Pileus Git - ~andy/linux/blob - 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         BUG_ON(jl->j_trans_id == 0);
1342
1343         /* if flushall == 0, the lock is already held */
1344         if (flushall) {
1345                 reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s);
1346         } else if (mutex_trylock(&journal->j_flush_mutex)) {
1347                 BUG();
1348         }
1349
1350         count = 0;
1351         if (j_len_saved > journal->j_trans_max) {
1352                 reiserfs_panic(s, "journal-715", "length is %lu, trans id %lu",
1353                                j_len_saved, jl->j_trans_id);
1354                 return 0;
1355         }
1356
1357         /* if all the work is already done, get out of here */
1358         if (atomic_read(&(jl->j_nonzerolen)) <= 0 &&
1359             atomic_read(&(jl->j_commit_left)) <= 0) {
1360                 goto flush_older_and_return;
1361         }
1362
1363         /* start by putting the commit list on disk.  This will also flush
1364          ** the commit lists of any olders transactions
1365          */
1366         flush_commit_list(s, jl, 1);
1367
1368         if (!(jl->j_state & LIST_DIRTY)
1369             && !reiserfs_is_journal_aborted(journal))
1370                 BUG();
1371
1372         /* are we done now? */
1373         if (atomic_read(&(jl->j_nonzerolen)) <= 0 &&
1374             atomic_read(&(jl->j_commit_left)) <= 0) {
1375                 goto flush_older_and_return;
1376         }
1377
1378         /* loop through each cnode, see if we need to write it,
1379          ** or wait on a more recent transaction, or just ignore it
1380          */
1381         if (atomic_read(&(journal->j_wcount)) != 0) {
1382                 reiserfs_panic(s, "journal-844", "journal list is flushing, "
1383                                "wcount is not 0");
1384         }
1385         cn = jl->j_realblock;
1386         while (cn) {
1387                 was_jwait = 0;
1388                 was_dirty = 0;
1389                 saved_bh = NULL;
1390                 /* blocknr of 0 is no longer in the hash, ignore it */
1391                 if (cn->blocknr == 0) {
1392                         goto free_cnode;
1393                 }
1394
1395                 /* This transaction failed commit. Don't write out to the disk */
1396                 if (!(jl->j_state & LIST_DIRTY))
1397                         goto free_cnode;
1398
1399                 pjl = find_newer_jl_for_cn(cn);
1400                 /* the order is important here.  We check pjl to make sure we
1401                  ** don't clear BH_JDirty_wait if we aren't the one writing this
1402                  ** block to disk
1403                  */
1404                 if (!pjl && cn->bh) {
1405                         saved_bh = cn->bh;
1406
1407                         /* we do this to make sure nobody releases the buffer while
1408                          ** we are working with it
1409                          */
1410                         get_bh(saved_bh);
1411
1412                         if (buffer_journal_dirty(saved_bh)) {
1413                                 BUG_ON(!can_dirty(cn));
1414                                 was_jwait = 1;
1415                                 was_dirty = 1;
1416                         } else if (can_dirty(cn)) {
1417                                 /* everything with !pjl && jwait should be writable */
1418                                 BUG();
1419                         }
1420                 }
1421
1422                 /* if someone has this block in a newer transaction, just make
1423                  ** sure they are committed, and don't try writing it to disk
1424                  */
1425                 if (pjl) {
1426                         if (atomic_read(&pjl->j_commit_left))
1427                                 flush_commit_list(s, pjl, 1);
1428                         goto free_cnode;
1429                 }
1430
1431                 /* bh == NULL when the block got to disk on its own, OR,
1432                  ** the block got freed in a future transaction
1433                  */
1434                 if (saved_bh == NULL) {
1435                         goto free_cnode;
1436                 }
1437
1438                 /* this should never happen.  kupdate_one_transaction has this list
1439                  ** locked while it works, so we should never see a buffer here that
1440                  ** is not marked JDirty_wait
1441                  */
1442                 if ((!was_jwait) && !buffer_locked(saved_bh)) {
1443                         reiserfs_warning(s, "journal-813",
1444                                          "BAD! buffer %llu %cdirty %cjwait, "
1445                                          "not in a newer tranasction",
1446                                          (unsigned long long)saved_bh->
1447                                          b_blocknr, was_dirty ? ' ' : '!',
1448                                          was_jwait ? ' ' : '!');
1449                 }
1450                 if (was_dirty) {
1451                         /* we inc again because saved_bh gets decremented at free_cnode */
1452                         get_bh(saved_bh);
1453                         set_bit(BLOCK_NEEDS_FLUSH, &cn->state);
1454                         lock_buffer(saved_bh);
1455                         BUG_ON(cn->blocknr != saved_bh->b_blocknr);
1456                         if (buffer_dirty(saved_bh))
1457                                 submit_logged_buffer(saved_bh);
1458                         else
1459                                 unlock_buffer(saved_bh);
1460                         count++;
1461                 } else {
1462                         reiserfs_warning(s, "clm-2082",
1463                                          "Unable to flush buffer %llu in %s",
1464                                          (unsigned long long)saved_bh->
1465                                          b_blocknr, __func__);
1466                 }
1467               free_cnode:
1468                 last = cn;
1469                 cn = cn->next;
1470                 if (saved_bh) {
1471                         /* we incremented this to keep others from taking the buffer head away */
1472                         put_bh(saved_bh);
1473                         if (atomic_read(&(saved_bh->b_count)) < 0) {
1474                                 reiserfs_warning(s, "journal-945",
1475                                                  "saved_bh->b_count < 0");
1476                         }
1477                 }
1478         }
1479         if (count > 0) {
1480                 cn = jl->j_realblock;
1481                 while (cn) {
1482                         if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) {
1483                                 if (!cn->bh) {
1484                                         reiserfs_panic(s, "journal-1011",
1485                                                        "cn->bh is NULL");
1486                                 }
1487
1488                                 depth = reiserfs_write_unlock_nested(s);
1489                                 __wait_on_buffer(cn->bh);
1490                                 reiserfs_write_lock_nested(s, depth);
1491
1492                                 if (!cn->bh) {
1493                                         reiserfs_panic(s, "journal-1012",
1494                                                        "cn->bh is NULL");
1495                                 }
1496                                 if (unlikely(!buffer_uptodate(cn->bh))) {
1497 #ifdef CONFIG_REISERFS_CHECK
1498                                         reiserfs_warning(s, "journal-949",
1499                                                          "buffer write failed");
1500 #endif
1501                                         err = -EIO;
1502                                 }
1503                                 /* note, we must clear the JDirty_wait bit after the up to date
1504                                  ** check, otherwise we race against our flushpage routine
1505                                  */
1506                                 BUG_ON(!test_clear_buffer_journal_dirty
1507                                        (cn->bh));
1508
1509                                 /* drop one ref for us */
1510                                 put_bh(cn->bh);
1511                                 /* drop one ref for journal_mark_dirty */
1512                                 release_buffer_page(cn->bh);
1513                         }
1514                         cn = cn->next;
1515                 }
1516         }
1517
1518         if (err)
1519                 reiserfs_abort(s, -EIO,
1520                                "Write error while pushing transaction to disk in %s",
1521                                __func__);
1522       flush_older_and_return:
1523
1524         /* before we can update the journal header block, we _must_ flush all
1525          ** real blocks from all older transactions to disk.  This is because
1526          ** once the header block is updated, this transaction will not be
1527          ** replayed after a crash
1528          */
1529         if (flushall) {
1530                 flush_older_journal_lists(s, jl);
1531         }
1532
1533         err = journal->j_errno;
1534         /* before we can remove everything from the hash tables for this
1535          ** transaction, we must make sure it can never be replayed
1536          **
1537          ** since we are only called from do_journal_end, we know for sure there
1538          ** are no allocations going on while we are flushing journal lists.  So,
1539          ** we only need to update the journal header block for the last list
1540          ** being flushed
1541          */
1542         if (!err && flushall) {
1543                 err =
1544                     update_journal_header_block(s,
1545                                                 (jl->j_start + jl->j_len +
1546                                                  2) % SB_ONDISK_JOURNAL_SIZE(s),
1547                                                 jl->j_trans_id);
1548                 if (err)
1549                         reiserfs_abort(s, -EIO,
1550                                        "Write error while updating journal header in %s",
1551                                        __func__);
1552         }
1553         remove_all_from_journal_list(s, jl, 0);
1554         list_del_init(&jl->j_list);
1555         journal->j_num_lists--;
1556         del_from_work_list(s, jl);
1557
1558         if (journal->j_last_flush_id != 0 &&
1559             (jl->j_trans_id - journal->j_last_flush_id) != 1) {
1560                 reiserfs_warning(s, "clm-2201", "last flush %lu, current %lu",
1561                                  journal->j_last_flush_id, jl->j_trans_id);
1562         }
1563         journal->j_last_flush_id = jl->j_trans_id;
1564
1565         /* not strictly required since we are freeing the list, but it should
1566          * help find code using dead lists later on
1567          */
1568         jl->j_len = 0;
1569         atomic_set(&(jl->j_nonzerolen), 0);
1570         jl->j_start = 0;
1571         jl->j_realblock = NULL;
1572         jl->j_commit_bh = NULL;
1573         jl->j_trans_id = 0;
1574         jl->j_state = 0;
1575         put_journal_list(s, jl);
1576         if (flushall)
1577                 mutex_unlock(&journal->j_flush_mutex);
1578         return err;
1579 }
1580
1581 static int write_one_transaction(struct super_block *s,
1582                                  struct reiserfs_journal_list *jl,
1583                                  struct buffer_chunk *chunk)
1584 {
1585         struct reiserfs_journal_cnode *cn;
1586         int ret = 0;
1587
1588         jl->j_state |= LIST_TOUCHED;
1589         del_from_work_list(s, jl);
1590         if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0) {
1591                 return 0;
1592         }
1593
1594         cn = jl->j_realblock;
1595         while (cn) {
1596                 /* if the blocknr == 0, this has been cleared from the hash,
1597                  ** skip it
1598                  */
1599                 if (cn->blocknr == 0) {
1600                         goto next;
1601                 }
1602                 if (cn->bh && can_dirty(cn) && buffer_dirty(cn->bh)) {
1603                         struct buffer_head *tmp_bh;
1604                         /* we can race against journal_mark_freed when we try
1605                          * to lock_buffer(cn->bh), so we have to inc the buffer
1606                          * count, and recheck things after locking
1607                          */
1608                         tmp_bh = cn->bh;
1609                         get_bh(tmp_bh);
1610                         lock_buffer(tmp_bh);
1611                         if (cn->bh && can_dirty(cn) && buffer_dirty(tmp_bh)) {
1612                                 if (!buffer_journal_dirty(tmp_bh) ||
1613                                     buffer_journal_prepared(tmp_bh))
1614                                         BUG();
1615                                 add_to_chunk(chunk, tmp_bh, NULL, write_chunk);
1616                                 ret++;
1617                         } else {
1618                                 /* note, cn->bh might be null now */
1619                                 unlock_buffer(tmp_bh);
1620                         }
1621                         put_bh(tmp_bh);
1622                 }
1623               next:
1624                 cn = cn->next;
1625                 cond_resched();
1626         }
1627         return ret;
1628 }
1629
1630 /* used by flush_commit_list */
1631 static int dirty_one_transaction(struct super_block *s,
1632                                  struct reiserfs_journal_list *jl)
1633 {
1634         struct reiserfs_journal_cnode *cn;
1635         struct reiserfs_journal_list *pjl;
1636         int ret = 0;
1637
1638         jl->j_state |= LIST_DIRTY;
1639         cn = jl->j_realblock;
1640         while (cn) {
1641                 /* look for a more recent transaction that logged this
1642                  ** buffer.  Only the most recent transaction with a buffer in
1643                  ** it is allowed to send that buffer to disk
1644                  */
1645                 pjl = find_newer_jl_for_cn(cn);
1646                 if (!pjl && cn->blocknr && cn->bh
1647                     && buffer_journal_dirty(cn->bh)) {
1648                         BUG_ON(!can_dirty(cn));
1649                         /* if the buffer is prepared, it will either be logged
1650                          * or restored.  If restored, we need to make sure
1651                          * it actually gets marked dirty
1652                          */
1653                         clear_buffer_journal_new(cn->bh);
1654                         if (buffer_journal_prepared(cn->bh)) {
1655                                 set_buffer_journal_restore_dirty(cn->bh);
1656                         } else {
1657                                 set_buffer_journal_test(cn->bh);
1658                                 mark_buffer_dirty(cn->bh);
1659                         }
1660                 }
1661                 cn = cn->next;
1662         }
1663         return ret;
1664 }
1665
1666 static int kupdate_transactions(struct super_block *s,
1667                                 struct reiserfs_journal_list *jl,
1668                                 struct reiserfs_journal_list **next_jl,
1669                                 unsigned int *next_trans_id,
1670                                 int num_blocks, int num_trans)
1671 {
1672         int ret = 0;
1673         int written = 0;
1674         int transactions_flushed = 0;
1675         unsigned int orig_trans_id = jl->j_trans_id;
1676         struct buffer_chunk chunk;
1677         struct list_head *entry;
1678         struct reiserfs_journal *journal = SB_JOURNAL(s);
1679         chunk.nr = 0;
1680
1681         reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s);
1682         if (!journal_list_still_alive(s, orig_trans_id)) {
1683                 goto done;
1684         }
1685
1686         /* we've got j_flush_mutex held, nobody is going to delete any
1687          * of these lists out from underneath us
1688          */
1689         while ((num_trans && transactions_flushed < num_trans) ||
1690                (!num_trans && written < num_blocks)) {
1691
1692                 if (jl->j_len == 0 || (jl->j_state & LIST_TOUCHED) ||
1693                     atomic_read(&jl->j_commit_left)
1694                     || !(jl->j_state & LIST_DIRTY)) {
1695                         del_from_work_list(s, jl);
1696                         break;
1697                 }
1698                 ret = write_one_transaction(s, jl, &chunk);
1699
1700                 if (ret < 0)
1701                         goto done;
1702                 transactions_flushed++;
1703                 written += ret;
1704                 entry = jl->j_list.next;
1705
1706                 /* did we wrap? */
1707                 if (entry == &journal->j_journal_list) {
1708                         break;
1709                 }
1710                 jl = JOURNAL_LIST_ENTRY(entry);
1711
1712                 /* don't bother with older transactions */
1713                 if (jl->j_trans_id <= orig_trans_id)
1714                         break;
1715         }
1716         if (chunk.nr) {
1717                 write_chunk(&chunk);
1718         }
1719
1720       done:
1721         mutex_unlock(&journal->j_flush_mutex);
1722         return ret;
1723 }
1724
1725 /* for o_sync and fsync heavy applications, they tend to use
1726 ** all the journa list slots with tiny transactions.  These
1727 ** trigger lots and lots of calls to update the header block, which
1728 ** adds seeks and slows things down.
1729 **
1730 ** This function tries to clear out a large chunk of the journal lists
1731 ** at once, which makes everything faster since only the newest journal
1732 ** list updates the header block
1733 */
1734 static int flush_used_journal_lists(struct super_block *s,
1735                                     struct reiserfs_journal_list *jl)
1736 {
1737         unsigned long len = 0;
1738         unsigned long cur_len;
1739         int ret;
1740         int i;
1741         int limit = 256;
1742         struct reiserfs_journal_list *tjl;
1743         struct reiserfs_journal_list *flush_jl;
1744         unsigned int trans_id;
1745         struct reiserfs_journal *journal = SB_JOURNAL(s);
1746
1747         flush_jl = tjl = jl;
1748
1749         /* in data logging mode, try harder to flush a lot of blocks */
1750         if (reiserfs_data_log(s))
1751                 limit = 1024;
1752         /* flush for 256 transactions or limit blocks, whichever comes first */
1753         for (i = 0; i < 256 && len < limit; i++) {
1754                 if (atomic_read(&tjl->j_commit_left) ||
1755                     tjl->j_trans_id < jl->j_trans_id) {
1756                         break;
1757                 }
1758                 cur_len = atomic_read(&tjl->j_nonzerolen);
1759                 if (cur_len > 0) {
1760                         tjl->j_state &= ~LIST_TOUCHED;
1761                 }
1762                 len += cur_len;
1763                 flush_jl = tjl;
1764                 if (tjl->j_list.next == &journal->j_journal_list)
1765                         break;
1766                 tjl = JOURNAL_LIST_ENTRY(tjl->j_list.next);
1767         }
1768         /* try to find a group of blocks we can flush across all the
1769          ** transactions, but only bother if we've actually spanned
1770          ** across multiple lists
1771          */
1772         if (flush_jl != jl) {
1773                 ret = kupdate_transactions(s, jl, &tjl, &trans_id, len, i);
1774         }
1775         flush_journal_list(s, flush_jl, 1);
1776         return 0;
1777 }
1778
1779 /*
1780 ** removes any nodes in table with name block and dev as bh.
1781 ** only touchs the hnext and hprev pointers.
1782 */
1783 void remove_journal_hash(struct super_block *sb,
1784                          struct reiserfs_journal_cnode **table,
1785                          struct reiserfs_journal_list *jl,
1786                          unsigned long block, int remove_freed)
1787 {
1788         struct reiserfs_journal_cnode *cur;
1789         struct reiserfs_journal_cnode **head;
1790
1791         head = &(journal_hash(table, sb, block));
1792         if (!head) {
1793                 return;
1794         }
1795         cur = *head;
1796         while (cur) {
1797                 if (cur->blocknr == block && cur->sb == sb
1798                     && (jl == NULL || jl == cur->jlist)
1799                     && (!test_bit(BLOCK_FREED, &cur->state) || remove_freed)) {
1800                         if (cur->hnext) {
1801                                 cur->hnext->hprev = cur->hprev;
1802                         }
1803                         if (cur->hprev) {
1804                                 cur->hprev->hnext = cur->hnext;
1805                         } else {
1806                                 *head = cur->hnext;
1807                         }
1808                         cur->blocknr = 0;
1809                         cur->sb = NULL;
1810                         cur->state = 0;
1811                         if (cur->bh && cur->jlist)      /* anybody who clears the cur->bh will also dec the nonzerolen */
1812                                 atomic_dec(&(cur->jlist->j_nonzerolen));
1813                         cur->bh = NULL;
1814                         cur->jlist = NULL;
1815                 }
1816                 cur = cur->hnext;
1817         }
1818 }
1819
1820 static void free_journal_ram(struct super_block *sb)
1821 {
1822         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1823         kfree(journal->j_current_jl);
1824         journal->j_num_lists--;
1825
1826         vfree(journal->j_cnode_free_orig);
1827         free_list_bitmaps(sb, journal->j_list_bitmap);
1828         free_bitmap_nodes(sb);  /* must be after free_list_bitmaps */
1829         if (journal->j_header_bh) {
1830                 brelse(journal->j_header_bh);
1831         }
1832         /* j_header_bh is on the journal dev, make sure not to release the journal
1833          * dev until we brelse j_header_bh
1834          */
1835         release_journal_dev(sb, journal);
1836         vfree(journal);
1837 }
1838
1839 /*
1840 ** call on unmount.  Only set error to 1 if you haven't made your way out
1841 ** of read_super() yet.  Any other caller must keep error at 0.
1842 */
1843 static int do_journal_release(struct reiserfs_transaction_handle *th,
1844                               struct super_block *sb, int error)
1845 {
1846         struct reiserfs_transaction_handle myth;
1847         int flushed = 0;
1848         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1849
1850         /* we only want to flush out transactions if we were called with error == 0
1851          */
1852         if (!error && !(sb->s_flags & MS_RDONLY)) {
1853                 /* end the current trans */
1854                 BUG_ON(!th->t_trans_id);
1855                 do_journal_end(th, sb, 10, FLUSH_ALL);
1856
1857                 /* make sure something gets logged to force our way into the flush code */
1858                 if (!journal_join(&myth, sb, 1)) {
1859                         reiserfs_prepare_for_journal(sb,
1860                                                      SB_BUFFER_WITH_SB(sb),
1861                                                      1);
1862                         journal_mark_dirty(&myth, sb,
1863                                            SB_BUFFER_WITH_SB(sb));
1864                         do_journal_end(&myth, sb, 1, FLUSH_ALL);
1865                         flushed = 1;
1866                 }
1867         }
1868
1869         /* this also catches errors during the do_journal_end above */
1870         if (!error && reiserfs_is_journal_aborted(journal)) {
1871                 memset(&myth, 0, sizeof(myth));
1872                 if (!journal_join_abort(&myth, sb, 1)) {
1873                         reiserfs_prepare_for_journal(sb,
1874                                                      SB_BUFFER_WITH_SB(sb),
1875                                                      1);
1876                         journal_mark_dirty(&myth, sb,
1877                                            SB_BUFFER_WITH_SB(sb));
1878                         do_journal_end(&myth, sb, 1, FLUSH_ALL);
1879                 }
1880         }
1881
1882         reiserfs_mounted_fs_count--;
1883         /* wait for all commits to finish */
1884         cancel_delayed_work(&SB_JOURNAL(sb)->j_work);
1885
1886         /*
1887          * We must release the write lock here because
1888          * the workqueue job (flush_async_commit) needs this lock
1889          */
1890         reiserfs_write_unlock(sb);
1891
1892         cancel_delayed_work_sync(&REISERFS_SB(sb)->old_work);
1893         flush_workqueue(commit_wq);
1894
1895         if (!reiserfs_mounted_fs_count) {
1896                 destroy_workqueue(commit_wq);
1897                 commit_wq = NULL;
1898         }
1899
1900         free_journal_ram(sb);
1901
1902         reiserfs_write_lock(sb);
1903
1904         return 0;
1905 }
1906
1907 /*
1908 ** call on unmount.  flush all journal trans, release all alloc'd ram
1909 */
1910 int journal_release(struct reiserfs_transaction_handle *th,
1911                     struct super_block *sb)
1912 {
1913         return do_journal_release(th, sb, 0);
1914 }
1915
1916 /*
1917 ** only call from an error condition inside reiserfs_read_super!
1918 */
1919 int journal_release_error(struct reiserfs_transaction_handle *th,
1920                           struct super_block *sb)
1921 {
1922         return do_journal_release(th, sb, 1);
1923 }
1924
1925 /* compares description block with commit block.  returns 1 if they differ, 0 if they are the same */
1926 static int journal_compare_desc_commit(struct super_block *sb,
1927                                        struct reiserfs_journal_desc *desc,
1928                                        struct reiserfs_journal_commit *commit)
1929 {
1930         if (get_commit_trans_id(commit) != get_desc_trans_id(desc) ||
1931             get_commit_trans_len(commit) != get_desc_trans_len(desc) ||
1932             get_commit_trans_len(commit) > SB_JOURNAL(sb)->j_trans_max ||
1933             get_commit_trans_len(commit) <= 0) {
1934                 return 1;
1935         }
1936         return 0;
1937 }
1938
1939 /* returns 0 if it did not find a description block
1940 ** returns -1 if it found a corrupt commit block
1941 ** returns 1 if both desc and commit were valid
1942 ** NOTE: only called during fs mount
1943 */
1944 static int journal_transaction_is_valid(struct super_block *sb,
1945                                         struct buffer_head *d_bh,
1946                                         unsigned int *oldest_invalid_trans_id,
1947                                         unsigned long *newest_mount_id)
1948 {
1949         struct reiserfs_journal_desc *desc;
1950         struct reiserfs_journal_commit *commit;
1951         struct buffer_head *c_bh;
1952         unsigned long offset;
1953
1954         if (!d_bh)
1955                 return 0;
1956
1957         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
1958         if (get_desc_trans_len(desc) > 0
1959             && !memcmp(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8)) {
1960                 if (oldest_invalid_trans_id && *oldest_invalid_trans_id
1961                     && get_desc_trans_id(desc) > *oldest_invalid_trans_id) {
1962                         reiserfs_debug(sb, REISERFS_DEBUG_CODE,
1963                                        "journal-986: transaction "
1964                                        "is valid returning because trans_id %d is greater than "
1965                                        "oldest_invalid %lu",
1966                                        get_desc_trans_id(desc),
1967                                        *oldest_invalid_trans_id);
1968                         return 0;
1969                 }
1970                 if (newest_mount_id
1971                     && *newest_mount_id > get_desc_mount_id(desc)) {
1972                         reiserfs_debug(sb, REISERFS_DEBUG_CODE,
1973                                        "journal-1087: transaction "
1974                                        "is valid returning because mount_id %d is less than "
1975                                        "newest_mount_id %lu",
1976                                        get_desc_mount_id(desc),
1977                                        *newest_mount_id);
1978                         return -1;
1979                 }
1980                 if (get_desc_trans_len(desc) > SB_JOURNAL(sb)->j_trans_max) {
1981                         reiserfs_warning(sb, "journal-2018",
1982                                          "Bad transaction length %d "
1983                                          "encountered, ignoring transaction",
1984                                          get_desc_trans_len(desc));
1985                         return -1;
1986                 }
1987                 offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
1988
1989                 /* ok, we have a journal description block, lets see if the transaction was valid */
1990                 c_bh =
1991                     journal_bread(sb,
1992                                   SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
1993                                   ((offset + get_desc_trans_len(desc) +
1994                                     1) % SB_ONDISK_JOURNAL_SIZE(sb)));
1995                 if (!c_bh)
1996                         return 0;
1997                 commit = (struct reiserfs_journal_commit *)c_bh->b_data;
1998                 if (journal_compare_desc_commit(sb, desc, commit)) {
1999                         reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2000                                        "journal_transaction_is_valid, commit offset %ld had bad "
2001                                        "time %d or length %d",
2002                                        c_bh->b_blocknr -
2003                                        SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2004                                        get_commit_trans_id(commit),
2005                                        get_commit_trans_len(commit));
2006                         brelse(c_bh);
2007                         if (oldest_invalid_trans_id) {
2008                                 *oldest_invalid_trans_id =
2009                                     get_desc_trans_id(desc);
2010                                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2011                                                "journal-1004: "
2012                                                "transaction_is_valid setting oldest invalid trans_id "
2013                                                "to %d",
2014                                                get_desc_trans_id(desc));
2015                         }
2016                         return -1;
2017                 }
2018                 brelse(c_bh);
2019                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2020                                "journal-1006: found valid "
2021                                "transaction start offset %llu, len %d id %d",
2022                                d_bh->b_blocknr -
2023                                SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2024                                get_desc_trans_len(desc),
2025                                get_desc_trans_id(desc));
2026                 return 1;
2027         } else {
2028                 return 0;
2029         }
2030 }
2031
2032 static void brelse_array(struct buffer_head **heads, int num)
2033 {
2034         int i;
2035         for (i = 0; i < num; i++) {
2036                 brelse(heads[i]);
2037         }
2038 }
2039
2040 /*
2041 ** given the start, and values for the oldest acceptable transactions,
2042 ** this either reads in a replays a transaction, or returns because the
2043 ** transaction is invalid, or too old.
2044 ** NOTE: only called during fs mount
2045 */
2046 static int journal_read_transaction(struct super_block *sb,
2047                                     unsigned long cur_dblock,
2048                                     unsigned long oldest_start,
2049                                     unsigned int oldest_trans_id,
2050                                     unsigned long newest_mount_id)
2051 {
2052         struct reiserfs_journal *journal = SB_JOURNAL(sb);
2053         struct reiserfs_journal_desc *desc;
2054         struct reiserfs_journal_commit *commit;
2055         unsigned int trans_id = 0;
2056         struct buffer_head *c_bh;
2057         struct buffer_head *d_bh;
2058         struct buffer_head **log_blocks = NULL;
2059         struct buffer_head **real_blocks = NULL;
2060         unsigned int trans_offset;
2061         int i;
2062         int trans_half;
2063
2064         d_bh = journal_bread(sb, cur_dblock);
2065         if (!d_bh)
2066                 return 1;
2067         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2068         trans_offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2069         reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1037: "
2070                        "journal_read_transaction, offset %llu, len %d mount_id %d",
2071                        d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2072                        get_desc_trans_len(desc), get_desc_mount_id(desc));
2073         if (get_desc_trans_id(desc) < oldest_trans_id) {
2074                 reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1039: "
2075                                "journal_read_trans skipping because %lu is too old",
2076                                cur_dblock -
2077                                SB_ONDISK_JOURNAL_1st_BLOCK(sb));
2078                 brelse(d_bh);
2079                 return 1;
2080         }
2081         if (get_desc_mount_id(desc) != newest_mount_id) {
2082                 reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1146: "
2083                                "journal_read_trans skipping because %d is != "
2084                                "newest_mount_id %lu", get_desc_mount_id(desc),
2085                                newest_mount_id);
2086                 brelse(d_bh);
2087                 return 1;
2088         }
2089         c_bh = journal_bread(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2090                              ((trans_offset + get_desc_trans_len(desc) + 1) %
2091                               SB_ONDISK_JOURNAL_SIZE(sb)));
2092         if (!c_bh) {
2093                 brelse(d_bh);
2094                 return 1;
2095         }
2096         commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2097         if (journal_compare_desc_commit(sb, desc, commit)) {
2098                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2099                                "journal_read_transaction, "
2100                                "commit offset %llu had bad time %d or length %d",
2101                                c_bh->b_blocknr -
2102                                SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2103                                get_commit_trans_id(commit),
2104                                get_commit_trans_len(commit));
2105                 brelse(c_bh);
2106                 brelse(d_bh);
2107                 return 1;
2108         }
2109
2110         if (bdev_read_only(sb->s_bdev)) {
2111                 reiserfs_warning(sb, "clm-2076",
2112                                  "device is readonly, unable to replay log");
2113                 brelse(c_bh);
2114                 brelse(d_bh);
2115                 return -EROFS;
2116         }
2117
2118         trans_id = get_desc_trans_id(desc);
2119         /* now we know we've got a good transaction, and it was inside the valid time ranges */
2120         log_blocks = kmalloc(get_desc_trans_len(desc) *
2121                              sizeof(struct buffer_head *), GFP_NOFS);
2122         real_blocks = kmalloc(get_desc_trans_len(desc) *
2123                               sizeof(struct buffer_head *), GFP_NOFS);
2124         if (!log_blocks || !real_blocks) {
2125                 brelse(c_bh);
2126                 brelse(d_bh);
2127                 kfree(log_blocks);
2128                 kfree(real_blocks);
2129                 reiserfs_warning(sb, "journal-1169",
2130                                  "kmalloc failed, unable to mount FS");
2131                 return -1;
2132         }
2133         /* get all the buffer heads */
2134         trans_half = journal_trans_half(sb->s_blocksize);
2135         for (i = 0; i < get_desc_trans_len(desc); i++) {
2136                 log_blocks[i] =
2137                     journal_getblk(sb,
2138                                    SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2139                                    (trans_offset + 1 +
2140                                     i) % SB_ONDISK_JOURNAL_SIZE(sb));
2141                 if (i < trans_half) {
2142                         real_blocks[i] =
2143                             sb_getblk(sb,
2144                                       le32_to_cpu(desc->j_realblock[i]));
2145                 } else {
2146                         real_blocks[i] =
2147                             sb_getblk(sb,
2148                                       le32_to_cpu(commit->
2149                                                   j_realblock[i - trans_half]));
2150                 }
2151                 if (real_blocks[i]->b_blocknr > SB_BLOCK_COUNT(sb)) {
2152                         reiserfs_warning(sb, "journal-1207",
2153                                          "REPLAY FAILURE fsck required! "
2154                                          "Block to replay is outside of "
2155                                          "filesystem");
2156                         goto abort_replay;
2157                 }
2158                 /* make sure we don't try to replay onto log or reserved area */
2159                 if (is_block_in_log_or_reserved_area
2160                     (sb, real_blocks[i]->b_blocknr)) {
2161                         reiserfs_warning(sb, "journal-1204",
2162                                          "REPLAY FAILURE fsck required! "
2163                                          "Trying to replay onto a log block");
2164                       abort_replay:
2165                         brelse_array(log_blocks, i);
2166                         brelse_array(real_blocks, i);
2167                         brelse(c_bh);
2168                         brelse(d_bh);
2169                         kfree(log_blocks);
2170                         kfree(real_blocks);
2171                         return -1;
2172                 }
2173         }
2174         /* read in the log blocks, memcpy to the corresponding real block */
2175         ll_rw_block(READ, get_desc_trans_len(desc), log_blocks);
2176         for (i = 0; i < get_desc_trans_len(desc); i++) {
2177
2178                 wait_on_buffer(log_blocks[i]);
2179                 if (!buffer_uptodate(log_blocks[i])) {
2180                         reiserfs_warning(sb, "journal-1212",
2181                                          "REPLAY FAILURE fsck required! "
2182                                          "buffer write failed");
2183                         brelse_array(log_blocks + i,
2184                                      get_desc_trans_len(desc) - i);
2185                         brelse_array(real_blocks, get_desc_trans_len(desc));
2186                         brelse(c_bh);
2187                         brelse(d_bh);
2188                         kfree(log_blocks);
2189                         kfree(real_blocks);
2190                         return -1;
2191                 }
2192                 memcpy(real_blocks[i]->b_data, log_blocks[i]->b_data,
2193                        real_blocks[i]->b_size);
2194                 set_buffer_uptodate(real_blocks[i]);
2195                 brelse(log_blocks[i]);
2196         }
2197         /* flush out the real blocks */
2198         for (i = 0; i < get_desc_trans_len(desc); i++) {
2199                 set_buffer_dirty(real_blocks[i]);
2200                 write_dirty_buffer(real_blocks[i], WRITE);
2201         }
2202         for (i = 0; i < get_desc_trans_len(desc); i++) {
2203                 wait_on_buffer(real_blocks[i]);
2204                 if (!buffer_uptodate(real_blocks[i])) {
2205                         reiserfs_warning(sb, "journal-1226",
2206                                          "REPLAY FAILURE, fsck required! "
2207                                          "buffer write failed");
2208                         brelse_array(real_blocks + i,
2209                                      get_desc_trans_len(desc) - i);
2210                         brelse(c_bh);
2211                         brelse(d_bh);
2212                         kfree(log_blocks);
2213                         kfree(real_blocks);
2214                         return -1;
2215                 }
2216                 brelse(real_blocks[i]);
2217         }
2218         cur_dblock =
2219             SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2220             ((trans_offset + get_desc_trans_len(desc) +
2221               2) % SB_ONDISK_JOURNAL_SIZE(sb));
2222         reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2223                        "journal-1095: setting journal " "start to offset %ld",
2224                        cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb));
2225
2226         /* init starting values for the first transaction, in case this is the last transaction to be replayed. */
2227         journal->j_start = cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2228         journal->j_last_flush_trans_id = trans_id;
2229         journal->j_trans_id = trans_id + 1;
2230         /* check for trans_id overflow */
2231         if (journal->j_trans_id == 0)
2232                 journal->j_trans_id = 10;
2233         brelse(c_bh);
2234         brelse(d_bh);
2235         kfree(log_blocks);
2236         kfree(real_blocks);
2237         return 0;
2238 }
2239
2240 /* This function reads blocks starting from block and to max_block of bufsize
2241    size (but no more than BUFNR blocks at a time). This proved to improve
2242    mounting speed on self-rebuilding raid5 arrays at least.
2243    Right now it is only used from journal code. But later we might use it
2244    from other places.
2245    Note: Do not use journal_getblk/sb_getblk functions here! */
2246 static struct buffer_head *reiserfs_breada(struct block_device *dev,
2247                                            b_blocknr_t block, int bufsize,
2248                                            b_blocknr_t max_block)
2249 {
2250         struct buffer_head *bhlist[BUFNR];
2251         unsigned int blocks = BUFNR;
2252         struct buffer_head *bh;
2253         int i, j;
2254
2255         bh = __getblk(dev, block, bufsize);
2256         if (buffer_uptodate(bh))
2257                 return (bh);
2258
2259         if (block + BUFNR > max_block) {
2260                 blocks = max_block - block;
2261         }
2262         bhlist[0] = bh;
2263         j = 1;
2264         for (i = 1; i < blocks; i++) {
2265                 bh = __getblk(dev, block + i, bufsize);
2266                 if (buffer_uptodate(bh)) {
2267                         brelse(bh);
2268                         break;
2269                 } else
2270                         bhlist[j++] = bh;
2271         }
2272         ll_rw_block(READ, j, bhlist);
2273         for (i = 1; i < j; i++)
2274                 brelse(bhlist[i]);
2275         bh = bhlist[0];
2276         wait_on_buffer(bh);
2277         if (buffer_uptodate(bh))
2278                 return bh;
2279         brelse(bh);
2280         return NULL;
2281 }
2282
2283 /*
2284 ** read and replay the log
2285 ** on a clean unmount, the journal header's next unflushed pointer will
2286 ** be to an invalid transaction.  This tests that before finding all the
2287 ** transactions in the log, which makes normal mount times fast.
2288 ** After a crash, this starts with the next unflushed transaction, and
2289 ** replays until it finds one too old, or invalid.
2290 ** On exit, it sets things up so the first transaction will work correctly.
2291 ** NOTE: only called during fs mount
2292 */
2293 static int journal_read(struct super_block *sb)
2294 {
2295         struct reiserfs_journal *journal = SB_JOURNAL(sb);
2296         struct reiserfs_journal_desc *desc;
2297         unsigned int oldest_trans_id = 0;
2298         unsigned int oldest_invalid_trans_id = 0;
2299         time_t start;
2300         unsigned long oldest_start = 0;
2301         unsigned long cur_dblock = 0;
2302         unsigned long newest_mount_id = 9;
2303         struct buffer_head *d_bh;
2304         struct reiserfs_journal_header *jh;
2305         int valid_journal_header = 0;
2306         int replay_count = 0;
2307         int continue_replay = 1;
2308         int ret;
2309         char b[BDEVNAME_SIZE];
2310
2311         cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2312         reiserfs_info(sb, "checking transaction log (%s)\n",
2313                       bdevname(journal->j_dev_bd, b));
2314         start = get_seconds();
2315
2316         /* step 1, read in the journal header block.  Check the transaction it says
2317          ** is the first unflushed, and if that transaction is not valid,
2318          ** replay is done
2319          */
2320         journal->j_header_bh = journal_bread(sb,
2321                                              SB_ONDISK_JOURNAL_1st_BLOCK(sb)
2322                                              + SB_ONDISK_JOURNAL_SIZE(sb));
2323         if (!journal->j_header_bh) {
2324                 return 1;
2325         }
2326         jh = (struct reiserfs_journal_header *)(journal->j_header_bh->b_data);
2327         if (le32_to_cpu(jh->j_first_unflushed_offset) <
2328             SB_ONDISK_JOURNAL_SIZE(sb)
2329             && le32_to_cpu(jh->j_last_flush_trans_id) > 0) {
2330                 oldest_start =
2331                     SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2332                     le32_to_cpu(jh->j_first_unflushed_offset);
2333                 oldest_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2334                 newest_mount_id = le32_to_cpu(jh->j_mount_id);
2335                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2336                                "journal-1153: found in "
2337                                "header: first_unflushed_offset %d, last_flushed_trans_id "
2338                                "%lu", le32_to_cpu(jh->j_first_unflushed_offset),
2339                                le32_to_cpu(jh->j_last_flush_trans_id));
2340                 valid_journal_header = 1;
2341
2342                 /* now, we try to read the first unflushed offset.  If it is not valid,
2343                  ** there is nothing more we can do, and it makes no sense to read
2344                  ** through the whole log.
2345                  */
2346                 d_bh =
2347                     journal_bread(sb,
2348                                   SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2349                                   le32_to_cpu(jh->j_first_unflushed_offset));
2350                 ret = journal_transaction_is_valid(sb, d_bh, NULL, NULL);
2351                 if (!ret) {
2352                         continue_replay = 0;
2353                 }
2354                 brelse(d_bh);
2355                 goto start_log_replay;
2356         }
2357
2358         /* ok, there are transactions that need to be replayed.  start with the first log block, find
2359          ** all the valid transactions, and pick out the oldest.
2360          */
2361         while (continue_replay
2362                && cur_dblock <
2363                (SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2364                 SB_ONDISK_JOURNAL_SIZE(sb))) {
2365                 /* Note that it is required for blocksize of primary fs device and journal
2366                    device to be the same */
2367                 d_bh =
2368                     reiserfs_breada(journal->j_dev_bd, cur_dblock,
2369                                     sb->s_blocksize,
2370                                     SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2371                                     SB_ONDISK_JOURNAL_SIZE(sb));
2372                 ret =
2373                     journal_transaction_is_valid(sb, d_bh,
2374                                                  &oldest_invalid_trans_id,
2375                                                  &newest_mount_id);
2376                 if (ret == 1) {
2377                         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2378                         if (oldest_start == 0) {        /* init all oldest_ values */
2379                                 oldest_trans_id = get_desc_trans_id(desc);
2380                                 oldest_start = d_bh->b_blocknr;
2381                                 newest_mount_id = get_desc_mount_id(desc);
2382                                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2383                                                "journal-1179: Setting "
2384                                                "oldest_start to offset %llu, trans_id %lu",
2385                                                oldest_start -
2386                                                SB_ONDISK_JOURNAL_1st_BLOCK
2387                                                (sb), oldest_trans_id);
2388                         } else if (oldest_trans_id > get_desc_trans_id(desc)) {
2389                                 /* one we just read was older */
2390                                 oldest_trans_id = get_desc_trans_id(desc);
2391                                 oldest_start = d_bh->b_blocknr;
2392                                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2393                                                "journal-1180: Resetting "
2394                                                "oldest_start to offset %lu, trans_id %lu",
2395                                                oldest_start -
2396                                                SB_ONDISK_JOURNAL_1st_BLOCK
2397                                                (sb), oldest_trans_id);
2398                         }
2399                         if (newest_mount_id < get_desc_mount_id(desc)) {
2400                                 newest_mount_id = get_desc_mount_id(desc);
2401                                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2402                                                "journal-1299: Setting "
2403                                                "newest_mount_id to %d",
2404                                                get_desc_mount_id(desc));
2405                         }
2406                         cur_dblock += get_desc_trans_len(desc) + 2;
2407                 } else {
2408                         cur_dblock++;
2409                 }
2410                 brelse(d_bh);
2411         }
2412
2413       start_log_replay:
2414         cur_dblock = oldest_start;
2415         if (oldest_trans_id) {
2416                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2417                                "journal-1206: Starting replay "
2418                                "from offset %llu, trans_id %lu",
2419                                cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2420                                oldest_trans_id);
2421
2422         }
2423         replay_count = 0;
2424         while (continue_replay && oldest_trans_id > 0) {
2425                 ret =
2426                     journal_read_transaction(sb, cur_dblock, oldest_start,
2427                                              oldest_trans_id, newest_mount_id);
2428                 if (ret < 0) {
2429                         return ret;
2430                 } else if (ret != 0) {
2431                         break;
2432                 }
2433                 cur_dblock =
2434                     SB_ONDISK_JOURNAL_1st_BLOCK(sb) + journal->j_start;
2435                 replay_count++;
2436                 if (cur_dblock == oldest_start)
2437                         break;
2438         }
2439
2440         if (oldest_trans_id == 0) {
2441                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2442                                "journal-1225: No valid " "transactions found");
2443         }
2444         /* j_start does not get set correctly if we don't replay any transactions.
2445          ** if we had a valid journal_header, set j_start to the first unflushed transaction value,
2446          ** copy the trans_id from the header
2447          */
2448         if (valid_journal_header && replay_count == 0) {
2449                 journal->j_start = le32_to_cpu(jh->j_first_unflushed_offset);
2450                 journal->j_trans_id =
2451                     le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2452                 /* check for trans_id overflow */
2453                 if (journal->j_trans_id == 0)
2454                         journal->j_trans_id = 10;
2455                 journal->j_last_flush_trans_id =
2456                     le32_to_cpu(jh->j_last_flush_trans_id);
2457                 journal->j_mount_id = le32_to_cpu(jh->j_mount_id) + 1;
2458         } else {
2459                 journal->j_mount_id = newest_mount_id + 1;
2460         }
2461         reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1299: Setting "
2462                        "newest_mount_id to %lu", journal->j_mount_id);
2463         journal->j_first_unflushed_offset = journal->j_start;
2464         if (replay_count > 0) {
2465                 reiserfs_info(sb,
2466                               "replayed %d transactions in %lu seconds\n",
2467                               replay_count, get_seconds() - start);
2468         }
2469         /* needed to satisfy the locking in _update_journal_header_block */
2470         reiserfs_write_lock(sb);
2471         if (!bdev_read_only(sb->s_bdev) &&
2472             _update_journal_header_block(sb, journal->j_start,
2473                                          journal->j_last_flush_trans_id)) {
2474                 reiserfs_write_unlock(sb);
2475                 /* replay failed, caller must call free_journal_ram and abort
2476                  ** the mount
2477                  */
2478                 return -1;
2479         }
2480         reiserfs_write_unlock(sb);
2481         return 0;
2482 }
2483
2484 static struct reiserfs_journal_list *alloc_journal_list(struct super_block *s)
2485 {
2486         struct reiserfs_journal_list *jl;
2487         jl = kzalloc(sizeof(struct reiserfs_journal_list),
2488                      GFP_NOFS | __GFP_NOFAIL);
2489         INIT_LIST_HEAD(&jl->j_list);
2490         INIT_LIST_HEAD(&jl->j_working_list);
2491         INIT_LIST_HEAD(&jl->j_tail_bh_list);
2492         INIT_LIST_HEAD(&jl->j_bh_list);
2493         mutex_init(&jl->j_commit_mutex);
2494         SB_JOURNAL(s)->j_num_lists++;
2495         get_journal_list(jl);
2496         return jl;
2497 }
2498
2499 static void journal_list_init(struct super_block *sb)
2500 {
2501         SB_JOURNAL(sb)->j_current_jl = alloc_journal_list(sb);
2502 }
2503
2504 static void release_journal_dev(struct super_block *super,
2505                                struct reiserfs_journal *journal)
2506 {
2507         if (journal->j_dev_bd != NULL) {
2508                 blkdev_put(journal->j_dev_bd, journal->j_dev_mode);
2509                 journal->j_dev_bd = NULL;
2510         }
2511 }
2512
2513 static int journal_init_dev(struct super_block *super,
2514                             struct reiserfs_journal *journal,
2515                             const char *jdev_name)
2516 {
2517         int result;
2518         dev_t jdev;
2519         fmode_t blkdev_mode = FMODE_READ | FMODE_WRITE | FMODE_EXCL;
2520         char b[BDEVNAME_SIZE];
2521
2522         result = 0;
2523
2524         journal->j_dev_bd = NULL;
2525         jdev = SB_ONDISK_JOURNAL_DEVICE(super) ?
2526             new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super)) : super->s_dev;
2527
2528         if (bdev_read_only(super->s_bdev))
2529                 blkdev_mode = FMODE_READ;
2530
2531         /* there is no "jdev" option and journal is on separate device */
2532         if ((!jdev_name || !jdev_name[0])) {
2533                 if (jdev == super->s_dev)
2534                         blkdev_mode &= ~FMODE_EXCL;
2535                 journal->j_dev_bd = blkdev_get_by_dev(jdev, blkdev_mode,
2536                                                       journal);
2537                 journal->j_dev_mode = blkdev_mode;
2538                 if (IS_ERR(journal->j_dev_bd)) {
2539                         result = PTR_ERR(journal->j_dev_bd);
2540                         journal->j_dev_bd = NULL;
2541                         reiserfs_warning(super, "sh-458",
2542                                          "cannot init journal device '%s': %i",
2543                                          __bdevname(jdev, b), result);
2544                         return result;
2545                 } else if (jdev != super->s_dev)
2546                         set_blocksize(journal->j_dev_bd, super->s_blocksize);
2547
2548                 return 0;
2549         }
2550
2551         journal->j_dev_mode = blkdev_mode;
2552         journal->j_dev_bd = blkdev_get_by_path(jdev_name, blkdev_mode, journal);
2553         if (IS_ERR(journal->j_dev_bd)) {
2554                 result = PTR_ERR(journal->j_dev_bd);
2555                 journal->j_dev_bd = NULL;
2556                 reiserfs_warning(super,
2557                                  "journal_init_dev: Cannot open '%s': %i",
2558                                  jdev_name, result);
2559                 return result;
2560         }
2561
2562         set_blocksize(journal->j_dev_bd, super->s_blocksize);
2563         reiserfs_info(super,
2564                       "journal_init_dev: journal device: %s\n",
2565                       bdevname(journal->j_dev_bd, b));
2566         return 0;
2567 }
2568
2569 /**
2570  * When creating/tuning a file system user can assign some
2571  * journal params within boundaries which depend on the ratio
2572  * blocksize/standard_blocksize.
2573  *
2574  * For blocks >= standard_blocksize transaction size should
2575  * be not less then JOURNAL_TRANS_MIN_DEFAULT, and not more
2576  * then JOURNAL_TRANS_MAX_DEFAULT.
2577  *
2578  * For blocks < standard_blocksize these boundaries should be
2579  * decreased proportionally.
2580  */
2581 #define REISERFS_STANDARD_BLKSIZE (4096)
2582
2583 static int check_advise_trans_params(struct super_block *sb,
2584                                      struct reiserfs_journal *journal)
2585 {
2586         if (journal->j_trans_max) {
2587                 /* Non-default journal params.
2588                    Do sanity check for them. */
2589                 int ratio = 1;
2590                 if (sb->s_blocksize < REISERFS_STANDARD_BLKSIZE)
2591                         ratio = REISERFS_STANDARD_BLKSIZE / sb->s_blocksize;
2592
2593                 if (journal->j_trans_max > JOURNAL_TRANS_MAX_DEFAULT / ratio ||
2594                     journal->j_trans_max < JOURNAL_TRANS_MIN_DEFAULT / ratio ||
2595                     SB_ONDISK_JOURNAL_SIZE(sb) / journal->j_trans_max <
2596                     JOURNAL_MIN_RATIO) {
2597                         reiserfs_warning(sb, "sh-462",
2598                                          "bad transaction max size (%u). "
2599                                          "FSCK?", journal->j_trans_max);
2600                         return 1;
2601                 }
2602                 if (journal->j_max_batch != (journal->j_trans_max) *
2603                         JOURNAL_MAX_BATCH_DEFAULT/JOURNAL_TRANS_MAX_DEFAULT) {
2604                         reiserfs_warning(sb, "sh-463",
2605                                          "bad transaction max batch (%u). "
2606                                          "FSCK?", journal->j_max_batch);
2607                         return 1;
2608                 }
2609         } else {
2610                 /* Default journal params.
2611                    The file system was created by old version
2612                    of mkreiserfs, so some fields contain zeros,
2613                    and we need to advise proper values for them */
2614                 if (sb->s_blocksize != REISERFS_STANDARD_BLKSIZE) {
2615                         reiserfs_warning(sb, "sh-464", "bad blocksize (%u)",
2616                                          sb->s_blocksize);
2617                         return 1;
2618                 }
2619                 journal->j_trans_max = JOURNAL_TRANS_MAX_DEFAULT;
2620                 journal->j_max_batch = JOURNAL_MAX_BATCH_DEFAULT;
2621                 journal->j_max_commit_age = JOURNAL_MAX_COMMIT_AGE;
2622         }
2623         return 0;
2624 }
2625
2626 /*
2627 ** must be called once on fs mount.  calls journal_read for you
2628 */
2629 int journal_init(struct super_block *sb, const char *j_dev_name,
2630                  int old_format, unsigned int commit_max_age)
2631 {
2632         int num_cnodes = SB_ONDISK_JOURNAL_SIZE(sb) * 2;
2633         struct buffer_head *bhjh;
2634         struct reiserfs_super_block *rs;
2635         struct reiserfs_journal_header *jh;
2636         struct reiserfs_journal *journal;
2637         struct reiserfs_journal_list *jl;
2638         char b[BDEVNAME_SIZE];
2639         int ret;
2640
2641         journal = SB_JOURNAL(sb) = vzalloc(sizeof(struct reiserfs_journal));
2642         if (!journal) {
2643                 reiserfs_warning(sb, "journal-1256",
2644                                  "unable to get memory for journal structure");
2645                 return 1;
2646         }
2647         INIT_LIST_HEAD(&journal->j_bitmap_nodes);
2648         INIT_LIST_HEAD(&journal->j_prealloc_list);
2649         INIT_LIST_HEAD(&journal->j_working_list);
2650         INIT_LIST_HEAD(&journal->j_journal_list);
2651         journal->j_persistent_trans = 0;
2652         if (reiserfs_allocate_list_bitmaps(sb, journal->j_list_bitmap,
2653                                            reiserfs_bmap_count(sb)))
2654                 goto free_and_return;
2655
2656         allocate_bitmap_nodes(sb);
2657
2658         /* reserved for journal area support */
2659         SB_JOURNAL_1st_RESERVED_BLOCK(sb) = (old_format ?
2660                                                  REISERFS_OLD_DISK_OFFSET_IN_BYTES
2661                                                  / sb->s_blocksize +
2662                                                  reiserfs_bmap_count(sb) +
2663                                                  1 :
2664                                                  REISERFS_DISK_OFFSET_IN_BYTES /
2665                                                  sb->s_blocksize + 2);
2666
2667         /* Sanity check to see is the standard journal fitting within first bitmap
2668            (actual for small blocksizes) */
2669         if (!SB_ONDISK_JOURNAL_DEVICE(sb) &&
2670             (SB_JOURNAL_1st_RESERVED_BLOCK(sb) +
2671              SB_ONDISK_JOURNAL_SIZE(sb) > sb->s_blocksize * 8)) {
2672                 reiserfs_warning(sb, "journal-1393",
2673                                  "journal does not fit for area addressed "
2674                                  "by first of bitmap blocks. It starts at "
2675                                  "%u and its size is %u. Block size %ld",
2676                                  SB_JOURNAL_1st_RESERVED_BLOCK(sb),
2677                                  SB_ONDISK_JOURNAL_SIZE(sb),
2678                                  sb->s_blocksize);
2679                 goto free_and_return;
2680         }
2681
2682         if (journal_init_dev(sb, journal, j_dev_name) != 0) {
2683                 reiserfs_warning(sb, "sh-462",
2684                                  "unable to initialize jornal device");
2685                 goto free_and_return;
2686         }
2687
2688         rs = SB_DISK_SUPER_BLOCK(sb);
2689
2690         /* read journal header */
2691         bhjh = journal_bread(sb,
2692                              SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2693                              SB_ONDISK_JOURNAL_SIZE(sb));
2694         if (!bhjh) {
2695                 reiserfs_warning(sb, "sh-459",
2696                                  "unable to read journal header");
2697                 goto free_and_return;
2698         }
2699         jh = (struct reiserfs_journal_header *)(bhjh->b_data);
2700
2701         /* make sure that journal matches to the super block */
2702         if (is_reiserfs_jr(rs)
2703             && (le32_to_cpu(jh->jh_journal.jp_journal_magic) !=
2704                 sb_jp_journal_magic(rs))) {
2705                 reiserfs_warning(sb, "sh-460",
2706                                  "journal header magic %x (device %s) does "
2707                                  "not match to magic found in super block %x",
2708                                  jh->jh_journal.jp_journal_magic,
2709                                  bdevname(journal->j_dev_bd, b),
2710                                  sb_jp_journal_magic(rs));
2711                 brelse(bhjh);
2712                 goto free_and_return;
2713         }
2714
2715         journal->j_trans_max = le32_to_cpu(jh->jh_journal.jp_journal_trans_max);
2716         journal->j_max_batch = le32_to_cpu(jh->jh_journal.jp_journal_max_batch);
2717         journal->j_max_commit_age =
2718             le32_to_cpu(jh->jh_journal.jp_journal_max_commit_age);
2719         journal->j_max_trans_age = JOURNAL_MAX_TRANS_AGE;
2720
2721         if (check_advise_trans_params(sb, journal) != 0)
2722                 goto free_and_return;
2723         journal->j_default_max_commit_age = journal->j_max_commit_age;
2724
2725         if (commit_max_age != 0) {
2726                 journal->j_max_commit_age = commit_max_age;
2727                 journal->j_max_trans_age = commit_max_age;
2728         }
2729
2730         reiserfs_info(sb, "journal params: device %s, size %u, "
2731                       "journal first block %u, max trans len %u, max batch %u, "
2732                       "max commit age %u, max trans age %u\n",
2733                       bdevname(journal->j_dev_bd, b),
2734                       SB_ONDISK_JOURNAL_SIZE(sb),
2735                       SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2736                       journal->j_trans_max,
2737                       journal->j_max_batch,
2738                       journal->j_max_commit_age, journal->j_max_trans_age);
2739
2740         brelse(bhjh);
2741
2742         journal->j_list_bitmap_index = 0;
2743         journal_list_init(sb);
2744
2745         memset(journal->j_list_hash_table, 0,
2746                JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
2747
2748         INIT_LIST_HEAD(&journal->j_dirty_buffers);
2749         spin_lock_init(&journal->j_dirty_buffers_lock);
2750
2751         journal->j_start = 0;
2752         journal->j_len = 0;
2753         journal->j_len_alloc = 0;
2754         atomic_set(&(journal->j_wcount), 0);
2755         atomic_set(&(journal->j_async_throttle), 0);
2756         journal->j_bcount = 0;
2757         journal->j_trans_start_time = 0;
2758         journal->j_last = NULL;
2759         journal->j_first = NULL;
2760         init_waitqueue_head(&(journal->j_join_wait));
2761         mutex_init(&journal->j_mutex);
2762         mutex_init(&journal->j_flush_mutex);
2763
2764         journal->j_trans_id = 10;
2765         journal->j_mount_id = 10;
2766         journal->j_state = 0;
2767         atomic_set(&(journal->j_jlock), 0);
2768         journal->j_cnode_free_list = allocate_cnodes(num_cnodes);
2769         journal->j_cnode_free_orig = journal->j_cnode_free_list;
2770         journal->j_cnode_free = journal->j_cnode_free_list ? num_cnodes : 0;
2771         journal->j_cnode_used = 0;
2772         journal->j_must_wait = 0;
2773
2774         if (journal->j_cnode_free == 0) {
2775                 reiserfs_warning(sb, "journal-2004", "Journal cnode memory "
2776                                  "allocation failed (%ld bytes). Journal is "
2777                                  "too large for available memory. Usually "
2778                                  "this is due to a journal that is too large.",
2779                                  sizeof (struct reiserfs_journal_cnode) * num_cnodes);
2780                 goto free_and_return;
2781         }
2782
2783         init_journal_hash(sb);
2784         jl = journal->j_current_jl;
2785
2786         /*
2787          * get_list_bitmap() may call flush_commit_list() which
2788          * requires the lock. Calling flush_commit_list() shouldn't happen
2789          * this early but I like to be paranoid.
2790          */
2791         reiserfs_write_lock(sb);
2792         jl->j_list_bitmap = get_list_bitmap(sb, jl);
2793         reiserfs_write_unlock(sb);
2794         if (!jl->j_list_bitmap) {
2795                 reiserfs_warning(sb, "journal-2005",
2796                                  "get_list_bitmap failed for journal list 0");
2797                 goto free_and_return;
2798         }
2799
2800         ret = journal_read(sb);
2801         if (ret < 0) {
2802                 reiserfs_warning(sb, "reiserfs-2006",
2803                                  "Replay Failure, unable to mount");
2804                 goto free_and_return;
2805         }
2806
2807         reiserfs_mounted_fs_count++;
2808         if (reiserfs_mounted_fs_count <= 1)
2809                 commit_wq = alloc_workqueue("reiserfs", WQ_MEM_RECLAIM, 0);
2810
2811         INIT_DELAYED_WORK(&journal->j_work, flush_async_commits);
2812         journal->j_work_sb = sb;
2813         return 0;
2814       free_and_return:
2815         free_journal_ram(sb);
2816         return 1;
2817 }
2818
2819 /*
2820 ** test for a polite end of the current transaction.  Used by file_write, and should
2821 ** be used by delete to make sure they don't write more than can fit inside a single
2822 ** transaction
2823 */
2824 int journal_transaction_should_end(struct reiserfs_transaction_handle *th,
2825                                    int new_alloc)
2826 {
2827         struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2828         time_t now = get_seconds();
2829         /* cannot restart while nested */
2830         BUG_ON(!th->t_trans_id);
2831         if (th->t_refcount > 1)
2832                 return 0;
2833         if (journal->j_must_wait > 0 ||
2834             (journal->j_len_alloc + new_alloc) >= journal->j_max_batch ||
2835             atomic_read(&(journal->j_jlock)) ||
2836             (now - journal->j_trans_start_time) > journal->j_max_trans_age ||
2837             journal->j_cnode_free < (journal->j_trans_max * 3)) {
2838                 return 1;
2839         }
2840
2841         journal->j_len_alloc += new_alloc;
2842         th->t_blocks_allocated += new_alloc ;
2843         return 0;
2844 }
2845
2846 /* this must be called inside a transaction
2847 */
2848 void reiserfs_block_writes(struct reiserfs_transaction_handle *th)
2849 {
2850         struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2851         BUG_ON(!th->t_trans_id);
2852         journal->j_must_wait = 1;
2853         set_bit(J_WRITERS_BLOCKED, &journal->j_state);
2854         return;
2855 }
2856
2857 /* this must be called without a transaction started
2858 */
2859 void reiserfs_allow_writes(struct super_block *s)
2860 {
2861         struct reiserfs_journal *journal = SB_JOURNAL(s);
2862         clear_bit(J_WRITERS_BLOCKED, &journal->j_state);
2863         wake_up(&journal->j_join_wait);
2864 }
2865
2866 /* this must be called without a transaction started
2867 */
2868 void reiserfs_wait_on_write_block(struct super_block *s)
2869 {
2870         struct reiserfs_journal *journal = SB_JOURNAL(s);
2871         wait_event(journal->j_join_wait,
2872                    !test_bit(J_WRITERS_BLOCKED, &journal->j_state));
2873 }
2874
2875 static void queue_log_writer(struct super_block *s)
2876 {
2877         wait_queue_t wait;
2878         struct reiserfs_journal *journal = SB_JOURNAL(s);
2879         set_bit(J_WRITERS_QUEUED, &journal->j_state);
2880
2881         /*
2882          * we don't want to use wait_event here because
2883          * we only want to wait once.
2884          */
2885         init_waitqueue_entry(&wait, current);
2886         add_wait_queue(&journal->j_join_wait, &wait);
2887         set_current_state(TASK_UNINTERRUPTIBLE);
2888         if (test_bit(J_WRITERS_QUEUED, &journal->j_state)) {
2889                 int depth = reiserfs_write_unlock_nested(s);
2890                 schedule();
2891                 reiserfs_write_lock_nested(s, depth);
2892         }
2893         __set_current_state(TASK_RUNNING);
2894         remove_wait_queue(&journal->j_join_wait, &wait);
2895 }
2896
2897 static void wake_queued_writers(struct super_block *s)
2898 {
2899         struct reiserfs_journal *journal = SB_JOURNAL(s);
2900         if (test_and_clear_bit(J_WRITERS_QUEUED, &journal->j_state))
2901                 wake_up(&journal->j_join_wait);
2902 }
2903
2904 static void let_transaction_grow(struct super_block *sb, unsigned int trans_id)
2905 {
2906         struct reiserfs_journal *journal = SB_JOURNAL(sb);
2907         unsigned long bcount = journal->j_bcount;
2908         while (1) {
2909                 int depth;
2910
2911                 depth = reiserfs_write_unlock_nested(sb);
2912                 schedule_timeout_uninterruptible(1);
2913                 reiserfs_write_lock_nested(sb, depth);
2914
2915                 journal->j_current_jl->j_state |= LIST_COMMIT_PENDING;
2916                 while ((atomic_read(&journal->j_wcount) > 0 ||
2917                         atomic_read(&journal->j_jlock)) &&
2918                        journal->j_trans_id == trans_id) {
2919                         queue_log_writer(sb);
2920                 }
2921                 if (journal->j_trans_id != trans_id)
2922                         break;
2923                 if (bcount == journal->j_bcount)
2924                         break;
2925                 bcount = journal->j_bcount;
2926         }
2927 }
2928
2929 /* join == true if you must join an existing transaction.
2930 ** join == false if you can deal with waiting for others to finish
2931 **
2932 ** this will block until the transaction is joinable.  send the number of blocks you
2933 ** expect to use in nblocks.
2934 */
2935 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
2936                               struct super_block *sb, unsigned long nblocks,
2937                               int join)
2938 {
2939         time_t now = get_seconds();
2940         unsigned int old_trans_id;
2941         struct reiserfs_journal *journal = SB_JOURNAL(sb);
2942         struct reiserfs_transaction_handle myth;
2943         int sched_count = 0;
2944         int retval;
2945         int depth;
2946
2947         reiserfs_check_lock_depth(sb, "journal_begin");
2948         BUG_ON(nblocks > journal->j_trans_max);
2949
2950         PROC_INFO_INC(sb, journal.journal_being);
2951         /* set here for journal_join */
2952         th->t_refcount = 1;
2953         th->t_super = sb;
2954
2955       relock:
2956         lock_journal(sb);
2957         if (join != JBEGIN_ABORT && reiserfs_is_journal_aborted(journal)) {
2958                 unlock_journal(sb);
2959                 retval = journal->j_errno;
2960                 goto out_fail;
2961         }
2962         journal->j_bcount++;
2963
2964         if (test_bit(J_WRITERS_BLOCKED, &journal->j_state)) {
2965                 unlock_journal(sb);
2966                 depth = reiserfs_write_unlock_nested(sb);
2967                 reiserfs_wait_on_write_block(sb);
2968                 reiserfs_write_lock_nested(sb, depth);
2969                 PROC_INFO_INC(sb, journal.journal_relock_writers);
2970                 goto relock;
2971         }
2972         now = get_seconds();
2973
2974         /* if there is no room in the journal OR
2975          ** if this transaction is too old, and we weren't called joinable, wait for it to finish before beginning
2976          ** we don't sleep if there aren't other writers
2977          */
2978
2979         if ((!join && journal->j_must_wait > 0) ||
2980             (!join
2981              && (journal->j_len_alloc + nblocks + 2) >= journal->j_max_batch)
2982             || (!join && atomic_read(&journal->j_wcount) > 0
2983                 && journal->j_trans_start_time > 0
2984                 && (now - journal->j_trans_start_time) >
2985                 journal->j_max_trans_age) || (!join
2986                                               && atomic_read(&journal->j_jlock))
2987             || (!join && journal->j_cnode_free < (journal->j_trans_max * 3))) {
2988
2989                 old_trans_id = journal->j_trans_id;
2990                 unlock_journal(sb);     /* allow others to finish this transaction */
2991
2992                 if (!join && (journal->j_len_alloc + nblocks + 2) >=
2993                     journal->j_max_batch &&
2994                     ((journal->j_len + nblocks + 2) * 100) <
2995                     (journal->j_len_alloc * 75)) {
2996                         if (atomic_read(&journal->j_wcount) > 10) {
2997                                 sched_count++;
2998                                 queue_log_writer(sb);
2999                                 goto relock;
3000                         }
3001                 }
3002                 /* don't mess with joining the transaction if all we have to do is
3003                  * wait for someone else to do a commit
3004                  */
3005                 if (atomic_read(&journal->j_jlock)) {
3006                         while (journal->j_trans_id == old_trans_id &&
3007                                atomic_read(&journal->j_jlock)) {
3008                                 queue_log_writer(sb);
3009                         }
3010                         goto relock;
3011                 }
3012                 retval = journal_join(&myth, sb, 1);
3013                 if (retval)
3014                         goto out_fail;
3015
3016                 /* someone might have ended the transaction while we joined */
3017                 if (old_trans_id != journal->j_trans_id) {
3018                         retval = do_journal_end(&myth, sb, 1, 0);
3019                 } else {
3020                         retval = do_journal_end(&myth, sb, 1, COMMIT_NOW);
3021                 }
3022
3023                 if (retval)
3024                         goto out_fail;
3025
3026                 PROC_INFO_INC(sb, journal.journal_relock_wcount);
3027                 goto relock;
3028         }
3029         /* we are the first writer, set trans_id */
3030         if (journal->j_trans_start_time == 0) {
3031                 journal->j_trans_start_time = get_seconds();
3032         }
3033         atomic_inc(&(journal->j_wcount));
3034         journal->j_len_alloc += nblocks;
3035         th->t_blocks_logged = 0;
3036         th->t_blocks_allocated = nblocks;
3037         th->t_trans_id = journal->j_trans_id;
3038         unlock_journal(sb);
3039         INIT_LIST_HEAD(&th->t_list);
3040         return 0;
3041
3042       out_fail:
3043         memset(th, 0, sizeof(*th));
3044         /* Re-set th->t_super, so we can properly keep track of how many
3045          * persistent transactions there are. We need to do this so if this
3046          * call is part of a failed restart_transaction, we can free it later */
3047         th->t_super = sb;
3048         return retval;
3049 }
3050
3051 struct reiserfs_transaction_handle *reiserfs_persistent_transaction(struct
3052                                                                     super_block
3053                                                                     *s,
3054                                                                     int nblocks)
3055 {
3056         int ret;
3057         struct reiserfs_transaction_handle *th;
3058
3059         /* if we're nesting into an existing transaction.  It will be
3060          ** persistent on its own
3061          */
3062         if (reiserfs_transaction_running(s)) {
3063                 th = current->journal_info;
3064                 th->t_refcount++;
3065                 BUG_ON(th->t_refcount < 2);
3066
3067                 return th;
3068         }
3069         th = kmalloc(sizeof(struct reiserfs_transaction_handle), GFP_NOFS);
3070         if (!th)
3071                 return NULL;
3072         ret = journal_begin(th, s, nblocks);
3073         if (ret) {
3074                 kfree(th);
3075                 return NULL;
3076         }
3077
3078         SB_JOURNAL(s)->j_persistent_trans++;
3079         return th;
3080 }
3081
3082 int reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle *th)
3083 {
3084         struct super_block *s = th->t_super;
3085         int ret = 0;
3086         if (th->t_trans_id)
3087                 ret = journal_end(th, th->t_super, th->t_blocks_allocated);
3088         else
3089                 ret = -EIO;
3090         if (th->t_refcount == 0) {
3091                 SB_JOURNAL(s)->j_persistent_trans--;
3092                 kfree(th);
3093         }
3094         return ret;
3095 }
3096
3097 static int journal_join(struct reiserfs_transaction_handle *th,
3098                         struct super_block *sb, unsigned long nblocks)
3099 {
3100         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3101
3102         /* this keeps do_journal_end from NULLing out the current->journal_info
3103          ** pointer
3104          */
3105         th->t_handle_save = cur_th;
3106         BUG_ON(cur_th && cur_th->t_refcount > 1);
3107         return do_journal_begin_r(th, sb, nblocks, JBEGIN_JOIN);
3108 }
3109
3110 int journal_join_abort(struct reiserfs_transaction_handle *th,
3111                        struct super_block *sb, unsigned long nblocks)
3112 {
3113         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3114
3115         /* this keeps do_journal_end from NULLing out the current->journal_info
3116          ** pointer
3117          */
3118         th->t_handle_save = cur_th;
3119         BUG_ON(cur_th && cur_th->t_refcount > 1);
3120         return do_journal_begin_r(th, sb, nblocks, JBEGIN_ABORT);
3121 }
3122
3123 int journal_begin(struct reiserfs_transaction_handle *th,
3124                   struct super_block *sb, unsigned long nblocks)
3125 {
3126         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3127         int ret;
3128
3129         th->t_handle_save = NULL;
3130         if (cur_th) {
3131                 /* we are nesting into the current transaction */
3132                 if (cur_th->t_super == sb) {
3133                         BUG_ON(!cur_th->t_refcount);
3134                         cur_th->t_refcount++;
3135                         memcpy(th, cur_th, sizeof(*th));
3136                         if (th->t_refcount <= 1)
3137                                 reiserfs_warning(sb, "reiserfs-2005",
3138                                                  "BAD: refcount <= 1, but "
3139                                                  "journal_info != 0");
3140                         return 0;
3141                 } else {
3142                         /* we've ended up with a handle from a different filesystem.
3143                          ** save it and restore on journal_end.  This should never
3144                          ** really happen...
3145                          */
3146                         reiserfs_warning(sb, "clm-2100",
3147                                          "nesting info a different FS");
3148                         th->t_handle_save = current->journal_info;
3149                         current->journal_info = th;
3150                 }
3151         } else {
3152                 current->journal_info = th;
3153         }
3154         ret = do_journal_begin_r(th, sb, nblocks, JBEGIN_REG);
3155         BUG_ON(current->journal_info != th);
3156
3157         /* I guess this boils down to being the reciprocal of clm-2100 above.
3158          * If do_journal_begin_r fails, we need to put it back, since journal_end
3159          * won't be called to do it. */
3160         if (ret)
3161                 current->journal_info = th->t_handle_save;
3162         else
3163                 BUG_ON(!th->t_refcount);
3164
3165         return ret;
3166 }
3167
3168 /*
3169 ** puts bh into the current transaction.  If it was already there, reorders removes the
3170 ** old pointers from the hash, and puts new ones in (to make sure replay happen in the right order).
3171 **
3172 ** if it was dirty, cleans and files onto the clean list.  I can't let it be dirty again until the
3173 ** transaction is committed.
3174 **
3175 ** if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len.
3176 */
3177 int journal_mark_dirty(struct reiserfs_transaction_handle *th,
3178                        struct super_block *sb, struct buffer_head *bh)
3179 {
3180         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3181         struct reiserfs_journal_cnode *cn = NULL;
3182         int count_already_incd = 0;
3183         int prepared = 0;
3184         BUG_ON(!th->t_trans_id);
3185
3186         PROC_INFO_INC(sb, journal.mark_dirty);
3187         if (th->t_trans_id != journal->j_trans_id) {
3188                 reiserfs_panic(th->t_super, "journal-1577",
3189                                "handle trans id %ld != current trans id %ld",
3190                                th->t_trans_id, journal->j_trans_id);
3191         }
3192
3193         prepared = test_clear_buffer_journal_prepared(bh);
3194         clear_buffer_journal_restore_dirty(bh);
3195         /* already in this transaction, we are done */
3196         if (buffer_journaled(bh)) {
3197                 PROC_INFO_INC(sb, journal.mark_dirty_already);
3198                 return 0;
3199         }
3200
3201         /* this must be turned into a panic instead of a warning.  We can't allow
3202          ** a dirty or journal_dirty or locked buffer to be logged, as some changes
3203          ** could get to disk too early.  NOT GOOD.
3204          */
3205         if (!prepared || buffer_dirty(bh)) {
3206                 reiserfs_warning(sb, "journal-1777",
3207                                  "buffer %llu bad state "
3208                                  "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT",
3209                                  (unsigned long long)bh->b_blocknr,
3210                                  prepared ? ' ' : '!',
3211                                  buffer_locked(bh) ? ' ' : '!',
3212                                  buffer_dirty(bh) ? ' ' : '!',
3213                                  buffer_journal_dirty(bh) ? ' ' : '!');
3214         }
3215
3216         if (atomic_read(&(journal->j_wcount)) <= 0) {
3217                 reiserfs_warning(sb, "journal-1409",
3218                                  "returning because j_wcount was %d",
3219                                  atomic_read(&(journal->j_wcount)));
3220                 return 1;
3221         }
3222         /* this error means I've screwed up, and we've overflowed the transaction.
3223          ** Nothing can be done here, except make the FS readonly or panic.
3224          */
3225         if (journal->j_len >= journal->j_trans_max) {
3226                 reiserfs_panic(th->t_super, "journal-1413",
3227                                "j_len (%lu) is too big",
3228                                journal->j_len);
3229         }
3230
3231         if (buffer_journal_dirty(bh)) {
3232                 count_already_incd = 1;
3233                 PROC_INFO_INC(sb, journal.mark_dirty_notjournal);
3234                 clear_buffer_journal_dirty(bh);
3235         }
3236
3237         if (journal->j_len > journal->j_len_alloc) {
3238                 journal->j_len_alloc = journal->j_len + JOURNAL_PER_BALANCE_CNT;
3239         }
3240
3241         set_buffer_journaled(bh);
3242
3243         /* now put this guy on the end */
3244         if (!cn) {
3245                 cn = get_cnode(sb);
3246                 if (!cn) {
3247                         reiserfs_panic(sb, "journal-4", "get_cnode failed!");
3248                 }
3249
3250                 if (th->t_blocks_logged == th->t_blocks_allocated) {
3251                         th->t_blocks_allocated += JOURNAL_PER_BALANCE_CNT;
3252                         journal->j_len_alloc += JOURNAL_PER_BALANCE_CNT;
3253                 }
3254                 th->t_blocks_logged++;
3255                 journal->j_len++;
3256
3257                 cn->bh = bh;
3258                 cn->blocknr = bh->b_blocknr;
3259                 cn->sb = sb;
3260                 cn->jlist = NULL;
3261                 insert_journal_hash(journal->j_hash_table, cn);
3262                 if (!count_already_incd) {
3263                         get_bh(bh);
3264                 }
3265         }
3266         cn->next = NULL;
3267         cn->prev = journal->j_last;
3268         cn->bh = bh;
3269         if (journal->j_last) {
3270                 journal->j_last->next = cn;
3271                 journal->j_last = cn;
3272         } else {
3273                 journal->j_first = cn;
3274                 journal->j_last = cn;
3275         }
3276         reiserfs_schedule_old_flush(sb);
3277         return 0;
3278 }
3279
3280 int journal_end(struct reiserfs_transaction_handle *th,
3281                 struct super_block *sb, unsigned long nblocks)
3282 {
3283         if (!current->journal_info && th->t_refcount > 1)
3284                 reiserfs_warning(sb, "REISER-NESTING",
3285                                  "th NULL, refcount %d", th->t_refcount);
3286
3287         if (!th->t_trans_id) {
3288                 WARN_ON(1);
3289                 return -EIO;
3290         }
3291
3292         th->t_refcount--;
3293         if (th->t_refcount > 0) {
3294                 struct reiserfs_transaction_handle *cur_th =
3295                     current->journal_info;
3296
3297                 /* we aren't allowed to close a nested transaction on a different
3298                  ** filesystem from the one in the task struct
3299                  */
3300                 BUG_ON(cur_th->t_super != th->t_super);
3301
3302                 if (th != cur_th) {
3303                         memcpy(current->journal_info, th, sizeof(*th));
3304                         th->t_trans_id = 0;
3305                 }
3306                 return 0;
3307         } else {
3308                 return do_journal_end(th, sb, nblocks, 0);
3309         }
3310 }
3311
3312 /* removes from the current transaction, relsing and descrementing any counters.
3313 ** also files the removed buffer directly onto the clean list
3314 **
3315 ** called by journal_mark_freed when a block has been deleted
3316 **
3317 ** returns 1 if it cleaned and relsed the buffer. 0 otherwise
3318 */
3319 static int remove_from_transaction(struct super_block *sb,
3320                                    b_blocknr_t blocknr, int already_cleaned)
3321 {
3322         struct buffer_head *bh;
3323         struct reiserfs_journal_cnode *cn;
3324         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3325         int ret = 0;
3326
3327         cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr);
3328         if (!cn || !cn->bh) {
3329                 return ret;
3330         }
3331         bh = cn->bh;
3332         if (cn->prev) {
3333                 cn->prev->next = cn->next;
3334         }
3335         if (cn->next) {
3336                 cn->next->prev = cn->prev;
3337         }
3338         if (cn == journal->j_first) {
3339                 journal->j_first = cn->next;
3340         }
3341         if (cn == journal->j_last) {
3342                 journal->j_last = cn->prev;
3343         }
3344         if (bh)
3345                 remove_journal_hash(sb, journal->j_hash_table, NULL,
3346                                     bh->b_blocknr, 0);
3347         clear_buffer_journaled(bh);     /* don't log this one */
3348
3349         if (!already_cleaned) {
3350                 clear_buffer_journal_dirty(bh);
3351                 clear_buffer_dirty(bh);
3352                 clear_buffer_journal_test(bh);
3353                 put_bh(bh);
3354                 if (atomic_read(&(bh->b_count)) < 0) {
3355                         reiserfs_warning(sb, "journal-1752",
3356                                          "b_count < 0");
3357                 }
3358                 ret = 1;
3359         }
3360         journal->j_len--;
3361         journal->j_len_alloc--;
3362         free_cnode(sb, cn);
3363         return ret;
3364 }
3365
3366 /*
3367 ** for any cnode in a journal list, it can only be dirtied of all the
3368 ** transactions that include it are committed to disk.
3369 ** this checks through each transaction, and returns 1 if you are allowed to dirty,
3370 ** and 0 if you aren't
3371 **
3372 ** it is called by dirty_journal_list, which is called after flush_commit_list has gotten all the log
3373 ** blocks for a given transaction on disk
3374 **
3375 */
3376 static int can_dirty(struct reiserfs_journal_cnode *cn)
3377 {
3378         struct super_block *sb = cn->sb;
3379         b_blocknr_t blocknr = cn->blocknr;
3380         struct reiserfs_journal_cnode *cur = cn->hprev;
3381         int can_dirty = 1;
3382
3383         /* first test hprev.  These are all newer than cn, so any node here
3384          ** with the same block number and dev means this node can't be sent
3385          ** to disk right now.
3386          */
3387         while (cur && can_dirty) {
3388                 if (cur->jlist && cur->bh && cur->blocknr && cur->sb == sb &&
3389                     cur->blocknr == blocknr) {
3390                         can_dirty = 0;
3391                 }
3392                 cur = cur->hprev;
3393         }
3394         /* then test hnext.  These are all older than cn.  As long as they
3395          ** are committed to the log, it is safe to write cn to disk
3396          */
3397         cur = cn->hnext;
3398         while (cur && can_dirty) {
3399                 if (cur->jlist && cur->jlist->j_len > 0 &&
3400                     atomic_read(&(cur->jlist->j_commit_left)) > 0 && cur->bh &&
3401                     cur->blocknr && cur->sb == sb && cur->blocknr == blocknr) {
3402                         can_dirty = 0;
3403                 }
3404                 cur = cur->hnext;
3405         }
3406         return can_dirty;
3407 }
3408
3409 /* syncs the commit blocks, but does not force the real buffers to disk
3410 ** will wait until the current transaction is done/committed before returning
3411 */
3412 int journal_end_sync(struct reiserfs_transaction_handle *th,
3413                      struct super_block *sb, unsigned long nblocks)
3414 {
3415         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3416
3417         BUG_ON(!th->t_trans_id);
3418         /* you can sync while nested, very, very bad */
3419         BUG_ON(th->t_refcount > 1);
3420         if (journal->j_len == 0) {
3421                 reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3422                                              1);
3423                 journal_mark_dirty(th, sb, SB_BUFFER_WITH_SB(sb));
3424         }
3425         return do_journal_end(th, sb, nblocks, COMMIT_NOW | WAIT);
3426 }
3427
3428 /*
3429 ** writeback the pending async commits to disk
3430 */
3431 static void flush_async_commits(struct work_struct *work)
3432 {
3433         struct reiserfs_journal *journal =
3434                 container_of(work, struct reiserfs_journal, j_work.work);
3435         struct super_block *sb = journal->j_work_sb;
3436         struct reiserfs_journal_list *jl;
3437         struct list_head *entry;
3438
3439         reiserfs_write_lock(sb);
3440         if (!list_empty(&journal->j_journal_list)) {
3441                 /* last entry is the youngest, commit it and you get everything */
3442                 entry = journal->j_journal_list.prev;
3443                 jl = JOURNAL_LIST_ENTRY(entry);
3444                 flush_commit_list(sb, jl, 1);
3445         }
3446         reiserfs_write_unlock(sb);
3447 }
3448
3449 /*
3450 ** flushes any old transactions to disk
3451 ** ends the current transaction if it is too old
3452 */
3453 void reiserfs_flush_old_commits(struct super_block *sb)
3454 {
3455         time_t now;
3456         struct reiserfs_transaction_handle th;
3457         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3458
3459         now = get_seconds();
3460         /* safety check so we don't flush while we are replaying the log during
3461          * mount
3462          */
3463         if (list_empty(&journal->j_journal_list))
3464                 return;
3465
3466         /* check the current transaction.  If there are no writers, and it is
3467          * too old, finish it, and force the commit blocks to disk
3468          */
3469         if (atomic_read(&journal->j_wcount) <= 0 &&
3470             journal->j_trans_start_time > 0 &&
3471             journal->j_len > 0 &&
3472             (now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3473                 if (!journal_join(&th, sb, 1)) {
3474                         reiserfs_prepare_for_journal(sb,
3475                                                      SB_BUFFER_WITH_SB(sb),
3476                                                      1);
3477                         journal_mark_dirty(&th, sb,
3478                                            SB_BUFFER_WITH_SB(sb));
3479
3480                         /* we're only being called from kreiserfsd, it makes no sense to do
3481                          ** an async commit so that kreiserfsd can do it later
3482                          */
3483                         do_journal_end(&th, sb, 1, COMMIT_NOW | WAIT);
3484                 }
3485         }
3486 }
3487
3488 /*
3489 ** returns 0 if do_journal_end should return right away, returns 1 if do_journal_end should finish the commit
3490 **
3491 ** if the current transaction is too old, but still has writers, this will wait on j_join_wait until all
3492 ** the writers are done.  By the time it wakes up, the transaction it was called has already ended, so it just
3493 ** flushes the commit list and returns 0.
3494 **
3495 ** Won't batch when flush or commit_now is set.  Also won't batch when others are waiting on j_join_wait.
3496 **
3497 ** Note, we can't allow the journal_end to proceed while there are still writers in the log.
3498 */
3499 static int check_journal_end(struct reiserfs_transaction_handle *th,
3500                              struct super_block *sb, unsigned long nblocks,
3501                              int flags)
3502 {
3503
3504         time_t now;
3505         int flush = flags & FLUSH_ALL;
3506         int commit_now = flags & COMMIT_NOW;
3507         int wait_on_commit = flags & WAIT;
3508         struct reiserfs_journal_list *jl;
3509         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3510
3511         BUG_ON(!th->t_trans_id);
3512
3513         if (th->t_trans_id != journal->j_trans_id) {
3514                 reiserfs_panic(th->t_super, "journal-1577",
3515                                "handle trans id %ld != current trans id %ld",
3516                                th->t_trans_id, journal->j_trans_id);
3517         }
3518
3519         journal->j_len_alloc -= (th->t_blocks_allocated - th->t_blocks_logged);
3520         if (atomic_read(&(journal->j_wcount)) > 0) {    /* <= 0 is allowed.  unmounting might not call begin */
3521                 atomic_dec(&(journal->j_wcount));
3522         }
3523
3524         /* BUG, deal with case where j_len is 0, but people previously freed blocks need to be released
3525          ** will be dealt with by next transaction that actually writes something, but should be taken
3526          ** care of in this trans
3527          */
3528         BUG_ON(journal->j_len == 0);
3529
3530         /* if wcount > 0, and we are called to with flush or commit_now,
3531          ** we wait on j_join_wait.  We will wake up when the last writer has
3532          ** finished the transaction, and started it on its way to the disk.
3533          ** Then, we flush the commit or journal list, and just return 0
3534          ** because the rest of journal end was already done for this transaction.
3535          */
3536         if (atomic_read(&(journal->j_wcount)) > 0) {
3537                 if (flush || commit_now) {
3538                         unsigned trans_id;
3539
3540                         jl = journal->j_current_jl;
3541                         trans_id = jl->j_trans_id;
3542                         if (wait_on_commit)
3543                                 jl->j_state |= LIST_COMMIT_PENDING;
3544                         atomic_set(&(journal->j_jlock), 1);
3545                         if (flush) {
3546                                 journal->j_next_full_flush = 1;
3547                         }
3548                         unlock_journal(sb);
3549
3550                         /* sleep while the current transaction is still j_jlocked */
3551                         while (journal->j_trans_id == trans_id) {
3552                                 if (atomic_read(&journal->j_jlock)) {
3553                                         queue_log_writer(sb);
3554                                 } else {
3555                                         lock_journal(sb);
3556                                         if (journal->j_trans_id == trans_id) {
3557                                                 atomic_set(&(journal->j_jlock),
3558                                                            1);
3559                                         }
3560                                         unlock_journal(sb);
3561                                 }
3562                         }
3563                         BUG_ON(journal->j_trans_id == trans_id);
3564
3565                         if (commit_now
3566                             && journal_list_still_alive(sb, trans_id)
3567                             && wait_on_commit) {
3568                                 flush_commit_list(sb, jl, 1);
3569                         }
3570                         return 0;
3571                 }
3572                 unlock_journal(sb);
3573                 return 0;
3574         }
3575
3576         /* deal with old transactions where we are the last writers */
3577         now = get_seconds();
3578         if ((now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3579                 commit_now = 1;
3580                 journal->j_next_async_flush = 1;
3581         }
3582         /* don't batch when someone is waiting on j_join_wait */
3583         /* don't batch when syncing the commit or flushing the whole trans */
3584         if (!(journal->j_must_wait > 0) && !(atomic_read(&(journal->j_jlock)))
3585             && !flush && !commit_now && (journal->j_len < journal->j_max_batch)
3586             && journal->j_len_alloc < journal->j_max_batch
3587             && journal->j_cnode_free > (journal->j_trans_max * 3)) {
3588                 journal->j_bcount++;
3589                 unlock_journal(sb);
3590                 return 0;
3591         }
3592
3593         if (journal->j_start > SB_ONDISK_JOURNAL_SIZE(sb)) {
3594                 reiserfs_panic(sb, "journal-003",
3595                                "j_start (%ld) is too high",
3596                                journal->j_start);
3597         }
3598         return 1;
3599 }
3600
3601 /*
3602 ** Does all the work that makes deleting blocks safe.
3603 ** when deleting a block mark BH_JNew, just remove it from the current transaction, clean it's buffer_head and move on.
3604 **
3605 ** otherwise:
3606 ** set a bit for the block in the journal bitmap.  That will prevent it from being allocated for unformatted nodes
3607 ** before this transaction has finished.
3608 **
3609 ** mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers.  That will prevent any old transactions with
3610 ** this block from trying to flush to the real location.  Since we aren't removing the cnode from the journal_list_hash,
3611 ** the block can't be reallocated yet.
3612 **
3613 ** Then remove it from the current transaction, decrementing any counters and filing it on the clean list.
3614 */
3615 int journal_mark_freed(struct reiserfs_transaction_handle *th,
3616                        struct super_block *sb, b_blocknr_t blocknr)
3617 {
3618         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3619         struct reiserfs_journal_cnode *cn = NULL;
3620         struct buffer_head *bh = NULL;
3621         struct reiserfs_list_bitmap *jb = NULL;
3622         int cleaned = 0;
3623         BUG_ON(!th->t_trans_id);
3624
3625         cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr);
3626         if (cn && cn->bh) {
3627                 bh = cn->bh;
3628                 get_bh(bh);
3629         }
3630         /* if it is journal new, we just remove it from this transaction */
3631         if (bh && buffer_journal_new(bh)) {
3632                 clear_buffer_journal_new(bh);
3633                 clear_prepared_bits(bh);
3634                 reiserfs_clean_and_file_buffer(bh);
3635                 cleaned = remove_from_transaction(sb, blocknr, cleaned);
3636         } else {
3637                 /* set the bit for this block in the journal bitmap for this transaction */
3638                 jb = journal->j_current_jl->j_list_bitmap;
3639                 if (!jb) {
3640                         reiserfs_panic(sb, "journal-1702",
3641                                        "journal_list_bitmap is NULL");
3642                 }
3643                 set_bit_in_list_bitmap(sb, blocknr, jb);
3644
3645                 /* Note, the entire while loop is not allowed to schedule.  */
3646
3647                 if (bh) {
3648                         clear_prepared_bits(bh);
3649                         reiserfs_clean_and_file_buffer(bh);
3650                 }
3651                 cleaned = remove_from_transaction(sb, blocknr, cleaned);
3652
3653                 /* find all older transactions with this block, make sure they don't try to write it out */
3654                 cn = get_journal_hash_dev(sb, journal->j_list_hash_table,
3655                                           blocknr);
3656                 while (cn) {
3657                         if (sb == cn->sb && blocknr == cn->blocknr) {
3658                                 set_bit(BLOCK_FREED, &cn->state);
3659                                 if (cn->bh) {
3660                                         if (!cleaned) {
3661                                                 /* remove_from_transaction will brelse the buffer if it was
3662                                                  ** in the current trans
3663                                                  */
3664                                                 clear_buffer_journal_dirty(cn->
3665                                                                            bh);
3666                                                 clear_buffer_dirty(cn->bh);
3667                                                 clear_buffer_journal_test(cn->
3668                                                                           bh);
3669                                                 cleaned = 1;
3670                                                 put_bh(cn->bh);
3671                                                 if (atomic_read
3672                                                     (&(cn->bh->b_count)) < 0) {
3673                                                         reiserfs_warning(sb,
3674                                                                  "journal-2138",
3675                                                                  "cn->bh->b_count < 0");
3676                                                 }
3677                                         }
3678                                         if (cn->jlist) {        /* since we are clearing the bh, we MUST dec nonzerolen */
3679                                                 atomic_dec(&
3680                                                            (cn->jlist->
3681                                                             j_nonzerolen));
3682                                         }
3683                                         cn->bh = NULL;
3684                                 }
3685                         }
3686                         cn = cn->hnext;
3687                 }
3688         }
3689
3690         if (bh)
3691                 release_buffer_page(bh); /* get_hash grabs the buffer */
3692         return 0;
3693 }
3694
3695 void reiserfs_update_inode_transaction(struct inode *inode)
3696 {
3697         struct reiserfs_journal *journal = SB_JOURNAL(inode->i_sb);
3698         REISERFS_I(inode)->i_jl = journal->j_current_jl;
3699         REISERFS_I(inode)->i_trans_id = journal->j_trans_id;
3700 }
3701
3702 /*
3703  * returns -1 on error, 0 if no commits/barriers were done and 1
3704  * if a transaction was actually committed and the barrier was done
3705  */
3706 static int __commit_trans_jl(struct inode *inode, unsigned long id,
3707                              struct reiserfs_journal_list *jl)
3708 {
3709         struct reiserfs_transaction_handle th;
3710         struct super_block *sb = inode->i_sb;
3711         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3712         int ret = 0;
3713
3714         /* is it from the current transaction, or from an unknown transaction? */
3715         if (id == journal->j_trans_id) {
3716                 jl = journal->j_current_jl;
3717                 /* try to let other writers come in and grow this transaction */
3718                 let_transaction_grow(sb, id);
3719                 if (journal->j_trans_id != id) {
3720                         goto flush_commit_only;
3721                 }
3722
3723                 ret = journal_begin(&th, sb, 1);
3724                 if (ret)
3725                         return ret;
3726
3727                 /* someone might have ended this transaction while we joined */
3728                 if (journal->j_trans_id != id) {
3729                         reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3730                                                      1);
3731                         journal_mark_dirty(&th, sb, SB_BUFFER_WITH_SB(sb));
3732                         ret = journal_end(&th, sb, 1);
3733                         goto flush_commit_only;
3734                 }
3735
3736                 ret = journal_end_sync(&th, sb, 1);
3737                 if (!ret)
3738                         ret = 1;
3739
3740         } else {
3741                 /* this gets tricky, we have to make sure the journal list in
3742                  * the inode still exists.  We know the list is still around
3743                  * if we've got a larger transaction id than the oldest list
3744                  */
3745               flush_commit_only:
3746                 if (journal_list_still_alive(inode->i_sb, id)) {
3747                         /*
3748                          * we only set ret to 1 when we know for sure
3749                          * the barrier hasn't been started yet on the commit
3750                          * block.
3751                          */
3752                         if (atomic_read(&jl->j_commit_left) > 1)
3753                                 ret = 1;
3754                         flush_commit_list(sb, jl, 1);
3755                         if (journal->j_errno)
3756                                 ret = journal->j_errno;
3757                 }
3758         }
3759         /* otherwise the list is gone, and long since committed */
3760         return ret;
3761 }
3762
3763 int reiserfs_commit_for_inode(struct inode *inode)
3764 {
3765         unsigned int id = REISERFS_I(inode)->i_trans_id;
3766         struct reiserfs_journal_list *jl = REISERFS_I(inode)->i_jl;
3767
3768         /* for the whole inode, assume unset id means it was
3769          * changed in the current transaction.  More conservative
3770          */
3771         if (!id || !jl) {
3772                 reiserfs_update_inode_transaction(inode);
3773                 id = REISERFS_I(inode)->i_trans_id;
3774                 /* jl will be updated in __commit_trans_jl */
3775         }
3776
3777         return __commit_trans_jl(inode, id, jl);
3778 }
3779
3780 void reiserfs_restore_prepared_buffer(struct super_block *sb,
3781                                       struct buffer_head *bh)
3782 {
3783         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3784         PROC_INFO_INC(sb, journal.restore_prepared);
3785         if (!bh) {
3786                 return;
3787         }
3788         if (test_clear_buffer_journal_restore_dirty(bh) &&
3789             buffer_journal_dirty(bh)) {
3790                 struct reiserfs_journal_cnode *cn;
3791                 reiserfs_write_lock(sb);
3792                 cn = get_journal_hash_dev(sb,
3793                                           journal->j_list_hash_table,
3794                                           bh->b_blocknr);
3795                 if (cn && can_dirty(cn)) {
3796                         set_buffer_journal_test(bh);
3797                         mark_buffer_dirty(bh);
3798                 }
3799                 reiserfs_write_unlock(sb);
3800         }
3801         clear_buffer_journal_prepared(bh);
3802 }
3803
3804 extern struct tree_balance *cur_tb;
3805 /*
3806 ** before we can change a metadata block, we have to make sure it won't
3807 ** be written to disk while we are altering it.  So, we must:
3808 ** clean it
3809 ** wait on it.
3810 **
3811 */
3812 int reiserfs_prepare_for_journal(struct super_block *sb,
3813                                  struct buffer_head *bh, int wait)
3814 {
3815         PROC_INFO_INC(sb, journal.prepare);
3816
3817         if (!trylock_buffer(bh)) {
3818                 if (!wait)
3819                         return 0;
3820                 lock_buffer(bh);
3821         }
3822         set_buffer_journal_prepared(bh);
3823         if (test_clear_buffer_dirty(bh) && buffer_journal_dirty(bh)) {
3824                 clear_buffer_journal_test(bh);
3825                 set_buffer_journal_restore_dirty(bh);
3826         }
3827         unlock_buffer(bh);
3828         return 1;
3829 }
3830
3831 /*
3832 ** long and ugly.  If flush, will not return until all commit
3833 ** blocks and all real buffers in the trans are on disk.
3834 ** If no_async, won't return until all commit blocks are on disk.
3835 **
3836 ** keep reading, there are comments as you go along
3837 **
3838 ** If the journal is aborted, we just clean up. Things like flushing
3839 ** journal lists, etc just won't happen.
3840 */
3841 static int do_journal_end(struct reiserfs_transaction_handle *th,
3842                           struct super_block *sb, unsigned long nblocks,
3843                           int flags)
3844 {
3845         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3846         struct reiserfs_journal_cnode *cn, *next, *jl_cn;
3847         struct reiserfs_journal_cnode *last_cn = NULL;
3848         struct reiserfs_journal_desc *desc;
3849         struct reiserfs_journal_commit *commit;
3850         struct buffer_head *c_bh;       /* commit bh */
3851         struct buffer_head *d_bh;       /* desc bh */
3852         int cur_write_start = 0;        /* start index of current log write */
3853         int old_start;
3854         int i;
3855         int flush;
3856         int wait_on_commit;
3857         struct reiserfs_journal_list *jl, *temp_jl;
3858         struct list_head *entry, *safe;
3859         unsigned long jindex;
3860         unsigned int commit_trans_id;
3861         int trans_half;
3862         int depth;
3863
3864         BUG_ON(th->t_refcount > 1);
3865         BUG_ON(!th->t_trans_id);
3866
3867         /* protect flush_older_commits from doing mistakes if the
3868            transaction ID counter gets overflowed.  */
3869         if (th->t_trans_id == ~0U)
3870                 flags |= FLUSH_ALL | COMMIT_NOW | WAIT;
3871         flush = flags & FLUSH_ALL;
3872         wait_on_commit = flags & WAIT;
3873
3874         current->journal_info = th->t_handle_save;
3875         reiserfs_check_lock_depth(sb, "journal end");
3876         if (journal->j_len == 0) {
3877                 reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3878                                              1);
3879                 journal_mark_dirty(th, sb, SB_BUFFER_WITH_SB(sb));
3880         }
3881
3882         lock_journal(sb);
3883         if (journal->j_next_full_flush) {
3884                 flags |= FLUSH_ALL;
3885                 flush = 1;
3886         }
3887         if (journal->j_next_async_flush) {
3888                 flags |= COMMIT_NOW | WAIT;
3889                 wait_on_commit = 1;
3890         }
3891
3892         /* check_journal_end locks the journal, and unlocks if it does not return 1
3893          ** it tells us if we should continue with the journal_end, or just return
3894          */
3895         if (!check_journal_end(th, sb, nblocks, flags)) {
3896                 reiserfs_schedule_old_flush(sb);
3897                 wake_queued_writers(sb);
3898                 reiserfs_async_progress_wait(sb);
3899                 goto out;
3900         }
3901
3902         /* check_journal_end might set these, check again */
3903         if (journal->j_next_full_flush) {
3904                 flush = 1;
3905         }
3906
3907         /*
3908          ** j must wait means we have to flush the log blocks, and the real blocks for
3909          ** this transaction
3910          */
3911         if (journal->j_must_wait > 0) {
3912                 flush = 1;
3913         }
3914 #ifdef REISERFS_PREALLOCATE
3915         /* quota ops might need to nest, setup the journal_info pointer for them
3916          * and raise the refcount so that it is > 0. */
3917         current->journal_info = th;
3918         th->t_refcount++;
3919         reiserfs_discard_all_prealloc(th);      /* it should not involve new blocks into
3920                                                  * the transaction */
3921         th->t_refcount--;
3922         current->journal_info = th->t_handle_save;
3923 #endif
3924
3925         /* setup description block */
3926         d_bh =
3927             journal_getblk(sb,
3928                            SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
3929                            journal->j_start);
3930         set_buffer_uptodate(d_bh);
3931         desc = (struct reiserfs_journal_desc *)(d_bh)->b_data;
3932         memset(d_bh->b_data, 0, d_bh->b_size);
3933         memcpy(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8);
3934         set_desc_trans_id(desc, journal->j_trans_id);
3935
3936         /* setup commit block.  Don't write (keep it clean too) this one until after everyone else is written */
3937         c_bh = journal_getblk(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
3938                               ((journal->j_start + journal->j_len +
3939                                 1) % SB_ONDISK_JOURNAL_SIZE(sb)));
3940         commit = (struct reiserfs_journal_commit *)c_bh->b_data;
3941         memset(c_bh->b_data, 0, c_bh->b_size);
3942         set_commit_trans_id(commit, journal->j_trans_id);
3943         set_buffer_uptodate(c_bh);
3944
3945         /* init this journal list */
3946         jl = journal->j_current_jl;
3947
3948         /* we lock the commit before doing anything because
3949          * we want to make sure nobody tries to run flush_commit_list until
3950          * the new transaction is fully setup, and we've already flushed the
3951          * ordered bh list
3952          */
3953         reiserfs_mutex_lock_safe(&jl->j_commit_mutex, sb);
3954
3955         /* save the transaction id in case we need to commit it later */
3956         commit_trans_id = jl->j_trans_id;
3957
3958         atomic_set(&jl->j_older_commits_done, 0);
3959         jl->j_trans_id = journal->j_trans_id;
3960         jl->j_timestamp = journal->j_trans_start_time;
3961         jl->j_commit_bh = c_bh;
3962         jl->j_start = journal->j_start;
3963         jl->j_len = journal->j_len;
3964         atomic_set(&jl->j_nonzerolen, journal->j_len);
3965         atomic_set(&jl->j_commit_left, journal->j_len + 2);
3966         jl->j_realblock = NULL;
3967
3968         /* The ENTIRE FOR LOOP MUST not cause schedule to occur.
3969          **  for each real block, add it to the journal list hash,
3970          ** copy into real block index array in the commit or desc block
3971          */
3972         trans_half = journal_trans_half(sb->s_blocksize);
3973         for (i = 0, cn = journal->j_first; cn; cn = cn->next, i++) {
3974                 if (buffer_journaled(cn->bh)) {
3975                         jl_cn = get_cnode(sb);
3976                         if (!jl_cn) {
3977                                 reiserfs_panic(sb, "journal-1676",
3978                                                "get_cnode returned NULL");
3979                         }
3980                         if (i == 0) {
3981                                 jl->j_realblock = jl_cn;
3982                         }
3983                         jl_cn->prev = last_cn;
3984                         jl_cn->next = NULL;
3985                         if (last_cn) {
3986                                 last_cn->next = jl_cn;
3987                         }
3988                         last_cn = jl_cn;
3989                         /* make sure the block we are trying to log is not a block
3990                            of journal or reserved area */
3991
3992                         if (is_block_in_log_or_reserved_area
3993                             (sb, cn->bh->b_blocknr)) {
3994                                 reiserfs_panic(sb, "journal-2332",
3995                                                "Trying to log block %lu, "
3996                                                "which is a log block",
3997                                                cn->bh->b_blocknr);
3998                         }
3999                         jl_cn->blocknr = cn->bh->b_blocknr;
4000                         jl_cn->state = 0;
4001                         jl_cn->sb = sb;
4002                         jl_cn->bh = cn->bh;
4003                         jl_cn->jlist = jl;
4004                         insert_journal_hash(journal->j_list_hash_table, jl_cn);
4005                         if (i < trans_half) {
4006                                 desc->j_realblock[i] =
4007                                     cpu_to_le32(cn->bh->b_blocknr);
4008                         } else {
4009                                 commit->j_realblock[i - trans_half] =
4010                                     cpu_to_le32(cn->bh->b_blocknr);
4011                         }
4012                 } else {
4013                         i--;
4014                 }
4015         }
4016         set_desc_trans_len(desc, journal->j_len);
4017         set_desc_mount_id(desc, journal->j_mount_id);
4018         set_desc_trans_id(desc, journal->j_trans_id);
4019         set_commit_trans_len(commit, journal->j_len);
4020
4021         /* special check in case all buffers in the journal were marked for not logging */
4022         BUG_ON(journal->j_len == 0);
4023
4024         /* we're about to dirty all the log blocks, mark the description block
4025          * dirty now too.  Don't mark the commit block dirty until all the
4026          * others are on disk
4027          */
4028         mark_buffer_dirty(d_bh);
4029
4030         /* first data block is j_start + 1, so add one to cur_write_start wherever you use it */
4031         cur_write_start = journal->j_start;
4032         cn = journal->j_first;
4033         jindex = 1;             /* start at one so we don't get the desc again */
4034         while (cn) {
4035                 clear_buffer_journal_new(cn->bh);
4036                 /* copy all the real blocks into log area.  dirty log blocks */
4037                 if (buffer_journaled(cn->bh)) {
4038                         struct buffer_head *tmp_bh;
4039                         char *addr;
4040                         struct page *page;
4041                         tmp_bh =
4042                             journal_getblk(sb,
4043                                            SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4044                                            ((cur_write_start +
4045                                              jindex) %
4046                                             SB_ONDISK_JOURNAL_SIZE(sb)));
4047                         set_buffer_uptodate(tmp_bh);
4048                         page = cn->bh->b_page;
4049                         addr = kmap(page);
4050                         memcpy(tmp_bh->b_data,
4051                                addr + offset_in_page(cn->bh->b_data),
4052                                cn->bh->b_size);
4053                         kunmap(page);
4054                         mark_buffer_dirty(tmp_bh);
4055                         jindex++;
4056                         set_buffer_journal_dirty(cn->bh);
4057                         clear_buffer_journaled(cn->bh);
4058                 } else {
4059                         /* JDirty cleared sometime during transaction.  don't log this one */
4060                         reiserfs_warning(sb, "journal-2048",
4061                                          "BAD, buffer in journal hash, "
4062                                          "but not JDirty!");
4063                         brelse(cn->bh);
4064                 }
4065                 next = cn->next;
4066                 free_cnode(sb, cn);
4067                 cn = next;
4068                 reiserfs_cond_resched(sb);
4069         }
4070
4071         /* we are done  with both the c_bh and d_bh, but
4072          ** c_bh must be written after all other commit blocks,
4073          ** so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1.
4074          */
4075
4076         journal->j_current_jl = alloc_journal_list(sb);
4077
4078         /* now it is safe to insert this transaction on the main list */
4079         list_add_tail(&jl->j_list, &journal->j_journal_list);
4080         list_add_tail(&jl->j_working_list, &journal->j_working_list);
4081         journal->j_num_work_lists++;
4082
4083         /* reset journal values for the next transaction */
4084         old_start = journal->j_start;
4085         journal->j_start =
4086             (journal->j_start + journal->j_len +
4087              2) % SB_ONDISK_JOURNAL_SIZE(sb);
4088         atomic_set(&(journal->j_wcount), 0);
4089         journal->j_bcount = 0;
4090         journal->j_last = NULL;
4091         journal->j_first = NULL;
4092         journal->j_len = 0;
4093         journal->j_trans_start_time = 0;
4094         /* check for trans_id overflow */
4095         if (++journal->j_trans_id == 0)
4096                 journal->j_trans_id = 10;
4097         journal->j_current_jl->j_trans_id = journal->j_trans_id;
4098         journal->j_must_wait = 0;
4099         journal->j_len_alloc = 0;
4100         journal->j_next_full_flush = 0;
4101         journal->j_next_async_flush = 0;
4102         init_journal_hash(sb);
4103
4104         // make sure reiserfs_add_jh sees the new current_jl before we
4105         // write out the tails
4106         smp_mb();
4107
4108         /* tail conversion targets have to hit the disk before we end the
4109          * transaction.  Otherwise a later transaction might repack the tail
4110          * before this transaction commits, leaving the data block unflushed and
4111          * clean, if we crash before the later transaction commits, the data block
4112          * is lost.
4113          */
4114         if (!list_empty(&jl->j_tail_bh_list)) {
4115                 depth = reiserfs_write_unlock_nested(sb);
4116                 write_ordered_buffers(&journal->j_dirty_buffers_lock,
4117                                       journal, jl, &jl->j_tail_bh_list);
4118                 reiserfs_write_lock_nested(sb, depth);
4119         }
4120         BUG_ON(!list_empty(&jl->j_tail_bh_list));
4121         mutex_unlock(&jl->j_commit_mutex);
4122
4123         /* honor the flush wishes from the caller, simple commits can
4124          ** be done outside the journal lock, they are done below
4125          **
4126          ** if we don't flush the commit list right now, we put it into
4127          ** the work queue so the people waiting on the async progress work
4128          ** queue don't wait for this proc to flush journal lists and such.
4129          */
4130         if (flush) {
4131                 flush_commit_list(sb, jl, 1);
4132                 flush_journal_list(sb, jl, 1);
4133         } else if (!(jl->j_state & LIST_COMMIT_PENDING))
4134                 queue_delayed_work(commit_wq, &journal->j_work, HZ / 10);
4135
4136         /* if the next transaction has any chance of wrapping, flush
4137          ** transactions that might get overwritten.  If any journal lists are very
4138          ** old flush them as well.
4139          */
4140       first_jl:
4141         list_for_each_safe(entry, safe, &journal->j_journal_list) {
4142                 temp_jl = JOURNAL_LIST_ENTRY(entry);
4143                 if (journal->j_start <= temp_jl->j_start) {
4144                         if ((journal->j_start + journal->j_trans_max + 1) >=
4145                             temp_jl->j_start) {
4146                                 flush_used_journal_lists(sb, temp_jl);
4147                                 goto first_jl;
4148                         } else if ((journal->j_start +
4149                                     journal->j_trans_max + 1) <
4150                                    SB_ONDISK_JOURNAL_SIZE(sb)) {
4151                                 /* if we don't cross into the next transaction and we don't
4152                                  * wrap, there is no way we can overlap any later transactions
4153                                  * break now
4154                                  */
4155                                 break;
4156                         }
4157                 } else if ((journal->j_start +
4158                             journal->j_trans_max + 1) >
4159                            SB_ONDISK_JOURNAL_SIZE(sb)) {
4160                         if (((journal->j_start + journal->j_trans_max + 1) %
4161                              SB_ONDISK_JOURNAL_SIZE(sb)) >=
4162                             temp_jl->j_start) {
4163                                 flush_used_journal_lists(sb, temp_jl);
4164                                 goto first_jl;
4165                         } else {
4166                                 /* we don't overlap anything from out start to the end of the
4167                                  * log, and our wrapped portion doesn't overlap anything at
4168                                  * the start of the log.  We can break
4169                                  */
4170                                 break;
4171                         }
4172                 }
4173         }
4174
4175         journal->j_current_jl->j_list_bitmap =
4176             get_list_bitmap(sb, journal->j_current_jl);
4177
4178         if (!(journal->j_current_jl->j_list_bitmap)) {
4179                 reiserfs_panic(sb, "journal-1996",
4180                                "could not get a list bitmap");
4181         }
4182
4183         atomic_set(&(journal->j_jlock), 0);
4184         unlock_journal(sb);
4185         /* wake up any body waiting to join. */
4186         clear_bit(J_WRITERS_QUEUED, &journal->j_state);
4187         wake_up(&(journal->j_join_wait));
4188
4189         if (!flush && wait_on_commit &&
4190             journal_list_still_alive(sb, commit_trans_id)) {
4191                 flush_commit_list(sb, jl, 1);
4192         }
4193       out:
4194         reiserfs_check_lock_depth(sb, "journal end2");
4195
4196         memset(th, 0, sizeof(*th));
4197         /* Re-set th->t_super, so we can properly keep track of how many
4198          * persistent transactions there are. We need to do this so if this
4199          * call is part of a failed restart_transaction, we can free it later */
4200         th->t_super = sb;
4201
4202         return journal->j_errno;
4203 }
4204
4205 /* Send the file system read only and refuse new transactions */
4206 void reiserfs_abort_journal(struct super_block *sb, int errno)
4207 {
4208         struct reiserfs_journal *journal = SB_JOURNAL(sb);
4209         if (test_bit(J_ABORTED, &journal->j_state))
4210                 return;
4211
4212         if (!journal->j_errno)
4213                 journal->j_errno = errno;
4214
4215         sb->s_flags |= MS_RDONLY;
4216         set_bit(J_ABORTED, &journal->j_state);
4217
4218 #ifdef CONFIG_REISERFS_CHECK
4219         dump_stack();
4220 #endif
4221 }