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
  18 #include "gfs2.h"
  19 #include "lm_interface.h"
  20 #include "incore.h"
  21 #include "bmap.h"
  22 #include "glock.h"
  23 #include "log.h"
  24 #include "lops.h"
  25 #include "meta_io.h"
  26 #include "util.h"
  27 #include "dir.h"
  28
  29 #define PULL 1
  30
  31 /**
  32  * gfs2_struct2blk - compute stuff
  33  * @sdp: the filesystem
  34  * @nstruct: the number of structures
  35  * @ssize: the size of the structures
  36  *
  37  * Compute the number of log descriptor blocks needed to hold a certain number
  38  * of structures of a certain size.
  39  *
  40  * Returns: the number of blocks needed (minimum is always 1)
  41  */
  42
  43 unsigned int gfs2_struct2blk(struct gfs2_sbd *sdp, unsigned int nstruct,
  44                              unsigned int ssize)
  45 {
  46         unsigned int blks;
  47         unsigned int first, second;
  48
  49         blks = 1;
  50         first = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) / ssize;
  51
  52         if (nstruct > first) {
  53                 second = (sdp->sd_sb.sb_bsize -
  54                           sizeof(struct gfs2_meta_header)) / ssize;
  55                 blks += DIV_ROUND_UP(nstruct - first, second);
  56         }
  57
  58         return blks;
  59 }
  60
  61 void gfs2_ail1_start(struct gfs2_sbd *sdp, int flags)
  62 {
  63         struct list_head *head = &sdp->sd_ail1_list;
  64         u64 sync_gen;
  65         struct list_head *first, *tmp;
  66         struct gfs2_ail *first_ai, *ai;
  67
  68         gfs2_log_lock(sdp);
  69         if (list_empty(head)) {
  70                 gfs2_log_unlock(sdp);
  71                 return;
  72         }
  73         sync_gen = sdp->sd_ail_sync_gen++;
  74
  75         first = head->prev;
  76         first_ai = list_entry(first, struct gfs2_ail, ai_list);
  77         first_ai->ai_sync_gen = sync_gen;
  78         gfs2_ail1_start_one(sdp, first_ai);
  79
  80         if (flags & DIO_ALL)
  81                 first = NULL;
  82
  83         for (;;) {
  84                 if (first && (head->prev != first ||
  85                               gfs2_ail1_empty_one(sdp, first_ai, 0)))
  86                         break;
  87
  88                 for (tmp = head->prev; tmp != head; tmp = tmp->prev) {
  89                         ai = list_entry(tmp, struct gfs2_ail, ai_list);
  90                         if (ai->ai_sync_gen >= sync_gen)
  91                                 continue;
  92                         ai->ai_sync_gen = sync_gen;
  93                         gfs2_ail1_start_one(sdp, ai);
  94                         break;
  95                 }
  96
  97                 if (tmp == head)
  98                         break;
  99         }
 100
 101         gfs2_log_unlock(sdp);
 102 }
 103
 104 int gfs2_ail1_empty(struct gfs2_sbd *sdp, int flags)
 105 {
 106         struct gfs2_ail *ai, *s;
 107         int ret;
 108
 109         gfs2_log_lock(sdp);
 110
 111         list_for_each_entry_safe_reverse(ai, s, &sdp->sd_ail1_list, ai_list) {
 112                 if (gfs2_ail1_empty_one(sdp, ai, flags))
 113                         list_move(&ai->ai_list, &sdp->sd_ail2_list);
 114                 else if (!(flags & DIO_ALL))
 115                         break;
 116         }
 117
 118         ret = list_empty(&sdp->sd_ail1_list);
 119
 120         gfs2_log_unlock(sdp);
 121
 122         return ret;
 123 }
 124
 125 static void ail2_empty(struct gfs2_sbd *sdp, unsigned int new_tail)
 126 {
 127         struct gfs2_ail *ai, *safe;
 128         unsigned int old_tail = sdp->sd_log_tail;
 129         int wrap = (new_tail < old_tail);
 130         int a, b, rm;
 131
 132         gfs2_log_lock(sdp);
 133
 134         list_for_each_entry_safe(ai, safe, &sdp->sd_ail2_list, ai_list) {
 135                 a = (old_tail <= ai->ai_first);
 136                 b = (ai->ai_first < new_tail);
 137                 rm = (wrap) ? (a || b) : (a && b);
 138                 if (!rm)
 139                         continue;
 140
 141                 gfs2_ail2_empty_one(sdp, ai);
 142                 list_del(&ai->ai_list);
 143                 gfs2_assert_warn(sdp, list_empty(&ai->ai_ail1_list));
 144                 gfs2_assert_warn(sdp, list_empty(&ai->ai_ail2_list));
 145                 kfree(ai);
 146         }
 147
 148         gfs2_log_unlock(sdp);
 149 }
 150
 151 /**
 152  * gfs2_log_reserve - Make a log reservation
 153  * @sdp: The GFS2 superblock
 154  * @blks: The number of blocks to reserve
 155  *
 156  * Returns: errno
 157  */
 158
 159 int gfs2_log_reserve(struct gfs2_sbd *sdp, unsigned int blks)
 160 {
 161         unsigned int try = 0;
 162
 163         if (gfs2_assert_warn(sdp, blks) ||
 164             gfs2_assert_warn(sdp, blks <= sdp->sd_jdesc->jd_blocks))
 165                 return -EINVAL;
 166
 167         mutex_lock(&sdp->sd_log_reserve_mutex);
 168         gfs2_log_lock(sdp);
 169         while(sdp->sd_log_blks_free <= blks) {
 170                 gfs2_log_unlock(sdp);
 171                 gfs2_ail1_empty(sdp, 0);
 172                 gfs2_log_flush(sdp, NULL);
 173
 174                 if (try++)
 175                         gfs2_ail1_start(sdp, 0);
 176                 gfs2_log_lock(sdp);
 177         }
 178         sdp->sd_log_blks_free -= blks;
 179         gfs2_log_unlock(sdp);
 180         mutex_unlock(&sdp->sd_log_reserve_mutex);
 181
 182         down_read(&sdp->sd_log_flush_lock);
 183
 184         return 0;
 185 }
 186
 187 /**
 188  * gfs2_log_release - Release a given number of log blocks
 189  * @sdp: The GFS2 superblock
 190  * @blks: The number of blocks
 191  *
 192  */
 193
 194 void gfs2_log_release(struct gfs2_sbd *sdp, unsigned int blks)
 195 {
 196
 197         gfs2_log_lock(sdp);
 198         sdp->sd_log_blks_free += blks;
 199         gfs2_assert_withdraw(sdp,
 200                              sdp->sd_log_blks_free <= sdp->sd_jdesc->jd_blocks);
 201         gfs2_log_unlock(sdp);
 202         up_read(&sdp->sd_log_flush_lock);
 203 }
 204
 205 static u64 log_bmap(struct gfs2_sbd *sdp, unsigned int lbn)
 206 {
 207         int new = 0;
 208         u64 dbn;
 209         int error;
 210         int bdy;
 211
 212         error = gfs2_block_map(sdp->sd_jdesc->jd_inode, lbn, &new, &dbn, &bdy);
 213         if (error || !dbn)
 214                 printk(KERN_INFO "error=%d, dbn=%llu lbn=%u", error, (unsigned long long)dbn, lbn);
 215         gfs2_assert_withdraw(sdp, !error && dbn);
 216
 217         return dbn;
 218 }
 219
 220 /**
 221  * log_distance - Compute distance between two journal blocks
 222  * @sdp: The GFS2 superblock
 223  * @newer: The most recent journal block of the pair
 224  * @older: The older journal block of the pair
 225  *
 226  *   Compute the distance (in the journal direction) between two
 227  *   blocks in the journal
 228  *
 229  * Returns: the distance in blocks
 230  */
 231
 232 static inline unsigned int log_distance(struct gfs2_sbd *sdp, unsigned int newer,
 233                                         unsigned int older)
 234 {
 235         int dist;
 236
 237         dist = newer - older;
 238         if (dist < 0)
 239                 dist += sdp->sd_jdesc->jd_blocks;
 240
 241         return dist;
 242 }
 243
 244 static unsigned int current_tail(struct gfs2_sbd *sdp)
 245 {
 246         struct gfs2_ail *ai;
 247         unsigned int tail;
 248
 249         gfs2_log_lock(sdp);
 250
 251         if (list_empty(&sdp->sd_ail1_list)) {
 252                 tail = sdp->sd_log_head;
 253         } else {
 254                 ai = list_entry(sdp->sd_ail1_list.prev, struct gfs2_ail, ai_list);
 255                 tail = ai->ai_first;
 256         }
 257
 258         gfs2_log_unlock(sdp);
 259
 260         return tail;
 261 }
 262
 263 static inline void log_incr_head(struct gfs2_sbd *sdp)
 264 {
 265         if (sdp->sd_log_flush_head == sdp->sd_log_tail)
 266                 gfs2_assert_withdraw(sdp, sdp->sd_log_flush_head == sdp->sd_log_head);
 267
 268         if (++sdp->sd_log_flush_head == sdp->sd_jdesc->jd_blocks) {
 269                 sdp->sd_log_flush_head = 0;
 270                 sdp->sd_log_flush_wrapped = 1;
 271         }
 272 }
 273
 274 /**
 275  * gfs2_log_get_buf - Get and initialize a buffer to use for log control data
 276  * @sdp: The GFS2 superblock
 277  *
 278  * Returns: the buffer_head
 279  */
 280
 281 struct buffer_head *gfs2_log_get_buf(struct gfs2_sbd *sdp)
 282 {
 283         u64 blkno = log_bmap(sdp, sdp->sd_log_flush_head);
 284         struct gfs2_log_buf *lb;
 285         struct buffer_head *bh;
 286
 287         lb = kzalloc(sizeof(struct gfs2_log_buf), GFP_NOFS | __GFP_NOFAIL);
 288         list_add(&lb->lb_list, &sdp->sd_log_flush_list);
 289
 290         bh = lb->lb_bh = sb_getblk(sdp->sd_vfs, blkno);
 291         lock_buffer(bh);
 292         memset(bh->b_data, 0, bh->b_size);
 293         set_buffer_uptodate(bh);
 294         clear_buffer_dirty(bh);
 295         unlock_buffer(bh);
 296
 297         log_incr_head(sdp);
 298
 299         return bh;
 300 }
 301
 302 /**
 303  * gfs2_log_fake_buf - Build a fake buffer head to write metadata buffer to log
 304  * @sdp: the filesystem
 305  * @data: the data the buffer_head should point to
 306  *
 307  * Returns: the log buffer descriptor
 308  */
 309
 310 struct buffer_head *gfs2_log_fake_buf(struct gfs2_sbd *sdp,
 311                                       struct buffer_head *real)
 312 {
 313         u64 blkno = log_bmap(sdp, sdp->sd_log_flush_head);
 314         struct gfs2_log_buf *lb;
 315         struct buffer_head *bh;
 316
 317         lb = kzalloc(sizeof(struct gfs2_log_buf), GFP_NOFS | __GFP_NOFAIL);
 318         list_add(&lb->lb_list, &sdp->sd_log_flush_list);
 319         lb->lb_real = real;
 320
 321         bh = lb->lb_bh = alloc_buffer_head(GFP_NOFS | __GFP_NOFAIL);
 322         atomic_set(&bh->b_count, 1);
 323         bh->b_state = (1 << BH_Mapped) | (1 << BH_Uptodate);
 324         set_bh_page(bh, real->b_page, bh_offset(real));
 325         bh->b_blocknr = blkno;
 326         bh->b_size = sdp->sd_sb.sb_bsize;
 327         bh->b_bdev = sdp->sd_vfs->s_bdev;
 328
 329         log_incr_head(sdp);
 330
 331         return bh;
 332 }
 333
 334 static void log_pull_tail(struct gfs2_sbd *sdp, unsigned int new_tail, int pull)
 335 {
 336         unsigned int dist = log_distance(sdp, new_tail, sdp->sd_log_tail);
 337
 338         ail2_empty(sdp, new_tail);
 339
 340         gfs2_log_lock(sdp);
 341         sdp->sd_log_blks_free += dist - (pull ? 1 : 0);
 342         gfs2_assert_withdraw(sdp, sdp->sd_log_blks_free <= sdp->sd_jdesc->jd_blocks);
 343         gfs2_log_unlock(sdp);
 344
 345         sdp->sd_log_tail = new_tail;
 346 }
 347
 348 /**
 349  * log_write_header - Get and initialize a journal header buffer
 350  * @sdp: The GFS2 superblock
 351  *
 352  * Returns: the initialized log buffer descriptor
 353  */
 354
 355 static void log_write_header(struct gfs2_sbd *sdp, u32 flags, int pull)
 356 {
 357         u64 blkno = log_bmap(sdp, sdp->sd_log_flush_head);
 358         struct buffer_head *bh;
 359         struct gfs2_log_header *lh;
 360         unsigned int tail;
 361         u32 hash;
 362
 363         bh = sb_getblk(sdp->sd_vfs, blkno);
 364         lock_buffer(bh);
 365         memset(bh->b_data, 0, bh->b_size);
 366         set_buffer_uptodate(bh);
 367         clear_buffer_dirty(bh);
 368         unlock_buffer(bh);
 369
 370         gfs2_ail1_empty(sdp, 0);
 371         tail = current_tail(sdp);
 372
 373         lh = (struct gfs2_log_header *)bh->b_data;
 374         memset(lh, 0, sizeof(struct gfs2_log_header));
 375         lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 376         lh->lh_header.mh_type = cpu_to_be32(GFS2_METATYPE_LH);
 377         lh->lh_header.mh_format = cpu_to_be32(GFS2_FORMAT_LH);
 378         lh->lh_sequence = cpu_to_be64(sdp->sd_log_sequence++);
 379         lh->lh_flags = cpu_to_be32(flags);
 380         lh->lh_tail = cpu_to_be32(tail);
 381         lh->lh_blkno = cpu_to_be32(sdp->sd_log_flush_head);
 382         hash = gfs2_disk_hash(bh->b_data, sizeof(struct gfs2_log_header));
 383         lh->lh_hash = cpu_to_be32(hash);
 384
 385         set_buffer_dirty(bh);
 386         if (sync_dirty_buffer(bh))
 387                 gfs2_io_error_bh(sdp, bh);
 388         brelse(bh);
 389
 390         if (sdp->sd_log_tail != tail)
 391                 log_pull_tail(sdp, tail, pull);
 392         else
 393                 gfs2_assert_withdraw(sdp, !pull);
 394
 395         sdp->sd_log_idle = (tail == sdp->sd_log_flush_head);
 396         log_incr_head(sdp);
 397 }
 398
 399 static void log_flush_commit(struct gfs2_sbd *sdp)
 400 {
 401         struct list_head *head = &sdp->sd_log_flush_list;
 402         struct gfs2_log_buf *lb;
 403         struct buffer_head *bh;
 404
 405         while (!list_empty(head)) {
 406                 lb = list_entry(head->next, struct gfs2_log_buf, lb_list);
 407                 list_del(&lb->lb_list);
 408                 bh = lb->lb_bh;
 409
 410                 wait_on_buffer(bh);
 411                 if (!buffer_uptodate(bh))
 412                         gfs2_io_error_bh(sdp, bh);
 413                 if (lb->lb_real) {
 414                         while (atomic_read(&bh->b_count) != 1)  /* Grrrr... */
 415                                 schedule();
 416                         free_buffer_head(bh);
 417                 } else
 418                         brelse(bh);
 419                 kfree(lb);
 420         }
 421
 422         log_write_header(sdp, 0, 0);
 423 }
 424
 425 /**
 426  * gfs2_log_flush - flush incore transaction(s)
 427  * @sdp: the filesystem
 428  * @gl: The glock structure to flush.  If NULL, flush the whole incore log
 429  *
 430  */
 431
 432 void gfs2_log_flush(struct gfs2_sbd *sdp, struct gfs2_glock *gl)
 433 {
 434         struct gfs2_ail *ai;
 435
 436         down_write(&sdp->sd_log_flush_lock);
 437
 438         if (gl) {
 439                 gfs2_log_lock(sdp);
 440                 if (list_empty(&gl->gl_le.le_list)) {
 441                         gfs2_log_unlock(sdp);
 442                         up_write(&sdp->sd_log_flush_lock);
 443                         return;
 444                 }
 445                 gfs2_log_unlock(sdp);
 446         }
 447
 448         ai = kzalloc(sizeof(struct gfs2_ail), GFP_NOFS | __GFP_NOFAIL);
 449         INIT_LIST_HEAD(&ai->ai_ail1_list);
 450         INIT_LIST_HEAD(&ai->ai_ail2_list);
 451
 452         gfs2_assert_withdraw(sdp, sdp->sd_log_num_buf == sdp->sd_log_commited_buf);
 453         gfs2_assert_withdraw(sdp,
 454                         sdp->sd_log_num_revoke == sdp->sd_log_commited_revoke);
 455
 456         sdp->sd_log_flush_head = sdp->sd_log_head;
 457         sdp->sd_log_flush_wrapped = 0;
 458         ai->ai_first = sdp->sd_log_flush_head;
 459
 460         lops_before_commit(sdp);
 461         if (!list_empty(&sdp->sd_log_flush_list))
 462                 log_flush_commit(sdp);
 463         else if (sdp->sd_log_tail != current_tail(sdp) && !sdp->sd_log_idle)
 464                 log_write_header(sdp, 0, PULL);
 465         lops_after_commit(sdp, ai);
 466         sdp->sd_log_head = sdp->sd_log_flush_head;
 467
 468         sdp->sd_log_blks_free -= sdp->sd_log_num_hdrs;
 469
 470         sdp->sd_log_blks_reserved = 0;
 471         sdp->sd_log_commited_buf = 0;
 472         sdp->sd_log_num_hdrs = 0;
 473         sdp->sd_log_commited_revoke = 0;
 474
 475         gfs2_log_lock(sdp);
 476         if (!list_empty(&ai->ai_ail1_list)) {
 477                 list_add(&ai->ai_list, &sdp->sd_ail1_list);
 478                 ai = NULL;
 479         }
 480         gfs2_log_unlock(sdp);
 481
 482         sdp->sd_vfs->s_dirt = 0;
 483         up_write(&sdp->sd_log_flush_lock);
 484
 485         kfree(ai);
 486 }
 487
 488 static void log_refund(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 489 {
 490         unsigned int reserved = 0;
 491         unsigned int old;
 492
 493         gfs2_log_lock(sdp);
 494
 495         sdp->sd_log_commited_buf += tr->tr_num_buf_new - tr->tr_num_buf_rm;
 496         gfs2_assert_withdraw(sdp, ((int)sdp->sd_log_commited_buf) >= 0);
 497         sdp->sd_log_commited_revoke += tr->tr_num_revoke - tr->tr_num_revoke_rm;
 498         gfs2_assert_withdraw(sdp, ((int)sdp->sd_log_commited_revoke) >= 0);
 499
 500         if (sdp->sd_log_commited_buf)
 501                 reserved += sdp->sd_log_commited_buf;
 502         if (sdp->sd_log_commited_revoke)
 503                 reserved += gfs2_struct2blk(sdp, sdp->sd_log_commited_revoke,
 504                                             sizeof(u64));
 505         if (reserved)
 506                 reserved++;
 507
 508         old = sdp->sd_log_blks_free;
 509         sdp->sd_log_blks_free += tr->tr_reserved -
 510                                  (reserved - sdp->sd_log_blks_reserved);
 511
 512         gfs2_assert_withdraw(sdp, sdp->sd_log_blks_free >= old);
 513         gfs2_assert_withdraw(sdp,
 514                              sdp->sd_log_blks_free <= sdp->sd_jdesc->jd_blocks +
 515                              sdp->sd_log_num_hdrs);
 516
 517         sdp->sd_log_blks_reserved = reserved;
 518
 519         gfs2_log_unlock(sdp);
 520 }
 521
 522 /**
 523  * gfs2_log_commit - Commit a transaction to the log
 524  * @sdp: the filesystem
 525  * @tr: the transaction
 526  *
 527  * Returns: errno
 528  */
 529
 530 void gfs2_log_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 531 {
 532         log_refund(sdp, tr);
 533         lops_incore_commit(sdp, tr);
 534
 535         sdp->sd_vfs->s_dirt = 1;
 536         up_read(&sdp->sd_log_flush_lock);
 537
 538         gfs2_log_lock(sdp);
 539         if (sdp->sd_log_num_buf > gfs2_tune_get(sdp, gt_incore_log_blocks)) {
 540                 gfs2_log_unlock(sdp);
 541                 gfs2_log_flush(sdp, NULL);
 542         } else {
 543                 gfs2_log_unlock(sdp);
 544         }
 545 }
 546
 547 /**
 548  * gfs2_log_shutdown - write a shutdown header into a journal
 549  * @sdp: the filesystem
 550  *
 551  */
 552
 553 void gfs2_log_shutdown(struct gfs2_sbd *sdp)
 554 {
 555         down_write(&sdp->sd_log_flush_lock);
 556
 557         gfs2_assert_withdraw(sdp, !sdp->sd_log_blks_reserved);
 558         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_gl);
 559         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_buf);
 560         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_jdata);
 561         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
 562         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_rg);
 563         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_databuf);
 564         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_hdrs);
 565         gfs2_assert_withdraw(sdp, list_empty(&sdp->sd_ail1_list));
 566
 567         sdp->sd_log_flush_head = sdp->sd_log_head;
 568         sdp->sd_log_flush_wrapped = 0;
 569
 570         log_write_header(sdp, GFS2_LOG_HEAD_UNMOUNT, 0);
 571
 572         gfs2_assert_warn(sdp, sdp->sd_log_blks_free == sdp->sd_jdesc->jd_blocks);
 573         gfs2_assert_warn(sdp, sdp->sd_log_head == sdp->sd_log_tail);
 574         gfs2_assert_warn(sdp, list_empty(&sdp->sd_ail2_list));
 575
 576         sdp->sd_log_head = sdp->sd_log_flush_head;
 577         sdp->sd_log_tail = sdp->sd_log_head;
 578
 579         up_write(&sdp->sd_log_flush_lock);
 580 }
 581