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