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 v.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)) /
  51                 ssize;
  52
  53         if (nstruct > first) {
  54                 second = (sdp->sd_sb.sb_bsize -
  55                           sizeof(struct gfs2_meta_header)) / ssize;
  56                 blks += DIV_ROUND_UP(nstruct - first, second);
  57         }
  58
  59         return blks;
  60 }
  61
  62 void gfs2_ail1_start(struct gfs2_sbd *sdp, int flags)
  63 {
  64         struct list_head *head = &sdp->sd_ail1_list;
  65         uint64_t sync_gen;
  66         struct list_head *first, *tmp;
  67         struct gfs2_ail *first_ai, *ai;
  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);
  80
  81         if (flags & DIO_ALL)
  82                 first = NULL;
  83
  84         for (;;) {
  85                 if (first && (head->prev != first ||
  86                               gfs2_ail1_empty_one(sdp, first_ai, 0)))
  87                         break;
  88
  89                 for (tmp = head->prev; tmp != head; tmp = tmp->prev) {
  90                         ai = list_entry(tmp, struct gfs2_ail, 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);
  95                         break;
  96                 }
  97
  98                 if (tmp == head)
  99                         break;
 100         }
 101
 102         gfs2_log_unlock(sdp);
 103 }
 104
 105 int gfs2_ail1_empty(struct gfs2_sbd *sdp, int flags)
 106 {
 107         struct gfs2_ail *ai, *s;
 108         int ret;
 109
 110         gfs2_log_lock(sdp);
 111
 112         list_for_each_entry_safe_reverse(ai, s, &sdp->sd_ail1_list, ai_list) {
 113                 if (gfs2_ail1_empty_one(sdp, ai, flags))
 114                         list_move(&ai->ai_list, &sdp->sd_ail2_list);
 115                 else if (!(flags & DIO_ALL))
 116                         break;
 117         }
 118
 119         ret = list_empty(&sdp->sd_ail1_list);
 120
 121         gfs2_log_unlock(sdp);
 122
 123         return ret;
 124 }
 125
 126 static void ail2_empty(struct gfs2_sbd *sdp, unsigned int new_tail)
 127 {
 128         struct gfs2_ail *ai, *safe;
 129         unsigned int old_tail = sdp->sd_log_tail;
 130         int wrap = (new_tail < old_tail);
 131         int a, b, rm;
 132
 133         gfs2_log_lock(sdp);
 134
 135         list_for_each_entry_safe(ai, safe, &sdp->sd_ail2_list, ai_list) {
 136                 a = (old_tail <= ai->ai_first);
 137                 b = (ai->ai_first < new_tail);
 138                 rm = (wrap) ? (a || b) : (a && b);
 139                 if (!rm)
 140                         continue;
 141
 142                 gfs2_ail2_empty_one(sdp, ai);
 143                 list_del(&ai->ai_list);
 144                 gfs2_assert_warn(sdp, list_empty(&ai->ai_ail1_list));
 145                 gfs2_assert_warn(sdp, list_empty(&ai->ai_ail2_list));
 146                 kfree(ai);
 147         }
 148
 149         gfs2_log_unlock(sdp);
 150 }
 151
 152 /**
 153  * gfs2_log_reserve - Make a log reservation
 154  * @sdp: The GFS2 superblock
 155  * @blks: The number of blocks to reserve
 156  *
 157  * Returns: errno
 158  */
 159
 160 int gfs2_log_reserve(struct gfs2_sbd *sdp, unsigned int blks)
 161 {
 162         unsigned int try = 0;
 163
 164         if (gfs2_assert_warn(sdp, blks) ||
 165             gfs2_assert_warn(sdp, blks <= sdp->sd_jdesc->jd_blocks))
 166                 return -EINVAL;
 167
 168         mutex_lock(&sdp->sd_log_reserve_mutex);
 169         gfs2_log_lock(sdp);
 170         while(sdp->sd_log_blks_free <= blks) {
 171                 gfs2_log_unlock(sdp);
 172                 gfs2_ail1_empty(sdp, 0);
 173                 gfs2_log_flush(sdp, NULL);
 174
 175                 if (try++)
 176                         gfs2_ail1_start(sdp, 0);
 177                 gfs2_log_lock(sdp);
 178         }
 179         sdp->sd_log_blks_free -= blks;
 180         /* printk(KERN_INFO "reserved %u blocks (%u left)\n", blks, sdp->sd_log_blks_free); */
 181         gfs2_log_unlock(sdp);
 182         mutex_unlock(&sdp->sd_log_reserve_mutex);
 183
 184         down_read(&sdp->sd_log_flush_lock);
 185
 186         return 0;
 187 }
 188
 189 /**
 190  * gfs2_log_release - Release a given number of log blocks
 191  * @sdp: The GFS2 superblock
 192  * @blks: The number of blocks
 193  *
 194  */
 195
 196 void gfs2_log_release(struct gfs2_sbd *sdp, unsigned int blks)
 197 {
 198
 199         gfs2_log_lock(sdp);
 200         sdp->sd_log_blks_free += blks;
 201         /* printk(KERN_INFO "released %u blocks (%u left)\n", blks, sdp->sd_log_blks_free); */
 202         gfs2_assert_withdraw(sdp,
 203                              sdp->sd_log_blks_free <= sdp->sd_jdesc->jd_blocks);
 204         gfs2_log_unlock(sdp);
 205         up_read(&sdp->sd_log_flush_lock);
 206 }
 207
 208 static uint64_t log_bmap(struct gfs2_sbd *sdp, unsigned int lbn)
 209 {
 210         int new = 0;
 211         uint64_t dbn;
 212         int error;
 213         int bdy;
 214
 215         error = gfs2_block_map(sdp->sd_jdesc->jd_inode, lbn, &new, &dbn, &bdy);
 216         if (!(!error && dbn)) {
 217                 printk(KERN_INFO "error=%d, dbn=%llu lbn=%u", error, dbn, lbn);
 218         }
 219         gfs2_assert_withdraw(sdp, !error && dbn);
 220
 221         return dbn;
 222 }
 223
 224 /**
 225  * log_distance - Compute distance between two journal blocks
 226  * @sdp: The GFS2 superblock
 227  * @newer: The most recent journal block of the pair
 228  * @older: The older journal block of the pair
 229  *
 230  *   Compute the distance (in the journal direction) between two
 231  *   blocks in the journal
 232  *
 233  * Returns: the distance in blocks
 234  */
 235
 236 static inline unsigned int log_distance(struct gfs2_sbd *sdp,
 237                                         unsigned int newer,
 238                                         unsigned int older)
 239 {
 240         int dist;
 241
 242         dist = newer - older;
 243         if (dist < 0)
 244                 dist += sdp->sd_jdesc->jd_blocks;
 245
 246         return dist;
 247 }
 248
 249 static unsigned int current_tail(struct gfs2_sbd *sdp)
 250 {
 251         struct gfs2_ail *ai;
 252         unsigned int tail;
 253
 254         gfs2_log_lock(sdp);
 255
 256         if (list_empty(&sdp->sd_ail1_list))
 257                 tail = sdp->sd_log_head;
 258         else {
 259                 ai = list_entry(sdp->sd_ail1_list.prev,
 260                                 struct gfs2_ail, ai_list);
 261                 tail = ai->ai_first;
 262         }
 263
 264         gfs2_log_unlock(sdp);
 265
 266         return tail;
 267 }
 268
 269 static inline void log_incr_head(struct gfs2_sbd *sdp)
 270 {
 271         if (sdp->sd_log_flush_head == sdp->sd_log_tail)
 272                 gfs2_assert_withdraw(sdp,
 273                                 sdp->sd_log_flush_head == sdp->sd_log_head);
 274
 275         if (++sdp->sd_log_flush_head == sdp->sd_jdesc->jd_blocks) {
 276                 sdp->sd_log_flush_head = 0;
 277                 sdp->sd_log_flush_wrapped = 1;
 278         }
 279 }
 280
 281 /**
 282  * gfs2_log_get_buf - Get and initialize a buffer to use for log control data
 283  * @sdp: The GFS2 superblock
 284  *
 285  * Returns: the buffer_head
 286  */
 287
 288 struct buffer_head *gfs2_log_get_buf(struct gfs2_sbd *sdp)
 289 {
 290         uint64_t blkno = log_bmap(sdp, sdp->sd_log_flush_head);
 291         struct gfs2_log_buf *lb;
 292         struct buffer_head *bh;
 293
 294         lb = kzalloc(sizeof(struct gfs2_log_buf), GFP_NOFS | __GFP_NOFAIL);
 295         list_add(&lb->lb_list, &sdp->sd_log_flush_list);
 296
 297         bh = lb->lb_bh = sb_getblk(sdp->sd_vfs, blkno);
 298         lock_buffer(bh);
 299         memset(bh->b_data, 0, bh->b_size);
 300         set_buffer_uptodate(bh);
 301         clear_buffer_dirty(bh);
 302         unlock_buffer(bh);
 303
 304         log_incr_head(sdp);
 305
 306         return bh;
 307 }
 308
 309 /**
 310  * gfs2_log_fake_buf - Build a fake buffer head to write metadata buffer to log
 311  * @sdp: the filesystem
 312  * @data: the data the buffer_head should point to
 313  *
 314  * Returns: the log buffer descriptor
 315  */
 316
 317 struct buffer_head *gfs2_log_fake_buf(struct gfs2_sbd *sdp,
 318                                       struct buffer_head *real)
 319 {
 320         uint64_t blkno = log_bmap(sdp, sdp->sd_log_flush_head);
 321         struct gfs2_log_buf *lb;
 322         struct buffer_head *bh;
 323
 324         lb = kzalloc(sizeof(struct gfs2_log_buf), GFP_NOFS | __GFP_NOFAIL);
 325         list_add(&lb->lb_list, &sdp->sd_log_flush_list);
 326         lb->lb_real = real;
 327
 328         bh = lb->lb_bh = alloc_buffer_head(GFP_NOFS | __GFP_NOFAIL);
 329         atomic_set(&bh->b_count, 1);
 330         bh->b_state = (1 << BH_Mapped) | (1 << BH_Uptodate);
 331         set_bh_page(bh, real->b_page, bh_offset(real));
 332         bh->b_blocknr = blkno;
 333         bh->b_size = sdp->sd_sb.sb_bsize;
 334         bh->b_bdev = sdp->sd_vfs->s_bdev;
 335
 336         log_incr_head(sdp);
 337
 338         return bh;
 339 }
 340
 341 static void log_pull_tail(struct gfs2_sbd *sdp, unsigned int new_tail, int pull)
 342 {
 343         unsigned int dist = log_distance(sdp, new_tail, sdp->sd_log_tail);
 344
 345         ail2_empty(sdp, new_tail);
 346
 347         gfs2_log_lock(sdp);
 348         sdp->sd_log_blks_free += dist - ((pull) ? 1 : 0);
 349         /* printk(KERN_INFO "pull tail refunding %u blocks (%u left) pull=%d\n", dist - ((pull) ? 1 : 0), sdp->sd_log_blks_free, pull); */
 350         gfs2_assert_withdraw(sdp,
 351                              sdp->sd_log_blks_free <= sdp->sd_jdesc->jd_blocks);
 352         gfs2_log_unlock(sdp);
 353
 354         sdp->sd_log_tail = new_tail;
 355 }
 356
 357 /**
 358  * log_write_header - Get and initialize a journal header buffer
 359  * @sdp: The GFS2 superblock
 360  *
 361  * Returns: the initialized log buffer descriptor
 362  */
 363
 364 static void log_write_header(struct gfs2_sbd *sdp, uint32_t flags, int pull)
 365 {
 366         uint64_t blkno = log_bmap(sdp, sdp->sd_log_flush_head);
 367         struct buffer_head *bh;
 368         struct gfs2_log_header *lh;
 369         unsigned int tail;
 370         uint32_t hash;
 371
 372         /* printk(KERN_INFO "log write header start (flags=%08x, pull=%d)\n", flags, pull); */
 373
 374         bh = sb_getblk(sdp->sd_vfs, blkno);
 375         lock_buffer(bh);
 376         memset(bh->b_data, 0, bh->b_size);
 377         set_buffer_uptodate(bh);
 378         clear_buffer_dirty(bh);
 379         unlock_buffer(bh);
 380
 381         gfs2_ail1_empty(sdp, 0);
 382         tail = current_tail(sdp);
 383
 384         lh = (struct gfs2_log_header *)bh->b_data;
 385         memset(lh, 0, sizeof(struct gfs2_log_header));
 386         lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 387         lh->lh_header.mh_type = cpu_to_be32(GFS2_METATYPE_LH);
 388         lh->lh_header.mh_format = cpu_to_be32(GFS2_FORMAT_LH);
 389         lh->lh_sequence = be64_to_cpu(sdp->sd_log_sequence++);
 390         lh->lh_flags = be32_to_cpu(flags);
 391         lh->lh_tail = be32_to_cpu(tail);
 392         lh->lh_blkno = be32_to_cpu(sdp->sd_log_flush_head);
 393         hash = gfs2_disk_hash(bh->b_data, sizeof(struct gfs2_log_header));
 394         lh->lh_hash = cpu_to_be32(hash);
 395
 396         set_buffer_dirty(bh);
 397         if (sync_dirty_buffer(bh))
 398                 gfs2_io_error_bh(sdp, bh);
 399         brelse(bh);
 400
 401         if (sdp->sd_log_tail != tail)
 402                 log_pull_tail(sdp, tail, pull);
 403         else
 404                 gfs2_assert_withdraw(sdp, !pull);
 405
 406         sdp->sd_log_idle = (tail == sdp->sd_log_flush_head);
 407         log_incr_head(sdp);
 408
 409         /* printk(KERN_INFO "log write header out\n"); */
 410 }
 411
 412 static void log_flush_commit(struct gfs2_sbd *sdp)
 413 {
 414         struct list_head *head = &sdp->sd_log_flush_list;
 415         struct gfs2_log_buf *lb;
 416         struct buffer_head *bh;
 417 #if 0
 418         unsigned int d;
 419
 420         d = log_distance(sdp, sdp->sd_log_flush_head, sdp->sd_log_head);
 421
 422         gfs2_assert_withdraw(sdp, d + 1 == sdp->sd_log_blks_reserved);
 423 #endif
 424
 425         while (!list_empty(head)) {
 426                 lb = list_entry(head->next, struct gfs2_log_buf, lb_list);
 427                 list_del(&lb->lb_list);
 428                 bh = lb->lb_bh;
 429
 430                 wait_on_buffer(bh);
 431                 if (!buffer_uptodate(bh))
 432                         gfs2_io_error_bh(sdp, bh);
 433                 if (lb->lb_real) {
 434                         while (atomic_read(&bh->b_count) != 1)  /* Grrrr... */
 435                                 schedule();
 436                         free_buffer_head(bh);
 437                 } else
 438                         brelse(bh);
 439                 kfree(lb);
 440         }
 441
 442         log_write_header(sdp, 0, 0);
 443 }
 444
 445 /**
 446  * gfs2_log_flush - flush incore transaction(s)
 447  * @sdp: the filesystem
 448  * @gl: The glock structure to flush.  If NULL, flush the whole incore log
 449  *
 450  */
 451
 452 void gfs2_log_flush(struct gfs2_sbd *sdp, struct gfs2_glock *gl)
 453 {
 454         struct gfs2_ail *ai;
 455
 456         down_write(&sdp->sd_log_flush_lock);
 457
 458         if (gl) {
 459                 gfs2_log_lock(sdp);
 460                 if (list_empty(&gl->gl_le.le_list)) {
 461                         gfs2_log_unlock(sdp);
 462                         up_write(&sdp->sd_log_flush_lock);
 463                         return;
 464                 }
 465                 gfs2_log_unlock(sdp);
 466         }
 467
 468         ai = kzalloc(sizeof(struct gfs2_ail), GFP_NOFS | __GFP_NOFAIL);
 469         INIT_LIST_HEAD(&ai->ai_ail1_list);
 470         INIT_LIST_HEAD(&ai->ai_ail2_list);
 471
 472         gfs2_assert_withdraw(sdp,
 473                         sdp->sd_log_num_buf == sdp->sd_log_commited_buf);
 474         gfs2_assert_withdraw(sdp,
 475                         sdp->sd_log_num_revoke == sdp->sd_log_commited_revoke);
 476
 477         sdp->sd_log_flush_head = sdp->sd_log_head;
 478         sdp->sd_log_flush_wrapped = 0;
 479         ai->ai_first = sdp->sd_log_flush_head;
 480
 481         lops_before_commit(sdp);
 482         if (!list_empty(&sdp->sd_log_flush_list))
 483                 log_flush_commit(sdp);
 484         else if (sdp->sd_log_tail != current_tail(sdp) && !sdp->sd_log_idle)
 485                 log_write_header(sdp, 0, PULL);
 486         lops_after_commit(sdp, ai);
 487         sdp->sd_log_head = sdp->sd_log_flush_head;
 488
 489         /* printk(KERN_INFO "sd_log_num_hdrs %u\n", sdp->sd_log_num_hdrs); */
 490         sdp->sd_log_blks_free -= sdp->sd_log_num_hdrs;
 491
 492         sdp->sd_log_blks_reserved =
 493                 sdp->sd_log_commited_buf =
 494                 sdp->sd_log_num_hdrs =
 495                 sdp->sd_log_commited_revoke = 0;
 496
 497         gfs2_log_lock(sdp);
 498         if (!list_empty(&ai->ai_ail1_list)) {
 499                 list_add(&ai->ai_list, &sdp->sd_ail1_list);
 500                 ai = NULL;
 501         }
 502         gfs2_log_unlock(sdp);
 503
 504         sdp->sd_vfs->s_dirt = 0;
 505         up_write(&sdp->sd_log_flush_lock);
 506
 507         kfree(ai);
 508 }
 509
 510 static void log_refund(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 511 {
 512         unsigned int reserved = 1;
 513         unsigned int old;
 514
 515         gfs2_log_lock(sdp);
 516
 517         sdp->sd_log_commited_buf += tr->tr_num_buf_new - tr->tr_num_buf_rm;
 518         gfs2_assert_withdraw(sdp, ((int)sdp->sd_log_commited_buf) >= 0);
 519         sdp->sd_log_commited_revoke += tr->tr_num_revoke - tr->tr_num_revoke_rm;
 520         gfs2_assert_withdraw(sdp, ((int)sdp->sd_log_commited_revoke) >= 0);
 521
 522         if (sdp->sd_log_commited_buf)
 523                 reserved += sdp->sd_log_commited_buf;
 524         if (sdp->sd_log_commited_revoke)
 525                 reserved += gfs2_struct2blk(sdp, sdp->sd_log_commited_revoke,
 526                                             sizeof(uint64_t));
 527
 528         old = sdp->sd_log_blks_free;
 529         sdp->sd_log_blks_free += tr->tr_reserved -
 530                                  (reserved - sdp->sd_log_blks_reserved);
 531
 532         gfs2_assert_withdraw(sdp, sdp->sd_log_blks_free >= old);
 533         gfs2_assert_withdraw(sdp,
 534                              sdp->sd_log_blks_free <= sdp->sd_jdesc->jd_blocks +
 535                              sdp->sd_log_num_hdrs);
 536
 537         sdp->sd_log_blks_reserved = reserved;
 538
 539         gfs2_log_unlock(sdp);
 540 }
 541
 542 /**
 543  * gfs2_log_commit - Commit a transaction to the log
 544  * @sdp: the filesystem
 545  * @tr: the transaction
 546  *
 547  * Returns: errno
 548  */
 549
 550 void gfs2_log_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 551 {
 552         log_refund(sdp, tr);
 553         lops_incore_commit(sdp, tr);
 554
 555         sdp->sd_vfs->s_dirt = 1;
 556         up_read(&sdp->sd_log_flush_lock);
 557
 558         gfs2_log_lock(sdp);
 559         if (sdp->sd_log_num_buf > gfs2_tune_get(sdp, gt_incore_log_blocks)) {
 560                 gfs2_log_unlock(sdp);
 561                 gfs2_log_flush(sdp, NULL);
 562         } else
 563                 gfs2_log_unlock(sdp);
 564 }
 565
 566 /**
 567  * gfs2_log_shutdown - write a shutdown header into a journal
 568  * @sdp: the filesystem
 569  *
 570  */
 571
 572 void gfs2_log_shutdown(struct gfs2_sbd *sdp)
 573 {
 574         down_write(&sdp->sd_log_flush_lock);
 575
 576         gfs2_assert_withdraw(sdp, !sdp->sd_log_blks_reserved);
 577         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_gl);
 578         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_buf);
 579         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_jdata);
 580         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
 581         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_rg);
 582         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_databuf);
 583         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_hdrs);
 584         gfs2_assert_withdraw(sdp, list_empty(&sdp->sd_ail1_list));
 585
 586         sdp->sd_log_flush_head = sdp->sd_log_head;
 587         sdp->sd_log_flush_wrapped = 0;
 588
 589         log_write_header(sdp, GFS2_LOG_HEAD_UNMOUNT, 0);
 590
 591         /* printk(KERN_INFO "sd_log_blks_free %u, sd_jdesc->jd_blocks %u\n", sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks); */
 592         gfs2_assert_warn(sdp, sdp->sd_log_blks_free == sdp->sd_jdesc->jd_blocks);
 593         gfs2_assert_warn(sdp, sdp->sd_log_head == sdp->sd_log_tail);
 594         gfs2_assert_warn(sdp, list_empty(&sdp->sd_ail2_list));
 595
 596         sdp->sd_log_head = sdp->sd_log_flush_head;
 597         sdp->sd_log_tail = sdp->sd_log_head;
 598
 599         up_write(&sdp->sd_log_flush_lock);
 600 }
 601