fs/gfs2/log.c

   1 /*
   2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   3  * Copyright (C) 2004-2007 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/crc32c.h>
  18 #include <linux/delay.h>
  19 #include <linux/kthread.h>
  20 #include <linux/freezer.h>
  21 #include <linux/bio.h>
  22 #include <linux/blkdev.h>
  23 #include <linux/writeback.h>
  24 #include <linux/list_sort.h>
  25
  26 #include "gfs2.h"
  27 #include "incore.h"
  28 #include "bmap.h"
  29 #include "glock.h"
  30 #include "log.h"
  31 #include "lops.h"
  32 #include "meta_io.h"
  33 #include "util.h"
  34 #include "dir.h"
  35 #include "trace_gfs2.h"
  36
  37 /**
  38  * gfs2_struct2blk - compute stuff
  39  * @sdp: the filesystem
  40  * @nstruct: the number of structures
  41  * @ssize: the size of the structures
  42  *
  43  * Compute the number of log descriptor blocks needed to hold a certain number
  44  * of structures of a certain size.
  45  *
  46  * Returns: the number of blocks needed (minimum is always 1)
  47  */
  48
  49 unsigned int gfs2_struct2blk(struct gfs2_sbd *sdp, unsigned int nstruct,
  50                              unsigned int ssize)
  51 {
  52         unsigned int blks;
  53         unsigned int first, second;
  54
  55         blks = 1;
  56         first = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) / ssize;
  57
  58         if (nstruct > first) {
  59                 second = (sdp->sd_sb.sb_bsize -
  60                           sizeof(struct gfs2_meta_header)) / ssize;
  61                 blks += DIV_ROUND_UP(nstruct - first, second);
  62         }
  63
  64         return blks;
  65 }
  66
  67 /**
  68  * gfs2_remove_from_ail - Remove an entry from the ail lists, updating counters
  69  * @mapping: The associated mapping (maybe NULL)
  70  * @bd: The gfs2_bufdata to remove
  71  *
  72  * The ail lock _must_ be held when calling this function
  73  *
  74  */
  75
  76 static void gfs2_remove_from_ail(struct gfs2_bufdata *bd)
  77 {
  78         bd->bd_tr = NULL;
  79         list_del_init(&bd->bd_ail_st_list);
  80         list_del_init(&bd->bd_ail_gl_list);
  81         atomic_dec(&bd->bd_gl->gl_ail_count);
  82         brelse(bd->bd_bh);
  83 }
  84
  85 /**
  86  * gfs2_ail1_start_one - Start I/O on a part of the AIL
  87  * @sdp: the filesystem
  88  * @wbc: The writeback control structure
  89  * @ai: The ail structure
  90  *
  91  */
  92
  93 static int gfs2_ail1_start_one(struct gfs2_sbd *sdp,
  94                                struct writeback_control *wbc,
  95                                struct gfs2_trans *tr,
  96                                bool *withdraw)
  97 __releases(&sdp->sd_ail_lock)
  98 __acquires(&sdp->sd_ail_lock)
  99 {
 100         struct gfs2_glock *gl = NULL;
 101         struct address_space *mapping;
 102         struct gfs2_bufdata *bd, *s;
 103         struct buffer_head *bh;
 104
 105         list_for_each_entry_safe_reverse(bd, s, &tr->tr_ail1_list, bd_ail_st_list) {
 106                 bh = bd->bd_bh;
 107
 108                 gfs2_assert(sdp, bd->bd_tr == tr);
 109
 110                 if (!buffer_busy(bh)) {
 111                         if (!buffer_uptodate(bh)) {
 112                                 gfs2_io_error_bh(sdp, bh);
 113                                 *withdraw = true;
 114                         }
 115                         list_move(&bd->bd_ail_st_list, &tr->tr_ail2_list);
 116                         continue;
 117                 }
 118
 119                 if (!buffer_dirty(bh))
 120                         continue;
 121                 if (gl == bd->bd_gl)
 122                         continue;
 123                 gl = bd->bd_gl;
 124                 list_move(&bd->bd_ail_st_list, &tr->tr_ail1_list);
 125                 mapping = bh->b_page->mapping;
 126                 if (!mapping)
 127                         continue;
 128                 spin_unlock(&sdp->sd_ail_lock);
 129                 generic_writepages(mapping, wbc);
 130                 spin_lock(&sdp->sd_ail_lock);
 131                 if (wbc->nr_to_write <= 0)
 132                         break;
 133                 return 1;
 134         }
 135
 136         return 0;
 137 }
 138
 139
 140 /**
 141  * gfs2_ail1_flush - start writeback of some ail1 entries
 142  * @sdp: The super block
 143  * @wbc: The writeback control structure
 144  *
 145  * Writes back some ail1 entries, according to the limits in the
 146  * writeback control structure
 147  */
 148
 149 void gfs2_ail1_flush(struct gfs2_sbd *sdp, struct writeback_control *wbc)
 150 {
 151         struct list_head *head = &sdp->sd_ail1_list;
 152         struct gfs2_trans *tr;
 153         struct blk_plug plug;
 154         bool withdraw = false;
 155
 156         trace_gfs2_ail_flush(sdp, wbc, 1);
 157         blk_start_plug(&plug);
 158         spin_lock(&sdp->sd_ail_lock);
 159 restart:
 160         list_for_each_entry_reverse(tr, head, tr_list) {
 161                 if (wbc->nr_to_write <= 0)
 162                         break;
 163                 if (gfs2_ail1_start_one(sdp, wbc, tr, &withdraw))
 164                         goto restart;
 165         }
 166         spin_unlock(&sdp->sd_ail_lock);
 167         blk_finish_plug(&plug);
 168         if (withdraw)
 169                 gfs2_lm_withdraw(sdp, NULL);
 170         trace_gfs2_ail_flush(sdp, wbc, 0);
 171 }
 172
 173 /**
 174  * gfs2_ail1_start - start writeback of all ail1 entries
 175  * @sdp: The superblock
 176  */
 177
 178 static void gfs2_ail1_start(struct gfs2_sbd *sdp)
 179 {
 180         struct writeback_control wbc = {
 181                 .sync_mode = WB_SYNC_NONE,
 182                 .nr_to_write = LONG_MAX,
 183                 .range_start = 0,
 184                 .range_end = LLONG_MAX,
 185         };
 186
 187         return gfs2_ail1_flush(sdp, &wbc);
 188 }
 189
 190 /**
 191  * gfs2_ail1_empty_one - Check whether or not a trans in the AIL has been synced
 192  * @sdp: the filesystem
 193  * @ai: the AIL entry
 194  *
 195  */
 196
 197 static void gfs2_ail1_empty_one(struct gfs2_sbd *sdp, struct gfs2_trans *tr,
 198                                 bool *withdraw)
 199 {
 200         struct gfs2_bufdata *bd, *s;
 201         struct buffer_head *bh;
 202
 203         list_for_each_entry_safe_reverse(bd, s, &tr->tr_ail1_list,
 204                                          bd_ail_st_list) {
 205                 bh = bd->bd_bh;
 206                 gfs2_assert(sdp, bd->bd_tr == tr);
 207                 if (buffer_busy(bh))
 208                         continue;
 209                 if (!buffer_uptodate(bh)) {
 210                         gfs2_io_error_bh(sdp, bh);
 211                         *withdraw = true;
 212                 }
 213                 list_move(&bd->bd_ail_st_list, &tr->tr_ail2_list);
 214         }
 215 }
 216
 217 /**
 218  * gfs2_ail1_empty - Try to empty the ail1 lists
 219  * @sdp: The superblock
 220  *
 221  * Tries to empty the ail1 lists, starting with the oldest first
 222  */
 223
 224 static int gfs2_ail1_empty(struct gfs2_sbd *sdp)
 225 {
 226         struct gfs2_trans *tr, *s;
 227         int oldest_tr = 1;
 228         int ret;
 229         bool withdraw = false;
 230
 231         spin_lock(&sdp->sd_ail_lock);
 232         list_for_each_entry_safe_reverse(tr, s, &sdp->sd_ail1_list, tr_list) {
 233                 gfs2_ail1_empty_one(sdp, tr, &withdraw);
 234                 if (list_empty(&tr->tr_ail1_list) && oldest_tr)
 235                         list_move(&tr->tr_list, &sdp->sd_ail2_list);
 236                 else
 237                         oldest_tr = 0;
 238         }
 239         ret = list_empty(&sdp->sd_ail1_list);
 240         spin_unlock(&sdp->sd_ail_lock);
 241
 242         if (withdraw)
 243                 gfs2_lm_withdraw(sdp, "fatal: I/O error(s)\n");
 244
 245         return ret;
 246 }
 247
 248 static void gfs2_ail1_wait(struct gfs2_sbd *sdp)
 249 {
 250         struct gfs2_trans *tr;
 251         struct gfs2_bufdata *bd;
 252         struct buffer_head *bh;
 253
 254         spin_lock(&sdp->sd_ail_lock);
 255         list_for_each_entry_reverse(tr, &sdp->sd_ail1_list, tr_list) {
 256                 list_for_each_entry(bd, &tr->tr_ail1_list, bd_ail_st_list) {
 257                         bh = bd->bd_bh;
 258                         if (!buffer_locked(bh))
 259                                 continue;
 260                         get_bh(bh);
 261                         spin_unlock(&sdp->sd_ail_lock);
 262                         wait_on_buffer(bh);
 263                         brelse(bh);
 264                         return;
 265                 }
 266         }
 267         spin_unlock(&sdp->sd_ail_lock);
 268 }
 269
 270 /**
 271  * gfs2_ail2_empty_one - Check whether or not a trans in the AIL has been synced
 272  * @sdp: the filesystem
 273  * @ai: the AIL entry
 274  *
 275  */
 276
 277 static void gfs2_ail2_empty_one(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 278 {
 279         struct list_head *head = &tr->tr_ail2_list;
 280         struct gfs2_bufdata *bd;
 281
 282         while (!list_empty(head)) {
 283                 bd = list_entry(head->prev, struct gfs2_bufdata,
 284                                 bd_ail_st_list);
 285                 gfs2_assert(sdp, bd->bd_tr == tr);
 286                 gfs2_remove_from_ail(bd);
 287         }
 288 }
 289
 290 static void ail2_empty(struct gfs2_sbd *sdp, unsigned int new_tail)
 291 {
 292         struct gfs2_trans *tr, *safe;
 293         unsigned int old_tail = sdp->sd_log_tail;
 294         int wrap = (new_tail < old_tail);
 295         int a, b, rm;
 296
 297         spin_lock(&sdp->sd_ail_lock);
 298
 299         list_for_each_entry_safe(tr, safe, &sdp->sd_ail2_list, tr_list) {
 300                 a = (old_tail <= tr->tr_first);
 301                 b = (tr->tr_first < new_tail);
 302                 rm = (wrap) ? (a || b) : (a && b);
 303                 if (!rm)
 304                         continue;
 305
 306                 gfs2_ail2_empty_one(sdp, tr);
 307                 list_del(&tr->tr_list);
 308                 gfs2_assert_warn(sdp, list_empty(&tr->tr_ail1_list));
 309                 gfs2_assert_warn(sdp, list_empty(&tr->tr_ail2_list));
 310                 kfree(tr);
 311         }
 312
 313         spin_unlock(&sdp->sd_ail_lock);
 314 }
 315
 316 /**
 317  * gfs2_log_release - Release a given number of log blocks
 318  * @sdp: The GFS2 superblock
 319  * @blks: The number of blocks
 320  *
 321  */
 322
 323 void gfs2_log_release(struct gfs2_sbd *sdp, unsigned int blks)
 324 {
 325
 326         atomic_add(blks, &sdp->sd_log_blks_free);
 327         trace_gfs2_log_blocks(sdp, blks);
 328         gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <=
 329                                   sdp->sd_jdesc->jd_blocks);
 330         up_read(&sdp->sd_log_flush_lock);
 331 }
 332
 333 /**
 334  * gfs2_log_reserve - Make a log reservation
 335  * @sdp: The GFS2 superblock
 336  * @blks: The number of blocks to reserve
 337  *
 338  * Note that we never give out the last few blocks of the journal. Thats
 339  * due to the fact that there is a small number of header blocks
 340  * associated with each log flush. The exact number can't be known until
 341  * flush time, so we ensure that we have just enough free blocks at all
 342  * times to avoid running out during a log flush.
 343  *
 344  * We no longer flush the log here, instead we wake up logd to do that
 345  * for us. To avoid the thundering herd and to ensure that we deal fairly
 346  * with queued waiters, we use an exclusive wait. This means that when we
 347  * get woken with enough journal space to get our reservation, we need to
 348  * wake the next waiter on the list.
 349  *
 350  * Returns: errno
 351  */
 352
 353 int gfs2_log_reserve(struct gfs2_sbd *sdp, unsigned int blks)
 354 {
 355         int ret = 0;
 356         unsigned reserved_blks = 7 * (4096 / sdp->sd_vfs->s_blocksize);
 357         unsigned wanted = blks + reserved_blks;
 358         DEFINE_WAIT(wait);
 359         int did_wait = 0;
 360         unsigned int free_blocks;
 361
 362         if (gfs2_assert_warn(sdp, blks) ||
 363             gfs2_assert_warn(sdp, blks <= sdp->sd_jdesc->jd_blocks))
 364                 return -EINVAL;
 365         atomic_add(blks, &sdp->sd_log_blks_needed);
 366 retry:
 367         free_blocks = atomic_read(&sdp->sd_log_blks_free);
 368         if (unlikely(free_blocks <= wanted)) {
 369                 do {
 370                         prepare_to_wait_exclusive(&sdp->sd_log_waitq, &wait,
 371                                         TASK_UNINTERRUPTIBLE);
 372                         wake_up(&sdp->sd_logd_waitq);
 373                         did_wait = 1;
 374                         if (atomic_read(&sdp->sd_log_blks_free) <= wanted)
 375                                 io_schedule();
 376                         free_blocks = atomic_read(&sdp->sd_log_blks_free);
 377                 } while(free_blocks <= wanted);
 378                 finish_wait(&sdp->sd_log_waitq, &wait);
 379         }
 380         atomic_inc(&sdp->sd_reserving_log);
 381         if (atomic_cmpxchg(&sdp->sd_log_blks_free, free_blocks,
 382                                 free_blocks - blks) != free_blocks) {
 383                 if (atomic_dec_and_test(&sdp->sd_reserving_log))
 384                         wake_up(&sdp->sd_reserving_log_wait);
 385                 goto retry;
 386         }
 387         atomic_sub(blks, &sdp->sd_log_blks_needed);
 388         trace_gfs2_log_blocks(sdp, -blks);
 389
 390         /*
 391          * If we waited, then so might others, wake them up _after_ we get
 392          * our share of the log.
 393          */
 394         if (unlikely(did_wait))
 395                 wake_up(&sdp->sd_log_waitq);
 396
 397         down_read(&sdp->sd_log_flush_lock);
 398         if (unlikely(!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))) {
 399                 gfs2_log_release(sdp, blks);
 400                 ret = -EROFS;
 401         }
 402         if (atomic_dec_and_test(&sdp->sd_reserving_log))
 403                 wake_up(&sdp->sd_reserving_log_wait);
 404         return ret;
 405 }
 406
 407 /**
 408  * log_distance - Compute distance between two journal blocks
 409  * @sdp: The GFS2 superblock
 410  * @newer: The most recent journal block of the pair
 411  * @older: The older journal block of the pair
 412  *
 413  *   Compute the distance (in the journal direction) between two
 414  *   blocks in the journal
 415  *
 416  * Returns: the distance in blocks
 417  */
 418
 419 static inline unsigned int log_distance(struct gfs2_sbd *sdp, unsigned int newer,
 420                                         unsigned int older)
 421 {
 422         int dist;
 423
 424         dist = newer - older;
 425         if (dist < 0)
 426                 dist += sdp->sd_jdesc->jd_blocks;
 427
 428         return dist;
 429 }
 430
 431 /**
 432  * calc_reserved - Calculate the number of blocks to reserve when
 433  *                 refunding a transaction's unused buffers.
 434  * @sdp: The GFS2 superblock
 435  *
 436  * This is complex.  We need to reserve room for all our currently used
 437  * metadata buffers (e.g. normal file I/O rewriting file time stamps) and
 438  * all our journaled data buffers for journaled files (e.g. files in the
 439  * meta_fs like rindex, or files for which chattr +j was done.)
 440  * If we don't reserve enough space, gfs2_log_refund and gfs2_log_flush
 441  * will count it as free space (sd_log_blks_free) and corruption will follow.
 442  *
 443  * We can have metadata bufs and jdata bufs in the same journal.  So each
 444  * type gets its own log header, for which we need to reserve a block.
 445  * In fact, each type has the potential for needing more than one header
 446  * in cases where we have more buffers than will fit on a journal page.
 447  * Metadata journal entries take up half the space of journaled buffer entries.
 448  * Thus, metadata entries have buf_limit (502) and journaled buffers have
 449  * databuf_limit (251) before they cause a wrap around.
 450  *
 451  * Also, we need to reserve blocks for revoke journal entries and one for an
 452  * overall header for the lot.
 453  *
 454  * Returns: the number of blocks reserved
 455  */
 456 static unsigned int calc_reserved(struct gfs2_sbd *sdp)
 457 {
 458         unsigned int reserved = 0;
 459         unsigned int mbuf;
 460         unsigned int dbuf;
 461         struct gfs2_trans *tr = sdp->sd_log_tr;
 462
 463         if (tr) {
 464                 mbuf = tr->tr_num_buf_new - tr->tr_num_buf_rm;
 465                 dbuf = tr->tr_num_databuf_new - tr->tr_num_databuf_rm;
 466                 reserved = mbuf + dbuf;
 467                 /* Account for header blocks */
 468                 reserved += DIV_ROUND_UP(mbuf, buf_limit(sdp));
 469                 reserved += DIV_ROUND_UP(dbuf, databuf_limit(sdp));
 470         }
 471
 472         if (sdp->sd_log_commited_revoke > 0)
 473                 reserved += gfs2_struct2blk(sdp, sdp->sd_log_commited_revoke,
 474                                           sizeof(u64));
 475         /* One for the overall header */
 476         if (reserved)
 477                 reserved++;
 478         return reserved;
 479 }
 480
 481 static unsigned int current_tail(struct gfs2_sbd *sdp)
 482 {
 483         struct gfs2_trans *tr;
 484         unsigned int tail;
 485
 486         spin_lock(&sdp->sd_ail_lock);
 487
 488         if (list_empty(&sdp->sd_ail1_list)) {
 489                 tail = sdp->sd_log_head;
 490         } else {
 491                 tr = list_entry(sdp->sd_ail1_list.prev, struct gfs2_trans,
 492                                 tr_list);
 493                 tail = tr->tr_first;
 494         }
 495
 496         spin_unlock(&sdp->sd_ail_lock);
 497
 498         return tail;
 499 }
 500
 501 static void log_pull_tail(struct gfs2_sbd *sdp, unsigned int new_tail)
 502 {
 503         unsigned int dist = log_distance(sdp, new_tail, sdp->sd_log_tail);
 504
 505         ail2_empty(sdp, new_tail);
 506
 507         atomic_add(dist, &sdp->sd_log_blks_free);
 508         trace_gfs2_log_blocks(sdp, dist);
 509         gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <=
 510                              sdp->sd_jdesc->jd_blocks);
 511
 512         sdp->sd_log_tail = new_tail;
 513 }
 514
 515
 516 static void log_flush_wait(struct gfs2_sbd *sdp)
 517 {
 518         DEFINE_WAIT(wait);
 519
 520         if (atomic_read(&sdp->sd_log_in_flight)) {
 521                 do {
 522                         prepare_to_wait(&sdp->sd_log_flush_wait, &wait,
 523                                         TASK_UNINTERRUPTIBLE);
 524                         if (atomic_read(&sdp->sd_log_in_flight))
 525                                 io_schedule();
 526                 } while(atomic_read(&sdp->sd_log_in_flight));
 527                 finish_wait(&sdp->sd_log_flush_wait, &wait);
 528         }
 529 }
 530
 531 static int ip_cmp(void *priv, struct list_head *a, struct list_head *b)
 532 {
 533         struct gfs2_inode *ipa, *ipb;
 534
 535         ipa = list_entry(a, struct gfs2_inode, i_ordered);
 536         ipb = list_entry(b, struct gfs2_inode, i_ordered);
 537
 538         if (ipa->i_no_addr < ipb->i_no_addr)
 539                 return -1;
 540         if (ipa->i_no_addr > ipb->i_no_addr)
 541                 return 1;
 542         return 0;
 543 }
 544
 545 static void gfs2_ordered_write(struct gfs2_sbd *sdp)
 546 {
 547         struct gfs2_inode *ip;
 548         LIST_HEAD(written);
 549
 550         spin_lock(&sdp->sd_ordered_lock);
 551         list_sort(NULL, &sdp->sd_log_le_ordered, &ip_cmp);
 552         while (!list_empty(&sdp->sd_log_le_ordered)) {
 553                 ip = list_entry(sdp->sd_log_le_ordered.next, struct gfs2_inode, i_ordered);
 554                 if (ip->i_inode.i_mapping->nrpages == 0) {
 555                         test_and_clear_bit(GIF_ORDERED, &ip->i_flags);
 556                         list_del(&ip->i_ordered);
 557                         continue;
 558                 }
 559                 list_move(&ip->i_ordered, &written);
 560                 spin_unlock(&sdp->sd_ordered_lock);
 561                 filemap_fdatawrite(ip->i_inode.i_mapping);
 562                 spin_lock(&sdp->sd_ordered_lock);
 563         }
 564         list_splice(&written, &sdp->sd_log_le_ordered);
 565         spin_unlock(&sdp->sd_ordered_lock);
 566 }
 567
 568 static void gfs2_ordered_wait(struct gfs2_sbd *sdp)
 569 {
 570         struct gfs2_inode *ip;
 571
 572         spin_lock(&sdp->sd_ordered_lock);
 573         while (!list_empty(&sdp->sd_log_le_ordered)) {
 574                 ip = list_entry(sdp->sd_log_le_ordered.next, struct gfs2_inode, i_ordered);
 575                 list_del(&ip->i_ordered);
 576                 WARN_ON(!test_and_clear_bit(GIF_ORDERED, &ip->i_flags));
 577                 if (ip->i_inode.i_mapping->nrpages == 0)
 578                         continue;
 579                 spin_unlock(&sdp->sd_ordered_lock);
 580                 filemap_fdatawait(ip->i_inode.i_mapping);
 581                 spin_lock(&sdp->sd_ordered_lock);
 582         }
 583         spin_unlock(&sdp->sd_ordered_lock);
 584 }
 585
 586 void gfs2_ordered_del_inode(struct gfs2_inode *ip)
 587 {
 588         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 589
 590         spin_lock(&sdp->sd_ordered_lock);
 591         if (test_and_clear_bit(GIF_ORDERED, &ip->i_flags))
 592                 list_del(&ip->i_ordered);
 593         spin_unlock(&sdp->sd_ordered_lock);
 594 }
 595
 596 void gfs2_add_revoke(struct gfs2_sbd *sdp, struct gfs2_bufdata *bd)
 597 {
 598         struct buffer_head *bh = bd->bd_bh;
 599         struct gfs2_glock *gl = bd->bd_gl;
 600
 601         bh->b_private = NULL;
 602         bd->bd_blkno = bh->b_blocknr;
 603         gfs2_remove_from_ail(bd); /* drops ref on bh */
 604         bd->bd_bh = NULL;
 605         bd->bd_ops = &gfs2_revoke_lops;
 606         sdp->sd_log_num_revoke++;
 607         if (atomic_inc_return(&gl->gl_revokes) == 1)
 608                 gfs2_glock_hold(gl);
 609         set_bit(GLF_LFLUSH, &gl->gl_flags);
 610         list_add(&bd->bd_list, &sdp->sd_log_le_revoke);
 611 }
 612
 613 void gfs2_write_revokes(struct gfs2_sbd *sdp)
 614 {
 615         struct gfs2_trans *tr;
 616         struct gfs2_bufdata *bd, *tmp;
 617         int have_revokes = 0;
 618         int max_revokes = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) / sizeof(u64);
 619
 620         gfs2_ail1_empty(sdp);
 621         spin_lock(&sdp->sd_ail_lock);
 622         list_for_each_entry(tr, &sdp->sd_ail1_list, tr_list) {
 623                 list_for_each_entry(bd, &tr->tr_ail2_list, bd_ail_st_list) {
 624                         if (list_empty(&bd->bd_list)) {
 625                                 have_revokes = 1;
 626                                 goto done;
 627                         }
 628                 }
 629         }
 630 done:
 631         spin_unlock(&sdp->sd_ail_lock);
 632         if (have_revokes == 0)
 633                 return;
 634         while (sdp->sd_log_num_revoke > max_revokes)
 635                 max_revokes += (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header)) / sizeof(u64);
 636         max_revokes -= sdp->sd_log_num_revoke;
 637         if (!sdp->sd_log_num_revoke) {
 638                 atomic_dec(&sdp->sd_log_blks_free);
 639                 /* If no blocks have been reserved, we need to also
 640                  * reserve a block for the header */
 641                 if (!sdp->sd_log_blks_reserved)
 642                         atomic_dec(&sdp->sd_log_blks_free);
 643         }
 644         gfs2_log_lock(sdp);
 645         spin_lock(&sdp->sd_ail_lock);
 646         list_for_each_entry(tr, &sdp->sd_ail1_list, tr_list) {
 647                 list_for_each_entry_safe(bd, tmp, &tr->tr_ail2_list, bd_ail_st_list) {
 648                         if (max_revokes == 0)
 649                                 goto out_of_blocks;
 650                         if (!list_empty(&bd->bd_list))
 651                                 continue;
 652                         gfs2_add_revoke(sdp, bd);
 653                         max_revokes--;
 654                 }
 655         }
 656 out_of_blocks:
 657         spin_unlock(&sdp->sd_ail_lock);
 658         gfs2_log_unlock(sdp);
 659
 660         if (!sdp->sd_log_num_revoke) {
 661                 atomic_inc(&sdp->sd_log_blks_free);
 662                 if (!sdp->sd_log_blks_reserved)
 663                         atomic_inc(&sdp->sd_log_blks_free);
 664         }
 665 }
 666
 667 /**
 668  * write_log_header - Write a journal log header buffer at sd_log_flush_head
 669  * @sdp: The GFS2 superblock
 670  * @jd: journal descriptor of the journal to which we are writing
 671  * @seq: sequence number
 672  * @tail: tail of the log
 673  * @flags: log header flags GFS2_LOG_HEAD_*
 674  * @op_flags: flags to pass to the bio
 675  *
 676  * Returns: the initialized log buffer descriptor
 677  */
 678
 679 void gfs2_write_log_header(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd,
 680                            u64 seq, u32 tail, u32 flags, int op_flags)
 681 {
 682         struct gfs2_log_header *lh;
 683         u32 hash, crc;
 684         struct page *page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
 685         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 686         struct timespec64 tv;
 687         struct super_block *sb = sdp->sd_vfs;
 688         u64 addr;
 689
 690         lh = page_address(page);
 691         clear_page(lh);
 692
 693         lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 694         lh->lh_header.mh_type = cpu_to_be32(GFS2_METATYPE_LH);
 695         lh->lh_header.__pad0 = cpu_to_be64(0);
 696         lh->lh_header.mh_format = cpu_to_be32(GFS2_FORMAT_LH);
 697         lh->lh_header.mh_jid = cpu_to_be32(sdp->sd_jdesc->jd_jid);
 698         lh->lh_sequence = cpu_to_be64(seq);
 699         lh->lh_flags = cpu_to_be32(flags);
 700         lh->lh_tail = cpu_to_be32(tail);
 701         lh->lh_blkno = cpu_to_be32(sdp->sd_log_flush_head);
 702         hash = ~crc32(~0, lh, LH_V1_SIZE);
 703         lh->lh_hash = cpu_to_be32(hash);
 704
 705         ktime_get_coarse_real_ts64(&tv);
 706         lh->lh_nsec = cpu_to_be32(tv.tv_nsec);
 707         lh->lh_sec = cpu_to_be64(tv.tv_sec);
 708         addr = gfs2_log_bmap(sdp);
 709         lh->lh_addr = cpu_to_be64(addr);
 710         lh->lh_jinode = cpu_to_be64(GFS2_I(jd->jd_inode)->i_no_addr);
 711
 712         /* We may only write local statfs, quota, etc., when writing to our
 713            own journal. The values are left 0 when recovering a journal
 714            different from our own. */
 715         if (!(flags & GFS2_LOG_HEAD_RECOVERY)) {
 716                 lh->lh_statfs_addr =
 717                         cpu_to_be64(GFS2_I(sdp->sd_sc_inode)->i_no_addr);
 718                 lh->lh_quota_addr =
 719                         cpu_to_be64(GFS2_I(sdp->sd_qc_inode)->i_no_addr);
 720
 721                 spin_lock(&sdp->sd_statfs_spin);
 722                 lh->lh_local_total = cpu_to_be64(l_sc->sc_total);
 723                 lh->lh_local_free = cpu_to_be64(l_sc->sc_free);
 724                 lh->lh_local_dinodes = cpu_to_be64(l_sc->sc_dinodes);
 725                 spin_unlock(&sdp->sd_statfs_spin);
 726         }
 727
 728         BUILD_BUG_ON(offsetof(struct gfs2_log_header, lh_crc) != LH_V1_SIZE);
 729
 730         crc = crc32c(~0, (void *)lh + LH_V1_SIZE + 4,
 731                      sb->s_blocksize - LH_V1_SIZE - 4);
 732         lh->lh_crc = cpu_to_be32(crc);
 733
 734         gfs2_log_write(sdp, page, sb->s_blocksize, 0, addr);
 735         gfs2_log_flush_bio(sdp, REQ_OP_WRITE, op_flags);
 736         log_flush_wait(sdp);
 737 }
 738
 739 /**
 740  * log_write_header - Get and initialize a journal header buffer
 741  * @sdp: The GFS2 superblock
 742  * @flags: The log header flags, including log header origin
 743  *
 744  * Returns: the initialized log buffer descriptor
 745  */
 746
 747 static void log_write_header(struct gfs2_sbd *sdp, u32 flags)
 748 {
 749         unsigned int tail;
 750         int op_flags = REQ_PREFLUSH | REQ_FUA | REQ_META | REQ_SYNC;
 751         enum gfs2_freeze_state state = atomic_read(&sdp->sd_freeze_state);
 752
 753         gfs2_assert_withdraw(sdp, (state != SFS_FROZEN));
 754         tail = current_tail(sdp);
 755
 756         if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags)) {
 757                 gfs2_ordered_wait(sdp);
 758                 log_flush_wait(sdp);
 759                 op_flags = REQ_SYNC | REQ_META | REQ_PRIO;
 760         }
 761         sdp->sd_log_idle = (tail == sdp->sd_log_flush_head);
 762         gfs2_write_log_header(sdp, sdp->sd_jdesc, sdp->sd_log_sequence++, tail,
 763                               flags, op_flags);
 764
 765         if (sdp->sd_log_tail != tail)
 766                 log_pull_tail(sdp, tail);
 767 }
 768
 769 /**
 770  * gfs2_log_flush - flush incore transaction(s)
 771  * @sdp: the filesystem
 772  * @gl: The glock structure to flush.  If NULL, flush the whole incore log
 773  * @flags: The log header flags: GFS2_LOG_HEAD_FLUSH_* and debug flags
 774  *
 775  */
 776
 777 void gfs2_log_flush(struct gfs2_sbd *sdp, struct gfs2_glock *gl, u32 flags)
 778 {
 779         struct gfs2_trans *tr;
 780         enum gfs2_freeze_state state = atomic_read(&sdp->sd_freeze_state);
 781
 782         down_write(&sdp->sd_log_flush_lock);
 783
 784         /* Log might have been flushed while we waited for the flush lock */
 785         if (gl && !test_bit(GLF_LFLUSH, &gl->gl_flags)) {
 786                 up_write(&sdp->sd_log_flush_lock);
 787                 return;
 788         }
 789         trace_gfs2_log_flush(sdp, 1, flags);
 790
 791         if (flags & GFS2_LOG_HEAD_FLUSH_SHUTDOWN)
 792                 clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 793
 794         sdp->sd_log_flush_head = sdp->sd_log_head;
 795         tr = sdp->sd_log_tr;
 796         if (tr) {
 797                 sdp->sd_log_tr = NULL;
 798                 INIT_LIST_HEAD(&tr->tr_ail1_list);
 799                 INIT_LIST_HEAD(&tr->tr_ail2_list);
 800                 tr->tr_first = sdp->sd_log_flush_head;
 801                 if (unlikely (state == SFS_FROZEN))
 802                         gfs2_assert_withdraw(sdp, !tr->tr_num_buf_new && !tr->tr_num_databuf_new);
 803         }
 804
 805         if (unlikely(state == SFS_FROZEN))
 806                 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
 807         gfs2_assert_withdraw(sdp,
 808                         sdp->sd_log_num_revoke == sdp->sd_log_commited_revoke);
 809
 810         gfs2_ordered_write(sdp);
 811         lops_before_commit(sdp, tr);
 812         gfs2_log_flush_bio(sdp, REQ_OP_WRITE, 0);
 813
 814         if (sdp->sd_log_head != sdp->sd_log_flush_head) {
 815                 log_flush_wait(sdp);
 816                 log_write_header(sdp, flags);
 817         } else if (sdp->sd_log_tail != current_tail(sdp) && !sdp->sd_log_idle){
 818                 atomic_dec(&sdp->sd_log_blks_free); /* Adjust for unreserved buffer */
 819                 trace_gfs2_log_blocks(sdp, -1);
 820                 log_write_header(sdp, flags);
 821         }
 822         lops_after_commit(sdp, tr);
 823
 824         gfs2_log_lock(sdp);
 825         sdp->sd_log_head = sdp->sd_log_flush_head;
 826         sdp->sd_log_blks_reserved = 0;
 827         sdp->sd_log_commited_revoke = 0;
 828
 829         spin_lock(&sdp->sd_ail_lock);
 830         if (tr && !list_empty(&tr->tr_ail1_list)) {
 831                 list_add(&tr->tr_list, &sdp->sd_ail1_list);
 832                 tr = NULL;
 833         }
 834         spin_unlock(&sdp->sd_ail_lock);
 835         gfs2_log_unlock(sdp);
 836
 837         if (!(flags & GFS2_LOG_HEAD_FLUSH_NORMAL)) {
 838                 if (!sdp->sd_log_idle) {
 839                         for (;;) {
 840                                 gfs2_ail1_start(sdp);
 841                                 gfs2_ail1_wait(sdp);
 842                                 if (gfs2_ail1_empty(sdp))
 843                                         break;
 844                         }
 845                         atomic_dec(&sdp->sd_log_blks_free); /* Adjust for unreserved buffer */
 846                         trace_gfs2_log_blocks(sdp, -1);
 847                         log_write_header(sdp, flags);
 848                         sdp->sd_log_head = sdp->sd_log_flush_head;
 849                 }
 850                 if (flags & (GFS2_LOG_HEAD_FLUSH_SHUTDOWN |
 851                              GFS2_LOG_HEAD_FLUSH_FREEZE))
 852                         gfs2_log_shutdown(sdp);
 853                 if (flags & GFS2_LOG_HEAD_FLUSH_FREEZE)
 854                         atomic_set(&sdp->sd_freeze_state, SFS_FROZEN);
 855         }
 856
 857         trace_gfs2_log_flush(sdp, 0, flags);
 858         up_write(&sdp->sd_log_flush_lock);
 859
 860         kfree(tr);
 861 }
 862
 863 /**
 864  * gfs2_merge_trans - Merge a new transaction into a cached transaction
 865  * @old: Original transaction to be expanded
 866  * @new: New transaction to be merged
 867  */
 868
 869 static void gfs2_merge_trans(struct gfs2_trans *old, struct gfs2_trans *new)
 870 {
 871         WARN_ON_ONCE(!test_bit(TR_ATTACHED, &old->tr_flags));
 872
 873         old->tr_num_buf_new     += new->tr_num_buf_new;
 874         old->tr_num_databuf_new += new->tr_num_databuf_new;
 875         old->tr_num_buf_rm      += new->tr_num_buf_rm;
 876         old->tr_num_databuf_rm  += new->tr_num_databuf_rm;
 877         old->tr_num_revoke      += new->tr_num_revoke;
 878         old->tr_num_revoke_rm   += new->tr_num_revoke_rm;
 879
 880         list_splice_tail_init(&new->tr_databuf, &old->tr_databuf);
 881         list_splice_tail_init(&new->tr_buf, &old->tr_buf);
 882 }
 883
 884 static void log_refund(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 885 {
 886         unsigned int reserved;
 887         unsigned int unused;
 888         unsigned int maxres;
 889
 890         gfs2_log_lock(sdp);
 891
 892         if (sdp->sd_log_tr) {
 893                 gfs2_merge_trans(sdp->sd_log_tr, tr);
 894         } else if (tr->tr_num_buf_new || tr->tr_num_databuf_new) {
 895                 gfs2_assert_withdraw(sdp, test_bit(TR_ALLOCED, &tr->tr_flags));
 896                 sdp->sd_log_tr = tr;
 897                 set_bit(TR_ATTACHED, &tr->tr_flags);
 898         }
 899
 900         sdp->sd_log_commited_revoke += tr->tr_num_revoke - tr->tr_num_revoke_rm;
 901         reserved = calc_reserved(sdp);
 902         maxres = sdp->sd_log_blks_reserved + tr->tr_reserved;
 903         gfs2_assert_withdraw(sdp, maxres >= reserved);
 904         unused = maxres - reserved;
 905         atomic_add(unused, &sdp->sd_log_blks_free);
 906         trace_gfs2_log_blocks(sdp, unused);
 907         gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <=
 908                              sdp->sd_jdesc->jd_blocks);
 909         sdp->sd_log_blks_reserved = reserved;
 910
 911         gfs2_log_unlock(sdp);
 912 }
 913
 914 /**
 915  * gfs2_log_commit - Commit a transaction to the log
 916  * @sdp: the filesystem
 917  * @tr: the transaction
 918  *
 919  * We wake up gfs2_logd if the number of pinned blocks exceed thresh1
 920  * or the total number of used blocks (pinned blocks plus AIL blocks)
 921  * is greater than thresh2.
 922  *
 923  * At mount time thresh1 is 1/3rd of journal size, thresh2 is 2/3rd of
 924  * journal size.
 925  *
 926  * Returns: errno
 927  */
 928
 929 void gfs2_log_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 930 {
 931         log_refund(sdp, tr);
 932
 933         if (atomic_read(&sdp->sd_log_pinned) > atomic_read(&sdp->sd_log_thresh1) ||
 934             ((sdp->sd_jdesc->jd_blocks - atomic_read(&sdp->sd_log_blks_free)) >
 935             atomic_read(&sdp->sd_log_thresh2)))
 936                 wake_up(&sdp->sd_logd_waitq);
 937 }
 938
 939 /**
 940  * gfs2_log_shutdown - write a shutdown header into a journal
 941  * @sdp: the filesystem
 942  *
 943  */
 944
 945 void gfs2_log_shutdown(struct gfs2_sbd *sdp)
 946 {
 947         gfs2_assert_withdraw(sdp, !sdp->sd_log_blks_reserved);
 948         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
 949         gfs2_assert_withdraw(sdp, list_empty(&sdp->sd_ail1_list));
 950
 951         sdp->sd_log_flush_head = sdp->sd_log_head;
 952
 953         log_write_header(sdp, GFS2_LOG_HEAD_UNMOUNT | GFS2_LFC_SHUTDOWN);
 954
 955         gfs2_assert_warn(sdp, sdp->sd_log_head == sdp->sd_log_tail);
 956         gfs2_assert_warn(sdp, list_empty(&sdp->sd_ail2_list));
 957
 958         sdp->sd_log_head = sdp->sd_log_flush_head;
 959         sdp->sd_log_tail = sdp->sd_log_head;
 960 }
 961
 962 static inline int gfs2_jrnl_flush_reqd(struct gfs2_sbd *sdp)
 963 {
 964         return (atomic_read(&sdp->sd_log_pinned) +
 965                 atomic_read(&sdp->sd_log_blks_needed) >=
 966                 atomic_read(&sdp->sd_log_thresh1));
 967 }
 968
 969 static inline int gfs2_ail_flush_reqd(struct gfs2_sbd *sdp)
 970 {
 971         unsigned int used_blocks = sdp->sd_jdesc->jd_blocks - atomic_read(&sdp->sd_log_blks_free);
 972
 973         if (test_and_clear_bit(SDF_FORCE_AIL_FLUSH, &sdp->sd_flags))
 974                 return 1;
 975
 976         return used_blocks + atomic_read(&sdp->sd_log_blks_needed) >=
 977                 atomic_read(&sdp->sd_log_thresh2);
 978 }
 979
 980 /**
 981  * gfs2_logd - Update log tail as Active Items get flushed to in-place blocks
 982  * @sdp: Pointer to GFS2 superblock
 983  *
 984  * Also, periodically check to make sure that we're using the most recent
 985  * journal index.
 986  */
 987
 988 int gfs2_logd(void *data)
 989 {
 990         struct gfs2_sbd *sdp = data;
 991         unsigned long t = 1;
 992         DEFINE_WAIT(wait);
 993         bool did_flush;
 994
 995         while (!kthread_should_stop()) {
 996
 997                 /* Check for errors writing to the journal */
 998                 if (sdp->sd_log_error) {
 999                         gfs2_lm_withdraw(sdp,
1000                                          "GFS2: fsid=%s: error %d: "
1001                                          "withdrawing the file system to "
1002                                          "prevent further damage.\n",
1003                                          sdp->sd_fsname, sdp->sd_log_error);
1004                 }
1005
1006                 did_flush = false;
1007                 if (gfs2_jrnl_flush_reqd(sdp) || t == 0) {
1008                         gfs2_ail1_empty(sdp);
1009                         gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
1010                                        GFS2_LFC_LOGD_JFLUSH_REQD);
1011                         did_flush = true;
1012                 }
1013
1014                 if (gfs2_ail_flush_reqd(sdp)) {
1015                         gfs2_ail1_start(sdp);
1016                         gfs2_ail1_wait(sdp);
1017                         gfs2_ail1_empty(sdp);
1018                         gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
1019                                        GFS2_LFC_LOGD_AIL_FLUSH_REQD);
1020                         did_flush = true;
1021                 }
1022
1023                 if (!gfs2_ail_flush_reqd(sdp) || did_flush)
1024                         wake_up(&sdp->sd_log_waitq);
1025
1026                 t = gfs2_tune_get(sdp, gt_logd_secs) * HZ;
1027
1028                 try_to_freeze();
1029
1030                 do {
1031                         prepare_to_wait(&sdp->sd_logd_waitq, &wait,
1032                                         TASK_INTERRUPTIBLE);
1033                         if (!gfs2_ail_flush_reqd(sdp) &&
1034                             !gfs2_jrnl_flush_reqd(sdp) &&
1035                             !kthread_should_stop())
1036                                 t = schedule_timeout(t);
1037                 } while(t && !gfs2_ail_flush_reqd(sdp) &&
1038                         !gfs2_jrnl_flush_reqd(sdp) &&
1039                         !kthread_should_stop());
1040                 finish_wait(&sdp->sd_logd_waitq, &wait);
1041         }
1042
1043         return 0;
1044 }
1045