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         sector_t blocknr;
  58
  59         gfs2_log_lock(sdp);
  60         spin_lock(&sdp->sd_ail_lock);
  61         list_for_each_entry_safe(bd, tmp, head, bd_ail_gl_list) {
  62                 bh = bd->bd_bh;
  63                 if (bh->b_state & b_state) {
  64                         if (fsync)
  65                                 continue;
  66                         gfs2_ail_error(gl, bh);
  67                 }
  68                 blocknr = bh->b_blocknr;
  69                 bh->b_private = NULL;
  70                 gfs2_remove_from_ail(bd); /* drops ref on bh */
  71
  72                 bd->bd_bh = NULL;
  73                 bd->bd_blkno = blocknr;
  74
  75                 gfs2_trans_add_revoke(sdp, bd);
  76         }
  77         BUG_ON(!fsync && atomic_read(&gl->gl_ail_count));
  78         spin_unlock(&sdp->sd_ail_lock);
  79         gfs2_log_unlock(sdp);
  80 }
  81
  82
  83 static void gfs2_ail_empty_gl(struct gfs2_glock *gl)
  84 {
  85         struct gfs2_sbd *sdp = gl->gl_sbd;
  86         struct gfs2_trans tr;
  87
  88         memset(&tr, 0, sizeof(tr));
  89         tr.tr_revokes = atomic_read(&gl->gl_ail_count);
  90
  91         if (!tr.tr_revokes)
  92                 return;
  93
  94         /* A shortened, inline version of gfs2_trans_begin() */
  95         tr.tr_reserved = 1 + gfs2_struct2blk(sdp, tr.tr_revokes, sizeof(u64));
  96         tr.tr_ip = (unsigned long)__builtin_return_address(0);
  97         gfs2_log_reserve(sdp, tr.tr_reserved);
  98         BUG_ON(current->journal_info);
  99         current->journal_info = &tr;
 100
 101         __gfs2_ail_flush(gl, 0);
 102
 103         gfs2_trans_end(sdp);
 104         gfs2_log_flush(sdp, NULL);
 105 }
 106
 107 void gfs2_ail_flush(struct gfs2_glock *gl, bool fsync)
 108 {
 109         struct gfs2_sbd *sdp = gl->gl_sbd;
 110         unsigned int revokes = atomic_read(&gl->gl_ail_count);
 111         int ret;
 112
 113         if (!revokes)
 114                 return;
 115
 116         ret = gfs2_trans_begin(sdp, 0, revokes);
 117         if (ret)
 118                 return;
 119         __gfs2_ail_flush(gl, fsync);
 120         gfs2_trans_end(sdp);
 121         gfs2_log_flush(sdp, NULL);
 122 }
 123
 124 /**
 125  * rgrp_go_sync - sync out the metadata for this glock
 126  * @gl: the glock
 127  *
 128  * Called when demoting or unlocking an EX glock.  We must flush
 129  * to disk all dirty buffers/pages relating to this glock, and must not
 130  * not return to caller to demote/unlock the glock until I/O is complete.
 131  */
 132
 133 static void rgrp_go_sync(struct gfs2_glock *gl)
 134 {
 135         struct address_space *metamapping = gfs2_glock2aspace(gl);
 136         struct gfs2_rgrpd *rgd;
 137         int error;
 138
 139         if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags))
 140                 return;
 141         BUG_ON(gl->gl_state != LM_ST_EXCLUSIVE);
 142
 143         gfs2_log_flush(gl->gl_sbd, gl);
 144         filemap_fdatawrite(metamapping);
 145         error = filemap_fdatawait(metamapping);
 146         mapping_set_error(metamapping, error);
 147         gfs2_ail_empty_gl(gl);
 148
 149         spin_lock(&gl->gl_spin);
 150         rgd = gl->gl_object;
 151         if (rgd)
 152                 gfs2_free_clones(rgd);
 153         spin_unlock(&gl->gl_spin);
 154 }
 155
 156 /**
 157  * rgrp_go_inval - invalidate the metadata for this glock
 158  * @gl: the glock
 159  * @flags:
 160  *
 161  * We never used LM_ST_DEFERRED with resource groups, so that we
 162  * should always see the metadata flag set here.
 163  *
 164  */
 165
 166 static void rgrp_go_inval(struct gfs2_glock *gl, int flags)
 167 {
 168         struct address_space *mapping = gfs2_glock2aspace(gl);
 169
 170         BUG_ON(!(flags & DIO_METADATA));
 171         gfs2_assert_withdraw(gl->gl_sbd, !atomic_read(&gl->gl_ail_count));
 172         truncate_inode_pages(mapping, 0);
 173
 174         if (gl->gl_object) {
 175                 struct gfs2_rgrpd *rgd = (struct gfs2_rgrpd *)gl->gl_object;
 176                 rgd->rd_flags &= ~GFS2_RDF_UPTODATE;
 177         }
 178 }
 179
 180 /**
 181  * inode_go_sync - Sync the dirty data and/or metadata for an inode glock
 182  * @gl: the glock protecting the inode
 183  *
 184  */
 185
 186 static void inode_go_sync(struct gfs2_glock *gl)
 187 {
 188         struct gfs2_inode *ip = gl->gl_object;
 189         struct address_space *metamapping = gfs2_glock2aspace(gl);
 190         int error;
 191
 192         if (ip && !S_ISREG(ip->i_inode.i_mode))
 193                 ip = NULL;
 194         if (ip && test_and_clear_bit(GIF_SW_PAGED, &ip->i_flags))
 195                 unmap_shared_mapping_range(ip->i_inode.i_mapping, 0, 0);
 196         if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags))
 197                 return;
 198
 199         BUG_ON(gl->gl_state != LM_ST_EXCLUSIVE);
 200
 201         gfs2_log_flush(gl->gl_sbd, gl);
 202         filemap_fdatawrite(metamapping);
 203         if (ip) {
 204                 struct address_space *mapping = ip->i_inode.i_mapping;
 205                 filemap_fdatawrite(mapping);
 206                 error = filemap_fdatawait(mapping);
 207                 mapping_set_error(mapping, error);
 208         }
 209         error = filemap_fdatawait(metamapping);
 210         mapping_set_error(metamapping, error);
 211         gfs2_ail_empty_gl(gl);
 212         /*
 213          * Writeback of the data mapping may cause the dirty flag to be set
 214          * so we have to clear it again here.
 215          */
 216         smp_mb__before_clear_bit();
 217         clear_bit(GLF_DIRTY, &gl->gl_flags);
 218 }
 219
 220 /**
 221  * inode_go_inval - prepare a inode glock to be released
 222  * @gl: the glock
 223  * @flags:
 224  *
 225  * Normally we invlidate everything, but if we are moving into
 226  * LM_ST_DEFERRED from LM_ST_SHARED or LM_ST_EXCLUSIVE then we
 227  * can keep hold of the metadata, since it won't have changed.
 228  *
 229  */
 230
 231 static void inode_go_inval(struct gfs2_glock *gl, int flags)
 232 {
 233         struct gfs2_inode *ip = gl->gl_object;
 234
 235         gfs2_assert_withdraw(gl->gl_sbd, !atomic_read(&gl->gl_ail_count));
 236
 237         if (flags & DIO_METADATA) {
 238                 struct address_space *mapping = gfs2_glock2aspace(gl);
 239                 truncate_inode_pages(mapping, 0);
 240                 if (ip) {
 241                         set_bit(GIF_INVALID, &ip->i_flags);
 242                         forget_all_cached_acls(&ip->i_inode);
 243                         gfs2_dir_hash_inval(ip);
 244                 }
 245         }
 246
 247         if (ip == GFS2_I(gl->gl_sbd->sd_rindex)) {
 248                 gfs2_log_flush(gl->gl_sbd, NULL);
 249                 gl->gl_sbd->sd_rindex_uptodate = 0;
 250         }
 251         if (ip && S_ISREG(ip->i_inode.i_mode))
 252                 truncate_inode_pages(ip->i_inode.i_mapping, 0);
 253 }
 254
 255 /**
 256  * inode_go_demote_ok - Check to see if it's ok to unlock an inode glock
 257  * @gl: the glock
 258  *
 259  * Returns: 1 if it's ok
 260  */
 261
 262 static int inode_go_demote_ok(const struct gfs2_glock *gl)
 263 {
 264         struct gfs2_sbd *sdp = gl->gl_sbd;
 265         struct gfs2_holder *gh;
 266
 267         if (sdp->sd_jindex == gl->gl_object || sdp->sd_rindex == gl->gl_object)
 268                 return 0;
 269
 270         if (!list_empty(&gl->gl_holders)) {
 271                 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
 272                 if (gh->gh_list.next != &gl->gl_holders)
 273                         return 0;
 274         }
 275
 276         return 1;
 277 }
 278
 279 /**
 280  * gfs2_set_nlink - Set the inode's link count based on on-disk info
 281  * @inode: The inode in question
 282  * @nlink: The link count
 283  *
 284  * If the link count has hit zero, it must never be raised, whatever the
 285  * on-disk inode might say. When new struct inodes are created the link
 286  * count is set to 1, so that we can safely use this test even when reading
 287  * in on disk information for the first time.
 288  */
 289
 290 static void gfs2_set_nlink(struct inode *inode, u32 nlink)
 291 {
 292         /*
 293          * We will need to review setting the nlink count here in the
 294          * light of the forthcoming ro bind mount work. This is a reminder
 295          * to do that.
 296          */
 297         if ((inode->i_nlink != nlink) && (inode->i_nlink != 0)) {
 298                 if (nlink == 0)
 299                         clear_nlink(inode);
 300                 else
 301                         set_nlink(inode, nlink);
 302         }
 303 }
 304
 305 static int gfs2_dinode_in(struct gfs2_inode *ip, const void *buf)
 306 {
 307         const struct gfs2_dinode *str = buf;
 308         struct timespec atime;
 309         u16 height, depth;
 310
 311         if (unlikely(ip->i_no_addr != be64_to_cpu(str->di_num.no_addr)))
 312                 goto corrupt;
 313         ip->i_no_formal_ino = be64_to_cpu(str->di_num.no_formal_ino);
 314         ip->i_inode.i_mode = be32_to_cpu(str->di_mode);
 315         ip->i_inode.i_rdev = 0;
 316         switch (ip->i_inode.i_mode & S_IFMT) {
 317         case S_IFBLK:
 318         case S_IFCHR:
 319                 ip->i_inode.i_rdev = MKDEV(be32_to_cpu(str->di_major),
 320                                            be32_to_cpu(str->di_minor));
 321                 break;
 322         };
 323
 324         ip->i_inode.i_uid = be32_to_cpu(str->di_uid);
 325         ip->i_inode.i_gid = be32_to_cpu(str->di_gid);
 326         gfs2_set_nlink(&ip->i_inode, be32_to_cpu(str->di_nlink));
 327         i_size_write(&ip->i_inode, be64_to_cpu(str->di_size));
 328         gfs2_set_inode_blocks(&ip->i_inode, be64_to_cpu(str->di_blocks));
 329         atime.tv_sec = be64_to_cpu(str->di_atime);
 330         atime.tv_nsec = be32_to_cpu(str->di_atime_nsec);
 331         if (timespec_compare(&ip->i_inode.i_atime, &atime) < 0)
 332                 ip->i_inode.i_atime = atime;
 333         ip->i_inode.i_mtime.tv_sec = be64_to_cpu(str->di_mtime);
 334         ip->i_inode.i_mtime.tv_nsec = be32_to_cpu(str->di_mtime_nsec);
 335         ip->i_inode.i_ctime.tv_sec = be64_to_cpu(str->di_ctime);
 336         ip->i_inode.i_ctime.tv_nsec = be32_to_cpu(str->di_ctime_nsec);
 337
 338         ip->i_goal = be64_to_cpu(str->di_goal_meta);
 339         ip->i_generation = be64_to_cpu(str->di_generation);
 340
 341         ip->i_diskflags = be32_to_cpu(str->di_flags);
 342         ip->i_eattr = be64_to_cpu(str->di_eattr);
 343         /* i_diskflags and i_eattr must be set before gfs2_set_inode_flags() */
 344         gfs2_set_inode_flags(&ip->i_inode);
 345         height = be16_to_cpu(str->di_height);
 346         if (unlikely(height > GFS2_MAX_META_HEIGHT))
 347                 goto corrupt;
 348         ip->i_height = (u8)height;
 349
 350         depth = be16_to_cpu(str->di_depth);
 351         if (unlikely(depth > GFS2_DIR_MAX_DEPTH))
 352                 goto corrupt;
 353         ip->i_depth = (u8)depth;
 354         ip->i_entries = be32_to_cpu(str->di_entries);
 355
 356         if (S_ISREG(ip->i_inode.i_mode))
 357                 gfs2_set_aops(&ip->i_inode);
 358
 359         return 0;
 360 corrupt:
 361         gfs2_consist_inode(ip);
 362         return -EIO;
 363 }
 364
 365 /**
 366  * gfs2_inode_refresh - Refresh the incore copy of the dinode
 367  * @ip: The GFS2 inode
 368  *
 369  * Returns: errno
 370  */
 371
 372 int gfs2_inode_refresh(struct gfs2_inode *ip)
 373 {
 374         struct buffer_head *dibh;
 375         int error;
 376
 377         error = gfs2_meta_inode_buffer(ip, &dibh);
 378         if (error)
 379                 return error;
 380
 381         error = gfs2_dinode_in(ip, dibh->b_data);
 382         brelse(dibh);
 383         clear_bit(GIF_INVALID, &ip->i_flags);
 384
 385         return error;
 386 }
 387
 388 /**
 389  * inode_go_lock - operation done after an inode lock is locked by a process
 390  * @gl: the glock
 391  * @flags:
 392  *
 393  * Returns: errno
 394  */
 395
 396 static int inode_go_lock(struct gfs2_holder *gh)
 397 {
 398         struct gfs2_glock *gl = gh->gh_gl;
 399         struct gfs2_sbd *sdp = gl->gl_sbd;
 400         struct gfs2_inode *ip = gl->gl_object;
 401         int error = 0;
 402
 403         if (!ip || (gh->gh_flags & GL_SKIP))
 404                 return 0;
 405
 406         if (test_bit(GIF_INVALID, &ip->i_flags)) {
 407                 error = gfs2_inode_refresh(ip);
 408                 if (error)
 409                         return error;
 410         }
 411
 412         if ((ip->i_diskflags & GFS2_DIF_TRUNC_IN_PROG) &&
 413             (gl->gl_state == LM_ST_EXCLUSIVE) &&
 414             (gh->gh_state == LM_ST_EXCLUSIVE)) {
 415                 spin_lock(&sdp->sd_trunc_lock);
 416                 if (list_empty(&ip->i_trunc_list))
 417                         list_add(&sdp->sd_trunc_list, &ip->i_trunc_list);
 418                 spin_unlock(&sdp->sd_trunc_lock);
 419                 wake_up(&sdp->sd_quota_wait);
 420                 return 1;
 421         }
 422
 423         return error;
 424 }
 425
 426 /**
 427  * inode_go_dump - print information about an inode
 428  * @seq: The iterator
 429  * @ip: the inode
 430  *
 431  * Returns: 0 on success, -ENOBUFS when we run out of space
 432  */
 433
 434 static int inode_go_dump(struct seq_file *seq, const struct gfs2_glock *gl)
 435 {
 436         const struct gfs2_inode *ip = gl->gl_object;
 437         if (ip == NULL)
 438                 return 0;
 439         gfs2_print_dbg(seq, " I: n:%llu/%llu t:%u f:0x%02lx d:0x%08x s:%llu\n",
 440                   (unsigned long long)ip->i_no_formal_ino,
 441                   (unsigned long long)ip->i_no_addr,
 442                   IF2DT(ip->i_inode.i_mode), ip->i_flags,
 443                   (unsigned int)ip->i_diskflags,
 444                   (unsigned long long)i_size_read(&ip->i_inode));
 445         return 0;
 446 }
 447
 448 /**
 449  * trans_go_sync - promote/demote the transaction glock
 450  * @gl: the glock
 451  * @state: the requested state
 452  * @flags:
 453  *
 454  */
 455
 456 static void trans_go_sync(struct gfs2_glock *gl)
 457 {
 458         struct gfs2_sbd *sdp = gl->gl_sbd;
 459
 460         if (gl->gl_state != LM_ST_UNLOCKED &&
 461             test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
 462                 gfs2_meta_syncfs(sdp);
 463                 gfs2_log_shutdown(sdp);
 464         }
 465 }
 466
 467 /**
 468  * trans_go_xmote_bh - After promoting/demoting the transaction glock
 469  * @gl: the glock
 470  *
 471  */
 472
 473 static int trans_go_xmote_bh(struct gfs2_glock *gl, struct gfs2_holder *gh)
 474 {
 475         struct gfs2_sbd *sdp = gl->gl_sbd;
 476         struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
 477         struct gfs2_glock *j_gl = ip->i_gl;
 478         struct gfs2_log_header_host head;
 479         int error;
 480
 481         if (test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
 482                 j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
 483
 484                 error = gfs2_find_jhead(sdp->sd_jdesc, &head);
 485                 if (error)
 486                         gfs2_consist(sdp);
 487                 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT))
 488                         gfs2_consist(sdp);
 489
 490                 /*  Initialize some head of the log stuff  */
 491                 if (!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) {
 492                         sdp->sd_log_sequence = head.lh_sequence + 1;
 493                         gfs2_log_pointers_init(sdp, head.lh_blkno);
 494                 }
 495         }
 496         return 0;
 497 }
 498
 499 /**
 500  * trans_go_demote_ok
 501  * @gl: the glock
 502  *
 503  * Always returns 0
 504  */
 505
 506 static int trans_go_demote_ok(const struct gfs2_glock *gl)
 507 {
 508         return 0;
 509 }
 510
 511 /**
 512  * iopen_go_callback - schedule the dcache entry for the inode to be deleted
 513  * @gl: the glock
 514  *
 515  * gl_spin lock is held while calling this
 516  */
 517 static void iopen_go_callback(struct gfs2_glock *gl)
 518 {
 519         struct gfs2_inode *ip = (struct gfs2_inode *)gl->gl_object;
 520         struct gfs2_sbd *sdp = gl->gl_sbd;
 521
 522         if (sdp->sd_vfs->s_flags & MS_RDONLY)
 523                 return;
 524
 525         if (gl->gl_demote_state == LM_ST_UNLOCKED &&
 526             gl->gl_state == LM_ST_SHARED && ip) {
 527                 gfs2_glock_hold(gl);
 528                 if (queue_work(gfs2_delete_workqueue, &gl->gl_delete) == 0)
 529                         gfs2_glock_put_nolock(gl);
 530         }
 531 }
 532
 533 const struct gfs2_glock_operations gfs2_meta_glops = {
 534         .go_type = LM_TYPE_META,
 535 };
 536
 537 const struct gfs2_glock_operations gfs2_inode_glops = {
 538         .go_xmote_th = inode_go_sync,
 539         .go_inval = inode_go_inval,
 540         .go_demote_ok = inode_go_demote_ok,
 541         .go_lock = inode_go_lock,
 542         .go_dump = inode_go_dump,
 543         .go_type = LM_TYPE_INODE,
 544         .go_flags = GLOF_ASPACE,
 545 };
 546
 547 const struct gfs2_glock_operations gfs2_rgrp_glops = {
 548         .go_xmote_th = rgrp_go_sync,
 549         .go_inval = rgrp_go_inval,
 550         .go_lock = gfs2_rgrp_go_lock,
 551         .go_unlock = gfs2_rgrp_go_unlock,
 552         .go_dump = gfs2_rgrp_dump,
 553         .go_type = LM_TYPE_RGRP,
 554         .go_flags = GLOF_ASPACE,
 555 };
 556
 557 const struct gfs2_glock_operations gfs2_trans_glops = {
 558         .go_xmote_th = trans_go_sync,
 559         .go_xmote_bh = trans_go_xmote_bh,
 560         .go_demote_ok = trans_go_demote_ok,
 561         .go_type = LM_TYPE_NONDISK,
 562 };
 563
 564 const struct gfs2_glock_operations gfs2_iopen_glops = {
 565         .go_type = LM_TYPE_IOPEN,
 566         .go_callback = iopen_go_callback,
 567 };
 568
 569 const struct gfs2_glock_operations gfs2_flock_glops = {
 570         .go_type = LM_TYPE_FLOCK,
 571 };
 572
 573 const struct gfs2_glock_operations gfs2_nondisk_glops = {
 574         .go_type = LM_TYPE_NONDISK,
 575 };
 576
 577 const struct gfs2_glock_operations gfs2_quota_glops = {
 578         .go_type = LM_TYPE_QUOTA,
 579 };
 580
 581 const struct gfs2_glock_operations gfs2_journal_glops = {
 582         .go_type = LM_TYPE_JOURNAL,
 583 };
 584
 585 const struct gfs2_glock_operations *gfs2_glops_list[] = {
 586         [LM_TYPE_META] = &gfs2_meta_glops,
 587         [LM_TYPE_INODE] = &gfs2_inode_glops,
 588         [LM_TYPE_RGRP] = &gfs2_rgrp_glops,
 589         [LM_TYPE_IOPEN] = &gfs2_iopen_glops,
 590         [LM_TYPE_FLOCK] = &gfs2_flock_glops,
 591         [LM_TYPE_NONDISK] = &gfs2_nondisk_glops,
 592         [LM_TYPE_QUOTA] = &gfs2_quota_glops,
 593         [LM_TYPE_JOURNAL] = &gfs2_journal_glops,
 594 };
 595