);
3183 }
3184
3185 int journal_begin(struct reiserfs_transaction_handle *th,
3186                   struct super_block *p_s_sb, unsigned long nblocks)
3187 {
3188         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3189         int ret;
3190
3191         th->t_handle_save = NULL;
3192         if (cur_th) {
3193                 /* we are nesting into the current transaction */
3194                 if (cur_th->t_super == p_s_sb) {
3195                         BUG_ON(!cur_th->t_refcount);
3196                         cur_th->t_refcount++;
3197                         memcpy(th, cur_th, sizeof(*th));
3198                         if (th->t_refcount <= 1)
3199                                 reiserfs_warning(p_s_sb,
3200                                                  "BAD: refcount <= 1, but journal_info != 0");
3201                         return 0;
3202                 } else {
3203                         /* we've ended up with a handle from a different filesystem.
3204                          ** save it and restore on journal_end.  This should never
3205                          ** really happen...
3206                          */
3207                         reiserfs_warning(p_s_sb,
3208                                          "clm-2100: nesting info a different FS");
3209                         th->t_handle_save = current->journal_info;
3210                         current->journal_info = th;
3211                 }
3212         } else {
3213                 current->journal_info = th;
3214         }
3215         ret = do_journal_begin_r(th, p_s_sb, nblocks, JBEGIN_REG);
3216         BUG_ON(current->journal_info != th);
3217
3218         /* I guess this boils down to being the reciprocal of clm-2100 above.
3219          * If do_journal_begin_r fails, we need to put it back, since journal_end
3220          * won't be called to do it. */
3221         if (ret)
3222                 current->journal_info = th->t_handle_save;
3223         else
3224                 BUG_ON(!th->t_refcount);
3225
3226         return ret;
3227 }
3228
3229 /*
3230 ** puts bh into the current transaction.  If it was already there, reorders removes the
3231 ** old pointers from the hash, and puts new ones in (to make sure replay happen in the right order).
3232 **
3233 ** if it was dirty, cleans and files onto the clean list.  I can't let it be dirty again until the
3234 ** transaction is committed.
3235 **
3236 ** if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len.
3237 */
3238 int journal_mark_dirty(struct reiserfs_transaction_handle *th,
3239                        struct super_block *p_s_sb, struct buffer_head *bh)
3240 {
3241         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3242         struct reiserfs_journal_cnode *cn = NULL;
3243         int count_already_incd = 0;
3244         int prepared = 0;
3245         BUG_ON(!th->t_trans_id);
3246
3247         PROC_INFO_INC(p_s_sb, journal.mark_dirty);
3248         if (th->t_trans_id != journal->j_trans_id) {
3249                 reiserfs_panic(th->t_super,
3250                                "journal-1577: handle trans id %ld != current trans id %ld\n",
3251                                th->t_trans_id, journal->j_trans_id);
3252         }
3253
3254         p_s_sb->s_dirt = 1;
3255
3256         prepared = test_clear_buffer_journal_prepared(bh);
3257         clear_buffer_journal_restore_dirty(bh);
3258         /* already in this transaction, we are done */
3259         if (buffer_journaled(bh)) {
3260                 PROC_INFO_INC(p_s_sb, journal.mark_dirty_already);
3261                 return 0;
3262         }
3263
3264         /* this must be turned into a panic instead of a warning.  We can't allow
3265          ** a dirty or journal_dirty or locked buffer to be logged, as some changes
3266          ** could get to disk too early.  NOT GOOD.
3267          */
3268         if (!prepared || buffer_dirty(bh)) {
3269                 reiserfs_warning(p_s_sb, "journal-1777: buffer %llu bad state "
3270                                  "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT",
3271                                  (unsigned long long)bh->b_blocknr,
3272                                  prepared ? ' ' : '!',
3273                                  buffer_locked(bh) ? ' ' : '!',
3274                                  buffer_dirty(bh) ? ' ' : '!',
3275                                  buffer_journal_dirty(bh) ? ' ' : '!');
3276         }
3277
3278         if (atomic_read(&(journal->j_wcount)) <= 0) {
3279                 reiserfs_warning(p_s_sb,
3280                                  "journal-1409: journal_mark_dirty returning because j_wcount was %d",
3281                                  atomic_read(&(journal->j_wcount)));
3282                 return 1;
3283         }
3284         /* this error means I've screwed up, and we've overflowed the transaction.
3285          ** Nothing can be done here, except make the FS readonly or panic.
3286          */
3287         if (journal->j_len >= journal->j_trans_max) {
3288                 reiserfs_panic(th->t_super,
3289                                "journal-1413: journal_mark_dirty: j_len (%lu) is too big\n",
3290                                journal->j_len);
3291         }
3292
3293         if (buffer_journal_dirty(bh)) {
3294                 count_already_incd = 1;
3295                 PROC_INFO_INC(p_s_sb, journal.mark_dirty_notjournal);
3296                 clear_buffer_journal_dirty(bh);
3297         }
3298
3299         if (journal->j_len > journal->j_len_alloc) {
3300                 journal->j_len_alloc = journal->j_len + JOURNAL_PER_BALANCE_CNT;
3301         }
3302
3303         set_buffer_journaled(bh);
3304
3305         /* now put this guy on the end */
3306         if (!cn) {
3307                 cn = get_cnode(p_s_sb);
3308                 if (!cn) {
3309                         reiserfs_panic(p_s_sb, "get_cnode failed!\n");
3310                 }
3311
3312                 if (th->t_blocks_logged == th->t_blocks_allocated) {
3313                         th->t_blocks_allocated += JOURNAL_PER_BALANCE_CNT;
3314                         journal->j_len_alloc += JOURNAL_PER_BALANCE_CNT;
3315                 }
3316                 th->t_blocks_logged++;
3317                 journal->j_len++;
3318
3319                 cn->bh = bh;
3320                 cn->blocknr = bh->b_blocknr;
3321                 cn->sb = p_s_sb;
3322                 cn->jlist = NULL;
3323                 insert_journal_hash(journal->j_hash_table, cn);
3324                 if (!count_already_incd) {
3325                         get_bh(bh);
3326                 }
3327         }
3328         cn->next = NULL;
3329         cn->prev = journal->j_last;
3330         cn->bh = bh;
3331         if (journal->j_last) {
3332                 journal->j_last->next = cn;
3333                 journal->j_last = cn;
3334         } else {
3335                 journal->j_first = cn;
3336                 journal->j_last = cn;
3337         }
3338         return 0;
3339 }
3340
3341 int journal_end(struct reiserfs_transaction_handle *th,
3342                 struct super_block *p_s_sb, unsigned long nblocks)
3343 {
3344         if (!current->journal_info && th->t_refcount > 1)
3345                 reiserfs_warning(p_s_sb, "REISER-NESTING: th NULL, refcount %d",
3346                                  th->t_refcount);
3347
3348         if (!th->t_trans_id) {
3349                 WARN_ON(1);
3350                 return -EIO;
3351         }
3352
3353         th->t_refcount--;
3354         if (th->t_refcount > 0) {
3355                 struct reiserfs_transaction_handle *cur_th =
3356                     current->journal_info;
3357
3358                 /* we aren't allowed to close a nested transaction on a different
3359                  ** filesystem from the one in the task struct
3360                  */
3361                 BUG_ON(cur_th->t_super != th->t_super);
3362
3363                 if (th != cur_th) {
3364                         memcpy(current->journal_info, th, sizeof(*th));
3365                         th->t_trans_id = 0;
3366                 }
3367                 return 0;
3368         } else {
3369                 return do_journal_end(th, p_s_sb, nblocks, 0);
3370         }
3371 }
3372
3373 /* removes from the current transaction, relsing and descrementing any counters.
3374 ** also files the removed buffer directly onto the clean list
3375 **
3376 ** called by journal_mark_freed when a block has been deleted
3377 **
3378 ** returns 1 if it cleaned and relsed the buffer. 0 otherwise
3379 */
3380 static int remove_from_transaction(struct super_block *p_s_sb,
3381                                    b_blocknr_t blocknr, int already_cleaned)
3382 {
3383         struct buffer_head *bh;
3384         struct reiserfs_journal_cnode *cn;
3385         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3386         int ret = 0;
3387
3388         cn = get_journal_hash_dev(p_s_sb, journal->j_hash_table, blocknr);
3389         if (!cn || !cn->bh) {
3390                 return ret;
3391         }
3392         bh = cn->bh;
3393         if (cn->prev) {
3394                 cn->prev->next = cn->next;
3395         }
3396         if (cn->next) {
3397                 cn->next->prev = cn->prev;
3398         }
3399         if (cn == journal->j_first) {
3400                 journal->j_first = cn->next;
3401         }
3402         if (cn == journal->j_last) {
3403                 journal->j_last = cn->prev;
3404         }
3405         if (bh)
3406                 remove_journal_hash(p_s_sb, journal->j_hash_table, NULL,
3407                                     bh->b_blocknr, 0);
3408         clear_buffer_journaled(bh);     /* don't log this one */
3409
3410         if (!already_cleaned) {
3411                 clear_buffer_journal_dirty(bh);
3412                 clear_buffer_dirty(bh);
3413                 clear_buffer_journal_test(bh);
3414                 put_bh(bh);
3415                 if (atomic_read(&(bh->b_count)) < 0) {
3416                         reiserfs_warning(p_s_sb,
3417                                          "journal-1752: remove from trans, b_count < 0");
3418                 }
3419                 ret = 1;
3420         }
3421         journal->j_len--;
3422         journal->j_len_alloc--;
3423         free_cnode(p_s_sb, cn);
3424         return ret;
3425 }
3426
3427 /*
3428 ** for any cnode in a journal list, it can only be dirtied of all the
3429 ** transactions that include it are committed to disk.
3430 ** this checks through each transaction, and returns 1 if you are allowed to dirty,
3431 ** and 0 if you aren't
3432 **
3433 ** it is called by dirty_journal_list, which is called after flush_commit_list has gotten all the log
3434 ** blocks for a given transaction on disk
3435 **
3436 */
3437 static int can_dirty(struct reiserfs_journal_cnode *cn)
3438 {
3439         struct super_block *sb = cn->sb;
3440         b_blocknr_t blocknr = cn->blocknr;
3441         struct reiserfs_journal_cnode *cur = cn->hprev;
3442         int can_dirty = 1;
3443
3444         /* first test hprev.  These are all newer than cn, so any node here
3445          ** with the same block number and dev means this node can't be sent
3446          ** to disk right now.
3447          */
3448         while (cur && can_dirty) {
3449                 if (cur->jlist && cur->bh && cur->blocknr && cur->sb == sb &&
3450                     cur->blocknr == blocknr) {
3451                         can_dirty = 0;
3452                 }
3453                 cur = cur->hprev;
3454         }
3455         /* then test hnext.  These are all older than cn.  As long as they
3456          ** are committed to the log, it is safe to write cn to disk
3457          */
3458         cur = cn->hnext;
3459         while (cur && can_dirty) {
3460                 if (cur->jlist && cur->jlist->j_len > 0 &&
3461                     atomic_read(&(cur->jlist->j_commit_left)) > 0 && cur->bh &&
3462                     cur->blocknr && cur->sb == sb && cur->blocknr == blocknr) {
3463                         can_dirty = 0;
3464                 }
3465                 cur = cur->hnext;
3466         }
3467         return can_dirty;
3468 }
3469
3470 /* syncs the commit blocks, but does not force the real buffers to disk
3471 ** will wait until the current transaction is done/committed before returning
3472 */
3473 int journal_end_sync(struct reiserfs_transaction_handle *th,
3474                      struct super_block *p_s_sb, unsigned long nblocks)
3475 {
3476         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3477
3478         BUG_ON(!th->t_trans_id);
3479         /* you can sync while nested, very, very bad */
3480         BUG_ON(th->t_refcount > 1);
3481         if (journal->j_len == 0) {
3482                 reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb),
3483                                              1);
3484                 journal_mark_dirty(th, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb));
3485         }
3486         return do_journal_end(th, p_s_sb, nblocks, COMMIT_NOW | WAIT);
3487 }
3488
3489 /*
3490 ** writeback the pending async commits to disk
3491 */
3492 static void flush_async_commits(struct work_struct *work)
3493 {
3494         struct reiserfs_journal *journal =
3495                 container_of(work, struct reiserfs_journal, j_work.work);
3496         struct super_block *p_s_sb = journal->j_work_sb;
3497         struct reiserfs_journal_list *jl;
3498         struct list_head *entry;
3499
3500         lock_kernel();
3501         if (!list_empty(&journal->j_journal_list)) {
3502                 /* last entry is the youngest, commit it and you get everything */
3503                 entry = journal->j_journal_list.prev;
3504                 jl = JOURNAL_LIST_ENTRY(entry);
3505                 flush_commit_list(p_s_sb, jl, 1);
3506         }
3507         unlock_kernel();
3508 }
3509
3510 /*
3511 ** flushes any old transactions to disk
3512 ** ends the current transaction if it is too old
3513 */
3514 int reiserfs_flush_old_commits(struct super_block *p_s_sb)
3515 {
3516         time_t now;
3517         struct reiserfs_transaction_handle th;
3518         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3519
3520         now = get_seconds();
3521         /* safety check so we don't flush while we are replaying the log during
3522          * mount
3523          */
3524         if (list_empty(&journal->j_journal_list)) {
3525                 return 0;
3526         }
3527
3528         /* check the current transaction.  If there are no writers, and it is
3529          * too old, finish it, and force the commit blocks to disk
3530          */
3531         if (atomic_read(&journal->j_wcount) <= 0 &&
3532             journal->j_trans_start_time > 0 &&
3533             journal->j_len > 0 &&
3534             (now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3535                 if (!journal_join(&th, p_s_sb, 1)) {
3536                         reiserfs_prepare_for_journal(p_s_sb,
3537                                                      SB_BUFFER_WITH_SB(p_s_sb),
3538                                                      1);
3539                         journal_mark_dirty(&th, p_s_sb,
3540                                            SB_BUFFER_WITH_SB(p_s_sb));
3541
3542                         /* we're only being called from kreiserfsd, it makes no sense to do
3543                          ** an async commit so that kreiserfsd can do it later
3544                          */
3545                         do_journal_end(&th, p_s_sb, 1, COMMIT_NOW | WAIT);
3546                 }
3547         }
3548         return p_s_sb->s_dirt;
3549 }
3550
3551 /*
3552 ** returns 0 if do_journal_end should return right away, returns 1 if do_journal_end should finish the commit
3553 **
3554 ** if the current transaction is too old, but still has writers, this will wait on j_join_wait until all
3555 ** the writers are done.  By the time it wakes up, the transaction it was called has already ended, so it just
3556 ** flushes the commit list and returns 0.
3557 **
3558 ** Won't batch when flush or commit_now is set.  Also won't batch when others are waiting on j_join_wait.
3559 **
3560 ** Note, we can't allow the journal_end to proceed while there are still writers in the log.
3561 */
3562 static int check_journal_end(struct reiserfs_transaction_handle *th,
3563                              struct super_block *p_s_sb, unsigned long nblocks,
3564                              int flags)
3565 {
3566
3567         time_t now;
3568         int flush = flags & FLUSH_ALL;
3569         int commit_now = flags & COMMIT_NOW;
3570         int wait_on_commit = flags & WAIT;
3571         struct reiserfs_journal_list *jl;
3572         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3573
3574         BUG_ON(!th->t_trans_id);
3575
3576         if (th->t_trans_id != journal->j_trans_id) {
3577                 reiserfs_panic(th->t_super,
3578                                "journal-1577: handle trans id %ld != current trans id %ld\n",
3579                                th->t_trans_id, journal->j_trans_id);
3580         }
3581
3582         journal->j_len_alloc -= (th->t_blocks_allocated - th->t_blocks_logged);
3583         if (atomic_read(&(journal->j_wcount)) > 0) {    /* <= 0 is allowed.  unmounting might not call begin */
3584                 atomic_dec(&(journal->j_wcount));
3585         }
3586
3587         /* BUG, deal with case where j_len is 0, but people previously freed blocks need to be released
3588          ** will be dealt with by next transaction that actually writes something, but should be taken
3589          ** care of in this trans
3590          */
3591         BUG_ON(journal->j_len == 0);
3592
3593         /* if wcount > 0, and we are called to with flush or commit_now,
3594          ** we wait on j_join_wait.  We will wake up when the last writer has
3595          ** finished the transaction, and started it on its way to the disk.
3596          ** Then, we flush the commit or journal list, and just return 0
3597          ** because the rest of journal end was already done for this transaction.
3598          */
3599         if (atomic_read(&(journal->j_wcount)) > 0) {
3600                 if (flush || commit_now) {
3601                         unsigned trans_id;
3602
3603                         jl = journal->j_current_jl;
3604                         trans_id = jl->j_trans_id;
3605                         if (wait_on_commit)
3606                                 jl->j_state |= LIST_COMMIT_PENDING;
3607                         atomic_set(&(journal->j_jlock), 1);
3608                         if (flush) {
3609                                 journal->j_next_full_flush = 1;
3610                         }
3611                         unlock_journal(p_s_sb);
3612
3613                         /* sleep while the current transaction is still j_jlocked */
3614                         while (journal->j_trans_id == trans_id) {
3615                                 if (atomic_read(&journal->j_jlock)) {
3616                                         queue_log_writer(p_s_sb);
3617                                 } else {
3618                                         lock_journal(p_s_sb);
3619                                         if (journal->j_trans_id == trans_id) {
3620                                                 atomic_set(&(journal->j_jlock),
3621                                                            1);
3622                                         }
3623                                         unlock_journal(p_s_sb);
3624                                 }
3625                         }
3626                         BUG_ON(journal->j_trans_id == trans_id);
3627
3628                         if (commit_now
3629                             && journal_list_still_alive(p_s_sb, trans_id)
3630                             && wait_on_commit) {
3631                                 flush_commit_list(p_s_sb, jl, 1);
3632                         }
3633                         return 0;
3634                 }
3635                 unlock_journal(p_s_sb);
3636                 return 0;
3637         }
3638
3639         /* deal with old transactions where we are the last writers */
3640         now = get_seconds();
3641         if ((now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3642                 commit_now = 1;
3643                 journal->j_next_async_flush = 1;
3644         }
3645         /* don't batch when someone is waiting on j_join_wait */
3646         /* don't batch when syncing the commit or flushing the whole trans */
3647         if (!(journal->j_must_wait > 0) && !(atomic_read(&(journal->j_jlock)))
3648             && !flush && !commit_now && (journal->j_len < journal->j_max_batch)
3649             && journal->j_len_alloc < journal->j_max_batch
3650             && journal->j_cnode_free > (journal->j_trans_max * 3)) {
3651                 journal->j_bcount++;
3652                 unlock_journal(p_s_sb);
3653                 return 0;
3654         }
3655
3656         if (journal->j_start > SB_ONDISK_JOURNAL_SIZE(p_s_sb)) {
3657                 reiserfs_panic(p_s_sb,
3658                                "journal-003: journal_end: j_start (%ld) is too high\n",
3659                                journal->j_start);
3660         }
3661         return 1;
3662 }
3663
3664 /*
3665 ** Does all the work that makes deleting blocks safe.
3666 ** when deleting a block mark BH_JNew, just remove it from the current transaction, clean it's buffer_head and move on.
3667 **
3668 ** otherwise:
3669 ** set a bit for the block in the journal bitmap.  That will prevent it from being allocated for unformatted nodes
3670 ** before this transaction has finished.
3671 **
3672 ** mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers.  That will prevent any old transactions with
3673 ** this block from trying to flush to the real location.  Since we aren't removing the cnode from the journal_list_hash,
3674 ** the block can't be reallocated yet.
3675 **
3676 ** Then remove it from the current transaction, decrementing any counters and filing it on the clean list.
3677 */
3678 int journal_mark_freed(struct reiserfs_transaction_handle *th,
3679                        struct super_block *p_s_sb, b_blocknr_t blocknr)
3680 {
3681         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3682         struct reiserfs_journal_cnode *cn = NULL;
3683         struct buffer_head *bh = NULL;
3684         struct reiserfs_list_bitmap *jb = NULL;
3685         int cleaned = 0;
3686         BUG_ON(!th->t_trans_id);
3687
3688         cn = get_journal_hash_dev(p_s_sb, journal->j_hash_table, blocknr);
3689         if (cn && cn->bh) {
3690                 bh = cn->bh;
3691                 get_bh(bh);
3692         }
3693         /* if it is journal new, we just remove it from this transaction */
3694         if (bh && buffer_journal_new(bh)) {
3695                 clear_buffer_journal_new(bh);
3696                 clear_prepared_bits(bh);
3697                 reiserfs_clean_and_file_buffer(bh);
3698                 cleaned = remove_from_transaction(p_s_sb, blocknr, cleaned);
3699         } else {
3700                 /* set the bit for this block in the journal bitmap for this transaction */
3701                 jb = journal->j_current_jl->j_list_bitmap;
3702                 if (!jb) {
3703                         reiserfs_panic(p_s_sb,
3704                                        "journal-1702: journal_mark_freed, journal_list_bitmap is NULL\n");
3705                 }
3706                 set_bit_in_list_bitmap(p_s_sb, blocknr, jb);
3707
3708                 /* Note, the entire while loop is not allowed to schedule.  */
3709
3710                 if (bh) {
3711                         clear_prepared_bits(bh);
3712                         reiserfs_clean_and_file_buffer(bh);
3713                 }
3714                 cleaned = remove_from_transaction(p_s_sb, blocknr, cleaned);
3715
3716                 /* find all older transactions with this block, make sure they don't try to write it out */
3717                 cn = get_journal_hash_dev(p_s_sb, journal->j_list_hash_table,
3718                                           blocknr);
3719                 while (cn) {
3720                         if (p_s_sb == cn->sb && blocknr == cn->blocknr) {
3721                                 set_bit(BLOCK_FREED, &cn->state);
3722                                 if (cn->bh) {
3723                                         if (!cleaned) {
3724                                                 /* remove_from_transaction will brelse the buffer if it was
3725                                                  ** in the current trans
3726                                                  */
3727                                                 clear_buffer_journal_dirty(cn->
3728                                                                            bh);
3729                                                 clear_buffer_dirty(cn->bh);
3730                                                 clear_buffer_journal_test(cn->
3731                                                                           bh);
3732                                                 cleaned = 1;
3733                                                 put_bh(cn->bh);
3734                                                 if (atomic_read
3735                                                     (&(cn->bh->b_count)) < 0) {
3736                                                         reiserfs_warning(p_s_sb,
3737                                                                          "journal-2138: cn->bh->b_count < 0");
3738                                                 }
3739                                         }
3740                                         if (cn->jlist) {        /* since we are clearing the bh, we MUST dec nonzerolen */
3741                                                 atomic_dec(&
3742                                                            (cn->jlist->
3743                                                             j_nonzerolen));
3744                                         }
3745                                         cn->bh = NULL;
3746                                 }
3747                         }
3748                         cn = cn->hnext;
3749                 }
3750         }
3751
3752         if (bh)
3753                 release_buffer_page(bh); /* get_hash grabs the buffer */
3754         return 0;
3755 }
3756
3757 void reiserfs_update_inode_transaction(struct inode *inode)
3758 {
3759         struct reiserfs_journal *journal = SB_JOURNAL(inode->i_sb);
3760         REISERFS_I(inode)->i_jl = journal->j_current_jl;
3761         REISERFS_I(inode)->i_trans_id = journal->j_trans_id;
3762 }
3763
3764 /*
3765  * returns -1 on error, 0 if no commits/barriers were done and 1
3766  * if a transaction was actually committed and the barrier was done
3767  */
3768 static int __commit_trans_jl(struct inode *inode, unsigned long id,
3769                              struct reiserfs_journal_list *jl)
3770 {
3771         struct reiserfs_transaction_handle th;
3772         struct super_block *sb = inode->i_sb;
3773         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3774         int ret = 0;
3775
3776         /* is it from the current transaction, or from an unknown transaction? */
3777         if (id == journal->j_trans_id) {
3778                 jl = journal->j_current_jl;
3779                 /* try to let other writers come in and grow this transaction */
3780                 let_transaction_grow(sb, id);
3781                 if (journal->j_trans_id != id) {
3782                         goto flush_commit_only;
3783                 }
3784
3785                 ret = journal_begin(&th, sb, 1);
3786                 if (ret)
3787                         return ret;
3788
3789                 /* someone might have ended this transaction while we joined */
3790                 if (journal->j_trans_id != id) {
3791                         reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3792                                                      1);
3793                         journal_mark_dirty(&th, sb, SB_BUFFER_WITH_SB(sb));
3794                         ret = journal_end(&th, sb, 1);
3795                         goto flush_commit_only;
3796                 }
3797
3798                 ret = journal_end_sync(&th, sb, 1);
3799                 if (!ret)
3800                         ret = 1;
3801
3802         } else {
3803                 /* this gets tricky, we have to make sure the journal list in
3804                  * the inode still exists.  We know the list is still around
3805                  * if we've got a larger transaction id than the oldest list
3806                  */
3807               flush_commit_only:
3808                 if (journal_list_still_alive(inode->i_sb, id)) {
3809                         /*
3810                          * we only set ret to 1 when we know for sure
3811                          * the barrier hasn't been started yet on the commit
3812                          * block.
3813                          */
3814                         if (atomic_read(&jl->j_commit_left) > 1)
3815                                 ret = 1;
3816                         flush_commit_list(sb, jl, 1);
3817                         if (journal->j_errno)
3818                                 ret = journal->j_errno;
3819                 }
3820         }
3821         /* otherwise the list is gone, and long since committed */
3822         return ret;
3823 }
3824
3825 int reiserfs_commit_for_inode(struct inode *inode)
3826 {
3827         unsigned int id = REISERFS_I(inode)->i_trans_id;
3828         struct reiserfs_journal_list *jl = REISERFS_I(inode)->i_jl;
3829
3830         /* for the whole inode, assume unset id means it was
3831          * changed in the current transaction.  More conservative
3832          */
3833         if (!id || !jl) {
3834                 reiserfs_update_inode_transaction(inode);
3835                 id = REISERFS_I(inode)->i_trans_id;
3836                 /* jl will be updated in __commit_trans_jl */
3837         }
3838
3839         return __commit_trans_jl(inode, id, jl);
3840 }
3841
3842 void reiserfs_restore_prepared_buffer(struct super_block *p_s_sb,
3843                                       struct buffer_head *bh)
3844 {
3845         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3846         PROC_INFO_INC(p_s_sb, journal.restore_prepared);
3847         if (!bh) {
3848                 return;
3849         }
3850         if (test_clear_buffer_journal_restore_dirty(bh) &&
3851             buffer_journal_dirty(bh)) {
3852                 struct reiserfs_journal_cnode *cn;
3853                 cn = get_journal_hash_dev(p_s_sb,
3854                                           journal->j_list_hash_table,
3855                                           bh->b_blocknr);
3856                 if (cn && can_dirty(cn)) {
3857                         set_buffer_journal_test(bh);
3858                         mark_buffer_dirty(bh);
3859                 }
3860         }
3861         clear_buffer_journal_prepared(bh);
3862 }
3863
3864 extern struct tree_balance *cur_tb;
3865 /*
3866 ** before we can change a metadata block, we have to make sure it won't
3867 ** be written to disk while we are altering it.  So, we must:
3868 ** clean it
3869 ** wait on it.
3870 **
3871 */
3872 int reiserfs_prepare_for_journal(struct super_block *p_s_sb,
3873                                  struct buffer_head *bh, int wait)
3874 {
3875         PROC_INFO_INC(p_s_sb, journal.prepare);
3876
3877         if (!trylock_buffer(bh)) {
3878                 if (!wait)
3879                         return 0;
3880                 lock_buffer(bh);
3881         }
3882         set_buffer_journal_prepared(bh);
3883         if (test_clear_buffer_dirty(bh) && buffer_journal_dirty(bh)) {
3884                 clear_buffer_journal_test(bh);
3885                 set_buffer_journal_restore_dirty(bh);
3886         }
3887         unlock_buffer(bh);
3888         return 1;
3889 }
3890
3891 static void flush_old_journal_lists(struct super_block *s)
3892 {
3893         struct reiserfs_journal *journal = SB_JOURNAL(s);
3894         struct reiserfs_journal_list *jl;
3895         struct list_head *entry;
3896         time_t now = get_seconds();
3897
3898         while (!list_empty(&journal->j_journal_list)) {
3899                 entry = journal->j_journal_list.next;
3900                 jl = JOURNAL_LIST_ENTRY(entry);
3901                 /* this check should always be run, to send old lists to disk */
3902                 if (jl->j_timestamp < (now - (JOURNAL_MAX_TRANS_AGE * 4)) &&
3903                     atomic_read(&jl->j_commit_left) == 0 &&
3904                     test_transaction(s, jl)) {
3905                         flush_used_journal_lists(s, jl);
3906                 } else {
3907                         break;
3908                 }
3909         }
3910 }
3911
3912 /*
3913 ** long and ugly.  If flush, will not return until all commit
3914 ** blocks and all real buffers in the trans are on disk.
3915 ** If no_async, won't return until all commit blocks are on disk.
3916 **
3917 ** keep reading, there are comments as you go along
3918 **
3919 ** If the journal is aborted, we just clean up. Things like flushing
3920 ** journal lists, etc just won't happen.
3921 */
3922 static int do_journal_end(struct reiserfs_transaction_handle *th,
3923                           struct super_block *p_s_sb, unsigned long nblocks,
3924                           int flags)
3925 {
3926         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3927         struct reiserfs_journal_cnode *cn, *next, *jl_cn;
3928         struct reiserfs_journal_cnode *last_cn = NULL;
3929         struct reiserfs_journal_desc *desc;
3930         struct reiserfs_journal_commit *commit;
3931         struct buffer_head *c_bh;       /* commit bh */
3932         struct buffer_head *d_bh;       /* desc bh */
3933         int cur_write_start = 0;        /* start index of current log write */
3934         int old_start;
3935         int i;
3936         int flush;
3937         int wait_on_commit;
3938         struct reiserfs_journal_list *jl, *temp_jl;
3939         struct list_head *entry, *safe;
3940         unsigned long jindex;
3941         unsigned int commit_trans_id;
3942         int trans_half;
3943
3944         BUG_ON(th->t_refcount > 1);
3945         BUG_ON(!th->t_trans_id);
3946
3947         /* protect flush_older_commits from doing mistakes if the
3948            transaction ID counter gets overflowed.  */
3949         if (th->t_trans_id == ~0U)
3950                 flags |= FLUSH_ALL | COMMIT_NOW | WAIT;
3951         flush = flags & FLUSH_ALL;
3952         wait_on_commit = flags & WAIT;
3953
3954         put_fs_excl();
3955         current->journal_info = th->t_handle_save;
3956         reiserfs_check_lock_depth(p_s_sb, "journal end");
3957         if (journal->j_len == 0) {
3958                 reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb),
3959                                              1);
3960                 journal_mark_dirty(th, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb));
3961         }
3962
3963         lock_journal(p_s_sb);
3964         if (journal->j_next_full_flush) {
3965                 flags |= FLUSH_ALL;
3966                 flush = 1;
3967         }
3968         if (journal->j_next_async_flush) {
3969                 flags |= COMMIT_NOW | WAIT;
3970                 wait_on_commit = 1;
3971         }
3972
3973         /* check_journal_end locks the journal, and unlocks if it does not return 1
3974          ** it tells us if we should continue with the journal_end, or just return
3975          */
3976         if (!check_journal_end(th, p_s_sb, nblocks, flags)) {
3977                 p_s_sb->s_dirt = 1;
3978                 wake_queued_writers(p_s_sb);
3979                 reiserfs_async_progress_wait(p_s_sb);
3980                 goto out;
3981         }
3982
3983         /* check_journal_end might set these, check again */
3984         if (journal->j_next_full_flush) {
3985                 flush = 1;
3986         }
3987
3988         /*
3989          ** j must wait means we have to flush the log blocks, and the real blocks for
3990          ** this transaction
3991          */
3992         if (journal->j_must_wait > 0) {
3993                 flush = 1;
3994         }
3995 #ifdef REISERFS_PREALLOCATE
3996         /* quota ops might need to nest, setup the journal_info pointer for them
3997          * and raise the refcount so that it is > 0. */
3998         current->journal_info = th;
3999         th->t_refcount++;
4000         reiserfs_discard_all_prealloc(th);      /* it should not involve new blocks into
4001                                                  * the transaction */
4002         th->t_refcount--;
4003         current->journal_info = th->t_handle_save;
4004 #endif
4005
4006         /* setup description block */
4007         d_bh =
4008             journal_getblk(p_s_sb,
4009                            SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
4010                            journal->j_start);
4011         set_buffer_uptodate(d_bh);
4012         desc = (struct reiserfs_journal_desc *)(d_bh)->b_data;
4013         memset(d_bh->b_data, 0, d_bh->b_size);
4014         memcpy(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8);
4015         set_desc_trans_id(desc, journal->j_trans_id);
4016
4017         /* setup commit block.  Don't write (keep it clean too) this one until after everyone else is written */
4018         c_bh = journal_getblk(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
4019                               ((journal->j_start + journal->j_len +
4020                                 1) % SB_ONDISK_JOURNAL_SIZE(p_s_sb)));
4021         commit = (struct reiserfs_journal_commit *)c_bh->b_data;
4022         memset(c_bh->b_data, 0, c_bh->b_size);
4023         set_commit_trans_id(commit, journal->j_trans_id);
4024         set_buffer_uptodate(c_bh);
4025
4026         /* init this journal list */
4027         jl = journal->j_current_jl;
4028
4029         /* we lock the commit before doing anything because
4030          * we want to make sure nobody tries to run flush_commit_list until
4031          * the new transaction is fully setup, and we've already flushed the
4032          * ordered bh list
4033          */
4034         mutex_lock(&jl->j_commit_mutex);
4035
4036         /* save the transaction id in case we need to commit it later */
4037         commit_trans_id = jl->j_trans_id;
4038
4039         atomic_set(&jl->j_older_commits_done, 0);
4040         jl->j_trans_id = journal->j_trans_id;
4041         jl->j_timestamp = journal->j_trans_start_time;
4042         jl->j_commit_bh = c_bh;
4043         jl->j_start = journal->j_start;
4044         jl->j_len = journal->j_len;
4045         atomic_set(&jl->j_nonzerolen, journal->j_len);
4046         atomic_set(&jl->j_commit_left, journal->j_len + 2);
4047         jl->j_realblock = NULL;
4048
4049         /* The ENTIRE FOR LOOP MUST not cause schedule to occur.
4050          **  for each real block, add it to the journal list hash,
4051          ** copy into real block index array in the commit or desc block
4052          */
4053         trans_half = journal_trans_half(p_s_sb->s_blocksize);
4054         for (i = 0, cn = journal->j_first; cn; cn = cn->next, i++) {
4055                 if (buffer_journaled(cn->bh)) {
4056                         jl_cn = get_cnode(p_s_sb);
4057                         if (!jl_cn) {
4058                                 reiserfs_panic(p_s_sb,
4059                                                "journal-1676, get_cnode returned NULL\n");
4060                         }
4061                         if (i == 0) {
4062                                 jl->j_realblock = jl_cn;
4063                         }
4064                         jl_cn->prev = last_cn;
4065                         jl_cn->next = NULL;
4066                         if (last_cn) {
4067                                 last_cn->next = jl_cn;
4068                         }
4069                         last_cn = jl_cn;
4070                         /* make sure the block we are trying to log is not a block
4071                            of journal or reserved area */
4072
4073                         if (is_block_in_log_or_reserved_area
4074                             (p_s_sb, cn->bh->b_blocknr)) {
4075                                 reiserfs_panic(p_s_sb,
4076                                                "journal-2332: Trying to log block %lu, which is a log block\n",
4077                                                cn->bh->b_blocknr);
4078                         }
4079                         jl_cn->blocknr = cn->bh->b_blocknr;
4080                         jl_cn->state = 0;
4081                         jl_cn->sb = p_s_sb;
4082                         jl_cn->bh = cn->bh;
4083                         jl_cn->jlist = jl;
4084                         insert_journal_hash(journal->j_list_hash_table, jl_cn);
4085                         if (i < trans_half) {
4086                                 desc->j_realblock[i] =
4087                                     cpu_to_le32(cn->bh->b_blocknr);
4088                         } else {
4089                                 commit->j_realblock[i - trans_half] =
4090                                     cpu_to_le32(cn->bh->b_blocknr);
4091                         }
4092                 } else {
4093                         i--;
4094                 }
4095         }
4096         set_desc_trans_len(desc, journal->j_len);
4097         set_desc_mount_id(desc, journal->j_mount_id);
4098         set_desc_trans_id(desc, journal->j_trans_id);
4099         set_commit_trans_len(commit, journal->j_len);
4100
4101         /* special check in case all buffers in the journal were marked for not logging */
4102         BUG_ON(journal->j_len == 0);
4103
4104         /* we're about to dirty all the log blocks, mark the description block
4105          * dirty now too.  Don't mark the commit block dirty until all the
4106          * others are on disk
4107          */
4108         mark_buffer_dirty(d_bh);
4109
4110         /* first data block is j_start + 1, so add one to cur_write_start wherever you use it */
4111         cur_write_start = journal->j_start;
4112         cn = journal->j_first;
4113         jindex = 1;             /* start at one so we don't get the desc again */
4114         while (cn) {
4115                 clear_buffer_journal_new(cn->bh);
4116                 /* copy all the real blocks into log area.  dirty log blocks */
4117                 if (buffer_journaled(cn->bh)) {
4118                         struct buffer_head *tmp_bh;
4119                         char *addr;
4120                         struct page *page;
4121                         tmp_bh =
4122                             journal_getblk(p_s_sb,
4123                                            SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
4124                                            ((cur_write_start +
4125                                              jindex) %
4126                                             SB_ONDISK_JOURNAL_SIZE(p_s_sb)));
4127                         set_buffer_uptodate(tmp_bh);
4128                         page = cn->bh->b_page;
4129                         addr = kmap(page);
4130                         memcpy(tmp_bh->b_data,
4131                                addr + offset_in_page(cn->bh->b_data),
4132                                cn->bh->b_size);
4133                         kunmap(page);
4134                         mark_buffer_dirty(tmp_bh);
4135                         jindex++;
4136                         set_buffer_journal_dirty(cn->bh);
4137                         clear_buffer_journaled(cn->bh);
4138                 } else {
4139                         /* JDirty cleared sometime during transaction.  don't log this one */
4140                         reiserfs_warning(p_s_sb,
4141                                          "journal-2048: do_journal_end: BAD, buffer in journal hash, but not JDirty!");
4142                         brelse(cn->bh);
4143                 }
4144                 next = cn->next;
4145                 free_cnode(p_s_sb, cn);
4146                 cn = next;
4147                 cond_resched();
4148         }
4149
4150         /* we are done  with both the c_bh and d_bh, but
4151          ** c_bh must be written after all other commit blocks,
4152          ** so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1.
4153          */
4154
4155         journal->j_current_jl = alloc_journal_list(p_s_sb);
4156
4157         /* now it is safe to insert this transaction on the main list */
4158         list_add_tail(&jl->j_list, &journal->j_journal_list);
4159         list_add_tail(&jl->j_working_list, &journal->j_working_list);
4160         journal->j_num_work_lists++;
4161
4162         /* reset journal values for the next transaction */
4163         old_start = journal->j_start;
4164         journal->j_start =
4165             (journal->j_start + journal->j_len +
4166              2) % SB_ONDISK_JOURNAL_SIZE(p_s_sb);
4167         atomic_set(&(journal->j_wcount), 0);
4168         journal->j_bcount = 0;
4169         journal->j_last = NULL;
4170         journal->j_first = NULL;
4171         journal->j_len = 0;
4172         journal->j_trans_start_time = 0;
4173         /* check for trans_id overflow */
4174         if (++journal->j_trans_id == 0)
4175                 journal->j_trans_id = 10;
4176         journal->j_current_jl->j_trans_id = journal->j_trans_id;
4177         journal->j_must_wait = 0;
4178         journal->j_len_alloc = 0;
4179         journal->j_next_full_flush = 0;
4180         journal->j_next_async_flush = 0;
4181         init_journal_hash(p_s_sb);
4182
4183         // make sure reiserfs_add_jh sees the new current_jl before we
4184         // write out the tails
4185         smp_mb();
4186
4187         /* tail conversion targets have to hit the disk before we end the
4188          * transaction.  Otherwise a later transaction might repack the tail
4189          * before this transaction commits, leaving the data block unflushed and
4190          * clean, if we crash before the later transaction commits, the data block
4191          * is lost.
4192          */
4193         if (!list_empty(&jl->j_tail_bh_list)) {
4194                 unlock_kernel();
4195                 write_ordered_buffers(&journal->j_dirty_buffers_lock,
4196                                       journal, jl, &jl->j_tail_bh_list);
4197                 lock_kernel();
4198         }
4199         BUG_ON(!list_empty(&jl->j_tail_bh_list));
4200         mutex_unlock(&jl->j_commit_mutex);
4201
4202         /* honor the flush wishes from the caller, simple commits can
4203          ** be done outside the journal lock, they are done below
4204          **
4205          ** if we don't flush the commit list right now, we put it into
4206          ** the work queue so the people waiting on the async progress work
4207          ** queue don't wait for this proc to flush journal lists and such.
4208          */
4209         if (flush) {
4210                 flush_commit_list(p_s_sb, jl, 1);
4211                 flush_journal_list(p_s_sb, jl, 1);
4212         } else if (!(jl->j_state & LIST_COMMIT_PENDING))
4213                 queue_delayed_work(commit_wq, &journal->j_work, HZ / 10);
4214
4215         /* if the next transaction has any chance of wrapping, flush
4216          ** transactions that might get overwritten.  If any journal lists are very
4217          ** old flush them as well.
4218          */
4219       first_jl:
4220         list_for_each_safe(entry, safe, &journal->j_journal_list) {
4221                 temp_jl = JOURNAL_LIST_ENTRY(entry);
4222                 if (journal->j_start <= temp_jl->j_start) {
4223                         if ((journal->j_start + journal->j_trans_max + 1) >=
4224                             temp_jl->j_start) {
4225                                 flush_used_journal_lists(p_s_sb, temp_jl);
4226                                 goto first_jl;
4227                         } else if ((journal->j_start +
4228                                     journal->j_trans_max + 1) <
4229                                    SB_ONDISK_JOURNAL_SIZE(p_s_sb)) {
4230                                 /* if we don't cross into the next transaction and we don't
4231                                  * wrap, there is no way we can overlap any later transactions
4232                                  * break now
4233                                  */
4234                                 break;
4235                         }
4236                 } else if ((journal->j_start +
4237                             journal->j_trans_max + 1) >
4238                            SB_ONDISK_JOURNAL_SIZE(p_s_sb)) {
4239                         if (((journal->j_start + journal->j_trans_max + 1) %
4240                              SB_ONDISK_JOURNAL_SIZE(p_s_sb)) >=
4241                             temp_jl->j_start) {
4242                                 flush_used_journal_lists(p_s_sb, temp_jl);
4243                                 goto first_jl;
4244                         } else {
4245                                 /* we don't overlap anything from out start to the end of the
4246                                  * log, and our wrapped portion doesn't overlap anything at
4247                                  * the start of the log.  We can break
4248                                  */
4249                                 break;
4250                         }
4251                 }
4252         }
4253         flush_old_journal_lists(p_s_sb);
4254
4255         journal->j_current_jl->j_list_bitmap =
4256             get_list_bitmap(p_s_sb, journal->j_current_jl);
4257
4258         if (!(journal->j_current_jl->j_list_bitmap)) {
4259                 reiserfs_panic(p_s_sb,
4260                                "journal-1996: do_journal_end, could not get a list bitmap\n");
4261         }
4262
4263         atomic_set(&(journal->j_jlock), 0);
4264         unlock_journal(p_s_sb);
4265         /* wake up any body waiting to join. */
4266         clear_bit(J_WRITERS_QUEUED, &journal->j_state);
4267         wake_up(&(journal->j_join_wait));
4268
4269         if (!flush && wait_on_commit &&
4270             journal_list_still_alive(p_s_sb, commit_trans_id)) {
4271                 flush_commit_list(p_s_sb, jl, 1);
4272         }
4273       out:
4274         reiserfs_check_lock_depth(p_s_sb, "journal end2");
4275
4276         memset(th, 0, sizeof(*th));
4277         /* Re-set th->t_super, so we can properly keep track of how many
4278          * persistent transactions there are. We need to do this so if this
4279          * call is part of a failed restart_transaction, we can free it later */
4280         th->t_super = p_s_sb;
4281
4282         return journal->j_errno;
4283 }
4284
4285 static void __reiserfs_journal_abort_hard(struct super_block *sb)
4286 {
4287         struct reiserfs_journal *journal = SB_JOURNAL(sb);
4288         if (test_bit(J_ABORTED, &journal->j_state))
4289                 return;
4290
4291         printk(KERN_CRIT "REISERFS: Aborting journal for filesystem on %s\n",
4292                reiserfs_bdevname(sb));
4293
4294         sb->s_flags |= MS_RDONLY;
4295         set_bit(J_ABORTED, &journal->j_state);
4296
4297 #ifdef CONFIG_REISERFS_CHECK
4298         dump_stack();
4299 #endif
4300 }
4301
4302 static void __reiserfs_journal_abort_soft(struct super_block *sb, int errno)
4303 {
4304         struct reiserfs_journal *journal = SB_JOURNAL(sb);
4305         if (test_bit(J_ABORTED, &journal->j_state))
4306                 return;
4307
4308         if (!journal->j_errno)
4309                 journal->j_errno = errno;
4310
4311         __reiserfs_journal_abort_hard(sb);
4312 }
4313
4314 void reiserfs_journal_abort(struct super_block *sb, int errno)
4315 {
4316         __reiserfs_journal_abort_soft(sb, errno);
4317 }