fs/reiserfs/journal.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Write ahead logging implementation copyright Chris Mason 2000
   4  *
   5  * The background commits make this code very interrelated, and
   6  * overly complex.  I need to rethink things a bit....The major players:
   7  *
   8  * journal_begin -- call with the number of blocks you expect to log.
   9  *                  If the current transaction is too
  10  *                  old, it will block until the current transaction is
  11  *                  finished, and then start a new one.
  12  *                  Usually, your transaction will get joined in with
  13  *                  previous ones for speed.
  14  *
  15  * journal_join  -- same as journal_begin, but won't block on the current
  16  *                  transaction regardless of age.  Don't ever call
  17  *                  this.  Ever.  There are only two places it should be
  18  *                  called from, and they are both inside this file.
  19  *
  20  * journal_mark_dirty -- adds blocks into this transaction.  clears any flags
  21  *                       that might make them get sent to disk
  22  *                       and then marks them BH_JDirty.  Puts the buffer head
  23  *                       into the current transaction hash.
  24  *
  25  * journal_end -- if the current transaction is batchable, it does nothing
  26  *                   otherwise, it could do an async/synchronous commit, or
  27  *                   a full flush of all log and real blocks in the
  28  *                   transaction.
  29  *
  30  * flush_old_commits -- if the current transaction is too old, it is ended and
  31  *                      commit blocks are sent to disk.  Forces commit blocks
  32  *                      to disk for all backgrounded commits that have been
  33  *                      around too long.
  34  *                   -- Note, if you call this as an immediate flush from
  35  *                      from within kupdate, it will ignore the immediate flag
  36  */
  37
  38 #include <linux/time.h>
  39 #include <linux/semaphore.h>
  40 #include <linux/vmalloc.h>
  41 #include "reiserfs.h"
  42 #include <linux/kernel.h>
  43 #include <linux/errno.h>
  44 #include <linux/fcntl.h>
  45 #include <linux/stat.h>
  46 #include <linux/string.h>
  47 #include <linux/buffer_head.h>
  48 #include <linux/workqueue.h>
  49 #include <linux/writeback.h>
  50 #include <linux/blkdev.h>
  51 #include <linux/backing-dev.h>
  52 #include <linux/uaccess.h>
  53 #include <linux/slab.h>
  54
  55
  56 /* gets a struct reiserfs_journal_list * from a list head */
  57 #define JOURNAL_LIST_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
  58                                j_list))
  59
  60 /* must be correct to keep the desc and commit structs at 4k */
  61 #define JOURNAL_TRANS_HALF 1018
  62 #define BUFNR 64                /*read ahead */
  63
  64 /* cnode stat bits.  Move these into reiserfs_fs.h */
  65
  66 /* this block was freed, and can't be written.  */
  67 #define BLOCK_FREED 2
  68 /* this block was freed during this transaction, and can't be written */
  69 #define BLOCK_FREED_HOLDER 3
  70
  71 /* used in flush_journal_list */
  72 #define BLOCK_NEEDS_FLUSH 4
  73 #define BLOCK_DIRTIED 5
  74
  75 /* journal list state bits */
  76 #define LIST_TOUCHED 1
  77 #define LIST_DIRTY   2
  78 #define LIST_COMMIT_PENDING  4  /* someone will commit this list */
  79
  80 /* flags for do_journal_end */
  81 #define FLUSH_ALL   1           /* flush commit and real blocks */
  82 #define COMMIT_NOW  2           /* end and commit this transaction */
  83 #define WAIT        4           /* wait for the log blocks to hit the disk */
  84
  85 static int do_journal_end(struct reiserfs_transaction_handle *, int flags);
  86 static int flush_journal_list(struct super_block *s,
  87                               struct reiserfs_journal_list *jl, int flushall);
  88 static int flush_commit_list(struct super_block *s,
  89                              struct reiserfs_journal_list *jl, int flushall);
  90 static int can_dirty(struct reiserfs_journal_cnode *cn);
  91 static int journal_join(struct reiserfs_transaction_handle *th,
  92                         struct super_block *sb);
  93 static void release_journal_dev(struct super_block *super,
  94                                struct reiserfs_journal *journal);
  95 static void dirty_one_transaction(struct super_block *s,
  96                                  struct reiserfs_journal_list *jl);
  97 static void flush_async_commits(struct work_struct *work);
  98 static void queue_log_writer(struct super_block *s);
  99
 100 /* values for join in do_journal_begin_r */
 101 enum {
 102         JBEGIN_REG = 0,         /* regular journal begin */
 103         /* join the running transaction if at all possible */
 104         JBEGIN_JOIN = 1,
 105         /* called from cleanup code, ignores aborted flag */
 106         JBEGIN_ABORT = 2,
 107 };
 108
 109 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
 110                               struct super_block *sb,
 111                               unsigned long nblocks, int join);
 112
 113 static void init_journal_hash(struct super_block *sb)
 114 {
 115         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 116         memset(journal->j_hash_table, 0,
 117                JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
 118 }
 119
 120 /*
 121  * clears BH_Dirty and sticks the buffer on the clean list.  Called because
 122  * I can't allow refile_buffer to make schedule happen after I've freed a
 123  * block.  Look at remove_from_transaction and journal_mark_freed for
 124  * more details.
 125  */
 126 static int reiserfs_clean_and_file_buffer(struct buffer_head *bh)
 127 {
 128         if (bh) {
 129                 clear_buffer_dirty(bh);
 130                 clear_buffer_journal_test(bh);
 131         }
 132         return 0;
 133 }
 134
 135 static struct reiserfs_bitmap_node *allocate_bitmap_node(struct super_block
 136                                                          *sb)
 137 {
 138         struct reiserfs_bitmap_node *bn;
 139         static int id;
 140
 141         bn = kmalloc(sizeof(struct reiserfs_bitmap_node), GFP_NOFS);
 142         if (!bn) {
 143                 return NULL;
 144         }
 145         bn->data = kzalloc(sb->s_blocksize, GFP_NOFS);
 146         if (!bn->data) {
 147                 kfree(bn);
 148                 return NULL;
 149         }
 150         bn->id = id++;
 151         INIT_LIST_HEAD(&bn->list);
 152         return bn;
 153 }
 154
 155 static struct reiserfs_bitmap_node *get_bitmap_node(struct super_block *sb)
 156 {
 157         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 158         struct reiserfs_bitmap_node *bn = NULL;
 159         struct list_head *entry = journal->j_bitmap_nodes.next;
 160
 161         journal->j_used_bitmap_nodes++;
 162 repeat:
 163
 164         if (entry != &journal->j_bitmap_nodes) {
 165                 bn = list_entry(entry, struct reiserfs_bitmap_node, list);
 166                 list_del(entry);
 167                 memset(bn->data, 0, sb->s_blocksize);
 168                 journal->j_free_bitmap_nodes--;
 169                 return bn;
 170         }
 171         bn = allocate_bitmap_node(sb);
 172         if (!bn) {
 173                 yield();
 174                 goto repeat;
 175         }
 176         return bn;
 177 }
 178 static inline void free_bitmap_node(struct super_block *sb,
 179                                     struct reiserfs_bitmap_node *bn)
 180 {
 181         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 182         journal->j_used_bitmap_nodes--;
 183         if (journal->j_free_bitmap_nodes > REISERFS_MAX_BITMAP_NODES) {
 184                 kfree(bn->data);
 185                 kfree(bn);
 186         } else {
 187                 list_add(&bn->list, &journal->j_bitmap_nodes);
 188                 journal->j_free_bitmap_nodes++;
 189         }
 190 }
 191
 192 static void allocate_bitmap_nodes(struct super_block *sb)
 193 {
 194         int i;
 195         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 196         struct reiserfs_bitmap_node *bn = NULL;
 197         for (i = 0; i < REISERFS_MIN_BITMAP_NODES; i++) {
 198                 bn = allocate_bitmap_node(sb);
 199                 if (bn) {
 200                         list_add(&bn->list, &journal->j_bitmap_nodes);
 201                         journal->j_free_bitmap_nodes++;
 202                 } else {
 203                         /* this is ok, we'll try again when more are needed */
 204                         break;
 205                 }
 206         }
 207 }
 208
 209 static int set_bit_in_list_bitmap(struct super_block *sb,
 210                                   b_blocknr_t block,
 211                                   struct reiserfs_list_bitmap *jb)
 212 {
 213         unsigned int bmap_nr = block / (sb->s_blocksize << 3);
 214         unsigned int bit_nr = block % (sb->s_blocksize << 3);
 215
 216         if (!jb->bitmaps[bmap_nr]) {
 217                 jb->bitmaps[bmap_nr] = get_bitmap_node(sb);
 218         }
 219         set_bit(bit_nr, (unsigned long *)jb->bitmaps[bmap_nr]->data);
 220         return 0;
 221 }
 222
 223 static void cleanup_bitmap_list(struct super_block *sb,
 224                                 struct reiserfs_list_bitmap *jb)
 225 {
 226         int i;
 227         if (jb->bitmaps == NULL)
 228                 return;
 229
 230         for (i = 0; i < reiserfs_bmap_count(sb); i++) {
 231                 if (jb->bitmaps[i]) {
 232                         free_bitmap_node(sb, jb->bitmaps[i]);
 233                         jb->bitmaps[i] = NULL;
 234                 }
 235         }
 236 }
 237
 238 /*
 239  * only call this on FS unmount.
 240  */
 241 static int free_list_bitmaps(struct super_block *sb,
 242                              struct reiserfs_list_bitmap *jb_array)
 243 {
 244         int i;
 245         struct reiserfs_list_bitmap *jb;
 246         for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
 247                 jb = jb_array + i;
 248                 jb->journal_list = NULL;
 249                 cleanup_bitmap_list(sb, jb);
 250                 vfree(jb->bitmaps);
 251                 jb->bitmaps = NULL;
 252         }
 253         return 0;
 254 }
 255
 256 static int free_bitmap_nodes(struct super_block *sb)
 257 {
 258         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 259         struct list_head *next = journal->j_bitmap_nodes.next;
 260         struct reiserfs_bitmap_node *bn;
 261
 262         while (next != &journal->j_bitmap_nodes) {
 263                 bn = list_entry(next, struct reiserfs_bitmap_node, list);
 264                 list_del(next);
 265                 kfree(bn->data);
 266                 kfree(bn);
 267                 next = journal->j_bitmap_nodes.next;
 268                 journal->j_free_bitmap_nodes--;
 269         }
 270
 271         return 0;
 272 }
 273
 274 /*
 275  * get memory for JOURNAL_NUM_BITMAPS worth of bitmaps.
 276  * jb_array is the array to be filled in.
 277  */
 278 int reiserfs_allocate_list_bitmaps(struct super_block *sb,
 279                                    struct reiserfs_list_bitmap *jb_array,
 280                                    unsigned int bmap_nr)
 281 {
 282         int i;
 283         int failed = 0;
 284         struct reiserfs_list_bitmap *jb;
 285         int mem = bmap_nr * sizeof(struct reiserfs_bitmap_node *);
 286
 287         for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
 288                 jb = jb_array + i;
 289                 jb->journal_list = NULL;
 290                 jb->bitmaps = vzalloc(mem);
 291                 if (!jb->bitmaps) {
 292                         reiserfs_warning(sb, "clm-2000", "unable to "
 293                                          "allocate bitmaps for journal lists");
 294                         failed = 1;
 295                         break;
 296                 }
 297         }
 298         if (failed) {
 299                 free_list_bitmaps(sb, jb_array);
 300                 return -1;
 301         }
 302         return 0;
 303 }
 304
 305 /*
 306  * find an available list bitmap.  If you can't find one, flush a commit list
 307  * and try again
 308  */
 309 static struct reiserfs_list_bitmap *get_list_bitmap(struct super_block *sb,
 310                                                     struct reiserfs_journal_list
 311                                                     *jl)
 312 {
 313         int i, j;
 314         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 315         struct reiserfs_list_bitmap *jb = NULL;
 316
 317         for (j = 0; j < (JOURNAL_NUM_BITMAPS * 3); j++) {
 318                 i = journal->j_list_bitmap_index;
 319                 journal->j_list_bitmap_index = (i + 1) % JOURNAL_NUM_BITMAPS;
 320                 jb = journal->j_list_bitmap + i;
 321                 if (journal->j_list_bitmap[i].journal_list) {
 322                         flush_commit_list(sb,
 323                                           journal->j_list_bitmap[i].
 324                                           journal_list, 1);
 325                         if (!journal->j_list_bitmap[i].journal_list) {
 326                                 break;
 327                         }
 328                 } else {
 329                         break;
 330                 }
 331         }
 332         /* double check to make sure if flushed correctly */
 333         if (jb->journal_list)
 334                 return NULL;
 335         jb->journal_list = jl;
 336         return jb;
 337 }
 338
 339 /*
 340  * allocates a new chunk of X nodes, and links them all together as a list.
 341  * Uses the cnode->next and cnode->prev pointers
 342  * returns NULL on failure
 343  */
 344 static struct reiserfs_journal_cnode *allocate_cnodes(int num_cnodes)
 345 {
 346         struct reiserfs_journal_cnode *head;
 347         int i;
 348         if (num_cnodes <= 0) {
 349                 return NULL;
 350         }
 351         head = vzalloc(array_size(num_cnodes,
 352                                   sizeof(struct reiserfs_journal_cnode)));
 353         if (!head) {
 354                 return NULL;
 355         }
 356         head[0].prev = NULL;
 357         head[0].next = head + 1;
 358         for (i = 1; i < num_cnodes; i++) {
 359                 head[i].prev = head + (i - 1);
 360                 head[i].next = head + (i + 1);  /* if last one, overwrite it after the if */
 361         }
 362         head[num_cnodes - 1].next = NULL;
 363         return head;
 364 }
 365
 366 /* pulls a cnode off the free list, or returns NULL on failure */
 367 static struct reiserfs_journal_cnode *get_cnode(struct super_block *sb)
 368 {
 369         struct reiserfs_journal_cnode *cn;
 370         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 371
 372         reiserfs_check_lock_depth(sb, "get_cnode");
 373
 374         if (journal->j_cnode_free <= 0) {
 375                 return NULL;
 376         }
 377         journal->j_cnode_used++;
 378         journal->j_cnode_free--;
 379         cn = journal->j_cnode_free_list;
 380         if (!cn) {
 381                 return cn;
 382         }
 383         if (cn->next) {
 384                 cn->next->prev = NULL;
 385         }
 386         journal->j_cnode_free_list = cn->next;
 387         memset(cn, 0, sizeof(struct reiserfs_journal_cnode));
 388         return cn;
 389 }
 390
 391 /*
 392  * returns a cnode to the free list
 393  */
 394 static void free_cnode(struct super_block *sb,
 395                        struct reiserfs_journal_cnode *cn)
 396 {
 397         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 398
 399         reiserfs_check_lock_depth(sb, "free_cnode");
 400
 401         journal->j_cnode_used--;
 402         journal->j_cnode_free++;
 403         /* memset(cn, 0, sizeof(struct reiserfs_journal_cnode)) ; */
 404         cn->next = journal->j_cnode_free_list;
 405         if (journal->j_cnode_free_list) {
 406                 journal->j_cnode_free_list->prev = cn;
 407         }
 408         cn->prev = NULL;        /* not needed with the memset, but I might kill the memset, and forget to do this */
 409         journal->j_cnode_free_list = cn;
 410 }
 411
 412 static void clear_prepared_bits(struct buffer_head *bh)
 413 {
 414         clear_buffer_journal_prepared(bh);
 415         clear_buffer_journal_restore_dirty(bh);
 416 }
 417
 418 /*
 419  * return a cnode with same dev, block number and size in table,
 420  * or null if not found
 421  */
 422 static inline struct reiserfs_journal_cnode *get_journal_hash_dev(struct
 423                                                                   super_block
 424                                                                   *sb,
 425                                                                   struct
 426                                                                   reiserfs_journal_cnode
 427                                                                   **table,
 428                                                                   long bl)
 429 {
 430         struct reiserfs_journal_cnode *cn;
 431         cn = journal_hash(table, sb, bl);
 432         while (cn) {
 433                 if (cn->blocknr == bl && cn->sb == sb)
 434                         return cn;
 435                 cn = cn->hnext;
 436         }
 437         return (struct reiserfs_journal_cnode *)0;
 438 }
 439
 440 /*
 441  * this actually means 'can this block be reallocated yet?'.  If you set
 442  * search_all, a block can only be allocated if it is not in the current
 443  * transaction, was not freed by the current transaction, and has no chance
 444  * of ever 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
 447  * in the current transaction.  Since deleting a block removes it from the
 448  * current transaction, this case should never happen.  If you don't set
 449  * search_all, make sure you never write the block without logging it.
 450  *
 451  * next_zero_bit is a suggestion about the next block to try for find_forward.
 452  * when bl is rejected because it is set in a journal list bitmap, we search
 453  * for the next zero bit in the bitmap that rejected bl.  Then, we return
 454  * that through next_zero_bit for find_forward to try.
 455  *
 456  * Just because we return something in next_zero_bit does not mean we won't
 457  * reject it on the next call to reiserfs_in_journal
 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         /*
 473          * If we aren't doing a search_all, this is a metablock, and it
 474          * will be logged before use.  if we crash before the transaction
 475          * that freed it commits,  this transaction won't have committed
 476          * either, and the block will never be written
 477          */
 478         if (search_all) {
 479                 for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
 480                         PROC_INFO_INC(sb, journal.in_journal_bitmap);
 481                         jb = journal->j_list_bitmap + i;
 482                         if (jb->journal_list && jb->bitmaps[bmap_nr] &&
 483                             test_bit(bit_nr,
 484                                      (unsigned long *)jb->bitmaps[bmap_nr]->
 485                                      data)) {
 486                                 *next_zero_bit =
 487                                     find_next_zero_bit((unsigned long *)
 488                                                        (jb->bitmaps[bmap_nr]->
 489                                                         data),
 490                                                        sb->s_blocksize << 3,
 491                                                        bit_nr + 1);
 492                                 return 1;
 493                         }
 494                 }
 495         }
 496
 497         bl = bmap_nr * (sb->s_blocksize << 3) + bit_nr;
 498         /* is it in any old transactions? */
 499         if (search_all
 500             && (cn =
 501                 get_journal_hash_dev(sb, journal->j_list_hash_table, bl))) {
 502                 return 1;
 503         }
 504
 505         /* is it in the current transaction.  This should never happen */
 506         if ((cn = get_journal_hash_dev(sb, journal->j_hash_table, bl))) {
 507                 BUG();
 508                 return 1;
 509         }
 510
 511         PROC_INFO_INC(sb, journal.in_journal_reusable);
 512         /* safe for reuse */
 513         return 0;
 514 }
 515
 516 /* insert cn into table */
 517 static inline void insert_journal_hash(struct reiserfs_journal_cnode **table,
 518                                        struct reiserfs_journal_cnode *cn)
 519 {
 520         struct reiserfs_journal_cnode *cn_orig;
 521
 522         cn_orig = journal_hash(table, cn->sb, cn->blocknr);
 523         cn->hnext = cn_orig;
 524         cn->hprev = NULL;
 525         if (cn_orig) {
 526                 cn_orig->hprev = cn;
 527         }
 528         journal_hash(table, cn->sb, cn->blocknr) = cn;
 529 }
 530
 531 /* lock the current transaction */
 532 static inline void lock_journal(struct super_block *sb)
 533 {
 534         PROC_INFO_INC(sb, journal.lock_journal);
 535
 536         reiserfs_mutex_lock_safe(&SB_JOURNAL(sb)->j_mutex, sb);
 537 }
 538
 539 /* unlock the current transaction */
 540 static inline void unlock_journal(struct super_block *sb)
 541 {
 542         mutex_unlock(&SB_JOURNAL(sb)->j_mutex);
 543 }
 544
 545 static inline void get_journal_list(struct reiserfs_journal_list *jl)
 546 {
 547         jl->j_refcount++;
 548 }
 549
 550 static inline void put_journal_list(struct super_block *s,
 551                                     struct reiserfs_journal_list *jl)
 552 {
 553         if (jl->j_refcount < 1) {
 554                 reiserfs_panic(s, "journal-2", "trans id %u, refcount at %d",
 555                                jl->j_trans_id, jl->j_refcount);
 556         }
 557         if (--jl->j_refcount == 0)
 558                 kfree(jl);
 559 }
 560
 561 /*
 562  * this used to be much more involved, and I'm keeping it just in case
 563  * things get ugly again.  it gets called by flush_commit_list, and
 564  * cleans up any data stored about blocks freed during a transaction.
 565  */
 566 static void cleanup_freed_for_journal_list(struct super_block *sb,
 567                                            struct reiserfs_journal_list *jl)
 568 {
 569
 570         struct reiserfs_list_bitmap *jb = jl->j_list_bitmap;
 571         if (jb) {
 572                 cleanup_bitmap_list(sb, jb);
 573         }
 574         jl->j_list_bitmap->journal_list = NULL;
 575         jl->j_list_bitmap = NULL;
 576 }
 577
 578 static int journal_list_still_alive(struct super_block *s,
 579                                     unsigned int trans_id)
 580 {
 581         struct reiserfs_journal *journal = SB_JOURNAL(s);
 582         struct list_head *entry = &journal->j_journal_list;
 583         struct reiserfs_journal_list *jl;
 584
 585         if (!list_empty(entry)) {
 586                 jl = JOURNAL_LIST_ENTRY(entry->next);
 587                 if (jl->j_trans_id <= trans_id) {
 588                         return 1;
 589                 }
 590         }
 591         return 0;
 592 }
 593
 594 /*
 595  * If page->mapping was null, we failed to truncate this page for
 596  * some reason.  Most likely because it was truncated after being
 597  * logged via data=journal.
 598  *
 599  * This does a check to see if the buffer belongs to one of these
 600  * lost pages before doing the final put_bh.  If page->mapping was
 601  * null, it tries to free buffers on the page, which should make the
 602  * final put_page drop the page from the lru.
 603  */
 604 static void release_buffer_page(struct buffer_head *bh)
 605 {
 606         struct page *page = bh->b_page;
 607         if (!page->mapping && trylock_page(page)) {
 608                 get_page(page);
 609                 put_bh(bh);
 610                 if (!page->mapping)
 611                         try_to_free_buffers(page);
 612                 unlock_page(page);
 613                 put_page(page);
 614         } else {
 615                 put_bh(bh);
 616         }
 617 }
 618
 619 static void reiserfs_end_buffer_io_sync(struct buffer_head *bh, int uptodate)
 620 {
 621         if (buffer_journaled(bh)) {
 622                 reiserfs_warning(NULL, "clm-2084",
 623                                  "pinned buffer %lu:%pg sent to disk",
 624                                  bh->b_blocknr, bh->b_bdev);
 625         }
 626         if (uptodate)
 627                 set_buffer_uptodate(bh);
 628         else
 629                 clear_buffer_uptodate(bh);
 630
 631         unlock_buffer(bh);
 632         release_buffer_page(bh);
 633 }
 634
 635 static void reiserfs_end_ordered_io(struct buffer_head *bh, int uptodate)
 636 {
 637         if (uptodate)
 638                 set_buffer_uptodate(bh);
 639         else
 640                 clear_buffer_uptodate(bh);
 641         unlock_buffer(bh);
 642         put_bh(bh);
 643 }
 644
 645 static void submit_logged_buffer(struct buffer_head *bh)
 646 {
 647         get_bh(bh);
 648         bh->b_end_io = reiserfs_end_buffer_io_sync;
 649         clear_buffer_journal_new(bh);
 650         clear_buffer_dirty(bh);
 651         if (!test_clear_buffer_journal_test(bh))
 652                 BUG();
 653         if (!buffer_uptodate(bh))
 654                 BUG();
 655         submit_bh(REQ_OP_WRITE, 0, bh);
 656 }
 657
 658 static void submit_ordered_buffer(struct buffer_head *bh)
 659 {
 660         get_bh(bh);
 661         bh->b_end_io = reiserfs_end_ordered_io;
 662         clear_buffer_dirty(bh);
 663         if (!buffer_uptodate(bh))
 664                 BUG();
 665         submit_bh(REQ_OP_WRITE, 0, bh);
 666 }
 667
 668 #define CHUNK_SIZE 32
 669 struct buffer_chunk {
 670         struct buffer_head *bh[CHUNK_SIZE];
 671         int nr;
 672 };
 673
 674 static void write_chunk(struct buffer_chunk *chunk)
 675 {
 676         int i;
 677         for (i = 0; i < chunk->nr; i++) {
 678                 submit_logged_buffer(chunk->bh[i]);
 679         }
 680         chunk->nr = 0;
 681 }
 682
 683 static void write_ordered_chunk(struct buffer_chunk *chunk)
 684 {
 685         int i;
 686         for (i = 0; i < chunk->nr; i++) {
 687                 submit_ordered_buffer(chunk->bh[i]);
 688         }
 689         chunk->nr = 0;
 690 }
 691
 692 static int add_to_chunk(struct buffer_chunk *chunk, struct buffer_head *bh,
 693                         spinlock_t * lock, void (fn) (struct buffer_chunk *))
 694 {
 695         int ret = 0;
 696         BUG_ON(chunk->nr >= CHUNK_SIZE);
 697         chunk->bh[chunk->nr++] = bh;
 698         if (chunk->nr >= CHUNK_SIZE) {
 699                 ret = 1;
 700                 if (lock) {
 701                         spin_unlock(lock);
 702                         fn(chunk);
 703                         spin_lock(lock);
 704                 } else {
 705                         fn(chunk);
 706                 }
 707         }
 708         return ret;
 709 }
 710
 711 static atomic_t nr_reiserfs_jh = ATOMIC_INIT(0);
 712 static struct reiserfs_jh *alloc_jh(void)
 713 {
 714         struct reiserfs_jh *jh;
 715         while (1) {
 716                 jh = kmalloc(sizeof(*jh), GFP_NOFS);
 717                 if (jh) {
 718                         atomic_inc(&nr_reiserfs_jh);
 719                         return jh;
 720                 }
 721                 yield();
 722         }
 723 }
 724
 725 /*
 726  * we want to free the jh when the buffer has been written
 727  * and waited on
 728  */
 729 void reiserfs_free_jh(struct buffer_head *bh)
 730 {
 731         struct reiserfs_jh *jh;
 732
 733         jh = bh->b_private;
 734         if (jh) {
 735                 bh->b_private = NULL;
 736                 jh->bh = NULL;
 737                 list_del_init(&jh->list);
 738                 kfree(jh);
 739                 if (atomic_read(&nr_reiserfs_jh) <= 0)
 740                         BUG();
 741                 atomic_dec(&nr_reiserfs_jh);
 742                 put_bh(bh);
 743         }
 744 }
 745
 746 static inline int __add_jh(struct reiserfs_journal *j, struct buffer_head *bh,
 747                            int tail)
 748 {
 749         struct reiserfs_jh *jh;
 750
 751         if (bh->b_private) {
 752                 spin_lock(&j->j_dirty_buffers_lock);
 753                 if (!bh->b_private) {
 754                         spin_unlock(&j->j_dirty_buffers_lock);
 755                         goto no_jh;
 756                 }
 757                 jh = bh->b_private;
 758                 list_del_init(&jh->list);
 759         } else {
 760 no_jh:
 761                 get_bh(bh);
 762                 jh = alloc_jh();
 763                 spin_lock(&j->j_dirty_buffers_lock);
 764                 /*
 765                  * buffer must be locked for __add_jh, should be able to have
 766                  * two adds at the same time
 767                  */
 768                 BUG_ON(bh->b_private);
 769                 jh->bh = bh;
 770                 bh->b_private = jh;
 771         }
 772         jh->jl = j->j_current_jl;
 773         if (tail)
 774                 list_add_tail(&jh->list, &jh->jl->j_tail_bh_list);
 775         else {
 776                 list_add_tail(&jh->list, &jh->jl->j_bh_list);
 777         }
 778         spin_unlock(&j->j_dirty_buffers_lock);
 779         return 0;
 780 }
 781
 782 int reiserfs_add_tail_list(struct inode *inode, struct buffer_head *bh)
 783 {
 784         return __add_jh(SB_JOURNAL(inode->i_sb), bh, 1);
 785 }
 786 int reiserfs_add_ordered_list(struct inode *inode, struct buffer_head *bh)
 787 {
 788         return __add_jh(SB_JOURNAL(inode->i_sb), bh, 0);
 789 }
 790
 791 #define JH_ENTRY(l) list_entry((l), struct reiserfs_jh, list)
 792 static int write_ordered_buffers(spinlock_t * lock,
 793                                  struct reiserfs_journal *j,
 794                                  struct reiserfs_journal_list *jl,
 795                                  struct list_head *list)
 796 {
 797         struct buffer_head *bh;
 798         struct reiserfs_jh *jh;
 799         int ret = j->j_errno;
 800         struct buffer_chunk chunk;
 801         struct list_head tmp;
 802         INIT_LIST_HEAD(&tmp);
 803
 804         chunk.nr = 0;
 805         spin_lock(lock);
 806         while (!list_empty(list)) {
 807                 jh = JH_ENTRY(list->next);
 808                 bh = jh->bh;
 809                 get_bh(bh);
 810                 if (!trylock_buffer(bh)) {
 811                         if (!buffer_dirty(bh)) {
 812                                 list_move(&jh->list, &tmp);
 813                                 goto loop_next;
 814                         }
 815                         spin_unlock(lock);
 816                         if (chunk.nr)
 817                                 write_ordered_chunk(&chunk);
 818                         wait_on_buffer(bh);
 819                         cond_resched();
 820                         spin_lock(lock);
 821                         goto loop_next;
 822                 }
 823                 /*
 824                  * in theory, dirty non-uptodate buffers should never get here,
 825                  * but the upper layer io error paths still have a few quirks.
 826                  * Handle them here as gracefully as we can
 827                  */
 828                 if (!buffer_uptodate(bh) && buffer_dirty(bh)) {
 829                         clear_buffer_dirty(bh);
 830                         ret = -EIO;
 831                 }
 832                 if (buffer_dirty(bh)) {
 833                         list_move(&jh->list, &tmp);
 834                         add_to_chunk(&chunk, bh, lock, write_ordered_chunk);
 835                 } else {
 836                         reiserfs_free_jh(bh);
 837                         unlock_buffer(bh);
 838                 }
 839 loop_next:
 840                 put_bh(bh);
 841                 cond_resched_lock(lock);
 842         }
 843         if (chunk.nr) {
 844                 spin_unlock(lock);
 845                 write_ordered_chunk(&chunk);
 846                 spin_lock(lock);
 847         }
 848         while (!list_empty(&tmp)) {
 849                 jh = JH_ENTRY(tmp.prev);
 850                 bh = jh->bh;
 851                 get_bh(bh);
 852                 reiserfs_free_jh(bh);
 853
 854                 if (buffer_locked(bh)) {
 855                         spin_unlock(lock);
 856                         wait_on_buffer(bh);
 857                         spin_lock(lock);
 858                 }
 859                 if (!buffer_uptodate(bh)) {
 860                         ret = -EIO;
 861                 }
 862                 /*
 863                  * ugly interaction with invalidatepage here.
 864                  * reiserfs_invalidate_page will pin any buffer that has a
 865                  * valid journal head from an older transaction.  If someone
 866                  * else sets our buffer dirty after we write it in the first
 867                  * loop, and then someone truncates the page away, nobody
 868                  * will ever write the buffer. We're safe if we write the
 869                  * page one last time after freeing the journal header.
 870                  */
 871                 if (buffer_dirty(bh) && unlikely(bh->b_page->mapping == NULL)) {
 872                         spin_unlock(lock);
 873                         ll_rw_block(REQ_OP_WRITE, 0, 1, &bh);
 874                         spin_lock(lock);
 875                 }
 876                 put_bh(bh);
 877                 cond_resched_lock(lock);
 878         }
 879         spin_unlock(lock);
 880         return ret;
 881 }
 882
 883 static int flush_older_commits(struct super_block *s,
 884                                struct reiserfs_journal_list *jl)
 885 {
 886         struct reiserfs_journal *journal = SB_JOURNAL(s);
 887         struct reiserfs_journal_list *other_jl;
 888         struct reiserfs_journal_list *first_jl;
 889         struct list_head *entry;
 890         unsigned int trans_id = jl->j_trans_id;
 891         unsigned int other_trans_id;
 892
 893 find_first:
 894         /*
 895          * first we walk backwards to find the oldest uncommitted transation
 896          */
 897         first_jl = jl;
 898         entry = jl->j_list.prev;
 899         while (1) {
 900                 other_jl = JOURNAL_LIST_ENTRY(entry);
 901                 if (entry == &journal->j_journal_list ||
 902                     atomic_read(&other_jl->j_older_commits_done))
 903                         break;
 904
 905                 first_jl = other_jl;
 906                 entry = other_jl->j_list.prev;
 907         }
 908
 909         /* if we didn't find any older uncommitted transactions, return now */
 910         if (first_jl == jl) {
 911                 return 0;
 912         }
 913
 914         entry = &first_jl->j_list;
 915         while (1) {
 916                 other_jl = JOURNAL_LIST_ENTRY(entry);
 917                 other_trans_id = other_jl->j_trans_id;
 918
 919                 if (other_trans_id < trans_id) {
 920                         if (atomic_read(&other_jl->j_commit_left) != 0) {
 921                                 flush_commit_list(s, other_jl, 0);
 922
 923                                 /* list we were called with is gone, return */
 924                                 if (!journal_list_still_alive(s, trans_id))
 925                                         return 1;
 926
 927                                 /*
 928                                  * the one we just flushed is gone, this means
 929                                  * all older lists are also gone, so first_jl
 930                                  * is no longer valid either.  Go back to the
 931                                  * beginning.
 932                                  */
 933                                 if (!journal_list_still_alive
 934                                     (s, other_trans_id)) {
 935                                         goto find_first;
 936                                 }
 937                         }
 938                         entry = entry->next;
 939                         if (entry == &journal->j_journal_list)
 940                                 return 0;
 941                 } else {
 942                         return 0;
 943                 }
 944         }
 945         return 0;
 946 }
 947
 948 static int reiserfs_async_progress_wait(struct super_block *s)
 949 {
 950         struct reiserfs_journal *j = SB_JOURNAL(s);
 951
 952         if (atomic_read(&j->j_async_throttle)) {
 953                 int depth;
 954
 955                 depth = reiserfs_write_unlock_nested(s);
 956                 congestion_wait(BLK_RW_ASYNC, HZ / 10);
 957                 reiserfs_write_lock_nested(s, depth);
 958         }
 959
 960         return 0;
 961 }
 962
 963 /*
 964  * if this journal list still has commit blocks unflushed, send them to disk.
 965  *
 966  * log areas must be flushed in order (transaction 2 can't commit before
 967  * transaction 1) Before the commit block can by written, every other log
 968  * block must be safely on disk
 969  */
 970 static int flush_commit_list(struct super_block *s,
 971                              struct reiserfs_journal_list *jl, int flushall)
 972 {
 973         int i;
 974         b_blocknr_t bn;
 975         struct buffer_head *tbh = NULL;
 976         unsigned int trans_id = jl->j_trans_id;
 977         struct reiserfs_journal *journal = SB_JOURNAL(s);
 978         int retval = 0;
 979         int write_len;
 980         int depth;
 981
 982         reiserfs_check_lock_depth(s, "flush_commit_list");
 983
 984         if (atomic_read(&jl->j_older_commits_done)) {
 985                 return 0;
 986         }
 987
 988         /*
 989          * before we can put our commit blocks on disk, we have to make
 990          * sure everyone older than us is on disk too
 991          */
 992         BUG_ON(jl->j_len <= 0);
 993         BUG_ON(trans_id == journal->j_trans_id);
 994
 995         get_journal_list(jl);
 996         if (flushall) {
 997                 if (flush_older_commits(s, jl) == 1) {
 998                         /*
 999                          * list disappeared during flush_older_commits.
1000                          * return
1001                          */
1002                         goto put_jl;
1003                 }
1004         }
1005
1006         /* make sure nobody is trying to flush this one at the same time */
1007         reiserfs_mutex_lock_safe(&jl->j_commit_mutex, s);
1008
1009         if (!journal_list_still_alive(s, trans_id)) {
1010                 mutex_unlock(&jl->j_commit_mutex);
1011                 goto put_jl;
1012         }
1013         BUG_ON(jl->j_trans_id == 0);
1014
1015         /* this commit is done, exit */
1016         if (atomic_read(&jl->j_commit_left) <= 0) {
1017                 if (flushall) {
1018                         atomic_set(&jl->j_older_commits_done, 1);
1019                 }
1020                 mutex_unlock(&jl->j_commit_mutex);
1021                 goto put_jl;
1022         }
1023
1024         if (!list_empty(&jl->j_bh_list)) {
1025                 int ret;
1026
1027                 /*
1028                  * We might sleep in numerous places inside
1029                  * write_ordered_buffers. Relax the write lock.
1030                  */
1031                 depth = reiserfs_write_unlock_nested(s);
1032                 ret = write_ordered_buffers(&journal->j_dirty_buffers_lock,
1033                                             journal, jl, &jl->j_bh_list);
1034                 if (ret < 0 && retval == 0)
1035                         retval = ret;
1036                 reiserfs_write_lock_nested(s, depth);
1037         }
1038         BUG_ON(!list_empty(&jl->j_bh_list));
1039         /*
1040          * for the description block and all the log blocks, submit any buffers
1041          * that haven't already reached the disk.  Try to write at least 256
1042          * log blocks. later on, we will only wait on blocks that correspond
1043          * to this transaction, but while we're unplugging we might as well
1044          * get a chunk of data on there.
1045          */
1046         atomic_inc(&journal->j_async_throttle);
1047         write_len = jl->j_len + 1;
1048         if (write_len < 256)
1049                 write_len = 256;
1050         for (i = 0 ; i < write_len ; i++) {
1051                 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + (jl->j_start + i) %
1052                     SB_ONDISK_JOURNAL_SIZE(s);
1053                 tbh = journal_find_get_block(s, bn);
1054                 if (tbh) {
1055                         if (buffer_dirty(tbh)) {
1056                             depth = reiserfs_write_unlock_nested(s);
1057                             ll_rw_block(REQ_OP_WRITE, 0, 1, &tbh);
1058                             reiserfs_write_lock_nested(s, depth);
1059                         }
1060                         put_bh(tbh) ;
1061                 }
1062         }
1063         atomic_dec(&journal->j_async_throttle);
1064
1065         for (i = 0; i < (jl->j_len + 1); i++) {
1066                 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) +
1067                     (jl->j_start + i) % SB_ONDISK_JOURNAL_SIZE(s);
1068                 tbh = journal_find_get_block(s, bn);
1069
1070                 depth = reiserfs_write_unlock_nested(s);
1071                 __wait_on_buffer(tbh);
1072                 reiserfs_write_lock_nested(s, depth);
1073                 /*
1074                  * since we're using ll_rw_blk above, it might have skipped
1075                  * over a locked buffer.  Double check here
1076                  */
1077                 /* redundant, sync_dirty_buffer() checks */
1078                 if (buffer_dirty(tbh)) {
1079                         depth = reiserfs_write_unlock_nested(s);
1080                         sync_dirty_buffer(tbh);
1081                         reiserfs_write_lock_nested(s, depth);
1082                 }
1083                 if (unlikely(!buffer_uptodate(tbh))) {
1084 #ifdef CONFIG_REISERFS_CHECK
1085                         reiserfs_warning(s, "journal-601",
1086                                          "buffer write failed");
1087 #endif
1088                         retval = -EIO;
1089                 }
1090                 /* once for journal_find_get_block */
1091                 put_bh(tbh);
1092                 /* once due to original getblk in do_journal_end */
1093                 put_bh(tbh);
1094                 atomic_dec(&jl->j_commit_left);
1095         }
1096
1097         BUG_ON(atomic_read(&jl->j_commit_left) != 1);
1098
1099         /*
1100          * If there was a write error in the journal - we can't commit
1101          * this transaction - it will be invalid and, if successful,
1102          * will just end up propagating the write error out to
1103          * the file system.
1104          */
1105         if (likely(!retval && !reiserfs_is_journal_aborted (journal))) {
1106                 if (buffer_dirty(jl->j_commit_bh))
1107                         BUG();
1108                 mark_buffer_dirty(jl->j_commit_bh) ;
1109                 depth = reiserfs_write_unlock_nested(s);
1110                 if (reiserfs_barrier_flush(s))
1111                         __sync_dirty_buffer(jl->j_commit_bh,
1112                                         REQ_SYNC | REQ_PREFLUSH | REQ_FUA);
1113                 else
1114                         sync_dirty_buffer(jl->j_commit_bh);
1115                 reiserfs_write_lock_nested(s, depth);
1116         }
1117
1118         /*
1119          * If there was a write error in the journal - we can't commit this
1120          * transaction - it will be invalid and, if successful, will just end
1121          * up propagating the write error out to the filesystem.
1122          */
1123         if (unlikely(!buffer_uptodate(jl->j_commit_bh))) {
1124 #ifdef CONFIG_REISERFS_CHECK
1125                 reiserfs_warning(s, "journal-615", "buffer write failed");
1126 #endif
1127                 retval = -EIO;
1128         }
1129         bforget(jl->j_commit_bh);
1130         if (journal->j_last_commit_id != 0 &&
1131             (jl->j_trans_id - journal->j_last_commit_id) != 1) {
1132                 reiserfs_warning(s, "clm-2200", "last commit %lu, current %lu",
1133                                  journal->j_last_commit_id, jl->j_trans_id);
1134         }
1135         journal->j_last_commit_id = jl->j_trans_id;
1136
1137         /*
1138          * now, every commit block is on the disk.  It is safe to allow
1139          * blocks freed during this transaction to be reallocated
1140          */
1141         cleanup_freed_for_journal_list(s, jl);
1142
1143         retval = retval ? retval : journal->j_errno;
1144
1145         /* mark the metadata dirty */
1146         if (!retval)
1147                 dirty_one_transaction(s, jl);
1148         atomic_dec(&jl->j_commit_left);
1149
1150         if (flushall) {
1151                 atomic_set(&jl->j_older_commits_done, 1);
1152         }
1153         mutex_unlock(&jl->j_commit_mutex);
1154 put_jl:
1155         put_journal_list(s, jl);
1156
1157         if (retval)
1158                 reiserfs_abort(s, retval, "Journal write error in %s",
1159                                __func__);
1160         return retval;
1161 }
1162
1163 /*
1164  * flush_journal_list frequently needs to find a newer transaction for a
1165  * given block.  This does that, or returns NULL if it can't find anything
1166  */
1167 static struct reiserfs_journal_list *find_newer_jl_for_cn(struct
1168                                                           reiserfs_journal_cnode
1169                                                           *cn)
1170 {
1171         struct super_block *sb = cn->sb;
1172         b_blocknr_t blocknr = cn->blocknr;
1173
1174         cn = cn->hprev;
1175         while (cn) {
1176                 if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist) {
1177                         return cn->jlist;
1178                 }
1179                 cn = cn->hprev;
1180         }
1181         return NULL;
1182 }
1183
1184 static void remove_journal_hash(struct super_block *,
1185                                 struct reiserfs_journal_cnode **,
1186                                 struct reiserfs_journal_list *, unsigned long,
1187                                 int);
1188
1189 /*
1190  * once all the real blocks have been flushed, it is safe to remove them
1191  * from the journal list for this transaction.  Aside from freeing the
1192  * cnode, this also allows the block to be reallocated for data blocks
1193  * if it had been deleted.
1194  */
1195 static void remove_all_from_journal_list(struct super_block *sb,
1196                                          struct reiserfs_journal_list *jl,
1197                                          int debug)
1198 {
1199         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1200         struct reiserfs_journal_cnode *cn, *last;
1201         cn = jl->j_realblock;
1202
1203         /*
1204          * which is better, to lock once around the whole loop, or
1205          * to lock for each call to remove_journal_hash?
1206          */
1207         while (cn) {
1208                 if (cn->blocknr != 0) {
1209                         if (debug) {
1210                                 reiserfs_warning(sb, "reiserfs-2201",
1211                                                  "block %u, bh is %d, state %ld",
1212                                                  cn->blocknr, cn->bh ? 1 : 0,
1213                                                  cn->state);
1214                         }
1215                         cn->state = 0;
1216                         remove_journal_hash(sb, journal->j_list_hash_table,
1217                                             jl, cn->blocknr, 1);
1218                 }
1219                 last = cn;
1220                 cn = cn->next;
1221                 free_cnode(sb, last);
1222         }
1223         jl->j_realblock = NULL;
1224 }
1225
1226 /*
1227  * if this timestamp is greater than the timestamp we wrote last to the
1228  * header block, write it to the header block.  once this is done, I can
1229  * safely say the log area for this transaction won't ever be replayed,
1230  * and I can start releasing blocks in this transaction for reuse as data
1231  * blocks.  called by flush_journal_list, before it calls
1232  * remove_all_from_journal_list
1233  */
1234 static int _update_journal_header_block(struct super_block *sb,
1235                                         unsigned long offset,
1236                                         unsigned int trans_id)
1237 {
1238         struct reiserfs_journal_header *jh;
1239         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1240         int depth;
1241
1242         if (reiserfs_is_journal_aborted(journal))
1243                 return -EIO;
1244
1245         if (trans_id >= journal->j_last_flush_trans_id) {
1246                 if (buffer_locked((journal->j_header_bh))) {
1247                         depth = reiserfs_write_unlock_nested(sb);
1248                         __wait_on_buffer(journal->j_header_bh);
1249                         reiserfs_write_lock_nested(sb, depth);
1250                         if (unlikely(!buffer_uptodate(journal->j_header_bh))) {
1251 #ifdef CONFIG_REISERFS_CHECK
1252                                 reiserfs_warning(sb, "journal-699",
1253                                                  "buffer write failed");
1254 #endif
1255                                 return -EIO;
1256                         }
1257                 }
1258                 journal->j_last_flush_trans_id = trans_id;
1259                 journal->j_first_unflushed_offset = offset;
1260                 jh = (struct reiserfs_journal_header *)(journal->j_header_bh->
1261                                                         b_data);
1262                 jh->j_last_flush_trans_id = cpu_to_le32(trans_id);
1263                 jh->j_first_unflushed_offset = cpu_to_le32(offset);
1264                 jh->j_mount_id = cpu_to_le32(journal->j_mount_id);
1265
1266                 set_buffer_dirty(journal->j_header_bh);
1267                 depth = reiserfs_write_unlock_nested(sb);
1268
1269                 if (reiserfs_barrier_flush(sb))
1270                         __sync_dirty_buffer(journal->j_header_bh,
1271                                         REQ_SYNC | REQ_PREFLUSH | REQ_FUA);
1272                 else
1273                         sync_dirty_buffer(journal->j_header_bh);
1274
1275                 reiserfs_write_lock_nested(sb, depth);
1276                 if (!buffer_uptodate(journal->j_header_bh)) {
1277                         reiserfs_warning(sb, "journal-837",
1278                                          "IO error during journal replay");
1279                         return -EIO;
1280                 }
1281         }
1282         return 0;
1283 }
1284
1285 static int update_journal_header_block(struct super_block *sb,
1286                                        unsigned long offset,
1287                                        unsigned int trans_id)
1288 {
1289         return _update_journal_header_block(sb, offset, trans_id);
1290 }
1291
1292 /*
1293 ** flush any and all journal lists older than you are
1294 ** can only be called from flush_journal_list
1295 */
1296 static int flush_older_journal_lists(struct super_block *sb,
1297                                      struct reiserfs_journal_list *jl)
1298 {
1299         struct list_head *entry;
1300         struct reiserfs_journal_list *other_jl;
1301         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1302         unsigned int trans_id = jl->j_trans_id;
1303
1304         /*
1305          * we know we are the only ones flushing things, no extra race
1306          * protection is required.
1307          */
1308 restart:
1309         entry = journal->j_journal_list.next;
1310         /* Did we wrap? */
1311         if (entry == &journal->j_journal_list)
1312                 return 0;
1313         other_jl = JOURNAL_LIST_ENTRY(entry);
1314         if (other_jl->j_trans_id < trans_id) {
1315                 BUG_ON(other_jl->j_refcount <= 0);
1316                 /* do not flush all */
1317                 flush_journal_list(sb, other_jl, 0);
1318
1319                 /* other_jl is now deleted from the list */
1320                 goto restart;
1321         }
1322         return 0;
1323 }
1324
1325 static void del_from_work_list(struct super_block *s,
1326                                struct reiserfs_journal_list *jl)
1327 {
1328         struct reiserfs_journal *journal = SB_JOURNAL(s);
1329         if (!list_empty(&jl->j_working_list)) {
1330                 list_del_init(&jl->j_working_list);
1331                 journal->j_num_work_lists--;
1332         }
1333 }
1334
1335 /*
1336  * flush a journal list, both commit and real blocks
1337  *
1338  * always set flushall to 1, unless you are calling from inside
1339  * flush_journal_list
1340  *
1341  * IMPORTANT.  This can only be called while there are no journal writers,
1342  * and the journal is locked.  That means it can only be called from
1343  * do_journal_end, or by journal_release
1344  */
1345 static int flush_journal_list(struct super_block *s,
1346                               struct reiserfs_journal_list *jl, int flushall)
1347 {
1348         struct reiserfs_journal_list *pjl;
1349         struct reiserfs_journal_cnode *cn;
1350         int count;
1351         int was_jwait = 0;
1352         int was_dirty = 0;
1353         struct buffer_head *saved_bh;
1354         unsigned long j_len_saved = jl->j_len;
1355         struct reiserfs_journal *journal = SB_JOURNAL(s);
1356         int err = 0;
1357         int depth;
1358
1359         BUG_ON(j_len_saved <= 0);
1360
1361         if (atomic_read(&journal->j_wcount) != 0) {
1362                 reiserfs_warning(s, "clm-2048", "called with wcount %d",
1363                                  atomic_read(&journal->j_wcount));
1364         }
1365
1366         /* if flushall == 0, the lock is already held */
1367         if (flushall) {
1368                 reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s);
1369         } else if (mutex_trylock(&journal->j_flush_mutex)) {
1370                 BUG();
1371         }
1372
1373         count = 0;
1374         if (j_len_saved > journal->j_trans_max) {
1375                 reiserfs_panic(s, "journal-715", "length is %lu, trans id %lu",
1376                                j_len_saved, jl->j_trans_id);
1377                 return 0;
1378         }
1379
1380         /* if all the work is already done, get out of here */
1381         if (atomic_read(&jl->j_nonzerolen) <= 0 &&
1382             atomic_read(&jl->j_commit_left) <= 0) {
1383                 goto flush_older_and_return;
1384         }
1385
1386         /*
1387          * start by putting the commit list on disk.  This will also flush
1388          * the commit lists of any olders transactions
1389          */
1390         flush_commit_list(s, jl, 1);
1391
1392         if (!(jl->j_state & LIST_DIRTY)
1393             && !reiserfs_is_journal_aborted(journal))
1394                 BUG();
1395
1396         /* are we done now? */
1397         if (atomic_read(&jl->j_nonzerolen) <= 0 &&
1398             atomic_read(&jl->j_commit_left) <= 0) {
1399                 goto flush_older_and_return;
1400         }
1401
1402         /*
1403          * loop through each cnode, see if we need to write it,
1404          * or wait on a more recent transaction, or just ignore it
1405          */
1406         if (atomic_read(&journal->j_wcount) != 0) {
1407                 reiserfs_panic(s, "journal-844", "journal list is flushing, "
1408                                "wcount is not 0");
1409         }
1410         cn = jl->j_realblock;
1411         while (cn) {
1412                 was_jwait = 0;
1413                 was_dirty = 0;
1414                 saved_bh = NULL;
1415                 /* blocknr of 0 is no longer in the hash, ignore it */
1416                 if (cn->blocknr == 0) {
1417                         goto free_cnode;
1418                 }
1419
1420                 /*
1421                  * This transaction failed commit.
1422                  * Don't write out to the disk
1423                  */
1424                 if (!(jl->j_state & LIST_DIRTY))
1425                         goto free_cnode;
1426
1427                 pjl = find_newer_jl_for_cn(cn);
1428                 /*
1429                  * the order is important here.  We check pjl to make sure we
1430                  * don't clear BH_JDirty_wait if we aren't the one writing this
1431                  * block to disk
1432                  */
1433                 if (!pjl && cn->bh) {
1434                         saved_bh = cn->bh;
1435
1436                         /*
1437                          * we do this to make sure nobody releases the
1438                          * buffer while we are working with it
1439                          */
1440                         get_bh(saved_bh);
1441
1442                         if (buffer_journal_dirty(saved_bh)) {
1443                                 BUG_ON(!can_dirty(cn));
1444                                 was_jwait = 1;
1445                                 was_dirty = 1;
1446                         } else if (can_dirty(cn)) {
1447                                 /*
1448                                  * everything with !pjl && jwait
1449                                  * should be writable
1450                                  */
1451                                 BUG();
1452                         }
1453                 }
1454
1455                 /*
1456                  * if someone has this block in a newer transaction, just make
1457                  * sure they are committed, and don't try writing it to disk
1458                  */
1459                 if (pjl) {
1460                         if (atomic_read(&pjl->j_commit_left))
1461                                 flush_commit_list(s, pjl, 1);
1462                         goto free_cnode;
1463                 }
1464
1465                 /*
1466                  * bh == NULL when the block got to disk on its own, OR,
1467                  * the block got freed in a future transaction
1468                  */
1469                 if (saved_bh == NULL) {
1470                         goto free_cnode;
1471                 }
1472
1473                 /*
1474                  * this should never happen.  kupdate_one_transaction has
1475                  * this list locked while it works, so we should never see a
1476                  * buffer here that is not marked JDirty_wait
1477                  */
1478                 if ((!was_jwait) && !buffer_locked(saved_bh)) {
1479                         reiserfs_warning(s, "journal-813",
1480                                          "BAD! buffer %llu %cdirty %cjwait, "
1481                                          "not in a newer transaction",
1482                                          (unsigned long long)saved_bh->
1483                                          b_blocknr, was_dirty ? ' ' : '!',
1484                                          was_jwait ? ' ' : '!');
1485                 }
1486                 if (was_dirty) {
1487                         /*
1488                          * we inc again because saved_bh gets decremented
1489                          * at free_cnode
1490                          */
1491                         get_bh(saved_bh);
1492                         set_bit(BLOCK_NEEDS_FLUSH, &cn->state);
1493                         lock_buffer(saved_bh);
1494                         BUG_ON(cn->blocknr != saved_bh->b_blocknr);
1495                         if (buffer_dirty(saved_bh))
1496                                 submit_logged_buffer(saved_bh);
1497                         else
1498                                 unlock_buffer(saved_bh);
1499                         count++;
1500                 } else {
1501                         reiserfs_warning(s, "clm-2082",
1502                                          "Unable to flush buffer %llu in %s",
1503                                          (unsigned long long)saved_bh->
1504                                          b_blocknr, __func__);
1505                 }
1506 free_cnode:
1507                 cn = cn->next;
1508                 if (saved_bh) {
1509                         /*
1510                          * we incremented this to keep others from
1511                          * taking the buffer head away
1512                          */
1513                         put_bh(saved_bh);
1514                         if (atomic_read(&saved_bh->b_count) < 0) {
1515                                 reiserfs_warning(s, "journal-945",
1516                                                  "saved_bh->b_count < 0");
1517                         }
1518                 }
1519         }
1520         if (count > 0) {
1521                 cn = jl->j_realblock;
1522                 while (cn) {
1523                         if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) {
1524                                 if (!cn->bh) {
1525                                         reiserfs_panic(s, "journal-1011",
1526                                                        "cn->bh is NULL");
1527                                 }
1528
1529                                 depth = reiserfs_write_unlock_nested(s);
1530                                 __wait_on_buffer(cn->bh);
1531                                 reiserfs_write_lock_nested(s, depth);
1532
1533                                 if (!cn->bh) {
1534                                         reiserfs_panic(s, "journal-1012",
1535                                                        "cn->bh is NULL");
1536                                 }
1537                                 if (unlikely(!buffer_uptodate(cn->bh))) {
1538 #ifdef CONFIG_REISERFS_CHECK
1539                                         reiserfs_warning(s, "journal-949",
1540                                                          "buffer write failed");
1541 #endif
1542                                         err = -EIO;
1543                                 }
1544                                 /*
1545                                  * note, we must clear the JDirty_wait bit
1546                                  * after the up to date check, otherwise we
1547                                  * race against our flushpage routine
1548                                  */
1549                                 BUG_ON(!test_clear_buffer_journal_dirty
1550                                        (cn->bh));
1551
1552                                 /* drop one ref for us */
1553                                 put_bh(cn->bh);
1554                                 /* drop one ref for journal_mark_dirty */
1555                                 release_buffer_page(cn->bh);
1556                         }
1557                         cn = cn->next;
1558                 }
1559         }
1560
1561         if (err)
1562                 reiserfs_abort(s, -EIO,
1563                                "Write error while pushing transaction to disk in %s",
1564                                __func__);
1565 flush_older_and_return:
1566
1567         /*
1568          * before we can update the journal header block, we _must_ flush all
1569          * real blocks from all older transactions to disk.  This is because
1570          * once the header block is updated, this transaction will not be
1571          * replayed after a crash
1572          */
1573         if (flushall) {
1574                 flush_older_journal_lists(s, jl);
1575         }
1576
1577         err = journal->j_errno;
1578         /*
1579          * before we can remove everything from the hash tables for this
1580          * transaction, we must make sure it can never be replayed
1581          *
1582          * since we are only called from do_journal_end, we know for sure there
1583          * are no allocations going on while we are flushing journal lists.  So,
1584          * we only need to update the journal header block for the last list
1585          * being flushed
1586          */
1587         if (!err && flushall) {
1588                 err =
1589                     update_journal_header_block(s,
1590                                                 (jl->j_start + jl->j_len +
1591                                                  2) % SB_ONDISK_JOURNAL_SIZE(s),
1592                                                 jl->j_trans_id);
1593                 if (err)
1594                         reiserfs_abort(s, -EIO,
1595                                        "Write error while updating journal header in %s",
1596                                        __func__);
1597         }
1598         remove_all_from_journal_list(s, jl, 0);
1599         list_del_init(&jl->j_list);
1600         journal->j_num_lists--;
1601         del_from_work_list(s, jl);
1602
1603         if (journal->j_last_flush_id != 0 &&
1604             (jl->j_trans_id - journal->j_last_flush_id) != 1) {
1605                 reiserfs_warning(s, "clm-2201", "last flush %lu, current %lu",
1606                                  journal->j_last_flush_id, jl->j_trans_id);
1607         }
1608         journal->j_last_flush_id = jl->j_trans_id;
1609
1610         /*
1611          * not strictly required since we are freeing the list, but it should
1612          * help find code using dead lists later on
1613          */
1614         jl->j_len = 0;
1615         atomic_set(&jl->j_nonzerolen, 0);
1616         jl->j_start = 0;
1617         jl->j_realblock = NULL;
1618         jl->j_commit_bh = NULL;
1619         jl->j_trans_id = 0;
1620         jl->j_state = 0;
1621         put_journal_list(s, jl);
1622         if (flushall)
1623                 mutex_unlock(&journal->j_flush_mutex);
1624         return err;
1625 }
1626
1627 static int write_one_transaction(struct super_block *s,
1628                                  struct reiserfs_journal_list *jl,
1629                                  struct buffer_chunk *chunk)
1630 {
1631         struct reiserfs_journal_cnode *cn;
1632         int ret = 0;
1633
1634         jl->j_state |= LIST_TOUCHED;
1635         del_from_work_list(s, jl);
1636         if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0) {
1637                 return 0;
1638         }
1639
1640         cn = jl->j_realblock;
1641         while (cn) {
1642                 /*
1643                  * if the blocknr == 0, this has been cleared from the hash,
1644                  * skip it
1645                  */
1646                 if (cn->blocknr == 0) {
1647                         goto next;
1648                 }
1649                 if (cn->bh && can_dirty(cn) && buffer_dirty(cn->bh)) {
1650                         struct buffer_head *tmp_bh;
1651                         /*
1652                          * we can race against journal_mark_freed when we try
1653                          * to lock_buffer(cn->bh), so we have to inc the buffer
1654                          * count, and recheck things after locking
1655                          */
1656                         tmp_bh = cn->bh;
1657                         get_bh(tmp_bh);
1658                         lock_buffer(tmp_bh);
1659                         if (cn->bh && can_dirty(cn) && buffer_dirty(tmp_bh)) {
1660                                 if (!buffer_journal_dirty(tmp_bh) ||
1661                                     buffer_journal_prepared(tmp_bh))
1662                                         BUG();
1663                                 add_to_chunk(chunk, tmp_bh, NULL, write_chunk);
1664                                 ret++;
1665                         } else {
1666                                 /* note, cn->bh might be null now */
1667                                 unlock_buffer(tmp_bh);
1668                         }
1669                         put_bh(tmp_bh);
1670                 }
1671 next:
1672                 cn = cn->next;
1673                 cond_resched();
1674         }
1675         return ret;
1676 }
1677
1678 /* used by flush_commit_list */
1679 static void dirty_one_transaction(struct super_block *s,
1680                                  struct reiserfs_journal_list *jl)
1681 {
1682         struct reiserfs_journal_cnode *cn;
1683         struct reiserfs_journal_list *pjl;
1684
1685         jl->j_state |= LIST_DIRTY;
1686         cn = jl->j_realblock;
1687         while (cn) {
1688                 /*
1689                  * look for a more recent transaction that logged this
1690                  * buffer.  Only the most recent transaction with a buffer in
1691                  * it is allowed to send that buffer to disk
1692                  */
1693                 pjl = find_newer_jl_for_cn(cn);
1694                 if (!pjl && cn->blocknr && cn->bh
1695                     && buffer_journal_dirty(cn->bh)) {
1696                         BUG_ON(!can_dirty(cn));
1697                         /*
1698                          * if the buffer is prepared, it will either be logged
1699                          * or restored.  If restored, we need to make sure
1700                          * it actually gets marked dirty
1701                          */
1702                         clear_buffer_journal_new(cn->bh);
1703                         if (buffer_journal_prepared(cn->bh)) {
1704                                 set_buffer_journal_restore_dirty(cn->bh);
1705                         } else {
1706                                 set_buffer_journal_test(cn->bh);
1707                                 mark_buffer_dirty(cn->bh);
1708                         }
1709                 }
1710                 cn = cn->next;
1711         }
1712 }
1713
1714 static int kupdate_transactions(struct super_block *s,
1715                                 struct reiserfs_journal_list *jl,
1716                                 struct reiserfs_journal_list **next_jl,
1717                                 unsigned int *next_trans_id,
1718                                 int num_blocks, int num_trans)
1719 {
1720         int ret = 0;
1721         int written = 0;
1722         int transactions_flushed = 0;
1723         unsigned int orig_trans_id = jl->j_trans_id;
1724         struct buffer_chunk chunk;
1725         struct list_head *entry;
1726         struct reiserfs_journal *journal = SB_JOURNAL(s);
1727         chunk.nr = 0;
1728
1729         reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s);
1730         if (!journal_list_still_alive(s, orig_trans_id)) {
1731                 goto done;
1732         }
1733
1734         /*
1735          * we've got j_flush_mutex held, nobody is going to delete any
1736          * of these lists out from underneath us
1737          */
1738         while ((num_trans && transactions_flushed < num_trans) ||
1739                (!num_trans && written < num_blocks)) {
1740
1741                 if (jl->j_len == 0 || (jl->j_state & LIST_TOUCHED) ||
1742                     atomic_read(&jl->j_commit_left)
1743                     || !(jl->j_state & LIST_DIRTY)) {
1744                         del_from_work_list(s, jl);
1745                         break;
1746                 }
1747                 ret = write_one_transaction(s, jl, &chunk);
1748
1749                 if (ret < 0)
1750                         goto done;
1751                 transactions_flushed++;
1752                 written += ret;
1753                 entry = jl->j_list.next;
1754
1755                 /* did we wrap? */
1756                 if (entry == &journal->j_journal_list) {
1757                         break;
1758                 }
1759                 jl = JOURNAL_LIST_ENTRY(entry);
1760
1761                 /* don't bother with older transactions */
1762                 if (jl->j_trans_id <= orig_trans_id)
1763                         break;
1764         }
1765         if (chunk.nr) {
1766                 write_chunk(&chunk);
1767         }
1768
1769 done:
1770         mutex_unlock(&journal->j_flush_mutex);
1771         return ret;
1772 }
1773
1774 /*
1775  * for o_sync and fsync heavy applications, they tend to use
1776  * all the journa list slots with tiny transactions.  These
1777  * trigger lots and lots of calls to update the header block, which
1778  * adds seeks and slows things down.
1779  *
1780  * This function tries to clear out a large chunk of the journal lists
1781  * at once, which makes everything faster since only the newest journal
1782  * list updates the header block
1783  */
1784 static int flush_used_journal_lists(struct super_block *s,
1785                                     struct reiserfs_journal_list *jl)
1786 {
1787         unsigned long len = 0;
1788         unsigned long cur_len;
1789         int i;
1790         int limit = 256;
1791         struct reiserfs_journal_list *tjl;
1792         struct reiserfs_journal_list *flush_jl;
1793         unsigned int trans_id;
1794         struct reiserfs_journal *journal = SB_JOURNAL(s);
1795
1796         flush_jl = tjl = jl;
1797
1798         /* in data logging mode, try harder to flush a lot of blocks */
1799         if (reiserfs_data_log(s))
1800                 limit = 1024;
1801         /* flush for 256 transactions or limit blocks, whichever comes first */
1802         for (i = 0; i < 256 && len < limit; i++) {
1803                 if (atomic_read(&tjl->j_commit_left) ||
1804                     tjl->j_trans_id < jl->j_trans_id) {
1805                         break;
1806                 }
1807                 cur_len = atomic_read(&tjl->j_nonzerolen);
1808                 if (cur_len > 0) {
1809                         tjl->j_state &= ~LIST_TOUCHED;
1810                 }
1811                 len += cur_len;
1812                 flush_jl = tjl;
1813                 if (tjl->j_list.next == &journal->j_journal_list)
1814                         break;
1815                 tjl = JOURNAL_LIST_ENTRY(tjl->j_list.next);
1816         }
1817         get_journal_list(jl);
1818         get_journal_list(flush_jl);
1819
1820         /*
1821          * try to find a group of blocks we can flush across all the
1822          * transactions, but only bother if we've actually spanned
1823          * across multiple lists
1824          */
1825         if (flush_jl != jl)
1826                 kupdate_transactions(s, jl, &tjl, &trans_id, len, i);
1827
1828         flush_journal_list(s, flush_jl, 1);
1829         put_journal_list(s, flush_jl);
1830         put_journal_list(s, jl);
1831         return 0;
1832 }
1833
1834 /*
1835  * removes any nodes in table with name block and dev as bh.
1836  * only touchs the hnext and hprev pointers.
1837  */
1838 static void remove_journal_hash(struct super_block *sb,
1839                          struct reiserfs_journal_cnode **table,
1840                          struct reiserfs_journal_list *jl,
1841                          unsigned long block, int remove_freed)
1842 {
1843         struct reiserfs_journal_cnode *cur;
1844         struct reiserfs_journal_cnode **head;
1845
1846         head = &(journal_hash(table, sb, block));
1847         if (!head) {
1848                 return;
1849         }
1850         cur = *head;
1851         while (cur) {
1852                 if (cur->blocknr == block && cur->sb == sb
1853                     && (jl == NULL || jl == cur->jlist)
1854                     && (!test_bit(BLOCK_FREED, &cur->state) || remove_freed)) {
1855                         if (cur->hnext) {
1856                                 cur->hnext->hprev = cur->hprev;
1857                         }
1858                         if (cur->hprev) {
1859                                 cur->hprev->hnext = cur->hnext;
1860                         } else {
1861                                 *head = cur->hnext;
1862                         }
1863                         cur->blocknr = 0;
1864                         cur->sb = NULL;
1865                         cur->state = 0;
1866                         /*
1867                          * anybody who clears the cur->bh will also
1868                          * dec the nonzerolen
1869                          */
1870                         if (cur->bh && cur->jlist)
1871                                 atomic_dec(&cur->jlist->j_nonzerolen);
1872                         cur->bh = NULL;
1873                         cur->jlist = NULL;
1874                 }
1875                 cur = cur->hnext;
1876         }
1877 }
1878
1879 static void free_journal_ram(struct super_block *sb)
1880 {
1881         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1882         kfree(journal->j_current_jl);
1883         journal->j_num_lists--;
1884
1885         vfree(journal->j_cnode_free_orig);
1886         free_list_bitmaps(sb, journal->j_list_bitmap);
1887         free_bitmap_nodes(sb);  /* must be after free_list_bitmaps */
1888         if (journal->j_header_bh) {
1889                 brelse(journal->j_header_bh);
1890         }
1891         /*
1892          * j_header_bh is on the journal dev, make sure
1893          * not to release the journal dev until we brelse j_header_bh
1894          */
1895         release_journal_dev(sb, journal);
1896         vfree(journal);
1897 }
1898
1899 /*
1900  * call on unmount.  Only set error to 1 if you haven't made your way out
1901  * of read_super() yet.  Any other caller must keep error at 0.
1902  */
1903 static int do_journal_release(struct reiserfs_transaction_handle *th,
1904                               struct super_block *sb, int error)
1905 {
1906         struct reiserfs_transaction_handle myth;
1907         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1908
1909         /*
1910          * we only want to flush out transactions if we were
1911          * called with error == 0
1912          */
1913         if (!error && !sb_rdonly(sb)) {
1914                 /* end the current trans */
1915                 BUG_ON(!th->t_trans_id);
1916                 do_journal_end(th, FLUSH_ALL);
1917
1918                 /*
1919                  * make sure something gets logged to force
1920                  * our way into the flush code
1921                  */
1922                 if (!journal_join(&myth, sb)) {
1923                         reiserfs_prepare_for_journal(sb,
1924                                                      SB_BUFFER_WITH_SB(sb),
1925                                                      1);
1926                         journal_mark_dirty(&myth, SB_BUFFER_WITH_SB(sb));
1927                         do_journal_end(&myth, FLUSH_ALL);
1928                 }
1929         }
1930
1931         /* this also catches errors during the do_journal_end above */
1932         if (!error && reiserfs_is_journal_aborted(journal)) {
1933                 memset(&myth, 0, sizeof(myth));
1934                 if (!journal_join_abort(&myth, sb)) {
1935                         reiserfs_prepare_for_journal(sb,
1936                                                      SB_BUFFER_WITH_SB(sb),
1937                                                      1);
1938                         journal_mark_dirty(&myth, SB_BUFFER_WITH_SB(sb));
1939                         do_journal_end(&myth, FLUSH_ALL);
1940                 }
1941         }
1942
1943
1944         /*
1945          * We must release the write lock here because
1946          * the workqueue job (flush_async_commit) needs this lock
1947          */
1948         reiserfs_write_unlock(sb);
1949
1950         /*
1951          * Cancel flushing of old commits. Note that neither of these works
1952          * will be requeued because superblock is being shutdown and doesn't
1953          * have SB_ACTIVE set.
1954          */
1955         reiserfs_cancel_old_flush(sb);
1956         /* wait for all commits to finish */
1957         cancel_delayed_work_sync(&SB_JOURNAL(sb)->j_work);
1958
1959         free_journal_ram(sb);
1960
1961         reiserfs_write_lock(sb);
1962
1963         return 0;
1964 }
1965
1966 /* * call on unmount.  flush all journal trans, release all alloc'd ram */
1967 int journal_release(struct reiserfs_transaction_handle *th,
1968                     struct super_block *sb)
1969 {
1970         return do_journal_release(th, sb, 0);
1971 }
1972
1973 /* only call from an error condition inside reiserfs_read_super!  */
1974 int journal_release_error(struct reiserfs_transaction_handle *th,
1975                           struct super_block *sb)
1976 {
1977         return do_journal_release(th, sb, 1);
1978 }
1979
1980 /*
1981  * compares description block with commit block.
1982  * returns 1 if they differ, 0 if they are the same
1983  */
1984 static int journal_compare_desc_commit(struct super_block *sb,
1985                                        struct reiserfs_journal_desc *desc,
1986                                        struct reiserfs_journal_commit *commit)
1987 {
1988         if (get_commit_trans_id(commit) != get_desc_trans_id(desc) ||
1989             get_commit_trans_len(commit) != get_desc_trans_len(desc) ||
1990             get_commit_trans_len(commit) > SB_JOURNAL(sb)->j_trans_max ||
1991             get_commit_trans_len(commit) <= 0) {
1992                 return 1;
1993         }
1994         return 0;
1995 }
1996
1997 /*
1998  * returns 0 if it did not find a description block
1999  * returns -1 if it found a corrupt commit block
2000  * returns 1 if both desc and commit were valid
2001  * NOTE: only called during fs mount
2002  */
2003 static int journal_transaction_is_valid(struct super_block *sb,
2004                                         struct buffer_head *d_bh,
2005                                         unsigned int *oldest_invalid_trans_id,
2006                                         unsigned long *newest_mount_id)
2007 {
2008         struct reiserfs_journal_desc *desc;
2009         struct reiserfs_journal_commit *commit;
2010         struct buffer_head *c_bh;
2011         unsigned long offset;
2012
2013         if (!d_bh)
2014                 return 0;
2015
2016         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2017         if (get_desc_trans_len(desc) > 0
2018             && !memcmp(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8)) {
2019                 if (oldest_invalid_trans_id && *oldest_invalid_trans_id
2020                     && get_desc_trans_id(desc) > *oldest_invalid_trans_id) {
2021                         reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2022                                        "journal-986: transaction "
2023                                        "is valid returning because trans_id %d is greater than "
2024                                        "oldest_invalid %lu",
2025                                        get_desc_trans_id(desc),
2026                                        *oldest_invalid_trans_id);
2027                         return 0;
2028                 }
2029                 if (newest_mount_id
2030                     && *newest_mount_id > get_desc_mount_id(desc)) {
2031                         reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2032                                        "journal-1087: transaction "
2033                                        "is valid returning because mount_id %d is less than "
2034                                        "newest_mount_id %lu",
2035                                        get_desc_mount_id(desc),
2036                                        *newest_mount_id);
2037                         return -1;
2038                 }
2039                 if (get_desc_trans_len(desc) > SB_JOURNAL(sb)->j_trans_max) {
2040                         reiserfs_warning(sb, "journal-2018",
2041                                          "Bad transaction length %d "
2042                                          "encountered, ignoring transaction",
2043                                          get_desc_trans_len(desc));
2044                         return -1;
2045                 }
2046                 offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2047
2048                 /*
2049                  * ok, we have a journal description block,
2050                  * let's see if the transaction was valid
2051                  */
2052                 c_bh =
2053                     journal_bread(sb,
2054                                   SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2055                                   ((offset + get_desc_trans_len(desc) +
2056                                     1) % SB_ONDISK_JOURNAL_SIZE(sb)));
2057                 if (!c_bh)
2058                         return 0;
2059                 commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2060                 if (journal_compare_desc_commit(sb, desc, commit)) {
2061                         reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2062                                        "journal_transaction_is_valid, commit offset %ld had bad "
2063                                        "time %d or length %d",
2064                                        c_bh->b_blocknr -
2065                                        SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2066                                        get_commit_trans_id(commit),
2067                                        get_commit_trans_len(commit));
2068                         brelse(c_bh);
2069                         if (oldest_invalid_trans_id) {
2070                                 *oldest_invalid_trans_id =
2071                                     get_desc_trans_id(desc);
2072                                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2073                                                "journal-1004: "
2074                                                "transaction_is_valid setting oldest invalid trans_id "
2075                                                "to %d",
2076                                                get_desc_trans_id(desc));
2077                         }
2078                         return -1;
2079                 }
2080                 brelse(c_bh);
2081                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2082                                "journal-1006: found valid "
2083                                "transaction start offset %llu, len %d id %d",
2084                                d_bh->b_blocknr -
2085                                SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2086                                get_desc_trans_len(desc),
2087                                get_desc_trans_id(desc));
2088                 return 1;
2089         } else {
2090                 return 0;
2091         }
2092 }
2093
2094 static void brelse_array(struct buffer_head **heads, int num)
2095 {
2096         int i;
2097         for (i = 0; i < num; i++) {
2098                 brelse(heads[i]);
2099         }
2100 }
2101
2102 /*
2103  * given the start, and values for the oldest acceptable transactions,
2104  * this either reads in a replays a transaction, or returns because the
2105  * transaction is invalid, or too old.
2106  * NOTE: only called during fs mount
2107  */
2108 static int journal_read_transaction(struct super_block *sb,
2109                                     unsigned long cur_dblock,
2110                                     unsigned long oldest_start,
2111                                     unsigned int oldest_trans_id,
2112                                     unsigned long newest_mount_id)
2113 {
2114         struct reiserfs_journal *journal = SB_JOURNAL(sb);
2115         struct reiserfs_journal_desc *desc;
2116         struct reiserfs_journal_commit *commit;
2117         unsigned int trans_id = 0;
2118         struct buffer_head *c_bh;
2119         struct buffer_head *d_bh;
2120         struct buffer_head **log_blocks = NULL;
2121         struct buffer_head **real_blocks = NULL;
2122         unsigned int trans_offset;
2123         int i;
2124         int trans_half;
2125
2126         d_bh = journal_bread(sb, cur_dblock);
2127         if (!d_bh)
2128                 return 1;
2129         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2130         trans_offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2131         reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1037: "
2132                        "journal_read_transaction, offset %llu, len %d mount_id %d",
2133                        d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2134                        get_desc_trans_len(desc), get_desc_mount_id(desc));
2135         if (get_desc_trans_id(desc) < oldest_trans_id) {
2136                 reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1039: "
2137                                "journal_read_trans skipping because %lu is too old",
2138                                cur_dblock -
2139                                SB_ONDISK_JOURNAL_1st_BLOCK(sb));
2140                 brelse(d_bh);
2141                 return 1;
2142         }
2143         if (get_desc_mount_id(desc) != newest_mount_id) {
2144                 reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1146: "
2145                                "journal_read_trans skipping because %d is != "
2146                                "newest_mount_id %lu", get_desc_mount_id(desc),
2147                                newest_mount_id);
2148                 brelse(d_bh);
2149                 return 1;
2150         }
2151         c_bh = journal_bread(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2152                              ((trans_offset + get_desc_trans_len(desc) + 1) %
2153                               SB_ONDISK_JOURNAL_SIZE(sb)));
2154         if (!c_bh) {
2155                 brelse(d_bh);
2156                 return 1;
2157         }
2158         commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2159         if (journal_compare_desc_commit(sb, desc, commit)) {
2160                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2161                                "journal_read_transaction, "
2162                                "commit offset %llu had bad time %d or length %d",
2163                                c_bh->b_blocknr -
2164                                SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2165                                get_commit_trans_id(commit),
2166                                get_commit_trans_len(commit));
2167                 brelse(c_bh);
2168                 brelse(d_bh);
2169                 return 1;
2170         }
2171
2172         if (bdev_read_only(sb->s_bdev)) {
2173                 reiserfs_warning(sb, "clm-2076",
2174                                  "device is readonly, unable to replay log");
2175                 brelse(c_bh);
2176                 brelse(d_bh);
2177                 return -EROFS;
2178         }
2179
2180         trans_id = get_desc_trans_id(desc);
2181         /*
2182          * now we know we've got a good transaction, and it was
2183          * inside the valid time ranges
2184          */
2185         log_blocks = kmalloc_array(get_desc_trans_len(desc),
2186                                    sizeof(struct buffer_head *),
2187                                    GFP_NOFS);
2188         real_blocks = kmalloc_array(get_desc_trans_len(desc),
2189                                     sizeof(struct buffer_head *),
2190                                     GFP_NOFS);
2191         if (!log_blocks || !real_blocks) {
2192                 brelse(c_bh);
2193                 brelse(d_bh);
2194                 kfree(log_blocks);
2195                 kfree(real_blocks);
2196                 reiserfs_warning(sb, "journal-1169",
2197                                  "kmalloc failed, unable to mount FS");
2198                 return -1;
2199         }
2200         /* get all the buffer heads */
2201         trans_half = journal_trans_half(sb->s_blocksize);
2202         for (i = 0; i < get_desc_trans_len(desc); i++) {
2203                 log_blocks[i] =
2204                     journal_getblk(sb,
2205                                    SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2206                                    (trans_offset + 1 +
2207                                     i) % SB_ONDISK_JOURNAL_SIZE(sb));
2208                 if (i < trans_half) {
2209                         real_blocks[i] =
2210                             sb_getblk(sb,
2211                                       le32_to_cpu(desc->j_realblock[i]));
2212                 } else {
2213                         real_blocks[i] =
2214                             sb_getblk(sb,
2215                                       le32_to_cpu(commit->
2216                                                   j_realblock[i - trans_half]));
2217                 }
2218                 if (real_blocks[i]->b_blocknr > SB_BLOCK_COUNT(sb)) {
2219                         reiserfs_warning(sb, "journal-1207",
2220                                          "REPLAY FAILURE fsck required! "
2221                                          "Block to replay is outside of "
2222                                          "filesystem");
2223                         goto abort_replay;
2224                 }
2225                 /* make sure we don't try to replay onto log or reserved area */
2226                 if (is_block_in_log_or_reserved_area
2227                     (sb, real_blocks[i]->b_blocknr)) {
2228                         reiserfs_warning(sb, "journal-1204",
2229                                          "REPLAY FAILURE fsck required! "
2230                                          "Trying to replay onto a log block");
2231 abort_replay:
2232                         brelse_array(log_blocks, i);
2233                         brelse_array(real_blocks, i);
2234                         brelse(c_bh);
2235                         brelse(d_bh);
2236                         kfree(log_blocks);
2237                         kfree(real_blocks);
2238                         return -1;
2239                 }
2240         }
2241         /* read in the log blocks, memcpy to the corresponding real block */
2242         ll_rw_block(REQ_OP_READ, 0, get_desc_trans_len(desc), log_blocks);
2243         for (i = 0; i < get_desc_trans_len(desc); i++) {
2244
2245                 wait_on_buffer(log_blocks[i]);
2246                 if (!buffer_uptodate(log_blocks[i])) {
2247                         reiserfs_warning(sb, "journal-1212",
2248                                          "REPLAY FAILURE fsck required! "
2249                                          "buffer write failed");
2250                         brelse_array(log_blocks + i,
2251                                      get_desc_trans_len(desc) - i);
2252                         brelse_array(real_blocks, get_desc_trans_len(desc));
2253                         brelse(c_bh);
2254                         brelse(d_bh);
2255                         kfree(log_blocks);
2256                         kfree(real_blocks);
2257                         return -1;
2258                 }
2259                 memcpy(real_blocks[i]->b_data, log_blocks[i]->b_data,
2260                        real_blocks[i]->b_size);
2261                 set_buffer_uptodate(real_blocks[i]);
2262                 brelse(log_blocks[i]);
2263         }
2264         /* flush out the real blocks */
2265         for (i = 0; i < get_desc_trans_len(desc); i++) {
2266                 set_buffer_dirty(real_blocks[i]);
2267                 write_dirty_buffer(real_blocks[i], 0);
2268         }
2269         for (i = 0; i < get_desc_trans_len(desc); i++) {
2270                 wait_on_buffer(real_blocks[i]);
2271                 if (!buffer_uptodate(real_blocks[i])) {
2272                         reiserfs_warning(sb, "journal-1226",
2273                                          "REPLAY FAILURE, fsck required! "
2274                                          "buffer write failed");
2275                         brelse_array(real_blocks + i,
2276                                      get_desc_trans_len(desc) - i);
2277                         brelse(c_bh);
2278                         brelse(d_bh);
2279                         kfree(log_blocks);
2280                         kfree(real_blocks);
2281                         return -1;
2282                 }
2283                 brelse(real_blocks[i]);
2284         }
2285         cur_dblock =
2286             SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2287             ((trans_offset + get_desc_trans_len(desc) +
2288               2) % SB_ONDISK_JOURNAL_SIZE(sb));
2289         reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2290                        "journal-1095: setting journal " "start to offset %ld",
2291                        cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb));
2292
2293         /*
2294          * init starting values for the first transaction, in case
2295          * this is the last transaction to be replayed.
2296          */
2297         journal->j_start = cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2298         journal->j_last_flush_trans_id = trans_id;
2299         journal->j_trans_id = trans_id + 1;
2300         /* check for trans_id overflow */
2301         if (journal->j_trans_id == 0)
2302                 journal->j_trans_id = 10;
2303         brelse(c_bh);
2304         brelse(d_bh);
2305         kfree(log_blocks);
2306         kfree(real_blocks);
2307         return 0;
2308 }
2309
2310 /*
2311  * This function reads blocks starting from block and to max_block of bufsize
2312  * size (but no more than BUFNR blocks at a time). This proved to improve
2313  * mounting speed on self-rebuilding raid5 arrays at least.
2314  * Right now it is only used from journal code. But later we might use it
2315  * from other places.
2316  * Note: Do not use journal_getblk/sb_getblk functions here!
2317  */
2318 static struct buffer_head *reiserfs_breada(struct block_device *dev,
2319                                            b_blocknr_t block, int bufsize,
2320                                            b_blocknr_t max_block)
2321 {
2322         struct buffer_head *bhlist[BUFNR];
2323         unsigned int blocks = BUFNR;
2324         struct buffer_head *bh;
2325         int i, j;
2326
2327         bh = __getblk(dev, block, bufsize);
2328         if (buffer_uptodate(bh))
2329                 return (bh);
2330
2331         if (block + BUFNR > max_block) {
2332                 blocks = max_block - block;
2333         }
2334         bhlist[0] = bh;
2335         j = 1;
2336         for (i = 1; i < blocks; i++) {
2337                 bh = __getblk(dev, block + i, bufsize);
2338                 if (buffer_uptodate(bh)) {
2339                         brelse(bh);
2340                         break;
2341                 } else
2342                         bhlist[j++] = bh;
2343         }
2344         ll_rw_block(REQ_OP_READ, 0, j, bhlist);
2345         for (i = 1; i < j; i++)
2346                 brelse(bhlist[i]);
2347         bh = bhlist[0];
2348         wait_on_buffer(bh);
2349         if (buffer_uptodate(bh))
2350                 return bh;
2351         brelse(bh);
2352         return NULL;
2353 }
2354
2355 /*
2356  * read and replay the log
2357  * on a clean unmount, the journal header's next unflushed pointer will be
2358  * to an invalid transaction.  This tests that before finding all the
2359  * transactions in the log, which makes normal mount times fast.
2360  *
2361  * After a crash, this starts with the next unflushed transaction, and
2362  * replays until it finds one too old, or invalid.
2363  *
2364  * On exit, it sets things up so the first transaction will work correctly.
2365  * NOTE: only called during fs mount
2366  */
2367 static int journal_read(struct super_block *sb)
2368 {
2369         struct reiserfs_journal *journal = SB_JOURNAL(sb);
2370         struct reiserfs_journal_desc *desc;
2371         unsigned int oldest_trans_id = 0;
2372         unsigned int oldest_invalid_trans_id = 0;
2373         time64_t start;
2374         unsigned long oldest_start = 0;
2375         unsigned long cur_dblock = 0;
2376         unsigned long newest_mount_id = 9;
2377         struct buffer_head *d_bh;
2378         struct reiserfs_journal_header *jh;
2379         int valid_journal_header = 0;
2380         int replay_count = 0;
2381         int continue_replay = 1;
2382         int ret;
2383
2384         cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2385         reiserfs_info(sb, "checking transaction log (%pg)\n",
2386                       journal->j_dev_bd);
2387         start = ktime_get_seconds();
2388
2389         /*
2390          * step 1, read in the journal header block.  Check the transaction
2391          * it says is the first unflushed, and if that transaction is not
2392          * valid, replay is done
2393          */
2394         journal->j_header_bh = journal_bread(sb,
2395                                              SB_ONDISK_JOURNAL_1st_BLOCK(sb)
2396                                              + SB_ONDISK_JOURNAL_SIZE(sb));
2397         if (!journal->j_header_bh) {
2398                 return 1;
2399         }
2400         jh = (struct reiserfs_journal_header *)(journal->j_header_bh->b_data);
2401         if (le32_to_cpu(jh->j_first_unflushed_offset) <
2402             SB_ONDISK_JOURNAL_SIZE(sb)
2403             && le32_to_cpu(jh->j_last_flush_trans_id) > 0) {
2404                 oldest_start =
2405                     SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2406                     le32_to_cpu(jh->j_first_unflushed_offset);
2407                 oldest_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2408                 newest_mount_id = le32_to_cpu(jh->j_mount_id);
2409                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2410                                "journal-1153: found in "
2411                                "header: first_unflushed_offset %d, last_flushed_trans_id "
2412                                "%lu", le32_to_cpu(jh->j_first_unflushed_offset),
2413                                le32_to_cpu(jh->j_last_flush_trans_id));
2414                 valid_journal_header = 1;
2415
2416                 /*
2417                  * now, we try to read the first unflushed offset.  If it
2418                  * is not valid, there is nothing more we can do, and it
2419                  * makes no sense to read through the whole log.
2420                  */
2421                 d_bh =
2422                     journal_bread(sb,
2423                                   SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2424                                   le32_to_cpu(jh->j_first_unflushed_offset));
2425                 ret = journal_transaction_is_valid(sb, d_bh, NULL, NULL);
2426                 if (!ret) {
2427                         continue_replay = 0;
2428                 }
2429                 brelse(d_bh);
2430                 goto start_log_replay;
2431         }
2432
2433         /*
2434          * ok, there are transactions that need to be replayed.  start
2435          * with the first log block, find all the valid transactions, and
2436          * pick out the oldest.
2437          */
2438         while (continue_replay
2439                && cur_dblock <
2440                (SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2441                 SB_ONDISK_JOURNAL_SIZE(sb))) {
2442                 /*
2443                  * Note that it is required for blocksize of primary fs
2444                  * device and journal device to be the same
2445                  */
2446                 d_bh =
2447                     reiserfs_breada(journal->j_dev_bd, cur_dblock,
2448                                     sb->s_blocksize,
2449                                     SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2450                                     SB_ONDISK_JOURNAL_SIZE(sb));
2451                 ret =
2452                     journal_transaction_is_valid(sb, d_bh,
2453                                                  &oldest_invalid_trans_id,
2454                                                  &newest_mount_id);
2455                 if (ret == 1) {
2456                         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2457                         if (oldest_start == 0) {        /* init all oldest_ values */
2458                                 oldest_trans_id = get_desc_trans_id(desc);
2459                                 oldest_start = d_bh->b_blocknr;
2460                                 newest_mount_id = get_desc_mount_id(desc);
2461                                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2462                                                "journal-1179: Setting "
2463                                                "oldest_start to offset %llu, trans_id %lu",
2464                                                oldest_start -
2465                                                SB_ONDISK_JOURNAL_1st_BLOCK
2466                                                (sb), oldest_trans_id);
2467                         } else if (oldest_trans_id > get_desc_trans_id(desc)) {
2468                                 /* one we just read was older */
2469                                 oldest_trans_id = get_desc_trans_id(desc);
2470                                 oldest_start = d_bh->b_blocknr;
2471                                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2472                                                "journal-1180: Resetting "
2473                                                "oldest_start to offset %lu, trans_id %lu",
2474                                                oldest_start -
2475                                                SB_ONDISK_JOURNAL_1st_BLOCK
2476                                                (sb), oldest_trans_id);
2477                         }
2478                         if (newest_mount_id < get_desc_mount_id(desc)) {
2479                                 newest_mount_id = get_desc_mount_id(desc);
2480                                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2481                                                "journal-1299: Setting "
2482                                                "newest_mount_id to %d",
2483                                                get_desc_mount_id(desc));
2484                         }
2485                         cur_dblock += get_desc_trans_len(desc) + 2;
2486                 } else {
2487                         cur_dblock++;
2488                 }
2489                 brelse(d_bh);
2490         }
2491
2492 start_log_replay:
2493         cur_dblock = oldest_start;
2494         if (oldest_trans_id) {
2495                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2496                                "journal-1206: Starting replay "
2497                                "from offset %llu, trans_id %lu",
2498                                cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2499                                oldest_trans_id);
2500
2501         }
2502         replay_count = 0;
2503         while (continue_replay && oldest_trans_id > 0) {
2504                 ret =
2505                     journal_read_transaction(sb, cur_dblock, oldest_start,
2506                                              oldest_trans_id, newest_mount_id);
2507                 if (ret < 0) {
2508                         return ret;
2509                 } else if (ret != 0) {
2510                         break;
2511                 }
2512                 cur_dblock =
2513                     SB_ONDISK_JOURNAL_1st_BLOCK(sb) + journal->j_start;
2514                 replay_count++;
2515                 if (cur_dblock == oldest_start)
2516                         break;
2517         }
2518
2519         if (oldest_trans_id == 0) {
2520                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2521                                "journal-1225: No valid " "transactions found");
2522         }
2523         /*
2524          * j_start does not get set correctly if we don't replay any
2525          * transactions.  if we had a valid journal_header, set j_start
2526          * to the first unflushed transaction value, copy the trans_id
2527          * from the header
2528          */
2529         if (valid_journal_header && replay_count == 0) {
2530                 journal->j_start = le32_to_cpu(jh->j_first_unflushed_offset);
2531                 journal->j_trans_id =
2532                     le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2533                 /* check for trans_id overflow */
2534                 if (journal->j_trans_id == 0)
2535                         journal->j_trans_id = 10;
2536                 journal->j_last_flush_trans_id =
2537                     le32_to_cpu(jh->j_last_flush_trans_id);
2538                 journal->j_mount_id = le32_to_cpu(jh->j_mount_id) + 1;
2539         } else {
2540                 journal->j_mount_id = newest_mount_id + 1;
2541         }
2542         reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1299: Setting "
2543                        "newest_mount_id to %lu", journal->j_mount_id);
2544         journal->j_first_unflushed_offset = journal->j_start;
2545         if (replay_count > 0) {
2546                 reiserfs_info(sb,
2547                               "replayed %d transactions in %lu seconds\n",
2548                               replay_count, ktime_get_seconds() - start);
2549         }
2550         /* needed to satisfy the locking in _update_journal_header_block */
2551         reiserfs_write_lock(sb);
2552         if (!bdev_read_only(sb->s_bdev) &&
2553             _update_journal_header_block(sb, journal->j_start,
2554                                          journal->j_last_flush_trans_id)) {
2555                 reiserfs_write_unlock(sb);
2556                 /*
2557                  * replay failed, caller must call free_journal_ram and abort
2558                  * the mount
2559                  */
2560                 return -1;
2561         }
2562         reiserfs_write_unlock(sb);
2563         return 0;
2564 }
2565
2566 static struct reiserfs_journal_list *alloc_journal_list(struct super_block *s)
2567 {
2568         struct reiserfs_journal_list *jl;
2569         jl = kzalloc(sizeof(struct reiserfs_journal_list),
2570                      GFP_NOFS | __GFP_NOFAIL);
2571         INIT_LIST_HEAD(&jl->j_list);
2572         INIT_LIST_HEAD(&jl->j_working_list);
2573         INIT_LIST_HEAD(&jl->j_tail_bh_list);
2574         INIT_LIST_HEAD(&jl->j_bh_list);
2575         mutex_init(&jl->j_commit_mutex);
2576         SB_JOURNAL(s)->j_num_lists++;
2577         get_journal_list(jl);
2578         return jl;
2579 }
2580
2581 static void journal_list_init(struct super_block *sb)
2582 {
2583         SB_JOURNAL(sb)->j_current_jl = alloc_journal_list(sb);
2584 }
2585
2586 static void release_journal_dev(struct super_block *super,
2587                                struct reiserfs_journal *journal)
2588 {
2589         if (journal->j_dev_bd != NULL) {
2590                 blkdev_put(journal->j_dev_bd, journal->j_dev_mode);
2591                 journal->j_dev_bd = NULL;
2592         }
2593 }
2594
2595 static int journal_init_dev(struct super_block *super,
2596                             struct reiserfs_journal *journal,
2597                             const char *jdev_name)
2598 {
2599         int result;
2600         dev_t jdev;
2601         fmode_t blkdev_mode = FMODE_READ | FMODE_WRITE | FMODE_EXCL;
2602         char b[BDEVNAME_SIZE];
2603
2604         result = 0;
2605
2606         journal->j_dev_bd = NULL;
2607         jdev = SB_ONDISK_JOURNAL_DEVICE(super) ?
2608             new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super)) : super->s_dev;
2609
2610         if (bdev_read_only(super->s_bdev))
2611                 blkdev_mode = FMODE_READ;
2612
2613         /* there is no "jdev" option and journal is on separate device */
2614         if ((!jdev_name || !jdev_name[0])) {
2615                 if (jdev == super->s_dev)
2616                         blkdev_mode &= ~FMODE_EXCL;
2617                 journal->j_dev_bd = blkdev_get_by_dev(jdev, blkdev_mode,
2618                                                       journal);
2619                 journal->j_dev_mode = blkdev_mode;
2620                 if (IS_ERR(journal->j_dev_bd)) {
2621                         result = PTR_ERR(journal->j_dev_bd);
2622                         journal->j_dev_bd = NULL;
2623                         reiserfs_warning(super, "sh-458",
2624                                          "cannot init journal device '%s': %i",
2625                                          __bdevname(jdev, b), result);
2626                         return result;
2627                 } else if (jdev != super->s_dev)
2628                         set_blocksize(journal->j_dev_bd, super->s_blocksize);
2629
2630                 return 0;
2631         }
2632
2633         journal->j_dev_mode = blkdev_mode;
2634         journal->j_dev_bd = blkdev_get_by_path(jdev_name, blkdev_mode, journal);
2635         if (IS_ERR(journal->j_dev_bd)) {
2636                 result = PTR_ERR(journal->j_dev_bd);
2637                 journal->j_dev_bd = NULL;
2638                 reiserfs_warning(super, "sh-457",
2639                                  "journal_init_dev: Cannot open '%s': %i",
2640                                  jdev_name, result);
2641                 return result;
2642         }
2643
2644         set_blocksize(journal->j_dev_bd, super->s_blocksize);
2645         reiserfs_info(super,
2646                       "journal_init_dev: journal device: %pg\n",
2647                       journal->j_dev_bd);
2648         return 0;
2649 }
2650
2651 /*
2652  * When creating/tuning a file system user can assign some
2653  * journal params within boundaries which depend on the ratio
2654  * blocksize/standard_blocksize.
2655  *
2656  * For blocks >= standard_blocksize transaction size should
2657  * be not less then JOURNAL_TRANS_MIN_DEFAULT, and not more
2658  * then JOURNAL_TRANS_MAX_DEFAULT.
2659  *
2660  * For blocks < standard_blocksize these boundaries should be
2661  * decreased proportionally.
2662  */
2663 #define REISERFS_STANDARD_BLKSIZE (4096)
2664
2665 static int check_advise_trans_params(struct super_block *sb,
2666                                      struct reiserfs_journal *journal)
2667 {
2668         if (journal->j_trans_max) {
2669                 /* Non-default journal params.  Do sanity check for them. */
2670                 int ratio = 1;
2671                 if (sb->s_blocksize < REISERFS_STANDARD_BLKSIZE)
2672                         ratio = REISERFS_STANDARD_BLKSIZE / sb->s_blocksize;
2673
2674                 if (journal->j_trans_max > JOURNAL_TRANS_MAX_DEFAULT / ratio ||
2675                     journal->j_trans_max < JOURNAL_TRANS_MIN_DEFAULT / ratio ||
2676                     SB_ONDISK_JOURNAL_SIZE(sb) / journal->j_trans_max <
2677                     JOURNAL_MIN_RATIO) {
2678                         reiserfs_warning(sb, "sh-462",
2679                                          "bad transaction max size (%u). "
2680                                          "FSCK?", journal->j_trans_max);
2681                         return 1;
2682                 }
2683                 if (journal->j_max_batch != (journal->j_trans_max) *
2684                         JOURNAL_MAX_BATCH_DEFAULT/JOURNAL_TRANS_MAX_DEFAULT) {
2685                         reiserfs_warning(sb, "sh-463",
2686                                          "bad transaction max batch (%u). "
2687                                          "FSCK?", journal->j_max_batch);
2688                         return 1;
2689                 }
2690         } else {
2691                 /*
2692                  * Default journal params.
2693                  * The file system was created by old version
2694                  * of mkreiserfs, so some fields contain zeros,
2695                  * and we need to advise proper values for them
2696                  */
2697                 if (sb->s_blocksize != REISERFS_STANDARD_BLKSIZE) {
2698                         reiserfs_warning(sb, "sh-464", "bad blocksize (%u)",
2699                                          sb->s_blocksize);
2700                         return 1;
2701                 }
2702                 journal->j_trans_max = JOURNAL_TRANS_MAX_DEFAULT;
2703                 journal->j_max_batch = JOURNAL_MAX_BATCH_DEFAULT;
2704                 journal->j_max_commit_age = JOURNAL_MAX_COMMIT_AGE;
2705         }
2706         return 0;
2707 }
2708
2709 /* must be called once on fs mount.  calls journal_read for you */
2710 int journal_init(struct super_block *sb, const char *j_dev_name,
2711                  int old_format, unsigned int commit_max_age)
2712 {
2713         int num_cnodes = SB_ONDISK_JOURNAL_SIZE(sb) * 2;
2714         struct buffer_head *bhjh;
2715         struct reiserfs_super_block *rs;
2716         struct reiserfs_journal_header *jh;
2717         struct reiserfs_journal *journal;
2718         struct reiserfs_journal_list *jl;
2719         int ret;
2720
2721         journal = SB_JOURNAL(sb) = vzalloc(sizeof(struct reiserfs_journal));
2722         if (!journal) {
2723                 reiserfs_warning(sb, "journal-1256",
2724                                  "unable to get memory for journal structure");
2725                 return 1;
2726         }
2727         INIT_LIST_HEAD(&journal->j_bitmap_nodes);
2728         INIT_LIST_HEAD(&journal->j_prealloc_list);
2729         INIT_LIST_HEAD(&journal->j_working_list);
2730         INIT_LIST_HEAD(&journal->j_journal_list);
2731         journal->j_persistent_trans = 0;
2732         if (reiserfs_allocate_list_bitmaps(sb, journal->j_list_bitmap,
2733                                            reiserfs_bmap_count(sb)))
2734                 goto free_and_return;
2735
2736         allocate_bitmap_nodes(sb);
2737
2738         /* reserved for journal area support */
2739         SB_JOURNAL_1st_RESERVED_BLOCK(sb) = (old_format ?
2740                                                  REISERFS_OLD_DISK_OFFSET_IN_BYTES
2741                                                  / sb->s_blocksize +
2742                                                  reiserfs_bmap_count(sb) +
2743                                                  1 :
2744                                                  REISERFS_DISK_OFFSET_IN_BYTES /
2745                                                  sb->s_blocksize + 2);
2746
2747         /*
2748          * Sanity check to see is the standard journal fitting
2749          * within first bitmap (actual for small blocksizes)
2750          */
2751         if (!SB_ONDISK_JOURNAL_DEVICE(sb) &&
2752             (SB_JOURNAL_1st_RESERVED_BLOCK(sb) +
2753              SB_ONDISK_JOURNAL_SIZE(sb) > sb->s_blocksize * 8)) {
2754                 reiserfs_warning(sb, "journal-1393",
2755                                  "journal does not fit for area addressed "
2756                                  "by first of bitmap blocks. It starts at "
2757                                  "%u and its size is %u. Block size %ld",
2758                                  SB_JOURNAL_1st_RESERVED_BLOCK(sb),
2759                                  SB_ONDISK_JOURNAL_SIZE(sb),
2760                                  sb->s_blocksize);
2761                 goto free_and_return;
2762         }
2763
2764         if (journal_init_dev(sb, journal, j_dev_name) != 0) {
2765                 reiserfs_warning(sb, "sh-462",
2766                                  "unable to initialize journal device");
2767                 goto free_and_return;
2768         }
2769
2770         rs = SB_DISK_SUPER_BLOCK(sb);
2771
2772         /* read journal header */
2773         bhjh = journal_bread(sb,
2774                              SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2775                              SB_ONDISK_JOURNAL_SIZE(sb));
2776         if (!bhjh) {
2777                 reiserfs_warning(sb, "sh-459",
2778                                  "unable to read journal header");
2779                 goto free_and_return;
2780         }
2781         jh = (struct reiserfs_journal_header *)(bhjh->b_data);
2782
2783         /* make sure that journal matches to the super block */
2784         if (is_reiserfs_jr(rs)
2785             && (le32_to_cpu(jh->jh_journal.jp_journal_magic) !=
2786                 sb_jp_journal_magic(rs))) {
2787                 reiserfs_warning(sb, "sh-460",
2788                                  "journal header magic %x (device %pg) does "
2789                                  "not match to magic found in super block %x",
2790                                  jh->jh_journal.jp_journal_magic,
2791                                  journal->j_dev_bd,
2792                                  sb_jp_journal_magic(rs));
2793                 brelse(bhjh);
2794                 goto free_and_return;
2795         }
2796
2797         journal->j_trans_max = le32_to_cpu(jh->jh_journal.jp_journal_trans_max);
2798         journal->j_max_batch = le32_to_cpu(jh->jh_journal.jp_journal_max_batch);
2799         journal->j_max_commit_age =
2800             le32_to_cpu(jh->jh_journal.jp_journal_max_commit_age);
2801         journal->j_max_trans_age = JOURNAL_MAX_TRANS_AGE;
2802
2803         if (check_advise_trans_params(sb, journal) != 0)
2804                 goto free_and_return;
2805         journal->j_default_max_commit_age = journal->j_max_commit_age;
2806
2807         if (commit_max_age != 0) {
2808                 journal->j_max_commit_age = commit_max_age;
2809                 journal->j_max_trans_age = commit_max_age;
2810         }
2811
2812         reiserfs_info(sb, "journal params: device %pg, size %u, "
2813                       "journal first block %u, max trans len %u, max batch %u, "
2814                       "max commit age %u, max trans age %u\n",
2815                       journal->j_dev_bd,
2816                       SB_ONDISK_JOURNAL_SIZE(sb),
2817                       SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2818                       journal->j_trans_max,
2819                       journal->j_max_batch,
2820                       journal->j_max_commit_age, journal->j_max_trans_age);
2821
2822         brelse(bhjh);
2823
2824         journal->j_list_bitmap_index = 0;
2825         journal_list_init(sb);
2826
2827         memset(journal->j_list_hash_table, 0,
2828                JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
2829
2830         INIT_LIST_HEAD(&journal->j_dirty_buffers);
2831         spin_lock_init(&journal->j_dirty_buffers_lock);
2832
2833         journal->j_start = 0;
2834         journal->j_len = 0;
2835         journal->j_len_alloc = 0;
2836         atomic_set(&journal->j_wcount, 0);
2837         atomic_set(&journal->j_async_throttle, 0);
2838         journal->j_bcount = 0;
2839         journal->j_trans_start_time = 0;
2840         journal->j_last = NULL;
2841         journal->j_first = NULL;
2842         init_waitqueue_head(&journal->j_join_wait);
2843         mutex_init(&journal->j_mutex);
2844         mutex_init(&journal->j_flush_mutex);
2845
2846         journal->j_trans_id = 10;
2847         journal->j_mount_id = 10;
2848         journal->j_state = 0;
2849         atomic_set(&journal->j_jlock, 0);
2850         journal->j_cnode_free_list = allocate_cnodes(num_cnodes);
2851         journal->j_cnode_free_orig = journal->j_cnode_free_list;
2852         journal->j_cnode_free = journal->j_cnode_free_list ? num_cnodes : 0;
2853         journal->j_cnode_used = 0;
2854         journal->j_must_wait = 0;
2855
2856         if (journal->j_cnode_free == 0) {
2857                 reiserfs_warning(sb, "journal-2004", "Journal cnode memory "
2858                                  "allocation failed (%ld bytes). Journal is "
2859                                  "too large for available memory. Usually "
2860                                  "this is due to a journal that is too large.",
2861                                  sizeof (struct reiserfs_journal_cnode) * num_cnodes);
2862                 goto free_and_return;
2863         }
2864
2865         init_journal_hash(sb);
2866         jl = journal->j_current_jl;
2867
2868         /*
2869          * get_list_bitmap() may call flush_commit_list() which
2870          * requires the lock. Calling flush_commit_list() shouldn't happen
2871          * this early but I like to be paranoid.
2872          */
2873         reiserfs_write_lock(sb);
2874         jl->j_list_bitmap = get_list_bitmap(sb, jl);
2875         reiserfs_write_unlock(sb);
2876         if (!jl->j_list_bitmap) {
2877                 reiserfs_warning(sb, "journal-2005",
2878                                  "get_list_bitmap failed for journal list 0");
2879                 goto free_and_return;
2880         }
2881
2882         ret = journal_read(sb);
2883         if (ret < 0) {
2884                 reiserfs_warning(sb, "reiserfs-2006",
2885                                  "Replay Failure, unable to mount");
2886                 goto free_and_return;
2887         }
2888
2889         INIT_DELAYED_WORK(&journal->j_work, flush_async_commits);
2890         journal->j_work_sb = sb;
2891         return 0;
2892 free_and_return:
2893         free_journal_ram(sb);
2894         return 1;
2895 }
2896
2897 /*
2898  * test for a polite end of the current transaction.  Used by file_write,
2899  * and should be used by delete to make sure they don't write more than
2900  * can fit inside a single transaction
2901  */
2902 int journal_transaction_should_end(struct reiserfs_transaction_handle *th,
2903                                    int new_alloc)
2904 {
2905         struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2906         time64_t now = ktime_get_seconds();
2907         /* cannot restart while nested */
2908         BUG_ON(!th->t_trans_id);
2909         if (th->t_refcount > 1)
2910                 return 0;
2911         if (journal->j_must_wait > 0 ||
2912             (journal->j_len_alloc + new_alloc) >= journal->j_max_batch ||
2913             atomic_read(&journal->j_jlock) ||
2914             (now - journal->j_trans_start_time) > journal->j_max_trans_age ||
2915             journal->j_cnode_free < (journal->j_trans_max * 3)) {
2916                 return 1;
2917         }
2918
2919         journal->j_len_alloc += new_alloc;
2920         th->t_blocks_allocated += new_alloc ;
2921         return 0;
2922 }
2923
2924 /* this must be called inside a transaction */
2925 void reiserfs_block_writes(struct reiserfs_transaction_handle *th)
2926 {
2927         struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2928         BUG_ON(!th->t_trans_id);
2929         journal->j_must_wait = 1;
2930         set_bit(J_WRITERS_BLOCKED, &journal->j_state);
2931         return;
2932 }
2933
2934 /* this must be called without a transaction started */
2935 void reiserfs_allow_writes(struct super_block *s)
2936 {
2937         struct reiserfs_journal *journal = SB_JOURNAL(s);
2938         clear_bit(J_WRITERS_BLOCKED, &journal->j_state);
2939         wake_up(&journal->j_join_wait);
2940 }
2941
2942 /* this must be called without a transaction started */
2943 void reiserfs_wait_on_write_block(struct super_block *s)
2944 {
2945         struct reiserfs_journal *journal = SB_JOURNAL(s);
2946         wait_event(journal->j_join_wait,
2947                    !test_bit(J_WRITERS_BLOCKED, &journal->j_state));
2948 }
2949
2950 static void queue_log_writer(struct super_block *s)
2951 {
2952         wait_queue_entry_t wait;
2953         struct reiserfs_journal *journal = SB_JOURNAL(s);
2954         set_bit(J_WRITERS_QUEUED, &journal->j_state);
2955
2956         /*
2957          * we don't want to use wait_event here because
2958          * we only want to wait once.
2959          */
2960         init_waitqueue_entry(&wait, current);
2961         add_wait_queue(&journal->j_join_wait, &wait);
2962         set_current_state(TASK_UNINTERRUPTIBLE);
2963         if (test_bit(J_WRITERS_QUEUED, &journal->j_state)) {
2964                 int depth = reiserfs_write_unlock_nested(s);
2965                 schedule();
2966                 reiserfs_write_lock_nested(s, depth);
2967         }
2968         __set_current_state(TASK_RUNNING);
2969         remove_wait_queue(&journal->j_join_wait, &wait);
2970 }
2971
2972 static void wake_queued_writers(struct super_block *s)
2973 {
2974         struct reiserfs_journal *journal = SB_JOURNAL(s);
2975         if (test_and_clear_bit(J_WRITERS_QUEUED, &journal->j_state))
2976                 wake_up(&journal->j_join_wait);
2977 }
2978
2979 static void let_transaction_grow(struct super_block *sb, unsigned int trans_id)
2980 {
2981         struct reiserfs_journal *journal = SB_JOURNAL(sb);
2982         unsigned long bcount = journal->j_bcount;
2983         while (1) {
2984                 int depth;
2985
2986                 depth = reiserfs_write_unlock_nested(sb);
2987                 schedule_timeout_uninterruptible(1);
2988                 reiserfs_write_lock_nested(sb, depth);
2989
2990                 journal->j_current_jl->j_state |= LIST_COMMIT_PENDING;
2991                 while ((atomic_read(&journal->j_wcount) > 0 ||
2992                         atomic_read(&journal->j_jlock)) &&
2993                        journal->j_trans_id == trans_id) {
2994                         queue_log_writer(sb);
2995                 }
2996                 if (journal->j_trans_id != trans_id)
2997                         break;
2998                 if (bcount == journal->j_bcount)
2999                         break;
3000                 bcount = journal->j_bcount;
3001         }
3002 }
3003
3004 /*
3005  * join == true if you must join an existing transaction.
3006  * join == false if you can deal with waiting for others to finish
3007  *
3008  * this will block until the transaction is joinable.  send the number of
3009  * blocks you expect to use in nblocks.
3010 */
3011 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
3012                               struct super_block *sb, unsigned long nblocks,
3013                               int join)
3014 {
3015         time64_t now = ktime_get_seconds();
3016         unsigned int old_trans_id;
3017         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3018         struct reiserfs_transaction_handle myth;
3019         int sched_count = 0;
3020         int retval;
3021         int depth;
3022
3023         reiserfs_check_lock_depth(sb, "journal_begin");
3024         BUG_ON(nblocks > journal->j_trans_max);
3025
3026         PROC_INFO_INC(sb, journal.journal_being);
3027         /* set here for journal_join */
3028         th->t_refcount = 1;
3029         th->t_super = sb;
3030
3031 relock:
3032         lock_journal(sb);
3033         if (join != JBEGIN_ABORT && reiserfs_is_journal_aborted(journal)) {
3034                 unlock_journal(sb);
3035                 retval = journal->j_errno;
3036                 goto out_fail;
3037         }
3038         journal->j_bcount++;
3039
3040         if (test_bit(J_WRITERS_BLOCKED, &journal->j_state)) {
3041                 unlock_journal(sb);
3042                 depth = reiserfs_write_unlock_nested(sb);
3043                 reiserfs_wait_on_write_block(sb);
3044                 reiserfs_write_lock_nested(sb, depth);
3045                 PROC_INFO_INC(sb, journal.journal_relock_writers);
3046                 goto relock;
3047         }
3048         now = ktime_get_seconds();
3049
3050         /*
3051          * if there is no room in the journal OR
3052          * if this transaction is too old, and we weren't called joinable,
3053          * wait for it to finish before beginning we don't sleep if there
3054          * aren't other writers
3055          */
3056
3057         if ((!join && journal->j_must_wait > 0) ||
3058             (!join
3059              && (journal->j_len_alloc + nblocks + 2) >= journal->j_max_batch)
3060             || (!join && atomic_read(&journal->j_wcount) > 0
3061                 && journal->j_trans_start_time > 0
3062                 && (now - journal->j_trans_start_time) >
3063                 journal->j_max_trans_age) || (!join
3064                                               && atomic_read(&journal->j_jlock))
3065             || (!join && journal->j_cnode_free < (journal->j_trans_max * 3))) {
3066
3067                 old_trans_id = journal->j_trans_id;
3068                 /* allow others to finish this transaction */
3069                 unlock_journal(sb);
3070
3071                 if (!join && (journal->j_len_alloc + nblocks + 2) >=
3072                     journal->j_max_batch &&
3073                     ((journal->j_len + nblocks + 2) * 100) <
3074                     (journal->j_len_alloc * 75)) {
3075                         if (atomic_read(&journal->j_wcount) > 10) {
3076                                 sched_count++;
3077                                 queue_log_writer(sb);
3078                                 goto relock;
3079                         }
3080                 }
3081                 /*
3082                  * don't mess with joining the transaction if all we
3083                  * have to do is wait for someone else to do a commit
3084                  */
3085                 if (atomic_read(&journal->j_jlock)) {
3086                         while (journal->j_trans_id == old_trans_id &&
3087                                atomic_read(&journal->j_jlock)) {
3088                                 queue_log_writer(sb);
3089                         }
3090                         goto relock;
3091                 }
3092                 retval = journal_join(&myth, sb);
3093                 if (retval)
3094                         goto out_fail;
3095
3096                 /* someone might have ended the transaction while we joined */
3097                 if (old_trans_id != journal->j_trans_id) {
3098                         retval = do_journal_end(&myth, 0);
3099                 } else {
3100                         retval = do_journal_end(&myth, COMMIT_NOW);
3101                 }
3102
3103                 if (retval)
3104                         goto out_fail;
3105
3106                 PROC_INFO_INC(sb, journal.journal_relock_wcount);
3107                 goto relock;
3108         }
3109         /* we are the first writer, set trans_id */
3110         if (journal->j_trans_start_time == 0) {
3111                 journal->j_trans_start_time = ktime_get_seconds();
3112         }
3113         atomic_inc(&journal->j_wcount);
3114         journal->j_len_alloc += nblocks;
3115         th->t_blocks_logged = 0;
3116         th->t_blocks_allocated = nblocks;
3117         th->t_trans_id = journal->j_trans_id;
3118         unlock_journal(sb);
3119         INIT_LIST_HEAD(&th->t_list);
3120         return 0;
3121
3122 out_fail:
3123         memset(th, 0, sizeof(*th));
3124         /*
3125          * Re-set th->t_super, so we can properly keep track of how many
3126          * persistent transactions there are. We need to do this so if this
3127          * call is part of a failed restart_transaction, we can free it later
3128          */
3129         th->t_super = sb;
3130         return retval;
3131 }
3132
3133 struct reiserfs_transaction_handle *reiserfs_persistent_transaction(struct
3134                                                                     super_block
3135                                                                     *s,
3136                                                                     int nblocks)
3137 {
3138         int ret;
3139         struct reiserfs_transaction_handle *th;
3140
3141         /*
3142          * if we're nesting into an existing transaction.  It will be
3143          * persistent on its own
3144          */
3145         if (reiserfs_transaction_running(s)) {
3146                 th = current->journal_info;
3147                 th->t_refcount++;
3148                 BUG_ON(th->t_refcount < 2);
3149
3150                 return th;
3151         }
3152         th = kmalloc(sizeof(struct reiserfs_transaction_handle), GFP_NOFS);
3153         if (!th)
3154                 return NULL;
3155         ret = journal_begin(th, s, nblocks);
3156         if (ret) {
3157                 kfree(th);
3158                 return NULL;
3159         }
3160
3161         SB_JOURNAL(s)->j_persistent_trans++;
3162         return th;
3163 }
3164
3165 int reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle *th)
3166 {
3167         struct super_block *s = th->t_super;
3168         int ret = 0;
3169         if (th->t_trans_id)
3170                 ret = journal_end(th);
3171         else
3172                 ret = -EIO;
3173         if (th->t_refcount == 0) {
3174                 SB_JOURNAL(s)->j_persistent_trans--;
3175                 kfree(th);
3176         }
3177         return ret;
3178 }
3179
3180 static int journal_join(struct reiserfs_transaction_handle *th,
3181                         struct super_block *sb)
3182 {
3183         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3184
3185         /*
3186          * this keeps do_journal_end from NULLing out the
3187          * current->journal_info pointer
3188          */
3189         th->t_handle_save = cur_th;
3190         BUG_ON(cur_th && cur_th->t_refcount > 1);
3191         return do_journal_begin_r(th, sb, 1, JBEGIN_JOIN);
3192 }
3193
3194 int journal_join_abort(struct reiserfs_transaction_handle *th,
3195                        struct super_block *sb)
3196 {
3197         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3198
3199         /*
3200          * this keeps do_journal_end from NULLing out the
3201          * current->journal_info pointer
3202          */
3203         th->t_handle_save = cur_th;
3204         BUG_ON(cur_th && cur_th->t_refcount > 1);
3205         return do_journal_begin_r(th, sb, 1, JBEGIN_ABORT);
3206 }
3207
3208 int journal_begin(struct reiserfs_transaction_handle *th,
3209                   struct super_block *sb, unsigned long nblocks)
3210 {
3211         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3212         int ret;
3213
3214         th->t_handle_save = NULL;
3215         if (cur_th) {
3216                 /* we are nesting into the current transaction */
3217                 if (cur_th->t_super == sb) {
3218                         BUG_ON(!cur_th->t_refcount);
3219                         cur_th->t_refcount++;
3220                         memcpy(th, cur_th, sizeof(*th));
3221                         if (th->t_refcount <= 1)
3222                                 reiserfs_warning(sb, "reiserfs-2005",
3223                                                  "BAD: refcount <= 1, but "
3224                                                  "journal_info != 0");
3225                         return 0;
3226                 } else {
3227                         /*
3228                          * we've ended up with a handle from a different
3229                          * filesystem.  save it and restore on journal_end.
3230                          * This should never really happen...
3231                          */
3232                         reiserfs_warning(sb, "clm-2100",
3233                                          "nesting info a different FS");
3234                         th->t_handle_save = current->journal_info;
3235                         current->journal_info = th;
3236                 }
3237         } else {
3238                 current->journal_info = th;
3239         }
3240         ret = do_journal_begin_r(th, sb, nblocks, JBEGIN_REG);
3241         BUG_ON(current->journal_info != th);
3242
3243         /*
3244          * I guess this boils down to being the reciprocal of clm-2100 above.
3245          * If do_journal_begin_r fails, we need to put it back, since
3246          * journal_end won't be called to do it. */
3247         if (ret)
3248                 current->journal_info = th->t_handle_save;
3249         else
3250                 BUG_ON(!th->t_refcount);
3251
3252         return ret;
3253 }
3254
3255 /*
3256  * puts bh into the current transaction.  If it was already there, reorders
3257  * removes the old pointers from the hash, and puts new ones in (to make
3258  * sure replay happen in the right order).
3259  *
3260  * if it was dirty, cleans and files onto the clean list.  I can't let it
3261  * be dirty again until the transaction is committed.
3262  *
3263  * if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len.
3264  */
3265 int journal_mark_dirty(struct reiserfs_transaction_handle *th,
3266                        struct buffer_head *bh)
3267 {
3268         struct super_block *sb = th->t_super;
3269         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3270         struct reiserfs_journal_cnode *cn = NULL;
3271         int count_already_incd = 0;
3272         int prepared = 0;
3273         BUG_ON(!th->t_trans_id);
3274
3275         PROC_INFO_INC(sb, journal.mark_dirty);
3276         if (th->t_trans_id != journal->j_trans_id) {
3277                 reiserfs_panic(th->t_super, "journal-1577",
3278                                "handle trans id %ld != current trans id %ld",
3279                                th->t_trans_id, journal->j_trans_id);
3280         }
3281
3282         prepared = test_clear_buffer_journal_prepared(bh);
3283         clear_buffer_journal_restore_dirty(bh);
3284         /* already in this transaction, we are done */
3285         if (buffer_journaled(bh)) {
3286                 PROC_INFO_INC(sb, journal.mark_dirty_already);
3287                 return 0;
3288         }
3289
3290         /*
3291          * this must be turned into a panic instead of a warning.  We can't
3292          * allow a dirty or journal_dirty or locked buffer to be logged, as
3293          * some changes could get to disk too early.  NOT GOOD.
3294          */
3295         if (!prepared || buffer_dirty(bh)) {
3296                 reiserfs_warning(sb, "journal-1777",
3297                                  "buffer %llu bad state "
3298                                  "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT",
3299                                  (unsigned long long)bh->b_blocknr,
3300                                  prepared ? ' ' : '!',
3301                                  buffer_locked(bh) ? ' ' : '!',
3302                                  buffer_dirty(bh) ? ' ' : '!',
3303                                  buffer_journal_dirty(bh) ? ' ' : '!');
3304         }
3305
3306         if (atomic_read(&journal->j_wcount) <= 0) {
3307                 reiserfs_warning(sb, "journal-1409",
3308                                  "returning because j_wcount was %d",
3309                                  atomic_read(&journal->j_wcount));
3310                 return 1;
3311         }
3312         /*
3313          * this error means I've screwed up, and we've overflowed
3314          * the transaction.  Nothing can be done here, except make the
3315          * FS readonly or panic.
3316          */
3317         if (journal->j_len >= journal->j_trans_max) {
3318                 reiserfs_panic(th->t_super, "journal-1413",
3319                                "j_len (%lu) is too big",
3320                                journal->j_len);
3321         }
3322
3323         if (buffer_journal_dirty(bh)) {
3324                 count_already_incd = 1;
3325                 PROC_INFO_INC(sb, journal.mark_dirty_notjournal);
3326                 clear_buffer_journal_dirty(bh);
3327         }
3328
3329         if (journal->j_len > journal->j_len_alloc) {
3330                 journal->j_len_alloc = journal->j_len + JOURNAL_PER_BALANCE_CNT;
3331         }
3332
3333         set_buffer_journaled(bh);
3334
3335         /* now put this guy on the end */
3336         if (!cn) {
3337                 cn = get_cnode(sb);
3338                 if (!cn) {
3339                         reiserfs_panic(sb, "journal-4", "get_cnode failed!");
3340                 }
3341
3342                 if (th->t_blocks_logged == th->t_blocks_allocated) {
3343                         th->t_blocks_allocated += JOURNAL_PER_BALANCE_CNT;
3344                         journal->j_len_alloc += JOURNAL_PER_BALANCE_CNT;
3345                 }
3346                 th->t_blocks_logged++;
3347                 journal->j_len++;
3348
3349                 cn->bh = bh;
3350                 cn->blocknr = bh->b_blocknr;
3351                 cn->sb = sb;
3352                 cn->jlist = NULL;
3353                 insert_journal_hash(journal->j_hash_table, cn);
3354                 if (!count_already_incd) {
3355                         get_bh(bh);
3356                 }
3357         }
3358         cn->next = NULL;
3359         cn->prev = journal->j_last;
3360         cn->bh = bh;
3361         if (journal->j_last) {
3362                 journal->j_last->next = cn;
3363                 journal->j_last = cn;
3364         } else {
3365                 journal->j_first = cn;
3366                 journal->j_last = cn;
3367         }
3368         reiserfs_schedule_old_flush(sb);
3369         return 0;
3370 }
3371
3372 int journal_end(struct reiserfs_transaction_handle *th)
3373 {
3374         struct super_block *sb = th->t_super;
3375         if (!current->journal_info && th->t_refcount > 1)
3376                 reiserfs_warning(sb, "REISER-NESTING",
3377                                  "th NULL, refcount %d", th->t_refcount);
3378
3379         if (!th->t_trans_id) {
3380                 WARN_ON(1);
3381                 return -EIO;
3382         }
3383
3384         th->t_refcount--;
3385         if (th->t_refcount > 0) {
3386                 struct reiserfs_transaction_handle *cur_th =
3387                     current->journal_info;
3388
3389                 /*
3390                  * we aren't allowed to close a nested transaction on a
3391                  * different filesystem from the one in the task struct
3392                  */
3393                 BUG_ON(cur_th->t_super != th->t_super);
3394
3395                 if (th != cur_th) {
3396                         memcpy(current->journal_info, th, sizeof(*th));
3397                         th->t_trans_id = 0;
3398                 }
3399                 return 0;
3400         } else {
3401                 return do_journal_end(th, 0);
3402         }
3403 }
3404
3405 /*
3406  * removes from the current transaction, relsing and descrementing any counters.
3407  * also files the removed buffer directly onto the clean list
3408  *
3409  * called by journal_mark_freed when a block has been deleted
3410  *
3411  * returns 1 if it cleaned and relsed the buffer. 0 otherwise
3412  */
3413 static int remove_from_transaction(struct super_block *sb,
3414                                    b_blocknr_t blocknr, int already_cleaned)
3415 {
3416         struct buffer_head *bh;
3417         struct reiserfs_journal_cnode *cn;
3418         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3419         int ret = 0;
3420
3421         cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr);
3422         if (!cn || !cn->bh) {
3423                 return ret;
3424         }
3425         bh = cn->bh;
3426         if (cn->prev) {
3427                 cn->prev->next = cn->next;
3428         }
3429         if (cn->next) {
3430                 cn->next->prev = cn->prev;
3431         }
3432         if (cn == journal->j_first) {
3433                 journal->j_first = cn->next;
3434         }
3435         if (cn == journal->j_last) {
3436                 journal->j_last = cn->prev;
3437         }
3438         remove_journal_hash(sb, journal->j_hash_table, NULL,
3439                             bh->b_blocknr, 0);
3440         clear_buffer_journaled(bh);     /* don't log this one */
3441
3442         if (!already_cleaned) {
3443                 clear_buffer_journal_dirty(bh);
3444                 clear_buffer_dirty(bh);
3445                 clear_buffer_journal_test(bh);
3446                 put_bh(bh);
3447                 if (atomic_read(&bh->b_count) < 0) {
3448                         reiserfs_warning(sb, "journal-1752",
3449                                          "b_count < 0");
3450                 }
3451                 ret = 1;
3452         }
3453         journal->j_len--;
3454         journal->j_len_alloc--;
3455         free_cnode(sb, cn);
3456         return ret;
3457 }
3458
3459 /*
3460  * for any cnode in a journal list, it can only be dirtied of all the
3461  * transactions that include it are committed to disk.
3462  * this checks through each transaction, and returns 1 if you are allowed
3463  * to dirty, and 0 if you aren't
3464  *
3465  * it is called by dirty_journal_list, which is called after
3466  * flush_commit_list has gotten all the log blocks for a given
3467  * transaction on disk
3468  *
3469  */
3470 static int can_dirty(struct reiserfs_journal_cnode *cn)
3471 {
3472         struct super_block *sb = cn->sb;
3473         b_blocknr_t blocknr = cn->blocknr;
3474         struct reiserfs_journal_cnode *cur = cn->hprev;
3475         int can_dirty = 1;
3476
3477         /*
3478          * first test hprev.  These are all newer than cn, so any node here
3479          * with the same block number and dev means this node can't be sent
3480          * to disk right now.
3481          */
3482         while (cur && can_dirty) {
3483                 if (cur->jlist && cur->bh && cur->blocknr && cur->sb == sb &&
3484                     cur->blocknr == blocknr) {
3485                         can_dirty = 0;
3486                 }
3487                 cur = cur->hprev;
3488         }
3489         /*
3490          * then test hnext.  These are all older than cn.  As long as they
3491          * are committed to the log, it is safe to write cn to disk
3492          */
3493         cur = cn->hnext;
3494         while (cur && can_dirty) {
3495                 if (cur->jlist && cur->jlist->j_len > 0 &&
3496                     atomic_read(&cur->jlist->j_commit_left) > 0 && cur->bh &&
3497                     cur->blocknr && cur->sb == sb && cur->blocknr == blocknr) {
3498                         can_dirty = 0;
3499                 }
3500                 cur = cur->hnext;
3501         }
3502         return can_dirty;
3503 }
3504
3505 /*
3506  * syncs the commit blocks, but does not force the real buffers to disk
3507  * will wait until the current transaction is done/committed before returning
3508  */
3509 int journal_end_sync(struct reiserfs_transaction_handle *th)
3510 {
3511         struct super_block *sb = th->t_super;
3512         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3513
3514         BUG_ON(!th->t_trans_id);
3515         /* you can sync while nested, very, very bad */
3516         BUG_ON(th->t_refcount > 1);
3517         if (journal->j_len == 0) {
3518                 reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3519                                              1);
3520                 journal_mark_dirty(th, SB_BUFFER_WITH_SB(sb));
3521         }
3522         return do_journal_end(th, COMMIT_NOW | WAIT);
3523 }
3524
3525 /* writeback the pending async commits to disk */
3526 static void flush_async_commits(struct work_struct *work)
3527 {
3528         struct reiserfs_journal *journal =
3529                 container_of(work, struct reiserfs_journal, j_work.work);
3530         struct super_block *sb = journal->j_work_sb;
3531         struct reiserfs_journal_list *jl;
3532         struct list_head *entry;
3533
3534         reiserfs_write_lock(sb);
3535         if (!list_empty(&journal->j_journal_list)) {
3536                 /* last entry is the youngest, commit it and you get everything */
3537                 entry = journal->j_journal_list.prev;
3538                 jl = JOURNAL_LIST_ENTRY(entry);
3539                 flush_commit_list(sb, jl, 1);
3540         }
3541         reiserfs_write_unlock(sb);
3542 }
3543
3544 /*
3545  * flushes any old transactions to disk
3546  * ends the current transaction if it is too old
3547  */
3548 void reiserfs_flush_old_commits(struct super_block *sb)
3549 {
3550         time64_t now;
3551         struct reiserfs_transaction_handle th;
3552         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3553
3554         now = ktime_get_seconds();
3555         /*
3556          * safety check so we don't flush while we are replaying the log during
3557          * mount
3558          */
3559         if (list_empty(&journal->j_journal_list))
3560                 return;
3561
3562         /*
3563          * check the current transaction.  If there are no writers, and it is
3564          * too old, finish it, and force the commit blocks to disk
3565          */
3566         if (atomic_read(&journal->j_wcount) <= 0 &&
3567             journal->j_trans_start_time > 0 &&
3568             journal->j_len > 0 &&
3569             (now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3570                 if (!journal_join(&th, sb)) {
3571                         reiserfs_prepare_for_journal(sb,
3572                                                      SB_BUFFER_WITH_SB(sb),
3573                                                      1);
3574                         journal_mark_dirty(&th, SB_BUFFER_WITH_SB(sb));
3575
3576                         /*
3577                          * we're only being called from kreiserfsd, it makes
3578                          * no sense to do an async commit so that kreiserfsd
3579                          * can do it later
3580                          */
3581                         do_journal_end(&th, COMMIT_NOW | WAIT);
3582                 }
3583         }
3584 }
3585
3586 /*
3587  * returns 0 if do_journal_end should return right away, returns 1 if
3588  * do_journal_end should finish the commit
3589  *
3590  * if the current transaction is too old, but still has writers, this will
3591  * wait on j_join_wait until all the writers are done.  By the time it
3592  * wakes up, the transaction it was called has already ended, so it just
3593  * flushes the commit list and returns 0.
3594  *
3595  * Won't batch when flush or commit_now is set.  Also won't batch when
3596  * others are waiting on j_join_wait.
3597  *
3598  * Note, we can't allow the journal_end to proceed while there are still
3599  * writers in the log.
3600  */
3601 static int check_journal_end(struct reiserfs_transaction_handle *th, int flags)
3602 {
3603
3604         time64_t now;
3605         int flush = flags & FLUSH_ALL;
3606         int commit_now = flags & COMMIT_NOW;
3607         int wait_on_commit = flags & WAIT;
3608         struct reiserfs_journal_list *jl;
3609         struct super_block *sb = th->t_super;
3610         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3611
3612         BUG_ON(!th->t_trans_id);
3613
3614         if (th->t_trans_id != journal->j_trans_id) {
3615                 reiserfs_panic(th->t_super, "journal-1577",
3616                                "handle trans id %ld != current trans id %ld",
3617                                th->t_trans_id, journal->j_trans_id);
3618         }
3619
3620         journal->j_len_alloc -= (th->t_blocks_allocated - th->t_blocks_logged);
3621         /* <= 0 is allowed.  unmounting might not call begin */
3622         if (atomic_read(&journal->j_wcount) > 0)
3623                 atomic_dec(&journal->j_wcount);
3624
3625         /*
3626          * BUG, deal with case where j_len is 0, but people previously
3627          * freed blocks need to be released will be dealt with by next
3628          * transaction that actually writes something, but should be taken
3629          * care of in this trans
3630          */
3631         BUG_ON(journal->j_len == 0);
3632
3633         /*
3634          * if wcount > 0, and we are called to with flush or commit_now,
3635          * we wait on j_join_wait.  We will wake up when the last writer has
3636          * finished the transaction, and started it on its way to the disk.
3637          * Then, we flush the commit or journal list, and just return 0
3638          * because the rest of journal end was already done for this
3639          * transaction.
3640          */
3641         if (atomic_read(&journal->j_wcount) > 0) {
3642                 if (flush || commit_now) {
3643                         unsigned trans_id;
3644
3645                         jl = journal->j_current_jl;
3646                         trans_id = jl->j_trans_id;
3647                         if (wait_on_commit)
3648                                 jl->j_state |= LIST_COMMIT_PENDING;
3649                         atomic_set(&journal->j_jlock, 1);
3650                         if (flush) {
3651                                 journal->j_next_full_flush = 1;
3652                         }
3653                         unlock_journal(sb);
3654
3655                         /*
3656                          * sleep while the current transaction is
3657                          * still j_jlocked
3658                          */
3659                         while (journal->j_trans_id == trans_id) {
3660                                 if (atomic_read(&journal->j_jlock)) {
3661                                         queue_log_writer(sb);
3662                                 } else {
3663                                         lock_journal(sb);
3664                                         if (journal->j_trans_id == trans_id) {
3665                                                 atomic_set(&journal->j_jlock,
3666                                                            1);
3667                                         }
3668                                         unlock_journal(sb);
3669                                 }
3670                         }
3671                         BUG_ON(journal->j_trans_id == trans_id);
3672
3673                         if (commit_now
3674                             && journal_list_still_alive(sb, trans_id)
3675                             && wait_on_commit) {
3676                                 flush_commit_list(sb, jl, 1);
3677                         }
3678                         return 0;
3679                 }
3680                 unlock_journal(sb);
3681                 return 0;
3682         }
3683
3684         /* deal with old transactions where we are the last writers */
3685         now = ktime_get_seconds();
3686         if ((now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3687                 commit_now = 1;
3688                 journal->j_next_async_flush = 1;
3689         }
3690         /* don't batch when someone is waiting on j_join_wait */
3691         /* don't batch when syncing the commit or flushing the whole trans */
3692         if (!(journal->j_must_wait > 0) && !(atomic_read(&journal->j_jlock))
3693             && !flush && !commit_now && (journal->j_len < journal->j_max_batch)
3694             && journal->j_len_alloc < journal->j_max_batch
3695             && journal->j_cnode_free > (journal->j_trans_max * 3)) {
3696                 journal->j_bcount++;
3697                 unlock_journal(sb);
3698                 return 0;
3699         }
3700
3701         if (journal->j_start > SB_ONDISK_JOURNAL_SIZE(sb)) {
3702                 reiserfs_panic(sb, "journal-003",
3703                                "j_start (%ld) is too high",
3704                                journal->j_start);
3705         }
3706         return 1;
3707 }
3708
3709 /*
3710  * Does all the work that makes deleting blocks safe.
3711  * when deleting a block mark BH_JNew, just remove it from the current
3712  * transaction, clean it's buffer_head and move on.
3713  *
3714  * otherwise:
3715  * set a bit for the block in the journal bitmap.  That will prevent it from
3716  * being allocated for unformatted nodes before this transaction has finished.
3717  *
3718  * mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers.
3719  * That will prevent any old transactions with this block from trying to flush
3720  * to the real location.  Since we aren't removing the cnode from the
3721  * journal_list_hash, *the block can't be reallocated yet.
3722  *
3723  * Then remove it from the current transaction, decrementing any counters and
3724  * filing it on the clean list.
3725  */
3726 int journal_mark_freed(struct reiserfs_transaction_handle *th,
3727                        struct super_block *sb, b_blocknr_t blocknr)
3728 {
3729         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3730         struct reiserfs_journal_cnode *cn = NULL;
3731         struct buffer_head *bh = NULL;
3732         struct reiserfs_list_bitmap *jb = NULL;
3733         int cleaned = 0;
3734         BUG_ON(!th->t_trans_id);
3735
3736         cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr);
3737         if (cn && cn->bh) {
3738                 bh = cn->bh;
3739                 get_bh(bh);
3740         }
3741         /* if it is journal new, we just remove it from this transaction */
3742         if (bh && buffer_journal_new(bh)) {
3743                 clear_buffer_journal_new(bh);
3744                 clear_prepared_bits(bh);
3745                 reiserfs_clean_and_file_buffer(bh);
3746                 cleaned = remove_from_transaction(sb, blocknr, cleaned);
3747         } else {
3748                 /*
3749                  * set the bit for this block in the journal bitmap
3750                  * for this transaction
3751                  */
3752                 jb = journal->j_current_jl->j_list_bitmap;
3753                 if (!jb) {
3754                         reiserfs_panic(sb, "journal-1702",
3755                                        "journal_list_bitmap is NULL");
3756                 }
3757                 set_bit_in_list_bitmap(sb, blocknr, jb);
3758
3759                 /* Note, the entire while loop is not allowed to schedule.  */
3760
3761                 if (bh) {
3762                         clear_prepared_bits(bh);
3763                         reiserfs_clean_and_file_buffer(bh);
3764                 }
3765                 cleaned = remove_from_transaction(sb, blocknr, cleaned);
3766
3767                 /*
3768                  * find all older transactions with this block,
3769                  * make sure they don't try to write it out
3770                  */
3771                 cn = get_journal_hash_dev(sb, journal->j_list_hash_table,
3772                                           blocknr);
3773                 while (cn) {
3774                         if (sb == cn->sb && blocknr == cn->blocknr) {
3775                                 set_bit(BLOCK_FREED, &cn->state);
3776                                 if (cn->bh) {
3777                                         /*
3778                                          * remove_from_transaction will brelse
3779                                          * the buffer if it was in the current
3780                                          * trans
3781                                          */
3782                                         if (!cleaned) {
3783                                                 clear_buffer_journal_dirty(cn->
3784                                                                            bh);
3785                                                 clear_buffer_dirty(cn->bh);
3786                                                 clear_buffer_journal_test(cn->
3787                                                                           bh);
3788                                                 cleaned = 1;
3789                                                 put_bh(cn->bh);
3790                                                 if (atomic_read
3791                                                     (&cn->bh->b_count) < 0) {
3792                                                         reiserfs_warning(sb,
3793                                                                  "journal-2138",
3794                                                                  "cn->bh->b_count < 0");
3795                                                 }
3796                                         }
3797                                         /*
3798                                          * since we are clearing the bh,
3799                                          * we MUST dec nonzerolen
3800                                          */
3801                                         if (cn->jlist) {
3802                                                 atomic_dec(&cn->jlist->
3803                                                            j_nonzerolen);
3804                                         }
3805                                         cn->bh = NULL;
3806                                 }
3807                         }
3808                         cn = cn->hnext;
3809                 }
3810         }
3811
3812         if (bh)
3813                 release_buffer_page(bh); /* get_hash grabs the buffer */
3814         return 0;
3815 }
3816
3817 void reiserfs_update_inode_transaction(struct inode *inode)
3818 {
3819         struct reiserfs_journal *journal = SB_JOURNAL(inode->i_sb);
3820         REISERFS_I(inode)->i_jl = journal->j_current_jl;
3821         REISERFS_I(inode)->i_trans_id = journal->j_trans_id;
3822 }
3823
3824 /*
3825  * returns -1 on error, 0 if no commits/barriers were done and 1
3826  * if a transaction was actually committed and the barrier was done
3827  */
3828 static int __commit_trans_jl(struct inode *inode, unsigned long id,
3829                              struct reiserfs_journal_list *jl)
3830 {
3831         struct reiserfs_transaction_handle th;
3832         struct super_block *sb = inode->i_sb;
3833         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3834         int ret = 0;
3835
3836         /*
3837          * is it from the current transaction,
3838          * or from an unknown transaction?
3839          */
3840         if (id == journal->j_trans_id) {
3841                 jl = journal->j_current_jl;
3842                 /*
3843                  * try to let other writers come in and
3844                  * grow this transaction
3845                  */
3846                 let_transaction_grow(sb, id);
3847                 if (journal->j_trans_id != id) {
3848                         goto flush_commit_only;
3849                 }
3850
3851                 ret = journal_begin(&th, sb, 1);
3852                 if (ret)
3853                         return ret;
3854
3855                 /* someone might have ended this transaction while we joined */
3856                 if (journal->j_trans_id != id) {
3857                         reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3858                                                      1);
3859                         journal_mark_dirty(&th, SB_BUFFER_WITH_SB(sb));
3860                         ret = journal_end(&th);
3861                         goto flush_commit_only;
3862                 }
3863
3864                 ret = journal_end_sync(&th);
3865                 if (!ret)
3866                         ret = 1;
3867
3868         } else {
3869                 /*
3870                  * this gets tricky, we have to make sure the journal list in
3871                  * the inode still exists.  We know the list is still around
3872                  * if we've got a larger transaction id than the oldest list
3873                  */
3874 flush_commit_only:
3875                 if (journal_list_still_alive(inode->i_sb, id)) {
3876                         /*
3877                          * we only set ret to 1 when we know for sure
3878                          * the barrier hasn't been started yet on the commit
3879                          * block.
3880                          */
3881                         if (atomic_read(&jl->j_commit_left) > 1)
3882                                 ret = 1;
3883                         flush_commit_list(sb, jl, 1);
3884                         if (journal->j_errno)
3885                                 ret = journal->j_errno;
3886                 }
3887         }
3888         /* otherwise the list is gone, and long since committed */
3889         return ret;
3890 }
3891
3892 int reiserfs_commit_for_inode(struct inode *inode)
3893 {
3894         unsigned int id = REISERFS_I(inode)->i_trans_id;
3895         struct reiserfs_journal_list *jl = REISERFS_I(inode)->i_jl;
3896
3897         /*
3898          * for the whole inode, assume unset id means it was
3899          * changed in the current transaction.  More conservative
3900          */
3901         if (!id || !jl) {
3902                 reiserfs_update_inode_transaction(inode);
3903                 id = REISERFS_I(inode)->i_trans_id;
3904                 /* jl will be updated in __commit_trans_jl */
3905         }
3906
3907         return __commit_trans_jl(inode, id, jl);
3908 }
3909
3910 void reiserfs_restore_prepared_buffer(struct super_block *sb,
3911                                       struct buffer_head *bh)
3912 {
3913         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3914         PROC_INFO_INC(sb, journal.restore_prepared);
3915         if (!bh) {
3916                 return;
3917         }
3918         if (test_clear_buffer_journal_restore_dirty(bh) &&
3919             buffer_journal_dirty(bh)) {
3920                 struct reiserfs_journal_cnode *cn;
3921                 reiserfs_write_lock(sb);
3922                 cn = get_journal_hash_dev(sb,
3923                                           journal->j_list_hash_table,
3924                                           bh->b_blocknr);
3925                 if (cn && can_dirty(cn)) {
3926                         set_buffer_journal_test(bh);
3927                         mark_buffer_dirty(bh);
3928                 }
3929                 reiserfs_write_unlock(sb);
3930         }
3931         clear_buffer_journal_prepared(bh);
3932 }
3933
3934 extern struct tree_balance *cur_tb;
3935 /*
3936  * before we can change a metadata block, we have to make sure it won't
3937  * be written to disk while we are altering it.  So, we must:
3938  * clean it
3939  * wait on it.
3940  */
3941 int reiserfs_prepare_for_journal(struct super_block *sb,
3942                                  struct buffer_head *bh, int wait)
3943 {
3944         PROC_INFO_INC(sb, journal.prepare);
3945
3946         if (!trylock_buffer(bh)) {
3947                 if (!wait)
3948                         return 0;
3949                 lock_buffer(bh);
3950         }
3951         set_buffer_journal_prepared(bh);
3952         if (test_clear_buffer_dirty(bh) && buffer_journal_dirty(bh)) {
3953                 clear_buffer_journal_test(bh);
3954                 set_buffer_journal_restore_dirty(bh);
3955         }
3956         unlock_buffer(bh);
3957         return 1;
3958 }
3959
3960 /*
3961  * long and ugly.  If flush, will not return until all commit
3962  * blocks and all real buffers in the trans are on disk.
3963  * If no_async, won't return until all commit blocks are on disk.
3964  *
3965  * keep reading, there are comments as you go along
3966  *
3967  * If the journal is aborted, we just clean up. Things like flushing
3968  * journal lists, etc just won't happen.
3969  */
3970 static int do_journal_end(struct reiserfs_transaction_handle *th, int flags)
3971 {
3972         struct super_block *sb = th->t_super;
3973         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3974         struct reiserfs_journal_cnode *cn, *next, *jl_cn;
3975         struct reiserfs_journal_cnode *last_cn = NULL;
3976         struct reiserfs_journal_desc *desc;
3977         struct reiserfs_journal_commit *commit;
3978         struct buffer_head *c_bh;       /* commit bh */
3979         struct buffer_head *d_bh;       /* desc bh */
3980         int cur_write_start = 0;        /* start index of current log write */
3981         int i;
3982         int flush;
3983         int wait_on_commit;
3984         struct reiserfs_journal_list *jl, *temp_jl;
3985         struct list_head *entry, *safe;
3986         unsigned long jindex;
3987         unsigned int commit_trans_id;
3988         int trans_half;
3989         int depth;
3990
3991         BUG_ON(th->t_refcount > 1);
3992         BUG_ON(!th->t_trans_id);
3993         BUG_ON(!th->t_super);
3994
3995         /*
3996          * protect flush_older_commits from doing mistakes if the
3997          * transaction ID counter gets overflowed.
3998          */
3999         if (th->t_trans_id == ~0U)
4000                 flags |= FLUSH_ALL | COMMIT_NOW | WAIT;
4001         flush = flags & FLUSH_ALL;
4002         wait_on_commit = flags & WAIT;
4003
4004         current->journal_info = th->t_handle_save;
4005         reiserfs_check_lock_depth(sb, "journal end");
4006         if (journal->j_len == 0) {
4007                 reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
4008                                              1);
4009                 journal_mark_dirty(th, SB_BUFFER_WITH_SB(sb));
4010         }
4011
4012         lock_journal(sb);
4013         if (journal->j_next_full_flush) {
4014                 flags |= FLUSH_ALL;
4015                 flush = 1;
4016         }
4017         if (journal->j_next_async_flush) {
4018                 flags |= COMMIT_NOW | WAIT;
4019                 wait_on_commit = 1;
4020         }
4021
4022         /*
4023          * check_journal_end locks the journal, and unlocks if it does
4024          * not return 1 it tells us if we should continue with the
4025          * journal_end, or just return
4026          */
4027         if (!check_journal_end(th, flags)) {
4028                 reiserfs_schedule_old_flush(sb);
4029                 wake_queued_writers(sb);
4030                 reiserfs_async_progress_wait(sb);
4031                 goto out;
4032         }
4033
4034         /* check_journal_end might set these, check again */
4035         if (journal->j_next_full_flush) {
4036                 flush = 1;
4037         }
4038
4039         /*
4040          * j must wait means we have to flush the log blocks, and the
4041          * real blocks for this transaction
4042          */
4043         if (journal->j_must_wait > 0) {
4044                 flush = 1;
4045         }
4046 #ifdef REISERFS_PREALLOCATE
4047         /*
4048          * quota ops might need to nest, setup the journal_info pointer
4049          * for them and raise the refcount so that it is > 0.
4050          */
4051         current->journal_info = th;
4052         th->t_refcount++;
4053
4054         /* it should not involve new blocks into the transaction */
4055         reiserfs_discard_all_prealloc(th);
4056
4057         th->t_refcount--;
4058         current->journal_info = th->t_handle_save;
4059 #endif
4060
4061         /* setup description block */
4062         d_bh =
4063             journal_getblk(sb,
4064                            SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4065                            journal->j_start);
4066         set_buffer_uptodate(d_bh);
4067         desc = (struct reiserfs_journal_desc *)(d_bh)->b_data;
4068         memset(d_bh->b_data, 0, d_bh->b_size);
4069         memcpy(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8);
4070         set_desc_trans_id(desc, journal->j_trans_id);
4071
4072         /*
4073          * setup commit block.  Don't write (keep it clean too) this one
4074          * until after everyone else is written
4075          */
4076         c_bh = journal_getblk(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4077                               ((journal->j_start + journal->j_len +
4078                                 1) % SB_ONDISK_JOURNAL_SIZE(sb)));
4079         commit = (struct reiserfs_journal_commit *)c_bh->b_data;
4080         memset(c_bh->b_data, 0, c_bh->b_size);
4081         set_commit_trans_id(commit, journal->j_trans_id);
4082         set_buffer_uptodate(c_bh);
4083
4084         /* init this journal list */
4085         jl = journal->j_current_jl;
4086
4087         /*
4088          * we lock the commit before doing anything because
4089          * we want to make sure nobody tries to run flush_commit_list until
4090          * the new transaction is fully setup, and we've already flushed the
4091          * ordered bh list
4092          */
4093         reiserfs_mutex_lock_safe(&jl->j_commit_mutex, sb);
4094
4095         /* save the transaction id in case we need to commit it later */
4096         commit_trans_id = jl->j_trans_id;
4097
4098         atomic_set(&jl->j_older_commits_done, 0);
4099         jl->j_trans_id = journal->j_trans_id;
4100         jl->j_timestamp = journal->j_trans_start_time;
4101         jl->j_commit_bh = c_bh;
4102         jl->j_start = journal->j_start;
4103         jl->j_len = journal->j_len;
4104         atomic_set(&jl->j_nonzerolen, journal->j_len);
4105         atomic_set(&jl->j_commit_left, journal->j_len + 2);
4106         jl->j_realblock = NULL;
4107
4108         /*
4109          * The ENTIRE FOR LOOP MUST not cause schedule to occur.
4110          * for each real block, add it to the journal list hash,
4111          * copy into real block index array in the commit or desc block
4112          */
4113         trans_half = journal_trans_half(sb->s_blocksize);
4114         for (i = 0, cn = journal->j_first; cn; cn = cn->next, i++) {
4115                 if (buffer_journaled(cn->bh)) {
4116                         jl_cn = get_cnode(sb);
4117                         if (!jl_cn) {
4118                                 reiserfs_panic(sb, "journal-1676",
4119                                                "get_cnode returned NULL");
4120                         }
4121                         if (i == 0) {
4122                                 jl->j_realblock = jl_cn;
4123                         }
4124                         jl_cn->prev = last_cn;
4125                         jl_cn->next = NULL;
4126                         if (last_cn) {
4127                                 last_cn->next = jl_cn;
4128                         }
4129                         last_cn = jl_cn;
4130                         /*
4131                          * make sure the block we are trying to log
4132                          * is not a block of journal or reserved area
4133                          */
4134                         if (is_block_in_log_or_reserved_area
4135                             (sb, cn->bh->b_blocknr)) {
4136                                 reiserfs_panic(sb, "journal-2332",
4137                                                "Trying to log block %lu, "
4138                                                "which is a log block",
4139                                                cn->bh->b_blocknr);
4140                         }
4141                         jl_cn->blocknr = cn->bh->b_blocknr;
4142                         jl_cn->state = 0;
4143                         jl_cn->sb = sb;
4144                         jl_cn->bh = cn->bh;
4145                         jl_cn->jlist = jl;
4146                         insert_journal_hash(journal->j_list_hash_table, jl_cn);
4147                         if (i < trans_half) {
4148                                 desc->j_realblock[i] =
4149                                     cpu_to_le32(cn->bh->b_blocknr);
4150                         } else {
4151                                 commit->j_realblock[i - trans_half] =
4152                                     cpu_to_le32(cn->bh->b_blocknr);
4153                         }
4154                 } else {
4155                         i--;
4156                 }
4157         }
4158         set_desc_trans_len(desc, journal->j_len);
4159         set_desc_mount_id(desc, journal->j_mount_id);
4160         set_desc_trans_id(desc, journal->j_trans_id);
4161         set_commit_trans_len(commit, journal->j_len);
4162
4163         /*
4164          * special check in case all buffers in the journal
4165          * were marked for not logging
4166          */
4167         BUG_ON(journal->j_len == 0);
4168
4169         /*
4170          * we're about to dirty all the log blocks, mark the description block
4171          * dirty now too.  Don't mark the commit block dirty until all the
4172          * others are on disk
4173          */
4174         mark_buffer_dirty(d_bh);
4175
4176         /*
4177          * first data block is j_start + 1, so add one to
4178          * cur_write_start wherever you use it
4179          */
4180         cur_write_start = journal->j_start;
4181         cn = journal->j_first;
4182         jindex = 1;     /* start at one so we don't get the desc again */
4183         while (cn) {
4184                 clear_buffer_journal_new(cn->bh);
4185                 /* copy all the real blocks into log area.  dirty log blocks */
4186                 if (buffer_journaled(cn->bh)) {
4187                         struct buffer_head *tmp_bh;
4188                         char *addr;
4189                         struct page *page;
4190                         tmp_bh =
4191                             journal_getblk(sb,
4192                                            SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4193                                            ((cur_write_start +
4194                                              jindex) %
4195                                             SB_ONDISK_JOURNAL_SIZE(sb)));
4196                         set_buffer_uptodate(tmp_bh);
4197                         page = cn->bh->b_page;
4198                         addr = kmap(page);
4199                         memcpy(tmp_bh->b_data,
4200                                addr + offset_in_page(cn->bh->b_data),
4201                                cn->bh->b_size);
4202                         kunmap(page);
4203                         mark_buffer_dirty(tmp_bh);
4204                         jindex++;
4205                         set_buffer_journal_dirty(cn->bh);
4206                         clear_buffer_journaled(cn->bh);
4207                 } else {
4208                         /*
4209                          * JDirty cleared sometime during transaction.
4210                          * don't log this one
4211                          */
4212                         reiserfs_warning(sb, "journal-2048",
4213                                          "BAD, buffer in journal hash, "
4214                                          "but not JDirty!");
4215                         brelse(cn->bh);
4216                 }
4217                 next = cn->next;
4218                 free_cnode(sb, cn);
4219                 cn = next;
4220                 reiserfs_cond_resched(sb);
4221         }
4222
4223         /*
4224          * we are done with both the c_bh and d_bh, but
4225          * c_bh must be written after all other commit blocks,
4226          * so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1.
4227          */
4228
4229         journal->j_current_jl = alloc_journal_list(sb);
4230
4231         /* now it is safe to insert this transaction on the main list */
4232         list_add_tail(&jl->j_list, &journal->j_journal_list);
4233         list_add_tail(&jl->j_working_list, &journal->j_working_list);
4234         journal->j_num_work_lists++;
4235
4236         /* reset journal values for the next transaction */
4237         journal->j_start =
4238             (journal->j_start + journal->j_len +
4239              2) % SB_ONDISK_JOURNAL_SIZE(sb);
4240         atomic_set(&journal->j_wcount, 0);
4241         journal->j_bcount = 0;
4242         journal->j_last = NULL;
4243         journal->j_first = NULL;
4244         journal->j_len = 0;
4245         journal->j_trans_start_time = 0;
4246         /* check for trans_id overflow */
4247         if (++journal->j_trans_id == 0)
4248                 journal->j_trans_id = 10;
4249         journal->j_current_jl->j_trans_id = journal->j_trans_id;
4250         journal->j_must_wait = 0;
4251         journal->j_len_alloc = 0;
4252         journal->j_next_full_flush = 0;
4253         journal->j_next_async_flush = 0;
4254         init_journal_hash(sb);
4255
4256         /*
4257          * make sure reiserfs_add_jh sees the new current_jl before we
4258          * write out the tails
4259          */
4260         smp_mb();
4261
4262         /*
4263          * tail conversion targets have to hit the disk before we end the
4264          * transaction.  Otherwise a later transaction might repack the tail
4265          * before this transaction commits, leaving the data block unflushed
4266          * and clean, if we crash before the later transaction commits, the
4267          * data block is lost.
4268          */
4269         if (!list_empty(&jl->j_tail_bh_list)) {
4270                 depth = reiserfs_write_unlock_nested(sb);
4271                 write_ordered_buffers(&journal->j_dirty_buffers_lock,
4272                                       journal, jl, &jl->j_tail_bh_list);
4273                 reiserfs_write_lock_nested(sb, depth);
4274         }
4275         BUG_ON(!list_empty(&jl->j_tail_bh_list));
4276         mutex_unlock(&jl->j_commit_mutex);
4277
4278         /*
4279          * honor the flush wishes from the caller, simple commits can
4280          * be done outside the journal lock, they are done below
4281          *
4282          * if we don't flush the commit list right now, we put it into
4283          * the work queue so the people waiting on the async progress work
4284          * queue don't wait for this proc to flush journal lists and such.
4285          */
4286         if (flush) {
4287                 flush_commit_list(sb, jl, 1);
4288                 flush_journal_list(sb, jl, 1);
4289         } else if (!(jl->j_state & LIST_COMMIT_PENDING)) {
4290                 /*
4291                  * Avoid queueing work when sb is being shut down. Transaction
4292                  * will be flushed on journal shutdown.
4293                  */
4294                 if (sb->s_flags & SB_ACTIVE)
4295                         queue_delayed_work(REISERFS_SB(sb)->commit_wq,
4296                                            &journal->j_work, HZ / 10);
4297         }
4298
4299         /*
4300          * if the next transaction has any chance of wrapping, flush
4301          * transactions that might get overwritten.  If any journal lists
4302          * are very old flush them as well.
4303          */
4304 first_jl:
4305         list_for_each_safe(entry, safe, &journal->j_journal_list) {
4306                 temp_jl = JOURNAL_LIST_ENTRY(entry);
4307                 if (journal->j_start <= temp_jl->j_start) {
4308                         if ((journal->j_start + journal->j_trans_max + 1) >=
4309                             temp_jl->j_start) {
4310                                 flush_used_journal_lists(sb, temp_jl);
4311                                 goto first_jl;
4312                         } else if ((journal->j_start +
4313                                     journal->j_trans_max + 1) <
4314                                    SB_ONDISK_JOURNAL_SIZE(sb)) {
4315                                 /*
4316                                  * if we don't cross into the next
4317                                  * transaction and we don't wrap, there is
4318                                  * no way we can overlap any later transactions
4319                                  * break now
4320                                  */
4321                                 break;
4322                         }
4323                 } else if ((journal->j_start +
4324                             journal->j_trans_max + 1) >
4325                            SB_ONDISK_JOURNAL_SIZE(sb)) {
4326                         if (((journal->j_start + journal->j_trans_max + 1) %
4327                              SB_ONDISK_JOURNAL_SIZE(sb)) >=
4328                             temp_jl->j_start) {
4329                                 flush_used_journal_lists(sb, temp_jl);
4330                                 goto first_jl;
4331                         } else {
4332                                 /*
4333                                 * we don't overlap anything from out start
4334                                 * to the end of the log, and our wrapped
4335                                 * portion doesn't overlap anything at
4336                                 * the start of the log.  We can break
4337                                 */
4338                                 break;
4339                         }
4340                 }
4341         }
4342
4343         journal->j_current_jl->j_list_bitmap =
4344             get_list_bitmap(sb, journal->j_current_jl);
4345
4346         if (!(journal->j_current_jl->j_list_bitmap)) {
4347                 reiserfs_panic(sb, "journal-1996",
4348                                "could not get a list bitmap");
4349         }
4350
4351         atomic_set(&journal->j_jlock, 0);
4352         unlock_journal(sb);
4353         /* wake up any body waiting to join. */
4354         clear_bit(J_WRITERS_QUEUED, &journal->j_state);
4355         wake_up(&journal->j_join_wait);
4356
4357         if (!flush && wait_on_commit &&
4358             journal_list_still_alive(sb, commit_trans_id)) {
4359                 flush_commit_list(sb, jl, 1);
4360         }
4361 out:
4362         reiserfs_check_lock_depth(sb, "journal end2");
4363
4364         memset(th, 0, sizeof(*th));
4365         /*
4366          * Re-set th->t_super, so we can properly keep track of how many
4367          * persistent transactions there are. We need to do this so if this
4368          * call is part of a failed restart_transaction, we can free it later
4369          */
4370         th->t_super = sb;
4371
4372         return journal->j_errno;
4373 }
4374
4375 /* Send the file system read only and refuse new transactions */
4376 void reiserfs_abort_journal(struct super_block *sb, int errno)
4377 {
4378         struct reiserfs_journal *journal = SB_JOURNAL(sb);
4379         if (test_bit(J_ABORTED, &journal->j_state))
4380                 return;
4381
4382         if (!journal->j_errno)
4383                 journal->j_errno = errno;
4384
4385         sb->s_flags |= SB_RDONLY;
4386         set_bit(J_ABORTED, &journal->j_state);
4387
4388 #ifdef CONFIG_REISERFS_CHECK
4389         dump_stack();
4390 #endif
4391 }