fs/gfs2/log.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   4  * Copyright (C) 2004-2007 Red Hat, Inc.  All rights reserved.
   5  */
   6
   7 #include <linux/sched.h>
   8 #include <linux/slab.h>
   9 #include <linux/spinlock.h>
  10 #include <linux/completion.h>
  11 #include <linux/buffer_head.h>
  12 #include <linux/gfs2_ondisk.h>
  13 #include <linux/crc32.h>
  14 #include <linux/crc32c.h>
  15 #include <linux/delay.h>
  16 #include <linux/kthread.h>
  17 #include <linux/freezer.h>
  18 #include <linux/bio.h>
  19 #include <linux/blkdev.h>
  20 #include <linux/writeback.h>
  21 #include <linux/list_sort.h>
  22
  23 #include "gfs2.h"
  24 #include "incore.h"
  25 #include "bmap.h"
  26 #include "glock.h"
  27 #include "log.h"
  28 #include "lops.h"
  29 #include "meta_io.h"
  30 #include "util.h"
  31 #include "dir.h"
  32 #include "trace_gfs2.h"
  33
  34 /**
  35  * gfs2_struct2blk - compute stuff
  36  * @sdp: the filesystem
  37  * @nstruct: the number of structures
  38  * @ssize: the size of the structures
  39  *
  40  * Compute the number of log descriptor blocks needed to hold a certain number
  41  * of structures of a certain size.
  42  *
  43  * Returns: the number of blocks needed (minimum is always 1)
  44  */
  45
  46 unsigned int gfs2_struct2blk(struct gfs2_sbd *sdp, unsigned int nstruct,
  47                              unsigned int ssize)
  48 {
  49         unsigned int blks;
  50         unsigned int first, second;
  51
  52         blks = 1;
  53         first = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) / ssize;
  54
  55         if (nstruct > first) {
  56                 second = (sdp->sd_sb.sb_bsize -
  57                           sizeof(struct gfs2_meta_header)) / ssize;
  58                 blks += DIV_ROUND_UP(nstruct - first, second);
  59         }
  60
  61         return blks;
  62 }
  63
  64 /**
  65  * gfs2_remove_from_ail - Remove an entry from the ail lists, updating counters
  66  * @mapping: The associated mapping (maybe NULL)
  67  * @bd: The gfs2_bufdata to remove
  68  *
  69  * The ail lock _must_ be held when calling this function
  70  *
  71  */
  72
  73 static void gfs2_remove_from_ail(struct gfs2_bufdata *bd)
  74 {
  75         bd->bd_tr = NULL;
  76         list_del_init(&bd->bd_ail_st_list);
  77         list_del_init(&bd->bd_ail_gl_list);
  78         atomic_dec(&bd->bd_gl->gl_ail_count);
  79         brelse(bd->bd_bh);
  80 }
  81
  82 /**
  83  * gfs2_ail1_start_one - Start I/O on a part of the AIL
  84  * @sdp: the filesystem
  85  * @wbc: The writeback control structure
  86  * @ai: The ail structure
  87  *
  88  */
  89
  90 static int gfs2_ail1_start_one(struct gfs2_sbd *sdp,
  91                                struct writeback_control *wbc,
  92                                struct gfs2_trans *tr,
  93                                bool *withdraw)
  94 __releases(&sdp->sd_ail_lock)
  95 __acquires(&sdp->sd_ail_lock)
  96 {
  97         struct gfs2_glock *gl = NULL;
  98         struct address_space *mapping;
  99         struct gfs2_bufdata *bd, *s;
 100         struct buffer_head *bh;
 101
 102         list_for_each_entry_safe_reverse(bd, s, &tr->tr_ail1_list, bd_ail_st_list) {
 103                 bh = bd->bd_bh;
 104
 105                 gfs2_assert(sdp, bd->bd_tr == tr);
 106
 107                 if (!buffer_busy(bh)) {
 108                         if (!buffer_uptodate(bh) &&
 109                             !test_and_set_bit(SDF_AIL1_IO_ERROR,
 110                                               &sdp->sd_flags)) {
 111                                 gfs2_io_error_bh(sdp, bh);
 112                                 *withdraw = true;
 113                         }
 114                         list_move(&bd->bd_ail_st_list, &tr->tr_ail2_list);
 115                         continue;
 116                 }
 117
 118                 if (!buffer_dirty(bh))
 119                         continue;
 120                 if (gl == bd->bd_gl)
 121                         continue;
 122                 gl = bd->bd_gl;
 123                 list_move(&bd->bd_ail_st_list, &tr->tr_ail1_list);
 124                 mapping = bh->b_page->mapping;
 125                 if (!mapping)
 126                         continue;
 127                 spin_unlock(&sdp->sd_ail_lock);
 128                 generic_writepages(mapping, wbc);
 129                 spin_lock(&sdp->sd_ail_lock);
 130                 if (wbc->nr_to_write <= 0)
 131                         break;
 132                 return 1;
 133         }
 134
 135         return 0;
 136 }
 137
 138
 139 /**
 140  * gfs2_ail1_flush - start writeback of some ail1 entries
 141  * @sdp: The super block
 142  * @wbc: The writeback control structure
 143  *
 144  * Writes back some ail1 entries, according to the limits in the
 145  * writeback control structure
 146  */
 147
 148 void gfs2_ail1_flush(struct gfs2_sbd *sdp, struct writeback_control *wbc)
 149 {
 150         struct list_head *head = &sdp->sd_ail1_list;
 151         struct gfs2_trans *tr;
 152         struct blk_plug plug;
 153         bool withdraw = false;
 154
 155         trace_gfs2_ail_flush(sdp, wbc, 1);
 156         blk_start_plug(&plug);
 157         spin_lock(&sdp->sd_ail_lock);
 158 restart:
 159         list_for_each_entry_reverse(tr, head, tr_list) {
 160                 if (wbc->nr_to_write <= 0)
 161                         break;
 162                 if (gfs2_ail1_start_one(sdp, wbc, tr, &withdraw))
 163                         goto restart;
 164         }
 165         spin_unlock(&sdp->sd_ail_lock);
 166         blk_finish_plug(&plug);
 167         if (withdraw)
 168                 gfs2_lm_withdraw(sdp, NULL);
 169         trace_gfs2_ail_flush(sdp, wbc, 0);
 170 }
 171
 172 /**
 173  * gfs2_ail1_start - start writeback of all ail1 entries
 174  * @sdp: The superblock
 175  */
 176
 177 static void gfs2_ail1_start(struct gfs2_sbd *sdp)
 178 {
 179         struct writeback_control wbc = {
 180                 .sync_mode = WB_SYNC_NONE,
 181                 .nr_to_write = LONG_MAX,
 182                 .range_start = 0,
 183                 .range_end = LLONG_MAX,
 184         };
 185
 186         return gfs2_ail1_flush(sdp, &wbc);
 187 }
 188
 189 /**
 190  * gfs2_ail1_empty_one - Check whether or not a trans in the AIL has been synced
 191  * @sdp: the filesystem
 192  * @ai: the AIL entry
 193  *
 194  */
 195
 196 static void gfs2_ail1_empty_one(struct gfs2_sbd *sdp, struct gfs2_trans *tr,
 197                                 bool *withdraw)
 198 {
 199         struct gfs2_bufdata *bd, *s;
 200         struct buffer_head *bh;
 201
 202         list_for_each_entry_safe_reverse(bd, s, &tr->tr_ail1_list,
 203                                          bd_ail_st_list) {
 204                 bh = bd->bd_bh;
 205                 gfs2_assert(sdp, bd->bd_tr == tr);
 206                 if (buffer_busy(bh))
 207                         continue;
 208                 if (!buffer_uptodate(bh) &&
 209                     !test_and_set_bit(SDF_AIL1_IO_ERROR, &sdp->sd_flags)) {
 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 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_ordered, &ip_cmp);
 552         while (!list_empty(&sdp->sd_log_ordered)) {
 553                 ip = list_entry(sdp->sd_log_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_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_ordered)) {
 574                 ip = list_entry(sdp->sd_log_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         sdp->sd_log_num_revoke++;
 602         if (atomic_inc_return(&gl->gl_revokes) == 1)
 603                 gfs2_glock_hold(gl);
 604         bh->b_private = NULL;
 605         bd->bd_blkno = bh->b_blocknr;
 606         gfs2_remove_from_ail(bd); /* drops ref on bh */
 607         bd->bd_bh = NULL;
 608         set_bit(GLF_LFLUSH, &gl->gl_flags);
 609         list_add(&bd->bd_list, &sdp->sd_log_revokes);
 610 }
 611
 612 void gfs2_glock_remove_revoke(struct gfs2_glock *gl)
 613 {
 614         if (atomic_dec_return(&gl->gl_revokes) == 0) {
 615                 clear_bit(GLF_LFLUSH, &gl->gl_flags);
 616                 gfs2_glock_queue_put(gl);
 617         }
 618 }
 619
 620 void gfs2_write_revokes(struct gfs2_sbd *sdp)
 621 {
 622         struct gfs2_trans *tr;
 623         struct gfs2_bufdata *bd, *tmp;
 624         int have_revokes = 0;
 625         int max_revokes = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) / sizeof(u64);
 626
 627         gfs2_ail1_empty(sdp);
 628         spin_lock(&sdp->sd_ail_lock);
 629         list_for_each_entry_reverse(tr, &sdp->sd_ail1_list, tr_list) {
 630                 list_for_each_entry(bd, &tr->tr_ail2_list, bd_ail_st_list) {
 631                         if (list_empty(&bd->bd_list)) {
 632                                 have_revokes = 1;
 633                                 goto done;
 634                         }
 635                 }
 636         }
 637 done:
 638         spin_unlock(&sdp->sd_ail_lock);
 639         if (have_revokes == 0)
 640                 return;
 641         while (sdp->sd_log_num_revoke > max_revokes)
 642                 max_revokes += (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header)) / sizeof(u64);
 643         max_revokes -= sdp->sd_log_num_revoke;
 644         if (!sdp->sd_log_num_revoke) {
 645                 atomic_dec(&sdp->sd_log_blks_free);
 646                 /* If no blocks have been reserved, we need to also
 647                  * reserve a block for the header */
 648                 if (!sdp->sd_log_blks_reserved)
 649                         atomic_dec(&sdp->sd_log_blks_free);
 650         }
 651         gfs2_log_lock(sdp);
 652         spin_lock(&sdp->sd_ail_lock);
 653         list_for_each_entry_reverse(tr, &sdp->sd_ail1_list, tr_list) {
 654                 list_for_each_entry_safe(bd, tmp, &tr->tr_ail2_list, bd_ail_st_list) {
 655                         if (max_revokes == 0)
 656                                 goto out_of_blocks;
 657                         if (!list_empty(&bd->bd_list))
 658                                 continue;
 659                         gfs2_add_revoke(sdp, bd);
 660                         max_revokes--;
 661                 }
 662         }
 663 out_of_blocks:
 664         spin_unlock(&sdp->sd_ail_lock);
 665         gfs2_log_unlock(sdp);
 666
 667         if (!sdp->sd_log_num_revoke) {
 668                 atomic_inc(&sdp->sd_log_blks_free);
 669                 if (!sdp->sd_log_blks_reserved)
 670                         atomic_inc(&sdp->sd_log_blks_free);
 671         }
 672 }
 673
 674 /**
 675  * gfs2_write_log_header - Write a journal log header buffer at lblock
 676  * @sdp: The GFS2 superblock
 677  * @jd: journal descriptor of the journal to which we are writing
 678  * @seq: sequence number
 679  * @tail: tail of the log
 680  * @lblock: value for lh_blkno (block number relative to start of journal)
 681  * @flags: log header flags GFS2_LOG_HEAD_*
 682  * @op_flags: flags to pass to the bio
 683  *
 684  * Returns: the initialized log buffer descriptor
 685  */
 686
 687 void gfs2_write_log_header(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd,
 688                            u64 seq, u32 tail, u32 lblock, u32 flags,
 689                            int op_flags)
 690 {
 691         struct gfs2_log_header *lh;
 692         u32 hash, crc;
 693         struct page *page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
 694         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 695         struct timespec64 tv;
 696         struct super_block *sb = sdp->sd_vfs;
 697         u64 dblock;
 698
 699         lh = page_address(page);
 700         clear_page(lh);
 701
 702         lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 703         lh->lh_header.mh_type = cpu_to_be32(GFS2_METATYPE_LH);
 704         lh->lh_header.__pad0 = cpu_to_be64(0);
 705         lh->lh_header.mh_format = cpu_to_be32(GFS2_FORMAT_LH);
 706         lh->lh_header.mh_jid = cpu_to_be32(sdp->sd_jdesc->jd_jid);
 707         lh->lh_sequence = cpu_to_be64(seq);
 708         lh->lh_flags = cpu_to_be32(flags);
 709         lh->lh_tail = cpu_to_be32(tail);
 710         lh->lh_blkno = cpu_to_be32(lblock);
 711         hash = ~crc32(~0, lh, LH_V1_SIZE);
 712         lh->lh_hash = cpu_to_be32(hash);
 713
 714         ktime_get_coarse_real_ts64(&tv);
 715         lh->lh_nsec = cpu_to_be32(tv.tv_nsec);
 716         lh->lh_sec = cpu_to_be64(tv.tv_sec);
 717         if (!list_empty(&jd->extent_list))
 718                 dblock = gfs2_log_bmap(sdp);
 719         else {
 720                 int ret = gfs2_lblk_to_dblk(jd->jd_inode, lblock, &dblock);
 721                 if (gfs2_assert_withdraw(sdp, ret == 0))
 722                         return;
 723         }
 724         lh->lh_addr = cpu_to_be64(dblock);
 725         lh->lh_jinode = cpu_to_be64(GFS2_I(jd->jd_inode)->i_no_addr);
 726
 727         /* We may only write local statfs, quota, etc., when writing to our
 728            own journal. The values are left 0 when recovering a journal
 729            different from our own. */
 730         if (!(flags & GFS2_LOG_HEAD_RECOVERY)) {
 731                 lh->lh_statfs_addr =
 732                         cpu_to_be64(GFS2_I(sdp->sd_sc_inode)->i_no_addr);
 733                 lh->lh_quota_addr =
 734                         cpu_to_be64(GFS2_I(sdp->sd_qc_inode)->i_no_addr);
 735
 736                 spin_lock(&sdp->sd_statfs_spin);
 737                 lh->lh_local_total = cpu_to_be64(l_sc->sc_total);
 738                 lh->lh_local_free = cpu_to_be64(l_sc->sc_free);
 739                 lh->lh_local_dinodes = cpu_to_be64(l_sc->sc_dinodes);
 740                 spin_unlock(&sdp->sd_statfs_spin);
 741         }
 742
 743         BUILD_BUG_ON(offsetof(struct gfs2_log_header, lh_crc) != LH_V1_SIZE);
 744
 745         crc = crc32c(~0, (void *)lh + LH_V1_SIZE + 4,
 746                      sb->s_blocksize - LH_V1_SIZE - 4);
 747         lh->lh_crc = cpu_to_be32(crc);
 748
 749         gfs2_log_write(sdp, page, sb->s_blocksize, 0, dblock);
 750         gfs2_log_submit_bio(&sdp->sd_log_bio, REQ_OP_WRITE | op_flags);
 751         log_flush_wait(sdp);
 752 }
 753
 754 /**
 755  * log_write_header - Get and initialize a journal header buffer
 756  * @sdp: The GFS2 superblock
 757  * @flags: The log header flags, including log header origin
 758  *
 759  * Returns: the initialized log buffer descriptor
 760  */
 761
 762 static void log_write_header(struct gfs2_sbd *sdp, u32 flags)
 763 {
 764         unsigned int tail;
 765         int op_flags = REQ_PREFLUSH | REQ_FUA | REQ_META | REQ_SYNC;
 766         enum gfs2_freeze_state state = atomic_read(&sdp->sd_freeze_state);
 767
 768         gfs2_assert_withdraw(sdp, (state != SFS_FROZEN));
 769         tail = current_tail(sdp);
 770
 771         if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags)) {
 772                 gfs2_ordered_wait(sdp);
 773                 log_flush_wait(sdp);
 774                 op_flags = REQ_SYNC | REQ_META | REQ_PRIO;
 775         }
 776         sdp->sd_log_idle = (tail == sdp->sd_log_flush_head);
 777         gfs2_write_log_header(sdp, sdp->sd_jdesc, sdp->sd_log_sequence++, tail,
 778                               sdp->sd_log_flush_head, flags, op_flags);
 779
 780         if (sdp->sd_log_tail != tail)
 781                 log_pull_tail(sdp, tail);
 782 }
 783
 784 /**
 785  * gfs2_log_flush - flush incore transaction(s)
 786  * @sdp: the filesystem
 787  * @gl: The glock structure to flush.  If NULL, flush the whole incore log
 788  * @flags: The log header flags: GFS2_LOG_HEAD_FLUSH_* and debug flags
 789  *
 790  */
 791
 792 void gfs2_log_flush(struct gfs2_sbd *sdp, struct gfs2_glock *gl, u32 flags)
 793 {
 794         struct gfs2_trans *tr;
 795         enum gfs2_freeze_state state = atomic_read(&sdp->sd_freeze_state);
 796
 797         down_write(&sdp->sd_log_flush_lock);
 798
 799         /* Log might have been flushed while we waited for the flush lock */
 800         if (gl && !test_bit(GLF_LFLUSH, &gl->gl_flags)) {
 801                 up_write(&sdp->sd_log_flush_lock);
 802                 return;
 803         }
 804         trace_gfs2_log_flush(sdp, 1, flags);
 805
 806         if (flags & GFS2_LOG_HEAD_FLUSH_SHUTDOWN)
 807                 clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 808
 809         sdp->sd_log_flush_head = sdp->sd_log_head;
 810         tr = sdp->sd_log_tr;
 811         if (tr) {
 812                 sdp->sd_log_tr = NULL;
 813                 tr->tr_first = sdp->sd_log_flush_head;
 814                 if (unlikely (state == SFS_FROZEN))
 815                         gfs2_assert_withdraw(sdp, !tr->tr_num_buf_new && !tr->tr_num_databuf_new);
 816         }
 817
 818         if (unlikely(state == SFS_FROZEN))
 819                 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
 820         gfs2_assert_withdraw(sdp,
 821                         sdp->sd_log_num_revoke == sdp->sd_log_commited_revoke);
 822
 823         gfs2_ordered_write(sdp);
 824         lops_before_commit(sdp, tr);
 825         gfs2_log_submit_bio(&sdp->sd_log_bio, REQ_OP_WRITE);
 826
 827         if (sdp->sd_log_head != sdp->sd_log_flush_head) {
 828                 log_flush_wait(sdp);
 829                 log_write_header(sdp, flags);
 830         } else if (sdp->sd_log_tail != current_tail(sdp) && !sdp->sd_log_idle){
 831                 atomic_dec(&sdp->sd_log_blks_free); /* Adjust for unreserved buffer */
 832                 trace_gfs2_log_blocks(sdp, -1);
 833                 log_write_header(sdp, flags);
 834         }
 835         lops_after_commit(sdp, tr);
 836
 837         gfs2_log_lock(sdp);
 838         sdp->sd_log_head = sdp->sd_log_flush_head;
 839         sdp->sd_log_blks_reserved = 0;
 840         sdp->sd_log_commited_revoke = 0;
 841
 842         spin_lock(&sdp->sd_ail_lock);
 843         if (tr && !list_empty(&tr->tr_ail1_list)) {
 844                 list_add(&tr->tr_list, &sdp->sd_ail1_list);
 845                 tr = NULL;
 846         }
 847         spin_unlock(&sdp->sd_ail_lock);
 848         gfs2_log_unlock(sdp);
 849
 850         if (!(flags & GFS2_LOG_HEAD_FLUSH_NORMAL)) {
 851                 if (!sdp->sd_log_idle) {
 852                         for (;;) {
 853                                 gfs2_ail1_start(sdp);
 854                                 gfs2_ail1_wait(sdp);
 855                                 if (gfs2_ail1_empty(sdp))
 856                                         break;
 857                         }
 858                         atomic_dec(&sdp->sd_log_blks_free); /* Adjust for unreserved buffer */
 859                         trace_gfs2_log_blocks(sdp, -1);
 860                         log_write_header(sdp, flags);
 861                         sdp->sd_log_head = sdp->sd_log_flush_head;
 862                 }
 863                 if (flags & (GFS2_LOG_HEAD_FLUSH_SHUTDOWN |
 864                              GFS2_LOG_HEAD_FLUSH_FREEZE))
 865                         gfs2_log_shutdown(sdp);
 866                 if (flags & GFS2_LOG_HEAD_FLUSH_FREEZE)
 867                         atomic_set(&sdp->sd_freeze_state, SFS_FROZEN);
 868         }
 869
 870         trace_gfs2_log_flush(sdp, 0, flags);
 871         up_write(&sdp->sd_log_flush_lock);
 872
 873         kfree(tr);
 874 }
 875
 876 /**
 877  * gfs2_merge_trans - Merge a new transaction into a cached transaction
 878  * @old: Original transaction to be expanded
 879  * @new: New transaction to be merged
 880  */
 881
 882 static void gfs2_merge_trans(struct gfs2_sbd *sdp, struct gfs2_trans *new)
 883 {
 884         struct gfs2_trans *old = sdp->sd_log_tr;
 885
 886         WARN_ON_ONCE(!test_bit(TR_ATTACHED, &old->tr_flags));
 887
 888         old->tr_num_buf_new     += new->tr_num_buf_new;
 889         old->tr_num_databuf_new += new->tr_num_databuf_new;
 890         old->tr_num_buf_rm      += new->tr_num_buf_rm;
 891         old->tr_num_databuf_rm  += new->tr_num_databuf_rm;
 892         old->tr_num_revoke      += new->tr_num_revoke;
 893
 894         list_splice_tail_init(&new->tr_databuf, &old->tr_databuf);
 895         list_splice_tail_init(&new->tr_buf, &old->tr_buf);
 896
 897         spin_lock(&sdp->sd_ail_lock);
 898         list_splice_tail_init(&new->tr_ail1_list, &old->tr_ail1_list);
 899         list_splice_tail_init(&new->tr_ail2_list, &old->tr_ail2_list);
 900         spin_unlock(&sdp->sd_ail_lock);
 901 }
 902
 903 static void log_refund(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 904 {
 905         unsigned int reserved;
 906         unsigned int unused;
 907         unsigned int maxres;
 908
 909         gfs2_log_lock(sdp);
 910
 911         if (sdp->sd_log_tr) {
 912                 gfs2_merge_trans(sdp, tr);
 913         } else if (tr->tr_num_buf_new || tr->tr_num_databuf_new) {
 914                 gfs2_assert_withdraw(sdp, test_bit(TR_ALLOCED, &tr->tr_flags));
 915                 sdp->sd_log_tr = tr;
 916                 set_bit(TR_ATTACHED, &tr->tr_flags);
 917         }
 918
 919         sdp->sd_log_commited_revoke += tr->tr_num_revoke;
 920         reserved = calc_reserved(sdp);
 921         maxres = sdp->sd_log_blks_reserved + tr->tr_reserved;
 922         gfs2_assert_withdraw(sdp, maxres >= reserved);
 923         unused = maxres - reserved;
 924         atomic_add(unused, &sdp->sd_log_blks_free);
 925         trace_gfs2_log_blocks(sdp, unused);
 926         gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <=
 927                              sdp->sd_jdesc->jd_blocks);
 928         sdp->sd_log_blks_reserved = reserved;
 929
 930         gfs2_log_unlock(sdp);
 931 }
 932
 933 /**
 934  * gfs2_log_commit - Commit a transaction to the log
 935  * @sdp: the filesystem
 936  * @tr: the transaction
 937  *
 938  * We wake up gfs2_logd if the number of pinned blocks exceed thresh1
 939  * or the total number of used blocks (pinned blocks plus AIL blocks)
 940  * is greater than thresh2.
 941  *
 942  * At mount time thresh1 is 1/3rd of journal size, thresh2 is 2/3rd of
 943  * journal size.
 944  *
 945  * Returns: errno
 946  */
 947
 948 void gfs2_log_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 949 {
 950         log_refund(sdp, tr);
 951
 952         if (atomic_read(&sdp->sd_log_pinned) > atomic_read(&sdp->sd_log_thresh1) ||
 953             ((sdp->sd_jdesc->jd_blocks - atomic_read(&sdp->sd_log_blks_free)) >
 954             atomic_read(&sdp->sd_log_thresh2)))
 955                 wake_up(&sdp->sd_logd_waitq);
 956 }
 957
 958 /**
 959  * gfs2_log_shutdown - write a shutdown header into a journal
 960  * @sdp: the filesystem
 961  *
 962  */
 963
 964 void gfs2_log_shutdown(struct gfs2_sbd *sdp)
 965 {
 966         gfs2_assert_withdraw(sdp, !sdp->sd_log_blks_reserved);
 967         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
 968         gfs2_assert_withdraw(sdp, list_empty(&sdp->sd_ail1_list));
 969
 970         sdp->sd_log_flush_head = sdp->sd_log_head;
 971
 972         log_write_header(sdp, GFS2_LOG_HEAD_UNMOUNT | GFS2_LFC_SHUTDOWN);
 973
 974         gfs2_assert_warn(sdp, sdp->sd_log_head == sdp->sd_log_tail);
 975         gfs2_assert_warn(sdp, list_empty(&sdp->sd_ail2_list));
 976
 977         sdp->sd_log_head = sdp->sd_log_flush_head;
 978         sdp->sd_log_tail = sdp->sd_log_head;
 979 }
 980
 981 static inline int gfs2_jrnl_flush_reqd(struct gfs2_sbd *sdp)
 982 {
 983         return (atomic_read(&sdp->sd_log_pinned) +
 984                 atomic_read(&sdp->sd_log_blks_needed) >=
 985                 atomic_read(&sdp->sd_log_thresh1));
 986 }
 987
 988 static inline int gfs2_ail_flush_reqd(struct gfs2_sbd *sdp)
 989 {
 990         unsigned int used_blocks = sdp->sd_jdesc->jd_blocks - atomic_read(&sdp->sd_log_blks_free);
 991
 992         if (test_and_clear_bit(SDF_FORCE_AIL_FLUSH, &sdp->sd_flags))
 993                 return 1;
 994
 995         return used_blocks + atomic_read(&sdp->sd_log_blks_needed) >=
 996                 atomic_read(&sdp->sd_log_thresh2);
 997 }
 998
 999 /**
1000  * gfs2_logd - Update log tail as Active Items get flushed to in-place blocks
1001  * @sdp: Pointer to GFS2 superblock
1002  *
1003  * Also, periodically check to make sure that we're using the most recent
1004  * journal index.
1005  */
1006
1007 int gfs2_logd(void *data)
1008 {
1009         struct gfs2_sbd *sdp = data;
1010         unsigned long t = 1;
1011         DEFINE_WAIT(wait);
1012         bool did_flush;
1013
1014         while (!kthread_should_stop()) {
1015
1016                 /* Check for errors writing to the journal */
1017                 if (sdp->sd_log_error) {
1018                         gfs2_lm_withdraw(sdp,
1019                                          "GFS2: fsid=%s: error %d: "
1020                                          "withdrawing the file system to "
1021                                          "prevent further damage.\n",
1022                                          sdp->sd_fsname, sdp->sd_log_error);
1023                 }
1024
1025                 did_flush = false;
1026                 if (gfs2_jrnl_flush_reqd(sdp) || t == 0) {
1027                         gfs2_ail1_empty(sdp);
1028                         gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
1029                                        GFS2_LFC_LOGD_JFLUSH_REQD);
1030                         did_flush = true;
1031                 }
1032
1033                 if (gfs2_ail_flush_reqd(sdp)) {
1034                         gfs2_ail1_start(sdp);
1035                         gfs2_ail1_wait(sdp);
1036                         gfs2_ail1_empty(sdp);
1037                         gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
1038                                        GFS2_LFC_LOGD_AIL_FLUSH_REQD);
1039                         did_flush = true;
1040                 }
1041
1042                 if (!gfs2_ail_flush_reqd(sdp) || did_flush)
1043                         wake_up(&sdp->sd_log_waitq);
1044
1045                 t = gfs2_tune_get(sdp, gt_logd_secs) * HZ;
1046
1047                 try_to_freeze();
1048
1049                 do {
1050                         prepare_to_wait(&sdp->sd_logd_waitq, &wait,
1051                                         TASK_INTERRUPTIBLE);
1052                         if (!gfs2_ail_flush_reqd(sdp) &&
1053                             !gfs2_jrnl_flush_reqd(sdp) &&
1054                             !kthread_should_stop())
1055                                 t = schedule_timeout(t);
1056                 } while(t && !gfs2_ail_flush_reqd(sdp) &&
1057                         !gfs2_jrnl_flush_reqd(sdp) &&
1058                         !kthread_should_stop());
1059                 finish_wait(&sdp->sd_logd_waitq, &wait);
1060         }
1061
1062         return 0;
1063 }
1064