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 <asm/uaccess.h>
  38 #include <asm/system.h>
  39
  40 #include <linux/time.h>
  41 #include <asm/semaphore.h>
  42
  43 #include <linux/vmalloc.h>
  44 #include <linux/reiserfs_fs.h>
  45
  46 #include <linux/kernel.h>
  47 #include <linux/errno.h>
  48 #include <linux/fcntl.h>
  49 #include <linux/stat.h>
  50 #include <linux/string.h>
  51 #include <linux/smp_lock.h>
  52 #include <linux/buffer_head.h>
  53 #include <linux/workqueue.h>
  54 #include <linux/writeback.h>
  55 #include <linux/blkdev.h>
  56
  57 /* gets a struct reiserfs_journal_list * from a list head */
  58 #define JOURNAL_LIST_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
  59                                j_list))
  60 #define JOURNAL_WORK_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
  61                                j_working_list))
  62
  63 /* the number of mounted filesystems.  This is used to decide when to
  64 ** start and kill the commit workqueue
  65 */
  66 static int reiserfs_mounted_fs_count;
  67
  68 static struct workqueue_struct *commit_wq;
  69
  70 #define JOURNAL_TRANS_HALF 1018 /* must be correct to keep the desc and commit
  71                                    structs at 4k */
  72 #define BUFNR 64                /*read ahead */
  73
  74 /* cnode stat bits.  Move these into reiserfs_fs.h */
  75
  76 #define BLOCK_FREED 2           /* this block was freed, and can't be written.  */
  77 #define BLOCK_FREED_HOLDER 3    /* this block was freed during this transaction, and can't be written */
  78
  79 #define BLOCK_NEEDS_FLUSH 4     /* used in flush_journal_list */
  80 #define BLOCK_DIRTIED 5
  81
  82 /* journal list state bits */
  83 #define LIST_TOUCHED 1
  84 #define LIST_DIRTY   2
  85 #define LIST_COMMIT_PENDING  4  /* someone will commit this list */
  86
  87 /* flags for do_journal_end */
  88 #define FLUSH_ALL   1           /* flush commit and real blocks */
  89 #define COMMIT_NOW  2           /* end and commit this transaction */
  90 #define WAIT        4           /* wait for the log blocks to hit the disk */
  91
  92 static int do_journal_end(struct reiserfs_transaction_handle *,
  93                           struct super_block *, unsigned long nblocks,
  94                           int flags);
  95 static int flush_journal_list(struct super_block *s,
  96                               struct reiserfs_journal_list *jl, int flushall);
  97 static int flush_commit_list(struct super_block *s,
  98                              struct reiserfs_journal_list *jl, int flushall);
  99 static int can_dirty(struct reiserfs_journal_cnode *cn);
 100 static int journal_join(struct reiserfs_transaction_handle *th,
 101                         struct super_block *p_s_sb, unsigned long nblocks);
 102 static int release_journal_dev(struct super_block *super,
 103                                struct reiserfs_journal *journal);
 104 static int dirty_one_transaction(struct super_block *s,
 105                                  struct reiserfs_journal_list *jl);
 106 static void flush_async_commits(void *p);
 107 static void queue_log_writer(struct super_block *s);
 108
 109 /* values for join in do_journal_begin_r */
 110 enum {
 111         JBEGIN_REG = 0,         /* regular journal begin */
 112         JBEGIN_JOIN = 1,        /* join the running transaction if at all possible */
 113         JBEGIN_ABORT = 2,       /* called from cleanup code, ignores aborted flag */
 114 };
 115
 116 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
 117                               struct super_block *p_s_sb,
 118                               unsigned long nblocks, int join);
 119
 120 static void init_journal_hash(struct super_block *p_s_sb)
 121 {
 122         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 123         memset(journal->j_hash_table, 0,
 124                JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
 125 }
 126
 127 /*
 128 ** clears BH_Dirty and sticks the buffer on the clean list.  Called because I can't allow refile_buffer to
 129 ** make schedule happen after I've freed a block.  Look at remove_from_transaction and journal_mark_freed for
 130 ** more details.
 131 */
 132 static int reiserfs_clean_and_file_buffer(struct buffer_head *bh)
 133 {
 134         if (bh) {
 135                 clear_buffer_dirty(bh);
 136                 clear_buffer_journal_test(bh);
 137         }
 138         return 0;
 139 }
 140
 141 static void disable_barrier(struct super_block *s)
 142 {
 143         REISERFS_SB(s)->s_mount_opt &= ~(1 << REISERFS_BARRIER_FLUSH);
 144         printk("reiserfs: disabling flush barriers on %s\n",
 145                reiserfs_bdevname(s));
 146 }
 147
 148 static struct reiserfs_bitmap_node *allocate_bitmap_node(struct super_block
 149                                                          *p_s_sb)
 150 {
 151         struct reiserfs_bitmap_node *bn;
 152         static int id;
 153
 154         bn = kmalloc(sizeof(struct reiserfs_bitmap_node), GFP_NOFS);
 155         if (!bn) {
 156                 return NULL;
 157         }
 158         bn->data = kzalloc(p_s_sb->s_blocksize, GFP_NOFS);
 159         if (!bn->data) {
 160                 kfree(bn);
 161                 return NULL;
 162         }
 163         bn->id = id++;
 164         INIT_LIST_HEAD(&bn->list);
 165         return bn;
 166 }
 167
 168 static struct reiserfs_bitmap_node *get_bitmap_node(struct super_block *p_s_sb)
 169 {
 170         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 171         struct reiserfs_bitmap_node *bn = NULL;
 172         struct list_head *entry = journal->j_bitmap_nodes.next;
 173
 174         journal->j_used_bitmap_nodes++;
 175       repeat:
 176
 177         if (entry != &journal->j_bitmap_nodes) {
 178                 bn = list_entry(entry, struct reiserfs_bitmap_node, list);
 179                 list_del(entry);
 180                 memset(bn->data, 0, p_s_sb->s_blocksize);
 181                 journal->j_free_bitmap_nodes--;
 182                 return bn;
 183         }
 184         bn = allocate_bitmap_node(p_s_sb);
 185         if (!bn) {
 186                 yield();
 187                 goto repeat;
 188         }
 189         return bn;
 190 }
 191 static inline void free_bitmap_node(struct super_block *p_s_sb,
 192                                     struct reiserfs_bitmap_node *bn)
 193 {
 194         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 195         journal->j_used_bitmap_nodes--;
 196         if (journal->j_free_bitmap_nodes > REISERFS_MAX_BITMAP_NODES) {
 197                 kfree(bn->data);
 198                 kfree(bn);
 199         } else {
 200                 list_add(&bn->list, &journal->j_bitmap_nodes);
 201                 journal->j_free_bitmap_nodes++;
 202         }
 203 }
 204
 205 static void allocate_bitmap_nodes(struct super_block *p_s_sb)
 206 {
 207         int i;
 208         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 209         struct reiserfs_bitmap_node *bn = NULL;
 210         for (i = 0; i < REISERFS_MIN_BITMAP_NODES; i++) {
 211                 bn = allocate_bitmap_node(p_s_sb);
 212                 if (bn) {
 213                         list_add(&bn->list, &journal->j_bitmap_nodes);
 214                         journal->j_free_bitmap_nodes++;
 215                 } else {
 216                         break;  // this is ok, we'll try again when more are needed
 217                 }
 218         }
 219 }
 220
 221 static int set_bit_in_list_bitmap(struct super_block *p_s_sb, int block,
 222                                   struct reiserfs_list_bitmap *jb)
 223 {
 224         int bmap_nr = block / (p_s_sb->s_blocksize << 3);
 225         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 < SB_BMAP_NR(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                                    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                         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         down(&SB_JOURNAL(p_s_sb)->j_lock);
 560 }
 561
 562 /* unlock the current transaction */
 563 static inline void unlock_journal(struct super_block *p_s_sb)
 564 {
 565         up(&SB_JOURNAL(p_s_sb)->j_lock);
 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 %lu, 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 long 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 static void reiserfs_end_buffer_io_sync(struct buffer_head *bh, int uptodate)
 618 {
 619         char b[BDEVNAME_SIZE];
 620
 621         if (buffer_journaled(bh)) {
 622                 reiserfs_warning(NULL,
 623                                  "clm-2084: pinned buffer %lu:%s sent to disk",
 624                                  bh->b_blocknr, bdevname(bh->b_bdev, b));
 625         }
 626         if (uptodate)
 627                 set_buffer_uptodate(bh);
 628         else
 629                 clear_buffer_uptodate(bh);
 630         unlock_buffer(bh);
 631         put_bh(bh);
 632 }
 633
 634 static void reiserfs_end_ordered_io(struct buffer_head *bh, int uptodate)
 635 {
 636         if (uptodate)
 637                 set_buffer_uptodate(bh);
 638         else
 639                 clear_buffer_uptodate(bh);
 640         unlock_buffer(bh);
 641         put_bh(bh);
 642 }
 643
 644 static void submit_logged_buffer(struct buffer_head *bh)
 645 {
 646         get_bh(bh);
 647         bh->b_end_io = reiserfs_end_buffer_io_sync;
 648         clear_buffer_journal_new(bh);
 649         clear_buffer_dirty(bh);
 650         if (!test_clear_buffer_journal_test(bh))
 651                 BUG();
 652         if (!buffer_uptodate(bh))
 653                 BUG();
 654         submit_bh(WRITE, bh);
 655 }
 656
 657 static void submit_ordered_buffer(struct buffer_head *bh)
 658 {
 659         get_bh(bh);
 660         bh->b_end_io = reiserfs_end_ordered_io;
 661         clear_buffer_dirty(bh);
 662         if (!buffer_uptodate(bh))
 663                 BUG();
 664         submit_bh(WRITE, bh);
 665 }
 666
 667 static int submit_barrier_buffer(struct buffer_head *bh)
 668 {
 669         get_bh(bh);
 670         bh->b_end_io = reiserfs_end_ordered_io;
 671         clear_buffer_dirty(bh);
 672         if (!buffer_uptodate(bh))
 673                 BUG();
 674         return submit_bh(WRITE_BARRIER, bh);
 675 }
 676
 677 static void check_barrier_completion(struct super_block *s,
 678                                      struct buffer_head *bh)
 679 {
 680         if (buffer_eopnotsupp(bh)) {
 681                 clear_buffer_eopnotsupp(bh);
 682                 disable_barrier(s);
 683                 set_buffer_uptodate(bh);
 684                 set_buffer_dirty(bh);
 685                 sync_dirty_buffer(bh);
 686         }
 687 }
 688
 689 #define CHUNK_SIZE 32
 690 struct buffer_chunk {
 691         struct buffer_head *bh[CHUNK_SIZE];
 692         int nr;
 693 };
 694
 695 static void write_chunk(struct buffer_chunk *chunk)
 696 {
 697         int i;
 698         get_fs_excl();
 699         for (i = 0; i < chunk->nr; i++) {
 700                 submit_logged_buffer(chunk->bh[i]);
 701         }
 702         chunk->nr = 0;
 703         put_fs_excl();
 704 }
 705
 706 static void write_ordered_chunk(struct buffer_chunk *chunk)
 707 {
 708         int i;
 709         get_fs_excl();
 710         for (i = 0; i < chunk->nr; i++) {
 711                 submit_ordered_buffer(chunk->bh[i]);
 712         }
 713         chunk->nr = 0;
 714         put_fs_excl();
 715 }
 716
 717 static int add_to_chunk(struct buffer_chunk *chunk, struct buffer_head *bh,
 718                         spinlock_t * lock, void (fn) (struct buffer_chunk *))
 719 {
 720         int ret = 0;
 721         if (chunk->nr >= CHUNK_SIZE)
 722                 BUG();
 723         chunk->bh[chunk->nr++] = bh;
 724         if (chunk->nr >= CHUNK_SIZE) {
 725                 ret = 1;
 726                 if (lock)
 727                         spin_unlock(lock);
 728                 fn(chunk);
 729                 if (lock)
 730                         spin_lock(lock);
 731         }
 732         return ret;
 733 }
 734
 735 static atomic_t nr_reiserfs_jh = ATOMIC_INIT(0);
 736 static struct reiserfs_jh *alloc_jh(void)
 737 {
 738         struct reiserfs_jh *jh;
 739         while (1) {
 740                 jh = kmalloc(sizeof(*jh), GFP_NOFS);
 741                 if (jh) {
 742                         atomic_inc(&nr_reiserfs_jh);
 743                         return jh;
 744                 }
 745                 yield();
 746         }
 747 }
 748
 749 /*
 750  * we want to free the jh when the buffer has been written
 751  * and waited on
 752  */
 753 void reiserfs_free_jh(struct buffer_head *bh)
 754 {
 755         struct reiserfs_jh *jh;
 756
 757         jh = bh->b_private;
 758         if (jh) {
 759                 bh->b_private = NULL;
 760                 jh->bh = NULL;
 761                 list_del_init(&jh->list);
 762                 kfree(jh);
 763                 if (atomic_read(&nr_reiserfs_jh) <= 0)
 764                         BUG();
 765                 atomic_dec(&nr_reiserfs_jh);
 766                 put_bh(bh);
 767         }
 768 }
 769
 770 static inline int __add_jh(struct reiserfs_journal *j, struct buffer_head *bh,
 771                            int tail)
 772 {
 773         struct reiserfs_jh *jh;
 774
 775         if (bh->b_private) {
 776                 spin_lock(&j->j_dirty_buffers_lock);
 777                 if (!bh->b_private) {
 778                         spin_unlock(&j->j_dirty_buffers_lock);
 779                         goto no_jh;
 780                 }
 781                 jh = bh->b_private;
 782                 list_del_init(&jh->list);
 783         } else {
 784               no_jh:
 785                 get_bh(bh);
 786                 jh = alloc_jh();
 787                 spin_lock(&j->j_dirty_buffers_lock);
 788                 /* buffer must be locked for __add_jh, should be able to have
 789                  * two adds at the same time
 790                  */
 791                 if (bh->b_private)
 792                         BUG();
 793                 jh->bh = bh;
 794                 bh->b_private = jh;
 795         }
 796         jh->jl = j->j_current_jl;
 797         if (tail)
 798                 list_add_tail(&jh->list, &jh->jl->j_tail_bh_list);
 799         else {
 800                 list_add_tail(&jh->list, &jh->jl->j_bh_list);
 801         }
 802         spin_unlock(&j->j_dirty_buffers_lock);
 803         return 0;
 804 }
 805
 806 int reiserfs_add_tail_list(struct inode *inode, struct buffer_head *bh)
 807 {
 808         return __add_jh(SB_JOURNAL(inode->i_sb), bh, 1);
 809 }
 810 int reiserfs_add_ordered_list(struct inode *inode, struct buffer_head *bh)
 811 {
 812         return __add_jh(SB_JOURNAL(inode->i_sb), bh, 0);
 813 }
 814
 815 #define JH_ENTRY(l) list_entry((l), struct reiserfs_jh, list)
 816 static int write_ordered_buffers(spinlock_t * lock,
 817                                  struct reiserfs_journal *j,
 818                                  struct reiserfs_journal_list *jl,
 819                                  struct list_head *list)
 820 {
 821         struct buffer_head *bh;
 822         struct reiserfs_jh *jh;
 823         int ret = j->j_errno;
 824         struct buffer_chunk chunk;
 825         struct list_head tmp;
 826         INIT_LIST_HEAD(&tmp);
 827
 828         chunk.nr = 0;
 829         spin_lock(lock);
 830         while (!list_empty(list)) {
 831                 jh = JH_ENTRY(list->next);
 832                 bh = jh->bh;
 833                 get_bh(bh);
 834                 if (test_set_buffer_locked(bh)) {
 835                         if (!buffer_dirty(bh)) {
 836                                 list_move(&jh->list, &tmp);
 837                                 goto loop_next;
 838                         }
 839                         spin_unlock(lock);
 840                         if (chunk.nr)
 841                                 write_ordered_chunk(&chunk);
 842                         wait_on_buffer(bh);
 843                         cond_resched();
 844                         spin_lock(lock);
 845                         goto loop_next;
 846                 }
 847                 /* in theory, dirty non-uptodate buffers should never get here,
 848                  * but the upper layer io error paths still have a few quirks.
 849                  * Handle them here as gracefully as we can
 850                  */
 851                 if (!buffer_uptodate(bh) && buffer_dirty(bh)) {
 852                         clear_buffer_dirty(bh);
 853                         ret = -EIO;
 854                 }
 855                 if (buffer_dirty(bh)) {
 856                         list_move(&jh->list, &tmp);
 857                         add_to_chunk(&chunk, bh, lock, write_ordered_chunk);
 858                 } else {
 859                         reiserfs_free_jh(bh);
 860                         unlock_buffer(bh);
 861                 }
 862               loop_next:
 863                 put_bh(bh);
 864                 cond_resched_lock(lock);
 865         }
 866         if (chunk.nr) {
 867                 spin_unlock(lock);
 868                 write_ordered_chunk(&chunk);
 869                 spin_lock(lock);
 870         }
 871         while (!list_empty(&tmp)) {
 872                 jh = JH_ENTRY(tmp.prev);
 873                 bh = jh->bh;
 874                 get_bh(bh);
 875                 reiserfs_free_jh(bh);
 876
 877                 if (buffer_locked(bh)) {
 878                         spin_unlock(lock);
 879                         wait_on_buffer(bh);
 880                         spin_lock(lock);
 881                 }
 882                 if (!buffer_uptodate(bh)) {
 883                         ret = -EIO;
 884                 }
 885                 /* ugly interaction with invalidatepage here.
 886                  * reiserfs_invalidate_page will pin any buffer that has a valid
 887                  * journal head from an older transaction.  If someone else sets
 888                  * our buffer dirty after we write it in the first loop, and
 889                  * then someone truncates the page away, nobody will ever write
 890                  * the buffer. We're safe if we write the page one last time
 891                  * after freeing the journal header.
 892                  */
 893                 if (buffer_dirty(bh) && unlikely(bh->b_page->mapping == NULL)) {
 894                         spin_unlock(lock);
 895                         ll_rw_block(WRITE, 1, &bh);
 896                         spin_lock(lock);
 897                 }
 898                 put_bh(bh);
 899                 cond_resched_lock(lock);
 900         }
 901         spin_unlock(lock);
 902         return ret;
 903 }
 904
 905 static int flush_older_commits(struct super_block *s,
 906                                struct reiserfs_journal_list *jl)
 907 {
 908         struct reiserfs_journal *journal = SB_JOURNAL(s);
 909         struct reiserfs_journal_list *other_jl;
 910         struct reiserfs_journal_list *first_jl;
 911         struct list_head *entry;
 912         unsigned long trans_id = jl->j_trans_id;
 913         unsigned long other_trans_id;
 914         unsigned long first_trans_id;
 915
 916       find_first:
 917         /*
 918          * first we walk backwards to find the oldest uncommitted transation
 919          */
 920         first_jl = jl;
 921         entry = jl->j_list.prev;
 922         while (1) {
 923                 other_jl = JOURNAL_LIST_ENTRY(entry);
 924                 if (entry == &journal->j_journal_list ||
 925                     atomic_read(&other_jl->j_older_commits_done))
 926                         break;
 927
 928                 first_jl = other_jl;
 929                 entry = other_jl->j_list.prev;
 930         }
 931
 932         /* if we didn't find any older uncommitted transactions, return now */
 933         if (first_jl == jl) {
 934                 return 0;
 935         }
 936
 937         first_trans_id = first_jl->j_trans_id;
 938
 939         entry = &first_jl->j_list;
 940         while (1) {
 941                 other_jl = JOURNAL_LIST_ENTRY(entry);
 942                 other_trans_id = other_jl->j_trans_id;
 943
 944                 if (other_trans_id < trans_id) {
 945                         if (atomic_read(&other_jl->j_commit_left) != 0) {
 946                                 flush_commit_list(s, other_jl, 0);
 947
 948                                 /* list we were called with is gone, return */
 949                                 if (!journal_list_still_alive(s, trans_id))
 950                                         return 1;
 951
 952                                 /* the one we just flushed is gone, this means all
 953                                  * older lists are also gone, so first_jl is no longer
 954                                  * valid either.  Go back to the beginning.
 955                                  */
 956                                 if (!journal_list_still_alive
 957                                     (s, other_trans_id)) {
 958                                         goto find_first;
 959                                 }
 960                         }
 961                         entry = entry->next;
 962                         if (entry == &journal->j_journal_list)
 963                                 return 0;
 964                 } else {
 965                         return 0;
 966                 }
 967         }
 968         return 0;
 969 }
 970 int reiserfs_async_progress_wait(struct super_block *s)
 971 {
 972         DEFINE_WAIT(wait);
 973         struct reiserfs_journal *j = SB_JOURNAL(s);
 974         if (atomic_read(&j->j_async_throttle))
 975                 blk_congestion_wait(WRITE, HZ / 10);
 976         return 0;
 977 }
 978
 979 /*
 980 ** if this journal list still has commit blocks unflushed, send them to disk.
 981 **
 982 ** log areas must be flushed in order (transaction 2 can't commit before transaction 1)
 983 ** Before the commit block can by written, every other log block must be safely on disk
 984 **
 985 */
 986 static int flush_commit_list(struct super_block *s,
 987                              struct reiserfs_journal_list *jl, int flushall)
 988 {
 989         int i;
 990         int bn;
 991         struct buffer_head *tbh = NULL;
 992         unsigned long trans_id = jl->j_trans_id;
 993         struct reiserfs_journal *journal = SB_JOURNAL(s);
 994         int barrier = 0;
 995         int retval = 0;
 996         int write_len;
 997
 998         reiserfs_check_lock_depth(s, "flush_commit_list");
 999
1000         if (atomic_read(&jl->j_older_commits_done)) {
1001                 return 0;
1002         }
1003
1004         get_fs_excl();
1005
1006         /* before we can put our commit blocks on disk, we have to make sure everyone older than
1007          ** us is on disk too
1008          */
1009         BUG_ON(jl->j_len <= 0);
1010         BUG_ON(trans_id == journal->j_trans_id);
1011
1012         get_journal_list(jl);
1013         if (flushall) {
1014                 if (flush_older_commits(s, jl) == 1) {
1015                         /* list disappeared during flush_older_commits.  return */
1016                         goto put_jl;
1017                 }
1018         }
1019
1020         /* make sure nobody is trying to flush this one at the same time */
1021         down(&jl->j_commit_lock);
1022         if (!journal_list_still_alive(s, trans_id)) {
1023                 up(&jl->j_commit_lock);
1024                 goto put_jl;
1025         }
1026         BUG_ON(jl->j_trans_id == 0);
1027
1028         /* this commit is done, exit */
1029         if (atomic_read(&(jl->j_commit_left)) <= 0) {
1030                 if (flushall) {
1031                         atomic_set(&(jl->j_older_commits_done), 1);
1032                 }
1033                 up(&jl->j_commit_lock);
1034                 goto put_jl;
1035         }
1036
1037         if (!list_empty(&jl->j_bh_list)) {
1038                 int ret;
1039                 unlock_kernel();
1040                 ret = write_ordered_buffers(&journal->j_dirty_buffers_lock,
1041                                             journal, jl, &jl->j_bh_list);
1042                 if (ret < 0 && retval == 0)
1043                         retval = ret;
1044                 lock_kernel();
1045         }
1046         BUG_ON(!list_empty(&jl->j_bh_list));
1047         /*
1048          * for the description block and all the log blocks, submit any buffers
1049          * that haven't already reached the disk.  Try to write at least 256
1050          * log blocks. later on, we will only wait on blocks that correspond
1051          * to this transaction, but while we're unplugging we might as well
1052          * get a chunk of data on there.
1053          */
1054         atomic_inc(&journal->j_async_throttle);
1055         write_len = jl->j_len + 1;
1056         if (write_len < 256)
1057                 write_len = 256;
1058         for (i = 0 ; i < write_len ; i++) {
1059                 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + (jl->j_start + i) %
1060                     SB_ONDISK_JOURNAL_SIZE(s);
1061                 tbh = journal_find_get_block(s, bn);
1062                 if (tbh) {
1063                         if (buffer_dirty(tbh))
1064                             ll_rw_block(WRITE, 1, &tbh) ;
1065                         put_bh(tbh) ;
1066                 }
1067         }
1068         atomic_dec(&journal->j_async_throttle);
1069
1070         /* We're skipping the commit if there's an error */
1071         if (retval || reiserfs_is_journal_aborted(journal))
1072                 barrier = 0;
1073
1074         /* wait on everything written so far before writing the commit
1075          * if we are in barrier mode, send the commit down now
1076          */
1077         barrier = reiserfs_barrier_flush(s);
1078         if (barrier) {
1079                 int ret;
1080                 lock_buffer(jl->j_commit_bh);
1081                 ret = submit_barrier_buffer(jl->j_commit_bh);
1082                 if (ret == -EOPNOTSUPP) {
1083                         set_buffer_uptodate(jl->j_commit_bh);
1084                         disable_barrier(s);
1085                         barrier = 0;
1086                 }
1087         }
1088         for (i = 0; i < (jl->j_len + 1); i++) {
1089                 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) +
1090                     (jl->j_start + i) % SB_ONDISK_JOURNAL_SIZE(s);
1091                 tbh = journal_find_get_block(s, bn);
1092                 wait_on_buffer(tbh);
1093                 // since we're using ll_rw_blk above, it might have skipped over
1094                 // a locked buffer.  Double check here
1095                 //
1096                 if (buffer_dirty(tbh))  /* redundant, sync_dirty_buffer() checks */
1097                         sync_dirty_buffer(tbh);
1098                 if (unlikely(!buffer_uptodate(tbh))) {
1099 #ifdef CONFIG_REISERFS_CHECK
1100                         reiserfs_warning(s, "journal-601, buffer write failed");
1101 #endif
1102                         retval = -EIO;
1103                 }
1104                 put_bh(tbh);    /* once for journal_find_get_block */
1105                 put_bh(tbh);    /* once due to original getblk in do_journal_end */
1106                 atomic_dec(&(jl->j_commit_left));
1107         }
1108
1109         BUG_ON(atomic_read(&(jl->j_commit_left)) != 1);
1110
1111         if (!barrier) {
1112                 /* If there was a write error in the journal - we can't commit
1113                  * this transaction - it will be invalid and, if successful,
1114                  * will just end up propogating the write error out to
1115                  * the file system. */
1116                 if (likely(!retval && !reiserfs_is_journal_aborted (journal))) {
1117                         if (buffer_dirty(jl->j_commit_bh))
1118                                 BUG();
1119                         mark_buffer_dirty(jl->j_commit_bh) ;
1120                         sync_dirty_buffer(jl->j_commit_bh) ;
1121                 }
1122         } else
1123                 wait_on_buffer(jl->j_commit_bh);
1124
1125         check_barrier_completion(s, jl->j_commit_bh);
1126
1127         /* If there was a write error in the journal - we can't commit this
1128          * transaction - it will be invalid and, if successful, will just end
1129          * up propogating the write error out to the filesystem. */
1130         if (unlikely(!buffer_uptodate(jl->j_commit_bh))) {
1131 #ifdef CONFIG_REISERFS_CHECK
1132                 reiserfs_warning(s, "journal-615: buffer write failed");
1133 #endif
1134                 retval = -EIO;
1135         }
1136         bforget(jl->j_commit_bh);
1137         if (journal->j_last_commit_id != 0 &&
1138             (jl->j_trans_id - journal->j_last_commit_id) != 1) {
1139                 reiserfs_warning(s, "clm-2200: last commit %lu, current %lu",
1140                                  journal->j_last_commit_id, jl->j_trans_id);
1141         }
1142         journal->j_last_commit_id = jl->j_trans_id;
1143
1144         /* now, every commit block is on the disk.  It is safe to allow blocks freed during this transaction to be reallocated */
1145         cleanup_freed_for_journal_list(s, jl);
1146
1147         retval = retval ? retval : journal->j_errno;
1148
1149         /* mark the metadata dirty */
1150         if (!retval)
1151                 dirty_one_transaction(s, jl);
1152         atomic_dec(&(jl->j_commit_left));
1153
1154         if (flushall) {
1155                 atomic_set(&(jl->j_older_commits_done), 1);
1156         }
1157         up(&jl->j_commit_lock);
1158       put_jl:
1159         put_journal_list(s, jl);
1160
1161         if (retval)
1162                 reiserfs_abort(s, retval, "Journal write error in %s",
1163                                __FUNCTION__);
1164         put_fs_excl();
1165         return retval;
1166 }
1167
1168 /*
1169 ** flush_journal_list frequently needs to find a newer transaction for a given block.  This does that, or
1170 ** returns NULL if it can't find anything
1171 */
1172 static struct reiserfs_journal_list *find_newer_jl_for_cn(struct
1173                                                           reiserfs_journal_cnode
1174                                                           *cn)
1175 {
1176         struct super_block *sb = cn->sb;
1177         b_blocknr_t blocknr = cn->blocknr;
1178
1179         cn = cn->hprev;
1180         while (cn) {
1181                 if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist) {
1182                         return cn->jlist;
1183                 }
1184                 cn = cn->hprev;
1185         }
1186         return NULL;
1187 }
1188
1189 static int newer_jl_done(struct reiserfs_journal_cnode *cn)
1190 {
1191         struct super_block *sb = cn->sb;
1192         b_blocknr_t blocknr = cn->blocknr;
1193
1194         cn = cn->hprev;
1195         while (cn) {
1196                 if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist &&
1197                     atomic_read(&cn->jlist->j_commit_left) != 0)
1198                                     return 0;
1199                 cn = cn->hprev;
1200         }
1201         return 1;
1202 }
1203
1204 static void remove_journal_hash(struct super_block *,
1205                                 struct reiserfs_journal_cnode **,
1206                                 struct reiserfs_journal_list *, unsigned long,
1207                                 int);
1208
1209 /*
1210 ** once all the real blocks have been flushed, it is safe to remove them from the
1211 ** journal list for this transaction.  Aside from freeing the cnode, this also allows the
1212 ** block to be reallocated for data blocks if it had been deleted.
1213 */
1214 static void remove_all_from_journal_list(struct super_block *p_s_sb,
1215                                          struct reiserfs_journal_list *jl,
1216                                          int debug)
1217 {
1218         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1219         struct reiserfs_journal_cnode *cn, *last;
1220         cn = jl->j_realblock;
1221
1222         /* which is better, to lock once around the whole loop, or
1223          ** to lock for each call to remove_journal_hash?
1224          */
1225         while (cn) {
1226                 if (cn->blocknr != 0) {
1227                         if (debug) {
1228                                 reiserfs_warning(p_s_sb,
1229                                                  "block %u, bh is %d, state %ld",
1230                                                  cn->blocknr, cn->bh ? 1 : 0,
1231                                                  cn->state);
1232                         }
1233                         cn->state = 0;
1234                         remove_journal_hash(p_s_sb, journal->j_list_hash_table,
1235                                             jl, cn->blocknr, 1);
1236                 }
1237                 last = cn;
1238                 cn = cn->next;
1239                 free_cnode(p_s_sb, last);
1240         }
1241         jl->j_realblock = NULL;
1242 }
1243
1244 /*
1245 ** if this timestamp is greater than the timestamp we wrote last to the header block, write it to the header block.
1246 ** once this is done, I can safely say the log area for this transaction won't ever be replayed, and I can start
1247 ** releasing blocks in this transaction for reuse as data blocks.
1248 ** called by flush_journal_list, before it calls remove_all_from_journal_list
1249 **
1250 */
1251 static int _update_journal_header_block(struct super_block *p_s_sb,
1252                                         unsigned long offset,
1253                                         unsigned long trans_id)
1254 {
1255         struct reiserfs_journal_header *jh;
1256         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1257
1258         if (reiserfs_is_journal_aborted(journal))
1259                 return -EIO;
1260
1261         if (trans_id >= journal->j_last_flush_trans_id) {
1262                 if (buffer_locked((journal->j_header_bh))) {
1263                         wait_on_buffer((journal->j_header_bh));
1264                         if (unlikely(!buffer_uptodate(journal->j_header_bh))) {
1265 #ifdef CONFIG_REISERFS_CHECK
1266                                 reiserfs_warning(p_s_sb,
1267                                                  "journal-699: buffer write failed");
1268 #endif
1269                                 return -EIO;
1270                         }
1271                 }
1272                 journal->j_last_flush_trans_id = trans_id;
1273                 journal->j_first_unflushed_offset = offset;
1274                 jh = (struct reiserfs_journal_header *)(journal->j_header_bh->
1275                                                         b_data);
1276                 jh->j_last_flush_trans_id = cpu_to_le32(trans_id);
1277                 jh->j_first_unflushed_offset = cpu_to_le32(offset);
1278                 jh->j_mount_id = cpu_to_le32(journal->j_mount_id);
1279
1280                 if (reiserfs_barrier_flush(p_s_sb)) {
1281                         int ret;
1282                         lock_buffer(journal->j_header_bh);
1283                         ret = submit_barrier_buffer(journal->j_header_bh);
1284                         if (ret == -EOPNOTSUPP) {
1285                                 set_buffer_uptodate(journal->j_header_bh);
1286                                 disable_barrier(p_s_sb);
1287                                 goto sync;
1288                         }
1289                         wait_on_buffer(journal->j_header_bh);
1290                         check_barrier_completion(p_s_sb, journal->j_header_bh);
1291                 } else {
1292                       sync:
1293                         set_buffer_dirty(journal->j_header_bh);
1294                         sync_dirty_buffer(journal->j_header_bh);
1295                 }
1296                 if (!buffer_uptodate(journal->j_header_bh)) {
1297                         reiserfs_warning(p_s_sb,
1298                                          "journal-837: IO error during journal replay");
1299                         return -EIO;
1300                 }
1301         }
1302         return 0;
1303 }
1304
1305 static int update_journal_header_block(struct super_block *p_s_sb,
1306                                        unsigned long offset,
1307                                        unsigned long trans_id)
1308 {
1309         return _update_journal_header_block(p_s_sb, offset, trans_id);
1310 }
1311
1312 /*
1313 ** flush any and all journal lists older than you are
1314 ** can only be called from flush_journal_list
1315 */
1316 static int flush_older_journal_lists(struct super_block *p_s_sb,
1317                                      struct reiserfs_journal_list *jl)
1318 {
1319         struct list_head *entry;
1320         struct reiserfs_journal_list *other_jl;
1321         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1322         unsigned long trans_id = jl->j_trans_id;
1323
1324         /* we know we are the only ones flushing things, no extra race
1325          * protection is required.
1326          */
1327       restart:
1328         entry = journal->j_journal_list.next;
1329         /* Did we wrap? */
1330         if (entry == &journal->j_journal_list)
1331                 return 0;
1332         other_jl = JOURNAL_LIST_ENTRY(entry);
1333         if (other_jl->j_trans_id < trans_id) {
1334                 BUG_ON(other_jl->j_refcount <= 0);
1335                 /* do not flush all */
1336                 flush_journal_list(p_s_sb, other_jl, 0);
1337
1338                 /* other_jl is now deleted from the list */
1339                 goto restart;
1340         }
1341         return 0;
1342 }
1343
1344 static void del_from_work_list(struct super_block *s,
1345                                struct reiserfs_journal_list *jl)
1346 {
1347         struct reiserfs_journal *journal = SB_JOURNAL(s);
1348         if (!list_empty(&jl->j_working_list)) {
1349                 list_del_init(&jl->j_working_list);
1350                 journal->j_num_work_lists--;
1351         }
1352 }
1353
1354 /* flush a journal list, both commit and real blocks
1355 **
1356 ** always set flushall to 1, unless you are calling from inside
1357 ** flush_journal_list
1358 **
1359 ** IMPORTANT.  This can only be called while there are no journal writers,
1360 ** and the journal is locked.  That means it can only be called from
1361 ** do_journal_end, or by journal_release
1362 */
1363 static int flush_journal_list(struct super_block *s,
1364                               struct reiserfs_journal_list *jl, int flushall)
1365 {
1366         struct reiserfs_journal_list *pjl;
1367         struct reiserfs_journal_cnode *cn, *last;
1368         int count;
1369         int was_jwait = 0;
1370         int was_dirty = 0;
1371         struct buffer_head *saved_bh;
1372         unsigned long j_len_saved = jl->j_len;
1373         struct reiserfs_journal *journal = SB_JOURNAL(s);
1374         int err = 0;
1375
1376         BUG_ON(j_len_saved <= 0);
1377
1378         if (atomic_read(&journal->j_wcount) != 0) {
1379                 reiserfs_warning(s,
1380                                  "clm-2048: flush_journal_list called with wcount %d",
1381                                  atomic_read(&journal->j_wcount));
1382         }
1383         BUG_ON(jl->j_trans_id == 0);
1384
1385         /* if flushall == 0, the lock is already held */
1386         if (flushall) {
1387                 down(&journal->j_flush_sem);
1388         } else if (!down_trylock(&journal->j_flush_sem)) {
1389                 BUG();
1390         }
1391
1392         count = 0;
1393         if (j_len_saved > journal->j_trans_max) {
1394                 reiserfs_panic(s,
1395                                "journal-715: flush_journal_list, length is %lu, trans id %lu\n",
1396                                j_len_saved, jl->j_trans_id);
1397                 return 0;
1398         }
1399
1400         get_fs_excl();
1401
1402         /* if all the work is already done, get out of here */
1403         if (atomic_read(&(jl->j_nonzerolen)) <= 0 &&
1404             atomic_read(&(jl->j_commit_left)) <= 0) {
1405                 goto flush_older_and_return;
1406         }
1407
1408         /* start by putting the commit list on disk.  This will also flush
1409          ** the commit lists of any olders transactions
1410          */
1411         flush_commit_list(s, jl, 1);
1412
1413         if (!(jl->j_state & LIST_DIRTY)
1414             && !reiserfs_is_journal_aborted(journal))
1415                 BUG();
1416
1417         /* are we done now? */
1418         if (atomic_read(&(jl->j_nonzerolen)) <= 0 &&
1419             atomic_read(&(jl->j_commit_left)) <= 0) {
1420                 goto flush_older_and_return;
1421         }
1422
1423         /* loop through each cnode, see if we need to write it,
1424          ** or wait on a more recent transaction, or just ignore it
1425          */
1426         if (atomic_read(&(journal->j_wcount)) != 0) {
1427                 reiserfs_panic(s,
1428                                "journal-844: panic journal list is flushing, wcount is not 0\n");
1429         }
1430         cn = jl->j_realblock;
1431         while (cn) {
1432                 was_jwait = 0;
1433                 was_dirty = 0;
1434                 saved_bh = NULL;
1435                 /* blocknr of 0 is no longer in the hash, ignore it */
1436                 if (cn->blocknr == 0) {
1437                         goto free_cnode;
1438                 }
1439
1440                 /* This transaction failed commit. Don't write out to the disk */
1441                 if (!(jl->j_state & LIST_DIRTY))
1442                         goto free_cnode;
1443
1444                 pjl = find_newer_jl_for_cn(cn);
1445                 /* the order is important here.  We check pjl to make sure we
1446                  ** don't clear BH_JDirty_wait if we aren't the one writing this
1447                  ** block to disk
1448                  */
1449                 if (!pjl && cn->bh) {
1450                         saved_bh = cn->bh;
1451
1452                         /* we do this to make sure nobody releases the buffer while
1453                          ** we are working with it
1454                          */
1455                         get_bh(saved_bh);
1456
1457                         if (buffer_journal_dirty(saved_bh)) {
1458                                 BUG_ON(!can_dirty(cn));
1459                                 was_jwait = 1;
1460                                 was_dirty = 1;
1461                         } else if (can_dirty(cn)) {
1462                                 /* everything with !pjl && jwait should be writable */
1463                                 BUG();
1464                         }
1465                 }
1466
1467                 /* if someone has this block in a newer transaction, just make
1468                  ** sure they are commited, and don't try writing it to disk
1469                  */
1470                 if (pjl) {
1471                         if (atomic_read(&pjl->j_commit_left))
1472                                 flush_commit_list(s, pjl, 1);
1473                         goto free_cnode;
1474                 }
1475
1476                 /* bh == NULL when the block got to disk on its own, OR,
1477                  ** the block got freed in a future transaction
1478                  */
1479                 if (saved_bh == NULL) {
1480                         goto free_cnode;
1481                 }
1482
1483                 /* this should never happen.  kupdate_one_transaction has this list
1484                  ** locked while it works, so we should never see a buffer here that
1485                  ** is not marked JDirty_wait
1486                  */
1487                 if ((!was_jwait) && !buffer_locked(saved_bh)) {
1488                         reiserfs_warning(s,
1489                                          "journal-813: BAD! buffer %llu %cdirty %cjwait, "
1490                                          "not in a newer tranasction",
1491                                          (unsigned long long)saved_bh->
1492                                          b_blocknr, was_dirty ? ' ' : '!',
1493                                          was_jwait ? ' ' : '!');
1494                 }
1495                 if (was_dirty) {
1496                         /* we inc again because saved_bh gets decremented at free_cnode */
1497                         get_bh(saved_bh);
1498                         set_bit(BLOCK_NEEDS_FLUSH, &cn->state);
1499                         lock_buffer(saved_bh);
1500                         BUG_ON(cn->blocknr != saved_bh->b_blocknr);
1501                         if (buffer_dirty(saved_bh))
1502                                 submit_logged_buffer(saved_bh);
1503                         else
1504                                 unlock_buffer(saved_bh);
1505                         count++;
1506                 } else {
1507                         reiserfs_warning(s,
1508                                          "clm-2082: Unable to flush buffer %llu in %s",
1509                                          (unsigned long long)saved_bh->
1510                                          b_blocknr, __FUNCTION__);
1511                 }
1512               free_cnode:
1513                 last = cn;
1514                 cn = cn->next;
1515                 if (saved_bh) {
1516                         /* we incremented this to keep others from taking the buffer head away */
1517                         put_bh(saved_bh);
1518                         if (atomic_read(&(saved_bh->b_count)) < 0) {
1519                                 reiserfs_warning(s,
1520                                                  "journal-945: saved_bh->b_count < 0");
1521                         }
1522                 }
1523         }
1524         if (count > 0) {
1525                 cn = jl->j_realblock;
1526                 while (cn) {
1527                         if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) {
1528                                 if (!cn->bh) {
1529                                         reiserfs_panic(s,
1530                                                        "journal-1011: cn->bh is NULL\n");
1531                                 }
1532                                 wait_on_buffer(cn->bh);
1533                                 if (!cn->bh) {
1534                                         reiserfs_panic(s,
1535                                                        "journal-1012: cn->bh is NULL\n");
1536                                 }
1537                                 if (unlikely(!buffer_uptodate(cn->bh))) {
1538 #ifdef CONFIG_REISERFS_CHECK
1539                                         reiserfs_warning(s,
1540                                                          "journal-949: buffer write failed\n");
1541 #endif
1542                                         err = -EIO;
1543                                 }
1544                                 /* note, we must clear the JDirty_wait bit after the up to date
1545                                  ** check, otherwise we race against our flushpage routine
1546                                  */
1547                                 BUG_ON(!test_clear_buffer_journal_dirty
1548                                        (cn->bh));
1549
1550                                 /* undo the inc from journal_mark_dirty */
1551                                 put_bh(cn->bh);
1552                                 brelse(cn->bh);
1553                         }
1554                         cn = cn->next;
1555                 }
1556         }
1557
1558         if (err)
1559                 reiserfs_abort(s, -EIO,
1560                                "Write error while pushing transaction to disk in %s",
1561                                __FUNCTION__);
1562       flush_older_and_return:
1563
1564         /* before we can update the journal header block, we _must_ flush all
1565          ** real blocks from all older transactions to disk.  This is because
1566          ** once the header block is updated, this transaction will not be
1567          ** replayed after a crash
1568          */
1569         if (flushall) {
1570                 flush_older_journal_lists(s, jl);
1571         }
1572
1573         err = journal->j_errno;
1574         /* before we can remove everything from the hash tables for this
1575          ** transaction, we must make sure it can never be replayed
1576          **
1577          ** since we are only called from do_journal_end, we know for sure there
1578          ** are no allocations going on while we are flushing journal lists.  So,
1579          ** we only need to update the journal header block for the last list
1580          ** being flushed
1581          */
1582         if (!err && flushall) {
1583                 err =
1584                     update_journal_header_block(s,
1585                                                 (jl->j_start + jl->j_len +
1586                                                  2) % SB_ONDISK_JOURNAL_SIZE(s),
1587                                                 jl->j_trans_id);
1588                 if (err)
1589                         reiserfs_abort(s, -EIO,
1590                                        "Write error while updating journal header in %s",
1591                                        __FUNCTION__);
1592         }
1593         remove_all_from_journal_list(s, jl, 0);
1594         list_del_init(&jl->j_list);
1595         journal->j_num_lists--;
1596         del_from_work_list(s, jl);
1597
1598         if (journal->j_last_flush_id != 0 &&
1599             (jl->j_trans_id - journal->j_last_flush_id) != 1) {
1600                 reiserfs_warning(s, "clm-2201: last flush %lu, current %lu",
1601                                  journal->j_last_flush_id, jl->j_trans_id);
1602         }
1603         journal->j_last_flush_id = jl->j_trans_id;
1604
1605         /* not strictly required since we are freeing the list, but it should
1606          * help find code using dead lists later on
1607          */
1608         jl->j_len = 0;
1609         atomic_set(&(jl->j_nonzerolen), 0);
1610         jl->j_start = 0;
1611         jl->j_realblock = NULL;
1612         jl->j_commit_bh = NULL;
1613         jl->j_trans_id = 0;
1614         jl->j_state = 0;
1615         put_journal_list(s, jl);
1616         if (flushall)
1617                 up(&journal->j_flush_sem);
1618         put_fs_excl();
1619         return err;
1620 }
1621
1622 static int test_transaction(struct super_block *s,
1623                             struct reiserfs_journal_list *jl)
1624 {
1625         struct reiserfs_journal_cnode *cn;
1626
1627         if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0)
1628                 return 1;
1629
1630         cn = jl->j_realblock;
1631         while (cn) {
1632                 /* if the blocknr == 0, this has been cleared from the hash,
1633                  ** skip it
1634                  */
1635                 if (cn->blocknr == 0) {
1636                         goto next;
1637                 }
1638                 if (cn->bh && !newer_jl_done(cn))
1639                         return 0;
1640               next:
1641                 cn = cn->next;
1642                 cond_resched();
1643         }
1644         return 0;
1645 }
1646
1647 static int write_one_transaction(struct super_block *s,
1648                                  struct reiserfs_journal_list *jl,
1649                                  struct buffer_chunk *chunk)
1650 {
1651         struct reiserfs_journal_cnode *cn;
1652         int ret = 0;
1653
1654         jl->j_state |= LIST_TOUCHED;
1655         del_from_work_list(s, jl);
1656         if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0) {
1657                 return 0;
1658         }
1659
1660         cn = jl->j_realblock;
1661         while (cn) {
1662                 /* if the blocknr == 0, this has been cleared from the hash,
1663                  ** skip it
1664                  */
1665                 if (cn->blocknr == 0) {
1666                         goto next;
1667                 }
1668                 if (cn->bh && can_dirty(cn) && buffer_dirty(cn->bh)) {
1669                         struct buffer_head *tmp_bh;
1670                         /* we can race against journal_mark_freed when we try
1671                          * to lock_buffer(cn->bh), so we have to inc the buffer
1672                          * count, and recheck things after locking
1673                          */
1674                         tmp_bh = cn->bh;
1675                         get_bh(tmp_bh);
1676                         lock_buffer(tmp_bh);
1677                         if (cn->bh && can_dirty(cn) && buffer_dirty(tmp_bh)) {
1678                                 if (!buffer_journal_dirty(tmp_bh) ||
1679                                     buffer_journal_prepared(tmp_bh))
1680                                         BUG();
1681                                 add_to_chunk(chunk, tmp_bh, NULL, write_chunk);
1682                                 ret++;
1683                         } else {
1684                                 /* note, cn->bh might be null now */
1685                                 unlock_buffer(tmp_bh);
1686                         }
1687                         put_bh(tmp_bh);
1688                 }
1689               next:
1690                 cn = cn->next;
1691                 cond_resched();
1692         }
1693         return ret;
1694 }
1695
1696 /* used by flush_commit_list */
1697 static int dirty_one_transaction(struct super_block *s,
1698                                  struct reiserfs_journal_list *jl)
1699 {
1700         struct reiserfs_journal_cnode *cn;
1701         struct reiserfs_journal_list *pjl;
1702         int ret = 0;
1703
1704         jl->j_state |= LIST_DIRTY;
1705         cn = jl->j_realblock;
1706         while (cn) {
1707                 /* look for a more recent transaction that logged this
1708                  ** buffer.  Only the most recent transaction with a buffer in
1709                  ** it is allowed to send that buffer to disk
1710                  */
1711                 pjl = find_newer_jl_for_cn(cn);
1712                 if (!pjl && cn->blocknr && cn->bh
1713                     && buffer_journal_dirty(cn->bh)) {
1714                         BUG_ON(!can_dirty(cn));
1715                         /* if the buffer is prepared, it will either be logged
1716                          * or restored.  If restored, we need to make sure
1717                          * it actually gets marked dirty
1718                          */
1719                         clear_buffer_journal_new(cn->bh);
1720                         if (buffer_journal_prepared(cn->bh)) {
1721                                 set_buffer_journal_restore_dirty(cn->bh);
1722                         } else {
1723                                 set_buffer_journal_test(cn->bh);
1724                                 mark_buffer_dirty(cn->bh);
1725                         }
1726                 }
1727                 cn = cn->next;
1728         }
1729         return ret;
1730 }
1731
1732 static int kupdate_transactions(struct super_block *s,
1733                                 struct reiserfs_journal_list *jl,
1734                                 struct reiserfs_journal_list **next_jl,
1735                                 unsigned long *next_trans_id,
1736                                 int num_blocks, int num_trans)
1737 {
1738         int ret = 0;
1739         int written = 0;
1740         int transactions_flushed = 0;
1741         unsigned long orig_trans_id = jl->j_trans_id;
1742         struct buffer_chunk chunk;
1743         struct list_head *entry;
1744         struct reiserfs_journal *journal = SB_JOURNAL(s);
1745         chunk.nr = 0;
1746
1747         down(&journal->j_flush_sem);
1748         if (!journal_list_still_alive(s, orig_trans_id)) {
1749                 goto done;
1750         }
1751
1752         /* we've got j_flush_sem held, nobody is going to delete any
1753          * of these lists out from underneath us
1754          */
1755         while ((num_trans && transactions_flushed < num_trans) ||
1756                (!num_trans && written < num_blocks)) {
1757
1758                 if (jl->j_len == 0 || (jl->j_state & LIST_TOUCHED) ||
1759                     atomic_read(&jl->j_commit_left)
1760                     || !(jl->j_state & LIST_DIRTY)) {
1761                         del_from_work_list(s, jl);
1762                         break;
1763                 }
1764                 ret = write_one_transaction(s, jl, &chunk);
1765
1766                 if (ret < 0)
1767                         goto done;
1768                 transactions_flushed++;
1769                 written += ret;
1770                 entry = jl->j_list.next;
1771
1772                 /* did we wrap? */
1773                 if (entry == &journal->j_journal_list) {
1774                         break;
1775                 }
1776                 jl = JOURNAL_LIST_ENTRY(entry);
1777
1778                 /* don't bother with older transactions */
1779                 if (jl->j_trans_id <= orig_trans_id)
1780                         break;
1781         }
1782         if (chunk.nr) {
1783                 write_chunk(&chunk);
1784         }
1785
1786       done:
1787         up(&journal->j_flush_sem);
1788         return ret;
1789 }
1790
1791 /* for o_sync and fsync heavy applications, they tend to use
1792 ** all the journa list slots with tiny transactions.  These
1793 ** trigger lots and lots of calls to update the header block, which
1794 ** adds seeks and slows things down.
1795 **
1796 ** This function tries to clear out a large chunk of the journal lists
1797 ** at once, which makes everything faster since only the newest journal
1798 ** list updates the header block
1799 */
1800 static int flush_used_journal_lists(struct super_block *s,
1801                                     struct reiserfs_journal_list *jl)
1802 {
1803         unsigned long len = 0;
1804         unsigned long cur_len;
1805         int ret;
1806         int i;
1807         int limit = 256;
1808         struct reiserfs_journal_list *tjl;
1809         struct reiserfs_journal_list *flush_jl;
1810         unsigned long trans_id;
1811         struct reiserfs_journal *journal = SB_JOURNAL(s);
1812
1813         flush_jl = tjl = jl;
1814
1815         /* in data logging mode, try harder to flush a lot of blocks */
1816         if (reiserfs_data_log(s))
1817                 limit = 1024;
1818         /* flush for 256 transactions or limit blocks, whichever comes first */
1819         for (i = 0; i < 256 && len < limit; i++) {
1820                 if (atomic_read(&tjl->j_commit_left) ||
1821                     tjl->j_trans_id < jl->j_trans_id) {
1822                         break;
1823                 }
1824                 cur_len = atomic_read(&tjl->j_nonzerolen);
1825                 if (cur_len > 0) {
1826                         tjl->j_state &= ~LIST_TOUCHED;
1827                 }
1828                 len += cur_len;
1829                 flush_jl = tjl;
1830                 if (tjl->j_list.next == &journal->j_journal_list)
1831                         break;
1832                 tjl = JOURNAL_LIST_ENTRY(tjl->j_list.next);
1833         }
1834         /* try to find a group of blocks we can flush across all the
1835          ** transactions, but only bother if we've actually spanned
1836          ** across multiple lists
1837          */
1838         if (flush_jl != jl) {
1839                 ret = kupdate_transactions(s, jl, &tjl, &trans_id, len, i);
1840         }
1841         flush_journal_list(s, flush_jl, 1);
1842         return 0;
1843 }
1844
1845 /*
1846 ** removes any nodes in table with name block and dev as bh.
1847 ** only touchs the hnext and hprev pointers.
1848 */
1849 void remove_journal_hash(struct super_block *sb,
1850                          struct reiserfs_journal_cnode **table,
1851                          struct reiserfs_journal_list *jl,
1852                          unsigned long block, int remove_freed)
1853 {
1854         struct reiserfs_journal_cnode *cur;
1855         struct reiserfs_journal_cnode **head;
1856
1857         head = &(journal_hash(table, sb, block));
1858         if (!head) {
1859                 return;
1860         }
1861         cur = *head;
1862         while (cur) {
1863                 if (cur->blocknr == block && cur->sb == sb
1864                     && (jl == NULL || jl == cur->jlist)
1865                     && (!test_bit(BLOCK_FREED, &cur->state) || remove_freed)) {
1866                         if (cur->hnext) {
1867                                 cur->hnext->hprev = cur->hprev;
1868                         }
1869                         if (cur->hprev) {
1870                                 cur->hprev->hnext = cur->hnext;
1871                         } else {
1872                                 *head = cur->hnext;
1873                         }
1874                         cur->blocknr = 0;
1875                         cur->sb = NULL;
1876                         cur->state = 0;
1877                         if (cur->bh && cur->jlist)      /* anybody who clears the cur->bh will also dec the nonzerolen */
1878                                 atomic_dec(&(cur->jlist->j_nonzerolen));
1879                         cur->bh = NULL;
1880                         cur->jlist = NULL;
1881                 }
1882                 cur = cur->hnext;
1883         }
1884 }
1885
1886 static void free_journal_ram(struct super_block *p_s_sb)
1887 {
1888         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1889         kfree(journal->j_current_jl);
1890         journal->j_num_lists--;
1891
1892         vfree(journal->j_cnode_free_orig);
1893         free_list_bitmaps(p_s_sb, journal->j_list_bitmap);
1894         free_bitmap_nodes(p_s_sb);      /* must be after free_list_bitmaps */
1895         if (journal->j_header_bh) {
1896                 brelse(journal->j_header_bh);
1897         }
1898         /* j_header_bh is on the journal dev, make sure not to release the journal
1899          * dev until we brelse j_header_bh
1900          */
1901         release_journal_dev(p_s_sb, journal);
1902         vfree(journal);
1903 }
1904
1905 /*
1906 ** call on unmount.  Only set error to 1 if you haven't made your way out
1907 ** of read_super() yet.  Any other caller must keep error at 0.
1908 */
1909 static int do_journal_release(struct reiserfs_transaction_handle *th,
1910                               struct super_block *p_s_sb, int error)
1911 {
1912         struct reiserfs_transaction_handle myth;
1913         int flushed = 0;
1914         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1915
1916         /* we only want to flush out transactions if we were called with error == 0
1917          */
1918         if (!error && !(p_s_sb->s_flags & MS_RDONLY)) {
1919                 /* end the current trans */
1920                 BUG_ON(!th->t_trans_id);
1921                 do_journal_end(th, p_s_sb, 10, FLUSH_ALL);
1922
1923                 /* make sure something gets logged to force our way into the flush code */
1924                 if (!journal_join(&myth, p_s_sb, 1)) {
1925                         reiserfs_prepare_for_journal(p_s_sb,
1926                                                      SB_BUFFER_WITH_SB(p_s_sb),
1927                                                      1);
1928                         journal_mark_dirty(&myth, p_s_sb,
1929                                            SB_BUFFER_WITH_SB(p_s_sb));
1930                         do_journal_end(&myth, p_s_sb, 1, FLUSH_ALL);
1931                         flushed = 1;
1932                 }
1933         }
1934
1935         /* this also catches errors during the do_journal_end above */
1936         if (!error && reiserfs_is_journal_aborted(journal)) {
1937                 memset(&myth, 0, sizeof(myth));
1938                 if (!journal_join_abort(&myth, p_s_sb, 1)) {
1939                         reiserfs_prepare_for_journal(p_s_sb,
1940                                                      SB_BUFFER_WITH_SB(p_s_sb),
1941                                                      1);
1942                         journal_mark_dirty(&myth, p_s_sb,
1943                                            SB_BUFFER_WITH_SB(p_s_sb));
1944                         do_journal_end(&myth, p_s_sb, 1, FLUSH_ALL);
1945                 }
1946         }
1947
1948         reiserfs_mounted_fs_count--;
1949         /* wait for all commits to finish */
1950         cancel_delayed_work(&SB_JOURNAL(p_s_sb)->j_work);
1951         flush_workqueue(commit_wq);
1952         if (!reiserfs_mounted_fs_count) {
1953                 destroy_workqueue(commit_wq);
1954                 commit_wq = NULL;
1955         }
1956
1957         free_journal_ram(p_s_sb);
1958
1959         return 0;
1960 }
1961
1962 /*
1963 ** call on unmount.  flush all journal trans, release all alloc'd ram
1964 */
1965 int journal_release(struct reiserfs_transaction_handle *th,
1966                     struct super_block *p_s_sb)
1967 {
1968         return do_journal_release(th, p_s_sb, 0);
1969 }
1970
1971 /*
1972 ** only call from an error condition inside reiserfs_read_super!
1973 */
1974 int journal_release_error(struct reiserfs_transaction_handle *th,
1975                           struct super_block *p_s_sb)
1976 {
1977         return do_journal_release(th, p_s_sb, 1);
1978 }
1979
1980 /* compares description block with commit block.  returns 1 if they differ, 0 if they are the same */
1981 static int journal_compare_desc_commit(struct super_block *p_s_sb,
1982                                        struct reiserfs_journal_desc *desc,
1983                                        struct reiserfs_journal_commit *commit)
1984 {
1985         if (get_commit_trans_id(commit) != get_desc_trans_id(desc) ||
1986             get_commit_trans_len(commit) != get_desc_trans_len(desc) ||
1987             get_commit_trans_len(commit) > SB_JOURNAL(p_s_sb)->j_trans_max ||
1988             get_commit_trans_len(commit) <= 0) {
1989                 return 1;
1990         }
1991         return 0;
1992 }
1993
1994 /* returns 0 if it did not find a description block
1995 ** returns -1 if it found a corrupt commit block
1996 ** returns 1 if both desc and commit were valid
1997 */
1998 static int journal_transaction_is_valid(struct super_block *p_s_sb,
1999                                         struct buffer_head *d_bh,
2000                                         unsigned long *oldest_invalid_trans_id,
2001                                         unsigned long *newest_mount_id)
2002 {
2003         struct reiserfs_journal_desc *desc;
2004         struct reiserfs_journal_commit *commit;
2005         struct buffer_head *c_bh;
2006         unsigned long offset;
2007
2008         if (!d_bh)
2009                 return 0;
2010
2011         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2012         if (get_desc_trans_len(desc) > 0
2013             && !memcmp(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8)) {
2014                 if (oldest_invalid_trans_id && *oldest_invalid_trans_id
2015                     && get_desc_trans_id(desc) > *oldest_invalid_trans_id) {
2016                         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2017                                        "journal-986: transaction "
2018                                        "is valid returning because trans_id %d is greater than "
2019                                        "oldest_invalid %lu",
2020                                        get_desc_trans_id(desc),
2021                                        *oldest_invalid_trans_id);
2022                         return 0;
2023                 }
2024                 if (newest_mount_id
2025                     && *newest_mount_id > get_desc_mount_id(desc)) {
2026                         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2027                                        "journal-1087: transaction "
2028                                        "is valid returning because mount_id %d is less than "
2029                                        "newest_mount_id %lu",
2030                                        get_desc_mount_id(desc),
2031                                        *newest_mount_id);
2032                         return -1;
2033                 }
2034                 if (get_desc_trans_len(desc) > SB_JOURNAL(p_s_sb)->j_trans_max) {
2035                         reiserfs_warning(p_s_sb,
2036                                          "journal-2018: Bad transaction length %d encountered, ignoring transaction",
2037                                          get_desc_trans_len(desc));
2038                         return -1;
2039                 }
2040                 offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb);
2041
2042                 /* ok, we have a journal description block, lets see if the transaction was valid */
2043                 c_bh =
2044                     journal_bread(p_s_sb,
2045                                   SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2046                                   ((offset + get_desc_trans_len(desc) +
2047                                     1) % SB_ONDISK_JOURNAL_SIZE(p_s_sb)));
2048                 if (!c_bh)
2049                         return 0;
2050                 commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2051                 if (journal_compare_desc_commit(p_s_sb, desc, commit)) {
2052                         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2053                                        "journal_transaction_is_valid, commit offset %ld had bad "
2054                                        "time %d or length %d",
2055                                        c_bh->b_blocknr -
2056                                        SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2057                                        get_commit_trans_id(commit),
2058                                        get_commit_trans_len(commit));
2059                         brelse(c_bh);
2060                         if (oldest_invalid_trans_id) {
2061                                 *oldest_invalid_trans_id =
2062                                     get_desc_trans_id(desc);
2063                                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2064                                                "journal-1004: "
2065                                                "transaction_is_valid setting oldest invalid trans_id "
2066                                                "to %d",
2067                                                get_desc_trans_id(desc));
2068                         }
2069                         return -1;
2070                 }
2071                 brelse(c_bh);
2072                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2073                                "journal-1006: found valid "
2074                                "transaction start offset %llu, len %d id %d",
2075                                d_bh->b_blocknr -
2076                                SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2077                                get_desc_trans_len(desc),
2078                                get_desc_trans_id(desc));
2079                 return 1;
2080         } else {
2081                 return 0;
2082         }
2083 }
2084
2085 static void brelse_array(struct buffer_head **heads, int num)
2086 {
2087         int i;
2088         for (i = 0; i < num; i++) {
2089                 brelse(heads[i]);
2090         }
2091 }
2092
2093 /*
2094 ** given the start, and values for the oldest acceptable transactions,
2095 ** this either reads in a replays a transaction, or returns because the transaction
2096 ** is invalid, or too old.
2097 */
2098 static int journal_read_transaction(struct super_block *p_s_sb,
2099                                     unsigned long cur_dblock,
2100                                     unsigned long oldest_start,
2101                                     unsigned long oldest_trans_id,
2102                                     unsigned long newest_mount_id)
2103 {
2104         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
2105         struct reiserfs_journal_desc *desc;
2106         struct reiserfs_journal_commit *commit;
2107         unsigned long trans_id = 0;
2108         struct buffer_head *c_bh;
2109         struct buffer_head *d_bh;
2110         struct buffer_head **log_blocks = NULL;
2111         struct buffer_head **real_blocks = NULL;
2112         unsigned long trans_offset;
2113         int i;
2114         int trans_half;
2115
2116         d_bh = journal_bread(p_s_sb, cur_dblock);
2117         if (!d_bh)
2118                 return 1;
2119         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2120         trans_offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb);
2121         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1037: "
2122                        "journal_read_transaction, offset %llu, len %d mount_id %d",
2123                        d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2124                        get_desc_trans_len(desc), get_desc_mount_id(desc));
2125         if (get_desc_trans_id(desc) < oldest_trans_id) {
2126                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1039: "
2127                                "journal_read_trans skipping because %lu is too old",
2128                                cur_dblock -
2129                                SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb));
2130                 brelse(d_bh);
2131                 return 1;
2132         }
2133         if (get_desc_mount_id(desc) != newest_mount_id) {
2134                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1146: "
2135                                "journal_read_trans skipping because %d is != "
2136                                "newest_mount_id %lu", get_desc_mount_id(desc),
2137                                newest_mount_id);
2138                 brelse(d_bh);
2139                 return 1;
2140         }
2141         c_bh = journal_bread(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2142                              ((trans_offset + get_desc_trans_len(desc) + 1) %
2143                               SB_ONDISK_JOURNAL_SIZE(p_s_sb)));
2144         if (!c_bh) {
2145                 brelse(d_bh);
2146                 return 1;
2147         }
2148         commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2149         if (journal_compare_desc_commit(p_s_sb, desc, commit)) {
2150                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2151                                "journal_read_transaction, "
2152                                "commit offset %llu had bad time %d or length %d",
2153                                c_bh->b_blocknr -
2154                                SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2155                                get_commit_trans_id(commit),
2156                                get_commit_trans_len(commit));
2157                 brelse(c_bh);
2158                 brelse(d_bh);
2159                 return 1;
2160         }
2161         trans_id = get_desc_trans_id(desc);
2162         /* now we know we've got a good transaction, and it was inside the valid time ranges */
2163         log_blocks = kmalloc(get_desc_trans_len(desc) *
2164                              sizeof(struct buffer_head *), GFP_NOFS);
2165         real_blocks = kmalloc(get_desc_trans_len(desc) *
2166                               sizeof(struct buffer_head *), GFP_NOFS);
2167         if (!log_blocks || !real_blocks) {
2168                 brelse(c_bh);
2169                 brelse(d_bh);
2170                 kfree(log_blocks);
2171                 kfree(real_blocks);
2172                 reiserfs_warning(p_s_sb,
2173                                  "journal-1169: kmalloc failed, unable to mount FS");
2174                 return -1;
2175         }
2176         /* get all the buffer heads */
2177         trans_half = journal_trans_half(p_s_sb->s_blocksize);
2178         for (i = 0; i < get_desc_trans_len(desc); i++) {
2179                 log_blocks[i] =
2180                     journal_getblk(p_s_sb,
2181                                    SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2182                                    (trans_offset + 1 +
2183                                     i) % SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2184                 if (i < trans_half) {
2185                         real_blocks[i] =
2186                             sb_getblk(p_s_sb,
2187                                       le32_to_cpu(desc->j_realblock[i]));
2188                 } else {
2189                         real_blocks[i] =
2190                             sb_getblk(p_s_sb,
2191                                       le32_to_cpu(commit->
2192                                                   j_realblock[i - trans_half]));
2193                 }
2194                 if (real_blocks[i]->b_blocknr > SB_BLOCK_COUNT(p_s_sb)) {
2195                         reiserfs_warning(p_s_sb,
2196                                          "journal-1207: REPLAY FAILURE fsck required! Block to replay is outside of filesystem");
2197                         goto abort_replay;
2198                 }
2199                 /* make sure we don't try to replay onto log or reserved area */
2200                 if (is_block_in_log_or_reserved_area
2201                     (p_s_sb, real_blocks[i]->b_blocknr)) {
2202                         reiserfs_warning(p_s_sb,
2203                                          "journal-1204: REPLAY FAILURE fsck required! Trying to replay onto a log block");
2204                       abort_replay:
2205                         brelse_array(log_blocks, i);
2206                         brelse_array(real_blocks, i);
2207                         brelse(c_bh);
2208                         brelse(d_bh);
2209                         kfree(log_blocks);
2210                         kfree(real_blocks);
2211                         return -1;
2212                 }
2213         }
2214         /* read in the log blocks, memcpy to the corresponding real block */
2215         ll_rw_block(READ, get_desc_trans_len(desc), log_blocks);
2216         for (i = 0; i < get_desc_trans_len(desc); i++) {
2217                 wait_on_buffer(log_blocks[i]);
2218                 if (!buffer_uptodate(log_blocks[i])) {
2219                         reiserfs_warning(p_s_sb,
2220                                          "journal-1212: REPLAY FAILURE fsck required! buffer write failed");
2221                         brelse_array(log_blocks + i,
2222                                      get_desc_trans_len(desc) - i);
2223                         brelse_array(real_blocks, get_desc_trans_len(desc));
2224                         brelse(c_bh);
2225                         brelse(d_bh);
2226                         kfree(log_blocks);
2227                         kfree(real_blocks);
2228                         return -1;
2229                 }
2230                 memcpy(real_blocks[i]->b_data, log_blocks[i]->b_data,
2231                        real_blocks[i]->b_size);
2232                 set_buffer_uptodate(real_blocks[i]);
2233                 brelse(log_blocks[i]);
2234         }
2235         /* flush out the real blocks */
2236         for (i = 0; i < get_desc_trans_len(desc); i++) {
2237                 set_buffer_dirty(real_blocks[i]);
2238                 ll_rw_block(SWRITE, 1, real_blocks + i);
2239         }
2240         for (i = 0; i < get_desc_trans_len(desc); i++) {
2241                 wait_on_buffer(real_blocks[i]);
2242                 if (!buffer_uptodate(real_blocks[i])) {
2243                         reiserfs_warning(p_s_sb,
2244                                          "journal-1226: REPLAY FAILURE, fsck required! buffer write failed");
2245                         brelse_array(real_blocks + i,
2246                                      get_desc_trans_len(desc) - i);
2247                         brelse(c_bh);
2248                         brelse(d_bh);
2249                         kfree(log_blocks);
2250                         kfree(real_blocks);
2251                         return -1;
2252                 }
2253                 brelse(real_blocks[i]);
2254         }
2255         cur_dblock =
2256             SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2257             ((trans_offset + get_desc_trans_len(desc) +
2258               2) % SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2259         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2260                        "journal-1095: setting journal " "start to offset %ld",
2261                        cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb));
2262
2263         /* init starting values for the first transaction, in case this is the last transaction to be replayed. */
2264         journal->j_start = cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb);
2265         journal->j_last_flush_trans_id = trans_id;
2266         journal->j_trans_id = trans_id + 1;
2267         /* check for trans_id overflow */
2268         if (journal->j_trans_id == 0)
2269                 journal->j_trans_id = 10;
2270         brelse(c_bh);
2271         brelse(d_bh);
2272         kfree(log_blocks);
2273         kfree(real_blocks);
2274         return 0;
2275 }
2276
2277 /* This function reads blocks starting from block and to max_block of bufsize
2278    size (but no more than BUFNR blocks at a time). This proved to improve
2279    mounting speed on self-rebuilding raid5 arrays at least.
2280    Right now it is only used from journal code. But later we might use it
2281    from other places.
2282    Note: Do not use journal_getblk/sb_getblk functions here! */
2283 static struct buffer_head *reiserfs_breada(struct block_device *dev, int block,
2284                                            int bufsize, unsigned int max_block)
2285 {
2286         struct buffer_head *bhlist[BUFNR];
2287         unsigned int blocks = BUFNR;
2288         struct buffer_head *bh;
2289         int i, j;
2290
2291         bh = __getblk(dev, block, bufsize);
2292         if (buffer_uptodate(bh))
2293                 return (bh);
2294
2295         if (block + BUFNR > max_block) {
2296                 blocks = max_block - block;
2297         }
2298         bhlist[0] = bh;
2299         j = 1;
2300         for (i = 1; i < blocks; i++) {
2301                 bh = __getblk(dev, block + i, bufsize);
2302                 if (buffer_uptodate(bh)) {
2303                         brelse(bh);
2304                         break;
2305                 } else
2306                         bhlist[j++] = bh;
2307         }
2308         ll_rw_block(READ, j, bhlist);
2309         for (i = 1; i < j; i++)
2310                 brelse(bhlist[i]);
2311         bh = bhlist[0];
2312         wait_on_buffer(bh);
2313         if (buffer_uptodate(bh))
2314                 return bh;
2315         brelse(bh);
2316         return NULL;
2317 }
2318
2319 /*
2320 ** read and replay the log
2321 ** on a clean unmount, the journal header's next unflushed pointer will be to an invalid
2322 ** transaction.  This tests that before finding all the transactions in the log, which makes normal mount times fast.
2323 **
2324 ** After a crash, this starts with the next unflushed transaction, and replays until it finds one too old, or invalid.
2325 **
2326 ** On exit, it sets things up so the first transaction will work correctly.
2327 */
2328 static int journal_read(struct super_block *p_s_sb)
2329 {
2330         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
2331         struct reiserfs_journal_desc *desc;
2332         unsigned long oldest_trans_id = 0;
2333         unsigned long oldest_invalid_trans_id = 0;
2334         time_t start;
2335         unsigned long oldest_start = 0;
2336         unsigned long cur_dblock = 0;
2337         unsigned long newest_mount_id = 9;
2338         struct buffer_head *d_bh;
2339         struct reiserfs_journal_header *jh;
2340         int valid_journal_header = 0;
2341         int replay_count = 0;
2342         int continue_replay = 1;
2343         int ret;
2344         char b[BDEVNAME_SIZE];
2345
2346         cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb);
2347         reiserfs_info(p_s_sb, "checking transaction log (%s)\n",
2348                       bdevname(journal->j_dev_bd, b));
2349         start = get_seconds();
2350
2351         /* step 1, read in the journal header block.  Check the transaction it says
2352          ** is the first unflushed, and if that transaction is not valid,
2353          ** replay is done
2354          */
2355         journal->j_header_bh = journal_bread(p_s_sb,
2356                                              SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb)
2357                                              + SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2358         if (!journal->j_header_bh) {
2359                 return 1;
2360         }
2361         jh = (struct reiserfs_journal_header *)(journal->j_header_bh->b_data);
2362         if (le32_to_cpu(jh->j_first_unflushed_offset) <
2363             SB_ONDISK_JOURNAL_SIZE(p_s_sb)
2364             && le32_to_cpu(jh->j_last_flush_trans_id) > 0) {
2365                 oldest_start =
2366                     SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2367                     le32_to_cpu(jh->j_first_unflushed_offset);
2368                 oldest_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2369                 newest_mount_id = le32_to_cpu(jh->j_mount_id);
2370                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2371                                "journal-1153: found in "
2372                                "header: first_unflushed_offset %d, last_flushed_trans_id "
2373                                "%lu", le32_to_cpu(jh->j_first_unflushed_offset),
2374                                le32_to_cpu(jh->j_last_flush_trans_id));
2375                 valid_journal_header = 1;
2376
2377                 /* now, we try to read the first unflushed offset.  If it is not valid,
2378                  ** there is nothing more we can do, and it makes no sense to read
2379                  ** through the whole log.
2380                  */
2381                 d_bh =
2382                     journal_bread(p_s_sb,
2383                                   SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2384                                   le32_to_cpu(jh->j_first_unflushed_offset));
2385                 ret = journal_transaction_is_valid(p_s_sb, d_bh, NULL, NULL);
2386                 if (!ret) {
2387                         continue_replay = 0;
2388                 }
2389                 brelse(d_bh);
2390                 goto start_log_replay;
2391         }
2392
2393         if (continue_replay && bdev_read_only(p_s_sb->s_bdev)) {
2394                 reiserfs_warning(p_s_sb,
2395                                  "clm-2076: device is readonly, unable to replay log");
2396                 return -1;
2397         }
2398
2399         /* ok, there are transactions that need to be replayed.  start with the first log block, find
2400          ** all the valid transactions, and pick out the oldest.
2401          */
2402         while (continue_replay
2403                && cur_dblock <
2404                (SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2405                 SB_ONDISK_JOURNAL_SIZE(p_s_sb))) {
2406                 /* Note that it is required for blocksize of primary fs device and journal
2407                    device to be the same */
2408                 d_bh =
2409                     reiserfs_breada(journal->j_dev_bd, cur_dblock,
2410                                     p_s_sb->s_blocksize,
2411                                     SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2412                                     SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2413                 ret =
2414                     journal_transaction_is_valid(p_s_sb, d_bh,
2415                                                  &oldest_invalid_trans_id,
2416                                                  &newest_mount_id);
2417                 if (ret == 1) {
2418                         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2419                         if (oldest_start == 0) {        /* init all oldest_ values */
2420                                 oldest_trans_id = get_desc_trans_id(desc);
2421                                 oldest_start = d_bh->b_blocknr;
2422                                 newest_mount_id = get_desc_mount_id(desc);
2423                                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2424                                                "journal-1179: Setting "
2425                                                "oldest_start to offset %llu, trans_id %lu",
2426                                                oldest_start -
2427                                                SB_ONDISK_JOURNAL_1st_BLOCK
2428                                                (p_s_sb), oldest_trans_id);
2429                         } else if (oldest_trans_id > get_desc_trans_id(desc)) {
2430                                 /* one we just read was older */
2431                                 oldest_trans_id = get_desc_trans_id(desc);
2432                                 oldest_start = d_bh->b_blocknr;
2433                                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2434                                                "journal-1180: Resetting "
2435                                                "oldest_start to offset %lu, trans_id %lu",
2436                                                oldest_start -
2437                                                SB_ONDISK_JOURNAL_1st_BLOCK
2438                                                (p_s_sb), oldest_trans_id);
2439                         }
2440                         if (newest_mount_id < get_desc_mount_id(desc)) {
2441                                 newest_mount_id = get_desc_mount_id(desc);
2442                                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2443                                                "journal-1299: Setting "
2444                                                "newest_mount_id to %d",
2445                                                get_desc_mount_id(desc));
2446                         }
2447                         cur_dblock += get_desc_trans_len(desc) + 2;
2448                 } else {
2449                         cur_dblock++;
2450                 }
2451                 brelse(d_bh);
2452         }
2453
2454       start_log_replay:
2455         cur_dblock = oldest_start;
2456         if (oldest_trans_id) {
2457                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2458                                "journal-1206: Starting replay "
2459                                "from offset %llu, trans_id %lu",
2460                                cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2461                                oldest_trans_id);
2462
2463         }
2464         replay_count = 0;
2465         while (continue_replay && oldest_trans_id > 0) {
2466                 ret =
2467                     journal_read_transaction(p_s_sb, cur_dblock, oldest_start,
2468                                              oldest_trans_id, newest_mount_id);
2469                 if (ret < 0) {
2470                         return ret;
2471                 } else if (ret != 0) {
2472                         break;
2473                 }
2474                 cur_dblock =
2475                     SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + journal->j_start;
2476                 replay_count++;
2477                 if (cur_dblock == oldest_start)
2478                         break;
2479         }
2480
2481         if (oldest_trans_id == 0) {
2482                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2483                                "journal-1225: No valid " "transactions found");
2484         }
2485         /* j_start does not get set correctly if we don't replay any transactions.
2486          ** if we had a valid journal_header, set j_start to the first unflushed transaction value,
2487          ** copy the trans_id from the header
2488          */
2489         if (valid_journal_header && replay_count == 0) {
2490                 journal->j_start = le32_to_cpu(jh->j_first_unflushed_offset);
2491                 journal->j_trans_id =
2492                     le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2493                 /* check for trans_id overflow */
2494                 if (journal->j_trans_id == 0)
2495                         journal->j_trans_id = 10;
2496                 journal->j_last_flush_trans_id =
2497                     le32_to_cpu(jh->j_last_flush_trans_id);
2498                 journal->j_mount_id = le32_to_cpu(jh->j_mount_id) + 1;
2499         } else {
2500                 journal->j_mount_id = newest_mount_id + 1;
2501         }
2502         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1299: Setting "
2503                        "newest_mount_id to %lu", journal->j_mount_id);
2504         journal->j_first_unflushed_offset = journal->j_start;
2505         if (replay_count > 0) {
2506                 reiserfs_info(p_s_sb,
2507                               "replayed %d transactions in %lu seconds\n",
2508                               replay_count, get_seconds() - start);
2509         }
2510         if (!bdev_read_only(p_s_sb->s_bdev) &&
2511             _update_journal_header_block(p_s_sb, journal->j_start,
2512                                          journal->j_last_flush_trans_id)) {
2513                 /* replay failed, caller must call free_journal_ram and abort
2514                  ** the mount
2515                  */
2516                 return -1;
2517         }
2518         return 0;
2519 }
2520
2521 static struct reiserfs_journal_list *alloc_journal_list(struct super_block *s)
2522 {
2523         struct reiserfs_journal_list *jl;
2524         jl = kzalloc(sizeof(struct reiserfs_journal_list),
2525                      GFP_NOFS | __GFP_NOFAIL);
2526         INIT_LIST_HEAD(&jl->j_list);
2527         INIT_LIST_HEAD(&jl->j_working_list);
2528         INIT_LIST_HEAD(&jl->j_tail_bh_list);
2529         INIT_LIST_HEAD(&jl->j_bh_list);
2530         sema_init(&jl->j_commit_lock, 1);
2531         SB_JOURNAL(s)->j_num_lists++;
2532         get_journal_list(jl);
2533         return jl;
2534 }
2535
2536 static void journal_list_init(struct super_block *p_s_sb)
2537 {
2538         SB_JOURNAL(p_s_sb)->j_current_jl = alloc_journal_list(p_s_sb);
2539 }
2540
2541 static int release_journal_dev(struct super_block *super,
2542                                struct reiserfs_journal *journal)
2543 {
2544         int result;
2545
2546         result = 0;
2547
2548         if (journal->j_dev_file != NULL) {
2549                 result = filp_close(journal->j_dev_file, NULL);
2550                 journal->j_dev_file = NULL;
2551                 journal->j_dev_bd = NULL;
2552         } else if (journal->j_dev_bd != NULL) {
2553                 result = blkdev_put(journal->j_dev_bd);
2554                 journal->j_dev_bd = NULL;
2555         }
2556
2557         if (result != 0) {
2558                 reiserfs_warning(super,
2559                                  "sh-457: release_journal_dev: Cannot release journal device: %i",
2560                                  result);
2561         }
2562         return result;
2563 }
2564
2565 static int journal_init_dev(struct super_block *super,
2566                             struct reiserfs_journal *journal,
2567                             const char *jdev_name)
2568 {
2569         int result;
2570         dev_t jdev;
2571         int blkdev_mode = FMODE_READ | FMODE_WRITE;
2572         char b[BDEVNAME_SIZE];
2573
2574         result = 0;
2575
2576         journal->j_dev_bd = NULL;
2577         journal->j_dev_file = NULL;
2578         jdev = SB_ONDISK_JOURNAL_DEVICE(super) ?
2579             new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super)) : super->s_dev;
2580
2581         if (bdev_read_only(super->s_bdev))
2582                 blkdev_mode = FMODE_READ;
2583
2584         /* there is no "jdev" option and journal is on separate device */
2585         if ((!jdev_name || !jdev_name[0])) {
2586                 journal->j_dev_bd = open_by_devnum(jdev, blkdev_mode);
2587                 if (IS_ERR(journal->j_dev_bd)) {
2588                         result = PTR_ERR(journal->j_dev_bd);
2589                         journal->j_dev_bd = NULL;
2590                         reiserfs_warning(super, "sh-458: journal_init_dev: "
2591                                          "cannot init journal device '%s': %i",
2592                                          __bdevname(jdev, b), result);
2593                         return result;
2594                 } else if (jdev != super->s_dev)
2595                         set_blocksize(journal->j_dev_bd, super->s_blocksize);
2596                 return 0;
2597         }
2598
2599         journal->j_dev_file = filp_open(jdev_name, 0, 0);
2600         if (!IS_ERR(journal->j_dev_file)) {
2601                 struct inode *jdev_inode = journal->j_dev_file->f_mapping->host;
2602                 if (!S_ISBLK(jdev_inode->i_mode)) {
2603                         reiserfs_warning(super, "journal_init_dev: '%s' is "
2604                                          "not a block device", jdev_name);
2605                         result = -ENOTBLK;
2606                         release_journal_dev(super, journal);
2607                 } else {
2608                         /* ok */
2609                         journal->j_dev_bd = I_BDEV(jdev_inode);
2610                         set_blocksize(journal->j_dev_bd, super->s_blocksize);
2611                         reiserfs_info(super,
2612                                       "journal_init_dev: journal device: %s\n",
2613                                       bdevname(journal->j_dev_bd, b));
2614                 }
2615         } else {
2616                 result = PTR_ERR(journal->j_dev_file);
2617                 journal->j_dev_file = NULL;
2618                 reiserfs_warning(super,
2619                                  "journal_init_dev: Cannot open '%s': %i",
2620                                  jdev_name, result);
2621         }
2622         return result;
2623 }
2624
2625 /*
2626 ** must be called once on fs mount.  calls journal_read for you
2627 */
2628 int journal_init(struct super_block *p_s_sb, const char *j_dev_name,
2629                  int old_format, unsigned int commit_max_age)
2630 {
2631         int num_cnodes = SB_ONDISK_JOURNAL_SIZE(p_s_sb) * 2;
2632         struct buffer_head *bhjh;
2633         struct reiserfs_super_block *rs;
2634         struct reiserfs_journal_header *jh;
2635         struct reiserfs_journal *journal;
2636         struct reiserfs_journal_list *jl;
2637         char b[BDEVNAME_SIZE];
2638
2639         journal = SB_JOURNAL(p_s_sb) = vmalloc(sizeof(struct reiserfs_journal));
2640         if (!journal) {
2641                 reiserfs_warning(p_s_sb,
2642                                  "journal-1256: unable to get memory for journal structure");
2643                 return 1;
2644         }
2645         memset(journal, 0, sizeof(struct reiserfs_journal));
2646         INIT_LIST_HEAD(&journal->j_bitmap_nodes);
2647         INIT_LIST_HEAD(&journal->j_prealloc_list);
2648         INIT_LIST_HEAD(&journal->j_working_list);
2649         INIT_LIST_HEAD(&journal->j_journal_list);
2650         journal->j_persistent_trans = 0;
2651         if (reiserfs_allocate_list_bitmaps(p_s_sb,
2652                                            journal->j_list_bitmap,
2653                                            SB_BMAP_NR(p_s_sb)))
2654                 goto free_and_return;
2655         allocate_bitmap_nodes(p_s_sb);
2656
2657         /* reserved for journal area support */
2658         SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb) = (old_format ?
2659                                                  REISERFS_OLD_DISK_OFFSET_IN_BYTES
2660                                                  / p_s_sb->s_blocksize +
2661                                                  SB_BMAP_NR(p_s_sb) +
2662                                                  1 :
2663                                                  REISERFS_DISK_OFFSET_IN_BYTES /
2664                                                  p_s_sb->s_blocksize + 2);
2665
2666         /* Sanity check to see is the standard journal fitting withing first bitmap
2667            (actual for small blocksizes) */
2668         if (!SB_ONDISK_JOURNAL_DEVICE(p_s_sb) &&
2669             (SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb) +
2670              SB_ONDISK_JOURNAL_SIZE(p_s_sb) > p_s_sb->s_blocksize * 8)) {
2671                 reiserfs_warning(p_s_sb,
2672                                  "journal-1393: journal does not fit for area "
2673                                  "addressed by first of bitmap blocks. It starts at "
2674                                  "%u and its size is %u. Block size %ld",
2675                                  SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb),
2676                                  SB_ONDISK_JOURNAL_SIZE(p_s_sb),
2677                                  p_s_sb->s_blocksize);
2678                 goto free_and_return;
2679         }
2680
2681         if (journal_init_dev(p_s_sb, journal, j_dev_name) != 0) {
2682                 reiserfs_warning(p_s_sb,
2683                                  "sh-462: unable to initialize jornal device");
2684                 goto free_and_return;
2685         }
2686
2687         rs = SB_DISK_SUPER_BLOCK(p_s_sb);
2688
2689         /* read journal header */
2690         bhjh = journal_bread(p_s_sb,
2691                              SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2692                              SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2693         if (!bhjh) {
2694                 reiserfs_warning(p_s_sb,
2695                                  "sh-459: unable to read journal header");
2696                 goto free_and_return;
2697         }
2698         jh = (struct reiserfs_journal_header *)(bhjh->b_data);
2699
2700         /* make sure that journal matches to the super block */
2701         if (is_reiserfs_jr(rs)
2702             && (le32_to_cpu(jh->jh_journal.jp_journal_magic) !=
2703                 sb_jp_journal_magic(rs))) {
2704                 reiserfs_warning(p_s_sb,
2705                                  "sh-460: journal header magic %x "
2706                                  "(device %s) does not match to magic found in super "
2707                                  "block %x", jh->jh_journal.jp_journal_magic,
2708                                  bdevname(journal->j_dev_bd, b),
2709                                  sb_jp_journal_magic(rs));
2710                 brelse(bhjh);
2711                 goto free_and_return;
2712         }
2713
2714         journal->j_trans_max = le32_to_cpu(jh->jh_journal.jp_journal_trans_max);
2715         journal->j_max_batch = le32_to_cpu(jh->jh_journal.jp_journal_max_batch);
2716         journal->j_max_commit_age =
2717             le32_to_cpu(jh->jh_journal.jp_journal_max_commit_age);
2718         journal->j_max_trans_age = JOURNAL_MAX_TRANS_AGE;
2719
2720         if (journal->j_trans_max) {
2721                 /* make sure these parameters are available, assign it if they are not */
2722                 __u32 initial = journal->j_trans_max;
2723                 __u32 ratio = 1;
2724
2725                 if (p_s_sb->s_blocksize < 4096)
2726                         ratio = 4096 / p_s_sb->s_blocksize;
2727
2728                 if (SB_ONDISK_JOURNAL_SIZE(p_s_sb) / journal->j_trans_max <
2729                     JOURNAL_MIN_RATIO)
2730                         journal->j_trans_max =
2731                             SB_ONDISK_JOURNAL_SIZE(p_s_sb) / JOURNAL_MIN_RATIO;
2732                 if (journal->j_trans_max > JOURNAL_TRANS_MAX_DEFAULT / ratio)
2733                         journal->j_trans_max =
2734                             JOURNAL_TRANS_MAX_DEFAULT / ratio;
2735                 if (journal->j_trans_max < JOURNAL_TRANS_MIN_DEFAULT / ratio)
2736                         journal->j_trans_max =
2737                             JOURNAL_TRANS_MIN_DEFAULT / ratio;
2738
2739                 if (journal->j_trans_max != initial)
2740                         reiserfs_warning(p_s_sb,
2741                                          "sh-461: journal_init: wrong transaction max size (%u). Changed to %u",
2742                                          initial, journal->j_trans_max);
2743
2744                 journal->j_max_batch = journal->j_trans_max *
2745                     JOURNAL_MAX_BATCH_DEFAULT / JOURNAL_TRANS_MAX_DEFAULT;
2746         }
2747
2748         if (!journal->j_trans_max) {
2749                 /*we have the file system was created by old version of mkreiserfs
2750                    so this field contains zero value */
2751                 journal->j_trans_max = JOURNAL_TRANS_MAX_DEFAULT;
2752                 journal->j_max_batch = JOURNAL_MAX_BATCH_DEFAULT;
2753                 journal->j_max_commit_age = JOURNAL_MAX_COMMIT_AGE;
2754
2755                 /* for blocksize >= 4096 - max transaction size is 1024. For block size < 4096
2756                    trans max size is decreased proportionally */
2757                 if (p_s_sb->s_blocksize < 4096) {
2758                         journal->j_trans_max /= (4096 / p_s_sb->s_blocksize);
2759                         journal->j_max_batch = (journal->j_trans_max) * 9 / 10;
2760                 }
2761         }
2762
2763         journal->j_default_max_commit_age = journal->j_max_commit_age;
2764
2765         if (commit_max_age != 0) {
2766                 journal->j_max_commit_age = commit_max_age;
2767                 journal->j_max_trans_age = commit_max_age;
2768         }
2769
2770         reiserfs_info(p_s_sb, "journal params: device %s, size %u, "
2771                       "journal first block %u, max trans len %u, max batch %u, "
2772                       "max commit age %u, max trans age %u\n",
2773                       bdevname(journal->j_dev_bd, b),
2774                       SB_ONDISK_JOURNAL_SIZE(p_s_sb),
2775                       SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2776                       journal->j_trans_max,
2777                       journal->j_max_batch,
2778                       journal->j_max_commit_age, journal->j_max_trans_age);
2779
2780         brelse(bhjh);
2781
2782         journal->j_list_bitmap_index = 0;
2783         journal_list_init(p_s_sb);
2784
2785         memset(journal->j_list_hash_table, 0,
2786                JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
2787
2788         INIT_LIST_HEAD(&journal->j_dirty_buffers);
2789         spin_lock_init(&journal->j_dirty_buffers_lock);
2790
2791         journal->j_start = 0;
2792         journal->j_len = 0;
2793         journal->j_len_alloc = 0;
2794         atomic_set(&(journal->j_wcount), 0);
2795         atomic_set(&(journal->j_async_throttle), 0);
2796         journal->j_bcount = 0;
2797         journal->j_trans_start_time = 0;
2798         journal->j_last = NULL;
2799         journal->j_first = NULL;
2800         init_waitqueue_head(&(journal->j_join_wait));
2801         sema_init(&journal->j_lock, 1);
2802         sema_init(&journal->j_flush_sem, 1);
2803
2804         journal->j_trans_id = 10;
2805         journal->j_mount_id = 10;
2806         journal->j_state = 0;
2807         atomic_set(&(journal->j_jlock), 0);
2808         journal->j_cnode_free_list = allocate_cnodes(num_cnodes);
2809         journal->j_cnode_free_orig = journal->j_cnode_free_list;
2810         journal->j_cnode_free = journal->j_cnode_free_list ? num_cnodes : 0;
2811         journal->j_cnode_used = 0;
2812         journal->j_must_wait = 0;
2813
2814         if (journal->j_cnode_free == 0) {
2815                 reiserfs_warning(p_s_sb, "journal-2004: Journal cnode memory "
2816                                  "allocation failed (%ld bytes). Journal is "
2817                                  "too large for available memory. Usually "
2818                                  "this is due to a journal that is too large.",
2819                                  sizeof (struct reiserfs_journal_cnode) * num_cnodes);
2820                 goto free_and_return;
2821         }
2822
2823         init_journal_hash(p_s_sb);
2824         jl = journal->j_current_jl;
2825         jl->j_list_bitmap = get_list_bitmap(p_s_sb, jl);
2826         if (!jl->j_list_bitmap) {
2827                 reiserfs_warning(p_s_sb,
2828                                  "journal-2005, get_list_bitmap failed for journal list 0");
2829                 goto free_and_return;
2830         }
2831         if (journal_read(p_s_sb) < 0) {
2832                 reiserfs_warning(p_s_sb, "Replay Failure, unable to mount");
2833                 goto free_and_return;
2834         }
2835
2836         reiserfs_mounted_fs_count++;
2837         if (reiserfs_mounted_fs_count <= 1)
2838                 commit_wq = create_workqueue("reiserfs");
2839
2840         INIT_WORK(&journal->j_work, flush_async_commits, p_s_sb);
2841         return 0;
2842       free_and_return:
2843         free_journal_ram(p_s_sb);
2844         return 1;
2845 }
2846
2847 /*
2848 ** test for a polite end of the current transaction.  Used by file_write, and should
2849 ** be used by delete to make sure they don't write more than can fit inside a single
2850 ** transaction
2851 */
2852 int journal_transaction_should_end(struct reiserfs_transaction_handle *th,
2853                                    int new_alloc)
2854 {
2855         struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2856         time_t now = get_seconds();
2857         /* cannot restart while nested */
2858         BUG_ON(!th->t_trans_id);
2859         if (th->t_refcount > 1)
2860                 return 0;
2861         if (journal->j_must_wait > 0 ||
2862             (journal->j_len_alloc + new_alloc) >= journal->j_max_batch ||
2863             atomic_read(&(journal->j_jlock)) ||
2864             (now - journal->j_trans_start_time) > journal->j_max_trans_age ||
2865             journal->j_cnode_free < (journal->j_trans_max * 3)) {
2866                 return 1;
2867         }
2868         /* protected by the BKL here */
2869         journal->j_len_alloc += new_alloc;
2870         th->t_blocks_allocated += new_alloc ;
2871         return 0;
2872 }
2873
2874 /* this must be called inside a transaction, and requires the
2875 ** kernel_lock to be held
2876 */
2877 void reiserfs_block_writes(struct reiserfs_transaction_handle *th)
2878 {
2879         struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2880         BUG_ON(!th->t_trans_id);
2881         journal->j_must_wait = 1;
2882         set_bit(J_WRITERS_BLOCKED, &journal->j_state);
2883         return;
2884 }
2885
2886 /* this must be called without a transaction started, and does not
2887 ** require BKL
2888 */
2889 void reiserfs_allow_writes(struct super_block *s)
2890 {
2891         struct reiserfs_journal *journal = SB_JOURNAL(s);
2892         clear_bit(J_WRITERS_BLOCKED, &journal->j_state);
2893         wake_up(&journal->j_join_wait);
2894 }
2895
2896 /* this must be called without a transaction started, and does not
2897 ** require BKL
2898 */
2899 void reiserfs_wait_on_write_block(struct super_block *s)
2900 {
2901         struct reiserfs_journal *journal = SB_JOURNAL(s);
2902         wait_event(journal->j_join_wait,
2903                    !test_bit(J_WRITERS_BLOCKED, &journal->j_state));
2904 }
2905
2906 static void queue_log_writer(struct super_block *s)
2907 {
2908         wait_queue_t wait;
2909         struct reiserfs_journal *journal = SB_JOURNAL(s);
2910         set_bit(J_WRITERS_QUEUED, &journal->j_state);
2911
2912         /*
2913          * we don't want to use wait_event here because
2914          * we only want to wait once.
2915          */
2916         init_waitqueue_entry(&wait, current);
2917         add_wait_queue(&journal->j_join_wait, &wait);
2918         set_current_state(TASK_UNINTERRUPTIBLE);
2919         if (test_bit(J_WRITERS_QUEUED, &journal->j_state))
2920                 schedule();
2921         current->state = TASK_RUNNING;
2922         remove_wait_queue(&journal->j_join_wait, &wait);
2923 }
2924
2925 static void wake_queued_writers(struct super_block *s)
2926 {
2927         struct reiserfs_journal *journal = SB_JOURNAL(s);
2928         if (test_and_clear_bit(J_WRITERS_QUEUED, &journal->j_state))
2929                 wake_up(&journal->j_join_wait);
2930 }
2931
2932 static void let_transaction_grow(struct super_block *sb, unsigned long trans_id)
2933 {
2934         struct reiserfs_journal *journal = SB_JOURNAL(sb);
2935         unsigned long bcount = journal->j_bcount;
2936         while (1) {
2937                 schedule_timeout_uninterruptible(1);
2938                 journal->j_current_jl->j_state |= LIST_COMMIT_PENDING;
2939                 while ((atomic_read(&journal->j_wcount) > 0 ||
2940                         atomic_read(&journal->j_jlock)) &&
2941                        journal->j_trans_id == trans_id) {
2942                         queue_log_writer(sb);
2943                 }
2944                 if (journal->j_trans_id != trans_id)
2945                         break;
2946                 if (bcount == journal->j_bcount)
2947                         break;
2948                 bcount = journal->j_bcount;
2949         }
2950 }
2951
2952 /* join == true if you must join an existing transaction.
2953 ** join == false if you can deal with waiting for others to finish
2954 **
2955 ** this will block until the transaction is joinable.  send the number of blocks you
2956 ** expect to use in nblocks.
2957 */
2958 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
2959                               struct super_block *p_s_sb, unsigned long nblocks,
2960                               int join)
2961 {
2962         time_t now = get_seconds();
2963         int old_trans_id;
2964         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
2965         struct reiserfs_transaction_handle myth;
2966         int sched_count = 0;
2967         int retval;
2968
2969         reiserfs_check_lock_depth(p_s_sb, "journal_begin");
2970         if (nblocks > journal->j_trans_max)
2971                 BUG();
2972
2973         PROC_INFO_INC(p_s_sb, journal.journal_being);
2974         /* set here for journal_join */
2975         th->t_refcount = 1;
2976         th->t_super = p_s_sb;
2977
2978       relock:
2979         lock_journal(p_s_sb);
2980         if (join != JBEGIN_ABORT && reiserfs_is_journal_aborted(journal)) {
2981                 unlock_journal(p_s_sb);
2982                 retval = journal->j_errno;
2983                 goto out_fail;
2984         }
2985         journal->j_bcount++;
2986
2987         if (test_bit(J_WRITERS_BLOCKED, &journal->j_state)) {
2988                 unlock_journal(p_s_sb);
2989                 reiserfs_wait_on_write_block(p_s_sb);
2990                 PROC_INFO_INC(p_s_sb, journal.journal_relock_writers);
2991                 goto relock;
2992         }
2993         now = get_seconds();
2994
2995         /* if there is no room in the journal OR
2996          ** if this transaction is too old, and we weren't called joinable, wait for it to finish before beginning
2997          ** we don't sleep if there aren't other writers
2998          */
2999
3000         if ((!join && journal->j_must_wait > 0) ||
3001             (!join
3002              && (journal->j_len_alloc + nblocks + 2) >= journal->j_max_batch)
3003             || (!join && atomic_read(&journal->j_wcount) > 0
3004                 && journal->j_trans_start_time > 0
3005                 && (now - journal->j_trans_start_time) >
3006                 journal->j_max_trans_age) || (!join
3007                                               && atomic_read(&journal->j_jlock))
3008             || (!join && journal->j_cnode_free < (journal->j_trans_max * 3))) {
3009
3010                 old_trans_id = journal->j_trans_id;
3011                 unlock_journal(p_s_sb); /* allow others to finish this transaction */
3012
3013                 if (!join && (journal->j_len_alloc + nblocks + 2) >=
3014                     journal->j_max_batch &&
3015                     ((journal->j_len + nblocks + 2) * 100) <
3016                     (journal->j_len_alloc * 75)) {
3017                         if (atomic_read(&journal->j_wcount) > 10) {
3018                                 sched_count++;
3019                                 queue_log_writer(p_s_sb);
3020                                 goto relock;
3021                         }
3022                 }
3023                 /* don't mess with joining the transaction if all we have to do is
3024                  * wait for someone else to do a commit
3025                  */
3026                 if (atomic_read(&journal->j_jlock)) {
3027                         while (journal->j_trans_id == old_trans_id &&
3028                                atomic_read(&journal->j_jlock)) {
3029                                 queue_log_writer(p_s_sb);
3030                         }
3031                         goto relock;
3032                 }
3033                 retval = journal_join(&myth, p_s_sb, 1);
3034                 if (retval)
3035                         goto out_fail;
3036
3037                 /* someone might have ended the transaction while we joined */
3038                 if (old_trans_id != journal->j_trans_id) {
3039                         retval = do_journal_end(&myth, p_s_sb, 1, 0);
3040                 } else {
3041                         retval = do_journal_end(&myth, p_s_sb, 1, COMMIT_NOW);
3042                 }
3043
3044                 if (retval)
3045                         goto out_fail;
3046
3047                 PROC_INFO_INC(p_s_sb, journal.journal_relock_wcount);
3048                 goto relock;
3049         }
3050         /* we are the first writer, set trans_id */
3051         if (journal->j_trans_start_time == 0) {
3052                 journal->j_trans_start_time = get_seconds();
3053         }
3054         atomic_inc(&(journal->j_wcount));
3055         journal->j_len_alloc += nblocks;
3056         th->t_blocks_logged = 0;
3057         th->t_blocks_allocated = nblocks;
3058         th->t_trans_id = journal->j_trans_id;
3059         unlock_journal(p_s_sb);
3060         INIT_LIST_HEAD(&th->t_list);
3061         get_fs_excl();
3062         return 0;
3063
3064       out_fail:
3065         memset(th, 0, sizeof(*th));
3066         /* Re-set th->t_super, so we can properly keep track of how many
3067          * persistent transactions there are. We need to do this so if this
3068          * call is part of a failed restart_transaction, we can free it later */
3069         th->t_super = p_s_sb;
3070         return retval;
3071 }
3072
3073 struct reiserfs_transaction_handle *reiserfs_persistent_transaction(struct
3074                                                                     super_block
3075                                                                     *s,
3076                                                                     int nblocks)
3077 {
3078         int ret;
3079         struct reiserfs_transaction_handle *th;
3080
3081         /* if we're nesting into an existing transaction.  It will be
3082          ** persistent on its own
3083          */
3084         if (reiserfs_transaction_running(s)) {
3085                 th = current->journal_info;
3086                 th->t_refcount++;
3087                 if (th->t_refcount < 2) {
3088                         BUG();
3089                 }
3090                 return th;
3091         }
3092         th = kmalloc(sizeof(struct reiserfs_transaction_handle), GFP_NOFS);
3093         if (!th)
3094                 return NULL;
3095         ret = journal_begin(th, s, nblocks);
3096         if (ret) {
3097                 kfree(th);
3098                 return NULL;
3099         }
3100
3101         SB_JOURNAL(s)->j_persistent_trans++;
3102         return th;
3103 }
3104
3105 int reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle *th)
3106 {
3107         struct super_block *s = th->t_super;
3108         int ret = 0;
3109         if (th->t_trans_id)
3110                 ret = journal_end(th, th->t_super, th->t_blocks_allocated);
3111         else
3112                 ret = -EIO;
3113         if (th->t_refcount == 0) {
3114                 SB_JOURNAL(s)->j_persistent_trans--;
3115                 kfree(th);
3116         }
3117         return ret;
3118 }
3119
3120 static int journal_join(struct reiserfs_transaction_handle *th,
3121                         struct super_block *p_s_sb, unsigned long nblocks)
3122 {
3123         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3124
3125         /* this keeps do_journal_end from NULLing out the current->journal_info
3126          ** pointer
3127          */
3128         th->t_handle_save = cur_th;
3129         if (cur_th && cur_th->t_refcount > 1) {
3130                 BUG();
3131         }
3132         return do_journal_begin_r(th, p_s_sb, nblocks, JBEGIN_JOIN);
3133 }
3134
3135 int journal_join_abort(struct reiserfs_transaction_handle *th,
3136                        struct super_block *p_s_sb, unsigned long nblocks)
3137 {
3138         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3139
3140         /* this keeps do_journal_end from NULLing out the current->journal_info
3141          ** pointer
3142          */
3143         th->t_handle_save = cur_th;
3144         if (cur_th && cur_th->t_refcount > 1) {
3145                 BUG();
3146         }
3147         return do_journal_begin_r(th, p_s_sb, nblocks, JBEGIN_ABORT);
3148 }
3149
3150 int journal_begin(struct reiserfs_transaction_handle *th,
3151                   struct super_block *p_s_sb, unsigned long nblocks)
3152 {
3153         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3154         int ret;
3155
3156         th->t_handle_save = NULL;
3157         if (cur_th) {
3158                 /* we are nesting into the current transaction */
3159                 if (cur_th->t_super == p_s_sb) {
3160                         BUG_ON(!cur_th->t_refcount);
3161                         cur_th->t_refcount++;
3162                         memcpy(th, cur_th, sizeof(*th));
3163                         if (th->t_refcount <= 1)
3164                                 reiserfs_warning(p_s_sb,
3165                                                  "BAD: refcount <= 1, but journal_info != 0");
3166                         return 0;
3167                 } else {
3168                         /* we've ended up with a handle from a different filesystem.
3169                          ** save it and restore on journal_end.  This should never
3170                          ** really happen...
3171                          */
3172                         reiserfs_warning(p_s_sb,
3173                                          "clm-2100: nesting info a different FS");
3174                         th->t_handle_save = current->journal_info;
3175                         current->journal_info = th;
3176                 }
3177         } else {
3178                 current->journal_info = th;
3179         }
3180         ret = do_journal_begin_r(th, p_s_sb, nblocks, JBEGIN_REG);
3181         if (current->journal_info != th)
3182                 BUG();
3183
3184         /* I guess this boils down to being the reciprocal of clm-2100 above.
3185          * If do_journal_begin_r fails, we need to put it back, since journal_end
3186          * won't be called to do it. */
3187         if (ret)
3188                 current->journal_info = th->t_handle_save;
3189         else
3190                 BUG_ON(!th->t_refcount);
3191
3192         return ret;
3193 }
3194
3195 /*
3196 ** puts bh into the current transaction.  If it was already there, reorders removes the
3197 ** old pointers from the hash, and puts new ones in (to make sure replay happen in the right order).
3198 **
3199 ** if it was dirty, cleans and files onto the clean list.  I can't let it be dirty again until the
3200 ** transaction is committed.
3201 **
3202 ** if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len.
3203 */
3204 int journal_mark_dirty(struct reiserfs_transaction_handle *th,
3205                        struct super_block *p_s_sb, struct buffer_head *bh)
3206 {
3207         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3208         struct reiserfs_journal_cnode *cn = NULL;
3209         int count_already_incd = 0;
3210         int prepared = 0;
3211         BUG_ON(!th->t_trans_id);
3212
3213         PROC_INFO_INC(p_s_sb, journal.mark_dirty);
3214         if (th->t_trans_id != journal->j_trans_id) {
3215                 reiserfs_panic(th->t_super,
3216                                "journal-1577: handle trans id %ld != current trans id %ld\n",
3217                                th->t_trans_id, journal->j_trans_id);
3218         }
3219
3220         p_s_sb->s_dirt = 1;
3221
3222         prepared = test_clear_buffer_journal_prepared(bh);
3223         clear_buffer_journal_restore_dirty(bh);
3224         /* already in this transaction, we are done */
3225         if (buffer_journaled(bh)) {
3226                 PROC_INFO_INC(p_s_sb, journal.mark_dirty_already);
3227                 return 0;
3228         }
3229
3230         /* this must be turned into a panic instead of a warning.  We can't allow
3231          ** a dirty or journal_dirty or locked buffer to be logged, as some changes
3232          ** could get to disk too early.  NOT GOOD.
3233          */
3234         if (!prepared || buffer_dirty(bh)) {
3235                 reiserfs_warning(p_s_sb, "journal-1777: buffer %llu bad state "
3236                                  "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT",
3237                                  (unsigned long long)bh->b_blocknr,
3238                                  prepared ? ' ' : '!',
3239                                  buffer_locked(bh) ? ' ' : '!',
3240                                  buffer_dirty(bh) ? ' ' : '!',
3241                                  buffer_journal_dirty(bh) ? ' ' : '!');
3242         }
3243
3244         if (atomic_read(&(journal->j_wcount)) <= 0) {
3245                 reiserfs_warning(p_s_sb,
3246                                  "journal-1409: journal_mark_dirty returning because j_wcount was %d",
3247                                  atomic_read(&(journal->j_wcount)));
3248                 return 1;
3249         }
3250         /* this error means I've screwed up, and we've overflowed the transaction.
3251          ** Nothing can be done here, except make the FS readonly or panic.
3252          */
3253         if (journal->j_len >= journal->j_trans_max) {
3254                 reiserfs_panic(th->t_super,
3255                                "journal-1413: journal_mark_dirty: j_len (%lu) is too big\n",
3256                                journal->j_len);
3257         }
3258
3259         if (buffer_journal_dirty(bh)) {
3260                 count_already_incd = 1;
3261                 PROC_INFO_INC(p_s_sb, journal.mark_dirty_notjournal);
3262                 clear_buffer_journal_dirty(bh);
3263         }
3264
3265         if (journal->j_len > journal->j_len_alloc) {
3266                 journal->j_len_alloc = journal->j_len + JOURNAL_PER_BALANCE_CNT;
3267         }
3268
3269         set_buffer_journaled(bh);
3270
3271         /* now put this guy on the end */
3272         if (!cn) {
3273                 cn = get_cnode(p_s_sb);
3274                 if (!cn) {
3275                         reiserfs_panic(p_s_sb, "get_cnode failed!\n");
3276                 }
3277
3278                 if (th->t_blocks_logged == th->t_blocks_allocated) {
3279                         th->t_blocks_allocated += JOURNAL_PER_BALANCE_CNT;
3280                         journal->j_len_alloc += JOURNAL_PER_BALANCE_CNT;
3281                 }
3282                 th->t_blocks_logged++;
3283                 journal->j_len++;
3284
3285                 cn->bh = bh;
3286                 cn->blocknr = bh->b_blocknr;
3287                 cn->sb = p_s_sb;
3288                 cn->jlist = NULL;
3289                 insert_journal_hash(journal->j_hash_table, cn);
3290                 if (!count_already_incd) {
3291                         get_bh(bh);
3292                 }
3293         }
3294         cn->next = NULL;
3295         cn->prev = journal->j_last;
3296         cn->bh = bh;
3297         if (journal->j_last) {
3298                 journal->j_last->next = cn;
3299                 journal->j_last = cn;
3300         } else {
3301                 journal->j_first = cn;
3302                 journal->j_last = cn;
3303         }
3304         return 0;
3305 }
3306
3307 int journal_end(struct reiserfs_transaction_handle *th,
3308                 struct super_block *p_s_sb, unsigned long nblocks)
3309 {
3310         if (!current->journal_info && th->t_refcount > 1)
3311                 reiserfs_warning(p_s_sb, "REISER-NESTING: th NULL, refcount %d",
3312                                  th->t_refcount);
3313
3314         if (!th->t_trans_id) {
3315                 WARN_ON(1);
3316                 return -EIO;
3317         }
3318
3319         th->t_refcount--;
3320         if (th->t_refcount > 0) {
3321                 struct reiserfs_transaction_handle *cur_th =
3322                     current->journal_info;
3323
3324                 /* we aren't allowed to close a nested transaction on a different
3325                  ** filesystem from the one in the task struct
3326                  */
3327                 if (cur_th->t_super != th->t_super)
3328                         BUG();
3329
3330                 if (th != cur_th) {
3331                         memcpy(current->journal_info, th, sizeof(*th));
3332                         th->t_trans_id = 0;
3333                 }
3334                 return 0;
3335         } else {
3336                 return do_journal_end(th, p_s_sb, nblocks, 0);
3337         }
3338 }
3339
3340 /* removes from the current transaction, relsing and descrementing any counters.
3341 ** also files the removed buffer directly onto the clean list
3342 **
3343 ** called by journal_mark_freed when a block has been deleted
3344 **
3345 ** returns 1 if it cleaned and relsed the buffer. 0 otherwise
3346 */
3347 static int remove_from_transaction(struct super_block *p_s_sb,
3348                                    b_blocknr_t blocknr, int already_cleaned)
3349 {
3350         struct buffer_head *bh;
3351         struct reiserfs_journal_cnode *cn;
3352         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3353         int ret = 0;
3354
3355         cn = get_journal_hash_dev(p_s_sb, journal->j_hash_table, blocknr);
3356         if (!cn || !cn->bh) {
3357                 return ret;
3358         }
3359         bh = cn->bh;
3360         if (cn->prev) {
3361                 cn->prev->next = cn->next;
3362         }
3363         if (cn->next) {
3364                 cn->next->prev = cn->prev;
3365         }
3366         if (cn == journal->j_first) {
3367                 journal->j_first = cn->next;
3368         }
3369         if (cn == journal->j_last) {
3370                 journal->j_last = cn->prev;
3371         }
3372         if (bh)
3373                 remove_journal_hash(p_s_sb, journal->j_hash_table, NULL,
3374                                     bh->b_blocknr, 0);
3375         clear_buffer_journaled(bh);     /* don't log this one */
3376
3377         if (!already_cleaned) {
3378                 clear_buffer_journal_dirty(bh);
3379                 clear_buffer_dirty(bh);
3380                 clear_buffer_journal_test(bh);
3381                 put_bh(bh);
3382                 if (atomic_read(&(bh->b_count)) < 0) {
3383                         reiserfs_warning(p_s_sb,
3384                                          "journal-1752: remove from trans, b_count < 0");
3385                 }
3386                 ret = 1;
3387         }
3388         journal->j_len--;
3389         journal->j_len_alloc--;
3390         free_cnode(p_s_sb, cn);
3391         return ret;
3392 }
3393
3394 /*
3395 ** for any cnode in a journal list, it can only be dirtied of all the
3396 ** transactions that include it are commited to disk.
3397 ** this checks through each transaction, and returns 1 if you are allowed to dirty,
3398 ** and 0 if you aren't
3399 **
3400 ** it is called by dirty_journal_list, which is called after flush_commit_list has gotten all the log
3401 ** blocks for a given transaction on disk
3402 **
3403 */
3404 static int can_dirty(struct reiserfs_journal_cnode *cn)
3405 {
3406         struct super_block *sb = cn->sb;
3407         b_blocknr_t blocknr = cn->blocknr;
3408         struct reiserfs_journal_cnode *cur = cn->hprev;
3409         int can_dirty = 1;
3410
3411         /* first test hprev.  These are all newer than cn, so any node here
3412          ** with the same block number and dev means this node can't be sent
3413          ** to disk right now.
3414          */
3415         while (cur && can_dirty) {
3416                 if (cur->jlist && cur->bh && cur->blocknr && cur->sb == sb &&
3417                     cur->blocknr == blocknr) {
3418                         can_dirty = 0;
3419                 }
3420                 cur = cur->hprev;
3421         }
3422         /* then test hnext.  These are all older than cn.  As long as they
3423          ** are committed to the log, it is safe to write cn to disk
3424          */
3425         cur = cn->hnext;
3426         while (cur && can_dirty) {
3427                 if (cur->jlist && cur->jlist->j_len > 0 &&
3428                     atomic_read(&(cur->jlist->j_commit_left)) > 0 && cur->bh &&
3429                     cur->blocknr && cur->sb == sb && cur->blocknr == blocknr) {
3430                         can_dirty = 0;
3431                 }
3432                 cur = cur->hnext;
3433         }
3434         return can_dirty;
3435 }
3436
3437 /* syncs the commit blocks, but does not force the real buffers to disk
3438 ** will wait until the current transaction is done/commited before returning
3439 */
3440 int journal_end_sync(struct reiserfs_transaction_handle *th,
3441                      struct super_block *p_s_sb, unsigned long nblocks)
3442 {
3443         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3444
3445         BUG_ON(!th->t_trans_id);
3446         /* you can sync while nested, very, very bad */
3447         if (th->t_refcount > 1) {
3448                 BUG();
3449         }
3450         if (journal->j_len == 0) {
3451                 reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb),
3452                                              1);
3453                 journal_mark_dirty(th, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb));
3454         }
3455         return do_journal_end(th, p_s_sb, nblocks, COMMIT_NOW | WAIT);
3456 }
3457
3458 /*
3459 ** writeback the pending async commits to disk
3460 */
3461 static void flush_async_commits(void *p)
3462 {
3463         struct super_block *p_s_sb = p;
3464         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3465         struct reiserfs_journal_list *jl;
3466         struct list_head *entry;
3467
3468         lock_kernel();
3469         if (!list_empty(&journal->j_journal_list)) {
3470                 /* last entry is the youngest, commit it and you get everything */
3471                 entry = journal->j_journal_list.prev;
3472                 jl = JOURNAL_LIST_ENTRY(entry);
3473                 flush_commit_list(p_s_sb, jl, 1);
3474         }
3475         unlock_kernel();
3476 }
3477
3478 /*
3479 ** flushes any old transactions to disk
3480 ** ends the current transaction if it is too old
3481 */
3482 int reiserfs_flush_old_commits(struct super_block *p_s_sb)
3483 {
3484         time_t now;
3485         struct reiserfs_transaction_handle th;
3486         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3487
3488         now = get_seconds();
3489         /* safety check so we don't flush while we are replaying the log during
3490          * mount
3491          */
3492         if (list_empty(&journal->j_journal_list)) {
3493                 return 0;
3494         }
3495
3496         /* check the current transaction.  If there are no writers, and it is
3497          * too old, finish it, and force the commit blocks to disk
3498          */
3499         if (atomic_read(&journal->j_wcount) <= 0 &&
3500             journal->j_trans_start_time > 0 &&
3501             journal->j_len > 0 &&
3502             (now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3503                 if (!journal_join(&th, p_s_sb, 1)) {
3504                         reiserfs_prepare_for_journal(p_s_sb,
3505                                                      SB_BUFFER_WITH_SB(p_s_sb),
3506                                                      1);
3507                         journal_mark_dirty(&th, p_s_sb,
3508                                            SB_BUFFER_WITH_SB(p_s_sb));
3509
3510                         /* we're only being called from kreiserfsd, it makes no sense to do
3511                          ** an async commit so that kreiserfsd can do it later
3512                          */
3513                         do_journal_end(&th, p_s_sb, 1, COMMIT_NOW | WAIT);
3514                 }
3515         }
3516         return p_s_sb->s_dirt;
3517 }
3518
3519 /*
3520 ** returns 0 if do_journal_end should return right away, returns 1 if do_journal_end should finish the commit
3521 **
3522 ** if the current transaction is too old, but still has writers, this will wait on j_join_wait until all
3523 ** the writers are done.  By the time it wakes up, the transaction it was called has already ended, so it just
3524 ** flushes the commit list and returns 0.
3525 **
3526 ** Won't batch when flush or commit_now is set.  Also won't batch when others are waiting on j_join_wait.
3527 **
3528 ** Note, we can't allow the journal_end to proceed while there are still writers in the log.
3529 */
3530 static int check_journal_end(struct reiserfs_transaction_handle *th,
3531                              struct super_block *p_s_sb, unsigned long nblocks,
3532                              int flags)
3533 {
3534
3535         time_t now;
3536         int flush = flags & FLUSH_ALL;
3537         int commit_now = flags & COMMIT_NOW;
3538         int wait_on_commit = flags & WAIT;
3539         struct reiserfs_journal_list *jl;
3540         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3541
3542         BUG_ON(!th->t_trans_id);
3543
3544         if (th->t_trans_id != journal->j_trans_id) {
3545                 reiserfs_panic(th->t_super,
3546                                "journal-1577: handle trans id %ld != current trans id %ld\n",
3547                                th->t_trans_id, journal->j_trans_id);
3548         }
3549
3550         journal->j_len_alloc -= (th->t_blocks_allocated - th->t_blocks_logged);
3551         if (atomic_read(&(journal->j_wcount)) > 0) {    /* <= 0 is allowed.  unmounting might not call begin */
3552                 atomic_dec(&(journal->j_wcount));
3553         }
3554
3555         /* BUG, deal with case where j_len is 0, but people previously freed blocks need to be released
3556          ** will be dealt with by next transaction that actually writes something, but should be taken
3557          ** care of in this trans
3558          */
3559         if (journal->j_len == 0) {
3560                 BUG();
3561         }
3562         /* if wcount > 0, and we are called to with flush or commit_now,
3563          ** we wait on j_join_wait.  We will wake up when the last writer has
3564          ** finished the transaction, and started it on its way to the disk.
3565          ** Then, we flush the commit or journal list, and just return 0
3566          ** because the rest of journal end was already done for this transaction.
3567          */
3568         if (atomic_read(&(journal->j_wcount)) > 0) {
3569                 if (flush || commit_now) {
3570                         unsigned trans_id;
3571
3572                         jl = journal->j_current_jl;
3573                         trans_id = jl->j_trans_id;
3574                         if (wait_on_commit)
3575                                 jl->j_state |= LIST_COMMIT_PENDING;
3576                         atomic_set(&(journal->j_jlock), 1);
3577                         if (flush) {
3578                                 journal->j_next_full_flush = 1;
3579                         }
3580                         unlock_journal(p_s_sb);
3581
3582                         /* sleep while the current transaction is still j_jlocked */
3583                         while (journal->j_trans_id == trans_id) {
3584                                 if (atomic_read(&journal->j_jlock)) {
3585                                         queue_log_writer(p_s_sb);
3586                                 } else {
3587                                         lock_journal(p_s_sb);
3588                                         if (journal->j_trans_id == trans_id) {
3589                                                 atomic_set(&(journal->j_jlock),
3590                                                            1);
3591                                         }
3592                                         unlock_journal(p_s_sb);
3593                                 }
3594                         }
3595                         if (journal->j_trans_id == trans_id) {
3596                                 BUG();
3597                         }
3598                         if (commit_now
3599                             && journal_list_still_alive(p_s_sb, trans_id)
3600                             && wait_on_commit) {
3601                                 flush_commit_list(p_s_sb, jl, 1);
3602                         }
3603                         return 0;
3604                 }
3605                 unlock_journal(p_s_sb);
3606                 return 0;
3607         }
3608
3609         /* deal with old transactions where we are the last writers */
3610         now = get_seconds();
3611         if ((now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3612                 commit_now = 1;
3613                 journal->j_next_async_flush = 1;
3614         }
3615         /* don't batch when someone is waiting on j_join_wait */
3616         /* don't batch when syncing the commit or flushing the whole trans */
3617         if (!(journal->j_must_wait > 0) && !(atomic_read(&(journal->j_jlock)))
3618             && !flush && !commit_now && (journal->j_len < journal->j_max_batch)
3619             && journal->j_len_alloc < journal->j_max_batch
3620             && journal->j_cnode_free > (journal->j_trans_max * 3)) {
3621                 journal->j_bcount++;
3622                 unlock_journal(p_s_sb);
3623                 return 0;
3624         }
3625
3626         if (journal->j_start > SB_ONDISK_JOURNAL_SIZE(p_s_sb)) {
3627                 reiserfs_panic(p_s_sb,
3628                                "journal-003: journal_end: j_start (%ld) is too high\n",
3629                                journal->j_start);
3630         }
3631         return 1;
3632 }
3633
3634 /*
3635 ** Does all the work that makes deleting blocks safe.
3636 ** when deleting a block mark BH_JNew, just remove it from the current transaction, clean it's buffer_head and move on.
3637 **
3638 ** otherwise:
3639 ** set a bit for the block in the journal bitmap.  That will prevent it from being allocated for unformatted nodes
3640 ** before this transaction has finished.
3641 **
3642 ** mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers.  That will prevent any old transactions with
3643 ** this block from trying to flush to the real location.  Since we aren't removing the cnode from the journal_list_hash,
3644 ** the block can't be reallocated yet.
3645 **
3646 ** Then remove it from the current transaction, decrementing any counters and filing it on the clean list.
3647 */
3648 int journal_mark_freed(struct reiserfs_transaction_handle *th,
3649                        struct super_block *p_s_sb, b_blocknr_t blocknr)
3650 {
3651         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3652         struct reiserfs_journal_cnode *cn = NULL;
3653         struct buffer_head *bh = NULL;
3654         struct reiserfs_list_bitmap *jb = NULL;
3655         int cleaned = 0;
3656         BUG_ON(!th->t_trans_id);
3657
3658         cn = get_journal_hash_dev(p_s_sb, journal->j_hash_table, blocknr);
3659         if (cn && cn->bh) {
3660                 bh = cn->bh;
3661                 get_bh(bh);
3662         }
3663         /* if it is journal new, we just remove it from this transaction */
3664         if (bh && buffer_journal_new(bh)) {
3665                 clear_buffer_journal_new(bh);
3666                 clear_prepared_bits(bh);
3667                 reiserfs_clean_and_file_buffer(bh);
3668                 cleaned = remove_from_transaction(p_s_sb, blocknr, cleaned);
3669         } else {
3670                 /* set the bit for this block in the journal bitmap for this transaction */
3671                 jb = journal->j_current_jl->j_list_bitmap;
3672                 if (!jb) {
3673                         reiserfs_panic(p_s_sb,
3674                                        "journal-1702: journal_mark_freed, journal_list_bitmap is NULL\n");
3675                 }
3676                 set_bit_in_list_bitmap(p_s_sb, blocknr, jb);
3677
3678                 /* Note, the entire while loop is not allowed to schedule.  */
3679
3680                 if (bh) {
3681                         clear_prepared_bits(bh);
3682                         reiserfs_clean_and_file_buffer(bh);
3683                 }
3684                 cleaned = remove_from_transaction(p_s_sb, blocknr, cleaned);
3685
3686                 /* find all older transactions with this block, make sure they don't try to write it out */
3687                 cn = get_journal_hash_dev(p_s_sb, journal->j_list_hash_table,
3688                                           blocknr);
3689                 while (cn) {
3690                         if (p_s_sb == cn->sb && blocknr == cn->blocknr) {
3691                                 set_bit(BLOCK_FREED, &cn->state);
3692                                 if (cn->bh) {
3693                                         if (!cleaned) {
3694                                                 /* remove_from_transaction will brelse the buffer if it was
3695                                                  ** in the current trans
3696                                                  */
3697                                                 clear_buffer_journal_dirty(cn->
3698                                                                            bh);
3699                                                 clear_buffer_dirty(cn->bh);
3700                                                 clear_buffer_journal_test(cn->
3701                                                                           bh);
3702                                                 cleaned = 1;
3703                                                 put_bh(cn->bh);
3704                                                 if (atomic_read
3705                                                     (&(cn->bh->b_count)) < 0) {
3706                                                         reiserfs_warning(p_s_sb,
3707                                                                          "journal-2138: cn->bh->b_count < 0");
3708                                                 }
3709                                         }
3710                                         if (cn->jlist) {        /* since we are clearing the bh, we MUST dec nonzerolen */
3711                                                 atomic_dec(&
3712                                                            (cn->jlist->
3713                                                             j_nonzerolen));
3714                                         }
3715                                         cn->bh = NULL;
3716                                 }
3717                         }
3718                         cn = cn->hnext;
3719                 }
3720         }
3721
3722         if (bh) {
3723                 put_bh(bh);     /* get_hash grabs the buffer */
3724                 if (atomic_read(&(bh->b_count)) < 0) {
3725                         reiserfs_warning(p_s_sb,
3726                                          "journal-2165: bh->b_count < 0");
3727                 }
3728         }
3729         return 0;
3730 }
3731
3732 void reiserfs_update_inode_transaction(struct inode *inode)
3733 {
3734         struct reiserfs_journal *journal = SB_JOURNAL(inode->i_sb);
3735         REISERFS_I(inode)->i_jl = journal->j_current_jl;
3736         REISERFS_I(inode)->i_trans_id = journal->j_trans_id;
3737 }
3738
3739 /*
3740  * returns -1 on error, 0 if no commits/barriers were done and 1
3741  * if a transaction was actually committed and the barrier was done
3742  */
3743 static int __commit_trans_jl(struct inode *inode, unsigned long id,
3744                              struct reiserfs_journal_list *jl)
3745 {
3746         struct reiserfs_transaction_handle th;
3747         struct super_block *sb = inode->i_sb;
3748         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3749         int ret = 0;
3750
3751         /* is it from the current transaction, or from an unknown transaction? */
3752         if (id == journal->j_trans_id) {
3753                 jl = journal->j_current_jl;
3754                 /* try to let other writers come in and grow this transaction */
3755                 let_transaction_grow(sb, id);
3756                 if (journal->j_trans_id != id) {
3757                         goto flush_commit_only;
3758                 }
3759
3760                 ret = journal_begin(&th, sb, 1);
3761                 if (ret)
3762                         return ret;
3763
3764                 /* someone might have ended this transaction while we joined */
3765                 if (journal->j_trans_id != id) {
3766                         reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3767                                                      1);
3768                         journal_mark_dirty(&th, sb, SB_BUFFER_WITH_SB(sb));
3769                         ret = journal_end(&th, sb, 1);
3770                         goto flush_commit_only;
3771                 }
3772
3773                 ret = journal_end_sync(&th, sb, 1);
3774                 if (!ret)
3775                         ret = 1;
3776
3777         } else {
3778                 /* this gets tricky, we have to make sure the journal list in
3779                  * the inode still exists.  We know the list is still around
3780                  * if we've got a larger transaction id than the oldest list
3781                  */
3782               flush_commit_only:
3783                 if (journal_list_still_alive(inode->i_sb, id)) {
3784                         /*
3785                          * we only set ret to 1 when we know for sure
3786                          * the barrier hasn't been started yet on the commit
3787                          * block.
3788                          */
3789                         if (atomic_read(&jl->j_commit_left) > 1)
3790                                 ret = 1;
3791                         flush_commit_list(sb, jl, 1);
3792                         if (journal->j_errno)
3793                                 ret = journal->j_errno;
3794                 }
3795         }
3796         /* otherwise the list is gone, and long since committed */
3797         return ret;
3798 }
3799
3800 int reiserfs_commit_for_inode(struct inode *inode)
3801 {
3802         unsigned long id = REISERFS_I(inode)->i_trans_id;
3803         struct reiserfs_journal_list *jl = REISERFS_I(inode)->i_jl;
3804
3805         /* for the whole inode, assume unset id means it was
3806          * changed in the current transaction.  More conservative
3807          */
3808         if (!id || !jl) {
3809                 reiserfs_update_inode_transaction(inode);
3810                 id = REISERFS_I(inode)->i_trans_id;
3811                 /* jl will be updated in __commit_trans_jl */
3812         }
3813
3814         return __commit_trans_jl(inode, id, jl);
3815 }
3816
3817 void reiserfs_restore_prepared_buffer(struct super_block *p_s_sb,
3818                                       struct buffer_head *bh)
3819 {
3820         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3821         PROC_INFO_INC(p_s_sb, journal.restore_prepared);
3822         if (!bh) {
3823                 return;
3824         }
3825         if (test_clear_buffer_journal_restore_dirty(bh) &&
3826             buffer_journal_dirty(bh)) {
3827                 struct reiserfs_journal_cnode *cn;
3828                 cn = get_journal_hash_dev(p_s_sb,
3829                                           journal->j_list_hash_table,
3830                                           bh->b_blocknr);
3831                 if (cn && can_dirty(cn)) {
3832                         set_buffer_journal_test(bh);
3833                         mark_buffer_dirty(bh);
3834                 }
3835         }
3836         clear_buffer_journal_prepared(bh);
3837 }
3838
3839 extern struct tree_balance *cur_tb;
3840 /*
3841 ** before we can change a metadata block, we have to make sure it won't
3842 ** be written to disk while we are altering it.  So, we must:
3843 ** clean it
3844 ** wait on it.
3845 **
3846 */
3847 int reiserfs_prepare_for_journal(struct super_block *p_s_sb,
3848                                  struct buffer_head *bh, int wait)
3849 {
3850         PROC_INFO_INC(p_s_sb, journal.prepare);
3851
3852         if (test_set_buffer_locked(bh)) {
3853                 if (!wait)
3854                         return 0;
3855                 lock_buffer(bh);
3856         }
3857         set_buffer_journal_prepared(bh);
3858         if (test_clear_buffer_dirty(bh) && buffer_journal_dirty(bh)) {
3859                 clear_buffer_journal_test(bh);
3860                 set_buffer_journal_restore_dirty(bh);
3861         }
3862         unlock_buffer(bh);
3863         return 1;
3864 }
3865
3866 static void flush_old_journal_lists(struct super_block *s)
3867 {
3868         struct reiserfs_journal *journal = SB_JOURNAL(s);
3869         struct reiserfs_journal_list *jl;
3870         struct list_head *entry;
3871         time_t now = get_seconds();
3872
3873         while (!list_empty(&journal->j_journal_list)) {
3874                 entry = journal->j_journal_list.next;
3875                 jl = JOURNAL_LIST_ENTRY(entry);
3876                 /* this check should always be run, to send old lists to disk */
3877                 if (jl->j_timestamp < (now - (JOURNAL_MAX_TRANS_AGE * 4)) &&
3878                     atomic_read(&jl->j_commit_left) == 0 &&
3879                     test_transaction(s, jl)) {
3880                         flush_used_journal_lists(s, jl);
3881                 } else {
3882                         break;
3883                 }
3884         }
3885 }
3886
3887 /*
3888 ** long and ugly.  If flush, will not return until all commit
3889 ** blocks and all real buffers in the trans are on disk.
3890 ** If no_async, won't return until all commit blocks are on disk.
3891 **
3892 ** keep reading, there are comments as you go along
3893 **
3894 ** If the journal is aborted, we just clean up. Things like flushing
3895 ** journal lists, etc just won't happen.
3896 */
3897 static int do_journal_end(struct reiserfs_transaction_handle *th,
3898                           struct super_block *p_s_sb, unsigned long nblocks,
3899                           int flags)
3900 {
3901         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3902         struct reiserfs_journal_cnode *cn, *next, *jl_cn;
3903         struct reiserfs_journal_cnode *last_cn = NULL;
3904         struct reiserfs_journal_desc *desc;
3905         struct reiserfs_journal_commit *commit;
3906         struct buffer_head *c_bh;       /* commit bh */
3907         struct buffer_head *d_bh;       /* desc bh */
3908         int cur_write_start = 0;        /* start index of current log write */
3909         int old_start;
3910         int i;
3911         int flush;
3912         int wait_on_commit;
3913         struct reiserfs_journal_list *jl, *temp_jl;
3914         struct list_head *entry, *safe;
3915         unsigned long jindex;
3916         unsigned long commit_trans_id;
3917         int trans_half;
3918
3919         BUG_ON(th->t_refcount > 1);
3920         BUG_ON(!th->t_trans_id);
3921
3922         /* protect flush_older_commits from doing mistakes if the
3923            transaction ID counter gets overflowed.  */
3924         if (th->t_trans_id == ~0UL)
3925                 flags |= FLUSH_ALL | COMMIT_NOW | WAIT;
3926         flush = flags & FLUSH_ALL;
3927         wait_on_commit = flags & WAIT;
3928
3929         put_fs_excl();
3930         current->journal_info = th->t_handle_save;
3931         reiserfs_check_lock_depth(p_s_sb, "journal end");
3932         if (journal->j_len == 0) {
3933                 reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb),
3934                                              1);
3935                 journal_mark_dirty(th, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb));
3936         }
3937
3938         lock_journal(p_s_sb);
3939         if (journal->j_next_full_flush) {
3940                 flags |= FLUSH_ALL;
3941                 flush = 1;
3942         }
3943         if (journal->j_next_async_flush) {
3944                 flags |= COMMIT_NOW | WAIT;
3945                 wait_on_commit = 1;
3946         }
3947
3948         /* check_journal_end locks the journal, and unlocks if it does not return 1
3949          ** it tells us if we should continue with the journal_end, or just return
3950          */
3951         if (!check_journal_end(th, p_s_sb, nblocks, flags)) {
3952                 p_s_sb->s_dirt = 1;
3953                 wake_queued_writers(p_s_sb);
3954                 reiserfs_async_progress_wait(p_s_sb);
3955                 goto out;
3956         }
3957
3958         /* check_journal_end might set these, check again */
3959         if (journal->j_next_full_flush) {
3960                 flush = 1;
3961         }
3962
3963         /*
3964          ** j must wait means we have to flush the log blocks, and the real blocks for
3965          ** this transaction
3966          */
3967         if (journal->j_must_wait > 0) {
3968                 flush = 1;
3969         }
3970 #ifdef REISERFS_PREALLOCATE
3971         /* quota ops might need to nest, setup the journal_info pointer for them
3972          * and raise the refcount so that it is > 0. */
3973         current->journal_info = th;
3974         th->t_refcount++;
3975         reiserfs_discard_all_prealloc(th);      /* it should not involve new blocks into
3976                                                  * the transaction */
3977         th->t_refcount--;
3978         current->journal_info = th->t_handle_save;
3979 #endif
3980
3981         /* setup description block */
3982         d_bh =
3983             journal_getblk(p_s_sb,
3984                            SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
3985                            journal->j_start);
3986         set_buffer_uptodate(d_bh);
3987         desc = (struct reiserfs_journal_desc *)(d_bh)->b_data;
3988         memset(d_bh->b_data, 0, d_bh->b_size);
3989         memcpy(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8);
3990         set_desc_trans_id(desc, journal->j_trans_id);
3991
3992         /* setup commit block.  Don't write (keep it clean too) this one until after everyone else is written */
3993         c_bh = journal_getblk(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
3994                               ((journal->j_start + journal->j_len +
3995                                 1) % SB_ONDISK_JOURNAL_SIZE(p_s_sb)));
3996         commit = (struct reiserfs_journal_commit *)c_bh->b_data;
3997         memset(c_bh->b_data, 0, c_bh->b_size);
3998         set_commit_trans_id(commit, journal->j_trans_id);
3999         set_buffer_uptodate(c_bh);
4000
4001         /* init this journal list */
4002         jl = journal->j_current_jl;
4003
4004         /* we lock the commit before doing anything because
4005          * we want to make sure nobody tries to run flush_commit_list until
4006          * the new transaction is fully setup, and we've already flushed the
4007          * ordered bh list
4008          */
4009         down(&jl->j_commit_lock);
4010
4011         /* save the transaction id in case we need to commit it later */
4012         commit_trans_id = jl->j_trans_id;
4013
4014         atomic_set(&jl->j_older_commits_done, 0);
4015         jl->j_trans_id = journal->j_trans_id;
4016         jl->j_timestamp = journal->j_trans_start_time;
4017         jl->j_commit_bh = c_bh;
4018         jl->j_start = journal->j_start;
4019         jl->j_len = journal->j_len;
4020         atomic_set(&jl->j_nonzerolen, journal->j_len);
4021         atomic_set(&jl->j_commit_left, journal->j_len + 2);
4022         jl->j_realblock = NULL;
4023
4024         /* The ENTIRE FOR LOOP MUST not cause schedule to occur.
4025          **  for each real block, add it to the journal list hash,
4026          ** copy into real block index array in the commit or desc block
4027          */
4028         trans_half = journal_trans_half(p_s_sb->s_blocksize);
4029         for (i = 0, cn = journal->j_first; cn; cn = cn->next, i++) {
4030                 if (buffer_journaled(cn->bh)) {
4031                         jl_cn = get_cnode(p_s_sb);
4032                         if (!jl_cn) {
4033                                 reiserfs_panic(p_s_sb,
4034                                                "journal-1676, get_cnode returned NULL\n");
4035                         }
4036                         if (i == 0) {
4037                                 jl->j_realblock = jl_cn;
4038                         }
4039                         jl_cn->prev = last_cn;
4040                         jl_cn->next = NULL;
4041                         if (last_cn) {
4042                                 last_cn->next = jl_cn;
4043                         }
4044                         last_cn = jl_cn;
4045                         /* make sure the block we are trying to log is not a block
4046                            of journal or reserved area */
4047
4048                         if (is_block_in_log_or_reserved_area
4049                             (p_s_sb, cn->bh->b_blocknr)) {
4050                                 reiserfs_panic(p_s_sb,
4051                                                "journal-2332: Trying to log block %lu, which is a log block\n",
4052                                                cn->bh->b_blocknr);
4053                         }
4054                         jl_cn->blocknr = cn->bh->b_blocknr;
4055                         jl_cn->state = 0;
4056                         jl_cn->sb = p_s_sb;
4057                         jl_cn->bh = cn->bh;
4058                         jl_cn->jlist = jl;
4059                         insert_journal_hash(journal->j_list_hash_table, jl_cn);
4060                         if (i < trans_half) {
4061                                 desc->j_realblock[i] =
4062                                     cpu_to_le32(cn->bh->b_blocknr);
4063                         } else {
4064                                 commit->j_realblock[i - trans_half] =
4065                                     cpu_to_le32(cn->bh->b_blocknr);
4066                         }
4067                 } else {
4068                         i--;
4069                 }
4070         }
4071         set_desc_trans_len(desc, journal->j_len);
4072         set_desc_mount_id(desc, journal->j_mount_id);
4073         set_desc_trans_id(desc, journal->j_trans_id);
4074         set_commit_trans_len(commit, journal->j_len);
4075
4076         /* special check in case all buffers in the journal were marked for not logging */
4077         if (journal->j_len == 0) {
4078                 BUG();
4079         }
4080
4081         /* we're about to dirty all the log blocks, mark the description block
4082          * dirty now too.  Don't mark the commit block dirty until all the
4083          * others are on disk
4084          */
4085         mark_buffer_dirty(d_bh);
4086
4087         /* first data block is j_start + 1, so add one to cur_write_start wherever you use it */
4088         cur_write_start = journal->j_start;
4089         cn = journal->j_first;
4090         jindex = 1;             /* start at one so we don't get the desc again */
4091         while (cn) {
4092                 clear_buffer_journal_new(cn->bh);
4093                 /* copy all the real blocks into log area.  dirty log blocks */
4094                 if (buffer_journaled(cn->bh)) {
4095                         struct buffer_head *tmp_bh;
4096                         char *addr;
4097                         struct page *page;
4098                         tmp_bh =
4099                             journal_getblk(p_s_sb,
4100                                            SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
4101                                            ((cur_write_start +
4102                                              jindex) %
4103                                             SB_ONDISK_JOURNAL_SIZE(p_s_sb)));
4104                         set_buffer_uptodate(tmp_bh);
4105                         page = cn->bh->b_page;
4106                         addr = kmap(page);
4107                         memcpy(tmp_bh->b_data,
4108                                addr + offset_in_page(cn->bh->b_data),
4109                                cn->bh->b_size);
4110                         kunmap(page);
4111                         mark_buffer_dirty(tmp_bh);
4112                         jindex++;
4113                         set_buffer_journal_dirty(cn->bh);
4114                         clear_buffer_journaled(cn->bh);
4115                 } else {
4116                         /* JDirty cleared sometime during transaction.  don't log this one */
4117                         reiserfs_warning(p_s_sb,
4118                                          "journal-2048: do_journal_end: BAD, buffer in journal hash, but not JDirty!");
4119                         brelse(cn->bh);
4120                 }
4121                 next = cn->next;
4122                 free_cnode(p_s_sb, cn);
4123                 cn = next;
4124                 cond_resched();
4125         }
4126
4127         /* we are done  with both the c_bh and d_bh, but
4128          ** c_bh must be written after all other commit blocks,
4129          ** so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1.
4130          */
4131
4132         journal->j_current_jl = alloc_journal_list(p_s_sb);
4133
4134         /* now it is safe to insert this transaction on the main list */
4135         list_add_tail(&jl->j_list, &journal->j_journal_list);
4136         list_add_tail(&jl->j_working_list, &journal->j_working_list);
4137         journal->j_num_work_lists++;
4138
4139         /* reset journal values for the next transaction */
4140         old_start = journal->j_start;
4141         journal->j_start =
4142             (journal->j_start + journal->j_len +
4143              2) % SB_ONDISK_JOURNAL_SIZE(p_s_sb);
4144         atomic_set(&(journal->j_wcount), 0);
4145         journal->j_bcount = 0;
4146         journal->j_last = NULL;
4147         journal->j_first = NULL;
4148         journal->j_len = 0;
4149         journal->j_trans_start_time = 0;
4150         /* check for trans_id overflow */
4151         if (++journal->j_trans_id == 0)
4152                 journal->j_trans_id = 10;
4153         journal->j_current_jl->j_trans_id = journal->j_trans_id;
4154         journal->j_must_wait = 0;
4155         journal->j_len_alloc = 0;
4156         journal->j_next_full_flush = 0;
4157         journal->j_next_async_flush = 0;
4158         init_journal_hash(p_s_sb);
4159
4160         // make sure reiserfs_add_jh sees the new current_jl before we
4161         // write out the tails
4162         smp_mb();
4163
4164         /* tail conversion targets have to hit the disk before we end the
4165          * transaction.  Otherwise a later transaction might repack the tail
4166          * before this transaction commits, leaving the data block unflushed and
4167          * clean, if we crash before the later transaction commits, the data block
4168          * is lost.
4169          */
4170         if (!list_empty(&jl->j_tail_bh_list)) {
4171                 unlock_kernel();
4172                 write_ordered_buffers(&journal->j_dirty_buffers_lock,
4173                                       journal, jl, &jl->j_tail_bh_list);
4174                 lock_kernel();
4175         }
4176         if (!list_empty(&jl->j_tail_bh_list))
4177                 BUG();
4178         up(&jl->j_commit_lock);
4179
4180         /* honor the flush wishes from the caller, simple commits can
4181          ** be done outside the journal lock, they are done below
4182          **
4183          ** if we don't flush the commit list right now, we put it into
4184          ** the work queue so the people waiting on the async progress work
4185          ** queue don't wait for this proc to flush journal lists and such.
4186          */
4187         if (flush) {
4188                 flush_commit_list(p_s_sb, jl, 1);
4189                 flush_journal_list(p_s_sb, jl, 1);
4190         } else if (!(jl->j_state & LIST_COMMIT_PENDING))
4191                 queue_delayed_work(commit_wq, &journal->j_work, HZ / 10);
4192
4193         /* if the next transaction has any chance of wrapping, flush
4194          ** transactions that might get overwritten.  If any journal lists are very
4195          ** old flush them as well.
4196          */
4197       first_jl:
4198         list_for_each_safe(entry, safe, &journal->j_journal_list) {
4199                 temp_jl = JOURNAL_LIST_ENTRY(entry);
4200                 if (journal->j_start <= temp_jl->j_start) {
4201                         if ((journal->j_start + journal->j_trans_max + 1) >=
4202                             temp_jl->j_start) {
4203                                 flush_used_journal_lists(p_s_sb, temp_jl);
4204                                 goto first_jl;
4205                         } else if ((journal->j_start +
4206                                     journal->j_trans_max + 1) <
4207                                    SB_ONDISK_JOURNAL_SIZE(p_s_sb)) {
4208                                 /* if we don't cross into the next transaction and we don't
4209                                  * wrap, there is no way we can overlap any later transactions
4210                                  * break now
4211                                  */
4212                                 break;
4213                         }
4214                 } else if ((journal->j_start +
4215                             journal->j_trans_max + 1) >
4216                            SB_ONDISK_JOURNAL_SIZE(p_s_sb)) {
4217                         if (((journal->j_start + journal->j_trans_max + 1) %
4218                              SB_ONDISK_JOURNAL_SIZE(p_s_sb)) >=
4219                             temp_jl->j_start) {
4220                                 flush_used_journal_lists(p_s_sb, temp_jl);
4221                                 goto first_jl;
4222                         } else {
4223                                 /* we don't overlap anything from out start to the end of the
4224                                  * log, and our wrapped portion doesn't overlap anything at
4225                                  * the start of the log.  We can break
4226                                  */
4227                                 break;
4228                         }
4229                 }
4230         }
4231         flush_old_journal_lists(p_s_sb);
4232
4233         journal->j_current_jl->j_list_bitmap =
4234             get_list_bitmap(p_s_sb, journal->j_current_jl);
4235
4236         if (!(journal->j_current_jl->j_list_bitmap)) {
4237                 reiserfs_panic(p_s_sb,
4238                                "journal-1996: do_journal_end, could not get a list bitmap\n");
4239         }
4240
4241         atomic_set(&(journal->j_jlock), 0);
4242         unlock_journal(p_s_sb);
4243         /* wake up any body waiting to join. */
4244         clear_bit(J_WRITERS_QUEUED, &journal->j_state);
4245         wake_up(&(journal->j_join_wait));
4246
4247         if (!flush && wait_on_commit &&
4248             journal_list_still_alive(p_s_sb, commit_trans_id)) {
4249                 flush_commit_list(p_s_sb, jl, 1);
4250         }
4251       out:
4252         reiserfs_check_lock_depth(p_s_sb, "journal end2");
4253
4254         memset(th, 0, sizeof(*th));
4255         /* Re-set th->t_super, so we can properly keep track of how many
4256          * persistent transactions there are. We need to do this so if this
4257          * call is part of a failed restart_transaction, we can free it later */
4258         th->t_super = p_s_sb;
4259
4260         return journal->j_errno;
4261 }
4262
4263 static void __reiserfs_journal_abort_hard(struct super_block *sb)
4264 {
4265         struct reiserfs_journal *journal = SB_JOURNAL(sb);
4266         if (test_bit(J_ABORTED, &journal->j_state))
4267                 return;
4268
4269         printk(KERN_CRIT "REISERFS: Aborting journal for filesystem on %s\n",
4270                reiserfs_bdevname(sb));
4271
4272         sb->s_flags |= MS_RDONLY;
4273         set_bit(J_ABORTED, &journal->j_state);
4274
4275 #ifdef CONFIG_REISERFS_CHECK
4276         dump_stack();
4277 #endif
4278 }
4279
4280 static void __reiserfs_journal_abort_soft(struct super_block *sb, int errno)
4281 {
4282         struct reiserfs_journal *journal = SB_JOURNAL(sb);
4283         if (test_bit(J_ABORTED, &journal->j_state))
4284                 return;
4285
4286         if (!journal->j_errno)
4287                 journal->j_errno = errno;
4288
4289         __reiserfs_journal_abort_hard(sb);
4290 }
4291
4292 void reiserfs_journal_abort(struct super_block *sb, int errno)
4293 {
4294         return __reiserfs_journal_abort_soft(sb, errno);
4295 }