fs/reiserfs/journal.c

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