fs/gfs2/glops.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   4  * Copyright (C) 2004-2008 Red Hat, Inc.  All rights reserved.
   5  */
   6
   7 #include <linux/spinlock.h>
   8 #include <linux/completion.h>
   9 #include <linux/buffer_head.h>
  10 #include <linux/gfs2_ondisk.h>
  11 #include <linux/bio.h>
  12 #include <linux/posix_acl.h>
  13 #include <linux/security.h>
  14
  15 #include "gfs2.h"
  16 #include "incore.h"
  17 #include "bmap.h"
  18 #include "glock.h"
  19 #include "glops.h"
  20 #include "inode.h"
  21 #include "log.h"
  22 #include "meta_io.h"
  23 #include "recovery.h"
  24 #include "rgrp.h"
  25 #include "util.h"
  26 #include "trans.h"
  27 #include "dir.h"
  28 #include "lops.h"
  29
  30 struct workqueue_struct *gfs2_freeze_wq;
  31
  32 extern struct workqueue_struct *gfs2_control_wq;
  33
  34 static void gfs2_ail_error(struct gfs2_glock *gl, const struct buffer_head *bh)
  35 {
  36         fs_err(gl->gl_name.ln_sbd,
  37                "AIL buffer %p: blocknr %llu state 0x%08lx mapping %p page "
  38                "state 0x%lx\n",
  39                bh, (unsigned long long)bh->b_blocknr, bh->b_state,
  40                bh->b_page->mapping, bh->b_page->flags);
  41         fs_err(gl->gl_name.ln_sbd, "AIL glock %u:%llu mapping %p\n",
  42                gl->gl_name.ln_type, gl->gl_name.ln_number,
  43                gfs2_glock2aspace(gl));
  44         gfs2_lm(gl->gl_name.ln_sbd, "AIL error\n");
  45         gfs2_withdraw(gl->gl_name.ln_sbd);
  46 }
  47
  48 /**
  49  * __gfs2_ail_flush - remove all buffers for a given lock from the AIL
  50  * @gl: the glock
  51  * @fsync: set when called from fsync (not all buffers will be clean)
  52  *
  53  * None of the buffers should be dirty, locked, or pinned.
  54  */
  55
  56 static void __gfs2_ail_flush(struct gfs2_glock *gl, bool fsync,
  57                              unsigned int nr_revokes)
  58 {
  59         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
  60         struct list_head *head = &gl->gl_ail_list;
  61         struct gfs2_bufdata *bd, *tmp;
  62         struct buffer_head *bh;
  63         const unsigned long b_state = (1UL << BH_Dirty)|(1UL << BH_Pinned)|(1UL << BH_Lock);
  64
  65         gfs2_log_lock(sdp);
  66         spin_lock(&sdp->sd_ail_lock);
  67         list_for_each_entry_safe_reverse(bd, tmp, head, bd_ail_gl_list) {
  68                 if (nr_revokes == 0)
  69                         break;
  70                 bh = bd->bd_bh;
  71                 if (bh->b_state & b_state) {
  72                         if (fsync)
  73                                 continue;
  74                         gfs2_ail_error(gl, bh);
  75                 }
  76                 gfs2_trans_add_revoke(sdp, bd);
  77                 nr_revokes--;
  78         }
  79         GLOCK_BUG_ON(gl, !fsync && atomic_read(&gl->gl_ail_count));
  80         spin_unlock(&sdp->sd_ail_lock);
  81         gfs2_log_unlock(sdp);
  82 }
  83
  84
  85 static int gfs2_ail_empty_gl(struct gfs2_glock *gl)
  86 {
  87         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
  88         struct gfs2_trans tr;
  89         int ret;
  90
  91         memset(&tr, 0, sizeof(tr));
  92         INIT_LIST_HEAD(&tr.tr_buf);
  93         INIT_LIST_HEAD(&tr.tr_databuf);
  94         INIT_LIST_HEAD(&tr.tr_ail1_list);
  95         INIT_LIST_HEAD(&tr.tr_ail2_list);
  96         tr.tr_revokes = atomic_read(&gl->gl_ail_count);
  97
  98         if (!tr.tr_revokes) {
  99                 bool have_revokes;
 100                 bool log_in_flight;
 101
 102                 /*
 103                  * We have nothing on the ail, but there could be revokes on
 104                  * the sdp revoke queue, in which case, we still want to flush
 105                  * the log and wait for it to finish.
 106                  *
 107                  * If the sdp revoke list is empty too, we might still have an
 108                  * io outstanding for writing revokes, so we should wait for
 109                  * it before returning.
 110                  *
 111                  * If none of these conditions are true, our revokes are all
 112                  * flushed and we can return.
 113                  */
 114                 gfs2_log_lock(sdp);
 115                 have_revokes = !list_empty(&sdp->sd_log_revokes);
 116                 log_in_flight = atomic_read(&sdp->sd_log_in_flight);
 117                 gfs2_log_unlock(sdp);
 118                 if (have_revokes)
 119                         goto flush;
 120                 if (log_in_flight)
 121                         log_flush_wait(sdp);
 122                 return 0;
 123         }
 124
 125         /* A shortened, inline version of gfs2_trans_begin()
 126          * tr->alloced is not set since the transaction structure is
 127          * on the stack */
 128         tr.tr_reserved = 1 + gfs2_struct2blk(sdp, tr.tr_revokes);
 129         tr.tr_ip = _RET_IP_;
 130         ret = gfs2_log_reserve(sdp, tr.tr_reserved);
 131         if (ret < 0)
 132                 return ret;
 133         WARN_ON_ONCE(current->journal_info);
 134         current->journal_info = &tr;
 135
 136         __gfs2_ail_flush(gl, 0, tr.tr_revokes);
 137
 138         gfs2_trans_end(sdp);
 139 flush:
 140         gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
 141                        GFS2_LFC_AIL_EMPTY_GL);
 142         return 0;
 143 }
 144
 145 void gfs2_ail_flush(struct gfs2_glock *gl, bool fsync)
 146 {
 147         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 148         unsigned int revokes = atomic_read(&gl->gl_ail_count);
 149         unsigned int max_revokes = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) / sizeof(u64);
 150         int ret;
 151
 152         if (!revokes)
 153                 return;
 154
 155         while (revokes > max_revokes)
 156                 max_revokes += (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header)) / sizeof(u64);
 157
 158         ret = gfs2_trans_begin(sdp, 0, max_revokes);
 159         if (ret)
 160                 return;
 161         __gfs2_ail_flush(gl, fsync, max_revokes);
 162         gfs2_trans_end(sdp);
 163         gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
 164                        GFS2_LFC_AIL_FLUSH);
 165 }
 166
 167 /**
 168  * rgrp_go_sync - sync out the metadata for this glock
 169  * @gl: the glock
 170  *
 171  * Called when demoting or unlocking an EX glock.  We must flush
 172  * to disk all dirty buffers/pages relating to this glock, and must not
 173  * return to caller to demote/unlock the glock until I/O is complete.
 174  */
 175
 176 static int rgrp_go_sync(struct gfs2_glock *gl)
 177 {
 178         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 179         struct address_space *mapping = &sdp->sd_aspace;
 180         struct gfs2_rgrpd *rgd = gfs2_glock2rgrp(gl);
 181         const unsigned bsize = sdp->sd_sb.sb_bsize;
 182         loff_t start = (rgd->rd_addr * bsize) & PAGE_MASK;
 183         loff_t end = PAGE_ALIGN((rgd->rd_addr + rgd->rd_length) * bsize) - 1;
 184         int error;
 185
 186         if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags))
 187                 return 0;
 188         GLOCK_BUG_ON(gl, gl->gl_state != LM_ST_EXCLUSIVE);
 189
 190         gfs2_log_flush(sdp, gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
 191                        GFS2_LFC_RGRP_GO_SYNC);
 192         filemap_fdatawrite_range(mapping, start, end);
 193         error = filemap_fdatawait_range(mapping, start, end);
 194         WARN_ON_ONCE(error && !gfs2_withdrawn(sdp));
 195         mapping_set_error(mapping, error);
 196         if (!error)
 197                 error = gfs2_ail_empty_gl(gl);
 198         gfs2_free_clones(rgd);
 199         return error;
 200 }
 201
 202 /**
 203  * rgrp_go_inval - invalidate the metadata for this glock
 204  * @gl: the glock
 205  * @flags:
 206  *
 207  * We never used LM_ST_DEFERRED with resource groups, so that we
 208  * should always see the metadata flag set here.
 209  *
 210  */
 211
 212 static void rgrp_go_inval(struct gfs2_glock *gl, int flags)
 213 {
 214         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 215         struct address_space *mapping = &sdp->sd_aspace;
 216         struct gfs2_rgrpd *rgd = gfs2_glock2rgrp(gl);
 217         const unsigned bsize = sdp->sd_sb.sb_bsize;
 218         loff_t start = (rgd->rd_addr * bsize) & PAGE_MASK;
 219         loff_t end = PAGE_ALIGN((rgd->rd_addr + rgd->rd_length) * bsize) - 1;
 220
 221         gfs2_rgrp_brelse(rgd);
 222         WARN_ON_ONCE(!(flags & DIO_METADATA));
 223         truncate_inode_pages_range(mapping, start, end);
 224         rgd->rd_flags &= ~GFS2_RDF_UPTODATE;
 225 }
 226
 227 static void gfs2_rgrp_go_dump(struct seq_file *seq, struct gfs2_glock *gl,
 228                               const char *fs_id_buf)
 229 {
 230         struct gfs2_rgrpd *rgd = gfs2_glock2rgrp(gl);
 231
 232         if (rgd)
 233                 gfs2_rgrp_dump(seq, rgd, fs_id_buf);
 234 }
 235
 236 static struct gfs2_inode *gfs2_glock2inode(struct gfs2_glock *gl)
 237 {
 238         struct gfs2_inode *ip;
 239
 240         spin_lock(&gl->gl_lockref.lock);
 241         ip = gl->gl_object;
 242         if (ip)
 243                 set_bit(GIF_GLOP_PENDING, &ip->i_flags);
 244         spin_unlock(&gl->gl_lockref.lock);
 245         return ip;
 246 }
 247
 248 struct gfs2_rgrpd *gfs2_glock2rgrp(struct gfs2_glock *gl)
 249 {
 250         struct gfs2_rgrpd *rgd;
 251
 252         spin_lock(&gl->gl_lockref.lock);
 253         rgd = gl->gl_object;
 254         spin_unlock(&gl->gl_lockref.lock);
 255
 256         return rgd;
 257 }
 258
 259 static void gfs2_clear_glop_pending(struct gfs2_inode *ip)
 260 {
 261         if (!ip)
 262                 return;
 263
 264         clear_bit_unlock(GIF_GLOP_PENDING, &ip->i_flags);
 265         wake_up_bit(&ip->i_flags, GIF_GLOP_PENDING);
 266 }
 267
 268 /**
 269  * inode_go_sync - Sync the dirty data and/or metadata for an inode glock
 270  * @gl: the glock protecting the inode
 271  *
 272  */
 273
 274 static int inode_go_sync(struct gfs2_glock *gl)
 275 {
 276         struct gfs2_inode *ip = gfs2_glock2inode(gl);
 277         int isreg = ip && S_ISREG(ip->i_inode.i_mode);
 278         struct address_space *metamapping = gfs2_glock2aspace(gl);
 279         int error = 0, ret;
 280
 281         if (isreg) {
 282                 if (test_and_clear_bit(GIF_SW_PAGED, &ip->i_flags))
 283                         unmap_shared_mapping_range(ip->i_inode.i_mapping, 0, 0);
 284                 inode_dio_wait(&ip->i_inode);
 285         }
 286         if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags))
 287                 goto out;
 288
 289         GLOCK_BUG_ON(gl, gl->gl_state != LM_ST_EXCLUSIVE);
 290
 291         gfs2_log_flush(gl->gl_name.ln_sbd, gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
 292                        GFS2_LFC_INODE_GO_SYNC);
 293         filemap_fdatawrite(metamapping);
 294         if (isreg) {
 295                 struct address_space *mapping = ip->i_inode.i_mapping;
 296                 filemap_fdatawrite(mapping);
 297                 error = filemap_fdatawait(mapping);
 298                 mapping_set_error(mapping, error);
 299         }
 300         ret = filemap_fdatawait(metamapping);
 301         mapping_set_error(metamapping, ret);
 302         if (!error)
 303                 error = ret;
 304         gfs2_ail_empty_gl(gl);
 305         /*
 306          * Writeback of the data mapping may cause the dirty flag to be set
 307          * so we have to clear it again here.
 308          */
 309         smp_mb__before_atomic();
 310         clear_bit(GLF_DIRTY, &gl->gl_flags);
 311
 312 out:
 313         gfs2_clear_glop_pending(ip);
 314         return error;
 315 }
 316
 317 /**
 318  * inode_go_inval - prepare a inode glock to be released
 319  * @gl: the glock
 320  * @flags:
 321  *
 322  * Normally we invalidate everything, but if we are moving into
 323  * LM_ST_DEFERRED from LM_ST_SHARED or LM_ST_EXCLUSIVE then we
 324  * can keep hold of the metadata, since it won't have changed.
 325  *
 326  */
 327
 328 static void inode_go_inval(struct gfs2_glock *gl, int flags)
 329 {
 330         struct gfs2_inode *ip = gfs2_glock2inode(gl);
 331
 332         if (flags & DIO_METADATA) {
 333                 struct address_space *mapping = gfs2_glock2aspace(gl);
 334                 truncate_inode_pages(mapping, 0);
 335                 if (ip) {
 336                         set_bit(GIF_INVALID, &ip->i_flags);
 337                         forget_all_cached_acls(&ip->i_inode);
 338                         security_inode_invalidate_secctx(&ip->i_inode);
 339                         gfs2_dir_hash_inval(ip);
 340                 }
 341         }
 342
 343         if (ip == GFS2_I(gl->gl_name.ln_sbd->sd_rindex)) {
 344                 gfs2_log_flush(gl->gl_name.ln_sbd, NULL,
 345                                GFS2_LOG_HEAD_FLUSH_NORMAL |
 346                                GFS2_LFC_INODE_GO_INVAL);
 347                 gl->gl_name.ln_sbd->sd_rindex_uptodate = 0;
 348         }
 349         if (ip && S_ISREG(ip->i_inode.i_mode))
 350                 truncate_inode_pages(ip->i_inode.i_mapping, 0);
 351
 352         gfs2_clear_glop_pending(ip);
 353 }
 354
 355 /**
 356  * inode_go_demote_ok - Check to see if it's ok to unlock an inode glock
 357  * @gl: the glock
 358  *
 359  * Returns: 1 if it's ok
 360  */
 361
 362 static int inode_go_demote_ok(const struct gfs2_glock *gl)
 363 {
 364         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 365
 366         if (sdp->sd_jindex == gl->gl_object || sdp->sd_rindex == gl->gl_object)
 367                 return 0;
 368
 369         return 1;
 370 }
 371
 372 static int gfs2_dinode_in(struct gfs2_inode *ip, const void *buf)
 373 {
 374         const struct gfs2_dinode *str = buf;
 375         struct timespec64 atime;
 376         u16 height, depth;
 377
 378         if (unlikely(ip->i_no_addr != be64_to_cpu(str->di_num.no_addr)))
 379                 goto corrupt;
 380         ip->i_no_formal_ino = be64_to_cpu(str->di_num.no_formal_ino);
 381         ip->i_inode.i_mode = be32_to_cpu(str->di_mode);
 382         ip->i_inode.i_rdev = 0;
 383         switch (ip->i_inode.i_mode & S_IFMT) {
 384         case S_IFBLK:
 385         case S_IFCHR:
 386                 ip->i_inode.i_rdev = MKDEV(be32_to_cpu(str->di_major),
 387                                            be32_to_cpu(str->di_minor));
 388                 break;
 389         }
 390
 391         i_uid_write(&ip->i_inode, be32_to_cpu(str->di_uid));
 392         i_gid_write(&ip->i_inode, be32_to_cpu(str->di_gid));
 393         set_nlink(&ip->i_inode, be32_to_cpu(str->di_nlink));
 394         i_size_write(&ip->i_inode, be64_to_cpu(str->di_size));
 395         gfs2_set_inode_blocks(&ip->i_inode, be64_to_cpu(str->di_blocks));
 396         atime.tv_sec = be64_to_cpu(str->di_atime);
 397         atime.tv_nsec = be32_to_cpu(str->di_atime_nsec);
 398         if (timespec64_compare(&ip->i_inode.i_atime, &atime) < 0)
 399                 ip->i_inode.i_atime = atime;
 400         ip->i_inode.i_mtime.tv_sec = be64_to_cpu(str->di_mtime);
 401         ip->i_inode.i_mtime.tv_nsec = be32_to_cpu(str->di_mtime_nsec);
 402         ip->i_inode.i_ctime.tv_sec = be64_to_cpu(str->di_ctime);
 403         ip->i_inode.i_ctime.tv_nsec = be32_to_cpu(str->di_ctime_nsec);
 404
 405         ip->i_goal = be64_to_cpu(str->di_goal_meta);
 406         ip->i_generation = be64_to_cpu(str->di_generation);
 407
 408         ip->i_diskflags = be32_to_cpu(str->di_flags);
 409         ip->i_eattr = be64_to_cpu(str->di_eattr);
 410         /* i_diskflags and i_eattr must be set before gfs2_set_inode_flags() */
 411         gfs2_set_inode_flags(&ip->i_inode);
 412         height = be16_to_cpu(str->di_height);
 413         if (unlikely(height > GFS2_MAX_META_HEIGHT))
 414                 goto corrupt;
 415         ip->i_height = (u8)height;
 416
 417         depth = be16_to_cpu(str->di_depth);
 418         if (unlikely(depth > GFS2_DIR_MAX_DEPTH))
 419                 goto corrupt;
 420         ip->i_depth = (u8)depth;
 421         ip->i_entries = be32_to_cpu(str->di_entries);
 422
 423         if (S_ISREG(ip->i_inode.i_mode))
 424                 gfs2_set_aops(&ip->i_inode);
 425
 426         return 0;
 427 corrupt:
 428         gfs2_consist_inode(ip);
 429         return -EIO;
 430 }
 431
 432 /**
 433  * gfs2_inode_refresh - Refresh the incore copy of the dinode
 434  * @ip: The GFS2 inode
 435  *
 436  * Returns: errno
 437  */
 438
 439 int gfs2_inode_refresh(struct gfs2_inode *ip)
 440 {
 441         struct buffer_head *dibh;
 442         int error;
 443
 444         error = gfs2_meta_inode_buffer(ip, &dibh);
 445         if (error)
 446                 return error;
 447
 448         error = gfs2_dinode_in(ip, dibh->b_data);
 449         brelse(dibh);
 450         clear_bit(GIF_INVALID, &ip->i_flags);
 451
 452         return error;
 453 }
 454
 455 /**
 456  * inode_go_lock - operation done after an inode lock is locked by a process
 457  * @gl: the glock
 458  * @flags:
 459  *
 460  * Returns: errno
 461  */
 462
 463 static int inode_go_lock(struct gfs2_holder *gh)
 464 {
 465         struct gfs2_glock *gl = gh->gh_gl;
 466         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 467         struct gfs2_inode *ip = gl->gl_object;
 468         int error = 0;
 469
 470         if (!ip || (gh->gh_flags & GL_SKIP))
 471                 return 0;
 472
 473         if (test_bit(GIF_INVALID, &ip->i_flags)) {
 474                 error = gfs2_inode_refresh(ip);
 475                 if (error)
 476                         return error;
 477         }
 478
 479         if (gh->gh_state != LM_ST_DEFERRED)
 480                 inode_dio_wait(&ip->i_inode);
 481
 482         if ((ip->i_diskflags & GFS2_DIF_TRUNC_IN_PROG) &&
 483             (gl->gl_state == LM_ST_EXCLUSIVE) &&
 484             (gh->gh_state == LM_ST_EXCLUSIVE)) {
 485                 spin_lock(&sdp->sd_trunc_lock);
 486                 if (list_empty(&ip->i_trunc_list))
 487                         list_add(&ip->i_trunc_list, &sdp->sd_trunc_list);
 488                 spin_unlock(&sdp->sd_trunc_lock);
 489                 wake_up(&sdp->sd_quota_wait);
 490                 return 1;
 491         }
 492
 493         return error;
 494 }
 495
 496 /**
 497  * inode_go_dump - print information about an inode
 498  * @seq: The iterator
 499  * @ip: the inode
 500  * @fs_id_buf: file system id (may be empty)
 501  *
 502  */
 503
 504 static void inode_go_dump(struct seq_file *seq, struct gfs2_glock *gl,
 505                           const char *fs_id_buf)
 506 {
 507         struct gfs2_inode *ip = gl->gl_object;
 508         struct inode *inode = &ip->i_inode;
 509         unsigned long nrpages;
 510
 511         if (ip == NULL)
 512                 return;
 513
 514         xa_lock_irq(&inode->i_data.i_pages);
 515         nrpages = inode->i_data.nrpages;
 516         xa_unlock_irq(&inode->i_data.i_pages);
 517
 518         gfs2_print_dbg(seq, "%s I: n:%llu/%llu t:%u f:0x%02lx d:0x%08x s:%llu "
 519                        "p:%lu\n", fs_id_buf,
 520                   (unsigned long long)ip->i_no_formal_ino,
 521                   (unsigned long long)ip->i_no_addr,
 522                   IF2DT(ip->i_inode.i_mode), ip->i_flags,
 523                   (unsigned int)ip->i_diskflags,
 524                   (unsigned long long)i_size_read(inode), nrpages);
 525 }
 526
 527 /**
 528  * freeze_go_sync - promote/demote the freeze glock
 529  * @gl: the glock
 530  * @state: the requested state
 531  * @flags:
 532  *
 533  */
 534
 535 static int freeze_go_sync(struct gfs2_glock *gl)
 536 {
 537         int error = 0;
 538         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 539
 540         if (gl->gl_req == LM_ST_EXCLUSIVE && !gfs2_withdrawn(sdp)) {
 541                 atomic_set(&sdp->sd_freeze_state, SFS_STARTING_FREEZE);
 542                 error = freeze_super(sdp->sd_vfs);
 543                 if (error) {
 544                         fs_info(sdp, "GFS2: couldn't freeze filesystem: %d\n",
 545                                 error);
 546                         if (gfs2_withdrawn(sdp)) {
 547                                 atomic_set(&sdp->sd_freeze_state, SFS_UNFROZEN);
 548                                 return 0;
 549                         }
 550                         gfs2_assert_withdraw(sdp, 0);
 551                 }
 552                 queue_work(gfs2_freeze_wq, &sdp->sd_freeze_work);
 553                 if (test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))
 554                         gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_FREEZE |
 555                                        GFS2_LFC_FREEZE_GO_SYNC);
 556                 else /* read-only mounts */
 557                         atomic_set(&sdp->sd_freeze_state, SFS_FROZEN);
 558         }
 559         return 0;
 560 }
 561
 562 /**
 563  * freeze_go_xmote_bh - After promoting/demoting the freeze glock
 564  * @gl: the glock
 565  *
 566  */
 567
 568 static int freeze_go_xmote_bh(struct gfs2_glock *gl, struct gfs2_holder *gh)
 569 {
 570         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 571         struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
 572         struct gfs2_glock *j_gl = ip->i_gl;
 573         struct gfs2_log_header_host head;
 574         int error;
 575
 576         if (test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
 577                 j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
 578
 579                 error = gfs2_find_jhead(sdp->sd_jdesc, &head, false);
 580                 if (error)
 581                         gfs2_consist(sdp);
 582                 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT))
 583                         gfs2_consist(sdp);
 584
 585                 /*  Initialize some head of the log stuff  */
 586                 if (!gfs2_withdrawn(sdp)) {
 587                         sdp->sd_log_sequence = head.lh_sequence + 1;
 588                         gfs2_log_pointers_init(sdp, head.lh_blkno);
 589                 }
 590         }
 591         return 0;
 592 }
 593
 594 /**
 595  * trans_go_demote_ok
 596  * @gl: the glock
 597  *
 598  * Always returns 0
 599  */
 600
 601 static int freeze_go_demote_ok(const struct gfs2_glock *gl)
 602 {
 603         return 0;
 604 }
 605
 606 /**
 607  * iopen_go_callback - schedule the dcache entry for the inode to be deleted
 608  * @gl: the glock
 609  *
 610  * gl_lockref.lock lock is held while calling this
 611  */
 612 static void iopen_go_callback(struct gfs2_glock *gl, bool remote)
 613 {
 614         struct gfs2_inode *ip = gl->gl_object;
 615         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 616
 617         if (!remote || sb_rdonly(sdp->sd_vfs))
 618                 return;
 619
 620         if (gl->gl_demote_state == LM_ST_UNLOCKED &&
 621             gl->gl_state == LM_ST_SHARED && ip) {
 622                 gl->gl_lockref.count++;
 623                 if (!queue_delayed_work(gfs2_delete_workqueue,
 624                                         &gl->gl_delete, 0))
 625                         gl->gl_lockref.count--;
 626         }
 627 }
 628
 629 static int iopen_go_demote_ok(const struct gfs2_glock *gl)
 630 {
 631        return !gfs2_delete_work_queued(gl);
 632 }
 633
 634 /**
 635  * inode_go_free - wake up anyone waiting for dlm's unlock ast to free it
 636  * @gl: glock being freed
 637  *
 638  * For now, this is only used for the journal inode glock. In withdraw
 639  * situations, we need to wait for the glock to be freed so that we know
 640  * other nodes may proceed with recovery / journal replay.
 641  */
 642 static void inode_go_free(struct gfs2_glock *gl)
 643 {
 644         /* Note that we cannot reference gl_object because it's already set
 645          * to NULL by this point in its lifecycle. */
 646         if (!test_bit(GLF_FREEING, &gl->gl_flags))
 647                 return;
 648         clear_bit_unlock(GLF_FREEING, &gl->gl_flags);
 649         wake_up_bit(&gl->gl_flags, GLF_FREEING);
 650 }
 651
 652 /**
 653  * nondisk_go_callback - used to signal when a node did a withdraw
 654  * @gl: the nondisk glock
 655  * @remote: true if this came from a different cluster node
 656  *
 657  */
 658 static void nondisk_go_callback(struct gfs2_glock *gl, bool remote)
 659 {
 660         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 661
 662         /* Ignore the callback unless it's from another node, and it's the
 663            live lock. */
 664         if (!remote || gl->gl_name.ln_number != GFS2_LIVE_LOCK)
 665                 return;
 666
 667         /* First order of business is to cancel the demote request. We don't
 668          * really want to demote a nondisk glock. At best it's just to inform
 669          * us of another node's withdraw. We'll keep it in SH mode. */
 670         clear_bit(GLF_DEMOTE, &gl->gl_flags);
 671         clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
 672
 673         /* Ignore the unlock if we're withdrawn, unmounting, or in recovery. */
 674         if (test_bit(SDF_NORECOVERY, &sdp->sd_flags) ||
 675             test_bit(SDF_WITHDRAWN, &sdp->sd_flags) ||
 676             test_bit(SDF_REMOTE_WITHDRAW, &sdp->sd_flags))
 677                 return;
 678
 679         /* We only care when a node wants us to unlock, because that means
 680          * they want a journal recovered. */
 681         if (gl->gl_demote_state != LM_ST_UNLOCKED)
 682                 return;
 683
 684         if (sdp->sd_args.ar_spectator) {
 685                 fs_warn(sdp, "Spectator node cannot recover journals.\n");
 686                 return;
 687         }
 688
 689         fs_warn(sdp, "Some node has withdrawn; checking for recovery.\n");
 690         set_bit(SDF_REMOTE_WITHDRAW, &sdp->sd_flags);
 691         /*
 692          * We can't call remote_withdraw directly here or gfs2_recover_journal
 693          * because this is called from the glock unlock function and the
 694          * remote_withdraw needs to enqueue and dequeue the same "live" glock
 695          * we were called from. So we queue it to the control work queue in
 696          * lock_dlm.
 697          */
 698         queue_delayed_work(gfs2_control_wq, &sdp->sd_control_work, 0);
 699 }
 700
 701 const struct gfs2_glock_operations gfs2_meta_glops = {
 702         .go_type = LM_TYPE_META,
 703         .go_flags = GLOF_NONDISK,
 704 };
 705
 706 const struct gfs2_glock_operations gfs2_inode_glops = {
 707         .go_sync = inode_go_sync,
 708         .go_inval = inode_go_inval,
 709         .go_demote_ok = inode_go_demote_ok,
 710         .go_lock = inode_go_lock,
 711         .go_dump = inode_go_dump,
 712         .go_type = LM_TYPE_INODE,
 713         .go_flags = GLOF_ASPACE | GLOF_LRU | GLOF_LVB,
 714         .go_free = inode_go_free,
 715 };
 716
 717 const struct gfs2_glock_operations gfs2_rgrp_glops = {
 718         .go_sync = rgrp_go_sync,
 719         .go_inval = rgrp_go_inval,
 720         .go_lock = gfs2_rgrp_go_lock,
 721         .go_dump = gfs2_rgrp_go_dump,
 722         .go_type = LM_TYPE_RGRP,
 723         .go_flags = GLOF_LVB,
 724 };
 725
 726 const struct gfs2_glock_operations gfs2_freeze_glops = {
 727         .go_sync = freeze_go_sync,
 728         .go_xmote_bh = freeze_go_xmote_bh,
 729         .go_demote_ok = freeze_go_demote_ok,
 730         .go_type = LM_TYPE_NONDISK,
 731         .go_flags = GLOF_NONDISK,
 732 };
 733
 734 const struct gfs2_glock_operations gfs2_iopen_glops = {
 735         .go_type = LM_TYPE_IOPEN,
 736         .go_callback = iopen_go_callback,
 737         .go_demote_ok = iopen_go_demote_ok,
 738         .go_flags = GLOF_LRU | GLOF_NONDISK,
 739 };
 740
 741 const struct gfs2_glock_operations gfs2_flock_glops = {
 742         .go_type = LM_TYPE_FLOCK,
 743         .go_flags = GLOF_LRU | GLOF_NONDISK,
 744 };
 745
 746 const struct gfs2_glock_operations gfs2_nondisk_glops = {
 747         .go_type = LM_TYPE_NONDISK,
 748         .go_flags = GLOF_NONDISK,
 749         .go_callback = nondisk_go_callback,
 750 };
 751
 752 const struct gfs2_glock_operations gfs2_quota_glops = {
 753         .go_type = LM_TYPE_QUOTA,
 754         .go_flags = GLOF_LVB | GLOF_LRU | GLOF_NONDISK,
 755 };
 756
 757 const struct gfs2_glock_operations gfs2_journal_glops = {
 758         .go_type = LM_TYPE_JOURNAL,
 759         .go_flags = GLOF_NONDISK,
 760 };
 761
 762 const struct gfs2_glock_operations *gfs2_glops_list[] = {
 763         [LM_TYPE_META] = &gfs2_meta_glops,
 764         [LM_TYPE_INODE] = &gfs2_inode_glops,
 765         [LM_TYPE_RGRP] = &gfs2_rgrp_glops,
 766         [LM_TYPE_IOPEN] = &gfs2_iopen_glops,
 767         [LM_TYPE_FLOCK] = &gfs2_flock_glops,
 768         [LM_TYPE_NONDISK] = &gfs2_nondisk_glops,
 769         [LM_TYPE_QUOTA] = &gfs2_quota_glops,
 770         [LM_TYPE_JOURNAL] = &gfs2_journal_glops,
 771 };
 772