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