fs/gfs2/glops.c

   1 /*
   2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   3  * Copyright (C) 2004-2008 Red Hat, Inc.  All rights reserved.
   4  *
   5  * This copyrighted material is made available to anyone wishing to use,
   6  * modify, copy, or redistribute it subject to the terms and conditions
   7  * of the GNU General Public License version 2.
   8  */
   9
  10 #include <linux/spinlock.h>
  11 #include <linux/completion.h>
  12 #include <linux/buffer_head.h>
  13 #include <linux/gfs2_ondisk.h>
  14 #include <linux/bio.h>
  15 #include <linux/posix_acl.h>
  16
  17 #include "gfs2.h"
  18 #include "incore.h"
  19 #include "bmap.h"
  20 #include "glock.h"
  21 #include "glops.h"
  22 #include "inode.h"
  23 #include "log.h"
  24 #include "meta_io.h"
  25 #include "recovery.h"
  26 #include "rgrp.h"
  27 #include "util.h"
  28 #include "trans.h"
  29 #include "dir.h"
  30
  31 static void gfs2_ail_error(struct gfs2_glock *gl, const struct buffer_head *bh)
  32 {
  33         fs_err(gl->gl_sbd, "AIL buffer %p: blocknr %llu state 0x%08lx mapping %p page state 0x%lx\n",
  34                bh, (unsigned long long)bh->b_blocknr, bh->b_state,
  35                bh->b_page->mapping, bh->b_page->flags);
  36         fs_err(gl->gl_sbd, "AIL glock %u:%llu mapping %p\n",
  37                gl->gl_name.ln_type, gl->gl_name.ln_number,
  38                gfs2_glock2aspace(gl));
  39         gfs2_lm_withdraw(gl->gl_sbd, "AIL error\n");
  40 }
  41
  42 /**
  43  * __gfs2_ail_flush - remove all buffers for a given lock from the AIL
  44  * @gl: the glock
  45  * @fsync: set when called from fsync (not all buffers will be clean)
  46  *
  47  * None of the buffers should be dirty, locked, or pinned.
  48  */
  49
  50 static void __gfs2_ail_flush(struct gfs2_glock *gl, bool fsync)
  51 {
  52         struct gfs2_sbd *sdp = gl->gl_sbd;
  53         struct list_head *head = &gl->gl_ail_list;
  54         struct gfs2_bufdata *bd, *tmp;
  55         struct buffer_head *bh;
  56         const unsigned long b_state = (1UL << BH_Dirty)|(1UL << BH_Pinned)|(1UL << BH_Lock);
  57
  58         gfs2_log_lock(sdp);
  59         spin_lock(&sdp->sd_ail_lock);
  60         list_for_each_entry_safe(bd, tmp, head, bd_ail_gl_list) {
  61                 bh = bd->bd_bh;
  62                 if (bh->b_state & b_state) {
  63                         if (fsync)
  64                                 continue;
  65                         gfs2_ail_error(gl, bh);
  66                 }
  67                 gfs2_trans_add_revoke(sdp, bd);
  68         }
  69         GLOCK_BUG_ON(gl, !fsync && atomic_read(&gl->gl_ail_count));
  70         spin_unlock(&sdp->sd_ail_lock);
  71         gfs2_log_unlock(sdp);
  72 }
  73
  74
  75 static void gfs2_ail_empty_gl(struct gfs2_glock *gl)
  76 {
  77         struct gfs2_sbd *sdp = gl->gl_sbd;
  78         struct gfs2_trans tr;
  79
  80         memset(&tr, 0, sizeof(tr));
  81         tr.tr_revokes = atomic_read(&gl->gl_ail_count);
  82
  83         if (!tr.tr_revokes)
  84                 return;
  85
  86         /* A shortened, inline version of gfs2_trans_begin() */
  87         tr.tr_reserved = 1 + gfs2_struct2blk(sdp, tr.tr_revokes, sizeof(u64));
  88         tr.tr_ip = (unsigned long)__builtin_return_address(0);
  89         sb_start_intwrite(sdp->sd_vfs);
  90         gfs2_log_reserve(sdp, tr.tr_reserved);
  91         WARN_ON_ONCE(current->journal_info);
  92         current->journal_info = &tr;
  93
  94         __gfs2_ail_flush(gl, 0);
  95
  96         gfs2_trans_end(sdp);
  97         gfs2_log_flush(sdp, NULL);
  98 }
  99
 100 void gfs2_ail_flush(struct gfs2_glock *gl, bool fsync)
 101 {
 102         struct gfs2_sbd *sdp = gl->gl_sbd;
 103         unsigned int revokes = atomic_read(&gl->gl_ail_count);
 104         int ret;
 105
 106         if (!revokes)
 107                 return;
 108
 109         ret = gfs2_trans_begin(sdp, 0, revokes);
 110         if (ret)
 111                 return;
 112         __gfs2_ail_flush(gl, fsync);
 113         gfs2_trans_end(sdp);
 114         gfs2_log_flush(sdp, NULL);
 115 }
 116
 117 /**
 118  * rgrp_go_sync - sync out the metadata for this glock
 119  * @gl: the glock
 120  *
 121  * Called when demoting or unlocking an EX glock.  We must flush
 122  * to disk all dirty buffers/pages relating to this glock, and must not
 123  * not return to caller to demote/unlock the glock until I/O is complete.
 124  */
 125
 126 static void rgrp_go_sync(struct gfs2_glock *gl)
 127 {
 128         struct address_space *metamapping = gfs2_glock2aspace(gl);
 129         struct gfs2_rgrpd *rgd;
 130         int error;
 131
 132         if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags))
 133                 return;
 134         GLOCK_BUG_ON(gl, gl->gl_state != LM_ST_EXCLUSIVE);
 135
 136         gfs2_log_flush(gl->gl_sbd, gl);
 137         filemap_fdatawrite(metamapping);
 138         error = filemap_fdatawait(metamapping);
 139         mapping_set_error(metamapping, error);
 140         gfs2_ail_empty_gl(gl);
 141
 142         spin_lock(&gl->gl_spin);
 143         rgd = gl->gl_object;
 144         if (rgd)
 145                 gfs2_free_clones(rgd);
 146         spin_unlock(&gl->gl_spin);
 147 }
 148
 149 /**
 150  * rgrp_go_inval - invalidate the metadata for this glock
 151  * @gl: the glock
 152  * @flags:
 153  *
 154  * We never used LM_ST_DEFERRED with resource groups, so that we
 155  * should always see the metadata flag set here.
 156  *
 157  */
 158
 159 static void rgrp_go_inval(struct gfs2_glock *gl, int flags)
 160 {
 161         struct address_space *mapping = gfs2_glock2aspace(gl);
 162
 163         WARN_ON_ONCE(!(flags & DIO_METADATA));
 164         gfs2_assert_withdraw(gl->gl_sbd, !atomic_read(&gl->gl_ail_count));
 165         truncate_inode_pages(mapping, 0);
 166
 167         if (gl->gl_object) {
 168                 struct gfs2_rgrpd *rgd = (struct gfs2_rgrpd *)gl->gl_object;
 169                 rgd->rd_flags &= ~GFS2_RDF_UPTODATE;
 170         }
 171 }
 172
 173 /**
 174  * inode_go_sync - Sync the dirty data and/or metadata for an inode glock
 175  * @gl: the glock protecting the inode
 176  *
 177  */
 178
 179 static void inode_go_sync(struct gfs2_glock *gl)
 180 {
 181         struct gfs2_inode *ip = gl->gl_object;
 182         struct address_space *metamapping = gfs2_glock2aspace(gl);
 183         int error;
 184
 185         if (ip && !S_ISREG(ip->i_inode.i_mode))
 186                 ip = NULL;
 187         if (ip && test_and_clear_bit(GIF_SW_PAGED, &ip->i_flags))
 188                 unmap_shared_mapping_range(ip->i_inode.i_mapping, 0, 0);
 189         if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags))
 190                 return;
 191
 192         GLOCK_BUG_ON(gl, gl->gl_state != LM_ST_EXCLUSIVE);
 193
 194         gfs2_log_flush(gl->gl_sbd, gl);
 195         filemap_fdatawrite(metamapping);
 196         if (ip) {
 197                 struct address_space *mapping = ip->i_inode.i_mapping;
 198                 filemap_fdatawrite(mapping);
 199                 error = filemap_fdatawait(mapping);
 200                 mapping_set_error(mapping, error);
 201         }
 202         error = filemap_fdatawait(metamapping);
 203         mapping_set_error(metamapping, error);
 204         gfs2_ail_empty_gl(gl);
 205         /*
 206          * Writeback of the data mapping may cause the dirty flag to be set
 207          * so we have to clear it again here.
 208          */
 209         smp_mb__before_clear_bit();
 210         clear_bit(GLF_DIRTY, &gl->gl_flags);
 211 }
 212
 213 /**
 214  * inode_go_inval - prepare a inode glock to be released
 215  * @gl: the glock
 216  * @flags:
 217  *
 218  * Normally we invlidate everything, but if we are moving into
 219  * LM_ST_DEFERRED from LM_ST_SHARED or LM_ST_EXCLUSIVE then we
 220  * can keep hold of the metadata, since it won't have changed.
 221  *
 222  */
 223
 224 static void inode_go_inval(struct gfs2_glock *gl, int flags)
 225 {
 226         struct gfs2_inode *ip = gl->gl_object;
 227
 228         gfs2_assert_withdraw(gl->gl_sbd, !atomic_read(&gl->gl_ail_count));
 229
 230         if (flags & DIO_METADATA) {
 231                 struct address_space *mapping = gfs2_glock2aspace(gl);
 232                 truncate_inode_pages(mapping, 0);
 233                 if (ip) {
 234                         set_bit(GIF_INVALID, &ip->i_flags);
 235                         forget_all_cached_acls(&ip->i_inode);
 236                         gfs2_dir_hash_inval(ip);
 237                 }
 238         }
 239
 240         if (ip == GFS2_I(gl->gl_sbd->sd_rindex)) {
 241                 gfs2_log_flush(gl->gl_sbd, NULL);
 242                 gl->gl_sbd->sd_rindex_uptodate = 0;
 243         }
 244         if (ip && S_ISREG(ip->i_inode.i_mode))
 245                 truncate_inode_pages(ip->i_inode.i_mapping, 0);
 246 }
 247
 248 /**
 249  * inode_go_demote_ok - Check to see if it's ok to unlock an inode glock
 250  * @gl: the glock
 251  *
 252  * Returns: 1 if it's ok
 253  */
 254
 255 static int inode_go_demote_ok(const struct gfs2_glock *gl)
 256 {
 257         struct gfs2_sbd *sdp = gl->gl_sbd;
 258         struct gfs2_holder *gh;
 259
 260         if (sdp->sd_jindex == gl->gl_object || sdp->sd_rindex == gl->gl_object)
 261                 return 0;
 262
 263         if (!list_empty(&gl->gl_holders)) {
 264                 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
 265                 if (gh->gh_list.next != &gl->gl_holders)
 266                         return 0;
 267         }
 268
 269         return 1;
 270 }
 271
 272 /**
 273  * gfs2_set_nlink - Set the inode's link count based on on-disk info
 274  * @inode: The inode in question
 275  * @nlink: The link count
 276  *
 277  * If the link count has hit zero, it must never be raised, whatever the
 278  * on-disk inode might say. When new struct inodes are created the link
 279  * count is set to 1, so that we can safely use this test even when reading
 280  * in on disk information for the first time.
 281  */
 282
 283 static void gfs2_set_nlink(struct inode *inode, u32 nlink)
 284 {
 285         /*
 286          * We will need to review setting the nlink count here in the
 287          * light of the forthcoming ro bind mount work. This is a reminder
 288          * to do that.
 289          */
 290         if ((inode->i_nlink != nlink) && (inode->i_nlink != 0)) {
 291                 if (nlink == 0)
 292                         clear_nlink(inode);
 293                 else
 294                         set_nlink(inode, nlink);
 295         }
 296 }
 297
 298 static int gfs2_dinode_in(struct gfs2_inode *ip, const void *buf)
 299 {
 300         const struct gfs2_dinode *str = buf;
 301         struct timespec atime;
 302         u16 height, depth;
 303
 304         if (unlikely(ip->i_no_addr != be64_to_cpu(str->di_num.no_addr)))
 305                 goto corrupt;
 306         ip->i_no_formal_ino = be64_to_cpu(str->di_num.no_formal_ino);
 307         ip->i_inode.i_mode = be32_to_cpu(str->di_mode);
 308         ip->i_inode.i_rdev = 0;
 309         switch (ip->i_inode.i_mode & S_IFMT) {
 310         case S_IFBLK:
 311         case S_IFCHR:
 312                 ip->i_inode.i_rdev = MKDEV(be32_to_cpu(str->di_major),
 313                                            be32_to_cpu(str->di_minor));
 314                 break;
 315         };
 316
 317         i_uid_write(&ip->i_inode, be32_to_cpu(str->di_uid));
 318         i_gid_write(&ip->i_inode, be32_to_cpu(str->di_gid));
 319         gfs2_set_nlink(&ip->i_inode, be32_to_cpu(str->di_nlink));
 320         i_size_write(&ip->i_inode, be64_to_cpu(str->di_size));
 321         gfs2_set_inode_blocks(&ip->i_inode, be64_to_cpu(str->di_blocks));
 322         atime.tv_sec = be64_to_cpu(str->di_atime);
 323         atime.tv_nsec = be32_to_cpu(str->di_atime_nsec);
 324         if (timespec_compare(&ip->i_inode.i_atime, &atime) < 0)
 325                 ip->i_inode.i_atime = atime;
 326         ip->i_inode.i_mtime.tv_sec = be64_to_cpu(str->di_mtime);
 327         ip->i_inode.i_mtime.tv_nsec = be32_to_cpu(str->di_mtime_nsec);
 328         ip->i_inode.i_ctime.tv_sec = be64_to_cpu(str->di_ctime);
 329         ip->i_inode.i_ctime.tv_nsec = be32_to_cpu(str->di_ctime_nsec);
 330
 331         ip->i_goal = be64_to_cpu(str->di_goal_meta);
 332         ip->i_generation = be64_to_cpu(str->di_generation);
 333
 334         ip->i_diskflags = be32_to_cpu(str->di_flags);
 335         ip->i_eattr = be64_to_cpu(str->di_eattr);
 336         /* i_diskflags and i_eattr must be set before gfs2_set_inode_flags() */
 337         gfs2_set_inode_flags(&ip->i_inode);
 338         height = be16_to_cpu(str->di_height);
 339         if (unlikely(height > GFS2_MAX_META_HEIGHT))
 340                 goto corrupt;
 341         ip->i_height = (u8)height;
 342
 343         depth = be16_to_cpu(str->di_depth);
 344         if (unlikely(depth > GFS2_DIR_MAX_DEPTH))
 345                 goto corrupt;
 346         ip->i_depth = (u8)depth;
 347         ip->i_entries = be32_to_cpu(str->di_entries);
 348
 349         if (S_ISREG(ip->i_inode.i_mode))
 350                 gfs2_set_aops(&ip->i_inode);
 351
 352         return 0;
 353 corrupt:
 354         gfs2_consist_inode(ip);
 355         return -EIO;
 356 }
 357
 358 /**
 359  * gfs2_inode_refresh - Refresh the incore copy of the dinode
 360  * @ip: The GFS2 inode
 361  *
 362  * Returns: errno
 363  */
 364
 365 int gfs2_inode_refresh(struct gfs2_inode *ip)
 366 {
 367         struct buffer_head *dibh;
 368         int error;
 369
 370         error = gfs2_meta_inode_buffer(ip, &dibh);
 371         if (error)
 372                 return error;
 373
 374         error = gfs2_dinode_in(ip, dibh->b_data);
 375         brelse(dibh);
 376         clear_bit(GIF_INVALID, &ip->i_flags);
 377
 378         return error;
 379 }
 380
 381 /**
 382  * inode_go_lock - operation done after an inode lock is locked by a process
 383  * @gl: the glock
 384  * @flags:
 385  *
 386  * Returns: errno
 387  */
 388
 389 static int inode_go_lock(struct gfs2_holder *gh)
 390 {
 391         struct gfs2_glock *gl = gh->gh_gl;
 392         struct gfs2_sbd *sdp = gl->gl_sbd;
 393         struct gfs2_inode *ip = gl->gl_object;
 394         int error = 0;
 395
 396         if (!ip || (gh->gh_flags & GL_SKIP))
 397                 return 0;
 398
 399         if (test_bit(GIF_INVALID, &ip->i_flags)) {
 400                 error = gfs2_inode_refresh(ip);
 401                 if (error)
 402                         return error;
 403         }
 404
 405         if ((ip->i_diskflags & GFS2_DIF_TRUNC_IN_PROG) &&
 406             (gl->gl_state == LM_ST_EXCLUSIVE) &&
 407             (gh->gh_state == LM_ST_EXCLUSIVE)) {
 408                 spin_lock(&sdp->sd_trunc_lock);
 409                 if (list_empty(&ip->i_trunc_list))
 410                         list_add(&sdp->sd_trunc_list, &ip->i_trunc_list);
 411                 spin_unlock(&sdp->sd_trunc_lock);
 412                 wake_up(&sdp->sd_quota_wait);
 413                 return 1;
 414         }
 415
 416         return error;
 417 }
 418
 419 /**
 420  * inode_go_dump - print information about an inode
 421  * @seq: The iterator
 422  * @ip: the inode
 423  *
 424  * Returns: 0 on success, -ENOBUFS when we run out of space
 425  */
 426
 427 static int inode_go_dump(struct seq_file *seq, const struct gfs2_glock *gl)
 428 {
 429         const struct gfs2_inode *ip = gl->gl_object;
 430         if (ip == NULL)
 431                 return 0;
 432         gfs2_print_dbg(seq, " I: n:%llu/%llu t:%u f:0x%02lx d:0x%08x s:%llu\n",
 433                   (unsigned long long)ip->i_no_formal_ino,
 434                   (unsigned long long)ip->i_no_addr,
 435                   IF2DT(ip->i_inode.i_mode), ip->i_flags,
 436                   (unsigned int)ip->i_diskflags,
 437                   (unsigned long long)i_size_read(&ip->i_inode));
 438         return 0;
 439 }
 440
 441 /**
 442  * trans_go_sync - promote/demote the transaction glock
 443  * @gl: the glock
 444  * @state: the requested state
 445  * @flags:
 446  *
 447  */
 448
 449 static void trans_go_sync(struct gfs2_glock *gl)
 450 {
 451         struct gfs2_sbd *sdp = gl->gl_sbd;
 452
 453         if (gl->gl_state != LM_ST_UNLOCKED &&
 454             test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
 455                 gfs2_meta_syncfs(sdp);
 456                 gfs2_log_shutdown(sdp);
 457         }
 458 }
 459
 460 /**
 461  * trans_go_xmote_bh - After promoting/demoting the transaction glock
 462  * @gl: the glock
 463  *
 464  */
 465
 466 static int trans_go_xmote_bh(struct gfs2_glock *gl, struct gfs2_holder *gh)
 467 {
 468         struct gfs2_sbd *sdp = gl->gl_sbd;
 469         struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
 470         struct gfs2_glock *j_gl = ip->i_gl;
 471         struct gfs2_log_header_host head;
 472         int error;
 473
 474         if (test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
 475                 j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
 476
 477                 error = gfs2_find_jhead(sdp->sd_jdesc, &head);
 478                 if (error)
 479                         gfs2_consist(sdp);
 480                 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT))
 481                         gfs2_consist(sdp);
 482
 483                 /*  Initialize some head of the log stuff  */
 484                 if (!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) {
 485                         sdp->sd_log_sequence = head.lh_sequence + 1;
 486                         gfs2_log_pointers_init(sdp, head.lh_blkno);
 487                 }
 488         }
 489         return 0;
 490 }
 491
 492 /**
 493  * trans_go_demote_ok
 494  * @gl: the glock
 495  *
 496  * Always returns 0
 497  */
 498
 499 static int trans_go_demote_ok(const struct gfs2_glock *gl)
 500 {
 501         return 0;
 502 }
 503
 504 /**
 505  * iopen_go_callback - schedule the dcache entry for the inode to be deleted
 506  * @gl: the glock
 507  *
 508  * gl_spin lock is held while calling this
 509  */
 510 static void iopen_go_callback(struct gfs2_glock *gl, bool remote)
 511 {
 512         struct gfs2_inode *ip = (struct gfs2_inode *)gl->gl_object;
 513         struct gfs2_sbd *sdp = gl->gl_sbd;
 514
 515         if (!remote || (sdp->sd_vfs->s_flags & MS_RDONLY))
 516                 return;
 517
 518         if (gl->gl_demote_state == LM_ST_UNLOCKED &&
 519             gl->gl_state == LM_ST_SHARED && ip) {
 520                 gfs2_glock_hold(gl);
 521                 if (queue_work(gfs2_delete_workqueue, &gl->gl_delete) == 0)
 522                         gfs2_glock_put_nolock(gl);
 523         }
 524 }
 525
 526 const struct gfs2_glock_operations gfs2_meta_glops = {
 527         .go_type = LM_TYPE_META,
 528 };
 529
 530 const struct gfs2_glock_operations gfs2_inode_glops = {
 531         .go_sync = inode_go_sync,
 532         .go_inval = inode_go_inval,
 533         .go_demote_ok = inode_go_demote_ok,
 534         .go_lock = inode_go_lock,
 535         .go_dump = inode_go_dump,
 536         .go_type = LM_TYPE_INODE,
 537         .go_flags = GLOF_ASPACE,
 538 };
 539
 540 const struct gfs2_glock_operations gfs2_rgrp_glops = {
 541         .go_sync = rgrp_go_sync,
 542         .go_inval = rgrp_go_inval,
 543         .go_lock = gfs2_rgrp_go_lock,
 544         .go_unlock = gfs2_rgrp_go_unlock,
 545         .go_dump = gfs2_rgrp_dump,
 546         .go_type = LM_TYPE_RGRP,
 547         .go_flags = GLOF_ASPACE | GLOF_LVB,
 548 };
 549
 550 const struct gfs2_glock_operations gfs2_trans_glops = {
 551         .go_sync = trans_go_sync,
 552         .go_xmote_bh = trans_go_xmote_bh,
 553         .go_demote_ok = trans_go_demote_ok,
 554         .go_type = LM_TYPE_NONDISK,
 555 };
 556
 557 const struct gfs2_glock_operations gfs2_iopen_glops = {
 558         .go_type = LM_TYPE_IOPEN,
 559         .go_callback = iopen_go_callback,
 560 };
 561
 562 const struct gfs2_glock_operations gfs2_flock_glops = {
 563         .go_type = LM_TYPE_FLOCK,
 564 };
 565
 566 const struct gfs2_glock_operations gfs2_nondisk_glops = {
 567         .go_type = LM_TYPE_NONDISK,
 568 };
 569
 570 const struct gfs2_glock_operations gfs2_quota_glops = {
 571         .go_type = LM_TYPE_QUOTA,
 572         .go_flags = GLOF_LVB,
 573 };
 574
 575 const struct gfs2_glock_operations gfs2_journal_glops = {
 576         .go_type = LM_TYPE_JOURNAL,
 577 };
 578
 579 const struct gfs2_glock_operations *gfs2_glops_list[] = {
 580         [LM_TYPE_META] = &gfs2_meta_glops,
 581         [LM_TYPE_INODE] = &gfs2_inode_glops,
 582         [LM_TYPE_RGRP] = &gfs2_rgrp_glops,
 583         [LM_TYPE_IOPEN] = &gfs2_iopen_glops,
 584         [LM_TYPE_FLOCK] = &gfs2_flock_glops,
 585         [LM_TYPE_NONDISK] = &gfs2_nondisk_glops,
 586         [LM_TYPE_QUOTA] = &gfs2_quota_glops,
 587         [LM_TYPE_JOURNAL] = &gfs2_journal_glops,
 588 };
 589