fs/gfs2/super.c

   1 /*
   2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   3  * Copyright (C) 2004-2007 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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  11
  12 #include <linux/bio.h>
  13 #include <linux/sched.h>
  14 #include <linux/slab.h>
  15 #include <linux/spinlock.h>
  16 #include <linux/completion.h>
  17 #include <linux/buffer_head.h>
  18 #include <linux/statfs.h>
  19 #include <linux/seq_file.h>
  20 #include <linux/mount.h>
  21 #include <linux/kthread.h>
  22 #include <linux/delay.h>
  23 #include <linux/gfs2_ondisk.h>
  24 #include <linux/crc32.h>
  25 #include <linux/time.h>
  26 #include <linux/wait.h>
  27 #include <linux/writeback.h>
  28 #include <linux/backing-dev.h>
  29
  30 #include "gfs2.h"
  31 #include "incore.h"
  32 #include "bmap.h"
  33 #include "dir.h"
  34 #include "glock.h"
  35 #include "glops.h"
  36 #include "inode.h"
  37 #include "log.h"
  38 #include "meta_io.h"
  39 #include "quota.h"
  40 #include "recovery.h"
  41 #include "rgrp.h"
  42 #include "super.h"
  43 #include "trans.h"
  44 #include "util.h"
  45 #include "sys.h"
  46 #include "xattr.h"
  47
  48 #define args_neq(a1, a2, x) ((a1)->ar_##x != (a2)->ar_##x)
  49
  50 enum {
  51         Opt_lockproto,
  52         Opt_locktable,
  53         Opt_hostdata,
  54         Opt_spectator,
  55         Opt_ignore_local_fs,
  56         Opt_localflocks,
  57         Opt_localcaching,
  58         Opt_debug,
  59         Opt_nodebug,
  60         Opt_upgrade,
  61         Opt_acl,
  62         Opt_noacl,
  63         Opt_quota_off,
  64         Opt_quota_account,
  65         Opt_quota_on,
  66         Opt_quota,
  67         Opt_noquota,
  68         Opt_suiddir,
  69         Opt_nosuiddir,
  70         Opt_data_writeback,
  71         Opt_data_ordered,
  72         Opt_meta,
  73         Opt_discard,
  74         Opt_nodiscard,
  75         Opt_commit,
  76         Opt_err_withdraw,
  77         Opt_err_panic,
  78         Opt_statfs_quantum,
  79         Opt_statfs_percent,
  80         Opt_quota_quantum,
  81         Opt_barrier,
  82         Opt_nobarrier,
  83         Opt_rgrplvb,
  84         Opt_norgrplvb,
  85         Opt_error,
  86 };
  87
  88 static const match_table_t tokens = {
  89         {Opt_lockproto, "lockproto=%s"},
  90         {Opt_locktable, "locktable=%s"},
  91         {Opt_hostdata, "hostdata=%s"},
  92         {Opt_spectator, "spectator"},
  93         {Opt_spectator, "norecovery"},
  94         {Opt_ignore_local_fs, "ignore_local_fs"},
  95         {Opt_localflocks, "localflocks"},
  96         {Opt_localcaching, "localcaching"},
  97         {Opt_debug, "debug"},
  98         {Opt_nodebug, "nodebug"},
  99         {Opt_upgrade, "upgrade"},
 100         {Opt_acl, "acl"},
 101         {Opt_noacl, "noacl"},
 102         {Opt_quota_off, "quota=off"},
 103         {Opt_quota_account, "quota=account"},
 104         {Opt_quota_on, "quota=on"},
 105         {Opt_quota, "quota"},
 106         {Opt_noquota, "noquota"},
 107         {Opt_suiddir, "suiddir"},
 108         {Opt_nosuiddir, "nosuiddir"},
 109         {Opt_data_writeback, "data=writeback"},
 110         {Opt_data_ordered, "data=ordered"},
 111         {Opt_meta, "meta"},
 112         {Opt_discard, "discard"},
 113         {Opt_nodiscard, "nodiscard"},
 114         {Opt_commit, "commit=%d"},
 115         {Opt_err_withdraw, "errors=withdraw"},
 116         {Opt_err_panic, "errors=panic"},
 117         {Opt_statfs_quantum, "statfs_quantum=%d"},
 118         {Opt_statfs_percent, "statfs_percent=%d"},
 119         {Opt_quota_quantum, "quota_quantum=%d"},
 120         {Opt_barrier, "barrier"},
 121         {Opt_nobarrier, "nobarrier"},
 122         {Opt_rgrplvb, "rgrplvb"},
 123         {Opt_norgrplvb, "norgrplvb"},
 124         {Opt_error, NULL}
 125 };
 126
 127 /**
 128  * gfs2_mount_args - Parse mount options
 129  * @args: The structure into which the parsed options will be written
 130  * @options: The options to parse
 131  *
 132  * Return: errno
 133  */
 134
 135 int gfs2_mount_args(struct gfs2_args *args, char *options)
 136 {
 137         char *o;
 138         int token;
 139         substring_t tmp[MAX_OPT_ARGS];
 140         int rv;
 141
 142         /* Split the options into tokens with the "," character and
 143            process them */
 144
 145         while (1) {
 146                 o = strsep(&options, ",");
 147                 if (o == NULL)
 148                         break;
 149                 if (*o == '\0')
 150                         continue;
 151
 152                 token = match_token(o, tokens, tmp);
 153                 switch (token) {
 154                 case Opt_lockproto:
 155                         match_strlcpy(args->ar_lockproto, &tmp[0],
 156                                       GFS2_LOCKNAME_LEN);
 157                         break;
 158                 case Opt_locktable:
 159                         match_strlcpy(args->ar_locktable, &tmp[0],
 160                                       GFS2_LOCKNAME_LEN);
 161                         break;
 162                 case Opt_hostdata:
 163                         match_strlcpy(args->ar_hostdata, &tmp[0],
 164                                       GFS2_LOCKNAME_LEN);
 165                         break;
 166                 case Opt_spectator:
 167                         args->ar_spectator = 1;
 168                         break;
 169                 case Opt_ignore_local_fs:
 170                         /* Retained for backwards compat only */
 171                         break;
 172                 case Opt_localflocks:
 173                         args->ar_localflocks = 1;
 174                         break;
 175                 case Opt_localcaching:
 176                         /* Retained for backwards compat only */
 177                         break;
 178                 case Opt_debug:
 179                         if (args->ar_errors == GFS2_ERRORS_PANIC) {
 180                                 pr_warn("-o debug and -o errors=panic are mutually exclusive\n");
 181                                 return -EINVAL;
 182                         }
 183                         args->ar_debug = 1;
 184                         break;
 185                 case Opt_nodebug:
 186                         args->ar_debug = 0;
 187                         break;
 188                 case Opt_upgrade:
 189                         /* Retained for backwards compat only */
 190                         break;
 191                 case Opt_acl:
 192                         args->ar_posix_acl = 1;
 193                         break;
 194                 case Opt_noacl:
 195                         args->ar_posix_acl = 0;
 196                         break;
 197                 case Opt_quota_off:
 198                 case Opt_noquota:
 199                         args->ar_quota = GFS2_QUOTA_OFF;
 200                         break;
 201                 case Opt_quota_account:
 202                         args->ar_quota = GFS2_QUOTA_ACCOUNT;
 203                         break;
 204                 case Opt_quota_on:
 205                 case Opt_quota:
 206                         args->ar_quota = GFS2_QUOTA_ON;
 207                         break;
 208                 case Opt_suiddir:
 209                         args->ar_suiddir = 1;
 210                         break;
 211                 case Opt_nosuiddir:
 212                         args->ar_suiddir = 0;
 213                         break;
 214                 case Opt_data_writeback:
 215                         args->ar_data = GFS2_DATA_WRITEBACK;
 216                         break;
 217                 case Opt_data_ordered:
 218                         args->ar_data = GFS2_DATA_ORDERED;
 219                         break;
 220                 case Opt_meta:
 221                         args->ar_meta = 1;
 222                         break;
 223                 case Opt_discard:
 224                         args->ar_discard = 1;
 225                         break;
 226                 case Opt_nodiscard:
 227                         args->ar_discard = 0;
 228                         break;
 229                 case Opt_commit:
 230                         rv = match_int(&tmp[0], &args->ar_commit);
 231                         if (rv || args->ar_commit <= 0) {
 232                                 pr_warn("commit mount option requires a positive numeric argument\n");
 233                                 return rv ? rv : -EINVAL;
 234                         }
 235                         break;
 236                 case Opt_statfs_quantum:
 237                         rv = match_int(&tmp[0], &args->ar_statfs_quantum);
 238                         if (rv || args->ar_statfs_quantum < 0) {
 239                                 pr_warn("statfs_quantum mount option requires a non-negative numeric argument\n");
 240                                 return rv ? rv : -EINVAL;
 241                         }
 242                         break;
 243                 case Opt_quota_quantum:
 244                         rv = match_int(&tmp[0], &args->ar_quota_quantum);
 245                         if (rv || args->ar_quota_quantum <= 0) {
 246                                 pr_warn("quota_quantum mount option requires a positive numeric argument\n");
 247                                 return rv ? rv : -EINVAL;
 248                         }
 249                         break;
 250                 case Opt_statfs_percent:
 251                         rv = match_int(&tmp[0], &args->ar_statfs_percent);
 252                         if (rv || args->ar_statfs_percent < 0 ||
 253                             args->ar_statfs_percent > 100) {
 254                                 pr_warn("statfs_percent mount option requires a numeric argument between 0 and 100\n");
 255                                 return rv ? rv : -EINVAL;
 256                         }
 257                         break;
 258                 case Opt_err_withdraw:
 259                         args->ar_errors = GFS2_ERRORS_WITHDRAW;
 260                         break;
 261                 case Opt_err_panic:
 262                         if (args->ar_debug) {
 263                                 pr_warn("-o debug and -o errors=panic are mutually exclusive\n");
 264                                 return -EINVAL;
 265                         }
 266                         args->ar_errors = GFS2_ERRORS_PANIC;
 267                         break;
 268                 case Opt_barrier:
 269                         args->ar_nobarrier = 0;
 270                         break;
 271                 case Opt_nobarrier:
 272                         args->ar_nobarrier = 1;
 273                         break;
 274                 case Opt_rgrplvb:
 275                         args->ar_rgrplvb = 1;
 276                         break;
 277                 case Opt_norgrplvb:
 278                         args->ar_rgrplvb = 0;
 279                         break;
 280                 case Opt_error:
 281                 default:
 282                         pr_warn("invalid mount option: %s\n", o);
 283                         return -EINVAL;
 284                 }
 285         }
 286
 287         return 0;
 288 }
 289
 290 /**
 291  * gfs2_jindex_free - Clear all the journal index information
 292  * @sdp: The GFS2 superblock
 293  *
 294  */
 295
 296 void gfs2_jindex_free(struct gfs2_sbd *sdp)
 297 {
 298         struct list_head list;
 299         struct gfs2_jdesc *jd;
 300
 301         spin_lock(&sdp->sd_jindex_spin);
 302         list_add(&list, &sdp->sd_jindex_list);
 303         list_del_init(&sdp->sd_jindex_list);
 304         sdp->sd_journals = 0;
 305         spin_unlock(&sdp->sd_jindex_spin);
 306
 307         while (!list_empty(&list)) {
 308                 jd = list_entry(list.next, struct gfs2_jdesc, jd_list);
 309                 gfs2_free_journal_extents(jd);
 310                 list_del(&jd->jd_list);
 311                 iput(jd->jd_inode);
 312                 kfree(jd);
 313         }
 314 }
 315
 316 static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
 317 {
 318         struct gfs2_jdesc *jd;
 319         int found = 0;
 320
 321         list_for_each_entry(jd, head, jd_list) {
 322                 if (jd->jd_jid == jid) {
 323                         found = 1;
 324                         break;
 325                 }
 326         }
 327
 328         if (!found)
 329                 jd = NULL;
 330
 331         return jd;
 332 }
 333
 334 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
 335 {
 336         struct gfs2_jdesc *jd;
 337
 338         spin_lock(&sdp->sd_jindex_spin);
 339         jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
 340         spin_unlock(&sdp->sd_jindex_spin);
 341
 342         return jd;
 343 }
 344
 345 int gfs2_jdesc_check(struct gfs2_jdesc *jd)
 346 {
 347         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 348         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 349         u64 size = i_size_read(jd->jd_inode);
 350
 351         if (gfs2_check_internal_file_size(jd->jd_inode, 8 << 20, 1 << 30))
 352                 return -EIO;
 353
 354         jd->jd_blocks = size >> sdp->sd_sb.sb_bsize_shift;
 355
 356         if (gfs2_write_alloc_required(ip, 0, size)) {
 357                 gfs2_consist_inode(ip);
 358                 return -EIO;
 359         }
 360
 361         return 0;
 362 }
 363
 364 static int init_threads(struct gfs2_sbd *sdp)
 365 {
 366         struct task_struct *p;
 367         int error = 0;
 368
 369         p = kthread_run(gfs2_logd, sdp, "gfs2_logd");
 370         if (IS_ERR(p)) {
 371                 error = PTR_ERR(p);
 372                 fs_err(sdp, "can't start logd thread: %d\n", error);
 373                 return error;
 374         }
 375         sdp->sd_logd_process = p;
 376
 377         p = kthread_run(gfs2_quotad, sdp, "gfs2_quotad");
 378         if (IS_ERR(p)) {
 379                 error = PTR_ERR(p);
 380                 fs_err(sdp, "can't start quotad thread: %d\n", error);
 381                 goto fail;
 382         }
 383         sdp->sd_quotad_process = p;
 384         return 0;
 385
 386 fail:
 387         kthread_stop(sdp->sd_logd_process);
 388         return error;
 389 }
 390
 391 /**
 392  * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
 393  * @sdp: the filesystem
 394  *
 395  * Returns: errno
 396  */
 397
 398 int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
 399 {
 400         struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
 401         struct gfs2_glock *j_gl = ip->i_gl;
 402         struct gfs2_holder thaw_gh;
 403         struct gfs2_log_header_host head;
 404         int error;
 405
 406         error = init_threads(sdp);
 407         if (error)
 408                 return error;
 409
 410         error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED, 0,
 411                                    &thaw_gh);
 412         if (error)
 413                 goto fail_threads;
 414
 415         j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
 416
 417         error = gfs2_find_jhead(sdp->sd_jdesc, &head);
 418         if (error)
 419                 goto fail;
 420
 421         if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 422                 gfs2_consist(sdp);
 423                 error = -EIO;
 424                 goto fail;
 425         }
 426
 427         /*  Initialize some head of the log stuff  */
 428         sdp->sd_log_sequence = head.lh_sequence + 1;
 429         gfs2_log_pointers_init(sdp, head.lh_blkno);
 430
 431         error = gfs2_quota_init(sdp);
 432         if (error)
 433                 goto fail;
 434
 435         set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 436
 437         gfs2_glock_dq_uninit(&thaw_gh);
 438
 439         return 0;
 440
 441 fail:
 442         thaw_gh.gh_flags |= GL_NOCACHE;
 443         gfs2_glock_dq_uninit(&thaw_gh);
 444 fail_threads:
 445         kthread_stop(sdp->sd_quotad_process);
 446         kthread_stop(sdp->sd_logd_process);
 447         return error;
 448 }
 449
 450 void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
 451 {
 452         const struct gfs2_statfs_change *str = buf;
 453
 454         sc->sc_total = be64_to_cpu(str->sc_total);
 455         sc->sc_free = be64_to_cpu(str->sc_free);
 456         sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
 457 }
 458
 459 static void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
 460 {
 461         struct gfs2_statfs_change *str = buf;
 462
 463         str->sc_total = cpu_to_be64(sc->sc_total);
 464         str->sc_free = cpu_to_be64(sc->sc_free);
 465         str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
 466 }
 467
 468 int gfs2_statfs_init(struct gfs2_sbd *sdp)
 469 {
 470         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 471         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 472         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 473         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 474         struct buffer_head *m_bh, *l_bh;
 475         struct gfs2_holder gh;
 476         int error;
 477
 478         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 479                                    &gh);
 480         if (error)
 481                 return error;
 482
 483         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 484         if (error)
 485                 goto out;
 486
 487         if (sdp->sd_args.ar_spectator) {
 488                 spin_lock(&sdp->sd_statfs_spin);
 489                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 490                                       sizeof(struct gfs2_dinode));
 491                 spin_unlock(&sdp->sd_statfs_spin);
 492         } else {
 493                 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 494                 if (error)
 495                         goto out_m_bh;
 496
 497                 spin_lock(&sdp->sd_statfs_spin);
 498                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 499                                       sizeof(struct gfs2_dinode));
 500                 gfs2_statfs_change_in(l_sc, l_bh->b_data +
 501                                       sizeof(struct gfs2_dinode));
 502                 spin_unlock(&sdp->sd_statfs_spin);
 503
 504                 brelse(l_bh);
 505         }
 506
 507 out_m_bh:
 508         brelse(m_bh);
 509 out:
 510         gfs2_glock_dq_uninit(&gh);
 511         return 0;
 512 }
 513
 514 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
 515                         s64 dinodes)
 516 {
 517         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 518         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 519         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 520         struct buffer_head *l_bh;
 521         s64 x, y;
 522         int need_sync = 0;
 523         int error;
 524
 525         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 526         if (error)
 527                 return;
 528
 529         gfs2_trans_add_meta(l_ip->i_gl, l_bh);
 530
 531         spin_lock(&sdp->sd_statfs_spin);
 532         l_sc->sc_total += total;
 533         l_sc->sc_free += free;
 534         l_sc->sc_dinodes += dinodes;
 535         gfs2_statfs_change_out(l_sc, l_bh->b_data + sizeof(struct gfs2_dinode));
 536         if (sdp->sd_args.ar_statfs_percent) {
 537                 x = 100 * l_sc->sc_free;
 538                 y = m_sc->sc_free * sdp->sd_args.ar_statfs_percent;
 539                 if (x >= y || x <= -y)
 540                         need_sync = 1;
 541         }
 542         spin_unlock(&sdp->sd_statfs_spin);
 543
 544         brelse(l_bh);
 545         if (need_sync)
 546                 gfs2_wake_up_statfs(sdp);
 547 }
 548
 549 void update_statfs(struct gfs2_sbd *sdp, struct buffer_head *m_bh,
 550                    struct buffer_head *l_bh)
 551 {
 552         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 553         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 554         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 555         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 556
 557         gfs2_trans_add_meta(l_ip->i_gl, l_bh);
 558
 559         spin_lock(&sdp->sd_statfs_spin);
 560         m_sc->sc_total += l_sc->sc_total;
 561         m_sc->sc_free += l_sc->sc_free;
 562         m_sc->sc_dinodes += l_sc->sc_dinodes;
 563         memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
 564         memset(l_bh->b_data + sizeof(struct gfs2_dinode),
 565                0, sizeof(struct gfs2_statfs_change));
 566         spin_unlock(&sdp->sd_statfs_spin);
 567
 568         gfs2_trans_add_meta(m_ip->i_gl, m_bh);
 569         gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
 570 }
 571
 572 int gfs2_statfs_sync(struct super_block *sb, int type)
 573 {
 574         struct gfs2_sbd *sdp = sb->s_fs_info;
 575         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 576         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 577         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 578         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 579         struct gfs2_holder gh;
 580         struct buffer_head *m_bh, *l_bh;
 581         int error;
 582
 583         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 584                                    &gh);
 585         if (error)
 586                 return error;
 587
 588         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 589         if (error)
 590                 goto out;
 591
 592         spin_lock(&sdp->sd_statfs_spin);
 593         gfs2_statfs_change_in(m_sc, m_bh->b_data +
 594                               sizeof(struct gfs2_dinode));
 595         if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
 596                 spin_unlock(&sdp->sd_statfs_spin);
 597                 goto out_bh;
 598         }
 599         spin_unlock(&sdp->sd_statfs_spin);
 600
 601         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 602         if (error)
 603                 goto out_bh;
 604
 605         error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
 606         if (error)
 607                 goto out_bh2;
 608
 609         update_statfs(sdp, m_bh, l_bh);
 610         sdp->sd_statfs_force_sync = 0;
 611
 612         gfs2_trans_end(sdp);
 613
 614 out_bh2:
 615         brelse(l_bh);
 616 out_bh:
 617         brelse(m_bh);
 618 out:
 619         gfs2_glock_dq_uninit(&gh);
 620         return error;
 621 }
 622
 623 struct lfcc {
 624         struct list_head list;
 625         struct gfs2_holder gh;
 626 };
 627
 628 /**
 629  * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
 630  *                            journals are clean
 631  * @sdp: the file system
 632  * @state: the state to put the transaction lock into
 633  * @t_gh: the hold on the transaction lock
 634  *
 635  * Returns: errno
 636  */
 637
 638 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp,
 639                                     struct gfs2_holder *freeze_gh)
 640 {
 641         struct gfs2_inode *ip;
 642         struct gfs2_jdesc *jd;
 643         struct lfcc *lfcc;
 644         LIST_HEAD(list);
 645         struct gfs2_log_header_host lh;
 646         struct gfs2_inode *dip = GFS2_I(sdp->sd_root_dir->d_inode);
 647         int error;
 648
 649         error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0,
 650                                    &sdp->sd_freeze_root_gh);
 651         if (error)
 652                 return error;
 653         atomic_set(&sdp->sd_frozen_root, 1);
 654         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 655                 lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
 656                 if (!lfcc) {
 657                         error = -ENOMEM;
 658                         goto out;
 659                 }
 660                 ip = GFS2_I(jd->jd_inode);
 661                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
 662                 if (error) {
 663                         kfree(lfcc);
 664                         goto out;
 665                 }
 666                 list_add(&lfcc->list, &list);
 667         }
 668
 669         error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_EXCLUSIVE,
 670                                    GL_NOCACHE, freeze_gh);
 671
 672         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 673                 error = gfs2_jdesc_check(jd);
 674                 if (error)
 675                         break;
 676                 error = gfs2_find_jhead(jd, &lh);
 677                 if (error)
 678                         break;
 679                 if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 680                         error = -EBUSY;
 681                         break;
 682                 }
 683         }
 684
 685         if (error)
 686                 gfs2_glock_dq_uninit(freeze_gh);
 687
 688 out:
 689         while (!list_empty(&list)) {
 690                 lfcc = list_entry(list.next, struct lfcc, list);
 691                 list_del(&lfcc->list);
 692                 gfs2_glock_dq_uninit(&lfcc->gh);
 693                 kfree(lfcc);
 694         }
 695         if (error) {
 696                 atomic_dec(&sdp->sd_frozen_root);
 697                 wait_event(sdp->sd_frozen_root_wait, atomic_read(&sdp->sd_frozen_root) == 0);
 698                 gfs2_glock_dq_uninit(&sdp->sd_freeze_root_gh);
 699         }
 700         return error;
 701 }
 702
 703 void gfs2_dinode_out(const struct gfs2_inode *ip, void *buf)
 704 {
 705         struct gfs2_dinode *str = buf;
 706
 707         str->di_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 708         str->di_header.mh_type = cpu_to_be32(GFS2_METATYPE_DI);
 709         str->di_header.mh_format = cpu_to_be32(GFS2_FORMAT_DI);
 710         str->di_num.no_addr = cpu_to_be64(ip->i_no_addr);
 711         str->di_num.no_formal_ino = cpu_to_be64(ip->i_no_formal_ino);
 712         str->di_mode = cpu_to_be32(ip->i_inode.i_mode);
 713         str->di_uid = cpu_to_be32(i_uid_read(&ip->i_inode));
 714         str->di_gid = cpu_to_be32(i_gid_read(&ip->i_inode));
 715         str->di_nlink = cpu_to_be32(ip->i_inode.i_nlink);
 716         str->di_size = cpu_to_be64(i_size_read(&ip->i_inode));
 717         str->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(&ip->i_inode));
 718         str->di_atime = cpu_to_be64(ip->i_inode.i_atime.tv_sec);
 719         str->di_mtime = cpu_to_be64(ip->i_inode.i_mtime.tv_sec);
 720         str->di_ctime = cpu_to_be64(ip->i_inode.i_ctime.tv_sec);
 721
 722         str->di_goal_meta = cpu_to_be64(ip->i_goal);
 723         str->di_goal_data = cpu_to_be64(ip->i_goal);
 724         str->di_generation = cpu_to_be64(ip->i_generation);
 725
 726         str->di_flags = cpu_to_be32(ip->i_diskflags);
 727         str->di_height = cpu_to_be16(ip->i_height);
 728         str->di_payload_format = cpu_to_be32(S_ISDIR(ip->i_inode.i_mode) &&
 729                                              !(ip->i_diskflags & GFS2_DIF_EXHASH) ?
 730                                              GFS2_FORMAT_DE : 0);
 731         str->di_depth = cpu_to_be16(ip->i_depth);
 732         str->di_entries = cpu_to_be32(ip->i_entries);
 733
 734         str->di_eattr = cpu_to_be64(ip->i_eattr);
 735         str->di_atime_nsec = cpu_to_be32(ip->i_inode.i_atime.tv_nsec);
 736         str->di_mtime_nsec = cpu_to_be32(ip->i_inode.i_mtime.tv_nsec);
 737         str->di_ctime_nsec = cpu_to_be32(ip->i_inode.i_ctime.tv_nsec);
 738 }
 739
 740 /**
 741  * gfs2_write_inode - Make sure the inode is stable on the disk
 742  * @inode: The inode
 743  * @wbc: The writeback control structure
 744  *
 745  * Returns: errno
 746  */
 747
 748 static int gfs2_write_inode(struct inode *inode, struct writeback_control *wbc)
 749 {
 750         struct gfs2_inode *ip = GFS2_I(inode);
 751         struct gfs2_sbd *sdp = GFS2_SB(inode);
 752         struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl);
 753         struct backing_dev_info *bdi = metamapping->backing_dev_info;
 754         int ret = 0;
 755
 756         if (wbc->sync_mode == WB_SYNC_ALL)
 757                 gfs2_log_flush(GFS2_SB(inode), ip->i_gl, NORMAL_FLUSH);
 758         if (bdi->dirty_exceeded)
 759                 gfs2_ail1_flush(sdp, wbc);
 760         else
 761                 filemap_fdatawrite(metamapping);
 762         if (wbc->sync_mode == WB_SYNC_ALL)
 763                 ret = filemap_fdatawait(metamapping);
 764         if (ret)
 765                 mark_inode_dirty_sync(inode);
 766         return ret;
 767 }
 768
 769 /**
 770  * gfs2_dirty_inode - check for atime updates
 771  * @inode: The inode in question
 772  * @flags: The type of dirty
 773  *
 774  * Unfortunately it can be called under any combination of inode
 775  * glock and transaction lock, so we have to check carefully.
 776  *
 777  * At the moment this deals only with atime - it should be possible
 778  * to expand that role in future, once a review of the locking has
 779  * been carried out.
 780  */
 781
 782 static void gfs2_dirty_inode(struct inode *inode, int flags)
 783 {
 784         struct gfs2_inode *ip = GFS2_I(inode);
 785         struct gfs2_sbd *sdp = GFS2_SB(inode);
 786         struct buffer_head *bh;
 787         struct gfs2_holder gh;
 788         int need_unlock = 0;
 789         int need_endtrans = 0;
 790         int ret;
 791
 792         if (!(flags & (I_DIRTY_DATASYNC|I_DIRTY_SYNC)))
 793                 return;
 794
 795         if (!gfs2_glock_is_locked_by_me(ip->i_gl)) {
 796                 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
 797                 if (ret) {
 798                         fs_err(sdp, "dirty_inode: glock %d\n", ret);
 799                         return;
 800                 }
 801                 need_unlock = 1;
 802         } else if (WARN_ON_ONCE(ip->i_gl->gl_state != LM_ST_EXCLUSIVE))
 803                 return;
 804
 805         if (current->journal_info == NULL) {
 806                 ret = gfs2_trans_begin(sdp, RES_DINODE, 0);
 807                 if (ret) {
 808                         fs_err(sdp, "dirty_inode: gfs2_trans_begin %d\n", ret);
 809                         goto out;
 810                 }
 811                 need_endtrans = 1;
 812         }
 813
 814         ret = gfs2_meta_inode_buffer(ip, &bh);
 815         if (ret == 0) {
 816                 gfs2_trans_add_meta(ip->i_gl, bh);
 817                 gfs2_dinode_out(ip, bh->b_data);
 818                 brelse(bh);
 819         }
 820
 821         if (need_endtrans)
 822                 gfs2_trans_end(sdp);
 823 out:
 824         if (need_unlock)
 825                 gfs2_glock_dq_uninit(&gh);
 826 }
 827
 828 /**
 829  * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
 830  * @sdp: the filesystem
 831  *
 832  * Returns: errno
 833  */
 834
 835 static int gfs2_make_fs_ro(struct gfs2_sbd *sdp)
 836 {
 837         struct gfs2_holder thaw_gh;
 838         int error;
 839
 840         error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED, GL_NOCACHE,
 841                                    &thaw_gh);
 842         if (error && !test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
 843                 return error;
 844
 845         down_write(&sdp->sd_log_flush_lock);
 846         clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 847         up_write(&sdp->sd_log_flush_lock);
 848
 849         kthread_stop(sdp->sd_quotad_process);
 850         kthread_stop(sdp->sd_logd_process);
 851
 852         flush_workqueue(gfs2_delete_workqueue);
 853         gfs2_quota_sync(sdp->sd_vfs, 0);
 854         gfs2_statfs_sync(sdp->sd_vfs, 0);
 855
 856         gfs2_log_flush(sdp, NULL, SHUTDOWN_FLUSH);
 857         gfs2_assert_warn(sdp, atomic_read(&sdp->sd_log_blks_free) == sdp->sd_jdesc->jd_blocks);
 858
 859         if (thaw_gh.gh_gl)
 860                 gfs2_glock_dq_uninit(&thaw_gh);
 861
 862         gfs2_quota_cleanup(sdp);
 863
 864         return error;
 865 }
 866
 867 static int gfs2_umount_recovery_wait(void *word)
 868 {
 869         schedule();
 870         return 0;
 871 }
 872
 873 /**
 874  * gfs2_put_super - Unmount the filesystem
 875  * @sb: The VFS superblock
 876  *
 877  */
 878
 879 static void gfs2_put_super(struct super_block *sb)
 880 {
 881         struct gfs2_sbd *sdp = sb->s_fs_info;
 882         int error;
 883         struct gfs2_jdesc *jd;
 884
 885         /* No more recovery requests */
 886         set_bit(SDF_NORECOVERY, &sdp->sd_flags);
 887         smp_mb();
 888
 889         /* Wait on outstanding recovery */
 890 restart:
 891         spin_lock(&sdp->sd_jindex_spin);
 892         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 893                 if (!test_bit(JDF_RECOVERY, &jd->jd_flags))
 894                         continue;
 895                 spin_unlock(&sdp->sd_jindex_spin);
 896                 wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
 897                             gfs2_umount_recovery_wait, TASK_UNINTERRUPTIBLE);
 898                 goto restart;
 899         }
 900         spin_unlock(&sdp->sd_jindex_spin);
 901
 902         if (!(sb->s_flags & MS_RDONLY)) {
 903                 error = gfs2_make_fs_ro(sdp);
 904                 if (error)
 905                         gfs2_io_error(sdp);
 906         }
 907         /*  At this point, we're through modifying the disk  */
 908
 909         /*  Release stuff  */
 910
 911         iput(sdp->sd_jindex);
 912         iput(sdp->sd_statfs_inode);
 913         iput(sdp->sd_rindex);
 914         iput(sdp->sd_quota_inode);
 915
 916         gfs2_glock_put(sdp->sd_rename_gl);
 917         gfs2_glock_put(sdp->sd_freeze_gl);
 918
 919         if (!sdp->sd_args.ar_spectator) {
 920                 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
 921                 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
 922                 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
 923                 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
 924                 iput(sdp->sd_sc_inode);
 925                 iput(sdp->sd_qc_inode);
 926         }
 927
 928         gfs2_glock_dq_uninit(&sdp->sd_live_gh);
 929         gfs2_clear_rgrpd(sdp);
 930         gfs2_jindex_free(sdp);
 931         /*  Take apart glock structures and buffer lists  */
 932         gfs2_gl_hash_clear(sdp);
 933         /*  Unmount the locking protocol  */
 934         gfs2_lm_unmount(sdp);
 935
 936         /*  At this point, we're through participating in the lockspace  */
 937         gfs2_sys_fs_del(sdp);
 938 }
 939
 940 /**
 941  * gfs2_sync_fs - sync the filesystem
 942  * @sb: the superblock
 943  *
 944  * Flushes the log to disk.
 945  */
 946
 947 static int gfs2_sync_fs(struct super_block *sb, int wait)
 948 {
 949         struct gfs2_sbd *sdp = sb->s_fs_info;
 950
 951         gfs2_quota_sync(sb, -1);
 952         if (wait && sdp && !atomic_read(&sdp->sd_log_freeze))
 953                 gfs2_log_flush(sdp, NULL, NORMAL_FLUSH);
 954         return 0;
 955 }
 956
 957 /**
 958  * gfs2_freeze - prevent further writes to the filesystem
 959  * @sb: the VFS structure for the filesystem
 960  *
 961  */
 962
 963 static int gfs2_freeze(struct super_block *sb)
 964 {
 965         struct gfs2_sbd *sdp = sb->s_fs_info;
 966         int error;
 967
 968         if (test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
 969                 return -EINVAL;
 970
 971         for (;;) {
 972                 error = gfs2_lock_fs_check_clean(sdp, &sdp->sd_freeze_gh);
 973                 if (!error)
 974                         break;
 975
 976                 switch (error) {
 977                 case -EBUSY:
 978                         fs_err(sdp, "waiting for recovery before freeze\n");
 979                         break;
 980
 981                 default:
 982                         fs_err(sdp, "error freezing FS: %d\n", error);
 983                         break;
 984                 }
 985
 986                 fs_err(sdp, "retrying...\n");
 987                 msleep(1000);
 988         }
 989         return 0;
 990 }
 991
 992 /**
 993  * gfs2_unfreeze - reallow writes to the filesystem
 994  * @sb: the VFS structure for the filesystem
 995  *
 996  */
 997
 998 static int gfs2_unfreeze(struct super_block *sb)
 999 {
1000         struct gfs2_sbd *sdp = sb->s_fs_info;
1001
1002         gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
1003         atomic_dec(&sdp->sd_frozen_root);
1004         wait_event(sdp->sd_frozen_root_wait, atomic_read(&sdp->sd_frozen_root) == 0);
1005         gfs2_glock_dq_uninit(&sdp->sd_freeze_root_gh);
1006         return 0;
1007 }
1008
1009 /**
1010  * statfs_fill - fill in the sg for a given RG
1011  * @rgd: the RG
1012  * @sc: the sc structure
1013  *
1014  * Returns: 0 on success, -ESTALE if the LVB is invalid
1015  */
1016
1017 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
1018                             struct gfs2_statfs_change_host *sc)
1019 {
1020         gfs2_rgrp_verify(rgd);
1021         sc->sc_total += rgd->rd_data;
1022         sc->sc_free += rgd->rd_free;
1023         sc->sc_dinodes += rgd->rd_dinodes;
1024         return 0;
1025 }
1026
1027 /**
1028  * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
1029  * @sdp: the filesystem
1030  * @sc: the sc info that will be returned
1031  *
1032  * Any error (other than a signal) will cause this routine to fall back
1033  * to the synchronous version.
1034  *
1035  * FIXME: This really shouldn't busy wait like this.
1036  *
1037  * Returns: errno
1038  */
1039
1040 static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
1041 {
1042         struct gfs2_rgrpd *rgd_next;
1043         struct gfs2_holder *gha, *gh;
1044         unsigned int slots = 64;
1045         unsigned int x;
1046         int done;
1047         int error = 0, err;
1048
1049         memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
1050         gha = kcalloc(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
1051         if (!gha)
1052                 return -ENOMEM;
1053
1054         rgd_next = gfs2_rgrpd_get_first(sdp);
1055
1056         for (;;) {
1057                 done = 1;
1058
1059                 for (x = 0; x < slots; x++) {
1060                         gh = gha + x;
1061
1062                         if (gh->gh_gl && gfs2_glock_poll(gh)) {
1063                                 err = gfs2_glock_wait(gh);
1064                                 if (err) {
1065                                         gfs2_holder_uninit(gh);
1066                                         error = err;
1067                                 } else {
1068                                         if (!error)
1069                                                 error = statfs_slow_fill(
1070                                                         gh->gh_gl->gl_object, sc);
1071                                         gfs2_glock_dq_uninit(gh);
1072                                 }
1073                         }
1074
1075                         if (gh->gh_gl)
1076                                 done = 0;
1077                         else if (rgd_next && !error) {
1078                                 error = gfs2_glock_nq_init(rgd_next->rd_gl,
1079                                                            LM_ST_SHARED,
1080                                                            GL_ASYNC,
1081                                                            gh);
1082                                 rgd_next = gfs2_rgrpd_get_next(rgd_next);
1083                                 done = 0;
1084                         }
1085
1086                         if (signal_pending(current))
1087                                 error = -ERESTARTSYS;
1088                 }
1089
1090                 if (done)
1091                         break;
1092
1093                 yield();
1094         }
1095
1096         kfree(gha);
1097         return error;
1098 }
1099
1100 /**
1101  * gfs2_statfs_i - Do a statfs
1102  * @sdp: the filesystem
1103  * @sg: the sg structure
1104  *
1105  * Returns: errno
1106  */
1107
1108 static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
1109 {
1110         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
1111         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
1112
1113         spin_lock(&sdp->sd_statfs_spin);
1114
1115         *sc = *m_sc;
1116         sc->sc_total += l_sc->sc_total;
1117         sc->sc_free += l_sc->sc_free;
1118         sc->sc_dinodes += l_sc->sc_dinodes;
1119
1120         spin_unlock(&sdp->sd_statfs_spin);
1121
1122         if (sc->sc_free < 0)
1123                 sc->sc_free = 0;
1124         if (sc->sc_free > sc->sc_total)
1125                 sc->sc_free = sc->sc_total;
1126         if (sc->sc_dinodes < 0)
1127                 sc->sc_dinodes = 0;
1128
1129         return 0;
1130 }
1131
1132 /**
1133  * gfs2_statfs - Gather and return stats about the filesystem
1134  * @sb: The superblock
1135  * @statfsbuf: The buffer
1136  *
1137  * Returns: 0 on success or error code
1138  */
1139
1140 static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
1141 {
1142         struct super_block *sb = dentry->d_inode->i_sb;
1143         struct gfs2_sbd *sdp = sb->s_fs_info;
1144         struct gfs2_statfs_change_host sc;
1145         int error;
1146
1147         error = gfs2_rindex_update(sdp);
1148         if (error)
1149                 return error;
1150
1151         if (gfs2_tune_get(sdp, gt_statfs_slow))
1152                 error = gfs2_statfs_slow(sdp, &sc);
1153         else
1154                 error = gfs2_statfs_i(sdp, &sc);
1155
1156         if (error)
1157                 return error;
1158
1159         buf->f_type = GFS2_MAGIC;
1160         buf->f_bsize = sdp->sd_sb.sb_bsize;
1161         buf->f_blocks = sc.sc_total;
1162         buf->f_bfree = sc.sc_free;
1163         buf->f_bavail = sc.sc_free;
1164         buf->f_files = sc.sc_dinodes + sc.sc_free;
1165         buf->f_ffree = sc.sc_free;
1166         buf->f_namelen = GFS2_FNAMESIZE;
1167
1168         return 0;
1169 }
1170
1171 /**
1172  * gfs2_remount_fs - called when the FS is remounted
1173  * @sb:  the filesystem
1174  * @flags:  the remount flags
1175  * @data:  extra data passed in (not used right now)
1176  *
1177  * Returns: errno
1178  */
1179
1180 static int gfs2_remount_fs(struct super_block *sb, int *flags, char *data)
1181 {
1182         struct gfs2_sbd *sdp = sb->s_fs_info;
1183         struct gfs2_args args = sdp->sd_args; /* Default to current settings */
1184         struct gfs2_tune *gt = &sdp->sd_tune;
1185         int error;
1186
1187         sync_filesystem(sb);
1188
1189         spin_lock(&gt->gt_spin);
1190         args.ar_commit = gt->gt_logd_secs;
1191         args.ar_quota_quantum = gt->gt_quota_quantum;
1192         if (gt->gt_statfs_slow)
1193                 args.ar_statfs_quantum = 0;
1194         else
1195                 args.ar_statfs_quantum = gt->gt_statfs_quantum;
1196         spin_unlock(&gt->gt_spin);
1197         error = gfs2_mount_args(&args, data);
1198         if (error)
1199                 return error;
1200
1201         /* Not allowed to change locking details */
1202         if (strcmp(args.ar_lockproto, sdp->sd_args.ar_lockproto) ||
1203             strcmp(args.ar_locktable, sdp->sd_args.ar_locktable) ||
1204             strcmp(args.ar_hostdata, sdp->sd_args.ar_hostdata))
1205                 return -EINVAL;
1206
1207         /* Some flags must not be changed */
1208         if (args_neq(&args, &sdp->sd_args, spectator) ||
1209             args_neq(&args, &sdp->sd_args, localflocks) ||
1210             args_neq(&args, &sdp->sd_args, meta))
1211                 return -EINVAL;
1212
1213         if (sdp->sd_args.ar_spectator)
1214                 *flags |= MS_RDONLY;
1215
1216         if ((sb->s_flags ^ *flags) & MS_RDONLY) {
1217                 if (*flags & MS_RDONLY)
1218                         error = gfs2_make_fs_ro(sdp);
1219                 else
1220                         error = gfs2_make_fs_rw(sdp);
1221                 if (error)
1222                         return error;
1223         }
1224
1225         sdp->sd_args = args;
1226         if (sdp->sd_args.ar_posix_acl)
1227                 sb->s_flags |= MS_POSIXACL;
1228         else
1229                 sb->s_flags &= ~MS_POSIXACL;
1230         if (sdp->sd_args.ar_nobarrier)
1231                 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1232         else
1233                 clear_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1234         spin_lock(&gt->gt_spin);
1235         gt->gt_logd_secs = args.ar_commit;
1236         gt->gt_quota_quantum = args.ar_quota_quantum;
1237         if (args.ar_statfs_quantum) {
1238                 gt->gt_statfs_slow = 0;
1239                 gt->gt_statfs_quantum = args.ar_statfs_quantum;
1240         }
1241         else {
1242                 gt->gt_statfs_slow = 1;
1243                 gt->gt_statfs_quantum = 30;
1244         }
1245         spin_unlock(&gt->gt_spin);
1246
1247         gfs2_online_uevent(sdp);
1248         return 0;
1249 }
1250
1251 /**
1252  * gfs2_drop_inode - Drop an inode (test for remote unlink)
1253  * @inode: The inode to drop
1254  *
1255  * If we've received a callback on an iopen lock then its because a
1256  * remote node tried to deallocate the inode but failed due to this node
1257  * still having the inode open. Here we mark the link count zero
1258  * since we know that it must have reached zero if the GLF_DEMOTE flag
1259  * is set on the iopen glock. If we didn't do a disk read since the
1260  * remote node removed the final link then we might otherwise miss
1261  * this event. This check ensures that this node will deallocate the
1262  * inode's blocks, or alternatively pass the baton on to another
1263  * node for later deallocation.
1264  */
1265
1266 static int gfs2_drop_inode(struct inode *inode)
1267 {
1268         struct gfs2_inode *ip = GFS2_I(inode);
1269
1270         if (!test_bit(GIF_FREE_VFS_INODE, &ip->i_flags) && inode->i_nlink) {
1271                 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1272                 if (gl && test_bit(GLF_DEMOTE, &gl->gl_flags))
1273                         clear_nlink(inode);
1274         }
1275         return generic_drop_inode(inode);
1276 }
1277
1278 static int is_ancestor(const struct dentry *d1, const struct dentry *d2)
1279 {
1280         do {
1281                 if (d1 == d2)
1282                         return 1;
1283                 d1 = d1->d_parent;
1284         } while (!IS_ROOT(d1));
1285         return 0;
1286 }
1287
1288 /**
1289  * gfs2_show_options - Show mount options for /proc/mounts
1290  * @s: seq_file structure
1291  * @root: root of this (sub)tree
1292  *
1293  * Returns: 0 on success or error code
1294  */
1295
1296 static int gfs2_show_options(struct seq_file *s, struct dentry *root)
1297 {
1298         struct gfs2_sbd *sdp = root->d_sb->s_fs_info;
1299         struct gfs2_args *args = &sdp->sd_args;
1300         int val;
1301
1302         if (is_ancestor(root, sdp->sd_master_dir))
1303                 seq_printf(s, ",meta");
1304         if (args->ar_lockproto[0])
1305                 seq_printf(s, ",lockproto=%s", args->ar_lockproto);
1306         if (args->ar_locktable[0])
1307                 seq_printf(s, ",locktable=%s", args->ar_locktable);
1308         if (args->ar_hostdata[0])
1309                 seq_printf(s, ",hostdata=%s", args->ar_hostdata);
1310         if (args->ar_spectator)
1311                 seq_printf(s, ",spectator");
1312         if (args->ar_localflocks)
1313                 seq_printf(s, ",localflocks");
1314         if (args->ar_debug)
1315                 seq_printf(s, ",debug");
1316         if (args->ar_posix_acl)
1317                 seq_printf(s, ",acl");
1318         if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
1319                 char *state;
1320                 switch (args->ar_quota) {
1321                 case GFS2_QUOTA_OFF:
1322                         state = "off";
1323                         break;
1324                 case GFS2_QUOTA_ACCOUNT:
1325                         state = "account";
1326                         break;
1327                 case GFS2_QUOTA_ON:
1328                         state = "on";
1329                         break;
1330                 default:
1331                         state = "unknown";
1332                         break;
1333                 }
1334                 seq_printf(s, ",quota=%s", state);
1335         }
1336         if (args->ar_suiddir)
1337                 seq_printf(s, ",suiddir");
1338         if (args->ar_data != GFS2_DATA_DEFAULT) {
1339                 char *state;
1340                 switch (args->ar_data) {
1341                 case GFS2_DATA_WRITEBACK:
1342                         state = "writeback";
1343                         break;
1344                 case GFS2_DATA_ORDERED:
1345                         state = "ordered";
1346                         break;
1347                 default:
1348                         state = "unknown";
1349                         break;
1350                 }
1351                 seq_printf(s, ",data=%s", state);
1352         }
1353         if (args->ar_discard)
1354                 seq_printf(s, ",discard");
1355         val = sdp->sd_tune.gt_logd_secs;
1356         if (val != 30)
1357                 seq_printf(s, ",commit=%d", val);
1358         val = sdp->sd_tune.gt_statfs_quantum;
1359         if (val != 30)
1360                 seq_printf(s, ",statfs_quantum=%d", val);
1361         else if (sdp->sd_tune.gt_statfs_slow)
1362                 seq_puts(s, ",statfs_quantum=0");
1363         val = sdp->sd_tune.gt_quota_quantum;
1364         if (val != 60)
1365                 seq_printf(s, ",quota_quantum=%d", val);
1366         if (args->ar_statfs_percent)
1367                 seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent);
1368         if (args->ar_errors != GFS2_ERRORS_DEFAULT) {
1369                 const char *state;
1370
1371                 switch (args->ar_errors) {
1372                 case GFS2_ERRORS_WITHDRAW:
1373                         state = "withdraw";
1374                         break;
1375                 case GFS2_ERRORS_PANIC:
1376                         state = "panic";
1377                         break;
1378                 default:
1379                         state = "unknown";
1380                         break;
1381                 }
1382                 seq_printf(s, ",errors=%s", state);
1383         }
1384         if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags))
1385                 seq_printf(s, ",nobarrier");
1386         if (test_bit(SDF_DEMOTE, &sdp->sd_flags))
1387                 seq_printf(s, ",demote_interface_used");
1388         if (args->ar_rgrplvb)
1389                 seq_printf(s, ",rgrplvb");
1390         return 0;
1391 }
1392
1393 static void gfs2_final_release_pages(struct gfs2_inode *ip)
1394 {
1395         struct inode *inode = &ip->i_inode;
1396         struct gfs2_glock *gl = ip->i_gl;
1397
1398         truncate_inode_pages(gfs2_glock2aspace(ip->i_gl), 0);
1399         truncate_inode_pages(&inode->i_data, 0);
1400
1401         if (atomic_read(&gl->gl_revokes) == 0) {
1402                 clear_bit(GLF_LFLUSH, &gl->gl_flags);
1403                 clear_bit(GLF_DIRTY, &gl->gl_flags);
1404         }
1405 }
1406
1407 static int gfs2_dinode_dealloc(struct gfs2_inode *ip)
1408 {
1409         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1410         struct gfs2_rgrpd *rgd;
1411         struct gfs2_holder gh;
1412         int error;
1413
1414         if (gfs2_get_inode_blocks(&ip->i_inode) != 1) {
1415                 gfs2_consist_inode(ip);
1416                 return -EIO;
1417         }
1418
1419         error = gfs2_rindex_update(sdp);
1420         if (error)
1421                 return error;
1422
1423         error = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
1424         if (error)
1425                 return error;
1426
1427         rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1428         if (!rgd) {
1429                 gfs2_consist_inode(ip);
1430                 error = -EIO;
1431                 goto out_qs;
1432         }
1433
1434         error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &gh);
1435         if (error)
1436                 goto out_qs;
1437
1438         error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA,
1439                                  sdp->sd_jdesc->jd_blocks);
1440         if (error)
1441                 goto out_rg_gunlock;
1442
1443         gfs2_free_di(rgd, ip);
1444
1445         gfs2_final_release_pages(ip);
1446
1447         gfs2_trans_end(sdp);
1448
1449 out_rg_gunlock:
1450         gfs2_glock_dq_uninit(&gh);
1451 out_qs:
1452         gfs2_quota_unhold(ip);
1453         return error;
1454 }
1455
1456 /**
1457  * gfs2_evict_inode - Remove an inode from cache
1458  * @inode: The inode to evict
1459  *
1460  * There are three cases to consider:
1461  * 1. i_nlink == 0, we are final opener (and must deallocate)
1462  * 2. i_nlink == 0, we are not the final opener (and cannot deallocate)
1463  * 3. i_nlink > 0
1464  *
1465  * If the fs is read only, then we have to treat all cases as per #3
1466  * since we are unable to do any deallocation. The inode will be
1467  * deallocated by the next read/write node to attempt an allocation
1468  * in the same resource group
1469  *
1470  * We have to (at the moment) hold the inodes main lock to cover
1471  * the gap between unlocking the shared lock on the iopen lock and
1472  * taking the exclusive lock. I'd rather do a shared -> exclusive
1473  * conversion on the iopen lock, but we can change that later. This
1474  * is safe, just less efficient.
1475  */
1476
1477 static void gfs2_evict_inode(struct inode *inode)
1478 {
1479         struct super_block *sb = inode->i_sb;
1480         struct gfs2_sbd *sdp = sb->s_fs_info;
1481         struct gfs2_inode *ip = GFS2_I(inode);
1482         struct gfs2_holder gh;
1483         int error;
1484
1485         if (test_bit(GIF_FREE_VFS_INODE, &ip->i_flags)) {
1486                 clear_inode(inode);
1487                 return;
1488         }
1489
1490         if (inode->i_nlink || (sb->s_flags & MS_RDONLY))
1491                 goto out;
1492
1493         /* Must not read inode block until block type has been verified */
1494         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, &gh);
1495         if (unlikely(error)) {
1496                 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1497                 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1498                 goto out;
1499         }
1500
1501         if (!test_bit(GIF_ALLOC_FAILED, &ip->i_flags)) {
1502                 error = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED);
1503                 if (error)
1504                         goto out_truncate;
1505         }
1506
1507         if (test_bit(GIF_INVALID, &ip->i_flags)) {
1508                 error = gfs2_inode_refresh(ip);
1509                 if (error)
1510                         goto out_truncate;
1511         }
1512
1513         ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1514         gfs2_glock_dq_wait(&ip->i_iopen_gh);
1515         gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE, &ip->i_iopen_gh);
1516         error = gfs2_glock_nq(&ip->i_iopen_gh);
1517         if (error)
1518                 goto out_truncate;
1519
1520         /* Case 1 starts here */
1521
1522         if (S_ISDIR(inode->i_mode) &&
1523             (ip->i_diskflags & GFS2_DIF_EXHASH)) {
1524                 error = gfs2_dir_exhash_dealloc(ip);
1525                 if (error)
1526                         goto out_unlock;
1527         }
1528
1529         if (ip->i_eattr) {
1530                 error = gfs2_ea_dealloc(ip);
1531                 if (error)
1532                         goto out_unlock;
1533         }
1534
1535         if (!gfs2_is_stuffed(ip)) {
1536                 error = gfs2_file_dealloc(ip);
1537                 if (error)
1538                         goto out_unlock;
1539         }
1540
1541         error = gfs2_dinode_dealloc(ip);
1542         goto out_unlock;
1543
1544 out_truncate:
1545         gfs2_log_flush(sdp, ip->i_gl, NORMAL_FLUSH);
1546         if (test_bit(GLF_DIRTY, &ip->i_gl->gl_flags)) {
1547                 struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl);
1548                 filemap_fdatawrite(metamapping);
1549                 filemap_fdatawait(metamapping);
1550         }
1551         write_inode_now(inode, 1);
1552         gfs2_ail_flush(ip->i_gl, 0);
1553
1554         /* Case 2 starts here */
1555         error = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
1556         if (error)
1557                 goto out_unlock;
1558         /* Needs to be done before glock release & also in a transaction */
1559         truncate_inode_pages(&inode->i_data, 0);
1560         gfs2_trans_end(sdp);
1561
1562 out_unlock:
1563         /* Error path for case 1 */
1564         if (gfs2_rs_active(ip->i_res))
1565                 gfs2_rs_deltree(ip->i_res);
1566
1567         if (test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1568                 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1569                 gfs2_glock_dq(&ip->i_iopen_gh);
1570         }
1571         gfs2_holder_uninit(&ip->i_iopen_gh);
1572         gfs2_glock_dq_uninit(&gh);
1573         if (error && error != GLR_TRYFAILED && error != -EROFS)
1574                 fs_warn(sdp, "gfs2_evict_inode: %d\n", error);
1575 out:
1576         /* Case 3 starts here */
1577         truncate_inode_pages_final(&inode->i_data);
1578         gfs2_rs_delete(ip, NULL);
1579         gfs2_ordered_del_inode(ip);
1580         clear_inode(inode);
1581         gfs2_dir_hash_inval(ip);
1582         ip->i_gl->gl_object = NULL;
1583         flush_delayed_work(&ip->i_gl->gl_work);
1584         gfs2_glock_add_to_lru(ip->i_gl);
1585         gfs2_glock_put(ip->i_gl);
1586         ip->i_gl = NULL;
1587         if (ip->i_iopen_gh.gh_gl) {
1588                 ip->i_iopen_gh.gh_gl->gl_object = NULL;
1589                 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1590                 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1591         }
1592 }
1593
1594 static struct inode *gfs2_alloc_inode(struct super_block *sb)
1595 {
1596         struct gfs2_inode *ip;
1597
1598         ip = kmem_cache_alloc(gfs2_inode_cachep, GFP_KERNEL);
1599         if (ip) {
1600                 ip->i_flags = 0;
1601                 ip->i_gl = NULL;
1602                 ip->i_rgd = NULL;
1603                 ip->i_res = NULL;
1604         }
1605         return &ip->i_inode;
1606 }
1607
1608 static void gfs2_i_callback(struct rcu_head *head)
1609 {
1610         struct inode *inode = container_of(head, struct inode, i_rcu);
1611         kmem_cache_free(gfs2_inode_cachep, inode);
1612 }
1613
1614 static void gfs2_destroy_inode(struct inode *inode)
1615 {
1616         call_rcu(&inode->i_rcu, gfs2_i_callback);
1617 }
1618
1619 const struct super_operations gfs2_super_ops = {
1620         .alloc_inode            = gfs2_alloc_inode,
1621         .destroy_inode          = gfs2_destroy_inode,
1622         .write_inode            = gfs2_write_inode,
1623         .dirty_inode            = gfs2_dirty_inode,
1624         .evict_inode            = gfs2_evict_inode,
1625         .put_super              = gfs2_put_super,
1626         .sync_fs                = gfs2_sync_fs,
1627         .freeze_fs              = gfs2_freeze,
1628         .unfreeze_fs            = gfs2_unfreeze,
1629         .statfs                 = gfs2_statfs,
1630         .remount_fs             = gfs2_remount_fs,
1631         .drop_inode             = gfs2_drop_inode,
1632         .show_options           = gfs2_show_options,
1633 };
1634