fs/gfs2/super.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   4  * Copyright (C) 2004-2007 Red Hat, Inc.  All rights reserved.
   5  */
   6
   7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
   8
   9 #include <linux/bio.h>
  10 #include <linux/sched/signal.h>
  11 #include <linux/slab.h>
  12 #include <linux/spinlock.h>
  13 #include <linux/completion.h>
  14 #include <linux/buffer_head.h>
  15 #include <linux/statfs.h>
  16 #include <linux/seq_file.h>
  17 #include <linux/mount.h>
  18 #include <linux/kthread.h>
  19 #include <linux/delay.h>
  20 #include <linux/gfs2_ondisk.h>
  21 #include <linux/crc32.h>
  22 #include <linux/time.h>
  23 #include <linux/wait.h>
  24 #include <linux/writeback.h>
  25 #include <linux/backing-dev.h>
  26 #include <linux/kernel.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 #include "lops.h"
  46
  47 enum dinode_demise {
  48         SHOULD_DELETE_DINODE,
  49         SHOULD_NOT_DELETE_DINODE,
  50         SHOULD_DEFER_EVICTION,
  51 };
  52
  53 /**
  54  * gfs2_jindex_free - Clear all the journal index information
  55  * @sdp: The GFS2 superblock
  56  *
  57  */
  58
  59 void gfs2_jindex_free(struct gfs2_sbd *sdp)
  60 {
  61         struct list_head list;
  62         struct gfs2_jdesc *jd;
  63
  64         spin_lock(&sdp->sd_jindex_spin);
  65         list_add(&list, &sdp->sd_jindex_list);
  66         list_del_init(&sdp->sd_jindex_list);
  67         sdp->sd_journals = 0;
  68         spin_unlock(&sdp->sd_jindex_spin);
  69
  70         sdp->sd_jdesc = NULL;
  71         while (!list_empty(&list)) {
  72                 jd = list_first_entry(&list, struct gfs2_jdesc, jd_list);
  73                 gfs2_free_journal_extents(jd);
  74                 list_del(&jd->jd_list);
  75                 iput(jd->jd_inode);
  76                 jd->jd_inode = NULL;
  77                 kfree(jd);
  78         }
  79 }
  80
  81 static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
  82 {
  83         struct gfs2_jdesc *jd;
  84         int found = 0;
  85
  86         list_for_each_entry(jd, head, jd_list) {
  87                 if (jd->jd_jid == jid) {
  88                         found = 1;
  89                         break;
  90                 }
  91         }
  92
  93         if (!found)
  94                 jd = NULL;
  95
  96         return jd;
  97 }
  98
  99 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
 100 {
 101         struct gfs2_jdesc *jd;
 102
 103         spin_lock(&sdp->sd_jindex_spin);
 104         jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
 105         spin_unlock(&sdp->sd_jindex_spin);
 106
 107         return jd;
 108 }
 109
 110 int gfs2_jdesc_check(struct gfs2_jdesc *jd)
 111 {
 112         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 113         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 114         u64 size = i_size_read(jd->jd_inode);
 115
 116         if (gfs2_check_internal_file_size(jd->jd_inode, 8 << 20, BIT(30)))
 117                 return -EIO;
 118
 119         jd->jd_blocks = size >> sdp->sd_sb.sb_bsize_shift;
 120
 121         if (gfs2_write_alloc_required(ip, 0, size)) {
 122                 gfs2_consist_inode(ip);
 123                 return -EIO;
 124         }
 125
 126         return 0;
 127 }
 128
 129 static int init_threads(struct gfs2_sbd *sdp)
 130 {
 131         struct task_struct *p;
 132         int error = 0;
 133
 134         p = kthread_run(gfs2_logd, sdp, "gfs2_logd");
 135         if (IS_ERR(p)) {
 136                 error = PTR_ERR(p);
 137                 fs_err(sdp, "can't start logd thread: %d\n", error);
 138                 return error;
 139         }
 140         sdp->sd_logd_process = p;
 141
 142         p = kthread_run(gfs2_quotad, sdp, "gfs2_quotad");
 143         if (IS_ERR(p)) {
 144                 error = PTR_ERR(p);
 145                 fs_err(sdp, "can't start quotad thread: %d\n", error);
 146                 goto fail;
 147         }
 148         sdp->sd_quotad_process = p;
 149         return 0;
 150
 151 fail:
 152         kthread_stop(sdp->sd_logd_process);
 153         sdp->sd_logd_process = NULL;
 154         return error;
 155 }
 156
 157 /**
 158  * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
 159  * @sdp: the filesystem
 160  *
 161  * Returns: errno
 162  */
 163
 164 int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
 165 {
 166         struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
 167         struct gfs2_glock *j_gl = ip->i_gl;
 168         struct gfs2_holder freeze_gh;
 169         struct gfs2_log_header_host head;
 170         int error;
 171
 172         error = init_threads(sdp);
 173         if (error)
 174                 return error;
 175
 176         error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED,
 177                                    LM_FLAG_NOEXP | GL_EXACT,
 178                                    &freeze_gh);
 179         if (error)
 180                 goto fail_threads;
 181
 182         j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
 183         if (gfs2_withdrawn(sdp)) {
 184                 error = -EIO;
 185                 goto fail;
 186         }
 187
 188         error = gfs2_find_jhead(sdp->sd_jdesc, &head, false);
 189         if (error || gfs2_withdrawn(sdp))
 190                 goto fail;
 191
 192         if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 193                 gfs2_consist(sdp);
 194                 error = -EIO;
 195                 goto fail;
 196         }
 197
 198         /*  Initialize some head of the log stuff  */
 199         sdp->sd_log_sequence = head.lh_sequence + 1;
 200         gfs2_log_pointers_init(sdp, head.lh_blkno);
 201
 202         error = gfs2_quota_init(sdp);
 203         if (error || gfs2_withdrawn(sdp))
 204                 goto fail;
 205
 206         set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 207
 208         gfs2_glock_dq_uninit(&freeze_gh);
 209
 210         return 0;
 211
 212 fail:
 213         gfs2_glock_dq_uninit(&freeze_gh);
 214 fail_threads:
 215         if (sdp->sd_quotad_process)
 216                 kthread_stop(sdp->sd_quotad_process);
 217         sdp->sd_quotad_process = NULL;
 218         if (sdp->sd_logd_process)
 219                 kthread_stop(sdp->sd_logd_process);
 220         sdp->sd_logd_process = NULL;
 221         return error;
 222 }
 223
 224 void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
 225 {
 226         const struct gfs2_statfs_change *str = buf;
 227
 228         sc->sc_total = be64_to_cpu(str->sc_total);
 229         sc->sc_free = be64_to_cpu(str->sc_free);
 230         sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
 231 }
 232
 233 static void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
 234 {
 235         struct gfs2_statfs_change *str = buf;
 236
 237         str->sc_total = cpu_to_be64(sc->sc_total);
 238         str->sc_free = cpu_to_be64(sc->sc_free);
 239         str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
 240 }
 241
 242 int gfs2_statfs_init(struct gfs2_sbd *sdp)
 243 {
 244         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 245         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 246         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 247         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 248         struct buffer_head *m_bh, *l_bh;
 249         struct gfs2_holder gh;
 250         int error;
 251
 252         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 253                                    &gh);
 254         if (error)
 255                 return error;
 256
 257         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 258         if (error)
 259                 goto out;
 260
 261         if (sdp->sd_args.ar_spectator) {
 262                 spin_lock(&sdp->sd_statfs_spin);
 263                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 264                                       sizeof(struct gfs2_dinode));
 265                 spin_unlock(&sdp->sd_statfs_spin);
 266         } else {
 267                 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 268                 if (error)
 269                         goto out_m_bh;
 270
 271                 spin_lock(&sdp->sd_statfs_spin);
 272                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 273                                       sizeof(struct gfs2_dinode));
 274                 gfs2_statfs_change_in(l_sc, l_bh->b_data +
 275                                       sizeof(struct gfs2_dinode));
 276                 spin_unlock(&sdp->sd_statfs_spin);
 277
 278                 brelse(l_bh);
 279         }
 280
 281 out_m_bh:
 282         brelse(m_bh);
 283 out:
 284         gfs2_glock_dq_uninit(&gh);
 285         return 0;
 286 }
 287
 288 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
 289                         s64 dinodes)
 290 {
 291         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 292         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 293         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 294         struct buffer_head *l_bh;
 295         s64 x, y;
 296         int need_sync = 0;
 297         int error;
 298
 299         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 300         if (error)
 301                 return;
 302
 303         gfs2_trans_add_meta(l_ip->i_gl, l_bh);
 304
 305         spin_lock(&sdp->sd_statfs_spin);
 306         l_sc->sc_total += total;
 307         l_sc->sc_free += free;
 308         l_sc->sc_dinodes += dinodes;
 309         gfs2_statfs_change_out(l_sc, l_bh->b_data + sizeof(struct gfs2_dinode));
 310         if (sdp->sd_args.ar_statfs_percent) {
 311                 x = 100 * l_sc->sc_free;
 312                 y = m_sc->sc_free * sdp->sd_args.ar_statfs_percent;
 313                 if (x >= y || x <= -y)
 314                         need_sync = 1;
 315         }
 316         spin_unlock(&sdp->sd_statfs_spin);
 317
 318         brelse(l_bh);
 319         if (need_sync)
 320                 gfs2_wake_up_statfs(sdp);
 321 }
 322
 323 void update_statfs(struct gfs2_sbd *sdp, struct buffer_head *m_bh,
 324                    struct buffer_head *l_bh)
 325 {
 326         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 327         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 328         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 329         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 330
 331         gfs2_trans_add_meta(l_ip->i_gl, l_bh);
 332         gfs2_trans_add_meta(m_ip->i_gl, m_bh);
 333
 334         spin_lock(&sdp->sd_statfs_spin);
 335         m_sc->sc_total += l_sc->sc_total;
 336         m_sc->sc_free += l_sc->sc_free;
 337         m_sc->sc_dinodes += l_sc->sc_dinodes;
 338         memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
 339         memset(l_bh->b_data + sizeof(struct gfs2_dinode),
 340                0, sizeof(struct gfs2_statfs_change));
 341         gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
 342         spin_unlock(&sdp->sd_statfs_spin);
 343 }
 344
 345 int gfs2_statfs_sync(struct super_block *sb, int type)
 346 {
 347         struct gfs2_sbd *sdp = sb->s_fs_info;
 348         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 349         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 350         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 351         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 352         struct gfs2_holder gh;
 353         struct buffer_head *m_bh, *l_bh;
 354         int error;
 355
 356         sb_start_write(sb);
 357         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 358                                    &gh);
 359         if (error)
 360                 goto out;
 361
 362         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 363         if (error)
 364                 goto out_unlock;
 365
 366         spin_lock(&sdp->sd_statfs_spin);
 367         gfs2_statfs_change_in(m_sc, m_bh->b_data +
 368                               sizeof(struct gfs2_dinode));
 369         if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
 370                 spin_unlock(&sdp->sd_statfs_spin);
 371                 goto out_bh;
 372         }
 373         spin_unlock(&sdp->sd_statfs_spin);
 374
 375         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 376         if (error)
 377                 goto out_bh;
 378
 379         error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
 380         if (error)
 381                 goto out_bh2;
 382
 383         update_statfs(sdp, m_bh, l_bh);
 384         sdp->sd_statfs_force_sync = 0;
 385
 386         gfs2_trans_end(sdp);
 387
 388 out_bh2:
 389         brelse(l_bh);
 390 out_bh:
 391         brelse(m_bh);
 392 out_unlock:
 393         gfs2_glock_dq_uninit(&gh);
 394 out:
 395         sb_end_write(sb);
 396         return error;
 397 }
 398
 399 struct lfcc {
 400         struct list_head list;
 401         struct gfs2_holder gh;
 402 };
 403
 404 /**
 405  * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
 406  *                            journals are clean
 407  * @sdp: the file system
 408  * @state: the state to put the transaction lock into
 409  * @t_gh: the hold on the transaction lock
 410  *
 411  * Returns: errno
 412  */
 413
 414 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp)
 415 {
 416         struct gfs2_inode *ip;
 417         struct gfs2_jdesc *jd;
 418         struct lfcc *lfcc;
 419         LIST_HEAD(list);
 420         struct gfs2_log_header_host lh;
 421         int error;
 422
 423         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 424                 lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
 425                 if (!lfcc) {
 426                         error = -ENOMEM;
 427                         goto out;
 428                 }
 429                 ip = GFS2_I(jd->jd_inode);
 430                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
 431                 if (error) {
 432                         kfree(lfcc);
 433                         goto out;
 434                 }
 435                 list_add(&lfcc->list, &list);
 436         }
 437
 438         error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_EXCLUSIVE,
 439                                    LM_FLAG_NOEXP, &sdp->sd_freeze_gh);
 440         if (error)
 441                 goto out;
 442
 443         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 444                 error = gfs2_jdesc_check(jd);
 445                 if (error)
 446                         break;
 447                 error = gfs2_find_jhead(jd, &lh, false);
 448                 if (error)
 449                         break;
 450                 if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 451                         error = -EBUSY;
 452                         break;
 453                 }
 454         }
 455
 456         if (error)
 457                 gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
 458
 459 out:
 460         while (!list_empty(&list)) {
 461                 lfcc = list_first_entry(&list, struct lfcc, list);
 462                 list_del(&lfcc->list);
 463                 gfs2_glock_dq_uninit(&lfcc->gh);
 464                 kfree(lfcc);
 465         }
 466         return error;
 467 }
 468
 469 void gfs2_dinode_out(const struct gfs2_inode *ip, void *buf)
 470 {
 471         struct gfs2_dinode *str = buf;
 472
 473         str->di_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 474         str->di_header.mh_type = cpu_to_be32(GFS2_METATYPE_DI);
 475         str->di_header.mh_format = cpu_to_be32(GFS2_FORMAT_DI);
 476         str->di_num.no_addr = cpu_to_be64(ip->i_no_addr);
 477         str->di_num.no_formal_ino = cpu_to_be64(ip->i_no_formal_ino);
 478         str->di_mode = cpu_to_be32(ip->i_inode.i_mode);
 479         str->di_uid = cpu_to_be32(i_uid_read(&ip->i_inode));
 480         str->di_gid = cpu_to_be32(i_gid_read(&ip->i_inode));
 481         str->di_nlink = cpu_to_be32(ip->i_inode.i_nlink);
 482         str->di_size = cpu_to_be64(i_size_read(&ip->i_inode));
 483         str->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(&ip->i_inode));
 484         str->di_atime = cpu_to_be64(ip->i_inode.i_atime.tv_sec);
 485         str->di_mtime = cpu_to_be64(ip->i_inode.i_mtime.tv_sec);
 486         str->di_ctime = cpu_to_be64(ip->i_inode.i_ctime.tv_sec);
 487
 488         str->di_goal_meta = cpu_to_be64(ip->i_goal);
 489         str->di_goal_data = cpu_to_be64(ip->i_goal);
 490         str->di_generation = cpu_to_be64(ip->i_generation);
 491
 492         str->di_flags = cpu_to_be32(ip->i_diskflags);
 493         str->di_height = cpu_to_be16(ip->i_height);
 494         str->di_payload_format = cpu_to_be32(S_ISDIR(ip->i_inode.i_mode) &&
 495                                              !(ip->i_diskflags & GFS2_DIF_EXHASH) ?
 496                                              GFS2_FORMAT_DE : 0);
 497         str->di_depth = cpu_to_be16(ip->i_depth);
 498         str->di_entries = cpu_to_be32(ip->i_entries);
 499
 500         str->di_eattr = cpu_to_be64(ip->i_eattr);
 501         str->di_atime_nsec = cpu_to_be32(ip->i_inode.i_atime.tv_nsec);
 502         str->di_mtime_nsec = cpu_to_be32(ip->i_inode.i_mtime.tv_nsec);
 503         str->di_ctime_nsec = cpu_to_be32(ip->i_inode.i_ctime.tv_nsec);
 504 }
 505
 506 /**
 507  * gfs2_write_inode - Make sure the inode is stable on the disk
 508  * @inode: The inode
 509  * @wbc: The writeback control structure
 510  *
 511  * Returns: errno
 512  */
 513
 514 static int gfs2_write_inode(struct inode *inode, struct writeback_control *wbc)
 515 {
 516         struct gfs2_inode *ip = GFS2_I(inode);
 517         struct gfs2_sbd *sdp = GFS2_SB(inode);
 518         struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl);
 519         struct backing_dev_info *bdi = inode_to_bdi(metamapping->host);
 520         int ret = 0;
 521         bool flush_all = (wbc->sync_mode == WB_SYNC_ALL || gfs2_is_jdata(ip));
 522
 523         if (flush_all)
 524                 gfs2_log_flush(GFS2_SB(inode), ip->i_gl,
 525                                GFS2_LOG_HEAD_FLUSH_NORMAL |
 526                                GFS2_LFC_WRITE_INODE);
 527         if (bdi->wb.dirty_exceeded)
 528                 gfs2_ail1_flush(sdp, wbc);
 529         else
 530                 filemap_fdatawrite(metamapping);
 531         if (flush_all)
 532                 ret = filemap_fdatawait(metamapping);
 533         if (ret)
 534                 mark_inode_dirty_sync(inode);
 535         else {
 536                 spin_lock(&inode->i_lock);
 537                 if (!(inode->i_flags & I_DIRTY))
 538                         gfs2_ordered_del_inode(ip);
 539                 spin_unlock(&inode->i_lock);
 540         }
 541         return ret;
 542 }
 543
 544 /**
 545  * gfs2_dirty_inode - check for atime updates
 546  * @inode: The inode in question
 547  * @flags: The type of dirty
 548  *
 549  * Unfortunately it can be called under any combination of inode
 550  * glock and transaction lock, so we have to check carefully.
 551  *
 552  * At the moment this deals only with atime - it should be possible
 553  * to expand that role in future, once a review of the locking has
 554  * been carried out.
 555  */
 556
 557 static void gfs2_dirty_inode(struct inode *inode, int flags)
 558 {
 559         struct gfs2_inode *ip = GFS2_I(inode);
 560         struct gfs2_sbd *sdp = GFS2_SB(inode);
 561         struct buffer_head *bh;
 562         struct gfs2_holder gh;
 563         int need_unlock = 0;
 564         int need_endtrans = 0;
 565         int ret;
 566
 567         if (!(flags & I_DIRTY_INODE))
 568                 return;
 569         if (unlikely(gfs2_withdrawn(sdp)))
 570                 return;
 571         if (!gfs2_glock_is_locked_by_me(ip->i_gl)) {
 572                 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
 573                 if (ret) {
 574                         fs_err(sdp, "dirty_inode: glock %d\n", ret);
 575                         gfs2_dump_glock(NULL, ip->i_gl, true);
 576                         return;
 577                 }
 578                 need_unlock = 1;
 579         } else if (WARN_ON_ONCE(ip->i_gl->gl_state != LM_ST_EXCLUSIVE))
 580                 return;
 581
 582         if (current->journal_info == NULL) {
 583                 ret = gfs2_trans_begin(sdp, RES_DINODE, 0);
 584                 if (ret) {
 585                         fs_err(sdp, "dirty_inode: gfs2_trans_begin %d\n", ret);
 586                         goto out;
 587                 }
 588                 need_endtrans = 1;
 589         }
 590
 591         ret = gfs2_meta_inode_buffer(ip, &bh);
 592         if (ret == 0) {
 593                 gfs2_trans_add_meta(ip->i_gl, bh);
 594                 gfs2_dinode_out(ip, bh->b_data);
 595                 brelse(bh);
 596         }
 597
 598         if (need_endtrans)
 599                 gfs2_trans_end(sdp);
 600 out:
 601         if (need_unlock)
 602                 gfs2_glock_dq_uninit(&gh);
 603 }
 604
 605 /**
 606  * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
 607  * @sdp: the filesystem
 608  *
 609  * Returns: errno
 610  */
 611
 612 int gfs2_make_fs_ro(struct gfs2_sbd *sdp)
 613 {
 614         struct gfs2_holder freeze_gh;
 615         int error = 0;
 616         int log_write_allowed = test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 617
 618         gfs2_holder_mark_uninitialized(&freeze_gh);
 619         if (sdp->sd_freeze_gl &&
 620             !gfs2_glock_is_locked_by_me(sdp->sd_freeze_gl)) {
 621                 if (!log_write_allowed) {
 622                         error = gfs2_glock_nq_init(sdp->sd_freeze_gl,
 623                                                    LM_ST_SHARED, LM_FLAG_TRY |
 624                                                    LM_FLAG_NOEXP | GL_EXACT,
 625                                                    &freeze_gh);
 626                         if (error == GLR_TRYFAILED)
 627                                 error = 0;
 628                 } else {
 629                         error = gfs2_glock_nq_init(sdp->sd_freeze_gl,
 630                                                    LM_ST_SHARED,
 631                                                    LM_FLAG_NOEXP | GL_EXACT,
 632                                                    &freeze_gh);
 633                         if (error && !gfs2_withdrawn(sdp))
 634                                 return error;
 635                 }
 636         }
 637
 638         gfs2_flush_delete_work(sdp);
 639         if (!log_write_allowed && current == sdp->sd_quotad_process)
 640                 fs_warn(sdp, "The quotad daemon is withdrawing.\n");
 641         else if (sdp->sd_quotad_process)
 642                 kthread_stop(sdp->sd_quotad_process);
 643         sdp->sd_quotad_process = NULL;
 644
 645         if (!log_write_allowed && current == sdp->sd_logd_process)
 646                 fs_warn(sdp, "The logd daemon is withdrawing.\n");
 647         else if (sdp->sd_logd_process)
 648                 kthread_stop(sdp->sd_logd_process);
 649         sdp->sd_logd_process = NULL;
 650
 651         if (log_write_allowed) {
 652                 gfs2_quota_sync(sdp->sd_vfs, 0);
 653                 gfs2_statfs_sync(sdp->sd_vfs, 0);
 654
 655                 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SHUTDOWN |
 656                                GFS2_LFC_MAKE_FS_RO);
 657                 wait_event(sdp->sd_reserving_log_wait,
 658                            atomic_read(&sdp->sd_reserving_log) == 0);
 659                 gfs2_assert_warn(sdp, atomic_read(&sdp->sd_log_blks_free) ==
 660                                  sdp->sd_jdesc->jd_blocks);
 661         } else {
 662                 wait_event_timeout(sdp->sd_reserving_log_wait,
 663                                    atomic_read(&sdp->sd_reserving_log) == 0,
 664                                    HZ * 5);
 665         }
 666         if (gfs2_holder_initialized(&freeze_gh))
 667                 gfs2_glock_dq_uninit(&freeze_gh);
 668
 669         gfs2_quota_cleanup(sdp);
 670
 671         if (!log_write_allowed)
 672                 sdp->sd_vfs->s_flags |= SB_RDONLY;
 673
 674         return error;
 675 }
 676
 677 /**
 678  * gfs2_put_super - Unmount the filesystem
 679  * @sb: The VFS superblock
 680  *
 681  */
 682
 683 static void gfs2_put_super(struct super_block *sb)
 684 {
 685         struct gfs2_sbd *sdp = sb->s_fs_info;
 686         int error;
 687         struct gfs2_jdesc *jd;
 688
 689         /* No more recovery requests */
 690         set_bit(SDF_NORECOVERY, &sdp->sd_flags);
 691         smp_mb();
 692
 693         /* Wait on outstanding recovery */
 694 restart:
 695         spin_lock(&sdp->sd_jindex_spin);
 696         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 697                 if (!test_bit(JDF_RECOVERY, &jd->jd_flags))
 698                         continue;
 699                 spin_unlock(&sdp->sd_jindex_spin);
 700                 wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
 701                             TASK_UNINTERRUPTIBLE);
 702                 goto restart;
 703         }
 704         spin_unlock(&sdp->sd_jindex_spin);
 705
 706         if (!sb_rdonly(sb)) {
 707                 error = gfs2_make_fs_ro(sdp);
 708                 if (error)
 709                         gfs2_io_error(sdp);
 710         }
 711         WARN_ON(gfs2_withdrawing(sdp));
 712
 713         /*  At this point, we're through modifying the disk  */
 714
 715         /*  Release stuff  */
 716
 717         iput(sdp->sd_jindex);
 718         iput(sdp->sd_statfs_inode);
 719         iput(sdp->sd_rindex);
 720         iput(sdp->sd_quota_inode);
 721
 722         gfs2_glock_put(sdp->sd_rename_gl);
 723         gfs2_glock_put(sdp->sd_freeze_gl);
 724
 725         if (!sdp->sd_args.ar_spectator) {
 726                 if (gfs2_holder_initialized(&sdp->sd_journal_gh))
 727                         gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
 728                 if (gfs2_holder_initialized(&sdp->sd_jinode_gh))
 729                         gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
 730                 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
 731                 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
 732                 iput(sdp->sd_sc_inode);
 733                 iput(sdp->sd_qc_inode);
 734         }
 735
 736         gfs2_glock_dq_uninit(&sdp->sd_live_gh);
 737         gfs2_clear_rgrpd(sdp);
 738         gfs2_jindex_free(sdp);
 739         /*  Take apart glock structures and buffer lists  */
 740         gfs2_gl_hash_clear(sdp);
 741         gfs2_delete_debugfs_file(sdp);
 742         /*  Unmount the locking protocol  */
 743         gfs2_lm_unmount(sdp);
 744
 745         /*  At this point, we're through participating in the lockspace  */
 746         gfs2_sys_fs_del(sdp);
 747 }
 748
 749 /**
 750  * gfs2_sync_fs - sync the filesystem
 751  * @sb: the superblock
 752  *
 753  * Flushes the log to disk.
 754  */
 755
 756 static int gfs2_sync_fs(struct super_block *sb, int wait)
 757 {
 758         struct gfs2_sbd *sdp = sb->s_fs_info;
 759
 760         gfs2_quota_sync(sb, -1);
 761         if (wait)
 762                 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
 763                                GFS2_LFC_SYNC_FS);
 764         return sdp->sd_log_error;
 765 }
 766
 767 void gfs2_freeze_func(struct work_struct *work)
 768 {
 769         int error;
 770         struct gfs2_holder freeze_gh;
 771         struct gfs2_sbd *sdp = container_of(work, struct gfs2_sbd, sd_freeze_work);
 772         struct super_block *sb = sdp->sd_vfs;
 773
 774         atomic_inc(&sb->s_active);
 775         error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED,
 776                                    LM_FLAG_NOEXP | GL_EXACT, &freeze_gh);
 777         if (error) {
 778                 fs_info(sdp, "GFS2: couldn't get freeze lock : %d\n", error);
 779                 gfs2_assert_withdraw(sdp, 0);
 780         } else {
 781                 atomic_set(&sdp->sd_freeze_state, SFS_UNFROZEN);
 782                 error = thaw_super(sb);
 783                 if (error) {
 784                         fs_info(sdp, "GFS2: couldn't thaw filesystem: %d\n",
 785                                 error);
 786                         gfs2_assert_withdraw(sdp, 0);
 787                 }
 788                 gfs2_glock_dq_uninit(&freeze_gh);
 789         }
 790         deactivate_super(sb);
 791         clear_bit_unlock(SDF_FS_FROZEN, &sdp->sd_flags);
 792         wake_up_bit(&sdp->sd_flags, SDF_FS_FROZEN);
 793         return;
 794 }
 795
 796 /**
 797  * gfs2_freeze - prevent further writes to the filesystem
 798  * @sb: the VFS structure for the filesystem
 799  *
 800  */
 801
 802 static int gfs2_freeze(struct super_block *sb)
 803 {
 804         struct gfs2_sbd *sdp = sb->s_fs_info;
 805         int error = 0;
 806
 807         mutex_lock(&sdp->sd_freeze_mutex);
 808         if (atomic_read(&sdp->sd_freeze_state) != SFS_UNFROZEN)
 809                 goto out;
 810
 811         for (;;) {
 812                 if (gfs2_withdrawn(sdp)) {
 813                         error = -EINVAL;
 814                         goto out;
 815                 }
 816
 817                 error = gfs2_lock_fs_check_clean(sdp);
 818                 if (!error)
 819                         break;
 820
 821                 if (error == -EBUSY)
 822                         fs_err(sdp, "waiting for recovery before freeze\n");
 823                 else if (error == -EIO) {
 824                         fs_err(sdp, "Fatal IO error: cannot freeze gfs2 due "
 825                                "to recovery error.\n");
 826                         goto out;
 827                 } else {
 828                         fs_err(sdp, "error freezing FS: %d\n", error);
 829                 }
 830                 fs_err(sdp, "retrying...\n");
 831                 msleep(1000);
 832         }
 833         set_bit(SDF_FS_FROZEN, &sdp->sd_flags);
 834 out:
 835         mutex_unlock(&sdp->sd_freeze_mutex);
 836         return error;
 837 }
 838
 839 /**
 840  * gfs2_unfreeze - reallow writes to the filesystem
 841  * @sb: the VFS structure for the filesystem
 842  *
 843  */
 844
 845 static int gfs2_unfreeze(struct super_block *sb)
 846 {
 847         struct gfs2_sbd *sdp = sb->s_fs_info;
 848
 849         mutex_lock(&sdp->sd_freeze_mutex);
 850         if (atomic_read(&sdp->sd_freeze_state) != SFS_FROZEN ||
 851             !gfs2_holder_initialized(&sdp->sd_freeze_gh)) {
 852                 mutex_unlock(&sdp->sd_freeze_mutex);
 853                 return 0;
 854         }
 855
 856         gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
 857         mutex_unlock(&sdp->sd_freeze_mutex);
 858         return wait_on_bit(&sdp->sd_flags, SDF_FS_FROZEN, TASK_INTERRUPTIBLE);
 859 }
 860
 861 /**
 862  * statfs_fill - fill in the sg for a given RG
 863  * @rgd: the RG
 864  * @sc: the sc structure
 865  *
 866  * Returns: 0 on success, -ESTALE if the LVB is invalid
 867  */
 868
 869 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
 870                             struct gfs2_statfs_change_host *sc)
 871 {
 872         gfs2_rgrp_verify(rgd);
 873         sc->sc_total += rgd->rd_data;
 874         sc->sc_free += rgd->rd_free;
 875         sc->sc_dinodes += rgd->rd_dinodes;
 876         return 0;
 877 }
 878
 879 /**
 880  * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
 881  * @sdp: the filesystem
 882  * @sc: the sc info that will be returned
 883  *
 884  * Any error (other than a signal) will cause this routine to fall back
 885  * to the synchronous version.
 886  *
 887  * FIXME: This really shouldn't busy wait like this.
 888  *
 889  * Returns: errno
 890  */
 891
 892 static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
 893 {
 894         struct gfs2_rgrpd *rgd_next;
 895         struct gfs2_holder *gha, *gh;
 896         unsigned int slots = 64;
 897         unsigned int x;
 898         int done;
 899         int error = 0, err;
 900
 901         memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
 902         gha = kmalloc_array(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
 903         if (!gha)
 904                 return -ENOMEM;
 905         for (x = 0; x < slots; x++)
 906                 gfs2_holder_mark_uninitialized(gha + x);
 907
 908         rgd_next = gfs2_rgrpd_get_first(sdp);
 909
 910         for (;;) {
 911                 done = 1;
 912
 913                 for (x = 0; x < slots; x++) {
 914                         gh = gha + x;
 915
 916                         if (gfs2_holder_initialized(gh) && gfs2_glock_poll(gh)) {
 917                                 err = gfs2_glock_wait(gh);
 918                                 if (err) {
 919                                         gfs2_holder_uninit(gh);
 920                                         error = err;
 921                                 } else {
 922                                         if (!error) {
 923                                                 struct gfs2_rgrpd *rgd =
 924                                                         gfs2_glock2rgrp(gh->gh_gl);
 925
 926                                                 error = statfs_slow_fill(rgd, sc);
 927                                         }
 928                                         gfs2_glock_dq_uninit(gh);
 929                                 }
 930                         }
 931
 932                         if (gfs2_holder_initialized(gh))
 933                                 done = 0;
 934                         else if (rgd_next && !error) {
 935                                 error = gfs2_glock_nq_init(rgd_next->rd_gl,
 936                                                            LM_ST_SHARED,
 937                                                            GL_ASYNC,
 938                                                            gh);
 939                                 rgd_next = gfs2_rgrpd_get_next(rgd_next);
 940                                 done = 0;
 941                         }
 942
 943                         if (signal_pending(current))
 944                                 error = -ERESTARTSYS;
 945                 }
 946
 947                 if (done)
 948                         break;
 949
 950                 yield();
 951         }
 952
 953         kfree(gha);
 954         return error;
 955 }
 956
 957 /**
 958  * gfs2_statfs_i - Do a statfs
 959  * @sdp: the filesystem
 960  * @sg: the sg structure
 961  *
 962  * Returns: errno
 963  */
 964
 965 static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
 966 {
 967         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 968         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 969
 970         spin_lock(&sdp->sd_statfs_spin);
 971
 972         *sc = *m_sc;
 973         sc->sc_total += l_sc->sc_total;
 974         sc->sc_free += l_sc->sc_free;
 975         sc->sc_dinodes += l_sc->sc_dinodes;
 976
 977         spin_unlock(&sdp->sd_statfs_spin);
 978
 979         if (sc->sc_free < 0)
 980                 sc->sc_free = 0;
 981         if (sc->sc_free > sc->sc_total)
 982                 sc->sc_free = sc->sc_total;
 983         if (sc->sc_dinodes < 0)
 984                 sc->sc_dinodes = 0;
 985
 986         return 0;
 987 }
 988
 989 /**
 990  * gfs2_statfs - Gather and return stats about the filesystem
 991  * @sb: The superblock
 992  * @statfsbuf: The buffer
 993  *
 994  * Returns: 0 on success or error code
 995  */
 996
 997 static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
 998 {
 999         struct super_block *sb = dentry->d_sb;
1000         struct gfs2_sbd *sdp = sb->s_fs_info;
1001         struct gfs2_statfs_change_host sc;
1002         int error;
1003
1004         error = gfs2_rindex_update(sdp);
1005         if (error)
1006                 return error;
1007
1008         if (gfs2_tune_get(sdp, gt_statfs_slow))
1009                 error = gfs2_statfs_slow(sdp, &sc);
1010         else
1011                 error = gfs2_statfs_i(sdp, &sc);
1012
1013         if (error)
1014                 return error;
1015
1016         buf->f_type = GFS2_MAGIC;
1017         buf->f_bsize = sdp->sd_sb.sb_bsize;
1018         buf->f_blocks = sc.sc_total;
1019         buf->f_bfree = sc.sc_free;
1020         buf->f_bavail = sc.sc_free;
1021         buf->f_files = sc.sc_dinodes + sc.sc_free;
1022         buf->f_ffree = sc.sc_free;
1023         buf->f_namelen = GFS2_FNAMESIZE;
1024
1025         return 0;
1026 }
1027
1028 /**
1029  * gfs2_drop_inode - Drop an inode (test for remote unlink)
1030  * @inode: The inode to drop
1031  *
1032  * If we've received a callback on an iopen lock then it's because a
1033  * remote node tried to deallocate the inode but failed due to this node
1034  * still having the inode open. Here we mark the link count zero
1035  * since we know that it must have reached zero if the GLF_DEMOTE flag
1036  * is set on the iopen glock. If we didn't do a disk read since the
1037  * remote node removed the final link then we might otherwise miss
1038  * this event. This check ensures that this node will deallocate the
1039  * inode's blocks, or alternatively pass the baton on to another
1040  * node for later deallocation.
1041  */
1042
1043 static int gfs2_drop_inode(struct inode *inode)
1044 {
1045         struct gfs2_inode *ip = GFS2_I(inode);
1046
1047         if (!test_bit(GIF_FREE_VFS_INODE, &ip->i_flags) &&
1048             inode->i_nlink &&
1049             gfs2_holder_initialized(&ip->i_iopen_gh)) {
1050                 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1051                 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
1052                         clear_nlink(inode);
1053         }
1054
1055         /*
1056          * When under memory pressure when an inode's link count has dropped to
1057          * zero, defer deleting the inode to the delete workqueue.  This avoids
1058          * calling into DLM under memory pressure, which can deadlock.
1059          */
1060         if (!inode->i_nlink &&
1061             unlikely(current->flags & PF_MEMALLOC) &&
1062             gfs2_holder_initialized(&ip->i_iopen_gh)) {
1063                 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1064
1065                 gfs2_glock_hold(gl);
1066                 if (!gfs2_queue_delete_work(gl, 0))
1067                         gfs2_glock_queue_put(gl);
1068                 return false;
1069         }
1070
1071         return generic_drop_inode(inode);
1072 }
1073
1074 static int is_ancestor(const struct dentry *d1, const struct dentry *d2)
1075 {
1076         do {
1077                 if (d1 == d2)
1078                         return 1;
1079                 d1 = d1->d_parent;
1080         } while (!IS_ROOT(d1));
1081         return 0;
1082 }
1083
1084 /**
1085  * gfs2_show_options - Show mount options for /proc/mounts
1086  * @s: seq_file structure
1087  * @root: root of this (sub)tree
1088  *
1089  * Returns: 0 on success or error code
1090  */
1091
1092 static int gfs2_show_options(struct seq_file *s, struct dentry *root)
1093 {
1094         struct gfs2_sbd *sdp = root->d_sb->s_fs_info;
1095         struct gfs2_args *args = &sdp->sd_args;
1096         int val;
1097
1098         if (is_ancestor(root, sdp->sd_master_dir))
1099                 seq_puts(s, ",meta");
1100         if (args->ar_lockproto[0])
1101                 seq_show_option(s, "lockproto", args->ar_lockproto);
1102         if (args->ar_locktable[0])
1103                 seq_show_option(s, "locktable", args->ar_locktable);
1104         if (args->ar_hostdata[0])
1105                 seq_show_option(s, "hostdata", args->ar_hostdata);
1106         if (args->ar_spectator)
1107                 seq_puts(s, ",spectator");
1108         if (args->ar_localflocks)
1109                 seq_puts(s, ",localflocks");
1110         if (args->ar_debug)
1111                 seq_puts(s, ",debug");
1112         if (args->ar_posix_acl)
1113                 seq_puts(s, ",acl");
1114         if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
1115                 char *state;
1116                 switch (args->ar_quota) {
1117                 case GFS2_QUOTA_OFF:
1118                         state = "off";
1119                         break;
1120                 case GFS2_QUOTA_ACCOUNT:
1121                         state = "account";
1122                         break;
1123                 case GFS2_QUOTA_ON:
1124                         state = "on";
1125                         break;
1126                 default:
1127                         state = "unknown";
1128                         break;
1129                 }
1130                 seq_printf(s, ",quota=%s", state);
1131         }
1132         if (args->ar_suiddir)
1133                 seq_puts(s, ",suiddir");
1134         if (args->ar_data != GFS2_DATA_DEFAULT) {
1135                 char *state;
1136                 switch (args->ar_data) {
1137                 case GFS2_DATA_WRITEBACK:
1138                         state = "writeback";
1139                         break;
1140                 case GFS2_DATA_ORDERED:
1141                         state = "ordered";
1142                         break;
1143                 default:
1144                         state = "unknown";
1145                         break;
1146                 }
1147                 seq_printf(s, ",data=%s", state);
1148         }
1149         if (args->ar_discard)
1150                 seq_puts(s, ",discard");
1151         val = sdp->sd_tune.gt_logd_secs;
1152         if (val != 30)
1153                 seq_printf(s, ",commit=%d", val);
1154         val = sdp->sd_tune.gt_statfs_quantum;
1155         if (val != 30)
1156                 seq_printf(s, ",statfs_quantum=%d", val);
1157         else if (sdp->sd_tune.gt_statfs_slow)
1158                 seq_puts(s, ",statfs_quantum=0");
1159         val = sdp->sd_tune.gt_quota_quantum;
1160         if (val != 60)
1161                 seq_printf(s, ",quota_quantum=%d", val);
1162         if (args->ar_statfs_percent)
1163                 seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent);
1164         if (args->ar_errors != GFS2_ERRORS_DEFAULT) {
1165                 const char *state;
1166
1167                 switch (args->ar_errors) {
1168                 case GFS2_ERRORS_WITHDRAW:
1169                         state = "withdraw";
1170                         break;
1171                 case GFS2_ERRORS_PANIC:
1172                         state = "panic";
1173                         break;
1174                 default:
1175                         state = "unknown";
1176                         break;
1177                 }
1178                 seq_printf(s, ",errors=%s", state);
1179         }
1180         if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags))
1181                 seq_puts(s, ",nobarrier");
1182         if (test_bit(SDF_DEMOTE, &sdp->sd_flags))
1183                 seq_puts(s, ",demote_interface_used");
1184         if (args->ar_rgrplvb)
1185                 seq_puts(s, ",rgrplvb");
1186         if (args->ar_loccookie)
1187                 seq_puts(s, ",loccookie");
1188         return 0;
1189 }
1190
1191 static void gfs2_final_release_pages(struct gfs2_inode *ip)
1192 {
1193         struct inode *inode = &ip->i_inode;
1194         struct gfs2_glock *gl = ip->i_gl;
1195
1196         truncate_inode_pages(gfs2_glock2aspace(ip->i_gl), 0);
1197         truncate_inode_pages(&inode->i_data, 0);
1198
1199         if (atomic_read(&gl->gl_revokes) == 0) {
1200                 clear_bit(GLF_LFLUSH, &gl->gl_flags);
1201                 clear_bit(GLF_DIRTY, &gl->gl_flags);
1202         }
1203 }
1204
1205 static int gfs2_dinode_dealloc(struct gfs2_inode *ip)
1206 {
1207         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1208         struct gfs2_rgrpd *rgd;
1209         struct gfs2_holder gh;
1210         int error;
1211
1212         if (gfs2_get_inode_blocks(&ip->i_inode) != 1) {
1213                 gfs2_consist_inode(ip);
1214                 return -EIO;
1215         }
1216
1217         error = gfs2_rindex_update(sdp);
1218         if (error)
1219                 return error;
1220
1221         error = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
1222         if (error)
1223                 return error;
1224
1225         rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1226         if (!rgd) {
1227                 gfs2_consist_inode(ip);
1228                 error = -EIO;
1229                 goto out_qs;
1230         }
1231
1232         error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &gh);
1233         if (error)
1234                 goto out_qs;
1235
1236         error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA,
1237                                  sdp->sd_jdesc->jd_blocks);
1238         if (error)
1239                 goto out_rg_gunlock;
1240
1241         gfs2_free_di(rgd, ip);
1242
1243         gfs2_final_release_pages(ip);
1244
1245         gfs2_trans_end(sdp);
1246
1247 out_rg_gunlock:
1248         gfs2_glock_dq_uninit(&gh);
1249 out_qs:
1250         gfs2_quota_unhold(ip);
1251         return error;
1252 }
1253
1254 /**
1255  * gfs2_glock_put_eventually
1256  * @gl: The glock to put
1257  *
1258  * When under memory pressure, trigger a deferred glock put to make sure we
1259  * won't call into DLM and deadlock.  Otherwise, put the glock directly.
1260  */
1261
1262 static void gfs2_glock_put_eventually(struct gfs2_glock *gl)
1263 {
1264         if (current->flags & PF_MEMALLOC)
1265                 gfs2_glock_queue_put(gl);
1266         else
1267                 gfs2_glock_put(gl);
1268 }
1269
1270 static bool gfs2_upgrade_iopen_glock(struct inode *inode)
1271 {
1272         struct gfs2_inode *ip = GFS2_I(inode);
1273         struct gfs2_sbd *sdp = GFS2_SB(inode);
1274         struct gfs2_holder *gh = &ip->i_iopen_gh;
1275         long timeout = 5 * HZ;
1276         int error;
1277
1278         gh->gh_flags |= GL_NOCACHE;
1279         gfs2_glock_dq_wait(gh);
1280
1281         /*
1282          * If there are no other lock holders, we'll get the lock immediately.
1283          * Otherwise, the other nodes holding the lock will be notified about
1284          * our locking request.  If they don't have the inode open, they'll
1285          * evict the cached inode and release the lock.  Otherwise, if they
1286          * poke the inode glock, we'll take this as an indication that they
1287          * still need the iopen glock and that they'll take care of deleting
1288          * the inode when they're done.  As a last resort, if another node
1289          * keeps holding the iopen glock without showing any activity on the
1290          * inode glock, we'll eventually time out.
1291          *
1292          * Note that we're passing the LM_FLAG_TRY_1CB flag to the first
1293          * locking request as an optimization to notify lock holders as soon as
1294          * possible.  Without that flag, they'd be notified implicitly by the
1295          * second locking request.
1296          */
1297
1298         gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE, gh);
1299         error = gfs2_glock_nq(gh);
1300         if (error != GLR_TRYFAILED)
1301                 return !error;
1302
1303         gfs2_holder_reinit(LM_ST_EXCLUSIVE, GL_ASYNC | GL_NOCACHE, gh);
1304         error = gfs2_glock_nq(gh);
1305         if (error)
1306                 return false;
1307
1308         timeout = wait_event_interruptible_timeout(sdp->sd_async_glock_wait,
1309                 !test_bit(HIF_WAIT, &gh->gh_iflags) ||
1310                 test_bit(GLF_DEMOTE, &ip->i_gl->gl_flags),
1311                 timeout);
1312         if (!test_bit(HIF_HOLDER, &gh->gh_iflags)) {
1313                 gfs2_glock_dq(gh);
1314                 return false;
1315         }
1316         return true;
1317 }
1318
1319 /**
1320  * evict_should_delete - determine whether the inode is eligible for deletion
1321  * @inode: The inode to evict
1322  *
1323  * This function determines whether the evicted inode is eligible to be deleted
1324  * and locks the inode glock.
1325  *
1326  * Returns: the fate of the dinode
1327  */
1328 static enum dinode_demise evict_should_delete(struct inode *inode,
1329                                               struct gfs2_holder *gh)
1330 {
1331         struct gfs2_inode *ip = GFS2_I(inode);
1332         struct super_block *sb = inode->i_sb;
1333         struct gfs2_sbd *sdp = sb->s_fs_info;
1334         int ret;
1335
1336         if (test_bit(GIF_ALLOC_FAILED, &ip->i_flags)) {
1337                 BUG_ON(!gfs2_glock_is_locked_by_me(ip->i_gl));
1338                 goto should_delete;
1339         }
1340
1341         if (test_bit(GIF_DEFERRED_DELETE, &ip->i_flags))
1342                 return SHOULD_DEFER_EVICTION;
1343
1344         /* Deletes should never happen under memory pressure anymore.  */
1345         if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
1346                 return SHOULD_DEFER_EVICTION;
1347
1348         /* Must not read inode block until block type has been verified */
1349         ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, gh);
1350         if (unlikely(ret)) {
1351                 glock_clear_object(ip->i_iopen_gh.gh_gl, ip);
1352                 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1353                 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1354                 return SHOULD_DEFER_EVICTION;
1355         }
1356
1357         if (gfs2_inode_already_deleted(ip->i_gl, ip->i_no_formal_ino))
1358                 return SHOULD_NOT_DELETE_DINODE;
1359         ret = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED);
1360         if (ret)
1361                 return SHOULD_NOT_DELETE_DINODE;
1362
1363         if (test_bit(GIF_INVALID, &ip->i_flags)) {
1364                 ret = gfs2_inode_refresh(ip);
1365                 if (ret)
1366                         return SHOULD_NOT_DELETE_DINODE;
1367         }
1368
1369         /*
1370          * The inode may have been recreated in the meantime.
1371          */
1372         if (inode->i_nlink)
1373                 return SHOULD_NOT_DELETE_DINODE;
1374
1375 should_delete:
1376         if (gfs2_holder_initialized(&ip->i_iopen_gh) &&
1377             test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1378                 if (!gfs2_upgrade_iopen_glock(inode)) {
1379                         gfs2_holder_uninit(&ip->i_iopen_gh);
1380                         return SHOULD_NOT_DELETE_DINODE;
1381                 }
1382         }
1383         return SHOULD_DELETE_DINODE;
1384 }
1385
1386 /**
1387  * evict_unlinked_inode - delete the pieces of an unlinked evicted inode
1388  * @inode: The inode to evict
1389  */
1390 static int evict_unlinked_inode(struct inode *inode)
1391 {
1392         struct gfs2_inode *ip = GFS2_I(inode);
1393         int ret;
1394
1395         if (S_ISDIR(inode->i_mode) &&
1396             (ip->i_diskflags & GFS2_DIF_EXHASH)) {
1397                 ret = gfs2_dir_exhash_dealloc(ip);
1398                 if (ret)
1399                         goto out;
1400         }
1401
1402         if (ip->i_eattr) {
1403                 ret = gfs2_ea_dealloc(ip);
1404                 if (ret)
1405                         goto out;
1406         }
1407
1408         if (!gfs2_is_stuffed(ip)) {
1409                 ret = gfs2_file_dealloc(ip);
1410                 if (ret)
1411                         goto out;
1412         }
1413
1414         /* We're about to clear the bitmap for the dinode, but as soon as we
1415            do, gfs2_create_inode can create another inode at the same block
1416            location and try to set gl_object again. We clear gl_object here so
1417            that subsequent inode creates don't see an old gl_object. */
1418         glock_clear_object(ip->i_gl, ip);
1419         ret = gfs2_dinode_dealloc(ip);
1420         gfs2_inode_remember_delete(ip->i_gl, ip->i_no_formal_ino);
1421 out:
1422         return ret;
1423 }
1424
1425 /**
1426  * gfs2_evict_inode - Remove an inode from cache
1427  * @inode: The inode to evict
1428  *
1429  * There are three cases to consider:
1430  * 1. i_nlink == 0, we are final opener (and must deallocate)
1431  * 2. i_nlink == 0, we are not the final opener (and cannot deallocate)
1432  * 3. i_nlink > 0
1433  *
1434  * If the fs is read only, then we have to treat all cases as per #3
1435  * since we are unable to do any deallocation. The inode will be
1436  * deallocated by the next read/write node to attempt an allocation
1437  * in the same resource group
1438  *
1439  * We have to (at the moment) hold the inodes main lock to cover
1440  * the gap between unlocking the shared lock on the iopen lock and
1441  * taking the exclusive lock. I'd rather do a shared -> exclusive
1442  * conversion on the iopen lock, but we can change that later. This
1443  * is safe, just less efficient.
1444  */
1445
1446 static void gfs2_evict_inode(struct inode *inode)
1447 {
1448         struct super_block *sb = inode->i_sb;
1449         struct gfs2_sbd *sdp = sb->s_fs_info;
1450         struct gfs2_inode *ip = GFS2_I(inode);
1451         struct gfs2_holder gh;
1452         struct address_space *metamapping;
1453         int ret;
1454
1455         if (test_bit(GIF_FREE_VFS_INODE, &ip->i_flags)) {
1456                 clear_inode(inode);
1457                 return;
1458         }
1459
1460         if (inode->i_nlink || sb_rdonly(sb))
1461                 goto out;
1462
1463         gfs2_holder_mark_uninitialized(&gh);
1464         ret = evict_should_delete(inode, &gh);
1465         if (ret == SHOULD_DEFER_EVICTION)
1466                 goto out;
1467         if (ret == SHOULD_NOT_DELETE_DINODE)
1468                 goto out_truncate;
1469
1470         ret = evict_unlinked_inode(inode);
1471         goto out_unlock;
1472
1473 out_truncate:
1474         gfs2_log_flush(sdp, ip->i_gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
1475                        GFS2_LFC_EVICT_INODE);
1476         metamapping = gfs2_glock2aspace(ip->i_gl);
1477         if (test_bit(GLF_DIRTY, &ip->i_gl->gl_flags)) {
1478                 filemap_fdatawrite(metamapping);
1479                 filemap_fdatawait(metamapping);
1480         }
1481         write_inode_now(inode, 1);
1482         gfs2_ail_flush(ip->i_gl, 0);
1483
1484         ret = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
1485         if (ret)
1486                 goto out_unlock;
1487         /* Needs to be done before glock release & also in a transaction */
1488         truncate_inode_pages(&inode->i_data, 0);
1489         truncate_inode_pages(metamapping, 0);
1490         gfs2_trans_end(sdp);
1491
1492 out_unlock:
1493         if (gfs2_rs_active(&ip->i_res))
1494                 gfs2_rs_deltree(&ip->i_res);
1495
1496         if (gfs2_holder_initialized(&gh)) {
1497                 glock_clear_object(ip->i_gl, ip);
1498                 gfs2_glock_dq_uninit(&gh);
1499         }
1500         if (ret && ret != GLR_TRYFAILED && ret != -EROFS)
1501                 fs_warn(sdp, "gfs2_evict_inode: %d\n", ret);
1502 out:
1503         truncate_inode_pages_final(&inode->i_data);
1504         if (ip->i_qadata)
1505                 gfs2_assert_warn(sdp, ip->i_qadata->qa_ref == 0);
1506         gfs2_rs_delete(ip, NULL);
1507         gfs2_ordered_del_inode(ip);
1508         clear_inode(inode);
1509         gfs2_dir_hash_inval(ip);
1510         if (ip->i_gl) {
1511                 glock_clear_object(ip->i_gl, ip);
1512                 wait_on_bit_io(&ip->i_flags, GIF_GLOP_PENDING, TASK_UNINTERRUPTIBLE);
1513                 gfs2_glock_add_to_lru(ip->i_gl);
1514                 gfs2_glock_put_eventually(ip->i_gl);
1515                 ip->i_gl = NULL;
1516         }
1517         if (gfs2_holder_initialized(&ip->i_iopen_gh)) {
1518                 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1519
1520                 glock_clear_object(gl, ip);
1521                 if (test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1522                         ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1523                         gfs2_glock_dq(&ip->i_iopen_gh);
1524                 }
1525                 gfs2_glock_hold(gl);
1526                 gfs2_holder_uninit(&ip->i_iopen_gh);
1527                 gfs2_glock_put_eventually(gl);
1528         }
1529 }
1530
1531 static struct inode *gfs2_alloc_inode(struct super_block *sb)
1532 {
1533         struct gfs2_inode *ip;
1534
1535         ip = kmem_cache_alloc(gfs2_inode_cachep, GFP_KERNEL);
1536         if (!ip)
1537                 return NULL;
1538         ip->i_flags = 0;
1539         ip->i_gl = NULL;
1540         gfs2_holder_mark_uninitialized(&ip->i_iopen_gh);
1541         memset(&ip->i_res, 0, sizeof(ip->i_res));
1542         RB_CLEAR_NODE(&ip->i_res.rs_node);
1543         ip->i_rahead = 0;
1544         return &ip->i_inode;
1545 }
1546
1547 static void gfs2_free_inode(struct inode *inode)
1548 {
1549         kmem_cache_free(gfs2_inode_cachep, GFS2_I(inode));
1550 }
1551
1552 const struct super_operations gfs2_super_ops = {
1553         .alloc_inode            = gfs2_alloc_inode,
1554         .free_inode             = gfs2_free_inode,
1555         .write_inode            = gfs2_write_inode,
1556         .dirty_inode            = gfs2_dirty_inode,
1557         .evict_inode            = gfs2_evict_inode,
1558         .put_super              = gfs2_put_super,
1559         .sync_fs                = gfs2_sync_fs,
1560         .freeze_super           = gfs2_freeze,
1561         .thaw_super             = gfs2_unfreeze,
1562         .statfs                 = gfs2_statfs,
1563         .drop_inode             = gfs2_drop_inode,
1564         .show_options           = gfs2_show_options,
1565 };
1566