fs/gfs2/log.c

   1 /*
   2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   3  * Copyright (C) 2004-2006 Red Hat, Inc.  All rights reserved.
   4  *
   5  * This copyrighted material is made available to anyone wishing to use,
   6  * modify, copy, or redistribute it subject to the terms and conditions
   7  * of the GNU General Public License version 2.
   8  */
   9
  10 #include <linux/sched.h>
  11 #include <linux/slab.h>
  12 #include <linux/spinlock.h>
  13 #include <linux/completion.h>
  14 #include <linux/buffer_head.h>
  15 #include <linux/gfs2_ondisk.h>
  16 #include <linux/crc32.h>
  17 #include <linux/lm_interface.h>
  18 #include <linux/delay.h>
  19 #include <linux/kthread.h>
  20 #include <linux/freezer.h>
  21
  22 #include "gfs2.h"
  23 #include "incore.h"
  24 #include "bmap.h"
  25 #include "glock.h"
  26 #include "log.h"
  27 #include "lops.h"
  28 #include "meta_io.h"
  29 #include "util.h"
  30 #include "dir.h"
  31 #include "super.h"
  32
  33 #define PULL 1
  34
  35 /**
  36  * gfs2_struct2blk - compute stuff
  37  * @sdp: the filesystem
  38  * @nstruct: the number of structures
  39  * @ssize: the size of the structures
  40  *
  41  * Compute the number of log descriptor blocks needed to hold a certain number
  42  * of structures of a certain size.
  43  *
  44  * Returns: the number of blocks needed (minimum is always 1)
  45  */
  46
  47 unsigned int gfs2_struct2blk(struct gfs2_sbd *sdp, unsigned int nstruct,
  48                              unsigned int ssize)
  49 {
  50         unsigned int blks;
  51         unsigned int first, second;
  52
  53         blks = 1;
  54         first = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) / ssize;
  55
  56         if (nstruct > first) {
  57                 second = (sdp->sd_sb.sb_bsize -
  58                           sizeof(struct gfs2_meta_header)) / ssize;
  59                 blks += DIV_ROUND_UP(nstruct - first, second);
  60         }
  61
  62         return blks;
  63 }
  64
  65 /**
  66  * gfs2_remove_from_ail - Remove an entry from the ail lists, updating counters
  67  * @mapping: The associated mapping (maybe NULL)
  68  * @bd: The gfs2_bufdata to remove
  69  *
  70  * The log lock _must_ be held when calling this function
  71  *
  72  */
  73
  74 void gfs2_remove_from_ail(struct gfs2_bufdata *bd)
  75 {
  76         bd->bd_ail = NULL;
  77         list_del_init(&bd->bd_ail_st_list);
  78         list_del_init(&bd->bd_ail_gl_list);
  79         atomic_dec(&bd->bd_gl->gl_ail_count);
  80         brelse(bd->bd_bh);
  81 }
  82
  83 /**
  84  * gfs2_ail1_start_one - Start I/O on a part of the AIL
  85  * @sdp: the filesystem
  86  * @tr: the part of the AIL
  87  *
  88  */
  89
  90 static void gfs2_ail1_start_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
  91 {
  92         struct gfs2_bufdata *bd, *s;
  93         struct buffer_head *bh;
  94         int retry;
  95
  96         BUG_ON(!spin_is_locked(&sdp->sd_log_lock));
  97
  98         do {
  99                 retry = 0;
 100
 101                 list_for_each_entry_safe_reverse(bd, s, &ai->ai_ail1_list,
 102                                                  bd_ail_st_list) {
 103                         bh = bd->bd_bh;
 104
 105                         gfs2_assert(sdp, bd->bd_ail == ai);
 106
 107                         if (!buffer_busy(bh)) {
 108                                 if (!buffer_uptodate(bh))
 109                                         gfs2_io_error_bh(sdp, bh);
 110                                 list_move(&bd->bd_ail_st_list, &ai->ai_ail2_list);
 111                                 continue;
 112                         }
 113
 114                         if (!buffer_dirty(bh))
 115                                 continue;
 116
 117                         list_move(&bd->bd_ail_st_list, &ai->ai_ail1_list);
 118
 119                         get_bh(bh);
 120                         gfs2_log_unlock(sdp);
 121                         lock_buffer(bh);
 122                         if (test_clear_buffer_dirty(bh)) {
 123                                 bh->b_end_io = end_buffer_write_sync;
 124                                 submit_bh(WRITE, bh);
 125                         } else {
 126                                 unlock_buffer(bh);
 127                                 brelse(bh);
 128                         }
 129                         gfs2_log_lock(sdp);
 130
 131                         retry = 1;
 132                         break;
 133                 }
 134         } while (retry);
 135 }
 136
 137 /**
 138  * gfs2_ail1_empty_one - Check whether or not a trans in the AIL has been synced
 139  * @sdp: the filesystem
 140  * @ai: the AIL entry
 141  *
 142  */
 143
 144 static int gfs2_ail1_empty_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai, int flags)
 145 {
 146         struct gfs2_bufdata *bd, *s;
 147         struct buffer_head *bh;
 148
 149         list_for_each_entry_safe_reverse(bd, s, &ai->ai_ail1_list,
 150                                          bd_ail_st_list) {
 151                 bh = bd->bd_bh;
 152
 153                 gfs2_assert(sdp, bd->bd_ail == ai);
 154
 155                 if (buffer_busy(bh)) {
 156                         if (flags & DIO_ALL)
 157                                 continue;
 158                         else
 159                                 break;
 160                 }
 161
 162                 if (!buffer_uptodate(bh))
 163                         gfs2_io_error_bh(sdp, bh);
 164
 165                 list_move(&bd->bd_ail_st_list, &ai->ai_ail2_list);
 166         }
 167
 168         return list_empty(&ai->ai_ail1_list);
 169 }
 170
 171 static void gfs2_ail1_start(struct gfs2_sbd *sdp, int flags)
 172 {
 173         struct list_head *head;
 174         u64 sync_gen;
 175         struct list_head *first;
 176         struct gfs2_ail *first_ai, *ai, *tmp;
 177         int done = 0;
 178
 179         gfs2_log_lock(sdp);
 180         head = &sdp->sd_ail1_list;
 181         if (list_empty(head)) {
 182                 gfs2_log_unlock(sdp);
 183                 return;
 184         }
 185         sync_gen = sdp->sd_ail_sync_gen++;
 186
 187         first = head->prev;
 188         first_ai = list_entry(first, struct gfs2_ail, ai_list);
 189         first_ai->ai_sync_gen = sync_gen;
 190         gfs2_ail1_start_one(sdp, first_ai); /* This may drop log lock */
 191
 192         if (flags & DIO_ALL)
 193                 first = NULL;
 194
 195         while(!done) {
 196                 if (first && (head->prev != first ||
 197                               gfs2_ail1_empty_one(sdp, first_ai, 0)))
 198                         break;
 199
 200                 done = 1;
 201                 list_for_each_entry_safe_reverse(ai, tmp, head, ai_list) {
 202                         if (ai->ai_sync_gen >= sync_gen)
 203                                 continue;
 204                         ai->ai_sync_gen = sync_gen;
 205                         gfs2_ail1_start_one(sdp, ai); /* This may drop log lock */
 206                         done = 0;
 207                         break;
 208                 }
 209         }
 210
 211         gfs2_log_unlock(sdp);
 212 }
 213
 214 static int gfs2_ail1_empty(struct gfs2_sbd *sdp, int flags)
 215 {
 216         struct gfs2_ail *ai, *s;
 217         int ret;
 218
 219         gfs2_log_lock(sdp);
 220
 221         list_for_each_entry_safe_reverse(ai, s, &sdp->sd_ail1_list, ai_list) {
 222                 if (gfs2_ail1_empty_one(sdp, ai, flags))
 223                         list_move(&ai->ai_list, &sdp->sd_ail2_list);
 224                 else if (!(flags & DIO_ALL))
 225                         break;
 226         }
 227
 228         ret = list_empty(&sdp->sd_ail1_list);
 229
 230         gfs2_log_unlock(sdp);
 231
 232         return ret;
 233 }
 234
 235
 236 /**
 237  * gfs2_ail2_empty_one - Check whether or not a trans in the AIL has been synced
 238  * @sdp: the filesystem
 239  * @ai: the AIL entry
 240  *
 241  */
 242
 243 static void gfs2_ail2_empty_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
 244 {
 245         struct list_head *head = &ai->ai_ail2_list;
 246         struct gfs2_bufdata *bd;
 247
 248         while (!list_empty(head)) {
 249                 bd = list_entry(head->prev, struct gfs2_bufdata,
 250                                 bd_ail_st_list);
 251                 gfs2_assert(sdp, bd->bd_ail == ai);
 252                 gfs2_remove_from_ail(bd);
 253         }
 254 }
 255
 256 static void ail2_empty(struct gfs2_sbd *sdp, unsigned int new_tail)
 257 {
 258         struct gfs2_ail *ai, *safe;
 259         unsigned int old_tail = sdp->sd_log_tail;
 260         int wrap = (new_tail < old_tail);
 261         int a, b, rm;
 262
 263         gfs2_log_lock(sdp);
 264
 265         list_for_each_entry_safe(ai, safe, &sdp->sd_ail2_list, ai_list) {
 266                 a = (old_tail <= ai->ai_first);
 267                 b = (ai->ai_first < new_tail);
 268                 rm = (wrap) ? (a || b) : (a && b);
 269                 if (!rm)
 270                         continue;
 271
 272                 gfs2_ail2_empty_one(sdp, ai);
 273                 list_del(&ai->ai_list);
 274                 gfs2_assert_warn(sdp, list_empty(&ai->ai_ail1_list));
 275                 gfs2_assert_warn(sdp, list_empty(&ai->ai_ail2_list));
 276                 kfree(ai);
 277         }
 278
 279         gfs2_log_unlock(sdp);
 280 }
 281
 282 /**
 283  * gfs2_log_reserve - Make a log reservation
 284  * @sdp: The GFS2 superblock
 285  * @blks: The number of blocks to reserve
 286  *
 287  * Note that we never give out the last few blocks of the journal. Thats
 288  * due to the fact that there is a small number of header blocks
 289  * associated with each log flush. The exact number can't be known until
 290  * flush time, so we ensure that we have just enough free blocks at all
 291  * times to avoid running out during a log flush.
 292  *
 293  * Returns: errno
 294  */
 295
 296 int gfs2_log_reserve(struct gfs2_sbd *sdp, unsigned int blks)
 297 {
 298         unsigned int try = 0;
 299         unsigned reserved_blks = 6 * (4096 / sdp->sd_vfs->s_blocksize);
 300
 301         if (gfs2_assert_warn(sdp, blks) ||
 302             gfs2_assert_warn(sdp, blks <= sdp->sd_jdesc->jd_blocks))
 303                 return -EINVAL;
 304
 305         mutex_lock(&sdp->sd_log_reserve_mutex);
 306         gfs2_log_lock(sdp);
 307         while(atomic_read(&sdp->sd_log_blks_free) <= (blks + reserved_blks)) {
 308                 gfs2_log_unlock(sdp);
 309                 gfs2_ail1_empty(sdp, 0);
 310                 gfs2_log_flush(sdp, NULL);
 311
 312                 if (try++)
 313                         gfs2_ail1_start(sdp, 0);
 314                 gfs2_log_lock(sdp);
 315         }
 316         atomic_sub(blks, &sdp->sd_log_blks_free);
 317         gfs2_log_unlock(sdp);
 318         mutex_unlock(&sdp->sd_log_reserve_mutex);
 319
 320         down_read(&sdp->sd_log_flush_lock);
 321
 322         return 0;
 323 }
 324
 325 /**
 326  * gfs2_log_release - Release a given number of log blocks
 327  * @sdp: The GFS2 superblock
 328  * @blks: The number of blocks
 329  *
 330  */
 331
 332 void gfs2_log_release(struct gfs2_sbd *sdp, unsigned int blks)
 333 {
 334
 335         gfs2_log_lock(sdp);
 336         atomic_add(blks, &sdp->sd_log_blks_free);
 337         gfs2_assert_withdraw(sdp,
 338                              atomic_read(&sdp->sd_log_blks_free) <= sdp->sd_jdesc->jd_blocks);
 339         gfs2_log_unlock(sdp);
 340         up_read(&sdp->sd_log_flush_lock);
 341 }
 342
 343 static u64 log_bmap(struct gfs2_sbd *sdp, unsigned int lbn)
 344 {
 345         struct inode *inode = sdp->sd_jdesc->jd_inode;
 346         int error;
 347         struct buffer_head bh_map = { .b_state = 0, .b_blocknr = 0 };
 348
 349         bh_map.b_size = 1 << inode->i_blkbits;
 350         error = gfs2_block_map(inode, lbn, 0, &bh_map);
 351         if (error || !bh_map.b_blocknr)
 352                 printk(KERN_INFO "error=%d, dbn=%llu lbn=%u", error,
 353                        (unsigned long long)bh_map.b_blocknr, lbn);
 354         gfs2_assert_withdraw(sdp, !error && bh_map.b_blocknr);
 355
 356         return bh_map.b_blocknr;
 357 }
 358
 359 /**
 360  * log_distance - Compute distance between two journal blocks
 361  * @sdp: The GFS2 superblock
 362  * @newer: The most recent journal block of the pair
 363  * @older: The older journal block of the pair
 364  *
 365  *   Compute the distance (in the journal direction) between two
 366  *   blocks in the journal
 367  *
 368  * Returns: the distance in blocks
 369  */
 370
 371 static inline unsigned int log_distance(struct gfs2_sbd *sdp, unsigned int newer,
 372                                         unsigned int older)
 373 {
 374         int dist;
 375
 376         dist = newer - older;
 377         if (dist < 0)
 378                 dist += sdp->sd_jdesc->jd_blocks;
 379
 380         return dist;
 381 }
 382
 383 /**
 384  * calc_reserved - Calculate the number of blocks to reserve when
 385  *                 refunding a transaction's unused buffers.
 386  * @sdp: The GFS2 superblock
 387  *
 388  * This is complex.  We need to reserve room for all our currently used
 389  * metadata buffers (e.g. normal file I/O rewriting file time stamps) and
 390  * all our journaled data buffers for journaled files (e.g. files in the
 391  * meta_fs like rindex, or files for which chattr +j was done.)
 392  * If we don't reserve enough space, gfs2_log_refund and gfs2_log_flush
 393  * will count it as free space (sd_log_blks_free) and corruption will follow.
 394  *
 395  * We can have metadata bufs and jdata bufs in the same journal.  So each
 396  * type gets its own log header, for which we need to reserve a block.
 397  * In fact, each type has the potential for needing more than one header
 398  * in cases where we have more buffers than will fit on a journal page.
 399  * Metadata journal entries take up half the space of journaled buffer entries.
 400  * Thus, metadata entries have buf_limit (502) and journaled buffers have
 401  * databuf_limit (251) before they cause a wrap around.
 402  *
 403  * Also, we need to reserve blocks for revoke journal entries and one for an
 404  * overall header for the lot.
 405  *
 406  * Returns: the number of blocks reserved
 407  */
 408 static unsigned int calc_reserved(struct gfs2_sbd *sdp)
 409 {
 410         unsigned int reserved = 0;
 411         unsigned int mbuf_limit, metabufhdrs_needed;
 412         unsigned int dbuf_limit, databufhdrs_needed;
 413         unsigned int revokes = 0;
 414
 415         mbuf_limit = buf_limit(sdp);
 416         metabufhdrs_needed = (sdp->sd_log_commited_buf +
 417                               (mbuf_limit - 1)) / mbuf_limit;
 418         dbuf_limit = databuf_limit(sdp);
 419         databufhdrs_needed = (sdp->sd_log_commited_databuf +
 420                               (dbuf_limit - 1)) / dbuf_limit;
 421
 422         if (sdp->sd_log_commited_revoke)
 423                 revokes = gfs2_struct2blk(sdp, sdp->sd_log_commited_revoke,
 424                                           sizeof(u64));
 425
 426         reserved = sdp->sd_log_commited_buf + metabufhdrs_needed +
 427                 sdp->sd_log_commited_databuf + databufhdrs_needed +
 428                 revokes;
 429         /* One for the overall header */
 430         if (reserved)
 431                 reserved++;
 432         return reserved;
 433 }
 434
 435 static unsigned int current_tail(struct gfs2_sbd *sdp)
 436 {
 437         struct gfs2_ail *ai;
 438         unsigned int tail;
 439
 440         gfs2_log_lock(sdp);
 441
 442         if (list_empty(&sdp->sd_ail1_list)) {
 443                 tail = sdp->sd_log_head;
 444         } else {
 445                 ai = list_entry(sdp->sd_ail1_list.prev, struct gfs2_ail, ai_list);
 446                 tail = ai->ai_first;
 447         }
 448
 449         gfs2_log_unlock(sdp);
 450
 451         return tail;
 452 }
 453
 454 void gfs2_log_incr_head(struct gfs2_sbd *sdp)
 455 {
 456         if (sdp->sd_log_flush_head == sdp->sd_log_tail)
 457                 BUG_ON(sdp->sd_log_flush_head != sdp->sd_log_head);
 458
 459         if (++sdp->sd_log_flush_head == sdp->sd_jdesc->jd_blocks) {
 460                 sdp->sd_log_flush_head = 0;
 461                 sdp->sd_log_flush_wrapped = 1;
 462         }
 463 }
 464
 465 /**
 466  * gfs2_log_write_endio - End of I/O for a log buffer
 467  * @bh: The buffer head
 468  * @uptodate: I/O Status
 469  *
 470  */
 471
 472 static void gfs2_log_write_endio(struct buffer_head *bh, int uptodate)
 473 {
 474         struct gfs2_sbd *sdp = bh->b_private;
 475         bh->b_private = NULL;
 476
 477         end_buffer_write_sync(bh, uptodate);
 478         if (atomic_dec_and_test(&sdp->sd_log_in_flight))
 479                 wake_up(&sdp->sd_log_flush_wait);
 480 }
 481
 482 /**
 483  * gfs2_log_get_buf - Get and initialize a buffer to use for log control data
 484  * @sdp: The GFS2 superblock
 485  *
 486  * Returns: the buffer_head
 487  */
 488
 489 struct buffer_head *gfs2_log_get_buf(struct gfs2_sbd *sdp)
 490 {
 491         u64 blkno = log_bmap(sdp, sdp->sd_log_flush_head);
 492         struct buffer_head *bh;
 493
 494         bh = sb_getblk(sdp->sd_vfs, blkno);
 495         lock_buffer(bh);
 496         memset(bh->b_data, 0, bh->b_size);
 497         set_buffer_uptodate(bh);
 498         clear_buffer_dirty(bh);
 499         gfs2_log_incr_head(sdp);
 500         atomic_inc(&sdp->sd_log_in_flight);
 501         bh->b_private = sdp;
 502         bh->b_end_io = gfs2_log_write_endio;
 503
 504         return bh;
 505 }
 506
 507 /**
 508  * gfs2_fake_write_endio -
 509  * @bh: The buffer head
 510  * @uptodate: The I/O Status
 511  *
 512  */
 513
 514 static void gfs2_fake_write_endio(struct buffer_head *bh, int uptodate)
 515 {
 516         struct buffer_head *real_bh = bh->b_private;
 517         struct gfs2_bufdata *bd = real_bh->b_private;
 518         struct gfs2_sbd *sdp = bd->bd_gl->gl_sbd;
 519
 520         end_buffer_write_sync(bh, uptodate);
 521         free_buffer_head(bh);
 522         unlock_buffer(real_bh);
 523         brelse(real_bh);
 524         if (atomic_dec_and_test(&sdp->sd_log_in_flight))
 525                 wake_up(&sdp->sd_log_flush_wait);
 526 }
 527
 528 /**
 529  * gfs2_log_fake_buf - Build a fake buffer head to write metadata buffer to log
 530  * @sdp: the filesystem
 531  * @data: the data the buffer_head should point to
 532  *
 533  * Returns: the log buffer descriptor
 534  */
 535
 536 struct buffer_head *gfs2_log_fake_buf(struct gfs2_sbd *sdp,
 537                                       struct buffer_head *real)
 538 {
 539         u64 blkno = log_bmap(sdp, sdp->sd_log_flush_head);
 540         struct buffer_head *bh;
 541
 542         bh = alloc_buffer_head(GFP_NOFS | __GFP_NOFAIL);
 543         atomic_set(&bh->b_count, 1);
 544         bh->b_state = (1 << BH_Mapped) | (1 << BH_Uptodate) | (1 << BH_Lock);
 545         set_bh_page(bh, real->b_page, bh_offset(real));
 546         bh->b_blocknr = blkno;
 547         bh->b_size = sdp->sd_sb.sb_bsize;
 548         bh->b_bdev = sdp->sd_vfs->s_bdev;
 549         bh->b_private = real;
 550         bh->b_end_io = gfs2_fake_write_endio;
 551
 552         gfs2_log_incr_head(sdp);
 553         atomic_inc(&sdp->sd_log_in_flight);
 554
 555         return bh;
 556 }
 557
 558 static void log_pull_tail(struct gfs2_sbd *sdp, unsigned int new_tail)
 559 {
 560         unsigned int dist = log_distance(sdp, new_tail, sdp->sd_log_tail);
 561
 562         ail2_empty(sdp, new_tail);
 563
 564         gfs2_log_lock(sdp);
 565         atomic_add(dist, &sdp->sd_log_blks_free);
 566         gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <= sdp->sd_jdesc->jd_blocks);
 567         gfs2_log_unlock(sdp);
 568
 569         sdp->sd_log_tail = new_tail;
 570 }
 571
 572 /**
 573  * log_write_header - Get and initialize a journal header buffer
 574  * @sdp: The GFS2 superblock
 575  *
 576  * Returns: the initialized log buffer descriptor
 577  */
 578
 579 static void log_write_header(struct gfs2_sbd *sdp, u32 flags, int pull)
 580 {
 581         u64 blkno = log_bmap(sdp, sdp->sd_log_flush_head);
 582         struct buffer_head *bh;
 583         struct gfs2_log_header *lh;
 584         unsigned int tail;
 585         u32 hash;
 586
 587         bh = sb_getblk(sdp->sd_vfs, blkno);
 588         lock_buffer(bh);
 589         memset(bh->b_data, 0, bh->b_size);
 590         set_buffer_uptodate(bh);
 591         clear_buffer_dirty(bh);
 592         unlock_buffer(bh);
 593
 594         gfs2_ail1_empty(sdp, 0);
 595         tail = current_tail(sdp);
 596
 597         lh = (struct gfs2_log_header *)bh->b_data;
 598         memset(lh, 0, sizeof(struct gfs2_log_header));
 599         lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 600         lh->lh_header.mh_type = cpu_to_be32(GFS2_METATYPE_LH);
 601         lh->lh_header.mh_format = cpu_to_be32(GFS2_FORMAT_LH);
 602         lh->lh_sequence = cpu_to_be64(sdp->sd_log_sequence++);
 603         lh->lh_flags = cpu_to_be32(flags);
 604         lh->lh_tail = cpu_to_be32(tail);
 605         lh->lh_blkno = cpu_to_be32(sdp->sd_log_flush_head);
 606         hash = gfs2_disk_hash(bh->b_data, sizeof(struct gfs2_log_header));
 607         lh->lh_hash = cpu_to_be32(hash);
 608
 609         set_buffer_dirty(bh);
 610         if (sync_dirty_buffer(bh))
 611                 gfs2_io_error_bh(sdp, bh);
 612         brelse(bh);
 613
 614         if (sdp->sd_log_tail != tail)
 615                 log_pull_tail(sdp, tail);
 616         else
 617                 gfs2_assert_withdraw(sdp, !pull);
 618
 619         sdp->sd_log_idle = (tail == sdp->sd_log_flush_head);
 620         gfs2_log_incr_head(sdp);
 621 }
 622
 623 static void log_flush_commit(struct gfs2_sbd *sdp)
 624 {
 625         DEFINE_WAIT(wait);
 626
 627         if (atomic_read(&sdp->sd_log_in_flight)) {
 628                 do {
 629                         prepare_to_wait(&sdp->sd_log_flush_wait, &wait,
 630                                         TASK_UNINTERRUPTIBLE);
 631                         if (atomic_read(&sdp->sd_log_in_flight))
 632                                 io_schedule();
 633                 } while(atomic_read(&sdp->sd_log_in_flight));
 634                 finish_wait(&sdp->sd_log_flush_wait, &wait);
 635         }
 636
 637         log_write_header(sdp, 0, 0);
 638 }
 639
 640 static void gfs2_ordered_write(struct gfs2_sbd *sdp)
 641 {
 642         struct gfs2_bufdata *bd;
 643         struct buffer_head *bh;
 644         LIST_HEAD(written);
 645
 646         gfs2_log_lock(sdp);
 647         while (!list_empty(&sdp->sd_log_le_ordered)) {
 648                 bd = list_entry(sdp->sd_log_le_ordered.next, struct gfs2_bufdata, bd_le.le_list);
 649                 list_move(&bd->bd_le.le_list, &written);
 650                 bh = bd->bd_bh;
 651                 if (!buffer_dirty(bh))
 652                         continue;
 653                 get_bh(bh);
 654                 gfs2_log_unlock(sdp);
 655                 lock_buffer(bh);
 656                 if (buffer_mapped(bh) && test_clear_buffer_dirty(bh)) {
 657                         bh->b_end_io = end_buffer_write_sync;
 658                         submit_bh(WRITE, bh);
 659                 } else {
 660                         unlock_buffer(bh);
 661                         brelse(bh);
 662                 }
 663                 gfs2_log_lock(sdp);
 664         }
 665         list_splice(&written, &sdp->sd_log_le_ordered);
 666         gfs2_log_unlock(sdp);
 667 }
 668
 669 static void gfs2_ordered_wait(struct gfs2_sbd *sdp)
 670 {
 671         struct gfs2_bufdata *bd;
 672         struct buffer_head *bh;
 673
 674         gfs2_log_lock(sdp);
 675         while (!list_empty(&sdp->sd_log_le_ordered)) {
 676                 bd = list_entry(sdp->sd_log_le_ordered.prev, struct gfs2_bufdata, bd_le.le_list);
 677                 bh = bd->bd_bh;
 678                 if (buffer_locked(bh)) {
 679                         get_bh(bh);
 680                         gfs2_log_unlock(sdp);
 681                         wait_on_buffer(bh);
 682                         brelse(bh);
 683                         gfs2_log_lock(sdp);
 684                         continue;
 685                 }
 686                 list_del_init(&bd->bd_le.le_list);
 687         }
 688         gfs2_log_unlock(sdp);
 689 }
 690
 691 /**
 692  * gfs2_log_flush - flush incore transaction(s)
 693  * @sdp: the filesystem
 694  * @gl: The glock structure to flush.  If NULL, flush the whole incore log
 695  *
 696  */
 697
 698 void __gfs2_log_flush(struct gfs2_sbd *sdp, struct gfs2_glock *gl)
 699 {
 700         struct gfs2_ail *ai;
 701
 702         down_write(&sdp->sd_log_flush_lock);
 703
 704         /* Log might have been flushed while we waited for the flush lock */
 705         if (gl && !test_bit(GLF_LFLUSH, &gl->gl_flags)) {
 706                 up_write(&sdp->sd_log_flush_lock);
 707                 return;
 708         }
 709
 710         ai = kzalloc(sizeof(struct gfs2_ail), GFP_NOFS | __GFP_NOFAIL);
 711         INIT_LIST_HEAD(&ai->ai_ail1_list);
 712         INIT_LIST_HEAD(&ai->ai_ail2_list);
 713
 714         if (sdp->sd_log_num_buf != sdp->sd_log_commited_buf) {
 715                 printk(KERN_INFO "GFS2: log buf %u %u\n", sdp->sd_log_num_buf,
 716                        sdp->sd_log_commited_buf);
 717                 gfs2_assert_withdraw(sdp, 0);
 718         }
 719         if (sdp->sd_log_num_databuf != sdp->sd_log_commited_databuf) {
 720                 printk(KERN_INFO "GFS2: log databuf %u %u\n",
 721                        sdp->sd_log_num_databuf, sdp->sd_log_commited_databuf);
 722                 gfs2_assert_withdraw(sdp, 0);
 723         }
 724         gfs2_assert_withdraw(sdp,
 725                         sdp->sd_log_num_revoke == sdp->sd_log_commited_revoke);
 726
 727         sdp->sd_log_flush_head = sdp->sd_log_head;
 728         sdp->sd_log_flush_wrapped = 0;
 729         ai->ai_first = sdp->sd_log_flush_head;
 730
 731         gfs2_ordered_write(sdp);
 732         lops_before_commit(sdp);
 733         gfs2_ordered_wait(sdp);
 734
 735         if (sdp->sd_log_head != sdp->sd_log_flush_head)
 736                 log_flush_commit(sdp);
 737         else if (sdp->sd_log_tail != current_tail(sdp) && !sdp->sd_log_idle){
 738                 gfs2_log_lock(sdp);
 739                 atomic_dec(&sdp->sd_log_blks_free); /* Adjust for unreserved buffer */
 740                 gfs2_log_unlock(sdp);
 741                 log_write_header(sdp, 0, PULL);
 742         }
 743         lops_after_commit(sdp, ai);
 744
 745         gfs2_log_lock(sdp);
 746         sdp->sd_log_head = sdp->sd_log_flush_head;
 747         sdp->sd_log_blks_reserved = 0;
 748         sdp->sd_log_commited_buf = 0;
 749         sdp->sd_log_commited_databuf = 0;
 750         sdp->sd_log_commited_revoke = 0;
 751
 752         if (!list_empty(&ai->ai_ail1_list)) {
 753                 list_add(&ai->ai_list, &sdp->sd_ail1_list);
 754                 ai = NULL;
 755         }
 756         gfs2_log_unlock(sdp);
 757
 758         sdp->sd_vfs->s_dirt = 0;
 759         up_write(&sdp->sd_log_flush_lock);
 760
 761         kfree(ai);
 762 }
 763
 764 static void log_refund(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 765 {
 766         unsigned int reserved;
 767         unsigned int old;
 768
 769         gfs2_log_lock(sdp);
 770
 771         sdp->sd_log_commited_buf += tr->tr_num_buf_new - tr->tr_num_buf_rm;
 772         sdp->sd_log_commited_databuf += tr->tr_num_databuf_new -
 773                 tr->tr_num_databuf_rm;
 774         gfs2_assert_withdraw(sdp, (((int)sdp->sd_log_commited_buf) >= 0) ||
 775                              (((int)sdp->sd_log_commited_databuf) >= 0));
 776         sdp->sd_log_commited_revoke += tr->tr_num_revoke - tr->tr_num_revoke_rm;
 777         gfs2_assert_withdraw(sdp, ((int)sdp->sd_log_commited_revoke) >= 0);
 778         reserved = calc_reserved(sdp);
 779         old = atomic_read(&sdp->sd_log_blks_free);
 780         atomic_add(tr->tr_reserved - (reserved - sdp->sd_log_blks_reserved),
 781                    &sdp->sd_log_blks_free);
 782
 783         gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) >= old);
 784         gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <=
 785                              sdp->sd_jdesc->jd_blocks);
 786
 787         sdp->sd_log_blks_reserved = reserved;
 788
 789         gfs2_log_unlock(sdp);
 790 }
 791
 792 /**
 793  * gfs2_log_commit - Commit a transaction to the log
 794  * @sdp: the filesystem
 795  * @tr: the transaction
 796  *
 797  * Returns: errno
 798  */
 799
 800 void gfs2_log_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 801 {
 802         log_refund(sdp, tr);
 803         lops_incore_commit(sdp, tr);
 804
 805         sdp->sd_vfs->s_dirt = 1;
 806         up_read(&sdp->sd_log_flush_lock);
 807
 808         gfs2_log_lock(sdp);
 809         if (sdp->sd_log_num_buf > gfs2_tune_get(sdp, gt_incore_log_blocks))
 810                 wake_up_process(sdp->sd_logd_process);
 811         gfs2_log_unlock(sdp);
 812 }
 813
 814 /**
 815  * gfs2_log_shutdown - write a shutdown header into a journal
 816  * @sdp: the filesystem
 817  *
 818  */
 819
 820 void gfs2_log_shutdown(struct gfs2_sbd *sdp)
 821 {
 822         down_write(&sdp->sd_log_flush_lock);
 823
 824         gfs2_assert_withdraw(sdp, !sdp->sd_log_blks_reserved);
 825         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_buf);
 826         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
 827         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_rg);
 828         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_databuf);
 829         gfs2_assert_withdraw(sdp, list_empty(&sdp->sd_ail1_list));
 830
 831         sdp->sd_log_flush_head = sdp->sd_log_head;
 832         sdp->sd_log_flush_wrapped = 0;
 833
 834         log_write_header(sdp, GFS2_LOG_HEAD_UNMOUNT,
 835                          (sdp->sd_log_tail == current_tail(sdp)) ? 0 : PULL);
 836
 837         gfs2_assert_warn(sdp, atomic_read(&sdp->sd_log_blks_free) == sdp->sd_jdesc->jd_blocks);
 838         gfs2_assert_warn(sdp, sdp->sd_log_head == sdp->sd_log_tail);
 839         gfs2_assert_warn(sdp, list_empty(&sdp->sd_ail2_list));
 840
 841         sdp->sd_log_head = sdp->sd_log_flush_head;
 842         sdp->sd_log_tail = sdp->sd_log_head;
 843
 844         up_write(&sdp->sd_log_flush_lock);
 845 }
 846
 847
 848 /**
 849  * gfs2_meta_syncfs - sync all the buffers in a filesystem
 850  * @sdp: the filesystem
 851  *
 852  */
 853
 854 void gfs2_meta_syncfs(struct gfs2_sbd *sdp)
 855 {
 856         gfs2_log_flush(sdp, NULL);
 857         for (;;) {
 858                 gfs2_ail1_start(sdp, DIO_ALL);
 859                 if (gfs2_ail1_empty(sdp, DIO_ALL))
 860                         break;
 861                 msleep(10);
 862         }
 863 }
 864
 865
 866 /**
 867  * gfs2_logd - Update log tail as Active Items get flushed to in-place blocks
 868  * @sdp: Pointer to GFS2 superblock
 869  *
 870  * Also, periodically check to make sure that we're using the most recent
 871  * journal index.
 872  */
 873
 874 int gfs2_logd(void *data)
 875 {
 876         struct gfs2_sbd *sdp = data;
 877         struct gfs2_holder ji_gh;
 878         unsigned long t;
 879         int need_flush;
 880
 881         while (!kthread_should_stop()) {
 882                 /* Advance the log tail */
 883
 884                 t = sdp->sd_log_flush_time +
 885                     gfs2_tune_get(sdp, gt_log_flush_secs) * HZ;
 886
 887                 gfs2_ail1_empty(sdp, DIO_ALL);
 888                 gfs2_log_lock(sdp);
 889                 need_flush = sdp->sd_log_num_buf > gfs2_tune_get(sdp, gt_incore_log_blocks);
 890                 gfs2_log_unlock(sdp);
 891                 if (need_flush || time_after_eq(jiffies, t)) {
 892                         gfs2_log_flush(sdp, NULL);
 893                         sdp->sd_log_flush_time = jiffies;
 894                 }
 895
 896                 /* Check for latest journal index */
 897
 898                 t = sdp->sd_jindex_refresh_time +
 899                     gfs2_tune_get(sdp, gt_jindex_refresh_secs) * HZ;
 900
 901                 if (time_after_eq(jiffies, t)) {
 902                         if (!gfs2_jindex_hold(sdp, &ji_gh))
 903                                 gfs2_glock_dq_uninit(&ji_gh);
 904                         sdp->sd_jindex_refresh_time = jiffies;
 905                 }
 906
 907                 t = gfs2_tune_get(sdp, gt_logd_secs) * HZ;
 908                 if (freezing(current))
 909                         refrigerator();
 910                 schedule_timeout_interruptible(t);
 911         }
 912
 913         return 0;
 914 }
 915