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         unsigned int revokes;
  90         int ret;
  91
  92         revokes = atomic_read(&gl->gl_ail_count);
  93
  94         if (!revokes) {
  95                 bool have_revokes;
  96                 bool log_in_flight;
  97
  98                 /*
  99                  * We have nothing on the ail, but there could be revokes on
 100                  * the sdp revoke queue, in which case, we still want to flush
 101                  * the log and wait for it to finish.
 102                  *
 103                  * If the sdp revoke list is empty too, we might still have an
 104                  * io outstanding for writing revokes, so we should wait for
 105                  * it before returning.
 106                  *
 107                  * If none of these conditions are true, our revokes are all
 108                  * flushed and we can return.
 109                  */
 110                 gfs2_log_lock(sdp);
 111                 have_revokes = !list_empty(&sdp->sd_log_revokes);
 112                 log_in_flight = atomic_read(&sdp->sd_log_in_flight);
 113                 gfs2_log_unlock(sdp);
 114                 if (have_revokes)
 115                         goto flush;
 116                 if (log_in_flight)
 117                         log_flush_wait(sdp);
 118                 return 0;
 119         }
 120
 121         memset(&tr, 0, sizeof(tr));
 122         set_bit(TR_ONSTACK, &tr.tr_flags);
 123         ret = __gfs2_trans_begin(&tr, sdp, 0, revokes, _RET_IP_);
 124         if (ret)
 125                 goto flush;
 126         __gfs2_ail_flush(gl, 0, revokes);
 127         gfs2_trans_end(sdp);
 128
 129 flush:
 130         gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
 131                        GFS2_LFC_AIL_EMPTY_GL);
 132         return 0;
 133 }
 134
 135 void gfs2_ail_flush(struct gfs2_glock *gl, bool fsync)
 136 {
 137         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 138         unsigned int revokes = atomic_read(&gl->gl_ail_count);
 139         int ret;
 140
 141         if (!revokes)
 142                 return;
 143
 144         ret = gfs2_trans_begin(sdp, 0, revokes);
 145         if (ret)
 146                 return;
 147         __gfs2_ail_flush(gl, fsync, revokes);
 148         gfs2_trans_end(sdp);
 149         gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
 150                        GFS2_LFC_AIL_FLUSH);
 151 }
 152
 153 /**
 154  * gfs2_rgrp_metasync - sync out the metadata of a resource group
 155  * @gl: the glock protecting the resource group
 156  *
 157  */
 158
 159 static int gfs2_rgrp_metasync(struct gfs2_glock *gl)
 160 {
 161         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 162         struct address_space *metamapping = &sdp->sd_aspace;
 163         struct gfs2_rgrpd *rgd = gfs2_glock2rgrp(gl);
 164         const unsigned bsize = sdp->sd_sb.sb_bsize;
 165         loff_t start = (rgd->rd_addr * bsize) & PAGE_MASK;
 166         loff_t end = PAGE_ALIGN((rgd->rd_addr + rgd->rd_length) * bsize) - 1;
 167         int error;
 168
 169         filemap_fdatawrite_range(metamapping, start, end);
 170         error = filemap_fdatawait_range(metamapping, start, end);
 171         WARN_ON_ONCE(error && !gfs2_withdrawn(sdp));
 172         mapping_set_error(metamapping, error);
 173         if (error)
 174                 gfs2_io_error(sdp);
 175         return error;
 176 }
 177
 178 /**
 179  * rgrp_go_sync - sync out the metadata for this glock
 180  * @gl: the glock
 181  *
 182  * Called when demoting or unlocking an EX glock.  We must flush
 183  * to disk all dirty buffers/pages relating to this glock, and must not
 184  * return to caller to demote/unlock the glock until I/O is complete.
 185  */
 186
 187 static int rgrp_go_sync(struct gfs2_glock *gl)
 188 {
 189         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 190         struct gfs2_rgrpd *rgd = gfs2_glock2rgrp(gl);
 191         int error;
 192
 193         if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags))
 194                 return 0;
 195         GLOCK_BUG_ON(gl, gl->gl_state != LM_ST_EXCLUSIVE);
 196
 197         gfs2_log_flush(sdp, gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
 198                        GFS2_LFC_RGRP_GO_SYNC);
 199         error = gfs2_rgrp_metasync(gl);
 200         if (!error)
 201                 error = gfs2_ail_empty_gl(gl);
 202         gfs2_free_clones(rgd);
 203         return error;
 204 }
 205
 206 /**
 207  * rgrp_go_inval - invalidate the metadata for this glock
 208  * @gl: the glock
 209  * @flags:
 210  *
 211  * We never used LM_ST_DEFERRED with resource groups, so that we
 212  * should always see the metadata flag set here.
 213  *
 214  */
 215
 216 static void rgrp_go_inval(struct gfs2_glock *gl, int flags)
 217 {
 218         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 219         struct address_space *mapping = &sdp->sd_aspace;
 220         struct gfs2_rgrpd *rgd = gfs2_glock2rgrp(gl);
 221         const unsigned bsize = sdp->sd_sb.sb_bsize;
 222         loff_t start = (rgd->rd_addr * bsize) & PAGE_MASK;
 223         loff_t end = PAGE_ALIGN((rgd->rd_addr + rgd->rd_length) * bsize) - 1;
 224
 225         gfs2_rgrp_brelse(rgd);
 226         WARN_ON_ONCE(!(flags & DIO_METADATA));
 227         truncate_inode_pages_range(mapping, start, end);
 228         rgd->rd_flags &= ~GFS2_RDF_UPTODATE;
 229 }
 230
 231 static void gfs2_rgrp_go_dump(struct seq_file *seq, struct gfs2_glock *gl,
 232                               const char *fs_id_buf)
 233 {
 234         struct gfs2_rgrpd *rgd = gl->gl_object;
 235
 236         if (rgd)
 237                 gfs2_rgrp_dump(seq, rgd, fs_id_buf);
 238 }
 239
 240 static struct gfs2_inode *gfs2_glock2inode(struct gfs2_glock *gl)
 241 {
 242         struct gfs2_inode *ip;
 243
 244         spin_lock(&gl->gl_lockref.lock);
 245         ip = gl->gl_object;
 246         if (ip)
 247                 set_bit(GIF_GLOP_PENDING, &ip->i_flags);
 248         spin_unlock(&gl->gl_lockref.lock);
 249         return ip;
 250 }
 251
 252 struct gfs2_rgrpd *gfs2_glock2rgrp(struct gfs2_glock *gl)
 253 {
 254         struct gfs2_rgrpd *rgd;
 255
 256         spin_lock(&gl->gl_lockref.lock);
 257         rgd = gl->gl_object;
 258         spin_unlock(&gl->gl_lockref.lock);
 259
 260         return rgd;
 261 }
 262
 263 static void gfs2_clear_glop_pending(struct gfs2_inode *ip)
 264 {
 265         if (!ip)
 266                 return;
 267
 268         clear_bit_unlock(GIF_GLOP_PENDING, &ip->i_flags);
 269         wake_up_bit(&ip->i_flags, GIF_GLOP_PENDING);
 270 }
 271
 272 /**
 273  * gfs2_inode_metasync - sync out the metadata of an inode
 274  * @gl: the glock protecting the inode
 275  *
 276  */
 277 int gfs2_inode_metasync(struct gfs2_glock *gl)
 278 {
 279         struct address_space *metamapping = gfs2_glock2aspace(gl);
 280         int error;
 281
 282         filemap_fdatawrite(metamapping);
 283         error = filemap_fdatawait(metamapping);
 284         if (error)
 285                 gfs2_io_error(gl->gl_name.ln_sbd);
 286         return error;
 287 }
 288
 289 /**
 290  * inode_go_sync - Sync the dirty metadata of an inode
 291  * @gl: the glock protecting the inode
 292  *
 293  */
 294
 295 static int inode_go_sync(struct gfs2_glock *gl)
 296 {
 297         struct gfs2_inode *ip = gfs2_glock2inode(gl);
 298         int isreg = ip && S_ISREG(ip->i_inode.i_mode);
 299         struct address_space *metamapping = gfs2_glock2aspace(gl);
 300         int error = 0, ret;
 301
 302         if (isreg) {
 303                 if (test_and_clear_bit(GIF_SW_PAGED, &ip->i_flags))
 304                         unmap_shared_mapping_range(ip->i_inode.i_mapping, 0, 0);
 305                 inode_dio_wait(&ip->i_inode);
 306         }
 307         if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags))
 308                 goto out;
 309
 310         GLOCK_BUG_ON(gl, gl->gl_state != LM_ST_EXCLUSIVE);
 311
 312         gfs2_log_flush(gl->gl_name.ln_sbd, gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
 313                        GFS2_LFC_INODE_GO_SYNC);
 314         filemap_fdatawrite(metamapping);
 315         if (isreg) {
 316                 struct address_space *mapping = ip->i_inode.i_mapping;
 317                 filemap_fdatawrite(mapping);
 318                 error = filemap_fdatawait(mapping);
 319                 mapping_set_error(mapping, error);
 320         }
 321         ret = gfs2_inode_metasync(gl);
 322         if (!error)
 323                 error = ret;
 324         gfs2_ail_empty_gl(gl);
 325         /*
 326          * Writeback of the data mapping may cause the dirty flag to be set
 327          * so we have to clear it again here.
 328          */
 329         smp_mb__before_atomic();
 330         clear_bit(GLF_DIRTY, &gl->gl_flags);
 331
 332 out:
 333         gfs2_clear_glop_pending(ip);
 334         return error;
 335 }
 336
 337 /**
 338  * inode_go_inval - prepare a inode glock to be released
 339  * @gl: the glock
 340  * @flags:
 341  *
 342  * Normally we invalidate everything, but if we are moving into
 343  * LM_ST_DEFERRED from LM_ST_SHARED or LM_ST_EXCLUSIVE then we
 344  * can keep hold of the metadata, since it won't have changed.
 345  *
 346  */
 347
 348 static void inode_go_inval(struct gfs2_glock *gl, int flags)
 349 {
 350         struct gfs2_inode *ip = gfs2_glock2inode(gl);
 351
 352         if (flags & DIO_METADATA) {
 353                 struct address_space *mapping = gfs2_glock2aspace(gl);
 354                 truncate_inode_pages(mapping, 0);
 355                 if (ip) {
 356                         set_bit(GIF_INVALID, &ip->i_flags);
 357                         forget_all_cached_acls(&ip->i_inode);
 358                         security_inode_invalidate_secctx(&ip->i_inode);
 359                         gfs2_dir_hash_inval(ip);
 360                 }
 361         }
 362
 363         if (ip == GFS2_I(gl->gl_name.ln_sbd->sd_rindex)) {
 364                 gfs2_log_flush(gl->gl_name.ln_sbd, NULL,
 365                                GFS2_LOG_HEAD_FLUSH_NORMAL |
 366                                GFS2_LFC_INODE_GO_INVAL);
 367                 gl->gl_name.ln_sbd->sd_rindex_uptodate = 0;
 368         }
 369         if (ip && S_ISREG(ip->i_inode.i_mode))
 370                 truncate_inode_pages(ip->i_inode.i_mapping, 0);
 371
 372         gfs2_clear_glop_pending(ip);
 373 }
 374
 375 /**
 376  * inode_go_demote_ok - Check to see if it's ok to unlock an inode glock
 377  * @gl: the glock
 378  *
 379  * Returns: 1 if it's ok
 380  */
 381
 382 static int inode_go_demote_ok(const struct gfs2_glock *gl)
 383 {
 384         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 385
 386         if (sdp->sd_jindex == gl->gl_object || sdp->sd_rindex == gl->gl_object)
 387                 return 0;
 388
 389         return 1;
 390 }
 391
 392 static int gfs2_dinode_in(struct gfs2_inode *ip, const void *buf)
 393 {
 394         const struct gfs2_dinode *str = buf;
 395         struct timespec64 atime;
 396         u16 height, depth;
 397
 398         if (unlikely(ip->i_no_addr != be64_to_cpu(str->di_num.no_addr)))
 399                 goto corrupt;
 400         ip->i_no_formal_ino = be64_to_cpu(str->di_num.no_formal_ino);
 401         ip->i_inode.i_mode = be32_to_cpu(str->di_mode);
 402         ip->i_inode.i_rdev = 0;
 403         switch (ip->i_inode.i_mode & S_IFMT) {
 404         case S_IFBLK:
 405         case S_IFCHR:
 406                 ip->i_inode.i_rdev = MKDEV(be32_to_cpu(str->di_major),
 407                                            be32_to_cpu(str->di_minor));
 408                 break;
 409         }
 410
 411         i_uid_write(&ip->i_inode, be32_to_cpu(str->di_uid));
 412         i_gid_write(&ip->i_inode, be32_to_cpu(str->di_gid));
 413         set_nlink(&ip->i_inode, be32_to_cpu(str->di_nlink));
 414         i_size_write(&ip->i_inode, be64_to_cpu(str->di_size));
 415         gfs2_set_inode_blocks(&ip->i_inode, be64_to_cpu(str->di_blocks));
 416         atime.tv_sec = be64_to_cpu(str->di_atime);
 417         atime.tv_nsec = be32_to_cpu(str->di_atime_nsec);
 418         if (timespec64_compare(&ip->i_inode.i_atime, &atime) < 0)
 419                 ip->i_inode.i_atime = atime;
 420         ip->i_inode.i_mtime.tv_sec = be64_to_cpu(str->di_mtime);
 421         ip->i_inode.i_mtime.tv_nsec = be32_to_cpu(str->di_mtime_nsec);
 422         ip->i_inode.i_ctime.tv_sec = be64_to_cpu(str->di_ctime);
 423         ip->i_inode.i_ctime.tv_nsec = be32_to_cpu(str->di_ctime_nsec);
 424
 425         ip->i_goal = be64_to_cpu(str->di_goal_meta);
 426         ip->i_generation = be64_to_cpu(str->di_generation);
 427
 428         ip->i_diskflags = be32_to_cpu(str->di_flags);
 429         ip->i_eattr = be64_to_cpu(str->di_eattr);
 430         /* i_diskflags and i_eattr must be set before gfs2_set_inode_flags() */
 431         gfs2_set_inode_flags(&ip->i_inode);
 432         height = be16_to_cpu(str->di_height);
 433         if (unlikely(height > GFS2_MAX_META_HEIGHT))
 434                 goto corrupt;
 435         ip->i_height = (u8)height;
 436
 437         depth = be16_to_cpu(str->di_depth);
 438         if (unlikely(depth > GFS2_DIR_MAX_DEPTH))
 439                 goto corrupt;
 440         ip->i_depth = (u8)depth;
 441         ip->i_entries = be32_to_cpu(str->di_entries);
 442
 443         if (S_ISREG(ip->i_inode.i_mode))
 444                 gfs2_set_aops(&ip->i_inode);
 445
 446         return 0;
 447 corrupt:
 448         gfs2_consist_inode(ip);
 449         return -EIO;
 450 }
 451
 452 /**
 453  * gfs2_inode_refresh - Refresh the incore copy of the dinode
 454  * @ip: The GFS2 inode
 455  *
 456  * Returns: errno
 457  */
 458
 459 int gfs2_inode_refresh(struct gfs2_inode *ip)
 460 {
 461         struct buffer_head *dibh;
 462         int error;
 463
 464         error = gfs2_meta_inode_buffer(ip, &dibh);
 465         if (error)
 466                 return error;
 467
 468         error = gfs2_dinode_in(ip, dibh->b_data);
 469         brelse(dibh);
 470         clear_bit(GIF_INVALID, &ip->i_flags);
 471
 472         return error;
 473 }
 474
 475 /**
 476  * inode_go_lock - operation done after an inode lock is locked by a process
 477  * @gl: the glock
 478  * @flags:
 479  *
 480  * Returns: errno
 481  */
 482
 483 static int inode_go_lock(struct gfs2_holder *gh)
 484 {
 485         struct gfs2_glock *gl = gh->gh_gl;
 486         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 487         struct gfs2_inode *ip = gl->gl_object;
 488         int error = 0;
 489
 490         if (!ip || (gh->gh_flags & GL_SKIP))
 491                 return 0;
 492
 493         if (test_bit(GIF_INVALID, &ip->i_flags)) {
 494                 error = gfs2_inode_refresh(ip);
 495                 if (error)
 496                         return error;
 497         }
 498
 499         if (gh->gh_state != LM_ST_DEFERRED)
 500                 inode_dio_wait(&ip->i_inode);
 501
 502         if ((ip->i_diskflags & GFS2_DIF_TRUNC_IN_PROG) &&
 503             (gl->gl_state == LM_ST_EXCLUSIVE) &&
 504             (gh->gh_state == LM_ST_EXCLUSIVE)) {
 505                 spin_lock(&sdp->sd_trunc_lock);
 506                 if (list_empty(&ip->i_trunc_list))
 507                         list_add(&ip->i_trunc_list, &sdp->sd_trunc_list);
 508                 spin_unlock(&sdp->sd_trunc_lock);
 509                 wake_up(&sdp->sd_quota_wait);
 510                 return 1;
 511         }
 512
 513         return error;
 514 }
 515
 516 /**
 517  * inode_go_dump - print information about an inode
 518  * @seq: The iterator
 519  * @ip: the inode
 520  * @fs_id_buf: file system id (may be empty)
 521  *
 522  */
 523
 524 static void inode_go_dump(struct seq_file *seq, struct gfs2_glock *gl,
 525                           const char *fs_id_buf)
 526 {
 527         struct gfs2_inode *ip = gl->gl_object;
 528         struct inode *inode = &ip->i_inode;
 529         unsigned long nrpages;
 530
 531         if (ip == NULL)
 532                 return;
 533
 534         xa_lock_irq(&inode->i_data.i_pages);
 535         nrpages = inode->i_data.nrpages;
 536         xa_unlock_irq(&inode->i_data.i_pages);
 537
 538         gfs2_print_dbg(seq, "%s I: n:%llu/%llu t:%u f:0x%02lx d:0x%08x s:%llu "
 539                        "p:%lu\n", fs_id_buf,
 540                   (unsigned long long)ip->i_no_formal_ino,
 541                   (unsigned long long)ip->i_no_addr,
 542                   IF2DT(ip->i_inode.i_mode), ip->i_flags,
 543                   (unsigned int)ip->i_diskflags,
 544                   (unsigned long long)i_size_read(inode), nrpages);
 545 }
 546
 547 /**
 548  * freeze_go_sync - promote/demote the freeze glock
 549  * @gl: the glock
 550  * @state: the requested state
 551  * @flags:
 552  *
 553  */
 554
 555 static int freeze_go_sync(struct gfs2_glock *gl)
 556 {
 557         int error = 0;
 558         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 559
 560         /*
 561          * We need to check gl_state == LM_ST_SHARED here and not gl_req ==
 562          * LM_ST_EXCLUSIVE. That's because when any node does a freeze,
 563          * all the nodes should have the freeze glock in SH mode and they all
 564          * call do_xmote: One for EX and the others for UN. They ALL must
 565          * freeze locally, and they ALL must queue freeze work. The freeze_work
 566          * calls freeze_func, which tries to reacquire the freeze glock in SH,
 567          * effectively waiting for the thaw on the node who holds it in EX.
 568          * Once thawed, the work func acquires the freeze glock in
 569          * SH and everybody goes back to thawed.
 570          */
 571         if (gl->gl_state == LM_ST_SHARED && !gfs2_withdrawn(sdp) &&
 572             !test_bit(SDF_NORECOVERY, &sdp->sd_flags)) {
 573                 atomic_set(&sdp->sd_freeze_state, SFS_STARTING_FREEZE);
 574                 error = freeze_super(sdp->sd_vfs);
 575                 if (error) {
 576                         fs_info(sdp, "GFS2: couldn't freeze filesystem: %d\n",
 577                                 error);
 578                         if (gfs2_withdrawn(sdp)) {
 579                                 atomic_set(&sdp->sd_freeze_state, SFS_UNFROZEN);
 580                                 return 0;
 581                         }
 582                         gfs2_assert_withdraw(sdp, 0);
 583                 }
 584                 queue_work(gfs2_freeze_wq, &sdp->sd_freeze_work);
 585                 if (test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))
 586                         gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_FREEZE |
 587                                        GFS2_LFC_FREEZE_GO_SYNC);
 588                 else /* read-only mounts */
 589                         atomic_set(&sdp->sd_freeze_state, SFS_FROZEN);
 590         }
 591         return 0;
 592 }
 593
 594 /**
 595  * freeze_go_xmote_bh - After promoting/demoting the freeze glock
 596  * @gl: the glock
 597  *
 598  */
 599
 600 static int freeze_go_xmote_bh(struct gfs2_glock *gl, struct gfs2_holder *gh)
 601 {
 602         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 603         struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
 604         struct gfs2_glock *j_gl = ip->i_gl;
 605         struct gfs2_log_header_host head;
 606         int error;
 607
 608         if (test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
 609                 j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
 610
 611                 error = gfs2_find_jhead(sdp->sd_jdesc, &head, false);
 612                 if (error)
 613                         gfs2_consist(sdp);
 614                 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT))
 615                         gfs2_consist(sdp);
 616
 617                 /*  Initialize some head of the log stuff  */
 618                 if (!gfs2_withdrawn(sdp)) {
 619                         sdp->sd_log_sequence = head.lh_sequence + 1;
 620                         gfs2_log_pointers_init(sdp, head.lh_blkno);
 621                 }
 622         }
 623         return 0;
 624 }
 625
 626 /**
 627  * trans_go_demote_ok
 628  * @gl: the glock
 629  *
 630  * Always returns 0
 631  */
 632
 633 static int freeze_go_demote_ok(const struct gfs2_glock *gl)
 634 {
 635         return 0;
 636 }
 637
 638 /**
 639  * iopen_go_callback - schedule the dcache entry for the inode to be deleted
 640  * @gl: the glock
 641  *
 642  * gl_lockref.lock lock is held while calling this
 643  */
 644 static void iopen_go_callback(struct gfs2_glock *gl, bool remote)
 645 {
 646         struct gfs2_inode *ip = gl->gl_object;
 647         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 648
 649         if (!remote || sb_rdonly(sdp->sd_vfs))
 650                 return;
 651
 652         if (gl->gl_demote_state == LM_ST_UNLOCKED &&
 653             gl->gl_state == LM_ST_SHARED && ip) {
 654                 gl->gl_lockref.count++;
 655                 if (!queue_delayed_work(gfs2_delete_workqueue,
 656                                         &gl->gl_delete, 0))
 657                         gl->gl_lockref.count--;
 658         }
 659 }
 660
 661 static int iopen_go_demote_ok(const struct gfs2_glock *gl)
 662 {
 663        return !gfs2_delete_work_queued(gl);
 664 }
 665
 666 /**
 667  * inode_go_free - wake up anyone waiting for dlm's unlock ast to free it
 668  * @gl: glock being freed
 669  *
 670  * For now, this is only used for the journal inode glock. In withdraw
 671  * situations, we need to wait for the glock to be freed so that we know
 672  * other nodes may proceed with recovery / journal replay.
 673  */
 674 static void inode_go_free(struct gfs2_glock *gl)
 675 {
 676         /* Note that we cannot reference gl_object because it's already set
 677          * to NULL by this point in its lifecycle. */
 678         if (!test_bit(GLF_FREEING, &gl->gl_flags))
 679                 return;
 680         clear_bit_unlock(GLF_FREEING, &gl->gl_flags);
 681         wake_up_bit(&gl->gl_flags, GLF_FREEING);
 682 }
 683
 684 /**
 685  * nondisk_go_callback - used to signal when a node did a withdraw
 686  * @gl: the nondisk glock
 687  * @remote: true if this came from a different cluster node
 688  *
 689  */
 690 static void nondisk_go_callback(struct gfs2_glock *gl, bool remote)
 691 {
 692         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 693
 694         /* Ignore the callback unless it's from another node, and it's the
 695            live lock. */
 696         if (!remote || gl->gl_name.ln_number != GFS2_LIVE_LOCK)
 697                 return;
 698
 699         /* First order of business is to cancel the demote request. We don't
 700          * really want to demote a nondisk glock. At best it's just to inform
 701          * us of another node's withdraw. We'll keep it in SH mode. */
 702         clear_bit(GLF_DEMOTE, &gl->gl_flags);
 703         clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
 704
 705         /* Ignore the unlock if we're withdrawn, unmounting, or in recovery. */
 706         if (test_bit(SDF_NORECOVERY, &sdp->sd_flags) ||
 707             test_bit(SDF_WITHDRAWN, &sdp->sd_flags) ||
 708             test_bit(SDF_REMOTE_WITHDRAW, &sdp->sd_flags))
 709                 return;
 710
 711         /* We only care when a node wants us to unlock, because that means
 712          * they want a journal recovered. */
 713         if (gl->gl_demote_state != LM_ST_UNLOCKED)
 714                 return;
 715
 716         if (sdp->sd_args.ar_spectator) {
 717                 fs_warn(sdp, "Spectator node cannot recover journals.\n");
 718                 return;
 719         }
 720
 721         fs_warn(sdp, "Some node has withdrawn; checking for recovery.\n");
 722         set_bit(SDF_REMOTE_WITHDRAW, &sdp->sd_flags);
 723         /*
 724          * We can't call remote_withdraw directly here or gfs2_recover_journal
 725          * because this is called from the glock unlock function and the
 726          * remote_withdraw needs to enqueue and dequeue the same "live" glock
 727          * we were called from. So we queue it to the control work queue in
 728          * lock_dlm.
 729          */
 730         queue_delayed_work(gfs2_control_wq, &sdp->sd_control_work, 0);
 731 }
 732
 733 const struct gfs2_glock_operations gfs2_meta_glops = {
 734         .go_type = LM_TYPE_META,
 735         .go_flags = GLOF_NONDISK,
 736 };
 737
 738 const struct gfs2_glock_operations gfs2_inode_glops = {
 739         .go_sync = inode_go_sync,
 740         .go_inval = inode_go_inval,
 741         .go_demote_ok = inode_go_demote_ok,
 742         .go_lock = inode_go_lock,
 743         .go_dump = inode_go_dump,
 744         .go_type = LM_TYPE_INODE,
 745         .go_flags = GLOF_ASPACE | GLOF_LRU | GLOF_LVB,
 746         .go_free = inode_go_free,
 747 };
 748
 749 const struct gfs2_glock_operations gfs2_rgrp_glops = {
 750         .go_sync = rgrp_go_sync,
 751         .go_inval = rgrp_go_inval,
 752         .go_lock = gfs2_rgrp_go_lock,
 753         .go_dump = gfs2_rgrp_go_dump,
 754         .go_type = LM_TYPE_RGRP,
 755         .go_flags = GLOF_LVB,
 756 };
 757
 758 const struct gfs2_glock_operations gfs2_freeze_glops = {
 759         .go_sync = freeze_go_sync,
 760         .go_xmote_bh = freeze_go_xmote_bh,
 761         .go_demote_ok = freeze_go_demote_ok,
 762         .go_type = LM_TYPE_NONDISK,
 763         .go_flags = GLOF_NONDISK,
 764 };
 765
 766 const struct gfs2_glock_operations gfs2_iopen_glops = {
 767         .go_type = LM_TYPE_IOPEN,
 768         .go_callback = iopen_go_callback,
 769         .go_demote_ok = iopen_go_demote_ok,
 770         .go_flags = GLOF_LRU | GLOF_NONDISK,
 771         .go_subclass = 1,
 772 };
 773
 774 const struct gfs2_glock_operations gfs2_flock_glops = {
 775         .go_type = LM_TYPE_FLOCK,
 776         .go_flags = GLOF_LRU | GLOF_NONDISK,
 777 };
 778
 779 const struct gfs2_glock_operations gfs2_nondisk_glops = {
 780         .go_type = LM_TYPE_NONDISK,
 781         .go_flags = GLOF_NONDISK,
 782         .go_callback = nondisk_go_callback,
 783 };
 784
 785 const struct gfs2_glock_operations gfs2_quota_glops = {
 786         .go_type = LM_TYPE_QUOTA,
 787         .go_flags = GLOF_LVB | GLOF_LRU | GLOF_NONDISK,
 788 };
 789
 790 const struct gfs2_glock_operations gfs2_journal_glops = {
 791         .go_type = LM_TYPE_JOURNAL,
 792         .go_flags = GLOF_NONDISK,
 793 };
 794
 795 const struct gfs2_glock_operations *gfs2_glops_list[] = {
 796         [LM_TYPE_META] = &gfs2_meta_glops,
 797         [LM_TYPE_INODE] = &gfs2_inode_glops,
 798         [LM_TYPE_RGRP] = &gfs2_rgrp_glops,
 799         [LM_TYPE_IOPEN] = &gfs2_iopen_glops,
 800         [LM_TYPE_FLOCK] = &gfs2_flock_glops,
 801         [LM_TYPE_NONDISK] = &gfs2_nondisk_glops,
 802         [LM_TYPE_QUOTA] = &gfs2_quota_glops,
 803         [LM_TYPE_JOURNAL] = &gfs2_journal_glops,
 804 };
 805