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
  85         list_for_each_entry(jd, head, jd_list) {
  86                 if (jd->jd_jid == jid)
  87                         return jd;
  88         }
  89         return NULL;
  90 }
  91
  92 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
  93 {
  94         struct gfs2_jdesc *jd;
  95
  96         spin_lock(&sdp->sd_jindex_spin);
  97         jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
  98         spin_unlock(&sdp->sd_jindex_spin);
  99
 100         return jd;
 101 }
 102
 103 int gfs2_jdesc_check(struct gfs2_jdesc *jd)
 104 {
 105         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 106         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 107         u64 size = i_size_read(jd->jd_inode);
 108
 109         if (gfs2_check_internal_file_size(jd->jd_inode, 8 << 20, BIT(30)))
 110                 return -EIO;
 111
 112         jd->jd_blocks = size >> sdp->sd_sb.sb_bsize_shift;
 113
 114         if (gfs2_write_alloc_required(ip, 0, size)) {
 115                 gfs2_consist_inode(ip);
 116                 return -EIO;
 117         }
 118
 119         return 0;
 120 }
 121
 122 /**
 123  * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
 124  * @sdp: the filesystem
 125  *
 126  * Returns: errno
 127  */
 128
 129 int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
 130 {
 131         struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
 132         struct gfs2_glock *j_gl = ip->i_gl;
 133         struct gfs2_log_header_host head;
 134         int error;
 135
 136         j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
 137         if (gfs2_withdrawn(sdp))
 138                 return -EIO;
 139
 140         error = gfs2_find_jhead(sdp->sd_jdesc, &head, false);
 141         if (error) {
 142                 gfs2_consist(sdp);
 143                 return error;
 144         }
 145
 146         if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 147                 gfs2_consist(sdp);
 148                 return -EIO;
 149         }
 150
 151         /*  Initialize some head of the log stuff  */
 152         sdp->sd_log_sequence = head.lh_sequence + 1;
 153         gfs2_log_pointers_init(sdp, head.lh_blkno);
 154
 155         error = gfs2_quota_init(sdp);
 156         if (!error && gfs2_withdrawn(sdp))
 157                 error = -EIO;
 158         if (!error)
 159                 set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 160         return error;
 161 }
 162
 163 void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
 164 {
 165         const struct gfs2_statfs_change *str = buf;
 166
 167         sc->sc_total = be64_to_cpu(str->sc_total);
 168         sc->sc_free = be64_to_cpu(str->sc_free);
 169         sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
 170 }
 171
 172 void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
 173 {
 174         struct gfs2_statfs_change *str = buf;
 175
 176         str->sc_total = cpu_to_be64(sc->sc_total);
 177         str->sc_free = cpu_to_be64(sc->sc_free);
 178         str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
 179 }
 180
 181 int gfs2_statfs_init(struct gfs2_sbd *sdp)
 182 {
 183         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 184         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 185         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 186         struct buffer_head *m_bh;
 187         struct gfs2_holder gh;
 188         int error;
 189
 190         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 191                                    &gh);
 192         if (error)
 193                 return error;
 194
 195         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 196         if (error)
 197                 goto out;
 198
 199         if (sdp->sd_args.ar_spectator) {
 200                 spin_lock(&sdp->sd_statfs_spin);
 201                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 202                                       sizeof(struct gfs2_dinode));
 203                 spin_unlock(&sdp->sd_statfs_spin);
 204         } else {
 205                 spin_lock(&sdp->sd_statfs_spin);
 206                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 207                                       sizeof(struct gfs2_dinode));
 208                 gfs2_statfs_change_in(l_sc, sdp->sd_sc_bh->b_data +
 209                                       sizeof(struct gfs2_dinode));
 210                 spin_unlock(&sdp->sd_statfs_spin);
 211
 212         }
 213
 214         brelse(m_bh);
 215 out:
 216         gfs2_glock_dq_uninit(&gh);
 217         return 0;
 218 }
 219
 220 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
 221                         s64 dinodes)
 222 {
 223         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 224         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 225         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 226         s64 x, y;
 227         int need_sync = 0;
 228
 229         gfs2_trans_add_meta(l_ip->i_gl, sdp->sd_sc_bh);
 230
 231         spin_lock(&sdp->sd_statfs_spin);
 232         l_sc->sc_total += total;
 233         l_sc->sc_free += free;
 234         l_sc->sc_dinodes += dinodes;
 235         gfs2_statfs_change_out(l_sc, sdp->sd_sc_bh->b_data +
 236                                sizeof(struct gfs2_dinode));
 237         if (sdp->sd_args.ar_statfs_percent) {
 238                 x = 100 * l_sc->sc_free;
 239                 y = m_sc->sc_free * sdp->sd_args.ar_statfs_percent;
 240                 if (x >= y || x <= -y)
 241                         need_sync = 1;
 242         }
 243         spin_unlock(&sdp->sd_statfs_spin);
 244
 245         if (need_sync)
 246                 gfs2_wake_up_statfs(sdp);
 247 }
 248
 249 void update_statfs(struct gfs2_sbd *sdp, struct buffer_head *m_bh)
 250 {
 251         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 252         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 253         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 254         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 255
 256         gfs2_trans_add_meta(l_ip->i_gl, sdp->sd_sc_bh);
 257         gfs2_trans_add_meta(m_ip->i_gl, m_bh);
 258
 259         spin_lock(&sdp->sd_statfs_spin);
 260         m_sc->sc_total += l_sc->sc_total;
 261         m_sc->sc_free += l_sc->sc_free;
 262         m_sc->sc_dinodes += l_sc->sc_dinodes;
 263         memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
 264         memset(sdp->sd_sc_bh->b_data + sizeof(struct gfs2_dinode),
 265                0, sizeof(struct gfs2_statfs_change));
 266         gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
 267         spin_unlock(&sdp->sd_statfs_spin);
 268 }
 269
 270 int gfs2_statfs_sync(struct super_block *sb, int type)
 271 {
 272         struct gfs2_sbd *sdp = sb->s_fs_info;
 273         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 274         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 275         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 276         struct gfs2_holder gh;
 277         struct buffer_head *m_bh;
 278         int error;
 279
 280         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 281                                    &gh);
 282         if (error)
 283                 goto out;
 284
 285         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 286         if (error)
 287                 goto out_unlock;
 288
 289         spin_lock(&sdp->sd_statfs_spin);
 290         gfs2_statfs_change_in(m_sc, m_bh->b_data +
 291                               sizeof(struct gfs2_dinode));
 292         if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
 293                 spin_unlock(&sdp->sd_statfs_spin);
 294                 goto out_bh;
 295         }
 296         spin_unlock(&sdp->sd_statfs_spin);
 297
 298         error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
 299         if (error)
 300                 goto out_bh;
 301
 302         update_statfs(sdp, m_bh);
 303         sdp->sd_statfs_force_sync = 0;
 304
 305         gfs2_trans_end(sdp);
 306
 307 out_bh:
 308         brelse(m_bh);
 309 out_unlock:
 310         gfs2_glock_dq_uninit(&gh);
 311 out:
 312         return error;
 313 }
 314
 315 struct lfcc {
 316         struct list_head list;
 317         struct gfs2_holder gh;
 318 };
 319
 320 /**
 321  * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
 322  *                            journals are clean
 323  * @sdp: the file system
 324  *
 325  * Returns: errno
 326  */
 327
 328 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp)
 329 {
 330         struct gfs2_inode *ip;
 331         struct gfs2_jdesc *jd;
 332         struct lfcc *lfcc;
 333         LIST_HEAD(list);
 334         struct gfs2_log_header_host lh;
 335         int error, error2;
 336
 337         /*
 338          * Grab all the journal glocks in SH mode.  We are *probably* doing
 339          * that to prevent recovery.
 340          */
 341
 342         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 343                 lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
 344                 if (!lfcc) {
 345                         error = -ENOMEM;
 346                         goto out;
 347                 }
 348                 ip = GFS2_I(jd->jd_inode);
 349                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
 350                 if (error) {
 351                         kfree(lfcc);
 352                         goto out;
 353                 }
 354                 list_add(&lfcc->list, &list);
 355         }
 356
 357         gfs2_freeze_unlock(&sdp->sd_freeze_gh);
 358
 359         error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_EXCLUSIVE,
 360                                    LM_FLAG_NOEXP | GL_NOPID,
 361                                    &sdp->sd_freeze_gh);
 362         if (error)
 363                 goto relock_shared;
 364
 365         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 366                 error = gfs2_jdesc_check(jd);
 367                 if (error)
 368                         break;
 369                 error = gfs2_find_jhead(jd, &lh, false);
 370                 if (error)
 371                         break;
 372                 if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 373                         error = -EBUSY;
 374                         break;
 375                 }
 376         }
 377
 378         if (!error)
 379                 goto out;  /* success */
 380
 381         gfs2_freeze_unlock(&sdp->sd_freeze_gh);
 382
 383 relock_shared:
 384         error2 = gfs2_freeze_lock_shared(sdp);
 385         gfs2_assert_withdraw(sdp, !error2);
 386
 387 out:
 388         while (!list_empty(&list)) {
 389                 lfcc = list_first_entry(&list, struct lfcc, list);
 390                 list_del(&lfcc->list);
 391                 gfs2_glock_dq_uninit(&lfcc->gh);
 392                 kfree(lfcc);
 393         }
 394         return error;
 395 }
 396
 397 void gfs2_dinode_out(const struct gfs2_inode *ip, void *buf)
 398 {
 399         const struct inode *inode = &ip->i_inode;
 400         struct gfs2_dinode *str = buf;
 401
 402         str->di_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 403         str->di_header.mh_type = cpu_to_be32(GFS2_METATYPE_DI);
 404         str->di_header.mh_format = cpu_to_be32(GFS2_FORMAT_DI);
 405         str->di_num.no_addr = cpu_to_be64(ip->i_no_addr);
 406         str->di_num.no_formal_ino = cpu_to_be64(ip->i_no_formal_ino);
 407         str->di_mode = cpu_to_be32(inode->i_mode);
 408         str->di_uid = cpu_to_be32(i_uid_read(inode));
 409         str->di_gid = cpu_to_be32(i_gid_read(inode));
 410         str->di_nlink = cpu_to_be32(inode->i_nlink);
 411         str->di_size = cpu_to_be64(i_size_read(inode));
 412         str->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(inode));
 413         str->di_atime = cpu_to_be64(inode->i_atime.tv_sec);
 414         str->di_mtime = cpu_to_be64(inode->i_mtime.tv_sec);
 415         str->di_ctime = cpu_to_be64(inode_get_ctime(inode).tv_sec);
 416
 417         str->di_goal_meta = cpu_to_be64(ip->i_goal);
 418         str->di_goal_data = cpu_to_be64(ip->i_goal);
 419         str->di_generation = cpu_to_be64(ip->i_generation);
 420
 421         str->di_flags = cpu_to_be32(ip->i_diskflags);
 422         str->di_height = cpu_to_be16(ip->i_height);
 423         str->di_payload_format = cpu_to_be32(S_ISDIR(inode->i_mode) &&
 424                                              !(ip->i_diskflags & GFS2_DIF_EXHASH) ?
 425                                              GFS2_FORMAT_DE : 0);
 426         str->di_depth = cpu_to_be16(ip->i_depth);
 427         str->di_entries = cpu_to_be32(ip->i_entries);
 428
 429         str->di_eattr = cpu_to_be64(ip->i_eattr);
 430         str->di_atime_nsec = cpu_to_be32(inode->i_atime.tv_nsec);
 431         str->di_mtime_nsec = cpu_to_be32(inode->i_mtime.tv_nsec);
 432         str->di_ctime_nsec = cpu_to_be32(inode_get_ctime(inode).tv_nsec);
 433 }
 434
 435 /**
 436  * gfs2_write_inode - Make sure the inode is stable on the disk
 437  * @inode: The inode
 438  * @wbc: The writeback control structure
 439  *
 440  * Returns: errno
 441  */
 442
 443 static int gfs2_write_inode(struct inode *inode, struct writeback_control *wbc)
 444 {
 445         struct gfs2_inode *ip = GFS2_I(inode);
 446         struct gfs2_sbd *sdp = GFS2_SB(inode);
 447         struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl);
 448         struct backing_dev_info *bdi = inode_to_bdi(metamapping->host);
 449         int ret = 0;
 450         bool flush_all = (wbc->sync_mode == WB_SYNC_ALL || gfs2_is_jdata(ip));
 451
 452         if (flush_all)
 453                 gfs2_log_flush(GFS2_SB(inode), ip->i_gl,
 454                                GFS2_LOG_HEAD_FLUSH_NORMAL |
 455                                GFS2_LFC_WRITE_INODE);
 456         if (bdi->wb.dirty_exceeded)
 457                 gfs2_ail1_flush(sdp, wbc);
 458         else
 459                 filemap_fdatawrite(metamapping);
 460         if (flush_all)
 461                 ret = filemap_fdatawait(metamapping);
 462         if (ret)
 463                 mark_inode_dirty_sync(inode);
 464         else {
 465                 spin_lock(&inode->i_lock);
 466                 if (!(inode->i_flags & I_DIRTY))
 467                         gfs2_ordered_del_inode(ip);
 468                 spin_unlock(&inode->i_lock);
 469         }
 470         return ret;
 471 }
 472
 473 /**
 474  * gfs2_dirty_inode - check for atime updates
 475  * @inode: The inode in question
 476  * @flags: The type of dirty
 477  *
 478  * Unfortunately it can be called under any combination of inode
 479  * glock and freeze glock, so we have to check carefully.
 480  *
 481  * At the moment this deals only with atime - it should be possible
 482  * to expand that role in future, once a review of the locking has
 483  * been carried out.
 484  */
 485
 486 static void gfs2_dirty_inode(struct inode *inode, int flags)
 487 {
 488         struct gfs2_inode *ip = GFS2_I(inode);
 489         struct gfs2_sbd *sdp = GFS2_SB(inode);
 490         struct buffer_head *bh;
 491         struct gfs2_holder gh;
 492         int need_unlock = 0;
 493         int need_endtrans = 0;
 494         int ret;
 495
 496         if (unlikely(!ip->i_gl)) {
 497                 /* This can only happen during incomplete inode creation. */
 498                 BUG_ON(!test_bit(GIF_ALLOC_FAILED, &ip->i_flags));
 499                 return;
 500         }
 501
 502         if (unlikely(gfs2_withdrawn(sdp)))
 503                 return;
 504         if (!gfs2_glock_is_locked_by_me(ip->i_gl)) {
 505                 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
 506                 if (ret) {
 507                         fs_err(sdp, "dirty_inode: glock %d\n", ret);
 508                         gfs2_dump_glock(NULL, ip->i_gl, true);
 509                         return;
 510                 }
 511                 need_unlock = 1;
 512         } else if (WARN_ON_ONCE(ip->i_gl->gl_state != LM_ST_EXCLUSIVE))
 513                 return;
 514
 515         if (current->journal_info == NULL) {
 516                 ret = gfs2_trans_begin(sdp, RES_DINODE, 0);
 517                 if (ret) {
 518                         fs_err(sdp, "dirty_inode: gfs2_trans_begin %d\n", ret);
 519                         goto out;
 520                 }
 521                 need_endtrans = 1;
 522         }
 523
 524         ret = gfs2_meta_inode_buffer(ip, &bh);
 525         if (ret == 0) {
 526                 gfs2_trans_add_meta(ip->i_gl, bh);
 527                 gfs2_dinode_out(ip, bh->b_data);
 528                 brelse(bh);
 529         }
 530
 531         if (need_endtrans)
 532                 gfs2_trans_end(sdp);
 533 out:
 534         if (need_unlock)
 535                 gfs2_glock_dq_uninit(&gh);
 536 }
 537
 538 /**
 539  * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
 540  * @sdp: the filesystem
 541  *
 542  * Returns: errno
 543  */
 544
 545 void gfs2_make_fs_ro(struct gfs2_sbd *sdp)
 546 {
 547         int log_write_allowed = test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 548
 549         if (!test_bit(SDF_DEACTIVATING, &sdp->sd_flags))
 550                 gfs2_flush_delete_work(sdp);
 551
 552         if (!log_write_allowed && current == sdp->sd_quotad_process)
 553                 fs_warn(sdp, "The quotad daemon is withdrawing.\n");
 554         else if (sdp->sd_quotad_process)
 555                 kthread_stop(sdp->sd_quotad_process);
 556         sdp->sd_quotad_process = NULL;
 557
 558         if (!log_write_allowed && current == sdp->sd_logd_process)
 559                 fs_warn(sdp, "The logd daemon is withdrawing.\n");
 560         else if (sdp->sd_logd_process)
 561                 kthread_stop(sdp->sd_logd_process);
 562         sdp->sd_logd_process = NULL;
 563
 564         if (log_write_allowed) {
 565                 gfs2_quota_sync(sdp->sd_vfs, 0);
 566                 gfs2_statfs_sync(sdp->sd_vfs, 0);
 567
 568                 /* We do two log flushes here. The first one commits dirty inodes
 569                  * and rgrps to the journal, but queues up revokes to the ail list.
 570                  * The second flush writes out and removes the revokes.
 571                  *
 572                  * The first must be done before the FLUSH_SHUTDOWN code
 573                  * clears the LIVE flag, otherwise it will not be able to start
 574                  * a transaction to write its revokes, and the error will cause
 575                  * a withdraw of the file system. */
 576                 gfs2_log_flush(sdp, NULL, GFS2_LFC_MAKE_FS_RO);
 577                 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SHUTDOWN |
 578                                GFS2_LFC_MAKE_FS_RO);
 579                 wait_event_timeout(sdp->sd_log_waitq,
 580                                    gfs2_log_is_empty(sdp),
 581                                    HZ * 5);
 582                 gfs2_assert_warn(sdp, gfs2_log_is_empty(sdp));
 583         } else {
 584                 wait_event_timeout(sdp->sd_log_waitq,
 585                                    gfs2_log_is_empty(sdp),
 586                                    HZ * 5);
 587         }
 588         gfs2_quota_cleanup(sdp);
 589
 590         if (!log_write_allowed)
 591                 sdp->sd_vfs->s_flags |= SB_RDONLY;
 592 }
 593
 594 /**
 595  * gfs2_put_super - Unmount the filesystem
 596  * @sb: The VFS superblock
 597  *
 598  */
 599
 600 static void gfs2_put_super(struct super_block *sb)
 601 {
 602         struct gfs2_sbd *sdp = sb->s_fs_info;
 603         struct gfs2_jdesc *jd;
 604
 605         /* No more recovery requests */
 606         set_bit(SDF_NORECOVERY, &sdp->sd_flags);
 607         smp_mb();
 608
 609         /* Wait on outstanding recovery */
 610 restart:
 611         spin_lock(&sdp->sd_jindex_spin);
 612         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 613                 if (!test_bit(JDF_RECOVERY, &jd->jd_flags))
 614                         continue;
 615                 spin_unlock(&sdp->sd_jindex_spin);
 616                 wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
 617                             TASK_UNINTERRUPTIBLE);
 618                 goto restart;
 619         }
 620         spin_unlock(&sdp->sd_jindex_spin);
 621
 622         if (!sb_rdonly(sb)) {
 623                 gfs2_make_fs_ro(sdp);
 624         }
 625         WARN_ON(gfs2_withdrawing(sdp));
 626
 627         /*  At this point, we're through modifying the disk  */
 628
 629         /*  Release stuff  */
 630
 631         gfs2_freeze_unlock(&sdp->sd_freeze_gh);
 632
 633         iput(sdp->sd_jindex);
 634         iput(sdp->sd_statfs_inode);
 635         iput(sdp->sd_rindex);
 636         iput(sdp->sd_quota_inode);
 637
 638         gfs2_glock_put(sdp->sd_rename_gl);
 639         gfs2_glock_put(sdp->sd_freeze_gl);
 640
 641         if (!sdp->sd_args.ar_spectator) {
 642                 if (gfs2_holder_initialized(&sdp->sd_journal_gh))
 643                         gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
 644                 if (gfs2_holder_initialized(&sdp->sd_jinode_gh))
 645                         gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
 646                 brelse(sdp->sd_sc_bh);
 647                 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
 648                 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
 649                 free_local_statfs_inodes(sdp);
 650                 iput(sdp->sd_qc_inode);
 651         }
 652
 653         gfs2_glock_dq_uninit(&sdp->sd_live_gh);
 654         gfs2_clear_rgrpd(sdp);
 655         gfs2_jindex_free(sdp);
 656         /*  Take apart glock structures and buffer lists  */
 657         gfs2_gl_hash_clear(sdp);
 658         truncate_inode_pages_final(&sdp->sd_aspace);
 659         gfs2_delete_debugfs_file(sdp);
 660         /*  Unmount the locking protocol  */
 661         gfs2_lm_unmount(sdp);
 662
 663         /*  At this point, we're through participating in the lockspace  */
 664         gfs2_sys_fs_del(sdp);
 665         free_sbd(sdp);
 666 }
 667
 668 /**
 669  * gfs2_sync_fs - sync the filesystem
 670  * @sb: the superblock
 671  * @wait: true to wait for completion
 672  *
 673  * Flushes the log to disk.
 674  */
 675
 676 static int gfs2_sync_fs(struct super_block *sb, int wait)
 677 {
 678         struct gfs2_sbd *sdp = sb->s_fs_info;
 679
 680         gfs2_quota_sync(sb, -1);
 681         if (wait)
 682                 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
 683                                GFS2_LFC_SYNC_FS);
 684         return sdp->sd_log_error;
 685 }
 686
 687 static int gfs2_freeze_locally(struct gfs2_sbd *sdp)
 688 {
 689         struct super_block *sb = sdp->sd_vfs;
 690         int error;
 691
 692         error = freeze_super(sb, FREEZE_HOLDER_USERSPACE);
 693         if (error)
 694                 return error;
 695
 696         if (test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
 697                 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_FREEZE |
 698                                GFS2_LFC_FREEZE_GO_SYNC);
 699                 if (gfs2_withdrawn(sdp)) {
 700                         error = thaw_super(sb, FREEZE_HOLDER_USERSPACE);
 701                         if (error)
 702                                 return error;
 703                         return -EIO;
 704                 }
 705         }
 706         return 0;
 707 }
 708
 709 static int gfs2_do_thaw(struct gfs2_sbd *sdp)
 710 {
 711         struct super_block *sb = sdp->sd_vfs;
 712         int error;
 713
 714         error = gfs2_freeze_lock_shared(sdp);
 715         if (error)
 716                 goto fail;
 717         error = thaw_super(sb, FREEZE_HOLDER_USERSPACE);
 718         if (!error)
 719                 return 0;
 720
 721 fail:
 722         fs_info(sdp, "GFS2: couldn't thaw filesystem: %d\n", error);
 723         gfs2_assert_withdraw(sdp, 0);
 724         return error;
 725 }
 726
 727 void gfs2_freeze_func(struct work_struct *work)
 728 {
 729         struct gfs2_sbd *sdp = container_of(work, struct gfs2_sbd, sd_freeze_work);
 730         struct super_block *sb = sdp->sd_vfs;
 731         int error;
 732
 733         mutex_lock(&sdp->sd_freeze_mutex);
 734         error = -EBUSY;
 735         if (test_bit(SDF_FROZEN, &sdp->sd_flags))
 736                 goto freeze_failed;
 737
 738         error = gfs2_freeze_locally(sdp);
 739         if (error)
 740                 goto freeze_failed;
 741
 742         gfs2_freeze_unlock(&sdp->sd_freeze_gh);
 743         set_bit(SDF_FROZEN, &sdp->sd_flags);
 744
 745         error = gfs2_do_thaw(sdp);
 746         if (error)
 747                 goto out;
 748
 749         clear_bit(SDF_FROZEN, &sdp->sd_flags);
 750         goto out;
 751
 752 freeze_failed:
 753         fs_info(sdp, "GFS2: couldn't freeze filesystem: %d\n", error);
 754
 755 out:
 756         mutex_unlock(&sdp->sd_freeze_mutex);
 757         deactivate_super(sb);
 758 }
 759
 760 /**
 761  * gfs2_freeze_super - prevent further writes to the filesystem
 762  * @sb: the VFS structure for the filesystem
 763  *
 764  */
 765
 766 static int gfs2_freeze_super(struct super_block *sb, enum freeze_holder who)
 767 {
 768         struct gfs2_sbd *sdp = sb->s_fs_info;
 769         int error;
 770
 771         if (!mutex_trylock(&sdp->sd_freeze_mutex))
 772                 return -EBUSY;
 773         error = -EBUSY;
 774         if (test_bit(SDF_FROZEN, &sdp->sd_flags))
 775                 goto out;
 776
 777         for (;;) {
 778                 error = gfs2_freeze_locally(sdp);
 779                 if (error) {
 780                         fs_info(sdp, "GFS2: couldn't freeze filesystem: %d\n",
 781                                 error);
 782                         goto out;
 783                 }
 784
 785                 error = gfs2_lock_fs_check_clean(sdp);
 786                 if (!error)
 787                         break;  /* success */
 788
 789                 error = gfs2_do_thaw(sdp);
 790                 if (error)
 791                         goto out;
 792
 793                 if (error == -EBUSY)
 794                         fs_err(sdp, "waiting for recovery before freeze\n");
 795                 else if (error == -EIO) {
 796                         fs_err(sdp, "Fatal IO error: cannot freeze gfs2 due "
 797                                "to recovery error.\n");
 798                         goto out;
 799                 } else {
 800                         fs_err(sdp, "error freezing FS: %d\n", error);
 801                 }
 802                 fs_err(sdp, "retrying...\n");
 803                 msleep(1000);
 804         }
 805
 806 out:
 807         if (!error) {
 808                 set_bit(SDF_FREEZE_INITIATOR, &sdp->sd_flags);
 809                 set_bit(SDF_FROZEN, &sdp->sd_flags);
 810         }
 811         mutex_unlock(&sdp->sd_freeze_mutex);
 812         return error;
 813 }
 814
 815 /**
 816  * gfs2_thaw_super - reallow writes to the filesystem
 817  * @sb: the VFS structure for the filesystem
 818  *
 819  */
 820
 821 static int gfs2_thaw_super(struct super_block *sb, enum freeze_holder who)
 822 {
 823         struct gfs2_sbd *sdp = sb->s_fs_info;
 824         int error;
 825
 826         if (!mutex_trylock(&sdp->sd_freeze_mutex))
 827                 return -EBUSY;
 828         error = -EINVAL;
 829         if (!test_bit(SDF_FREEZE_INITIATOR, &sdp->sd_flags))
 830                 goto out;
 831
 832         gfs2_freeze_unlock(&sdp->sd_freeze_gh);
 833
 834         error = gfs2_do_thaw(sdp);
 835
 836         if (!error) {
 837                 clear_bit(SDF_FREEZE_INITIATOR, &sdp->sd_flags);
 838                 clear_bit(SDF_FROZEN, &sdp->sd_flags);
 839         }
 840 out:
 841         mutex_unlock(&sdp->sd_freeze_mutex);
 842         return error;
 843 }
 844
 845 void gfs2_thaw_freeze_initiator(struct super_block *sb)
 846 {
 847         struct gfs2_sbd *sdp = sb->s_fs_info;
 848
 849         mutex_lock(&sdp->sd_freeze_mutex);
 850         if (!test_bit(SDF_FREEZE_INITIATOR, &sdp->sd_flags))
 851                 goto out;
 852
 853         gfs2_freeze_unlock(&sdp->sd_freeze_gh);
 854
 855 out:
 856         mutex_unlock(&sdp->sd_freeze_mutex);
 857 }
 858
 859 /**
 860  * statfs_slow_fill - fill in the sg for a given RG
 861  * @rgd: the RG
 862  * @sc: the sc structure
 863  *
 864  * Returns: 0 on success, -ESTALE if the LVB is invalid
 865  */
 866
 867 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
 868                             struct gfs2_statfs_change_host *sc)
 869 {
 870         gfs2_rgrp_verify(rgd);
 871         sc->sc_total += rgd->rd_data;
 872         sc->sc_free += rgd->rd_free;
 873         sc->sc_dinodes += rgd->rd_dinodes;
 874         return 0;
 875 }
 876
 877 /**
 878  * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
 879  * @sdp: the filesystem
 880  * @sc: the sc info that will be returned
 881  *
 882  * Any error (other than a signal) will cause this routine to fall back
 883  * to the synchronous version.
 884  *
 885  * FIXME: This really shouldn't busy wait like this.
 886  *
 887  * Returns: errno
 888  */
 889
 890 static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
 891 {
 892         struct gfs2_rgrpd *rgd_next;
 893         struct gfs2_holder *gha, *gh;
 894         unsigned int slots = 64;
 895         unsigned int x;
 896         int done;
 897         int error = 0, err;
 898
 899         memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
 900         gha = kmalloc_array(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
 901         if (!gha)
 902                 return -ENOMEM;
 903         for (x = 0; x < slots; x++)
 904                 gfs2_holder_mark_uninitialized(gha + x);
 905
 906         rgd_next = gfs2_rgrpd_get_first(sdp);
 907
 908         for (;;) {
 909                 done = 1;
 910
 911                 for (x = 0; x < slots; x++) {
 912                         gh = gha + x;
 913
 914                         if (gfs2_holder_initialized(gh) && gfs2_glock_poll(gh)) {
 915                                 err = gfs2_glock_wait(gh);
 916                                 if (err) {
 917                                         gfs2_holder_uninit(gh);
 918                                         error = err;
 919                                 } else {
 920                                         if (!error) {
 921                                                 struct gfs2_rgrpd *rgd =
 922                                                         gfs2_glock2rgrp(gh->gh_gl);
 923
 924                                                 error = statfs_slow_fill(rgd, sc);
 925                                         }
 926                                         gfs2_glock_dq_uninit(gh);
 927                                 }
 928                         }
 929
 930                         if (gfs2_holder_initialized(gh))
 931                                 done = 0;
 932                         else if (rgd_next && !error) {
 933                                 error = gfs2_glock_nq_init(rgd_next->rd_gl,
 934                                                            LM_ST_SHARED,
 935                                                            GL_ASYNC,
 936                                                            gh);
 937                                 rgd_next = gfs2_rgrpd_get_next(rgd_next);
 938                                 done = 0;
 939                         }
 940
 941                         if (signal_pending(current))
 942                                 error = -ERESTARTSYS;
 943                 }
 944
 945                 if (done)
 946                         break;
 947
 948                 yield();
 949         }
 950
 951         kfree(gha);
 952         return error;
 953 }
 954
 955 /**
 956  * gfs2_statfs_i - Do a statfs
 957  * @sdp: the filesystem
 958  * @sc: the sc structure
 959  *
 960  * Returns: errno
 961  */
 962
 963 static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
 964 {
 965         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 966         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 967
 968         spin_lock(&sdp->sd_statfs_spin);
 969
 970         *sc = *m_sc;
 971         sc->sc_total += l_sc->sc_total;
 972         sc->sc_free += l_sc->sc_free;
 973         sc->sc_dinodes += l_sc->sc_dinodes;
 974
 975         spin_unlock(&sdp->sd_statfs_spin);
 976
 977         if (sc->sc_free < 0)
 978                 sc->sc_free = 0;
 979         if (sc->sc_free > sc->sc_total)
 980                 sc->sc_free = sc->sc_total;
 981         if (sc->sc_dinodes < 0)
 982                 sc->sc_dinodes = 0;
 983
 984         return 0;
 985 }
 986
 987 /**
 988  * gfs2_statfs - Gather and return stats about the filesystem
 989  * @dentry: The name of the link
 990  * @buf: The buffer
 991  *
 992  * Returns: 0 on success or error code
 993  */
 994
 995 static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
 996 {
 997         struct super_block *sb = dentry->d_sb;
 998         struct gfs2_sbd *sdp = sb->s_fs_info;
 999         struct gfs2_statfs_change_host sc;
1000         int error;
1001
1002         error = gfs2_rindex_update(sdp);
1003         if (error)
1004                 return error;
1005
1006         if (gfs2_tune_get(sdp, gt_statfs_slow))
1007                 error = gfs2_statfs_slow(sdp, &sc);
1008         else
1009                 error = gfs2_statfs_i(sdp, &sc);
1010
1011         if (error)
1012                 return error;
1013
1014         buf->f_type = GFS2_MAGIC;
1015         buf->f_bsize = sdp->sd_sb.sb_bsize;
1016         buf->f_blocks = sc.sc_total;
1017         buf->f_bfree = sc.sc_free;
1018         buf->f_bavail = sc.sc_free;
1019         buf->f_files = sc.sc_dinodes + sc.sc_free;
1020         buf->f_ffree = sc.sc_free;
1021         buf->f_namelen = GFS2_FNAMESIZE;
1022
1023         return 0;
1024 }
1025
1026 /**
1027  * gfs2_drop_inode - Drop an inode (test for remote unlink)
1028  * @inode: The inode to drop
1029  *
1030  * If we've received a callback on an iopen lock then it's because a
1031  * remote node tried to deallocate the inode but failed due to this node
1032  * still having the inode open. Here we mark the link count zero
1033  * since we know that it must have reached zero if the GLF_DEMOTE flag
1034  * is set on the iopen glock. If we didn't do a disk read since the
1035  * remote node removed the final link then we might otherwise miss
1036  * this event. This check ensures that this node will deallocate the
1037  * inode's blocks, or alternatively pass the baton on to another
1038  * node for later deallocation.
1039  */
1040
1041 static int gfs2_drop_inode(struct inode *inode)
1042 {
1043         struct gfs2_inode *ip = GFS2_I(inode);
1044         struct gfs2_sbd *sdp = GFS2_SB(inode);
1045
1046         if (inode->i_nlink &&
1047             gfs2_holder_initialized(&ip->i_iopen_gh)) {
1048                 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1049                 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
1050                         clear_nlink(inode);
1051         }
1052
1053         /*
1054          * When under memory pressure when an inode's link count has dropped to
1055          * zero, defer deleting the inode to the delete workqueue.  This avoids
1056          * calling into DLM under memory pressure, which can deadlock.
1057          */
1058         if (!inode->i_nlink &&
1059             unlikely(current->flags & PF_MEMALLOC) &&
1060             gfs2_holder_initialized(&ip->i_iopen_gh)) {
1061                 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1062
1063                 gfs2_glock_hold(gl);
1064                 if (!gfs2_queue_try_to_evict(gl))
1065                         gfs2_glock_queue_put(gl);
1066                 return 0;
1067         }
1068
1069         /*
1070          * No longer cache inodes when trying to evict them all.
1071          */
1072         if (test_bit(SDF_EVICTING, &sdp->sd_flags))
1073                 return 1;
1074
1075         return generic_drop_inode(inode);
1076 }
1077
1078 static int is_ancestor(const struct dentry *d1, const struct dentry *d2)
1079 {
1080         do {
1081                 if (d1 == d2)
1082                         return 1;
1083                 d1 = d1->d_parent;
1084         } while (!IS_ROOT(d1));
1085         return 0;
1086 }
1087
1088 /**
1089  * gfs2_show_options - Show mount options for /proc/mounts
1090  * @s: seq_file structure
1091  * @root: root of this (sub)tree
1092  *
1093  * Returns: 0 on success or error code
1094  */
1095
1096 static int gfs2_show_options(struct seq_file *s, struct dentry *root)
1097 {
1098         struct gfs2_sbd *sdp = root->d_sb->s_fs_info;
1099         struct gfs2_args *args = &sdp->sd_args;
1100         unsigned int logd_secs, statfs_slow, statfs_quantum, quota_quantum;
1101
1102         spin_lock(&sdp->sd_tune.gt_spin);
1103         logd_secs = sdp->sd_tune.gt_logd_secs;
1104         quota_quantum = sdp->sd_tune.gt_quota_quantum;
1105         statfs_quantum = sdp->sd_tune.gt_statfs_quantum;
1106         statfs_slow = sdp->sd_tune.gt_statfs_slow;
1107         spin_unlock(&sdp->sd_tune.gt_spin);
1108
1109         if (is_ancestor(root, sdp->sd_master_dir))
1110                 seq_puts(s, ",meta");
1111         if (args->ar_lockproto[0])
1112                 seq_show_option(s, "lockproto", args->ar_lockproto);
1113         if (args->ar_locktable[0])
1114                 seq_show_option(s, "locktable", args->ar_locktable);
1115         if (args->ar_hostdata[0])
1116                 seq_show_option(s, "hostdata", args->ar_hostdata);
1117         if (args->ar_spectator)
1118                 seq_puts(s, ",spectator");
1119         if (args->ar_localflocks)
1120                 seq_puts(s, ",localflocks");
1121         if (args->ar_debug)
1122                 seq_puts(s, ",debug");
1123         if (args->ar_posix_acl)
1124                 seq_puts(s, ",acl");
1125         if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
1126                 char *state;
1127                 switch (args->ar_quota) {
1128                 case GFS2_QUOTA_OFF:
1129                         state = "off";
1130                         break;
1131                 case GFS2_QUOTA_ACCOUNT:
1132                         state = "account";
1133                         break;
1134                 case GFS2_QUOTA_ON:
1135                         state = "on";
1136                         break;
1137                 default:
1138                         state = "unknown";
1139                         break;
1140                 }
1141                 seq_printf(s, ",quota=%s", state);
1142         }
1143         if (args->ar_suiddir)
1144                 seq_puts(s, ",suiddir");
1145         if (args->ar_data != GFS2_DATA_DEFAULT) {
1146                 char *state;
1147                 switch (args->ar_data) {
1148                 case GFS2_DATA_WRITEBACK:
1149                         state = "writeback";
1150                         break;
1151                 case GFS2_DATA_ORDERED:
1152                         state = "ordered";
1153                         break;
1154                 default:
1155                         state = "unknown";
1156                         break;
1157                 }
1158                 seq_printf(s, ",data=%s", state);
1159         }
1160         if (args->ar_discard)
1161                 seq_puts(s, ",discard");
1162         if (logd_secs != 30)
1163                 seq_printf(s, ",commit=%d", logd_secs);
1164         if (statfs_quantum != 30)
1165                 seq_printf(s, ",statfs_quantum=%d", statfs_quantum);
1166         else if (statfs_slow)
1167                 seq_puts(s, ",statfs_quantum=0");
1168         if (quota_quantum != 60)
1169                 seq_printf(s, ",quota_quantum=%d", quota_quantum);
1170         if (args->ar_statfs_percent)
1171                 seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent);
1172         if (args->ar_errors != GFS2_ERRORS_DEFAULT) {
1173                 const char *state;
1174
1175                 switch (args->ar_errors) {
1176                 case GFS2_ERRORS_WITHDRAW:
1177                         state = "withdraw";
1178                         break;
1179                 case GFS2_ERRORS_PANIC:
1180                         state = "panic";
1181                         break;
1182                 default:
1183                         state = "unknown";
1184                         break;
1185                 }
1186                 seq_printf(s, ",errors=%s", state);
1187         }
1188         if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags))
1189                 seq_puts(s, ",nobarrier");
1190         if (test_bit(SDF_DEMOTE, &sdp->sd_flags))
1191                 seq_puts(s, ",demote_interface_used");
1192         if (args->ar_rgrplvb)
1193                 seq_puts(s, ",rgrplvb");
1194         if (args->ar_loccookie)
1195                 seq_puts(s, ",loccookie");
1196         return 0;
1197 }
1198
1199 static void gfs2_final_release_pages(struct gfs2_inode *ip)
1200 {
1201         struct inode *inode = &ip->i_inode;
1202         struct gfs2_glock *gl = ip->i_gl;
1203
1204         if (unlikely(!gl)) {
1205                 /* This can only happen during incomplete inode creation. */
1206                 BUG_ON(!test_bit(GIF_ALLOC_FAILED, &ip->i_flags));
1207                 return;
1208         }
1209
1210         truncate_inode_pages(gfs2_glock2aspace(gl), 0);
1211         truncate_inode_pages(&inode->i_data, 0);
1212
1213         if (atomic_read(&gl->gl_revokes) == 0) {
1214                 clear_bit(GLF_LFLUSH, &gl->gl_flags);
1215                 clear_bit(GLF_DIRTY, &gl->gl_flags);
1216         }
1217 }
1218
1219 static int gfs2_dinode_dealloc(struct gfs2_inode *ip)
1220 {
1221         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1222         struct gfs2_rgrpd *rgd;
1223         struct gfs2_holder gh;
1224         int error;
1225
1226         if (gfs2_get_inode_blocks(&ip->i_inode) != 1) {
1227                 gfs2_consist_inode(ip);
1228                 return -EIO;
1229         }
1230
1231         gfs2_rindex_update(sdp);
1232
1233         error = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
1234         if (error)
1235                 return error;
1236
1237         rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1238         if (!rgd) {
1239                 gfs2_consist_inode(ip);
1240                 error = -EIO;
1241                 goto out_qs;
1242         }
1243
1244         error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE,
1245                                    LM_FLAG_NODE_SCOPE, &gh);
1246         if (error)
1247                 goto out_qs;
1248
1249         error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA,
1250                                  sdp->sd_jdesc->jd_blocks);
1251         if (error)
1252                 goto out_rg_gunlock;
1253
1254         gfs2_free_di(rgd, ip);
1255
1256         gfs2_final_release_pages(ip);
1257
1258         gfs2_trans_end(sdp);
1259
1260 out_rg_gunlock:
1261         gfs2_glock_dq_uninit(&gh);
1262 out_qs:
1263         gfs2_quota_unhold(ip);
1264         return error;
1265 }
1266
1267 /**
1268  * gfs2_glock_put_eventually
1269  * @gl: The glock to put
1270  *
1271  * When under memory pressure, trigger a deferred glock put to make sure we
1272  * won't call into DLM and deadlock.  Otherwise, put the glock directly.
1273  */
1274
1275 static void gfs2_glock_put_eventually(struct gfs2_glock *gl)
1276 {
1277         if (current->flags & PF_MEMALLOC)
1278                 gfs2_glock_queue_put(gl);
1279         else
1280                 gfs2_glock_put(gl);
1281 }
1282
1283 static bool gfs2_upgrade_iopen_glock(struct inode *inode)
1284 {
1285         struct gfs2_inode *ip = GFS2_I(inode);
1286         struct gfs2_sbd *sdp = GFS2_SB(inode);
1287         struct gfs2_holder *gh = &ip->i_iopen_gh;
1288         long timeout = 5 * HZ;
1289         int error;
1290
1291         gh->gh_flags |= GL_NOCACHE;
1292         gfs2_glock_dq_wait(gh);
1293
1294         /*
1295          * If there are no other lock holders, we will immediately get
1296          * exclusive access to the iopen glock here.
1297          *
1298          * Otherwise, the other nodes holding the lock will be notified about
1299          * our locking request.  If they do not have the inode open, they are
1300          * expected to evict the cached inode and release the lock, allowing us
1301          * to proceed.
1302          *
1303          * Otherwise, if they cannot evict the inode, they are expected to poke
1304          * the inode glock (note: not the iopen glock).  We will notice that
1305          * and stop waiting for the iopen glock immediately.  The other node(s)
1306          * are then expected to take care of deleting the inode when they no
1307          * longer use it.
1308          *
1309          * As a last resort, if another node keeps holding the iopen glock
1310          * without showing any activity on the inode glock, we will eventually
1311          * time out and fail the iopen glock upgrade.
1312          *
1313          * Note that we're passing the LM_FLAG_TRY_1CB flag to the first
1314          * locking request as an optimization to notify lock holders as soon as
1315          * possible.  Without that flag, they'd be notified implicitly by the
1316          * second locking request.
1317          */
1318
1319         gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE, gh);
1320         error = gfs2_glock_nq(gh);
1321         if (error != GLR_TRYFAILED)
1322                 return !error;
1323
1324         gfs2_holder_reinit(LM_ST_EXCLUSIVE, GL_ASYNC | GL_NOCACHE, gh);
1325         error = gfs2_glock_nq(gh);
1326         if (error)
1327                 return false;
1328
1329         timeout = wait_event_interruptible_timeout(sdp->sd_async_glock_wait,
1330                 !test_bit(HIF_WAIT, &gh->gh_iflags) ||
1331                 test_bit(GLF_DEMOTE, &ip->i_gl->gl_flags),
1332                 timeout);
1333         if (!test_bit(HIF_HOLDER, &gh->gh_iflags)) {
1334                 gfs2_glock_dq(gh);
1335                 return false;
1336         }
1337         return gfs2_glock_holder_ready(gh) == 0;
1338 }
1339
1340 /**
1341  * evict_should_delete - determine whether the inode is eligible for deletion
1342  * @inode: The inode to evict
1343  * @gh: The glock holder structure
1344  *
1345  * This function determines whether the evicted inode is eligible to be deleted
1346  * and locks the inode glock.
1347  *
1348  * Returns: the fate of the dinode
1349  */
1350 static enum dinode_demise evict_should_delete(struct inode *inode,
1351                                               struct gfs2_holder *gh)
1352 {
1353         struct gfs2_inode *ip = GFS2_I(inode);
1354         struct super_block *sb = inode->i_sb;
1355         struct gfs2_sbd *sdp = sb->s_fs_info;
1356         int ret;
1357
1358         if (unlikely(test_bit(GIF_ALLOC_FAILED, &ip->i_flags)))
1359                 goto should_delete;
1360
1361         if (test_bit(GIF_DEFERRED_DELETE, &ip->i_flags))
1362                 return SHOULD_DEFER_EVICTION;
1363
1364         /* Deletes should never happen under memory pressure anymore.  */
1365         if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
1366                 return SHOULD_DEFER_EVICTION;
1367
1368         /* Must not read inode block until block type has been verified */
1369         ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, gh);
1370         if (unlikely(ret)) {
1371                 glock_clear_object(ip->i_iopen_gh.gh_gl, ip);
1372                 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1373                 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1374                 return SHOULD_DEFER_EVICTION;
1375         }
1376
1377         if (gfs2_inode_already_deleted(ip->i_gl, ip->i_no_formal_ino))
1378                 return SHOULD_NOT_DELETE_DINODE;
1379         ret = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED);
1380         if (ret)
1381                 return SHOULD_NOT_DELETE_DINODE;
1382
1383         ret = gfs2_instantiate(gh);
1384         if (ret)
1385                 return SHOULD_NOT_DELETE_DINODE;
1386
1387         /*
1388          * The inode may have been recreated in the meantime.
1389          */
1390         if (inode->i_nlink)
1391                 return SHOULD_NOT_DELETE_DINODE;
1392
1393 should_delete:
1394         if (gfs2_holder_initialized(&ip->i_iopen_gh) &&
1395             test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1396                 if (!gfs2_upgrade_iopen_glock(inode)) {
1397                         gfs2_holder_uninit(&ip->i_iopen_gh);
1398                         return SHOULD_NOT_DELETE_DINODE;
1399                 }
1400         }
1401         return SHOULD_DELETE_DINODE;
1402 }
1403
1404 /**
1405  * evict_unlinked_inode - delete the pieces of an unlinked evicted inode
1406  * @inode: The inode to evict
1407  */
1408 static int evict_unlinked_inode(struct inode *inode)
1409 {
1410         struct gfs2_inode *ip = GFS2_I(inode);
1411         int ret;
1412
1413         if (S_ISDIR(inode->i_mode) &&
1414             (ip->i_diskflags & GFS2_DIF_EXHASH)) {
1415                 ret = gfs2_dir_exhash_dealloc(ip);
1416                 if (ret)
1417                         goto out;
1418         }
1419
1420         if (ip->i_eattr) {
1421                 ret = gfs2_ea_dealloc(ip);
1422                 if (ret)
1423                         goto out;
1424         }
1425
1426         if (!gfs2_is_stuffed(ip)) {
1427                 ret = gfs2_file_dealloc(ip);
1428                 if (ret)
1429                         goto out;
1430         }
1431
1432         /*
1433          * As soon as we clear the bitmap for the dinode, gfs2_create_inode()
1434          * can get called to recreate it, or even gfs2_inode_lookup() if the
1435          * inode was recreated on another node in the meantime.
1436          *
1437          * However, inserting the new inode into the inode hash table will not
1438          * succeed until the old inode is removed, and that only happens after
1439          * ->evict_inode() returns.  The new inode is attached to its inode and
1440          *  iopen glocks after inserting it into the inode hash table, so at
1441          *  that point we can be sure that both glocks are unused.
1442          */
1443
1444         ret = gfs2_dinode_dealloc(ip);
1445         if (!ret && ip->i_gl)
1446                 gfs2_inode_remember_delete(ip->i_gl, ip->i_no_formal_ino);
1447
1448 out:
1449         return ret;
1450 }
1451
1452 /*
1453  * evict_linked_inode - evict an inode whose dinode has not been unlinked
1454  * @inode: The inode to evict
1455  */
1456 static int evict_linked_inode(struct inode *inode)
1457 {
1458         struct super_block *sb = inode->i_sb;
1459         struct gfs2_sbd *sdp = sb->s_fs_info;
1460         struct gfs2_inode *ip = GFS2_I(inode);
1461         struct address_space *metamapping;
1462         int ret;
1463
1464         gfs2_log_flush(sdp, ip->i_gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
1465                        GFS2_LFC_EVICT_INODE);
1466         metamapping = gfs2_glock2aspace(ip->i_gl);
1467         if (test_bit(GLF_DIRTY, &ip->i_gl->gl_flags)) {
1468                 filemap_fdatawrite(metamapping);
1469                 filemap_fdatawait(metamapping);
1470         }
1471         write_inode_now(inode, 1);
1472         gfs2_ail_flush(ip->i_gl, 0);
1473
1474         ret = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
1475         if (ret)
1476                 return ret;
1477
1478         /* Needs to be done before glock release & also in a transaction */
1479         truncate_inode_pages(&inode->i_data, 0);
1480         truncate_inode_pages(metamapping, 0);
1481         gfs2_trans_end(sdp);
1482         return 0;
1483 }
1484
1485 /**
1486  * gfs2_evict_inode - Remove an inode from cache
1487  * @inode: The inode to evict
1488  *
1489  * There are three cases to consider:
1490  * 1. i_nlink == 0, we are final opener (and must deallocate)
1491  * 2. i_nlink == 0, we are not the final opener (and cannot deallocate)
1492  * 3. i_nlink > 0
1493  *
1494  * If the fs is read only, then we have to treat all cases as per #3
1495  * since we are unable to do any deallocation. The inode will be
1496  * deallocated by the next read/write node to attempt an allocation
1497  * in the same resource group
1498  *
1499  * We have to (at the moment) hold the inodes main lock to cover
1500  * the gap between unlocking the shared lock on the iopen lock and
1501  * taking the exclusive lock. I'd rather do a shared -> exclusive
1502  * conversion on the iopen lock, but we can change that later. This
1503  * is safe, just less efficient.
1504  */
1505
1506 static void gfs2_evict_inode(struct inode *inode)
1507 {
1508         struct super_block *sb = inode->i_sb;
1509         struct gfs2_sbd *sdp = sb->s_fs_info;
1510         struct gfs2_inode *ip = GFS2_I(inode);
1511         struct gfs2_holder gh;
1512         int ret;
1513
1514         if (inode->i_nlink || sb_rdonly(sb) || !ip->i_no_addr)
1515                 goto out;
1516
1517         /*
1518          * In case of an incomplete mount, gfs2_evict_inode() may be called for
1519          * system files without having an active journal to write to.  In that
1520          * case, skip the filesystem evict.
1521          */
1522         if (!sdp->sd_jdesc)
1523                 goto out;
1524
1525         gfs2_holder_mark_uninitialized(&gh);
1526         ret = evict_should_delete(inode, &gh);
1527         if (ret == SHOULD_DEFER_EVICTION)
1528                 goto out;
1529         if (ret == SHOULD_DELETE_DINODE)
1530                 ret = evict_unlinked_inode(inode);
1531         else
1532                 ret = evict_linked_inode(inode);
1533
1534         if (gfs2_rs_active(&ip->i_res))
1535                 gfs2_rs_deltree(&ip->i_res);
1536
1537         if (gfs2_holder_initialized(&gh))
1538                 gfs2_glock_dq_uninit(&gh);
1539         if (ret && ret != GLR_TRYFAILED && ret != -EROFS)
1540                 fs_warn(sdp, "gfs2_evict_inode: %d\n", ret);
1541 out:
1542         truncate_inode_pages_final(&inode->i_data);
1543         if (ip->i_qadata)
1544                 gfs2_assert_warn(sdp, ip->i_qadata->qa_ref == 0);
1545         gfs2_rs_deltree(&ip->i_res);
1546         gfs2_ordered_del_inode(ip);
1547         clear_inode(inode);
1548         gfs2_dir_hash_inval(ip);
1549         if (gfs2_holder_initialized(&ip->i_iopen_gh)) {
1550                 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1551
1552                 glock_clear_object(gl, ip);
1553                 gfs2_glock_hold(gl);
1554                 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1555                 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1556                 gfs2_glock_put_eventually(gl);
1557         }
1558         if (ip->i_gl) {
1559                 glock_clear_object(ip->i_gl, ip);
1560                 wait_on_bit_io(&ip->i_flags, GIF_GLOP_PENDING, TASK_UNINTERRUPTIBLE);
1561                 gfs2_glock_add_to_lru(ip->i_gl);
1562                 gfs2_glock_put_eventually(ip->i_gl);
1563                 ip->i_gl = NULL;
1564         }
1565 }
1566
1567 static struct inode *gfs2_alloc_inode(struct super_block *sb)
1568 {
1569         struct gfs2_inode *ip;
1570
1571         ip = alloc_inode_sb(sb, gfs2_inode_cachep, GFP_KERNEL);
1572         if (!ip)
1573                 return NULL;
1574         ip->i_no_addr = 0;
1575         ip->i_flags = 0;
1576         ip->i_gl = NULL;
1577         gfs2_holder_mark_uninitialized(&ip->i_iopen_gh);
1578         memset(&ip->i_res, 0, sizeof(ip->i_res));
1579         RB_CLEAR_NODE(&ip->i_res.rs_node);
1580         ip->i_rahead = 0;
1581         return &ip->i_inode;
1582 }
1583
1584 static void gfs2_free_inode(struct inode *inode)
1585 {
1586         kmem_cache_free(gfs2_inode_cachep, GFS2_I(inode));
1587 }
1588
1589 extern void free_local_statfs_inodes(struct gfs2_sbd *sdp)
1590 {
1591         struct local_statfs_inode *lsi, *safe;
1592
1593         /* Run through the statfs inodes list to iput and free memory */
1594         list_for_each_entry_safe(lsi, safe, &sdp->sd_sc_inodes_list, si_list) {
1595                 if (lsi->si_jid == sdp->sd_jdesc->jd_jid)
1596                         sdp->sd_sc_inode = NULL; /* belongs to this node */
1597                 if (lsi->si_sc_inode)
1598                         iput(lsi->si_sc_inode);
1599                 list_del(&lsi->si_list);
1600                 kfree(lsi);
1601         }
1602 }
1603
1604 extern struct inode *find_local_statfs_inode(struct gfs2_sbd *sdp,
1605                                              unsigned int index)
1606 {
1607         struct local_statfs_inode *lsi;
1608
1609         /* Return the local (per node) statfs inode in the
1610          * sdp->sd_sc_inodes_list corresponding to the 'index'. */
1611         list_for_each_entry(lsi, &sdp->sd_sc_inodes_list, si_list) {
1612                 if (lsi->si_jid == index)
1613                         return lsi->si_sc_inode;
1614         }
1615         return NULL;
1616 }
1617
1618 const struct super_operations gfs2_super_ops = {
1619         .alloc_inode            = gfs2_alloc_inode,
1620         .free_inode             = gfs2_free_inode,
1621         .write_inode            = gfs2_write_inode,
1622         .dirty_inode            = gfs2_dirty_inode,
1623         .evict_inode            = gfs2_evict_inode,
1624         .put_super              = gfs2_put_super,
1625         .sync_fs                = gfs2_sync_fs,
1626         .freeze_super           = gfs2_freeze_super,
1627         .thaw_super             = gfs2_thaw_super,
1628         .statfs                 = gfs2_statfs,
1629         .drop_inode             = gfs2_drop_inode,
1630         .show_options           = gfs2_show_options,
1631 };
1632