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 t_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_trans_gl, LM_ST_SHARED, 0, &t_gh);
 411         if (error)
 412                 goto fail_threads;
 413
 414         j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
 415
 416         error = gfs2_find_jhead(sdp->sd_jdesc, &head);
 417         if (error)
 418                 goto fail;
 419
 420         if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 421                 gfs2_consist(sdp);
 422                 error = -EIO;
 423                 goto fail;
 424         }
 425
 426         /*  Initialize some head of the log stuff  */
 427         sdp->sd_log_sequence = head.lh_sequence + 1;
 428         gfs2_log_pointers_init(sdp, head.lh_blkno);
 429
 430         error = gfs2_quota_init(sdp);
 431         if (error)
 432                 goto fail;
 433
 434         set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 435
 436         gfs2_glock_dq_uninit(&t_gh);
 437
 438         return 0;
 439
 440 fail:
 441         t_gh.gh_flags |= GL_NOCACHE;
 442         gfs2_glock_dq_uninit(&t_gh);
 443 fail_threads:
 444         kthread_stop(sdp->sd_quotad_process);
 445         kthread_stop(sdp->sd_logd_process);
 446         return error;
 447 }
 448
 449 void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
 450 {
 451         const struct gfs2_statfs_change *str = buf;
 452
 453         sc->sc_total = be64_to_cpu(str->sc_total);
 454         sc->sc_free = be64_to_cpu(str->sc_free);
 455         sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
 456 }
 457
 458 static void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
 459 {
 460         struct gfs2_statfs_change *str = buf;
 461
 462         str->sc_total = cpu_to_be64(sc->sc_total);
 463         str->sc_free = cpu_to_be64(sc->sc_free);
 464         str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
 465 }
 466
 467 int gfs2_statfs_init(struct gfs2_sbd *sdp)
 468 {
 469         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 470         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 471         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 472         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 473         struct buffer_head *m_bh, *l_bh;
 474         struct gfs2_holder gh;
 475         int error;
 476
 477         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 478                                    &gh);
 479         if (error)
 480                 return error;
 481
 482         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 483         if (error)
 484                 goto out;
 485
 486         if (sdp->sd_args.ar_spectator) {
 487                 spin_lock(&sdp->sd_statfs_spin);
 488                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 489                                       sizeof(struct gfs2_dinode));
 490                 spin_unlock(&sdp->sd_statfs_spin);
 491         } else {
 492                 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 493                 if (error)
 494                         goto out_m_bh;
 495
 496                 spin_lock(&sdp->sd_statfs_spin);
 497                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 498                                       sizeof(struct gfs2_dinode));
 499                 gfs2_statfs_change_in(l_sc, l_bh->b_data +
 500                                       sizeof(struct gfs2_dinode));
 501                 spin_unlock(&sdp->sd_statfs_spin);
 502
 503                 brelse(l_bh);
 504         }
 505
 506 out_m_bh:
 507         brelse(m_bh);
 508 out:
 509         gfs2_glock_dq_uninit(&gh);
 510         return 0;
 511 }
 512
 513 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
 514                         s64 dinodes)
 515 {
 516         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 517         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 518         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 519         struct buffer_head *l_bh;
 520         s64 x, y;
 521         int need_sync = 0;
 522         int error;
 523
 524         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 525         if (error)
 526                 return;
 527
 528         gfs2_trans_add_meta(l_ip->i_gl, l_bh);
 529
 530         spin_lock(&sdp->sd_statfs_spin);
 531         l_sc->sc_total += total;
 532         l_sc->sc_free += free;
 533         l_sc->sc_dinodes += dinodes;
 534         gfs2_statfs_change_out(l_sc, l_bh->b_data + sizeof(struct gfs2_dinode));
 535         if (sdp->sd_args.ar_statfs_percent) {
 536                 x = 100 * l_sc->sc_free;
 537                 y = m_sc->sc_free * sdp->sd_args.ar_statfs_percent;
 538                 if (x >= y || x <= -y)
 539                         need_sync = 1;
 540         }
 541         spin_unlock(&sdp->sd_statfs_spin);
 542
 543         brelse(l_bh);
 544         if (need_sync)
 545                 gfs2_wake_up_statfs(sdp);
 546 }
 547
 548 void update_statfs(struct gfs2_sbd *sdp, struct buffer_head *m_bh,
 549                    struct buffer_head *l_bh)
 550 {
 551         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 552         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 553         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 554         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 555
 556         gfs2_trans_add_meta(l_ip->i_gl, l_bh);
 557
 558         spin_lock(&sdp->sd_statfs_spin);
 559         m_sc->sc_total += l_sc->sc_total;
 560         m_sc->sc_free += l_sc->sc_free;
 561         m_sc->sc_dinodes += l_sc->sc_dinodes;
 562         memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
 563         memset(l_bh->b_data + sizeof(struct gfs2_dinode),
 564                0, sizeof(struct gfs2_statfs_change));
 565         spin_unlock(&sdp->sd_statfs_spin);
 566
 567         gfs2_trans_add_meta(m_ip->i_gl, m_bh);
 568         gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
 569 }
 570
 571 int gfs2_statfs_sync(struct super_block *sb, int type)
 572 {
 573         struct gfs2_sbd *sdp = sb->s_fs_info;
 574         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 575         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 576         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 577         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 578         struct gfs2_holder gh;
 579         struct buffer_head *m_bh, *l_bh;
 580         int error;
 581
 582         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 583                                    &gh);
 584         if (error)
 585                 return error;
 586
 587         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 588         if (error)
 589                 goto out;
 590
 591         spin_lock(&sdp->sd_statfs_spin);
 592         gfs2_statfs_change_in(m_sc, m_bh->b_data +
 593                               sizeof(struct gfs2_dinode));
 594         if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
 595                 spin_unlock(&sdp->sd_statfs_spin);
 596                 goto out_bh;
 597         }
 598         spin_unlock(&sdp->sd_statfs_spin);
 599
 600         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 601         if (error)
 602                 goto out_bh;
 603
 604         error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
 605         if (error)
 606                 goto out_bh2;
 607
 608         update_statfs(sdp, m_bh, l_bh);
 609         sdp->sd_statfs_force_sync = 0;
 610
 611         gfs2_trans_end(sdp);
 612
 613 out_bh2:
 614         brelse(l_bh);
 615 out_bh:
 616         brelse(m_bh);
 617 out:
 618         gfs2_glock_dq_uninit(&gh);
 619         return error;
 620 }
 621
 622 struct lfcc {
 623         struct list_head list;
 624         struct gfs2_holder gh;
 625 };
 626
 627 /**
 628  * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
 629  *                            journals are clean
 630  * @sdp: the file system
 631  * @state: the state to put the transaction lock into
 632  * @t_gh: the hold on the transaction lock
 633  *
 634  * Returns: errno
 635  */
 636
 637 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp,
 638                                     struct gfs2_holder *t_gh)
 639 {
 640         struct gfs2_inode *ip;
 641         struct gfs2_jdesc *jd;
 642         struct lfcc *lfcc;
 643         LIST_HEAD(list);
 644         struct gfs2_log_header_host lh;
 645         int error;
 646
 647         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 648                 lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
 649                 if (!lfcc) {
 650                         error = -ENOMEM;
 651                         goto out;
 652                 }
 653                 ip = GFS2_I(jd->jd_inode);
 654                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
 655                 if (error) {
 656                         kfree(lfcc);
 657                         goto out;
 658                 }
 659                 list_add(&lfcc->list, &list);
 660         }
 661
 662         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_DEFERRED,
 663                                    GL_NOCACHE, t_gh);
 664
 665         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 666                 error = gfs2_jdesc_check(jd);
 667                 if (error)
 668                         break;
 669                 error = gfs2_find_jhead(jd, &lh);
 670                 if (error)
 671                         break;
 672                 if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 673                         error = -EBUSY;
 674                         break;
 675                 }
 676         }
 677
 678         if (error)
 679                 gfs2_glock_dq_uninit(t_gh);
 680
 681 out:
 682         while (!list_empty(&list)) {
 683                 lfcc = list_entry(list.next, struct lfcc, list);
 684                 list_del(&lfcc->list);
 685                 gfs2_glock_dq_uninit(&lfcc->gh);
 686                 kfree(lfcc);
 687         }
 688         return error;
 689 }
 690
 691 void gfs2_dinode_out(const struct gfs2_inode *ip, void *buf)
 692 {
 693         struct gfs2_dinode *str = buf;
 694
 695         str->di_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 696         str->di_header.mh_type = cpu_to_be32(GFS2_METATYPE_DI);
 697         str->di_header.mh_format = cpu_to_be32(GFS2_FORMAT_DI);
 698         str->di_num.no_addr = cpu_to_be64(ip->i_no_addr);
 699         str->di_num.no_formal_ino = cpu_to_be64(ip->i_no_formal_ino);
 700         str->di_mode = cpu_to_be32(ip->i_inode.i_mode);
 701         str->di_uid = cpu_to_be32(i_uid_read(&ip->i_inode));
 702         str->di_gid = cpu_to_be32(i_gid_read(&ip->i_inode));
 703         str->di_nlink = cpu_to_be32(ip->i_inode.i_nlink);
 704         str->di_size = cpu_to_be64(i_size_read(&ip->i_inode));
 705         str->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(&ip->i_inode));
 706         str->di_atime = cpu_to_be64(ip->i_inode.i_atime.tv_sec);
 707         str->di_mtime = cpu_to_be64(ip->i_inode.i_mtime.tv_sec);
 708         str->di_ctime = cpu_to_be64(ip->i_inode.i_ctime.tv_sec);
 709
 710         str->di_goal_meta = cpu_to_be64(ip->i_goal);
 711         str->di_goal_data = cpu_to_be64(ip->i_goal);
 712         str->di_generation = cpu_to_be64(ip->i_generation);
 713
 714         str->di_flags = cpu_to_be32(ip->i_diskflags);
 715         str->di_height = cpu_to_be16(ip->i_height);
 716         str->di_payload_format = cpu_to_be32(S_ISDIR(ip->i_inode.i_mode) &&
 717                                              !(ip->i_diskflags & GFS2_DIF_EXHASH) ?
 718                                              GFS2_FORMAT_DE : 0);
 719         str->di_depth = cpu_to_be16(ip->i_depth);
 720         str->di_entries = cpu_to_be32(ip->i_entries);
 721
 722         str->di_eattr = cpu_to_be64(ip->i_eattr);
 723         str->di_atime_nsec = cpu_to_be32(ip->i_inode.i_atime.tv_nsec);
 724         str->di_mtime_nsec = cpu_to_be32(ip->i_inode.i_mtime.tv_nsec);
 725         str->di_ctime_nsec = cpu_to_be32(ip->i_inode.i_ctime.tv_nsec);
 726 }
 727
 728 /**
 729  * gfs2_write_inode - Make sure the inode is stable on the disk
 730  * @inode: The inode
 731  * @wbc: The writeback control structure
 732  *
 733  * Returns: errno
 734  */
 735
 736 static int gfs2_write_inode(struct inode *inode, struct writeback_control *wbc)
 737 {
 738         struct gfs2_inode *ip = GFS2_I(inode);
 739         struct gfs2_sbd *sdp = GFS2_SB(inode);
 740         struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl);
 741         struct backing_dev_info *bdi = metamapping->backing_dev_info;
 742         int ret = 0;
 743
 744         if (wbc->sync_mode == WB_SYNC_ALL)
 745                 gfs2_log_flush(GFS2_SB(inode), ip->i_gl);
 746         if (bdi->dirty_exceeded)
 747                 gfs2_ail1_flush(sdp, wbc);
 748         else
 749                 filemap_fdatawrite(metamapping);
 750         if (wbc->sync_mode == WB_SYNC_ALL)
 751                 ret = filemap_fdatawait(metamapping);
 752         if (ret)
 753                 mark_inode_dirty_sync(inode);
 754         return ret;
 755 }
 756
 757 /**
 758  * gfs2_dirty_inode - check for atime updates
 759  * @inode: The inode in question
 760  * @flags: The type of dirty
 761  *
 762  * Unfortunately it can be called under any combination of inode
 763  * glock and transaction lock, so we have to check carefully.
 764  *
 765  * At the moment this deals only with atime - it should be possible
 766  * to expand that role in future, once a review of the locking has
 767  * been carried out.
 768  */
 769
 770 static void gfs2_dirty_inode(struct inode *inode, int flags)
 771 {
 772         struct gfs2_inode *ip = GFS2_I(inode);
 773         struct gfs2_sbd *sdp = GFS2_SB(inode);
 774         struct buffer_head *bh;
 775         struct gfs2_holder gh;
 776         int need_unlock = 0;
 777         int need_endtrans = 0;
 778         int ret;
 779
 780         if (!(flags & (I_DIRTY_DATASYNC|I_DIRTY_SYNC)))
 781                 return;
 782
 783         if (!gfs2_glock_is_locked_by_me(ip->i_gl)) {
 784                 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
 785                 if (ret) {
 786                         fs_err(sdp, "dirty_inode: glock %d\n", ret);
 787                         return;
 788                 }
 789                 need_unlock = 1;
 790         } else if (WARN_ON_ONCE(ip->i_gl->gl_state != LM_ST_EXCLUSIVE))
 791                 return;
 792
 793         if (current->journal_info == NULL) {
 794                 ret = gfs2_trans_begin(sdp, RES_DINODE, 0);
 795                 if (ret) {
 796                         fs_err(sdp, "dirty_inode: gfs2_trans_begin %d\n", ret);
 797                         goto out;
 798                 }
 799                 need_endtrans = 1;
 800         }
 801
 802         ret = gfs2_meta_inode_buffer(ip, &bh);
 803         if (ret == 0) {
 804                 gfs2_trans_add_meta(ip->i_gl, bh);
 805                 gfs2_dinode_out(ip, bh->b_data);
 806                 brelse(bh);
 807         }
 808
 809         if (need_endtrans)
 810                 gfs2_trans_end(sdp);
 811 out:
 812         if (need_unlock)
 813                 gfs2_glock_dq_uninit(&gh);
 814 }
 815
 816 /**
 817  * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
 818  * @sdp: the filesystem
 819  *
 820  * Returns: errno
 821  */
 822
 823 static int gfs2_make_fs_ro(struct gfs2_sbd *sdp)
 824 {
 825         struct gfs2_holder t_gh;
 826         int error;
 827
 828         kthread_stop(sdp->sd_quotad_process);
 829         kthread_stop(sdp->sd_logd_process);
 830
 831         flush_workqueue(gfs2_delete_workqueue);
 832         gfs2_quota_sync(sdp->sd_vfs, 0);
 833         gfs2_statfs_sync(sdp->sd_vfs, 0);
 834
 835         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, GL_NOCACHE,
 836                                    &t_gh);
 837         if (error && !test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
 838                 return error;
 839
 840         gfs2_meta_syncfs(sdp);
 841         gfs2_log_shutdown(sdp);
 842
 843         clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 844
 845         if (t_gh.gh_gl)
 846                 gfs2_glock_dq_uninit(&t_gh);
 847
 848         gfs2_quota_cleanup(sdp);
 849
 850         return error;
 851 }
 852
 853 static int gfs2_umount_recovery_wait(void *word)
 854 {
 855         schedule();
 856         return 0;
 857 }
 858
 859 /**
 860  * gfs2_put_super - Unmount the filesystem
 861  * @sb: The VFS superblock
 862  *
 863  */
 864
 865 static void gfs2_put_super(struct super_block *sb)
 866 {
 867         struct gfs2_sbd *sdp = sb->s_fs_info;
 868         int error;
 869         struct gfs2_jdesc *jd;
 870
 871         /* No more recovery requests */
 872         set_bit(SDF_NORECOVERY, &sdp->sd_flags);
 873         smp_mb();
 874
 875         /* Wait on outstanding recovery */
 876 restart:
 877         spin_lock(&sdp->sd_jindex_spin);
 878         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 879                 if (!test_bit(JDF_RECOVERY, &jd->jd_flags))
 880                         continue;
 881                 spin_unlock(&sdp->sd_jindex_spin);
 882                 wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
 883                             gfs2_umount_recovery_wait, TASK_UNINTERRUPTIBLE);
 884                 goto restart;
 885         }
 886         spin_unlock(&sdp->sd_jindex_spin);
 887
 888         if (!(sb->s_flags & MS_RDONLY)) {
 889                 error = gfs2_make_fs_ro(sdp);
 890                 if (error)
 891                         gfs2_io_error(sdp);
 892         }
 893         /*  At this point, we're through modifying the disk  */
 894
 895         /*  Release stuff  */
 896
 897         iput(sdp->sd_jindex);
 898         iput(sdp->sd_statfs_inode);
 899         iput(sdp->sd_rindex);
 900         iput(sdp->sd_quota_inode);
 901
 902         gfs2_glock_put(sdp->sd_rename_gl);
 903         gfs2_glock_put(sdp->sd_trans_gl);
 904
 905         if (!sdp->sd_args.ar_spectator) {
 906                 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
 907                 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
 908                 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
 909                 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
 910                 iput(sdp->sd_sc_inode);
 911                 iput(sdp->sd_qc_inode);
 912         }
 913
 914         gfs2_glock_dq_uninit(&sdp->sd_live_gh);
 915         gfs2_clear_rgrpd(sdp);
 916         gfs2_jindex_free(sdp);
 917         /*  Take apart glock structures and buffer lists  */
 918         gfs2_gl_hash_clear(sdp);
 919         /*  Unmount the locking protocol  */
 920         gfs2_lm_unmount(sdp);
 921
 922         /*  At this point, we're through participating in the lockspace  */
 923         gfs2_sys_fs_del(sdp);
 924 }
 925
 926 /**
 927  * gfs2_sync_fs - sync the filesystem
 928  * @sb: the superblock
 929  *
 930  * Flushes the log to disk.
 931  */
 932
 933 static int gfs2_sync_fs(struct super_block *sb, int wait)
 934 {
 935         struct gfs2_sbd *sdp = sb->s_fs_info;
 936
 937         gfs2_quota_sync(sb, -1);
 938         if (wait && sdp)
 939                 gfs2_log_flush(sdp, NULL);
 940         return 0;
 941 }
 942
 943 /**
 944  * gfs2_freeze - prevent further writes to the filesystem
 945  * @sb: the VFS structure for the filesystem
 946  *
 947  */
 948
 949 static int gfs2_freeze(struct super_block *sb)
 950 {
 951         struct gfs2_sbd *sdp = sb->s_fs_info;
 952         int error;
 953
 954         if (test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
 955                 return -EINVAL;
 956
 957         for (;;) {
 958                 error = gfs2_lock_fs_check_clean(sdp, &sdp->sd_freeze_gh);
 959                 if (!error)
 960                         break;
 961
 962                 switch (error) {
 963                 case -EBUSY:
 964                         fs_err(sdp, "waiting for recovery before freeze\n");
 965                         break;
 966
 967                 default:
 968                         fs_err(sdp, "error freezing FS: %d\n", error);
 969                         break;
 970                 }
 971
 972                 fs_err(sdp, "retrying...\n");
 973                 msleep(1000);
 974         }
 975         return 0;
 976 }
 977
 978 /**
 979  * gfs2_unfreeze - reallow writes to the filesystem
 980  * @sb: the VFS structure for the filesystem
 981  *
 982  */
 983
 984 static int gfs2_unfreeze(struct super_block *sb)
 985 {
 986         struct gfs2_sbd *sdp = sb->s_fs_info;
 987
 988         gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
 989         return 0;
 990 }
 991
 992 /**
 993  * statfs_fill - fill in the sg for a given RG
 994  * @rgd: the RG
 995  * @sc: the sc structure
 996  *
 997  * Returns: 0 on success, -ESTALE if the LVB is invalid
 998  */
 999
1000 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
1001                             struct gfs2_statfs_change_host *sc)
1002 {
1003         gfs2_rgrp_verify(rgd);
1004         sc->sc_total += rgd->rd_data;
1005         sc->sc_free += rgd->rd_free;
1006         sc->sc_dinodes += rgd->rd_dinodes;
1007         return 0;
1008 }
1009
1010 /**
1011  * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
1012  * @sdp: the filesystem
1013  * @sc: the sc info that will be returned
1014  *
1015  * Any error (other than a signal) will cause this routine to fall back
1016  * to the synchronous version.
1017  *
1018  * FIXME: This really shouldn't busy wait like this.
1019  *
1020  * Returns: errno
1021  */
1022
1023 static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
1024 {
1025         struct gfs2_rgrpd *rgd_next;
1026         struct gfs2_holder *gha, *gh;
1027         unsigned int slots = 64;
1028         unsigned int x;
1029         int done;
1030         int error = 0, err;
1031
1032         memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
1033         gha = kcalloc(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
1034         if (!gha)
1035                 return -ENOMEM;
1036
1037         rgd_next = gfs2_rgrpd_get_first(sdp);
1038
1039         for (;;) {
1040                 done = 1;
1041
1042                 for (x = 0; x < slots; x++) {
1043                         gh = gha + x;
1044
1045                         if (gh->gh_gl && gfs2_glock_poll(gh)) {
1046                                 err = gfs2_glock_wait(gh);
1047                                 if (err) {
1048                                         gfs2_holder_uninit(gh);
1049                                         error = err;
1050                                 } else {
1051                                         if (!error)
1052                                                 error = statfs_slow_fill(
1053                                                         gh->gh_gl->gl_object, sc);
1054                                         gfs2_glock_dq_uninit(gh);
1055                                 }
1056                         }
1057
1058                         if (gh->gh_gl)
1059                                 done = 0;
1060                         else if (rgd_next && !error) {
1061                                 error = gfs2_glock_nq_init(rgd_next->rd_gl,
1062                                                            LM_ST_SHARED,
1063                                                            GL_ASYNC,
1064                                                            gh);
1065                                 rgd_next = gfs2_rgrpd_get_next(rgd_next);
1066                                 done = 0;
1067                         }
1068
1069                         if (signal_pending(current))
1070                                 error = -ERESTARTSYS;
1071                 }
1072
1073                 if (done)
1074                         break;
1075
1076                 yield();
1077         }
1078
1079         kfree(gha);
1080         return error;
1081 }
1082
1083 /**
1084  * gfs2_statfs_i - Do a statfs
1085  * @sdp: the filesystem
1086  * @sg: the sg structure
1087  *
1088  * Returns: errno
1089  */
1090
1091 static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
1092 {
1093         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
1094         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
1095
1096         spin_lock(&sdp->sd_statfs_spin);
1097
1098         *sc = *m_sc;
1099         sc->sc_total += l_sc->sc_total;
1100         sc->sc_free += l_sc->sc_free;
1101         sc->sc_dinodes += l_sc->sc_dinodes;
1102
1103         spin_unlock(&sdp->sd_statfs_spin);
1104
1105         if (sc->sc_free < 0)
1106                 sc->sc_free = 0;
1107         if (sc->sc_free > sc->sc_total)
1108                 sc->sc_free = sc->sc_total;
1109         if (sc->sc_dinodes < 0)
1110                 sc->sc_dinodes = 0;
1111
1112         return 0;
1113 }
1114
1115 /**
1116  * gfs2_statfs - Gather and return stats about the filesystem
1117  * @sb: The superblock
1118  * @statfsbuf: The buffer
1119  *
1120  * Returns: 0 on success or error code
1121  */
1122
1123 static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
1124 {
1125         struct super_block *sb = dentry->d_inode->i_sb;
1126         struct gfs2_sbd *sdp = sb->s_fs_info;
1127         struct gfs2_statfs_change_host sc;
1128         int error;
1129
1130         error = gfs2_rindex_update(sdp);
1131         if (error)
1132                 return error;
1133
1134         if (gfs2_tune_get(sdp, gt_statfs_slow))
1135                 error = gfs2_statfs_slow(sdp, &sc);
1136         else
1137                 error = gfs2_statfs_i(sdp, &sc);
1138
1139         if (error)
1140                 return error;
1141
1142         buf->f_type = GFS2_MAGIC;
1143         buf->f_bsize = sdp->sd_sb.sb_bsize;
1144         buf->f_blocks = sc.sc_total;
1145         buf->f_bfree = sc.sc_free;
1146         buf->f_bavail = sc.sc_free;
1147         buf->f_files = sc.sc_dinodes + sc.sc_free;
1148         buf->f_ffree = sc.sc_free;
1149         buf->f_namelen = GFS2_FNAMESIZE;
1150
1151         return 0;
1152 }
1153
1154 /**
1155  * gfs2_remount_fs - called when the FS is remounted
1156  * @sb:  the filesystem
1157  * @flags:  the remount flags
1158  * @data:  extra data passed in (not used right now)
1159  *
1160  * Returns: errno
1161  */
1162
1163 static int gfs2_remount_fs(struct super_block *sb, int *flags, char *data)
1164 {
1165         struct gfs2_sbd *sdp = sb->s_fs_info;
1166         struct gfs2_args args = sdp->sd_args; /* Default to current settings */
1167         struct gfs2_tune *gt = &sdp->sd_tune;
1168         int error;
1169
1170         spin_lock(&gt->gt_spin);
1171         args.ar_commit = gt->gt_logd_secs;
1172         args.ar_quota_quantum = gt->gt_quota_quantum;
1173         if (gt->gt_statfs_slow)
1174                 args.ar_statfs_quantum = 0;
1175         else
1176                 args.ar_statfs_quantum = gt->gt_statfs_quantum;
1177         spin_unlock(&gt->gt_spin);
1178         error = gfs2_mount_args(&args, data);
1179         if (error)
1180                 return error;
1181
1182         /* Not allowed to change locking details */
1183         if (strcmp(args.ar_lockproto, sdp->sd_args.ar_lockproto) ||
1184             strcmp(args.ar_locktable, sdp->sd_args.ar_locktable) ||
1185             strcmp(args.ar_hostdata, sdp->sd_args.ar_hostdata))
1186                 return -EINVAL;
1187
1188         /* Some flags must not be changed */
1189         if (args_neq(&args, &sdp->sd_args, spectator) ||
1190             args_neq(&args, &sdp->sd_args, localflocks) ||
1191             args_neq(&args, &sdp->sd_args, meta))
1192                 return -EINVAL;
1193
1194         if (sdp->sd_args.ar_spectator)
1195                 *flags |= MS_RDONLY;
1196
1197         if ((sb->s_flags ^ *flags) & MS_RDONLY) {
1198                 if (*flags & MS_RDONLY)
1199                         error = gfs2_make_fs_ro(sdp);
1200                 else
1201                         error = gfs2_make_fs_rw(sdp);
1202                 if (error)
1203                         return error;
1204         }
1205
1206         sdp->sd_args = args;
1207         if (sdp->sd_args.ar_posix_acl)
1208                 sb->s_flags |= MS_POSIXACL;
1209         else
1210                 sb->s_flags &= ~MS_POSIXACL;
1211         if (sdp->sd_args.ar_nobarrier)
1212                 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1213         else
1214                 clear_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1215         spin_lock(&gt->gt_spin);
1216         gt->gt_logd_secs = args.ar_commit;
1217         gt->gt_quota_quantum = args.ar_quota_quantum;
1218         if (args.ar_statfs_quantum) {
1219                 gt->gt_statfs_slow = 0;
1220                 gt->gt_statfs_quantum = args.ar_statfs_quantum;
1221         }
1222         else {
1223                 gt->gt_statfs_slow = 1;
1224                 gt->gt_statfs_quantum = 30;
1225         }
1226         spin_unlock(&gt->gt_spin);
1227
1228         gfs2_online_uevent(sdp);
1229         return 0;
1230 }
1231
1232 /**
1233  * gfs2_drop_inode - Drop an inode (test for remote unlink)
1234  * @inode: The inode to drop
1235  *
1236  * If we've received a callback on an iopen lock then its because a
1237  * remote node tried to deallocate the inode but failed due to this node
1238  * still having the inode open. Here we mark the link count zero
1239  * since we know that it must have reached zero if the GLF_DEMOTE flag
1240  * is set on the iopen glock. If we didn't do a disk read since the
1241  * remote node removed the final link then we might otherwise miss
1242  * this event. This check ensures that this node will deallocate the
1243  * inode's blocks, or alternatively pass the baton on to another
1244  * node for later deallocation.
1245  */
1246
1247 static int gfs2_drop_inode(struct inode *inode)
1248 {
1249         struct gfs2_inode *ip = GFS2_I(inode);
1250
1251         if (!test_bit(GIF_FREE_VFS_INODE, &ip->i_flags) && inode->i_nlink) {
1252                 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1253                 if (gl && test_bit(GLF_DEMOTE, &gl->gl_flags))
1254                         clear_nlink(inode);
1255         }
1256         return generic_drop_inode(inode);
1257 }
1258
1259 static int is_ancestor(const struct dentry *d1, const struct dentry *d2)
1260 {
1261         do {
1262                 if (d1 == d2)
1263                         return 1;
1264                 d1 = d1->d_parent;
1265         } while (!IS_ROOT(d1));
1266         return 0;
1267 }
1268
1269 /**
1270  * gfs2_show_options - Show mount options for /proc/mounts
1271  * @s: seq_file structure
1272  * @root: root of this (sub)tree
1273  *
1274  * Returns: 0 on success or error code
1275  */
1276
1277 static int gfs2_show_options(struct seq_file *s, struct dentry *root)
1278 {
1279         struct gfs2_sbd *sdp = root->d_sb->s_fs_info;
1280         struct gfs2_args *args = &sdp->sd_args;
1281         int val;
1282
1283         if (is_ancestor(root, sdp->sd_master_dir))
1284                 seq_printf(s, ",meta");
1285         if (args->ar_lockproto[0])
1286                 seq_printf(s, ",lockproto=%s", args->ar_lockproto);
1287         if (args->ar_locktable[0])
1288                 seq_printf(s, ",locktable=%s", args->ar_locktable);
1289         if (args->ar_hostdata[0])
1290                 seq_printf(s, ",hostdata=%s", args->ar_hostdata);
1291         if (args->ar_spectator)
1292                 seq_printf(s, ",spectator");
1293         if (args->ar_localflocks)
1294                 seq_printf(s, ",localflocks");
1295         if (args->ar_debug)
1296                 seq_printf(s, ",debug");
1297         if (args->ar_posix_acl)
1298                 seq_printf(s, ",acl");
1299         if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
1300                 char *state;
1301                 switch (args->ar_quota) {
1302                 case GFS2_QUOTA_OFF:
1303                         state = "off";
1304                         break;
1305                 case GFS2_QUOTA_ACCOUNT:
1306                         state = "account";
1307                         break;
1308                 case GFS2_QUOTA_ON:
1309                         state = "on";
1310                         break;
1311                 default:
1312                         state = "unknown";
1313                         break;
1314                 }
1315                 seq_printf(s, ",quota=%s", state);
1316         }
1317         if (args->ar_suiddir)
1318                 seq_printf(s, ",suiddir");
1319         if (args->ar_data != GFS2_DATA_DEFAULT) {
1320                 char *state;
1321                 switch (args->ar_data) {
1322                 case GFS2_DATA_WRITEBACK:
1323                         state = "writeback";
1324                         break;
1325                 case GFS2_DATA_ORDERED:
1326                         state = "ordered";
1327                         break;
1328                 default:
1329                         state = "unknown";
1330                         break;
1331                 }
1332                 seq_printf(s, ",data=%s", state);
1333         }
1334         if (args->ar_discard)
1335                 seq_printf(s, ",discard");
1336         val = sdp->sd_tune.gt_logd_secs;
1337         if (val != 30)
1338                 seq_printf(s, ",commit=%d", val);
1339         val = sdp->sd_tune.gt_statfs_quantum;
1340         if (val != 30)
1341                 seq_printf(s, ",statfs_quantum=%d", val);
1342         else if (sdp->sd_tune.gt_statfs_slow)
1343                 seq_puts(s, ",statfs_quantum=0");
1344         val = sdp->sd_tune.gt_quota_quantum;
1345         if (val != 60)
1346                 seq_printf(s, ",quota_quantum=%d", val);
1347         if (args->ar_statfs_percent)
1348                 seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent);
1349         if (args->ar_errors != GFS2_ERRORS_DEFAULT) {
1350                 const char *state;
1351
1352                 switch (args->ar_errors) {
1353                 case GFS2_ERRORS_WITHDRAW:
1354                         state = "withdraw";
1355                         break;
1356                 case GFS2_ERRORS_PANIC:
1357                         state = "panic";
1358                         break;
1359                 default:
1360                         state = "unknown";
1361                         break;
1362                 }
1363                 seq_printf(s, ",errors=%s", state);
1364         }
1365         if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags))
1366                 seq_printf(s, ",nobarrier");
1367         if (test_bit(SDF_DEMOTE, &sdp->sd_flags))
1368                 seq_printf(s, ",demote_interface_used");
1369         if (args->ar_rgrplvb)
1370                 seq_printf(s, ",rgrplvb");
1371         return 0;
1372 }
1373
1374 static void gfs2_final_release_pages(struct gfs2_inode *ip)
1375 {
1376         struct inode *inode = &ip->i_inode;
1377         struct gfs2_glock *gl = ip->i_gl;
1378
1379         truncate_inode_pages(gfs2_glock2aspace(ip->i_gl), 0);
1380         truncate_inode_pages(&inode->i_data, 0);
1381
1382         if (atomic_read(&gl->gl_revokes) == 0) {
1383                 clear_bit(GLF_LFLUSH, &gl->gl_flags);
1384                 clear_bit(GLF_DIRTY, &gl->gl_flags);
1385         }
1386 }
1387
1388 static int gfs2_dinode_dealloc(struct gfs2_inode *ip)
1389 {
1390         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1391         struct gfs2_rgrpd *rgd;
1392         struct gfs2_holder gh;
1393         int error;
1394
1395         if (gfs2_get_inode_blocks(&ip->i_inode) != 1) {
1396                 gfs2_consist_inode(ip);
1397                 return -EIO;
1398         }
1399
1400         error = gfs2_rindex_update(sdp);
1401         if (error)
1402                 return error;
1403
1404         error = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
1405         if (error)
1406                 return error;
1407
1408         rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1409         if (!rgd) {
1410                 gfs2_consist_inode(ip);
1411                 error = -EIO;
1412                 goto out_qs;
1413         }
1414
1415         error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &gh);
1416         if (error)
1417                 goto out_qs;
1418
1419         error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA,
1420                                  sdp->sd_jdesc->jd_blocks);
1421         if (error)
1422                 goto out_rg_gunlock;
1423
1424         gfs2_free_di(rgd, ip);
1425
1426         gfs2_final_release_pages(ip);
1427
1428         gfs2_trans_end(sdp);
1429
1430 out_rg_gunlock:
1431         gfs2_glock_dq_uninit(&gh);
1432 out_qs:
1433         gfs2_quota_unhold(ip);
1434         return error;
1435 }
1436
1437 /**
1438  * gfs2_evict_inode - Remove an inode from cache
1439  * @inode: The inode to evict
1440  *
1441  * There are three cases to consider:
1442  * 1. i_nlink == 0, we are final opener (and must deallocate)
1443  * 2. i_nlink == 0, we are not the final opener (and cannot deallocate)
1444  * 3. i_nlink > 0
1445  *
1446  * If the fs is read only, then we have to treat all cases as per #3
1447  * since we are unable to do any deallocation. The inode will be
1448  * deallocated by the next read/write node to attempt an allocation
1449  * in the same resource group
1450  *
1451  * We have to (at the moment) hold the inodes main lock to cover
1452  * the gap between unlocking the shared lock on the iopen lock and
1453  * taking the exclusive lock. I'd rather do a shared -> exclusive
1454  * conversion on the iopen lock, but we can change that later. This
1455  * is safe, just less efficient.
1456  */
1457
1458 static void gfs2_evict_inode(struct inode *inode)
1459 {
1460         struct super_block *sb = inode->i_sb;
1461         struct gfs2_sbd *sdp = sb->s_fs_info;
1462         struct gfs2_inode *ip = GFS2_I(inode);
1463         struct gfs2_holder gh;
1464         int error;
1465
1466         if (test_bit(GIF_FREE_VFS_INODE, &ip->i_flags)) {
1467                 clear_inode(inode);
1468                 return;
1469         }
1470
1471         if (inode->i_nlink || (sb->s_flags & MS_RDONLY))
1472                 goto out;
1473
1474         /* Must not read inode block until block type has been verified */
1475         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, &gh);
1476         if (unlikely(error)) {
1477                 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1478                 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1479                 goto out;
1480         }
1481
1482         if (!test_bit(GIF_ALLOC_FAILED, &ip->i_flags)) {
1483                 error = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED);
1484                 if (error)
1485                         goto out_truncate;
1486         }
1487
1488         if (test_bit(GIF_INVALID, &ip->i_flags)) {
1489                 error = gfs2_inode_refresh(ip);
1490                 if (error)
1491                         goto out_truncate;
1492         }
1493
1494         ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1495         gfs2_glock_dq_wait(&ip->i_iopen_gh);
1496         gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE, &ip->i_iopen_gh);
1497         error = gfs2_glock_nq(&ip->i_iopen_gh);
1498         if (error)
1499                 goto out_truncate;
1500
1501         /* Case 1 starts here */
1502
1503         if (S_ISDIR(inode->i_mode) &&
1504             (ip->i_diskflags & GFS2_DIF_EXHASH)) {
1505                 error = gfs2_dir_exhash_dealloc(ip);
1506                 if (error)
1507                         goto out_unlock;
1508         }
1509
1510         if (ip->i_eattr) {
1511                 error = gfs2_ea_dealloc(ip);
1512                 if (error)
1513                         goto out_unlock;
1514         }
1515
1516         if (!gfs2_is_stuffed(ip)) {
1517                 error = gfs2_file_dealloc(ip);
1518                 if (error)
1519                         goto out_unlock;
1520         }
1521
1522         error = gfs2_dinode_dealloc(ip);
1523         goto out_unlock;
1524
1525 out_truncate:
1526         gfs2_log_flush(sdp, ip->i_gl);
1527         if (test_bit(GLF_DIRTY, &ip->i_gl->gl_flags)) {
1528                 struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl);
1529                 filemap_fdatawrite(metamapping);
1530                 filemap_fdatawait(metamapping);
1531         }
1532         write_inode_now(inode, 1);
1533         gfs2_ail_flush(ip->i_gl, 0);
1534
1535         /* Case 2 starts here */
1536         error = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
1537         if (error)
1538                 goto out_unlock;
1539         /* Needs to be done before glock release & also in a transaction */
1540         truncate_inode_pages(&inode->i_data, 0);
1541         gfs2_trans_end(sdp);
1542
1543 out_unlock:
1544         /* Error path for case 1 */
1545         if (gfs2_rs_active(ip->i_res))
1546                 gfs2_rs_deltree(ip->i_res);
1547
1548         if (test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1549                 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1550                 gfs2_glock_dq(&ip->i_iopen_gh);
1551         }
1552         gfs2_holder_uninit(&ip->i_iopen_gh);
1553         gfs2_glock_dq_uninit(&gh);
1554         if (error && error != GLR_TRYFAILED && error != -EROFS)
1555                 fs_warn(sdp, "gfs2_evict_inode: %d\n", error);
1556 out:
1557         /* Case 3 starts here */
1558         truncate_inode_pages_final(&inode->i_data);
1559         gfs2_rs_delete(ip, NULL);
1560         gfs2_ordered_del_inode(ip);
1561         clear_inode(inode);
1562         gfs2_dir_hash_inval(ip);
1563         ip->i_gl->gl_object = NULL;
1564         flush_delayed_work(&ip->i_gl->gl_work);
1565         gfs2_glock_add_to_lru(ip->i_gl);
1566         gfs2_glock_put(ip->i_gl);
1567         ip->i_gl = NULL;
1568         if (ip->i_iopen_gh.gh_gl) {
1569                 ip->i_iopen_gh.gh_gl->gl_object = NULL;
1570                 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1571                 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1572         }
1573 }
1574
1575 static struct inode *gfs2_alloc_inode(struct super_block *sb)
1576 {
1577         struct gfs2_inode *ip;
1578
1579         ip = kmem_cache_alloc(gfs2_inode_cachep, GFP_KERNEL);
1580         if (ip) {
1581                 ip->i_flags = 0;
1582                 ip->i_gl = NULL;
1583                 ip->i_rgd = NULL;
1584                 ip->i_res = NULL;
1585         }
1586         return &ip->i_inode;
1587 }
1588
1589 static void gfs2_i_callback(struct rcu_head *head)
1590 {
1591         struct inode *inode = container_of(head, struct inode, i_rcu);
1592         kmem_cache_free(gfs2_inode_cachep, inode);
1593 }
1594
1595 static void gfs2_destroy_inode(struct inode *inode)
1596 {
1597         call_rcu(&inode->i_rcu, gfs2_i_callback);
1598 }
1599
1600 const struct super_operations gfs2_super_ops = {
1601         .alloc_inode            = gfs2_alloc_inode,
1602         .destroy_inode          = gfs2_destroy_inode,
1603         .write_inode            = gfs2_write_inode,
1604         .dirty_inode            = gfs2_dirty_inode,
1605         .evict_inode            = gfs2_evict_inode,
1606         .put_super              = gfs2_put_super,
1607         .sync_fs                = gfs2_sync_fs,
1608         .freeze_fs              = gfs2_freeze,
1609         .unfreeze_fs            = gfs2_unfreeze,
1610         .statfs                 = gfs2_statfs,
1611         .remount_fs             = gfs2_remount_fs,
1612         .drop_inode             = gfs2_drop_inode,
1613         .show_options           = gfs2_show_options,
1614 };
1615