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_get_atime_sec(inode));
 414         str->di_mtime = cpu_to_be64(inode_get_mtime_sec(inode));
 415         str->di_ctime = cpu_to_be64(inode_get_ctime_sec(inode));
 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_get_atime_nsec(inode));
 431         str->di_mtime_nsec = cpu_to_be32(inode_get_mtime_nsec(inode));
 432         str->di_ctime_nsec = cpu_to_be32(inode_get_ctime_nsec(inode));
 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_KILL, &sdp->sd_flags))
 550                 gfs2_flush_delete_work(sdp);
 551
 552         gfs2_destroy_threads(sdp);
 553
 554         if (log_write_allowed) {
 555                 gfs2_quota_sync(sdp->sd_vfs, 0);
 556                 gfs2_statfs_sync(sdp->sd_vfs, 0);
 557
 558                 /* We do two log flushes here. The first one commits dirty inodes
 559                  * and rgrps to the journal, but queues up revokes to the ail list.
 560                  * The second flush writes out and removes the revokes.
 561                  *
 562                  * The first must be done before the FLUSH_SHUTDOWN code
 563                  * clears the LIVE flag, otherwise it will not be able to start
 564                  * a transaction to write its revokes, and the error will cause
 565                  * a withdraw of the file system. */
 566                 gfs2_log_flush(sdp, NULL, GFS2_LFC_MAKE_FS_RO);
 567                 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SHUTDOWN |
 568                                GFS2_LFC_MAKE_FS_RO);
 569                 wait_event_timeout(sdp->sd_log_waitq,
 570                                    gfs2_log_is_empty(sdp),
 571                                    HZ * 5);
 572                 gfs2_assert_warn(sdp, gfs2_log_is_empty(sdp));
 573         }
 574         gfs2_quota_cleanup(sdp);
 575 }
 576
 577 /**
 578  * gfs2_put_super - Unmount the filesystem
 579  * @sb: The VFS superblock
 580  *
 581  */
 582
 583 static void gfs2_put_super(struct super_block *sb)
 584 {
 585         struct gfs2_sbd *sdp = sb->s_fs_info;
 586         struct gfs2_jdesc *jd;
 587
 588         /* No more recovery requests */
 589         set_bit(SDF_NORECOVERY, &sdp->sd_flags);
 590         smp_mb();
 591
 592         /* Wait on outstanding recovery */
 593 restart:
 594         spin_lock(&sdp->sd_jindex_spin);
 595         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 596                 if (!test_bit(JDF_RECOVERY, &jd->jd_flags))
 597                         continue;
 598                 spin_unlock(&sdp->sd_jindex_spin);
 599                 wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
 600                             TASK_UNINTERRUPTIBLE);
 601                 goto restart;
 602         }
 603         spin_unlock(&sdp->sd_jindex_spin);
 604
 605         if (!sb_rdonly(sb))
 606                 gfs2_make_fs_ro(sdp);
 607         else {
 608                 if (gfs2_withdrawn(sdp))
 609                         gfs2_destroy_threads(sdp);
 610
 611                 gfs2_quota_cleanup(sdp);
 612         }
 613
 614         WARN_ON(gfs2_withdrawing(sdp));
 615
 616         /*  At this point, we're through modifying the disk  */
 617
 618         /*  Release stuff  */
 619
 620         gfs2_freeze_unlock(&sdp->sd_freeze_gh);
 621
 622         iput(sdp->sd_jindex);
 623         iput(sdp->sd_statfs_inode);
 624         iput(sdp->sd_rindex);
 625         iput(sdp->sd_quota_inode);
 626
 627         gfs2_glock_put(sdp->sd_rename_gl);
 628         gfs2_glock_put(sdp->sd_freeze_gl);
 629
 630         if (!sdp->sd_args.ar_spectator) {
 631                 if (gfs2_holder_initialized(&sdp->sd_journal_gh))
 632                         gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
 633                 if (gfs2_holder_initialized(&sdp->sd_jinode_gh))
 634                         gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
 635                 brelse(sdp->sd_sc_bh);
 636                 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
 637                 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
 638                 free_local_statfs_inodes(sdp);
 639                 iput(sdp->sd_qc_inode);
 640         }
 641
 642         gfs2_glock_dq_uninit(&sdp->sd_live_gh);
 643         gfs2_clear_rgrpd(sdp);
 644         gfs2_jindex_free(sdp);
 645         /*  Take apart glock structures and buffer lists  */
 646         gfs2_gl_hash_clear(sdp);
 647         truncate_inode_pages_final(&sdp->sd_aspace);
 648         gfs2_delete_debugfs_file(sdp);
 649         /*  Unmount the locking protocol  */
 650         gfs2_lm_unmount(sdp);
 651
 652         /*  At this point, we're through participating in the lockspace  */
 653         gfs2_sys_fs_del(sdp);
 654         free_sbd(sdp);
 655 }
 656
 657 /**
 658  * gfs2_sync_fs - sync the filesystem
 659  * @sb: the superblock
 660  * @wait: true to wait for completion
 661  *
 662  * Flushes the log to disk.
 663  */
 664
 665 static int gfs2_sync_fs(struct super_block *sb, int wait)
 666 {
 667         struct gfs2_sbd *sdp = sb->s_fs_info;
 668
 669         gfs2_quota_sync(sb, -1);
 670         if (wait)
 671                 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
 672                                GFS2_LFC_SYNC_FS);
 673         return sdp->sd_log_error;
 674 }
 675
 676 static int gfs2_freeze_locally(struct gfs2_sbd *sdp)
 677 {
 678         struct super_block *sb = sdp->sd_vfs;
 679         int error;
 680
 681         error = freeze_super(sb, FREEZE_HOLDER_USERSPACE);
 682         if (error)
 683                 return error;
 684
 685         if (test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
 686                 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_FREEZE |
 687                                GFS2_LFC_FREEZE_GO_SYNC);
 688                 if (gfs2_withdrawn(sdp)) {
 689                         error = thaw_super(sb, FREEZE_HOLDER_USERSPACE);
 690                         if (error)
 691                                 return error;
 692                         return -EIO;
 693                 }
 694         }
 695         return 0;
 696 }
 697
 698 static int gfs2_do_thaw(struct gfs2_sbd *sdp)
 699 {
 700         struct super_block *sb = sdp->sd_vfs;
 701         int error;
 702
 703         error = gfs2_freeze_lock_shared(sdp);
 704         if (error)
 705                 goto fail;
 706         error = thaw_super(sb, FREEZE_HOLDER_USERSPACE);
 707         if (!error)
 708                 return 0;
 709
 710 fail:
 711         fs_info(sdp, "GFS2: couldn't thaw filesystem: %d\n", error);
 712         gfs2_assert_withdraw(sdp, 0);
 713         return error;
 714 }
 715
 716 void gfs2_freeze_func(struct work_struct *work)
 717 {
 718         struct gfs2_sbd *sdp = container_of(work, struct gfs2_sbd, sd_freeze_work);
 719         struct super_block *sb = sdp->sd_vfs;
 720         int error;
 721
 722         mutex_lock(&sdp->sd_freeze_mutex);
 723         error = -EBUSY;
 724         if (test_bit(SDF_FROZEN, &sdp->sd_flags))
 725                 goto freeze_failed;
 726
 727         error = gfs2_freeze_locally(sdp);
 728         if (error)
 729                 goto freeze_failed;
 730
 731         gfs2_freeze_unlock(&sdp->sd_freeze_gh);
 732         set_bit(SDF_FROZEN, &sdp->sd_flags);
 733
 734         error = gfs2_do_thaw(sdp);
 735         if (error)
 736                 goto out;
 737
 738         clear_bit(SDF_FROZEN, &sdp->sd_flags);
 739         goto out;
 740
 741 freeze_failed:
 742         fs_info(sdp, "GFS2: couldn't freeze filesystem: %d\n", error);
 743
 744 out:
 745         mutex_unlock(&sdp->sd_freeze_mutex);
 746         deactivate_super(sb);
 747 }
 748
 749 /**
 750  * gfs2_freeze_super - prevent further writes to the filesystem
 751  * @sb: the VFS structure for the filesystem
 752  *
 753  */
 754
 755 static int gfs2_freeze_super(struct super_block *sb, enum freeze_holder who)
 756 {
 757         struct gfs2_sbd *sdp = sb->s_fs_info;
 758         int error;
 759
 760         if (!mutex_trylock(&sdp->sd_freeze_mutex))
 761                 return -EBUSY;
 762         error = -EBUSY;
 763         if (test_bit(SDF_FROZEN, &sdp->sd_flags))
 764                 goto out;
 765
 766         for (;;) {
 767                 error = gfs2_freeze_locally(sdp);
 768                 if (error) {
 769                         fs_info(sdp, "GFS2: couldn't freeze filesystem: %d\n",
 770                                 error);
 771                         goto out;
 772                 }
 773
 774                 error = gfs2_lock_fs_check_clean(sdp);
 775                 if (!error)
 776                         break;  /* success */
 777
 778                 error = gfs2_do_thaw(sdp);
 779                 if (error)
 780                         goto out;
 781
 782                 if (error == -EBUSY)
 783                         fs_err(sdp, "waiting for recovery before freeze\n");
 784                 else if (error == -EIO) {
 785                         fs_err(sdp, "Fatal IO error: cannot freeze gfs2 due "
 786                                "to recovery error.\n");
 787                         goto out;
 788                 } else {
 789                         fs_err(sdp, "error freezing FS: %d\n", error);
 790                 }
 791                 fs_err(sdp, "retrying...\n");
 792                 msleep(1000);
 793         }
 794
 795 out:
 796         if (!error) {
 797                 set_bit(SDF_FREEZE_INITIATOR, &sdp->sd_flags);
 798                 set_bit(SDF_FROZEN, &sdp->sd_flags);
 799         }
 800         mutex_unlock(&sdp->sd_freeze_mutex);
 801         return error;
 802 }
 803
 804 /**
 805  * gfs2_thaw_super - reallow writes to the filesystem
 806  * @sb: the VFS structure for the filesystem
 807  *
 808  */
 809
 810 static int gfs2_thaw_super(struct super_block *sb, enum freeze_holder who)
 811 {
 812         struct gfs2_sbd *sdp = sb->s_fs_info;
 813         int error;
 814
 815         if (!mutex_trylock(&sdp->sd_freeze_mutex))
 816                 return -EBUSY;
 817         error = -EINVAL;
 818         if (!test_bit(SDF_FREEZE_INITIATOR, &sdp->sd_flags))
 819                 goto out;
 820
 821         gfs2_freeze_unlock(&sdp->sd_freeze_gh);
 822
 823         error = gfs2_do_thaw(sdp);
 824
 825         if (!error) {
 826                 clear_bit(SDF_FREEZE_INITIATOR, &sdp->sd_flags);
 827                 clear_bit(SDF_FROZEN, &sdp->sd_flags);
 828         }
 829 out:
 830         mutex_unlock(&sdp->sd_freeze_mutex);
 831         return error;
 832 }
 833
 834 void gfs2_thaw_freeze_initiator(struct super_block *sb)
 835 {
 836         struct gfs2_sbd *sdp = sb->s_fs_info;
 837
 838         mutex_lock(&sdp->sd_freeze_mutex);
 839         if (!test_bit(SDF_FREEZE_INITIATOR, &sdp->sd_flags))
 840                 goto out;
 841
 842         gfs2_freeze_unlock(&sdp->sd_freeze_gh);
 843
 844 out:
 845         mutex_unlock(&sdp->sd_freeze_mutex);
 846 }
 847
 848 /**
 849  * statfs_slow_fill - fill in the sg for a given RG
 850  * @rgd: the RG
 851  * @sc: the sc structure
 852  *
 853  * Returns: 0 on success, -ESTALE if the LVB is invalid
 854  */
 855
 856 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
 857                             struct gfs2_statfs_change_host *sc)
 858 {
 859         gfs2_rgrp_verify(rgd);
 860         sc->sc_total += rgd->rd_data;
 861         sc->sc_free += rgd->rd_free;
 862         sc->sc_dinodes += rgd->rd_dinodes;
 863         return 0;
 864 }
 865
 866 /**
 867  * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
 868  * @sdp: the filesystem
 869  * @sc: the sc info that will be returned
 870  *
 871  * Any error (other than a signal) will cause this routine to fall back
 872  * to the synchronous version.
 873  *
 874  * FIXME: This really shouldn't busy wait like this.
 875  *
 876  * Returns: errno
 877  */
 878
 879 static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
 880 {
 881         struct gfs2_rgrpd *rgd_next;
 882         struct gfs2_holder *gha, *gh;
 883         unsigned int slots = 64;
 884         unsigned int x;
 885         int done;
 886         int error = 0, err;
 887
 888         memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
 889         gha = kmalloc_array(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
 890         if (!gha)
 891                 return -ENOMEM;
 892         for (x = 0; x < slots; x++)
 893                 gfs2_holder_mark_uninitialized(gha + x);
 894
 895         rgd_next = gfs2_rgrpd_get_first(sdp);
 896
 897         for (;;) {
 898                 done = 1;
 899
 900                 for (x = 0; x < slots; x++) {
 901                         gh = gha + x;
 902
 903                         if (gfs2_holder_initialized(gh) && gfs2_glock_poll(gh)) {
 904                                 err = gfs2_glock_wait(gh);
 905                                 if (err) {
 906                                         gfs2_holder_uninit(gh);
 907                                         error = err;
 908                                 } else {
 909                                         if (!error) {
 910                                                 struct gfs2_rgrpd *rgd =
 911                                                         gfs2_glock2rgrp(gh->gh_gl);
 912
 913                                                 error = statfs_slow_fill(rgd, sc);
 914                                         }
 915                                         gfs2_glock_dq_uninit(gh);
 916                                 }
 917                         }
 918
 919                         if (gfs2_holder_initialized(gh))
 920                                 done = 0;
 921                         else if (rgd_next && !error) {
 922                                 error = gfs2_glock_nq_init(rgd_next->rd_gl,
 923                                                            LM_ST_SHARED,
 924                                                            GL_ASYNC,
 925                                                            gh);
 926                                 rgd_next = gfs2_rgrpd_get_next(rgd_next);
 927                                 done = 0;
 928                         }
 929
 930                         if (signal_pending(current))
 931                                 error = -ERESTARTSYS;
 932                 }
 933
 934                 if (done)
 935                         break;
 936
 937                 yield();
 938         }
 939
 940         kfree(gha);
 941         return error;
 942 }
 943
 944 /**
 945  * gfs2_statfs_i - Do a statfs
 946  * @sdp: the filesystem
 947  * @sc: the sc structure
 948  *
 949  * Returns: errno
 950  */
 951
 952 static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
 953 {
 954         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 955         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 956
 957         spin_lock(&sdp->sd_statfs_spin);
 958
 959         *sc = *m_sc;
 960         sc->sc_total += l_sc->sc_total;
 961         sc->sc_free += l_sc->sc_free;
 962         sc->sc_dinodes += l_sc->sc_dinodes;
 963
 964         spin_unlock(&sdp->sd_statfs_spin);
 965
 966         if (sc->sc_free < 0)
 967                 sc->sc_free = 0;
 968         if (sc->sc_free > sc->sc_total)
 969                 sc->sc_free = sc->sc_total;
 970         if (sc->sc_dinodes < 0)
 971                 sc->sc_dinodes = 0;
 972
 973         return 0;
 974 }
 975
 976 /**
 977  * gfs2_statfs - Gather and return stats about the filesystem
 978  * @dentry: The name of the link
 979  * @buf: The buffer
 980  *
 981  * Returns: 0 on success or error code
 982  */
 983
 984 static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
 985 {
 986         struct super_block *sb = dentry->d_sb;
 987         struct gfs2_sbd *sdp = sb->s_fs_info;
 988         struct gfs2_statfs_change_host sc;
 989         int error;
 990
 991         error = gfs2_rindex_update(sdp);
 992         if (error)
 993                 return error;
 994
 995         if (gfs2_tune_get(sdp, gt_statfs_slow))
 996                 error = gfs2_statfs_slow(sdp, &sc);
 997         else
 998                 error = gfs2_statfs_i(sdp, &sc);
 999
1000         if (error)
1001                 return error;
1002
1003         buf->f_type = GFS2_MAGIC;
1004         buf->f_bsize = sdp->sd_sb.sb_bsize;
1005         buf->f_blocks = sc.sc_total;
1006         buf->f_bfree = sc.sc_free;
1007         buf->f_bavail = sc.sc_free;
1008         buf->f_files = sc.sc_dinodes + sc.sc_free;
1009         buf->f_ffree = sc.sc_free;
1010         buf->f_namelen = GFS2_FNAMESIZE;
1011         buf->f_fsid = uuid_to_fsid(sb->s_uuid.b);
1012
1013         return 0;
1014 }
1015
1016 /**
1017  * gfs2_drop_inode - Drop an inode (test for remote unlink)
1018  * @inode: The inode to drop
1019  *
1020  * If we've received a callback on an iopen lock then it's because a
1021  * remote node tried to deallocate the inode but failed due to this node
1022  * still having the inode open. Here we mark the link count zero
1023  * since we know that it must have reached zero if the GLF_DEMOTE flag
1024  * is set on the iopen glock. If we didn't do a disk read since the
1025  * remote node removed the final link then we might otherwise miss
1026  * this event. This check ensures that this node will deallocate the
1027  * inode's blocks, or alternatively pass the baton on to another
1028  * node for later deallocation.
1029  */
1030
1031 static int gfs2_drop_inode(struct inode *inode)
1032 {
1033         struct gfs2_inode *ip = GFS2_I(inode);
1034         struct gfs2_sbd *sdp = GFS2_SB(inode);
1035
1036         if (inode->i_nlink &&
1037             gfs2_holder_initialized(&ip->i_iopen_gh)) {
1038                 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1039                 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
1040                         clear_nlink(inode);
1041         }
1042
1043         /*
1044          * When under memory pressure when an inode's link count has dropped to
1045          * zero, defer deleting the inode to the delete workqueue.  This avoids
1046          * calling into DLM under memory pressure, which can deadlock.
1047          */
1048         if (!inode->i_nlink &&
1049             unlikely(current->flags & PF_MEMALLOC) &&
1050             gfs2_holder_initialized(&ip->i_iopen_gh)) {
1051                 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1052
1053                 gfs2_glock_hold(gl);
1054                 if (!gfs2_queue_try_to_evict(gl))
1055                         gfs2_glock_queue_put(gl);
1056                 return 0;
1057         }
1058
1059         /*
1060          * No longer cache inodes when trying to evict them all.
1061          */
1062         if (test_bit(SDF_EVICTING, &sdp->sd_flags))
1063                 return 1;
1064
1065         return generic_drop_inode(inode);
1066 }
1067
1068 static int is_ancestor(const struct dentry *d1, const struct dentry *d2)
1069 {
1070         do {
1071                 if (d1 == d2)
1072                         return 1;
1073                 d1 = d1->d_parent;
1074         } while (!IS_ROOT(d1));
1075         return 0;
1076 }
1077
1078 /**
1079  * gfs2_show_options - Show mount options for /proc/mounts
1080  * @s: seq_file structure
1081  * @root: root of this (sub)tree
1082  *
1083  * Returns: 0 on success or error code
1084  */
1085
1086 static int gfs2_show_options(struct seq_file *s, struct dentry *root)
1087 {
1088         struct gfs2_sbd *sdp = root->d_sb->s_fs_info;
1089         struct gfs2_args *args = &sdp->sd_args;
1090         unsigned int logd_secs, statfs_slow, statfs_quantum, quota_quantum;
1091
1092         spin_lock(&sdp->sd_tune.gt_spin);
1093         logd_secs = sdp->sd_tune.gt_logd_secs;
1094         quota_quantum = sdp->sd_tune.gt_quota_quantum;
1095         statfs_quantum = sdp->sd_tune.gt_statfs_quantum;
1096         statfs_slow = sdp->sd_tune.gt_statfs_slow;
1097         spin_unlock(&sdp->sd_tune.gt_spin);
1098
1099         if (is_ancestor(root, sdp->sd_master_dir))
1100                 seq_puts(s, ",meta");
1101         if (args->ar_lockproto[0])
1102                 seq_show_option(s, "lockproto", args->ar_lockproto);
1103         if (args->ar_locktable[0])
1104                 seq_show_option(s, "locktable", args->ar_locktable);
1105         if (args->ar_hostdata[0])
1106                 seq_show_option(s, "hostdata", args->ar_hostdata);
1107         if (args->ar_spectator)
1108                 seq_puts(s, ",spectator");
1109         if (args->ar_localflocks)
1110                 seq_puts(s, ",localflocks");
1111         if (args->ar_debug)
1112                 seq_puts(s, ",debug");
1113         if (args->ar_posix_acl)
1114                 seq_puts(s, ",acl");
1115         if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
1116                 char *state;
1117                 switch (args->ar_quota) {
1118                 case GFS2_QUOTA_OFF:
1119                         state = "off";
1120                         break;
1121                 case GFS2_QUOTA_ACCOUNT:
1122                         state = "account";
1123                         break;
1124                 case GFS2_QUOTA_ON:
1125                         state = "on";
1126                         break;
1127                 case GFS2_QUOTA_QUIET:
1128                         state = "quiet";
1129                         break;
1130                 default:
1131                         state = "unknown";
1132                         break;
1133                 }
1134                 seq_printf(s, ",quota=%s", state);
1135         }
1136         if (args->ar_suiddir)
1137                 seq_puts(s, ",suiddir");
1138         if (args->ar_data != GFS2_DATA_DEFAULT) {
1139                 char *state;
1140                 switch (args->ar_data) {
1141                 case GFS2_DATA_WRITEBACK:
1142                         state = "writeback";
1143                         break;
1144                 case GFS2_DATA_ORDERED:
1145                         state = "ordered";
1146                         break;
1147                 default:
1148                         state = "unknown";
1149                         break;
1150                 }
1151                 seq_printf(s, ",data=%s", state);
1152         }
1153         if (args->ar_discard)
1154                 seq_puts(s, ",discard");
1155         if (logd_secs != 30)
1156                 seq_printf(s, ",commit=%d", logd_secs);
1157         if (statfs_quantum != 30)
1158                 seq_printf(s, ",statfs_quantum=%d", statfs_quantum);
1159         else if (statfs_slow)
1160                 seq_puts(s, ",statfs_quantum=0");
1161         if (quota_quantum != 60)
1162                 seq_printf(s, ",quota_quantum=%d", quota_quantum);
1163         if (args->ar_statfs_percent)
1164                 seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent);
1165         if (args->ar_errors != GFS2_ERRORS_DEFAULT) {
1166                 const char *state;
1167
1168                 switch (args->ar_errors) {
1169                 case GFS2_ERRORS_WITHDRAW:
1170                         state = "withdraw";
1171                         break;
1172                 case GFS2_ERRORS_PANIC:
1173                         state = "panic";
1174                         break;
1175                 default:
1176                         state = "unknown";
1177                         break;
1178                 }
1179                 seq_printf(s, ",errors=%s", state);
1180         }
1181         if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags))
1182                 seq_puts(s, ",nobarrier");
1183         if (test_bit(SDF_DEMOTE, &sdp->sd_flags))
1184                 seq_puts(s, ",demote_interface_used");
1185         if (args->ar_rgrplvb)
1186                 seq_puts(s, ",rgrplvb");
1187         if (args->ar_loccookie)
1188                 seq_puts(s, ",loccookie");
1189         return 0;
1190 }
1191
1192 static void gfs2_final_release_pages(struct gfs2_inode *ip)
1193 {
1194         struct inode *inode = &ip->i_inode;
1195         struct gfs2_glock *gl = ip->i_gl;
1196
1197         if (unlikely(!gl)) {
1198                 /* This can only happen during incomplete inode creation. */
1199                 BUG_ON(!test_bit(GIF_ALLOC_FAILED, &ip->i_flags));
1200                 return;
1201         }
1202
1203         truncate_inode_pages(gfs2_glock2aspace(gl), 0);
1204         truncate_inode_pages(&inode->i_data, 0);
1205
1206         if (atomic_read(&gl->gl_revokes) == 0) {
1207                 clear_bit(GLF_LFLUSH, &gl->gl_flags);
1208                 clear_bit(GLF_DIRTY, &gl->gl_flags);
1209         }
1210 }
1211
1212 static int gfs2_dinode_dealloc(struct gfs2_inode *ip)
1213 {
1214         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1215         struct gfs2_rgrpd *rgd;
1216         struct gfs2_holder gh;
1217         int error;
1218
1219         if (gfs2_get_inode_blocks(&ip->i_inode) != 1) {
1220                 gfs2_consist_inode(ip);
1221                 return -EIO;
1222         }
1223
1224         gfs2_rindex_update(sdp);
1225
1226         error = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
1227         if (error)
1228                 return error;
1229
1230         rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1231         if (!rgd) {
1232                 gfs2_consist_inode(ip);
1233                 error = -EIO;
1234                 goto out_qs;
1235         }
1236
1237         error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE,
1238                                    LM_FLAG_NODE_SCOPE, &gh);
1239         if (error)
1240                 goto out_qs;
1241
1242         error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA,
1243                                  sdp->sd_jdesc->jd_blocks);
1244         if (error)
1245                 goto out_rg_gunlock;
1246
1247         gfs2_free_di(rgd, ip);
1248
1249         gfs2_final_release_pages(ip);
1250
1251         gfs2_trans_end(sdp);
1252
1253 out_rg_gunlock:
1254         gfs2_glock_dq_uninit(&gh);
1255 out_qs:
1256         gfs2_quota_unhold(ip);
1257         return error;
1258 }
1259
1260 /**
1261  * gfs2_glock_put_eventually
1262  * @gl: The glock to put
1263  *
1264  * When under memory pressure, trigger a deferred glock put to make sure we
1265  * won't call into DLM and deadlock.  Otherwise, put the glock directly.
1266  */
1267
1268 static void gfs2_glock_put_eventually(struct gfs2_glock *gl)
1269 {
1270         if (current->flags & PF_MEMALLOC)
1271                 gfs2_glock_queue_put(gl);
1272         else
1273                 gfs2_glock_put(gl);
1274 }
1275
1276 static bool gfs2_upgrade_iopen_glock(struct inode *inode)
1277 {
1278         struct gfs2_inode *ip = GFS2_I(inode);
1279         struct gfs2_sbd *sdp = GFS2_SB(inode);
1280         struct gfs2_holder *gh = &ip->i_iopen_gh;
1281         long timeout = 5 * HZ;
1282         int error;
1283
1284         gh->gh_flags |= GL_NOCACHE;
1285         gfs2_glock_dq_wait(gh);
1286
1287         /*
1288          * If there are no other lock holders, we will immediately get
1289          * exclusive access to the iopen glock here.
1290          *
1291          * Otherwise, the other nodes holding the lock will be notified about
1292          * our locking request.  If they do not have the inode open, they are
1293          * expected to evict the cached inode and release the lock, allowing us
1294          * to proceed.
1295          *
1296          * Otherwise, if they cannot evict the inode, they are expected to poke
1297          * the inode glock (note: not the iopen glock).  We will notice that
1298          * and stop waiting for the iopen glock immediately.  The other node(s)
1299          * are then expected to take care of deleting the inode when they no
1300          * longer use it.
1301          *
1302          * As a last resort, if another node keeps holding the iopen glock
1303          * without showing any activity on the inode glock, we will eventually
1304          * time out and fail the iopen glock upgrade.
1305          */
1306
1307         gfs2_holder_reinit(LM_ST_EXCLUSIVE, GL_ASYNC | GL_NOCACHE, gh);
1308         error = gfs2_glock_nq(gh);
1309         if (error)
1310                 return false;
1311
1312         timeout = wait_event_interruptible_timeout(sdp->sd_async_glock_wait,
1313                 !test_bit(HIF_WAIT, &gh->gh_iflags) ||
1314                 test_bit(GLF_DEMOTE, &ip->i_gl->gl_flags),
1315                 timeout);
1316         if (!test_bit(HIF_HOLDER, &gh->gh_iflags)) {
1317                 gfs2_glock_dq(gh);
1318                 return false;
1319         }
1320         return gfs2_glock_holder_ready(gh) == 0;
1321 }
1322
1323 /**
1324  * evict_should_delete - determine whether the inode is eligible for deletion
1325  * @inode: The inode to evict
1326  * @gh: The glock holder structure
1327  *
1328  * This function determines whether the evicted inode is eligible to be deleted
1329  * and locks the inode glock.
1330  *
1331  * Returns: the fate of the dinode
1332  */
1333 static enum dinode_demise evict_should_delete(struct inode *inode,
1334                                               struct gfs2_holder *gh)
1335 {
1336         struct gfs2_inode *ip = GFS2_I(inode);
1337         struct super_block *sb = inode->i_sb;
1338         struct gfs2_sbd *sdp = sb->s_fs_info;
1339         int ret;
1340
1341         if (unlikely(test_bit(GIF_ALLOC_FAILED, &ip->i_flags)))
1342                 goto should_delete;
1343
1344         if (test_bit(GIF_DEFERRED_DELETE, &ip->i_flags))
1345                 return SHOULD_DEFER_EVICTION;
1346
1347         /* Deletes should never happen under memory pressure anymore.  */
1348         if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
1349                 return SHOULD_DEFER_EVICTION;
1350
1351         /* Must not read inode block until block type has been verified */
1352         ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, gh);
1353         if (unlikely(ret)) {
1354                 glock_clear_object(ip->i_iopen_gh.gh_gl, ip);
1355                 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1356                 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1357                 return SHOULD_DEFER_EVICTION;
1358         }
1359
1360         if (gfs2_inode_already_deleted(ip->i_gl, ip->i_no_formal_ino))
1361                 return SHOULD_NOT_DELETE_DINODE;
1362         ret = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED);
1363         if (ret)
1364                 return SHOULD_NOT_DELETE_DINODE;
1365
1366         ret = gfs2_instantiate(gh);
1367         if (ret)
1368                 return SHOULD_NOT_DELETE_DINODE;
1369
1370         /*
1371          * The inode may have been recreated in the meantime.
1372          */
1373         if (inode->i_nlink)
1374                 return SHOULD_NOT_DELETE_DINODE;
1375
1376 should_delete:
1377         if (gfs2_holder_initialized(&ip->i_iopen_gh) &&
1378             test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1379                 if (!gfs2_upgrade_iopen_glock(inode)) {
1380                         gfs2_holder_uninit(&ip->i_iopen_gh);
1381                         return SHOULD_NOT_DELETE_DINODE;
1382                 }
1383         }
1384         return SHOULD_DELETE_DINODE;
1385 }
1386
1387 /**
1388  * evict_unlinked_inode - delete the pieces of an unlinked evicted inode
1389  * @inode: The inode to evict
1390  */
1391 static int evict_unlinked_inode(struct inode *inode)
1392 {
1393         struct gfs2_inode *ip = GFS2_I(inode);
1394         int ret;
1395
1396         if (S_ISDIR(inode->i_mode) &&
1397             (ip->i_diskflags & GFS2_DIF_EXHASH)) {
1398                 ret = gfs2_dir_exhash_dealloc(ip);
1399                 if (ret)
1400                         goto out;
1401         }
1402
1403         if (ip->i_eattr) {
1404                 ret = gfs2_ea_dealloc(ip);
1405                 if (ret)
1406                         goto out;
1407         }
1408
1409         if (!gfs2_is_stuffed(ip)) {
1410                 ret = gfs2_file_dealloc(ip);
1411                 if (ret)
1412                         goto out;
1413         }
1414
1415         /*
1416          * As soon as we clear the bitmap for the dinode, gfs2_create_inode()
1417          * can get called to recreate it, or even gfs2_inode_lookup() if the
1418          * inode was recreated on another node in the meantime.
1419          *
1420          * However, inserting the new inode into the inode hash table will not
1421          * succeed until the old inode is removed, and that only happens after
1422          * ->evict_inode() returns.  The new inode is attached to its inode and
1423          *  iopen glocks after inserting it into the inode hash table, so at
1424          *  that point we can be sure that both glocks are unused.
1425          */
1426
1427         ret = gfs2_dinode_dealloc(ip);
1428         if (!ret && ip->i_gl)
1429                 gfs2_inode_remember_delete(ip->i_gl, ip->i_no_formal_ino);
1430
1431 out:
1432         return ret;
1433 }
1434
1435 /*
1436  * evict_linked_inode - evict an inode whose dinode has not been unlinked
1437  * @inode: The inode to evict
1438  */
1439 static int evict_linked_inode(struct inode *inode)
1440 {
1441         struct super_block *sb = inode->i_sb;
1442         struct gfs2_sbd *sdp = sb->s_fs_info;
1443         struct gfs2_inode *ip = GFS2_I(inode);
1444         struct address_space *metamapping;
1445         int ret;
1446
1447         gfs2_log_flush(sdp, ip->i_gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
1448                        GFS2_LFC_EVICT_INODE);
1449         metamapping = gfs2_glock2aspace(ip->i_gl);
1450         if (test_bit(GLF_DIRTY, &ip->i_gl->gl_flags)) {
1451                 filemap_fdatawrite(metamapping);
1452                 filemap_fdatawait(metamapping);
1453         }
1454         write_inode_now(inode, 1);
1455         gfs2_ail_flush(ip->i_gl, 0);
1456
1457         ret = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
1458         if (ret)
1459                 return ret;
1460
1461         /* Needs to be done before glock release & also in a transaction */
1462         truncate_inode_pages(&inode->i_data, 0);
1463         truncate_inode_pages(metamapping, 0);
1464         gfs2_trans_end(sdp);
1465         return 0;
1466 }
1467
1468 /**
1469  * gfs2_evict_inode - Remove an inode from cache
1470  * @inode: The inode to evict
1471  *
1472  * There are three cases to consider:
1473  * 1. i_nlink == 0, we are final opener (and must deallocate)
1474  * 2. i_nlink == 0, we are not the final opener (and cannot deallocate)
1475  * 3. i_nlink > 0
1476  *
1477  * If the fs is read only, then we have to treat all cases as per #3
1478  * since we are unable to do any deallocation. The inode will be
1479  * deallocated by the next read/write node to attempt an allocation
1480  * in the same resource group
1481  *
1482  * We have to (at the moment) hold the inodes main lock to cover
1483  * the gap between unlocking the shared lock on the iopen lock and
1484  * taking the exclusive lock. I'd rather do a shared -> exclusive
1485  * conversion on the iopen lock, but we can change that later. This
1486  * is safe, just less efficient.
1487  */
1488
1489 static void gfs2_evict_inode(struct inode *inode)
1490 {
1491         struct super_block *sb = inode->i_sb;
1492         struct gfs2_sbd *sdp = sb->s_fs_info;
1493         struct gfs2_inode *ip = GFS2_I(inode);
1494         struct gfs2_holder gh;
1495         int ret;
1496
1497         if (inode->i_nlink || sb_rdonly(sb) || !ip->i_no_addr)
1498                 goto out;
1499
1500         /*
1501          * In case of an incomplete mount, gfs2_evict_inode() may be called for
1502          * system files without having an active journal to write to.  In that
1503          * case, skip the filesystem evict.
1504          */
1505         if (!sdp->sd_jdesc)
1506                 goto out;
1507
1508         gfs2_holder_mark_uninitialized(&gh);
1509         ret = evict_should_delete(inode, &gh);
1510         if (ret == SHOULD_DEFER_EVICTION)
1511                 goto out;
1512         if (ret == SHOULD_DELETE_DINODE)
1513                 ret = evict_unlinked_inode(inode);
1514         else
1515                 ret = evict_linked_inode(inode);
1516
1517         if (gfs2_rs_active(&ip->i_res))
1518                 gfs2_rs_deltree(&ip->i_res);
1519
1520         if (gfs2_holder_initialized(&gh))
1521                 gfs2_glock_dq_uninit(&gh);
1522         if (ret && ret != GLR_TRYFAILED && ret != -EROFS)
1523                 fs_warn(sdp, "gfs2_evict_inode: %d\n", ret);
1524 out:
1525         truncate_inode_pages_final(&inode->i_data);
1526         if (ip->i_qadata)
1527                 gfs2_assert_warn(sdp, ip->i_qadata->qa_ref == 0);
1528         gfs2_rs_deltree(&ip->i_res);
1529         gfs2_ordered_del_inode(ip);
1530         clear_inode(inode);
1531         gfs2_dir_hash_inval(ip);
1532         if (gfs2_holder_initialized(&ip->i_iopen_gh)) {
1533                 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1534
1535                 glock_clear_object(gl, ip);
1536                 gfs2_glock_hold(gl);
1537                 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1538                 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1539                 gfs2_glock_put_eventually(gl);
1540         }
1541         if (ip->i_gl) {
1542                 glock_clear_object(ip->i_gl, ip);
1543                 wait_on_bit_io(&ip->i_flags, GIF_GLOP_PENDING, TASK_UNINTERRUPTIBLE);
1544                 gfs2_glock_add_to_lru(ip->i_gl);
1545                 gfs2_glock_put_eventually(ip->i_gl);
1546                 rcu_assign_pointer(ip->i_gl, NULL);
1547         }
1548 }
1549
1550 static struct inode *gfs2_alloc_inode(struct super_block *sb)
1551 {
1552         struct gfs2_inode *ip;
1553
1554         ip = alloc_inode_sb(sb, gfs2_inode_cachep, GFP_KERNEL);
1555         if (!ip)
1556                 return NULL;
1557         ip->i_no_addr = 0;
1558         ip->i_flags = 0;
1559         ip->i_gl = NULL;
1560         gfs2_holder_mark_uninitialized(&ip->i_iopen_gh);
1561         memset(&ip->i_res, 0, sizeof(ip->i_res));
1562         RB_CLEAR_NODE(&ip->i_res.rs_node);
1563         ip->i_rahead = 0;
1564         return &ip->i_inode;
1565 }
1566
1567 static void gfs2_free_inode(struct inode *inode)
1568 {
1569         kmem_cache_free(gfs2_inode_cachep, GFS2_I(inode));
1570 }
1571
1572 void free_local_statfs_inodes(struct gfs2_sbd *sdp)
1573 {
1574         struct local_statfs_inode *lsi, *safe;
1575
1576         /* Run through the statfs inodes list to iput and free memory */
1577         list_for_each_entry_safe(lsi, safe, &sdp->sd_sc_inodes_list, si_list) {
1578                 if (lsi->si_jid == sdp->sd_jdesc->jd_jid)
1579                         sdp->sd_sc_inode = NULL; /* belongs to this node */
1580                 if (lsi->si_sc_inode)
1581                         iput(lsi->si_sc_inode);
1582                 list_del(&lsi->si_list);
1583                 kfree(lsi);
1584         }
1585 }
1586
1587 struct inode *find_local_statfs_inode(struct gfs2_sbd *sdp,
1588                                       unsigned int index)
1589 {
1590         struct local_statfs_inode *lsi;
1591
1592         /* Return the local (per node) statfs inode in the
1593          * sdp->sd_sc_inodes_list corresponding to the 'index'. */
1594         list_for_each_entry(lsi, &sdp->sd_sc_inodes_list, si_list) {
1595                 if (lsi->si_jid == index)
1596                         return lsi->si_sc_inode;
1597         }
1598         return NULL;
1599 }
1600
1601 const struct super_operations gfs2_super_ops = {
1602         .alloc_inode            = gfs2_alloc_inode,
1603         .free_inode             = gfs2_free_inode,
1604         .write_inode            = gfs2_write_inode,
1605         .dirty_inode            = gfs2_dirty_inode,
1606         .evict_inode            = gfs2_evict_inode,
1607         .put_super              = gfs2_put_super,
1608         .sync_fs                = gfs2_sync_fs,
1609         .freeze_super           = gfs2_freeze_super,
1610         .thaw_super             = gfs2_thaw_super,
1611         .statfs                 = gfs2_statfs,
1612         .drop_inode             = gfs2_drop_inode,
1613         .show_options           = gfs2_show_options,
1614 };
1615