fs/reiserfs/journal.c

   1 /*
   2 ** Write ahead logging implementation copyright Chris Mason 2000
   3 **
   4 ** The background commits make this code very interelated, and
   5 ** overly complex.  I need to rethink things a bit....The major players:
   6 **
   7 ** journal_begin -- call with the number of blocks you expect to log.
   8 **                  If the current transaction is too
   9 **                  old, it will block until the current transaction is
  10 **                  finished, and then start a new one.
  11 **                  Usually, your transaction will get joined in with
  12 **                  previous ones for speed.
  13 **
  14 ** journal_join  -- same as journal_begin, but won't block on the current
  15 **                  transaction regardless of age.  Don't ever call
  16 **                  this.  Ever.  There are only two places it should be
  17 **                  called from, and they are both inside this file.
  18 **
  19 ** journal_mark_dirty -- adds blocks into this transaction.  clears any flags
  20 **                       that might make them get sent to disk
  21 **                       and then marks them BH_JDirty.  Puts the buffer head
  22 **                       into the current transaction hash.
  23 **
  24 ** journal_end -- if the current transaction is batchable, it does nothing
  25 **                   otherwise, it could do an async/synchronous commit, or
  26 **                   a full flush of all log and real blocks in the
  27 **                   transaction.
  28 **
  29 ** flush_old_commits -- if the current transaction is too old, it is ended and
  30 **                      commit blocks are sent to disk.  Forces commit blocks
  31 **                      to disk for all backgrounded commits that have been
  32 **                      around too long.
  33 **                   -- Note, if you call this as an immediate flush from
  34 **                      from within kupdate, it will ignore the immediate flag
  35 */
  36
  37 #include <linux/time.h>
  38 #include <linux/semaphore.h>
  39 #include <linux/vmalloc.h>
  40 #include <linux/reiserfs_fs.h>
  41 #include <linux/kernel.h>
  42 #include <linux/errno.h>
  43 #include <linux/fcntl.h>
  44 #include <linux/stat.h>
  45 #include <linux/string.h>
  46 #include <linux/smp_lock.h>
  47 #include <linux/buffer_head.h>
  48 #include <linux/workqueue.h>
  49 #include <linux/writeback.h>
  50 #include <linux/blkdev.h>
  51 #include <linux/backing-dev.h>
  52 #include <linux/uaccess.h>
  53 #include <linux/slab.h>
  54
  55 #include <asm/system.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 *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(struct work_struct *work);
 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 *sb,
 118                               unsigned long nblocks, int join);
 119
 120 static void init_journal_hash(struct super_block *sb)
 121 {
 122         struct reiserfs_journal *journal = SB_JOURNAL(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                                                          *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(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 *sb)
 169 {
 170         struct reiserfs_journal *journal = SB_JOURNAL(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, sb->s_blocksize);
 181                 journal->j_free_bitmap_nodes--;
 182                 return bn;
 183         }
 184         bn = allocate_bitmap_node(sb);
 185         if (!bn) {
 186                 yield();
 187                 goto repeat;
 188         }
 189         return bn;
 190 }
 191 static inline void free_bitmap_node(struct super_block *sb,
 192                                     struct reiserfs_bitmap_node *bn)
 193 {
 194         struct reiserfs_journal *journal = SB_JOURNAL(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 *sb)
 206 {
 207         int i;
 208         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 209         struct reiserfs_bitmap_node *bn = NULL;
 210         for (i = 0; i < REISERFS_MIN_BITMAP_NODES; i++) {
 211                 bn = allocate_bitmap_node(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 *sb,
 222                                   b_blocknr_t block,
 223                                   struct reiserfs_list_bitmap *jb)
 224 {
 225         unsigned int bmap_nr = block / (sb->s_blocksize << 3);
 226         unsigned int bit_nr = block % (sb->s_blocksize << 3);
 227
 228         if (!jb->bitmaps[bmap_nr]) {
 229                 jb->bitmaps[bmap_nr] = get_bitmap_node(sb);
 230         }
 231         set_bit(bit_nr, (unsigned long *)jb->bitmaps[bmap_nr]->data);
 232         return 0;
 233 }
 234
 235 static void cleanup_bitmap_list(struct super_block *sb,
 236                                 struct reiserfs_list_bitmap *jb)
 237 {
 238         int i;
 239         if (jb->bitmaps == NULL)
 240                 return;
 241
 242         for (i = 0; i < reiserfs_bmap_count(sb); i++) {
 243                 if (jb->bitmaps[i]) {
 244                         free_bitmap_node(sb, jb->bitmaps[i]);
 245                         jb->bitmaps[i] = NULL;
 246                 }
 247         }
 248 }
 249
 250 /*
 251 ** only call this on FS unmount.
 252 */
 253 static int free_list_bitmaps(struct super_block *sb,
 254                              struct reiserfs_list_bitmap *jb_array)
 255 {
 256         int i;
 257         struct reiserfs_list_bitmap *jb;
 258         for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
 259                 jb = jb_array + i;
 260                 jb->journal_list = NULL;
 261                 cleanup_bitmap_list(sb, jb);
 262                 vfree(jb->bitmaps);
 263                 jb->bitmaps = NULL;
 264         }
 265         return 0;
 266 }
 267
 268 static int free_bitmap_nodes(struct super_block *sb)
 269 {
 270         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 271         struct list_head *next = journal->j_bitmap_nodes.next;
 272         struct reiserfs_bitmap_node *bn;
 273
 274         while (next != &journal->j_bitmap_nodes) {
 275                 bn = list_entry(next, struct reiserfs_bitmap_node, list);
 276                 list_del(next);
 277                 kfree(bn->data);
 278                 kfree(bn);
 279                 next = journal->j_bitmap_nodes.next;
 280                 journal->j_free_bitmap_nodes--;
 281         }
 282
 283         return 0;
 284 }
 285
 286 /*
 287 ** get memory for JOURNAL_NUM_BITMAPS worth of bitmaps.
 288 ** jb_array is the array to be filled in.
 289 */
 290 int reiserfs_allocate_list_bitmaps(struct super_block *sb,
 291                                    struct reiserfs_list_bitmap *jb_array,
 292                                    unsigned int bmap_nr)
 293 {
 294         int i;
 295         int failed = 0;
 296         struct reiserfs_list_bitmap *jb;
 297         int mem = bmap_nr * sizeof(struct reiserfs_bitmap_node *);
 298
 299         for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
 300                 jb = jb_array + i;
 301                 jb->journal_list = NULL;
 302                 jb->bitmaps = vmalloc(mem);
 303                 if (!jb->bitmaps) {
 304                         reiserfs_warning(sb, "clm-2000", "unable to "
 305                                          "allocate bitmaps for journal lists");
 306                         failed = 1;
 307                         break;
 308                 }
 309                 memset(jb->bitmaps, 0, mem);
 310         }
 311         if (failed) {
 312                 free_list_bitmaps(sb, jb_array);
 313                 return -1;
 314         }
 315         return 0;
 316 }
 317
 318 /*
 319 ** find an available list bitmap.  If you can't find one, flush a commit list
 320 ** and try again
 321 */
 322 static struct reiserfs_list_bitmap *get_list_bitmap(struct super_block *sb,
 323                                                     struct reiserfs_journal_list
 324                                                     *jl)
 325 {
 326         int i, j;
 327         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 328         struct reiserfs_list_bitmap *jb = NULL;
 329
 330         for (j = 0; j < (JOURNAL_NUM_BITMAPS * 3); j++) {
 331                 i = journal->j_list_bitmap_index;
 332                 journal->j_list_bitmap_index = (i + 1) % JOURNAL_NUM_BITMAPS;
 333                 jb = journal->j_list_bitmap + i;
 334                 if (journal->j_list_bitmap[i].journal_list) {
 335                         flush_commit_list(sb,
 336                                           journal->j_list_bitmap[i].
 337                                           journal_list, 1);
 338                         if (!journal->j_list_bitmap[i].journal_list) {
 339                                 break;
 340                         }
 341                 } else {
 342                         break;
 343                 }
 344         }
 345         if (jb->journal_list) { /* double check to make sure if flushed correctly */
 346                 return NULL;
 347         }
 348         jb->journal_list = jl;
 349         return jb;
 350 }
 351
 352 /*
 353 ** allocates a new chunk of X nodes, and links them all together as a list.
 354 ** Uses the cnode->next and cnode->prev pointers
 355 ** returns NULL on failure
 356 */
 357 static struct reiserfs_journal_cnode *allocate_cnodes(int num_cnodes)
 358 {
 359         struct reiserfs_journal_cnode *head;
 360         int i;
 361         if (num_cnodes <= 0) {
 362                 return NULL;
 363         }
 364         head = vmalloc(num_cnodes * sizeof(struct reiserfs_journal_cnode));
 365         if (!head) {
 366                 return NULL;
 367         }
 368         memset(head, 0, num_cnodes * sizeof(struct reiserfs_journal_cnode));
 369         head[0].prev = NULL;
 370         head[0].next = head + 1;
 371         for (i = 1; i < num_cnodes; i++) {
 372                 head[i].prev = head + (i - 1);
 373                 head[i].next = head + (i + 1);  /* if last one, overwrite it after the if */
 374         }
 375         head[num_cnodes - 1].next = NULL;
 376         return head;
 377 }
 378
 379 /*
 380 ** pulls a cnode off the free list, or returns NULL on failure
 381 */
 382 static struct reiserfs_journal_cnode *get_cnode(struct super_block *sb)
 383 {
 384         struct reiserfs_journal_cnode *cn;
 385         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 386
 387         reiserfs_check_lock_depth(sb, "get_cnode");
 388
 389         if (journal->j_cnode_free <= 0) {
 390                 return NULL;
 391         }
 392         journal->j_cnode_used++;
 393         journal->j_cnode_free--;
 394         cn = journal->j_cnode_free_list;
 395         if (!cn) {
 396                 return cn;
 397         }
 398         if (cn->next) {
 399                 cn->next->prev = NULL;
 400         }
 401         journal->j_cnode_free_list = cn->next;
 402         memset(cn, 0, sizeof(struct reiserfs_journal_cnode));
 403         return cn;
 404 }
 405
 406 /*
 407 ** returns a cnode to the free list
 408 */
 409 static void free_cnode(struct super_block *sb,
 410                        struct reiserfs_journal_cnode *cn)
 411 {
 412         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 413
 414         reiserfs_check_lock_depth(sb, "free_cnode");
 415
 416         journal->j_cnode_used--;
 417         journal->j_cnode_free++;
 418         /* memset(cn, 0, sizeof(struct reiserfs_journal_cnode)) ; */
 419         cn->next = journal->j_cnode_free_list;
 420         if (journal->j_cnode_free_list) {
 421                 journal->j_cnode_free_list->prev = cn;
 422         }
 423         cn->prev = NULL;        /* not needed with the memset, but I might kill the memset, and forget to do this */
 424         journal->j_cnode_free_list = cn;
 425 }
 426
 427 static void clear_prepared_bits(struct buffer_head *bh)
 428 {
 429         clear_buffer_journal_prepared(bh);
 430         clear_buffer_journal_restore_dirty(bh);
 431 }
 432
 433 /* return a cnode with same dev, block number and size in table, or null if not found */
 434 static inline struct reiserfs_journal_cnode *get_journal_hash_dev(struct
 435                                                                   super_block
 436                                                                   *sb,
 437                                                                   struct
 438                                                                   reiserfs_journal_cnode
 439                                                                   **table,
 440                                                                   long bl)
 441 {
 442         struct reiserfs_journal_cnode *cn;
 443         cn = journal_hash(table, sb, bl);
 444         while (cn) {
 445                 if (cn->blocknr == bl && cn->sb == sb)
 446                         return cn;
 447                 cn = cn->hnext;
 448         }
 449         return (struct reiserfs_journal_cnode *)0;
 450 }
 451
 452 /*
 453 ** this actually means 'can this block be reallocated yet?'.  If you set search_all, a block can only be allocated
 454 ** if it is not in the current transaction, was not freed by the current transaction, and has no chance of ever
 455 ** being overwritten by a replay after crashing.
 456 **
 457 ** If you don't set search_all, a block can only be allocated if it is not in the current transaction.  Since deleting
 458 ** a block removes it from the current transaction, this case should never happen.  If you don't set search_all, make
 459 ** sure you never write the block without logging it.
 460 **
 461 ** next_zero_bit is a suggestion about the next block to try for find_forward.
 462 ** when bl is rejected because it is set in a journal list bitmap, we search
 463 ** for the next zero bit in the bitmap that rejected bl.  Then, we return that
 464 ** through next_zero_bit for find_forward to try.
 465 **
 466 ** Just because we return something in next_zero_bit does not mean we won't
 467 ** reject it on the next call to reiserfs_in_journal
 468 **
 469 */
 470 int reiserfs_in_journal(struct super_block *sb,
 471                         unsigned int bmap_nr, int bit_nr, int search_all,
 472                         b_blocknr_t * next_zero_bit)
 473 {
 474         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 475         struct reiserfs_journal_cnode *cn;
 476         struct reiserfs_list_bitmap *jb;
 477         int i;
 478         unsigned long bl;
 479
 480         *next_zero_bit = 0;     /* always start this at zero. */
 481
 482         PROC_INFO_INC(sb, journal.in_journal);
 483         /* If we aren't doing a search_all, this is a metablock, and it will be logged before use.
 484          ** if we crash before the transaction that freed it commits,  this transaction won't
 485          ** have committed either, and the block will never be written
 486          */
 487         if (search_all) {
 488                 for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
 489                         PROC_INFO_INC(sb, journal.in_journal_bitmap);
 490                         jb = journal->j_list_bitmap + i;
 491                         if (jb->journal_list && jb->bitmaps[bmap_nr] &&
 492                             test_bit(bit_nr,
 493                                      (unsigned long *)jb->bitmaps[bmap_nr]->
 494                                      data)) {
 495                                 *next_zero_bit =
 496                                     find_next_zero_bit((unsigned long *)
 497                                                        (jb->bitmaps[bmap_nr]->
 498                                                         data),
 499                                                        sb->s_blocksize << 3,
 500                                                        bit_nr + 1);
 501                                 return 1;
 502                         }
 503                 }
 504         }
 505
 506         bl = bmap_nr * (sb->s_blocksize << 3) + bit_nr;
 507         /* is it in any old transactions? */
 508         if (search_all
 509             && (cn =
 510                 get_journal_hash_dev(sb, journal->j_list_hash_table, bl))) {
 511                 return 1;
 512         }
 513
 514         /* is it in the current transaction.  This should never happen */
 515         if ((cn = get_journal_hash_dev(sb, journal->j_hash_table, bl))) {
 516                 BUG();
 517                 return 1;
 518         }
 519
 520         PROC_INFO_INC(sb, journal.in_journal_reusable);
 521         /* safe for reuse */
 522         return 0;
 523 }
 524
 525 /* insert cn into table
 526 */
 527 static inline void insert_journal_hash(struct reiserfs_journal_cnode **table,
 528                                        struct reiserfs_journal_cnode *cn)
 529 {
 530         struct reiserfs_journal_cnode *cn_orig;
 531
 532         cn_orig = journal_hash(table, cn->sb, cn->blocknr);
 533         cn->hnext = cn_orig;
 534         cn->hprev = NULL;
 535         if (cn_orig) {
 536                 cn_orig->hprev = cn;
 537         }
 538         journal_hash(table, cn->sb, cn->blocknr) = cn;
 539 }
 540
 541 /* lock the current transaction */
 542 static inline void lock_journal(struct super_block *sb)
 543 {
 544         PROC_INFO_INC(sb, journal.lock_journal);
 545
 546         reiserfs_mutex_lock_safe(&SB_JOURNAL(sb)->j_mutex, sb);
 547 }
 548
 549 /* unlock the current transaction */
 550 static inline void unlock_journal(struct super_block *sb)
 551 {
 552         mutex_unlock(&SB_JOURNAL(sb)->j_mutex);
 553 }
 554
 555 static inline void get_journal_list(struct reiserfs_journal_list *jl)
 556 {
 557         jl->j_refcount++;
 558 }
 559
 560 static inline void put_journal_list(struct super_block *s,
 561                                     struct reiserfs_journal_list *jl)
 562 {
 563         if (jl->j_refcount < 1) {
 564                 reiserfs_panic(s, "journal-2", "trans id %u, refcount at %d",
 565                                jl->j_trans_id, jl->j_refcount);
 566         }
 567         if (--jl->j_refcount == 0)
 568                 kfree(jl);
 569 }
 570
 571 /*
 572 ** this used to be much more involved, and I'm keeping it just in case things get ugly again.
 573 ** it gets called by flush_commit_list, and cleans up any data stored about blocks freed during a
 574 ** transaction.
 575 */
 576 static void cleanup_freed_for_journal_list(struct super_block *sb,
 577                                            struct reiserfs_journal_list *jl)
 578 {
 579
 580         struct reiserfs_list_bitmap *jb = jl->j_list_bitmap;
 581         if (jb) {
 582                 cleanup_bitmap_list(sb, jb);
 583         }
 584         jl->j_list_bitmap->journal_list = NULL;
 585         jl->j_list_bitmap = NULL;
 586 }
 587
 588 static int journal_list_still_alive(struct super_block *s,
 589                                     unsigned int trans_id)
 590 {
 591         struct reiserfs_journal *journal = SB_JOURNAL(s);
 592         struct list_head *entry = &journal->j_journal_list;
 593         struct reiserfs_journal_list *jl;
 594
 595         if (!list_empty(entry)) {
 596                 jl = JOURNAL_LIST_ENTRY(entry->next);
 597                 if (jl->j_trans_id <= trans_id) {
 598                         return 1;
 599                 }
 600         }
 601         return 0;
 602 }
 603
 604 /*
 605  * If page->mapping was null, we failed to truncate this page for
 606  * some reason.  Most likely because it was truncated after being
 607  * logged via data=journal.
 608  *
 609  * This does a check to see if the buffer belongs to one of these
 610  * lost pages before doing the final put_bh.  If page->mapping was
 611  * null, it tries to free buffers on the page, which should make the
 612  * final page_cache_release drop the page from the lru.
 613  */
 614 static void release_buffer_page(struct buffer_head *bh)
 615 {
 616         struct page *page = bh->b_page;
 617         if (!page->mapping && trylock_page(page)) {
 618                 page_cache_get(page);
 619                 put_bh(bh);
 620                 if (!page->mapping)
 621                         try_to_free_buffers(page);
 622                 unlock_page(page);
 623                 page_cache_release(page);
 624         } else {
 625                 put_bh(bh);
 626         }
 627 }
 628
 629 static void reiserfs_end_buffer_io_sync(struct buffer_head *bh, int uptodate)
 630 {
 631         char b[BDEVNAME_SIZE];
 632
 633         if (buffer_journaled(bh)) {
 634                 reiserfs_warning(NULL, "clm-2084",
 635                                  "pinned buffer %lu:%s sent to disk",
 636                                  bh->b_blocknr, bdevname(bh->b_bdev, b));
 637         }
 638         if (uptodate)
 639                 set_buffer_uptodate(bh);
 640         else
 641                 clear_buffer_uptodate(bh);
 642
 643         unlock_buffer(bh);
 644         release_buffer_page(bh);
 645 }
 646
 647 static void reiserfs_end_ordered_io(struct buffer_head *bh, int uptodate)
 648 {
 649         if (uptodate)
 650                 set_buffer_uptodate(bh);
 651         else
 652                 clear_buffer_uptodate(bh);
 653         unlock_buffer(bh);
 654         put_bh(bh);
 655 }
 656
 657 static void submit_logged_buffer(struct buffer_head *bh)
 658 {
 659         get_bh(bh);
 660         bh->b_end_io = reiserfs_end_buffer_io_sync;
 661         clear_buffer_journal_new(bh);
 662         clear_buffer_dirty(bh);
 663         if (!test_clear_buffer_journal_test(bh))
 664                 BUG();
 665         if (!buffer_uptodate(bh))
 666                 BUG();
 667         submit_bh(WRITE, bh);
 668 }
 669
 670 static void submit_ordered_buffer(struct buffer_head *bh)
 671 {
 672         get_bh(bh);
 673         bh->b_end_io = reiserfs_end_ordered_io;
 674         clear_buffer_dirty(bh);
 675         if (!buffer_uptodate(bh))
 676                 BUG();
 677         submit_bh(WRITE, bh);
 678 }
 679
 680 static int submit_barrier_buffer(struct buffer_head *bh)
 681 {
 682         get_bh(bh);
 683         bh->b_end_io = reiserfs_end_ordered_io;
 684         clear_buffer_dirty(bh);
 685         if (!buffer_uptodate(bh))
 686                 BUG();
 687         return submit_bh(WRITE_BARRIER, bh);
 688 }
 689
 690 static void check_barrier_completion(struct super_block *s,
 691                                      struct buffer_head *bh)
 692 {
 693         if (buffer_eopnotsupp(bh)) {
 694                 clear_buffer_eopnotsupp(bh);
 695                 disable_barrier(s);
 696                 set_buffer_uptodate(bh);
 697                 set_buffer_dirty(bh);
 698                 reiserfs_write_unlock(s);
 699                 sync_dirty_buffer(bh);
 700                 reiserfs_write_lock(s);
 701         }
 702 }
 703
 704 #define CHUNK_SIZE 32
 705 struct buffer_chunk {
 706         struct buffer_head *bh[CHUNK_SIZE];
 707         int nr;
 708 };
 709
 710 static void write_chunk(struct buffer_chunk *chunk)
 711 {
 712         int i;
 713         get_fs_excl();
 714         for (i = 0; i < chunk->nr; i++) {
 715                 submit_logged_buffer(chunk->bh[i]);
 716         }
 717         chunk->nr = 0;
 718         put_fs_excl();
 719 }
 720
 721 static void write_ordered_chunk(struct buffer_chunk *chunk)
 722 {
 723         int i;
 724         get_fs_excl();
 725         for (i = 0; i < chunk->nr; i++) {
 726                 submit_ordered_buffer(chunk->bh[i]);
 727         }
 728         chunk->nr = 0;
 729         put_fs_excl();
 730 }
 731
 732 static int add_to_chunk(struct buffer_chunk *chunk, struct buffer_head *bh,
 733                         spinlock_t * lock, void (fn) (struct buffer_chunk *))
 734 {
 735         int ret = 0;
 736         BUG_ON(chunk->nr >= CHUNK_SIZE);
 737         chunk->bh[chunk->nr++] = bh;
 738         if (chunk->nr >= CHUNK_SIZE) {
 739                 ret = 1;
 740                 if (lock)
 741                         spin_unlock(lock);
 742                 fn(chunk);
 743                 if (lock)
 744                         spin_lock(lock);
 745         }
 746         return ret;
 747 }
 748
 749 static atomic_t nr_reiserfs_jh = ATOMIC_INIT(0);
 750 static struct reiserfs_jh *alloc_jh(void)
 751 {
 752         struct reiserfs_jh *jh;
 753         while (1) {
 754                 jh = kmalloc(sizeof(*jh), GFP_NOFS);
 755                 if (jh) {
 756                         atomic_inc(&nr_reiserfs_jh);
 757                         return jh;
 758                 }
 759                 yield();
 760         }
 761 }
 762
 763 /*
 764  * we want to free the jh when the buffer has been written
 765  * and waited on
 766  */
 767 void reiserfs_free_jh(struct buffer_head *bh)
 768 {
 769         struct reiserfs_jh *jh;
 770
 771         jh = bh->b_private;
 772         if (jh) {
 773                 bh->b_private = NULL;
 774                 jh->bh = NULL;
 775                 list_del_init(&jh->list);
 776                 kfree(jh);
 777                 if (atomic_read(&nr_reiserfs_jh) <= 0)
 778                         BUG();
 779                 atomic_dec(&nr_reiserfs_jh);
 780                 put_bh(bh);
 781         }
 782 }
 783
 784 static inline int __add_jh(struct reiserfs_journal *j, struct buffer_head *bh,
 785                            int tail)
 786 {
 787         struct reiserfs_jh *jh;
 788
 789         if (bh->b_private) {
 790                 spin_lock(&j->j_dirty_buffers_lock);
 791                 if (!bh->b_private) {
 792                         spin_unlock(&j->j_dirty_buffers_lock);
 793                         goto no_jh;
 794                 }
 795                 jh = bh->b_private;
 796                 list_del_init(&jh->list);
 797         } else {
 798               no_jh:
 799                 get_bh(bh);
 800                 jh = alloc_jh();
 801                 spin_lock(&j->j_dirty_buffers_lock);
 802                 /* buffer must be locked for __add_jh, should be able to have
 803                  * two adds at the same time
 804                  */
 805                 BUG_ON(bh->b_private);
 806                 jh->bh = bh;
 807                 bh->b_private = jh;
 808         }
 809         jh->jl = j->j_current_jl;
 810         if (tail)
 811                 list_add_tail(&jh->list, &jh->jl->j_tail_bh_list);
 812         else {
 813                 list_add_tail(&jh->list, &jh->jl->j_bh_list);
 814         }
 815         spin_unlock(&j->j_dirty_buffers_lock);
 816         return 0;
 817 }
 818
 819 int reiserfs_add_tail_list(struct inode *inode, struct buffer_head *bh)
 820 {
 821         return __add_jh(SB_JOURNAL(inode->i_sb), bh, 1);
 822 }
 823 int reiserfs_add_ordered_list(struct inode *inode, struct buffer_head *bh)
 824 {
 825         return __add_jh(SB_JOURNAL(inode->i_sb), bh, 0);
 826 }
 827
 828 #define JH_ENTRY(l) list_entry((l), struct reiserfs_jh, list)
 829 static int write_ordered_buffers(spinlock_t * lock,
 830                                  struct reiserfs_journal *j,
 831                                  struct reiserfs_journal_list *jl,
 832                                  struct list_head *list)
 833 {
 834         struct buffer_head *bh;
 835         struct reiserfs_jh *jh;
 836         int ret = j->j_errno;
 837         struct buffer_chunk chunk;
 838         struct list_head tmp;
 839         INIT_LIST_HEAD(&tmp);
 840
 841         chunk.nr = 0;
 842         spin_lock(lock);
 843         while (!list_empty(list)) {
 844                 jh = JH_ENTRY(list->next);
 845                 bh = jh->bh;
 846                 get_bh(bh);
 847                 if (!trylock_buffer(bh)) {
 848                         if (!buffer_dirty(bh)) {
 849                                 list_move(&jh->list, &tmp);
 850                                 goto loop_next;
 851                         }
 852                         spin_unlock(lock);
 853                         if (chunk.nr)
 854                                 write_ordered_chunk(&chunk);
 855                         wait_on_buffer(bh);
 856                         cond_resched();
 857                         spin_lock(lock);
 858                         goto loop_next;
 859                 }
 860                 /* in theory, dirty non-uptodate buffers should never get here,
 861                  * but the upper layer io error paths still have a few quirks.
 862                  * Handle them here as gracefully as we can
 863                  */
 864                 if (!buffer_uptodate(bh) && buffer_dirty(bh)) {
 865                         clear_buffer_dirty(bh);
 866                         ret = -EIO;
 867                 }
 868                 if (buffer_dirty(bh)) {
 869                         list_move(&jh->list, &tmp);
 870                         add_to_chunk(&chunk, bh, lock, write_ordered_chunk);
 871                 } else {
 872                         reiserfs_free_jh(bh);
 873                         unlock_buffer(bh);
 874                 }
 875               loop_next:
 876                 put_bh(bh);
 877                 cond_resched_lock(lock);
 878         }
 879         if (chunk.nr) {
 880                 spin_unlock(lock);
 881                 write_ordered_chunk(&chunk);
 882                 spin_lock(lock);
 883         }
 884         while (!list_empty(&tmp)) {
 885                 jh = JH_ENTRY(tmp.prev);
 886                 bh = jh->bh;
 887                 get_bh(bh);
 888                 reiserfs_free_jh(bh);
 889
 890                 if (buffer_locked(bh)) {
 891                         spin_unlock(lock);
 892                         wait_on_buffer(bh);
 893                         spin_lock(lock);
 894                 }
 895                 if (!buffer_uptodate(bh)) {
 896                         ret = -EIO;
 897                 }
 898                 /* ugly interaction with invalidatepage here.
 899                  * reiserfs_invalidate_page will pin any buffer that has a valid
 900                  * journal head from an older transaction.  If someone else sets
 901                  * our buffer dirty after we write it in the first loop, and
 902                  * then someone truncates the page away, nobody will ever write
 903                  * the buffer. We're safe if we write the page one last time
 904                  * after freeing the journal header.
 905                  */
 906                 if (buffer_dirty(bh) && unlikely(bh->b_page->mapping == NULL)) {
 907                         spin_unlock(lock);
 908                         ll_rw_block(WRITE, 1, &bh);
 909                         spin_lock(lock);
 910                 }
 911                 put_bh(bh);
 912                 cond_resched_lock(lock);
 913         }
 914         spin_unlock(lock);
 915         return ret;
 916 }
 917
 918 static int flush_older_commits(struct super_block *s,
 919                                struct reiserfs_journal_list *jl)
 920 {
 921         struct reiserfs_journal *journal = SB_JOURNAL(s);
 922         struct reiserfs_journal_list *other_jl;
 923         struct reiserfs_journal_list *first_jl;
 924         struct list_head *entry;
 925         unsigned int trans_id = jl->j_trans_id;
 926         unsigned int other_trans_id;
 927         unsigned int first_trans_id;
 928
 929       find_first:
 930         /*
 931          * first we walk backwards to find the oldest uncommitted transation
 932          */
 933         first_jl = jl;
 934         entry = jl->j_list.prev;
 935         while (1) {
 936                 other_jl = JOURNAL_LIST_ENTRY(entry);
 937                 if (entry == &journal->j_journal_list ||
 938                     atomic_read(&other_jl->j_older_commits_done))
 939                         break;
 940
 941                 first_jl = other_jl;
 942                 entry = other_jl->j_list.prev;
 943         }
 944
 945         /* if we didn't find any older uncommitted transactions, return now */
 946         if (first_jl == jl) {
 947                 return 0;
 948         }
 949
 950         first_trans_id = first_jl->j_trans_id;
 951
 952         entry = &first_jl->j_list;
 953         while (1) {
 954                 other_jl = JOURNAL_LIST_ENTRY(entry);
 955                 other_trans_id = other_jl->j_trans_id;
 956
 957                 if (other_trans_id < trans_id) {
 958                         if (atomic_read(&other_jl->j_commit_left) != 0) {
 959                                 flush_commit_list(s, other_jl, 0);
 960
 961                                 /* list we were called with is gone, return */
 962                                 if (!journal_list_still_alive(s, trans_id))
 963                                         return 1;
 964
 965                                 /* the one we just flushed is gone, this means all
 966                                  * older lists are also gone, so first_jl is no longer
 967                                  * valid either.  Go back to the beginning.
 968                                  */
 969                                 if (!journal_list_still_alive
 970                                     (s, other_trans_id)) {
 971                                         goto find_first;
 972                                 }
 973                         }
 974                         entry = entry->next;
 975                         if (entry == &journal->j_journal_list)
 976                                 return 0;
 977                 } else {
 978                         return 0;
 979                 }
 980         }
 981         return 0;
 982 }
 983
 984 static int reiserfs_async_progress_wait(struct super_block *s)
 985 {
 986         DEFINE_WAIT(wait);
 987         struct reiserfs_journal *j = SB_JOURNAL(s);
 988
 989         if (atomic_read(&j->j_async_throttle)) {
 990                 reiserfs_write_unlock(s);
 991                 congestion_wait(BLK_RW_ASYNC, HZ / 10);
 992                 reiserfs_write_lock(s);
 993         }
 994
 995         return 0;
 996 }
 997
 998 /*
 999 ** if this journal list still has commit blocks unflushed, send them to disk.
1000 **
1001 ** log areas must be flushed in order (transaction 2 can't commit before transaction 1)
1002 ** Before the commit block can by written, every other log block must be safely on disk
1003 **
1004 */
1005 static int flush_commit_list(struct super_block *s,
1006                              struct reiserfs_journal_list *jl, int flushall)
1007 {
1008         int i;
1009         b_blocknr_t bn;
1010         struct buffer_head *tbh = NULL;
1011         unsigned int trans_id = jl->j_trans_id;
1012         struct reiserfs_journal *journal = SB_JOURNAL(s);
1013         int barrier = 0;
1014         int retval = 0;
1015         int write_len;
1016
1017         reiserfs_check_lock_depth(s, "flush_commit_list");
1018
1019         if (atomic_read(&jl->j_older_commits_done)) {
1020                 return 0;
1021         }
1022
1023         get_fs_excl();
1024
1025         /* before we can put our commit blocks on disk, we have to make sure everyone older than
1026          ** us is on disk too
1027          */
1028         BUG_ON(jl->j_len <= 0);
1029         BUG_ON(trans_id == journal->j_trans_id);
1030
1031         get_journal_list(jl);
1032         if (flushall) {
1033                 if (flush_older_commits(s, jl) == 1) {
1034                         /* list disappeared during flush_older_commits.  return */
1035                         goto put_jl;
1036                 }
1037         }
1038
1039         /* make sure nobody is trying to flush this one at the same time */
1040         reiserfs_mutex_lock_safe(&jl->j_commit_mutex, s);
1041
1042         if (!journal_list_still_alive(s, trans_id)) {
1043                 mutex_unlock(&jl->j_commit_mutex);
1044                 goto put_jl;
1045         }
1046         BUG_ON(jl->j_trans_id == 0);
1047
1048         /* this commit is done, exit */
1049         if (atomic_read(&(jl->j_commit_left)) <= 0) {
1050                 if (flushall) {
1051                         atomic_set(&(jl->j_older_commits_done), 1);
1052                 }
1053                 mutex_unlock(&jl->j_commit_mutex);
1054                 goto put_jl;
1055         }
1056
1057         if (!list_empty(&jl->j_bh_list)) {
1058                 int ret;
1059
1060                 /*
1061                  * We might sleep in numerous places inside
1062                  * write_ordered_buffers. Relax the write lock.
1063                  */
1064                 reiserfs_write_unlock(s);
1065                 ret = write_ordered_buffers(&journal->j_dirty_buffers_lock,
1066                                             journal, jl, &jl->j_bh_list);
1067                 if (ret < 0 && retval == 0)
1068                         retval = ret;
1069                 reiserfs_write_lock(s);
1070         }
1071         BUG_ON(!list_empty(&jl->j_bh_list));
1072         /*
1073          * for the description block and all the log blocks, submit any buffers
1074          * that haven't already reached the disk.  Try to write at least 256
1075          * log blocks. later on, we will only wait on blocks that correspond
1076          * to this transaction, but while we're unplugging we might as well
1077          * get a chunk of data on there.
1078          */
1079         atomic_inc(&journal->j_async_throttle);
1080         write_len = jl->j_len + 1;
1081         if (write_len < 256)
1082                 write_len = 256;
1083         for (i = 0 ; i < write_len ; i++) {
1084                 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + (jl->j_start + i) %
1085                     SB_ONDISK_JOURNAL_SIZE(s);
1086                 tbh = journal_find_get_block(s, bn);
1087                 if (tbh) {
1088                         if (buffer_dirty(tbh)) {
1089                             reiserfs_write_unlock(s);
1090                             ll_rw_block(WRITE, 1, &tbh);
1091                             reiserfs_write_lock(s);
1092                         }
1093                         put_bh(tbh) ;
1094                 }
1095         }
1096         atomic_dec(&journal->j_async_throttle);
1097
1098         /* We're skipping the commit if there's an error */
1099         if (retval || reiserfs_is_journal_aborted(journal))
1100                 barrier = 0;
1101
1102         /* wait on everything written so far before writing the commit
1103          * if we are in barrier mode, send the commit down now
1104          */
1105         barrier = reiserfs_barrier_flush(s);
1106         if (barrier) {
1107                 int ret;
1108                 lock_buffer(jl->j_commit_bh);
1109                 ret = submit_barrier_buffer(jl->j_commit_bh);
1110                 if (ret == -EOPNOTSUPP) {
1111                         set_buffer_uptodate(jl->j_commit_bh);
1112                         disable_barrier(s);
1113                         barrier = 0;
1114                 }
1115         }
1116         for (i = 0; i < (jl->j_len + 1); i++) {
1117                 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) +
1118                     (jl->j_start + i) % SB_ONDISK_JOURNAL_SIZE(s);
1119                 tbh = journal_find_get_block(s, bn);
1120
1121                 reiserfs_write_unlock(s);
1122                 wait_on_buffer(tbh);
1123                 reiserfs_write_lock(s);
1124                 // since we're using ll_rw_blk above, it might have skipped over
1125                 // a locked buffer.  Double check here
1126                 //
1127                 /* redundant, sync_dirty_buffer() checks */
1128                 if (buffer_dirty(tbh)) {
1129                         reiserfs_write_unlock(s);
1130                         sync_dirty_buffer(tbh);
1131                         reiserfs_write_lock(s);
1132                 }
1133                 if (unlikely(!buffer_uptodate(tbh))) {
1134 #ifdef CONFIG_REISERFS_CHECK
1135                         reiserfs_warning(s, "journal-601",
1136                                          "buffer write failed");
1137 #endif
1138                         retval = -EIO;
1139                 }
1140                 put_bh(tbh);    /* once for journal_find_get_block */
1141                 put_bh(tbh);    /* once due to original getblk in do_journal_end */
1142                 atomic_dec(&(jl->j_commit_left));
1143         }
1144
1145         BUG_ON(atomic_read(&(jl->j_commit_left)) != 1);
1146
1147         if (!barrier) {
1148                 /* If there was a write error in the journal - we can't commit
1149                  * this transaction - it will be invalid and, if successful,
1150                  * will just end up propagating the write error out to
1151                  * the file system. */
1152                 if (likely(!retval && !reiserfs_is_journal_aborted (journal))) {
1153                         if (buffer_dirty(jl->j_commit_bh))
1154                                 BUG();
1155                         mark_buffer_dirty(jl->j_commit_bh) ;
1156                         reiserfs_write_unlock(s);
1157                         sync_dirty_buffer(jl->j_commit_bh) ;
1158                         reiserfs_write_lock(s);
1159                 }
1160         } else {
1161                 reiserfs_write_unlock(s);
1162                 wait_on_buffer(jl->j_commit_bh);
1163                 reiserfs_write_lock(s);
1164         }
1165
1166         check_barrier_completion(s, jl->j_commit_bh);
1167
1168         /* If there was a write error in the journal - we can't commit this
1169          * transaction - it will be invalid and, if successful, will just end
1170          * up propagating the write error out to the filesystem. */
1171         if (unlikely(!buffer_uptodate(jl->j_commit_bh))) {
1172 #ifdef CONFIG_REISERFS_CHECK
1173                 reiserfs_warning(s, "journal-615", "buffer write failed");
1174 #endif
1175                 retval = -EIO;
1176         }
1177         bforget(jl->j_commit_bh);
1178         if (journal->j_last_commit_id != 0 &&
1179             (jl->j_trans_id - journal->j_last_commit_id) != 1) {
1180                 reiserfs_warning(s, "clm-2200", "last commit %lu, current %lu",
1181                                  journal->j_last_commit_id, jl->j_trans_id);
1182         }
1183         journal->j_last_commit_id = jl->j_trans_id;
1184
1185         /* now, every commit block is on the disk.  It is safe to allow blocks freed during this transaction to be reallocated */
1186         cleanup_freed_for_journal_list(s, jl);
1187
1188         retval = retval ? retval : journal->j_errno;
1189
1190         /* mark the metadata dirty */
1191         if (!retval)
1192                 dirty_one_transaction(s, jl);
1193         atomic_dec(&(jl->j_commit_left));
1194
1195         if (flushall) {
1196                 atomic_set(&(jl->j_older_commits_done), 1);
1197         }
1198         mutex_unlock(&jl->j_commit_mutex);
1199       put_jl:
1200         put_journal_list(s, jl);
1201
1202         if (retval)
1203                 reiserfs_abort(s, retval, "Journal write error in %s",
1204                                __func__);
1205         put_fs_excl();
1206         return retval;
1207 }
1208
1209 /*
1210 ** flush_journal_list frequently needs to find a newer transaction for a given block.  This does that, or
1211 ** returns NULL if it can't find anything
1212 */
1213 static struct reiserfs_journal_list *find_newer_jl_for_cn(struct
1214                                                           reiserfs_journal_cnode
1215                                                           *cn)
1216 {
1217         struct super_block *sb = cn->sb;
1218         b_blocknr_t blocknr = cn->blocknr;
1219
1220         cn = cn->hprev;
1221         while (cn) {
1222                 if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist) {
1223                         return cn->jlist;
1224                 }
1225                 cn = cn->hprev;
1226         }
1227         return NULL;
1228 }
1229
1230 static int newer_jl_done(struct reiserfs_journal_cnode *cn)
1231 {
1232         struct super_block *sb = cn->sb;
1233         b_blocknr_t blocknr = cn->blocknr;
1234
1235         cn = cn->hprev;
1236         while (cn) {
1237                 if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist &&
1238                     atomic_read(&cn->jlist->j_commit_left) != 0)
1239                                     return 0;
1240                 cn = cn->hprev;
1241         }
1242         return 1;
1243 }
1244
1245 static void remove_journal_hash(struct super_block *,
1246                                 struct reiserfs_journal_cnode **,
1247                                 struct reiserfs_journal_list *, unsigned long,
1248                                 int);
1249
1250 /*
1251 ** once all the real blocks have been flushed, it is safe to remove them from the
1252 ** journal list for this transaction.  Aside from freeing the cnode, this also allows the
1253 ** block to be reallocated for data blocks if it had been deleted.
1254 */
1255 static void remove_all_from_journal_list(struct super_block *sb,
1256                                          struct reiserfs_journal_list *jl,
1257                                          int debug)
1258 {
1259         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1260         struct reiserfs_journal_cnode *cn, *last;
1261         cn = jl->j_realblock;
1262
1263         /* which is better, to lock once around the whole loop, or
1264          ** to lock for each call to remove_journal_hash?
1265          */
1266         while (cn) {
1267                 if (cn->blocknr != 0) {
1268                         if (debug) {
1269                                 reiserfs_warning(sb, "reiserfs-2201",
1270                                                  "block %u, bh is %d, state %ld",
1271                                                  cn->blocknr, cn->bh ? 1 : 0,
1272                                                  cn->state);
1273                         }
1274                         cn->state = 0;
1275                         remove_journal_hash(sb, journal->j_list_hash_table,
1276                                             jl, cn->blocknr, 1);
1277                 }
1278                 last = cn;
1279                 cn = cn->next;
1280                 free_cnode(sb, last);
1281         }
1282         jl->j_realblock = NULL;
1283 }
1284
1285 /*
1286 ** if this timestamp is greater than the timestamp we wrote last to the header block, write it to the header block.
1287 ** once this is done, I can safely say the log area for this transaction won't ever be replayed, and I can start
1288 ** releasing blocks in this transaction for reuse as data blocks.
1289 ** called by flush_journal_list, before it calls remove_all_from_journal_list
1290 **
1291 */
1292 static int _update_journal_header_block(struct super_block *sb,
1293                                         unsigned long offset,
1294                                         unsigned int trans_id)
1295 {
1296         struct reiserfs_journal_header *jh;
1297         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1298
1299         if (reiserfs_is_journal_aborted(journal))
1300                 return -EIO;
1301
1302         if (trans_id >= journal->j_last_flush_trans_id) {
1303                 if (buffer_locked((journal->j_header_bh))) {
1304                         reiserfs_write_unlock(sb);
1305                         wait_on_buffer((journal->j_header_bh));
1306                         reiserfs_write_lock(sb);
1307                         if (unlikely(!buffer_uptodate(journal->j_header_bh))) {
1308 #ifdef CONFIG_REISERFS_CHECK
1309                                 reiserfs_warning(sb, "journal-699",
1310                                                  "buffer write failed");
1311 #endif
1312                                 return -EIO;
1313                         }
1314                 }
1315                 journal->j_last_flush_trans_id = trans_id;
1316                 journal->j_first_unflushed_offset = offset;
1317                 jh = (struct reiserfs_journal_header *)(journal->j_header_bh->
1318                                                         b_data);
1319                 jh->j_last_flush_trans_id = cpu_to_le32(trans_id);
1320                 jh->j_first_unflushed_offset = cpu_to_le32(offset);
1321                 jh->j_mount_id = cpu_to_le32(journal->j_mount_id);
1322
1323                 if (reiserfs_barrier_flush(sb)) {
1324                         int ret;
1325                         lock_buffer(journal->j_header_bh);
1326                         ret = submit_barrier_buffer(journal->j_header_bh);
1327                         if (ret == -EOPNOTSUPP) {
1328                                 set_buffer_uptodate(journal->j_header_bh);
1329                                 disable_barrier(sb);
1330                                 goto sync;
1331                         }
1332                         reiserfs_write_unlock(sb);
1333                         wait_on_buffer(journal->j_header_bh);
1334                         reiserfs_write_lock(sb);
1335                         check_barrier_completion(sb, journal->j_header_bh);
1336                 } else {
1337                       sync:
1338                         set_buffer_dirty(journal->j_header_bh);
1339                         reiserfs_write_unlock(sb);
1340                         sync_dirty_buffer(journal->j_header_bh);
1341                         reiserfs_write_lock(sb);
1342                 }
1343                 if (!buffer_uptodate(journal->j_header_bh)) {
1344                         reiserfs_warning(sb, "journal-837",
1345                                          "IO error during journal replay");
1346                         return -EIO;
1347                 }
1348         }
1349         return 0;
1350 }
1351
1352 static int update_journal_header_block(struct super_block *sb,
1353                                        unsigned long offset,
1354                                        unsigned int trans_id)
1355 {
1356         return _update_journal_header_block(sb, offset, trans_id);
1357 }
1358
1359 /*
1360 ** flush any and all journal lists older than you are
1361 ** can only be called from flush_journal_list
1362 */
1363 static int flush_older_journal_lists(struct super_block *sb,
1364                                      struct reiserfs_journal_list *jl)
1365 {
1366         struct list_head *entry;
1367         struct reiserfs_journal_list *other_jl;
1368         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1369         unsigned int trans_id = jl->j_trans_id;
1370
1371         /* we know we are the only ones flushing things, no extra race
1372          * protection is required.
1373          */
1374       restart:
1375         entry = journal->j_journal_list.next;
1376         /* Did we wrap? */
1377         if (entry == &journal->j_journal_list)
1378                 return 0;
1379         other_jl = JOURNAL_LIST_ENTRY(entry);
1380         if (other_jl->j_trans_id < trans_id) {
1381                 BUG_ON(other_jl->j_refcount <= 0);
1382                 /* do not flush all */
1383                 flush_journal_list(sb, other_jl, 0);
1384
1385                 /* other_jl is now deleted from the list */
1386                 goto restart;
1387         }
1388         return 0;
1389 }
1390
1391 static void del_from_work_list(struct super_block *s,
1392                                struct reiserfs_journal_list *jl)
1393 {
1394         struct reiserfs_journal *journal = SB_JOURNAL(s);
1395         if (!list_empty(&jl->j_working_list)) {
1396                 list_del_init(&jl->j_working_list);
1397                 journal->j_num_work_lists--;
1398         }
1399 }
1400
1401 /* flush a journal list, both commit and real blocks
1402 **
1403 ** always set flushall to 1, unless you are calling from inside
1404 ** flush_journal_list
1405 **
1406 ** IMPORTANT.  This can only be called while there are no journal writers,
1407 ** and the journal is locked.  That means it can only be called from
1408 ** do_journal_end, or by journal_release
1409 */
1410 static int flush_journal_list(struct super_block *s,
1411                               struct reiserfs_journal_list *jl, int flushall)
1412 {
1413         struct reiserfs_journal_list *pjl;
1414         struct reiserfs_journal_cnode *cn, *last;
1415         int count;
1416         int was_jwait = 0;
1417         int was_dirty = 0;
1418         struct buffer_head *saved_bh;
1419         unsigned long j_len_saved = jl->j_len;
1420         struct reiserfs_journal *journal = SB_JOURNAL(s);
1421         int err = 0;
1422
1423         BUG_ON(j_len_saved <= 0);
1424
1425         if (atomic_read(&journal->j_wcount) != 0) {
1426                 reiserfs_warning(s, "clm-2048", "called with wcount %d",
1427                                  atomic_read(&journal->j_wcount));
1428         }
1429         BUG_ON(jl->j_trans_id == 0);
1430
1431         /* if flushall == 0, the lock is already held */
1432         if (flushall) {
1433                 reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s);
1434         } else if (mutex_trylock(&journal->j_flush_mutex)) {
1435                 BUG();
1436         }
1437
1438         count = 0;
1439         if (j_len_saved > journal->j_trans_max) {
1440                 reiserfs_panic(s, "journal-715", "length is %lu, trans id %lu",
1441                                j_len_saved, jl->j_trans_id);
1442                 return 0;
1443         }
1444
1445         get_fs_excl();
1446
1447         /* if all the work is already done, get out of here */
1448         if (atomic_read(&(jl->j_nonzerolen)) <= 0 &&
1449             atomic_read(&(jl->j_commit_left)) <= 0) {
1450                 goto flush_older_and_return;
1451         }
1452
1453         /* start by putting the commit list on disk.  This will also flush
1454          ** the commit lists of any olders transactions
1455          */
1456         flush_commit_list(s, jl, 1);
1457
1458         if (!(jl->j_state & LIST_DIRTY)
1459             && !reiserfs_is_journal_aborted(journal))
1460                 BUG();
1461
1462         /* are we done now? */
1463         if (atomic_read(&(jl->j_nonzerolen)) <= 0 &&
1464             atomic_read(&(jl->j_commit_left)) <= 0) {
1465                 goto flush_older_and_return;
1466         }
1467
1468         /* loop through each cnode, see if we need to write it,
1469          ** or wait on a more recent transaction, or just ignore it
1470          */
1471         if (atomic_read(&(journal->j_wcount)) != 0) {
1472                 reiserfs_panic(s, "journal-844", "journal list is flushing, "
1473                                "wcount is not 0");
1474         }
1475         cn = jl->j_realblock;
1476         while (cn) {
1477                 was_jwait = 0;
1478                 was_dirty = 0;
1479                 saved_bh = NULL;
1480                 /* blocknr of 0 is no longer in the hash, ignore it */
1481                 if (cn->blocknr == 0) {
1482                         goto free_cnode;
1483                 }
1484
1485                 /* This transaction failed commit. Don't write out to the disk */
1486                 if (!(jl->j_state & LIST_DIRTY))
1487                         goto free_cnode;
1488
1489                 pjl = find_newer_jl_for_cn(cn);
1490                 /* the order is important here.  We check pjl to make sure we
1491                  ** don't clear BH_JDirty_wait if we aren't the one writing this
1492                  ** block to disk
1493                  */
1494                 if (!pjl && cn->bh) {
1495                         saved_bh = cn->bh;
1496
1497                         /* we do this to make sure nobody releases the buffer while
1498                          ** we are working with it
1499                          */
1500                         get_bh(saved_bh);
1501
1502                         if (buffer_journal_dirty(saved_bh)) {
1503                                 BUG_ON(!can_dirty(cn));
1504                                 was_jwait = 1;
1505                                 was_dirty = 1;
1506                         } else if (can_dirty(cn)) {
1507                                 /* everything with !pjl && jwait should be writable */
1508                                 BUG();
1509                         }
1510                 }
1511
1512                 /* if someone has this block in a newer transaction, just make
1513                  ** sure they are committed, and don't try writing it to disk
1514                  */
1515                 if (pjl) {
1516                         if (atomic_read(&pjl->j_commit_left))
1517                                 flush_commit_list(s, pjl, 1);
1518                         goto free_cnode;
1519                 }
1520
1521                 /* bh == NULL when the block got to disk on its own, OR,
1522                  ** the block got freed in a future transaction
1523                  */
1524                 if (saved_bh == NULL) {
1525                         goto free_cnode;
1526                 }
1527
1528                 /* this should never happen.  kupdate_one_transaction has this list
1529                  ** locked while it works, so we should never see a buffer here that
1530                  ** is not marked JDirty_wait
1531                  */
1532                 if ((!was_jwait) && !buffer_locked(saved_bh)) {
1533                         reiserfs_warning(s, "journal-813",
1534                                          "BAD! buffer %llu %cdirty %cjwait, "
1535                                          "not in a newer tranasction",
1536                                          (unsigned long long)saved_bh->
1537                                          b_blocknr, was_dirty ? ' ' : '!',
1538                                          was_jwait ? ' ' : '!');
1539                 }
1540                 if (was_dirty) {
1541                         /* we inc again because saved_bh gets decremented at free_cnode */
1542                         get_bh(saved_bh);
1543                         set_bit(BLOCK_NEEDS_FLUSH, &cn->state);
1544                         lock_buffer(saved_bh);
1545                         BUG_ON(cn->blocknr != saved_bh->b_blocknr);
1546                         if (buffer_dirty(saved_bh))
1547                                 submit_logged_buffer(saved_bh);
1548                         else
1549                                 unlock_buffer(saved_bh);
1550                         count++;
1551                 } else {
1552                         reiserfs_warning(s, "clm-2082",
1553                                          "Unable to flush buffer %llu in %s",
1554                                          (unsigned long long)saved_bh->
1555                                          b_blocknr, __func__);
1556                 }
1557               free_cnode:
1558                 last = cn;
1559                 cn = cn->next;
1560                 if (saved_bh) {
1561                         /* we incremented this to keep others from taking the buffer head away */
1562                         put_bh(saved_bh);
1563                         if (atomic_read(&(saved_bh->b_count)) < 0) {
1564                                 reiserfs_warning(s, "journal-945",
1565                                                  "saved_bh->b_count < 0");
1566                         }
1567                 }
1568         }
1569         if (count > 0) {
1570                 cn = jl->j_realblock;
1571                 while (cn) {
1572                         if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) {
1573                                 if (!cn->bh) {
1574                                         reiserfs_panic(s, "journal-1011",
1575                                                        "cn->bh is NULL");
1576                                 }
1577
1578                                 reiserfs_write_unlock(s);
1579                                 wait_on_buffer(cn->bh);
1580                                 reiserfs_write_lock(s);
1581
1582                                 if (!cn->bh) {
1583                                         reiserfs_panic(s, "journal-1012",
1584                                                        "cn->bh is NULL");
1585                                 }
1586                                 if (unlikely(!buffer_uptodate(cn->bh))) {
1587 #ifdef CONFIG_REISERFS_CHECK
1588                                         reiserfs_warning(s, "journal-949",
1589                                                          "buffer write failed");
1590 #endif
1591                                         err = -EIO;
1592                                 }
1593                                 /* note, we must clear the JDirty_wait bit after the up to date
1594                                  ** check, otherwise we race against our flushpage routine
1595                                  */
1596                                 BUG_ON(!test_clear_buffer_journal_dirty
1597                                        (cn->bh));
1598
1599                                 /* drop one ref for us */
1600                                 put_bh(cn->bh);
1601                                 /* drop one ref for journal_mark_dirty */
1602                                 release_buffer_page(cn->bh);
1603                         }
1604                         cn = cn->next;
1605                 }
1606         }
1607
1608         if (err)
1609                 reiserfs_abort(s, -EIO,
1610                                "Write error while pushing transaction to disk in %s",
1611                                __func__);
1612       flush_older_and_return:
1613
1614         /* before we can update the journal header block, we _must_ flush all
1615          ** real blocks from all older transactions to disk.  This is because
1616          ** once the header block is updated, this transaction will not be
1617          ** replayed after a crash
1618          */
1619         if (flushall) {
1620                 flush_older_journal_lists(s, jl);
1621         }
1622
1623         err = journal->j_errno;
1624         /* before we can remove everything from the hash tables for this
1625          ** transaction, we must make sure it can never be replayed
1626          **
1627          ** since we are only called from do_journal_end, we know for sure there
1628          ** are no allocations going on while we are flushing journal lists.  So,
1629          ** we only need to update the journal header block for the last list
1630          ** being flushed
1631          */
1632         if (!err && flushall) {
1633                 err =
1634                     update_journal_header_block(s,
1635                                                 (jl->j_start + jl->j_len +
1636                                                  2) % SB_ONDISK_JOURNAL_SIZE(s),
1637                                                 jl->j_trans_id);
1638                 if (err)
1639                         reiserfs_abort(s, -EIO,
1640                                        "Write error while updating journal header in %s",
1641                                        __func__);
1642         }
1643         remove_all_from_journal_list(s, jl, 0);
1644         list_del_init(&jl->j_list);
1645         journal->j_num_lists--;
1646         del_from_work_list(s, jl);
1647
1648         if (journal->j_last_flush_id != 0 &&
1649             (jl->j_trans_id - journal->j_last_flush_id) != 1) {
1650                 reiserfs_warning(s, "clm-2201", "last flush %lu, current %lu",
1651                                  journal->j_last_flush_id, jl->j_trans_id);
1652         }
1653         journal->j_last_flush_id = jl->j_trans_id;
1654
1655         /* not strictly required since we are freeing the list, but it should
1656          * help find code using dead lists later on
1657          */
1658         jl->j_len = 0;
1659         atomic_set(&(jl->j_nonzerolen), 0);
1660         jl->j_start = 0;
1661         jl->j_realblock = NULL;
1662         jl->j_commit_bh = NULL;
1663         jl->j_trans_id = 0;
1664         jl->j_state = 0;
1665         put_journal_list(s, jl);
1666         if (flushall)
1667                 mutex_unlock(&journal->j_flush_mutex);
1668         put_fs_excl();
1669         return err;
1670 }
1671
1672 static int test_transaction(struct super_block *s,
1673                             struct reiserfs_journal_list *jl)
1674 {
1675         struct reiserfs_journal_cnode *cn;
1676
1677         if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0)
1678                 return 1;
1679
1680         cn = jl->j_realblock;
1681         while (cn) {
1682                 /* if the blocknr == 0, this has been cleared from the hash,
1683                  ** skip it
1684                  */
1685                 if (cn->blocknr == 0) {
1686                         goto next;
1687                 }
1688                 if (cn->bh && !newer_jl_done(cn))
1689                         return 0;
1690               next:
1691                 cn = cn->next;
1692                 cond_resched();
1693         }
1694         return 0;
1695 }
1696
1697 static int write_one_transaction(struct super_block *s,
1698                                  struct reiserfs_journal_list *jl,
1699                                  struct buffer_chunk *chunk)
1700 {
1701         struct reiserfs_journal_cnode *cn;
1702         int ret = 0;
1703
1704         jl->j_state |= LIST_TOUCHED;
1705         del_from_work_list(s, jl);
1706         if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0) {
1707                 return 0;
1708         }
1709
1710         cn = jl->j_realblock;
1711         while (cn) {
1712                 /* if the blocknr == 0, this has been cleared from the hash,
1713                  ** skip it
1714                  */
1715                 if (cn->blocknr == 0) {
1716                         goto next;
1717                 }
1718                 if (cn->bh && can_dirty(cn) && buffer_dirty(cn->bh)) {
1719                         struct buffer_head *tmp_bh;
1720                         /* we can race against journal_mark_freed when we try
1721                          * to lock_buffer(cn->bh), so we have to inc the buffer
1722                          * count, and recheck things after locking
1723                          */
1724                         tmp_bh = cn->bh;
1725                         get_bh(tmp_bh);
1726                         lock_buffer(tmp_bh);
1727                         if (cn->bh && can_dirty(cn) && buffer_dirty(tmp_bh)) {
1728                                 if (!buffer_journal_dirty(tmp_bh) ||
1729                                     buffer_journal_prepared(tmp_bh))
1730                                         BUG();
1731                                 add_to_chunk(chunk, tmp_bh, NULL, write_chunk);
1732                                 ret++;
1733                         } else {
1734                                 /* note, cn->bh might be null now */
1735                                 unlock_buffer(tmp_bh);
1736                         }
1737                         put_bh(tmp_bh);
1738                 }
1739               next:
1740                 cn = cn->next;
1741                 cond_resched();
1742         }
1743         return ret;
1744 }
1745
1746 /* used by flush_commit_list */
1747 static int dirty_one_transaction(struct super_block *s,
1748                                  struct reiserfs_journal_list *jl)
1749 {
1750         struct reiserfs_journal_cnode *cn;
1751         struct reiserfs_journal_list *pjl;
1752         int ret = 0;
1753
1754         jl->j_state |= LIST_DIRTY;
1755         cn = jl->j_realblock;
1756         while (cn) {
1757                 /* look for a more recent transaction that logged this
1758                  ** buffer.  Only the most recent transaction with a buffer in
1759                  ** it is allowed to send that buffer to disk
1760                  */
1761                 pjl = find_newer_jl_for_cn(cn);
1762                 if (!pjl && cn->blocknr && cn->bh
1763                     && buffer_journal_dirty(cn->bh)) {
1764                         BUG_ON(!can_dirty(cn));
1765                         /* if the buffer is prepared, it will either be logged
1766                          * or restored.  If restored, we need to make sure
1767                          * it actually gets marked dirty
1768                          */
1769                         clear_buffer_journal_new(cn->bh);
1770                         if (buffer_journal_prepared(cn->bh)) {
1771                                 set_buffer_journal_restore_dirty(cn->bh);
1772                         } else {
1773                                 set_buffer_journal_test(cn->bh);
1774                                 mark_buffer_dirty(cn->bh);
1775                         }
1776                 }
1777                 cn = cn->next;
1778         }
1779         return ret;
1780 }
1781
1782 static int kupdate_transactions(struct super_block *s,
1783                                 struct reiserfs_journal_list *jl,
1784                                 struct reiserfs_journal_list **next_jl,
1785                                 unsigned int *next_trans_id,
1786                                 int num_blocks, int num_trans)
1787 {
1788         int ret = 0;
1789         int written = 0;
1790         int transactions_flushed = 0;
1791         unsigned int orig_trans_id = jl->j_trans_id;
1792         struct buffer_chunk chunk;
1793         struct list_head *entry;
1794         struct reiserfs_journal *journal = SB_JOURNAL(s);
1795         chunk.nr = 0;
1796
1797         reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s);
1798         if (!journal_list_still_alive(s, orig_trans_id)) {
1799                 goto done;
1800         }
1801
1802         /* we've got j_flush_mutex held, nobody is going to delete any
1803          * of these lists out from underneath us
1804          */
1805         while ((num_trans && transactions_flushed < num_trans) ||
1806                (!num_trans && written < num_blocks)) {
1807
1808                 if (jl->j_len == 0 || (jl->j_state & LIST_TOUCHED) ||
1809                     atomic_read(&jl->j_commit_left)
1810                     || !(jl->j_state & LIST_DIRTY)) {
1811                         del_from_work_list(s, jl);
1812                         break;
1813                 }
1814                 ret = write_one_transaction(s, jl, &chunk);
1815
1816                 if (ret < 0)
1817                         goto done;
1818                 transactions_flushed++;
1819                 written += ret;
1820                 entry = jl->j_list.next;
1821
1822                 /* did we wrap? */
1823                 if (entry == &journal->j_journal_list) {
1824                         break;
1825                 }
1826                 jl = JOURNAL_LIST_ENTRY(entry);
1827
1828                 /* don't bother with older transactions */
1829                 if (jl->j_trans_id <= orig_trans_id)
1830                         break;
1831         }
1832         if (chunk.nr) {
1833                 write_chunk(&chunk);
1834         }
1835
1836       done:
1837         mutex_unlock(&journal->j_flush_mutex);
1838         return ret;
1839 }
1840
1841 /* for o_sync and fsync heavy applications, they tend to use
1842 ** all the journa list slots with tiny transactions.  These
1843 ** trigger lots and lots of calls to update the header block, which
1844 ** adds seeks and slows things down.
1845 **
1846 ** This function tries to clear out a large chunk of the journal lists
1847 ** at once, which makes everything faster since only the newest journal
1848 ** list updates the header block
1849 */
1850 static int flush_used_journal_lists(struct super_block *s,
1851                                     struct reiserfs_journal_list *jl)
1852 {
1853         unsigned long len = 0;
1854         unsigned long cur_len;
1855         int ret;
1856         int i;
1857         int limit = 256;
1858         struct reiserfs_journal_list *tjl;
1859         struct reiserfs_journal_list *flush_jl;
1860         unsigned int trans_id;
1861         struct reiserfs_journal *journal = SB_JOURNAL(s);
1862
1863         flush_jl = tjl = jl;
1864
1865         /* in data logging mode, try harder to flush a lot of blocks */
1866         if (reiserfs_data_log(s))
1867                 limit = 1024;
1868         /* flush for 256 transactions or limit blocks, whichever comes first */
1869         for (i = 0; i < 256 && len < limit; i++) {
1870                 if (atomic_read(&tjl->j_commit_left) ||
1871                     tjl->j_trans_id < jl->j_trans_id) {
1872                         break;
1873                 }
1874                 cur_len = atomic_read(&tjl->j_nonzerolen);
1875                 if (cur_len > 0) {
1876                         tjl->j_state &= ~LIST_TOUCHED;
1877                 }
1878                 len += cur_len;
1879                 flush_jl = tjl;
1880                 if (tjl->j_list.next == &journal->j_journal_list)
1881                         break;
1882                 tjl = JOURNAL_LIST_ENTRY(tjl->j_list.next);
1883         }
1884         /* try to find a group of blocks we can flush across all the
1885          ** transactions, but only bother if we've actually spanned
1886          ** across multiple lists
1887          */
1888         if (flush_jl != jl) {
1889                 ret = kupdate_transactions(s, jl, &tjl, &trans_id, len, i);
1890         }
1891         flush_journal_list(s, flush_jl, 1);
1892         return 0;
1893 }
1894
1895 /*
1896 ** removes any nodes in table with name block and dev as bh.
1897 ** only touchs the hnext and hprev pointers.
1898 */
1899 void remove_journal_hash(struct super_block *sb,
1900                          struct reiserfs_journal_cnode **table,
1901                          struct reiserfs_journal_list *jl,
1902                          unsigned long block, int remove_freed)
1903 {
1904         struct reiserfs_journal_cnode *cur;
1905         struct reiserfs_journal_cnode **head;
1906
1907         head = &(journal_hash(table, sb, block));
1908         if (!head) {
1909                 return;
1910         }
1911         cur = *head;
1912         while (cur) {
1913                 if (cur->blocknr == block && cur->sb == sb
1914                     && (jl == NULL || jl == cur->jlist)
1915                     && (!test_bit(BLOCK_FREED, &cur->state) || remove_freed)) {
1916                         if (cur->hnext) {
1917                                 cur->hnext->hprev = cur->hprev;
1918                         }
1919                         if (cur->hprev) {
1920                                 cur->hprev->hnext = cur->hnext;
1921                         } else {
1922                                 *head = cur->hnext;
1923                         }
1924                         cur->blocknr = 0;
1925                         cur->sb = NULL;
1926                         cur->state = 0;
1927                         if (cur->bh && cur->jlist)      /* anybody who clears the cur->bh will also dec the nonzerolen */
1928                                 atomic_dec(&(cur->jlist->j_nonzerolen));
1929                         cur->bh = NULL;
1930                         cur->jlist = NULL;
1931                 }
1932                 cur = cur->hnext;
1933         }
1934 }
1935
1936 static void free_journal_ram(struct super_block *sb)
1937 {
1938         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1939         kfree(journal->j_current_jl);
1940         journal->j_num_lists--;
1941
1942         vfree(journal->j_cnode_free_orig);
1943         free_list_bitmaps(sb, journal->j_list_bitmap);
1944         free_bitmap_nodes(sb);  /* must be after free_list_bitmaps */
1945         if (journal->j_header_bh) {
1946                 brelse(journal->j_header_bh);
1947         }
1948         /* j_header_bh is on the journal dev, make sure not to release the journal
1949          * dev until we brelse j_header_bh
1950          */
1951         release_journal_dev(sb, journal);
1952         vfree(journal);
1953 }
1954
1955 /*
1956 ** call on unmount.  Only set error to 1 if you haven't made your way out
1957 ** of read_super() yet.  Any other caller must keep error at 0.
1958 */
1959 static int do_journal_release(struct reiserfs_transaction_handle *th,
1960                               struct super_block *sb, int error)
1961 {
1962         struct reiserfs_transaction_handle myth;
1963         int flushed = 0;
1964         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1965
1966         /* we only want to flush out transactions if we were called with error == 0
1967          */
1968         if (!error && !(sb->s_flags & MS_RDONLY)) {
1969                 /* end the current trans */
1970                 BUG_ON(!th->t_trans_id);
1971                 do_journal_end(th, sb, 10, FLUSH_ALL);
1972
1973                 /* make sure something gets logged to force our way into the flush code */
1974                 if (!journal_join(&myth, sb, 1)) {
1975                         reiserfs_prepare_for_journal(sb,
1976                                                      SB_BUFFER_WITH_SB(sb),
1977                                                      1);
1978                         journal_mark_dirty(&myth, sb,
1979                                            SB_BUFFER_WITH_SB(sb));
1980                         do_journal_end(&myth, sb, 1, FLUSH_ALL);
1981                         flushed = 1;
1982                 }
1983         }
1984
1985         /* this also catches errors during the do_journal_end above */
1986         if (!error && reiserfs_is_journal_aborted(journal)) {
1987                 memset(&myth, 0, sizeof(myth));
1988                 if (!journal_join_abort(&myth, sb, 1)) {
1989                         reiserfs_prepare_for_journal(sb,
1990                                                      SB_BUFFER_WITH_SB(sb),
1991                                                      1);
1992                         journal_mark_dirty(&myth, sb,
1993                                            SB_BUFFER_WITH_SB(sb));
1994                         do_journal_end(&myth, sb, 1, FLUSH_ALL);
1995                 }
1996         }
1997
1998         reiserfs_mounted_fs_count--;
1999         /* wait for all commits to finish */
2000         cancel_delayed_work(&SB_JOURNAL(sb)->j_work);
2001
2002         /*
2003          * We must release the write lock here because
2004          * the workqueue job (flush_async_commit) needs this lock
2005          */
2006         reiserfs_write_unlock(sb);
2007         flush_workqueue(commit_wq);
2008
2009         if (!reiserfs_mounted_fs_count) {
2010                 destroy_workqueue(commit_wq);
2011                 commit_wq = NULL;
2012         }
2013
2014         free_journal_ram(sb);
2015
2016         reiserfs_write_lock(sb);
2017
2018         return 0;
2019 }
2020
2021 /*
2022 ** call on unmount.  flush all journal trans, release all alloc'd ram
2023 */
2024 int journal_release(struct reiserfs_transaction_handle *th,
2025                     struct super_block *sb)
2026 {
2027         return do_journal_release(th, sb, 0);
2028 }
2029
2030 /*
2031 ** only call from an error condition inside reiserfs_read_super!
2032 */
2033 int journal_release_error(struct reiserfs_transaction_handle *th,
2034                           struct super_block *sb)
2035 {
2036         return do_journal_release(th, sb, 1);
2037 }
2038
2039 /* compares description block with commit block.  returns 1 if they differ, 0 if they are the same */
2040 static int journal_compare_desc_commit(struct super_block *sb,
2041                                        struct reiserfs_journal_desc *desc,
2042                                        struct reiserfs_journal_commit *commit)
2043 {
2044         if (get_commit_trans_id(commit) != get_desc_trans_id(desc) ||
2045             get_commit_trans_len(commit) != get_desc_trans_len(desc) ||
2046             get_commit_trans_len(commit) > SB_JOURNAL(sb)->j_trans_max ||
2047             get_commit_trans_len(commit) <= 0) {
2048                 return 1;
2049         }
2050         return 0;
2051 }
2052
2053 /* returns 0 if it did not find a description block
2054 ** returns -1 if it found a corrupt commit block
2055 ** returns 1 if both desc and commit were valid
2056 */
2057 static int journal_transaction_is_valid(struct super_block *sb,
2058                                         struct buffer_head *d_bh,
2059                                         unsigned int *oldest_invalid_trans_id,
2060                                         unsigned long *newest_mount_id)
2061 {
2062         struct reiserfs_journal_desc *desc;
2063         struct reiserfs_journal_commit *commit;
2064         struct buffer_head *c_bh;
2065         unsigned long offset;
2066
2067         if (!d_bh)
2068                 return 0;
2069
2070         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2071         if (get_desc_trans_len(desc) > 0
2072             && !memcmp(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8)) {
2073                 if (oldest_invalid_trans_id && *oldest_invalid_trans_id
2074                     && get_desc_trans_id(desc) > *oldest_invalid_trans_id) {
2075                         reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2076                                        "journal-986: transaction "
2077                                        "is valid returning because trans_id %d is greater than "
2078                                        "oldest_invalid %lu",
2079                                        get_desc_trans_id(desc),
2080                                        *oldest_invalid_trans_id);
2081                         return 0;
2082                 }
2083                 if (newest_mount_id
2084                     && *newest_mount_id > get_desc_mount_id(desc)) {
2085                         reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2086                                        "journal-1087: transaction "
2087                                        "is valid returning because mount_id %d is less than "
2088                                        "newest_mount_id %lu",
2089                                        get_desc_mount_id(desc),
2090                                        *newest_mount_id);
2091                         return -1;
2092                 }
2093                 if (get_desc_trans_len(desc) > SB_JOURNAL(sb)->j_trans_max) {
2094                         reiserfs_warning(sb, "journal-2018",
2095                                          "Bad transaction length %d "
2096                                          "encountered, ignoring transaction",
2097                                          get_desc_trans_len(desc));
2098                         return -1;
2099                 }
2100                 offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2101
2102                 /* ok, we have a journal description block, lets see if the transaction was valid */
2103                 c_bh =
2104                     journal_bread(sb,
2105                                   SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2106                                   ((offset + get_desc_trans_len(desc) +
2107                                     1) % SB_ONDISK_JOURNAL_SIZE(sb)));
2108                 if (!c_bh)
2109                         return 0;
2110                 commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2111                 if (journal_compare_desc_commit(sb, desc, commit)) {
2112                         reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2113                                        "journal_transaction_is_valid, commit offset %ld had bad "
2114                                        "time %d or length %d",
2115                                        c_bh->b_blocknr -
2116                                        SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2117                                        get_commit_trans_id(commit),
2118                                        get_commit_trans_len(commit));
2119                         brelse(c_bh);
2120                         if (oldest_invalid_trans_id) {
2121                                 *oldest_invalid_trans_id =
2122                                     get_desc_trans_id(desc);
2123                                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2124                                                "journal-1004: "
2125                                                "transaction_is_valid setting oldest invalid trans_id "
2126                                                "to %d",
2127                                                get_desc_trans_id(desc));
2128                         }
2129                         return -1;
2130                 }
2131                 brelse(c_bh);
2132                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2133                                "journal-1006: found valid "
2134                                "transaction start offset %llu, len %d id %d",
2135                                d_bh->b_blocknr -
2136                                SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2137                                get_desc_trans_len(desc),
2138                                get_desc_trans_id(desc));
2139                 return 1;
2140         } else {
2141                 return 0;
2142         }
2143 }
2144
2145 static void brelse_array(struct buffer_head **heads, int num)
2146 {
2147         int i;
2148         for (i = 0; i < num; i++) {
2149                 brelse(heads[i]);
2150         }
2151 }
2152
2153 /*
2154 ** given the start, and values for the oldest acceptable transactions,
2155 ** this either reads in a replays a transaction, or returns because the transaction
2156 ** is invalid, or too old.
2157 */
2158 static int journal_read_transaction(struct super_block *sb,
2159                                     unsigned long cur_dblock,
2160                                     unsigned long oldest_start,
2161                                     unsigned int oldest_trans_id,
2162                                     unsigned long newest_mount_id)
2163 {
2164         struct reiserfs_journal *journal = SB_JOURNAL(sb);
2165         struct reiserfs_journal_desc *desc;
2166         struct reiserfs_journal_commit *commit;
2167         unsigned int trans_id = 0;
2168         struct buffer_head *c_bh;
2169         struct buffer_head *d_bh;
2170         struct buffer_head **log_blocks = NULL;
2171         struct buffer_head **real_blocks = NULL;
2172         unsigned int trans_offset;
2173         int i;
2174         int trans_half;
2175
2176         d_bh = journal_bread(sb, cur_dblock);
2177         if (!d_bh)
2178                 return 1;
2179         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2180         trans_offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2181         reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1037: "
2182                        "journal_read_transaction, offset %llu, len %d mount_id %d",
2183                        d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2184                        get_desc_trans_len(desc), get_desc_mount_id(desc));
2185         if (get_desc_trans_id(desc) < oldest_trans_id) {
2186                 reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1039: "
2187                                "journal_read_trans skipping because %lu is too old",
2188                                cur_dblock -
2189                                SB_ONDISK_JOURNAL_1st_BLOCK(sb));
2190                 brelse(d_bh);
2191                 return 1;
2192         }
2193         if (get_desc_mount_id(desc) != newest_mount_id) {
2194                 reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1146: "
2195                                "journal_read_trans skipping because %d is != "
2196                                "newest_mount_id %lu", get_desc_mount_id(desc),
2197                                newest_mount_id);
2198                 brelse(d_bh);
2199                 return 1;
2200         }
2201         c_bh = journal_bread(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2202                              ((trans_offset + get_desc_trans_len(desc) + 1) %
2203                               SB_ONDISK_JOURNAL_SIZE(sb)));
2204         if (!c_bh) {
2205                 brelse(d_bh);
2206                 return 1;
2207         }
2208         commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2209         if (journal_compare_desc_commit(sb, desc, commit)) {
2210                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2211                                "journal_read_transaction, "
2212                                "commit offset %llu had bad time %d or length %d",
2213                                c_bh->b_blocknr -
2214                                SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2215                                get_commit_trans_id(commit),
2216                                get_commit_trans_len(commit));
2217                 brelse(c_bh);
2218                 brelse(d_bh);
2219                 return 1;
2220         }
2221
2222         if (bdev_read_only(sb->s_bdev)) {
2223                 reiserfs_warning(sb, "clm-2076",
2224                                  "device is readonly, unable to replay log");
2225                 brelse(c_bh);
2226                 brelse(d_bh);
2227                 return -EROFS;
2228         }
2229
2230         trans_id = get_desc_trans_id(desc);
2231         /* now we know we've got a good transaction, and it was inside the valid time ranges */
2232         log_blocks = kmalloc(get_desc_trans_len(desc) *
2233                              sizeof(struct buffer_head *), GFP_NOFS);
2234         real_blocks = kmalloc(get_desc_trans_len(desc) *
2235                               sizeof(struct buffer_head *), GFP_NOFS);
2236         if (!log_blocks || !real_blocks) {
2237                 brelse(c_bh);
2238                 brelse(d_bh);
2239                 kfree(log_blocks);
2240                 kfree(real_blocks);
2241                 reiserfs_warning(sb, "journal-1169",
2242                                  "kmalloc failed, unable to mount FS");
2243                 return -1;
2244         }
2245         /* get all the buffer heads */
2246         trans_half = journal_trans_half(sb->s_blocksize);
2247         for (i = 0; i < get_desc_trans_len(desc); i++) {
2248                 log_blocks[i] =
2249                     journal_getblk(sb,
2250                                    SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2251                                    (trans_offset + 1 +
2252                                     i) % SB_ONDISK_JOURNAL_SIZE(sb));
2253                 if (i < trans_half) {
2254                         real_blocks[i] =
2255                             sb_getblk(sb,
2256                                       le32_to_cpu(desc->j_realblock[i]));
2257                 } else {
2258                         real_blocks[i] =
2259                             sb_getblk(sb,
2260                                       le32_to_cpu(commit->
2261                                                   j_realblock[i - trans_half]));
2262                 }
2263                 if (real_blocks[i]->b_blocknr > SB_BLOCK_COUNT(sb)) {
2264                         reiserfs_warning(sb, "journal-1207",
2265                                          "REPLAY FAILURE fsck required! "
2266                                          "Block to replay is outside of "
2267                                          "filesystem");
2268                         goto abort_replay;
2269                 }
2270                 /* make sure we don't try to replay onto log or reserved area */
2271                 if (is_block_in_log_or_reserved_area
2272                     (sb, real_blocks[i]->b_blocknr)) {
2273                         reiserfs_warning(sb, "journal-1204",
2274                                          "REPLAY FAILURE fsck required! "
2275                                          "Trying to replay onto a log block");
2276                       abort_replay:
2277                         brelse_array(log_blocks, i);
2278                         brelse_array(real_blocks, i);
2279                         brelse(c_bh);
2280                         brelse(d_bh);
2281                         kfree(log_blocks);
2282                         kfree(real_blocks);
2283                         return -1;
2284                 }
2285         }
2286         /* read in the log blocks, memcpy to the corresponding real block */
2287         ll_rw_block(READ, get_desc_trans_len(desc), log_blocks);
2288         for (i = 0; i < get_desc_trans_len(desc); i++) {
2289
2290                 reiserfs_write_unlock(sb);
2291                 wait_on_buffer(log_blocks[i]);
2292                 reiserfs_write_lock(sb);
2293
2294                 if (!buffer_uptodate(log_blocks[i])) {
2295                         reiserfs_warning(sb, "journal-1212",
2296                                          "REPLAY FAILURE fsck required! "
2297                                          "buffer write failed");
2298                         brelse_array(log_blocks + i,
2299                                      get_desc_trans_len(desc) - i);
2300                         brelse_array(real_blocks, get_desc_trans_len(desc));
2301                         brelse(c_bh);
2302                         brelse(d_bh);
2303                         kfree(log_blocks);
2304                         kfree(real_blocks);
2305                         return -1;
2306                 }
2307                 memcpy(real_blocks[i]->b_data, log_blocks[i]->b_data,
2308                        real_blocks[i]->b_size);
2309                 set_buffer_uptodate(real_blocks[i]);
2310                 brelse(log_blocks[i]);
2311         }
2312         /* flush out the real blocks */
2313         for (i = 0; i < get_desc_trans_len(desc); i++) {
2314                 set_buffer_dirty(real_blocks[i]);
2315                 ll_rw_block(SWRITE, 1, real_blocks + i);
2316         }
2317         for (i = 0; i < get_desc_trans_len(desc); i++) {
2318                 wait_on_buffer(real_blocks[i]);
2319                 if (!buffer_uptodate(real_blocks[i])) {
2320                         reiserfs_warning(sb, "journal-1226",
2321                                          "REPLAY FAILURE, fsck required! "
2322                                          "buffer write failed");
2323                         brelse_array(real_blocks + i,
2324                                      get_desc_trans_len(desc) - i);
2325                         brelse(c_bh);
2326                         brelse(d_bh);
2327                         kfree(log_blocks);
2328                         kfree(real_blocks);
2329                         return -1;
2330                 }
2331                 brelse(real_blocks[i]);
2332         }
2333         cur_dblock =
2334             SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2335             ((trans_offset + get_desc_trans_len(desc) +
2336               2) % SB_ONDISK_JOURNAL_SIZE(sb));
2337         reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2338                        "journal-1095: setting journal " "start to offset %ld",
2339                        cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb));
2340
2341         /* init starting values for the first transaction, in case this is the last transaction to be replayed. */
2342         journal->j_start = cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2343         journal->j_last_flush_trans_id = trans_id;
2344         journal->j_trans_id = trans_id + 1;
2345         /* check for trans_id overflow */
2346         if (journal->j_trans_id == 0)
2347                 journal->j_trans_id = 10;
2348         brelse(c_bh);
2349         brelse(d_bh);
2350         kfree(log_blocks);
2351         kfree(real_blocks);
2352         return 0;
2353 }
2354
2355 /* This function reads blocks starting from block and to max_block of bufsize
2356    size (but no more than BUFNR blocks at a time). This proved to improve
2357    mounting speed on self-rebuilding raid5 arrays at least.
2358    Right now it is only used from journal code. But later we might use it
2359    from other places.
2360    Note: Do not use journal_getblk/sb_getblk functions here! */
2361 static struct buffer_head *reiserfs_breada(struct block_device *dev,
2362                                            b_blocknr_t block, int bufsize,
2363                                            b_blocknr_t max_block)
2364 {
2365         struct buffer_head *bhlist[BUFNR];
2366         unsigned int blocks = BUFNR;
2367         struct buffer_head *bh;
2368         int i, j;
2369
2370         bh = __getblk(dev, block, bufsize);
2371         if (buffer_uptodate(bh))
2372                 return (bh);
2373
2374         if (block + BUFNR > max_block) {
2375                 blocks = max_block - block;
2376         }
2377         bhlist[0] = bh;
2378         j = 1;
2379         for (i = 1; i < blocks; i++) {
2380                 bh = __getblk(dev, block + i, bufsize);
2381                 if (buffer_uptodate(bh)) {
2382                         brelse(bh);
2383                         break;
2384                 } else
2385                         bhlist[j++] = bh;
2386         }
2387         ll_rw_block(READ, j, bhlist);
2388         for (i = 1; i < j; i++)
2389                 brelse(bhlist[i]);
2390         bh = bhlist[0];
2391         wait_on_buffer(bh);
2392         if (buffer_uptodate(bh))
2393                 return bh;
2394         brelse(bh);
2395         return NULL;
2396 }
2397
2398 /*
2399 ** read and replay the log
2400 ** on a clean unmount, the journal header's next unflushed pointer will be to an invalid
2401 ** transaction.  This tests that before finding all the transactions in the log, which makes normal mount times fast.
2402 **
2403 ** After a crash, this starts with the next unflushed transaction, and replays until it finds one too old, or invalid.
2404 **
2405 ** On exit, it sets things up so the first transaction will work correctly.
2406 */
2407 static int journal_read(struct super_block *sb)
2408 {
2409         struct reiserfs_journal *journal = SB_JOURNAL(sb);
2410         struct reiserfs_journal_desc *desc;
2411         unsigned int oldest_trans_id = 0;
2412         unsigned int oldest_invalid_trans_id = 0;
2413         time_t start;
2414         unsigned long oldest_start = 0;
2415         unsigned long cur_dblock = 0;
2416         unsigned long newest_mount_id = 9;
2417         struct buffer_head *d_bh;
2418         struct reiserfs_journal_header *jh;
2419         int valid_journal_header = 0;
2420         int replay_count = 0;
2421         int continue_replay = 1;
2422         int ret;
2423         char b[BDEVNAME_SIZE];
2424
2425         cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2426         reiserfs_info(sb, "checking transaction log (%s)\n",
2427                       bdevname(journal->j_dev_bd, b));
2428         start = get_seconds();
2429
2430         /* step 1, read in the journal header block.  Check the transaction it says
2431          ** is the first unflushed, and if that transaction is not valid,
2432          ** replay is done
2433          */
2434         journal->j_header_bh = journal_bread(sb,
2435                                              SB_ONDISK_JOURNAL_1st_BLOCK(sb)
2436                                              + SB_ONDISK_JOURNAL_SIZE(sb));
2437         if (!journal->j_header_bh) {
2438                 return 1;
2439         }
2440         jh = (struct reiserfs_journal_header *)(journal->j_header_bh->b_data);
2441         if (le32_to_cpu(jh->j_first_unflushed_offset) <
2442             SB_ONDISK_JOURNAL_SIZE(sb)
2443             && le32_to_cpu(jh->j_last_flush_trans_id) > 0) {
2444                 oldest_start =
2445                     SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2446                     le32_to_cpu(jh->j_first_unflushed_offset);
2447                 oldest_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2448                 newest_mount_id = le32_to_cpu(jh->j_mount_id);
2449                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2450                                "journal-1153: found in "
2451                                "header: first_unflushed_offset %d, last_flushed_trans_id "
2452                                "%lu", le32_to_cpu(jh->j_first_unflushed_offset),
2453                                le32_to_cpu(jh->j_last_flush_trans_id));
2454                 valid_journal_header = 1;
2455
2456                 /* now, we try to read the first unflushed offset.  If it is not valid,
2457                  ** there is nothing more we can do, and it makes no sense to read
2458                  ** through the whole log.
2459                  */
2460                 d_bh =
2461                     journal_bread(sb,
2462                                   SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2463                                   le32_to_cpu(jh->j_first_unflushed_offset));
2464                 ret = journal_transaction_is_valid(sb, d_bh, NULL, NULL);
2465                 if (!ret) {
2466                         continue_replay = 0;
2467                 }
2468                 brelse(d_bh);
2469                 goto start_log_replay;
2470         }
2471
2472         /* ok, there are transactions that need to be replayed.  start with the first log block, find
2473          ** all the valid transactions, and pick out the oldest.
2474          */
2475         while (continue_replay
2476                && cur_dblock <
2477                (SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2478                 SB_ONDISK_JOURNAL_SIZE(sb))) {
2479                 /* Note that it is required for blocksize of primary fs device and journal
2480                    device to be the same */
2481                 d_bh =
2482                     reiserfs_breada(journal->j_dev_bd, cur_dblock,
2483                                     sb->s_blocksize,
2484                                     SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2485                                     SB_ONDISK_JOURNAL_SIZE(sb));
2486                 ret =
2487                     journal_transaction_is_valid(sb, d_bh,
2488                                                  &oldest_invalid_trans_id,
2489                                                  &newest_mount_id);
2490                 if (ret == 1) {
2491                         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2492                         if (oldest_start == 0) {        /* init all oldest_ values */
2493                                 oldest_trans_id = get_desc_trans_id(desc);
2494                                 oldest_start = d_bh->b_blocknr;
2495                                 newest_mount_id = get_desc_mount_id(desc);
2496                                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2497                                                "journal-1179: Setting "
2498                                                "oldest_start to offset %llu, trans_id %lu",
2499                                                oldest_start -
2500                                                SB_ONDISK_JOURNAL_1st_BLOCK
2501                                                (sb), oldest_trans_id);
2502                         } else if (oldest_trans_id > get_desc_trans_id(desc)) {
2503                                 /* one we just read was older */
2504                                 oldest_trans_id = get_desc_trans_id(desc);
2505                                 oldest_start = d_bh->b_blocknr;
2506                                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2507                                                "journal-1180: Resetting "
2508                                                "oldest_start to offset %lu, trans_id %lu",
2509                                                oldest_start -
2510                                                SB_ONDISK_JOURNAL_1st_BLOCK
2511                                                (sb), oldest_trans_id);
2512                         }
2513                         if (newest_mount_id < get_desc_mount_id(desc)) {
2514                                 newest_mount_id = get_desc_mount_id(desc);
2515                                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2516                                                "journal-1299: Setting "
2517                                                "newest_mount_id to %d",
2518                                                get_desc_mount_id(desc));
2519                         }
2520                         cur_dblock += get_desc_trans_len(desc) + 2;
2521                 } else {
2522                         cur_dblock++;
2523                 }
2524                 brelse(d_bh);
2525         }
2526
2527       start_log_replay:
2528         cur_dblock = oldest_start;
2529         if (oldest_trans_id) {
2530                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2531                                "journal-1206: Starting replay "
2532                                "from offset %llu, trans_id %lu",
2533                                cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2534                                oldest_trans_id);
2535
2536         }
2537         replay_count = 0;
2538         while (continue_replay && oldest_trans_id > 0) {
2539                 ret =
2540                     journal_read_transaction(sb, cur_dblock, oldest_start,
2541                                              oldest_trans_id, newest_mount_id);
2542                 if (ret < 0) {
2543                         return ret;
2544                 } else if (ret != 0) {
2545                         break;
2546                 }
2547                 cur_dblock =
2548                     SB_ONDISK_JOURNAL_1st_BLOCK(sb) + journal->j_start;
2549                 replay_count++;
2550                 if (cur_dblock == oldest_start)
2551                         break;
2552         }
2553
2554         if (oldest_trans_id == 0) {
2555                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2556                                "journal-1225: No valid " "transactions found");
2557         }
2558         /* j_start does not get set correctly if we don't replay any transactions.
2559          ** if we had a valid journal_header, set j_start to the first unflushed transaction value,
2560          ** copy the trans_id from the header
2561          */
2562         if (valid_journal_header && replay_count == 0) {
2563                 journal->j_start = le32_to_cpu(jh->j_first_unflushed_offset);
2564                 journal->j_trans_id =
2565                     le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2566                 /* check for trans_id overflow */
2567                 if (journal->j_trans_id == 0)
2568                         journal->j_trans_id = 10;
2569                 journal->j_last_flush_trans_id =
2570                     le32_to_cpu(jh->j_last_flush_trans_id);
2571                 journal->j_mount_id = le32_to_cpu(jh->j_mount_id) + 1;
2572         } else {
2573                 journal->j_mount_id = newest_mount_id + 1;
2574         }
2575         reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1299: Setting "
2576                        "newest_mount_id to %lu", journal->j_mount_id);
2577         journal->j_first_unflushed_offset = journal->j_start;
2578         if (replay_count > 0) {
2579                 reiserfs_info(sb,
2580                               "replayed %d transactions in %lu seconds\n",
2581                               replay_count, get_seconds() - start);
2582         }
2583         if (!bdev_read_only(sb->s_bdev) &&
2584             _update_journal_header_block(sb, journal->j_start,
2585                                          journal->j_last_flush_trans_id)) {
2586                 /* replay failed, caller must call free_journal_ram and abort
2587                  ** the mount
2588                  */
2589                 return -1;
2590         }
2591         return 0;
2592 }
2593
2594 static struct reiserfs_journal_list *alloc_journal_list(struct super_block *s)
2595 {
2596         struct reiserfs_journal_list *jl;
2597         jl = kzalloc(sizeof(struct reiserfs_journal_list),
2598                      GFP_NOFS | __GFP_NOFAIL);
2599         INIT_LIST_HEAD(&jl->j_list);
2600         INIT_LIST_HEAD(&jl->j_working_list);
2601         INIT_LIST_HEAD(&jl->j_tail_bh_list);
2602         INIT_LIST_HEAD(&jl->j_bh_list);
2603         mutex_init(&jl->j_commit_mutex);
2604         SB_JOURNAL(s)->j_num_lists++;
2605         get_journal_list(jl);
2606         return jl;
2607 }
2608
2609 static void journal_list_init(struct super_block *sb)
2610 {
2611         SB_JOURNAL(sb)->j_current_jl = alloc_journal_list(sb);
2612 }
2613
2614 static int release_journal_dev(struct super_block *super,
2615                                struct reiserfs_journal *journal)
2616 {
2617         int result;
2618
2619         result = 0;
2620
2621         if (journal->j_dev_bd != NULL) {
2622                 if (journal->j_dev_bd->bd_dev != super->s_dev)
2623                         bd_release(journal->j_dev_bd);
2624                 result = blkdev_put(journal->j_dev_bd, journal->j_dev_mode);
2625                 journal->j_dev_bd = NULL;
2626         }
2627
2628         if (result != 0) {
2629                 reiserfs_warning(super, "sh-457",
2630                                  "Cannot release journal device: %i", result);
2631         }
2632         return result;
2633 }
2634
2635 static int journal_init_dev(struct super_block *super,
2636                             struct reiserfs_journal *journal,
2637                             const char *jdev_name)
2638 {
2639         int result;
2640         dev_t jdev;
2641         fmode_t blkdev_mode = FMODE_READ | FMODE_WRITE;
2642         char b[BDEVNAME_SIZE];
2643
2644         result = 0;
2645
2646         journal->j_dev_bd = NULL;
2647         jdev = SB_ONDISK_JOURNAL_DEVICE(super) ?
2648             new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super)) : super->s_dev;
2649
2650         if (bdev_read_only(super->s_bdev))
2651                 blkdev_mode = FMODE_READ;
2652
2653         /* there is no "jdev" option and journal is on separate device */
2654         if ((!jdev_name || !jdev_name[0])) {
2655                 journal->j_dev_bd = open_by_devnum(jdev, blkdev_mode);
2656                 journal->j_dev_mode = blkdev_mode;
2657                 if (IS_ERR(journal->j_dev_bd)) {
2658                         result = PTR_ERR(journal->j_dev_bd);
2659                         journal->j_dev_bd = NULL;
2660                         reiserfs_warning(super, "sh-458",
2661                                          "cannot init journal device '%s': %i",
2662                                          __bdevname(jdev, b), result);
2663                         return result;
2664                 } else if (jdev != super->s_dev) {
2665                         result = bd_claim(journal->j_dev_bd, journal);
2666                         if (result) {
2667                                 blkdev_put(journal->j_dev_bd, blkdev_mode);
2668                                 return result;
2669                         }
2670
2671                         set_blocksize(journal->j_dev_bd, super->s_blocksize);
2672                 }
2673
2674                 return 0;
2675         }
2676
2677         journal->j_dev_mode = blkdev_mode;
2678         journal->j_dev_bd = open_bdev_exclusive(jdev_name,
2679                                                 blkdev_mode, journal);
2680         if (IS_ERR(journal->j_dev_bd)) {
2681                 result = PTR_ERR(journal->j_dev_bd);
2682                 journal->j_dev_bd = NULL;
2683                 reiserfs_warning(super,
2684                                  "journal_init_dev: Cannot open '%s': %i",
2685                                  jdev_name, result);
2686                 return result;
2687         }
2688
2689         set_blocksize(journal->j_dev_bd, super->s_blocksize);
2690         reiserfs_info(super,
2691                       "journal_init_dev: journal device: %s\n",
2692                       bdevname(journal->j_dev_bd, b));
2693         return 0;
2694 }
2695
2696 /**
2697  * When creating/tuning a file system user can assign some
2698  * journal params within boundaries which depend on the ratio
2699  * blocksize/standard_blocksize.
2700  *
2701  * For blocks >= standard_blocksize transaction size should
2702  * be not less then JOURNAL_TRANS_MIN_DEFAULT, and not more
2703  * then JOURNAL_TRANS_MAX_DEFAULT.
2704  *
2705  * For blocks < standard_blocksize these boundaries should be
2706  * decreased proportionally.
2707  */
2708 #define REISERFS_STANDARD_BLKSIZE (4096)
2709
2710 static int check_advise_trans_params(struct super_block *sb,
2711                                      struct reiserfs_journal *journal)
2712 {
2713         if (journal->j_trans_max) {
2714                 /* Non-default journal params.
2715                    Do sanity check for them. */
2716                 int ratio = 1;
2717                 if (sb->s_blocksize < REISERFS_STANDARD_BLKSIZE)
2718                         ratio = REISERFS_STANDARD_BLKSIZE / sb->s_blocksize;
2719
2720                 if (journal->j_trans_max > JOURNAL_TRANS_MAX_DEFAULT / ratio ||
2721                     journal->j_trans_max < JOURNAL_TRANS_MIN_DEFAULT / ratio ||
2722                     SB_ONDISK_JOURNAL_SIZE(sb) / journal->j_trans_max <
2723                     JOURNAL_MIN_RATIO) {
2724                         reiserfs_warning(sb, "sh-462",
2725                                          "bad transaction max size (%u). "
2726                                          "FSCK?", journal->j_trans_max);
2727                         return 1;
2728                 }
2729                 if (journal->j_max_batch != (journal->j_trans_max) *
2730                         JOURNAL_MAX_BATCH_DEFAULT/JOURNAL_TRANS_MAX_DEFAULT) {
2731                         reiserfs_warning(sb, "sh-463",
2732                                          "bad transaction max batch (%u). "
2733                                          "FSCK?", journal->j_max_batch);
2734                         return 1;
2735                 }
2736         } else {
2737                 /* Default journal params.
2738                    The file system was created by old version
2739                    of mkreiserfs, so some fields contain zeros,
2740                    and we need to advise proper values for them */
2741                 if (sb->s_blocksize != REISERFS_STANDARD_BLKSIZE) {
2742                         reiserfs_warning(sb, "sh-464", "bad blocksize (%u)",
2743                                          sb->s_blocksize);
2744                         return 1;
2745                 }
2746                 journal->j_trans_max = JOURNAL_TRANS_MAX_DEFAULT;
2747                 journal->j_max_batch = JOURNAL_MAX_BATCH_DEFAULT;
2748                 journal->j_max_commit_age = JOURNAL_MAX_COMMIT_AGE;
2749         }
2750         return 0;
2751 }
2752
2753 /*
2754 ** must be called once on fs mount.  calls journal_read for you
2755 */
2756 int journal_init(struct super_block *sb, const char *j_dev_name,
2757                  int old_format, unsigned int commit_max_age)
2758 {
2759         int num_cnodes = SB_ONDISK_JOURNAL_SIZE(sb) * 2;
2760         struct buffer_head *bhjh;
2761         struct reiserfs_super_block *rs;
2762         struct reiserfs_journal_header *jh;
2763         struct reiserfs_journal *journal;
2764         struct reiserfs_journal_list *jl;
2765         char b[BDEVNAME_SIZE];
2766         int ret;
2767
2768         /*
2769          * Unlock here to avoid various RECLAIM-FS-ON <-> IN-RECLAIM-FS
2770          * dependency inversion warnings.
2771          */
2772         reiserfs_write_unlock(sb);
2773         journal = SB_JOURNAL(sb) = vmalloc(sizeof(struct reiserfs_journal));
2774         if (!journal) {
2775                 reiserfs_warning(sb, "journal-1256",
2776                                  "unable to get memory for journal structure");
2777                 reiserfs_write_lock(sb);
2778                 return 1;
2779         }
2780         memset(journal, 0, sizeof(struct reiserfs_journal));
2781         INIT_LIST_HEAD(&journal->j_bitmap_nodes);
2782         INIT_LIST_HEAD(&journal->j_prealloc_list);
2783         INIT_LIST_HEAD(&journal->j_working_list);
2784         INIT_LIST_HEAD(&journal->j_journal_list);
2785         journal->j_persistent_trans = 0;
2786         ret = reiserfs_allocate_list_bitmaps(sb, journal->j_list_bitmap,
2787                                            reiserfs_bmap_count(sb));
2788         reiserfs_write_lock(sb);
2789         if (ret)
2790                 goto free_and_return;
2791
2792         allocate_bitmap_nodes(sb);
2793
2794         /* reserved for journal area support */
2795         SB_JOURNAL_1st_RESERVED_BLOCK(sb) = (old_format ?
2796                                                  REISERFS_OLD_DISK_OFFSET_IN_BYTES
2797                                                  / sb->s_blocksize +
2798                                                  reiserfs_bmap_count(sb) +
2799                                                  1 :
2800                                                  REISERFS_DISK_OFFSET_IN_BYTES /
2801                                                  sb->s_blocksize + 2);
2802
2803         /* Sanity check to see is the standard journal fitting withing first bitmap
2804            (actual for small blocksizes) */
2805         if (!SB_ONDISK_JOURNAL_DEVICE(sb) &&
2806             (SB_JOURNAL_1st_RESERVED_BLOCK(sb) +
2807              SB_ONDISK_JOURNAL_SIZE(sb) > sb->s_blocksize * 8)) {
2808                 reiserfs_warning(sb, "journal-1393",
2809                                  "journal does not fit for area addressed "
2810                                  "by first of bitmap blocks. It starts at "
2811                                  "%u and its size is %u. Block size %ld",
2812                                  SB_JOURNAL_1st_RESERVED_BLOCK(sb),
2813                                  SB_ONDISK_JOURNAL_SIZE(sb),
2814                                  sb->s_blocksize);
2815                 goto free_and_return;
2816         }
2817
2818         /*
2819          * We need to unlock here to avoid creating the following
2820          * dependency:
2821          * reiserfs_lock -> sysfs_mutex
2822          * Because the reiserfs mmap path creates the following dependency:
2823          * mm->mmap -> reiserfs_lock, hence we have
2824          * mm->mmap -> reiserfs_lock ->sysfs_mutex
2825          * This would ends up in a circular dependency with sysfs readdir path
2826          * which does sysfs_mutex -> mm->mmap_sem
2827          * This is fine because the reiserfs lock is useless in mount path,
2828          * at least until we call journal_begin. We keep it for paranoid
2829          * reasons.
2830          */
2831         reiserfs_write_unlock(sb);
2832         if (journal_init_dev(sb, journal, j_dev_name) != 0) {
2833                 reiserfs_write_lock(sb);
2834                 reiserfs_warning(sb, "sh-462",
2835                                  "unable to initialize jornal device");
2836                 goto free_and_return;
2837         }
2838         reiserfs_write_lock(sb);
2839
2840         rs = SB_DISK_SUPER_BLOCK(sb);
2841
2842         /* read journal header */
2843         bhjh = journal_bread(sb,
2844                              SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2845                              SB_ONDISK_JOURNAL_SIZE(sb));
2846         if (!bhjh) {
2847                 reiserfs_warning(sb, "sh-459",
2848                                  "unable to read journal header");
2849                 goto free_and_return;
2850         }
2851         jh = (struct reiserfs_journal_header *)(bhjh->b_data);
2852
2853         /* make sure that journal matches to the super block */
2854         if (is_reiserfs_jr(rs)
2855             && (le32_to_cpu(jh->jh_journal.jp_journal_magic) !=
2856                 sb_jp_journal_magic(rs))) {
2857                 reiserfs_warning(sb, "sh-460",
2858                                  "journal header magic %x (device %s) does "
2859                                  "not match to magic found in super block %x",
2860                                  jh->jh_journal.jp_journal_magic,
2861                                  bdevname(journal->j_dev_bd, b),
2862                                  sb_jp_journal_magic(rs));
2863                 brelse(bhjh);
2864                 goto free_and_return;
2865         }
2866
2867         journal->j_trans_max = le32_to_cpu(jh->jh_journal.jp_journal_trans_max);
2868         journal->j_max_batch = le32_to_cpu(jh->jh_journal.jp_journal_max_batch);
2869         journal->j_max_commit_age =
2870             le32_to_cpu(jh->jh_journal.jp_journal_max_commit_age);
2871         journal->j_max_trans_age = JOURNAL_MAX_TRANS_AGE;
2872
2873         if (check_advise_trans_params(sb, journal) != 0)
2874                 goto free_and_return;
2875         journal->j_default_max_commit_age = journal->j_max_commit_age;
2876
2877         if (commit_max_age != 0) {
2878                 journal->j_max_commit_age = commit_max_age;
2879                 journal->j_max_trans_age = commit_max_age;
2880         }
2881
2882         reiserfs_info(sb, "journal params: device %s, size %u, "
2883                       "journal first block %u, max trans len %u, max batch %u, "
2884                       "max commit age %u, max trans age %u\n",
2885                       bdevname(journal->j_dev_bd, b),
2886                       SB_ONDISK_JOURNAL_SIZE(sb),
2887                       SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2888                       journal->j_trans_max,
2889                       journal->j_max_batch,
2890                       journal->j_max_commit_age, journal->j_max_trans_age);
2891
2892         brelse(bhjh);
2893
2894         journal->j_list_bitmap_index = 0;
2895         journal_list_init(sb);
2896
2897         memset(journal->j_list_hash_table, 0,
2898                JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
2899
2900         INIT_LIST_HEAD(&journal->j_dirty_buffers);
2901         spin_lock_init(&journal->j_dirty_buffers_lock);
2902
2903         journal->j_start = 0;
2904         journal->j_len = 0;
2905         journal->j_len_alloc = 0;
2906         atomic_set(&(journal->j_wcount), 0);
2907         atomic_set(&(journal->j_async_throttle), 0);
2908         journal->j_bcount = 0;
2909         journal->j_trans_start_time = 0;
2910         journal->j_last = NULL;
2911         journal->j_first = NULL;
2912         init_waitqueue_head(&(journal->j_join_wait));
2913         mutex_init(&journal->j_mutex);
2914         mutex_init(&journal->j_flush_mutex);
2915
2916         journal->j_trans_id = 10;
2917         journal->j_mount_id = 10;
2918         journal->j_state = 0;
2919         atomic_set(&(journal->j_jlock), 0);
2920         reiserfs_write_unlock(sb);
2921         journal->j_cnode_free_list = allocate_cnodes(num_cnodes);
2922         reiserfs_write_lock(sb);
2923         journal->j_cnode_free_orig = journal->j_cnode_free_list;
2924         journal->j_cnode_free = journal->j_cnode_free_list ? num_cnodes : 0;
2925         journal->j_cnode_used = 0;
2926         journal->j_must_wait = 0;
2927
2928         if (journal->j_cnode_free == 0) {
2929                 reiserfs_warning(sb, "journal-2004", "Journal cnode memory "
2930                                  "allocation failed (%ld bytes). Journal is "
2931                                  "too large for available memory. Usually "
2932                                  "this is due to a journal that is too large.",
2933                                  sizeof (struct reiserfs_journal_cnode) * num_cnodes);
2934                 goto free_and_return;
2935         }
2936
2937         init_journal_hash(sb);
2938         jl = journal->j_current_jl;
2939         jl->j_list_bitmap = get_list_bitmap(sb, jl);
2940         if (!jl->j_list_bitmap) {
2941                 reiserfs_warning(sb, "journal-2005",
2942                                  "get_list_bitmap failed for journal list 0");
2943                 goto free_and_return;
2944         }
2945         if (journal_read(sb) < 0) {
2946                 reiserfs_warning(sb, "reiserfs-2006",
2947                                  "Replay Failure, unable to mount");
2948                 goto free_and_return;
2949         }
2950
2951         reiserfs_mounted_fs_count++;
2952         if (reiserfs_mounted_fs_count <= 1) {
2953                 reiserfs_write_unlock(sb);
2954                 commit_wq = create_workqueue("reiserfs");
2955                 reiserfs_write_lock(sb);
2956         }
2957
2958         INIT_DELAYED_WORK(&journal->j_work, flush_async_commits);
2959         journal->j_work_sb = sb;
2960         return 0;
2961       free_and_return:
2962         free_journal_ram(sb);
2963         return 1;
2964 }
2965
2966 /*
2967 ** test for a polite end of the current transaction.  Used by file_write, and should
2968 ** be used by delete to make sure they don't write more than can fit inside a single
2969 ** transaction
2970 */
2971 int journal_transaction_should_end(struct reiserfs_transaction_handle *th,
2972                                    int new_alloc)
2973 {
2974         struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2975         time_t now = get_seconds();
2976         /* cannot restart while nested */
2977         BUG_ON(!th->t_trans_id);
2978         if (th->t_refcount > 1)
2979                 return 0;
2980         if (journal->j_must_wait > 0 ||
2981             (journal->j_len_alloc + new_alloc) >= journal->j_max_batch ||
2982             atomic_read(&(journal->j_jlock)) ||
2983             (now - journal->j_trans_start_time) > journal->j_max_trans_age ||
2984             journal->j_cnode_free < (journal->j_trans_max * 3)) {
2985                 return 1;
2986         }
2987         /* protected by the BKL here */
2988         journal->j_len_alloc += new_alloc;
2989         th->t_blocks_allocated += new_alloc ;
2990         return 0;
2991 }
2992
2993 /* this must be called inside a transaction, and requires the
2994 ** kernel_lock to be held
2995 */
2996 void reiserfs_block_writes(struct reiserfs_transaction_handle *th)
2997 {
2998         struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2999         BUG_ON(!th->t_trans_id);
3000         journal->j_must_wait = 1;
3001         set_bit(J_WRITERS_BLOCKED, &journal->j_state);
3002         return;
3003 }
3004
3005 /* this must be called without a transaction started, and does not
3006 ** require BKL
3007 */
3008 void reiserfs_allow_writes(struct super_block *s)
3009 {
3010         struct reiserfs_journal *journal = SB_JOURNAL(s);
3011         clear_bit(J_WRITERS_BLOCKED, &journal->j_state);
3012         wake_up(&journal->j_join_wait);
3013 }
3014
3015 /* this must be called without a transaction started, and does not
3016 ** require BKL
3017 */
3018 void reiserfs_wait_on_write_block(struct super_block *s)
3019 {
3020         struct reiserfs_journal *journal = SB_JOURNAL(s);
3021         wait_event(journal->j_join_wait,
3022                    !test_bit(J_WRITERS_BLOCKED, &journal->j_state));
3023 }
3024
3025 static void queue_log_writer(struct super_block *s)
3026 {
3027         wait_queue_t wait;
3028         struct reiserfs_journal *journal = SB_JOURNAL(s);
3029         set_bit(J_WRITERS_QUEUED, &journal->j_state);
3030
3031         /*
3032          * we don't want to use wait_event here because
3033          * we only want to wait once.
3034          */
3035         init_waitqueue_entry(&wait, current);
3036         add_wait_queue(&journal->j_join_wait, &wait);
3037         set_current_state(TASK_UNINTERRUPTIBLE);
3038         if (test_bit(J_WRITERS_QUEUED, &journal->j_state)) {
3039                 reiserfs_write_unlock(s);
3040                 schedule();
3041                 reiserfs_write_lock(s);
3042         }
3043         __set_current_state(TASK_RUNNING);
3044         remove_wait_queue(&journal->j_join_wait, &wait);
3045 }
3046
3047 static void wake_queued_writers(struct super_block *s)
3048 {
3049         struct reiserfs_journal *journal = SB_JOURNAL(s);
3050         if (test_and_clear_bit(J_WRITERS_QUEUED, &journal->j_state))
3051                 wake_up(&journal->j_join_wait);
3052 }
3053
3054 static void let_transaction_grow(struct super_block *sb, unsigned int trans_id)
3055 {
3056         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3057         unsigned long bcount = journal->j_bcount;
3058         while (1) {
3059                 reiserfs_write_unlock(sb);
3060                 schedule_timeout_uninterruptible(1);
3061                 reiserfs_write_lock(sb);
3062                 journal->j_current_jl->j_state |= LIST_COMMIT_PENDING;
3063                 while ((atomic_read(&journal->j_wcount) > 0 ||
3064                         atomic_read(&journal->j_jlock)) &&
3065                        journal->j_trans_id == trans_id) {
3066                         queue_log_writer(sb);
3067                 }
3068                 if (journal->j_trans_id != trans_id)
3069                         break;
3070                 if (bcount == journal->j_bcount)
3071                         break;
3072                 bcount = journal->j_bcount;
3073         }
3074 }
3075
3076 /* join == true if you must join an existing transaction.
3077 ** join == false if you can deal with waiting for others to finish
3078 **
3079 ** this will block until the transaction is joinable.  send the number of blocks you
3080 ** expect to use in nblocks.
3081 */
3082 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
3083                               struct super_block *sb, unsigned long nblocks,
3084                               int join)
3085 {
3086         time_t now = get_seconds();
3087         unsigned int old_trans_id;
3088         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3089         struct reiserfs_transaction_handle myth;
3090         int sched_count = 0;
3091         int retval;
3092
3093         reiserfs_check_lock_depth(sb, "journal_begin");
3094         BUG_ON(nblocks > journal->j_trans_max);
3095
3096         PROC_INFO_INC(sb, journal.journal_being);
3097         /* set here for journal_join */
3098         th->t_refcount = 1;
3099         th->t_super = sb;
3100
3101       relock:
3102         lock_journal(sb);
3103         if (join != JBEGIN_ABORT && reiserfs_is_journal_aborted(journal)) {
3104                 unlock_journal(sb);
3105                 retval = journal->j_errno;
3106                 goto out_fail;
3107         }
3108         journal->j_bcount++;
3109
3110         if (test_bit(J_WRITERS_BLOCKED, &journal->j_state)) {
3111                 unlock_journal(sb);
3112                 reiserfs_write_unlock(sb);
3113                 reiserfs_wait_on_write_block(sb);
3114                 reiserfs_write_lock(sb);
3115                 PROC_INFO_INC(sb, journal.journal_relock_writers);
3116                 goto relock;
3117         }
3118         now = get_seconds();
3119
3120         /* if there is no room in the journal OR
3121          ** if this transaction is too old, and we weren't called joinable, wait for it to finish before beginning
3122          ** we don't sleep if there aren't other writers
3123          */
3124
3125         if ((!join && journal->j_must_wait > 0) ||
3126             (!join
3127              && (journal->j_len_alloc + nblocks + 2) >= journal->j_max_batch)
3128             || (!join && atomic_read(&journal->j_wcount) > 0
3129                 && journal->j_trans_start_time > 0
3130                 && (now - journal->j_trans_start_time) >
3131                 journal->j_max_trans_age) || (!join
3132                                               && atomic_read(&journal->j_jlock))
3133             || (!join && journal->j_cnode_free < (journal->j_trans_max * 3))) {
3134
3135                 old_trans_id = journal->j_trans_id;
3136                 unlock_journal(sb);     /* allow others to finish this transaction */
3137
3138                 if (!join && (journal->j_len_alloc + nblocks + 2) >=
3139                     journal->j_max_batch &&
3140                     ((journal->j_len + nblocks + 2) * 100) <
3141                     (journal->j_len_alloc * 75)) {
3142                         if (atomic_read(&journal->j_wcount) > 10) {
3143                                 sched_count++;
3144                                 queue_log_writer(sb);
3145                                 goto relock;
3146                         }
3147                 }
3148                 /* don't mess with joining the transaction if all we have to do is
3149                  * wait for someone else to do a commit
3150                  */
3151                 if (atomic_read(&journal->j_jlock)) {
3152                         while (journal->j_trans_id == old_trans_id &&
3153                                atomic_read(&journal->j_jlock)) {
3154                                 queue_log_writer(sb);
3155                         }
3156                         goto relock;
3157                 }
3158                 retval = journal_join(&myth, sb, 1);
3159                 if (retval)
3160                         goto out_fail;
3161
3162                 /* someone might have ended the transaction while we joined */
3163                 if (old_trans_id != journal->j_trans_id) {
3164                         retval = do_journal_end(&myth, sb, 1, 0);
3165                 } else {
3166                         retval = do_journal_end(&myth, sb, 1, COMMIT_NOW);
3167                 }
3168
3169                 if (retval)
3170                         goto out_fail;
3171
3172                 PROC_INFO_INC(sb, journal.journal_relock_wcount);
3173                 goto relock;
3174         }
3175         /* we are the first writer, set trans_id */
3176         if (journal->j_trans_start_time == 0) {
3177                 journal->j_trans_start_time = get_seconds();
3178         }
3179         atomic_inc(&(journal->j_wcount));
3180         journal->j_len_alloc += nblocks;
3181         th->t_blocks_logged = 0;
3182         th->t_blocks_allocated = nblocks;
3183         th->t_trans_id = journal->j_trans_id;
3184         unlock_journal(sb);
3185         INIT_LIST_HEAD(&th->t_list);
3186         get_fs_excl();
3187         return 0;
3188
3189       out_fail:
3190         memset(th, 0, sizeof(*th));
3191         /* Re-set th->t_super, so we can properly keep track of how many
3192          * persistent transactions there are. We need to do this so if this
3193          * call is part of a failed restart_transaction, we can free it later */
3194         th->t_super = sb;
3195         return retval;
3196 }
3197
3198 struct reiserfs_transaction_handle *reiserfs_persistent_transaction(struct
3199                                                                     super_block
3200                                                                     *s,
3201                                                                     int nblocks)
3202 {
3203         int ret;
3204         struct reiserfs_transaction_handle *th;
3205
3206         /* if we're nesting into an existing transaction.  It will be
3207          ** persistent on its own
3208          */
3209         if (reiserfs_transaction_running(s)) {
3210                 th = current->journal_info;
3211                 th->t_refcount++;
3212                 BUG_ON(th->t_refcount < 2);
3213
3214                 return th;
3215         }
3216         th = kmalloc(sizeof(struct reiserfs_transaction_handle), GFP_NOFS);
3217         if (!th)
3218                 return NULL;
3219         ret = journal_begin(th, s, nblocks);
3220         if (ret) {
3221                 kfree(th);
3222                 return NULL;
3223         }
3224
3225         SB_JOURNAL(s)->j_persistent_trans++;
3226         return th;
3227 }
3228
3229 int reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle *th)
3230 {
3231         struct super_block *s = th->t_super;
3232         int ret = 0;
3233         if (th->t_trans_id)
3234                 ret = journal_end(th, th->t_super, th->t_blocks_allocated);
3235         else
3236                 ret = -EIO;
3237         if (th->t_refcount == 0) {
3238                 SB_JOURNAL(s)->j_persistent_trans--;
3239                 kfree(th);
3240         }
3241         return ret;
3242 }
3243
3244 static int journal_join(struct reiserfs_transaction_handle *th,
3245                         struct super_block *sb, unsigned long nblocks)
3246 {
3247         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3248
3249         /* this keeps do_journal_end from NULLing out the current->journal_info
3250          ** pointer
3251          */
3252         th->t_handle_save = cur_th;
3253         BUG_ON(cur_th && cur_th->t_refcount > 1);
3254         return do_journal_begin_r(th, sb, nblocks, JBEGIN_JOIN);
3255 }
3256
3257 int journal_join_abort(struct reiserfs_transaction_handle *th,
3258                        struct super_block *sb, unsigned long nblocks)
3259 {
3260         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3261
3262         /* this keeps do_journal_end from NULLing out the current->journal_info
3263          ** pointer
3264          */
3265         th->t_handle_save = cur_th;
3266         BUG_ON(cur_th && cur_th->t_refcount > 1);
3267         return do_journal_begin_r(th, sb, nblocks, JBEGIN_ABORT);
3268 }
3269
3270 int journal_begin(struct reiserfs_transaction_handle *th,
3271                   struct super_block *sb, unsigned long nblocks)
3272 {
3273         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3274         int ret;
3275
3276         th->t_handle_save = NULL;
3277         if (cur_th) {
3278                 /* we are nesting into the current transaction */
3279                 if (cur_th->t_super == sb) {
3280                         BUG_ON(!cur_th->t_refcount);
3281                         cur_th->t_refcount++;
3282                         memcpy(th, cur_th, sizeof(*th));
3283                         if (th->t_refcount <= 1)
3284                                 reiserfs_warning(sb, "reiserfs-2005",
3285                                                  "BAD: refcount <= 1, but "
3286                                                  "journal_info != 0");
3287                         return 0;
3288                 } else {
3289                         /* we've ended up with a handle from a different filesystem.
3290                          ** save it and restore on journal_end.  This should never
3291                          ** really happen...
3292                          */
3293                         reiserfs_warning(sb, "clm-2100",
3294                                          "nesting info a different FS");
3295                         th->t_handle_save = current->journal_info;
3296                         current->journal_info = th;
3297                 }
3298         } else {
3299                 current->journal_info = th;
3300         }
3301         ret = do_journal_begin_r(th, sb, nblocks, JBEGIN_REG);
3302         BUG_ON(current->journal_info != th);
3303
3304         /* I guess this boils down to being the reciprocal of clm-2100 above.
3305          * If do_journal_begin_r fails, we need to put it back, since journal_end
3306          * won't be called to do it. */
3307         if (ret)
3308                 current->journal_info = th->t_handle_save;
3309         else
3310                 BUG_ON(!th->t_refcount);
3311
3312         return ret;
3313 }
3314
3315 /*
3316 ** puts bh into the current transaction.  If it was already there, reorders removes the
3317 ** old pointers from the hash, and puts new ones in (to make sure replay happen in the right order).
3318 **
3319 ** if it was dirty, cleans and files onto the clean list.  I can't let it be dirty again until the
3320 ** transaction is committed.
3321 **
3322 ** if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len.
3323 */
3324 int journal_mark_dirty(struct reiserfs_transaction_handle *th,
3325                        struct super_block *sb, struct buffer_head *bh)
3326 {
3327         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3328         struct reiserfs_journal_cnode *cn = NULL;
3329         int count_already_incd = 0;
3330         int prepared = 0;
3331         BUG_ON(!th->t_trans_id);
3332
3333         PROC_INFO_INC(sb, journal.mark_dirty);
3334         if (th->t_trans_id != journal->j_trans_id) {
3335                 reiserfs_panic(th->t_super, "journal-1577",
3336                                "handle trans id %ld != current trans id %ld",
3337                                th->t_trans_id, journal->j_trans_id);
3338         }
3339
3340         sb->s_dirt = 1;
3341
3342         prepared = test_clear_buffer_journal_prepared(bh);
3343         clear_buffer_journal_restore_dirty(bh);
3344         /* already in this transaction, we are done */
3345         if (buffer_journaled(bh)) {
3346                 PROC_INFO_INC(sb, journal.mark_dirty_already);
3347                 return 0;
3348         }
3349
3350         /* this must be turned into a panic instead of a warning.  We can't allow
3351          ** a dirty or journal_dirty or locked buffer to be logged, as some changes
3352          ** could get to disk too early.  NOT GOOD.
3353          */
3354         if (!prepared || buffer_dirty(bh)) {
3355                 reiserfs_warning(sb, "journal-1777",
3356                                  "buffer %llu bad state "
3357                                  "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT",
3358                                  (unsigned long long)bh->b_blocknr,
3359                                  prepared ? ' ' : '!',
3360                                  buffer_locked(bh) ? ' ' : '!',
3361                                  buffer_dirty(bh) ? ' ' : '!',
3362                                  buffer_journal_dirty(bh) ? ' ' : '!');
3363         }
3364
3365         if (atomic_read(&(journal->j_wcount)) <= 0) {
3366                 reiserfs_warning(sb, "journal-1409",
3367                                  "returning because j_wcount was %d",
3368                                  atomic_read(&(journal->j_wcount)));
3369                 return 1;
3370         }
3371         /* this error means I've screwed up, and we've overflowed the transaction.
3372          ** Nothing can be done here, except make the FS readonly or panic.
3373          */
3374         if (journal->j_len >= journal->j_trans_max) {
3375                 reiserfs_panic(th->t_super, "journal-1413",
3376                                "j_len (%lu) is too big",
3377                                journal->j_len);
3378         }
3379
3380         if (buffer_journal_dirty(bh)) {
3381                 count_already_incd = 1;
3382                 PROC_INFO_INC(sb, journal.mark_dirty_notjournal);
3383                 clear_buffer_journal_dirty(bh);
3384         }
3385
3386         if (journal->j_len > journal->j_len_alloc) {
3387                 journal->j_len_alloc = journal->j_len + JOURNAL_PER_BALANCE_CNT;
3388         }
3389
3390         set_buffer_journaled(bh);
3391
3392         /* now put this guy on the end */
3393         if (!cn) {
3394                 cn = get_cnode(sb);
3395                 if (!cn) {
3396                         reiserfs_panic(sb, "journal-4", "get_cnode failed!");
3397                 }
3398
3399                 if (th->t_blocks_logged == th->t_blocks_allocated) {
3400                         th->t_blocks_allocated += JOURNAL_PER_BALANCE_CNT;
3401                         journal->j_len_alloc += JOURNAL_PER_BALANCE_CNT;
3402                 }
3403                 th->t_blocks_logged++;
3404                 journal->j_len++;
3405
3406                 cn->bh = bh;
3407                 cn->blocknr = bh->b_blocknr;
3408                 cn->sb = sb;
3409                 cn->jlist = NULL;
3410                 insert_journal_hash(journal->j_hash_table, cn);
3411                 if (!count_already_incd) {
3412                         get_bh(bh);
3413                 }
3414         }
3415         cn->next = NULL;
3416         cn->prev = journal->j_last;
3417         cn->bh = bh;
3418         if (journal->j_last) {
3419                 journal->j_last->next = cn;
3420                 journal->j_last = cn;
3421         } else {
3422                 journal->j_first = cn;
3423                 journal->j_last = cn;
3424         }
3425         return 0;
3426 }
3427
3428 int journal_end(struct reiserfs_transaction_handle *th,
3429                 struct super_block *sb, unsigned long nblocks)
3430 {
3431         if (!current->journal_info && th->t_refcount > 1)
3432                 reiserfs_warning(sb, "REISER-NESTING",
3433                                  "th NULL, refcount %d", th->t_refcount);
3434
3435         if (!th->t_trans_id) {
3436                 WARN_ON(1);
3437                 return -EIO;
3438         }
3439
3440         th->t_refcount--;
3441         if (th->t_refcount > 0) {
3442                 struct reiserfs_transaction_handle *cur_th =
3443                     current->journal_info;
3444
3445                 /* we aren't allowed to close a nested transaction on a different
3446                  ** filesystem from the one in the task struct
3447                  */
3448                 BUG_ON(cur_th->t_super != th->t_super);
3449
3450                 if (th != cur_th) {
3451                         memcpy(current->journal_info, th, sizeof(*th));
3452                         th->t_trans_id = 0;
3453                 }
3454                 return 0;
3455         } else {
3456                 return do_journal_end(th, sb, nblocks, 0);
3457         }
3458 }
3459
3460 /* removes from the current transaction, relsing and descrementing any counters.
3461 ** also files the removed buffer directly onto the clean list
3462 **
3463 ** called by journal_mark_freed when a block has been deleted
3464 **
3465 ** returns 1 if it cleaned and relsed the buffer. 0 otherwise
3466 */
3467 static int remove_from_transaction(struct super_block *sb,
3468                                    b_blocknr_t blocknr, int already_cleaned)
3469 {
3470         struct buffer_head *bh;
3471         struct reiserfs_journal_cnode *cn;
3472         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3473         int ret = 0;
3474
3475         cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr);
3476         if (!cn || !cn->bh) {
3477                 return ret;
3478         }
3479         bh = cn->bh;
3480         if (cn->prev) {
3481                 cn->prev->next = cn->next;
3482         }
3483         if (cn->next) {
3484                 cn->next->prev = cn->prev;
3485         }
3486         if (cn == journal->j_first) {
3487                 journal->j_first = cn->next;
3488         }
3489         if (cn == journal->j_last) {
3490                 journal->j_last = cn->prev;
3491         }
3492         if (bh)
3493                 remove_journal_hash(sb, journal->j_hash_table, NULL,
3494                                     bh->b_blocknr, 0);
3495         clear_buffer_journaled(bh);     /* don't log this one */
3496
3497         if (!already_cleaned) {
3498                 clear_buffer_journal_dirty(bh);
3499                 clear_buffer_dirty(bh);
3500                 clear_buffer_journal_test(bh);
3501                 put_bh(bh);
3502                 if (atomic_read(&(bh->b_count)) < 0) {
3503                         reiserfs_warning(sb, "journal-1752",
3504                                          "b_count < 0");
3505                 }
3506                 ret = 1;
3507         }
3508         journal->j_len--;
3509         journal->j_len_alloc--;
3510         free_cnode(sb, cn);
3511         return ret;
3512 }
3513
3514 /*
3515 ** for any cnode in a journal list, it can only be dirtied of all the
3516 ** transactions that include it are committed to disk.
3517 ** this checks through each transaction, and returns 1 if you are allowed to dirty,
3518 ** and 0 if you aren't
3519 **
3520 ** it is called by dirty_journal_list, which is called after flush_commit_list has gotten all the log
3521 ** blocks for a given transaction on disk
3522 **
3523 */
3524 static int can_dirty(struct reiserfs_journal_cnode *cn)
3525 {
3526         struct super_block *sb = cn->sb;
3527         b_blocknr_t blocknr = cn->blocknr;
3528         struct reiserfs_journal_cnode *cur = cn->hprev;
3529         int can_dirty = 1;
3530
3531         /* first test hprev.  These are all newer than cn, so any node here
3532          ** with the same block number and dev means this node can't be sent
3533          ** to disk right now.
3534          */
3535         while (cur && can_dirty) {
3536                 if (cur->jlist && cur->bh && cur->blocknr && cur->sb == sb &&
3537                     cur->blocknr == blocknr) {
3538                         can_dirty = 0;
3539                 }
3540                 cur = cur->hprev;
3541         }
3542         /* then test hnext.  These are all older than cn.  As long as they
3543          ** are committed to the log, it is safe to write cn to disk
3544          */
3545         cur = cn->hnext;
3546         while (cur && can_dirty) {
3547                 if (cur->jlist && cur->jlist->j_len > 0 &&
3548                     atomic_read(&(cur->jlist->j_commit_left)) > 0 && cur->bh &&
3549                     cur->blocknr && cur->sb == sb && cur->blocknr == blocknr) {
3550                         can_dirty = 0;
3551                 }
3552                 cur = cur->hnext;
3553         }
3554         return can_dirty;
3555 }
3556
3557 /* syncs the commit blocks, but does not force the real buffers to disk
3558 ** will wait until the current transaction is done/committed before returning
3559 */
3560 int journal_end_sync(struct reiserfs_transaction_handle *th,
3561                      struct super_block *sb, unsigned long nblocks)
3562 {
3563         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3564
3565         BUG_ON(!th->t_trans_id);
3566         /* you can sync while nested, very, very bad */
3567         BUG_ON(th->t_refcount > 1);
3568         if (journal->j_len == 0) {
3569                 reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3570                                              1);
3571                 journal_mark_dirty(th, sb, SB_BUFFER_WITH_SB(sb));
3572         }
3573         return do_journal_end(th, sb, nblocks, COMMIT_NOW | WAIT);
3574 }
3575
3576 /*
3577 ** writeback the pending async commits to disk
3578 */
3579 static void flush_async_commits(struct work_struct *work)
3580 {
3581         struct reiserfs_journal *journal =
3582                 container_of(work, struct reiserfs_journal, j_work.work);
3583         struct super_block *sb = journal->j_work_sb;
3584         struct reiserfs_journal_list *jl;
3585         struct list_head *entry;
3586
3587         reiserfs_write_lock(sb);
3588         if (!list_empty(&journal->j_journal_list)) {
3589                 /* last entry is the youngest, commit it and you get everything */
3590                 entry = journal->j_journal_list.prev;
3591                 jl = JOURNAL_LIST_ENTRY(entry);
3592                 flush_commit_list(sb, jl, 1);
3593         }
3594         reiserfs_write_unlock(sb);
3595 }
3596
3597 /*
3598 ** flushes any old transactions to disk
3599 ** ends the current transaction if it is too old
3600 */
3601 int reiserfs_flush_old_commits(struct super_block *sb)
3602 {
3603         time_t now;
3604         struct reiserfs_transaction_handle th;
3605         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3606
3607         now = get_seconds();
3608         /* safety check so we don't flush while we are replaying the log during
3609          * mount
3610          */
3611         if (list_empty(&journal->j_journal_list)) {
3612                 return 0;
3613         }
3614
3615         /* check the current transaction.  If there are no writers, and it is
3616          * too old, finish it, and force the commit blocks to disk
3617          */
3618         if (atomic_read(&journal->j_wcount) <= 0 &&
3619             journal->j_trans_start_time > 0 &&
3620             journal->j_len > 0 &&
3621             (now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3622                 if (!journal_join(&th, sb, 1)) {
3623                         reiserfs_prepare_for_journal(sb,
3624                                                      SB_BUFFER_WITH_SB(sb),
3625                                                      1);
3626                         journal_mark_dirty(&th, sb,
3627                                            SB_BUFFER_WITH_SB(sb));
3628
3629                         /* we're only being called from kreiserfsd, it makes no sense to do
3630                          ** an async commit so that kreiserfsd can do it later
3631                          */
3632                         do_journal_end(&th, sb, 1, COMMIT_NOW | WAIT);
3633                 }
3634         }
3635         return sb->s_dirt;
3636 }
3637
3638 /*
3639 ** returns 0 if do_journal_end should return right away, returns 1 if do_journal_end should finish the commit
3640 **
3641 ** if the current transaction is too old, but still has writers, this will wait on j_join_wait until all
3642 ** the writers are done.  By the time it wakes up, the transaction it was called has already ended, so it just
3643 ** flushes the commit list and returns 0.
3644 **
3645 ** Won't batch when flush or commit_now is set.  Also won't batch when others are waiting on j_join_wait.
3646 **
3647 ** Note, we can't allow the journal_end to proceed while there are still writers in the log.
3648 */
3649 static int check_journal_end(struct reiserfs_transaction_handle *th,
3650                              struct super_block *sb, unsigned long nblocks,
3651                              int flags)
3652 {
3653
3654         time_t now;
3655         int flush = flags & FLUSH_ALL;
3656         int commit_now = flags & COMMIT_NOW;
3657         int wait_on_commit = flags & WAIT;
3658         struct reiserfs_journal_list *jl;
3659         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3660
3661         BUG_ON(!th->t_trans_id);
3662
3663         if (th->t_trans_id != journal->j_trans_id) {
3664                 reiserfs_panic(th->t_super, "journal-1577",
3665                                "handle trans id %ld != current trans id %ld",
3666                                th->t_trans_id, journal->j_trans_id);
3667         }
3668
3669         journal->j_len_alloc -= (th->t_blocks_allocated - th->t_blocks_logged);
3670         if (atomic_read(&(journal->j_wcount)) > 0) {    /* <= 0 is allowed.  unmounting might not call begin */
3671                 atomic_dec(&(journal->j_wcount));
3672         }
3673
3674         /* BUG, deal with case where j_len is 0, but people previously freed blocks need to be released
3675          ** will be dealt with by next transaction that actually writes something, but should be taken
3676          ** care of in this trans
3677          */
3678         BUG_ON(journal->j_len == 0);
3679
3680         /* if wcount > 0, and we are called to with flush or commit_now,
3681          ** we wait on j_join_wait.  We will wake up when the last writer has
3682          ** finished the transaction, and started it on its way to the disk.
3683          ** Then, we flush the commit or journal list, and just return 0
3684          ** because the rest of journal end was already done for this transaction.
3685          */
3686         if (atomic_read(&(journal->j_wcount)) > 0) {
3687                 if (flush || commit_now) {
3688                         unsigned trans_id;
3689
3690                         jl = journal->j_current_jl;
3691                         trans_id = jl->j_trans_id;
3692                         if (wait_on_commit)
3693                                 jl->j_state |= LIST_COMMIT_PENDING;
3694                         atomic_set(&(journal->j_jlock), 1);
3695                         if (flush) {
3696                                 journal->j_next_full_flush = 1;
3697                         }
3698                         unlock_journal(sb);
3699
3700                         /* sleep while the current transaction is still j_jlocked */
3701                         while (journal->j_trans_id == trans_id) {
3702                                 if (atomic_read(&journal->j_jlock)) {
3703                                         queue_log_writer(sb);
3704                                 } else {
3705                                         lock_journal(sb);
3706                                         if (journal->j_trans_id == trans_id) {
3707                                                 atomic_set(&(journal->j_jlock),
3708                                                            1);
3709                                         }
3710                                         unlock_journal(sb);
3711                                 }
3712                         }
3713                         BUG_ON(journal->j_trans_id == trans_id);
3714
3715                         if (commit_now
3716                             && journal_list_still_alive(sb, trans_id)
3717                             && wait_on_commit) {
3718                                 flush_commit_list(sb, jl, 1);
3719                         }
3720                         return 0;
3721                 }
3722                 unlock_journal(sb);
3723                 return 0;
3724         }
3725
3726         /* deal with old transactions where we are the last writers */
3727         now = get_seconds();
3728         if ((now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3729                 commit_now = 1;
3730                 journal->j_next_async_flush = 1;
3731         }
3732         /* don't batch when someone is waiting on j_join_wait */
3733         /* don't batch when syncing the commit or flushing the whole trans */
3734         if (!(journal->j_must_wait > 0) && !(atomic_read(&(journal->j_jlock)))
3735             && !flush && !commit_now && (journal->j_len < journal->j_max_batch)
3736             && journal->j_len_alloc < journal->j_max_batch
3737             && journal->j_cnode_free > (journal->j_trans_max * 3)) {
3738                 journal->j_bcount++;
3739                 unlock_journal(sb);
3740                 return 0;
3741         }
3742
3743         if (journal->j_start > SB_ONDISK_JOURNAL_SIZE(sb)) {
3744                 reiserfs_panic(sb, "journal-003",
3745                                "j_start (%ld) is too high",
3746                                journal->j_start);
3747         }
3748         return 1;
3749 }
3750
3751 /*
3752 ** Does all the work that makes deleting blocks safe.
3753 ** when deleting a block mark BH_JNew, just remove it from the current transaction, clean it's buffer_head and move on.
3754 **
3755 ** otherwise:
3756 ** set a bit for the block in the journal bitmap.  That will prevent it from being allocated for unformatted nodes
3757 ** before this transaction has finished.
3758 **
3759 ** mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers.  That will prevent any old transactions with
3760 ** this block from trying to flush to the real location.  Since we aren't removing the cnode from the journal_list_hash,
3761 ** the block can't be reallocated yet.
3762 **
3763 ** Then remove it from the current transaction, decrementing any counters and filing it on the clean list.
3764 */
3765 int journal_mark_freed(struct reiserfs_transaction_handle *th,
3766                        struct super_block *sb, b_blocknr_t blocknr)
3767 {
3768         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3769         struct reiserfs_journal_cnode *cn = NULL;
3770         struct buffer_head *bh = NULL;
3771         struct reiserfs_list_bitmap *jb = NULL;
3772         int cleaned = 0;
3773         BUG_ON(!th->t_trans_id);
3774
3775         cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr);
3776         if (cn && cn->bh) {
3777                 bh = cn->bh;
3778                 get_bh(bh);
3779         }
3780         /* if it is journal new, we just remove it from this transaction */
3781         if (bh && buffer_journal_new(bh)) {
3782                 clear_buffer_journal_new(bh);
3783                 clear_prepared_bits(bh);
3784                 reiserfs_clean_and_file_buffer(bh);
3785                 cleaned = remove_from_transaction(sb, blocknr, cleaned);
3786         } else {
3787                 /* set the bit for this block in the journal bitmap for this transaction */
3788                 jb = journal->j_current_jl->j_list_bitmap;
3789                 if (!jb) {
3790                         reiserfs_panic(sb, "journal-1702",
3791                                        "journal_list_bitmap is NULL");
3792                 }
3793                 set_bit_in_list_bitmap(sb, blocknr, jb);
3794
3795                 /* Note, the entire while loop is not allowed to schedule.  */
3796
3797                 if (bh) {
3798                         clear_prepared_bits(bh);
3799                         reiserfs_clean_and_file_buffer(bh);
3800                 }
3801                 cleaned = remove_from_transaction(sb, blocknr, cleaned);
3802
3803                 /* find all older transactions with this block, make sure they don't try to write it out */
3804                 cn = get_journal_hash_dev(sb, journal->j_list_hash_table,
3805                                           blocknr);
3806                 while (cn) {
3807                         if (sb == cn->sb && blocknr == cn->blocknr) {
3808                                 set_bit(BLOCK_FREED, &cn->state);
3809                                 if (cn->bh) {
3810                                         if (!cleaned) {
3811                                                 /* remove_from_transaction will brelse the buffer if it was
3812                                                  ** in the current trans
3813                                                  */
3814                                                 clear_buffer_journal_dirty(cn->
3815                                                                            bh);
3816                                                 clear_buffer_dirty(cn->bh);
3817                                                 clear_buffer_journal_test(cn->
3818                                                                           bh);
3819                                                 cleaned = 1;
3820                                                 put_bh(cn->bh);
3821                                                 if (atomic_read
3822                                                     (&(cn->bh->b_count)) < 0) {
3823                                                         reiserfs_warning(sb,
3824                                                                  "journal-2138",
3825                                                                  "cn->bh->b_count < 0");
3826                                                 }
3827                                         }
3828                                         if (cn->jlist) {        /* since we are clearing the bh, we MUST dec nonzerolen */
3829                                                 atomic_dec(&
3830                                                            (cn->jlist->
3831                                                             j_nonzerolen));
3832                                         }
3833                                         cn->bh = NULL;
3834                                 }
3835                         }
3836                         cn = cn->hnext;
3837                 }
3838         }
3839
3840         if (bh)
3841                 release_buffer_page(bh); /* get_hash grabs the buffer */
3842         return 0;
3843 }
3844
3845 void reiserfs_update_inode_transaction(struct inode *inode)
3846 {
3847         struct reiserfs_journal *journal = SB_JOURNAL(inode->i_sb);
3848         REISERFS_I(inode)->i_jl = journal->j_current_jl;
3849         REISERFS_I(inode)->i_trans_id = journal->j_trans_id;
3850 }
3851
3852 /*
3853  * returns -1 on error, 0 if no commits/barriers were done and 1
3854  * if a transaction was actually committed and the barrier was done
3855  */
3856 static int __commit_trans_jl(struct inode *inode, unsigned long id,
3857                              struct reiserfs_journal_list *jl)
3858 {
3859         struct reiserfs_transaction_handle th;
3860         struct super_block *sb = inode->i_sb;
3861         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3862         int ret = 0;
3863
3864         /* is it from the current transaction, or from an unknown transaction? */
3865         if (id == journal->j_trans_id) {
3866                 jl = journal->j_current_jl;
3867                 /* try to let other writers come in and grow this transaction */
3868                 let_transaction_grow(sb, id);
3869                 if (journal->j_trans_id != id) {
3870                         goto flush_commit_only;
3871                 }
3872
3873                 ret = journal_begin(&th, sb, 1);
3874                 if (ret)
3875                         return ret;
3876
3877                 /* someone might have ended this transaction while we joined */
3878                 if (journal->j_trans_id != id) {
3879                         reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3880                                                      1);
3881                         journal_mark_dirty(&th, sb, SB_BUFFER_WITH_SB(sb));
3882                         ret = journal_end(&th, sb, 1);
3883                         goto flush_commit_only;
3884                 }
3885
3886                 ret = journal_end_sync(&th, sb, 1);
3887                 if (!ret)
3888                         ret = 1;
3889
3890         } else {
3891                 /* this gets tricky, we have to make sure the journal list in
3892                  * the inode still exists.  We know the list is still around
3893                  * if we've got a larger transaction id than the oldest list
3894                  */
3895               flush_commit_only:
3896                 if (journal_list_still_alive(inode->i_sb, id)) {
3897                         /*
3898                          * we only set ret to 1 when we know for sure
3899                          * the barrier hasn't been started yet on the commit
3900                          * block.
3901                          */
3902                         if (atomic_read(&jl->j_commit_left) > 1)
3903                                 ret = 1;
3904                         flush_commit_list(sb, jl, 1);
3905                         if (journal->j_errno)
3906                                 ret = journal->j_errno;
3907                 }
3908         }
3909         /* otherwise the list is gone, and long since committed */
3910         return ret;
3911 }
3912
3913 int reiserfs_commit_for_inode(struct inode *inode)
3914 {
3915         unsigned int id = REISERFS_I(inode)->i_trans_id;
3916         struct reiserfs_journal_list *jl = REISERFS_I(inode)->i_jl;
3917
3918         /* for the whole inode, assume unset id means it was
3919          * changed in the current transaction.  More conservative
3920          */
3921         if (!id || !jl) {
3922                 reiserfs_update_inode_transaction(inode);
3923                 id = REISERFS_I(inode)->i_trans_id;
3924                 /* jl will be updated in __commit_trans_jl */
3925         }
3926
3927         return __commit_trans_jl(inode, id, jl);
3928 }
3929
3930 void reiserfs_restore_prepared_buffer(struct super_block *sb,
3931                                       struct buffer_head *bh)
3932 {
3933         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3934         PROC_INFO_INC(sb, journal.restore_prepared);
3935         if (!bh) {
3936                 return;
3937         }
3938         if (test_clear_buffer_journal_restore_dirty(bh) &&
3939             buffer_journal_dirty(bh)) {
3940                 struct reiserfs_journal_cnode *cn;
3941                 cn = get_journal_hash_dev(sb,
3942                                           journal->j_list_hash_table,
3943                                           bh->b_blocknr);
3944                 if (cn && can_dirty(cn)) {
3945                         set_buffer_journal_test(bh);
3946                         mark_buffer_dirty(bh);
3947                 }
3948         }
3949         clear_buffer_journal_prepared(bh);
3950 }
3951
3952 extern struct tree_balance *cur_tb;
3953 /*
3954 ** before we can change a metadata block, we have to make sure it won't
3955 ** be written to disk while we are altering it.  So, we must:
3956 ** clean it
3957 ** wait on it.
3958 **
3959 */
3960 int reiserfs_prepare_for_journal(struct super_block *sb,
3961                                  struct buffer_head *bh, int wait)
3962 {
3963         PROC_INFO_INC(sb, journal.prepare);
3964
3965         if (!trylock_buffer(bh)) {
3966                 if (!wait)
3967                         return 0;
3968                 lock_buffer(bh);
3969         }
3970         set_buffer_journal_prepared(bh);
3971         if (test_clear_buffer_dirty(bh) && buffer_journal_dirty(bh)) {
3972                 clear_buffer_journal_test(bh);
3973                 set_buffer_journal_restore_dirty(bh);
3974         }
3975         unlock_buffer(bh);
3976         return 1;
3977 }
3978
3979 static void flush_old_journal_lists(struct super_block *s)
3980 {
3981         struct reiserfs_journal *journal = SB_JOURNAL(s);
3982         struct reiserfs_journal_list *jl;
3983         struct list_head *entry;
3984         time_t now = get_seconds();
3985
3986         while (!list_empty(&journal->j_journal_list)) {
3987                 entry = journal->j_journal_list.next;
3988                 jl = JOURNAL_LIST_ENTRY(entry);
3989                 /* this check should always be run, to send old lists to disk */
3990                 if (jl->j_timestamp < (now - (JOURNAL_MAX_TRANS_AGE * 4)) &&
3991                     atomic_read(&jl->j_commit_left) == 0 &&
3992                     test_transaction(s, jl)) {
3993                         flush_used_journal_lists(s, jl);
3994                 } else {
3995                         break;
3996                 }
3997         }
3998 }
3999
4000 /*
4001 ** long and ugly.  If flush, will not return until all commit
4002 ** blocks and all real buffers in the trans are on disk.
4003 ** If no_async, won't return until all commit blocks are on disk.
4004 **
4005 ** keep reading, there are comments as you go along
4006 **
4007 ** If the journal is aborted, we just clean up. Things like flushing
4008 ** journal lists, etc just won't happen.
4009 */
4010 static int do_journal_end(struct reiserfs_transaction_handle *th,
4011                           struct super_block *sb, unsigned long nblocks,
4012                           int flags)
4013 {
4014         struct reiserfs_journal *journal = SB_JOURNAL(sb);
4015         struct reiserfs_journal_cnode *cn, *next, *jl_cn;
4016         struct reiserfs_journal_cnode *last_cn = NULL;
4017         struct reiserfs_journal_desc *desc;
4018         struct reiserfs_journal_commit *commit;
4019         struct buffer_head *c_bh;       /* commit bh */
4020         struct buffer_head *d_bh;       /* desc bh */
4021         int cur_write_start = 0;        /* start index of current log write */
4022         int old_start;
4023         int i;
4024         int flush;
4025         int wait_on_commit;
4026         struct reiserfs_journal_list *jl, *temp_jl;
4027         struct list_head *entry, *safe;
4028         unsigned long jindex;
4029         unsigned int commit_trans_id;
4030         int trans_half;
4031
4032         BUG_ON(th->t_refcount > 1);
4033         BUG_ON(!th->t_trans_id);
4034
4035         /* protect flush_older_commits from doing mistakes if the
4036            transaction ID counter gets overflowed.  */
4037         if (th->t_trans_id == ~0U)
4038                 flags |= FLUSH_ALL | COMMIT_NOW | WAIT;
4039         flush = flags & FLUSH_ALL;
4040         wait_on_commit = flags & WAIT;
4041
4042         put_fs_excl();
4043         current->journal_info = th->t_handle_save;
4044         reiserfs_check_lock_depth(sb, "journal end");
4045         if (journal->j_len == 0) {
4046                 reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
4047                                              1);
4048                 journal_mark_dirty(th, sb, SB_BUFFER_WITH_SB(sb));
4049         }
4050
4051         lock_journal(sb);
4052         if (journal->j_next_full_flush) {
4053                 flags |= FLUSH_ALL;
4054                 flush = 1;
4055         }
4056         if (journal->j_next_async_flush) {
4057                 flags |= COMMIT_NOW | WAIT;
4058                 wait_on_commit = 1;
4059         }
4060
4061         /* check_journal_end locks the journal, and unlocks if it does not return 1
4062          ** it tells us if we should continue with the journal_end, or just return
4063          */
4064         if (!check_journal_end(th, sb, nblocks, flags)) {
4065                 sb->s_dirt = 1;
4066                 wake_queued_writers(sb);
4067                 reiserfs_async_progress_wait(sb);
4068                 goto out;
4069         }
4070
4071         /* check_journal_end might set these, check again */
4072         if (journal->j_next_full_flush) {
4073                 flush = 1;
4074         }
4075
4076         /*
4077          ** j must wait means we have to flush the log blocks, and the real blocks for
4078          ** this transaction
4079          */
4080         if (journal->j_must_wait > 0) {
4081                 flush = 1;
4082         }
4083 #ifdef REISERFS_PREALLOCATE
4084         /* quota ops might need to nest, setup the journal_info pointer for them
4085          * and raise the refcount so that it is > 0. */
4086         current->journal_info = th;
4087         th->t_refcount++;
4088         reiserfs_discard_all_prealloc(th);      /* it should not involve new blocks into
4089                                                  * the transaction */
4090         th->t_refcount--;
4091         current->journal_info = th->t_handle_save;
4092 #endif
4093
4094         /* setup description block */
4095         d_bh =
4096             journal_getblk(sb,
4097                            SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4098                            journal->j_start);
4099         set_buffer_uptodate(d_bh);
4100         desc = (struct reiserfs_journal_desc *)(d_bh)->b_data;
4101         memset(d_bh->b_data, 0, d_bh->b_size);
4102         memcpy(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8);
4103         set_desc_trans_id(desc, journal->j_trans_id);
4104
4105         /* setup commit block.  Don't write (keep it clean too) this one until after everyone else is written */
4106         c_bh = journal_getblk(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4107                               ((journal->j_start + journal->j_len +
4108                                 1) % SB_ONDISK_JOURNAL_SIZE(sb)));
4109         commit = (struct reiserfs_journal_commit *)c_bh->b_data;
4110         memset(c_bh->b_data, 0, c_bh->b_size);
4111         set_commit_trans_id(commit, journal->j_trans_id);
4112         set_buffer_uptodate(c_bh);
4113
4114         /* init this journal list */
4115         jl = journal->j_current_jl;
4116
4117         /* we lock the commit before doing anything because
4118          * we want to make sure nobody tries to run flush_commit_list until
4119          * the new transaction is fully setup, and we've already flushed the
4120          * ordered bh list
4121          */
4122         reiserfs_mutex_lock_safe(&jl->j_commit_mutex, sb);
4123
4124         /* save the transaction id in case we need to commit it later */
4125         commit_trans_id = jl->j_trans_id;
4126
4127         atomic_set(&jl->j_older_commits_done, 0);
4128         jl->j_trans_id = journal->j_trans_id;
4129         jl->j_timestamp = journal->j_trans_start_time;
4130         jl->j_commit_bh = c_bh;
4131         jl->j_start = journal->j_start;
4132         jl->j_len = journal->j_len;
4133         atomic_set(&jl->j_nonzerolen, journal->j_len);
4134         atomic_set(&jl->j_commit_left, journal->j_len + 2);
4135         jl->j_realblock = NULL;
4136
4137         /* The ENTIRE FOR LOOP MUST not cause schedule to occur.
4138          **  for each real block, add it to the journal list hash,
4139          ** copy into real block index array in the commit or desc block
4140          */
4141         trans_half = journal_trans_half(sb->s_blocksize);
4142         for (i = 0, cn = journal->j_first; cn; cn = cn->next, i++) {
4143                 if (buffer_journaled(cn->bh)) {
4144                         jl_cn = get_cnode(sb);
4145                         if (!jl_cn) {
4146                                 reiserfs_panic(sb, "journal-1676",
4147                                                "get_cnode returned NULL");
4148                         }
4149                         if (i == 0) {
4150                                 jl->j_realblock = jl_cn;
4151                         }
4152                         jl_cn->prev = last_cn;
4153                         jl_cn->next = NULL;
4154                         if (last_cn) {
4155                                 last_cn->next = jl_cn;
4156                         }
4157                         last_cn = jl_cn;
4158                         /* make sure the block we are trying to log is not a block
4159                            of journal or reserved area */
4160
4161                         if (is_block_in_log_or_reserved_area
4162                             (sb, cn->bh->b_blocknr)) {
4163                                 reiserfs_panic(sb, "journal-2332",
4164                                                "Trying to log block %lu, "
4165                                                "which is a log block",
4166                                                cn->bh->b_blocknr);
4167                         }
4168                         jl_cn->blocknr = cn->bh->b_blocknr;
4169                         jl_cn->state = 0;
4170                         jl_cn->sb = sb;
4171                         jl_cn->bh = cn->bh;
4172                         jl_cn->jlist = jl;
4173                         insert_journal_hash(journal->j_list_hash_table, jl_cn);
4174                         if (i < trans_half) {
4175                                 desc->j_realblock[i] =
4176                                     cpu_to_le32(cn->bh->b_blocknr);
4177                         } else {
4178                                 commit->j_realblock[i - trans_half] =
4179                                     cpu_to_le32(cn->bh->b_blocknr);
4180                         }
4181                 } else {
4182                         i--;
4183                 }
4184         }
4185         set_desc_trans_len(desc, journal->j_len);
4186         set_desc_mount_id(desc, journal->j_mount_id);
4187         set_desc_trans_id(desc, journal->j_trans_id);
4188         set_commit_trans_len(commit, journal->j_len);
4189
4190         /* special check in case all buffers in the journal were marked for not logging */
4191         BUG_ON(journal->j_len == 0);
4192
4193         /* we're about to dirty all the log blocks, mark the description block
4194          * dirty now too.  Don't mark the commit block dirty until all the
4195          * others are on disk
4196          */
4197         mark_buffer_dirty(d_bh);
4198
4199         /* first data block is j_start + 1, so add one to cur_write_start wherever you use it */
4200         cur_write_start = journal->j_start;
4201         cn = journal->j_first;
4202         jindex = 1;             /* start at one so we don't get the desc again */
4203         while (cn) {
4204                 clear_buffer_journal_new(cn->bh);
4205                 /* copy all the real blocks into log area.  dirty log blocks */
4206                 if (buffer_journaled(cn->bh)) {
4207                         struct buffer_head *tmp_bh;
4208                         char *addr;
4209                         struct page *page;
4210                         tmp_bh =
4211                             journal_getblk(sb,
4212                                            SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4213                                            ((cur_write_start +
4214                                              jindex) %
4215                                             SB_ONDISK_JOURNAL_SIZE(sb)));
4216                         set_buffer_uptodate(tmp_bh);
4217                         page = cn->bh->b_page;
4218                         addr = kmap(page);
4219                         memcpy(tmp_bh->b_data,
4220                                addr + offset_in_page(cn->bh->b_data),
4221                                cn->bh->b_size);
4222                         kunmap(page);
4223                         mark_buffer_dirty(tmp_bh);
4224                         jindex++;
4225                         set_buffer_journal_dirty(cn->bh);
4226                         clear_buffer_journaled(cn->bh);
4227                 } else {
4228                         /* JDirty cleared sometime during transaction.  don't log this one */
4229                         reiserfs_warning(sb, "journal-2048",
4230                                          "BAD, buffer in journal hash, "
4231                                          "but not JDirty!");
4232                         brelse(cn->bh);
4233                 }
4234                 next = cn->next;
4235                 free_cnode(sb, cn);
4236                 cn = next;
4237                 reiserfs_write_unlock(sb);
4238                 cond_resched();
4239                 reiserfs_write_lock(sb);
4240         }
4241
4242         /* we are done  with both the c_bh and d_bh, but
4243          ** c_bh must be written after all other commit blocks,
4244          ** so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1.
4245          */
4246
4247         journal->j_current_jl = alloc_journal_list(sb);
4248
4249         /* now it is safe to insert this transaction on the main list */
4250         list_add_tail(&jl->j_list, &journal->j_journal_list);
4251         list_add_tail(&jl->j_working_list, &journal->j_working_list);
4252         journal->j_num_work_lists++;
4253
4254         /* reset journal values for the next transaction */
4255         old_start = journal->j_start;
4256         journal->j_start =
4257             (journal->j_start + journal->j_len +
4258              2) % SB_ONDISK_JOURNAL_SIZE(sb);
4259         atomic_set(&(journal->j_wcount), 0);
4260         journal->j_bcount = 0;
4261         journal->j_last = NULL;
4262         journal->j_first = NULL;
4263         journal->j_len = 0;
4264         journal->j_trans_start_time = 0;
4265         /* check for trans_id overflow */
4266         if (++journal->j_trans_id == 0)
4267                 journal->j_trans_id = 10;
4268         journal->j_current_jl->j_trans_id = journal->j_trans_id;
4269         journal->j_must_wait = 0;
4270         journal->j_len_alloc = 0;
4271         journal->j_next_full_flush = 0;
4272         journal->j_next_async_flush = 0;
4273         init_journal_hash(sb);
4274
4275         // make sure reiserfs_add_jh sees the new current_jl before we
4276         // write out the tails
4277         smp_mb();
4278
4279         /* tail conversion targets have to hit the disk before we end the
4280          * transaction.  Otherwise a later transaction might repack the tail
4281          * before this transaction commits, leaving the data block unflushed and
4282          * clean, if we crash before the later transaction commits, the data block
4283          * is lost.
4284          */
4285         if (!list_empty(&jl->j_tail_bh_list)) {
4286                 reiserfs_write_unlock(sb);
4287                 write_ordered_buffers(&journal->j_dirty_buffers_lock,
4288                                       journal, jl, &jl->j_tail_bh_list);
4289                 reiserfs_write_lock(sb);
4290         }
4291         BUG_ON(!list_empty(&jl->j_tail_bh_list));
4292         mutex_unlock(&jl->j_commit_mutex);
4293
4294         /* honor the flush wishes from the caller, simple commits can
4295          ** be done outside the journal lock, they are done below
4296          **
4297          ** if we don't flush the commit list right now, we put it into
4298          ** the work queue so the people waiting on the async progress work
4299          ** queue don't wait for this proc to flush journal lists and such.
4300          */
4301         if (flush) {
4302                 flush_commit_list(sb, jl, 1);
4303                 flush_journal_list(sb, jl, 1);
4304         } else if (!(jl->j_state & LIST_COMMIT_PENDING))
4305                 queue_delayed_work(commit_wq, &journal->j_work, HZ / 10);
4306
4307         /* if the next transaction has any chance of wrapping, flush
4308          ** transactions that might get overwritten.  If any journal lists are very
4309          ** old flush them as well.
4310          */
4311       first_jl:
4312         list_for_each_safe(entry, safe, &journal->j_journal_list) {
4313                 temp_jl = JOURNAL_LIST_ENTRY(entry);
4314                 if (journal->j_start <= temp_jl->j_start) {
4315                         if ((journal->j_start + journal->j_trans_max + 1) >=
4316                             temp_jl->j_start) {
4317                                 flush_used_journal_lists(sb, temp_jl);
4318                                 goto first_jl;
4319                         } else if ((journal->j_start +
4320                                     journal->j_trans_max + 1) <
4321                                    SB_ONDISK_JOURNAL_SIZE(sb)) {
4322                                 /* if we don't cross into the next transaction and we don't
4323                                  * wrap, there is no way we can overlap any later transactions
4324                                  * break now
4325                                  */
4326                                 break;
4327                         }
4328                 } else if ((journal->j_start +
4329                             journal->j_trans_max + 1) >
4330                            SB_ONDISK_JOURNAL_SIZE(sb)) {
4331                         if (((journal->j_start + journal->j_trans_max + 1) %
4332                              SB_ONDISK_JOURNAL_SIZE(sb)) >=
4333                             temp_jl->j_start) {
4334                                 flush_used_journal_lists(sb, temp_jl);
4335                                 goto first_jl;
4336                         } else {
4337                                 /* we don't overlap anything from out start to the end of the
4338                                  * log, and our wrapped portion doesn't overlap anything at
4339                                  * the start of the log.  We can break
4340                                  */
4341                                 break;
4342                         }
4343                 }
4344         }
4345         flush_old_journal_lists(sb);
4346
4347         journal->j_current_jl->j_list_bitmap =
4348             get_list_bitmap(sb, journal->j_current_jl);
4349
4350         if (!(journal->j_current_jl->j_list_bitmap)) {
4351                 reiserfs_panic(sb, "journal-1996",
4352                                "could not get a list bitmap");
4353         }
4354
4355         atomic_set(&(journal->j_jlock), 0);
4356         unlock_journal(sb);
4357         /* wake up any body waiting to join. */
4358         clear_bit(J_WRITERS_QUEUED, &journal->j_state);
4359         wake_up(&(journal->j_join_wait));
4360
4361         if (!flush && wait_on_commit &&
4362             journal_list_still_alive(sb, commit_trans_id)) {
4363                 flush_commit_list(sb, jl, 1);
4364         }
4365       out:
4366         reiserfs_check_lock_depth(sb, "journal end2");
4367
4368         memset(th, 0, sizeof(*th));
4369         /* Re-set th->t_super, so we can properly keep track of how many
4370          * persistent transactions there are. We need to do this so if this
4371          * call is part of a failed restart_transaction, we can free it later */
4372         th->t_super = sb;
4373
4374         return journal->j_errno;
4375 }
4376
4377 /* Send the file system read only and refuse new transactions */
4378 void reiserfs_abort_journal(struct super_block *sb, int errno)
4379 {
4380         struct reiserfs_journal *journal = SB_JOURNAL(sb);
4381         if (test_bit(J_ABORTED, &journal->j_state))
4382                 return;
4383
4384         if (!journal->j_errno)
4385                 journal->j_errno = errno;
4386
4387         sb->s_flags |= MS_RDONLY;
4388         set_bit(J_ABORTED, &journal->j_state);
4389
4390 #ifdef CONFIG_REISERFS_CHECK
4391         dump_stack();
4392 #endif
4393 }
4394