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
  30 /**
  31  * __gfs2_ail_flush - remove all buffers for a given lock from the AIL
  32  * @gl: the glock
  33  *
  34  * None of the buffers should be dirty, locked, or pinned.
  35  */
  36
  37 static void __gfs2_ail_flush(struct gfs2_glock *gl)
  38 {
  39         struct gfs2_sbd *sdp = gl->gl_sbd;
  40         struct list_head *head = &gl->gl_ail_list;
  41         struct gfs2_bufdata *bd;
  42         struct buffer_head *bh;
  43
  44         spin_lock(&sdp->sd_ail_lock);
  45         while (!list_empty(head)) {
  46                 bd = list_entry(head->next, struct gfs2_bufdata,
  47                                 bd_ail_gl_list);
  48                 bh = bd->bd_bh;
  49                 gfs2_remove_from_ail(bd);
  50                 spin_unlock(&sdp->sd_ail_lock);
  51
  52                 bd->bd_bh = NULL;
  53                 bh->b_private = NULL;
  54                 bd->bd_blkno = bh->b_blocknr;
  55                 gfs2_log_lock(sdp);
  56                 gfs2_assert_withdraw(sdp, !buffer_busy(bh));
  57                 gfs2_trans_add_revoke(sdp, bd);
  58                 gfs2_log_unlock(sdp);
  59
  60                 spin_lock(&sdp->sd_ail_lock);
  61         }
  62         gfs2_assert_withdraw(sdp, !atomic_read(&gl->gl_ail_count));
  63         spin_unlock(&sdp->sd_ail_lock);
  64 }
  65
  66
  67 static void gfs2_ail_empty_gl(struct gfs2_glock *gl)
  68 {
  69         struct gfs2_sbd *sdp = gl->gl_sbd;
  70         struct gfs2_trans tr;
  71
  72         memset(&tr, 0, sizeof(tr));
  73         tr.tr_revokes = atomic_read(&gl->gl_ail_count);
  74
  75         if (!tr.tr_revokes)
  76                 return;
  77
  78         /* A shortened, inline version of gfs2_trans_begin() */
  79         tr.tr_reserved = 1 + gfs2_struct2blk(sdp, tr.tr_revokes, sizeof(u64));
  80         tr.tr_ip = (unsigned long)__builtin_return_address(0);
  81         INIT_LIST_HEAD(&tr.tr_list_buf);
  82         gfs2_log_reserve(sdp, tr.tr_reserved);
  83         BUG_ON(current->journal_info);
  84         current->journal_info = &tr;
  85
  86         __gfs2_ail_flush(gl);
  87
  88         gfs2_trans_end(sdp);
  89         gfs2_log_flush(sdp, NULL);
  90 }
  91
  92 void gfs2_ail_flush(struct gfs2_glock *gl)
  93 {
  94         struct gfs2_sbd *sdp = gl->gl_sbd;
  95         unsigned int revokes = atomic_read(&gl->gl_ail_count);
  96         int ret;
  97
  98         if (!revokes)
  99                 return;
 100
 101         ret = gfs2_trans_begin(sdp, 0, revokes);
 102         if (ret)
 103                 return;
 104         __gfs2_ail_flush(gl);
 105         gfs2_trans_end(sdp);
 106         gfs2_log_flush(sdp, NULL);
 107 }
 108
 109 /**
 110  * rgrp_go_sync - sync out the metadata for this glock
 111  * @gl: the glock
 112  *
 113  * Called when demoting or unlocking an EX glock.  We must flush
 114  * to disk all dirty buffers/pages relating to this glock, and must not
 115  * not return to caller to demote/unlock the glock until I/O is complete.
 116  */
 117
 118 static void rgrp_go_sync(struct gfs2_glock *gl)
 119 {
 120         struct address_space *metamapping = gfs2_glock2aspace(gl);
 121         int error;
 122
 123         if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags))
 124                 return;
 125         BUG_ON(gl->gl_state != LM_ST_EXCLUSIVE);
 126
 127         gfs2_log_flush(gl->gl_sbd, gl);
 128         filemap_fdatawrite(metamapping);
 129         error = filemap_fdatawait(metamapping);
 130         mapping_set_error(metamapping, error);
 131         gfs2_ail_empty_gl(gl);
 132 }
 133
 134 /**
 135  * rgrp_go_inval - invalidate the metadata for this glock
 136  * @gl: the glock
 137  * @flags:
 138  *
 139  * We never used LM_ST_DEFERRED with resource groups, so that we
 140  * should always see the metadata flag set here.
 141  *
 142  */
 143
 144 static void rgrp_go_inval(struct gfs2_glock *gl, int flags)
 145 {
 146         struct address_space *mapping = gfs2_glock2aspace(gl);
 147
 148         BUG_ON(!(flags & DIO_METADATA));
 149         gfs2_assert_withdraw(gl->gl_sbd, !atomic_read(&gl->gl_ail_count));
 150         truncate_inode_pages(mapping, 0);
 151
 152         if (gl->gl_object) {
 153                 struct gfs2_rgrpd *rgd = (struct gfs2_rgrpd *)gl->gl_object;
 154                 rgd->rd_flags &= ~GFS2_RDF_UPTODATE;
 155         }
 156 }
 157
 158 /**
 159  * inode_go_sync - Sync the dirty data and/or metadata for an inode glock
 160  * @gl: the glock protecting the inode
 161  *
 162  */
 163
 164 static void inode_go_sync(struct gfs2_glock *gl)
 165 {
 166         struct gfs2_inode *ip = gl->gl_object;
 167         struct address_space *metamapping = gfs2_glock2aspace(gl);
 168         int error;
 169
 170         if (ip && !S_ISREG(ip->i_inode.i_mode))
 171                 ip = NULL;
 172         if (ip && test_and_clear_bit(GIF_SW_PAGED, &ip->i_flags))
 173                 unmap_shared_mapping_range(ip->i_inode.i_mapping, 0, 0);
 174         if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags))
 175                 return;
 176
 177         BUG_ON(gl->gl_state != LM_ST_EXCLUSIVE);
 178
 179         gfs2_log_flush(gl->gl_sbd, gl);
 180         filemap_fdatawrite(metamapping);
 181         if (ip) {
 182                 struct address_space *mapping = ip->i_inode.i_mapping;
 183                 filemap_fdatawrite(mapping);
 184                 error = filemap_fdatawait(mapping);
 185                 mapping_set_error(mapping, error);
 186         }
 187         error = filemap_fdatawait(metamapping);
 188         mapping_set_error(metamapping, error);
 189         gfs2_ail_empty_gl(gl);
 190         /*
 191          * Writeback of the data mapping may cause the dirty flag to be set
 192          * so we have to clear it again here.
 193          */
 194         smp_mb__before_clear_bit();
 195         clear_bit(GLF_DIRTY, &gl->gl_flags);
 196 }
 197
 198 /**
 199  * inode_go_inval - prepare a inode glock to be released
 200  * @gl: the glock
 201  * @flags:
 202  *
 203  * Normally we invlidate everything, but if we are moving into
 204  * LM_ST_DEFERRED from LM_ST_SHARED or LM_ST_EXCLUSIVE then we
 205  * can keep hold of the metadata, since it won't have changed.
 206  *
 207  */
 208
 209 static void inode_go_inval(struct gfs2_glock *gl, int flags)
 210 {
 211         struct gfs2_inode *ip = gl->gl_object;
 212
 213         gfs2_assert_withdraw(gl->gl_sbd, !atomic_read(&gl->gl_ail_count));
 214
 215         if (flags & DIO_METADATA) {
 216                 struct address_space *mapping = gfs2_glock2aspace(gl);
 217                 truncate_inode_pages(mapping, 0);
 218                 if (ip) {
 219                         set_bit(GIF_INVALID, &ip->i_flags);
 220                         forget_all_cached_acls(&ip->i_inode);
 221                 }
 222         }
 223
 224         if (ip == GFS2_I(gl->gl_sbd->sd_rindex))
 225                 gl->gl_sbd->sd_rindex_uptodate = 0;
 226         if (ip && S_ISREG(ip->i_inode.i_mode))
 227                 truncate_inode_pages(ip->i_inode.i_mapping, 0);
 228 }
 229
 230 /**
 231  * inode_go_demote_ok - Check to see if it's ok to unlock an inode glock
 232  * @gl: the glock
 233  *
 234  * Returns: 1 if it's ok
 235  */
 236
 237 static int inode_go_demote_ok(const struct gfs2_glock *gl)
 238 {
 239         struct gfs2_sbd *sdp = gl->gl_sbd;
 240         struct gfs2_holder *gh;
 241
 242         if (sdp->sd_jindex == gl->gl_object || sdp->sd_rindex == gl->gl_object)
 243                 return 0;
 244
 245         if (!list_empty(&gl->gl_holders)) {
 246                 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
 247                 if (gh->gh_list.next != &gl->gl_holders)
 248                         return 0;
 249         }
 250
 251         return 1;
 252 }
 253
 254 /**
 255  * gfs2_set_nlink - Set the inode's link count based on on-disk info
 256  * @inode: The inode in question
 257  * @nlink: The link count
 258  *
 259  * If the link count has hit zero, it must never be raised, whatever the
 260  * on-disk inode might say. When new struct inodes are created the link
 261  * count is set to 1, so that we can safely use this test even when reading
 262  * in on disk information for the first time.
 263  */
 264
 265 static void gfs2_set_nlink(struct inode *inode, u32 nlink)
 266 {
 267         /*
 268          * We will need to review setting the nlink count here in the
 269          * light of the forthcoming ro bind mount work. This is a reminder
 270          * to do that.
 271          */
 272         if ((inode->i_nlink != nlink) && (inode->i_nlink != 0)) {
 273                 if (nlink == 0)
 274                         clear_nlink(inode);
 275                 else
 276                         inode->i_nlink = nlink;
 277         }
 278 }
 279
 280 static int gfs2_dinode_in(struct gfs2_inode *ip, const void *buf)
 281 {
 282         const struct gfs2_dinode *str = buf;
 283         struct timespec atime;
 284         u16 height, depth;
 285
 286         if (unlikely(ip->i_no_addr != be64_to_cpu(str->di_num.no_addr)))
 287                 goto corrupt;
 288         ip->i_no_formal_ino = be64_to_cpu(str->di_num.no_formal_ino);
 289         ip->i_inode.i_mode = be32_to_cpu(str->di_mode);
 290         ip->i_inode.i_rdev = 0;
 291         switch (ip->i_inode.i_mode & S_IFMT) {
 292         case S_IFBLK:
 293         case S_IFCHR:
 294                 ip->i_inode.i_rdev = MKDEV(be32_to_cpu(str->di_major),
 295                                            be32_to_cpu(str->di_minor));
 296                 break;
 297         };
 298
 299         ip->i_inode.i_uid = be32_to_cpu(str->di_uid);
 300         ip->i_inode.i_gid = be32_to_cpu(str->di_gid);
 301         gfs2_set_nlink(&ip->i_inode, be32_to_cpu(str->di_nlink));
 302         i_size_write(&ip->i_inode, be64_to_cpu(str->di_size));
 303         gfs2_set_inode_blocks(&ip->i_inode, be64_to_cpu(str->di_blocks));
 304         atime.tv_sec = be64_to_cpu(str->di_atime);
 305         atime.tv_nsec = be32_to_cpu(str->di_atime_nsec);
 306         if (timespec_compare(&ip->i_inode.i_atime, &atime) < 0)
 307                 ip->i_inode.i_atime = atime;
 308         ip->i_inode.i_mtime.tv_sec = be64_to_cpu(str->di_mtime);
 309         ip->i_inode.i_mtime.tv_nsec = be32_to_cpu(str->di_mtime_nsec);
 310         ip->i_inode.i_ctime.tv_sec = be64_to_cpu(str->di_ctime);
 311         ip->i_inode.i_ctime.tv_nsec = be32_to_cpu(str->di_ctime_nsec);
 312
 313         ip->i_goal = be64_to_cpu(str->di_goal_meta);
 314         ip->i_generation = be64_to_cpu(str->di_generation);
 315
 316         ip->i_diskflags = be32_to_cpu(str->di_flags);
 317         gfs2_set_inode_flags(&ip->i_inode);
 318         height = be16_to_cpu(str->di_height);
 319         if (unlikely(height > GFS2_MAX_META_HEIGHT))
 320                 goto corrupt;
 321         ip->i_height = (u8)height;
 322
 323         depth = be16_to_cpu(str->di_depth);
 324         if (unlikely(depth > GFS2_DIR_MAX_DEPTH))
 325                 goto corrupt;
 326         ip->i_depth = (u8)depth;
 327         ip->i_entries = be32_to_cpu(str->di_entries);
 328
 329         ip->i_eattr = be64_to_cpu(str->di_eattr);
 330         if (S_ISREG(ip->i_inode.i_mode))
 331                 gfs2_set_aops(&ip->i_inode);
 332
 333         return 0;
 334 corrupt:
 335         gfs2_consist_inode(ip);
 336         return -EIO;
 337 }
 338
 339 /**
 340  * gfs2_inode_refresh - Refresh the incore copy of the dinode
 341  * @ip: The GFS2 inode
 342  *
 343  * Returns: errno
 344  */
 345
 346 int gfs2_inode_refresh(struct gfs2_inode *ip)
 347 {
 348         struct buffer_head *dibh;
 349         int error;
 350
 351         error = gfs2_meta_inode_buffer(ip, &dibh);
 352         if (error)
 353                 return error;
 354
 355         if (gfs2_metatype_check(GFS2_SB(&ip->i_inode), dibh, GFS2_METATYPE_DI)) {
 356                 brelse(dibh);
 357                 return -EIO;
 358         }
 359
 360         error = gfs2_dinode_in(ip, dibh->b_data);
 361         brelse(dibh);
 362         clear_bit(GIF_INVALID, &ip->i_flags);
 363
 364         return error;
 365 }
 366
 367 /**
 368  * inode_go_lock - operation done after an inode lock is locked by a process
 369  * @gl: the glock
 370  * @flags:
 371  *
 372  * Returns: errno
 373  */
 374
 375 static int inode_go_lock(struct gfs2_holder *gh)
 376 {
 377         struct gfs2_glock *gl = gh->gh_gl;
 378         struct gfs2_sbd *sdp = gl->gl_sbd;
 379         struct gfs2_inode *ip = gl->gl_object;
 380         int error = 0;
 381
 382         if (!ip || (gh->gh_flags & GL_SKIP))
 383                 return 0;
 384
 385         if (test_bit(GIF_INVALID, &ip->i_flags)) {
 386                 error = gfs2_inode_refresh(ip);
 387                 if (error)
 388                         return error;
 389         }
 390
 391         if ((ip->i_diskflags & GFS2_DIF_TRUNC_IN_PROG) &&
 392             (gl->gl_state == LM_ST_EXCLUSIVE) &&
 393             (gh->gh_state == LM_ST_EXCLUSIVE)) {
 394                 spin_lock(&sdp->sd_trunc_lock);
 395                 if (list_empty(&ip->i_trunc_list))
 396                         list_add(&sdp->sd_trunc_list, &ip->i_trunc_list);
 397                 spin_unlock(&sdp->sd_trunc_lock);
 398                 wake_up(&sdp->sd_quota_wait);
 399                 return 1;
 400         }
 401
 402         return error;
 403 }
 404
 405 /**
 406  * inode_go_dump - print information about an inode
 407  * @seq: The iterator
 408  * @ip: the inode
 409  *
 410  * Returns: 0 on success, -ENOBUFS when we run out of space
 411  */
 412
 413 static int inode_go_dump(struct seq_file *seq, const struct gfs2_glock *gl)
 414 {
 415         const struct gfs2_inode *ip = gl->gl_object;
 416         if (ip == NULL)
 417                 return 0;
 418         gfs2_print_dbg(seq, " I: n:%llu/%llu t:%u f:0x%02lx d:0x%08x s:%llu\n",
 419                   (unsigned long long)ip->i_no_formal_ino,
 420                   (unsigned long long)ip->i_no_addr,
 421                   IF2DT(ip->i_inode.i_mode), ip->i_flags,
 422                   (unsigned int)ip->i_diskflags,
 423                   (unsigned long long)i_size_read(&ip->i_inode));
 424         return 0;
 425 }
 426
 427 /**
 428  * rgrp_go_lock - operation done after an rgrp lock is locked by
 429  *    a first holder on this node.
 430  * @gl: the glock
 431  * @flags:
 432  *
 433  * Returns: errno
 434  */
 435
 436 static int rgrp_go_lock(struct gfs2_holder *gh)
 437 {
 438         return gfs2_rgrp_bh_get(gh->gh_gl->gl_object);
 439 }
 440
 441 /**
 442  * rgrp_go_unlock - operation done before an rgrp lock is unlocked by
 443  *    a last holder on this node.
 444  * @gl: the glock
 445  * @flags:
 446  *
 447  */
 448
 449 static void rgrp_go_unlock(struct gfs2_holder *gh)
 450 {
 451         gfs2_rgrp_bh_put(gh->gh_gl->gl_object);
 452 }
 453
 454 /**
 455  * trans_go_sync - promote/demote the transaction glock
 456  * @gl: the glock
 457  * @state: the requested state
 458  * @flags:
 459  *
 460  */
 461
 462 static void trans_go_sync(struct gfs2_glock *gl)
 463 {
 464         struct gfs2_sbd *sdp = gl->gl_sbd;
 465
 466         if (gl->gl_state != LM_ST_UNLOCKED &&
 467             test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
 468                 gfs2_meta_syncfs(sdp);
 469                 gfs2_log_shutdown(sdp);
 470         }
 471 }
 472
 473 /**
 474  * trans_go_xmote_bh - After promoting/demoting the transaction glock
 475  * @gl: the glock
 476  *
 477  */
 478
 479 static int trans_go_xmote_bh(struct gfs2_glock *gl, struct gfs2_holder *gh)
 480 {
 481         struct gfs2_sbd *sdp = gl->gl_sbd;
 482         struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
 483         struct gfs2_glock *j_gl = ip->i_gl;
 484         struct gfs2_log_header_host head;
 485         int error;
 486
 487         if (test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
 488                 j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
 489
 490                 error = gfs2_find_jhead(sdp->sd_jdesc, &head);
 491                 if (error)
 492                         gfs2_consist(sdp);
 493                 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT))
 494                         gfs2_consist(sdp);
 495
 496                 /*  Initialize some head of the log stuff  */
 497                 if (!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) {
 498                         sdp->sd_log_sequence = head.lh_sequence + 1;
 499                         gfs2_log_pointers_init(sdp, head.lh_blkno);
 500                 }
 501         }
 502         return 0;
 503 }
 504
 505 /**
 506  * trans_go_demote_ok
 507  * @gl: the glock
 508  *
 509  * Always returns 0
 510  */
 511
 512 static int trans_go_demote_ok(const struct gfs2_glock *gl)
 513 {
 514         return 0;
 515 }
 516
 517 /**
 518  * iopen_go_callback - schedule the dcache entry for the inode to be deleted
 519  * @gl: the glock
 520  *
 521  * gl_spin lock is held while calling this
 522  */
 523 static void iopen_go_callback(struct gfs2_glock *gl)
 524 {
 525         struct gfs2_inode *ip = (struct gfs2_inode *)gl->gl_object;
 526         struct gfs2_sbd *sdp = gl->gl_sbd;
 527
 528         if (sdp->sd_vfs->s_flags & MS_RDONLY)
 529                 return;
 530
 531         if (gl->gl_demote_state == LM_ST_UNLOCKED &&
 532             gl->gl_state == LM_ST_SHARED && ip) {
 533                 gfs2_glock_hold(gl);
 534                 if (queue_work(gfs2_delete_workqueue, &gl->gl_delete) == 0)
 535                         gfs2_glock_put_nolock(gl);
 536         }
 537 }
 538
 539 const struct gfs2_glock_operations gfs2_meta_glops = {
 540         .go_type = LM_TYPE_META,
 541 };
 542
 543 const struct gfs2_glock_operations gfs2_inode_glops = {
 544         .go_xmote_th = inode_go_sync,
 545         .go_inval = inode_go_inval,
 546         .go_demote_ok = inode_go_demote_ok,
 547         .go_lock = inode_go_lock,
 548         .go_dump = inode_go_dump,
 549         .go_type = LM_TYPE_INODE,
 550         .go_min_hold_time = HZ / 5,
 551         .go_flags = GLOF_ASPACE,
 552 };
 553
 554 const struct gfs2_glock_operations gfs2_rgrp_glops = {
 555         .go_xmote_th = rgrp_go_sync,
 556         .go_inval = rgrp_go_inval,
 557         .go_lock = rgrp_go_lock,
 558         .go_unlock = rgrp_go_unlock,
 559         .go_dump = gfs2_rgrp_dump,
 560         .go_type = LM_TYPE_RGRP,
 561         .go_min_hold_time = HZ / 5,
 562         .go_flags = GLOF_ASPACE,
 563 };
 564
 565 const struct gfs2_glock_operations gfs2_trans_glops = {
 566         .go_xmote_th = trans_go_sync,
 567         .go_xmote_bh = trans_go_xmote_bh,
 568         .go_demote_ok = trans_go_demote_ok,
 569         .go_type = LM_TYPE_NONDISK,
 570 };
 571
 572 const struct gfs2_glock_operations gfs2_iopen_glops = {
 573         .go_type = LM_TYPE_IOPEN,
 574         .go_callback = iopen_go_callback,
 575 };
 576
 577 const struct gfs2_glock_operations gfs2_flock_glops = {
 578         .go_type = LM_TYPE_FLOCK,
 579 };
 580
 581 const struct gfs2_glock_operations gfs2_nondisk_glops = {
 582         .go_type = LM_TYPE_NONDISK,
 583 };
 584
 585 const struct gfs2_glock_operations gfs2_quota_glops = {
 586         .go_type = LM_TYPE_QUOTA,
 587 };
 588
 589 const struct gfs2_glock_operations gfs2_journal_glops = {
 590         .go_type = LM_TYPE_JOURNAL,
 591 };
 592
 593 const struct gfs2_glock_operations *gfs2_glops_list[] = {
 594         [LM_TYPE_META] = &gfs2_meta_glops,
 595         [LM_TYPE_INODE] = &gfs2_inode_glops,
 596         [LM_TYPE_RGRP] = &gfs2_rgrp_glops,
 597         [LM_TYPE_IOPEN] = &gfs2_iopen_glops,
 598         [LM_TYPE_FLOCK] = &gfs2_flock_glops,
 599         [LM_TYPE_NONDISK] = &gfs2_nondisk_glops,
 600         [LM_TYPE_QUOTA] = &gfs2_quota_glops,
 601         [LM_TYPE_JOURNAL] = &gfs2_journal_glops,
 602 };
 603