linux/fs/gfs2/glops.c
<<
>>
Prefs
   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
  31static 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
  50static 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
  83static 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        sb_start_intwrite(sdp->sd_vfs);
  98        gfs2_log_reserve(sdp, tr.tr_reserved);
  99        BUG_ON(current->journal_info);
 100        current->journal_info = &tr;
 101
 102        __gfs2_ail_flush(gl, 0);
 103
 104        gfs2_trans_end(sdp);
 105        gfs2_log_flush(sdp, NULL);
 106}
 107
 108void gfs2_ail_flush(struct gfs2_glock *gl, bool fsync)
 109{
 110        struct gfs2_sbd *sdp = gl->gl_sbd;
 111        unsigned int revokes = atomic_read(&gl->gl_ail_count);
 112        int ret;
 113
 114        if (!revokes)
 115                return;
 116
 117        ret = gfs2_trans_begin(sdp, 0, revokes);
 118        if (ret)
 119                return;
 120        __gfs2_ail_flush(gl, fsync);
 121        gfs2_trans_end(sdp);
 122        gfs2_log_flush(sdp, NULL);
 123}
 124
 125/**
 126 * rgrp_go_sync - sync out the metadata for this glock
 127 * @gl: the glock
 128 *
 129 * Called when demoting or unlocking an EX glock.  We must flush
 130 * to disk all dirty buffers/pages relating to this glock, and must not
 131 * not return to caller to demote/unlock the glock until I/O is complete.
 132 */
 133
 134static void rgrp_go_sync(struct gfs2_glock *gl)
 135{
 136        struct address_space *metamapping = gfs2_glock2aspace(gl);
 137        struct gfs2_rgrpd *rgd;
 138        int error;
 139
 140        if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags))
 141                return;
 142        BUG_ON(gl->gl_state != LM_ST_EXCLUSIVE);
 143
 144        gfs2_log_flush(gl->gl_sbd, gl);
 145        filemap_fdatawrite(metamapping);
 146        error = filemap_fdatawait(metamapping);
 147        mapping_set_error(metamapping, error);
 148        gfs2_ail_empty_gl(gl);
 149
 150        spin_lock(&gl->gl_spin);
 151        rgd = gl->gl_object;
 152        if (rgd)
 153                gfs2_free_clones(rgd);
 154        spin_unlock(&gl->gl_spin);
 155}
 156
 157/**
 158 * rgrp_go_inval - invalidate the metadata for this glock
 159 * @gl: the glock
 160 * @flags:
 161 *
 162 * We never used LM_ST_DEFERRED with resource groups, so that we
 163 * should always see the metadata flag set here.
 164 *
 165 */
 166
 167static void rgrp_go_inval(struct gfs2_glock *gl, int flags)
 168{
 169        struct address_space *mapping = gfs2_glock2aspace(gl);
 170
 171        BUG_ON(!(flags & DIO_METADATA));
 172        gfs2_assert_withdraw(gl->gl_sbd, !atomic_read(&gl->gl_ail_count));
 173        truncate_inode_pages(mapping, 0);
 174
 175        if (gl->gl_object) {
 176                struct gfs2_rgrpd *rgd = (struct gfs2_rgrpd *)gl->gl_object;
 177                rgd->rd_flags &= ~GFS2_RDF_UPTODATE;
 178        }
 179}
 180
 181/**
 182 * inode_go_sync - Sync the dirty data and/or metadata for an inode glock
 183 * @gl: the glock protecting the inode
 184 *
 185 */
 186
 187static void inode_go_sync(struct gfs2_glock *gl)
 188{
 189        struct gfs2_inode *ip = gl->gl_object;
 190        struct address_space *metamapping = gfs2_glock2aspace(gl);
 191        int error;
 192
 193        if (ip && !S_ISREG(ip->i_inode.i_mode))
 194                ip = NULL;
 195        if (ip && test_and_clear_bit(GIF_SW_PAGED, &ip->i_flags))
 196                unmap_shared_mapping_range(ip->i_inode.i_mapping, 0, 0);
 197        if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags))
 198                return;
 199
 200        BUG_ON(gl->gl_state != LM_ST_EXCLUSIVE);
 201
 202        gfs2_log_flush(gl->gl_sbd, gl);
 203        filemap_fdatawrite(metamapping);
 204        if (ip) {
 205                struct address_space *mapping = ip->i_inode.i_mapping;
 206                filemap_fdatawrite(mapping);
 207                error = filemap_fdatawait(mapping);
 208                mapping_set_error(mapping, error);
 209        }
 210        error = filemap_fdatawait(metamapping);
 211        mapping_set_error(metamapping, error);
 212        gfs2_ail_empty_gl(gl);
 213        /*
 214         * Writeback of the data mapping may cause the dirty flag to be set
 215         * so we have to clear it again here.
 216         */
 217        smp_mb__before_clear_bit();
 218        clear_bit(GLF_DIRTY, &gl->gl_flags);
 219}
 220
 221/**
 222 * inode_go_inval - prepare a inode glock to be released
 223 * @gl: the glock
 224 * @flags:
 225 * 
 226 * Normally we invlidate everything, but if we are moving into
 227 * LM_ST_DEFERRED from LM_ST_SHARED or LM_ST_EXCLUSIVE then we
 228 * can keep hold of the metadata, since it won't have changed.
 229 *
 230 */
 231
 232static void inode_go_inval(struct gfs2_glock *gl, int flags)
 233{
 234        struct gfs2_inode *ip = gl->gl_object;
 235
 236        gfs2_assert_withdraw(gl->gl_sbd, !atomic_read(&gl->gl_ail_count));
 237
 238        if (flags & DIO_METADATA) {
 239                struct address_space *mapping = gfs2_glock2aspace(gl);
 240                truncate_inode_pages(mapping, 0);
 241                if (ip) {
 242                        set_bit(GIF_INVALID, &ip->i_flags);
 243                        forget_all_cached_acls(&ip->i_inode);
 244                        gfs2_dir_hash_inval(ip);
 245                }
 246        }
 247
 248        if (ip == GFS2_I(gl->gl_sbd->sd_rindex)) {
 249                gfs2_log_flush(gl->gl_sbd, NULL);
 250                gl->gl_sbd->sd_rindex_uptodate = 0;
 251        }
 252        if (ip && S_ISREG(ip->i_inode.i_mode))
 253                truncate_inode_pages(ip->i_inode.i_mapping, 0);
 254}
 255
 256/**
 257 * inode_go_demote_ok - Check to see if it's ok to unlock an inode glock
 258 * @gl: the glock
 259 *
 260 * Returns: 1 if it's ok
 261 */
 262
 263static int inode_go_demote_ok(const struct gfs2_glock *gl)
 264{
 265        struct gfs2_sbd *sdp = gl->gl_sbd;
 266        struct gfs2_holder *gh;
 267
 268        if (sdp->sd_jindex == gl->gl_object || sdp->sd_rindex == gl->gl_object)
 269                return 0;
 270
 271        if (!list_empty(&gl->gl_holders)) {
 272                gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
 273                if (gh->gh_list.next != &gl->gl_holders)
 274                        return 0;
 275        }
 276
 277        return 1;
 278}
 279
 280/**
 281 * gfs2_set_nlink - Set the inode's link count based on on-disk info
 282 * @inode: The inode in question
 283 * @nlink: The link count
 284 *
 285 * If the link count has hit zero, it must never be raised, whatever the
 286 * on-disk inode might say. When new struct inodes are created the link
 287 * count is set to 1, so that we can safely use this test even when reading
 288 * in on disk information for the first time.
 289 */
 290
 291static void gfs2_set_nlink(struct inode *inode, u32 nlink)
 292{
 293        /*
 294         * We will need to review setting the nlink count here in the
 295         * light of the forthcoming ro bind mount work. This is a reminder
 296         * to do that.
 297         */
 298        if ((inode->i_nlink != nlink) && (inode->i_nlink != 0)) {
 299                if (nlink == 0)
 300                        clear_nlink(inode);
 301                else
 302                        set_nlink(inode, nlink);
 303        }
 304}
 305
 306static int gfs2_dinode_in(struct gfs2_inode *ip, const void *buf)
 307{
 308        const struct gfs2_dinode *str = buf;
 309        struct timespec atime;
 310        u16 height, depth;
 311
 312        if (unlikely(ip->i_no_addr != be64_to_cpu(str->di_num.no_addr)))
 313                goto corrupt;
 314        ip->i_no_formal_ino = be64_to_cpu(str->di_num.no_formal_ino);
 315        ip->i_inode.i_mode = be32_to_cpu(str->di_mode);
 316        ip->i_inode.i_rdev = 0;
 317        switch (ip->i_inode.i_mode & S_IFMT) {
 318        case S_IFBLK:
 319        case S_IFCHR:
 320                ip->i_inode.i_rdev = MKDEV(be32_to_cpu(str->di_major),
 321                                           be32_to_cpu(str->di_minor));
 322                break;
 323        };
 324
 325        ip->i_inode.i_uid = be32_to_cpu(str->di_uid);
 326        ip->i_inode.i_gid = be32_to_cpu(str->di_gid);
 327        gfs2_set_nlink(&ip->i_inode, be32_to_cpu(str->di_nlink));
 328        i_size_write(&ip->i_inode, be64_to_cpu(str->di_size));
 329        gfs2_set_inode_blocks(&ip->i_inode, be64_to_cpu(str->di_blocks));
 330        atime.tv_sec = be64_to_cpu(str->di_atime);
 331        atime.tv_nsec = be32_to_cpu(str->di_atime_nsec);
 332        if (timespec_compare(&ip->i_inode.i_atime, &atime) < 0)
 333                ip->i_inode.i_atime = atime;
 334        ip->i_inode.i_mtime.tv_sec = be64_to_cpu(str->di_mtime);
 335        ip->i_inode.i_mtime.tv_nsec = be32_to_cpu(str->di_mtime_nsec);
 336        ip->i_inode.i_ctime.tv_sec = be64_to_cpu(str->di_ctime);
 337        ip->i_inode.i_ctime.tv_nsec = be32_to_cpu(str->di_ctime_nsec);
 338
 339        ip->i_goal = be64_to_cpu(str->di_goal_meta);
 340        ip->i_generation = be64_to_cpu(str->di_generation);
 341
 342        ip->i_diskflags = be32_to_cpu(str->di_flags);
 343        ip->i_eattr = be64_to_cpu(str->di_eattr);
 344        /* i_diskflags and i_eattr must be set before gfs2_set_inode_flags() */
 345        gfs2_set_inode_flags(&ip->i_inode);
 346        height = be16_to_cpu(str->di_height);
 347        if (unlikely(height > GFS2_MAX_META_HEIGHT))
 348                goto corrupt;
 349        ip->i_height = (u8)height;
 350
 351        depth = be16_to_cpu(str->di_depth);
 352        if (unlikely(depth > GFS2_DIR_MAX_DEPTH))
 353                goto corrupt;
 354        ip->i_depth = (u8)depth;
 355        ip->i_entries = be32_to_cpu(str->di_entries);
 356
 357        if (S_ISREG(ip->i_inode.i_mode))
 358                gfs2_set_aops(&ip->i_inode);
 359
 360        return 0;
 361corrupt:
 362        gfs2_consist_inode(ip);
 363        return -EIO;
 364}
 365
 366/**
 367 * gfs2_inode_refresh - Refresh the incore copy of the dinode
 368 * @ip: The GFS2 inode
 369 *
 370 * Returns: errno
 371 */
 372
 373int gfs2_inode_refresh(struct gfs2_inode *ip)
 374{
 375        struct buffer_head *dibh;
 376        int error;
 377
 378        error = gfs2_meta_inode_buffer(ip, &dibh);
 379        if (error)
 380                return error;
 381
 382        error = gfs2_dinode_in(ip, dibh->b_data);
 383        brelse(dibh);
 384        clear_bit(GIF_INVALID, &ip->i_flags);
 385
 386        return error;
 387}
 388
 389/**
 390 * inode_go_lock - operation done after an inode lock is locked by a process
 391 * @gl: the glock
 392 * @flags:
 393 *
 394 * Returns: errno
 395 */
 396
 397static int inode_go_lock(struct gfs2_holder *gh)
 398{
 399        struct gfs2_glock *gl = gh->gh_gl;
 400        struct gfs2_sbd *sdp = gl->gl_sbd;
 401        struct gfs2_inode *ip = gl->gl_object;
 402        int error = 0;
 403
 404        if (!ip || (gh->gh_flags & GL_SKIP))
 405                return 0;
 406
 407        if (test_bit(GIF_INVALID, &ip->i_flags)) {
 408                error = gfs2_inode_refresh(ip);
 409                if (error)
 410                        return error;
 411        }
 412
 413        if ((ip->i_diskflags & GFS2_DIF_TRUNC_IN_PROG) &&
 414            (gl->gl_state == LM_ST_EXCLUSIVE) &&
 415            (gh->gh_state == LM_ST_EXCLUSIVE)) {
 416                spin_lock(&sdp->sd_trunc_lock);
 417                if (list_empty(&ip->i_trunc_list))
 418                        list_add(&sdp->sd_trunc_list, &ip->i_trunc_list);
 419                spin_unlock(&sdp->sd_trunc_lock);
 420                wake_up(&sdp->sd_quota_wait);
 421                return 1;
 422        }
 423
 424        return error;
 425}
 426
 427/**
 428 * inode_go_dump - print information about an inode
 429 * @seq: The iterator
 430 * @ip: the inode
 431 *
 432 * Returns: 0 on success, -ENOBUFS when we run out of space
 433 */
 434
 435static int inode_go_dump(struct seq_file *seq, const struct gfs2_glock *gl)
 436{
 437        const struct gfs2_inode *ip = gl->gl_object;
 438        if (ip == NULL)
 439                return 0;
 440        gfs2_print_dbg(seq, " I: n:%llu/%llu t:%u f:0x%02lx d:0x%08x s:%llu\n",
 441                  (unsigned long long)ip->i_no_formal_ino,
 442                  (unsigned long long)ip->i_no_addr,
 443                  IF2DT(ip->i_inode.i_mode), ip->i_flags,
 444                  (unsigned int)ip->i_diskflags,
 445                  (unsigned long long)i_size_read(&ip->i_inode));
 446        return 0;
 447}
 448
 449/**
 450 * trans_go_sync - promote/demote the transaction glock
 451 * @gl: the glock
 452 * @state: the requested state
 453 * @flags:
 454 *
 455 */
 456
 457static void trans_go_sync(struct gfs2_glock *gl)
 458{
 459        struct gfs2_sbd *sdp = gl->gl_sbd;
 460
 461        if (gl->gl_state != LM_ST_UNLOCKED &&
 462            test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
 463                gfs2_meta_syncfs(sdp);
 464                gfs2_log_shutdown(sdp);
 465        }
 466}
 467
 468/**
 469 * trans_go_xmote_bh - After promoting/demoting the transaction glock
 470 * @gl: the glock
 471 *
 472 */
 473
 474static int trans_go_xmote_bh(struct gfs2_glock *gl, struct gfs2_holder *gh)
 475{
 476        struct gfs2_sbd *sdp = gl->gl_sbd;
 477        struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
 478        struct gfs2_glock *j_gl = ip->i_gl;
 479        struct gfs2_log_header_host head;
 480        int error;
 481
 482        if (test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
 483                j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
 484
 485                error = gfs2_find_jhead(sdp->sd_jdesc, &head);
 486                if (error)
 487                        gfs2_consist(sdp);
 488                if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT))
 489                        gfs2_consist(sdp);
 490
 491                /*  Initialize some head of the log stuff  */
 492                if (!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) {
 493                        sdp->sd_log_sequence = head.lh_sequence + 1;
 494                        gfs2_log_pointers_init(sdp, head.lh_blkno);
 495                }
 496        }
 497        return 0;
 498}
 499
 500/**
 501 * trans_go_demote_ok
 502 * @gl: the glock
 503 *
 504 * Always returns 0
 505 */
 506
 507static int trans_go_demote_ok(const struct gfs2_glock *gl)
 508{
 509        return 0;
 510}
 511
 512/**
 513 * iopen_go_callback - schedule the dcache entry for the inode to be deleted
 514 * @gl: the glock
 515 *
 516 * gl_spin lock is held while calling this
 517 */
 518static void iopen_go_callback(struct gfs2_glock *gl)
 519{
 520        struct gfs2_inode *ip = (struct gfs2_inode *)gl->gl_object;
 521        struct gfs2_sbd *sdp = gl->gl_sbd;
 522
 523        if (sdp->sd_vfs->s_flags & MS_RDONLY)
 524                return;
 525
 526        if (gl->gl_demote_state == LM_ST_UNLOCKED &&
 527            gl->gl_state == LM_ST_SHARED && ip) {
 528                gfs2_glock_hold(gl);
 529                if (queue_work(gfs2_delete_workqueue, &gl->gl_delete) == 0)
 530                        gfs2_glock_put_nolock(gl);
 531        }
 532}
 533
 534const struct gfs2_glock_operations gfs2_meta_glops = {
 535        .go_type = LM_TYPE_META,
 536};
 537
 538const struct gfs2_glock_operations gfs2_inode_glops = {
 539        .go_xmote_th = inode_go_sync,
 540        .go_inval = inode_go_inval,
 541        .go_demote_ok = inode_go_demote_ok,
 542        .go_lock = inode_go_lock,
 543        .go_dump = inode_go_dump,
 544        .go_type = LM_TYPE_INODE,
 545        .go_flags = GLOF_ASPACE,
 546};
 547
 548const struct gfs2_glock_operations gfs2_rgrp_glops = {
 549        .go_xmote_th = rgrp_go_sync,
 550        .go_inval = rgrp_go_inval,
 551        .go_lock = gfs2_rgrp_go_lock,
 552        .go_unlock = gfs2_rgrp_go_unlock,
 553        .go_dump = gfs2_rgrp_dump,
 554        .go_type = LM_TYPE_RGRP,
 555        .go_flags = GLOF_ASPACE,
 556};
 557
 558const struct gfs2_glock_operations gfs2_trans_glops = {
 559        .go_xmote_th = trans_go_sync,
 560        .go_xmote_bh = trans_go_xmote_bh,
 561        .go_demote_ok = trans_go_demote_ok,
 562        .go_type = LM_TYPE_NONDISK,
 563};
 564
 565const struct gfs2_glock_operations gfs2_iopen_glops = {
 566        .go_type = LM_TYPE_IOPEN,
 567        .go_callback = iopen_go_callback,
 568};
 569
 570const struct gfs2_glock_operations gfs2_flock_glops = {
 571        .go_type = LM_TYPE_FLOCK,
 572};
 573
 574const struct gfs2_glock_operations gfs2_nondisk_glops = {
 575        .go_type = LM_TYPE_NONDISK,
 576};
 577
 578const struct gfs2_glock_operations gfs2_quota_glops = {
 579        .go_type = LM_TYPE_QUOTA,
 580};
 581
 582const struct gfs2_glock_operations gfs2_journal_glops = {
 583        .go_type = LM_TYPE_JOURNAL,
 584};
 585
 586const struct gfs2_glock_operations *gfs2_glops_list[] = {
 587        [LM_TYPE_META] = &gfs2_meta_glops,
 588        [LM_TYPE_INODE] = &gfs2_inode_glops,
 589        [LM_TYPE_RGRP] = &gfs2_rgrp_glops,
 590        [LM_TYPE_IOPEN] = &gfs2_iopen_glops,
 591        [LM_TYPE_FLOCK] = &gfs2_flock_glops,
 592        [LM_TYPE_NONDISK] = &gfs2_nondisk_glops,
 593        [LM_TYPE_QUOTA] = &gfs2_quota_glops,
 594        [LM_TYPE_JOURNAL] = &gfs2_journal_glops,
 595};
 596
 597
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.