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  * gfs2_rgrp_metasync - sync out the metadata of a resource group
 169  * @gl: the glock protecting the resource group
 170  *
 171  */
 172
 173 static int gfs2_rgrp_metasync(struct gfs2_glock *gl)
 174 {
 175         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 176         struct address_space *metamapping = &sdp->sd_aspace;
 177         struct gfs2_rgrpd *rgd = gfs2_glock2rgrp(gl);
 178         const unsigned bsize = sdp->sd_sb.sb_bsize;
 179         loff_t start = (rgd->rd_addr * bsize) & PAGE_MASK;
 180         loff_t end = PAGE_ALIGN((rgd->rd_addr + rgd->rd_length) * bsize) - 1;
 181         int error;
 182
 183         filemap_fdatawrite_range(metamapping, start, end);
 184         error = filemap_fdatawait_range(metamapping, start, end);
 185         WARN_ON_ONCE(error && !gfs2_withdrawn(sdp));
 186         mapping_set_error(metamapping, error);
 187         if (error)
 188                 gfs2_io_error(sdp);
 189         return error;
 190 }
 191
 192 /**
 193  * rgrp_go_sync - sync out the metadata for this glock
 194  * @gl: the glock
 195  *
 196  * Called when demoting or unlocking an EX glock.  We must flush
 197  * to disk all dirty buffers/pages relating to this glock, and must not
 198  * return to caller to demote/unlock the glock until I/O is complete.
 199  */
 200
 201 static int rgrp_go_sync(struct gfs2_glock *gl)
 202 {
 203         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 204         struct gfs2_rgrpd *rgd = gfs2_glock2rgrp(gl);
 205         int error;
 206
 207         if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags))
 208                 return 0;
 209         GLOCK_BUG_ON(gl, gl->gl_state != LM_ST_EXCLUSIVE);
 210
 211         gfs2_log_flush(sdp, gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
 212                        GFS2_LFC_RGRP_GO_SYNC);
 213         error = gfs2_rgrp_metasync(gl);
 214         if (!error)
 215                 error = gfs2_ail_empty_gl(gl);
 216         gfs2_free_clones(rgd);
 217         return error;
 218 }
 219
 220 /**
 221  * rgrp_go_inval - invalidate the metadata for this glock
 222  * @gl: the glock
 223  * @flags:
 224  *
 225  * We never used LM_ST_DEFERRED with resource groups, so that we
 226  * should always see the metadata flag set here.
 227  *
 228  */
 229
 230 static void rgrp_go_inval(struct gfs2_glock *gl, int flags)
 231 {
 232         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 233         struct address_space *mapping = &sdp->sd_aspace;
 234         struct gfs2_rgrpd *rgd = gfs2_glock2rgrp(gl);
 235         const unsigned bsize = sdp->sd_sb.sb_bsize;
 236         loff_t start = (rgd->rd_addr * bsize) & PAGE_MASK;
 237         loff_t end = PAGE_ALIGN((rgd->rd_addr + rgd->rd_length) * bsize) - 1;
 238
 239         gfs2_rgrp_brelse(rgd);
 240         WARN_ON_ONCE(!(flags & DIO_METADATA));
 241         truncate_inode_pages_range(mapping, start, end);
 242         rgd->rd_flags &= ~GFS2_RDF_UPTODATE;
 243 }
 244
 245 static void gfs2_rgrp_go_dump(struct seq_file *seq, struct gfs2_glock *gl,
 246                               const char *fs_id_buf)
 247 {
 248         struct gfs2_rgrpd *rgd = gl->gl_object;
 249
 250         if (rgd)
 251                 gfs2_rgrp_dump(seq, rgd, fs_id_buf);
 252 }
 253
 254 static struct gfs2_inode *gfs2_glock2inode(struct gfs2_glock *gl)
 255 {
 256         struct gfs2_inode *ip;
 257
 258         spin_lock(&gl->gl_lockref.lock);
 259         ip = gl->gl_object;
 260         if (ip)
 261                 set_bit(GIF_GLOP_PENDING, &ip->i_flags);
 262         spin_unlock(&gl->gl_lockref.lock);
 263         return ip;
 264 }
 265
 266 struct gfs2_rgrpd *gfs2_glock2rgrp(struct gfs2_glock *gl)
 267 {
 268         struct gfs2_rgrpd *rgd;
 269
 270         spin_lock(&gl->gl_lockref.lock);
 271         rgd = gl->gl_object;
 272         spin_unlock(&gl->gl_lockref.lock);
 273
 274         return rgd;
 275 }
 276
 277 static void gfs2_clear_glop_pending(struct gfs2_inode *ip)
 278 {
 279         if (!ip)
 280                 return;
 281
 282         clear_bit_unlock(GIF_GLOP_PENDING, &ip->i_flags);
 283         wake_up_bit(&ip->i_flags, GIF_GLOP_PENDING);
 284 }
 285
 286 /**
 287  * gfs2_inode_metasync - sync out the metadata of an inode
 288  * @gl: the glock protecting the inode
 289  *
 290  */
 291 int gfs2_inode_metasync(struct gfs2_glock *gl)
 292 {
 293         struct address_space *metamapping = gfs2_glock2aspace(gl);
 294         int error;
 295
 296         filemap_fdatawrite(metamapping);
 297         error = filemap_fdatawait(metamapping);
 298         if (error)
 299                 gfs2_io_error(gl->gl_name.ln_sbd);
 300         return error;
 301 }
 302
 303 /**
 304  * inode_go_sync - Sync the dirty metadata of an inode
 305  * @gl: the glock protecting the inode
 306  *
 307  */
 308
 309 static int inode_go_sync(struct gfs2_glock *gl)
 310 {
 311         struct gfs2_inode *ip = gfs2_glock2inode(gl);
 312         int isreg = ip && S_ISREG(ip->i_inode.i_mode);
 313         struct address_space *metamapping = gfs2_glock2aspace(gl);
 314         int error = 0, ret;
 315
 316         if (isreg) {
 317                 if (test_and_clear_bit(GIF_SW_PAGED, &ip->i_flags))
 318                         unmap_shared_mapping_range(ip->i_inode.i_mapping, 0, 0);
 319                 inode_dio_wait(&ip->i_inode);
 320         }
 321         if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags))
 322                 goto out;
 323
 324         GLOCK_BUG_ON(gl, gl->gl_state != LM_ST_EXCLUSIVE);
 325
 326         gfs2_log_flush(gl->gl_name.ln_sbd, gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
 327                        GFS2_LFC_INODE_GO_SYNC);
 328         filemap_fdatawrite(metamapping);
 329         if (isreg) {
 330                 struct address_space *mapping = ip->i_inode.i_mapping;
 331                 filemap_fdatawrite(mapping);
 332                 error = filemap_fdatawait(mapping);
 333                 mapping_set_error(mapping, error);
 334         }
 335         ret = gfs2_inode_metasync(gl);
 336         if (!error)
 337                 error = ret;
 338         gfs2_ail_empty_gl(gl);
 339         /*
 340          * Writeback of the data mapping may cause the dirty flag to be set
 341          * so we have to clear it again here.
 342          */
 343         smp_mb__before_atomic();
 344         clear_bit(GLF_DIRTY, &gl->gl_flags);
 345
 346 out:
 347         gfs2_clear_glop_pending(ip);
 348         return error;
 349 }
 350
 351 /**
 352  * inode_go_inval - prepare a inode glock to be released
 353  * @gl: the glock
 354  * @flags:
 355  *
 356  * Normally we invalidate everything, but if we are moving into
 357  * LM_ST_DEFERRED from LM_ST_SHARED or LM_ST_EXCLUSIVE then we
 358  * can keep hold of the metadata, since it won't have changed.
 359  *
 360  */
 361
 362 static void inode_go_inval(struct gfs2_glock *gl, int flags)
 363 {
 364         struct gfs2_inode *ip = gfs2_glock2inode(gl);
 365
 366         if (flags & DIO_METADATA) {
 367                 struct address_space *mapping = gfs2_glock2aspace(gl);
 368                 truncate_inode_pages(mapping, 0);
 369                 if (ip) {
 370                         set_bit(GIF_INVALID, &ip->i_flags);
 371                         forget_all_cached_acls(&ip->i_inode);
 372                         security_inode_invalidate_secctx(&ip->i_inode);
 373                         gfs2_dir_hash_inval(ip);
 374                 }
 375         }
 376
 377         if (ip == GFS2_I(gl->gl_name.ln_sbd->sd_rindex)) {
 378                 gfs2_log_flush(gl->gl_name.ln_sbd, NULL,
 379                                GFS2_LOG_HEAD_FLUSH_NORMAL |
 380                                GFS2_LFC_INODE_GO_INVAL);
 381                 gl->gl_name.ln_sbd->sd_rindex_uptodate = 0;
 382         }
 383         if (ip && S_ISREG(ip->i_inode.i_mode))
 384                 truncate_inode_pages(ip->i_inode.i_mapping, 0);
 385
 386         gfs2_clear_glop_pending(ip);
 387 }
 388
 389 /**
 390  * inode_go_demote_ok - Check to see if it's ok to unlock an inode glock
 391  * @gl: the glock
 392  *
 393  * Returns: 1 if it's ok
 394  */
 395
 396 static int inode_go_demote_ok(const struct gfs2_glock *gl)
 397 {
 398         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 399
 400         if (sdp->sd_jindex == gl->gl_object || sdp->sd_rindex == gl->gl_object)
 401                 return 0;
 402
 403         return 1;
 404 }
 405
 406 static int gfs2_dinode_in(struct gfs2_inode *ip, const void *buf)
 407 {
 408         const struct gfs2_dinode *str = buf;
 409         struct timespec64 atime;
 410         u16 height, depth;
 411
 412         if (unlikely(ip->i_no_addr != be64_to_cpu(str->di_num.no_addr)))
 413                 goto corrupt;
 414         ip->i_no_formal_ino = be64_to_cpu(str->di_num.no_formal_ino);
 415         ip->i_inode.i_mode = be32_to_cpu(str->di_mode);
 416         ip->i_inode.i_rdev = 0;
 417         switch (ip->i_inode.i_mode & S_IFMT) {
 418         case S_IFBLK:
 419         case S_IFCHR:
 420                 ip->i_inode.i_rdev = MKDEV(be32_to_cpu(str->di_major),
 421                                            be32_to_cpu(str->di_minor));
 422                 break;
 423         }
 424
 425         i_uid_write(&ip->i_inode, be32_to_cpu(str->di_uid));
 426         i_gid_write(&ip->i_inode, be32_to_cpu(str->di_gid));
 427         set_nlink(&ip->i_inode, be32_to_cpu(str->di_nlink));
 428         i_size_write(&ip->i_inode, be64_to_cpu(str->di_size));
 429         gfs2_set_inode_blocks(&ip->i_inode, be64_to_cpu(str->di_blocks));
 430         atime.tv_sec = be64_to_cpu(str->di_atime);
 431         atime.tv_nsec = be32_to_cpu(str->di_atime_nsec);
 432         if (timespec64_compare(&ip->i_inode.i_atime, &atime) < 0)
 433                 ip->i_inode.i_atime = atime;
 434         ip->i_inode.i_mtime.tv_sec = be64_to_cpu(str->di_mtime);
 435         ip->i_inode.i_mtime.tv_nsec = be32_to_cpu(str->di_mtime_nsec);
 436         ip->i_inode.i_ctime.tv_sec = be64_to_cpu(str->di_ctime);
 437         ip->i_inode.i_ctime.tv_nsec = be32_to_cpu(str->di_ctime_nsec);
 438
 439         ip->i_goal = be64_to_cpu(str->di_goal_meta);
 440         ip->i_generation = be64_to_cpu(str->di_generation);
 441
 442         ip->i_diskflags = be32_to_cpu(str->di_flags);
 443         ip->i_eattr = be64_to_cpu(str->di_eattr);
 444         /* i_diskflags and i_eattr must be set before gfs2_set_inode_flags() */
 445         gfs2_set_inode_flags(&ip->i_inode);
 446         height = be16_to_cpu(str->di_height);
 447         if (unlikely(height > GFS2_MAX_META_HEIGHT))
 448                 goto corrupt;
 449         ip->i_height = (u8)height;
 450
 451         depth = be16_to_cpu(str->di_depth);
 452         if (unlikely(depth > GFS2_DIR_MAX_DEPTH))
 453                 goto corrupt;
 454         ip->i_depth = (u8)depth;
 455         ip->i_entries = be32_to_cpu(str->di_entries);
 456
 457         if (S_ISREG(ip->i_inode.i_mode))
 458                 gfs2_set_aops(&ip->i_inode);
 459
 460         return 0;
 461 corrupt:
 462         gfs2_consist_inode(ip);
 463         return -EIO;
 464 }
 465
 466 /**
 467  * gfs2_inode_refresh - Refresh the incore copy of the dinode
 468  * @ip: The GFS2 inode
 469  *
 470  * Returns: errno
 471  */
 472
 473 int gfs2_inode_refresh(struct gfs2_inode *ip)
 474 {
 475         struct buffer_head *dibh;
 476         int error;
 477
 478         error = gfs2_meta_inode_buffer(ip, &dibh);
 479         if (error)
 480                 return error;
 481
 482         error = gfs2_dinode_in(ip, dibh->b_data);
 483         brelse(dibh);
 484         clear_bit(GIF_INVALID, &ip->i_flags);
 485
 486         return error;
 487 }
 488
 489 /**
 490  * inode_go_lock - operation done after an inode lock is locked by a process
 491  * @gl: the glock
 492  * @flags:
 493  *
 494  * Returns: errno
 495  */
 496
 497 static int inode_go_lock(struct gfs2_holder *gh)
 498 {
 499         struct gfs2_glock *gl = gh->gh_gl;
 500         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 501         struct gfs2_inode *ip = gl->gl_object;
 502         int error = 0;
 503
 504         if (!ip || (gh->gh_flags & GL_SKIP))
 505                 return 0;
 506
 507         if (test_bit(GIF_INVALID, &ip->i_flags)) {
 508                 error = gfs2_inode_refresh(ip);
 509                 if (error)
 510                         return error;
 511         }
 512
 513         if (gh->gh_state != LM_ST_DEFERRED)
 514                 inode_dio_wait(&ip->i_inode);
 515
 516         if ((ip->i_diskflags & GFS2_DIF_TRUNC_IN_PROG) &&
 517             (gl->gl_state == LM_ST_EXCLUSIVE) &&
 518             (gh->gh_state == LM_ST_EXCLUSIVE)) {
 519                 spin_lock(&sdp->sd_trunc_lock);
 520                 if (list_empty(&ip->i_trunc_list))
 521                         list_add(&ip->i_trunc_list, &sdp->sd_trunc_list);
 522                 spin_unlock(&sdp->sd_trunc_lock);
 523                 wake_up(&sdp->sd_quota_wait);
 524                 return 1;
 525         }
 526
 527         return error;
 528 }
 529
 530 /**
 531  * inode_go_dump - print information about an inode
 532  * @seq: The iterator
 533  * @ip: the inode
 534  * @fs_id_buf: file system id (may be empty)
 535  *
 536  */
 537
 538 static void inode_go_dump(struct seq_file *seq, struct gfs2_glock *gl,
 539                           const char *fs_id_buf)
 540 {
 541         struct gfs2_inode *ip = gl->gl_object;
 542         struct inode *inode = &ip->i_inode;
 543         unsigned long nrpages;
 544
 545         if (ip == NULL)
 546                 return;
 547
 548         xa_lock_irq(&inode->i_data.i_pages);
 549         nrpages = inode->i_data.nrpages;
 550         xa_unlock_irq(&inode->i_data.i_pages);
 551
 552         gfs2_print_dbg(seq, "%s I: n:%llu/%llu t:%u f:0x%02lx d:0x%08x s:%llu "
 553                        "p:%lu\n", fs_id_buf,
 554                   (unsigned long long)ip->i_no_formal_ino,
 555                   (unsigned long long)ip->i_no_addr,
 556                   IF2DT(ip->i_inode.i_mode), ip->i_flags,
 557                   (unsigned int)ip->i_diskflags,
 558                   (unsigned long long)i_size_read(inode), nrpages);
 559 }
 560
 561 /**
 562  * freeze_go_sync - promote/demote the freeze glock
 563  * @gl: the glock
 564  * @state: the requested state
 565  * @flags:
 566  *
 567  */
 568
 569 static int freeze_go_sync(struct gfs2_glock *gl)
 570 {
 571         int error = 0;
 572         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 573
 574         /*
 575          * We need to check gl_state == LM_ST_SHARED here and not gl_req ==
 576          * LM_ST_EXCLUSIVE. That's because when any node does a freeze,
 577          * all the nodes should have the freeze glock in SH mode and they all
 578          * call do_xmote: One for EX and the others for UN. They ALL must
 579          * freeze locally, and they ALL must queue freeze work. The freeze_work
 580          * calls freeze_func, which tries to reacquire the freeze glock in SH,
 581          * effectively waiting for the thaw on the node who holds it in EX.
 582          * Once thawed, the work func acquires the freeze glock in
 583          * SH and everybody goes back to thawed.
 584          */
 585         if (gl->gl_state == LM_ST_SHARED && !gfs2_withdrawn(sdp) &&
 586             !test_bit(SDF_NORECOVERY, &sdp->sd_flags)) {
 587                 atomic_set(&sdp->sd_freeze_state, SFS_STARTING_FREEZE);
 588                 error = freeze_super(sdp->sd_vfs);
 589                 if (error) {
 590                         fs_info(sdp, "GFS2: couldn't freeze filesystem: %d\n",
 591                                 error);
 592                         if (gfs2_withdrawn(sdp)) {
 593                                 atomic_set(&sdp->sd_freeze_state, SFS_UNFROZEN);
 594                                 return 0;
 595                         }
 596                         gfs2_assert_withdraw(sdp, 0);
 597                 }
 598                 queue_work(gfs2_freeze_wq, &sdp->sd_freeze_work);
 599                 if (test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))
 600                         gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_FREEZE |
 601                                        GFS2_LFC_FREEZE_GO_SYNC);
 602                 else /* read-only mounts */
 603                         atomic_set(&sdp->sd_freeze_state, SFS_FROZEN);
 604         }
 605         return 0;
 606 }
 607
 608 /**
 609  * freeze_go_xmote_bh - After promoting/demoting the freeze glock
 610  * @gl: the glock
 611  *
 612  */
 613
 614 static int freeze_go_xmote_bh(struct gfs2_glock *gl, struct gfs2_holder *gh)
 615 {
 616         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 617         struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
 618         struct gfs2_glock *j_gl = ip->i_gl;
 619         struct gfs2_log_header_host head;
 620         int error;
 621
 622         if (test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
 623                 j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
 624
 625                 error = gfs2_find_jhead(sdp->sd_jdesc, &head, false);
 626                 if (error)
 627                         gfs2_consist(sdp);
 628                 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT))
 629                         gfs2_consist(sdp);
 630
 631                 /*  Initialize some head of the log stuff  */
 632                 if (!gfs2_withdrawn(sdp)) {
 633                         sdp->sd_log_sequence = head.lh_sequence + 1;
 634                         gfs2_log_pointers_init(sdp, head.lh_blkno);
 635                 }
 636         }
 637         return 0;
 638 }
 639
 640 /**
 641  * trans_go_demote_ok
 642  * @gl: the glock
 643  *
 644  * Always returns 0
 645  */
 646
 647 static int freeze_go_demote_ok(const struct gfs2_glock *gl)
 648 {
 649         return 0;
 650 }
 651
 652 /**
 653  * iopen_go_callback - schedule the dcache entry for the inode to be deleted
 654  * @gl: the glock
 655  *
 656  * gl_lockref.lock lock is held while calling this
 657  */
 658 static void iopen_go_callback(struct gfs2_glock *gl, bool remote)
 659 {
 660         struct gfs2_inode *ip = gl->gl_object;
 661         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 662
 663         if (!remote || sb_rdonly(sdp->sd_vfs))
 664                 return;
 665
 666         if (gl->gl_demote_state == LM_ST_UNLOCKED &&
 667             gl->gl_state == LM_ST_SHARED && ip) {
 668                 gl->gl_lockref.count++;
 669                 if (!queue_delayed_work(gfs2_delete_workqueue,
 670                                         &gl->gl_delete, 0))
 671                         gl->gl_lockref.count--;
 672         }
 673 }
 674
 675 static int iopen_go_demote_ok(const struct gfs2_glock *gl)
 676 {
 677        return !gfs2_delete_work_queued(gl);
 678 }
 679
 680 /**
 681  * inode_go_free - wake up anyone waiting for dlm's unlock ast to free it
 682  * @gl: glock being freed
 683  *
 684  * For now, this is only used for the journal inode glock. In withdraw
 685  * situations, we need to wait for the glock to be freed so that we know
 686  * other nodes may proceed with recovery / journal replay.
 687  */
 688 static void inode_go_free(struct gfs2_glock *gl)
 689 {
 690         /* Note that we cannot reference gl_object because it's already set
 691          * to NULL by this point in its lifecycle. */
 692         if (!test_bit(GLF_FREEING, &gl->gl_flags))
 693                 return;
 694         clear_bit_unlock(GLF_FREEING, &gl->gl_flags);
 695         wake_up_bit(&gl->gl_flags, GLF_FREEING);
 696 }
 697
 698 /**
 699  * nondisk_go_callback - used to signal when a node did a withdraw
 700  * @gl: the nondisk glock
 701  * @remote: true if this came from a different cluster node
 702  *
 703  */
 704 static void nondisk_go_callback(struct gfs2_glock *gl, bool remote)
 705 {
 706         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 707
 708         /* Ignore the callback unless it's from another node, and it's the
 709            live lock. */
 710         if (!remote || gl->gl_name.ln_number != GFS2_LIVE_LOCK)
 711                 return;
 712
 713         /* First order of business is to cancel the demote request. We don't
 714          * really want to demote a nondisk glock. At best it's just to inform
 715          * us of another node's withdraw. We'll keep it in SH mode. */
 716         clear_bit(GLF_DEMOTE, &gl->gl_flags);
 717         clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
 718
 719         /* Ignore the unlock if we're withdrawn, unmounting, or in recovery. */
 720         if (test_bit(SDF_NORECOVERY, &sdp->sd_flags) ||
 721             test_bit(SDF_WITHDRAWN, &sdp->sd_flags) ||
 722             test_bit(SDF_REMOTE_WITHDRAW, &sdp->sd_flags))
 723                 return;
 724
 725         /* We only care when a node wants us to unlock, because that means
 726          * they want a journal recovered. */
 727         if (gl->gl_demote_state != LM_ST_UNLOCKED)
 728                 return;
 729
 730         if (sdp->sd_args.ar_spectator) {
 731                 fs_warn(sdp, "Spectator node cannot recover journals.\n");
 732                 return;
 733         }
 734
 735         fs_warn(sdp, "Some node has withdrawn; checking for recovery.\n");
 736         set_bit(SDF_REMOTE_WITHDRAW, &sdp->sd_flags);
 737         /*
 738          * We can't call remote_withdraw directly here or gfs2_recover_journal
 739          * because this is called from the glock unlock function and the
 740          * remote_withdraw needs to enqueue and dequeue the same "live" glock
 741          * we were called from. So we queue it to the control work queue in
 742          * lock_dlm.
 743          */
 744         queue_delayed_work(gfs2_control_wq, &sdp->sd_control_work, 0);
 745 }
 746
 747 const struct gfs2_glock_operations gfs2_meta_glops = {
 748         .go_type = LM_TYPE_META,
 749         .go_flags = GLOF_NONDISK,
 750 };
 751
 752 const struct gfs2_glock_operations gfs2_inode_glops = {
 753         .go_sync = inode_go_sync,
 754         .go_inval = inode_go_inval,
 755         .go_demote_ok = inode_go_demote_ok,
 756         .go_lock = inode_go_lock,
 757         .go_dump = inode_go_dump,
 758         .go_type = LM_TYPE_INODE,
 759         .go_flags = GLOF_ASPACE | GLOF_LRU | GLOF_LVB,
 760         .go_free = inode_go_free,
 761 };
 762
 763 const struct gfs2_glock_operations gfs2_rgrp_glops = {
 764         .go_sync = rgrp_go_sync,
 765         .go_inval = rgrp_go_inval,
 766         .go_lock = gfs2_rgrp_go_lock,
 767         .go_dump = gfs2_rgrp_go_dump,
 768         .go_type = LM_TYPE_RGRP,
 769         .go_flags = GLOF_LVB,
 770 };
 771
 772 const struct gfs2_glock_operations gfs2_freeze_glops = {
 773         .go_sync = freeze_go_sync,
 774         .go_xmote_bh = freeze_go_xmote_bh,
 775         .go_demote_ok = freeze_go_demote_ok,
 776         .go_type = LM_TYPE_NONDISK,
 777         .go_flags = GLOF_NONDISK,
 778 };
 779
 780 const struct gfs2_glock_operations gfs2_iopen_glops = {
 781         .go_type = LM_TYPE_IOPEN,
 782         .go_callback = iopen_go_callback,
 783         .go_demote_ok = iopen_go_demote_ok,
 784         .go_flags = GLOF_LRU | GLOF_NONDISK,
 785         .go_subclass = 1,
 786 };
 787
 788 const struct gfs2_glock_operations gfs2_flock_glops = {
 789         .go_type = LM_TYPE_FLOCK,
 790         .go_flags = GLOF_LRU | GLOF_NONDISK,
 791 };
 792
 793 const struct gfs2_glock_operations gfs2_nondisk_glops = {
 794         .go_type = LM_TYPE_NONDISK,
 795         .go_flags = GLOF_NONDISK,
 796         .go_callback = nondisk_go_callback,
 797 };
 798
 799 const struct gfs2_glock_operations gfs2_quota_glops = {
 800         .go_type = LM_TYPE_QUOTA,
 801         .go_flags = GLOF_LVB | GLOF_LRU | GLOF_NONDISK,
 802 };
 803
 804 const struct gfs2_glock_operations gfs2_journal_glops = {
 805         .go_type = LM_TYPE_JOURNAL,
 806         .go_flags = GLOF_NONDISK,
 807 };
 808
 809 const struct gfs2_glock_operations *gfs2_glops_list[] = {
 810         [LM_TYPE_META] = &gfs2_meta_glops,
 811         [LM_TYPE_INODE] = &gfs2_inode_glops,
 812         [LM_TYPE_RGRP] = &gfs2_rgrp_glops,
 813         [LM_TYPE_IOPEN] = &gfs2_iopen_glops,
 814         [LM_TYPE_FLOCK] = &gfs2_flock_glops,
 815         [LM_TYPE_NONDISK] = &gfs2_nondisk_glops,
 816         [LM_TYPE_QUOTA] = &gfs2_quota_glops,
 817         [LM_TYPE_JOURNAL] = &gfs2_journal_glops,
 818 };
 819