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/signal.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_loccookie,
  87         Opt_noloccookie,
  88         Opt_error,
  89 };
  90
  91 static const match_table_t tokens = {
  92         {Opt_lockproto, "lockproto=%s"},
  93         {Opt_locktable, "locktable=%s"},
  94         {Opt_hostdata, "hostdata=%s"},
  95         {Opt_spectator, "spectator"},
  96         {Opt_spectator, "norecovery"},
  97         {Opt_ignore_local_fs, "ignore_local_fs"},
  98         {Opt_localflocks, "localflocks"},
  99         {Opt_localcaching, "localcaching"},
 100         {Opt_debug, "debug"},
 101         {Opt_nodebug, "nodebug"},
 102         {Opt_upgrade, "upgrade"},
 103         {Opt_acl, "acl"},
 104         {Opt_noacl, "noacl"},
 105         {Opt_quota_off, "quota=off"},
 106         {Opt_quota_account, "quota=account"},
 107         {Opt_quota_on, "quota=on"},
 108         {Opt_quota, "quota"},
 109         {Opt_noquota, "noquota"},
 110         {Opt_suiddir, "suiddir"},
 111         {Opt_nosuiddir, "nosuiddir"},
 112         {Opt_data_writeback, "data=writeback"},
 113         {Opt_data_ordered, "data=ordered"},
 114         {Opt_meta, "meta"},
 115         {Opt_discard, "discard"},
 116         {Opt_nodiscard, "nodiscard"},
 117         {Opt_commit, "commit=%d"},
 118         {Opt_err_withdraw, "errors=withdraw"},
 119         {Opt_err_panic, "errors=panic"},
 120         {Opt_statfs_quantum, "statfs_quantum=%d"},
 121         {Opt_statfs_percent, "statfs_percent=%d"},
 122         {Opt_quota_quantum, "quota_quantum=%d"},
 123         {Opt_barrier, "barrier"},
 124         {Opt_nobarrier, "nobarrier"},
 125         {Opt_rgrplvb, "rgrplvb"},
 126         {Opt_norgrplvb, "norgrplvb"},
 127         {Opt_loccookie, "loccookie"},
 128         {Opt_noloccookie, "noloccookie"},
 129         {Opt_error, NULL}
 130 };
 131
 132 /**
 133  * gfs2_mount_args - Parse mount options
 134  * @args: The structure into which the parsed options will be written
 135  * @options: The options to parse
 136  *
 137  * Return: errno
 138  */
 139
 140 int gfs2_mount_args(struct gfs2_args *args, char *options)
 141 {
 142         char *o;
 143         int token;
 144         substring_t tmp[MAX_OPT_ARGS];
 145         int rv;
 146
 147         /* Split the options into tokens with the "," character and
 148            process them */
 149
 150         while (1) {
 151                 o = strsep(&options, ",");
 152                 if (o == NULL)
 153                         break;
 154                 if (*o == '\0')
 155                         continue;
 156
 157                 token = match_token(o, tokens, tmp);
 158                 switch (token) {
 159                 case Opt_lockproto:
 160                         match_strlcpy(args->ar_lockproto, &tmp[0],
 161                                       GFS2_LOCKNAME_LEN);
 162                         break;
 163                 case Opt_locktable:
 164                         match_strlcpy(args->ar_locktable, &tmp[0],
 165                                       GFS2_LOCKNAME_LEN);
 166                         break;
 167                 case Opt_hostdata:
 168                         match_strlcpy(args->ar_hostdata, &tmp[0],
 169                                       GFS2_LOCKNAME_LEN);
 170                         break;
 171                 case Opt_spectator:
 172                         args->ar_spectator = 1;
 173                         break;
 174                 case Opt_ignore_local_fs:
 175                         /* Retained for backwards compat only */
 176                         break;
 177                 case Opt_localflocks:
 178                         args->ar_localflocks = 1;
 179                         break;
 180                 case Opt_localcaching:
 181                         /* Retained for backwards compat only */
 182                         break;
 183                 case Opt_debug:
 184                         if (args->ar_errors == GFS2_ERRORS_PANIC) {
 185                                 pr_warn("-o debug and -o errors=panic are mutually exclusive\n");
 186                                 return -EINVAL;
 187                         }
 188                         args->ar_debug = 1;
 189                         break;
 190                 case Opt_nodebug:
 191                         args->ar_debug = 0;
 192                         break;
 193                 case Opt_upgrade:
 194                         /* Retained for backwards compat only */
 195                         break;
 196                 case Opt_acl:
 197                         args->ar_posix_acl = 1;
 198                         break;
 199                 case Opt_noacl:
 200                         args->ar_posix_acl = 0;
 201                         break;
 202                 case Opt_quota_off:
 203                 case Opt_noquota:
 204                         args->ar_quota = GFS2_QUOTA_OFF;
 205                         break;
 206                 case Opt_quota_account:
 207                         args->ar_quota = GFS2_QUOTA_ACCOUNT;
 208                         break;
 209                 case Opt_quota_on:
 210                 case Opt_quota:
 211                         args->ar_quota = GFS2_QUOTA_ON;
 212                         break;
 213                 case Opt_suiddir:
 214                         args->ar_suiddir = 1;
 215                         break;
 216                 case Opt_nosuiddir:
 217                         args->ar_suiddir = 0;
 218                         break;
 219                 case Opt_data_writeback:
 220                         args->ar_data = GFS2_DATA_WRITEBACK;
 221                         break;
 222                 case Opt_data_ordered:
 223                         args->ar_data = GFS2_DATA_ORDERED;
 224                         break;
 225                 case Opt_meta:
 226                         args->ar_meta = 1;
 227                         break;
 228                 case Opt_discard:
 229                         args->ar_discard = 1;
 230                         break;
 231                 case Opt_nodiscard:
 232                         args->ar_discard = 0;
 233                         break;
 234                 case Opt_commit:
 235                         rv = match_int(&tmp[0], &args->ar_commit);
 236                         if (rv || args->ar_commit <= 0) {
 237                                 pr_warn("commit mount option requires a positive numeric argument\n");
 238                                 return rv ? rv : -EINVAL;
 239                         }
 240                         break;
 241                 case Opt_statfs_quantum:
 242                         rv = match_int(&tmp[0], &args->ar_statfs_quantum);
 243                         if (rv || args->ar_statfs_quantum < 0) {
 244                                 pr_warn("statfs_quantum mount option requires a non-negative numeric argument\n");
 245                                 return rv ? rv : -EINVAL;
 246                         }
 247                         break;
 248                 case Opt_quota_quantum:
 249                         rv = match_int(&tmp[0], &args->ar_quota_quantum);
 250                         if (rv || args->ar_quota_quantum <= 0) {
 251                                 pr_warn("quota_quantum mount option requires a positive numeric argument\n");
 252                                 return rv ? rv : -EINVAL;
 253                         }
 254                         break;
 255                 case Opt_statfs_percent:
 256                         rv = match_int(&tmp[0], &args->ar_statfs_percent);
 257                         if (rv || args->ar_statfs_percent < 0 ||
 258                             args->ar_statfs_percent > 100) {
 259                                 pr_warn("statfs_percent mount option requires a numeric argument between 0 and 100\n");
 260                                 return rv ? rv : -EINVAL;
 261                         }
 262                         break;
 263                 case Opt_err_withdraw:
 264                         args->ar_errors = GFS2_ERRORS_WITHDRAW;
 265                         break;
 266                 case Opt_err_panic:
 267                         if (args->ar_debug) {
 268                                 pr_warn("-o debug and -o errors=panic are mutually exclusive\n");
 269                                 return -EINVAL;
 270                         }
 271                         args->ar_errors = GFS2_ERRORS_PANIC;
 272                         break;
 273                 case Opt_barrier:
 274                         args->ar_nobarrier = 0;
 275                         break;
 276                 case Opt_nobarrier:
 277                         args->ar_nobarrier = 1;
 278                         break;
 279                 case Opt_rgrplvb:
 280                         args->ar_rgrplvb = 1;
 281                         break;
 282                 case Opt_norgrplvb:
 283                         args->ar_rgrplvb = 0;
 284                         break;
 285                 case Opt_loccookie:
 286                         args->ar_loccookie = 1;
 287                         break;
 288                 case Opt_noloccookie:
 289                         args->ar_loccookie = 0;
 290                         break;
 291                 case Opt_error:
 292                 default:
 293                         pr_warn("invalid mount option: %s\n", o);
 294                         return -EINVAL;
 295                 }
 296         }
 297
 298         return 0;
 299 }
 300
 301 /**
 302  * gfs2_jindex_free - Clear all the journal index information
 303  * @sdp: The GFS2 superblock
 304  *
 305  */
 306
 307 void gfs2_jindex_free(struct gfs2_sbd *sdp)
 308 {
 309         struct list_head list;
 310         struct gfs2_jdesc *jd;
 311
 312         spin_lock(&sdp->sd_jindex_spin);
 313         list_add(&list, &sdp->sd_jindex_list);
 314         list_del_init(&sdp->sd_jindex_list);
 315         sdp->sd_journals = 0;
 316         spin_unlock(&sdp->sd_jindex_spin);
 317
 318         while (!list_empty(&list)) {
 319                 jd = list_entry(list.next, struct gfs2_jdesc, jd_list);
 320                 gfs2_free_journal_extents(jd);
 321                 list_del(&jd->jd_list);
 322                 iput(jd->jd_inode);
 323                 kfree(jd);
 324         }
 325 }
 326
 327 static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
 328 {
 329         struct gfs2_jdesc *jd;
 330         int found = 0;
 331
 332         list_for_each_entry(jd, head, jd_list) {
 333                 if (jd->jd_jid == jid) {
 334                         found = 1;
 335                         break;
 336                 }
 337         }
 338
 339         if (!found)
 340                 jd = NULL;
 341
 342         return jd;
 343 }
 344
 345 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
 346 {
 347         struct gfs2_jdesc *jd;
 348
 349         spin_lock(&sdp->sd_jindex_spin);
 350         jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
 351         spin_unlock(&sdp->sd_jindex_spin);
 352
 353         return jd;
 354 }
 355
 356 int gfs2_jdesc_check(struct gfs2_jdesc *jd)
 357 {
 358         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 359         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 360         u64 size = i_size_read(jd->jd_inode);
 361
 362         if (gfs2_check_internal_file_size(jd->jd_inode, 8 << 20, BIT(30)))
 363                 return -EIO;
 364
 365         jd->jd_blocks = size >> sdp->sd_sb.sb_bsize_shift;
 366
 367         if (gfs2_write_alloc_required(ip, 0, size)) {
 368                 gfs2_consist_inode(ip);
 369                 return -EIO;
 370         }
 371
 372         return 0;
 373 }
 374
 375 static int init_threads(struct gfs2_sbd *sdp)
 376 {
 377         struct task_struct *p;
 378         int error = 0;
 379
 380         p = kthread_run(gfs2_logd, sdp, "gfs2_logd");
 381         if (IS_ERR(p)) {
 382                 error = PTR_ERR(p);
 383                 fs_err(sdp, "can't start logd thread: %d\n", error);
 384                 return error;
 385         }
 386         sdp->sd_logd_process = p;
 387
 388         p = kthread_run(gfs2_quotad, sdp, "gfs2_quotad");
 389         if (IS_ERR(p)) {
 390                 error = PTR_ERR(p);
 391                 fs_err(sdp, "can't start quotad thread: %d\n", error);
 392                 goto fail;
 393         }
 394         sdp->sd_quotad_process = p;
 395         return 0;
 396
 397 fail:
 398         kthread_stop(sdp->sd_logd_process);
 399         return error;
 400 }
 401
 402 /**
 403  * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
 404  * @sdp: the filesystem
 405  *
 406  * Returns: errno
 407  */
 408
 409 int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
 410 {
 411         struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
 412         struct gfs2_glock *j_gl = ip->i_gl;
 413         struct gfs2_holder freeze_gh;
 414         struct gfs2_log_header_host head;
 415         int error;
 416
 417         error = init_threads(sdp);
 418         if (error)
 419                 return error;
 420
 421         error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED, 0,
 422                                    &freeze_gh);
 423         if (error)
 424                 goto fail_threads;
 425
 426         j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
 427
 428         error = gfs2_find_jhead(sdp->sd_jdesc, &head);
 429         if (error)
 430                 goto fail;
 431
 432         if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 433                 gfs2_consist(sdp);
 434                 error = -EIO;
 435                 goto fail;
 436         }
 437
 438         /*  Initialize some head of the log stuff  */
 439         sdp->sd_log_sequence = head.lh_sequence + 1;
 440         gfs2_log_pointers_init(sdp, head.lh_blkno);
 441
 442         error = gfs2_quota_init(sdp);
 443         if (error)
 444                 goto fail;
 445
 446         set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 447
 448         gfs2_glock_dq_uninit(&freeze_gh);
 449
 450         return 0;
 451
 452 fail:
 453         freeze_gh.gh_flags |= GL_NOCACHE;
 454         gfs2_glock_dq_uninit(&freeze_gh);
 455 fail_threads:
 456         kthread_stop(sdp->sd_quotad_process);
 457         kthread_stop(sdp->sd_logd_process);
 458         return error;
 459 }
 460
 461 void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
 462 {
 463         const struct gfs2_statfs_change *str = buf;
 464
 465         sc->sc_total = be64_to_cpu(str->sc_total);
 466         sc->sc_free = be64_to_cpu(str->sc_free);
 467         sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
 468 }
 469
 470 static void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
 471 {
 472         struct gfs2_statfs_change *str = buf;
 473
 474         str->sc_total = cpu_to_be64(sc->sc_total);
 475         str->sc_free = cpu_to_be64(sc->sc_free);
 476         str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
 477 }
 478
 479 int gfs2_statfs_init(struct gfs2_sbd *sdp)
 480 {
 481         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 482         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 483         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 484         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 485         struct buffer_head *m_bh, *l_bh;
 486         struct gfs2_holder gh;
 487         int error;
 488
 489         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 490                                    &gh);
 491         if (error)
 492                 return error;
 493
 494         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 495         if (error)
 496                 goto out;
 497
 498         if (sdp->sd_args.ar_spectator) {
 499                 spin_lock(&sdp->sd_statfs_spin);
 500                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 501                                       sizeof(struct gfs2_dinode));
 502                 spin_unlock(&sdp->sd_statfs_spin);
 503         } else {
 504                 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 505                 if (error)
 506                         goto out_m_bh;
 507
 508                 spin_lock(&sdp->sd_statfs_spin);
 509                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 510                                       sizeof(struct gfs2_dinode));
 511                 gfs2_statfs_change_in(l_sc, l_bh->b_data +
 512                                       sizeof(struct gfs2_dinode));
 513                 spin_unlock(&sdp->sd_statfs_spin);
 514
 515                 brelse(l_bh);
 516         }
 517
 518 out_m_bh:
 519         brelse(m_bh);
 520 out:
 521         gfs2_glock_dq_uninit(&gh);
 522         return 0;
 523 }
 524
 525 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
 526                         s64 dinodes)
 527 {
 528         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 529         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 530         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 531         struct buffer_head *l_bh;
 532         s64 x, y;
 533         int need_sync = 0;
 534         int error;
 535
 536         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 537         if (error)
 538                 return;
 539
 540         gfs2_trans_add_meta(l_ip->i_gl, l_bh);
 541
 542         spin_lock(&sdp->sd_statfs_spin);
 543         l_sc->sc_total += total;
 544         l_sc->sc_free += free;
 545         l_sc->sc_dinodes += dinodes;
 546         gfs2_statfs_change_out(l_sc, l_bh->b_data + sizeof(struct gfs2_dinode));
 547         if (sdp->sd_args.ar_statfs_percent) {
 548                 x = 100 * l_sc->sc_free;
 549                 y = m_sc->sc_free * sdp->sd_args.ar_statfs_percent;
 550                 if (x >= y || x <= -y)
 551                         need_sync = 1;
 552         }
 553         spin_unlock(&sdp->sd_statfs_spin);
 554
 555         brelse(l_bh);
 556         if (need_sync)
 557                 gfs2_wake_up_statfs(sdp);
 558 }
 559
 560 void update_statfs(struct gfs2_sbd *sdp, struct buffer_head *m_bh,
 561                    struct buffer_head *l_bh)
 562 {
 563         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 564         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 565         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 566         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 567
 568         gfs2_trans_add_meta(l_ip->i_gl, l_bh);
 569         gfs2_trans_add_meta(m_ip->i_gl, m_bh);
 570
 571         spin_lock(&sdp->sd_statfs_spin);
 572         m_sc->sc_total += l_sc->sc_total;
 573         m_sc->sc_free += l_sc->sc_free;
 574         m_sc->sc_dinodes += l_sc->sc_dinodes;
 575         memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
 576         memset(l_bh->b_data + sizeof(struct gfs2_dinode),
 577                0, sizeof(struct gfs2_statfs_change));
 578         gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
 579         spin_unlock(&sdp->sd_statfs_spin);
 580 }
 581
 582 int gfs2_statfs_sync(struct super_block *sb, int type)
 583 {
 584         struct gfs2_sbd *sdp = sb->s_fs_info;
 585         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 586         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 587         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 588         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 589         struct gfs2_holder gh;
 590         struct buffer_head *m_bh, *l_bh;
 591         int error;
 592
 593         sb_start_write(sb);
 594         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 595                                    &gh);
 596         if (error)
 597                 goto out;
 598
 599         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 600         if (error)
 601                 goto out_unlock;
 602
 603         spin_lock(&sdp->sd_statfs_spin);
 604         gfs2_statfs_change_in(m_sc, m_bh->b_data +
 605                               sizeof(struct gfs2_dinode));
 606         if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
 607                 spin_unlock(&sdp->sd_statfs_spin);
 608                 goto out_bh;
 609         }
 610         spin_unlock(&sdp->sd_statfs_spin);
 611
 612         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 613         if (error)
 614                 goto out_bh;
 615
 616         error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
 617         if (error)
 618                 goto out_bh2;
 619
 620         update_statfs(sdp, m_bh, l_bh);
 621         sdp->sd_statfs_force_sync = 0;
 622
 623         gfs2_trans_end(sdp);
 624
 625 out_bh2:
 626         brelse(l_bh);
 627 out_bh:
 628         brelse(m_bh);
 629 out_unlock:
 630         gfs2_glock_dq_uninit(&gh);
 631 out:
 632         sb_end_write(sb);
 633         return error;
 634 }
 635
 636 struct lfcc {
 637         struct list_head list;
 638         struct gfs2_holder gh;
 639 };
 640
 641 /**
 642  * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
 643  *                            journals are clean
 644  * @sdp: the file system
 645  * @state: the state to put the transaction lock into
 646  * @t_gh: the hold on the transaction lock
 647  *
 648  * Returns: errno
 649  */
 650
 651 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp,
 652                                     struct gfs2_holder *freeze_gh)
 653 {
 654         struct gfs2_inode *ip;
 655         struct gfs2_jdesc *jd;
 656         struct lfcc *lfcc;
 657         LIST_HEAD(list);
 658         struct gfs2_log_header_host lh;
 659         int error;
 660
 661         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 662                 lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
 663                 if (!lfcc) {
 664                         error = -ENOMEM;
 665                         goto out;
 666                 }
 667                 ip = GFS2_I(jd->jd_inode);
 668                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
 669                 if (error) {
 670                         kfree(lfcc);
 671                         goto out;
 672                 }
 673                 list_add(&lfcc->list, &list);
 674         }
 675
 676         error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_EXCLUSIVE,
 677                                    GL_NOCACHE, freeze_gh);
 678
 679         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 680                 error = gfs2_jdesc_check(jd);
 681                 if (error)
 682                         break;
 683                 error = gfs2_find_jhead(jd, &lh);
 684                 if (error)
 685                         break;
 686                 if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 687                         error = -EBUSY;
 688                         break;
 689                 }
 690         }
 691
 692         if (error)
 693                 gfs2_glock_dq_uninit(freeze_gh);
 694
 695 out:
 696         while (!list_empty(&list)) {
 697                 lfcc = list_entry(list.next, struct lfcc, list);
 698                 list_del(&lfcc->list);
 699                 gfs2_glock_dq_uninit(&lfcc->gh);
 700                 kfree(lfcc);
 701         }
 702         return error;
 703 }
 704
 705 void gfs2_dinode_out(const struct gfs2_inode *ip, void *buf)
 706 {
 707         struct gfs2_dinode *str = buf;
 708
 709         str->di_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 710         str->di_header.mh_type = cpu_to_be32(GFS2_METATYPE_DI);
 711         str->di_header.mh_format = cpu_to_be32(GFS2_FORMAT_DI);
 712         str->di_num.no_addr = cpu_to_be64(ip->i_no_addr);
 713         str->di_num.no_formal_ino = cpu_to_be64(ip->i_no_formal_ino);
 714         str->di_mode = cpu_to_be32(ip->i_inode.i_mode);
 715         str->di_uid = cpu_to_be32(i_uid_read(&ip->i_inode));
 716         str->di_gid = cpu_to_be32(i_gid_read(&ip->i_inode));
 717         str->di_nlink = cpu_to_be32(ip->i_inode.i_nlink);
 718         str->di_size = cpu_to_be64(i_size_read(&ip->i_inode));
 719         str->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(&ip->i_inode));
 720         str->di_atime = cpu_to_be64(ip->i_inode.i_atime.tv_sec);
 721         str->di_mtime = cpu_to_be64(ip->i_inode.i_mtime.tv_sec);
 722         str->di_ctime = cpu_to_be64(ip->i_inode.i_ctime.tv_sec);
 723
 724         str->di_goal_meta = cpu_to_be64(ip->i_goal);
 725         str->di_goal_data = cpu_to_be64(ip->i_goal);
 726         str->di_generation = cpu_to_be64(ip->i_generation);
 727
 728         str->di_flags = cpu_to_be32(ip->i_diskflags);
 729         str->di_height = cpu_to_be16(ip->i_height);
 730         str->di_payload_format = cpu_to_be32(S_ISDIR(ip->i_inode.i_mode) &&
 731                                              !(ip->i_diskflags & GFS2_DIF_EXHASH) ?
 732                                              GFS2_FORMAT_DE : 0);
 733         str->di_depth = cpu_to_be16(ip->i_depth);
 734         str->di_entries = cpu_to_be32(ip->i_entries);
 735
 736         str->di_eattr = cpu_to_be64(ip->i_eattr);
 737         str->di_atime_nsec = cpu_to_be32(ip->i_inode.i_atime.tv_nsec);
 738         str->di_mtime_nsec = cpu_to_be32(ip->i_inode.i_mtime.tv_nsec);
 739         str->di_ctime_nsec = cpu_to_be32(ip->i_inode.i_ctime.tv_nsec);
 740 }
 741
 742 /**
 743  * gfs2_write_inode - Make sure the inode is stable on the disk
 744  * @inode: The inode
 745  * @wbc: The writeback control structure
 746  *
 747  * Returns: errno
 748  */
 749
 750 static int gfs2_write_inode(struct inode *inode, struct writeback_control *wbc)
 751 {
 752         struct gfs2_inode *ip = GFS2_I(inode);
 753         struct gfs2_sbd *sdp = GFS2_SB(inode);
 754         struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl);
 755         struct backing_dev_info *bdi = inode_to_bdi(metamapping->host);
 756         int ret = 0;
 757
 758         if (wbc->sync_mode == WB_SYNC_ALL)
 759                 gfs2_log_flush(GFS2_SB(inode), ip->i_gl, NORMAL_FLUSH);
 760         if (bdi->wb.dirty_exceeded)
 761                 gfs2_ail1_flush(sdp, wbc);
 762         else
 763                 filemap_fdatawrite(metamapping);
 764         if (wbc->sync_mode == WB_SYNC_ALL)
 765                 ret = filemap_fdatawait(metamapping);
 766         if (ret)
 767                 mark_inode_dirty_sync(inode);
 768         return ret;
 769 }
 770
 771 /**
 772  * gfs2_dirty_inode - check for atime updates
 773  * @inode: The inode in question
 774  * @flags: The type of dirty
 775  *
 776  * Unfortunately it can be called under any combination of inode
 777  * glock and transaction lock, so we have to check carefully.
 778  *
 779  * At the moment this deals only with atime - it should be possible
 780  * to expand that role in future, once a review of the locking has
 781  * been carried out.
 782  */
 783
 784 static void gfs2_dirty_inode(struct inode *inode, int flags)
 785 {
 786         struct gfs2_inode *ip = GFS2_I(inode);
 787         struct gfs2_sbd *sdp = GFS2_SB(inode);
 788         struct buffer_head *bh;
 789         struct gfs2_holder gh;
 790         int need_unlock = 0;
 791         int need_endtrans = 0;
 792         int ret;
 793
 794         if (!(flags & I_DIRTY_INODE))
 795                 return;
 796         if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
 797                 return;
 798         if (!gfs2_glock_is_locked_by_me(ip->i_gl)) {
 799                 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
 800                 if (ret) {
 801                         fs_err(sdp, "dirty_inode: glock %d\n", ret);
 802                         return;
 803                 }
 804                 need_unlock = 1;
 805         } else if (WARN_ON_ONCE(ip->i_gl->gl_state != LM_ST_EXCLUSIVE))
 806                 return;
 807
 808         if (current->journal_info == NULL) {
 809                 ret = gfs2_trans_begin(sdp, RES_DINODE, 0);
 810                 if (ret) {
 811                         fs_err(sdp, "dirty_inode: gfs2_trans_begin %d\n", ret);
 812                         goto out;
 813                 }
 814                 need_endtrans = 1;
 815         }
 816
 817         ret = gfs2_meta_inode_buffer(ip, &bh);
 818         if (ret == 0) {
 819                 gfs2_trans_add_meta(ip->i_gl, bh);
 820                 gfs2_dinode_out(ip, bh->b_data);
 821                 brelse(bh);
 822         }
 823
 824         if (need_endtrans)
 825                 gfs2_trans_end(sdp);
 826 out:
 827         if (need_unlock)
 828                 gfs2_glock_dq_uninit(&gh);
 829 }
 830
 831 /**
 832  * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
 833  * @sdp: the filesystem
 834  *
 835  * Returns: errno
 836  */
 837
 838 static int gfs2_make_fs_ro(struct gfs2_sbd *sdp)
 839 {
 840         struct gfs2_holder freeze_gh;
 841         int error;
 842
 843         error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED, GL_NOCACHE,
 844                                    &freeze_gh);
 845         if (error && !test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
 846                 return error;
 847
 848         flush_workqueue(gfs2_delete_workqueue);
 849         kthread_stop(sdp->sd_quotad_process);
 850         kthread_stop(sdp->sd_logd_process);
 851
 852         gfs2_quota_sync(sdp->sd_vfs, 0);
 853         gfs2_statfs_sync(sdp->sd_vfs, 0);
 854
 855         gfs2_log_flush(sdp, NULL, SHUTDOWN_FLUSH);
 856         wait_event(sdp->sd_reserving_log_wait, atomic_read(&sdp->sd_reserving_log) == 0);
 857         gfs2_assert_warn(sdp, atomic_read(&sdp->sd_log_blks_free) == sdp->sd_jdesc->jd_blocks);
 858
 859         if (gfs2_holder_initialized(&freeze_gh))
 860                 gfs2_glock_dq_uninit(&freeze_gh);
 861
 862         gfs2_quota_cleanup(sdp);
 863
 864         return error;
 865 }
 866
 867 /**
 868  * gfs2_put_super - Unmount the filesystem
 869  * @sb: The VFS superblock
 870  *
 871  */
 872
 873 static void gfs2_put_super(struct super_block *sb)
 874 {
 875         struct gfs2_sbd *sdp = sb->s_fs_info;
 876         int error;
 877         struct gfs2_jdesc *jd;
 878
 879         /* No more recovery requests */
 880         set_bit(SDF_NORECOVERY, &sdp->sd_flags);
 881         smp_mb();
 882
 883         /* Wait on outstanding recovery */
 884 restart:
 885         spin_lock(&sdp->sd_jindex_spin);
 886         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 887                 if (!test_bit(JDF_RECOVERY, &jd->jd_flags))
 888                         continue;
 889                 spin_unlock(&sdp->sd_jindex_spin);
 890                 wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
 891                             TASK_UNINTERRUPTIBLE);
 892                 goto restart;
 893         }
 894         spin_unlock(&sdp->sd_jindex_spin);
 895
 896         if (!sb_rdonly(sb)) {
 897                 error = gfs2_make_fs_ro(sdp);
 898                 if (error)
 899                         gfs2_io_error(sdp);
 900         }
 901         /*  At this point, we're through modifying the disk  */
 902
 903         /*  Release stuff  */
 904
 905         iput(sdp->sd_jindex);
 906         iput(sdp->sd_statfs_inode);
 907         iput(sdp->sd_rindex);
 908         iput(sdp->sd_quota_inode);
 909
 910         gfs2_glock_put(sdp->sd_rename_gl);
 911         gfs2_glock_put(sdp->sd_freeze_gl);
 912
 913         if (!sdp->sd_args.ar_spectator) {
 914                 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
 915                 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
 916                 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
 917                 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
 918                 iput(sdp->sd_sc_inode);
 919                 iput(sdp->sd_qc_inode);
 920         }
 921
 922         gfs2_glock_dq_uninit(&sdp->sd_live_gh);
 923         gfs2_clear_rgrpd(sdp);
 924         gfs2_jindex_free(sdp);
 925         /*  Take apart glock structures and buffer lists  */
 926         gfs2_gl_hash_clear(sdp);
 927         truncate_inode_pages_final(&sdp->sd_aspace);
 928         gfs2_delete_debugfs_file(sdp);
 929         /*  Unmount the locking protocol  */
 930         gfs2_lm_unmount(sdp);
 931
 932         /*  At this point, we're through participating in the lockspace  */
 933         gfs2_sys_fs_del(sdp);
 934 }
 935
 936 /**
 937  * gfs2_sync_fs - sync the filesystem
 938  * @sb: the superblock
 939  *
 940  * Flushes the log to disk.
 941  */
 942
 943 static int gfs2_sync_fs(struct super_block *sb, int wait)
 944 {
 945         struct gfs2_sbd *sdp = sb->s_fs_info;
 946
 947         gfs2_quota_sync(sb, -1);
 948         if (wait)
 949                 gfs2_log_flush(sdp, NULL, NORMAL_FLUSH);
 950         return sdp->sd_log_error;
 951 }
 952
 953 void gfs2_freeze_func(struct work_struct *work)
 954 {
 955         int error;
 956         struct gfs2_holder freeze_gh;
 957         struct gfs2_sbd *sdp = container_of(work, struct gfs2_sbd, sd_freeze_work);
 958         struct super_block *sb = sdp->sd_vfs;
 959
 960         atomic_inc(&sb->s_active);
 961         error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED, 0,
 962                                    &freeze_gh);
 963         if (error) {
 964                 printk(KERN_INFO "GFS2: couln't get freeze lock : %d\n", error);
 965                 gfs2_assert_withdraw(sdp, 0);
 966         }
 967         else {
 968                 atomic_set(&sdp->sd_freeze_state, SFS_UNFROZEN);
 969                 error = thaw_super(sb);
 970                 if (error) {
 971                         printk(KERN_INFO "GFS2: couldn't thaw filesystem: %d\n",
 972                                error);
 973                         gfs2_assert_withdraw(sdp, 0);
 974                 }
 975                 if (!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))
 976                         freeze_gh.gh_flags |= GL_NOCACHE;
 977                 gfs2_glock_dq_uninit(&freeze_gh);
 978         }
 979         deactivate_super(sb);
 980         return;
 981 }
 982
 983 /**
 984  * gfs2_freeze - prevent further writes to the filesystem
 985  * @sb: the VFS structure for the filesystem
 986  *
 987  */
 988
 989 static int gfs2_freeze(struct super_block *sb)
 990 {
 991         struct gfs2_sbd *sdp = sb->s_fs_info;
 992         int error;
 993
 994         mutex_lock(&sdp->sd_freeze_mutex);
 995         if (atomic_read(&sdp->sd_freeze_state) != SFS_UNFROZEN) {
 996                 error = -EBUSY;
 997                 goto out;
 998         }
 999
1000         if (test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) {
1001                 error = -EINVAL;
1002                 goto out;
1003         }
1004
1005         for (;;) {
1006                 error = gfs2_lock_fs_check_clean(sdp, &sdp->sd_freeze_gh);
1007                 if (!error)
1008                         break;
1009
1010                 switch (error) {
1011                 case -EBUSY:
1012                         fs_err(sdp, "waiting for recovery before freeze\n");
1013                         break;
1014
1015                 default:
1016                         fs_err(sdp, "error freezing FS: %d\n", error);
1017                         break;
1018                 }
1019
1020                 fs_err(sdp, "retrying...\n");
1021                 msleep(1000);
1022         }
1023         error = 0;
1024 out:
1025         mutex_unlock(&sdp->sd_freeze_mutex);
1026         return error;
1027 }
1028
1029 /**
1030  * gfs2_unfreeze - reallow writes to the filesystem
1031  * @sb: the VFS structure for the filesystem
1032  *
1033  */
1034
1035 static int gfs2_unfreeze(struct super_block *sb)
1036 {
1037         struct gfs2_sbd *sdp = sb->s_fs_info;
1038
1039         mutex_lock(&sdp->sd_freeze_mutex);
1040         if (atomic_read(&sdp->sd_freeze_state) != SFS_FROZEN ||
1041             !gfs2_holder_initialized(&sdp->sd_freeze_gh)) {
1042                 mutex_unlock(&sdp->sd_freeze_mutex);
1043                 return -EINVAL;
1044         }
1045
1046         gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
1047         mutex_unlock(&sdp->sd_freeze_mutex);
1048         return 0;
1049 }
1050
1051 /**
1052  * statfs_fill - fill in the sg for a given RG
1053  * @rgd: the RG
1054  * @sc: the sc structure
1055  *
1056  * Returns: 0 on success, -ESTALE if the LVB is invalid
1057  */
1058
1059 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
1060                             struct gfs2_statfs_change_host *sc)
1061 {
1062         gfs2_rgrp_verify(rgd);
1063         sc->sc_total += rgd->rd_data;
1064         sc->sc_free += rgd->rd_free;
1065         sc->sc_dinodes += rgd->rd_dinodes;
1066         return 0;
1067 }
1068
1069 /**
1070  * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
1071  * @sdp: the filesystem
1072  * @sc: the sc info that will be returned
1073  *
1074  * Any error (other than a signal) will cause this routine to fall back
1075  * to the synchronous version.
1076  *
1077  * FIXME: This really shouldn't busy wait like this.
1078  *
1079  * Returns: errno
1080  */
1081
1082 static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
1083 {
1084         struct gfs2_rgrpd *rgd_next;
1085         struct gfs2_holder *gha, *gh;
1086         unsigned int slots = 64;
1087         unsigned int x;
1088         int done;
1089         int error = 0, err;
1090
1091         memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
1092         gha = kmalloc(slots * sizeof(struct gfs2_holder), GFP_KERNEL);
1093         if (!gha)
1094                 return -ENOMEM;
1095         for (x = 0; x < slots; x++)
1096                 gfs2_holder_mark_uninitialized(gha + x);
1097
1098         rgd_next = gfs2_rgrpd_get_first(sdp);
1099
1100         for (;;) {
1101                 done = 1;
1102
1103                 for (x = 0; x < slots; x++) {
1104                         gh = gha + x;
1105
1106                         if (gfs2_holder_initialized(gh) && gfs2_glock_poll(gh)) {
1107                                 err = gfs2_glock_wait(gh);
1108                                 if (err) {
1109                                         gfs2_holder_uninit(gh);
1110                                         error = err;
1111                                 } else {
1112                                         if (!error) {
1113                                                 struct gfs2_rgrpd *rgd =
1114                                                         gfs2_glock2rgrp(gh->gh_gl);
1115
1116                                                 error = statfs_slow_fill(rgd, sc);
1117                                         }
1118                                         gfs2_glock_dq_uninit(gh);
1119                                 }
1120                         }
1121
1122                         if (gfs2_holder_initialized(gh))
1123                                 done = 0;
1124                         else if (rgd_next && !error) {
1125                                 error = gfs2_glock_nq_init(rgd_next->rd_gl,
1126                                                            LM_ST_SHARED,
1127                                                            GL_ASYNC,
1128                                                            gh);
1129                                 rgd_next = gfs2_rgrpd_get_next(rgd_next);
1130                                 done = 0;
1131                         }
1132
1133                         if (signal_pending(current))
1134                                 error = -ERESTARTSYS;
1135                 }
1136
1137                 if (done)
1138                         break;
1139
1140                 yield();
1141         }
1142
1143         kfree(gha);
1144         return error;
1145 }
1146
1147 /**
1148  * gfs2_statfs_i - Do a statfs
1149  * @sdp: the filesystem
1150  * @sg: the sg structure
1151  *
1152  * Returns: errno
1153  */
1154
1155 static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
1156 {
1157         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
1158         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
1159
1160         spin_lock(&sdp->sd_statfs_spin);
1161
1162         *sc = *m_sc;
1163         sc->sc_total += l_sc->sc_total;
1164         sc->sc_free += l_sc->sc_free;
1165         sc->sc_dinodes += l_sc->sc_dinodes;
1166
1167         spin_unlock(&sdp->sd_statfs_spin);
1168
1169         if (sc->sc_free < 0)
1170                 sc->sc_free = 0;
1171         if (sc->sc_free > sc->sc_total)
1172                 sc->sc_free = sc->sc_total;
1173         if (sc->sc_dinodes < 0)
1174                 sc->sc_dinodes = 0;
1175
1176         return 0;
1177 }
1178
1179 /**
1180  * gfs2_statfs - Gather and return stats about the filesystem
1181  * @sb: The superblock
1182  * @statfsbuf: The buffer
1183  *
1184  * Returns: 0 on success or error code
1185  */
1186
1187 static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
1188 {
1189         struct super_block *sb = dentry->d_sb;
1190         struct gfs2_sbd *sdp = sb->s_fs_info;
1191         struct gfs2_statfs_change_host sc;
1192         int error;
1193
1194         error = gfs2_rindex_update(sdp);
1195         if (error)
1196                 return error;
1197
1198         if (gfs2_tune_get(sdp, gt_statfs_slow))
1199                 error = gfs2_statfs_slow(sdp, &sc);
1200         else
1201                 error = gfs2_statfs_i(sdp, &sc);
1202
1203         if (error)
1204                 return error;
1205
1206         buf->f_type = GFS2_MAGIC;
1207         buf->f_bsize = sdp->sd_sb.sb_bsize;
1208         buf->f_blocks = sc.sc_total;
1209         buf->f_bfree = sc.sc_free;
1210         buf->f_bavail = sc.sc_free;
1211         buf->f_files = sc.sc_dinodes + sc.sc_free;
1212         buf->f_ffree = sc.sc_free;
1213         buf->f_namelen = GFS2_FNAMESIZE;
1214
1215         return 0;
1216 }
1217
1218 /**
1219  * gfs2_remount_fs - called when the FS is remounted
1220  * @sb:  the filesystem
1221  * @flags:  the remount flags
1222  * @data:  extra data passed in (not used right now)
1223  *
1224  * Returns: errno
1225  */
1226
1227 static int gfs2_remount_fs(struct super_block *sb, int *flags, char *data)
1228 {
1229         struct gfs2_sbd *sdp = sb->s_fs_info;
1230         struct gfs2_args args = sdp->sd_args; /* Default to current settings */
1231         struct gfs2_tune *gt = &sdp->sd_tune;
1232         int error;
1233
1234         sync_filesystem(sb);
1235
1236         spin_lock(&gt->gt_spin);
1237         args.ar_commit = gt->gt_logd_secs;
1238         args.ar_quota_quantum = gt->gt_quota_quantum;
1239         if (gt->gt_statfs_slow)
1240                 args.ar_statfs_quantum = 0;
1241         else
1242                 args.ar_statfs_quantum = gt->gt_statfs_quantum;
1243         spin_unlock(&gt->gt_spin);
1244         error = gfs2_mount_args(&args, data);
1245         if (error)
1246                 return error;
1247
1248         /* Not allowed to change locking details */
1249         if (strcmp(args.ar_lockproto, sdp->sd_args.ar_lockproto) ||
1250             strcmp(args.ar_locktable, sdp->sd_args.ar_locktable) ||
1251             strcmp(args.ar_hostdata, sdp->sd_args.ar_hostdata))
1252                 return -EINVAL;
1253
1254         /* Some flags must not be changed */
1255         if (args_neq(&args, &sdp->sd_args, spectator) ||
1256             args_neq(&args, &sdp->sd_args, localflocks) ||
1257             args_neq(&args, &sdp->sd_args, meta))
1258                 return -EINVAL;
1259
1260         if (sdp->sd_args.ar_spectator)
1261                 *flags |= MS_RDONLY;
1262
1263         if ((sb->s_flags ^ *flags) & MS_RDONLY) {
1264                 if (*flags & MS_RDONLY)
1265                         error = gfs2_make_fs_ro(sdp);
1266                 else
1267                         error = gfs2_make_fs_rw(sdp);
1268                 if (error)
1269                         return error;
1270         }
1271
1272         sdp->sd_args = args;
1273         if (sdp->sd_args.ar_posix_acl)
1274                 sb->s_flags |= MS_POSIXACL;
1275         else
1276                 sb->s_flags &= ~MS_POSIXACL;
1277         if (sdp->sd_args.ar_nobarrier)
1278                 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1279         else
1280                 clear_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1281         spin_lock(&gt->gt_spin);
1282         gt->gt_logd_secs = args.ar_commit;
1283         gt->gt_quota_quantum = args.ar_quota_quantum;
1284         if (args.ar_statfs_quantum) {
1285                 gt->gt_statfs_slow = 0;
1286                 gt->gt_statfs_quantum = args.ar_statfs_quantum;
1287         }
1288         else {
1289                 gt->gt_statfs_slow = 1;
1290                 gt->gt_statfs_quantum = 30;
1291         }
1292         spin_unlock(&gt->gt_spin);
1293
1294         gfs2_online_uevent(sdp);
1295         return 0;
1296 }
1297
1298 /**
1299  * gfs2_drop_inode - Drop an inode (test for remote unlink)
1300  * @inode: The inode to drop
1301  *
1302  * If we've received a callback on an iopen lock then it's because a
1303  * remote node tried to deallocate the inode but failed due to this node
1304  * still having the inode open. Here we mark the link count zero
1305  * since we know that it must have reached zero if the GLF_DEMOTE flag
1306  * is set on the iopen glock. If we didn't do a disk read since the
1307  * remote node removed the final link then we might otherwise miss
1308  * this event. This check ensures that this node will deallocate the
1309  * inode's blocks, or alternatively pass the baton on to another
1310  * node for later deallocation.
1311  */
1312
1313 static int gfs2_drop_inode(struct inode *inode)
1314 {
1315         struct gfs2_inode *ip = GFS2_I(inode);
1316
1317         if (!test_bit(GIF_FREE_VFS_INODE, &ip->i_flags) &&
1318             inode->i_nlink &&
1319             gfs2_holder_initialized(&ip->i_iopen_gh)) {
1320                 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1321                 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
1322                         clear_nlink(inode);
1323         }
1324
1325         /*
1326          * When under memory pressure when an inode's link count has dropped to
1327          * zero, defer deleting the inode to the delete workqueue.  This avoids
1328          * calling into DLM under memory pressure, which can deadlock.
1329          */
1330         if (!inode->i_nlink &&
1331             unlikely(current->flags & PF_MEMALLOC) &&
1332             gfs2_holder_initialized(&ip->i_iopen_gh)) {
1333                 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1334
1335                 gfs2_glock_hold(gl);
1336                 if (queue_work(gfs2_delete_workqueue, &gl->gl_delete) == 0)
1337                         gfs2_glock_queue_put(gl);
1338                 return false;
1339         }
1340
1341         return generic_drop_inode(inode);
1342 }
1343
1344 static int is_ancestor(const struct dentry *d1, const struct dentry *d2)
1345 {
1346         do {
1347                 if (d1 == d2)
1348                         return 1;
1349                 d1 = d1->d_parent;
1350         } while (!IS_ROOT(d1));
1351         return 0;
1352 }
1353
1354 /**
1355  * gfs2_show_options - Show mount options for /proc/mounts
1356  * @s: seq_file structure
1357  * @root: root of this (sub)tree
1358  *
1359  * Returns: 0 on success or error code
1360  */
1361
1362 static int gfs2_show_options(struct seq_file *s, struct dentry *root)
1363 {
1364         struct gfs2_sbd *sdp = root->d_sb->s_fs_info;
1365         struct gfs2_args *args = &sdp->sd_args;
1366         int val;
1367
1368         if (is_ancestor(root, sdp->sd_master_dir))
1369                 seq_puts(s, ",meta");
1370         if (args->ar_lockproto[0])
1371                 seq_show_option(s, "lockproto", args->ar_lockproto);
1372         if (args->ar_locktable[0])
1373                 seq_show_option(s, "locktable", args->ar_locktable);
1374         if (args->ar_hostdata[0])
1375                 seq_show_option(s, "hostdata", args->ar_hostdata);
1376         if (args->ar_spectator)
1377                 seq_puts(s, ",spectator");
1378         if (args->ar_localflocks)
1379                 seq_puts(s, ",localflocks");
1380         if (args->ar_debug)
1381                 seq_puts(s, ",debug");
1382         if (args->ar_posix_acl)
1383                 seq_puts(s, ",acl");
1384         if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
1385                 char *state;
1386                 switch (args->ar_quota) {
1387                 case GFS2_QUOTA_OFF:
1388                         state = "off";
1389                         break;
1390                 case GFS2_QUOTA_ACCOUNT:
1391                         state = "account";
1392                         break;
1393                 case GFS2_QUOTA_ON:
1394                         state = "on";
1395                         break;
1396                 default:
1397                         state = "unknown";
1398                         break;
1399                 }
1400                 seq_printf(s, ",quota=%s", state);
1401         }
1402         if (args->ar_suiddir)
1403                 seq_puts(s, ",suiddir");
1404         if (args->ar_data != GFS2_DATA_DEFAULT) {
1405                 char *state;
1406                 switch (args->ar_data) {
1407                 case GFS2_DATA_WRITEBACK:
1408                         state = "writeback";
1409                         break;
1410                 case GFS2_DATA_ORDERED:
1411                         state = "ordered";
1412                         break;
1413                 default:
1414                         state = "unknown";
1415                         break;
1416                 }
1417                 seq_printf(s, ",data=%s", state);
1418         }
1419         if (args->ar_discard)
1420                 seq_puts(s, ",discard");
1421         val = sdp->sd_tune.gt_logd_secs;
1422         if (val != 30)
1423                 seq_printf(s, ",commit=%d", val);
1424         val = sdp->sd_tune.gt_statfs_quantum;
1425         if (val != 30)
1426                 seq_printf(s, ",statfs_quantum=%d", val);
1427         else if (sdp->sd_tune.gt_statfs_slow)
1428                 seq_puts(s, ",statfs_quantum=0");
1429         val = sdp->sd_tune.gt_quota_quantum;
1430         if (val != 60)
1431                 seq_printf(s, ",quota_quantum=%d", val);
1432         if (args->ar_statfs_percent)
1433                 seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent);
1434         if (args->ar_errors != GFS2_ERRORS_DEFAULT) {
1435                 const char *state;
1436
1437                 switch (args->ar_errors) {
1438                 case GFS2_ERRORS_WITHDRAW:
1439                         state = "withdraw";
1440                         break;
1441                 case GFS2_ERRORS_PANIC:
1442                         state = "panic";
1443                         break;
1444                 default:
1445                         state = "unknown";
1446                         break;
1447                 }
1448                 seq_printf(s, ",errors=%s", state);
1449         }
1450         if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags))
1451                 seq_puts(s, ",nobarrier");
1452         if (test_bit(SDF_DEMOTE, &sdp->sd_flags))
1453                 seq_puts(s, ",demote_interface_used");
1454         if (args->ar_rgrplvb)
1455                 seq_puts(s, ",rgrplvb");
1456         if (args->ar_loccookie)
1457                 seq_puts(s, ",loccookie");
1458         return 0;
1459 }
1460
1461 static void gfs2_final_release_pages(struct gfs2_inode *ip)
1462 {
1463         struct inode *inode = &ip->i_inode;
1464         struct gfs2_glock *gl = ip->i_gl;
1465
1466         truncate_inode_pages(gfs2_glock2aspace(ip->i_gl), 0);
1467         truncate_inode_pages(&inode->i_data, 0);
1468
1469         if (atomic_read(&gl->gl_revokes) == 0) {
1470                 clear_bit(GLF_LFLUSH, &gl->gl_flags);
1471                 clear_bit(GLF_DIRTY, &gl->gl_flags);
1472         }
1473 }
1474
1475 static int gfs2_dinode_dealloc(struct gfs2_inode *ip)
1476 {
1477         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1478         struct gfs2_rgrpd *rgd;
1479         struct gfs2_holder gh;
1480         int error;
1481
1482         if (gfs2_get_inode_blocks(&ip->i_inode) != 1) {
1483                 gfs2_consist_inode(ip);
1484                 return -EIO;
1485         }
1486
1487         error = gfs2_rindex_update(sdp);
1488         if (error)
1489                 return error;
1490
1491         error = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
1492         if (error)
1493                 return error;
1494
1495         rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1496         if (!rgd) {
1497                 gfs2_consist_inode(ip);
1498                 error = -EIO;
1499                 goto out_qs;
1500         }
1501
1502         error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &gh);
1503         if (error)
1504                 goto out_qs;
1505
1506         error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA,
1507                                  sdp->sd_jdesc->jd_blocks);
1508         if (error)
1509                 goto out_rg_gunlock;
1510
1511         gfs2_free_di(rgd, ip);
1512
1513         gfs2_final_release_pages(ip);
1514
1515         gfs2_trans_end(sdp);
1516
1517 out_rg_gunlock:
1518         gfs2_glock_dq_uninit(&gh);
1519 out_qs:
1520         gfs2_quota_unhold(ip);
1521         return error;
1522 }
1523
1524 /**
1525  * gfs2_glock_put_eventually
1526  * @gl: The glock to put
1527  *
1528  * When under memory pressure, trigger a deferred glock put to make sure we
1529  * won't call into DLM and deadlock.  Otherwise, put the glock directly.
1530  */
1531
1532 static void gfs2_glock_put_eventually(struct gfs2_glock *gl)
1533 {
1534         if (current->flags & PF_MEMALLOC)
1535                 gfs2_glock_queue_put(gl);
1536         else
1537                 gfs2_glock_put(gl);
1538 }
1539
1540 /**
1541  * gfs2_evict_inode - Remove an inode from cache
1542  * @inode: The inode to evict
1543  *
1544  * There are three cases to consider:
1545  * 1. i_nlink == 0, we are final opener (and must deallocate)
1546  * 2. i_nlink == 0, we are not the final opener (and cannot deallocate)
1547  * 3. i_nlink > 0
1548  *
1549  * If the fs is read only, then we have to treat all cases as per #3
1550  * since we are unable to do any deallocation. The inode will be
1551  * deallocated by the next read/write node to attempt an allocation
1552  * in the same resource group
1553  *
1554  * We have to (at the moment) hold the inodes main lock to cover
1555  * the gap between unlocking the shared lock on the iopen lock and
1556  * taking the exclusive lock. I'd rather do a shared -> exclusive
1557  * conversion on the iopen lock, but we can change that later. This
1558  * is safe, just less efficient.
1559  */
1560
1561 static void gfs2_evict_inode(struct inode *inode)
1562 {
1563         struct super_block *sb = inode->i_sb;
1564         struct gfs2_sbd *sdp = sb->s_fs_info;
1565         struct gfs2_inode *ip = GFS2_I(inode);
1566         struct gfs2_holder gh;
1567         struct address_space *metamapping;
1568         int error;
1569
1570         if (test_bit(GIF_FREE_VFS_INODE, &ip->i_flags)) {
1571                 clear_inode(inode);
1572                 return;
1573         }
1574
1575         if (inode->i_nlink || sb_rdonly(sb))
1576                 goto out;
1577
1578         if (test_bit(GIF_ALLOC_FAILED, &ip->i_flags)) {
1579                 BUG_ON(!gfs2_glock_is_locked_by_me(ip->i_gl));
1580                 gfs2_holder_mark_uninitialized(&gh);
1581                 goto alloc_failed;
1582         }
1583
1584         /* Deletes should never happen under memory pressure anymore.  */
1585         if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
1586                 goto out;
1587
1588         /* Must not read inode block until block type has been verified */
1589         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, &gh);
1590         if (unlikely(error)) {
1591                 glock_clear_object(ip->i_iopen_gh.gh_gl, ip);
1592                 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1593                 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1594                 goto out;
1595         }
1596
1597         error = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED);
1598         if (error)
1599                 goto out_truncate;
1600
1601         if (test_bit(GIF_INVALID, &ip->i_flags)) {
1602                 error = gfs2_inode_refresh(ip);
1603                 if (error)
1604                         goto out_truncate;
1605         }
1606
1607         /*
1608          * The inode may have been recreated in the meantime.
1609          */
1610         if (inode->i_nlink)
1611                 goto out_truncate;
1612
1613 alloc_failed:
1614         if (gfs2_holder_initialized(&ip->i_iopen_gh) &&
1615             test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1616                 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1617                 gfs2_glock_dq_wait(&ip->i_iopen_gh);
1618                 gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE,
1619                                    &ip->i_iopen_gh);
1620                 error = gfs2_glock_nq(&ip->i_iopen_gh);
1621                 if (error)
1622                         goto out_truncate;
1623         }
1624
1625         /* Case 1 starts here */
1626
1627         if (S_ISDIR(inode->i_mode) &&
1628             (ip->i_diskflags & GFS2_DIF_EXHASH)) {
1629                 error = gfs2_dir_exhash_dealloc(ip);
1630                 if (error)
1631                         goto out_unlock;
1632         }
1633
1634         if (ip->i_eattr) {
1635                 error = gfs2_ea_dealloc(ip);
1636                 if (error)
1637                         goto out_unlock;
1638         }
1639
1640         if (!gfs2_is_stuffed(ip)) {
1641                 error = gfs2_file_dealloc(ip);
1642                 if (error)
1643                         goto out_unlock;
1644         }
1645
1646         /* We're about to clear the bitmap for the dinode, but as soon as we
1647            do, gfs2_create_inode can create another inode at the same block
1648            location and try to set gl_object again. We clear gl_object here so
1649            that subsequent inode creates don't see an old gl_object. */
1650         glock_clear_object(ip->i_gl, ip);
1651         error = gfs2_dinode_dealloc(ip);
1652         goto out_unlock;
1653
1654 out_truncate:
1655         gfs2_log_flush(sdp, ip->i_gl, NORMAL_FLUSH);
1656         metamapping = gfs2_glock2aspace(ip->i_gl);
1657         if (test_bit(GLF_DIRTY, &ip->i_gl->gl_flags)) {
1658                 filemap_fdatawrite(metamapping);
1659                 filemap_fdatawait(metamapping);
1660         }
1661         write_inode_now(inode, 1);
1662         gfs2_ail_flush(ip->i_gl, 0);
1663
1664         /* Case 2 starts here */
1665         error = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
1666         if (error)
1667                 goto out_unlock;
1668         /* Needs to be done before glock release & also in a transaction */
1669         truncate_inode_pages(&inode->i_data, 0);
1670         truncate_inode_pages(metamapping, 0);
1671         gfs2_trans_end(sdp);
1672
1673 out_unlock:
1674         /* Error path for case 1 */
1675         if (gfs2_rs_active(&ip->i_res))
1676                 gfs2_rs_deltree(&ip->i_res);
1677
1678         if (gfs2_holder_initialized(&ip->i_iopen_gh)) {
1679                 glock_clear_object(ip->i_iopen_gh.gh_gl, ip);
1680                 if (test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1681                         ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1682                         gfs2_glock_dq(&ip->i_iopen_gh);
1683                 }
1684                 gfs2_holder_uninit(&ip->i_iopen_gh);
1685         }
1686         if (gfs2_holder_initialized(&gh)) {
1687                 glock_clear_object(ip->i_gl, ip);
1688                 gfs2_glock_dq_uninit(&gh);
1689         }
1690         if (error && error != GLR_TRYFAILED && error != -EROFS)
1691                 fs_warn(sdp, "gfs2_evict_inode: %d\n", error);
1692 out:
1693         /* Case 3 starts here */
1694         truncate_inode_pages_final(&inode->i_data);
1695         gfs2_rsqa_delete(ip, NULL);
1696         gfs2_ordered_del_inode(ip);
1697         clear_inode(inode);
1698         gfs2_dir_hash_inval(ip);
1699         glock_clear_object(ip->i_gl, ip);
1700         wait_on_bit_io(&ip->i_flags, GIF_GLOP_PENDING, TASK_UNINTERRUPTIBLE);
1701         gfs2_glock_add_to_lru(ip->i_gl);
1702         gfs2_glock_put_eventually(ip->i_gl);
1703         ip->i_gl = NULL;
1704         if (gfs2_holder_initialized(&ip->i_iopen_gh)) {
1705                 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1706
1707                 glock_clear_object(gl, ip);
1708                 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1709                 gfs2_glock_hold(gl);
1710                 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1711                 gfs2_glock_put_eventually(gl);
1712         }
1713 }
1714
1715 static struct inode *gfs2_alloc_inode(struct super_block *sb)
1716 {
1717         struct gfs2_inode *ip;
1718
1719         ip = kmem_cache_alloc(gfs2_inode_cachep, GFP_KERNEL);
1720         if (ip) {
1721                 ip->i_flags = 0;
1722                 ip->i_gl = NULL;
1723                 ip->i_rgd = NULL;
1724                 memset(&ip->i_res, 0, sizeof(ip->i_res));
1725                 RB_CLEAR_NODE(&ip->i_res.rs_node);
1726                 ip->i_rahead = 0;
1727         }
1728         return &ip->i_inode;
1729 }
1730
1731 static void gfs2_i_callback(struct rcu_head *head)
1732 {
1733         struct inode *inode = container_of(head, struct inode, i_rcu);
1734         kmem_cache_free(gfs2_inode_cachep, inode);
1735 }
1736
1737 static void gfs2_destroy_inode(struct inode *inode)
1738 {
1739         call_rcu(&inode->i_rcu, gfs2_i_callback);
1740 }
1741
1742 const struct super_operations gfs2_super_ops = {
1743         .alloc_inode            = gfs2_alloc_inode,
1744         .destroy_inode          = gfs2_destroy_inode,
1745         .write_inode            = gfs2_write_inode,
1746         .dirty_inode            = gfs2_dirty_inode,
1747         .evict_inode            = gfs2_evict_inode,
1748         .put_super              = gfs2_put_super,
1749         .sync_fs                = gfs2_sync_fs,
1750         .freeze_super           = gfs2_freeze,
1751         .thaw_super             = gfs2_unfreeze,
1752         .statfs                 = gfs2_statfs,
1753         .remount_fs             = gfs2_remount_fs,
1754         .drop_inode             = gfs2_drop_inode,
1755         .show_options           = gfs2_show_options,
1756 };
1757