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 #include <linux/bio.h>
  11 #include <linux/sched.h>
  12 #include <linux/slab.h>
  13 #include <linux/spinlock.h>
  14 #include <linux/completion.h>
  15 #include <linux/buffer_head.h>
  16 #include <linux/statfs.h>
  17 #include <linux/seq_file.h>
  18 #include <linux/mount.h>
  19 #include <linux/kthread.h>
  20 #include <linux/delay.h>
  21 #include <linux/gfs2_ondisk.h>
  22 #include <linux/crc32.h>
  23 #include <linux/time.h>
  24 #include <linux/wait.h>
  25 #include <linux/writeback.h>
  26 #include <linux/backing-dev.h>
  27
  28 #include "gfs2.h"
  29 #include "incore.h"
  30 #include "bmap.h"
  31 #include "dir.h"
  32 #include "glock.h"
  33 #include "glops.h"
  34 #include "inode.h"
  35 #include "log.h"
  36 #include "meta_io.h"
  37 #include "quota.h"
  38 #include "recovery.h"
  39 #include "rgrp.h"
  40 #include "super.h"
  41 #include "trans.h"
  42 #include "util.h"
  43 #include "sys.h"
  44 #include "xattr.h"
  45
  46 #define args_neq(a1, a2, x) ((a1)->ar_##x != (a2)->ar_##x)
  47
  48 enum {
  49         Opt_lockproto,
  50         Opt_locktable,
  51         Opt_hostdata,
  52         Opt_spectator,
  53         Opt_ignore_local_fs,
  54         Opt_localflocks,
  55         Opt_localcaching,
  56         Opt_debug,
  57         Opt_nodebug,
  58         Opt_upgrade,
  59         Opt_acl,
  60         Opt_noacl,
  61         Opt_quota_off,
  62         Opt_quota_account,
  63         Opt_quota_on,
  64         Opt_quota,
  65         Opt_noquota,
  66         Opt_suiddir,
  67         Opt_nosuiddir,
  68         Opt_data_writeback,
  69         Opt_data_ordered,
  70         Opt_meta,
  71         Opt_discard,
  72         Opt_nodiscard,
  73         Opt_commit,
  74         Opt_err_withdraw,
  75         Opt_err_panic,
  76         Opt_statfs_quantum,
  77         Opt_statfs_percent,
  78         Opt_quota_quantum,
  79         Opt_barrier,
  80         Opt_nobarrier,
  81         Opt_error,
  82 };
  83
  84 static const match_table_t tokens = {
  85         {Opt_lockproto, "lockproto=%s"},
  86         {Opt_locktable, "locktable=%s"},
  87         {Opt_hostdata, "hostdata=%s"},
  88         {Opt_spectator, "spectator"},
  89         {Opt_spectator, "norecovery"},
  90         {Opt_ignore_local_fs, "ignore_local_fs"},
  91         {Opt_localflocks, "localflocks"},
  92         {Opt_localcaching, "localcaching"},
  93         {Opt_debug, "debug"},
  94         {Opt_nodebug, "nodebug"},
  95         {Opt_upgrade, "upgrade"},
  96         {Opt_acl, "acl"},
  97         {Opt_noacl, "noacl"},
  98         {Opt_quota_off, "quota=off"},
  99         {Opt_quota_account, "quota=account"},
 100         {Opt_quota_on, "quota=on"},
 101         {Opt_quota, "quota"},
 102         {Opt_noquota, "noquota"},
 103         {Opt_suiddir, "suiddir"},
 104         {Opt_nosuiddir, "nosuiddir"},
 105         {Opt_data_writeback, "data=writeback"},
 106         {Opt_data_ordered, "data=ordered"},
 107         {Opt_meta, "meta"},
 108         {Opt_discard, "discard"},
 109         {Opt_nodiscard, "nodiscard"},
 110         {Opt_commit, "commit=%d"},
 111         {Opt_err_withdraw, "errors=withdraw"},
 112         {Opt_err_panic, "errors=panic"},
 113         {Opt_statfs_quantum, "statfs_quantum=%d"},
 114         {Opt_statfs_percent, "statfs_percent=%d"},
 115         {Opt_quota_quantum, "quota_quantum=%d"},
 116         {Opt_barrier, "barrier"},
 117         {Opt_nobarrier, "nobarrier"},
 118         {Opt_error, NULL}
 119 };
 120
 121 /**
 122  * gfs2_mount_args - Parse mount options
 123  * @args: The structure into which the parsed options will be written
 124  * @options: The options to parse
 125  *
 126  * Return: errno
 127  */
 128
 129 int gfs2_mount_args(struct gfs2_args *args, char *options)
 130 {
 131         char *o;
 132         int token;
 133         substring_t tmp[MAX_OPT_ARGS];
 134         int rv;
 135
 136         /* Split the options into tokens with the "," character and
 137            process them */
 138
 139         while (1) {
 140                 o = strsep(&options, ",");
 141                 if (o == NULL)
 142                         break;
 143                 if (*o == '\0')
 144                         continue;
 145
 146                 token = match_token(o, tokens, tmp);
 147                 switch (token) {
 148                 case Opt_lockproto:
 149                         match_strlcpy(args->ar_lockproto, &tmp[0],
 150                                       GFS2_LOCKNAME_LEN);
 151                         break;
 152                 case Opt_locktable:
 153                         match_strlcpy(args->ar_locktable, &tmp[0],
 154                                       GFS2_LOCKNAME_LEN);
 155                         break;
 156                 case Opt_hostdata:
 157                         match_strlcpy(args->ar_hostdata, &tmp[0],
 158                                       GFS2_LOCKNAME_LEN);
 159                         break;
 160                 case Opt_spectator:
 161                         args->ar_spectator = 1;
 162                         break;
 163                 case Opt_ignore_local_fs:
 164                         /* Retained for backwards compat only */
 165                         break;
 166                 case Opt_localflocks:
 167                         args->ar_localflocks = 1;
 168                         break;
 169                 case Opt_localcaching:
 170                         /* Retained for backwards compat only */
 171                         break;
 172                 case Opt_debug:
 173                         if (args->ar_errors == GFS2_ERRORS_PANIC) {
 174                                 printk(KERN_WARNING "GFS2: -o debug and -o errors=panic "
 175                                        "are mutually exclusive.\n");
 176                                 return -EINVAL;
 177                         }
 178                         args->ar_debug = 1;
 179                         break;
 180                 case Opt_nodebug:
 181                         args->ar_debug = 0;
 182                         break;
 183                 case Opt_upgrade:
 184                         /* Retained for backwards compat only */
 185                         break;
 186                 case Opt_acl:
 187                         args->ar_posix_acl = 1;
 188                         break;
 189                 case Opt_noacl:
 190                         args->ar_posix_acl = 0;
 191                         break;
 192                 case Opt_quota_off:
 193                 case Opt_noquota:
 194                         args->ar_quota = GFS2_QUOTA_OFF;
 195                         break;
 196                 case Opt_quota_account:
 197                         args->ar_quota = GFS2_QUOTA_ACCOUNT;
 198                         break;
 199                 case Opt_quota_on:
 200                 case Opt_quota:
 201                         args->ar_quota = GFS2_QUOTA_ON;
 202                         break;
 203                 case Opt_suiddir:
 204                         args->ar_suiddir = 1;
 205                         break;
 206                 case Opt_nosuiddir:
 207                         args->ar_suiddir = 0;
 208                         break;
 209                 case Opt_data_writeback:
 210                         args->ar_data = GFS2_DATA_WRITEBACK;
 211                         break;
 212                 case Opt_data_ordered:
 213                         args->ar_data = GFS2_DATA_ORDERED;
 214                         break;
 215                 case Opt_meta:
 216                         args->ar_meta = 1;
 217                         break;
 218                 case Opt_discard:
 219                         args->ar_discard = 1;
 220                         break;
 221                 case Opt_nodiscard:
 222                         args->ar_discard = 0;
 223                         break;
 224                 case Opt_commit:
 225                         rv = match_int(&tmp[0], &args->ar_commit);
 226                         if (rv || args->ar_commit <= 0) {
 227                                 printk(KERN_WARNING "GFS2: commit mount option requires a positive numeric argument\n");
 228                                 return rv ? rv : -EINVAL;
 229                         }
 230                         break;
 231                 case Opt_statfs_quantum:
 232                         rv = match_int(&tmp[0], &args->ar_statfs_quantum);
 233                         if (rv || args->ar_statfs_quantum < 0) {
 234                                 printk(KERN_WARNING "GFS2: statfs_quantum mount option requires a non-negative numeric argument\n");
 235                                 return rv ? rv : -EINVAL;
 236                         }
 237                         break;
 238                 case Opt_quota_quantum:
 239                         rv = match_int(&tmp[0], &args->ar_quota_quantum);
 240                         if (rv || args->ar_quota_quantum <= 0) {
 241                                 printk(KERN_WARNING "GFS2: quota_quantum mount option requires a positive numeric argument\n");
 242                                 return rv ? rv : -EINVAL;
 243                         }
 244                         break;
 245                 case Opt_statfs_percent:
 246                         rv = match_int(&tmp[0], &args->ar_statfs_percent);
 247                         if (rv || args->ar_statfs_percent < 0 ||
 248                             args->ar_statfs_percent > 100) {
 249                                 printk(KERN_WARNING "statfs_percent mount option requires a numeric argument between 0 and 100\n");
 250                                 return rv ? rv : -EINVAL;
 251                         }
 252                         break;
 253                 case Opt_err_withdraw:
 254                         args->ar_errors = GFS2_ERRORS_WITHDRAW;
 255                         break;
 256                 case Opt_err_panic:
 257                         if (args->ar_debug) {
 258                                 printk(KERN_WARNING "GFS2: -o debug and -o errors=panic "
 259                                         "are mutually exclusive.\n");
 260                                 return -EINVAL;
 261                         }
 262                         args->ar_errors = GFS2_ERRORS_PANIC;
 263                         break;
 264                 case Opt_barrier:
 265                         args->ar_nobarrier = 0;
 266                         break;
 267                 case Opt_nobarrier:
 268                         args->ar_nobarrier = 1;
 269                         break;
 270                 case Opt_error:
 271                 default:
 272                         printk(KERN_WARNING "GFS2: invalid mount option: %s\n", o);
 273                         return -EINVAL;
 274                 }
 275         }
 276
 277         return 0;
 278 }
 279
 280 /**
 281  * gfs2_jindex_free - Clear all the journal index information
 282  * @sdp: The GFS2 superblock
 283  *
 284  */
 285
 286 void gfs2_jindex_free(struct gfs2_sbd *sdp)
 287 {
 288         struct list_head list, *head;
 289         struct gfs2_jdesc *jd;
 290         struct gfs2_journal_extent *jext;
 291
 292         spin_lock(&sdp->sd_jindex_spin);
 293         list_add(&list, &sdp->sd_jindex_list);
 294         list_del_init(&sdp->sd_jindex_list);
 295         sdp->sd_journals = 0;
 296         spin_unlock(&sdp->sd_jindex_spin);
 297
 298         while (!list_empty(&list)) {
 299                 jd = list_entry(list.next, struct gfs2_jdesc, jd_list);
 300                 head = &jd->extent_list;
 301                 while (!list_empty(head)) {
 302                         jext = list_entry(head->next,
 303                                           struct gfs2_journal_extent,
 304                                           extent_list);
 305                         list_del(&jext->extent_list);
 306                         kfree(jext);
 307                 }
 308                 list_del(&jd->jd_list);
 309                 iput(jd->jd_inode);
 310                 kfree(jd);
 311         }
 312 }
 313
 314 static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
 315 {
 316         struct gfs2_jdesc *jd;
 317         int found = 0;
 318
 319         list_for_each_entry(jd, head, jd_list) {
 320                 if (jd->jd_jid == jid) {
 321                         found = 1;
 322                         break;
 323                 }
 324         }
 325
 326         if (!found)
 327                 jd = NULL;
 328
 329         return jd;
 330 }
 331
 332 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
 333 {
 334         struct gfs2_jdesc *jd;
 335
 336         spin_lock(&sdp->sd_jindex_spin);
 337         jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
 338         spin_unlock(&sdp->sd_jindex_spin);
 339
 340         return jd;
 341 }
 342
 343 int gfs2_jdesc_check(struct gfs2_jdesc *jd)
 344 {
 345         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 346         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 347         u64 size = i_size_read(jd->jd_inode);
 348
 349         if (gfs2_check_internal_file_size(jd->jd_inode, 8 << 20, 1 << 30))
 350                 return -EIO;
 351
 352         jd->jd_blocks = size >> sdp->sd_sb.sb_bsize_shift;
 353
 354         if (gfs2_write_alloc_required(ip, 0, size)) {
 355                 gfs2_consist_inode(ip);
 356                 return -EIO;
 357         }
 358
 359         return 0;
 360 }
 361
 362 /**
 363  * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
 364  * @sdp: the filesystem
 365  *
 366  * Returns: errno
 367  */
 368
 369 int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
 370 {
 371         struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
 372         struct gfs2_glock *j_gl = ip->i_gl;
 373         struct gfs2_holder t_gh;
 374         struct gfs2_log_header_host head;
 375         int error;
 376
 377         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, 0, &t_gh);
 378         if (error)
 379                 return error;
 380
 381         j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
 382
 383         error = gfs2_find_jhead(sdp->sd_jdesc, &head);
 384         if (error)
 385                 goto fail;
 386
 387         if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 388                 gfs2_consist(sdp);
 389                 error = -EIO;
 390                 goto fail;
 391         }
 392
 393         /*  Initialize some head of the log stuff  */
 394         sdp->sd_log_sequence = head.lh_sequence + 1;
 395         gfs2_log_pointers_init(sdp, head.lh_blkno);
 396
 397         error = gfs2_quota_init(sdp);
 398         if (error)
 399                 goto fail;
 400
 401         set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 402
 403         gfs2_glock_dq_uninit(&t_gh);
 404
 405         return 0;
 406
 407 fail:
 408         t_gh.gh_flags |= GL_NOCACHE;
 409         gfs2_glock_dq_uninit(&t_gh);
 410
 411         return error;
 412 }
 413
 414 void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
 415 {
 416         const struct gfs2_statfs_change *str = buf;
 417
 418         sc->sc_total = be64_to_cpu(str->sc_total);
 419         sc->sc_free = be64_to_cpu(str->sc_free);
 420         sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
 421 }
 422
 423 static void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
 424 {
 425         struct gfs2_statfs_change *str = buf;
 426
 427         str->sc_total = cpu_to_be64(sc->sc_total);
 428         str->sc_free = cpu_to_be64(sc->sc_free);
 429         str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
 430 }
 431
 432 int gfs2_statfs_init(struct gfs2_sbd *sdp)
 433 {
 434         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 435         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 436         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 437         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 438         struct buffer_head *m_bh, *l_bh;
 439         struct gfs2_holder gh;
 440         int error;
 441
 442         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 443                                    &gh);
 444         if (error)
 445                 return error;
 446
 447         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 448         if (error)
 449                 goto out;
 450
 451         if (sdp->sd_args.ar_spectator) {
 452                 spin_lock(&sdp->sd_statfs_spin);
 453                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 454                                       sizeof(struct gfs2_dinode));
 455                 spin_unlock(&sdp->sd_statfs_spin);
 456         } else {
 457                 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 458                 if (error)
 459                         goto out_m_bh;
 460
 461                 spin_lock(&sdp->sd_statfs_spin);
 462                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 463                                       sizeof(struct gfs2_dinode));
 464                 gfs2_statfs_change_in(l_sc, l_bh->b_data +
 465                                       sizeof(struct gfs2_dinode));
 466                 spin_unlock(&sdp->sd_statfs_spin);
 467
 468                 brelse(l_bh);
 469         }
 470
 471 out_m_bh:
 472         brelse(m_bh);
 473 out:
 474         gfs2_glock_dq_uninit(&gh);
 475         return 0;
 476 }
 477
 478 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
 479                         s64 dinodes)
 480 {
 481         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 482         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 483         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 484         struct buffer_head *l_bh;
 485         s64 x, y;
 486         int need_sync = 0;
 487         int error;
 488
 489         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 490         if (error)
 491                 return;
 492
 493         gfs2_trans_add_bh(l_ip->i_gl, l_bh, 1);
 494
 495         spin_lock(&sdp->sd_statfs_spin);
 496         l_sc->sc_total += total;
 497         l_sc->sc_free += free;
 498         l_sc->sc_dinodes += dinodes;
 499         gfs2_statfs_change_out(l_sc, l_bh->b_data + sizeof(struct gfs2_dinode));
 500         if (sdp->sd_args.ar_statfs_percent) {
 501                 x = 100 * l_sc->sc_free;
 502                 y = m_sc->sc_free * sdp->sd_args.ar_statfs_percent;
 503                 if (x >= y || x <= -y)
 504                         need_sync = 1;
 505         }
 506         spin_unlock(&sdp->sd_statfs_spin);
 507
 508         brelse(l_bh);
 509         if (need_sync)
 510                 gfs2_wake_up_statfs(sdp);
 511 }
 512
 513 void update_statfs(struct gfs2_sbd *sdp, struct buffer_head *m_bh,
 514                    struct buffer_head *l_bh)
 515 {
 516         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 517         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 518         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 519         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 520
 521         gfs2_trans_add_bh(l_ip->i_gl, l_bh, 1);
 522
 523         spin_lock(&sdp->sd_statfs_spin);
 524         m_sc->sc_total += l_sc->sc_total;
 525         m_sc->sc_free += l_sc->sc_free;
 526         m_sc->sc_dinodes += l_sc->sc_dinodes;
 527         memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
 528         memset(l_bh->b_data + sizeof(struct gfs2_dinode),
 529                0, sizeof(struct gfs2_statfs_change));
 530         spin_unlock(&sdp->sd_statfs_spin);
 531
 532         gfs2_trans_add_bh(m_ip->i_gl, m_bh, 1);
 533         gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
 534 }
 535
 536 int gfs2_statfs_sync(struct super_block *sb, int type)
 537 {
 538         struct gfs2_sbd *sdp = sb->s_fs_info;
 539         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 540         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 541         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 542         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 543         struct gfs2_holder gh;
 544         struct buffer_head *m_bh, *l_bh;
 545         int error;
 546
 547         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 548                                    &gh);
 549         if (error)
 550                 return error;
 551
 552         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 553         if (error)
 554                 goto out;
 555
 556         spin_lock(&sdp->sd_statfs_spin);
 557         gfs2_statfs_change_in(m_sc, m_bh->b_data +
 558                               sizeof(struct gfs2_dinode));
 559         if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
 560                 spin_unlock(&sdp->sd_statfs_spin);
 561                 goto out_bh;
 562         }
 563         spin_unlock(&sdp->sd_statfs_spin);
 564
 565         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 566         if (error)
 567                 goto out_bh;
 568
 569         error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
 570         if (error)
 571                 goto out_bh2;
 572
 573         update_statfs(sdp, m_bh, l_bh);
 574         sdp->sd_statfs_force_sync = 0;
 575
 576         gfs2_trans_end(sdp);
 577
 578 out_bh2:
 579         brelse(l_bh);
 580 out_bh:
 581         brelse(m_bh);
 582 out:
 583         gfs2_glock_dq_uninit(&gh);
 584         return error;
 585 }
 586
 587 struct lfcc {
 588         struct list_head list;
 589         struct gfs2_holder gh;
 590 };
 591
 592 /**
 593  * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
 594  *                            journals are clean
 595  * @sdp: the file system
 596  * @state: the state to put the transaction lock into
 597  * @t_gh: the hold on the transaction lock
 598  *
 599  * Returns: errno
 600  */
 601
 602 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp,
 603                                     struct gfs2_holder *t_gh)
 604 {
 605         struct gfs2_inode *ip;
 606         struct gfs2_jdesc *jd;
 607         struct lfcc *lfcc;
 608         LIST_HEAD(list);
 609         struct gfs2_log_header_host lh;
 610         int error;
 611
 612         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 613                 lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
 614                 if (!lfcc) {
 615                         error = -ENOMEM;
 616                         goto out;
 617                 }
 618                 ip = GFS2_I(jd->jd_inode);
 619                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
 620                 if (error) {
 621                         kfree(lfcc);
 622                         goto out;
 623                 }
 624                 list_add(&lfcc->list, &list);
 625         }
 626
 627         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_DEFERRED,
 628                                    GL_NOCACHE, t_gh);
 629
 630         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 631                 error = gfs2_jdesc_check(jd);
 632                 if (error)
 633                         break;
 634                 error = gfs2_find_jhead(jd, &lh);
 635                 if (error)
 636                         break;
 637                 if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 638                         error = -EBUSY;
 639                         break;
 640                 }
 641         }
 642
 643         if (error)
 644                 gfs2_glock_dq_uninit(t_gh);
 645
 646 out:
 647         while (!list_empty(&list)) {
 648                 lfcc = list_entry(list.next, struct lfcc, list);
 649                 list_del(&lfcc->list);
 650                 gfs2_glock_dq_uninit(&lfcc->gh);
 651                 kfree(lfcc);
 652         }
 653         return error;
 654 }
 655
 656 /**
 657  * gfs2_freeze_fs - freezes the file system
 658  * @sdp: the file system
 659  *
 660  * This function flushes data and meta data for all machines by
 661  * acquiring the transaction log exclusively.  All journals are
 662  * ensured to be in a clean state as well.
 663  *
 664  * Returns: errno
 665  */
 666
 667 int gfs2_freeze_fs(struct gfs2_sbd *sdp)
 668 {
 669         int error = 0;
 670
 671         mutex_lock(&sdp->sd_freeze_lock);
 672
 673         if (!sdp->sd_freeze_count++) {
 674                 error = gfs2_lock_fs_check_clean(sdp, &sdp->sd_freeze_gh);
 675                 if (error)
 676                         sdp->sd_freeze_count--;
 677         }
 678
 679         mutex_unlock(&sdp->sd_freeze_lock);
 680
 681         return error;
 682 }
 683
 684 /**
 685  * gfs2_unfreeze_fs - unfreezes the file system
 686  * @sdp: the file system
 687  *
 688  * This function allows the file system to proceed by unlocking
 689  * the exclusively held transaction lock.  Other GFS2 nodes are
 690  * now free to acquire the lock shared and go on with their lives.
 691  *
 692  */
 693
 694 void gfs2_unfreeze_fs(struct gfs2_sbd *sdp)
 695 {
 696         mutex_lock(&sdp->sd_freeze_lock);
 697
 698         if (sdp->sd_freeze_count && !--sdp->sd_freeze_count)
 699                 gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
 700
 701         mutex_unlock(&sdp->sd_freeze_lock);
 702 }
 703
 704 void gfs2_dinode_out(const struct gfs2_inode *ip, void *buf)
 705 {
 706         struct gfs2_dinode *str = buf;
 707
 708         str->di_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 709         str->di_header.mh_type = cpu_to_be32(GFS2_METATYPE_DI);
 710         str->di_header.mh_format = cpu_to_be32(GFS2_FORMAT_DI);
 711         str->di_num.no_addr = cpu_to_be64(ip->i_no_addr);
 712         str->di_num.no_formal_ino = cpu_to_be64(ip->i_no_formal_ino);
 713         str->di_mode = cpu_to_be32(ip->i_inode.i_mode);
 714         str->di_uid = cpu_to_be32(ip->i_inode.i_uid);
 715         str->di_gid = cpu_to_be32(ip->i_inode.i_gid);
 716         str->di_nlink = cpu_to_be32(ip->i_inode.i_nlink);
 717         str->di_size = cpu_to_be64(i_size_read(&ip->i_inode));
 718         str->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(&ip->i_inode));
 719         str->di_atime = cpu_to_be64(ip->i_inode.i_atime.tv_sec);
 720         str->di_mtime = cpu_to_be64(ip->i_inode.i_mtime.tv_sec);
 721         str->di_ctime = cpu_to_be64(ip->i_inode.i_ctime.tv_sec);
 722
 723         str->di_goal_meta = cpu_to_be64(ip->i_goal);
 724         str->di_goal_data = cpu_to_be64(ip->i_goal);
 725         str->di_generation = cpu_to_be64(ip->i_generation);
 726
 727         str->di_flags = cpu_to_be32(ip->i_diskflags);
 728         str->di_height = cpu_to_be16(ip->i_height);
 729         str->di_payload_format = cpu_to_be32(S_ISDIR(ip->i_inode.i_mode) &&
 730                                              !(ip->i_diskflags & GFS2_DIF_EXHASH) ?
 731                                              GFS2_FORMAT_DE : 0);
 732         str->di_depth = cpu_to_be16(ip->i_depth);
 733         str->di_entries = cpu_to_be32(ip->i_entries);
 734
 735         str->di_eattr = cpu_to_be64(ip->i_eattr);
 736         str->di_atime_nsec = cpu_to_be32(ip->i_inode.i_atime.tv_nsec);
 737         str->di_mtime_nsec = cpu_to_be32(ip->i_inode.i_mtime.tv_nsec);
 738         str->di_ctime_nsec = cpu_to_be32(ip->i_inode.i_ctime.tv_nsec);
 739 }
 740
 741 /**
 742  * gfs2_write_inode - Make sure the inode is stable on the disk
 743  * @inode: The inode
 744  * @wbc: The writeback control structure
 745  *
 746  * Returns: errno
 747  */
 748
 749 static int gfs2_write_inode(struct inode *inode, struct writeback_control *wbc)
 750 {
 751         struct gfs2_inode *ip = GFS2_I(inode);
 752         struct gfs2_sbd *sdp = GFS2_SB(inode);
 753         struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl);
 754         struct backing_dev_info *bdi = metamapping->backing_dev_info;
 755         int ret = 0;
 756
 757         if (wbc->sync_mode == WB_SYNC_ALL)
 758                 gfs2_log_flush(GFS2_SB(inode), ip->i_gl);
 759         if (bdi->dirty_exceeded)
 760                 gfs2_ail1_flush(sdp, wbc);
 761         else
 762                 filemap_fdatawrite(metamapping);
 763         if (wbc->sync_mode == WB_SYNC_ALL)
 764                 ret = filemap_fdatawait(metamapping);
 765         if (ret)
 766                 mark_inode_dirty_sync(inode);
 767         return ret;
 768 }
 769
 770 /**
 771  * gfs2_dirty_inode - check for atime updates
 772  * @inode: The inode in question
 773  * @flags: The type of dirty
 774  *
 775  * Unfortunately it can be called under any combination of inode
 776  * glock and transaction lock, so we have to check carefully.
 777  *
 778  * At the moment this deals only with atime - it should be possible
 779  * to expand that role in future, once a review of the locking has
 780  * been carried out.
 781  */
 782
 783 static void gfs2_dirty_inode(struct inode *inode, int flags)
 784 {
 785         struct gfs2_inode *ip = GFS2_I(inode);
 786         struct gfs2_sbd *sdp = GFS2_SB(inode);
 787         struct buffer_head *bh;
 788         struct gfs2_holder gh;
 789         int need_unlock = 0;
 790         int need_endtrans = 0;
 791         int ret;
 792
 793         if (!(flags & (I_DIRTY_DATASYNC|I_DIRTY_SYNC)))
 794                 return;
 795
 796         if (!gfs2_glock_is_locked_by_me(ip->i_gl)) {
 797                 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
 798                 if (ret) {
 799                         fs_err(sdp, "dirty_inode: glock %d\n", ret);
 800                         return;
 801                 }
 802                 need_unlock = 1;
 803         }
 804
 805         if (current->journal_info == NULL) {
 806                 ret = gfs2_trans_begin(sdp, RES_DINODE, 0);
 807                 if (ret) {
 808                         fs_err(sdp, "dirty_inode: gfs2_trans_begin %d\n", ret);
 809                         goto out;
 810                 }
 811                 need_endtrans = 1;
 812         }
 813
 814         ret = gfs2_meta_inode_buffer(ip, &bh);
 815         if (ret == 0) {
 816                 gfs2_trans_add_bh(ip->i_gl, bh, 1);
 817                 gfs2_dinode_out(ip, bh->b_data);
 818                 brelse(bh);
 819         }
 820
 821         if (need_endtrans)
 822                 gfs2_trans_end(sdp);
 823 out:
 824         if (need_unlock)
 825                 gfs2_glock_dq_uninit(&gh);
 826 }
 827
 828 /**
 829  * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
 830  * @sdp: the filesystem
 831  *
 832  * Returns: errno
 833  */
 834
 835 static int gfs2_make_fs_ro(struct gfs2_sbd *sdp)
 836 {
 837         struct gfs2_holder t_gh;
 838         int error;
 839
 840         flush_workqueue(gfs2_delete_workqueue);
 841         gfs2_quota_sync(sdp->sd_vfs, 0, 1);
 842         gfs2_statfs_sync(sdp->sd_vfs, 0);
 843
 844         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, GL_NOCACHE,
 845                                    &t_gh);
 846         if (error && !test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
 847                 return error;
 848
 849         gfs2_meta_syncfs(sdp);
 850         gfs2_log_shutdown(sdp);
 851
 852         clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 853
 854         if (t_gh.gh_gl)
 855                 gfs2_glock_dq_uninit(&t_gh);
 856
 857         gfs2_quota_cleanup(sdp);
 858
 859         return error;
 860 }
 861
 862 static int gfs2_umount_recovery_wait(void *word)
 863 {
 864         schedule();
 865         return 0;
 866 }
 867
 868 /**
 869  * gfs2_put_super - Unmount the filesystem
 870  * @sb: The VFS superblock
 871  *
 872  */
 873
 874 static void gfs2_put_super(struct super_block *sb)
 875 {
 876         struct gfs2_sbd *sdp = sb->s_fs_info;
 877         int error;
 878         struct gfs2_jdesc *jd;
 879
 880         /*  Unfreeze the filesystem, if we need to  */
 881
 882         mutex_lock(&sdp->sd_freeze_lock);
 883         if (sdp->sd_freeze_count)
 884                 gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
 885         mutex_unlock(&sdp->sd_freeze_lock);
 886
 887         /* No more recovery requests */
 888         set_bit(SDF_NORECOVERY, &sdp->sd_flags);
 889         smp_mb();
 890
 891         /* Wait on outstanding recovery */
 892 restart:
 893         spin_lock(&sdp->sd_jindex_spin);
 894         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 895                 if (!test_bit(JDF_RECOVERY, &jd->jd_flags))
 896                         continue;
 897                 spin_unlock(&sdp->sd_jindex_spin);
 898                 wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
 899                             gfs2_umount_recovery_wait, TASK_UNINTERRUPTIBLE);
 900                 goto restart;
 901         }
 902         spin_unlock(&sdp->sd_jindex_spin);
 903
 904         kthread_stop(sdp->sd_quotad_process);
 905         kthread_stop(sdp->sd_logd_process);
 906
 907         if (!(sb->s_flags & MS_RDONLY)) {
 908                 error = gfs2_make_fs_ro(sdp);
 909                 if (error)
 910                         gfs2_io_error(sdp);
 911         }
 912         /*  At this point, we're through modifying the disk  */
 913
 914         /*  Release stuff  */
 915
 916         iput(sdp->sd_jindex);
 917         iput(sdp->sd_statfs_inode);
 918         iput(sdp->sd_rindex);
 919         iput(sdp->sd_quota_inode);
 920
 921         gfs2_glock_put(sdp->sd_rename_gl);
 922         gfs2_glock_put(sdp->sd_trans_gl);
 923
 924         if (!sdp->sd_args.ar_spectator) {
 925                 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
 926                 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
 927                 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
 928                 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
 929                 iput(sdp->sd_sc_inode);
 930                 iput(sdp->sd_qc_inode);
 931         }
 932
 933         gfs2_glock_dq_uninit(&sdp->sd_live_gh);
 934         gfs2_clear_rgrpd(sdp);
 935         gfs2_jindex_free(sdp);
 936         /*  Take apart glock structures and buffer lists  */
 937         gfs2_gl_hash_clear(sdp);
 938         /*  Unmount the locking protocol  */
 939         gfs2_lm_unmount(sdp);
 940
 941         /*  At this point, we're through participating in the lockspace  */
 942         gfs2_sys_fs_del(sdp);
 943 }
 944
 945 /**
 946  * gfs2_sync_fs - sync the filesystem
 947  * @sb: the superblock
 948  *
 949  * Flushes the log to disk.
 950  */
 951
 952 static int gfs2_sync_fs(struct super_block *sb, int wait)
 953 {
 954         struct gfs2_sbd *sdp = sb->s_fs_info;
 955         if (wait && sdp)
 956                 gfs2_log_flush(sdp, NULL);
 957         return 0;
 958 }
 959
 960 /**
 961  * gfs2_freeze - prevent further writes to the filesystem
 962  * @sb: the VFS structure for the filesystem
 963  *
 964  */
 965
 966 static int gfs2_freeze(struct super_block *sb)
 967 {
 968         struct gfs2_sbd *sdp = sb->s_fs_info;
 969         int error;
 970
 971         if (test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
 972                 return -EINVAL;
 973
 974         for (;;) {
 975                 error = gfs2_freeze_fs(sdp);
 976                 if (!error)
 977                         break;
 978
 979                 switch (error) {
 980                 case -EBUSY:
 981                         fs_err(sdp, "waiting for recovery before freeze\n");
 982                         break;
 983
 984                 default:
 985                         fs_err(sdp, "error freezing FS: %d\n", error);
 986                         break;
 987                 }
 988
 989                 fs_err(sdp, "retrying...\n");
 990                 msleep(1000);
 991         }
 992         return 0;
 993 }
 994
 995 /**
 996  * gfs2_unfreeze - reallow writes to the filesystem
 997  * @sb: the VFS structure for the filesystem
 998  *
 999  */
1000
1001 static int gfs2_unfreeze(struct super_block *sb)
1002 {
1003         gfs2_unfreeze_fs(sb->s_fs_info);
1004         return 0;
1005 }
1006
1007 /**
1008  * statfs_fill - fill in the sg for a given RG
1009  * @rgd: the RG
1010  * @sc: the sc structure
1011  *
1012  * Returns: 0 on success, -ESTALE if the LVB is invalid
1013  */
1014
1015 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
1016                             struct gfs2_statfs_change_host *sc)
1017 {
1018         gfs2_rgrp_verify(rgd);
1019         sc->sc_total += rgd->rd_data;
1020         sc->sc_free += rgd->rd_free;
1021         sc->sc_dinodes += rgd->rd_dinodes;
1022         return 0;
1023 }
1024
1025 /**
1026  * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
1027  * @sdp: the filesystem
1028  * @sc: the sc info that will be returned
1029  *
1030  * Any error (other than a signal) will cause this routine to fall back
1031  * to the synchronous version.
1032  *
1033  * FIXME: This really shouldn't busy wait like this.
1034  *
1035  * Returns: errno
1036  */
1037
1038 static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
1039 {
1040         struct gfs2_rgrpd *rgd_next;
1041         struct gfs2_holder *gha, *gh;
1042         unsigned int slots = 64;
1043         unsigned int x;
1044         int done;
1045         int error = 0, err;
1046
1047         memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
1048         gha = kcalloc(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
1049         if (!gha)
1050                 return -ENOMEM;
1051
1052         rgd_next = gfs2_rgrpd_get_first(sdp);
1053
1054         for (;;) {
1055                 done = 1;
1056
1057                 for (x = 0; x < slots; x++) {
1058                         gh = gha + x;
1059
1060                         if (gh->gh_gl && gfs2_glock_poll(gh)) {
1061                                 err = gfs2_glock_wait(gh);
1062                                 if (err) {
1063                                         gfs2_holder_uninit(gh);
1064                                         error = err;
1065                                 } else {
1066                                         if (!error)
1067                                                 error = statfs_slow_fill(
1068                                                         gh->gh_gl->gl_object, sc);
1069                                         gfs2_glock_dq_uninit(gh);
1070                                 }
1071                         }
1072
1073                         if (gh->gh_gl)
1074                                 done = 0;
1075                         else if (rgd_next && !error) {
1076                                 error = gfs2_glock_nq_init(rgd_next->rd_gl,
1077                                                            LM_ST_SHARED,
1078                                                            GL_ASYNC,
1079                                                            gh);
1080                                 rgd_next = gfs2_rgrpd_get_next(rgd_next);
1081                                 done = 0;
1082                         }
1083
1084                         if (signal_pending(current))
1085                                 error = -ERESTARTSYS;
1086                 }
1087
1088                 if (done)
1089                         break;
1090
1091                 yield();
1092         }
1093
1094         kfree(gha);
1095         return error;
1096 }
1097
1098 /**
1099  * gfs2_statfs_i - Do a statfs
1100  * @sdp: the filesystem
1101  * @sg: the sg structure
1102  *
1103  * Returns: errno
1104  */
1105
1106 static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
1107 {
1108         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
1109         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
1110
1111         spin_lock(&sdp->sd_statfs_spin);
1112
1113         *sc = *m_sc;
1114         sc->sc_total += l_sc->sc_total;
1115         sc->sc_free += l_sc->sc_free;
1116         sc->sc_dinodes += l_sc->sc_dinodes;
1117
1118         spin_unlock(&sdp->sd_statfs_spin);
1119
1120         if (sc->sc_free < 0)
1121                 sc->sc_free = 0;
1122         if (sc->sc_free > sc->sc_total)
1123                 sc->sc_free = sc->sc_total;
1124         if (sc->sc_dinodes < 0)
1125                 sc->sc_dinodes = 0;
1126
1127         return 0;
1128 }
1129
1130 /**
1131  * gfs2_statfs - Gather and return stats about the filesystem
1132  * @sb: The superblock
1133  * @statfsbuf: The buffer
1134  *
1135  * Returns: 0 on success or error code
1136  */
1137
1138 static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
1139 {
1140         struct super_block *sb = dentry->d_inode->i_sb;
1141         struct gfs2_sbd *sdp = sb->s_fs_info;
1142         struct gfs2_statfs_change_host sc;
1143         int error;
1144
1145         error = gfs2_rindex_update(sdp);
1146         if (error)
1147                 return error;
1148
1149         if (gfs2_tune_get(sdp, gt_statfs_slow))
1150                 error = gfs2_statfs_slow(sdp, &sc);
1151         else
1152                 error = gfs2_statfs_i(sdp, &sc);
1153
1154         if (error)
1155                 return error;
1156
1157         buf->f_type = GFS2_MAGIC;
1158         buf->f_bsize = sdp->sd_sb.sb_bsize;
1159         buf->f_blocks = sc.sc_total;
1160         buf->f_bfree = sc.sc_free;
1161         buf->f_bavail = sc.sc_free;
1162         buf->f_files = sc.sc_dinodes + sc.sc_free;
1163         buf->f_ffree = sc.sc_free;
1164         buf->f_namelen = GFS2_FNAMESIZE;
1165
1166         return 0;
1167 }
1168
1169 /**
1170  * gfs2_remount_fs - called when the FS is remounted
1171  * @sb:  the filesystem
1172  * @flags:  the remount flags
1173  * @data:  extra data passed in (not used right now)
1174  *
1175  * Returns: errno
1176  */
1177
1178 static int gfs2_remount_fs(struct super_block *sb, int *flags, char *data)
1179 {
1180         struct gfs2_sbd *sdp = sb->s_fs_info;
1181         struct gfs2_args args = sdp->sd_args; /* Default to current settings */
1182         struct gfs2_tune *gt = &sdp->sd_tune;
1183         int error;
1184
1185         spin_lock(&gt->gt_spin);
1186         args.ar_commit = gt->gt_logd_secs;
1187         args.ar_quota_quantum = gt->gt_quota_quantum;
1188         if (gt->gt_statfs_slow)
1189                 args.ar_statfs_quantum = 0;
1190         else
1191                 args.ar_statfs_quantum = gt->gt_statfs_quantum;
1192         spin_unlock(&gt->gt_spin);
1193         error = gfs2_mount_args(&args, data);
1194         if (error)
1195                 return error;
1196
1197         /* Not allowed to change locking details */
1198         if (strcmp(args.ar_lockproto, sdp->sd_args.ar_lockproto) ||
1199             strcmp(args.ar_locktable, sdp->sd_args.ar_locktable) ||
1200             strcmp(args.ar_hostdata, sdp->sd_args.ar_hostdata))
1201                 return -EINVAL;
1202
1203         /* Some flags must not be changed */
1204         if (args_neq(&args, &sdp->sd_args, spectator) ||
1205             args_neq(&args, &sdp->sd_args, localflocks) ||
1206             args_neq(&args, &sdp->sd_args, meta))
1207                 return -EINVAL;
1208
1209         if (sdp->sd_args.ar_spectator)
1210                 *flags |= MS_RDONLY;
1211
1212         if ((sb->s_flags ^ *flags) & MS_RDONLY) {
1213                 if (*flags & MS_RDONLY)
1214                         error = gfs2_make_fs_ro(sdp);
1215                 else
1216                         error = gfs2_make_fs_rw(sdp);
1217                 if (error)
1218                         return error;
1219         }
1220
1221         sdp->sd_args = args;
1222         if (sdp->sd_args.ar_posix_acl)
1223                 sb->s_flags |= MS_POSIXACL;
1224         else
1225                 sb->s_flags &= ~MS_POSIXACL;
1226         if (sdp->sd_args.ar_nobarrier)
1227                 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1228         else
1229                 clear_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1230         spin_lock(&gt->gt_spin);
1231         gt->gt_logd_secs = args.ar_commit;
1232         gt->gt_quota_quantum = args.ar_quota_quantum;
1233         if (args.ar_statfs_quantum) {
1234                 gt->gt_statfs_slow = 0;
1235                 gt->gt_statfs_quantum = args.ar_statfs_quantum;
1236         }
1237         else {
1238                 gt->gt_statfs_slow = 1;
1239                 gt->gt_statfs_quantum = 30;
1240         }
1241         spin_unlock(&gt->gt_spin);
1242
1243         gfs2_online_uevent(sdp);
1244         return 0;
1245 }
1246
1247 /**
1248  * gfs2_drop_inode - Drop an inode (test for remote unlink)
1249  * @inode: The inode to drop
1250  *
1251  * If we've received a callback on an iopen lock then its because a
1252  * remote node tried to deallocate the inode but failed due to this node
1253  * still having the inode open. Here we mark the link count zero
1254  * since we know that it must have reached zero if the GLF_DEMOTE flag
1255  * is set on the iopen glock. If we didn't do a disk read since the
1256  * remote node removed the final link then we might otherwise miss
1257  * this event. This check ensures that this node will deallocate the
1258  * inode's blocks, or alternatively pass the baton on to another
1259  * node for later deallocation.
1260  */
1261
1262 static int gfs2_drop_inode(struct inode *inode)
1263 {
1264         struct gfs2_inode *ip = GFS2_I(inode);
1265
1266         if (inode->i_nlink) {
1267                 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1268                 if (gl && test_bit(GLF_DEMOTE, &gl->gl_flags))
1269                         clear_nlink(inode);
1270         }
1271         return generic_drop_inode(inode);
1272 }
1273
1274 static int is_ancestor(const struct dentry *d1, const struct dentry *d2)
1275 {
1276         do {
1277                 if (d1 == d2)
1278                         return 1;
1279                 d1 = d1->d_parent;
1280         } while (!IS_ROOT(d1));
1281         return 0;
1282 }
1283
1284 /**
1285  * gfs2_show_options - Show mount options for /proc/mounts
1286  * @s: seq_file structure
1287  * @mnt: vfsmount
1288  *
1289  * Returns: 0 on success or error code
1290  */
1291
1292 static int gfs2_show_options(struct seq_file *s, struct vfsmount *mnt)
1293 {
1294         struct gfs2_sbd *sdp = mnt->mnt_sb->s_fs_info;
1295         struct gfs2_args *args = &sdp->sd_args;
1296         int val;
1297
1298         if (is_ancestor(mnt->mnt_root, sdp->sd_master_dir))
1299                 seq_printf(s, ",meta");
1300         if (args->ar_lockproto[0])
1301                 seq_show_option(s, "lockproto", args->ar_lockproto);
1302         if (args->ar_locktable[0])
1303                 seq_show_option(s, "locktable", args->ar_locktable);
1304         if (args->ar_hostdata[0])
1305                 seq_show_option(s, "hostdata", args->ar_hostdata);
1306         if (args->ar_spectator)
1307                 seq_printf(s, ",spectator");
1308         if (args->ar_localflocks)
1309                 seq_printf(s, ",localflocks");
1310         if (args->ar_debug)
1311                 seq_printf(s, ",debug");
1312         if (args->ar_posix_acl)
1313                 seq_printf(s, ",acl");
1314         if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
1315                 char *state;
1316                 switch (args->ar_quota) {
1317                 case GFS2_QUOTA_OFF:
1318                         state = "off";
1319                         break;
1320                 case GFS2_QUOTA_ACCOUNT:
1321                         state = "account";
1322                         break;
1323                 case GFS2_QUOTA_ON:
1324                         state = "on";
1325                         break;
1326                 default:
1327                         state = "unknown";
1328                         break;
1329                 }
1330                 seq_printf(s, ",quota=%s", state);
1331         }
1332         if (args->ar_suiddir)
1333                 seq_printf(s, ",suiddir");
1334         if (args->ar_data != GFS2_DATA_DEFAULT) {
1335                 char *state;
1336                 switch (args->ar_data) {
1337                 case GFS2_DATA_WRITEBACK:
1338                         state = "writeback";
1339                         break;
1340                 case GFS2_DATA_ORDERED:
1341                         state = "ordered";
1342                         break;
1343                 default:
1344                         state = "unknown";
1345                         break;
1346                 }
1347                 seq_printf(s, ",data=%s", state);
1348         }
1349         if (args->ar_discard)
1350                 seq_printf(s, ",discard");
1351         val = sdp->sd_tune.gt_logd_secs;
1352         if (val != 30)
1353                 seq_printf(s, ",commit=%d", val);
1354         val = sdp->sd_tune.gt_statfs_quantum;
1355         if (val != 30)
1356                 seq_printf(s, ",statfs_quantum=%d", val);
1357         val = sdp->sd_tune.gt_quota_quantum;
1358         if (val != 60)
1359                 seq_printf(s, ",quota_quantum=%d", val);
1360         if (args->ar_statfs_percent)
1361                 seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent);
1362         if (args->ar_errors != GFS2_ERRORS_DEFAULT) {
1363                 const char *state;
1364
1365                 switch (args->ar_errors) {
1366                 case GFS2_ERRORS_WITHDRAW:
1367                         state = "withdraw";
1368                         break;
1369                 case GFS2_ERRORS_PANIC:
1370                         state = "panic";
1371                         break;
1372                 default:
1373                         state = "unknown";
1374                         break;
1375                 }
1376                 seq_printf(s, ",errors=%s", state);
1377         }
1378         if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags))
1379                 seq_printf(s, ",nobarrier");
1380         if (test_bit(SDF_DEMOTE, &sdp->sd_flags))
1381                 seq_printf(s, ",demote_interface_used");
1382         return 0;
1383 }
1384
1385 static void gfs2_final_release_pages(struct gfs2_inode *ip)
1386 {
1387         struct inode *inode = &ip->i_inode;
1388         struct gfs2_glock *gl = ip->i_gl;
1389
1390         truncate_inode_pages(gfs2_glock2aspace(ip->i_gl), 0);
1391         truncate_inode_pages(&inode->i_data, 0);
1392
1393         if (atomic_read(&gl->gl_revokes) == 0) {
1394                 clear_bit(GLF_LFLUSH, &gl->gl_flags);
1395                 clear_bit(GLF_DIRTY, &gl->gl_flags);
1396         }
1397 }
1398
1399 static int gfs2_dinode_dealloc(struct gfs2_inode *ip)
1400 {
1401         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1402         struct gfs2_alloc *al;
1403         struct gfs2_rgrpd *rgd;
1404         int error;
1405
1406         if (gfs2_get_inode_blocks(&ip->i_inode) != 1) {
1407                 gfs2_consist_inode(ip);
1408                 return -EIO;
1409         }
1410
1411         al = gfs2_alloc_get(ip);
1412         if (!al)
1413                 return -ENOMEM;
1414
1415         error = gfs2_quota_hold(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
1416         if (error)
1417                 goto out;
1418
1419         rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr);
1420         if (!rgd) {
1421                 gfs2_consist_inode(ip);
1422                 error = -EIO;
1423                 goto out_qs;
1424         }
1425
1426         error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0,
1427                                    &al->al_rgd_gh);
1428         if (error)
1429                 goto out_qs;
1430
1431         error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA,
1432                                  sdp->sd_jdesc->jd_blocks);
1433         if (error)
1434                 goto out_rg_gunlock;
1435
1436         gfs2_free_di(rgd, ip);
1437
1438         gfs2_final_release_pages(ip);
1439
1440         gfs2_trans_end(sdp);
1441
1442 out_rg_gunlock:
1443         gfs2_glock_dq_uninit(&al->al_rgd_gh);
1444 out_qs:
1445         gfs2_quota_unhold(ip);
1446 out:
1447         gfs2_alloc_put(ip);
1448         return error;
1449 }
1450
1451 /**
1452  * gfs2_evict_inode - Remove an inode from cache
1453  * @inode: The inode to evict
1454  *
1455  * There are three cases to consider:
1456  * 1. i_nlink == 0, we are final opener (and must deallocate)
1457  * 2. i_nlink == 0, we are not the final opener (and cannot deallocate)
1458  * 3. i_nlink > 0
1459  *
1460  * If the fs is read only, then we have to treat all cases as per #3
1461  * since we are unable to do any deallocation. The inode will be
1462  * deallocated by the next read/write node to attempt an allocation
1463  * in the same resource group
1464  *
1465  * We have to (at the moment) hold the inodes main lock to cover
1466  * the gap between unlocking the shared lock on the iopen lock and
1467  * taking the exclusive lock. I'd rather do a shared -> exclusive
1468  * conversion on the iopen lock, but we can change that later. This
1469  * is safe, just less efficient.
1470  */
1471
1472 static void gfs2_evict_inode(struct inode *inode)
1473 {
1474         struct super_block *sb = inode->i_sb;
1475         struct gfs2_sbd *sdp = sb->s_fs_info;
1476         struct gfs2_inode *ip = GFS2_I(inode);
1477         struct gfs2_holder gh;
1478         int error;
1479
1480         if (inode->i_nlink || (sb->s_flags & MS_RDONLY))
1481                 goto out;
1482
1483         /* Must not read inode block until block type has been verified */
1484         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, &gh);
1485         if (unlikely(error)) {
1486                 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1487                 goto out;
1488         }
1489
1490         if (!test_bit(GIF_ALLOC_FAILED, &ip->i_flags)) {
1491                 error = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED);
1492                 if (error)
1493                         goto out_truncate;
1494         }
1495
1496         if (test_bit(GIF_INVALID, &ip->i_flags)) {
1497                 error = gfs2_inode_refresh(ip);
1498                 if (error)
1499                         goto out_truncate;
1500         }
1501
1502         ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1503         gfs2_glock_dq_wait(&ip->i_iopen_gh);
1504         gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE, &ip->i_iopen_gh);
1505         error = gfs2_glock_nq(&ip->i_iopen_gh);
1506         if (error)
1507                 goto out_truncate;
1508
1509         /* Case 1 starts here */
1510
1511         if (S_ISDIR(inode->i_mode) &&
1512             (ip->i_diskflags & GFS2_DIF_EXHASH)) {
1513                 error = gfs2_dir_exhash_dealloc(ip);
1514                 if (error)
1515                         goto out_unlock;
1516         }
1517
1518         if (ip->i_eattr) {
1519                 error = gfs2_ea_dealloc(ip);
1520                 if (error)
1521                         goto out_unlock;
1522         }
1523
1524         if (!gfs2_is_stuffed(ip)) {
1525                 error = gfs2_file_dealloc(ip);
1526                 if (error)
1527                         goto out_unlock;
1528         }
1529
1530         error = gfs2_dinode_dealloc(ip);
1531         goto out_unlock;
1532
1533 out_truncate:
1534         gfs2_log_flush(sdp, ip->i_gl);
1535         write_inode_now(inode, 1);
1536         gfs2_ail_flush(ip->i_gl, 0);
1537
1538         /* Case 2 starts here */
1539         error = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
1540         if (error)
1541                 goto out_unlock;
1542         /* Needs to be done before glock release & also in a transaction */
1543         truncate_inode_pages(&inode->i_data, 0);
1544         gfs2_trans_end(sdp);
1545
1546 out_unlock:
1547         /* Error path for case 1 */
1548         if (test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags))
1549                 gfs2_glock_dq(&ip->i_iopen_gh);
1550         gfs2_holder_uninit(&ip->i_iopen_gh);
1551         gfs2_glock_dq_uninit(&gh);
1552         if (error && error != GLR_TRYFAILED && error != -EROFS)
1553                 fs_warn(sdp, "gfs2_evict_inode: %d\n", error);
1554 out:
1555         /* Case 3 starts here */
1556         truncate_inode_pages(&inode->i_data, 0);
1557         end_writeback(inode);
1558         gfs2_dir_hash_inval(ip);
1559         ip->i_gl->gl_object = NULL;
1560         gfs2_glock_add_to_lru(ip->i_gl);
1561         gfs2_glock_put(ip->i_gl);
1562         ip->i_gl = NULL;
1563         if (ip->i_iopen_gh.gh_gl) {
1564                 ip->i_iopen_gh.gh_gl->gl_object = NULL;
1565                 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1566         }
1567 }
1568
1569 static struct inode *gfs2_alloc_inode(struct super_block *sb)
1570 {
1571         struct gfs2_inode *ip;
1572
1573         ip = kmem_cache_alloc(gfs2_inode_cachep, GFP_KERNEL);
1574         if (ip) {
1575                 ip->i_flags = 0;
1576                 ip->i_gl = NULL;
1577                 ip->i_rgd = NULL;
1578         }
1579         return &ip->i_inode;
1580 }
1581
1582 static void gfs2_i_callback(struct rcu_head *head)
1583 {
1584         struct inode *inode = container_of(head, struct inode, i_rcu);
1585         INIT_LIST_HEAD(&inode->i_dentry);
1586         kmem_cache_free(gfs2_inode_cachep, inode);
1587 }
1588
1589 static void gfs2_destroy_inode(struct inode *inode)
1590 {
1591         call_rcu(&inode->i_rcu, gfs2_i_callback);
1592 }
1593
1594 const struct super_operations gfs2_super_ops = {
1595         .alloc_inode            = gfs2_alloc_inode,
1596         .destroy_inode          = gfs2_destroy_inode,
1597         .write_inode            = gfs2_write_inode,
1598         .dirty_inode            = gfs2_dirty_inode,
1599         .evict_inode            = gfs2_evict_inode,
1600         .put_super              = gfs2_put_super,
1601         .sync_fs                = gfs2_sync_fs,
1602         .freeze_fs              = gfs2_freeze,
1603         .unfreeze_fs            = gfs2_unfreeze,
1604         .statfs                 = gfs2_statfs,
1605         .remount_fs             = gfs2_remount_fs,
1606         .drop_inode             = gfs2_drop_inode,
1607         .show_options           = gfs2_show_options,
1608 };
1609