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 __releases(&sdp->sd_ail_lock)
  97 __acquires(&sdp->sd_ail_lock)
  98 {
  99         struct gfs2_glock *gl = NULL;
 100         struct address_space *mapping;
 101         struct gfs2_bufdata *bd, *s;
 102         struct buffer_head *bh;
 103
 104         list_for_each_entry_safe_reverse(bd, s, &tr->tr_ail1_list, bd_ail_st_list) {
 105                 bh = bd->bd_bh;
 106
 107                 gfs2_assert(sdp, bd->bd_tr == tr);
 108
 109                 if (!buffer_busy(bh)) {
 110                         if (!buffer_uptodate(bh))
 111                                 gfs2_io_error_bh(sdp, bh);
 112                         list_move(&bd->bd_ail_st_list, &tr->tr_ail2_list);
 113                         continue;
 114                 }
 115
 116                 if (!buffer_dirty(bh))
 117                         continue;
 118                 if (gl == bd->bd_gl)
 119                         continue;
 120                 gl = bd->bd_gl;
 121                 list_move(&bd->bd_ail_st_list, &tr->tr_ail1_list);
 122                 mapping = bh->b_page->mapping;
 123                 if (!mapping)
 124                         continue;
 125                 spin_unlock(&sdp->sd_ail_lock);
 126                 generic_writepages(mapping, wbc);
 127                 spin_lock(&sdp->sd_ail_lock);
 128                 if (wbc->nr_to_write <= 0)
 129                         break;
 130                 return 1;
 131         }
 132
 133         return 0;
 134 }
 135
 136
 137 /**
 138  * gfs2_ail1_flush - start writeback of some ail1 entries
 139  * @sdp: The super block
 140  * @wbc: The writeback control structure
 141  *
 142  * Writes back some ail1 entries, according to the limits in the
 143  * writeback control structure
 144  */
 145
 146 void gfs2_ail1_flush(struct gfs2_sbd *sdp, struct writeback_control *wbc)
 147 {
 148         struct list_head *head = &sdp->sd_ail1_list;
 149         struct gfs2_trans *tr;
 150         struct blk_plug plug;
 151
 152         trace_gfs2_ail_flush(sdp, wbc, 1);
 153         blk_start_plug(&plug);
 154         spin_lock(&sdp->sd_ail_lock);
 155 restart:
 156         list_for_each_entry_reverse(tr, head, tr_list) {
 157                 if (wbc->nr_to_write <= 0)
 158                         break;
 159                 if (gfs2_ail1_start_one(sdp, wbc, tr))
 160                         goto restart;
 161         }
 162         spin_unlock(&sdp->sd_ail_lock);
 163         blk_finish_plug(&plug);
 164         trace_gfs2_ail_flush(sdp, wbc, 0);
 165 }
 166
 167 /**
 168  * gfs2_ail1_start - start writeback of all ail1 entries
 169  * @sdp: The superblock
 170  */
 171
 172 static void gfs2_ail1_start(struct gfs2_sbd *sdp)
 173 {
 174         struct writeback_control wbc = {
 175                 .sync_mode = WB_SYNC_NONE,
 176                 .nr_to_write = LONG_MAX,
 177                 .range_start = 0,
 178                 .range_end = LLONG_MAX,
 179         };
 180
 181         return gfs2_ail1_flush(sdp, &wbc);
 182 }
 183
 184 /**
 185  * gfs2_ail1_empty_one - Check whether or not a trans in the AIL has been synced
 186  * @sdp: the filesystem
 187  * @ai: the AIL entry
 188  *
 189  */
 190
 191 static void gfs2_ail1_empty_one(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 192 {
 193         struct gfs2_bufdata *bd, *s;
 194         struct buffer_head *bh;
 195
 196         list_for_each_entry_safe_reverse(bd, s, &tr->tr_ail1_list,
 197                                          bd_ail_st_list) {
 198                 bh = bd->bd_bh;
 199                 gfs2_assert(sdp, bd->bd_tr == tr);
 200                 if (buffer_busy(bh))
 201                         continue;
 202                 if (!buffer_uptodate(bh))
 203                         gfs2_io_error_bh(sdp, bh);
 204                 list_move(&bd->bd_ail_st_list, &tr->tr_ail2_list);
 205         }
 206
 207 }
 208
 209 /**
 210  * gfs2_ail1_empty - Try to empty the ail1 lists
 211  * @sdp: The superblock
 212  *
 213  * Tries to empty the ail1 lists, starting with the oldest first
 214  */
 215
 216 static int gfs2_ail1_empty(struct gfs2_sbd *sdp)
 217 {
 218         struct gfs2_trans *tr, *s;
 219         int oldest_tr = 1;
 220         int ret;
 221
 222         spin_lock(&sdp->sd_ail_lock);
 223         list_for_each_entry_safe_reverse(tr, s, &sdp->sd_ail1_list, tr_list) {
 224                 gfs2_ail1_empty_one(sdp, tr);
 225                 if (list_empty(&tr->tr_ail1_list) && oldest_tr)
 226                         list_move(&tr->tr_list, &sdp->sd_ail2_list);
 227                 else
 228                         oldest_tr = 0;
 229         }
 230         ret = list_empty(&sdp->sd_ail1_list);
 231         spin_unlock(&sdp->sd_ail_lock);
 232
 233         return ret;
 234 }
 235
 236 static void gfs2_ail1_wait(struct gfs2_sbd *sdp)
 237 {
 238         struct gfs2_trans *tr;
 239         struct gfs2_bufdata *bd;
 240         struct buffer_head *bh;
 241
 242         spin_lock(&sdp->sd_ail_lock);
 243         list_for_each_entry_reverse(tr, &sdp->sd_ail1_list, tr_list) {
 244                 list_for_each_entry(bd, &tr->tr_ail1_list, bd_ail_st_list) {
 245                         bh = bd->bd_bh;
 246                         if (!buffer_locked(bh))
 247                                 continue;
 248                         get_bh(bh);
 249                         spin_unlock(&sdp->sd_ail_lock);
 250                         wait_on_buffer(bh);
 251                         brelse(bh);
 252                         return;
 253                 }
 254         }
 255         spin_unlock(&sdp->sd_ail_lock);
 256 }
 257
 258 /**
 259  * gfs2_ail2_empty_one - Check whether or not a trans in the AIL has been synced
 260  * @sdp: the filesystem
 261  * @ai: the AIL entry
 262  *
 263  */
 264
 265 static void gfs2_ail2_empty_one(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 266 {
 267         struct list_head *head = &tr->tr_ail2_list;
 268         struct gfs2_bufdata *bd;
 269
 270         while (!list_empty(head)) {
 271                 bd = list_entry(head->prev, struct gfs2_bufdata,
 272                                 bd_ail_st_list);
 273                 gfs2_assert(sdp, bd->bd_tr == tr);
 274                 gfs2_remove_from_ail(bd);
 275         }
 276 }
 277
 278 static void ail2_empty(struct gfs2_sbd *sdp, unsigned int new_tail)
 279 {
 280         struct gfs2_trans *tr, *safe;
 281         unsigned int old_tail = sdp->sd_log_tail;
 282         int wrap = (new_tail < old_tail);
 283         int a, b, rm;
 284
 285         spin_lock(&sdp->sd_ail_lock);
 286
 287         list_for_each_entry_safe(tr, safe, &sdp->sd_ail2_list, tr_list) {
 288                 a = (old_tail <= tr->tr_first);
 289                 b = (tr->tr_first < new_tail);
 290                 rm = (wrap) ? (a || b) : (a && b);
 291                 if (!rm)
 292                         continue;
 293
 294                 gfs2_ail2_empty_one(sdp, tr);
 295                 list_del(&tr->tr_list);
 296                 gfs2_assert_warn(sdp, list_empty(&tr->tr_ail1_list));
 297                 gfs2_assert_warn(sdp, list_empty(&tr->tr_ail2_list));
 298                 kfree(tr);
 299         }
 300
 301         spin_unlock(&sdp->sd_ail_lock);
 302 }
 303
 304 /**
 305  * gfs2_log_release - Release a given number of log blocks
 306  * @sdp: The GFS2 superblock
 307  * @blks: The number of blocks
 308  *
 309  */
 310
 311 void gfs2_log_release(struct gfs2_sbd *sdp, unsigned int blks)
 312 {
 313
 314         atomic_add(blks, &sdp->sd_log_blks_free);
 315         trace_gfs2_log_blocks(sdp, blks);
 316         gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <=
 317                                   sdp->sd_jdesc->jd_blocks);
 318         up_read(&sdp->sd_log_flush_lock);
 319 }
 320
 321 /**
 322  * gfs2_log_reserve - Make a log reservation
 323  * @sdp: The GFS2 superblock
 324  * @blks: The number of blocks to reserve
 325  *
 326  * Note that we never give out the last few blocks of the journal. Thats
 327  * due to the fact that there is a small number of header blocks
 328  * associated with each log flush. The exact number can't be known until
 329  * flush time, so we ensure that we have just enough free blocks at all
 330  * times to avoid running out during a log flush.
 331  *
 332  * We no longer flush the log here, instead we wake up logd to do that
 333  * for us. To avoid the thundering herd and to ensure that we deal fairly
 334  * with queued waiters, we use an exclusive wait. This means that when we
 335  * get woken with enough journal space to get our reservation, we need to
 336  * wake the next waiter on the list.
 337  *
 338  * Returns: errno
 339  */
 340
 341 int gfs2_log_reserve(struct gfs2_sbd *sdp, unsigned int blks)
 342 {
 343         int ret = 0;
 344         unsigned reserved_blks = 7 * (4096 / sdp->sd_vfs->s_blocksize);
 345         unsigned wanted = blks + reserved_blks;
 346         DEFINE_WAIT(wait);
 347         int did_wait = 0;
 348         unsigned int free_blocks;
 349
 350         if (gfs2_assert_warn(sdp, blks) ||
 351             gfs2_assert_warn(sdp, blks <= sdp->sd_jdesc->jd_blocks))
 352                 return -EINVAL;
 353         atomic_add(blks, &sdp->sd_log_blks_needed);
 354 retry:
 355         free_blocks = atomic_read(&sdp->sd_log_blks_free);
 356         if (unlikely(free_blocks <= wanted)) {
 357                 do {
 358                         prepare_to_wait_exclusive(&sdp->sd_log_waitq, &wait,
 359                                         TASK_UNINTERRUPTIBLE);
 360                         wake_up(&sdp->sd_logd_waitq);
 361                         did_wait = 1;
 362                         if (atomic_read(&sdp->sd_log_blks_free) <= wanted)
 363                                 io_schedule();
 364                         free_blocks = atomic_read(&sdp->sd_log_blks_free);
 365                 } while(free_blocks <= wanted);
 366                 finish_wait(&sdp->sd_log_waitq, &wait);
 367         }
 368         atomic_inc(&sdp->sd_reserving_log);
 369         if (atomic_cmpxchg(&sdp->sd_log_blks_free, free_blocks,
 370                                 free_blocks - blks) != free_blocks) {
 371                 if (atomic_dec_and_test(&sdp->sd_reserving_log))
 372                         wake_up(&sdp->sd_reserving_log_wait);
 373                 goto retry;
 374         }
 375         atomic_sub(blks, &sdp->sd_log_blks_needed);
 376         trace_gfs2_log_blocks(sdp, -blks);
 377
 378         /*
 379          * If we waited, then so might others, wake them up _after_ we get
 380          * our share of the log.
 381          */
 382         if (unlikely(did_wait))
 383                 wake_up(&sdp->sd_log_waitq);
 384
 385         down_read(&sdp->sd_log_flush_lock);
 386         if (unlikely(!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))) {
 387                 gfs2_log_release(sdp, blks);
 388                 ret = -EROFS;
 389         }
 390         if (atomic_dec_and_test(&sdp->sd_reserving_log))
 391                 wake_up(&sdp->sd_reserving_log_wait);
 392         return ret;
 393 }
 394
 395 /**
 396  * log_distance - Compute distance between two journal blocks
 397  * @sdp: The GFS2 superblock
 398  * @newer: The most recent journal block of the pair
 399  * @older: The older journal block of the pair
 400  *
 401  *   Compute the distance (in the journal direction) between two
 402  *   blocks in the journal
 403  *
 404  * Returns: the distance in blocks
 405  */
 406
 407 static inline unsigned int log_distance(struct gfs2_sbd *sdp, unsigned int newer,
 408                                         unsigned int older)
 409 {
 410         int dist;
 411
 412         dist = newer - older;
 413         if (dist < 0)
 414                 dist += sdp->sd_jdesc->jd_blocks;
 415
 416         return dist;
 417 }
 418
 419 /**
 420  * calc_reserved - Calculate the number of blocks to reserve when
 421  *                 refunding a transaction's unused buffers.
 422  * @sdp: The GFS2 superblock
 423  *
 424  * This is complex.  We need to reserve room for all our currently used
 425  * metadata buffers (e.g. normal file I/O rewriting file time stamps) and
 426  * all our journaled data buffers for journaled files (e.g. files in the
 427  * meta_fs like rindex, or files for which chattr +j was done.)
 428  * If we don't reserve enough space, gfs2_log_refund and gfs2_log_flush
 429  * will count it as free space (sd_log_blks_free) and corruption will follow.
 430  *
 431  * We can have metadata bufs and jdata bufs in the same journal.  So each
 432  * type gets its own log header, for which we need to reserve a block.
 433  * In fact, each type has the potential for needing more than one header
 434  * in cases where we have more buffers than will fit on a journal page.
 435  * Metadata journal entries take up half the space of journaled buffer entries.
 436  * Thus, metadata entries have buf_limit (502) and journaled buffers have
 437  * databuf_limit (251) before they cause a wrap around.
 438  *
 439  * Also, we need to reserve blocks for revoke journal entries and one for an
 440  * overall header for the lot.
 441  *
 442  * Returns: the number of blocks reserved
 443  */
 444 static unsigned int calc_reserved(struct gfs2_sbd *sdp)
 445 {
 446         unsigned int reserved = 0;
 447         unsigned int mbuf;
 448         unsigned int dbuf;
 449         struct gfs2_trans *tr = sdp->sd_log_tr;
 450
 451         if (tr) {
 452                 mbuf = tr->tr_num_buf_new - tr->tr_num_buf_rm;
 453                 dbuf = tr->tr_num_databuf_new - tr->tr_num_databuf_rm;
 454                 reserved = mbuf + dbuf;
 455                 /* Account for header blocks */
 456                 reserved += DIV_ROUND_UP(mbuf, buf_limit(sdp));
 457                 reserved += DIV_ROUND_UP(dbuf, databuf_limit(sdp));
 458         }
 459
 460         if (sdp->sd_log_commited_revoke > 0)
 461                 reserved += gfs2_struct2blk(sdp, sdp->sd_log_commited_revoke,
 462                                           sizeof(u64));
 463         /* One for the overall header */
 464         if (reserved)
 465                 reserved++;
 466         return reserved;
 467 }
 468
 469 static unsigned int current_tail(struct gfs2_sbd *sdp)
 470 {
 471         struct gfs2_trans *tr;
 472         unsigned int tail;
 473
 474         spin_lock(&sdp->sd_ail_lock);
 475
 476         if (list_empty(&sdp->sd_ail1_list)) {
 477                 tail = sdp->sd_log_head;
 478         } else {
 479                 tr = list_entry(sdp->sd_ail1_list.prev, struct gfs2_trans,
 480                                 tr_list);
 481                 tail = tr->tr_first;
 482         }
 483
 484         spin_unlock(&sdp->sd_ail_lock);
 485
 486         return tail;
 487 }
 488
 489 static void log_pull_tail(struct gfs2_sbd *sdp, unsigned int new_tail)
 490 {
 491         unsigned int dist = log_distance(sdp, new_tail, sdp->sd_log_tail);
 492
 493         ail2_empty(sdp, new_tail);
 494
 495         atomic_add(dist, &sdp->sd_log_blks_free);
 496         trace_gfs2_log_blocks(sdp, dist);
 497         gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <=
 498                              sdp->sd_jdesc->jd_blocks);
 499
 500         sdp->sd_log_tail = new_tail;
 501 }
 502
 503
 504 static void log_flush_wait(struct gfs2_sbd *sdp)
 505 {
 506         DEFINE_WAIT(wait);
 507
 508         if (atomic_read(&sdp->sd_log_in_flight)) {
 509                 do {
 510                         prepare_to_wait(&sdp->sd_log_flush_wait, &wait,
 511                                         TASK_UNINTERRUPTIBLE);
 512                         if (atomic_read(&sdp->sd_log_in_flight))
 513                                 io_schedule();
 514                 } while(atomic_read(&sdp->sd_log_in_flight));
 515                 finish_wait(&sdp->sd_log_flush_wait, &wait);
 516         }
 517 }
 518
 519 static int ip_cmp(void *priv, struct list_head *a, struct list_head *b)
 520 {
 521         struct gfs2_inode *ipa, *ipb;
 522
 523         ipa = list_entry(a, struct gfs2_inode, i_ordered);
 524         ipb = list_entry(b, struct gfs2_inode, i_ordered);
 525
 526         if (ipa->i_no_addr < ipb->i_no_addr)
 527                 return -1;
 528         if (ipa->i_no_addr > ipb->i_no_addr)
 529                 return 1;
 530         return 0;
 531 }
 532
 533 static void gfs2_ordered_write(struct gfs2_sbd *sdp)
 534 {
 535         struct gfs2_inode *ip;
 536         LIST_HEAD(written);
 537
 538         spin_lock(&sdp->sd_ordered_lock);
 539         list_sort(NULL, &sdp->sd_log_le_ordered, &ip_cmp);
 540         while (!list_empty(&sdp->sd_log_le_ordered)) {
 541                 ip = list_entry(sdp->sd_log_le_ordered.next, struct gfs2_inode, i_ordered);
 542                 if (ip->i_inode.i_mapping->nrpages == 0) {
 543                         test_and_clear_bit(GIF_ORDERED, &ip->i_flags);
 544                         list_del(&ip->i_ordered);
 545                         continue;
 546                 }
 547                 list_move(&ip->i_ordered, &written);
 548                 spin_unlock(&sdp->sd_ordered_lock);
 549                 filemap_fdatawrite(ip->i_inode.i_mapping);
 550                 spin_lock(&sdp->sd_ordered_lock);
 551         }
 552         list_splice(&written, &sdp->sd_log_le_ordered);
 553         spin_unlock(&sdp->sd_ordered_lock);
 554 }
 555
 556 static void gfs2_ordered_wait(struct gfs2_sbd *sdp)
 557 {
 558         struct gfs2_inode *ip;
 559
 560         spin_lock(&sdp->sd_ordered_lock);
 561         while (!list_empty(&sdp->sd_log_le_ordered)) {
 562                 ip = list_entry(sdp->sd_log_le_ordered.next, struct gfs2_inode, i_ordered);
 563                 list_del(&ip->i_ordered);
 564                 WARN_ON(!test_and_clear_bit(GIF_ORDERED, &ip->i_flags));
 565                 if (ip->i_inode.i_mapping->nrpages == 0)
 566                         continue;
 567                 spin_unlock(&sdp->sd_ordered_lock);
 568                 filemap_fdatawait(ip->i_inode.i_mapping);
 569                 spin_lock(&sdp->sd_ordered_lock);
 570         }
 571         spin_unlock(&sdp->sd_ordered_lock);
 572 }
 573
 574 void gfs2_ordered_del_inode(struct gfs2_inode *ip)
 575 {
 576         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 577
 578         spin_lock(&sdp->sd_ordered_lock);
 579         if (test_and_clear_bit(GIF_ORDERED, &ip->i_flags))
 580                 list_del(&ip->i_ordered);
 581         spin_unlock(&sdp->sd_ordered_lock);
 582 }
 583
 584 void gfs2_add_revoke(struct gfs2_sbd *sdp, struct gfs2_bufdata *bd)
 585 {
 586         struct buffer_head *bh = bd->bd_bh;
 587         struct gfs2_glock *gl = bd->bd_gl;
 588
 589         bh->b_private = NULL;
 590         bd->bd_blkno = bh->b_blocknr;
 591         gfs2_remove_from_ail(bd); /* drops ref on bh */
 592         bd->bd_bh = NULL;
 593         bd->bd_ops = &gfs2_revoke_lops;
 594         sdp->sd_log_num_revoke++;
 595         atomic_inc(&gl->gl_revokes);
 596         set_bit(GLF_LFLUSH, &gl->gl_flags);
 597         list_add(&bd->bd_list, &sdp->sd_log_le_revoke);
 598 }
 599
 600 void gfs2_write_revokes(struct gfs2_sbd *sdp)
 601 {
 602         struct gfs2_trans *tr;
 603         struct gfs2_bufdata *bd, *tmp;
 604         int have_revokes = 0;
 605         int max_revokes = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) / sizeof(u64);
 606
 607         gfs2_ail1_empty(sdp);
 608         spin_lock(&sdp->sd_ail_lock);
 609         list_for_each_entry(tr, &sdp->sd_ail1_list, tr_list) {
 610                 list_for_each_entry(bd, &tr->tr_ail2_list, bd_ail_st_list) {
 611                         if (list_empty(&bd->bd_list)) {
 612                                 have_revokes = 1;
 613                                 goto done;
 614                         }
 615                 }
 616         }
 617 done:
 618         spin_unlock(&sdp->sd_ail_lock);
 619         if (have_revokes == 0)
 620                 return;
 621         while (sdp->sd_log_num_revoke > max_revokes)
 622                 max_revokes += (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header)) / sizeof(u64);
 623         max_revokes -= sdp->sd_log_num_revoke;
 624         if (!sdp->sd_log_num_revoke) {
 625                 atomic_dec(&sdp->sd_log_blks_free);
 626                 /* If no blocks have been reserved, we need to also
 627                  * reserve a block for the header */
 628                 if (!sdp->sd_log_blks_reserved)
 629                         atomic_dec(&sdp->sd_log_blks_free);
 630         }
 631         gfs2_log_lock(sdp);
 632         spin_lock(&sdp->sd_ail_lock);
 633         list_for_each_entry(tr, &sdp->sd_ail1_list, tr_list) {
 634                 list_for_each_entry_safe(bd, tmp, &tr->tr_ail2_list, bd_ail_st_list) {
 635                         if (max_revokes == 0)
 636                                 goto out_of_blocks;
 637                         if (!list_empty(&bd->bd_list))
 638                                 continue;
 639                         gfs2_add_revoke(sdp, bd);
 640                         max_revokes--;
 641                 }
 642         }
 643 out_of_blocks:
 644         spin_unlock(&sdp->sd_ail_lock);
 645         gfs2_log_unlock(sdp);
 646
 647         if (!sdp->sd_log_num_revoke) {
 648                 atomic_inc(&sdp->sd_log_blks_free);
 649                 if (!sdp->sd_log_blks_reserved)
 650                         atomic_inc(&sdp->sd_log_blks_free);
 651         }
 652 }
 653
 654 /**
 655  * write_log_header - Write a journal log header buffer at sd_log_flush_head
 656  * @sdp: The GFS2 superblock
 657  * @jd: journal descriptor of the journal to which we are writing
 658  * @seq: sequence number
 659  * @tail: tail of the log
 660  * @flags: log header flags GFS2_LOG_HEAD_*
 661  * @op_flags: flags to pass to the bio
 662  *
 663  * Returns: the initialized log buffer descriptor
 664  */
 665
 666 void gfs2_write_log_header(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd,
 667                            u64 seq, u32 tail, u32 flags, int op_flags)
 668 {
 669         struct gfs2_log_header *lh;
 670         u32 hash, crc;
 671         struct page *page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
 672         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 673         struct timespec64 tv;
 674         struct super_block *sb = sdp->sd_vfs;
 675         u64 addr;
 676
 677         lh = page_address(page);
 678         clear_page(lh);
 679
 680         lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 681         lh->lh_header.mh_type = cpu_to_be32(GFS2_METATYPE_LH);
 682         lh->lh_header.__pad0 = cpu_to_be64(0);
 683         lh->lh_header.mh_format = cpu_to_be32(GFS2_FORMAT_LH);
 684         lh->lh_header.mh_jid = cpu_to_be32(sdp->sd_jdesc->jd_jid);
 685         lh->lh_sequence = cpu_to_be64(seq);
 686         lh->lh_flags = cpu_to_be32(flags);
 687         lh->lh_tail = cpu_to_be32(tail);
 688         lh->lh_blkno = cpu_to_be32(sdp->sd_log_flush_head);
 689         hash = ~crc32(~0, lh, LH_V1_SIZE);
 690         lh->lh_hash = cpu_to_be32(hash);
 691
 692         tv = current_kernel_time64();
 693         lh->lh_nsec = cpu_to_be32(tv.tv_nsec);
 694         lh->lh_sec = cpu_to_be64(tv.tv_sec);
 695         addr = gfs2_log_bmap(sdp);
 696         lh->lh_addr = cpu_to_be64(addr);
 697         lh->lh_jinode = cpu_to_be64(GFS2_I(jd->jd_inode)->i_no_addr);
 698
 699         /* We may only write local statfs, quota, etc., when writing to our
 700            own journal. The values are left 0 when recovering a journal
 701            different from our own. */
 702         if (!(flags & GFS2_LOG_HEAD_RECOVERY)) {
 703                 lh->lh_statfs_addr =
 704                         cpu_to_be64(GFS2_I(sdp->sd_sc_inode)->i_no_addr);
 705                 lh->lh_quota_addr =
 706                         cpu_to_be64(GFS2_I(sdp->sd_qc_inode)->i_no_addr);
 707
 708                 spin_lock(&sdp->sd_statfs_spin);
 709                 lh->lh_local_total = cpu_to_be64(l_sc->sc_total);
 710                 lh->lh_local_free = cpu_to_be64(l_sc->sc_free);
 711                 lh->lh_local_dinodes = cpu_to_be64(l_sc->sc_dinodes);
 712                 spin_unlock(&sdp->sd_statfs_spin);
 713         }
 714
 715         BUILD_BUG_ON(offsetof(struct gfs2_log_header, lh_crc) != LH_V1_SIZE);
 716
 717         crc = crc32c(~0, (void *)lh + LH_V1_SIZE + 4,
 718                      sb->s_blocksize - LH_V1_SIZE - 4);
 719         lh->lh_crc = cpu_to_be32(crc);
 720
 721         gfs2_log_write(sdp, page, sb->s_blocksize, 0, addr);
 722         gfs2_log_flush_bio(sdp, REQ_OP_WRITE, op_flags);
 723         log_flush_wait(sdp);
 724 }
 725
 726 /**
 727  * log_write_header - Get and initialize a journal header buffer
 728  * @sdp: The GFS2 superblock
 729  * @flags: The log header flags, including log header origin
 730  *
 731  * Returns: the initialized log buffer descriptor
 732  */
 733
 734 static void log_write_header(struct gfs2_sbd *sdp, u32 flags)
 735 {
 736         unsigned int tail;
 737         int op_flags = REQ_PREFLUSH | REQ_FUA | REQ_META | REQ_SYNC;
 738         enum gfs2_freeze_state state = atomic_read(&sdp->sd_freeze_state);
 739
 740         gfs2_assert_withdraw(sdp, (state != SFS_FROZEN));
 741         tail = current_tail(sdp);
 742
 743         if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags)) {
 744                 gfs2_ordered_wait(sdp);
 745                 log_flush_wait(sdp);
 746                 op_flags = REQ_SYNC | REQ_META | REQ_PRIO;
 747         }
 748         sdp->sd_log_idle = (tail == sdp->sd_log_flush_head);
 749         gfs2_write_log_header(sdp, sdp->sd_jdesc, sdp->sd_log_sequence++, tail,
 750                               flags, op_flags);
 751
 752         if (sdp->sd_log_tail != tail)
 753                 log_pull_tail(sdp, tail);
 754 }
 755
 756 /**
 757  * gfs2_log_flush - flush incore transaction(s)
 758  * @sdp: the filesystem
 759  * @gl: The glock structure to flush.  If NULL, flush the whole incore log
 760  * @flags: The log header flags: GFS2_LOG_HEAD_FLUSH_* and debug flags
 761  *
 762  */
 763
 764 void gfs2_log_flush(struct gfs2_sbd *sdp, struct gfs2_glock *gl, u32 flags)
 765 {
 766         struct gfs2_trans *tr;
 767         enum gfs2_freeze_state state = atomic_read(&sdp->sd_freeze_state);
 768
 769         down_write(&sdp->sd_log_flush_lock);
 770
 771         /* Log might have been flushed while we waited for the flush lock */
 772         if (gl && !test_bit(GLF_LFLUSH, &gl->gl_flags)) {
 773                 up_write(&sdp->sd_log_flush_lock);
 774                 return;
 775         }
 776         trace_gfs2_log_flush(sdp, 1, flags);
 777
 778         if (flags & GFS2_LOG_HEAD_FLUSH_SHUTDOWN)
 779                 clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 780
 781         sdp->sd_log_flush_head = sdp->sd_log_head;
 782         tr = sdp->sd_log_tr;
 783         if (tr) {
 784                 sdp->sd_log_tr = NULL;
 785                 INIT_LIST_HEAD(&tr->tr_ail1_list);
 786                 INIT_LIST_HEAD(&tr->tr_ail2_list);
 787                 tr->tr_first = sdp->sd_log_flush_head;
 788                 if (unlikely (state == SFS_FROZEN))
 789                         gfs2_assert_withdraw(sdp, !tr->tr_num_buf_new && !tr->tr_num_databuf_new);
 790         }
 791
 792         if (unlikely(state == SFS_FROZEN))
 793                 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
 794         gfs2_assert_withdraw(sdp,
 795                         sdp->sd_log_num_revoke == sdp->sd_log_commited_revoke);
 796
 797         gfs2_ordered_write(sdp);
 798         lops_before_commit(sdp, tr);
 799         gfs2_log_flush_bio(sdp, REQ_OP_WRITE, 0);
 800
 801         if (sdp->sd_log_head != sdp->sd_log_flush_head) {
 802                 log_flush_wait(sdp);
 803                 log_write_header(sdp, flags);
 804         } else if (sdp->sd_log_tail != current_tail(sdp) && !sdp->sd_log_idle){
 805                 atomic_dec(&sdp->sd_log_blks_free); /* Adjust for unreserved buffer */
 806                 trace_gfs2_log_blocks(sdp, -1);
 807                 log_write_header(sdp, flags);
 808         }
 809         lops_after_commit(sdp, tr);
 810
 811         gfs2_log_lock(sdp);
 812         sdp->sd_log_head = sdp->sd_log_flush_head;
 813         sdp->sd_log_blks_reserved = 0;
 814         sdp->sd_log_commited_revoke = 0;
 815
 816         spin_lock(&sdp->sd_ail_lock);
 817         if (tr && !list_empty(&tr->tr_ail1_list)) {
 818                 list_add(&tr->tr_list, &sdp->sd_ail1_list);
 819                 tr = NULL;
 820         }
 821         spin_unlock(&sdp->sd_ail_lock);
 822         gfs2_log_unlock(sdp);
 823
 824         if (!(flags & GFS2_LOG_HEAD_FLUSH_NORMAL)) {
 825                 if (!sdp->sd_log_idle) {
 826                         for (;;) {
 827                                 gfs2_ail1_start(sdp);
 828                                 gfs2_ail1_wait(sdp);
 829                                 if (gfs2_ail1_empty(sdp))
 830                                         break;
 831                         }
 832                         atomic_dec(&sdp->sd_log_blks_free); /* Adjust for unreserved buffer */
 833                         trace_gfs2_log_blocks(sdp, -1);
 834                         log_write_header(sdp, flags);
 835                         sdp->sd_log_head = sdp->sd_log_flush_head;
 836                 }
 837                 if (flags & (GFS2_LOG_HEAD_FLUSH_SHUTDOWN |
 838                              GFS2_LOG_HEAD_FLUSH_FREEZE))
 839                         gfs2_log_shutdown(sdp);
 840                 if (flags & GFS2_LOG_HEAD_FLUSH_FREEZE)
 841                         atomic_set(&sdp->sd_freeze_state, SFS_FROZEN);
 842         }
 843
 844         trace_gfs2_log_flush(sdp, 0, flags);
 845         up_write(&sdp->sd_log_flush_lock);
 846
 847         kfree(tr);
 848 }
 849
 850 /**
 851  * gfs2_merge_trans - Merge a new transaction into a cached transaction
 852  * @old: Original transaction to be expanded
 853  * @new: New transaction to be merged
 854  */
 855
 856 static void gfs2_merge_trans(struct gfs2_trans *old, struct gfs2_trans *new)
 857 {
 858         WARN_ON_ONCE(!test_bit(TR_ATTACHED, &old->tr_flags));
 859
 860         old->tr_num_buf_new     += new->tr_num_buf_new;
 861         old->tr_num_databuf_new += new->tr_num_databuf_new;
 862         old->tr_num_buf_rm      += new->tr_num_buf_rm;
 863         old->tr_num_databuf_rm  += new->tr_num_databuf_rm;
 864         old->tr_num_revoke      += new->tr_num_revoke;
 865         old->tr_num_revoke_rm   += new->tr_num_revoke_rm;
 866
 867         list_splice_tail_init(&new->tr_databuf, &old->tr_databuf);
 868         list_splice_tail_init(&new->tr_buf, &old->tr_buf);
 869 }
 870
 871 static void log_refund(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 872 {
 873         unsigned int reserved;
 874         unsigned int unused;
 875         unsigned int maxres;
 876
 877         gfs2_log_lock(sdp);
 878
 879         if (sdp->sd_log_tr) {
 880                 gfs2_merge_trans(sdp->sd_log_tr, tr);
 881         } else if (tr->tr_num_buf_new || tr->tr_num_databuf_new) {
 882                 gfs2_assert_withdraw(sdp, test_bit(TR_ALLOCED, &tr->tr_flags));
 883                 sdp->sd_log_tr = tr;
 884                 set_bit(TR_ATTACHED, &tr->tr_flags);
 885         }
 886
 887         sdp->sd_log_commited_revoke += tr->tr_num_revoke - tr->tr_num_revoke_rm;
 888         reserved = calc_reserved(sdp);
 889         maxres = sdp->sd_log_blks_reserved + tr->tr_reserved;
 890         gfs2_assert_withdraw(sdp, maxres >= reserved);
 891         unused = maxres - reserved;
 892         atomic_add(unused, &sdp->sd_log_blks_free);
 893         trace_gfs2_log_blocks(sdp, unused);
 894         gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <=
 895                              sdp->sd_jdesc->jd_blocks);
 896         sdp->sd_log_blks_reserved = reserved;
 897
 898         gfs2_log_unlock(sdp);
 899 }
 900
 901 /**
 902  * gfs2_log_commit - Commit a transaction to the log
 903  * @sdp: the filesystem
 904  * @tr: the transaction
 905  *
 906  * We wake up gfs2_logd if the number of pinned blocks exceed thresh1
 907  * or the total number of used blocks (pinned blocks plus AIL blocks)
 908  * is greater than thresh2.
 909  *
 910  * At mount time thresh1 is 1/3rd of journal size, thresh2 is 2/3rd of
 911  * journal size.
 912  *
 913  * Returns: errno
 914  */
 915
 916 void gfs2_log_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 917 {
 918         log_refund(sdp, tr);
 919
 920         if (atomic_read(&sdp->sd_log_pinned) > atomic_read(&sdp->sd_log_thresh1) ||
 921             ((sdp->sd_jdesc->jd_blocks - atomic_read(&sdp->sd_log_blks_free)) >
 922             atomic_read(&sdp->sd_log_thresh2)))
 923                 wake_up(&sdp->sd_logd_waitq);
 924 }
 925
 926 /**
 927  * gfs2_log_shutdown - write a shutdown header into a journal
 928  * @sdp: the filesystem
 929  *
 930  */
 931
 932 void gfs2_log_shutdown(struct gfs2_sbd *sdp)
 933 {
 934         gfs2_assert_withdraw(sdp, !sdp->sd_log_blks_reserved);
 935         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
 936         gfs2_assert_withdraw(sdp, list_empty(&sdp->sd_ail1_list));
 937
 938         sdp->sd_log_flush_head = sdp->sd_log_head;
 939
 940         log_write_header(sdp, GFS2_LOG_HEAD_UNMOUNT | GFS2_LFC_SHUTDOWN);
 941
 942         gfs2_assert_warn(sdp, sdp->sd_log_head == sdp->sd_log_tail);
 943         gfs2_assert_warn(sdp, list_empty(&sdp->sd_ail2_list));
 944
 945         sdp->sd_log_head = sdp->sd_log_flush_head;
 946         sdp->sd_log_tail = sdp->sd_log_head;
 947 }
 948
 949 static inline int gfs2_jrnl_flush_reqd(struct gfs2_sbd *sdp)
 950 {
 951         return (atomic_read(&sdp->sd_log_pinned) +
 952                 atomic_read(&sdp->sd_log_blks_needed) >=
 953                 atomic_read(&sdp->sd_log_thresh1));
 954 }
 955
 956 static inline int gfs2_ail_flush_reqd(struct gfs2_sbd *sdp)
 957 {
 958         unsigned int used_blocks = sdp->sd_jdesc->jd_blocks - atomic_read(&sdp->sd_log_blks_free);
 959
 960         if (test_and_clear_bit(SDF_FORCE_AIL_FLUSH, &sdp->sd_flags))
 961                 return 1;
 962
 963         return used_blocks + atomic_read(&sdp->sd_log_blks_needed) >=
 964                 atomic_read(&sdp->sd_log_thresh2);
 965 }
 966
 967 /**
 968  * gfs2_logd - Update log tail as Active Items get flushed to in-place blocks
 969  * @sdp: Pointer to GFS2 superblock
 970  *
 971  * Also, periodically check to make sure that we're using the most recent
 972  * journal index.
 973  */
 974
 975 int gfs2_logd(void *data)
 976 {
 977         struct gfs2_sbd *sdp = data;
 978         unsigned long t = 1;
 979         DEFINE_WAIT(wait);
 980         bool did_flush;
 981
 982         while (!kthread_should_stop()) {
 983
 984                 /* Check for errors writing to the journal */
 985                 if (sdp->sd_log_error) {
 986                         gfs2_lm_withdraw(sdp,
 987                                          "GFS2: fsid=%s: error %d: "
 988                                          "withdrawing the file system to "
 989                                          "prevent further damage.\n",
 990                                          sdp->sd_fsname, sdp->sd_log_error);
 991                 }
 992
 993                 did_flush = false;
 994                 if (gfs2_jrnl_flush_reqd(sdp) || t == 0) {
 995                         gfs2_ail1_empty(sdp);
 996                         gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
 997                                        GFS2_LFC_LOGD_JFLUSH_REQD);
 998                         did_flush = true;
 999                 }
1000
1001                 if (gfs2_ail_flush_reqd(sdp)) {
1002                         gfs2_ail1_start(sdp);
1003                         gfs2_ail1_wait(sdp);
1004                         gfs2_ail1_empty(sdp);
1005                         gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
1006                                        GFS2_LFC_LOGD_AIL_FLUSH_REQD);
1007                         did_flush = true;
1008                 }
1009
1010                 if (!gfs2_ail_flush_reqd(sdp) || did_flush)
1011                         wake_up(&sdp->sd_log_waitq);
1012
1013                 t = gfs2_tune_get(sdp, gt_logd_secs) * HZ;
1014
1015                 try_to_freeze();
1016
1017                 do {
1018                         prepare_to_wait(&sdp->sd_logd_waitq, &wait,
1019                                         TASK_INTERRUPTIBLE);
1020                         if (!gfs2_ail_flush_reqd(sdp) &&
1021                             !gfs2_jrnl_flush_reqd(sdp) &&
1022                             !kthread_should_stop())
1023                                 t = schedule_timeout(t);
1024                 } while(t && !gfs2_ail_flush_reqd(sdp) &&
1025                         !gfs2_jrnl_flush_reqd(sdp) &&
1026                         !kthread_should_stop());
1027                 finish_wait(&sdp->sd_logd_waitq, &wait);
1028         }
1029
1030         return 0;
1031 }
1032