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_rgrplvb,
  82         Opt_norgrplvb,
  83         Opt_error,
  84 };
  85
  86 static const match_table_t tokens = {
  87         {Opt_lockproto, "lockproto=%s"},
  88         {Opt_locktable, "locktable=%s"},
  89         {Opt_hostdata, "hostdata=%s"},
  90         {Opt_spectator, "spectator"},
  91         {Opt_spectator, "norecovery"},
  92         {Opt_ignore_local_fs, "ignore_local_fs"},
  93         {Opt_localflocks, "localflocks"},
  94         {Opt_localcaching, "localcaching"},
  95         {Opt_debug, "debug"},
  96         {Opt_nodebug, "nodebug"},
  97         {Opt_upgrade, "upgrade"},
  98         {Opt_acl, "acl"},
  99         {Opt_noacl, "noacl"},
 100         {Opt_quota_off, "quota=off"},
 101         {Opt_quota_account, "quota=account"},
 102         {Opt_quota_on, "quota=on"},
 103         {Opt_quota, "quota"},
 104         {Opt_noquota, "noquota"},
 105         {Opt_suiddir, "suiddir"},
 106         {Opt_nosuiddir, "nosuiddir"},
 107         {Opt_data_writeback, "data=writeback"},
 108         {Opt_data_ordered, "data=ordered"},
 109         {Opt_meta, "meta"},
 110         {Opt_discard, "discard"},
 111         {Opt_nodiscard, "nodiscard"},
 112         {Opt_commit, "commit=%d"},
 113         {Opt_err_withdraw, "errors=withdraw"},
 114         {Opt_err_panic, "errors=panic"},
 115         {Opt_statfs_quantum, "statfs_quantum=%d"},
 116         {Opt_statfs_percent, "statfs_percent=%d"},
 117         {Opt_quota_quantum, "quota_quantum=%d"},
 118         {Opt_barrier, "barrier"},
 119         {Opt_nobarrier, "nobarrier"},
 120         {Opt_rgrplvb, "rgrplvb"},
 121         {Opt_norgrplvb, "norgrplvb"},
 122         {Opt_error, NULL}
 123 };
 124
 125 /**
 126  * gfs2_mount_args - Parse mount options
 127  * @args: The structure into which the parsed options will be written
 128  * @options: The options to parse
 129  *
 130  * Return: errno
 131  */
 132
 133 int gfs2_mount_args(struct gfs2_args *args, char *options)
 134 {
 135         char *o;
 136         int token;
 137         substring_t tmp[MAX_OPT_ARGS];
 138         int rv;
 139
 140         /* Split the options into tokens with the "," character and
 141            process them */
 142
 143         while (1) {
 144                 o = strsep(&options, ",");
 145                 if (o == NULL)
 146                         break;
 147                 if (*o == '\0')
 148                         continue;
 149
 150                 token = match_token(o, tokens, tmp);
 151                 switch (token) {
 152                 case Opt_lockproto:
 153                         match_strlcpy(args->ar_lockproto, &tmp[0],
 154                                       GFS2_LOCKNAME_LEN);
 155                         break;
 156                 case Opt_locktable:
 157                         match_strlcpy(args->ar_locktable, &tmp[0],
 158                                       GFS2_LOCKNAME_LEN);
 159                         break;
 160                 case Opt_hostdata:
 161                         match_strlcpy(args->ar_hostdata, &tmp[0],
 162                                       GFS2_LOCKNAME_LEN);
 163                         break;
 164                 case Opt_spectator:
 165                         args->ar_spectator = 1;
 166                         break;
 167                 case Opt_ignore_local_fs:
 168                         /* Retained for backwards compat only */
 169                         break;
 170                 case Opt_localflocks:
 171                         args->ar_localflocks = 1;
 172                         break;
 173                 case Opt_localcaching:
 174                         /* Retained for backwards compat only */
 175                         break;
 176                 case Opt_debug:
 177                         if (args->ar_errors == GFS2_ERRORS_PANIC) {
 178                                 printk(KERN_WARNING "GFS2: -o debug and -o errors=panic "
 179                                        "are mutually exclusive.\n");
 180                                 return -EINVAL;
 181                         }
 182                         args->ar_debug = 1;
 183                         break;
 184                 case Opt_nodebug:
 185                         args->ar_debug = 0;
 186                         break;
 187                 case Opt_upgrade:
 188                         /* Retained for backwards compat only */
 189                         break;
 190                 case Opt_acl:
 191                         args->ar_posix_acl = 1;
 192                         break;
 193                 case Opt_noacl:
 194                         args->ar_posix_acl = 0;
 195                         break;
 196                 case Opt_quota_off:
 197                 case Opt_noquota:
 198                         args->ar_quota = GFS2_QUOTA_OFF;
 199                         break;
 200                 case Opt_quota_account:
 201                         args->ar_quota = GFS2_QUOTA_ACCOUNT;
 202                         break;
 203                 case Opt_quota_on:
 204                 case Opt_quota:
 205                         args->ar_quota = GFS2_QUOTA_ON;
 206                         break;
 207                 case Opt_suiddir:
 208                         args->ar_suiddir = 1;
 209                         break;
 210                 case Opt_nosuiddir:
 211                         args->ar_suiddir = 0;
 212                         break;
 213                 case Opt_data_writeback:
 214                         args->ar_data = GFS2_DATA_WRITEBACK;
 215                         break;
 216                 case Opt_data_ordered:
 217                         args->ar_data = GFS2_DATA_ORDERED;
 218                         break;
 219                 case Opt_meta:
 220                         args->ar_meta = 1;
 221                         break;
 222                 case Opt_discard:
 223                         args->ar_discard = 1;
 224                         break;
 225                 case Opt_nodiscard:
 226                         args->ar_discard = 0;
 227                         break;
 228                 case Opt_commit:
 229                         rv = match_int(&tmp[0], &args->ar_commit);
 230                         if (rv || args->ar_commit <= 0) {
 231                                 printk(KERN_WARNING "GFS2: commit mount option requires a positive numeric argument\n");
 232                                 return rv ? rv : -EINVAL;
 233                         }
 234                         break;
 235                 case Opt_statfs_quantum:
 236                         rv = match_int(&tmp[0], &args->ar_statfs_quantum);
 237                         if (rv || args->ar_statfs_quantum < 0) {
 238                                 printk(KERN_WARNING "GFS2: statfs_quantum mount option requires a non-negative numeric argument\n");
 239                                 return rv ? rv : -EINVAL;
 240                         }
 241                         break;
 242                 case Opt_quota_quantum:
 243                         rv = match_int(&tmp[0], &args->ar_quota_quantum);
 244                         if (rv || args->ar_quota_quantum <= 0) {
 245                                 printk(KERN_WARNING "GFS2: quota_quantum mount option requires a positive numeric argument\n");
 246                                 return rv ? rv : -EINVAL;
 247                         }
 248                         break;
 249                 case Opt_statfs_percent:
 250                         rv = match_int(&tmp[0], &args->ar_statfs_percent);
 251                         if (rv || args->ar_statfs_percent < 0 ||
 252                             args->ar_statfs_percent > 100) {
 253                                 printk(KERN_WARNING "statfs_percent mount option requires a numeric argument between 0 and 100\n");
 254                                 return rv ? rv : -EINVAL;
 255                         }
 256                         break;
 257                 case Opt_err_withdraw:
 258                         args->ar_errors = GFS2_ERRORS_WITHDRAW;
 259                         break;
 260                 case Opt_err_panic:
 261                         if (args->ar_debug) {
 262                                 printk(KERN_WARNING "GFS2: -o debug and -o errors=panic "
 263                                         "are mutually exclusive.\n");
 264                                 return -EINVAL;
 265                         }
 266                         args->ar_errors = GFS2_ERRORS_PANIC;
 267                         break;
 268                 case Opt_barrier:
 269                         args->ar_nobarrier = 0;
 270                         break;
 271                 case Opt_nobarrier:
 272                         args->ar_nobarrier = 1;
 273                         break;
 274                 case Opt_rgrplvb:
 275                         args->ar_rgrplvb = 1;
 276                         break;
 277                 case Opt_norgrplvb:
 278                         args->ar_rgrplvb = 0;
 279                         break;
 280                 case Opt_error:
 281                 default:
 282                         printk(KERN_WARNING "GFS2: invalid mount option: %s\n", o);
 283                         return -EINVAL;
 284                 }
 285         }
 286
 287         return 0;
 288 }
 289
 290 /**
 291  * gfs2_jindex_free - Clear all the journal index information
 292  * @sdp: The GFS2 superblock
 293  *
 294  */
 295
 296 void gfs2_jindex_free(struct gfs2_sbd *sdp)
 297 {
 298         struct list_head list, *head;
 299         struct gfs2_jdesc *jd;
 300         struct gfs2_journal_extent *jext;
 301
 302         spin_lock(&sdp->sd_jindex_spin);
 303         list_add(&list, &sdp->sd_jindex_list);
 304         list_del_init(&sdp->sd_jindex_list);
 305         sdp->sd_journals = 0;
 306         spin_unlock(&sdp->sd_jindex_spin);
 307
 308         while (!list_empty(&list)) {
 309                 jd = list_entry(list.next, struct gfs2_jdesc, jd_list);
 310                 head = &jd->extent_list;
 311                 while (!list_empty(head)) {
 312                         jext = list_entry(head->next,
 313                                           struct gfs2_journal_extent,
 314                                           extent_list);
 315                         list_del(&jext->extent_list);
 316                         kfree(jext);
 317                 }
 318                 list_del(&jd->jd_list);
 319                 iput(jd->jd_inode);
 320                 kfree(jd);
 321         }
 322 }
 323
 324 static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
 325 {
 326         struct gfs2_jdesc *jd;
 327         int found = 0;
 328
 329         list_for_each_entry(jd, head, jd_list) {
 330                 if (jd->jd_jid == jid) {
 331                         found = 1;
 332                         break;
 333                 }
 334         }
 335
 336         if (!found)
 337                 jd = NULL;
 338
 339         return jd;
 340 }
 341
 342 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
 343 {
 344         struct gfs2_jdesc *jd;
 345
 346         spin_lock(&sdp->sd_jindex_spin);
 347         jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
 348         spin_unlock(&sdp->sd_jindex_spin);
 349
 350         return jd;
 351 }
 352
 353 int gfs2_jdesc_check(struct gfs2_jdesc *jd)
 354 {
 355         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 356         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 357         u64 size = i_size_read(jd->jd_inode);
 358
 359         if (gfs2_check_internal_file_size(jd->jd_inode, 8 << 20, 1 << 30))
 360                 return -EIO;
 361
 362         jd->jd_blocks = size >> sdp->sd_sb.sb_bsize_shift;
 363
 364         if (gfs2_write_alloc_required(ip, 0, size)) {
 365                 gfs2_consist_inode(ip);
 366                 return -EIO;
 367         }
 368
 369         return 0;
 370 }
 371
 372 static int init_threads(struct gfs2_sbd *sdp)
 373 {
 374         struct task_struct *p;
 375         int error = 0;
 376
 377         p = kthread_run(gfs2_logd, sdp, "gfs2_logd");
 378         if (IS_ERR(p)) {
 379                 error = PTR_ERR(p);
 380                 fs_err(sdp, "can't start logd thread: %d\n", error);
 381                 return error;
 382         }
 383         sdp->sd_logd_process = p;
 384
 385         p = kthread_run(gfs2_quotad, sdp, "gfs2_quotad");
 386         if (IS_ERR(p)) {
 387                 error = PTR_ERR(p);
 388                 fs_err(sdp, "can't start quotad thread: %d\n", error);
 389                 goto fail;
 390         }
 391         sdp->sd_quotad_process = p;
 392         return 0;
 393
 394 fail:
 395         kthread_stop(sdp->sd_logd_process);
 396         return error;
 397 }
 398
 399 /**
 400  * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
 401  * @sdp: the filesystem
 402  *
 403  * Returns: errno
 404  */
 405
 406 int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
 407 {
 408         struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
 409         struct gfs2_glock *j_gl = ip->i_gl;
 410         struct gfs2_holder t_gh;
 411         struct gfs2_log_header_host head;
 412         int error;
 413
 414         error = init_threads(sdp);
 415         if (error)
 416                 return error;
 417
 418         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, 0, &t_gh);
 419         if (error)
 420                 goto fail_threads;
 421
 422         j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
 423
 424         error = gfs2_find_jhead(sdp->sd_jdesc, &head);
 425         if (error)
 426                 goto fail;
 427
 428         if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 429                 gfs2_consist(sdp);
 430                 error = -EIO;
 431                 goto fail;
 432         }
 433
 434         /*  Initialize some head of the log stuff  */
 435         sdp->sd_log_sequence = head.lh_sequence + 1;
 436         gfs2_log_pointers_init(sdp, head.lh_blkno);
 437
 438         error = gfs2_quota_init(sdp);
 439         if (error)
 440                 goto fail;
 441
 442         set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 443
 444         gfs2_glock_dq_uninit(&t_gh);
 445
 446         return 0;
 447
 448 fail:
 449         t_gh.gh_flags |= GL_NOCACHE;
 450         gfs2_glock_dq_uninit(&t_gh);
 451 fail_threads:
 452         kthread_stop(sdp->sd_quotad_process);
 453         kthread_stop(sdp->sd_logd_process);
 454         return error;
 455 }
 456
 457 void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
 458 {
 459         const struct gfs2_statfs_change *str = buf;
 460
 461         sc->sc_total = be64_to_cpu(str->sc_total);
 462         sc->sc_free = be64_to_cpu(str->sc_free);
 463         sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
 464 }
 465
 466 static void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
 467 {
 468         struct gfs2_statfs_change *str = buf;
 469
 470         str->sc_total = cpu_to_be64(sc->sc_total);
 471         str->sc_free = cpu_to_be64(sc->sc_free);
 472         str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
 473 }
 474
 475 int gfs2_statfs_init(struct gfs2_sbd *sdp)
 476 {
 477         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 478         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 479         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 480         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 481         struct buffer_head *m_bh, *l_bh;
 482         struct gfs2_holder gh;
 483         int error;
 484
 485         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 486                                    &gh);
 487         if (error)
 488                 return error;
 489
 490         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 491         if (error)
 492                 goto out;
 493
 494         if (sdp->sd_args.ar_spectator) {
 495                 spin_lock(&sdp->sd_statfs_spin);
 496                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 497                                       sizeof(struct gfs2_dinode));
 498                 spin_unlock(&sdp->sd_statfs_spin);
 499         } else {
 500                 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 501                 if (error)
 502                         goto out_m_bh;
 503
 504                 spin_lock(&sdp->sd_statfs_spin);
 505                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 506                                       sizeof(struct gfs2_dinode));
 507                 gfs2_statfs_change_in(l_sc, l_bh->b_data +
 508                                       sizeof(struct gfs2_dinode));
 509                 spin_unlock(&sdp->sd_statfs_spin);
 510
 511                 brelse(l_bh);
 512         }
 513
 514 out_m_bh:
 515         brelse(m_bh);
 516 out:
 517         gfs2_glock_dq_uninit(&gh);
 518         return 0;
 519 }
 520
 521 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
 522                         s64 dinodes)
 523 {
 524         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 525         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 526         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 527         struct buffer_head *l_bh;
 528         s64 x, y;
 529         int need_sync = 0;
 530         int error;
 531
 532         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 533         if (error)
 534                 return;
 535
 536         gfs2_trans_add_meta(l_ip->i_gl, l_bh);
 537
 538         spin_lock(&sdp->sd_statfs_spin);
 539         l_sc->sc_total += total;
 540         l_sc->sc_free += free;
 541         l_sc->sc_dinodes += dinodes;
 542         gfs2_statfs_change_out(l_sc, l_bh->b_data + sizeof(struct gfs2_dinode));
 543         if (sdp->sd_args.ar_statfs_percent) {
 544                 x = 100 * l_sc->sc_free;
 545                 y = m_sc->sc_free * sdp->sd_args.ar_statfs_percent;
 546                 if (x >= y || x <= -y)
 547                         need_sync = 1;
 548         }
 549         spin_unlock(&sdp->sd_statfs_spin);
 550
 551         brelse(l_bh);
 552         if (need_sync)
 553                 gfs2_wake_up_statfs(sdp);
 554 }
 555
 556 void update_statfs(struct gfs2_sbd *sdp, struct buffer_head *m_bh,
 557                    struct buffer_head *l_bh)
 558 {
 559         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 560         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 561         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 562         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 563
 564         gfs2_trans_add_meta(l_ip->i_gl, l_bh);
 565
 566         spin_lock(&sdp->sd_statfs_spin);
 567         m_sc->sc_total += l_sc->sc_total;
 568         m_sc->sc_free += l_sc->sc_free;
 569         m_sc->sc_dinodes += l_sc->sc_dinodes;
 570         memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
 571         memset(l_bh->b_data + sizeof(struct gfs2_dinode),
 572                0, sizeof(struct gfs2_statfs_change));
 573         spin_unlock(&sdp->sd_statfs_spin);
 574
 575         gfs2_trans_add_meta(m_ip->i_gl, m_bh);
 576         gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
 577 }
 578
 579 int gfs2_statfs_sync(struct super_block *sb, int type)
 580 {
 581         struct gfs2_sbd *sdp = sb->s_fs_info;
 582         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 583         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 584         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 585         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 586         struct gfs2_holder gh;
 587         struct buffer_head *m_bh, *l_bh;
 588         int error;
 589
 590         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 591                                    &gh);
 592         if (error)
 593                 return error;
 594
 595         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 596         if (error)
 597                 goto out;
 598
 599         spin_lock(&sdp->sd_statfs_spin);
 600         gfs2_statfs_change_in(m_sc, m_bh->b_data +
 601                               sizeof(struct gfs2_dinode));
 602         if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
 603                 spin_unlock(&sdp->sd_statfs_spin);
 604                 goto out_bh;
 605         }
 606         spin_unlock(&sdp->sd_statfs_spin);
 607
 608         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 609         if (error)
 610                 goto out_bh;
 611
 612         error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
 613         if (error)
 614                 goto out_bh2;
 615
 616         update_statfs(sdp, m_bh, l_bh);
 617         sdp->sd_statfs_force_sync = 0;
 618
 619         gfs2_trans_end(sdp);
 620
 621 out_bh2:
 622         brelse(l_bh);
 623 out_bh:
 624         brelse(m_bh);
 625 out:
 626         gfs2_glock_dq_uninit(&gh);
 627         return error;
 628 }
 629
 630 struct lfcc {
 631         struct list_head list;
 632         struct gfs2_holder gh;
 633 };
 634
 635 /**
 636  * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
 637  *                            journals are clean
 638  * @sdp: the file system
 639  * @state: the state to put the transaction lock into
 640  * @t_gh: the hold on the transaction lock
 641  *
 642  * Returns: errno
 643  */
 644
 645 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp,
 646                                     struct gfs2_holder *t_gh)
 647 {
 648         struct gfs2_inode *ip;
 649         struct gfs2_jdesc *jd;
 650         struct lfcc *lfcc;
 651         LIST_HEAD(list);
 652         struct gfs2_log_header_host lh;
 653         int error;
 654
 655         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 656                 lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
 657                 if (!lfcc) {
 658                         error = -ENOMEM;
 659                         goto out;
 660                 }
 661                 ip = GFS2_I(jd->jd_inode);
 662                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
 663                 if (error) {
 664                         kfree(lfcc);
 665                         goto out;
 666                 }
 667                 list_add(&lfcc->list, &list);
 668         }
 669
 670         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_DEFERRED,
 671                                    GL_NOCACHE, t_gh);
 672
 673         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 674                 error = gfs2_jdesc_check(jd);
 675                 if (error)
 676                         break;
 677                 error = gfs2_find_jhead(jd, &lh);
 678                 if (error)
 679                         break;
 680                 if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 681                         error = -EBUSY;
 682                         break;
 683                 }
 684         }
 685
 686         if (error)
 687                 gfs2_glock_dq_uninit(t_gh);
 688
 689 out:
 690         while (!list_empty(&list)) {
 691                 lfcc = list_entry(list.next, struct lfcc, list);
 692                 list_del(&lfcc->list);
 693                 gfs2_glock_dq_uninit(&lfcc->gh);
 694                 kfree(lfcc);
 695         }
 696         return error;
 697 }
 698
 699 void gfs2_dinode_out(const struct gfs2_inode *ip, void *buf)
 700 {
 701         struct gfs2_dinode *str = buf;
 702
 703         str->di_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 704         str->di_header.mh_type = cpu_to_be32(GFS2_METATYPE_DI);
 705         str->di_header.mh_format = cpu_to_be32(GFS2_FORMAT_DI);
 706         str->di_num.no_addr = cpu_to_be64(ip->i_no_addr);
 707         str->di_num.no_formal_ino = cpu_to_be64(ip->i_no_formal_ino);
 708         str->di_mode = cpu_to_be32(ip->i_inode.i_mode);
 709         str->di_uid = cpu_to_be32(i_uid_read(&ip->i_inode));
 710         str->di_gid = cpu_to_be32(i_gid_read(&ip->i_inode));
 711         str->di_nlink = cpu_to_be32(ip->i_inode.i_nlink);
 712         str->di_size = cpu_to_be64(i_size_read(&ip->i_inode));
 713         str->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(&ip->i_inode));
 714         str->di_atime = cpu_to_be64(ip->i_inode.i_atime.tv_sec);
 715         str->di_mtime = cpu_to_be64(ip->i_inode.i_mtime.tv_sec);
 716         str->di_ctime = cpu_to_be64(ip->i_inode.i_ctime.tv_sec);
 717
 718         str->di_goal_meta = cpu_to_be64(ip->i_goal);
 719         str->di_goal_data = cpu_to_be64(ip->i_goal);
 720         str->di_generation = cpu_to_be64(ip->i_generation);
 721
 722         str->di_flags = cpu_to_be32(ip->i_diskflags);
 723         str->di_height = cpu_to_be16(ip->i_height);
 724         str->di_payload_format = cpu_to_be32(S_ISDIR(ip->i_inode.i_mode) &&
 725                                              !(ip->i_diskflags & GFS2_DIF_EXHASH) ?
 726                                              GFS2_FORMAT_DE : 0);
 727         str->di_depth = cpu_to_be16(ip->i_depth);
 728         str->di_entries = cpu_to_be32(ip->i_entries);
 729
 730         str->di_eattr = cpu_to_be64(ip->i_eattr);
 731         str->di_atime_nsec = cpu_to_be32(ip->i_inode.i_atime.tv_nsec);
 732         str->di_mtime_nsec = cpu_to_be32(ip->i_inode.i_mtime.tv_nsec);
 733         str->di_ctime_nsec = cpu_to_be32(ip->i_inode.i_ctime.tv_nsec);
 734 }
 735
 736 /**
 737  * gfs2_write_inode - Make sure the inode is stable on the disk
 738  * @inode: The inode
 739  * @wbc: The writeback control structure
 740  *
 741  * Returns: errno
 742  */
 743
 744 static int gfs2_write_inode(struct inode *inode, struct writeback_control *wbc)
 745 {
 746         struct gfs2_inode *ip = GFS2_I(inode);
 747         struct gfs2_sbd *sdp = GFS2_SB(inode);
 748         struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl);
 749         struct backing_dev_info *bdi = metamapping->backing_dev_info;
 750         int ret = 0;
 751
 752         if (wbc->sync_mode == WB_SYNC_ALL)
 753                 gfs2_log_flush(GFS2_SB(inode), ip->i_gl);
 754         if (bdi->dirty_exceeded)
 755                 gfs2_ail1_flush(sdp, wbc);
 756         else
 757                 filemap_fdatawrite(metamapping);
 758         if (wbc->sync_mode == WB_SYNC_ALL)
 759                 ret = filemap_fdatawait(metamapping);
 760         if (ret)
 761                 mark_inode_dirty_sync(inode);
 762         return ret;
 763 }
 764
 765 /**
 766  * gfs2_dirty_inode - check for atime updates
 767  * @inode: The inode in question
 768  * @flags: The type of dirty
 769  *
 770  * Unfortunately it can be called under any combination of inode
 771  * glock and transaction lock, so we have to check carefully.
 772  *
 773  * At the moment this deals only with atime - it should be possible
 774  * to expand that role in future, once a review of the locking has
 775  * been carried out.
 776  */
 777
 778 static void gfs2_dirty_inode(struct inode *inode, int flags)
 779 {
 780         struct gfs2_inode *ip = GFS2_I(inode);
 781         struct gfs2_sbd *sdp = GFS2_SB(inode);
 782         struct buffer_head *bh;
 783         struct gfs2_holder gh;
 784         int need_unlock = 0;
 785         int need_endtrans = 0;
 786         int ret;
 787
 788         if (!(flags & (I_DIRTY_DATASYNC|I_DIRTY_SYNC)))
 789                 return;
 790
 791         if (!gfs2_glock_is_locked_by_me(ip->i_gl)) {
 792                 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
 793                 if (ret) {
 794                         fs_err(sdp, "dirty_inode: glock %d\n", ret);
 795                         return;
 796                 }
 797                 need_unlock = 1;
 798         } else if (WARN_ON_ONCE(ip->i_gl->gl_state != LM_ST_EXCLUSIVE))
 799                 return;
 800
 801         if (current->journal_info == NULL) {
 802                 ret = gfs2_trans_begin(sdp, RES_DINODE, 0);
 803                 if (ret) {
 804                         fs_err(sdp, "dirty_inode: gfs2_trans_begin %d\n", ret);
 805                         goto out;
 806                 }
 807                 need_endtrans = 1;
 808         }
 809
 810         ret = gfs2_meta_inode_buffer(ip, &bh);
 811         if (ret == 0) {
 812                 gfs2_trans_add_meta(ip->i_gl, bh);
 813                 gfs2_dinode_out(ip, bh->b_data);
 814                 brelse(bh);
 815         }
 816
 817         if (need_endtrans)
 818                 gfs2_trans_end(sdp);
 819 out:
 820         if (need_unlock)
 821                 gfs2_glock_dq_uninit(&gh);
 822 }
 823
 824 /**
 825  * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
 826  * @sdp: the filesystem
 827  *
 828  * Returns: errno
 829  */
 830
 831 static int gfs2_make_fs_ro(struct gfs2_sbd *sdp)
 832 {
 833         struct gfs2_holder t_gh;
 834         int error;
 835
 836         kthread_stop(sdp->sd_quotad_process);
 837         kthread_stop(sdp->sd_logd_process);
 838
 839         flush_workqueue(gfs2_delete_workqueue);
 840         gfs2_quota_sync(sdp->sd_vfs, 0);
 841         gfs2_statfs_sync(sdp->sd_vfs, 0);
 842
 843         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, GL_NOCACHE,
 844                                    &t_gh);
 845         if (error && !test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
 846                 return error;
 847
 848         gfs2_meta_syncfs(sdp);
 849         gfs2_log_shutdown(sdp);
 850
 851         clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 852
 853         if (t_gh.gh_gl)
 854                 gfs2_glock_dq_uninit(&t_gh);
 855
 856         gfs2_quota_cleanup(sdp);
 857
 858         return error;
 859 }
 860
 861 static int gfs2_umount_recovery_wait(void *word)
 862 {
 863         schedule();
 864         return 0;
 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                             gfs2_umount_recovery_wait, TASK_UNINTERRUPTIBLE);
 892                 goto restart;
 893         }
 894         spin_unlock(&sdp->sd_jindex_spin);
 895
 896         if (!(sb->s_flags & MS_RDONLY)) {
 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_trans_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         /*  Unmount the locking protocol  */
 928         gfs2_lm_unmount(sdp);
 929
 930         /*  At this point, we're through participating in the lockspace  */
 931         gfs2_sys_fs_del(sdp);
 932 }
 933
 934 /**
 935  * gfs2_sync_fs - sync the filesystem
 936  * @sb: the superblock
 937  *
 938  * Flushes the log to disk.
 939  */
 940
 941 static int gfs2_sync_fs(struct super_block *sb, int wait)
 942 {
 943         struct gfs2_sbd *sdp = sb->s_fs_info;
 944
 945         gfs2_quota_sync(sb, -1);
 946         if (wait && sdp)
 947                 gfs2_log_flush(sdp, NULL);
 948         return 0;
 949 }
 950
 951 /**
 952  * gfs2_freeze - prevent further writes to the filesystem
 953  * @sb: the VFS structure for the filesystem
 954  *
 955  */
 956
 957 static int gfs2_freeze(struct super_block *sb)
 958 {
 959         struct gfs2_sbd *sdp = sb->s_fs_info;
 960         int error;
 961
 962         if (test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
 963                 return -EINVAL;
 964
 965         for (;;) {
 966                 error = gfs2_lock_fs_check_clean(sdp, &sdp->sd_freeze_gh);
 967                 if (!error)
 968                         break;
 969
 970                 switch (error) {
 971                 case -EBUSY:
 972                         fs_err(sdp, "waiting for recovery before freeze\n");
 973                         break;
 974
 975                 default:
 976                         fs_err(sdp, "error freezing FS: %d\n", error);
 977                         break;
 978                 }
 979
 980                 fs_err(sdp, "retrying...\n");
 981                 msleep(1000);
 982         }
 983         return 0;
 984 }
 985
 986 /**
 987  * gfs2_unfreeze - reallow writes to the filesystem
 988  * @sb: the VFS structure for the filesystem
 989  *
 990  */
 991
 992 static int gfs2_unfreeze(struct super_block *sb)
 993 {
 994         struct gfs2_sbd *sdp = sb->s_fs_info;
 995
 996         gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
 997         return 0;
 998 }
 999
1000 /**
1001  * statfs_fill - fill in the sg for a given RG
1002  * @rgd: the RG
1003  * @sc: the sc structure
1004  *
1005  * Returns: 0 on success, -ESTALE if the LVB is invalid
1006  */
1007
1008 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
1009                             struct gfs2_statfs_change_host *sc)
1010 {
1011         gfs2_rgrp_verify(rgd);
1012         sc->sc_total += rgd->rd_data;
1013         sc->sc_free += rgd->rd_free;
1014         sc->sc_dinodes += rgd->rd_dinodes;
1015         return 0;
1016 }
1017
1018 /**
1019  * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
1020  * @sdp: the filesystem
1021  * @sc: the sc info that will be returned
1022  *
1023  * Any error (other than a signal) will cause this routine to fall back
1024  * to the synchronous version.
1025  *
1026  * FIXME: This really shouldn't busy wait like this.
1027  *
1028  * Returns: errno
1029  */
1030
1031 static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
1032 {
1033         struct gfs2_rgrpd *rgd_next;
1034         struct gfs2_holder *gha, *gh;
1035         unsigned int slots = 64;
1036         unsigned int x;
1037         int done;
1038         int error = 0, err;
1039
1040         memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
1041         gha = kcalloc(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
1042         if (!gha)
1043                 return -ENOMEM;
1044
1045         rgd_next = gfs2_rgrpd_get_first(sdp);
1046
1047         for (;;) {
1048                 done = 1;
1049
1050                 for (x = 0; x < slots; x++) {
1051                         gh = gha + x;
1052
1053                         if (gh->gh_gl && gfs2_glock_poll(gh)) {
1054                                 err = gfs2_glock_wait(gh);
1055                                 if (err) {
1056                                         gfs2_holder_uninit(gh);
1057                                         error = err;
1058                                 } else {
1059                                         if (!error)
1060                                                 error = statfs_slow_fill(
1061                                                         gh->gh_gl->gl_object, sc);
1062                                         gfs2_glock_dq_uninit(gh);
1063                                 }
1064                         }
1065
1066                         if (gh->gh_gl)
1067                                 done = 0;
1068                         else if (rgd_next && !error) {
1069                                 error = gfs2_glock_nq_init(rgd_next->rd_gl,
1070                                                            LM_ST_SHARED,
1071                                                            GL_ASYNC,
1072                                                            gh);
1073                                 rgd_next = gfs2_rgrpd_get_next(rgd_next);
1074                                 done = 0;
1075                         }
1076
1077                         if (signal_pending(current))
1078                                 error = -ERESTARTSYS;
1079                 }
1080
1081                 if (done)
1082                         break;
1083
1084                 yield();
1085         }
1086
1087         kfree(gha);
1088         return error;
1089 }
1090
1091 /**
1092  * gfs2_statfs_i - Do a statfs
1093  * @sdp: the filesystem
1094  * @sg: the sg structure
1095  *
1096  * Returns: errno
1097  */
1098
1099 static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
1100 {
1101         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
1102         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
1103
1104         spin_lock(&sdp->sd_statfs_spin);
1105
1106         *sc = *m_sc;
1107         sc->sc_total += l_sc->sc_total;
1108         sc->sc_free += l_sc->sc_free;
1109         sc->sc_dinodes += l_sc->sc_dinodes;
1110
1111         spin_unlock(&sdp->sd_statfs_spin);
1112
1113         if (sc->sc_free < 0)
1114                 sc->sc_free = 0;
1115         if (sc->sc_free > sc->sc_total)
1116                 sc->sc_free = sc->sc_total;
1117         if (sc->sc_dinodes < 0)
1118                 sc->sc_dinodes = 0;
1119
1120         return 0;
1121 }
1122
1123 /**
1124  * gfs2_statfs - Gather and return stats about the filesystem
1125  * @sb: The superblock
1126  * @statfsbuf: The buffer
1127  *
1128  * Returns: 0 on success or error code
1129  */
1130
1131 static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
1132 {
1133         struct super_block *sb = dentry->d_inode->i_sb;
1134         struct gfs2_sbd *sdp = sb->s_fs_info;
1135         struct gfs2_statfs_change_host sc;
1136         int error;
1137
1138         error = gfs2_rindex_update(sdp);
1139         if (error)
1140                 return error;
1141
1142         if (gfs2_tune_get(sdp, gt_statfs_slow))
1143                 error = gfs2_statfs_slow(sdp, &sc);
1144         else
1145                 error = gfs2_statfs_i(sdp, &sc);
1146
1147         if (error)
1148                 return error;
1149
1150         buf->f_type = GFS2_MAGIC;
1151         buf->f_bsize = sdp->sd_sb.sb_bsize;
1152         buf->f_blocks = sc.sc_total;
1153         buf->f_bfree = sc.sc_free;
1154         buf->f_bavail = sc.sc_free;
1155         buf->f_files = sc.sc_dinodes + sc.sc_free;
1156         buf->f_ffree = sc.sc_free;
1157         buf->f_namelen = GFS2_FNAMESIZE;
1158
1159         return 0;
1160 }
1161
1162 /**
1163  * gfs2_remount_fs - called when the FS is remounted
1164  * @sb:  the filesystem
1165  * @flags:  the remount flags
1166  * @data:  extra data passed in (not used right now)
1167  *
1168  * Returns: errno
1169  */
1170
1171 static int gfs2_remount_fs(struct super_block *sb, int *flags, char *data)
1172 {
1173         struct gfs2_sbd *sdp = sb->s_fs_info;
1174         struct gfs2_args args = sdp->sd_args; /* Default to current settings */
1175         struct gfs2_tune *gt = &sdp->sd_tune;
1176         int error;
1177
1178         spin_lock(&gt->gt_spin);
1179         args.ar_commit = gt->gt_logd_secs;
1180         args.ar_quota_quantum = gt->gt_quota_quantum;
1181         if (gt->gt_statfs_slow)
1182                 args.ar_statfs_quantum = 0;
1183         else
1184                 args.ar_statfs_quantum = gt->gt_statfs_quantum;
1185         spin_unlock(&gt->gt_spin);
1186         error = gfs2_mount_args(&args, data);
1187         if (error)
1188                 return error;
1189
1190         /* Not allowed to change locking details */
1191         if (strcmp(args.ar_lockproto, sdp->sd_args.ar_lockproto) ||
1192             strcmp(args.ar_locktable, sdp->sd_args.ar_locktable) ||
1193             strcmp(args.ar_hostdata, sdp->sd_args.ar_hostdata))
1194                 return -EINVAL;
1195
1196         /* Some flags must not be changed */
1197         if (args_neq(&args, &sdp->sd_args, spectator) ||
1198             args_neq(&args, &sdp->sd_args, localflocks) ||
1199             args_neq(&args, &sdp->sd_args, meta))
1200                 return -EINVAL;
1201
1202         if (sdp->sd_args.ar_spectator)
1203                 *flags |= MS_RDONLY;
1204
1205         if ((sb->s_flags ^ *flags) & MS_RDONLY) {
1206                 if (*flags & MS_RDONLY)
1207                         error = gfs2_make_fs_ro(sdp);
1208                 else
1209                         error = gfs2_make_fs_rw(sdp);
1210                 if (error)
1211                         return error;
1212         }
1213
1214         sdp->sd_args = args;
1215         if (sdp->sd_args.ar_posix_acl)
1216                 sb->s_flags |= MS_POSIXACL;
1217         else
1218                 sb->s_flags &= ~MS_POSIXACL;
1219         if (sdp->sd_args.ar_nobarrier)
1220                 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1221         else
1222                 clear_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1223         spin_lock(&gt->gt_spin);
1224         gt->gt_logd_secs = args.ar_commit;
1225         gt->gt_quota_quantum = args.ar_quota_quantum;
1226         if (args.ar_statfs_quantum) {
1227                 gt->gt_statfs_slow = 0;
1228                 gt->gt_statfs_quantum = args.ar_statfs_quantum;
1229         }
1230         else {
1231                 gt->gt_statfs_slow = 1;
1232                 gt->gt_statfs_quantum = 30;
1233         }
1234         spin_unlock(&gt->gt_spin);
1235
1236         gfs2_online_uevent(sdp);
1237         return 0;
1238 }
1239
1240 /**
1241  * gfs2_drop_inode - Drop an inode (test for remote unlink)
1242  * @inode: The inode to drop
1243  *
1244  * If we've received a callback on an iopen lock then its because a
1245  * remote node tried to deallocate the inode but failed due to this node
1246  * still having the inode open. Here we mark the link count zero
1247  * since we know that it must have reached zero if the GLF_DEMOTE flag
1248  * is set on the iopen glock. If we didn't do a disk read since the
1249  * remote node removed the final link then we might otherwise miss
1250  * this event. This check ensures that this node will deallocate the
1251  * inode's blocks, or alternatively pass the baton on to another
1252  * node for later deallocation.
1253  */
1254
1255 static int gfs2_drop_inode(struct inode *inode)
1256 {
1257         struct gfs2_inode *ip = GFS2_I(inode);
1258
1259         if (inode->i_nlink) {
1260                 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1261                 if (gl && test_bit(GLF_DEMOTE, &gl->gl_flags))
1262                         clear_nlink(inode);
1263         }
1264         return generic_drop_inode(inode);
1265 }
1266
1267 static int is_ancestor(const struct dentry *d1, const struct dentry *d2)
1268 {
1269         do {
1270                 if (d1 == d2)
1271                         return 1;
1272                 d1 = d1->d_parent;
1273         } while (!IS_ROOT(d1));
1274         return 0;
1275 }
1276
1277 /**
1278  * gfs2_show_options - Show mount options for /proc/mounts
1279  * @s: seq_file structure
1280  * @root: root of this (sub)tree
1281  *
1282  * Returns: 0 on success or error code
1283  */
1284
1285 static int gfs2_show_options(struct seq_file *s, struct dentry *root)
1286 {
1287         struct gfs2_sbd *sdp = root->d_sb->s_fs_info;
1288         struct gfs2_args *args = &sdp->sd_args;
1289         int val;
1290
1291         if (is_ancestor(root, sdp->sd_master_dir))
1292                 seq_printf(s, ",meta");
1293         if (args->ar_lockproto[0])
1294                 seq_printf(s, ",lockproto=%s", args->ar_lockproto);
1295         if (args->ar_locktable[0])
1296                 seq_printf(s, ",locktable=%s", args->ar_locktable);
1297         if (args->ar_hostdata[0])
1298                 seq_printf(s, ",hostdata=%s", args->ar_hostdata);
1299         if (args->ar_spectator)
1300                 seq_printf(s, ",spectator");
1301         if (args->ar_localflocks)
1302                 seq_printf(s, ",localflocks");
1303         if (args->ar_debug)
1304                 seq_printf(s, ",debug");
1305         if (args->ar_posix_acl)
1306                 seq_printf(s, ",acl");
1307         if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
1308                 char *state;
1309                 switch (args->ar_quota) {
1310                 case GFS2_QUOTA_OFF:
1311                         state = "off";
1312                         break;
1313                 case GFS2_QUOTA_ACCOUNT:
1314                         state = "account";
1315                         break;
1316                 case GFS2_QUOTA_ON:
1317                         state = "on";
1318                         break;
1319                 default:
1320                         state = "unknown";
1321                         break;
1322                 }
1323                 seq_printf(s, ",quota=%s", state);
1324         }
1325         if (args->ar_suiddir)
1326                 seq_printf(s, ",suiddir");
1327         if (args->ar_data != GFS2_DATA_DEFAULT) {
1328                 char *state;
1329                 switch (args->ar_data) {
1330                 case GFS2_DATA_WRITEBACK:
1331                         state = "writeback";
1332                         break;
1333                 case GFS2_DATA_ORDERED:
1334                         state = "ordered";
1335                         break;
1336                 default:
1337                         state = "unknown";
1338                         break;
1339                 }
1340                 seq_printf(s, ",data=%s", state);
1341         }
1342         if (args->ar_discard)
1343                 seq_printf(s, ",discard");
1344         val = sdp->sd_tune.gt_logd_secs;
1345         if (val != 30)
1346                 seq_printf(s, ",commit=%d", val);
1347         val = sdp->sd_tune.gt_statfs_quantum;
1348         if (val != 30)
1349                 seq_printf(s, ",statfs_quantum=%d", val);
1350         else if (sdp->sd_tune.gt_statfs_slow)
1351                 seq_puts(s, ",statfs_quantum=0");
1352         val = sdp->sd_tune.gt_quota_quantum;
1353         if (val != 60)
1354                 seq_printf(s, ",quota_quantum=%d", val);
1355         if (args->ar_statfs_percent)
1356                 seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent);
1357         if (args->ar_errors != GFS2_ERRORS_DEFAULT) {
1358                 const char *state;
1359
1360                 switch (args->ar_errors) {
1361                 case GFS2_ERRORS_WITHDRAW:
1362                         state = "withdraw";
1363                         break;
1364                 case GFS2_ERRORS_PANIC:
1365                         state = "panic";
1366                         break;
1367                 default:
1368                         state = "unknown";
1369                         break;
1370                 }
1371                 seq_printf(s, ",errors=%s", state);
1372         }
1373         if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags))
1374                 seq_printf(s, ",nobarrier");
1375         if (test_bit(SDF_DEMOTE, &sdp->sd_flags))
1376                 seq_printf(s, ",demote_interface_used");
1377         if (args->ar_rgrplvb)
1378                 seq_printf(s, ",rgrplvb");
1379         return 0;
1380 }
1381
1382 static void gfs2_final_release_pages(struct gfs2_inode *ip)
1383 {
1384         struct inode *inode = &ip->i_inode;
1385         struct gfs2_glock *gl = ip->i_gl;
1386
1387         truncate_inode_pages(gfs2_glock2aspace(ip->i_gl), 0);
1388         truncate_inode_pages(&inode->i_data, 0);
1389
1390         if (atomic_read(&gl->gl_revokes) == 0) {
1391                 clear_bit(GLF_LFLUSH, &gl->gl_flags);
1392                 clear_bit(GLF_DIRTY, &gl->gl_flags);
1393         }
1394 }
1395
1396 static int gfs2_dinode_dealloc(struct gfs2_inode *ip)
1397 {
1398         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1399         struct gfs2_rgrpd *rgd;
1400         struct gfs2_holder gh;
1401         int error;
1402
1403         if (gfs2_get_inode_blocks(&ip->i_inode) != 1) {
1404                 gfs2_consist_inode(ip);
1405                 return -EIO;
1406         }
1407
1408         error = gfs2_rindex_update(sdp);
1409         if (error)
1410                 return error;
1411
1412         error = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
1413         if (error)
1414                 return error;
1415
1416         rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1417         if (!rgd) {
1418                 gfs2_consist_inode(ip);
1419                 error = -EIO;
1420                 goto out_qs;
1421         }
1422
1423         error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &gh);
1424         if (error)
1425                 goto out_qs;
1426
1427         error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA,
1428                                  sdp->sd_jdesc->jd_blocks);
1429         if (error)
1430                 goto out_rg_gunlock;
1431
1432         gfs2_free_di(rgd, ip);
1433
1434         gfs2_final_release_pages(ip);
1435
1436         gfs2_trans_end(sdp);
1437
1438 out_rg_gunlock:
1439         gfs2_glock_dq_uninit(&gh);
1440 out_qs:
1441         gfs2_quota_unhold(ip);
1442         return error;
1443 }
1444
1445 /**
1446  * gfs2_evict_inode - Remove an inode from cache
1447  * @inode: The inode to evict
1448  *
1449  * There are three cases to consider:
1450  * 1. i_nlink == 0, we are final opener (and must deallocate)
1451  * 2. i_nlink == 0, we are not the final opener (and cannot deallocate)
1452  * 3. i_nlink > 0
1453  *
1454  * If the fs is read only, then we have to treat all cases as per #3
1455  * since we are unable to do any deallocation. The inode will be
1456  * deallocated by the next read/write node to attempt an allocation
1457  * in the same resource group
1458  *
1459  * We have to (at the moment) hold the inodes main lock to cover
1460  * the gap between unlocking the shared lock on the iopen lock and
1461  * taking the exclusive lock. I'd rather do a shared -> exclusive
1462  * conversion on the iopen lock, but we can change that later. This
1463  * is safe, just less efficient.
1464  */
1465
1466 static void gfs2_evict_inode(struct inode *inode)
1467 {
1468         struct super_block *sb = inode->i_sb;
1469         struct gfs2_sbd *sdp = sb->s_fs_info;
1470         struct gfs2_inode *ip = GFS2_I(inode);
1471         struct gfs2_holder gh;
1472         int error;
1473
1474         if (inode->i_nlink || (sb->s_flags & MS_RDONLY))
1475                 goto out;
1476
1477         /* Must not read inode block until block type has been verified */
1478         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, &gh);
1479         if (unlikely(error)) {
1480                 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1481                 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1482                 goto out;
1483         }
1484
1485         if (!test_bit(GIF_ALLOC_FAILED, &ip->i_flags)) {
1486                 error = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED);
1487                 if (error)
1488                         goto out_truncate;
1489         }
1490
1491         if (test_bit(GIF_INVALID, &ip->i_flags)) {
1492                 error = gfs2_inode_refresh(ip);
1493                 if (error)
1494                         goto out_truncate;
1495         }
1496
1497         ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1498         gfs2_glock_dq_wait(&ip->i_iopen_gh);
1499         gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE, &ip->i_iopen_gh);
1500         error = gfs2_glock_nq(&ip->i_iopen_gh);
1501         if (error)
1502                 goto out_truncate;
1503
1504         /* Case 1 starts here */
1505
1506         if (S_ISDIR(inode->i_mode) &&
1507             (ip->i_diskflags & GFS2_DIF_EXHASH)) {
1508                 error = gfs2_dir_exhash_dealloc(ip);
1509                 if (error)
1510                         goto out_unlock;
1511         }
1512
1513         if (ip->i_eattr) {
1514                 error = gfs2_ea_dealloc(ip);
1515                 if (error)
1516                         goto out_unlock;
1517         }
1518
1519         if (!gfs2_is_stuffed(ip)) {
1520                 error = gfs2_file_dealloc(ip);
1521                 if (error)
1522                         goto out_unlock;
1523         }
1524
1525         error = gfs2_dinode_dealloc(ip);
1526         goto out_unlock;
1527
1528 out_truncate:
1529         gfs2_log_flush(sdp, ip->i_gl);
1530         if (test_bit(GLF_DIRTY, &ip->i_gl->gl_flags)) {
1531                 struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl);
1532                 filemap_fdatawrite(metamapping);
1533                 filemap_fdatawait(metamapping);
1534         }
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 (gfs2_rs_active(ip->i_res))
1549                 gfs2_rs_deltree(ip->i_res);
1550
1551         if (test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1552                 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1553                 gfs2_glock_dq(&ip->i_iopen_gh);
1554         }
1555         gfs2_holder_uninit(&ip->i_iopen_gh);
1556         gfs2_glock_dq_uninit(&gh);
1557         if (error && error != GLR_TRYFAILED && error != -EROFS)
1558                 fs_warn(sdp, "gfs2_evict_inode: %d\n", error);
1559 out:
1560         /* Case 3 starts here */
1561         truncate_inode_pages(&inode->i_data, 0);
1562         gfs2_rs_delete(ip, NULL);
1563         gfs2_ordered_del_inode(ip);
1564         clear_inode(inode);
1565         gfs2_dir_hash_inval(ip);
1566         ip->i_gl->gl_object = NULL;
1567         flush_delayed_work(&ip->i_gl->gl_work);
1568         gfs2_glock_add_to_lru(ip->i_gl);
1569         gfs2_glock_put(ip->i_gl);
1570         ip->i_gl = NULL;
1571         if (ip->i_iopen_gh.gh_gl) {
1572                 ip->i_iopen_gh.gh_gl->gl_object = NULL;
1573                 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1574                 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1575         }
1576 }
1577
1578 static struct inode *gfs2_alloc_inode(struct super_block *sb)
1579 {
1580         struct gfs2_inode *ip;
1581
1582         ip = kmem_cache_alloc(gfs2_inode_cachep, GFP_KERNEL);
1583         if (ip) {
1584                 ip->i_flags = 0;
1585                 ip->i_gl = NULL;
1586                 ip->i_rgd = NULL;
1587                 ip->i_res = NULL;
1588         }
1589         return &ip->i_inode;
1590 }
1591
1592 static void gfs2_i_callback(struct rcu_head *head)
1593 {
1594         struct inode *inode = container_of(head, struct inode, i_rcu);
1595         kmem_cache_free(gfs2_inode_cachep, inode);
1596 }
1597
1598 static void gfs2_destroy_inode(struct inode *inode)
1599 {
1600         call_rcu(&inode->i_rcu, gfs2_i_callback);
1601 }
1602
1603 const struct super_operations gfs2_super_ops = {
1604         .alloc_inode            = gfs2_alloc_inode,
1605         .destroy_inode          = gfs2_destroy_inode,
1606         .write_inode            = gfs2_write_inode,
1607         .dirty_inode            = gfs2_dirty_inode,
1608         .evict_inode            = gfs2_evict_inode,
1609         .put_super              = gfs2_put_super,
1610         .sync_fs                = gfs2_sync_fs,
1611         .freeze_fs              = gfs2_freeze,
1612         .unfreeze_fs            = gfs2_unfreeze,
1613         .statfs                 = gfs2_statfs,
1614         .remount_fs             = gfs2_remount_fs,
1615         .drop_inode             = gfs2_drop_inode,
1616         .show_options           = gfs2_show_options,
1617 };
1618