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 error;
 208         struct buffer_head bh_map;
 209
 210         error = gfs2_block_map(sdp->sd_jdesc->jd_inode, lbn, 0, &bh_map, 1);
 211         if (error || !bh_map.b_blocknr)
 212                 printk(KERN_INFO "error=%d, dbn=%llu lbn=%u", error, bh_map.b_blocknr, lbn);
 213         gfs2_assert_withdraw(sdp, !error && bh_map.b_blocknr);
 214
 215         return bh_map.b_blocknr;
 216 }
 217
 218 /**
 219  * log_distance - Compute distance between two journal blocks
 220  * @sdp: The GFS2 superblock
 221  * @newer: The most recent journal block of the pair
 222  * @older: The older journal block of the pair
 223  *
 224  *   Compute the distance (in the journal direction) between two
 225  *   blocks in the journal
 226  *
 227  * Returns: the distance in blocks
 228  */
 229
 230 static inline unsigned int log_distance(struct gfs2_sbd *sdp, unsigned int newer,
 231                                         unsigned int older)
 232 {
 233         int dist;
 234
 235         dist = newer - older;
 236         if (dist < 0)
 237                 dist += sdp->sd_jdesc->jd_blocks;
 238
 239         return dist;
 240 }
 241
 242 static unsigned int current_tail(struct gfs2_sbd *sdp)
 243 {
 244         struct gfs2_ail *ai;
 245         unsigned int tail;
 246
 247         gfs2_log_lock(sdp);
 248
 249         if (list_empty(&sdp->sd_ail1_list)) {
 250                 tail = sdp->sd_log_head;
 251         } else {
 252                 ai = list_entry(sdp->sd_ail1_list.prev, struct gfs2_ail, ai_list);
 253                 tail = ai->ai_first;
 254         }
 255
 256         gfs2_log_unlock(sdp);
 257
 258         return tail;
 259 }
 260
 261 static inline void log_incr_head(struct gfs2_sbd *sdp)
 262 {
 263         if (sdp->sd_log_flush_head == sdp->sd_log_tail)
 264                 gfs2_assert_withdraw(sdp, sdp->sd_log_flush_head == sdp->sd_log_head);
 265
 266         if (++sdp->sd_log_flush_head == sdp->sd_jdesc->jd_blocks) {
 267                 sdp->sd_log_flush_head = 0;
 268                 sdp->sd_log_flush_wrapped = 1;
 269         }
 270 }
 271
 272 /**
 273  * gfs2_log_get_buf - Get and initialize a buffer to use for log control data
 274  * @sdp: The GFS2 superblock
 275  *
 276  * Returns: the buffer_head
 277  */
 278
 279 struct buffer_head *gfs2_log_get_buf(struct gfs2_sbd *sdp)
 280 {
 281         u64 blkno = log_bmap(sdp, sdp->sd_log_flush_head);
 282         struct gfs2_log_buf *lb;
 283         struct buffer_head *bh;
 284
 285         lb = kzalloc(sizeof(struct gfs2_log_buf), GFP_NOFS | __GFP_NOFAIL);
 286         list_add(&lb->lb_list, &sdp->sd_log_flush_list);
 287
 288         bh = lb->lb_bh = sb_getblk(sdp->sd_vfs, blkno);
 289         lock_buffer(bh);
 290         memset(bh->b_data, 0, bh->b_size);
 291         set_buffer_uptodate(bh);
 292         clear_buffer_dirty(bh);
 293         unlock_buffer(bh);
 294
 295         log_incr_head(sdp);
 296
 297         return bh;
 298 }
 299
 300 /**
 301  * gfs2_log_fake_buf - Build a fake buffer head to write metadata buffer to log
 302  * @sdp: the filesystem
 303  * @data: the data the buffer_head should point to
 304  *
 305  * Returns: the log buffer descriptor
 306  */
 307
 308 struct buffer_head *gfs2_log_fake_buf(struct gfs2_sbd *sdp,
 309                                       struct buffer_head *real)
 310 {
 311         u64 blkno = log_bmap(sdp, sdp->sd_log_flush_head);
 312         struct gfs2_log_buf *lb;
 313         struct buffer_head *bh;
 314
 315         lb = kzalloc(sizeof(struct gfs2_log_buf), GFP_NOFS | __GFP_NOFAIL);
 316         list_add(&lb->lb_list, &sdp->sd_log_flush_list);
 317         lb->lb_real = real;
 318
 319         bh = lb->lb_bh = alloc_buffer_head(GFP_NOFS | __GFP_NOFAIL);
 320         atomic_set(&bh->b_count, 1);
 321         bh->b_state = (1 << BH_Mapped) | (1 << BH_Uptodate);
 322         set_bh_page(bh, real->b_page, bh_offset(real));
 323         bh->b_blocknr = blkno;
 324         bh->b_size = sdp->sd_sb.sb_bsize;
 325         bh->b_bdev = sdp->sd_vfs->s_bdev;
 326
 327         log_incr_head(sdp);
 328
 329         return bh;
 330 }
 331
 332 static void log_pull_tail(struct gfs2_sbd *sdp, unsigned int new_tail, int pull)
 333 {
 334         unsigned int dist = log_distance(sdp, new_tail, sdp->sd_log_tail);
 335
 336         ail2_empty(sdp, new_tail);
 337
 338         gfs2_log_lock(sdp);
 339         sdp->sd_log_blks_free += dist - (pull ? 1 : 0);
 340         gfs2_assert_withdraw(sdp, sdp->sd_log_blks_free <= sdp->sd_jdesc->jd_blocks);
 341         gfs2_log_unlock(sdp);
 342
 343         sdp->sd_log_tail = new_tail;
 344 }
 345
 346 /**
 347  * log_write_header - Get and initialize a journal header buffer
 348  * @sdp: The GFS2 superblock
 349  *
 350  * Returns: the initialized log buffer descriptor
 351  */
 352
 353 static void log_write_header(struct gfs2_sbd *sdp, u32 flags, int pull)
 354 {
 355         u64 blkno = log_bmap(sdp, sdp->sd_log_flush_head);
 356         struct buffer_head *bh;
 357         struct gfs2_log_header *lh;
 358         unsigned int tail;
 359         u32 hash;
 360
 361         bh = sb_getblk(sdp->sd_vfs, blkno);
 362         lock_buffer(bh);
 363         memset(bh->b_data, 0, bh->b_size);
 364         set_buffer_uptodate(bh);
 365         clear_buffer_dirty(bh);
 366         unlock_buffer(bh);
 367
 368         gfs2_ail1_empty(sdp, 0);
 369         tail = current_tail(sdp);
 370
 371         lh = (struct gfs2_log_header *)bh->b_data;
 372         memset(lh, 0, sizeof(struct gfs2_log_header));
 373         lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 374         lh->lh_header.mh_type = cpu_to_be32(GFS2_METATYPE_LH);
 375         lh->lh_header.mh_format = cpu_to_be32(GFS2_FORMAT_LH);
 376         lh->lh_sequence = cpu_to_be64(sdp->sd_log_sequence++);
 377         lh->lh_flags = cpu_to_be32(flags);
 378         lh->lh_tail = cpu_to_be32(tail);
 379         lh->lh_blkno = cpu_to_be32(sdp->sd_log_flush_head);
 380         hash = gfs2_disk_hash(bh->b_data, sizeof(struct gfs2_log_header));
 381         lh->lh_hash = cpu_to_be32(hash);
 382
 383         set_buffer_dirty(bh);
 384         if (sync_dirty_buffer(bh))
 385                 gfs2_io_error_bh(sdp, bh);
 386         brelse(bh);
 387
 388         if (sdp->sd_log_tail != tail)
 389                 log_pull_tail(sdp, tail, pull);
 390         else
 391                 gfs2_assert_withdraw(sdp, !pull);
 392
 393         sdp->sd_log_idle = (tail == sdp->sd_log_flush_head);
 394         log_incr_head(sdp);
 395 }
 396
 397 static void log_flush_commit(struct gfs2_sbd *sdp)
 398 {
 399         struct list_head *head = &sdp->sd_log_flush_list;
 400         struct gfs2_log_buf *lb;
 401         struct buffer_head *bh;
 402
 403         while (!list_empty(head)) {
 404                 lb = list_entry(head->next, struct gfs2_log_buf, lb_list);
 405                 list_del(&lb->lb_list);
 406                 bh = lb->lb_bh;
 407
 408                 wait_on_buffer(bh);
 409                 if (!buffer_uptodate(bh))
 410                         gfs2_io_error_bh(sdp, bh);
 411                 if (lb->lb_real) {
 412                         while (atomic_read(&bh->b_count) != 1)  /* Grrrr... */
 413                                 schedule();
 414                         free_buffer_head(bh);
 415                 } else
 416                         brelse(bh);
 417                 kfree(lb);
 418         }
 419
 420         log_write_header(sdp, 0, 0);
 421 }
 422
 423 /**
 424  * gfs2_log_flush - flush incore transaction(s)
 425  * @sdp: the filesystem
 426  * @gl: The glock structure to flush.  If NULL, flush the whole incore log
 427  *
 428  */
 429
 430 void gfs2_log_flush(struct gfs2_sbd *sdp, struct gfs2_glock *gl)
 431 {
 432         struct gfs2_ail *ai;
 433
 434         down_write(&sdp->sd_log_flush_lock);
 435
 436         if (gl) {
 437                 gfs2_log_lock(sdp);
 438                 if (list_empty(&gl->gl_le.le_list)) {
 439                         gfs2_log_unlock(sdp);
 440                         up_write(&sdp->sd_log_flush_lock);
 441                         return;
 442                 }
 443                 gfs2_log_unlock(sdp);
 444         }
 445
 446         ai = kzalloc(sizeof(struct gfs2_ail), GFP_NOFS | __GFP_NOFAIL);
 447         INIT_LIST_HEAD(&ai->ai_ail1_list);
 448         INIT_LIST_HEAD(&ai->ai_ail2_list);
 449
 450         gfs2_assert_withdraw(sdp, sdp->sd_log_num_buf == sdp->sd_log_commited_buf);
 451         gfs2_assert_withdraw(sdp,
 452                         sdp->sd_log_num_revoke == sdp->sd_log_commited_revoke);
 453
 454         sdp->sd_log_flush_head = sdp->sd_log_head;
 455         sdp->sd_log_flush_wrapped = 0;
 456         ai->ai_first = sdp->sd_log_flush_head;
 457
 458         lops_before_commit(sdp);
 459         if (!list_empty(&sdp->sd_log_flush_list))
 460                 log_flush_commit(sdp);
 461         else if (sdp->sd_log_tail != current_tail(sdp) && !sdp->sd_log_idle)
 462                 log_write_header(sdp, 0, PULL);
 463         lops_after_commit(sdp, ai);
 464         sdp->sd_log_head = sdp->sd_log_flush_head;
 465
 466         sdp->sd_log_blks_free -= sdp->sd_log_num_hdrs;
 467
 468         sdp->sd_log_blks_reserved = 0;
 469         sdp->sd_log_commited_buf = 0;
 470         sdp->sd_log_num_hdrs = 0;
 471         sdp->sd_log_commited_revoke = 0;
 472
 473         gfs2_log_lock(sdp);
 474         if (!list_empty(&ai->ai_ail1_list)) {
 475                 list_add(&ai->ai_list, &sdp->sd_ail1_list);
 476                 ai = NULL;
 477         }
 478         gfs2_log_unlock(sdp);
 479
 480         sdp->sd_vfs->s_dirt = 0;
 481         up_write(&sdp->sd_log_flush_lock);
 482
 483         kfree(ai);
 484 }
 485
 486 static void log_refund(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 487 {
 488         unsigned int reserved = 0;
 489         unsigned int old;
 490
 491         gfs2_log_lock(sdp);
 492
 493         sdp->sd_log_commited_buf += tr->tr_num_buf_new - tr->tr_num_buf_rm;
 494         gfs2_assert_withdraw(sdp, ((int)sdp->sd_log_commited_buf) >= 0);
 495         sdp->sd_log_commited_revoke += tr->tr_num_revoke - tr->tr_num_revoke_rm;
 496         gfs2_assert_withdraw(sdp, ((int)sdp->sd_log_commited_revoke) >= 0);
 497
 498         if (sdp->sd_log_commited_buf)
 499                 reserved += sdp->sd_log_commited_buf;
 500         if (sdp->sd_log_commited_revoke)
 501                 reserved += gfs2_struct2blk(sdp, sdp->sd_log_commited_revoke,
 502                                             sizeof(u64));
 503         if (reserved)
 504                 reserved++;
 505
 506         old = sdp->sd_log_blks_free;
 507         sdp->sd_log_blks_free += tr->tr_reserved -
 508                                  (reserved - sdp->sd_log_blks_reserved);
 509
 510         gfs2_assert_withdraw(sdp, sdp->sd_log_blks_free >= old);
 511         gfs2_assert_withdraw(sdp,
 512                              sdp->sd_log_blks_free <= sdp->sd_jdesc->jd_blocks +
 513                              sdp->sd_log_num_hdrs);
 514
 515         sdp->sd_log_blks_reserved = reserved;
 516
 517         gfs2_log_unlock(sdp);
 518 }
 519
 520 /**
 521  * gfs2_log_commit - Commit a transaction to the log
 522  * @sdp: the filesystem
 523  * @tr: the transaction
 524  *
 525  * Returns: errno
 526  */
 527
 528 void gfs2_log_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 529 {
 530         log_refund(sdp, tr);
 531         lops_incore_commit(sdp, tr);
 532
 533         sdp->sd_vfs->s_dirt = 1;
 534         up_read(&sdp->sd_log_flush_lock);
 535
 536         gfs2_log_lock(sdp);
 537         if (sdp->sd_log_num_buf > gfs2_tune_get(sdp, gt_incore_log_blocks)) {
 538                 gfs2_log_unlock(sdp);
 539                 gfs2_log_flush(sdp, NULL);
 540         } else {
 541                 gfs2_log_unlock(sdp);
 542         }
 543 }
 544
 545 /**
 546  * gfs2_log_shutdown - write a shutdown header into a journal
 547  * @sdp: the filesystem
 548  *
 549  */
 550
 551 void gfs2_log_shutdown(struct gfs2_sbd *sdp)
 552 {
 553         down_write(&sdp->sd_log_flush_lock);
 554
 555         gfs2_assert_withdraw(sdp, !sdp->sd_log_blks_reserved);
 556         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_gl);
 557         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_buf);
 558         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_jdata);
 559         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
 560         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_rg);
 561         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_databuf);
 562         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_hdrs);
 563         gfs2_assert_withdraw(sdp, list_empty(&sdp->sd_ail1_list));
 564
 565         sdp->sd_log_flush_head = sdp->sd_log_head;
 566         sdp->sd_log_flush_wrapped = 0;
 567
 568         log_write_header(sdp, GFS2_LOG_HEAD_UNMOUNT, 0);
 569
 570         gfs2_assert_warn(sdp, sdp->sd_log_blks_free == sdp->sd_jdesc->jd_blocks);
 571         gfs2_assert_warn(sdp, sdp->sd_log_head == sdp->sd_log_tail);
 572         gfs2_assert_warn(sdp, list_empty(&sdp->sd_ail2_list));
 573
 574         sdp->sd_log_head = sdp->sd_log_flush_head;
 575         sdp->sd_log_tail = sdp->sd_log_head;
 576
 577         up_write(&sdp->sd_log_flush_lock);
 578 }
 579