fs/reiserfs/journal.c

   1 /*
   2 ** Write ahead logging implementation copyright Chris Mason 2000
   3 **
   4 ** The background commits make this code very interelated, and
   5 ** overly complex.  I need to rethink things a bit....The major players:
   6 **
   7 ** journal_begin -- call with the number of blocks you expect to log.
   8 **                  If the current transaction is too
   9 **                  old, it will block until the current transaction is
  10 **                  finished, and then start a new one.
  11 **                  Usually, your transaction will get joined in with
  12 **                  previous ones for speed.
  13 **
  14 ** journal_join  -- same as journal_begin, but won't block on the current
  15 **                  transaction regardless of age.  Don't ever call
  16 **                  this.  Ever.  There are only two places it should be
  17 **                  called from, and they are both inside this file.
  18 **
  19 ** journal_mark_dirty -- adds blocks into this transaction.  clears any flags
  20 **                       that might make them get sent to disk
  21 **                       and then marks them BH_JDirty.  Puts the buffer head
  22 **                       into the current transaction hash.
  23 **
  24 ** journal_end -- if the current transaction is batchable, it does nothing
  25 **                   otherwise, it could do an async/synchronous commit, or
  26 **                   a full flush of all log and real blocks in the
  27 **                   transaction.
  28 **
  29 ** flush_old_commits -- if the current transaction is too old, it is ended and
  30 **                      commit blocks are sent to disk.  Forces commit blocks
  31 **                      to disk for all backgrounded commits that have been
  32 **                      around too long.
  33 **                   -- Note, if you call this as an immediate flush from
  34 **                      from within kupdate, it will ignore the immediate flag
  35 */
  36
  37 #include <asm/uaccess.h>
  38 #include <asm/system.h>
  39
  40 #include <linux/time.h>
  41 #include <asm/semaphore.h>
  42
  43 #include <linux/vmalloc.h>
  44 #include <linux/reiserfs_fs.h>
  45
  46 #include <linux/kernel.h>
  47 #include <linux/errno.h>
  48 #include <linux/fcntl.h>
  49 #include <linux/stat.h>
  50 #include <linux/string.h>
  51 #include <linux/smp_lock.h>
  52 #include <linux/buffer_head.h>
  53 #include <linux/workqueue.h>
  54 #include <linux/writeback.h>
  55 #include <linux/blkdev.h>
  56 #include <linux/backing-dev.h>
  57
  58 /* gets a struct reiserfs_journal_list * from a list head */
  59 #define JOURNAL_LIST_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
  60                                j_list))
  61 #define JOURNAL_WORK_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
  62                                j_working_list))
  63
  64 /* the number of mounted filesystems.  This is used to decide when to
  65 ** start and kill the commit workqueue
  66 */
  67 static int reiserfs_mounted_fs_count;
  68
  69 static struct workqueue_struct *commit_wq;
  70
  71 #define JOURNAL_TRANS_HALF 1018 /* must be correct to keep the desc and commit
  72                                    structs at 4k */
  73 #define BUFNR 64                /*read ahead */
  74
  75 /* cnode stat bits.  Move these into reiserfs_fs.h */
  76
  77 #define BLOCK_FREED 2           /* this block was freed, and can't be written.  */
  78 #define BLOCK_FREED_HOLDER 3    /* this block was freed during this transaction, and can't be written */
  79
  80 #define BLOCK_NEEDS_FLUSH 4     /* used in flush_journal_list */
  81 #define BLOCK_DIRTIED 5
  82
  83 /* journal list state bits */
  84 #define LIST_TOUCHED 1
  85 #define LIST_DIRTY   2
  86 #define LIST_COMMIT_PENDING  4  /* someone will commit this list */
  87
  88 /* flags for do_journal_end */
  89 #define FLUSH_ALL   1           /* flush commit and real blocks */
  90 #define COMMIT_NOW  2           /* end and commit this transaction */
  91 #define WAIT        4           /* wait for the log blocks to hit the disk */
  92
  93 static int do_journal_end(struct reiserfs_transaction_handle *,
  94                           struct super_block *, unsigned long nblocks,
  95                           int flags);
  96 static int flush_journal_list(struct super_block *s,
  97                               struct reiserfs_journal_list *jl, int flushall);
  98 static int flush_commit_list(struct super_block *s,
  99                              struct reiserfs_journal_list *jl, int flushall);
 100 static int can_dirty(struct reiserfs_journal_cnode *cn);
 101 static int journal_join(struct reiserfs_transaction_handle *th,
 102                         struct super_block *p_s_sb, unsigned long nblocks);
 103 static int release_journal_dev(struct super_block *super,
 104                                struct reiserfs_journal *journal);
 105 static int dirty_one_transaction(struct super_block *s,
 106                                  struct reiserfs_journal_list *jl);
 107 static void flush_async_commits(struct work_struct *work);
 108 static void queue_log_writer(struct super_block *s);
 109
 110 /* values for join in do_journal_begin_r */
 111 enum {
 112         JBEGIN_REG = 0,         /* regular journal begin */
 113         JBEGIN_JOIN = 1,        /* join the running transaction if at all possible */
 114         JBEGIN_ABORT = 2,       /* called from cleanup code, ignores aborted flag */
 115 };
 116
 117 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
 118                               struct super_block *p_s_sb,
 119                               unsigned long nblocks, int join);
 120
 121 static void init_journal_hash(struct super_block *p_s_sb)
 122 {
 123         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 124         memset(journal->j_hash_table, 0,
 125                JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
 126 }
 127
 128 /*
 129 ** clears BH_Dirty and sticks the buffer on the clean list.  Called because I can't allow refile_buffer to
 130 ** make schedule happen after I've freed a block.  Look at remove_from_transaction and journal_mark_freed for
 131 ** more details.
 132 */
 133 static int reiserfs_clean_and_file_buffer(struct buffer_head *bh)
 134 {
 135         if (bh) {
 136                 clear_buffer_dirty(bh);
 137                 clear_buffer_journal_test(bh);
 138         }
 139         return 0;
 140 }
 141
 142 static void disable_barrier(struct super_block *s)
 143 {
 144         REISERFS_SB(s)->s_mount_opt &= ~(1 << REISERFS_BARRIER_FLUSH);
 145         printk("reiserfs: disabling flush barriers on %s\n",
 146                reiserfs_bdevname(s));
 147 }
 148
 149 static struct reiserfs_bitmap_node *allocate_bitmap_node(struct super_block
 150                                                          *p_s_sb)
 151 {
 152         struct reiserfs_bitmap_node *bn;
 153         static int id;
 154
 155         bn = kmalloc(sizeof(struct reiserfs_bitmap_node), GFP_NOFS);
 156         if (!bn) {
 157                 return NULL;
 158         }
 159         bn->data = kzalloc(p_s_sb->s_blocksize, GFP_NOFS);
 160         if (!bn->data) {
 161                 kfree(bn);
 162                 return NULL;
 163         }
 164         bn->id = id++;
 165         INIT_LIST_HEAD(&bn->list);
 166         return bn;
 167 }
 168
 169 static struct reiserfs_bitmap_node *get_bitmap_node(struct super_block *p_s_sb)
 170 {
 171         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 172         struct reiserfs_bitmap_node *bn = NULL;
 173         struct list_head *entry = journal->j_bitmap_nodes.next;
 174
 175         journal->j_used_bitmap_nodes++;
 176       repeat:
 177
 178         if (entry != &journal->j_bitmap_nodes) {
 179                 bn = list_entry(entry, struct reiserfs_bitmap_node, list);
 180                 list_del(entry);
 181                 memset(bn->data, 0, p_s_sb->s_blocksize);
 182                 journal->j_free_bitmap_nodes--;
 183                 return bn;
 184         }
 185         bn = allocate_bitmap_node(p_s_sb);
 186         if (!bn) {
 187                 yield();
 188                 goto repeat;
 189         }
 190         return bn;
 191 }
 192 static inline void free_bitmap_node(struct super_block *p_s_sb,
 193                                     struct reiserfs_bitmap_node *bn)
 194 {
 195         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 196         journal->j_used_bitmap_nodes--;
 197         if (journal->j_free_bitmap_nodes > REISERFS_MAX_BITMAP_NODES) {
 198                 kfree(bn->data);
 199                 kfree(bn);
 200         } else {
 201                 list_add(&bn->list, &journal->j_bitmap_nodes);
 202                 journal->j_free_bitmap_nodes++;
 203         }
 204 }
 205
 206 static void allocate_bitmap_nodes(struct super_block *p_s_sb)
 207 {
 208         int i;
 209         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 210         struct reiserfs_bitmap_node *bn = NULL;
 211         for (i = 0; i < REISERFS_MIN_BITMAP_NODES; i++) {
 212                 bn = allocate_bitmap_node(p_s_sb);
 213                 if (bn) {
 214                         list_add(&bn->list, &journal->j_bitmap_nodes);
 215                         journal->j_free_bitmap_nodes++;
 216                 } else {
 217                         break;  // this is ok, we'll try again when more are needed
 218                 }
 219         }
 220 }
 221
 222 static int set_bit_in_list_bitmap(struct super_block *p_s_sb, int block,
 223                                   struct reiserfs_list_bitmap *jb)
 224 {
 225         int bmap_nr = block / (p_s_sb->s_blocksize << 3);
 226         int bit_nr = block % (p_s_sb->s_blocksize << 3);
 227
 228         if (!jb->bitmaps[bmap_nr]) {
 229                 jb->bitmaps[bmap_nr] = get_bitmap_node(p_s_sb);
 230         }
 231         set_bit(bit_nr, (unsigned long *)jb->bitmaps[bmap_nr]->data);
 232         return 0;
 233 }
 234
 235 static void cleanup_bitmap_list(struct super_block *p_s_sb,
 236                                 struct reiserfs_list_bitmap *jb)
 237 {
 238         int i;
 239         if (jb->bitmaps == NULL)
 240                 return;
 241
 242         for (i = 0; i < SB_BMAP_NR(p_s_sb); i++) {
 243                 if (jb->bitmaps[i]) {
 244                         free_bitmap_node(p_s_sb, jb->bitmaps[i]);
 245                         jb->bitmaps[i] = NULL;
 246                 }
 247         }
 248 }
 249
 250 /*
 251 ** only call this on FS unmount.
 252 */
 253 static int free_list_bitmaps(struct super_block *p_s_sb,
 254                              struct reiserfs_list_bitmap *jb_array)
 255 {
 256         int i;
 257         struct reiserfs_list_bitmap *jb;
 258         for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
 259                 jb = jb_array + i;
 260                 jb->journal_list = NULL;
 261                 cleanup_bitmap_list(p_s_sb, jb);
 262                 vfree(jb->bitmaps);
 263                 jb->bitmaps = NULL;
 264         }
 265         return 0;
 266 }
 267
 268 static int free_bitmap_nodes(struct super_block *p_s_sb)
 269 {
 270         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 271         struct list_head *next = journal->j_bitmap_nodes.next;
 272         struct reiserfs_bitmap_node *bn;
 273
 274         while (next != &journal->j_bitmap_nodes) {
 275                 bn = list_entry(next, struct reiserfs_bitmap_node, list);
 276                 list_del(next);
 277                 kfree(bn->data);
 278                 kfree(bn);
 279                 next = journal->j_bitmap_nodes.next;
 280                 journal->j_free_bitmap_nodes--;
 281         }
 282
 283         return 0;
 284 }
 285
 286 /*
 287 ** get memory for JOURNAL_NUM_BITMAPS worth of bitmaps.
 288 ** jb_array is the array to be filled in.
 289 */
 290 int reiserfs_allocate_list_bitmaps(struct super_block *p_s_sb,
 291                                    struct reiserfs_list_bitmap *jb_array,
 292                                    int bmap_nr)
 293 {
 294         int i;
 295         int failed = 0;
 296         struct reiserfs_list_bitmap *jb;
 297         int mem = bmap_nr * sizeof(struct reiserfs_bitmap_node *);
 298
 299         for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
 300                 jb = jb_array + i;
 301                 jb->journal_list = NULL;
 302                 jb->bitmaps = vmalloc(mem);
 303                 if (!jb->bitmaps) {
 304                         reiserfs_warning(p_s_sb,
 305                                          "clm-2000, unable to allocate bitmaps for journal lists");
 306                         failed = 1;
 307                         break;
 308                 }
 309                 memset(jb->bitmaps, 0, mem);
 310         }
 311         if (failed) {
 312                 free_list_bitmaps(p_s_sb, jb_array);
 313                 return -1;
 314         }
 315         return 0;
 316 }
 317
 318 /*
 319 ** find an available list bitmap.  If you can't find one, flush a commit list
 320 ** and try again
 321 */
 322 static struct reiserfs_list_bitmap *get_list_bitmap(struct super_block *p_s_sb,
 323                                                     struct reiserfs_journal_list
 324                                                     *jl)
 325 {
 326         int i, j;
 327         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 328         struct reiserfs_list_bitmap *jb = NULL;
 329
 330         for (j = 0; j < (JOURNAL_NUM_BITMAPS * 3); j++) {
 331                 i = journal->j_list_bitmap_index;
 332                 journal->j_list_bitmap_index = (i + 1) % JOURNAL_NUM_BITMAPS;
 333                 jb = journal->j_list_bitmap + i;
 334                 if (journal->j_list_bitmap[i].journal_list) {
 335                         flush_commit_list(p_s_sb,
 336                                           journal->j_list_bitmap[i].
 337                                           journal_list, 1);
 338                         if (!journal->j_list_bitmap[i].journal_list) {
 339                                 break;
 340                         }
 341                 } else {
 342                         break;
 343                 }
 344         }
 345         if (jb->journal_list) { /* double check to make sure if flushed correctly */
 346                 return NULL;
 347         }
 348         jb->journal_list = jl;
 349         return jb;
 350 }
 351
 352 /*
 353 ** allocates a new chunk of X nodes, and links them all together as a list.
 354 ** Uses the cnode->next and cnode->prev pointers
 355 ** returns NULL on failure
 356 */
 357 static struct reiserfs_journal_cnode *allocate_cnodes(int num_cnodes)
 358 {
 359         struct reiserfs_journal_cnode *head;
 360         int i;
 361         if (num_cnodes <= 0) {
 362                 return NULL;
 363         }
 364         head = vmalloc(num_cnodes * sizeof(struct reiserfs_journal_cnode));
 365         if (!head) {
 366                 return NULL;
 367         }
 368         memset(head, 0, num_cnodes * sizeof(struct reiserfs_journal_cnode));
 369         head[0].prev = NULL;
 370         head[0].next = head + 1;
 371         for (i = 1; i < num_cnodes; i++) {
 372                 head[i].prev = head + (i - 1);
 373                 head[i].next = head + (i + 1);  /* if last one, overwrite it after the if */
 374         }
 375         head[num_cnodes - 1].next = NULL;
 376         return head;
 377 }
 378
 379 /*
 380 ** pulls a cnode off the free list, or returns NULL on failure
 381 */
 382 static struct reiserfs_journal_cnode *get_cnode(struct super_block *p_s_sb)
 383 {
 384         struct reiserfs_journal_cnode *cn;
 385         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 386
 387         reiserfs_check_lock_depth(p_s_sb, "get_cnode");
 388
 389         if (journal->j_cnode_free <= 0) {
 390                 return NULL;
 391         }
 392         journal->j_cnode_used++;
 393         journal->j_cnode_free--;
 394         cn = journal->j_cnode_free_list;
 395         if (!cn) {
 396                 return cn;
 397         }
 398         if (cn->next) {
 399                 cn->next->prev = NULL;
 400         }
 401         journal->j_cnode_free_list = cn->next;
 402         memset(cn, 0, sizeof(struct reiserfs_journal_cnode));
 403         return cn;
 404 }
 405
 406 /*
 407 ** returns a cnode to the free list
 408 */
 409 static void free_cnode(struct super_block *p_s_sb,
 410                        struct reiserfs_journal_cnode *cn)
 411 {
 412         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 413
 414         reiserfs_check_lock_depth(p_s_sb, "free_cnode");
 415
 416         journal->j_cnode_used--;
 417         journal->j_cnode_free++;
 418         /* memset(cn, 0, sizeof(struct reiserfs_journal_cnode)) ; */
 419         cn->next = journal->j_cnode_free_list;
 420         if (journal->j_cnode_free_list) {
 421                 journal->j_cnode_free_list->prev = cn;
 422         }
 423         cn->prev = NULL;        /* not needed with the memset, but I might kill the memset, and forget to do this */
 424         journal->j_cnode_free_list = cn;
 425 }
 426
 427 static void clear_prepared_bits(struct buffer_head *bh)
 428 {
 429         clear_buffer_journal_prepared(bh);
 430         clear_buffer_journal_restore_dirty(bh);
 431 }
 432
 433 /* utility function to force a BUG if it is called without the big
 434 ** kernel lock held.  caller is the string printed just before calling BUG()
 435 */
 436 void reiserfs_check_lock_depth(struct super_block *sb, char *caller)
 437 {
 438 #ifdef CONFIG_SMP
 439         if (current->lock_depth < 0) {
 440                 reiserfs_panic(sb, "%s called without kernel lock held",
 441                                caller);
 442         }
 443 #else
 444         ;
 445 #endif
 446 }
 447
 448 /* return a cnode with same dev, block number and size in table, or null if not found */
 449 static inline struct reiserfs_journal_cnode *get_journal_hash_dev(struct
 450                                                                   super_block
 451                                                                   *sb,
 452                                                                   struct
 453                                                                   reiserfs_journal_cnode
 454                                                                   **table,
 455                                                                   long bl)
 456 {
 457         struct reiserfs_journal_cnode *cn;
 458         cn = journal_hash(table, sb, bl);
 459         while (cn) {
 460                 if (cn->blocknr == bl && cn->sb == sb)
 461                         return cn;
 462                 cn = cn->hnext;
 463         }
 464         return (struct reiserfs_journal_cnode *)0;
 465 }
 466
 467 /*
 468 ** this actually means 'can this block be reallocated yet?'.  If you set search_all, a block can only be allocated
 469 ** if it is not in the current transaction, was not freed by the current transaction, and has no chance of ever
 470 ** being overwritten by a replay after crashing.
 471 **
 472 ** If you don't set search_all, a block can only be allocated if it is not in the current transaction.  Since deleting
 473 ** a block removes it from the current transaction, this case should never happen.  If you don't set search_all, make
 474 ** sure you never write the block without logging it.
 475 **
 476 ** next_zero_bit is a suggestion about the next block to try for find_forward.
 477 ** when bl is rejected because it is set in a journal list bitmap, we search
 478 ** for the next zero bit in the bitmap that rejected bl.  Then, we return that
 479 ** through next_zero_bit for find_forward to try.
 480 **
 481 ** Just because we return something in next_zero_bit does not mean we won't
 482 ** reject it on the next call to reiserfs_in_journal
 483 **
 484 */
 485 int reiserfs_in_journal(struct super_block *p_s_sb,
 486                         int bmap_nr, int bit_nr, int search_all,
 487                         b_blocknr_t * next_zero_bit)
 488 {
 489         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 490         struct reiserfs_journal_cnode *cn;
 491         struct reiserfs_list_bitmap *jb;
 492         int i;
 493         unsigned long bl;
 494
 495         *next_zero_bit = 0;     /* always start this at zero. */
 496
 497         PROC_INFO_INC(p_s_sb, journal.in_journal);
 498         /* If we aren't doing a search_all, this is a metablock, and it will be logged before use.
 499          ** if we crash before the transaction that freed it commits,  this transaction won't
 500          ** have committed either, and the block will never be written
 501          */
 502         if (search_all) {
 503                 for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
 504                         PROC_INFO_INC(p_s_sb, journal.in_journal_bitmap);
 505                         jb = journal->j_list_bitmap + i;
 506                         if (jb->journal_list && jb->bitmaps[bmap_nr] &&
 507                             test_bit(bit_nr,
 508                                      (unsigned long *)jb->bitmaps[bmap_nr]->
 509                                      data)) {
 510                                 *next_zero_bit =
 511                                     find_next_zero_bit((unsigned long *)
 512                                                        (jb->bitmaps[bmap_nr]->
 513                                                         data),
 514                                                        p_s_sb->s_blocksize << 3,
 515                                                        bit_nr + 1);
 516                                 return 1;
 517                         }
 518                 }
 519         }
 520
 521         bl = bmap_nr * (p_s_sb->s_blocksize << 3) + bit_nr;
 522         /* is it in any old transactions? */
 523         if (search_all
 524             && (cn =
 525                 get_journal_hash_dev(p_s_sb, journal->j_list_hash_table, bl))) {
 526                 return 1;
 527         }
 528
 529         /* is it in the current transaction.  This should never happen */
 530         if ((cn = get_journal_hash_dev(p_s_sb, journal->j_hash_table, bl))) {
 531                 BUG();
 532                 return 1;
 533         }
 534
 535         PROC_INFO_INC(p_s_sb, journal.in_journal_reusable);
 536         /* safe for reuse */
 537         return 0;
 538 }
 539
 540 /* insert cn into table
 541 */
 542 static inline void insert_journal_hash(struct reiserfs_journal_cnode **table,
 543                                        struct reiserfs_journal_cnode *cn)
 544 {
 545         struct reiserfs_journal_cnode *cn_orig;
 546
 547         cn_orig = journal_hash(table, cn->sb, cn->blocknr);
 548         cn->hnext = cn_orig;
 549         cn->hprev = NULL;
 550         if (cn_orig) {
 551                 cn_orig->hprev = cn;
 552         }
 553         journal_hash(table, cn->sb, cn->blocknr) = cn;
 554 }
 555
 556 /* lock the current transaction */
 557 static inline void lock_journal(struct super_block *p_s_sb)
 558 {
 559         PROC_INFO_INC(p_s_sb, journal.lock_journal);
 560         down(&SB_JOURNAL(p_s_sb)->j_lock);
 561 }
 562
 563 /* unlock the current transaction */
 564 static inline void unlock_journal(struct super_block *p_s_sb)
 565 {
 566         up(&SB_JOURNAL(p_s_sb)->j_lock);
 567 }
 568
 569 static inline void get_journal_list(struct reiserfs_journal_list *jl)
 570 {
 571         jl->j_refcount++;
 572 }
 573
 574 static inline void put_journal_list(struct super_block *s,
 575                                     struct reiserfs_journal_list *jl)
 576 {
 577         if (jl->j_refcount < 1) {
 578                 reiserfs_panic(s, "trans id %lu, refcount at %d",
 579                                jl->j_trans_id, jl->j_refcount);
 580         }
 581         if (--jl->j_refcount == 0)
 582                 kfree(jl);
 583 }
 584
 585 /*
 586 ** this used to be much more involved, and I'm keeping it just in case things get ugly again.
 587 ** it gets called by flush_commit_list, and cleans up any data stored about blocks freed during a
 588 ** transaction.
 589 */
 590 static void cleanup_freed_for_journal_list(struct super_block *p_s_sb,
 591                                            struct reiserfs_journal_list *jl)
 592 {
 593
 594         struct reiserfs_list_bitmap *jb = jl->j_list_bitmap;
 595         if (jb) {
 596                 cleanup_bitmap_list(p_s_sb, jb);
 597         }
 598         jl->j_list_bitmap->journal_list = NULL;
 599         jl->j_list_bitmap = NULL;
 600 }
 601
 602 static int journal_list_still_alive(struct super_block *s,
 603                                     unsigned long trans_id)
 604 {
 605         struct reiserfs_journal *journal = SB_JOURNAL(s);
 606         struct list_head *entry = &journal->j_journal_list;
 607         struct reiserfs_journal_list *jl;
 608
 609         if (!list_empty(entry)) {
 610                 jl = JOURNAL_LIST_ENTRY(entry->next);
 611                 if (jl->j_trans_id <= trans_id) {
 612                         return 1;
 613                 }
 614         }
 615         return 0;
 616 }
 617
 618 static void reiserfs_end_buffer_io_sync(struct buffer_head *bh, int uptodate)
 619 {
 620         char b[BDEVNAME_SIZE];
 621
 622         if (buffer_journaled(bh)) {
 623                 reiserfs_warning(NULL,
 624                                  "clm-2084: pinned buffer %lu:%s sent to disk",
 625                                  bh->b_blocknr, bdevname(bh->b_bdev, b));
 626         }
 627         if (uptodate)
 628                 set_buffer_uptodate(bh);
 629         else
 630                 clear_buffer_uptodate(bh);
 631         unlock_buffer(bh);
 632         put_bh(bh);
 633 }
 634
 635 static void reiserfs_end_ordered_io(struct buffer_head *bh, int uptodate)
 636 {
 637         if (uptodate)
 638                 set_buffer_uptodate(bh);
 639         else
 640                 clear_buffer_uptodate(bh);
 641         unlock_buffer(bh);
 642         put_bh(bh);
 643 }
 644
 645 static void submit_logged_buffer(struct buffer_head *bh)
 646 {
 647         get_bh(bh);
 648         bh->b_end_io = reiserfs_end_buffer_io_sync;
 649         clear_buffer_journal_new(bh);
 650         clear_buffer_dirty(bh);
 651         if (!test_clear_buffer_journal_test(bh))
 652                 BUG();
 653         if (!buffer_uptodate(bh))
 654                 BUG();
 655         submit_bh(WRITE, bh);
 656 }
 657
 658 static void submit_ordered_buffer(struct buffer_head *bh)
 659 {
 660         get_bh(bh);
 661         bh->b_end_io = reiserfs_end_ordered_io;
 662         clear_buffer_dirty(bh);
 663         if (!buffer_uptodate(bh))
 664                 BUG();
 665         submit_bh(WRITE, bh);
 666 }
 667
 668 static int submit_barrier_buffer(struct buffer_head *bh)
 669 {
 670         get_bh(bh);
 671         bh->b_end_io = reiserfs_end_ordered_io;
 672         clear_buffer_dirty(bh);
 673         if (!buffer_uptodate(bh))
 674                 BUG();
 675         return submit_bh(WRITE_BARRIER, bh);
 676 }
 677
 678 static void check_barrier_completion(struct super_block *s,
 679                                      struct buffer_head *bh)
 680 {
 681         if (buffer_eopnotsupp(bh)) {
 682                 clear_buffer_eopnotsupp(bh);
 683                 disable_barrier(s);
 684                 set_buffer_uptodate(bh);
 685                 set_buffer_dirty(bh);
 686                 sync_dirty_buffer(bh);
 687         }
 688 }
 689
 690 #define CHUNK_SIZE 32
 691 struct buffer_chunk {
 692         struct buffer_head *bh[CHUNK_SIZE];
 693         int nr;
 694 };
 695
 696 static void write_chunk(struct buffer_chunk *chunk)
 697 {
 698         int i;
 699         get_fs_excl();
 700         for (i = 0; i < chunk->nr; i++) {
 701                 submit_logged_buffer(chunk->bh[i]);
 702         }
 703         chunk->nr = 0;
 704         put_fs_excl();
 705 }
 706
 707 static void write_ordered_chunk(struct buffer_chunk *chunk)
 708 {
 709         int i;
 710         get_fs_excl();
 711         for (i = 0; i < chunk->nr; i++) {
 712                 submit_ordered_buffer(chunk->bh[i]);
 713         }
 714         chunk->nr = 0;
 715         put_fs_excl();
 716 }
 717
 718 static int add_to_chunk(struct buffer_chunk *chunk, struct buffer_head *bh,
 719                         spinlock_t * lock, void (fn) (struct buffer_chunk *))
 720 {
 721         int ret = 0;
 722         BUG_ON(chunk->nr >= CHUNK_SIZE);
 723         chunk->bh[chunk->nr++] = bh;
 724         if (chunk->nr >= CHUNK_SIZE) {
 725                 ret = 1;
 726                 if (lock)
 727                         spin_unlock(lock);
 728                 fn(chunk);
 729                 if (lock)
 730                         spin_lock(lock);
 731         }
 732         return ret;
 733 }
 734
 735 static atomic_t nr_reiserfs_jh = ATOMIC_INIT(0);
 736 static struct reiserfs_jh *alloc_jh(void)
 737 {
 738         struct reiserfs_jh *jh;
 739         while (1) {
 740                 jh = kmalloc(sizeof(*jh), GFP_NOFS);
 741                 if (jh) {
 742                         atomic_inc(&nr_reiserfs_jh);
 743                         return jh;
 744                 }
 745                 yield();
 746         }
 747 }
 748
 749 /*
 750  * we want to free the jh when the buffer has been written
 751  * and waited on
 752  */
 753 void reiserfs_free_jh(struct buffer_head *bh)
 754 {
 755         struct reiserfs_jh *jh;
 756
 757         jh = bh->b_private;
 758         if (jh) {
 759                 bh->b_private = NULL;
 760                 jh->bh = NULL;
 761                 list_del_init(&jh->list);
 762                 kfree(jh);
 763                 if (atomic_read(&nr_reiserfs_jh) <= 0)
 764                         BUG();
 765                 atomic_dec(&nr_reiserfs_jh);
 766                 put_bh(bh);
 767         }
 768 }
 769
 770 static inline int __add_jh(struct reiserfs_journal *j, struct buffer_head *bh,
 771                            int tail)
 772 {
 773         struct reiserfs_jh *jh;
 774
 775         if (bh->b_private) {
 776                 spin_lock(&j->j_dirty_buffers_lock);
 777                 if (!bh->b_private) {
 778                         spin_unlock(&j->j_dirty_buffers_lock);
 779                         goto no_jh;
 780                 }
 781                 jh = bh->b_private;
 782                 list_del_init(&jh->list);
 783         } else {
 784               no_jh:
 785                 get_bh(bh);
 786                 jh = alloc_jh();
 787                 spin_lock(&j->j_dirty_buffers_lock);
 788                 /* buffer must be locked for __add_jh, should be able to have
 789                  * two adds at the same time
 790                  */
 791                 BUG_ON(bh->b_private);
 792                 jh->bh = bh;
 793                 bh->b_private = jh;
 794         }
 795         jh->jl = j->j_current_jl;
 796         if (tail)
 797                 list_add_tail(&jh->list, &jh->jl->j_tail_bh_list);
 798         else {
 799                 list_add_tail(&jh->list, &jh->jl->j_bh_list);
 800         }
 801         spin_unlock(&j->j_dirty_buffers_lock);
 802         return 0;
 803 }
 804
 805 int reiserfs_add_tail_list(struct inode *inode, struct buffer_head *bh)
 806 {
 807         return __add_jh(SB_JOURNAL(inode->i_sb), bh, 1);
 808 }
 809 int reiserfs_add_ordered_list(struct inode *inode, struct buffer_head *bh)
 810 {
 811         return __add_jh(SB_JOURNAL(inode->i_sb), bh, 0);
 812 }
 813
 814 #define JH_ENTRY(l) list_entry((l), struct reiserfs_jh, list)
 815 static int write_ordered_buffers(spinlock_t * lock,
 816                                  struct reiserfs_journal *j,
 817                                  struct reiserfs_journal_list *jl,
 818                                  struct list_head *list)
 819 {
 820         struct buffer_head *bh;
 821         struct reiserfs_jh *jh;
 822         int ret = j->j_errno;
 823         struct buffer_chunk chunk;
 824         struct list_head tmp;
 825         INIT_LIST_HEAD(&tmp);
 826
 827         chunk.nr = 0;
 828         spin_lock(lock);
 829         while (!list_empty(list)) {
 830                 jh = JH_ENTRY(list->next);
 831                 bh = jh->bh;
 832                 get_bh(bh);
 833                 if (test_set_buffer_locked(bh)) {
 834                         if (!buffer_dirty(bh)) {
 835                                 list_move(&jh->list, &tmp);
 836                                 goto loop_next;
 837                         }
 838                         spin_unlock(lock);
 839                         if (chunk.nr)
 840                                 write_ordered_chunk(&chunk);
 841                         wait_on_buffer(bh);
 842                         cond_resched();
 843                         spin_lock(lock);
 844                         goto loop_next;
 845                 }
 846                 /* in theory, dirty non-uptodate buffers should never get here,
 847                  * but the upper layer io error paths still have a few quirks.
 848                  * Handle them here as gracefully as we can
 849                  */
 850                 if (!buffer_uptodate(bh) && buffer_dirty(bh)) {
 851                         clear_buffer_dirty(bh);
 852                         ret = -EIO;
 853                 }
 854                 if (buffer_dirty(bh)) {
 855                         list_move(&jh->list, &tmp);
 856                         add_to_chunk(&chunk, bh, lock, write_ordered_chunk);
 857                 } else {
 858                         reiserfs_free_jh(bh);
 859                         unlock_buffer(bh);
 860                 }
 861               loop_next:
 862                 put_bh(bh);
 863                 cond_resched_lock(lock);
 864         }
 865         if (chunk.nr) {
 866                 spin_unlock(lock);
 867                 write_ordered_chunk(&chunk);
 868                 spin_lock(lock);
 869         }
 870         while (!list_empty(&tmp)) {
 871                 jh = JH_ENTRY(tmp.prev);
 872                 bh = jh->bh;
 873                 get_bh(bh);
 874                 reiserfs_free_jh(bh);
 875
 876                 if (buffer_locked(bh)) {
 877                         spin_unlock(lock);
 878                         wait_on_buffer(bh);
 879                         spin_lock(lock);
 880                 }
 881                 if (!buffer_uptodate(bh)) {
 882                         ret = -EIO;
 883                 }
 884                 /* ugly interaction with invalidatepage here.
 885                  * reiserfs_invalidate_page will pin any buffer that has a valid
 886                  * journal head from an older transaction.  If someone else sets
 887                  * our buffer dirty after we write it in the first loop, and
 888                  * then someone truncates the page away, nobody will ever write
 889                  * the buffer. We're safe if we write the page one last time
 890                  * after freeing the journal header.
 891                  */
 892                 if (buffer_dirty(bh) && unlikely(bh->b_page->mapping == NULL)) {
 893                         spin_unlock(lock);
 894                         ll_rw_block(WRITE, 1, &bh);
 895                         spin_lock(lock);
 896                 }
 897                 put_bh(bh);
 898                 cond_resched_lock(lock);
 899         }
 900         spin_unlock(lock);
 901         return ret;
 902 }
 903
 904 static int flush_older_commits(struct super_block *s,
 905                                struct reiserfs_journal_list *jl)
 906 {
 907         struct reiserfs_journal *journal = SB_JOURNAL(s);
 908         struct reiserfs_journal_list *other_jl;
 909         struct reiserfs_journal_list *first_jl;
 910         struct list_head *entry;
 911         unsigned long trans_id = jl->j_trans_id;
 912         unsigned long other_trans_id;
 913         unsigned long first_trans_id;
 914
 915       find_first:
 916         /*
 917          * first we walk backwards to find the oldest uncommitted transation
 918          */
 919         first_jl = jl;
 920         entry = jl->j_list.prev;
 921         while (1) {
 922                 other_jl = JOURNAL_LIST_ENTRY(entry);
 923                 if (entry == &journal->j_journal_list ||
 924                     atomic_read(&other_jl->j_older_commits_done))
 925                         break;
 926
 927                 first_jl = other_jl;
 928                 entry = other_jl->j_list.prev;
 929         }
 930
 931         /* if we didn't find any older uncommitted transactions, return now */
 932         if (first_jl == jl) {
 933                 return 0;
 934         }
 935
 936         first_trans_id = first_jl->j_trans_id;
 937
 938         entry = &first_jl->j_list;
 939         while (1) {
 940                 other_jl = JOURNAL_LIST_ENTRY(entry);
 941                 other_trans_id = other_jl->j_trans_id;
 942
 943                 if (other_trans_id < trans_id) {
 944                         if (atomic_read(&other_jl->j_commit_left) != 0) {
 945                                 flush_commit_list(s, other_jl, 0);
 946
 947                                 /* list we were called with is gone, return */
 948                                 if (!journal_list_still_alive(s, trans_id))
 949                                         return 1;
 950
 951                                 /* the one we just flushed is gone, this means all
 952                                  * older lists are also gone, so first_jl is no longer
 953                                  * valid either.  Go back to the beginning.
 954                                  */
 955                                 if (!journal_list_still_alive
 956                                     (s, other_trans_id)) {
 957                                         goto find_first;
 958                                 }
 959                         }
 960                         entry = entry->next;
 961                         if (entry == &journal->j_journal_list)
 962                                 return 0;
 963                 } else {
 964                         return 0;
 965                 }
 966         }
 967         return 0;
 968 }
 969 int reiserfs_async_progress_wait(struct super_block *s)
 970 {
 971         DEFINE_WAIT(wait);
 972         struct reiserfs_journal *j = SB_JOURNAL(s);
 973         if (atomic_read(&j->j_async_throttle))
 974                 congestion_wait(WRITE, HZ / 10);
 975         return 0;
 976 }
 977
 978 /*
 979 ** if this journal list still has commit blocks unflushed, send them to disk.
 980 **
 981 ** log areas must be flushed in order (transaction 2 can't commit before transaction 1)
 982 ** Before the commit block can by written, every other log block must be safely on disk
 983 **
 984 */
 985 static int flush_commit_list(struct super_block *s,
 986                              struct reiserfs_journal_list *jl, int flushall)
 987 {
 988         int i;
 989         int bn;
 990         struct buffer_head *tbh = NULL;
 991         unsigned long trans_id = jl->j_trans_id;
 992         struct reiserfs_journal *journal = SB_JOURNAL(s);
 993         int barrier = 0;
 994         int retval = 0;
 995         int write_len;
 996
 997         reiserfs_check_lock_depth(s, "flush_commit_list");
 998
 999         if (atomic_read(&jl->j_older_commits_done)) {
1000                 return 0;
1001         }
1002
1003         get_fs_excl();
1004
1005         /* before we can put our commit blocks on disk, we have to make sure everyone older than
1006          ** us is on disk too
1007          */
1008         BUG_ON(jl->j_len <= 0);
1009         BUG_ON(trans_id == journal->j_trans_id);
1010
1011         get_journal_list(jl);
1012         if (flushall) {
1013                 if (flush_older_commits(s, jl) == 1) {
1014                         /* list disappeared during flush_older_commits.  return */
1015                         goto put_jl;
1016                 }
1017         }
1018
1019         /* make sure nobody is trying to flush this one at the same time */
1020         down(&jl->j_commit_lock);
1021         if (!journal_list_still_alive(s, trans_id)) {
1022                 up(&jl->j_commit_lock);
1023                 goto put_jl;
1024         }
1025         BUG_ON(jl->j_trans_id == 0);
1026
1027         /* this commit is done, exit */
1028         if (atomic_read(&(jl->j_commit_left)) <= 0) {
1029                 if (flushall) {
1030                         atomic_set(&(jl->j_older_commits_done), 1);
1031                 }
1032                 up(&jl->j_commit_lock);
1033                 goto put_jl;
1034         }
1035
1036         if (!list_empty(&jl->j_bh_list)) {
1037                 int ret;
1038                 unlock_kernel();
1039                 ret = write_ordered_buffers(&journal->j_dirty_buffers_lock,
1040                                             journal, jl, &jl->j_bh_list);
1041                 if (ret < 0 && retval == 0)
1042                         retval = ret;
1043                 lock_kernel();
1044         }
1045         BUG_ON(!list_empty(&jl->j_bh_list));
1046         /*
1047          * for the description block and all the log blocks, submit any buffers
1048          * that haven't already reached the disk.  Try to write at least 256
1049          * log blocks. later on, we will only wait on blocks that correspond
1050          * to this transaction, but while we're unplugging we might as well
1051          * get a chunk of data on there.
1052          */
1053         atomic_inc(&journal->j_async_throttle);
1054         write_len = jl->j_len + 1;
1055         if (write_len < 256)
1056                 write_len = 256;
1057         for (i = 0 ; i < write_len ; i++) {
1058                 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + (jl->j_start + i) %
1059                     SB_ONDISK_JOURNAL_SIZE(s);
1060                 tbh = journal_find_get_block(s, bn);
1061                 if (tbh) {
1062                         if (buffer_dirty(tbh))
1063                             ll_rw_block(WRITE, 1, &tbh) ;
1064                         put_bh(tbh) ;
1065                 }
1066         }
1067         atomic_dec(&journal->j_async_throttle);
1068
1069         /* We're skipping the commit if there's an error */
1070         if (retval || reiserfs_is_journal_aborted(journal))
1071                 barrier = 0;
1072
1073         /* wait on everything written so far before writing the commit
1074          * if we are in barrier mode, send the commit down now
1075          */
1076         barrier = reiserfs_barrier_flush(s);
1077         if (barrier) {
1078                 int ret;
1079                 lock_buffer(jl->j_commit_bh);
1080                 ret = submit_barrier_buffer(jl->j_commit_bh);
1081                 if (ret == -EOPNOTSUPP) {
1082                         set_buffer_uptodate(jl->j_commit_bh);
1083                         disable_barrier(s);
1084                         barrier = 0;
1085                 }
1086         }
1087         for (i = 0; i < (jl->j_len + 1); i++) {
1088                 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) +
1089                     (jl->j_start + i) % SB_ONDISK_JOURNAL_SIZE(s);
1090                 tbh = journal_find_get_block(s, bn);
1091                 wait_on_buffer(tbh);
1092                 // since we're using ll_rw_blk above, it might have skipped over
1093                 // a locked buffer.  Double check here
1094                 //
1095                 if (buffer_dirty(tbh))  /* redundant, sync_dirty_buffer() checks */
1096                         sync_dirty_buffer(tbh);
1097                 if (unlikely(!buffer_uptodate(tbh))) {
1098 #ifdef CONFIG_REISERFS_CHECK
1099                         reiserfs_warning(s, "journal-601, buffer write failed");
1100 #endif
1101                         retval = -EIO;
1102                 }
1103                 put_bh(tbh);    /* once for journal_find_get_block */
1104                 put_bh(tbh);    /* once due to original getblk in do_journal_end */
1105                 atomic_dec(&(jl->j_commit_left));
1106         }
1107
1108         BUG_ON(atomic_read(&(jl->j_commit_left)) != 1);
1109
1110         if (!barrier) {
1111                 /* If there was a write error in the journal - we can't commit
1112                  * this transaction - it will be invalid and, if successful,
1113                  * will just end up propagating the write error out to
1114                  * the file system. */
1115                 if (likely(!retval && !reiserfs_is_journal_aborted (journal))) {
1116                         if (buffer_dirty(jl->j_commit_bh))
1117                                 BUG();
1118                         mark_buffer_dirty(jl->j_commit_bh) ;
1119                         sync_dirty_buffer(jl->j_commit_bh) ;
1120                 }
1121         } else
1122                 wait_on_buffer(jl->j_commit_bh);
1123
1124         check_barrier_completion(s, jl->j_commit_bh);
1125
1126         /* If there was a write error in the journal - we can't commit this
1127          * transaction - it will be invalid and, if successful, will just end
1128          * up propagating the write error out to the filesystem. */
1129         if (unlikely(!buffer_uptodate(jl->j_commit_bh))) {
1130 #ifdef CONFIG_REISERFS_CHECK
1131                 reiserfs_warning(s, "journal-615: buffer write failed");
1132 #endif
1133                 retval = -EIO;
1134         }
1135         bforget(jl->j_commit_bh);
1136         if (journal->j_last_commit_id != 0 &&
1137             (jl->j_trans_id - journal->j_last_commit_id) != 1) {
1138                 reiserfs_warning(s, "clm-2200: last commit %lu, current %lu",
1139                                  journal->j_last_commit_id, jl->j_trans_id);
1140         }
1141         journal->j_last_commit_id = jl->j_trans_id;
1142
1143         /* now, every commit block is on the disk.  It is safe to allow blocks freed during this transaction to be reallocated */
1144         cleanup_freed_for_journal_list(s, jl);
1145
1146         retval = retval ? retval : journal->j_errno;
1147
1148         /* mark the metadata dirty */
1149         if (!retval)
1150                 dirty_one_transaction(s, jl);
1151         atomic_dec(&(jl->j_commit_left));
1152
1153         if (flushall) {
1154                 atomic_set(&(jl->j_older_commits_done), 1);
1155         }
1156         up(&jl->j_commit_lock);
1157       put_jl:
1158         put_journal_list(s, jl);
1159
1160         if (retval)
1161                 reiserfs_abort(s, retval, "Journal write error in %s",
1162                                __FUNCTION__);
1163         put_fs_excl();
1164         return retval;
1165 }
1166
1167 /*
1168 ** flush_journal_list frequently needs to find a newer transaction for a given block.  This does that, or
1169 ** returns NULL if it can't find anything
1170 */
1171 static struct reiserfs_journal_list *find_newer_jl_for_cn(struct
1172                                                           reiserfs_journal_cnode
1173                                                           *cn)
1174 {
1175         struct super_block *sb = cn->sb;
1176         b_blocknr_t blocknr = cn->blocknr;
1177
1178         cn = cn->hprev;
1179         while (cn) {
1180                 if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist) {
1181                         return cn->jlist;
1182                 }
1183                 cn = cn->hprev;
1184         }
1185         return NULL;
1186 }
1187
1188 static int newer_jl_done(struct reiserfs_journal_cnode *cn)
1189 {
1190         struct super_block *sb = cn->sb;
1191         b_blocknr_t blocknr = cn->blocknr;
1192
1193         cn = cn->hprev;
1194         while (cn) {
1195                 if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist &&
1196                     atomic_read(&cn->jlist->j_commit_left) != 0)
1197                                     return 0;
1198                 cn = cn->hprev;
1199         }
1200         return 1;
1201 }
1202
1203 static void remove_journal_hash(struct super_block *,
1204                                 struct reiserfs_journal_cnode **,
1205                                 struct reiserfs_journal_list *, unsigned long,
1206                                 int);
1207
1208 /*
1209 ** once all the real blocks have been flushed, it is safe to remove them from the
1210 ** journal list for this transaction.  Aside from freeing the cnode, this also allows the
1211 ** block to be reallocated for data blocks if it had been deleted.
1212 */
1213 static void remove_all_from_journal_list(struct super_block *p_s_sb,
1214                                          struct reiserfs_journal_list *jl,
1215                                          int debug)
1216 {
1217         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1218         struct reiserfs_journal_cnode *cn, *last;
1219         cn = jl->j_realblock;
1220
1221         /* which is better, to lock once around the whole loop, or
1222          ** to lock for each call to remove_journal_hash?
1223          */
1224         while (cn) {
1225                 if (cn->blocknr != 0) {
1226                         if (debug) {
1227                                 reiserfs_warning(p_s_sb,
1228                                                  "block %u, bh is %d, state %ld",
1229                                                  cn->blocknr, cn->bh ? 1 : 0,
1230                                                  cn->state);
1231                         }
1232                         cn->state = 0;
1233                         remove_journal_hash(p_s_sb, journal->j_list_hash_table,
1234                                             jl, cn->blocknr, 1);
1235                 }
1236                 last = cn;
1237                 cn = cn->next;
1238                 free_cnode(p_s_sb, last);
1239         }
1240         jl->j_realblock = NULL;
1241 }
1242
1243 /*
1244 ** if this timestamp is greater than the timestamp we wrote last to the header block, write it to the header block.
1245 ** once this is done, I can safely say the log area for this transaction won't ever be replayed, and I can start
1246 ** releasing blocks in this transaction for reuse as data blocks.
1247 ** called by flush_journal_list, before it calls remove_all_from_journal_list
1248 **
1249 */
1250 static int _update_journal_header_block(struct super_block *p_s_sb,
1251                                         unsigned long offset,
1252                                         unsigned long trans_id)
1253 {
1254         struct reiserfs_journal_header *jh;
1255         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1256
1257         if (reiserfs_is_journal_aborted(journal))
1258                 return -EIO;
1259
1260         if (trans_id >= journal->j_last_flush_trans_id) {
1261                 if (buffer_locked((journal->j_header_bh))) {
1262                         wait_on_buffer((journal->j_header_bh));
1263                         if (unlikely(!buffer_uptodate(journal->j_header_bh))) {
1264 #ifdef CONFIG_REISERFS_CHECK
1265                                 reiserfs_warning(p_s_sb,
1266                                                  "journal-699: buffer write failed");
1267 #endif
1268                                 return -EIO;
1269                         }
1270                 }
1271                 journal->j_last_flush_trans_id = trans_id;
1272                 journal->j_first_unflushed_offset = offset;
1273                 jh = (struct reiserfs_journal_header *)(journal->j_header_bh->
1274                                                         b_data);
1275                 jh->j_last_flush_trans_id = cpu_to_le32(trans_id);
1276                 jh->j_first_unflushed_offset = cpu_to_le32(offset);
1277                 jh->j_mount_id = cpu_to_le32(journal->j_mount_id);
1278
1279                 if (reiserfs_barrier_flush(p_s_sb)) {
1280                         int ret;
1281                         lock_buffer(journal->j_header_bh);
1282                         ret = submit_barrier_buffer(journal->j_header_bh);
1283                         if (ret == -EOPNOTSUPP) {
1284                                 set_buffer_uptodate(journal->j_header_bh);
1285                                 disable_barrier(p_s_sb);
1286                                 goto sync;
1287                         }
1288                         wait_on_buffer(journal->j_header_bh);
1289                         check_barrier_completion(p_s_sb, journal->j_header_bh);
1290                 } else {
1291                       sync:
1292                         set_buffer_dirty(journal->j_header_bh);
1293                         sync_dirty_buffer(journal->j_header_bh);
1294                 }
1295                 if (!buffer_uptodate(journal->j_header_bh)) {
1296                         reiserfs_warning(p_s_sb,
1297                                          "journal-837: IO error during journal replay");
1298                         return -EIO;
1299                 }
1300         }
1301         return 0;
1302 }
1303
1304 static int update_journal_header_block(struct super_block *p_s_sb,
1305                                        unsigned long offset,
1306                                        unsigned long trans_id)
1307 {
1308         return _update_journal_header_block(p_s_sb, offset, trans_id);
1309 }
1310
1311 /*
1312 ** flush any and all journal lists older than you are
1313 ** can only be called from flush_journal_list
1314 */
1315 static int flush_older_journal_lists(struct super_block *p_s_sb,
1316                                      struct reiserfs_journal_list *jl)
1317 {
1318         struct list_head *entry;
1319         struct reiserfs_journal_list *other_jl;
1320         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1321         unsigned long trans_id = jl->j_trans_id;
1322
1323         /* we know we are the only ones flushing things, no extra race
1324          * protection is required.
1325          */
1326       restart:
1327         entry = journal->j_journal_list.next;
1328         /* Did we wrap? */
1329         if (entry == &journal->j_journal_list)
1330                 return 0;
1331         other_jl = JOURNAL_LIST_ENTRY(entry);
1332         if (other_jl->j_trans_id < trans_id) {
1333                 BUG_ON(other_jl->j_refcount <= 0);
1334                 /* do not flush all */
1335                 flush_journal_list(p_s_sb, other_jl, 0);
1336
1337                 /* other_jl is now deleted from the list */
1338                 goto restart;
1339         }
1340         return 0;
1341 }
1342
1343 static void del_from_work_list(struct super_block *s,
1344                                struct reiserfs_journal_list *jl)
1345 {
1346         struct reiserfs_journal *journal = SB_JOURNAL(s);
1347         if (!list_empty(&jl->j_working_list)) {
1348                 list_del_init(&jl->j_working_list);
1349                 journal->j_num_work_lists--;
1350         }
1351 }
1352
1353 /* flush a journal list, both commit and real blocks
1354 **
1355 ** always set flushall to 1, unless you are calling from inside
1356 ** flush_journal_list
1357 **
1358 ** IMPORTANT.  This can only be called while there are no journal writers,
1359 ** and the journal is locked.  That means it can only be called from
1360 ** do_journal_end, or by journal_release
1361 */
1362 static int flush_journal_list(struct super_block *s,
1363                               struct reiserfs_journal_list *jl, int flushall)
1364 {
1365         struct reiserfs_journal_list *pjl;
1366         struct reiserfs_journal_cnode *cn, *last;
1367         int count;
1368         int was_jwait = 0;
1369         int was_dirty = 0;
1370         struct buffer_head *saved_bh;
1371         unsigned long j_len_saved = jl->j_len;
1372         struct reiserfs_journal *journal = SB_JOURNAL(s);
1373         int err = 0;
1374
1375         BUG_ON(j_len_saved <= 0);
1376
1377         if (atomic_read(&journal->j_wcount) != 0) {
1378                 reiserfs_warning(s,
1379                                  "clm-2048: flush_journal_list called with wcount %d",
1380                                  atomic_read(&journal->j_wcount));
1381         }
1382         BUG_ON(jl->j_trans_id == 0);
1383
1384         /* if flushall == 0, the lock is already held */
1385         if (flushall) {
1386                 down(&journal->j_flush_sem);
1387         } else if (!down_trylock(&journal->j_flush_sem)) {
1388                 BUG();
1389         }
1390
1391         count = 0;
1392         if (j_len_saved > journal->j_trans_max) {
1393                 reiserfs_panic(s,
1394                                "journal-715: flush_journal_list, length is %lu, trans id %lu\n",
1395                                j_len_saved, jl->j_trans_id);
1396                 return 0;
1397         }
1398
1399         get_fs_excl();
1400
1401         /* if all the work is already done, get out of here */
1402         if (atomic_read(&(jl->j_nonzerolen)) <= 0 &&
1403             atomic_read(&(jl->j_commit_left)) <= 0) {
1404                 goto flush_older_and_return;
1405         }
1406
1407         /* start by putting the commit list on disk.  This will also flush
1408          ** the commit lists of any olders transactions
1409          */
1410         flush_commit_list(s, jl, 1);
1411
1412         if (!(jl->j_state & LIST_DIRTY)
1413             && !reiserfs_is_journal_aborted(journal))
1414                 BUG();
1415
1416         /* are we done now? */
1417         if (atomic_read(&(jl->j_nonzerolen)) <= 0 &&
1418             atomic_read(&(jl->j_commit_left)) <= 0) {
1419                 goto flush_older_and_return;
1420         }
1421
1422         /* loop through each cnode, see if we need to write it,
1423          ** or wait on a more recent transaction, or just ignore it
1424          */
1425         if (atomic_read(&(journal->j_wcount)) != 0) {
1426                 reiserfs_panic(s,
1427                                "journal-844: panic journal list is flushing, wcount is not 0\n");
1428         }
1429         cn = jl->j_realblock;
1430         while (cn) {
1431                 was_jwait = 0;
1432                 was_dirty = 0;
1433                 saved_bh = NULL;
1434                 /* blocknr of 0 is no longer in the hash, ignore it */
1435                 if (cn->blocknr == 0) {
1436                         goto free_cnode;
1437                 }
1438
1439                 /* This transaction failed commit. Don't write out to the disk */
1440                 if (!(jl->j_state & LIST_DIRTY))
1441                         goto free_cnode;
1442
1443                 pjl = find_newer_jl_for_cn(cn);
1444                 /* the order is important here.  We check pjl to make sure we
1445                  ** don't clear BH_JDirty_wait if we aren't the one writing this
1446                  ** block to disk
1447                  */
1448                 if (!pjl && cn->bh) {
1449                         saved_bh = cn->bh;
1450
1451                         /* we do this to make sure nobody releases the buffer while
1452                          ** we are working with it
1453                          */
1454                         get_bh(saved_bh);
1455
1456                         if (buffer_journal_dirty(saved_bh)) {
1457                                 BUG_ON(!can_dirty(cn));
1458                                 was_jwait = 1;
1459                                 was_dirty = 1;
1460                         } else if (can_dirty(cn)) {
1461                                 /* everything with !pjl && jwait should be writable */
1462                                 BUG();
1463                         }
1464                 }
1465
1466                 /* if someone has this block in a newer transaction, just make
1467                  ** sure they are committed, and don't try writing it to disk
1468                  */
1469                 if (pjl) {
1470                         if (atomic_read(&pjl->j_commit_left))
1471                                 flush_commit_list(s, pjl, 1);
1472                         goto free_cnode;
1473                 }
1474
1475                 /* bh == NULL when the block got to disk on its own, OR,
1476                  ** the block got freed in a future transaction
1477                  */
1478                 if (saved_bh == NULL) {
1479                         goto free_cnode;
1480                 }
1481
1482                 /* this should never happen.  kupdate_one_transaction has this list
1483                  ** locked while it works, so we should never see a buffer here that
1484                  ** is not marked JDirty_wait
1485                  */
1486                 if ((!was_jwait) && !buffer_locked(saved_bh)) {
1487                         reiserfs_warning(s,
1488                                          "journal-813: BAD! buffer %llu %cdirty %cjwait, "
1489                                          "not in a newer tranasction",
1490                                          (unsigned long long)saved_bh->
1491                                          b_blocknr, was_dirty ? ' ' : '!',
1492                                          was_jwait ? ' ' : '!');
1493                 }
1494                 if (was_dirty) {
1495                         /* we inc again because saved_bh gets decremented at free_cnode */
1496                         get_bh(saved_bh);
1497                         set_bit(BLOCK_NEEDS_FLUSH, &cn->state);
1498                         lock_buffer(saved_bh);
1499                         BUG_ON(cn->blocknr != saved_bh->b_blocknr);
1500                         if (buffer_dirty(saved_bh))
1501                                 submit_logged_buffer(saved_bh);
1502                         else
1503                                 unlock_buffer(saved_bh);
1504                         count++;
1505                 } else {
1506                         reiserfs_warning(s,
1507                                          "clm-2082: Unable to flush buffer %llu in %s",
1508                                          (unsigned long long)saved_bh->
1509                                          b_blocknr, __FUNCTION__);
1510                 }
1511               free_cnode:
1512                 last = cn;
1513                 cn = cn->next;
1514                 if (saved_bh) {
1515                         /* we incremented this to keep others from taking the buffer head away */
1516                         put_bh(saved_bh);
1517                         if (atomic_read(&(saved_bh->b_count)) < 0) {
1518                                 reiserfs_warning(s,
1519                                                  "journal-945: saved_bh->b_count < 0");
1520                         }
1521                 }
1522         }
1523         if (count > 0) {
1524                 cn = jl->j_realblock;
1525                 while (cn) {
1526                         if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) {
1527                                 if (!cn->bh) {
1528                                         reiserfs_panic(s,
1529                                                        "journal-1011: cn->bh is NULL\n");
1530                                 }
1531                                 wait_on_buffer(cn->bh);
1532                                 if (!cn->bh) {
1533                                         reiserfs_panic(s,
1534                                                        "journal-1012: cn->bh is NULL\n");
1535                                 }
1536                                 if (unlikely(!buffer_uptodate(cn->bh))) {
1537 #ifdef CONFIG_REISERFS_CHECK
1538                                         reiserfs_warning(s,
1539                                                          "journal-949: buffer write failed\n");
1540 #endif
1541                                         err = -EIO;
1542                                 }
1543                                 /* note, we must clear the JDirty_wait bit after the up to date
1544                                  ** check, otherwise we race against our flushpage routine
1545                                  */
1546                                 BUG_ON(!test_clear_buffer_journal_dirty
1547                                        (cn->bh));
1548
1549                                 /* undo the inc from journal_mark_dirty */
1550                                 put_bh(cn->bh);
1551                                 brelse(cn->bh);
1552                         }
1553                         cn = cn->next;
1554                 }
1555         }
1556
1557         if (err)
1558                 reiserfs_abort(s, -EIO,
1559                                "Write error while pushing transaction to disk in %s",
1560                                __FUNCTION__);
1561       flush_older_and_return:
1562
1563         /* before we can update the journal header block, we _must_ flush all
1564          ** real blocks from all older transactions to disk.  This is because
1565          ** once the header block is updated, this transaction will not be
1566          ** replayed after a crash
1567          */
1568         if (flushall) {
1569                 flush_older_journal_lists(s, jl);
1570         }
1571
1572         err = journal->j_errno;
1573         /* before we can remove everything from the hash tables for this
1574          ** transaction, we must make sure it can never be replayed
1575          **
1576          ** since we are only called from do_journal_end, we know for sure there
1577          ** are no allocations going on while we are flushing journal lists.  So,
1578          ** we only need to update the journal header block for the last list
1579          ** being flushed
1580          */
1581         if (!err && flushall) {
1582                 err =
1583                     update_journal_header_block(s,
1584                                                 (jl->j_start + jl->j_len +
1585                                                  2) % SB_ONDISK_JOURNAL_SIZE(s),
1586                                                 jl->j_trans_id);
1587                 if (err)
1588                         reiserfs_abort(s, -EIO,
1589                                        "Write error while updating journal header in %s",
1590                                        __FUNCTION__);
1591         }
1592         remove_all_from_journal_list(s, jl, 0);
1593         list_del_init(&jl->j_list);
1594         journal->j_num_lists--;
1595         del_from_work_list(s, jl);
1596
1597         if (journal->j_last_flush_id != 0 &&
1598             (jl->j_trans_id - journal->j_last_flush_id) != 1) {
1599                 reiserfs_warning(s, "clm-2201: last flush %lu, current %lu",
1600                                  journal->j_last_flush_id, jl->j_trans_id);
1601         }
1602         journal->j_last_flush_id = jl->j_trans_id;
1603
1604         /* not strictly required since we are freeing the list, but it should
1605          * help find code using dead lists later on
1606          */
1607         jl->j_len = 0;
1608         atomic_set(&(jl->j_nonzerolen), 0);
1609         jl->j_start = 0;
1610         jl->j_realblock = NULL;
1611         jl->j_commit_bh = NULL;
1612         jl->j_trans_id = 0;
1613         jl->j_state = 0;
1614         put_journal_list(s, jl);
1615         if (flushall)
1616                 up(&journal->j_flush_sem);
1617         put_fs_excl();
1618         return err;
1619 }
1620
1621 static int test_transaction(struct super_block *s,
1622                             struct reiserfs_journal_list *jl)
1623 {
1624         struct reiserfs_journal_cnode *cn;
1625
1626         if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0)
1627                 return 1;
1628
1629         cn = jl->j_realblock;
1630         while (cn) {
1631                 /* if the blocknr == 0, this has been cleared from the hash,
1632                  ** skip it
1633                  */
1634                 if (cn->blocknr == 0) {
1635                         goto next;
1636                 }
1637                 if (cn->bh && !newer_jl_done(cn))
1638                         return 0;
1639               next:
1640                 cn = cn->next;
1641                 cond_resched();
1642         }
1643         return 0;
1644 }
1645
1646 static int write_one_transaction(struct super_block *s,
1647                                  struct reiserfs_journal_list *jl,
1648                                  struct buffer_chunk *chunk)
1649 {
1650         struct reiserfs_journal_cnode *cn;
1651         int ret = 0;
1652
1653         jl->j_state |= LIST_TOUCHED;
1654         del_from_work_list(s, jl);
1655         if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0) {
1656                 return 0;
1657         }
1658
1659         cn = jl->j_realblock;
1660         while (cn) {
1661                 /* if the blocknr == 0, this has been cleared from the hash,
1662                  ** skip it
1663                  */
1664                 if (cn->blocknr == 0) {
1665                         goto next;
1666                 }
1667                 if (cn->bh && can_dirty(cn) && buffer_dirty(cn->bh)) {
1668                         struct buffer_head *tmp_bh;
1669                         /* we can race against journal_mark_freed when we try
1670                          * to lock_buffer(cn->bh), so we have to inc the buffer
1671                          * count, and recheck things after locking
1672                          */
1673                         tmp_bh = cn->bh;
1674                         get_bh(tmp_bh);
1675                         lock_buffer(tmp_bh);
1676                         if (cn->bh && can_dirty(cn) && buffer_dirty(tmp_bh)) {
1677                                 if (!buffer_journal_dirty(tmp_bh) ||
1678                                     buffer_journal_prepared(tmp_bh))
1679                                         BUG();
1680                                 add_to_chunk(chunk, tmp_bh, NULL, write_chunk);
1681                                 ret++;
1682                         } else {
1683                                 /* note, cn->bh might be null now */
1684                                 unlock_buffer(tmp_bh);
1685                         }
1686                         put_bh(tmp_bh);
1687                 }
1688               next:
1689                 cn = cn->next;
1690                 cond_resched();
1691         }
1692         return ret;
1693 }
1694
1695 /* used by flush_commit_list */
1696 static int dirty_one_transaction(struct super_block *s,
1697                                  struct reiserfs_journal_list *jl)
1698 {
1699         struct reiserfs_journal_cnode *cn;
1700         struct reiserfs_journal_list *pjl;
1701         int ret = 0;
1702
1703         jl->j_state |= LIST_DIRTY;
1704         cn = jl->j_realblock;
1705         while (cn) {
1706                 /* look for a more recent transaction that logged this
1707                  ** buffer.  Only the most recent transaction with a buffer in
1708                  ** it is allowed to send that buffer to disk
1709                  */
1710                 pjl = find_newer_jl_for_cn(cn);
1711                 if (!pjl && cn->blocknr && cn->bh
1712                     && buffer_journal_dirty(cn->bh)) {
1713                         BUG_ON(!can_dirty(cn));
1714                         /* if the buffer is prepared, it will either be logged
1715                          * or restored.  If restored, we need to make sure
1716                          * it actually gets marked dirty
1717                          */
1718                         clear_buffer_journal_new(cn->bh);
1719                         if (buffer_journal_prepared(cn->bh)) {
1720                                 set_buffer_journal_restore_dirty(cn->bh);
1721                         } else {
1722                                 set_buffer_journal_test(cn->bh);
1723                                 mark_buffer_dirty(cn->bh);
1724                         }
1725                 }
1726                 cn = cn->next;
1727         }
1728         return ret;
1729 }
1730
1731 static int kupdate_transactions(struct super_block *s,
1732                                 struct reiserfs_journal_list *jl,
1733                                 struct reiserfs_journal_list **next_jl,
1734                                 unsigned long *next_trans_id,
1735                                 int num_blocks, int num_trans)
1736 {
1737         int ret = 0;
1738         int written = 0;
1739         int transactions_flushed = 0;
1740         unsigned long orig_trans_id = jl->j_trans_id;
1741         struct buffer_chunk chunk;
1742         struct list_head *entry;
1743         struct reiserfs_journal *journal = SB_JOURNAL(s);
1744         chunk.nr = 0;
1745
1746         down(&journal->j_flush_sem);
1747         if (!journal_list_still_alive(s, orig_trans_id)) {
1748                 goto done;
1749         }
1750
1751         /* we've got j_flush_sem held, nobody is going to delete any
1752          * of these lists out from underneath us
1753          */
1754         while ((num_trans && transactions_flushed < num_trans) ||
1755                (!num_trans && written < num_blocks)) {
1756
1757                 if (jl->j_len == 0 || (jl->j_state & LIST_TOUCHED) ||
1758                     atomic_read(&jl->j_commit_left)
1759                     || !(jl->j_state & LIST_DIRTY)) {
1760                         del_from_work_list(s, jl);
1761                         break;
1762                 }
1763                 ret = write_one_transaction(s, jl, &chunk);
1764
1765                 if (ret < 0)
1766                         goto done;
1767                 transactions_flushed++;
1768                 written += ret;
1769                 entry = jl->j_list.next;
1770
1771                 /* did we wrap? */
1772                 if (entry == &journal->j_journal_list) {
1773                         break;
1774                 }
1775                 jl = JOURNAL_LIST_ENTRY(entry);
1776
1777                 /* don't bother with older transactions */
1778                 if (jl->j_trans_id <= orig_trans_id)
1779                         break;
1780         }
1781         if (chunk.nr) {
1782                 write_chunk(&chunk);
1783         }
1784
1785       done:
1786         up(&journal->j_flush_sem);
1787         return ret;
1788 }
1789
1790 /* for o_sync and fsync heavy applications, they tend to use
1791 ** all the journa list slots with tiny transactions.  These
1792 ** trigger lots and lots of calls to update the header block, which
1793 ** adds seeks and slows things down.
1794 **
1795 ** This function tries to clear out a large chunk of the journal lists
1796 ** at once, which makes everything faster since only the newest journal
1797 ** list updates the header block
1798 */
1799 static int flush_used_journal_lists(struct super_block *s,
1800                                     struct reiserfs_journal_list *jl)
1801 {
1802         unsigned long len = 0;
1803         unsigned long cur_len;
1804         int ret;
1805         int i;
1806         int limit = 256;
1807         struct reiserfs_journal_list *tjl;
1808         struct reiserfs_journal_list *flush_jl;
1809         unsigned long trans_id;
1810         struct reiserfs_journal *journal = SB_JOURNAL(s);
1811
1812         flush_jl = tjl = jl;
1813
1814         /* in data logging mode, try harder to flush a lot of blocks */
1815         if (reiserfs_data_log(s))
1816                 limit = 1024;
1817         /* flush for 256 transactions or limit blocks, whichever comes first */
1818         for (i = 0; i < 256 && len < limit; i++) {
1819                 if (atomic_read(&tjl->j_commit_left) ||
1820                     tjl->j_trans_id < jl->j_trans_id) {
1821                         break;
1822                 }
1823                 cur_len = atomic_read(&tjl->j_nonzerolen);
1824                 if (cur_len > 0) {
1825                         tjl->j_state &= ~LIST_TOUCHED;
1826                 }
1827                 len += cur_len;
1828                 flush_jl = tjl;
1829                 if (tjl->j_list.next == &journal->j_journal_list)
1830                         break;
1831                 tjl = JOURNAL_LIST_ENTRY(tjl->j_list.next);
1832         }
1833         /* try to find a group of blocks we can flush across all the
1834          ** transactions, but only bother if we've actually spanned
1835          ** across multiple lists
1836          */
1837         if (flush_jl != jl) {
1838                 ret = kupdate_transactions(s, jl, &tjl, &trans_id, len, i);
1839         }
1840         flush_journal_list(s, flush_jl, 1);
1841         return 0;
1842 }
1843
1844 /*
1845 ** removes any nodes in table with name block and dev as bh.
1846 ** only touchs the hnext and hprev pointers.
1847 */
1848 void remove_journal_hash(struct super_block *sb,
1849                          struct reiserfs_journal_cnode **table,
1850                          struct reiserfs_journal_list *jl,
1851                          unsigned long block, int remove_freed)
1852 {
1853         struct reiserfs_journal_cnode *cur;
1854         struct reiserfs_journal_cnode **head;
1855
1856         head = &(journal_hash(table, sb, block));
1857         if (!head) {
1858                 return;
1859         }
1860         cur = *head;
1861         while (cur) {
1862                 if (cur->blocknr == block && cur->sb == sb
1863                     && (jl == NULL || jl == cur->jlist)
1864                     && (!test_bit(BLOCK_FREED, &cur->state) || remove_freed)) {
1865                         if (cur->hnext) {
1866                                 cur->hnext->hprev = cur->hprev;
1867                         }
1868                         if (cur->hprev) {
1869                                 cur->hprev->hnext = cur->hnext;
1870                         } else {
1871                                 *head = cur->hnext;
1872                         }
1873                         cur->blocknr = 0;
1874                         cur->sb = NULL;
1875                         cur->state = 0;
1876                         if (cur->bh && cur->jlist)      /* anybody who clears the cur->bh will also dec the nonzerolen */
1877                                 atomic_dec(&(cur->jlist->j_nonzerolen));
1878                         cur->bh = NULL;
1879                         cur->jlist = NULL;
1880                 }
1881                 cur = cur->hnext;
1882         }
1883 }
1884
1885 static void free_journal_ram(struct super_block *p_s_sb)
1886 {
1887         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1888         kfree(journal->j_current_jl);
1889         journal->j_num_lists--;
1890
1891         vfree(journal->j_cnode_free_orig);
1892         free_list_bitmaps(p_s_sb, journal->j_list_bitmap);
1893         free_bitmap_nodes(p_s_sb);      /* must be after free_list_bitmaps */
1894         if (journal->j_header_bh) {
1895                 brelse(journal->j_header_bh);
1896         }
1897         /* j_header_bh is on the journal dev, make sure not to release the journal
1898          * dev until we brelse j_header_bh
1899          */
1900         release_journal_dev(p_s_sb, journal);
1901         vfree(journal);
1902 }
1903
1904 /*
1905 ** call on unmount.  Only set error to 1 if you haven't made your way out
1906 ** of read_super() yet.  Any other caller must keep error at 0.
1907 */
1908 static int do_journal_release(struct reiserfs_transaction_handle *th,
1909                               struct super_block *p_s_sb, int error)
1910 {
1911         struct reiserfs_transaction_handle myth;
1912         int flushed = 0;
1913         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1914
1915         /* we only want to flush out transactions if we were called with error == 0
1916          */
1917         if (!error && !(p_s_sb->s_flags & MS_RDONLY)) {
1918                 /* end the current trans */
1919                 BUG_ON(!th->t_trans_id);
1920                 do_journal_end(th, p_s_sb, 10, FLUSH_ALL);
1921
1922                 /* make sure something gets logged to force our way into the flush code */
1923                 if (!journal_join(&myth, p_s_sb, 1)) {
1924                         reiserfs_prepare_for_journal(p_s_sb,
1925                                                      SB_BUFFER_WITH_SB(p_s_sb),
1926                                                      1);
1927                         journal_mark_dirty(&myth, p_s_sb,
1928                                            SB_BUFFER_WITH_SB(p_s_sb));
1929                         do_journal_end(&myth, p_s_sb, 1, FLUSH_ALL);
1930                         flushed = 1;
1931                 }
1932         }
1933
1934         /* this also catches errors during the do_journal_end above */
1935         if (!error && reiserfs_is_journal_aborted(journal)) {
1936                 memset(&myth, 0, sizeof(myth));
1937                 if (!journal_join_abort(&myth, p_s_sb, 1)) {
1938                         reiserfs_prepare_for_journal(p_s_sb,
1939                                                      SB_BUFFER_WITH_SB(p_s_sb),
1940                                                      1);
1941                         journal_mark_dirty(&myth, p_s_sb,
1942                                            SB_BUFFER_WITH_SB(p_s_sb));
1943                         do_journal_end(&myth, p_s_sb, 1, FLUSH_ALL);
1944                 }
1945         }
1946
1947         reiserfs_mounted_fs_count--;
1948         /* wait for all commits to finish */
1949         cancel_delayed_work(&SB_JOURNAL(p_s_sb)->j_work);
1950         flush_workqueue(commit_wq);
1951         if (!reiserfs_mounted_fs_count) {
1952                 destroy_workqueue(commit_wq);
1953                 commit_wq = NULL;
1954         }
1955
1956         free_journal_ram(p_s_sb);
1957
1958         return 0;
1959 }
1960
1961 /*
1962 ** call on unmount.  flush all journal trans, release all alloc'd ram
1963 */
1964 int journal_release(struct reiserfs_transaction_handle *th,
1965                     struct super_block *p_s_sb)
1966 {
1967         return do_journal_release(th, p_s_sb, 0);
1968 }
1969
1970 /*
1971 ** only call from an error condition inside reiserfs_read_super!
1972 */
1973 int journal_release_error(struct reiserfs_transaction_handle *th,
1974                           struct super_block *p_s_sb)
1975 {
1976         return do_journal_release(th, p_s_sb, 1);
1977 }
1978
1979 /* compares description block with commit block.  returns 1 if they differ, 0 if they are the same */
1980 static int journal_compare_desc_commit(struct super_block *p_s_sb,
1981                                        struct reiserfs_journal_desc *desc,
1982                                        struct reiserfs_journal_commit *commit)
1983 {
1984         if (get_commit_trans_id(commit) != get_desc_trans_id(desc) ||
1985             get_commit_trans_len(commit) != get_desc_trans_len(desc) ||
1986             get_commit_trans_len(commit) > SB_JOURNAL(p_s_sb)->j_trans_max ||
1987             get_commit_trans_len(commit) <= 0) {
1988                 return 1;
1989         }
1990         return 0;
1991 }
1992
1993 /* returns 0 if it did not find a description block
1994 ** returns -1 if it found a corrupt commit block
1995 ** returns 1 if both desc and commit were valid
1996 */
1997 static int journal_transaction_is_valid(struct super_block *p_s_sb,
1998                                         struct buffer_head *d_bh,
1999                                         unsigned long *oldest_invalid_trans_id,
2000                                         unsigned long *newest_mount_id)
2001 {
2002         struct reiserfs_journal_desc *desc;
2003         struct reiserfs_journal_commit *commit;
2004         struct buffer_head *c_bh;
2005         unsigned long offset;
2006
2007         if (!d_bh)
2008                 return 0;
2009
2010         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2011         if (get_desc_trans_len(desc) > 0
2012             && !memcmp(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8)) {
2013                 if (oldest_invalid_trans_id && *oldest_invalid_trans_id
2014                     && get_desc_trans_id(desc) > *oldest_invalid_trans_id) {
2015                         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2016                                        "journal-986: transaction "
2017                                        "is valid returning because trans_id %d is greater than "
2018                                        "oldest_invalid %lu",
2019                                        get_desc_trans_id(desc),
2020                                        *oldest_invalid_trans_id);
2021                         return 0;
2022                 }
2023                 if (newest_mount_id
2024                     && *newest_mount_id > get_desc_mount_id(desc)) {
2025                         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2026                                        "journal-1087: transaction "
2027                                        "is valid returning because mount_id %d is less than "
2028                                        "newest_mount_id %lu",
2029                                        get_desc_mount_id(desc),
2030                                        *newest_mount_id);
2031                         return -1;
2032                 }
2033                 if (get_desc_trans_len(desc) > SB_JOURNAL(p_s_sb)->j_trans_max) {
2034                         reiserfs_warning(p_s_sb,
2035                                          "journal-2018: Bad transaction length %d encountered, ignoring transaction",
2036                                          get_desc_trans_len(desc));
2037                         return -1;
2038                 }
2039                 offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb);
2040
2041                 /* ok, we have a journal description block, lets see if the transaction was valid */
2042                 c_bh =
2043                     journal_bread(p_s_sb,
2044                                   SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2045                                   ((offset + get_desc_trans_len(desc) +
2046                                     1) % SB_ONDISK_JOURNAL_SIZE(p_s_sb)));
2047                 if (!c_bh)
2048                         return 0;
2049                 commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2050                 if (journal_compare_desc_commit(p_s_sb, desc, commit)) {
2051                         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2052                                        "journal_transaction_is_valid, commit offset %ld had bad "
2053                                        "time %d or length %d",
2054                                        c_bh->b_blocknr -
2055                                        SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2056                                        get_commit_trans_id(commit),
2057                                        get_commit_trans_len(commit));
2058                         brelse(c_bh);
2059                         if (oldest_invalid_trans_id) {
2060                                 *oldest_invalid_trans_id =
2061                                     get_desc_trans_id(desc);
2062                                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2063                                                "journal-1004: "
2064                                                "transaction_is_valid setting oldest invalid trans_id "
2065                                                "to %d",
2066                                                get_desc_trans_id(desc));
2067                         }
2068                         return -1;
2069                 }
2070                 brelse(c_bh);
2071                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2072                                "journal-1006: found valid "
2073                                "transaction start offset %llu, len %d id %d",
2074                                d_bh->b_blocknr -
2075                                SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2076                                get_desc_trans_len(desc),
2077                                get_desc_trans_id(desc));
2078                 return 1;
2079         } else {
2080                 return 0;
2081         }
2082 }
2083
2084 static void brelse_array(struct buffer_head **heads, int num)
2085 {
2086         int i;
2087         for (i = 0; i < num; i++) {
2088                 brelse(heads[i]);
2089         }
2090 }
2091
2092 /*
2093 ** given the start, and values for the oldest acceptable transactions,
2094 ** this either reads in a replays a transaction, or returns because the transaction
2095 ** is invalid, or too old.
2096 */
2097 static int journal_read_transaction(struct super_block *p_s_sb,
2098                                     unsigned long cur_dblock,
2099                                     unsigned long oldest_start,
2100                                     unsigned long oldest_trans_id,
2101                                     unsigned long newest_mount_id)
2102 {
2103         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
2104         struct reiserfs_journal_desc *desc;
2105         struct reiserfs_journal_commit *commit;
2106         unsigned long trans_id = 0;
2107         struct buffer_head *c_bh;
2108         struct buffer_head *d_bh;
2109         struct buffer_head **log_blocks = NULL;
2110         struct buffer_head **real_blocks = NULL;
2111         unsigned long trans_offset;
2112         int i;
2113         int trans_half;
2114
2115         d_bh = journal_bread(p_s_sb, cur_dblock);
2116         if (!d_bh)
2117                 return 1;
2118         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2119         trans_offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb);
2120         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1037: "
2121                        "journal_read_transaction, offset %llu, len %d mount_id %d",
2122                        d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2123                        get_desc_trans_len(desc), get_desc_mount_id(desc));
2124         if (get_desc_trans_id(desc) < oldest_trans_id) {
2125                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1039: "
2126                                "journal_read_trans skipping because %lu is too old",
2127                                cur_dblock -
2128                                SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb));
2129                 brelse(d_bh);
2130                 return 1;
2131         }
2132         if (get_desc_mount_id(desc) != newest_mount_id) {
2133                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1146: "
2134                                "journal_read_trans skipping because %d is != "
2135                                "newest_mount_id %lu", get_desc_mount_id(desc),
2136                                newest_mount_id);
2137                 brelse(d_bh);
2138                 return 1;
2139         }
2140         c_bh = journal_bread(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2141                              ((trans_offset + get_desc_trans_len(desc) + 1) %
2142                               SB_ONDISK_JOURNAL_SIZE(p_s_sb)));
2143         if (!c_bh) {
2144                 brelse(d_bh);
2145                 return 1;
2146         }
2147         commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2148         if (journal_compare_desc_commit(p_s_sb, desc, commit)) {
2149                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2150                                "journal_read_transaction, "
2151                                "commit offset %llu had bad time %d or length %d",
2152                                c_bh->b_blocknr -
2153                                SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2154                                get_commit_trans_id(commit),
2155                                get_commit_trans_len(commit));
2156                 brelse(c_bh);
2157                 brelse(d_bh);
2158                 return 1;
2159         }
2160         trans_id = get_desc_trans_id(desc);
2161         /* now we know we've got a good transaction, and it was inside the valid time ranges */
2162         log_blocks = kmalloc(get_desc_trans_len(desc) *
2163                              sizeof(struct buffer_head *), GFP_NOFS);
2164         real_blocks = kmalloc(get_desc_trans_len(desc) *
2165                               sizeof(struct buffer_head *), GFP_NOFS);
2166         if (!log_blocks || !real_blocks) {
2167                 brelse(c_bh);
2168                 brelse(d_bh);
2169                 kfree(log_blocks);
2170                 kfree(real_blocks);
2171                 reiserfs_warning(p_s_sb,
2172                                  "journal-1169: kmalloc failed, unable to mount FS");
2173                 return -1;
2174         }
2175         /* get all the buffer heads */
2176         trans_half = journal_trans_half(p_s_sb->s_blocksize);
2177         for (i = 0; i < get_desc_trans_len(desc); i++) {
2178                 log_blocks[i] =
2179                     journal_getblk(p_s_sb,
2180                                    SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2181                                    (trans_offset + 1 +
2182                                     i) % SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2183                 if (i < trans_half) {
2184                         real_blocks[i] =
2185                             sb_getblk(p_s_sb,
2186                                       le32_to_cpu(desc->j_realblock[i]));
2187                 } else {
2188                         real_blocks[i] =
2189                             sb_getblk(p_s_sb,
2190                                       le32_to_cpu(commit->
2191                                                   j_realblock[i - trans_half]));
2192                 }
2193                 if (real_blocks[i]->b_blocknr > SB_BLOCK_COUNT(p_s_sb)) {
2194                         reiserfs_warning(p_s_sb,
2195                                          "journal-1207: REPLAY FAILURE fsck required! Block to replay is outside of filesystem");
2196                         goto abort_replay;
2197                 }
2198                 /* make sure we don't try to replay onto log or reserved area */
2199                 if (is_block_in_log_or_reserved_area
2200                     (p_s_sb, real_blocks[i]->b_blocknr)) {
2201                         reiserfs_warning(p_s_sb,
2202                                          "journal-1204: REPLAY FAILURE fsck required! Trying to replay onto a log block");
2203                       abort_replay:
2204                         brelse_array(log_blocks, i);
2205                         brelse_array(real_blocks, i);
2206                         brelse(c_bh);
2207                         brelse(d_bh);
2208                         kfree(log_blocks);
2209                         kfree(real_blocks);
2210                         return -1;
2211                 }
2212         }
2213         /* read in the log blocks, memcpy to the corresponding real block */
2214         ll_rw_block(READ, get_desc_trans_len(desc), log_blocks);
2215         for (i = 0; i < get_desc_trans_len(desc); i++) {
2216                 wait_on_buffer(log_blocks[i]);
2217                 if (!buffer_uptodate(log_blocks[i])) {
2218                         reiserfs_warning(p_s_sb,
2219                                          "journal-1212: REPLAY FAILURE fsck required! buffer write failed");
2220                         brelse_array(log_blocks + i,
2221                                      get_desc_trans_len(desc) - i);
2222                         brelse_array(real_blocks, get_desc_trans_len(desc));
2223                         brelse(c_bh);
2224                         brelse(d_bh);
2225                         kfree(log_blocks);
2226                         kfree(real_blocks);
2227                         return -1;
2228                 }
2229                 memcpy(real_blocks[i]->b_data, log_blocks[i]->b_data,
2230                        real_blocks[i]->b_size);
2231                 set_buffer_uptodate(real_blocks[i]);
2232                 brelse(log_blocks[i]);
2233         }
2234         /* flush out the real blocks */
2235         for (i = 0; i < get_desc_trans_len(desc); i++) {
2236                 set_buffer_dirty(real_blocks[i]);
2237                 ll_rw_block(SWRITE, 1, real_blocks + i);
2238         }
2239         for (i = 0; i < get_desc_trans_len(desc); i++) {
2240                 wait_on_buffer(real_blocks[i]);
2241                 if (!buffer_uptodate(real_blocks[i])) {
2242                         reiserfs_warning(p_s_sb,
2243                                          "journal-1226: REPLAY FAILURE, fsck required! buffer write failed");
2244                         brelse_array(real_blocks + i,
2245                                      get_desc_trans_len(desc) - i);
2246                         brelse(c_bh);
2247                         brelse(d_bh);
2248                         kfree(log_blocks);
2249                         kfree(real_blocks);
2250                         return -1;
2251                 }
2252                 brelse(real_blocks[i]);
2253         }
2254         cur_dblock =
2255             SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2256             ((trans_offset + get_desc_trans_len(desc) +
2257               2) % SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2258         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2259                        "journal-1095: setting journal " "start to offset %ld",
2260                        cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb));
2261
2262         /* init starting values for the first transaction, in case this is the last transaction to be replayed. */
2263         journal->j_start = cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb);
2264         journal->j_last_flush_trans_id = trans_id;
2265         journal->j_trans_id = trans_id + 1;
2266         /* check for trans_id overflow */
2267         if (journal->j_trans_id == 0)
2268                 journal->j_trans_id = 10;
2269         brelse(c_bh);
2270         brelse(d_bh);
2271         kfree(log_blocks);
2272         kfree(real_blocks);
2273         return 0;
2274 }
2275
2276 /* This function reads blocks starting from block and to max_block of bufsize
2277    size (but no more than BUFNR blocks at a time). This proved to improve
2278    mounting speed on self-rebuilding raid5 arrays at least.
2279    Right now it is only used from journal code. But later we might use it
2280    from other places.
2281    Note: Do not use journal_getblk/sb_getblk functions here! */
2282 static struct buffer_head *reiserfs_breada(struct block_device *dev, int block,
2283                                            int bufsize, unsigned int max_block)
2284 {
2285         struct buffer_head *bhlist[BUFNR];
2286         unsigned int blocks = BUFNR;
2287         struct buffer_head *bh;
2288         int i, j;
2289
2290         bh = __getblk(dev, block, bufsize);
2291         if (buffer_uptodate(bh))
2292                 return (bh);
2293
2294         if (block + BUFNR > max_block) {
2295                 blocks = max_block - block;
2296         }
2297         bhlist[0] = bh;
2298         j = 1;
2299         for (i = 1; i < blocks; i++) {
2300                 bh = __getblk(dev, block + i, bufsize);
2301                 if (buffer_uptodate(bh)) {
2302                         brelse(bh);
2303                         break;
2304                 } else
2305                         bhlist[j++] = bh;
2306         }
2307         ll_rw_block(READ, j, bhlist);
2308         for (i = 1; i < j; i++)
2309                 brelse(bhlist[i]);
2310         bh = bhlist[0];
2311         wait_on_buffer(bh);
2312         if (buffer_uptodate(bh))
2313                 return bh;
2314         brelse(bh);
2315         return NULL;
2316 }
2317
2318 /*
2319 ** read and replay the log
2320 ** on a clean unmount, the journal header's next unflushed pointer will be to an invalid
2321 ** transaction.  This tests that before finding all the transactions in the log, which makes normal mount times fast.
2322 **
2323 ** After a crash, this starts with the next unflushed transaction, and replays until it finds one too old, or invalid.
2324 **
2325 ** On exit, it sets things up so the first transaction will work correctly.
2326 */
2327 static int journal_read(struct super_block *p_s_sb)
2328 {
2329         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
2330         struct reiserfs_journal_desc *desc;
2331         unsigned long oldest_trans_id = 0;
2332         unsigned long oldest_invalid_trans_id = 0;
2333         time_t start;
2334         unsigned long oldest_start = 0;
2335         unsigned long cur_dblock = 0;
2336         unsigned long newest_mount_id = 9;
2337         struct buffer_head *d_bh;
2338         struct reiserfs_journal_header *jh;
2339         int valid_journal_header = 0;
2340         int replay_count = 0;
2341         int continue_replay = 1;
2342         int ret;
2343         char b[BDEVNAME_SIZE];
2344
2345         cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb);
2346         reiserfs_info(p_s_sb, "checking transaction log (%s)\n",
2347                       bdevname(journal->j_dev_bd, b));
2348         start = get_seconds();
2349
2350         /* step 1, read in the journal header block.  Check the transaction it says
2351          ** is the first unflushed, and if that transaction is not valid,
2352          ** replay is done
2353          */
2354         journal->j_header_bh = journal_bread(p_s_sb,
2355                                              SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb)
2356                                              + SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2357         if (!journal->j_header_bh) {
2358                 return 1;
2359         }
2360         jh = (struct reiserfs_journal_header *)(journal->j_header_bh->b_data);
2361         if (le32_to_cpu(jh->j_first_unflushed_offset) <
2362             SB_ONDISK_JOURNAL_SIZE(p_s_sb)
2363             && le32_to_cpu(jh->j_last_flush_trans_id) > 0) {
2364                 oldest_start =
2365                     SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2366                     le32_to_cpu(jh->j_first_unflushed_offset);
2367                 oldest_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2368                 newest_mount_id = le32_to_cpu(jh->j_mount_id);
2369                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2370                                "journal-1153: found in "
2371                                "header: first_unflushed_offset %d, last_flushed_trans_id "
2372                                "%lu", le32_to_cpu(jh->j_first_unflushed_offset),
2373                                le32_to_cpu(jh->j_last_flush_trans_id));
2374                 valid_journal_header = 1;
2375
2376                 /* now, we try to read the first unflushed offset.  If it is not valid,
2377                  ** there is nothing more we can do, and it makes no sense to read
2378                  ** through the whole log.
2379                  */
2380                 d_bh =
2381                     journal_bread(p_s_sb,
2382                                   SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2383                                   le32_to_cpu(jh->j_first_unflushed_offset));
2384                 ret = journal_transaction_is_valid(p_s_sb, d_bh, NULL, NULL);
2385                 if (!ret) {
2386                         continue_replay = 0;
2387                 }
2388                 brelse(d_bh);
2389                 goto start_log_replay;
2390         }
2391
2392         if (continue_replay && bdev_read_only(p_s_sb->s_bdev)) {
2393                 reiserfs_warning(p_s_sb,
2394                                  "clm-2076: device is readonly, unable to replay log");
2395                 return -1;
2396         }
2397
2398         /* ok, there are transactions that need to be replayed.  start with the first log block, find
2399          ** all the valid transactions, and pick out the oldest.
2400          */
2401         while (continue_replay
2402                && cur_dblock <
2403                (SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2404                 SB_ONDISK_JOURNAL_SIZE(p_s_sb))) {
2405                 /* Note that it is required for blocksize of primary fs device and journal
2406                    device to be the same */
2407                 d_bh =
2408                     reiserfs_breada(journal->j_dev_bd, cur_dblock,
2409                                     p_s_sb->s_blocksize,
2410                                     SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2411                                     SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2412                 ret =
2413                     journal_transaction_is_valid(p_s_sb, d_bh,
2414                                                  &oldest_invalid_trans_id,
2415                                                  &newest_mount_id);
2416                 if (ret == 1) {
2417                         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2418                         if (oldest_start == 0) {        /* init all oldest_ values */
2419                                 oldest_trans_id = get_desc_trans_id(desc);
2420                                 oldest_start = d_bh->b_blocknr;
2421                                 newest_mount_id = get_desc_mount_id(desc);
2422                                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2423                                                "journal-1179: Setting "
2424                                                "oldest_start to offset %llu, trans_id %lu",
2425                                                oldest_start -
2426                                                SB_ONDISK_JOURNAL_1st_BLOCK
2427                                                (p_s_sb), oldest_trans_id);
2428                         } else if (oldest_trans_id > get_desc_trans_id(desc)) {
2429                                 /* one we just read was older */
2430                                 oldest_trans_id = get_desc_trans_id(desc);
2431                                 oldest_start = d_bh->b_blocknr;
2432                                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2433                                                "journal-1180: Resetting "
2434                                                "oldest_start to offset %lu, trans_id %lu",
2435                                                oldest_start -
2436                                                SB_ONDISK_JOURNAL_1st_BLOCK
2437                                                (p_s_sb), oldest_trans_id);
2438                         }
2439                         if (newest_mount_id < get_desc_mount_id(desc)) {
2440                                 newest_mount_id = get_desc_mount_id(desc);
2441                                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2442                                                "journal-1299: Setting "
2443                                                "newest_mount_id to %d",
2444                                                get_desc_mount_id(desc));
2445                         }
2446                         cur_dblock += get_desc_trans_len(desc) + 2;
2447                 } else {
2448                         cur_dblock++;
2449                 }
2450                 brelse(d_bh);
2451         }
2452
2453       start_log_replay:
2454         cur_dblock = oldest_start;
2455         if (oldest_trans_id) {
2456                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2457                                "journal-1206: Starting replay "
2458                                "from offset %llu, trans_id %lu",
2459                                cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2460                                oldest_trans_id);
2461
2462         }
2463         replay_count = 0;
2464         while (continue_replay && oldest_trans_id > 0) {
2465                 ret =
2466                     journal_read_transaction(p_s_sb, cur_dblock, oldest_start,
2467                                              oldest_trans_id, newest_mount_id);
2468                 if (ret < 0) {
2469                         return ret;
2470                 } else if (ret != 0) {
2471                         break;
2472                 }
2473                 cur_dblock =
2474                     SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + journal->j_start;
2475                 replay_count++;
2476                 if (cur_dblock == oldest_start)
2477                         break;
2478         }
2479
2480         if (oldest_trans_id == 0) {
2481                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2482                                "journal-1225: No valid " "transactions found");
2483         }
2484         /* j_start does not get set correctly if we don't replay any transactions.
2485          ** if we had a valid journal_header, set j_start to the first unflushed transaction value,
2486          ** copy the trans_id from the header
2487          */
2488         if (valid_journal_header && replay_count == 0) {
2489                 journal->j_start = le32_to_cpu(jh->j_first_unflushed_offset);
2490                 journal->j_trans_id =
2491                     le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2492                 /* check for trans_id overflow */
2493                 if (journal->j_trans_id == 0)
2494                         journal->j_trans_id = 10;
2495                 journal->j_last_flush_trans_id =
2496                     le32_to_cpu(jh->j_last_flush_trans_id);
2497                 journal->j_mount_id = le32_to_cpu(jh->j_mount_id) + 1;
2498         } else {
2499                 journal->j_mount_id = newest_mount_id + 1;
2500         }
2501         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1299: Setting "
2502                        "newest_mount_id to %lu", journal->j_mount_id);
2503         journal->j_first_unflushed_offset = journal->j_start;
2504         if (replay_count > 0) {
2505                 reiserfs_info(p_s_sb,
2506                               "replayed %d transactions in %lu seconds\n",
2507                               replay_count, get_seconds() - start);
2508         }
2509         if (!bdev_read_only(p_s_sb->s_bdev) &&
2510             _update_journal_header_block(p_s_sb, journal->j_start,
2511                                          journal->j_last_flush_trans_id)) {
2512                 /* replay failed, caller must call free_journal_ram and abort
2513                  ** the mount
2514                  */
2515                 return -1;
2516         }
2517         return 0;
2518 }
2519
2520 static struct reiserfs_journal_list *alloc_journal_list(struct super_block *s)
2521 {
2522         struct reiserfs_journal_list *jl;
2523         jl = kzalloc(sizeof(struct reiserfs_journal_list),
2524                      GFP_NOFS | __GFP_NOFAIL);
2525         INIT_LIST_HEAD(&jl->j_list);
2526         INIT_LIST_HEAD(&jl->j_working_list);
2527         INIT_LIST_HEAD(&jl->j_tail_bh_list);
2528         INIT_LIST_HEAD(&jl->j_bh_list);
2529         sema_init(&jl->j_commit_lock, 1);
2530         SB_JOURNAL(s)->j_num_lists++;
2531         get_journal_list(jl);
2532         return jl;
2533 }
2534
2535 static void journal_list_init(struct super_block *p_s_sb)
2536 {
2537         SB_JOURNAL(p_s_sb)->j_current_jl = alloc_journal_list(p_s_sb);
2538 }
2539
2540 static int release_journal_dev(struct super_block *super,
2541                                struct reiserfs_journal *journal)
2542 {
2543         int result;
2544
2545         result = 0;
2546
2547         if (journal->j_dev_file != NULL) {
2548                 result = filp_close(journal->j_dev_file, NULL);
2549                 journal->j_dev_file = NULL;
2550                 journal->j_dev_bd = NULL;
2551         } else if (journal->j_dev_bd != NULL) {
2552                 result = blkdev_put(journal->j_dev_bd);
2553                 journal->j_dev_bd = NULL;
2554         }
2555
2556         if (result != 0) {
2557                 reiserfs_warning(super,
2558                                  "sh-457: release_journal_dev: Cannot release journal device: %i",
2559                                  result);
2560         }
2561         return result;
2562 }
2563
2564 static int journal_init_dev(struct super_block *super,
2565                             struct reiserfs_journal *journal,
2566                             const char *jdev_name)
2567 {
2568         int result;
2569         dev_t jdev;
2570         int blkdev_mode = FMODE_READ | FMODE_WRITE;
2571         char b[BDEVNAME_SIZE];
2572
2573         result = 0;
2574
2575         journal->j_dev_bd = NULL;
2576         journal->j_dev_file = NULL;
2577         jdev = SB_ONDISK_JOURNAL_DEVICE(super) ?
2578             new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super)) : super->s_dev;
2579
2580         if (bdev_read_only(super->s_bdev))
2581                 blkdev_mode = FMODE_READ;
2582
2583         /* there is no "jdev" option and journal is on separate device */
2584         if ((!jdev_name || !jdev_name[0])) {
2585                 journal->j_dev_bd = open_by_devnum(jdev, blkdev_mode);
2586                 if (IS_ERR(journal->j_dev_bd)) {
2587                         result = PTR_ERR(journal->j_dev_bd);
2588                         journal->j_dev_bd = NULL;
2589                         reiserfs_warning(super, "sh-458: journal_init_dev: "
2590                                          "cannot init journal device '%s': %i",
2591                                          __bdevname(jdev, b), result);
2592                         return result;
2593                 } else if (jdev != super->s_dev)
2594                         set_blocksize(journal->j_dev_bd, super->s_blocksize);
2595                 return 0;
2596         }
2597
2598         journal->j_dev_file = filp_open(jdev_name, 0, 0);
2599         if (!IS_ERR(journal->j_dev_file)) {
2600                 struct inode *jdev_inode = journal->j_dev_file->f_mapping->host;
2601                 if (!S_ISBLK(jdev_inode->i_mode)) {
2602                         reiserfs_warning(super, "journal_init_dev: '%s' is "
2603                                          "not a block device", jdev_name);
2604                         result = -ENOTBLK;
2605                         release_journal_dev(super, journal);
2606                 } else {
2607                         /* ok */
2608                         journal->j_dev_bd = I_BDEV(jdev_inode);
2609                         set_blocksize(journal->j_dev_bd, super->s_blocksize);
2610                         reiserfs_info(super,
2611                                       "journal_init_dev: journal device: %s\n",
2612                                       bdevname(journal->j_dev_bd, b));
2613                 }
2614         } else {
2615                 result = PTR_ERR(journal->j_dev_file);
2616                 journal->j_dev_file = NULL;
2617                 reiserfs_warning(super,
2618                                  "journal_init_dev: Cannot open '%s': %i",
2619                                  jdev_name, result);
2620         }
2621         return result;
2622 }
2623
2624 /*
2625 ** must be called once on fs mount.  calls journal_read for you
2626 */
2627 int journal_init(struct super_block *p_s_sb, const char *j_dev_name,
2628                  int old_format, unsigned int commit_max_age)
2629 {
2630         int num_cnodes = SB_ONDISK_JOURNAL_SIZE(p_s_sb) * 2;
2631         struct buffer_head *bhjh;
2632         struct reiserfs_super_block *rs;
2633         struct reiserfs_journal_header *jh;
2634         struct reiserfs_journal *journal;
2635         struct reiserfs_journal_list *jl;
2636         char b[BDEVNAME_SIZE];
2637
2638         journal = SB_JOURNAL(p_s_sb) = vmalloc(sizeof(struct reiserfs_journal));
2639         if (!journal) {
2640                 reiserfs_warning(p_s_sb,
2641                                  "journal-1256: unable to get memory for journal structure");
2642                 return 1;
2643         }
2644         memset(journal, 0, sizeof(struct reiserfs_journal));
2645         INIT_LIST_HEAD(&journal->j_bitmap_nodes);
2646         INIT_LIST_HEAD(&journal->j_prealloc_list);
2647         INIT_LIST_HEAD(&journal->j_working_list);
2648         INIT_LIST_HEAD(&journal->j_journal_list);
2649         journal->j_persistent_trans = 0;
2650         if (reiserfs_allocate_list_bitmaps(p_s_sb,
2651                                            journal->j_list_bitmap,
2652                                            SB_BMAP_NR(p_s_sb)))
2653                 goto free_and_return;
2654         allocate_bitmap_nodes(p_s_sb);
2655
2656         /* reserved for journal area support */
2657         SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb) = (old_format ?
2658                                                  REISERFS_OLD_DISK_OFFSET_IN_BYTES
2659                                                  / p_s_sb->s_blocksize +
2660                                                  SB_BMAP_NR(p_s_sb) +
2661                                                  1 :
2662                                                  REISERFS_DISK_OFFSET_IN_BYTES /
2663                                                  p_s_sb->s_blocksize + 2);
2664
2665         /* Sanity check to see is the standard journal fitting withing first bitmap
2666            (actual for small blocksizes) */
2667         if (!SB_ONDISK_JOURNAL_DEVICE(p_s_sb) &&
2668             (SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb) +
2669              SB_ONDISK_JOURNAL_SIZE(p_s_sb) > p_s_sb->s_blocksize * 8)) {
2670                 reiserfs_warning(p_s_sb,
2671                                  "journal-1393: journal does not fit for area "
2672                                  "addressed by first of bitmap blocks. It starts at "
2673                                  "%u and its size is %u. Block size %ld",
2674                                  SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb),
2675                                  SB_ONDISK_JOURNAL_SIZE(p_s_sb),
2676                                  p_s_sb->s_blocksize);
2677                 goto free_and_return;
2678         }
2679
2680         if (journal_init_dev(p_s_sb, journal, j_dev_name) != 0) {
2681                 reiserfs_warning(p_s_sb,
2682                                  "sh-462: unable to initialize jornal device");
2683                 goto free_and_return;
2684         }
2685
2686         rs = SB_DISK_SUPER_BLOCK(p_s_sb);
2687
2688         /* read journal header */
2689         bhjh = journal_bread(p_s_sb,
2690                              SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2691                              SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2692         if (!bhjh) {
2693                 reiserfs_warning(p_s_sb,
2694                                  "sh-459: unable to read journal header");
2695                 goto free_and_return;
2696         }
2697         jh = (struct reiserfs_journal_header *)(bhjh->b_data);
2698
2699         /* make sure that journal matches to the super block */
2700         if (is_reiserfs_jr(rs)
2701             && (le32_to_cpu(jh->jh_journal.jp_journal_magic) !=
2702                 sb_jp_journal_magic(rs))) {
2703                 reiserfs_warning(p_s_sb,
2704                                  "sh-460: journal header magic %x "
2705                                  "(device %s) does not match to magic found in super "
2706                                  "block %x", jh->jh_journal.jp_journal_magic,
2707                                  bdevname(journal->j_dev_bd, b),
2708                                  sb_jp_journal_magic(rs));
2709                 brelse(bhjh);
2710                 goto free_and_return;
2711         }
2712
2713         journal->j_trans_max = le32_to_cpu(jh->jh_journal.jp_journal_trans_max);
2714         journal->j_max_batch = le32_to_cpu(jh->jh_journal.jp_journal_max_batch);
2715         journal->j_max_commit_age =
2716             le32_to_cpu(jh->jh_journal.jp_journal_max_commit_age);
2717         journal->j_max_trans_age = JOURNAL_MAX_TRANS_AGE;
2718
2719         if (journal->j_trans_max) {
2720                 /* make sure these parameters are available, assign it if they are not */
2721                 __u32 initial = journal->j_trans_max;
2722                 __u32 ratio = 1;
2723
2724                 if (p_s_sb->s_blocksize < 4096)
2725                         ratio = 4096 / p_s_sb->s_blocksize;
2726
2727                 if (SB_ONDISK_JOURNAL_SIZE(p_s_sb) / journal->j_trans_max <
2728                     JOURNAL_MIN_RATIO)
2729                         journal->j_trans_max =
2730                             SB_ONDISK_JOURNAL_SIZE(p_s_sb) / JOURNAL_MIN_RATIO;
2731                 if (journal->j_trans_max > JOURNAL_TRANS_MAX_DEFAULT / ratio)
2732                         journal->j_trans_max =
2733                             JOURNAL_TRANS_MAX_DEFAULT / ratio;
2734                 if (journal->j_trans_max < JOURNAL_TRANS_MIN_DEFAULT / ratio)
2735                         journal->j_trans_max =
2736                             JOURNAL_TRANS_MIN_DEFAULT / ratio;
2737
2738                 if (journal->j_trans_max != initial)
2739                         reiserfs_warning(p_s_sb,
2740                                          "sh-461: journal_init: wrong transaction max size (%u). Changed to %u",
2741                                          initial, journal->j_trans_max);
2742
2743                 journal->j_max_batch = journal->j_trans_max *
2744                     JOURNAL_MAX_BATCH_DEFAULT / JOURNAL_TRANS_MAX_DEFAULT;
2745         }
2746
2747         if (!journal->j_trans_max) {
2748                 /*we have the file system was created by old version of mkreiserfs
2749                    so this field contains zero value */
2750                 journal->j_trans_max = JOURNAL_TRANS_MAX_DEFAULT;
2751                 journal->j_max_batch = JOURNAL_MAX_BATCH_DEFAULT;
2752                 journal->j_max_commit_age = JOURNAL_MAX_COMMIT_AGE;
2753
2754                 /* for blocksize >= 4096 - max transaction size is 1024. For block size < 4096
2755                    trans max size is decreased proportionally */
2756                 if (p_s_sb->s_blocksize < 4096) {
2757                         journal->j_trans_max /= (4096 / p_s_sb->s_blocksize);
2758                         journal->j_max_batch = (journal->j_trans_max) * 9 / 10;
2759                 }
2760         }
2761
2762         journal->j_default_max_commit_age = journal->j_max_commit_age;
2763
2764         if (commit_max_age != 0) {
2765                 journal->j_max_commit_age = commit_max_age;
2766                 journal->j_max_trans_age = commit_max_age;
2767         }
2768
2769         reiserfs_info(p_s_sb, "journal params: device %s, size %u, "
2770                       "journal first block %u, max trans len %u, max batch %u, "
2771                       "max commit age %u, max trans age %u\n",
2772                       bdevname(journal->j_dev_bd, b),
2773                       SB_ONDISK_JOURNAL_SIZE(p_s_sb),
2774                       SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2775                       journal->j_trans_max,
2776                       journal->j_max_batch,
2777                       journal->j_max_commit_age, journal->j_max_trans_age);
2778
2779         brelse(bhjh);
2780
2781         journal->j_list_bitmap_index = 0;
2782         journal_list_init(p_s_sb);
2783
2784         memset(journal->j_list_hash_table, 0,
2785                JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
2786
2787         INIT_LIST_HEAD(&journal->j_dirty_buffers);
2788         spin_lock_init(&journal->j_dirty_buffers_lock);
2789
2790         journal->j_start = 0;
2791         journal->j_len = 0;
2792         journal->j_len_alloc = 0;
2793         atomic_set(&(journal->j_wcount), 0);
2794         atomic_set(&(journal->j_async_throttle), 0);
2795         journal->j_bcount = 0;
2796         journal->j_trans_start_time = 0;
2797         journal->j_last = NULL;
2798         journal->j_first = NULL;
2799         init_waitqueue_head(&(journal->j_join_wait));
2800         sema_init(&journal->j_lock, 1);
2801         sema_init(&journal->j_flush_sem, 1);
2802
2803         journal->j_trans_id = 10;
2804         journal->j_mount_id = 10;
2805         journal->j_state = 0;
2806         atomic_set(&(journal->j_jlock), 0);
2807         journal->j_cnode_free_list = allocate_cnodes(num_cnodes);
2808         journal->j_cnode_free_orig = journal->j_cnode_free_list;
2809         journal->j_cnode_free = journal->j_cnode_free_list ? num_cnodes : 0;
2810         journal->j_cnode_used = 0;
2811         journal->j_must_wait = 0;
2812
2813         if (journal->j_cnode_free == 0) {
2814                 reiserfs_warning(p_s_sb, "journal-2004: Journal cnode memory "
2815                                  "allocation failed (%ld bytes). Journal is "
2816                                  "too large for available memory. Usually "
2817                                  "this is due to a journal that is too large.",
2818                                  sizeof (struct reiserfs_journal_cnode) * num_cnodes);
2819                 goto free_and_return;
2820         }
2821
2822         init_journal_hash(p_s_sb);
2823         jl = journal->j_current_jl;
2824         jl->j_list_bitmap = get_list_bitmap(p_s_sb, jl);
2825         if (!jl->j_list_bitmap) {
2826                 reiserfs_warning(p_s_sb,
2827                                  "journal-2005, get_list_bitmap failed for journal list 0");
2828                 goto free_and_return;
2829         }
2830         if (journal_read(p_s_sb) < 0) {
2831                 reiserfs_warning(p_s_sb, "Replay Failure, unable to mount");
2832                 goto free_and_return;
2833         }
2834
2835         reiserfs_mounted_fs_count++;
2836         if (reiserfs_mounted_fs_count <= 1)
2837                 commit_wq = create_workqueue("reiserfs");
2838
2839         INIT_DELAYED_WORK(&journal->j_work, flush_async_commits);
2840         journal->j_work_sb = p_s_sb;
2841         return 0;
2842       free_and_return:
2843         free_journal_ram(p_s_sb);
2844         return 1;
2845 }
2846
2847 /*
2848 ** test for a polite end of the current transaction.  Used by file_write, and should
2849 ** be used by delete to make sure they don't write more than can fit inside a single
2850 ** transaction
2851 */
2852 int journal_transaction_should_end(struct reiserfs_transaction_handle *th,
2853                                    int new_alloc)
2854 {
2855         struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2856         time_t now = get_seconds();
2857         /* cannot restart while nested */
2858         BUG_ON(!th->t_trans_id);
2859         if (th->t_refcount > 1)
2860                 return 0;
2861         if (journal->j_must_wait > 0 ||
2862             (journal->j_len_alloc + new_alloc) >= journal->j_max_batch ||
2863             atomic_read(&(journal->j_jlock)) ||
2864             (now - journal->j_trans_start_time) > journal->j_max_trans_age ||
2865             journal->j_cnode_free < (journal->j_trans_max * 3)) {
2866                 return 1;
2867         }
2868         /* protected by the BKL here */
2869         journal->j_len_alloc += new_alloc;
2870         th->t_blocks_allocated += new_alloc ;
2871         return 0;
2872 }
2873
2874 /* this must be called inside a transaction, and requires the
2875 ** kernel_lock to be held
2876 */
2877 void reiserfs_block_writes(struct reiserfs_transaction_handle *th)
2878 {
2879         struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2880         BUG_ON(!th->t_trans_id);
2881         journal->j_must_wait = 1;
2882         set_bit(J_WRITERS_BLOCKED, &journal->j_state);
2883         return;
2884 }
2885
2886 /* this must be called without a transaction started, and does not
2887 ** require BKL
2888 */
2889 void reiserfs_allow_writes(struct super_block *s)
2890 {
2891         struct reiserfs_journal *journal = SB_JOURNAL(s);
2892         clear_bit(J_WRITERS_BLOCKED, &journal->j_state);
2893         wake_up(&journal->j_join_wait);
2894 }
2895
2896 /* this must be called without a transaction started, and does not
2897 ** require BKL
2898 */
2899 void reiserfs_wait_on_write_block(struct super_block *s)
2900 {
2901         struct reiserfs_journal *journal = SB_JOURNAL(s);
2902         wait_event(journal->j_join_wait,
2903                    !test_bit(J_WRITERS_BLOCKED, &journal->j_state));
2904 }
2905
2906 static void queue_log_writer(struct super_block *s)
2907 {
2908         wait_queue_t wait;
2909         struct reiserfs_journal *journal = SB_JOURNAL(s);
2910         set_bit(J_WRITERS_QUEUED, &journal->j_state);
2911
2912         /*
2913          * we don't want to use wait_event here because
2914          * we only want to wait once.
2915          */
2916         init_waitqueue_entry(&wait, current);
2917         add_wait_queue(&journal->j_join_wait, &wait);
2918         set_current_state(TASK_UNINTERRUPTIBLE);
2919         if (test_bit(J_WRITERS_QUEUED, &journal->j_state))
2920                 schedule();
2921         __set_current_state(TASK_RUNNING);
2922         remove_wait_queue(&journal->j_join_wait, &wait);
2923 }
2924
2925 static void wake_queued_writers(struct super_block *s)
2926 {
2927         struct reiserfs_journal *journal = SB_JOURNAL(s);
2928         if (test_and_clear_bit(J_WRITERS_QUEUED, &journal->j_state))
2929                 wake_up(&journal->j_join_wait);
2930 }
2931
2932 static void let_transaction_grow(struct super_block *sb, unsigned long trans_id)
2933 {
2934         struct reiserfs_journal *journal = SB_JOURNAL(sb);
2935         unsigned long bcount = journal->j_bcount;
2936         while (1) {
2937                 schedule_timeout_uninterruptible(1);
2938                 journal->j_current_jl->j_state |= LIST_COMMIT_PENDING;
2939                 while ((atomic_read(&journal->j_wcount) > 0 ||
2940                         atomic_read(&journal->j_jlock)) &&
2941                        journal->j_trans_id == trans_id) {
2942                         queue_log_writer(sb);
2943                 }
2944                 if (journal->j_trans_id != trans_id)
2945                         break;
2946                 if (bcount == journal->j_bcount)
2947                         break;
2948                 bcount = journal->j_bcount;
2949         }
2950 }
2951
2952 /* join == true if you must join an existing transaction.
2953 ** join == false if you can deal with waiting for others to finish
2954 **
2955 ** this will block until the transaction is joinable.  send the number of blocks you
2956 ** expect to use in nblocks.
2957 */
2958 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
2959                               struct super_block *p_s_sb, unsigned long nblocks,
2960                               int join)
2961 {
2962         time_t now = get_seconds();
2963         int old_trans_id;
2964         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
2965         struct reiserfs_transaction_handle myth;
2966         int sched_count = 0;
2967         int retval;
2968
2969         reiserfs_check_lock_depth(p_s_sb, "journal_begin");
2970         BUG_ON(nblocks > journal->j_trans_max);
2971
2972         PROC_INFO_INC(p_s_sb, journal.journal_being);
2973         /* set here for journal_join */
2974         th->t_refcount = 1;
2975         th->t_super = p_s_sb;
2976
2977       relock:
2978         lock_journal(p_s_sb);
2979         if (join != JBEGIN_ABORT && reiserfs_is_journal_aborted(journal)) {
2980                 unlock_journal(p_s_sb);
2981                 retval = journal->j_errno;
2982                 goto out_fail;
2983         }
2984         journal->j_bcount++;
2985
2986         if (test_bit(J_WRITERS_BLOCKED, &journal->j_state)) {
2987                 unlock_journal(p_s_sb);
2988                 reiserfs_wait_on_write_block(p_s_sb);
2989                 PROC_INFO_INC(p_s_sb, journal.journal_relock_writers);
2990                 goto relock;
2991         }
2992         now = get_seconds();
2993
2994         /* if there is no room in the journal OR
2995          ** if this transaction is too old, and we weren't called joinable, wait for it to finish before beginning
2996          ** we don't sleep if there aren't other writers
2997          */
2998
2999         if ((!join && journal->j_must_wait > 0) ||
3000             (!join
3001              && (journal->j_len_alloc + nblocks + 2) >= journal->j_max_batch)
3002             || (!join && atomic_read(&journal->j_wcount) > 0
3003                 && journal->j_trans_start_time > 0
3004                 && (now - journal->j_trans_start_time) >
3005                 journal->j_max_trans_age) || (!join
3006                                               && atomic_read(&journal->j_jlock))
3007             || (!join && journal->j_cnode_free < (journal->j_trans_max * 3))) {
3008
3009                 old_trans_id = journal->j_trans_id;
3010                 unlock_journal(p_s_sb); /* allow others to finish this transaction */
3011
3012                 if (!join && (journal->j_len_alloc + nblocks + 2) >=
3013                     journal->j_max_batch &&
3014                     ((journal->j_len + nblocks + 2) * 100) <
3015                     (journal->j_len_alloc * 75)) {
3016                         if (atomic_read(&journal->j_wcount) > 10) {
3017                                 sched_count++;
3018                                 queue_log_writer(p_s_sb);
3019                                 goto relock;
3020                         }
3021                 }
3022                 /* don't mess with joining the transaction if all we have to do is
3023                  * wait for someone else to do a commit
3024                  */
3025                 if (atomic_read(&journal->j_jlock)) {
3026                         while (journal->j_trans_id == old_trans_id &&
3027                                atomic_read(&journal->j_jlock)) {
3028                                 queue_log_writer(p_s_sb);
3029                         }
3030                         goto relock;
3031                 }
3032                 retval = journal_join(&myth, p_s_sb, 1);
3033                 if (retval)
3034                         goto out_fail;
3035
3036                 /* someone might have ended the transaction while we joined */
3037                 if (old_trans_id != journal->j_trans_id) {
3038                         retval = do_journal_end(&myth, p_s_sb, 1, 0);
3039                 } else {
3040                         retval = do_journal_end(&myth, p_s_sb, 1, COMMIT_NOW);
3041                 }
3042
3043                 if (retval)
3044                         goto out_fail;
3045
3046                 PROC_INFO_INC(p_s_sb, journal.journal_relock_wcount);
3047                 goto relock;
3048         }
3049         /* we are the first writer, set trans_id */
3050         if (journal->j_trans_start_time == 0) {
3051                 journal->j_trans_start_time = get_seconds();
3052         }
3053         atomic_inc(&(journal->j_wcount));
3054         journal->j_len_alloc += nblocks;
3055         th->t_blocks_logged = 0;
3056         th->t_blocks_allocated = nblocks;
3057         th->t_trans_id = journal->j_trans_id;
3058         unlock_journal(p_s_sb);
3059         INIT_LIST_HEAD(&th->t_list);
3060         get_fs_excl();
3061         return 0;
3062
3063       out_fail:
3064         memset(th, 0, sizeof(*th));
3065         /* Re-set th->t_super, so we can properly keep track of how many
3066          * persistent transactions there are. We need to do this so if this
3067          * call is part of a failed restart_transaction, we can free it later */
3068         th->t_super = p_s_sb;
3069         return retval;
3070 }
3071
3072 struct reiserfs_transaction_handle *reiserfs_persistent_transaction(struct
3073                                                                     super_block
3074                                                                     *s,
3075                                                                     int nblocks)
3076 {
3077         int ret;
3078         struct reiserfs_transaction_handle *th;
3079
3080         /* if we're nesting into an existing transaction.  It will be
3081          ** persistent on its own
3082          */
3083         if (reiserfs_transaction_running(s)) {
3084                 th = current->journal_info;
3085                 th->t_refcount++;
3086                 BUG_ON(th->t_refcount < 2);
3087
3088                 return th;
3089         }
3090         th = kmalloc(sizeof(struct reiserfs_transaction_handle), GFP_NOFS);
3091         if (!th)
3092                 return NULL;
3093         ret = journal_begin(th, s, nblocks);
3094         if (ret) {
3095                 kfree(th);
3096                 return NULL;
3097         }
3098
3099         SB_JOURNAL(s)->j_persistent_trans++;
3100         return th;
3101 }
3102
3103 int reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle *th)
3104 {
3105         struct super_block *s = th->t_super;
3106         int ret = 0;
3107         if (th->t_trans_id)
3108                 ret = journal_end(th, th->t_super, th->t_blocks_allocated);
3109         else
3110                 ret = -EIO;
3111         if (th->t_refcount == 0) {
3112                 SB_JOURNAL(s)->j_persistent_trans--;
3113                 kfree(th);
3114         }
3115         return ret;
3116 }
3117
3118 static int journal_join(struct reiserfs_transaction_handle *th,
3119                         struct super_block *p_s_sb, unsigned long nblocks)
3120 {
3121         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3122
3123         /* this keeps do_journal_end from NULLing out the current->journal_info
3124          ** pointer
3125          */
3126         th->t_handle_save = cur_th;
3127         BUG_ON(cur_th && cur_th->t_refcount > 1);
3128         return do_journal_begin_r(th, p_s_sb, nblocks, JBEGIN_JOIN);
3129 }
3130
3131 int journal_join_abort(struct reiserfs_transaction_handle *th,
3132                        struct super_block *p_s_sb, unsigned long nblocks)
3133 {
3134         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3135
3136         /* this keeps do_journal_end from NULLing out the current->journal_info
3137          ** pointer
3138          */
3139         th->t_handle_save = cur_th;
3140         BUG_ON(cur_th && cur_th->t_refcount > 1);
3141         return do_journal_begin_r(th, p_s_sb, nblocks, JBEGIN_ABORT);
3142 }
3143
3144 int journal_begin(struct reiserfs_transaction_handle *th,
3145                   struct super_block *p_s_sb, unsigned long nblocks)
3146 {
3147         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3148         int ret;
3149
3150         th->t_handle_save = NULL;
3151         if (cur_th) {
3152                 /* we are nesting into the current transaction */
3153                 if (cur_th->t_super == p_s_sb) {
3154                         BUG_ON(!cur_th->t_refcount);
3155                         cur_th->t_refcount++;
3156                         memcpy(th, cur_th, sizeof(*th));
3157                         if (th->t_refcount <= 1)
3158                                 reiserfs_warning(p_s_sb,
3159                                                  "BAD: refcount <= 1, but journal_info != 0");
3160                         return 0;
3161                 } else {
3162                         /* we've ended up with a handle from a different filesystem.
3163                          ** save it and restore on journal_end.  This should never
3164                          ** really happen...
3165                          */
3166                         reiserfs_warning(p_s_sb,
3167                                          "clm-2100: nesting info a different FS");
3168                         th->t_handle_save = current->journal_info;
3169                         current->journal_info = th;
3170                 }
3171         } else {
3172                 current->journal_info = th;
3173         }
3174         ret = do_journal_begin_r(th, p_s_sb, nblocks, JBEGIN_REG);
3175         BUG_ON(current->journal_info != th);
3176
3177         /* I guess this boils down to being the reciprocal of clm-2100 above.
3178          * If do_journal_begin_r fails, we need to put it back, since journal_end
3179          * won't be called to do it. */
3180         if (ret)
3181                 current->journal_info = th->t_handle_save;
3182         else
3183                 BUG_ON(!th->t_refcount);
3184
3185         return ret;
3186 }
3187
3188 /*
3189 ** puts bh into the current transaction.  If it was already there, reorders removes the
3190 ** old pointers from the hash, and puts new ones in (to make sure replay happen in the right order).
3191 **
3192 ** if it was dirty, cleans and files onto the clean list.  I can't let it be dirty again until the
3193 ** transaction is committed.
3194 **
3195 ** if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len.
3196 */
3197 int journal_mark_dirty(struct reiserfs_transaction_handle *th,
3198                        struct super_block *p_s_sb, struct buffer_head *bh)
3199 {
3200         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3201         struct reiserfs_journal_cnode *cn = NULL;
3202         int count_already_incd = 0;
3203         int prepared = 0;
3204         BUG_ON(!th->t_trans_id);
3205
3206         PROC_INFO_INC(p_s_sb, journal.mark_dirty);
3207         if (th->t_trans_id != journal->j_trans_id) {
3208                 reiserfs_panic(th->t_super,
3209                                "journal-1577: handle trans id %ld != current trans id %ld\n",
3210                                th->t_trans_id, journal->j_trans_id);
3211         }
3212
3213         p_s_sb->s_dirt = 1;
3214
3215         prepared = test_clear_buffer_journal_prepared(bh);
3216         clear_buffer_journal_restore_dirty(bh);
3217         /* already in this transaction, we are done */
3218         if (buffer_journaled(bh)) {
3219                 PROC_INFO_INC(p_s_sb, journal.mark_dirty_already);
3220                 return 0;
3221         }
3222
3223         /* this must be turned into a panic instead of a warning.  We can't allow
3224          ** a dirty or journal_dirty or locked buffer to be logged, as some changes
3225          ** could get to disk too early.  NOT GOOD.
3226          */
3227         if (!prepared || buffer_dirty(bh)) {
3228                 reiserfs_warning(p_s_sb, "journal-1777: buffer %llu bad state "
3229                                  "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT",
3230                                  (unsigned long long)bh->b_blocknr,
3231                                  prepared ? ' ' : '!',
3232                                  buffer_locked(bh) ? ' ' : '!',
3233                                  buffer_dirty(bh) ? ' ' : '!',
3234                                  buffer_journal_dirty(bh) ? ' ' : '!');
3235         }
3236
3237         if (atomic_read(&(journal->j_wcount)) <= 0) {
3238                 reiserfs_warning(p_s_sb,
3239                                  "journal-1409: journal_mark_dirty returning because j_wcount was %d",
3240                                  atomic_read(&(journal->j_wcount)));
3241                 return 1;
3242         }
3243         /* this error means I've screwed up, and we've overflowed the transaction.
3244          ** Nothing can be done here, except make the FS readonly or panic.
3245          */
3246         if (journal->j_len >= journal->j_trans_max) {
3247                 reiserfs_panic(th->t_super,
3248                                "journal-1413: journal_mark_dirty: j_len (%lu) is too big\n",
3249                                journal->j_len);
3250         }
3251
3252         if (buffer_journal_dirty(bh)) {
3253                 count_already_incd = 1;
3254                 PROC_INFO_INC(p_s_sb, journal.mark_dirty_notjournal);
3255                 clear_buffer_journal_dirty(bh);
3256         }
3257
3258         if (journal->j_len > journal->j_len_alloc) {
3259                 journal->j_len_alloc = journal->j_len + JOURNAL_PER_BALANCE_CNT;
3260         }
3261
3262         set_buffer_journaled(bh);
3263
3264         /* now put this guy on the end */
3265         if (!cn) {
3266                 cn = get_cnode(p_s_sb);
3267                 if (!cn) {
3268                         reiserfs_panic(p_s_sb, "get_cnode failed!\n");
3269                 }
3270
3271                 if (th->t_blocks_logged == th->t_blocks_allocated) {
3272                         th->t_blocks_allocated += JOURNAL_PER_BALANCE_CNT;
3273                         journal->j_len_alloc += JOURNAL_PER_BALANCE_CNT;
3274                 }
3275                 th->t_blocks_logged++;
3276                 journal->j_len++;
3277
3278                 cn->bh = bh;
3279                 cn->blocknr = bh->b_blocknr;
3280                 cn->sb = p_s_sb;
3281                 cn->jlist = NULL;
3282                 insert_journal_hash(journal->j_hash_table, cn);
3283                 if (!count_already_incd) {
3284                         get_bh(bh);
3285                 }
3286         }
3287         cn->next = NULL;
3288         cn->prev = journal->j_last;
3289         cn->bh = bh;
3290         if (journal->j_last) {
3291                 journal->j_last->next = cn;
3292                 journal->j_last = cn;
3293         } else {
3294                 journal->j_first = cn;
3295                 journal->j_last = cn;
3296         }
3297         return 0;
3298 }
3299
3300 int journal_end(struct reiserfs_transaction_handle *th,
3301                 struct super_block *p_s_sb, unsigned long nblocks)
3302 {
3303         if (!current->journal_info && th->t_refcount > 1)
3304                 reiserfs_warning(p_s_sb, "REISER-NESTING: th NULL, refcount %d",
3305                                  th->t_refcount);
3306
3307         if (!th->t_trans_id) {
3308                 WARN_ON(1);
3309                 return -EIO;
3310         }
3311
3312         th->t_refcount--;
3313         if (th->t_refcount > 0) {
3314                 struct reiserfs_transaction_handle *cur_th =
3315                     current->journal_info;
3316
3317                 /* we aren't allowed to close a nested transaction on a different
3318                  ** filesystem from the one in the task struct
3319                  */
3320                 BUG_ON(cur_th->t_super != th->t_super);
3321
3322                 if (th != cur_th) {
3323                         memcpy(current->journal_info, th, sizeof(*th));
3324                         th->t_trans_id = 0;
3325                 }
3326                 return 0;
3327         } else {
3328                 return do_journal_end(th, p_s_sb, nblocks, 0);
3329         }
3330 }
3331
3332 /* removes from the current transaction, relsing and descrementing any counters.
3333 ** also files the removed buffer directly onto the clean list
3334 **
3335 ** called by journal_mark_freed when a block has been deleted
3336 **
3337 ** returns 1 if it cleaned and relsed the buffer. 0 otherwise
3338 */
3339 static int remove_from_transaction(struct super_block *p_s_sb,
3340                                    b_blocknr_t blocknr, int already_cleaned)
3341 {
3342         struct buffer_head *bh;
3343         struct reiserfs_journal_cnode *cn;
3344         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3345         int ret = 0;
3346
3347         cn = get_journal_hash_dev(p_s_sb, journal->j_hash_table, blocknr);
3348         if (!cn || !cn->bh) {
3349                 return ret;
3350         }
3351         bh = cn->bh;
3352         if (cn->prev) {
3353                 cn->prev->next = cn->next;
3354         }
3355         if (cn->next) {
3356                 cn->next->prev = cn->prev;
3357         }
3358         if (cn == journal->j_first) {
3359                 journal->j_first = cn->next;
3360         }
3361         if (cn == journal->j_last) {
3362                 journal->j_last = cn->prev;
3363         }
3364         if (bh)
3365                 remove_journal_hash(p_s_sb, journal->j_hash_table, NULL,
3366                                     bh->b_blocknr, 0);
3367         clear_buffer_journaled(bh);     /* don't log this one */
3368
3369         if (!already_cleaned) {
3370                 clear_buffer_journal_dirty(bh);
3371                 clear_buffer_dirty(bh);
3372                 clear_buffer_journal_test(bh);
3373                 put_bh(bh);
3374                 if (atomic_read(&(bh->b_count)) < 0) {
3375                         reiserfs_warning(p_s_sb,
3376                                          "journal-1752: remove from trans, b_count < 0");
3377                 }
3378                 ret = 1;
3379         }
3380         journal->j_len--;
3381         journal->j_len_alloc--;
3382         free_cnode(p_s_sb, cn);
3383         return ret;
3384 }
3385
3386 /*
3387 ** for any cnode in a journal list, it can only be dirtied of all the
3388 ** transactions that include it are committed to disk.
3389 ** this checks through each transaction, and returns 1 if you are allowed to dirty,
3390 ** and 0 if you aren't
3391 **
3392 ** it is called by dirty_journal_list, which is called after flush_commit_list has gotten all the log
3393 ** blocks for a given transaction on disk
3394 **
3395 */
3396 static int can_dirty(struct reiserfs_journal_cnode *cn)
3397 {
3398         struct super_block *sb = cn->sb;
3399         b_blocknr_t blocknr = cn->blocknr;
3400         struct reiserfs_journal_cnode *cur = cn->hprev;
3401         int can_dirty = 1;
3402
3403         /* first test hprev.  These are all newer than cn, so any node here
3404          ** with the same block number and dev means this node can't be sent
3405          ** to disk right now.
3406          */
3407         while (cur && can_dirty) {
3408                 if (cur->jlist && cur->bh && cur->blocknr && cur->sb == sb &&
3409                     cur->blocknr == blocknr) {
3410                         can_dirty = 0;
3411                 }
3412                 cur = cur->hprev;
3413         }
3414         /* then test hnext.  These are all older than cn.  As long as they
3415          ** are committed to the log, it is safe to write cn to disk
3416          */
3417         cur = cn->hnext;
3418         while (cur && can_dirty) {
3419                 if (cur->jlist && cur->jlist->j_len > 0 &&
3420                     atomic_read(&(cur->jlist->j_commit_left)) > 0 && cur->bh &&
3421                     cur->blocknr && cur->sb == sb && cur->blocknr == blocknr) {
3422                         can_dirty = 0;
3423                 }
3424                 cur = cur->hnext;
3425         }
3426         return can_dirty;
3427 }
3428
3429 /* syncs the commit blocks, but does not force the real buffers to disk
3430 ** will wait until the current transaction is done/committed before returning
3431 */
3432 int journal_end_sync(struct reiserfs_transaction_handle *th,
3433                      struct super_block *p_s_sb, unsigned long nblocks)
3434 {
3435         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3436
3437         BUG_ON(!th->t_trans_id);
3438         /* you can sync while nested, very, very bad */
3439         BUG_ON(th->t_refcount > 1);
3440         if (journal->j_len == 0) {
3441                 reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb),
3442                                              1);
3443                 journal_mark_dirty(th, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb));
3444         }
3445         return do_journal_end(th, p_s_sb, nblocks, COMMIT_NOW | WAIT);
3446 }
3447
3448 /*
3449 ** writeback the pending async commits to disk
3450 */
3451 static void flush_async_commits(struct work_struct *work)
3452 {
3453         struct reiserfs_journal *journal =
3454                 container_of(work, struct reiserfs_journal, j_work.work);
3455         struct super_block *p_s_sb = journal->j_work_sb;
3456         struct reiserfs_journal_list *jl;
3457         struct list_head *entry;
3458
3459         lock_kernel();
3460         if (!list_empty(&journal->j_journal_list)) {
3461                 /* last entry is the youngest, commit it and you get everything */
3462                 entry = journal->j_journal_list.prev;
3463                 jl = JOURNAL_LIST_ENTRY(entry);
3464                 flush_commit_list(p_s_sb, jl, 1);
3465         }
3466         unlock_kernel();
3467 }
3468
3469 /*
3470 ** flushes any old transactions to disk
3471 ** ends the current transaction if it is too old
3472 */
3473 int reiserfs_flush_old_commits(struct super_block *p_s_sb)
3474 {
3475         time_t now;
3476         struct reiserfs_transaction_handle th;
3477         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3478
3479         now = get_seconds();
3480         /* safety check so we don't flush while we are replaying the log during
3481          * mount
3482          */
3483         if (list_empty(&journal->j_journal_list)) {
3484                 return 0;
3485         }
3486
3487         /* check the current transaction.  If there are no writers, and it is
3488          * too old, finish it, and force the commit blocks to disk
3489          */
3490         if (atomic_read(&journal->j_wcount) <= 0 &&
3491             journal->j_trans_start_time > 0 &&
3492             journal->j_len > 0 &&
3493             (now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3494                 if (!journal_join(&th, p_s_sb, 1)) {
3495                         reiserfs_prepare_for_journal(p_s_sb,
3496                                                      SB_BUFFER_WITH_SB(p_s_sb),
3497                                                      1);
3498                         journal_mark_dirty(&th, p_s_sb,
3499                                            SB_BUFFER_WITH_SB(p_s_sb));
3500
3501                         /* we're only being called from kreiserfsd, it makes no sense to do
3502                          ** an async commit so that kreiserfsd can do it later
3503                          */
3504                         do_journal_end(&th, p_s_sb, 1, COMMIT_NOW | WAIT);
3505                 }
3506         }
3507         return p_s_sb->s_dirt;
3508 }
3509
3510 /*
3511 ** returns 0 if do_journal_end should return right away, returns 1 if do_journal_end should finish the commit
3512 **
3513 ** if the current transaction is too old, but still has writers, this will wait on j_join_wait until all
3514 ** the writers are done.  By the time it wakes up, the transaction it was called has already ended, so it just
3515 ** flushes the commit list and returns 0.
3516 **
3517 ** Won't batch when flush or commit_now is set.  Also won't batch when others are waiting on j_join_wait.
3518 **
3519 ** Note, we can't allow the journal_end to proceed while there are still writers in the log.
3520 */
3521 static int check_journal_end(struct reiserfs_transaction_handle *th,
3522                              struct super_block *p_s_sb, unsigned long nblocks,
3523                              int flags)
3524 {
3525
3526         time_t now;
3527         int flush = flags & FLUSH_ALL;
3528         int commit_now = flags & COMMIT_NOW;
3529         int wait_on_commit = flags & WAIT;
3530         struct reiserfs_journal_list *jl;
3531         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3532
3533         BUG_ON(!th->t_trans_id);
3534
3535         if (th->t_trans_id != journal->j_trans_id) {
3536                 reiserfs_panic(th->t_super,
3537                                "journal-1577: handle trans id %ld != current trans id %ld\n",
3538                                th->t_trans_id, journal->j_trans_id);
3539         }
3540
3541         journal->j_len_alloc -= (th->t_blocks_allocated - th->t_blocks_logged);
3542         if (atomic_read(&(journal->j_wcount)) > 0) {    /* <= 0 is allowed.  unmounting might not call begin */
3543                 atomic_dec(&(journal->j_wcount));
3544         }
3545
3546         /* BUG, deal with case where j_len is 0, but people previously freed blocks need to be released
3547          ** will be dealt with by next transaction that actually writes something, but should be taken
3548          ** care of in this trans
3549          */
3550         BUG_ON(journal->j_len == 0);
3551
3552         /* if wcount > 0, and we are called to with flush or commit_now,
3553          ** we wait on j_join_wait.  We will wake up when the last writer has
3554          ** finished the transaction, and started it on its way to the disk.
3555          ** Then, we flush the commit or journal list, and just return 0
3556          ** because the rest of journal end was already done for this transaction.
3557          */
3558         if (atomic_read(&(journal->j_wcount)) > 0) {
3559                 if (flush || commit_now) {
3560                         unsigned trans_id;
3561
3562                         jl = journal->j_current_jl;
3563                         trans_id = jl->j_trans_id;
3564                         if (wait_on_commit)
3565                                 jl->j_state |= LIST_COMMIT_PENDING;
3566                         atomic_set(&(journal->j_jlock), 1);
3567                         if (flush) {
3568                                 journal->j_next_full_flush = 1;
3569                         }
3570                         unlock_journal(p_s_sb);
3571
3572                         /* sleep while the current transaction is still j_jlocked */
3573                         while (journal->j_trans_id == trans_id) {
3574                                 if (atomic_read(&journal->j_jlock)) {
3575                                         queue_log_writer(p_s_sb);
3576                                 } else {
3577                                         lock_journal(p_s_sb);
3578                                         if (journal->j_trans_id == trans_id) {
3579                                                 atomic_set(&(journal->j_jlock),
3580                                                            1);
3581                                         }
3582                                         unlock_journal(p_s_sb);
3583                                 }
3584                         }
3585                         BUG_ON(journal->j_trans_id == trans_id);
3586
3587                         if (commit_now
3588                             && journal_list_still_alive(p_s_sb, trans_id)
3589                             && wait_on_commit) {
3590                                 flush_commit_list(p_s_sb, jl, 1);
3591                         }
3592                         return 0;
3593                 }
3594                 unlock_journal(p_s_sb);
3595                 return 0;
3596         }
3597
3598         /* deal with old transactions where we are the last writers */
3599         now = get_seconds();
3600         if ((now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3601                 commit_now = 1;
3602                 journal->j_next_async_flush = 1;
3603         }
3604         /* don't batch when someone is waiting on j_join_wait */
3605         /* don't batch when syncing the commit or flushing the whole trans */
3606         if (!(journal->j_must_wait > 0) && !(atomic_read(&(journal->j_jlock)))
3607             && !flush && !commit_now && (journal->j_len < journal->j_max_batch)
3608             && journal->j_len_alloc < journal->j_max_batch
3609             && journal->j_cnode_free > (journal->j_trans_max * 3)) {
3610                 journal->j_bcount++;
3611                 unlock_journal(p_s_sb);
3612                 return 0;
3613         }
3614
3615         if (journal->j_start > SB_ONDISK_JOURNAL_SIZE(p_s_sb)) {
3616                 reiserfs_panic(p_s_sb,
3617                                "journal-003: journal_end: j_start (%ld) is too high\n",
3618                                journal->j_start);
3619         }
3620         return 1;
3621 }
3622
3623 /*
3624 ** Does all the work that makes deleting blocks safe.
3625 ** when deleting a block mark BH_JNew, just remove it from the current transaction, clean it's buffer_head and move on.
3626 **
3627 ** otherwise:
3628 ** set a bit for the block in the journal bitmap.  That will prevent it from being allocated for unformatted nodes
3629 ** before this transaction has finished.
3630 **
3631 ** mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers.  That will prevent any old transactions with
3632 ** this block from trying to flush to the real location.  Since we aren't removing the cnode from the journal_list_hash,
3633 ** the block can't be reallocated yet.
3634 **
3635 ** Then remove it from the current transaction, decrementing any counters and filing it on the clean list.
3636 */
3637 int journal_mark_freed(struct reiserfs_transaction_handle *th,
3638                        struct super_block *p_s_sb, b_blocknr_t blocknr)
3639 {
3640         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3641         struct reiserfs_journal_cnode *cn = NULL;
3642         struct buffer_head *bh = NULL;
3643         struct reiserfs_list_bitmap *jb = NULL;
3644         int cleaned = 0;
3645         BUG_ON(!th->t_trans_id);
3646
3647         cn = get_journal_hash_dev(p_s_sb, journal->j_hash_table, blocknr);
3648         if (cn && cn->bh) {
3649                 bh = cn->bh;
3650                 get_bh(bh);
3651         }
3652         /* if it is journal new, we just remove it from this transaction */
3653         if (bh && buffer_journal_new(bh)) {
3654                 clear_buffer_journal_new(bh);
3655                 clear_prepared_bits(bh);
3656                 reiserfs_clean_and_file_buffer(bh);
3657                 cleaned = remove_from_transaction(p_s_sb, blocknr, cleaned);
3658         } else {
3659                 /* set the bit for this block in the journal bitmap for this transaction */
3660                 jb = journal->j_current_jl->j_list_bitmap;
3661                 if (!jb) {
3662                         reiserfs_panic(p_s_sb,
3663                                        "journal-1702: journal_mark_freed, journal_list_bitmap is NULL\n");
3664                 }
3665                 set_bit_in_list_bitmap(p_s_sb, blocknr, jb);
3666
3667                 /* Note, the entire while loop is not allowed to schedule.  */
3668
3669                 if (bh) {
3670                         clear_prepared_bits(bh);
3671                         reiserfs_clean_and_file_buffer(bh);
3672                 }
3673                 cleaned = remove_from_transaction(p_s_sb, blocknr, cleaned);
3674
3675                 /* find all older transactions with this block, make sure they don't try to write it out */
3676                 cn = get_journal_hash_dev(p_s_sb, journal->j_list_hash_table,
3677                                           blocknr);
3678                 while (cn) {
3679                         if (p_s_sb == cn->sb && blocknr == cn->blocknr) {
3680                                 set_bit(BLOCK_FREED, &cn->state);
3681                                 if (cn->bh) {
3682                                         if (!cleaned) {
3683                                                 /* remove_from_transaction will brelse the buffer if it was
3684                                                  ** in the current trans
3685                                                  */
3686                                                 clear_buffer_journal_dirty(cn->
3687                                                                            bh);
3688                                                 clear_buffer_dirty(cn->bh);
3689                                                 clear_buffer_journal_test(cn->
3690                                                                           bh);
3691                                                 cleaned = 1;
3692                                                 put_bh(cn->bh);
3693                                                 if (atomic_read
3694                                                     (&(cn->bh->b_count)) < 0) {
3695                                                         reiserfs_warning(p_s_sb,
3696                                                                          "journal-2138: cn->bh->b_count < 0");
3697                                                 }
3698                                         }
3699                                         if (cn->jlist) {        /* since we are clearing the bh, we MUST dec nonzerolen */
3700                                                 atomic_dec(&
3701                                                            (cn->jlist->
3702                                                             j_nonzerolen));
3703                                         }
3704                                         cn->bh = NULL;
3705                                 }
3706                         }
3707                         cn = cn->hnext;
3708                 }
3709         }
3710
3711         if (bh) {
3712                 put_bh(bh);     /* get_hash grabs the buffer */
3713                 if (atomic_read(&(bh->b_count)) < 0) {
3714                         reiserfs_warning(p_s_sb,
3715                                          "journal-2165: bh->b_count < 0");
3716                 }
3717         }
3718         return 0;
3719 }
3720
3721 void reiserfs_update_inode_transaction(struct inode *inode)
3722 {
3723         struct reiserfs_journal *journal = SB_JOURNAL(inode->i_sb);
3724         REISERFS_I(inode)->i_jl = journal->j_current_jl;
3725         REISERFS_I(inode)->i_trans_id = journal->j_trans_id;
3726 }
3727
3728 /*
3729  * returns -1 on error, 0 if no commits/barriers were done and 1
3730  * if a transaction was actually committed and the barrier was done
3731  */
3732 static int __commit_trans_jl(struct inode *inode, unsigned long id,
3733                              struct reiserfs_journal_list *jl)
3734 {
3735         struct reiserfs_transaction_handle th;
3736         struct super_block *sb = inode->i_sb;
3737         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3738         int ret = 0;
3739
3740         /* is it from the current transaction, or from an unknown transaction? */
3741         if (id == journal->j_trans_id) {
3742                 jl = journal->j_current_jl;
3743                 /* try to let other writers come in and grow this transaction */
3744                 let_transaction_grow(sb, id);
3745                 if (journal->j_trans_id != id) {
3746                         goto flush_commit_only;
3747                 }
3748
3749                 ret = journal_begin(&th, sb, 1);
3750                 if (ret)
3751                         return ret;
3752
3753                 /* someone might have ended this transaction while we joined */
3754                 if (journal->j_trans_id != id) {
3755                         reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3756                                                      1);
3757                         journal_mark_dirty(&th, sb, SB_BUFFER_WITH_SB(sb));
3758                         ret = journal_end(&th, sb, 1);
3759                         goto flush_commit_only;
3760                 }
3761
3762                 ret = journal_end_sync(&th, sb, 1);
3763                 if (!ret)
3764                         ret = 1;
3765
3766         } else {
3767                 /* this gets tricky, we have to make sure the journal list in
3768                  * the inode still exists.  We know the list is still around
3769                  * if we've got a larger transaction id than the oldest list
3770                  */
3771               flush_commit_only:
3772                 if (journal_list_still_alive(inode->i_sb, id)) {
3773                         /*
3774                          * we only set ret to 1 when we know for sure
3775                          * the barrier hasn't been started yet on the commit
3776                          * block.
3777                          */
3778                         if (atomic_read(&jl->j_commit_left) > 1)
3779                                 ret = 1;
3780                         flush_commit_list(sb, jl, 1);
3781                         if (journal->j_errno)
3782                                 ret = journal->j_errno;
3783                 }
3784         }
3785         /* otherwise the list is gone, and long since committed */
3786         return ret;
3787 }
3788
3789 int reiserfs_commit_for_inode(struct inode *inode)
3790 {
3791         unsigned long id = REISERFS_I(inode)->i_trans_id;
3792         struct reiserfs_journal_list *jl = REISERFS_I(inode)->i_jl;
3793
3794         /* for the whole inode, assume unset id means it was
3795          * changed in the current transaction.  More conservative
3796          */
3797         if (!id || !jl) {
3798                 reiserfs_update_inode_transaction(inode);
3799                 id = REISERFS_I(inode)->i_trans_id;
3800                 /* jl will be updated in __commit_trans_jl */
3801         }
3802
3803         return __commit_trans_jl(inode, id, jl);
3804 }
3805
3806 void reiserfs_restore_prepared_buffer(struct super_block *p_s_sb,
3807                                       struct buffer_head *bh)
3808 {
3809         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3810         PROC_INFO_INC(p_s_sb, journal.restore_prepared);
3811         if (!bh) {
3812                 return;
3813         }
3814         if (test_clear_buffer_journal_restore_dirty(bh) &&
3815             buffer_journal_dirty(bh)) {
3816                 struct reiserfs_journal_cnode *cn;
3817                 cn = get_journal_hash_dev(p_s_sb,
3818                                           journal->j_list_hash_table,
3819                                           bh->b_blocknr);
3820                 if (cn && can_dirty(cn)) {
3821                         set_buffer_journal_test(bh);
3822                         mark_buffer_dirty(bh);
3823                 }
3824         }
3825         clear_buffer_journal_prepared(bh);
3826 }
3827
3828 extern struct tree_balance *cur_tb;
3829 /*
3830 ** before we can change a metadata block, we have to make sure it won't
3831 ** be written to disk while we are altering it.  So, we must:
3832 ** clean it
3833 ** wait on it.
3834 **
3835 */
3836 int reiserfs_prepare_for_journal(struct super_block *p_s_sb,
3837                                  struct buffer_head *bh, int wait)
3838 {
3839         PROC_INFO_INC(p_s_sb, journal.prepare);
3840
3841         if (test_set_buffer_locked(bh)) {
3842                 if (!wait)
3843                         return 0;
3844                 lock_buffer(bh);
3845         }
3846         set_buffer_journal_prepared(bh);
3847         if (test_clear_buffer_dirty(bh) && buffer_journal_dirty(bh)) {
3848                 clear_buffer_journal_test(bh);
3849                 set_buffer_journal_restore_dirty(bh);
3850         }
3851         unlock_buffer(bh);
3852         return 1;
3853 }
3854
3855 static void flush_old_journal_lists(struct super_block *s)
3856 {
3857         struct reiserfs_journal *journal = SB_JOURNAL(s);
3858         struct reiserfs_journal_list *jl;
3859         struct list_head *entry;
3860         time_t now = get_seconds();
3861
3862         while (!list_empty(&journal->j_journal_list)) {
3863                 entry = journal->j_journal_list.next;
3864                 jl = JOURNAL_LIST_ENTRY(entry);
3865                 /* this check should always be run, to send old lists to disk */
3866                 if (jl->j_timestamp < (now - (JOURNAL_MAX_TRANS_AGE * 4)) &&
3867                     atomic_read(&jl->j_commit_left) == 0 &&
3868                     test_transaction(s, jl)) {
3869                         flush_used_journal_lists(s, jl);
3870                 } else {
3871                         break;
3872                 }
3873         }
3874 }
3875
3876 /*
3877 ** long and ugly.  If flush, will not return until all commit
3878 ** blocks and all real buffers in the trans are on disk.
3879 ** If no_async, won't return until all commit blocks are on disk.
3880 **
3881 ** keep reading, there are comments as you go along
3882 **
3883 ** If the journal is aborted, we just clean up. Things like flushing
3884 ** journal lists, etc just won't happen.
3885 */
3886 static int do_journal_end(struct reiserfs_transaction_handle *th,
3887                           struct super_block *p_s_sb, unsigned long nblocks,
3888                           int flags)
3889 {
3890         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3891         struct reiserfs_journal_cnode *cn, *next, *jl_cn;
3892         struct reiserfs_journal_cnode *last_cn = NULL;
3893         struct reiserfs_journal_desc *desc;
3894         struct reiserfs_journal_commit *commit;
3895         struct buffer_head *c_bh;       /* commit bh */
3896         struct buffer_head *d_bh;       /* desc bh */
3897         int cur_write_start = 0;        /* start index of current log write */
3898         int old_start;
3899         int i;
3900         int flush;
3901         int wait_on_commit;
3902         struct reiserfs_journal_list *jl, *temp_jl;
3903         struct list_head *entry, *safe;
3904         unsigned long jindex;
3905         unsigned long commit_trans_id;
3906         int trans_half;
3907
3908         BUG_ON(th->t_refcount > 1);
3909         BUG_ON(!th->t_trans_id);
3910
3911         /* protect flush_older_commits from doing mistakes if the
3912            transaction ID counter gets overflowed.  */
3913         if (th->t_trans_id == ~0UL)
3914                 flags |= FLUSH_ALL | COMMIT_NOW | WAIT;
3915         flush = flags & FLUSH_ALL;
3916         wait_on_commit = flags & WAIT;
3917
3918         put_fs_excl();
3919         current->journal_info = th->t_handle_save;
3920         reiserfs_check_lock_depth(p_s_sb, "journal end");
3921         if (journal->j_len == 0) {
3922                 reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb),
3923                                              1);
3924                 journal_mark_dirty(th, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb));
3925         }
3926
3927         lock_journal(p_s_sb);
3928         if (journal->j_next_full_flush) {
3929                 flags |= FLUSH_ALL;
3930                 flush = 1;
3931         }
3932         if (journal->j_next_async_flush) {
3933                 flags |= COMMIT_NOW | WAIT;
3934                 wait_on_commit = 1;
3935         }
3936
3937         /* check_journal_end locks the journal, and unlocks if it does not return 1
3938          ** it tells us if we should continue with the journal_end, or just return
3939          */
3940         if (!check_journal_end(th, p_s_sb, nblocks, flags)) {
3941                 p_s_sb->s_dirt = 1;
3942                 wake_queued_writers(p_s_sb);
3943                 reiserfs_async_progress_wait(p_s_sb);
3944                 goto out;
3945         }
3946
3947         /* check_journal_end might set these, check again */
3948         if (journal->j_next_full_flush) {
3949                 flush = 1;
3950         }
3951
3952         /*
3953          ** j must wait means we have to flush the log blocks, and the real blocks for
3954          ** this transaction
3955          */
3956         if (journal->j_must_wait > 0) {
3957                 flush = 1;
3958         }
3959 #ifdef REISERFS_PREALLOCATE
3960         /* quota ops might need to nest, setup the journal_info pointer for them
3961          * and raise the refcount so that it is > 0. */
3962         current->journal_info = th;
3963         th->t_refcount++;
3964         reiserfs_discard_all_prealloc(th);      /* it should not involve new blocks into
3965                                                  * the transaction */
3966         th->t_refcount--;
3967         current->journal_info = th->t_handle_save;
3968 #endif
3969
3970         /* setup description block */
3971         d_bh =
3972             journal_getblk(p_s_sb,
3973                            SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
3974                            journal->j_start);
3975         set_buffer_uptodate(d_bh);
3976         desc = (struct reiserfs_journal_desc *)(d_bh)->b_data;
3977         memset(d_bh->b_data, 0, d_bh->b_size);
3978         memcpy(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8);
3979         set_desc_trans_id(desc, journal->j_trans_id);
3980
3981         /* setup commit block.  Don't write (keep it clean too) this one until after everyone else is written */
3982         c_bh = journal_getblk(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
3983                               ((journal->j_start + journal->j_len +
3984                                 1) % SB_ONDISK_JOURNAL_SIZE(p_s_sb)));
3985         commit = (struct reiserfs_journal_commit *)c_bh->b_data;
3986         memset(c_bh->b_data, 0, c_bh->b_size);
3987         set_commit_trans_id(commit, journal->j_trans_id);
3988         set_buffer_uptodate(c_bh);
3989
3990         /* init this journal list */
3991         jl = journal->j_current_jl;
3992
3993         /* we lock the commit before doing anything because
3994          * we want to make sure nobody tries to run flush_commit_list until
3995          * the new transaction is fully setup, and we've already flushed the
3996          * ordered bh list
3997          */
3998         down(&jl->j_commit_lock);
3999
4000         /* save the transaction id in case we need to commit it later */
4001         commit_trans_id = jl->j_trans_id;
4002
4003         atomic_set(&jl->j_older_commits_done, 0);
4004         jl->j_trans_id = journal->j_trans_id;
4005         jl->j_timestamp = journal->j_trans_start_time;
4006         jl->j_commit_bh = c_bh;
4007         jl->j_start = journal->j_start;
4008         jl->j_len = journal->j_len;
4009         atomic_set(&jl->j_nonzerolen, journal->j_len);
4010         atomic_set(&jl->j_commit_left, journal->j_len + 2);
4011         jl->j_realblock = NULL;
4012
4013         /* The ENTIRE FOR LOOP MUST not cause schedule to occur.
4014          **  for each real block, add it to the journal list hash,
4015          ** copy into real block index array in the commit or desc block
4016          */
4017         trans_half = journal_trans_half(p_s_sb->s_blocksize);
4018         for (i = 0, cn = journal->j_first; cn; cn = cn->next, i++) {
4019                 if (buffer_journaled(cn->bh)) {
4020                         jl_cn = get_cnode(p_s_sb);
4021                         if (!jl_cn) {
4022                                 reiserfs_panic(p_s_sb,
4023                                                "journal-1676, get_cnode returned NULL\n");
4024                         }
4025                         if (i == 0) {
4026                                 jl->j_realblock = jl_cn;
4027                         }
4028                         jl_cn->prev = last_cn;
4029                         jl_cn->next = NULL;
4030                         if (last_cn) {
4031                                 last_cn->next = jl_cn;
4032                         }
4033                         last_cn = jl_cn;
4034                         /* make sure the block we are trying to log is not a block
4035                            of journal or reserved area */
4036
4037                         if (is_block_in_log_or_reserved_area
4038                             (p_s_sb, cn->bh->b_blocknr)) {
4039                                 reiserfs_panic(p_s_sb,
4040                                                "journal-2332: Trying to log block %lu, which is a log block\n",
4041                                                cn->bh->b_blocknr);
4042                         }
4043                         jl_cn->blocknr = cn->bh->b_blocknr;
4044                         jl_cn->state = 0;
4045                         jl_cn->sb = p_s_sb;
4046                         jl_cn->bh = cn->bh;
4047                         jl_cn->jlist = jl;
4048                         insert_journal_hash(journal->j_list_hash_table, jl_cn);
4049                         if (i < trans_half) {
4050                                 desc->j_realblock[i] =
4051                                     cpu_to_le32(cn->bh->b_blocknr);
4052                         } else {
4053                                 commit->j_realblock[i - trans_half] =
4054                                     cpu_to_le32(cn->bh->b_blocknr);
4055                         }
4056                 } else {
4057                         i--;
4058                 }
4059         }
4060         set_desc_trans_len(desc, journal->j_len);
4061         set_desc_mount_id(desc, journal->j_mount_id);
4062         set_desc_trans_id(desc, journal->j_trans_id);
4063         set_commit_trans_len(commit, journal->j_len);
4064
4065         /* special check in case all buffers in the journal were marked for not logging */
4066         BUG_ON(journal->j_len == 0);
4067
4068         /* we're about to dirty all the log blocks, mark the description block
4069          * dirty now too.  Don't mark the commit block dirty until all the
4070          * others are on disk
4071          */
4072         mark_buffer_dirty(d_bh);
4073
4074         /* first data block is j_start + 1, so add one to cur_write_start wherever you use it */
4075         cur_write_start = journal->j_start;
4076         cn = journal->j_first;
4077         jindex = 1;             /* start at one so we don't get the desc again */
4078         while (cn) {
4079                 clear_buffer_journal_new(cn->bh);
4080                 /* copy all the real blocks into log area.  dirty log blocks */
4081                 if (buffer_journaled(cn->bh)) {
4082                         struct buffer_head *tmp_bh;
4083                         char *addr;
4084                         struct page *page;
4085                         tmp_bh =
4086                             journal_getblk(p_s_sb,
4087                                            SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
4088                                            ((cur_write_start +
4089                                              jindex) %
4090                                             SB_ONDISK_JOURNAL_SIZE(p_s_sb)));
4091                         set_buffer_uptodate(tmp_bh);
4092                         page = cn->bh->b_page;
4093                         addr = kmap(page);
4094                         memcpy(tmp_bh->b_data,
4095                                addr + offset_in_page(cn->bh->b_data),
4096                                cn->bh->b_size);
4097                         kunmap(page);
4098                         mark_buffer_dirty(tmp_bh);
4099                         jindex++;
4100                         set_buffer_journal_dirty(cn->bh);
4101                         clear_buffer_journaled(cn->bh);
4102                 } else {
4103                         /* JDirty cleared sometime during transaction.  don't log this one */
4104                         reiserfs_warning(p_s_sb,
4105                                          "journal-2048: do_journal_end: BAD, buffer in journal hash, but not JDirty!");
4106                         brelse(cn->bh);
4107                 }
4108                 next = cn->next;
4109                 free_cnode(p_s_sb, cn);
4110                 cn = next;
4111                 cond_resched();
4112         }
4113
4114         /* we are done  with both the c_bh and d_bh, but
4115          ** c_bh must be written after all other commit blocks,
4116          ** so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1.
4117          */
4118
4119         journal->j_current_jl = alloc_journal_list(p_s_sb);
4120
4121         /* now it is safe to insert this transaction on the main list */
4122         list_add_tail(&jl->j_list, &journal->j_journal_list);
4123         list_add_tail(&jl->j_working_list, &journal->j_working_list);
4124         journal->j_num_work_lists++;
4125
4126         /* reset journal values for the next transaction */
4127         old_start = journal->j_start;
4128         journal->j_start =
4129             (journal->j_start + journal->j_len +
4130              2) % SB_ONDISK_JOURNAL_SIZE(p_s_sb);
4131         atomic_set(&(journal->j_wcount), 0);
4132         journal->j_bcount = 0;
4133         journal->j_last = NULL;
4134         journal->j_first = NULL;
4135         journal->j_len = 0;
4136         journal->j_trans_start_time = 0;
4137         /* check for trans_id overflow */
4138         if (++journal->j_trans_id == 0)
4139                 journal->j_trans_id = 10;
4140         journal->j_current_jl->j_trans_id = journal->j_trans_id;
4141         journal->j_must_wait = 0;
4142         journal->j_len_alloc = 0;
4143         journal->j_next_full_flush = 0;
4144         journal->j_next_async_flush = 0;
4145         init_journal_hash(p_s_sb);
4146
4147         // make sure reiserfs_add_jh sees the new current_jl before we
4148         // write out the tails
4149         smp_mb();
4150
4151         /* tail conversion targets have to hit the disk before we end the
4152          * transaction.  Otherwise a later transaction might repack the tail
4153          * before this transaction commits, leaving the data block unflushed and
4154          * clean, if we crash before the later transaction commits, the data block
4155          * is lost.
4156          */
4157         if (!list_empty(&jl->j_tail_bh_list)) {
4158                 unlock_kernel();
4159                 write_ordered_buffers(&journal->j_dirty_buffers_lock,
4160                                       journal, jl, &jl->j_tail_bh_list);
4161                 lock_kernel();
4162         }
4163         BUG_ON(!list_empty(&jl->j_tail_bh_list));
4164         up(&jl->j_commit_lock);
4165
4166         /* honor the flush wishes from the caller, simple commits can
4167          ** be done outside the journal lock, they are done below
4168          **
4169          ** if we don't flush the commit list right now, we put it into
4170          ** the work queue so the people waiting on the async progress work
4171          ** queue don't wait for this proc to flush journal lists and such.
4172          */
4173         if (flush) {
4174                 flush_commit_list(p_s_sb, jl, 1);
4175                 flush_journal_list(p_s_sb, jl, 1);
4176         } else if (!(jl->j_state & LIST_COMMIT_PENDING))
4177                 queue_delayed_work(commit_wq, &journal->j_work, HZ / 10);
4178
4179         /* if the next transaction has any chance of wrapping, flush
4180          ** transactions that might get overwritten.  If any journal lists are very
4181          ** old flush them as well.
4182          */
4183       first_jl:
4184         list_for_each_safe(entry, safe, &journal->j_journal_list) {
4185                 temp_jl = JOURNAL_LIST_ENTRY(entry);
4186                 if (journal->j_start <= temp_jl->j_start) {
4187                         if ((journal->j_start + journal->j_trans_max + 1) >=
4188                             temp_jl->j_start) {
4189                                 flush_used_journal_lists(p_s_sb, temp_jl);
4190                                 goto first_jl;
4191                         } else if ((journal->j_start +
4192                                     journal->j_trans_max + 1) <
4193                                    SB_ONDISK_JOURNAL_SIZE(p_s_sb)) {
4194                                 /* if we don't cross into the next transaction and we don't
4195                                  * wrap, there is no way we can overlap any later transactions
4196                                  * break now
4197                                  */
4198                                 break;
4199                         }
4200                 } else if ((journal->j_start +
4201                             journal->j_trans_max + 1) >
4202                            SB_ONDISK_JOURNAL_SIZE(p_s_sb)) {
4203                         if (((journal->j_start + journal->j_trans_max + 1) %
4204                              SB_ONDISK_JOURNAL_SIZE(p_s_sb)) >=
4205                             temp_jl->j_start) {
4206                                 flush_used_journal_lists(p_s_sb, temp_jl);
4207                                 goto first_jl;
4208                         } else {
4209                                 /* we don't overlap anything from out start to the end of the
4210                                  * log, and our wrapped portion doesn't overlap anything at
4211                                  * the start of the log.  We can break
4212                                  */
4213                                 break;
4214                         }
4215                 }
4216         }
4217         flush_old_journal_lists(p_s_sb);
4218
4219         journal->j_current_jl->j_list_bitmap =
4220             get_list_bitmap(p_s_sb, journal->j_current_jl);
4221
4222         if (!(journal->j_current_jl->j_list_bitmap)) {
4223                 reiserfs_panic(p_s_sb,
4224                                "journal-1996: do_journal_end, could not get a list bitmap\n");
4225         }
4226
4227         atomic_set(&(journal->j_jlock), 0);
4228         unlock_journal(p_s_sb);
4229         /* wake up any body waiting to join. */
4230         clear_bit(J_WRITERS_QUEUED, &journal->j_state);
4231         wake_up(&(journal->j_join_wait));
4232
4233         if (!flush && wait_on_commit &&
4234             journal_list_still_alive(p_s_sb, commit_trans_id)) {
4235                 flush_commit_list(p_s_sb, jl, 1);
4236         }
4237       out:
4238         reiserfs_check_lock_depth(p_s_sb, "journal end2");
4239
4240         memset(th, 0, sizeof(*th));
4241         /* Re-set th->t_super, so we can properly keep track of how many
4242          * persistent transactions there are. We need to do this so if this
4243          * call is part of a failed restart_transaction, we can free it later */
4244         th->t_super = p_s_sb;
4245
4246         return journal->j_errno;
4247 }
4248
4249 static void __reiserfs_journal_abort_hard(struct super_block *sb)
4250 {
4251         struct reiserfs_journal *journal = SB_JOURNAL(sb);
4252         if (test_bit(J_ABORTED, &journal->j_state))
4253                 return;
4254
4255         printk(KERN_CRIT "REISERFS: Aborting journal for filesystem on %s\n",
4256                reiserfs_bdevname(sb));
4257
4258         sb->s_flags |= MS_RDONLY;
4259         set_bit(J_ABORTED, &journal->j_state);
4260
4261 #ifdef CONFIG_REISERFS_CHECK
4262         dump_stack();
4263 #endif
4264 }
4265
4266 static void __reiserfs_journal_abort_soft(struct super_block *sb, int errno)
4267 {
4268         struct reiserfs_journal *journal = SB_JOURNAL(sb);
4269         if (test_bit(J_ABORTED, &journal->j_state))
4270                 return;
4271
4272         if (!journal->j_errno)
4273                 journal->j_errno = errno;
4274
4275         __reiserfs_journal_abort_hard(sb);
4276 }
4277
4278 void reiserfs_journal_abort(struct super_block *sb, int errno)
4279 {
4280         return __reiserfs_journal_abort_soft(sb, errno);
4281 }