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 #include <linux/kernel.h>
  30
  31 #include "gfs2.h"
  32 #include "incore.h"
  33 #include "bmap.h"
  34 #include "dir.h"
  35 #include "glock.h"
  36 #include "glops.h"
  37 #include "inode.h"
  38 #include "log.h"
  39 #include "meta_io.h"
  40 #include "quota.h"
  41 #include "recovery.h"
  42 #include "rgrp.h"
  43 #include "super.h"
  44 #include "trans.h"
  45 #include "util.h"
  46 #include "sys.h"
  47 #include "xattr.h"
  48
  49 #define args_neq(a1, a2, x) ((a1)->ar_##x != (a2)->ar_##x)
  50
  51 enum {
  52         Opt_lockproto,
  53         Opt_locktable,
  54         Opt_hostdata,
  55         Opt_spectator,
  56         Opt_ignore_local_fs,
  57         Opt_localflocks,
  58         Opt_localcaching,
  59         Opt_debug,
  60         Opt_nodebug,
  61         Opt_upgrade,
  62         Opt_acl,
  63         Opt_noacl,
  64         Opt_quota_off,
  65         Opt_quota_account,
  66         Opt_quota_on,
  67         Opt_quota,
  68         Opt_noquota,
  69         Opt_suiddir,
  70         Opt_nosuiddir,
  71         Opt_data_writeback,
  72         Opt_data_ordered,
  73         Opt_meta,
  74         Opt_discard,
  75         Opt_nodiscard,
  76         Opt_commit,
  77         Opt_err_withdraw,
  78         Opt_err_panic,
  79         Opt_statfs_quantum,
  80         Opt_statfs_percent,
  81         Opt_quota_quantum,
  82         Opt_barrier,
  83         Opt_nobarrier,
  84         Opt_rgrplvb,
  85         Opt_norgrplvb,
  86         Opt_error,
  87 };
  88
  89 static const match_table_t tokens = {
  90         {Opt_lockproto, "lockproto=%s"},
  91         {Opt_locktable, "locktable=%s"},
  92         {Opt_hostdata, "hostdata=%s"},
  93         {Opt_spectator, "spectator"},
  94         {Opt_spectator, "norecovery"},
  95         {Opt_ignore_local_fs, "ignore_local_fs"},
  96         {Opt_localflocks, "localflocks"},
  97         {Opt_localcaching, "localcaching"},
  98         {Opt_debug, "debug"},
  99         {Opt_nodebug, "nodebug"},
 100         {Opt_upgrade, "upgrade"},
 101         {Opt_acl, "acl"},
 102         {Opt_noacl, "noacl"},
 103         {Opt_quota_off, "quota=off"},
 104         {Opt_quota_account, "quota=account"},
 105         {Opt_quota_on, "quota=on"},
 106         {Opt_quota, "quota"},
 107         {Opt_noquota, "noquota"},
 108         {Opt_suiddir, "suiddir"},
 109         {Opt_nosuiddir, "nosuiddir"},
 110         {Opt_data_writeback, "data=writeback"},
 111         {Opt_data_ordered, "data=ordered"},
 112         {Opt_meta, "meta"},
 113         {Opt_discard, "discard"},
 114         {Opt_nodiscard, "nodiscard"},
 115         {Opt_commit, "commit=%d"},
 116         {Opt_err_withdraw, "errors=withdraw"},
 117         {Opt_err_panic, "errors=panic"},
 118         {Opt_statfs_quantum, "statfs_quantum=%d"},
 119         {Opt_statfs_percent, "statfs_percent=%d"},
 120         {Opt_quota_quantum, "quota_quantum=%d"},
 121         {Opt_barrier, "barrier"},
 122         {Opt_nobarrier, "nobarrier"},
 123         {Opt_rgrplvb, "rgrplvb"},
 124         {Opt_norgrplvb, "norgrplvb"},
 125         {Opt_error, NULL}
 126 };
 127
 128 /**
 129  * gfs2_mount_args - Parse mount options
 130  * @args: The structure into which the parsed options will be written
 131  * @options: The options to parse
 132  *
 133  * Return: errno
 134  */
 135
 136 int gfs2_mount_args(struct gfs2_args *args, char *options)
 137 {
 138         char *o;
 139         int token;
 140         substring_t tmp[MAX_OPT_ARGS];
 141         int rv;
 142
 143         /* Split the options into tokens with the "," character and
 144            process them */
 145
 146         while (1) {
 147                 o = strsep(&options, ",");
 148                 if (o == NULL)
 149                         break;
 150                 if (*o == '\0')
 151                         continue;
 152
 153                 token = match_token(o, tokens, tmp);
 154                 switch (token) {
 155                 case Opt_lockproto:
 156                         match_strlcpy(args->ar_lockproto, &tmp[0],
 157                                       GFS2_LOCKNAME_LEN);
 158                         break;
 159                 case Opt_locktable:
 160                         match_strlcpy(args->ar_locktable, &tmp[0],
 161                                       GFS2_LOCKNAME_LEN);
 162                         break;
 163                 case Opt_hostdata:
 164                         match_strlcpy(args->ar_hostdata, &tmp[0],
 165                                       GFS2_LOCKNAME_LEN);
 166                         break;
 167                 case Opt_spectator:
 168                         args->ar_spectator = 1;
 169                         break;
 170                 case Opt_ignore_local_fs:
 171                         /* Retained for backwards compat only */
 172                         break;
 173                 case Opt_localflocks:
 174                         args->ar_localflocks = 1;
 175                         break;
 176                 case Opt_localcaching:
 177                         /* Retained for backwards compat only */
 178                         break;
 179                 case Opt_debug:
 180                         if (args->ar_errors == GFS2_ERRORS_PANIC) {
 181                                 pr_warn("-o debug and -o errors=panic are mutually exclusive\n");
 182                                 return -EINVAL;
 183                         }
 184                         args->ar_debug = 1;
 185                         break;
 186                 case Opt_nodebug:
 187                         args->ar_debug = 0;
 188                         break;
 189                 case Opt_upgrade:
 190                         /* Retained for backwards compat only */
 191                         break;
 192                 case Opt_acl:
 193                         args->ar_posix_acl = 1;
 194                         break;
 195                 case Opt_noacl:
 196                         args->ar_posix_acl = 0;
 197                         break;
 198                 case Opt_quota_off:
 199                 case Opt_noquota:
 200                         args->ar_quota = GFS2_QUOTA_OFF;
 201                         break;
 202                 case Opt_quota_account:
 203                         args->ar_quota = GFS2_QUOTA_ACCOUNT;
 204                         break;
 205                 case Opt_quota_on:
 206                 case Opt_quota:
 207                         args->ar_quota = GFS2_QUOTA_ON;
 208                         break;
 209                 case Opt_suiddir:
 210                         args->ar_suiddir = 1;
 211                         break;
 212                 case Opt_nosuiddir:
 213                         args->ar_suiddir = 0;
 214                         break;
 215                 case Opt_data_writeback:
 216                         args->ar_data = GFS2_DATA_WRITEBACK;
 217                         break;
 218                 case Opt_data_ordered:
 219                         args->ar_data = GFS2_DATA_ORDERED;
 220                         break;
 221                 case Opt_meta:
 222                         args->ar_meta = 1;
 223                         break;
 224                 case Opt_discard:
 225                         args->ar_discard = 1;
 226                         break;
 227                 case Opt_nodiscard:
 228                         args->ar_discard = 0;
 229                         break;
 230                 case Opt_commit:
 231                         rv = match_int(&tmp[0], &args->ar_commit);
 232                         if (rv || args->ar_commit <= 0) {
 233                                 pr_warn("commit mount option requires a positive numeric argument\n");
 234                                 return rv ? rv : -EINVAL;
 235                         }
 236                         break;
 237                 case Opt_statfs_quantum:
 238                         rv = match_int(&tmp[0], &args->ar_statfs_quantum);
 239                         if (rv || args->ar_statfs_quantum < 0) {
 240                                 pr_warn("statfs_quantum mount option requires a non-negative numeric argument\n");
 241                                 return rv ? rv : -EINVAL;
 242                         }
 243                         break;
 244                 case Opt_quota_quantum:
 245                         rv = match_int(&tmp[0], &args->ar_quota_quantum);
 246                         if (rv || args->ar_quota_quantum <= 0) {
 247                                 pr_warn("quota_quantum mount option requires a positive numeric argument\n");
 248                                 return rv ? rv : -EINVAL;
 249                         }
 250                         break;
 251                 case Opt_statfs_percent:
 252                         rv = match_int(&tmp[0], &args->ar_statfs_percent);
 253                         if (rv || args->ar_statfs_percent < 0 ||
 254                             args->ar_statfs_percent > 100) {
 255                                 pr_warn("statfs_percent mount option requires a numeric argument between 0 and 100\n");
 256                                 return rv ? rv : -EINVAL;
 257                         }
 258                         break;
 259                 case Opt_err_withdraw:
 260                         args->ar_errors = GFS2_ERRORS_WITHDRAW;
 261                         break;
 262                 case Opt_err_panic:
 263                         if (args->ar_debug) {
 264                                 pr_warn("-o debug and -o errors=panic are mutually exclusive\n");
 265                                 return -EINVAL;
 266                         }
 267                         args->ar_errors = GFS2_ERRORS_PANIC;
 268                         break;
 269                 case Opt_barrier:
 270                         args->ar_nobarrier = 0;
 271                         break;
 272                 case Opt_nobarrier:
 273                         args->ar_nobarrier = 1;
 274                         break;
 275                 case Opt_rgrplvb:
 276                         args->ar_rgrplvb = 1;
 277                         break;
 278                 case Opt_norgrplvb:
 279                         args->ar_rgrplvb = 0;
 280                         break;
 281                 case Opt_error:
 282                 default:
 283                         pr_warn("invalid mount option: %s\n", o);
 284                         return -EINVAL;
 285                 }
 286         }
 287
 288         return 0;
 289 }
 290
 291 /**
 292  * gfs2_jindex_free - Clear all the journal index information
 293  * @sdp: The GFS2 superblock
 294  *
 295  */
 296
 297 void gfs2_jindex_free(struct gfs2_sbd *sdp)
 298 {
 299         struct list_head list;
 300         struct gfs2_jdesc *jd;
 301
 302         spin_lock(&sdp->sd_jindex_spin);
 303         list_add(&list, &sdp->sd_jindex_list);
 304         list_del_init(&sdp->sd_jindex_list);
 305         sdp->sd_journals = 0;
 306         spin_unlock(&sdp->sd_jindex_spin);
 307
 308         while (!list_empty(&list)) {
 309                 jd = list_entry(list.next, struct gfs2_jdesc, jd_list);
 310                 gfs2_free_journal_extents(jd);
 311                 list_del(&jd->jd_list);
 312                 iput(jd->jd_inode);
 313                 kfree(jd);
 314         }
 315 }
 316
 317 static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
 318 {
 319         struct gfs2_jdesc *jd;
 320         int found = 0;
 321
 322         list_for_each_entry(jd, head, jd_list) {
 323                 if (jd->jd_jid == jid) {
 324                         found = 1;
 325                         break;
 326                 }
 327         }
 328
 329         if (!found)
 330                 jd = NULL;
 331
 332         return jd;
 333 }
 334
 335 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
 336 {
 337         struct gfs2_jdesc *jd;
 338
 339         spin_lock(&sdp->sd_jindex_spin);
 340         jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
 341         spin_unlock(&sdp->sd_jindex_spin);
 342
 343         return jd;
 344 }
 345
 346 int gfs2_jdesc_check(struct gfs2_jdesc *jd)
 347 {
 348         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 349         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 350         u64 size = i_size_read(jd->jd_inode);
 351
 352         if (gfs2_check_internal_file_size(jd->jd_inode, 8 << 20, 1 << 30))
 353                 return -EIO;
 354
 355         jd->jd_blocks = size >> sdp->sd_sb.sb_bsize_shift;
 356
 357         if (gfs2_write_alloc_required(ip, 0, size)) {
 358                 gfs2_consist_inode(ip);
 359                 return -EIO;
 360         }
 361
 362         return 0;
 363 }
 364
 365 static int init_threads(struct gfs2_sbd *sdp)
 366 {
 367         struct task_struct *p;
 368         int error = 0;
 369
 370         p = kthread_run(gfs2_logd, sdp, "gfs2_logd");
 371         if (IS_ERR(p)) {
 372                 error = PTR_ERR(p);
 373                 fs_err(sdp, "can't start logd thread: %d\n", error);
 374                 return error;
 375         }
 376         sdp->sd_logd_process = p;
 377
 378         p = kthread_run(gfs2_quotad, sdp, "gfs2_quotad");
 379         if (IS_ERR(p)) {
 380                 error = PTR_ERR(p);
 381                 fs_err(sdp, "can't start quotad thread: %d\n", error);
 382                 goto fail;
 383         }
 384         sdp->sd_quotad_process = p;
 385         return 0;
 386
 387 fail:
 388         kthread_stop(sdp->sd_logd_process);
 389         return error;
 390 }
 391
 392 /**
 393  * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
 394  * @sdp: the filesystem
 395  *
 396  * Returns: errno
 397  */
 398
 399 int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
 400 {
 401         struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
 402         struct gfs2_glock *j_gl = ip->i_gl;
 403         struct gfs2_holder freeze_gh;
 404         struct gfs2_log_header_host head;
 405         int error;
 406
 407         error = init_threads(sdp);
 408         if (error)
 409                 return error;
 410
 411         error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED, 0,
 412                                    &freeze_gh);
 413         if (error)
 414                 goto fail_threads;
 415
 416         j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
 417
 418         error = gfs2_find_jhead(sdp->sd_jdesc, &head);
 419         if (error)
 420                 goto fail;
 421
 422         if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 423                 gfs2_consist(sdp);
 424                 error = -EIO;
 425                 goto fail;
 426         }
 427
 428         /*  Initialize some head of the log stuff  */
 429         sdp->sd_log_sequence = head.lh_sequence + 1;
 430         gfs2_log_pointers_init(sdp, head.lh_blkno);
 431
 432         error = gfs2_quota_init(sdp);
 433         if (error)
 434                 goto fail;
 435
 436         set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 437
 438         gfs2_glock_dq_uninit(&freeze_gh);
 439
 440         return 0;
 441
 442 fail:
 443         freeze_gh.gh_flags |= GL_NOCACHE;
 444         gfs2_glock_dq_uninit(&freeze_gh);
 445 fail_threads:
 446         kthread_stop(sdp->sd_quotad_process);
 447         kthread_stop(sdp->sd_logd_process);
 448         return error;
 449 }
 450
 451 void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
 452 {
 453         const struct gfs2_statfs_change *str = buf;
 454
 455         sc->sc_total = be64_to_cpu(str->sc_total);
 456         sc->sc_free = be64_to_cpu(str->sc_free);
 457         sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
 458 }
 459
 460 static void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
 461 {
 462         struct gfs2_statfs_change *str = buf;
 463
 464         str->sc_total = cpu_to_be64(sc->sc_total);
 465         str->sc_free = cpu_to_be64(sc->sc_free);
 466         str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
 467 }
 468
 469 int gfs2_statfs_init(struct gfs2_sbd *sdp)
 470 {
 471         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 472         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 473         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 474         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 475         struct buffer_head *m_bh, *l_bh;
 476         struct gfs2_holder gh;
 477         int error;
 478
 479         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 480                                    &gh);
 481         if (error)
 482                 return error;
 483
 484         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 485         if (error)
 486                 goto out;
 487
 488         if (sdp->sd_args.ar_spectator) {
 489                 spin_lock(&sdp->sd_statfs_spin);
 490                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 491                                       sizeof(struct gfs2_dinode));
 492                 spin_unlock(&sdp->sd_statfs_spin);
 493         } else {
 494                 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 495                 if (error)
 496                         goto out_m_bh;
 497
 498                 spin_lock(&sdp->sd_statfs_spin);
 499                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 500                                       sizeof(struct gfs2_dinode));
 501                 gfs2_statfs_change_in(l_sc, l_bh->b_data +
 502                                       sizeof(struct gfs2_dinode));
 503                 spin_unlock(&sdp->sd_statfs_spin);
 504
 505                 brelse(l_bh);
 506         }
 507
 508 out_m_bh:
 509         brelse(m_bh);
 510 out:
 511         gfs2_glock_dq_uninit(&gh);
 512         return 0;
 513 }
 514
 515 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
 516                         s64 dinodes)
 517 {
 518         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 519         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 520         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 521         struct buffer_head *l_bh;
 522         s64 x, y;
 523         int need_sync = 0;
 524         int error;
 525
 526         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 527         if (error)
 528                 return;
 529
 530         gfs2_trans_add_meta(l_ip->i_gl, l_bh);
 531
 532         spin_lock(&sdp->sd_statfs_spin);
 533         l_sc->sc_total += total;
 534         l_sc->sc_free += free;
 535         l_sc->sc_dinodes += dinodes;
 536         gfs2_statfs_change_out(l_sc, l_bh->b_data + sizeof(struct gfs2_dinode));
 537         if (sdp->sd_args.ar_statfs_percent) {
 538                 x = 100 * l_sc->sc_free;
 539                 y = m_sc->sc_free * sdp->sd_args.ar_statfs_percent;
 540                 if (x >= y || x <= -y)
 541                         need_sync = 1;
 542         }
 543         spin_unlock(&sdp->sd_statfs_spin);
 544
 545         brelse(l_bh);
 546         if (need_sync)
 547                 gfs2_wake_up_statfs(sdp);
 548 }
 549
 550 void update_statfs(struct gfs2_sbd *sdp, struct buffer_head *m_bh,
 551                    struct buffer_head *l_bh)
 552 {
 553         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 554         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 555         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 556         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 557
 558         gfs2_trans_add_meta(l_ip->i_gl, l_bh);
 559
 560         spin_lock(&sdp->sd_statfs_spin);
 561         m_sc->sc_total += l_sc->sc_total;
 562         m_sc->sc_free += l_sc->sc_free;
 563         m_sc->sc_dinodes += l_sc->sc_dinodes;
 564         memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
 565         memset(l_bh->b_data + sizeof(struct gfs2_dinode),
 566                0, sizeof(struct gfs2_statfs_change));
 567         spin_unlock(&sdp->sd_statfs_spin);
 568
 569         gfs2_trans_add_meta(m_ip->i_gl, m_bh);
 570         gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
 571 }
 572
 573 int gfs2_statfs_sync(struct super_block *sb, int type)
 574 {
 575         struct gfs2_sbd *sdp = sb->s_fs_info;
 576         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 577         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 578         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 579         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 580         struct gfs2_holder gh;
 581         struct buffer_head *m_bh, *l_bh;
 582         int error;
 583
 584         sb_start_write(sb);
 585         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 586                                    &gh);
 587         if (error)
 588                 goto out;
 589
 590         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 591         if (error)
 592                 goto out_unlock;
 593
 594         spin_lock(&sdp->sd_statfs_spin);
 595         gfs2_statfs_change_in(m_sc, m_bh->b_data +
 596                               sizeof(struct gfs2_dinode));
 597         if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
 598                 spin_unlock(&sdp->sd_statfs_spin);
 599                 goto out_bh;
 600         }
 601         spin_unlock(&sdp->sd_statfs_spin);
 602
 603         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 604         if (error)
 605                 goto out_bh;
 606
 607         error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
 608         if (error)
 609                 goto out_bh2;
 610
 611         update_statfs(sdp, m_bh, l_bh);
 612         sdp->sd_statfs_force_sync = 0;
 613
 614         gfs2_trans_end(sdp);
 615
 616 out_bh2:
 617         brelse(l_bh);
 618 out_bh:
 619         brelse(m_bh);
 620 out_unlock:
 621         gfs2_glock_dq_uninit(&gh);
 622 out:
 623         sb_end_write(sb);
 624         return error;
 625 }
 626
 627 struct lfcc {
 628         struct list_head list;
 629         struct gfs2_holder gh;
 630 };
 631
 632 /**
 633  * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
 634  *                            journals are clean
 635  * @sdp: the file system
 636  * @state: the state to put the transaction lock into
 637  * @t_gh: the hold on the transaction lock
 638  *
 639  * Returns: errno
 640  */
 641
 642 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp,
 643                                     struct gfs2_holder *freeze_gh)
 644 {
 645         struct gfs2_inode *ip;
 646         struct gfs2_jdesc *jd;
 647         struct lfcc *lfcc;
 648         LIST_HEAD(list);
 649         struct gfs2_log_header_host lh;
 650         int error;
 651
 652         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 653                 lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
 654                 if (!lfcc) {
 655                         error = -ENOMEM;
 656                         goto out;
 657                 }
 658                 ip = GFS2_I(jd->jd_inode);
 659                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
 660                 if (error) {
 661                         kfree(lfcc);
 662                         goto out;
 663                 }
 664                 list_add(&lfcc->list, &list);
 665         }
 666
 667         error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_EXCLUSIVE,
 668                                    GL_NOCACHE, freeze_gh);
 669
 670         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 671                 error = gfs2_jdesc_check(jd);
 672                 if (error)
 673                         break;
 674                 error = gfs2_find_jhead(jd, &lh);
 675                 if (error)
 676                         break;
 677                 if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 678                         error = -EBUSY;
 679                         break;
 680                 }
 681         }
 682
 683         if (error)
 684                 gfs2_glock_dq_uninit(freeze_gh);
 685
 686 out:
 687         while (!list_empty(&list)) {
 688                 lfcc = list_entry(list.next, struct lfcc, list);
 689                 list_del(&lfcc->list);
 690                 gfs2_glock_dq_uninit(&lfcc->gh);
 691                 kfree(lfcc);
 692         }
 693         return error;
 694 }
 695
 696 void gfs2_dinode_out(const struct gfs2_inode *ip, void *buf)
 697 {
 698         struct gfs2_dinode *str = buf;
 699
 700         str->di_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 701         str->di_header.mh_type = cpu_to_be32(GFS2_METATYPE_DI);
 702         str->di_header.mh_format = cpu_to_be32(GFS2_FORMAT_DI);
 703         str->di_num.no_addr = cpu_to_be64(ip->i_no_addr);
 704         str->di_num.no_formal_ino = cpu_to_be64(ip->i_no_formal_ino);
 705         str->di_mode = cpu_to_be32(ip->i_inode.i_mode);
 706         str->di_uid = cpu_to_be32(i_uid_read(&ip->i_inode));
 707         str->di_gid = cpu_to_be32(i_gid_read(&ip->i_inode));
 708         str->di_nlink = cpu_to_be32(ip->i_inode.i_nlink);
 709         str->di_size = cpu_to_be64(i_size_read(&ip->i_inode));
 710         str->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(&ip->i_inode));
 711         str->di_atime = cpu_to_be64(ip->i_inode.i_atime.tv_sec);
 712         str->di_mtime = cpu_to_be64(ip->i_inode.i_mtime.tv_sec);
 713         str->di_ctime = cpu_to_be64(ip->i_inode.i_ctime.tv_sec);
 714
 715         str->di_goal_meta = cpu_to_be64(ip->i_goal);
 716         str->di_goal_data = cpu_to_be64(ip->i_goal);
 717         str->di_generation = cpu_to_be64(ip->i_generation);
 718
 719         str->di_flags = cpu_to_be32(ip->i_diskflags);
 720         str->di_height = cpu_to_be16(ip->i_height);
 721         str->di_payload_format = cpu_to_be32(S_ISDIR(ip->i_inode.i_mode) &&
 722                                              !(ip->i_diskflags & GFS2_DIF_EXHASH) ?
 723                                              GFS2_FORMAT_DE : 0);
 724         str->di_depth = cpu_to_be16(ip->i_depth);
 725         str->di_entries = cpu_to_be32(ip->i_entries);
 726
 727         str->di_eattr = cpu_to_be64(ip->i_eattr);
 728         str->di_atime_nsec = cpu_to_be32(ip->i_inode.i_atime.tv_nsec);
 729         str->di_mtime_nsec = cpu_to_be32(ip->i_inode.i_mtime.tv_nsec);
 730         str->di_ctime_nsec = cpu_to_be32(ip->i_inode.i_ctime.tv_nsec);
 731 }
 732
 733 /**
 734  * gfs2_write_inode - Make sure the inode is stable on the disk
 735  * @inode: The inode
 736  * @wbc: The writeback control structure
 737  *
 738  * Returns: errno
 739  */
 740
 741 static int gfs2_write_inode(struct inode *inode, struct writeback_control *wbc)
 742 {
 743         struct gfs2_inode *ip = GFS2_I(inode);
 744         struct gfs2_sbd *sdp = GFS2_SB(inode);
 745         struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl);
 746         struct backing_dev_info *bdi = inode_to_bdi(metamapping->host);
 747         int ret = 0;
 748
 749         if (wbc->sync_mode == WB_SYNC_ALL)
 750                 gfs2_log_flush(GFS2_SB(inode), ip->i_gl, NORMAL_FLUSH);
 751         if (bdi->wb.dirty_exceeded)
 752                 gfs2_ail1_flush(sdp, wbc);
 753         else
 754                 filemap_fdatawrite(metamapping);
 755         if (wbc->sync_mode == WB_SYNC_ALL)
 756                 ret = filemap_fdatawait(metamapping);
 757         if (ret)
 758                 mark_inode_dirty_sync(inode);
 759         return ret;
 760 }
 761
 762 /**
 763  * gfs2_dirty_inode - check for atime updates
 764  * @inode: The inode in question
 765  * @flags: The type of dirty
 766  *
 767  * Unfortunately it can be called under any combination of inode
 768  * glock and transaction lock, so we have to check carefully.
 769  *
 770  * At the moment this deals only with atime - it should be possible
 771  * to expand that role in future, once a review of the locking has
 772  * been carried out.
 773  */
 774
 775 static void gfs2_dirty_inode(struct inode *inode, int flags)
 776 {
 777         struct gfs2_inode *ip = GFS2_I(inode);
 778         struct gfs2_sbd *sdp = GFS2_SB(inode);
 779         struct buffer_head *bh;
 780         struct gfs2_holder gh;
 781         int need_unlock = 0;
 782         int need_endtrans = 0;
 783         int ret;
 784
 785         if (!(flags & (I_DIRTY_DATASYNC|I_DIRTY_SYNC)))
 786                 return;
 787
 788         if (!gfs2_glock_is_locked_by_me(ip->i_gl)) {
 789                 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
 790                 if (ret) {
 791                         fs_err(sdp, "dirty_inode: glock %d\n", ret);
 792                         return;
 793                 }
 794                 need_unlock = 1;
 795         } else if (WARN_ON_ONCE(ip->i_gl->gl_state != LM_ST_EXCLUSIVE))
 796                 return;
 797
 798         if (current->journal_info == NULL) {
 799                 ret = gfs2_trans_begin(sdp, RES_DINODE, 0);
 800                 if (ret) {
 801                         fs_err(sdp, "dirty_inode: gfs2_trans_begin %d\n", ret);
 802                         goto out;
 803                 }
 804                 need_endtrans = 1;
 805         }
 806
 807         ret = gfs2_meta_inode_buffer(ip, &bh);
 808         if (ret == 0) {
 809                 gfs2_trans_add_meta(ip->i_gl, bh);
 810                 gfs2_dinode_out(ip, bh->b_data);
 811                 brelse(bh);
 812         }
 813
 814         if (need_endtrans)
 815                 gfs2_trans_end(sdp);
 816 out:
 817         if (need_unlock)
 818                 gfs2_glock_dq_uninit(&gh);
 819 }
 820
 821 /**
 822  * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
 823  * @sdp: the filesystem
 824  *
 825  * Returns: errno
 826  */
 827
 828 static int gfs2_make_fs_ro(struct gfs2_sbd *sdp)
 829 {
 830         struct gfs2_holder freeze_gh;
 831         int error;
 832
 833         error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED, GL_NOCACHE,
 834                                    &freeze_gh);
 835         if (error && !test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
 836                 return error;
 837
 838         kthread_stop(sdp->sd_quotad_process);
 839         kthread_stop(sdp->sd_logd_process);
 840
 841         flush_workqueue(gfs2_delete_workqueue);
 842         gfs2_quota_sync(sdp->sd_vfs, 0);
 843         gfs2_statfs_sync(sdp->sd_vfs, 0);
 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         gfs2_log_flush(sdp, NULL, SHUTDOWN_FLUSH);
 850         wait_event(sdp->sd_reserving_log_wait, atomic_read(&sdp->sd_reserving_log) == 0);
 851         gfs2_assert_warn(sdp, atomic_read(&sdp->sd_log_blks_free) == sdp->sd_jdesc->jd_blocks);
 852
 853         if (freeze_gh.gh_gl)
 854                 gfs2_glock_dq_uninit(&freeze_gh);
 855
 856         gfs2_quota_cleanup(sdp);
 857
 858         return error;
 859 }
 860
 861 /**
 862  * gfs2_put_super - Unmount the filesystem
 863  * @sb: The VFS superblock
 864  *
 865  */
 866
 867 static void gfs2_put_super(struct super_block *sb)
 868 {
 869         struct gfs2_sbd *sdp = sb->s_fs_info;
 870         int error;
 871         struct gfs2_jdesc *jd;
 872
 873         /* No more recovery requests */
 874         set_bit(SDF_NORECOVERY, &sdp->sd_flags);
 875         smp_mb();
 876
 877         /* Wait on outstanding recovery */
 878 restart:
 879         spin_lock(&sdp->sd_jindex_spin);
 880         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 881                 if (!test_bit(JDF_RECOVERY, &jd->jd_flags))
 882                         continue;
 883                 spin_unlock(&sdp->sd_jindex_spin);
 884                 wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
 885                             TASK_UNINTERRUPTIBLE);
 886                 goto restart;
 887         }
 888         spin_unlock(&sdp->sd_jindex_spin);
 889
 890         if (!(sb->s_flags & MS_RDONLY)) {
 891                 error = gfs2_make_fs_ro(sdp);
 892                 if (error)
 893                         gfs2_io_error(sdp);
 894         }
 895         /*  At this point, we're through modifying the disk  */
 896
 897         /*  Release stuff  */
 898
 899         iput(sdp->sd_jindex);
 900         iput(sdp->sd_statfs_inode);
 901         iput(sdp->sd_rindex);
 902         iput(sdp->sd_quota_inode);
 903
 904         gfs2_glock_put(sdp->sd_rename_gl);
 905         gfs2_glock_put(sdp->sd_freeze_gl);
 906
 907         if (!sdp->sd_args.ar_spectator) {
 908                 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
 909                 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
 910                 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
 911                 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
 912                 iput(sdp->sd_sc_inode);
 913                 iput(sdp->sd_qc_inode);
 914         }
 915
 916         gfs2_glock_dq_uninit(&sdp->sd_live_gh);
 917         gfs2_clear_rgrpd(sdp);
 918         gfs2_jindex_free(sdp);
 919         /*  Take apart glock structures and buffer lists  */
 920         gfs2_gl_hash_clear(sdp);
 921         /*  Unmount the locking protocol  */
 922         gfs2_lm_unmount(sdp);
 923
 924         /*  At this point, we're through participating in the lockspace  */
 925         gfs2_sys_fs_del(sdp);
 926 }
 927
 928 /**
 929  * gfs2_sync_fs - sync the filesystem
 930  * @sb: the superblock
 931  *
 932  * Flushes the log to disk.
 933  */
 934
 935 static int gfs2_sync_fs(struct super_block *sb, int wait)
 936 {
 937         struct gfs2_sbd *sdp = sb->s_fs_info;
 938
 939         gfs2_quota_sync(sb, -1);
 940         if (wait && sdp)
 941                 gfs2_log_flush(sdp, NULL, NORMAL_FLUSH);
 942         return 0;
 943 }
 944
 945 void gfs2_freeze_func(struct work_struct *work)
 946 {
 947         int error;
 948         struct gfs2_holder freeze_gh;
 949         struct gfs2_sbd *sdp = container_of(work, struct gfs2_sbd, sd_freeze_work);
 950         struct super_block *sb = sdp->sd_vfs;
 951
 952         atomic_inc(&sb->s_active);
 953         error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED, 0,
 954                                    &freeze_gh);
 955         if (error) {
 956                 printk(KERN_INFO "GFS2: couln't get freeze lock : %d\n", error);
 957                 gfs2_assert_withdraw(sdp, 0);
 958         }
 959         else {
 960                 atomic_set(&sdp->sd_freeze_state, SFS_UNFROZEN);
 961                 error = thaw_super(sb);
 962                 if (error) {
 963                         printk(KERN_INFO "GFS2: couldn't thaw filesystem: %d\n",
 964                                error);
 965                         gfs2_assert_withdraw(sdp, 0);
 966                 }
 967                 if (!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))
 968                         freeze_gh.gh_flags |= GL_NOCACHE;
 969                 gfs2_glock_dq_uninit(&freeze_gh);
 970         }
 971         deactivate_super(sb);
 972         return;
 973 }
 974
 975 /**
 976  * gfs2_freeze - prevent further writes to the filesystem
 977  * @sb: the VFS structure for the filesystem
 978  *
 979  */
 980
 981 static int gfs2_freeze(struct super_block *sb)
 982 {
 983         struct gfs2_sbd *sdp = sb->s_fs_info;
 984         int error = 0;
 985
 986         mutex_lock(&sdp->sd_freeze_mutex);
 987         if (atomic_read(&sdp->sd_freeze_state) != SFS_UNFROZEN)
 988                 goto out;
 989
 990         if (test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) {
 991                 error = -EINVAL;
 992                 goto out;
 993         }
 994
 995         for (;;) {
 996                 error = gfs2_lock_fs_check_clean(sdp, &sdp->sd_freeze_gh);
 997                 if (!error)
 998                         break;
 999
1000                 switch (error) {
1001                 case -EBUSY:
1002                         fs_err(sdp, "waiting for recovery before freeze\n");
1003                         break;
1004
1005                 default:
1006                         fs_err(sdp, "error freezing FS: %d\n", error);
1007                         break;
1008                 }
1009
1010                 fs_err(sdp, "retrying...\n");
1011                 msleep(1000);
1012         }
1013         error = 0;
1014 out:
1015         mutex_unlock(&sdp->sd_freeze_mutex);
1016         return error;
1017 }
1018
1019 /**
1020  * gfs2_unfreeze - reallow writes to the filesystem
1021  * @sb: the VFS structure for the filesystem
1022  *
1023  */
1024
1025 static int gfs2_unfreeze(struct super_block *sb)
1026 {
1027         struct gfs2_sbd *sdp = sb->s_fs_info;
1028
1029         mutex_lock(&sdp->sd_freeze_mutex);
1030         if (atomic_read(&sdp->sd_freeze_state) != SFS_FROZEN ||
1031             sdp->sd_freeze_gh.gh_gl == NULL) {
1032                 mutex_unlock(&sdp->sd_freeze_mutex);
1033                 return 0;
1034         }
1035
1036         gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
1037         mutex_unlock(&sdp->sd_freeze_mutex);
1038         return 0;
1039 }
1040
1041 /**
1042  * statfs_fill - fill in the sg for a given RG
1043  * @rgd: the RG
1044  * @sc: the sc structure
1045  *
1046  * Returns: 0 on success, -ESTALE if the LVB is invalid
1047  */
1048
1049 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
1050                             struct gfs2_statfs_change_host *sc)
1051 {
1052         gfs2_rgrp_verify(rgd);
1053         sc->sc_total += rgd->rd_data;
1054         sc->sc_free += rgd->rd_free;
1055         sc->sc_dinodes += rgd->rd_dinodes;
1056         return 0;
1057 }
1058
1059 /**
1060  * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
1061  * @sdp: the filesystem
1062  * @sc: the sc info that will be returned
1063  *
1064  * Any error (other than a signal) will cause this routine to fall back
1065  * to the synchronous version.
1066  *
1067  * FIXME: This really shouldn't busy wait like this.
1068  *
1069  * Returns: errno
1070  */
1071
1072 static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
1073 {
1074         struct gfs2_rgrpd *rgd_next;
1075         struct gfs2_holder *gha, *gh;
1076         unsigned int slots = 64;
1077         unsigned int x;
1078         int done;
1079         int error = 0, err;
1080
1081         memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
1082         gha = kcalloc(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
1083         if (!gha)
1084                 return -ENOMEM;
1085
1086         rgd_next = gfs2_rgrpd_get_first(sdp);
1087
1088         for (;;) {
1089                 done = 1;
1090
1091                 for (x = 0; x < slots; x++) {
1092                         gh = gha + x;
1093
1094                         if (gh->gh_gl && gfs2_glock_poll(gh)) {
1095                                 err = gfs2_glock_wait(gh);
1096                                 if (err) {
1097                                         gfs2_holder_uninit(gh);
1098                                         error = err;
1099                                 } else {
1100                                         if (!error)
1101                                                 error = statfs_slow_fill(
1102                                                         gh->gh_gl->gl_object, sc);
1103                                         gfs2_glock_dq_uninit(gh);
1104                                 }
1105                         }
1106
1107                         if (gh->gh_gl)
1108                                 done = 0;
1109                         else if (rgd_next && !error) {
1110                                 error = gfs2_glock_nq_init(rgd_next->rd_gl,
1111                                                            LM_ST_SHARED,
1112                                                            GL_ASYNC,
1113                                                            gh);
1114                                 rgd_next = gfs2_rgrpd_get_next(rgd_next);
1115                                 done = 0;
1116                         }
1117
1118                         if (signal_pending(current))
1119                                 error = -ERESTARTSYS;
1120                 }
1121
1122                 if (done)
1123                         break;
1124
1125                 yield();
1126         }
1127
1128         kfree(gha);
1129         return error;
1130 }
1131
1132 /**
1133  * gfs2_statfs_i - Do a statfs
1134  * @sdp: the filesystem
1135  * @sg: the sg structure
1136  *
1137  * Returns: errno
1138  */
1139
1140 static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
1141 {
1142         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
1143         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
1144
1145         spin_lock(&sdp->sd_statfs_spin);
1146
1147         *sc = *m_sc;
1148         sc->sc_total += l_sc->sc_total;
1149         sc->sc_free += l_sc->sc_free;
1150         sc->sc_dinodes += l_sc->sc_dinodes;
1151
1152         spin_unlock(&sdp->sd_statfs_spin);
1153
1154         if (sc->sc_free < 0)
1155                 sc->sc_free = 0;
1156         if (sc->sc_free > sc->sc_total)
1157                 sc->sc_free = sc->sc_total;
1158         if (sc->sc_dinodes < 0)
1159                 sc->sc_dinodes = 0;
1160
1161         return 0;
1162 }
1163
1164 /**
1165  * gfs2_statfs - Gather and return stats about the filesystem
1166  * @sb: The superblock
1167  * @statfsbuf: The buffer
1168  *
1169  * Returns: 0 on success or error code
1170  */
1171
1172 static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
1173 {
1174         struct super_block *sb = d_inode(dentry)->i_sb;
1175         struct gfs2_sbd *sdp = sb->s_fs_info;
1176         struct gfs2_statfs_change_host sc;
1177         int error;
1178
1179         error = gfs2_rindex_update(sdp);
1180         if (error)
1181                 return error;
1182
1183         if (gfs2_tune_get(sdp, gt_statfs_slow))
1184                 error = gfs2_statfs_slow(sdp, &sc);
1185         else
1186                 error = gfs2_statfs_i(sdp, &sc);
1187
1188         if (error)
1189                 return error;
1190
1191         buf->f_type = GFS2_MAGIC;
1192         buf->f_bsize = sdp->sd_sb.sb_bsize;
1193         buf->f_blocks = sc.sc_total;
1194         buf->f_bfree = sc.sc_free;
1195         buf->f_bavail = sc.sc_free;
1196         buf->f_files = sc.sc_dinodes + sc.sc_free;
1197         buf->f_ffree = sc.sc_free;
1198         buf->f_namelen = GFS2_FNAMESIZE;
1199
1200         return 0;
1201 }
1202
1203 /**
1204  * gfs2_remount_fs - called when the FS is remounted
1205  * @sb:  the filesystem
1206  * @flags:  the remount flags
1207  * @data:  extra data passed in (not used right now)
1208  *
1209  * Returns: errno
1210  */
1211
1212 static int gfs2_remount_fs(struct super_block *sb, int *flags, char *data)
1213 {
1214         struct gfs2_sbd *sdp = sb->s_fs_info;
1215         struct gfs2_args args = sdp->sd_args; /* Default to current settings */
1216         struct gfs2_tune *gt = &sdp->sd_tune;
1217         int error;
1218
1219         sync_filesystem(sb);
1220
1221         spin_lock(&gt->gt_spin);
1222         args.ar_commit = gt->gt_logd_secs;
1223         args.ar_quota_quantum = gt->gt_quota_quantum;
1224         if (gt->gt_statfs_slow)
1225                 args.ar_statfs_quantum = 0;
1226         else
1227                 args.ar_statfs_quantum = gt->gt_statfs_quantum;
1228         spin_unlock(&gt->gt_spin);
1229         error = gfs2_mount_args(&args, data);
1230         if (error)
1231                 return error;
1232
1233         /* Not allowed to change locking details */
1234         if (strcmp(args.ar_lockproto, sdp->sd_args.ar_lockproto) ||
1235             strcmp(args.ar_locktable, sdp->sd_args.ar_locktable) ||
1236             strcmp(args.ar_hostdata, sdp->sd_args.ar_hostdata))
1237                 return -EINVAL;
1238
1239         /* Some flags must not be changed */
1240         if (args_neq(&args, &sdp->sd_args, spectator) ||
1241             args_neq(&args, &sdp->sd_args, localflocks) ||
1242             args_neq(&args, &sdp->sd_args, meta))
1243                 return -EINVAL;
1244
1245         if (sdp->sd_args.ar_spectator)
1246                 *flags |= MS_RDONLY;
1247
1248         if ((sb->s_flags ^ *flags) & MS_RDONLY) {
1249                 if (*flags & MS_RDONLY)
1250                         error = gfs2_make_fs_ro(sdp);
1251                 else
1252                         error = gfs2_make_fs_rw(sdp);
1253                 if (error)
1254                         return error;
1255         }
1256
1257         sdp->sd_args = args;
1258         if (sdp->sd_args.ar_posix_acl)
1259                 sb->s_flags |= MS_POSIXACL;
1260         else
1261                 sb->s_flags &= ~MS_POSIXACL;
1262         if (sdp->sd_args.ar_nobarrier)
1263                 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1264         else
1265                 clear_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1266         spin_lock(&gt->gt_spin);
1267         gt->gt_logd_secs = args.ar_commit;
1268         gt->gt_quota_quantum = args.ar_quota_quantum;
1269         if (args.ar_statfs_quantum) {
1270                 gt->gt_statfs_slow = 0;
1271                 gt->gt_statfs_quantum = args.ar_statfs_quantum;
1272         }
1273         else {
1274                 gt->gt_statfs_slow = 1;
1275                 gt->gt_statfs_quantum = 30;
1276         }
1277         spin_unlock(&gt->gt_spin);
1278
1279         gfs2_online_uevent(sdp);
1280         return 0;
1281 }
1282
1283 /**
1284  * gfs2_drop_inode - Drop an inode (test for remote unlink)
1285  * @inode: The inode to drop
1286  *
1287  * If we've received a callback on an iopen lock then its because a
1288  * remote node tried to deallocate the inode but failed due to this node
1289  * still having the inode open. Here we mark the link count zero
1290  * since we know that it must have reached zero if the GLF_DEMOTE flag
1291  * is set on the iopen glock. If we didn't do a disk read since the
1292  * remote node removed the final link then we might otherwise miss
1293  * this event. This check ensures that this node will deallocate the
1294  * inode's blocks, or alternatively pass the baton on to another
1295  * node for later deallocation.
1296  */
1297
1298 static int gfs2_drop_inode(struct inode *inode)
1299 {
1300         struct gfs2_inode *ip = GFS2_I(inode);
1301
1302         if (!test_bit(GIF_FREE_VFS_INODE, &ip->i_flags) && inode->i_nlink) {
1303                 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1304                 if (gl && test_bit(GLF_DEMOTE, &gl->gl_flags))
1305                         clear_nlink(inode);
1306         }
1307         return generic_drop_inode(inode);
1308 }
1309
1310 static int is_ancestor(const struct dentry *d1, const struct dentry *d2)
1311 {
1312         do {
1313                 if (d1 == d2)
1314                         return 1;
1315                 d1 = d1->d_parent;
1316         } while (!IS_ROOT(d1));
1317         return 0;
1318 }
1319
1320 /**
1321  * gfs2_show_options - Show mount options for /proc/mounts
1322  * @s: seq_file structure
1323  * @root: root of this (sub)tree
1324  *
1325  * Returns: 0 on success or error code
1326  */
1327
1328 static int gfs2_show_options(struct seq_file *s, struct dentry *root)
1329 {
1330         struct gfs2_sbd *sdp = root->d_sb->s_fs_info;
1331         struct gfs2_args *args = &sdp->sd_args;
1332         int val;
1333
1334         if (is_ancestor(root, sdp->sd_master_dir))
1335                 seq_puts(s, ",meta");
1336         if (args->ar_lockproto[0])
1337                 seq_show_option(s, "lockproto", args->ar_lockproto);
1338         if (args->ar_locktable[0])
1339                 seq_show_option(s, "locktable", args->ar_locktable);
1340         if (args->ar_hostdata[0])
1341                 seq_show_option(s, "hostdata", args->ar_hostdata);
1342         if (args->ar_spectator)
1343                 seq_puts(s, ",spectator");
1344         if (args->ar_localflocks)
1345                 seq_puts(s, ",localflocks");
1346         if (args->ar_debug)
1347                 seq_puts(s, ",debug");
1348         if (args->ar_posix_acl)
1349                 seq_puts(s, ",acl");
1350         if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
1351                 char *state;
1352                 switch (args->ar_quota) {
1353                 case GFS2_QUOTA_OFF:
1354                         state = "off";
1355                         break;
1356                 case GFS2_QUOTA_ACCOUNT:
1357                         state = "account";
1358                         break;
1359                 case GFS2_QUOTA_ON:
1360                         state = "on";
1361                         break;
1362                 default:
1363                         state = "unknown";
1364                         break;
1365                 }
1366                 seq_printf(s, ",quota=%s", state);
1367         }
1368         if (args->ar_suiddir)
1369                 seq_puts(s, ",suiddir");
1370         if (args->ar_data != GFS2_DATA_DEFAULT) {
1371                 char *state;
1372                 switch (args->ar_data) {
1373                 case GFS2_DATA_WRITEBACK:
1374                         state = "writeback";
1375                         break;
1376                 case GFS2_DATA_ORDERED:
1377                         state = "ordered";
1378                         break;
1379                 default:
1380                         state = "unknown";
1381                         break;
1382                 }
1383                 seq_printf(s, ",data=%s", state);
1384         }
1385         if (args->ar_discard)
1386                 seq_puts(s, ",discard");
1387         val = sdp->sd_tune.gt_logd_secs;
1388         if (val != 30)
1389                 seq_printf(s, ",commit=%d", val);
1390         val = sdp->sd_tune.gt_statfs_quantum;
1391         if (val != 30)
1392                 seq_printf(s, ",statfs_quantum=%d", val);
1393         else if (sdp->sd_tune.gt_statfs_slow)
1394                 seq_puts(s, ",statfs_quantum=0");
1395         val = sdp->sd_tune.gt_quota_quantum;
1396         if (val != 60)
1397                 seq_printf(s, ",quota_quantum=%d", val);
1398         if (args->ar_statfs_percent)
1399                 seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent);
1400         if (args->ar_errors != GFS2_ERRORS_DEFAULT) {
1401                 const char *state;
1402
1403                 switch (args->ar_errors) {
1404                 case GFS2_ERRORS_WITHDRAW:
1405                         state = "withdraw";
1406                         break;
1407                 case GFS2_ERRORS_PANIC:
1408                         state = "panic";
1409                         break;
1410                 default:
1411                         state = "unknown";
1412                         break;
1413                 }
1414                 seq_printf(s, ",errors=%s", state);
1415         }
1416         if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags))
1417                 seq_puts(s, ",nobarrier");
1418         if (test_bit(SDF_DEMOTE, &sdp->sd_flags))
1419                 seq_puts(s, ",demote_interface_used");
1420         if (args->ar_rgrplvb)
1421                 seq_puts(s, ",rgrplvb");
1422         return 0;
1423 }
1424
1425 static void gfs2_final_release_pages(struct gfs2_inode *ip)
1426 {
1427         struct inode *inode = &ip->i_inode;
1428         struct gfs2_glock *gl = ip->i_gl;
1429
1430         truncate_inode_pages(gfs2_glock2aspace(ip->i_gl), 0);
1431         truncate_inode_pages(&inode->i_data, 0);
1432
1433         if (atomic_read(&gl->gl_revokes) == 0) {
1434                 clear_bit(GLF_LFLUSH, &gl->gl_flags);
1435                 clear_bit(GLF_DIRTY, &gl->gl_flags);
1436         }
1437 }
1438
1439 static int gfs2_dinode_dealloc(struct gfs2_inode *ip)
1440 {
1441         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1442         struct gfs2_rgrpd *rgd;
1443         struct gfs2_holder gh;
1444         int error;
1445
1446         if (gfs2_get_inode_blocks(&ip->i_inode) != 1) {
1447                 gfs2_consist_inode(ip);
1448                 return -EIO;
1449         }
1450
1451         error = gfs2_rindex_update(sdp);
1452         if (error)
1453                 return error;
1454
1455         error = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
1456         if (error)
1457                 return error;
1458
1459         rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1460         if (!rgd) {
1461                 gfs2_consist_inode(ip);
1462                 error = -EIO;
1463                 goto out_qs;
1464         }
1465
1466         error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &gh);
1467         if (error)
1468                 goto out_qs;
1469
1470         error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA,
1471                                  sdp->sd_jdesc->jd_blocks);
1472         if (error)
1473                 goto out_rg_gunlock;
1474
1475         gfs2_free_di(rgd, ip);
1476
1477         gfs2_final_release_pages(ip);
1478
1479         gfs2_trans_end(sdp);
1480
1481 out_rg_gunlock:
1482         gfs2_glock_dq_uninit(&gh);
1483 out_qs:
1484         gfs2_quota_unhold(ip);
1485         return error;
1486 }
1487
1488 /**
1489  * gfs2_evict_inode - Remove an inode from cache
1490  * @inode: The inode to evict
1491  *
1492  * There are three cases to consider:
1493  * 1. i_nlink == 0, we are final opener (and must deallocate)
1494  * 2. i_nlink == 0, we are not the final opener (and cannot deallocate)
1495  * 3. i_nlink > 0
1496  *
1497  * If the fs is read only, then we have to treat all cases as per #3
1498  * since we are unable to do any deallocation. The inode will be
1499  * deallocated by the next read/write node to attempt an allocation
1500  * in the same resource group
1501  *
1502  * We have to (at the moment) hold the inodes main lock to cover
1503  * the gap between unlocking the shared lock on the iopen lock and
1504  * taking the exclusive lock. I'd rather do a shared -> exclusive
1505  * conversion on the iopen lock, but we can change that later. This
1506  * is safe, just less efficient.
1507  */
1508
1509 static void gfs2_evict_inode(struct inode *inode)
1510 {
1511         struct super_block *sb = inode->i_sb;
1512         struct gfs2_sbd *sdp = sb->s_fs_info;
1513         struct gfs2_inode *ip = GFS2_I(inode);
1514         struct gfs2_holder gh;
1515         int error;
1516
1517         if (test_bit(GIF_FREE_VFS_INODE, &ip->i_flags)) {
1518                 clear_inode(inode);
1519                 return;
1520         }
1521
1522         if (inode->i_nlink || (sb->s_flags & MS_RDONLY))
1523                 goto out;
1524
1525         /* Must not read inode block until block type has been verified */
1526         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, &gh);
1527         if (unlikely(error)) {
1528                 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1529                 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1530                 goto out;
1531         }
1532
1533         if (!test_bit(GIF_ALLOC_FAILED, &ip->i_flags)) {
1534                 error = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED);
1535                 if (error)
1536                         goto out_truncate;
1537         }
1538
1539         if (test_bit(GIF_INVALID, &ip->i_flags)) {
1540                 error = gfs2_inode_refresh(ip);
1541                 if (error)
1542                         goto out_truncate;
1543         }
1544
1545         ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1546         gfs2_glock_dq_wait(&ip->i_iopen_gh);
1547         gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE, &ip->i_iopen_gh);
1548         error = gfs2_glock_nq(&ip->i_iopen_gh);
1549         if (error)
1550                 goto out_truncate;
1551
1552         /* Case 1 starts here */
1553
1554         if (S_ISDIR(inode->i_mode) &&
1555             (ip->i_diskflags & GFS2_DIF_EXHASH)) {
1556                 error = gfs2_dir_exhash_dealloc(ip);
1557                 if (error)
1558                         goto out_unlock;
1559         }
1560
1561         if (ip->i_eattr) {
1562                 error = gfs2_ea_dealloc(ip);
1563                 if (error)
1564                         goto out_unlock;
1565         }
1566
1567         if (!gfs2_is_stuffed(ip)) {
1568                 error = gfs2_file_dealloc(ip);
1569                 if (error)
1570                         goto out_unlock;
1571         }
1572
1573         error = gfs2_dinode_dealloc(ip);
1574         goto out_unlock;
1575
1576 out_truncate:
1577         gfs2_log_flush(sdp, ip->i_gl, NORMAL_FLUSH);
1578         if (test_bit(GLF_DIRTY, &ip->i_gl->gl_flags)) {
1579                 struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl);
1580                 filemap_fdatawrite(metamapping);
1581                 filemap_fdatawait(metamapping);
1582         }
1583         write_inode_now(inode, 1);
1584         gfs2_ail_flush(ip->i_gl, 0);
1585
1586         /* Case 2 starts here */
1587         error = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
1588         if (error)
1589                 goto out_unlock;
1590         /* Needs to be done before glock release & also in a transaction */
1591         truncate_inode_pages(&inode->i_data, 0);
1592         gfs2_trans_end(sdp);
1593
1594 out_unlock:
1595         /* Error path for case 1 */
1596         if (gfs2_rs_active(ip->i_res))
1597                 gfs2_rs_deltree(ip->i_res);
1598
1599         if (test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1600                 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1601                 gfs2_glock_dq(&ip->i_iopen_gh);
1602         }
1603         gfs2_holder_uninit(&ip->i_iopen_gh);
1604         gfs2_glock_dq_uninit(&gh);
1605         if (error && error != GLR_TRYFAILED && error != -EROFS)
1606                 fs_warn(sdp, "gfs2_evict_inode: %d\n", error);
1607 out:
1608         /* Case 3 starts here */
1609         truncate_inode_pages_final(&inode->i_data);
1610         gfs2_rs_delete(ip, NULL);
1611         gfs2_ordered_del_inode(ip);
1612         clear_inode(inode);
1613         gfs2_dir_hash_inval(ip);
1614         ip->i_gl->gl_object = NULL;
1615         flush_delayed_work(&ip->i_gl->gl_work);
1616         gfs2_glock_add_to_lru(ip->i_gl);
1617         gfs2_glock_put(ip->i_gl);
1618         ip->i_gl = NULL;
1619         if (ip->i_iopen_gh.gh_gl) {
1620                 ip->i_iopen_gh.gh_gl->gl_object = NULL;
1621                 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1622                 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1623         }
1624 }
1625
1626 static struct inode *gfs2_alloc_inode(struct super_block *sb)
1627 {
1628         struct gfs2_inode *ip;
1629
1630         ip = kmem_cache_alloc(gfs2_inode_cachep, GFP_KERNEL);
1631         if (ip) {
1632                 ip->i_flags = 0;
1633                 ip->i_gl = NULL;
1634                 ip->i_rgd = NULL;
1635                 ip->i_res = NULL;
1636                 ip->i_rahead = 0;
1637         }
1638         return &ip->i_inode;
1639 }
1640
1641 static void gfs2_i_callback(struct rcu_head *head)
1642 {
1643         struct inode *inode = container_of(head, struct inode, i_rcu);
1644         kmem_cache_free(gfs2_inode_cachep, inode);
1645 }
1646
1647 static void gfs2_destroy_inode(struct inode *inode)
1648 {
1649         call_rcu(&inode->i_rcu, gfs2_i_callback);
1650 }
1651
1652 const struct super_operations gfs2_super_ops = {
1653         .alloc_inode            = gfs2_alloc_inode,
1654         .destroy_inode          = gfs2_destroy_inode,
1655         .write_inode            = gfs2_write_inode,
1656         .dirty_inode            = gfs2_dirty_inode,
1657         .evict_inode            = gfs2_evict_inode,
1658         .put_super              = gfs2_put_super,
1659         .sync_fs                = gfs2_sync_fs,
1660         .freeze_super           = gfs2_freeze,
1661         .thaw_super             = gfs2_unfreeze,
1662         .statfs                 = gfs2_statfs,
1663         .remount_fs             = gfs2_remount_fs,
1664         .drop_inode             = gfs2_drop_inode,
1665         .show_options           = gfs2_show_options,
1666 };
1667