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;
 336
 337         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 338                 lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
 339                 if (!lfcc) {
 340                         error = -ENOMEM;
 341                         goto out;
 342                 }
 343                 ip = GFS2_I(jd->jd_inode);
 344                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
 345                 if (error) {
 346                         kfree(lfcc);
 347                         goto out;
 348                 }
 349                 list_add(&lfcc->list, &list);
 350         }
 351
 352         error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_EXCLUSIVE,
 353                                    LM_FLAG_NOEXP | GL_NOPID,
 354                                    &sdp->sd_freeze_gh);
 355         if (error)
 356                 goto out;
 357
 358         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 359                 error = gfs2_jdesc_check(jd);
 360                 if (error)
 361                         break;
 362                 error = gfs2_find_jhead(jd, &lh, false);
 363                 if (error)
 364                         break;
 365                 if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 366                         error = -EBUSY;
 367                         break;
 368                 }
 369         }
 370
 371         if (error)
 372                 gfs2_freeze_unlock(&sdp->sd_freeze_gh);
 373
 374 out:
 375         while (!list_empty(&list)) {
 376                 lfcc = list_first_entry(&list, struct lfcc, list);
 377                 list_del(&lfcc->list);
 378                 gfs2_glock_dq_uninit(&lfcc->gh);
 379                 kfree(lfcc);
 380         }
 381         return error;
 382 }
 383
 384 void gfs2_dinode_out(const struct gfs2_inode *ip, void *buf)
 385 {
 386         const struct inode *inode = &ip->i_inode;
 387         struct gfs2_dinode *str = buf;
 388
 389         str->di_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 390         str->di_header.mh_type = cpu_to_be32(GFS2_METATYPE_DI);
 391         str->di_header.mh_format = cpu_to_be32(GFS2_FORMAT_DI);
 392         str->di_num.no_addr = cpu_to_be64(ip->i_no_addr);
 393         str->di_num.no_formal_ino = cpu_to_be64(ip->i_no_formal_ino);
 394         str->di_mode = cpu_to_be32(inode->i_mode);
 395         str->di_uid = cpu_to_be32(i_uid_read(inode));
 396         str->di_gid = cpu_to_be32(i_gid_read(inode));
 397         str->di_nlink = cpu_to_be32(inode->i_nlink);
 398         str->di_size = cpu_to_be64(i_size_read(inode));
 399         str->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(inode));
 400         str->di_atime = cpu_to_be64(inode->i_atime.tv_sec);
 401         str->di_mtime = cpu_to_be64(inode->i_mtime.tv_sec);
 402         str->di_ctime = cpu_to_be64(inode->i_ctime.tv_sec);
 403
 404         str->di_goal_meta = cpu_to_be64(ip->i_goal);
 405         str->di_goal_data = cpu_to_be64(ip->i_goal);
 406         str->di_generation = cpu_to_be64(ip->i_generation);
 407
 408         str->di_flags = cpu_to_be32(ip->i_diskflags);
 409         str->di_height = cpu_to_be16(ip->i_height);
 410         str->di_payload_format = cpu_to_be32(S_ISDIR(inode->i_mode) &&
 411                                              !(ip->i_diskflags & GFS2_DIF_EXHASH) ?
 412                                              GFS2_FORMAT_DE : 0);
 413         str->di_depth = cpu_to_be16(ip->i_depth);
 414         str->di_entries = cpu_to_be32(ip->i_entries);
 415
 416         str->di_eattr = cpu_to_be64(ip->i_eattr);
 417         str->di_atime_nsec = cpu_to_be32(inode->i_atime.tv_nsec);
 418         str->di_mtime_nsec = cpu_to_be32(inode->i_mtime.tv_nsec);
 419         str->di_ctime_nsec = cpu_to_be32(inode->i_ctime.tv_nsec);
 420 }
 421
 422 /**
 423  * gfs2_write_inode - Make sure the inode is stable on the disk
 424  * @inode: The inode
 425  * @wbc: The writeback control structure
 426  *
 427  * Returns: errno
 428  */
 429
 430 static int gfs2_write_inode(struct inode *inode, struct writeback_control *wbc)
 431 {
 432         struct gfs2_inode *ip = GFS2_I(inode);
 433         struct gfs2_sbd *sdp = GFS2_SB(inode);
 434         struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl);
 435         struct backing_dev_info *bdi = inode_to_bdi(metamapping->host);
 436         int ret = 0;
 437         bool flush_all = (wbc->sync_mode == WB_SYNC_ALL || gfs2_is_jdata(ip));
 438
 439         if (flush_all)
 440                 gfs2_log_flush(GFS2_SB(inode), ip->i_gl,
 441                                GFS2_LOG_HEAD_FLUSH_NORMAL |
 442                                GFS2_LFC_WRITE_INODE);
 443         if (bdi->wb.dirty_exceeded)
 444                 gfs2_ail1_flush(sdp, wbc);
 445         else
 446                 filemap_fdatawrite(metamapping);
 447         if (flush_all)
 448                 ret = filemap_fdatawait(metamapping);
 449         if (ret)
 450                 mark_inode_dirty_sync(inode);
 451         else {
 452                 spin_lock(&inode->i_lock);
 453                 if (!(inode->i_flags & I_DIRTY))
 454                         gfs2_ordered_del_inode(ip);
 455                 spin_unlock(&inode->i_lock);
 456         }
 457         return ret;
 458 }
 459
 460 /**
 461  * gfs2_dirty_inode - check for atime updates
 462  * @inode: The inode in question
 463  * @flags: The type of dirty
 464  *
 465  * Unfortunately it can be called under any combination of inode
 466  * glock and transaction lock, so we have to check carefully.
 467  *
 468  * At the moment this deals only with atime - it should be possible
 469  * to expand that role in future, once a review of the locking has
 470  * been carried out.
 471  */
 472
 473 static void gfs2_dirty_inode(struct inode *inode, int flags)
 474 {
 475         struct gfs2_inode *ip = GFS2_I(inode);
 476         struct gfs2_sbd *sdp = GFS2_SB(inode);
 477         struct buffer_head *bh;
 478         struct gfs2_holder gh;
 479         int need_unlock = 0;
 480         int need_endtrans = 0;
 481         int ret;
 482
 483         if (unlikely(gfs2_withdrawn(sdp)))
 484                 return;
 485         if (!gfs2_glock_is_locked_by_me(ip->i_gl)) {
 486                 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
 487                 if (ret) {
 488                         fs_err(sdp, "dirty_inode: glock %d\n", ret);
 489                         gfs2_dump_glock(NULL, ip->i_gl, true);
 490                         return;
 491                 }
 492                 need_unlock = 1;
 493         } else if (WARN_ON_ONCE(ip->i_gl->gl_state != LM_ST_EXCLUSIVE))
 494                 return;
 495
 496         if (current->journal_info == NULL) {
 497                 ret = gfs2_trans_begin(sdp, RES_DINODE, 0);
 498                 if (ret) {
 499                         fs_err(sdp, "dirty_inode: gfs2_trans_begin %d\n", ret);
 500                         goto out;
 501                 }
 502                 need_endtrans = 1;
 503         }
 504
 505         ret = gfs2_meta_inode_buffer(ip, &bh);
 506         if (ret == 0) {
 507                 gfs2_trans_add_meta(ip->i_gl, bh);
 508                 gfs2_dinode_out(ip, bh->b_data);
 509                 brelse(bh);
 510         }
 511
 512         if (need_endtrans)
 513                 gfs2_trans_end(sdp);
 514 out:
 515         if (need_unlock)
 516                 gfs2_glock_dq_uninit(&gh);
 517 }
 518
 519 /**
 520  * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
 521  * @sdp: the filesystem
 522  *
 523  * Returns: errno
 524  */
 525
 526 void gfs2_make_fs_ro(struct gfs2_sbd *sdp)
 527 {
 528         int log_write_allowed = test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 529
 530         gfs2_flush_delete_work(sdp);
 531         if (!log_write_allowed && current == sdp->sd_quotad_process)
 532                 fs_warn(sdp, "The quotad daemon is withdrawing.\n");
 533         else if (sdp->sd_quotad_process)
 534                 kthread_stop(sdp->sd_quotad_process);
 535         sdp->sd_quotad_process = NULL;
 536
 537         if (!log_write_allowed && current == sdp->sd_logd_process)
 538                 fs_warn(sdp, "The logd daemon is withdrawing.\n");
 539         else if (sdp->sd_logd_process)
 540                 kthread_stop(sdp->sd_logd_process);
 541         sdp->sd_logd_process = NULL;
 542
 543         if (log_write_allowed) {
 544                 gfs2_quota_sync(sdp->sd_vfs, 0);
 545                 gfs2_statfs_sync(sdp->sd_vfs, 0);
 546
 547                 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SHUTDOWN |
 548                                GFS2_LFC_MAKE_FS_RO);
 549                 wait_event_timeout(sdp->sd_log_waitq,
 550                                    gfs2_log_is_empty(sdp),
 551                                    HZ * 5);
 552                 gfs2_assert_warn(sdp, gfs2_log_is_empty(sdp));
 553         } else {
 554                 wait_event_timeout(sdp->sd_log_waitq,
 555                                    gfs2_log_is_empty(sdp),
 556                                    HZ * 5);
 557         }
 558         gfs2_quota_cleanup(sdp);
 559
 560         if (!log_write_allowed)
 561                 sdp->sd_vfs->s_flags |= SB_RDONLY;
 562 }
 563
 564 /**
 565  * gfs2_put_super - Unmount the filesystem
 566  * @sb: The VFS superblock
 567  *
 568  */
 569
 570 static void gfs2_put_super(struct super_block *sb)
 571 {
 572         struct gfs2_sbd *sdp = sb->s_fs_info;
 573         struct gfs2_jdesc *jd;
 574
 575         /* No more recovery requests */
 576         set_bit(SDF_NORECOVERY, &sdp->sd_flags);
 577         smp_mb();
 578
 579         /* Wait on outstanding recovery */
 580 restart:
 581         spin_lock(&sdp->sd_jindex_spin);
 582         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 583                 if (!test_bit(JDF_RECOVERY, &jd->jd_flags))
 584                         continue;
 585                 spin_unlock(&sdp->sd_jindex_spin);
 586                 wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
 587                             TASK_UNINTERRUPTIBLE);
 588                 goto restart;
 589         }
 590         spin_unlock(&sdp->sd_jindex_spin);
 591
 592         if (!sb_rdonly(sb)) {
 593                 gfs2_make_fs_ro(sdp);
 594         }
 595         WARN_ON(gfs2_withdrawing(sdp));
 596
 597         /*  At this point, we're through modifying the disk  */
 598
 599         /*  Release stuff  */
 600
 601         iput(sdp->sd_jindex);
 602         iput(sdp->sd_statfs_inode);
 603         iput(sdp->sd_rindex);
 604         iput(sdp->sd_quota_inode);
 605
 606         gfs2_glock_put(sdp->sd_rename_gl);
 607         gfs2_glock_put(sdp->sd_freeze_gl);
 608
 609         if (!sdp->sd_args.ar_spectator) {
 610                 if (gfs2_holder_initialized(&sdp->sd_journal_gh))
 611                         gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
 612                 if (gfs2_holder_initialized(&sdp->sd_jinode_gh))
 613                         gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
 614                 brelse(sdp->sd_sc_bh);
 615                 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
 616                 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
 617                 free_local_statfs_inodes(sdp);
 618                 iput(sdp->sd_qc_inode);
 619         }
 620
 621         gfs2_glock_dq_uninit(&sdp->sd_live_gh);
 622         gfs2_clear_rgrpd(sdp);
 623         gfs2_jindex_free(sdp);
 624         /*  Take apart glock structures and buffer lists  */
 625         gfs2_gl_hash_clear(sdp);
 626         truncate_inode_pages_final(&sdp->sd_aspace);
 627         gfs2_delete_debugfs_file(sdp);
 628         /*  Unmount the locking protocol  */
 629         gfs2_lm_unmount(sdp);
 630
 631         /*  At this point, we're through participating in the lockspace  */
 632         gfs2_sys_fs_del(sdp);
 633         free_sbd(sdp);
 634 }
 635
 636 /**
 637  * gfs2_sync_fs - sync the filesystem
 638  * @sb: the superblock
 639  * @wait: true to wait for completion
 640  *
 641  * Flushes the log to disk.
 642  */
 643
 644 static int gfs2_sync_fs(struct super_block *sb, int wait)
 645 {
 646         struct gfs2_sbd *sdp = sb->s_fs_info;
 647
 648         gfs2_quota_sync(sb, -1);
 649         if (wait)
 650                 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
 651                                GFS2_LFC_SYNC_FS);
 652         return sdp->sd_log_error;
 653 }
 654
 655 void gfs2_freeze_func(struct work_struct *work)
 656 {
 657         int error;
 658         struct gfs2_holder freeze_gh;
 659         struct gfs2_sbd *sdp = container_of(work, struct gfs2_sbd, sd_freeze_work);
 660         struct super_block *sb = sdp->sd_vfs;
 661
 662         atomic_inc(&sb->s_active);
 663         error = gfs2_freeze_lock(sdp, &freeze_gh, 0);
 664         if (error) {
 665                 gfs2_assert_withdraw(sdp, 0);
 666         } else {
 667                 atomic_set(&sdp->sd_freeze_state, SFS_UNFROZEN);
 668                 error = thaw_super(sb);
 669                 if (error) {
 670                         fs_info(sdp, "GFS2: couldn't thaw filesystem: %d\n",
 671                                 error);
 672                         gfs2_assert_withdraw(sdp, 0);
 673                 }
 674                 gfs2_freeze_unlock(&freeze_gh);
 675         }
 676         deactivate_super(sb);
 677         clear_bit_unlock(SDF_FS_FROZEN, &sdp->sd_flags);
 678         wake_up_bit(&sdp->sd_flags, SDF_FS_FROZEN);
 679         return;
 680 }
 681
 682 /**
 683  * gfs2_freeze - prevent further writes to the filesystem
 684  * @sb: the VFS structure for the filesystem
 685  *
 686  */
 687
 688 static int gfs2_freeze(struct super_block *sb)
 689 {
 690         struct gfs2_sbd *sdp = sb->s_fs_info;
 691         int error;
 692
 693         mutex_lock(&sdp->sd_freeze_mutex);
 694         if (atomic_read(&sdp->sd_freeze_state) != SFS_UNFROZEN) {
 695                 error = -EBUSY;
 696                 goto out;
 697         }
 698
 699         for (;;) {
 700                 if (gfs2_withdrawn(sdp)) {
 701                         error = -EINVAL;
 702                         goto out;
 703                 }
 704
 705                 error = gfs2_lock_fs_check_clean(sdp);
 706                 if (!error)
 707                         break;
 708
 709                 if (error == -EBUSY)
 710                         fs_err(sdp, "waiting for recovery before freeze\n");
 711                 else if (error == -EIO) {
 712                         fs_err(sdp, "Fatal IO error: cannot freeze gfs2 due "
 713                                "to recovery error.\n");
 714                         goto out;
 715                 } else {
 716                         fs_err(sdp, "error freezing FS: %d\n", error);
 717                 }
 718                 fs_err(sdp, "retrying...\n");
 719                 msleep(1000);
 720         }
 721         set_bit(SDF_FS_FROZEN, &sdp->sd_flags);
 722 out:
 723         mutex_unlock(&sdp->sd_freeze_mutex);
 724         return error;
 725 }
 726
 727 /**
 728  * gfs2_unfreeze - reallow writes to the filesystem
 729  * @sb: the VFS structure for the filesystem
 730  *
 731  */
 732
 733 static int gfs2_unfreeze(struct super_block *sb)
 734 {
 735         struct gfs2_sbd *sdp = sb->s_fs_info;
 736
 737         mutex_lock(&sdp->sd_freeze_mutex);
 738         if (atomic_read(&sdp->sd_freeze_state) != SFS_FROZEN ||
 739             !gfs2_holder_initialized(&sdp->sd_freeze_gh)) {
 740                 mutex_unlock(&sdp->sd_freeze_mutex);
 741                 return -EINVAL;
 742         }
 743
 744         gfs2_freeze_unlock(&sdp->sd_freeze_gh);
 745         mutex_unlock(&sdp->sd_freeze_mutex);
 746         return wait_on_bit(&sdp->sd_flags, SDF_FS_FROZEN, TASK_INTERRUPTIBLE);
 747 }
 748
 749 /**
 750  * statfs_slow_fill - fill in the sg for a given RG
 751  * @rgd: the RG
 752  * @sc: the sc structure
 753  *
 754  * Returns: 0 on success, -ESTALE if the LVB is invalid
 755  */
 756
 757 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
 758                             struct gfs2_statfs_change_host *sc)
 759 {
 760         gfs2_rgrp_verify(rgd);
 761         sc->sc_total += rgd->rd_data;
 762         sc->sc_free += rgd->rd_free;
 763         sc->sc_dinodes += rgd->rd_dinodes;
 764         return 0;
 765 }
 766
 767 /**
 768  * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
 769  * @sdp: the filesystem
 770  * @sc: the sc info that will be returned
 771  *
 772  * Any error (other than a signal) will cause this routine to fall back
 773  * to the synchronous version.
 774  *
 775  * FIXME: This really shouldn't busy wait like this.
 776  *
 777  * Returns: errno
 778  */
 779
 780 static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
 781 {
 782         struct gfs2_rgrpd *rgd_next;
 783         struct gfs2_holder *gha, *gh;
 784         unsigned int slots = 64;
 785         unsigned int x;
 786         int done;
 787         int error = 0, err;
 788
 789         memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
 790         gha = kmalloc_array(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
 791         if (!gha)
 792                 return -ENOMEM;
 793         for (x = 0; x < slots; x++)
 794                 gfs2_holder_mark_uninitialized(gha + x);
 795
 796         rgd_next = gfs2_rgrpd_get_first(sdp);
 797
 798         for (;;) {
 799                 done = 1;
 800
 801                 for (x = 0; x < slots; x++) {
 802                         gh = gha + x;
 803
 804                         if (gfs2_holder_initialized(gh) && gfs2_glock_poll(gh)) {
 805                                 err = gfs2_glock_wait(gh);
 806                                 if (err) {
 807                                         gfs2_holder_uninit(gh);
 808                                         error = err;
 809                                 } else {
 810                                         if (!error) {
 811                                                 struct gfs2_rgrpd *rgd =
 812                                                         gfs2_glock2rgrp(gh->gh_gl);
 813
 814                                                 error = statfs_slow_fill(rgd, sc);
 815                                         }
 816                                         gfs2_glock_dq_uninit(gh);
 817                                 }
 818                         }
 819
 820                         if (gfs2_holder_initialized(gh))
 821                                 done = 0;
 822                         else if (rgd_next && !error) {
 823                                 error = gfs2_glock_nq_init(rgd_next->rd_gl,
 824                                                            LM_ST_SHARED,
 825                                                            GL_ASYNC,
 826                                                            gh);
 827                                 rgd_next = gfs2_rgrpd_get_next(rgd_next);
 828                                 done = 0;
 829                         }
 830
 831                         if (signal_pending(current))
 832                                 error = -ERESTARTSYS;
 833                 }
 834
 835                 if (done)
 836                         break;
 837
 838                 yield();
 839         }
 840
 841         kfree(gha);
 842         return error;
 843 }
 844
 845 /**
 846  * gfs2_statfs_i - Do a statfs
 847  * @sdp: the filesystem
 848  * @sc: the sc structure
 849  *
 850  * Returns: errno
 851  */
 852
 853 static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
 854 {
 855         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 856         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 857
 858         spin_lock(&sdp->sd_statfs_spin);
 859
 860         *sc = *m_sc;
 861         sc->sc_total += l_sc->sc_total;
 862         sc->sc_free += l_sc->sc_free;
 863         sc->sc_dinodes += l_sc->sc_dinodes;
 864
 865         spin_unlock(&sdp->sd_statfs_spin);
 866
 867         if (sc->sc_free < 0)
 868                 sc->sc_free = 0;
 869         if (sc->sc_free > sc->sc_total)
 870                 sc->sc_free = sc->sc_total;
 871         if (sc->sc_dinodes < 0)
 872                 sc->sc_dinodes = 0;
 873
 874         return 0;
 875 }
 876
 877 /**
 878  * gfs2_statfs - Gather and return stats about the filesystem
 879  * @dentry: The name of the link
 880  * @buf: The buffer
 881  *
 882  * Returns: 0 on success or error code
 883  */
 884
 885 static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
 886 {
 887         struct super_block *sb = dentry->d_sb;
 888         struct gfs2_sbd *sdp = sb->s_fs_info;
 889         struct gfs2_statfs_change_host sc;
 890         int error;
 891
 892         error = gfs2_rindex_update(sdp);
 893         if (error)
 894                 return error;
 895
 896         if (gfs2_tune_get(sdp, gt_statfs_slow))
 897                 error = gfs2_statfs_slow(sdp, &sc);
 898         else
 899                 error = gfs2_statfs_i(sdp, &sc);
 900
 901         if (error)
 902                 return error;
 903
 904         buf->f_type = GFS2_MAGIC;
 905         buf->f_bsize = sdp->sd_sb.sb_bsize;
 906         buf->f_blocks = sc.sc_total;
 907         buf->f_bfree = sc.sc_free;
 908         buf->f_bavail = sc.sc_free;
 909         buf->f_files = sc.sc_dinodes + sc.sc_free;
 910         buf->f_ffree = sc.sc_free;
 911         buf->f_namelen = GFS2_FNAMESIZE;
 912
 913         return 0;
 914 }
 915
 916 /**
 917  * gfs2_drop_inode - Drop an inode (test for remote unlink)
 918  * @inode: The inode to drop
 919  *
 920  * If we've received a callback on an iopen lock then it's because a
 921  * remote node tried to deallocate the inode but failed due to this node
 922  * still having the inode open. Here we mark the link count zero
 923  * since we know that it must have reached zero if the GLF_DEMOTE flag
 924  * is set on the iopen glock. If we didn't do a disk read since the
 925  * remote node removed the final link then we might otherwise miss
 926  * this event. This check ensures that this node will deallocate the
 927  * inode's blocks, or alternatively pass the baton on to another
 928  * node for later deallocation.
 929  */
 930
 931 static int gfs2_drop_inode(struct inode *inode)
 932 {
 933         struct gfs2_inode *ip = GFS2_I(inode);
 934
 935         if (!test_bit(GIF_FREE_VFS_INODE, &ip->i_flags) &&
 936             inode->i_nlink &&
 937             gfs2_holder_initialized(&ip->i_iopen_gh)) {
 938                 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
 939                 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
 940                         clear_nlink(inode);
 941         }
 942
 943         /*
 944          * When under memory pressure when an inode's link count has dropped to
 945          * zero, defer deleting the inode to the delete workqueue.  This avoids
 946          * calling into DLM under memory pressure, which can deadlock.
 947          */
 948         if (!inode->i_nlink &&
 949             unlikely(current->flags & PF_MEMALLOC) &&
 950             gfs2_holder_initialized(&ip->i_iopen_gh)) {
 951                 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
 952
 953                 gfs2_glock_hold(gl);
 954                 if (!gfs2_queue_delete_work(gl, 0))
 955                         gfs2_glock_queue_put(gl);
 956                 return 0;
 957         }
 958
 959         return generic_drop_inode(inode);
 960 }
 961
 962 static int is_ancestor(const struct dentry *d1, const struct dentry *d2)
 963 {
 964         do {
 965                 if (d1 == d2)
 966                         return 1;
 967                 d1 = d1->d_parent;
 968         } while (!IS_ROOT(d1));
 969         return 0;
 970 }
 971
 972 /**
 973  * gfs2_show_options - Show mount options for /proc/mounts
 974  * @s: seq_file structure
 975  * @root: root of this (sub)tree
 976  *
 977  * Returns: 0 on success or error code
 978  */
 979
 980 static int gfs2_show_options(struct seq_file *s, struct dentry *root)
 981 {
 982         struct gfs2_sbd *sdp = root->d_sb->s_fs_info;
 983         struct gfs2_args *args = &sdp->sd_args;
 984         unsigned int logd_secs, statfs_slow, statfs_quantum, quota_quantum;
 985
 986         spin_lock(&sdp->sd_tune.gt_spin);
 987         logd_secs = sdp->sd_tune.gt_logd_secs;
 988         quota_quantum = sdp->sd_tune.gt_quota_quantum;
 989         statfs_quantum = sdp->sd_tune.gt_statfs_quantum;
 990         statfs_slow = sdp->sd_tune.gt_statfs_slow;
 991         spin_unlock(&sdp->sd_tune.gt_spin);
 992
 993         if (is_ancestor(root, sdp->sd_master_dir))
 994                 seq_puts(s, ",meta");
 995         if (args->ar_lockproto[0])
 996                 seq_show_option(s, "lockproto", args->ar_lockproto);
 997         if (args->ar_locktable[0])
 998                 seq_show_option(s, "locktable", args->ar_locktable);
 999         if (args->ar_hostdata[0])
1000                 seq_show_option(s, "hostdata", args->ar_hostdata);
1001         if (args->ar_spectator)
1002                 seq_puts(s, ",spectator");
1003         if (args->ar_localflocks)
1004                 seq_puts(s, ",localflocks");
1005         if (args->ar_debug)
1006                 seq_puts(s, ",debug");
1007         if (args->ar_posix_acl)
1008                 seq_puts(s, ",acl");
1009         if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
1010                 char *state;
1011                 switch (args->ar_quota) {
1012                 case GFS2_QUOTA_OFF:
1013                         state = "off";
1014                         break;
1015                 case GFS2_QUOTA_ACCOUNT:
1016                         state = "account";
1017                         break;
1018                 case GFS2_QUOTA_ON:
1019                         state = "on";
1020                         break;
1021                 default:
1022                         state = "unknown";
1023                         break;
1024                 }
1025                 seq_printf(s, ",quota=%s", state);
1026         }
1027         if (args->ar_suiddir)
1028                 seq_puts(s, ",suiddir");
1029         if (args->ar_data != GFS2_DATA_DEFAULT) {
1030                 char *state;
1031                 switch (args->ar_data) {
1032                 case GFS2_DATA_WRITEBACK:
1033                         state = "writeback";
1034                         break;
1035                 case GFS2_DATA_ORDERED:
1036                         state = "ordered";
1037                         break;
1038                 default:
1039                         state = "unknown";
1040                         break;
1041                 }
1042                 seq_printf(s, ",data=%s", state);
1043         }
1044         if (args->ar_discard)
1045                 seq_puts(s, ",discard");
1046         if (logd_secs != 30)
1047                 seq_printf(s, ",commit=%d", logd_secs);
1048         if (statfs_quantum != 30)
1049                 seq_printf(s, ",statfs_quantum=%d", statfs_quantum);
1050         else if (statfs_slow)
1051                 seq_puts(s, ",statfs_quantum=0");
1052         if (quota_quantum != 60)
1053                 seq_printf(s, ",quota_quantum=%d", quota_quantum);
1054         if (args->ar_statfs_percent)
1055                 seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent);
1056         if (args->ar_errors != GFS2_ERRORS_DEFAULT) {
1057                 const char *state;
1058
1059                 switch (args->ar_errors) {
1060                 case GFS2_ERRORS_WITHDRAW:
1061                         state = "withdraw";
1062                         break;
1063                 case GFS2_ERRORS_PANIC:
1064                         state = "panic";
1065                         break;
1066                 default:
1067                         state = "unknown";
1068                         break;
1069                 }
1070                 seq_printf(s, ",errors=%s", state);
1071         }
1072         if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags))
1073                 seq_puts(s, ",nobarrier");
1074         if (test_bit(SDF_DEMOTE, &sdp->sd_flags))
1075                 seq_puts(s, ",demote_interface_used");
1076         if (args->ar_rgrplvb)
1077                 seq_puts(s, ",rgrplvb");
1078         if (args->ar_loccookie)
1079                 seq_puts(s, ",loccookie");
1080         return 0;
1081 }
1082
1083 static void gfs2_final_release_pages(struct gfs2_inode *ip)
1084 {
1085         struct inode *inode = &ip->i_inode;
1086         struct gfs2_glock *gl = ip->i_gl;
1087
1088         truncate_inode_pages(gfs2_glock2aspace(ip->i_gl), 0);
1089         truncate_inode_pages(&inode->i_data, 0);
1090
1091         if (atomic_read(&gl->gl_revokes) == 0) {
1092                 clear_bit(GLF_LFLUSH, &gl->gl_flags);
1093                 clear_bit(GLF_DIRTY, &gl->gl_flags);
1094         }
1095 }
1096
1097 static int gfs2_dinode_dealloc(struct gfs2_inode *ip)
1098 {
1099         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1100         struct gfs2_rgrpd *rgd;
1101         struct gfs2_holder gh;
1102         int error;
1103
1104         if (gfs2_get_inode_blocks(&ip->i_inode) != 1) {
1105                 gfs2_consist_inode(ip);
1106                 return -EIO;
1107         }
1108
1109         error = gfs2_rindex_update(sdp);
1110         if (error)
1111                 return error;
1112
1113         error = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
1114         if (error)
1115                 return error;
1116
1117         rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1118         if (!rgd) {
1119                 gfs2_consist_inode(ip);
1120                 error = -EIO;
1121                 goto out_qs;
1122         }
1123
1124         error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE,
1125                                    LM_FLAG_NODE_SCOPE, &gh);
1126         if (error)
1127                 goto out_qs;
1128
1129         error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA,
1130                                  sdp->sd_jdesc->jd_blocks);
1131         if (error)
1132                 goto out_rg_gunlock;
1133
1134         gfs2_free_di(rgd, ip);
1135
1136         gfs2_final_release_pages(ip);
1137
1138         gfs2_trans_end(sdp);
1139
1140 out_rg_gunlock:
1141         gfs2_glock_dq_uninit(&gh);
1142 out_qs:
1143         gfs2_quota_unhold(ip);
1144         return error;
1145 }
1146
1147 /**
1148  * gfs2_glock_put_eventually
1149  * @gl: The glock to put
1150  *
1151  * When under memory pressure, trigger a deferred glock put to make sure we
1152  * won't call into DLM and deadlock.  Otherwise, put the glock directly.
1153  */
1154
1155 static void gfs2_glock_put_eventually(struct gfs2_glock *gl)
1156 {
1157         if (current->flags & PF_MEMALLOC)
1158                 gfs2_glock_queue_put(gl);
1159         else
1160                 gfs2_glock_put(gl);
1161 }
1162
1163 static bool gfs2_upgrade_iopen_glock(struct inode *inode)
1164 {
1165         struct gfs2_inode *ip = GFS2_I(inode);
1166         struct gfs2_sbd *sdp = GFS2_SB(inode);
1167         struct gfs2_holder *gh = &ip->i_iopen_gh;
1168         long timeout = 5 * HZ;
1169         int error;
1170
1171         gh->gh_flags |= GL_NOCACHE;
1172         gfs2_glock_dq_wait(gh);
1173
1174         /*
1175          * If there are no other lock holders, we'll get the lock immediately.
1176          * Otherwise, the other nodes holding the lock will be notified about
1177          * our locking request.  If they don't have the inode open, they'll
1178          * evict the cached inode and release the lock.  Otherwise, if they
1179          * poke the inode glock, we'll take this as an indication that they
1180          * still need the iopen glock and that they'll take care of deleting
1181          * the inode when they're done.  As a last resort, if another node
1182          * keeps holding the iopen glock without showing any activity on the
1183          * inode glock, we'll eventually time out.
1184          *
1185          * Note that we're passing the LM_FLAG_TRY_1CB flag to the first
1186          * locking request as an optimization to notify lock holders as soon as
1187          * possible.  Without that flag, they'd be notified implicitly by the
1188          * second locking request.
1189          */
1190
1191         gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE, gh);
1192         error = gfs2_glock_nq(gh);
1193         if (error != GLR_TRYFAILED)
1194                 return !error;
1195
1196         gfs2_holder_reinit(LM_ST_EXCLUSIVE, GL_ASYNC | GL_NOCACHE, gh);
1197         error = gfs2_glock_nq(gh);
1198         if (error)
1199                 return false;
1200
1201         timeout = wait_event_interruptible_timeout(sdp->sd_async_glock_wait,
1202                 !test_bit(HIF_WAIT, &gh->gh_iflags) ||
1203                 test_bit(GLF_DEMOTE, &ip->i_gl->gl_flags),
1204                 timeout);
1205         if (!test_bit(HIF_HOLDER, &gh->gh_iflags)) {
1206                 gfs2_glock_dq(gh);
1207                 return false;
1208         }
1209         return gfs2_glock_holder_ready(gh) == 0;
1210 }
1211
1212 /**
1213  * evict_should_delete - determine whether the inode is eligible for deletion
1214  * @inode: The inode to evict
1215  * @gh: The glock holder structure
1216  *
1217  * This function determines whether the evicted inode is eligible to be deleted
1218  * and locks the inode glock.
1219  *
1220  * Returns: the fate of the dinode
1221  */
1222 static enum dinode_demise evict_should_delete(struct inode *inode,
1223                                               struct gfs2_holder *gh)
1224 {
1225         struct gfs2_inode *ip = GFS2_I(inode);
1226         struct super_block *sb = inode->i_sb;
1227         struct gfs2_sbd *sdp = sb->s_fs_info;
1228         int ret;
1229
1230         if (test_bit(GIF_ALLOC_FAILED, &ip->i_flags)) {
1231                 BUG_ON(!gfs2_glock_is_locked_by_me(ip->i_gl));
1232                 goto should_delete;
1233         }
1234
1235         if (test_bit(GIF_DEFERRED_DELETE, &ip->i_flags))
1236                 return SHOULD_DEFER_EVICTION;
1237
1238         /* Deletes should never happen under memory pressure anymore.  */
1239         if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
1240                 return SHOULD_DEFER_EVICTION;
1241
1242         /* Must not read inode block until block type has been verified */
1243         ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, gh);
1244         if (unlikely(ret)) {
1245                 glock_clear_object(ip->i_iopen_gh.gh_gl, ip);
1246                 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1247                 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1248                 return SHOULD_DEFER_EVICTION;
1249         }
1250
1251         if (gfs2_inode_already_deleted(ip->i_gl, ip->i_no_formal_ino))
1252                 return SHOULD_NOT_DELETE_DINODE;
1253         ret = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED);
1254         if (ret)
1255                 return SHOULD_NOT_DELETE_DINODE;
1256
1257         ret = gfs2_instantiate(gh);
1258         if (ret)
1259                 return SHOULD_NOT_DELETE_DINODE;
1260
1261         /*
1262          * The inode may have been recreated in the meantime.
1263          */
1264         if (inode->i_nlink)
1265                 return SHOULD_NOT_DELETE_DINODE;
1266
1267 should_delete:
1268         if (gfs2_holder_initialized(&ip->i_iopen_gh) &&
1269             test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1270                 if (!gfs2_upgrade_iopen_glock(inode)) {
1271                         gfs2_holder_uninit(&ip->i_iopen_gh);
1272                         return SHOULD_NOT_DELETE_DINODE;
1273                 }
1274         }
1275         return SHOULD_DELETE_DINODE;
1276 }
1277
1278 /**
1279  * evict_unlinked_inode - delete the pieces of an unlinked evicted inode
1280  * @inode: The inode to evict
1281  */
1282 static int evict_unlinked_inode(struct inode *inode)
1283 {
1284         struct gfs2_inode *ip = GFS2_I(inode);
1285         int ret;
1286
1287         if (S_ISDIR(inode->i_mode) &&
1288             (ip->i_diskflags & GFS2_DIF_EXHASH)) {
1289                 ret = gfs2_dir_exhash_dealloc(ip);
1290                 if (ret)
1291                         goto out;
1292         }
1293
1294         if (ip->i_eattr) {
1295                 ret = gfs2_ea_dealloc(ip);
1296                 if (ret)
1297                         goto out;
1298         }
1299
1300         if (!gfs2_is_stuffed(ip)) {
1301                 ret = gfs2_file_dealloc(ip);
1302                 if (ret)
1303                         goto out;
1304         }
1305
1306         /* We're about to clear the bitmap for the dinode, but as soon as we
1307            do, gfs2_create_inode can create another inode at the same block
1308            location and try to set gl_object again. We clear gl_object here so
1309            that subsequent inode creates don't see an old gl_object. */
1310         glock_clear_object(ip->i_gl, ip);
1311         ret = gfs2_dinode_dealloc(ip);
1312         gfs2_inode_remember_delete(ip->i_gl, ip->i_no_formal_ino);
1313 out:
1314         return ret;
1315 }
1316
1317 /*
1318  * evict_linked_inode - evict an inode whose dinode has not been unlinked
1319  * @inode: The inode to evict
1320  */
1321 static int evict_linked_inode(struct inode *inode)
1322 {
1323         struct super_block *sb = inode->i_sb;
1324         struct gfs2_sbd *sdp = sb->s_fs_info;
1325         struct gfs2_inode *ip = GFS2_I(inode);
1326         struct address_space *metamapping;
1327         int ret;
1328
1329         gfs2_log_flush(sdp, ip->i_gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
1330                        GFS2_LFC_EVICT_INODE);
1331         metamapping = gfs2_glock2aspace(ip->i_gl);
1332         if (test_bit(GLF_DIRTY, &ip->i_gl->gl_flags)) {
1333                 filemap_fdatawrite(metamapping);
1334                 filemap_fdatawait(metamapping);
1335         }
1336         write_inode_now(inode, 1);
1337         gfs2_ail_flush(ip->i_gl, 0);
1338
1339         ret = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
1340         if (ret)
1341                 return ret;
1342
1343         /* Needs to be done before glock release & also in a transaction */
1344         truncate_inode_pages(&inode->i_data, 0);
1345         truncate_inode_pages(metamapping, 0);
1346         gfs2_trans_end(sdp);
1347         return 0;
1348 }
1349
1350 /**
1351  * gfs2_evict_inode - Remove an inode from cache
1352  * @inode: The inode to evict
1353  *
1354  * There are three cases to consider:
1355  * 1. i_nlink == 0, we are final opener (and must deallocate)
1356  * 2. i_nlink == 0, we are not the final opener (and cannot deallocate)
1357  * 3. i_nlink > 0
1358  *
1359  * If the fs is read only, then we have to treat all cases as per #3
1360  * since we are unable to do any deallocation. The inode will be
1361  * deallocated by the next read/write node to attempt an allocation
1362  * in the same resource group
1363  *
1364  * We have to (at the moment) hold the inodes main lock to cover
1365  * the gap between unlocking the shared lock on the iopen lock and
1366  * taking the exclusive lock. I'd rather do a shared -> exclusive
1367  * conversion on the iopen lock, but we can change that later. This
1368  * is safe, just less efficient.
1369  */
1370
1371 static void gfs2_evict_inode(struct inode *inode)
1372 {
1373         struct super_block *sb = inode->i_sb;
1374         struct gfs2_sbd *sdp = sb->s_fs_info;
1375         struct gfs2_inode *ip = GFS2_I(inode);
1376         struct gfs2_holder gh;
1377         int ret;
1378
1379         if (test_bit(GIF_FREE_VFS_INODE, &ip->i_flags)) {
1380                 clear_inode(inode);
1381                 return;
1382         }
1383
1384         if (inode->i_nlink || sb_rdonly(sb))
1385                 goto out;
1386
1387         /*
1388          * In case of an incomplete mount, gfs2_evict_inode() may be called for
1389          * system files without having an active journal to write to.  In that
1390          * case, skip the filesystem evict.
1391          */
1392         if (!sdp->sd_jdesc)
1393                 goto out;
1394
1395         gfs2_holder_mark_uninitialized(&gh);
1396         ret = evict_should_delete(inode, &gh);
1397         if (ret == SHOULD_DEFER_EVICTION)
1398                 goto out;
1399         if (ret == SHOULD_DELETE_DINODE)
1400                 ret = evict_unlinked_inode(inode);
1401         else
1402                 ret = evict_linked_inode(inode);
1403
1404         if (gfs2_rs_active(&ip->i_res))
1405                 gfs2_rs_deltree(&ip->i_res);
1406
1407         if (gfs2_holder_initialized(&gh)) {
1408                 glock_clear_object(ip->i_gl, ip);
1409                 gfs2_glock_dq_uninit(&gh);
1410         }
1411         if (ret && ret != GLR_TRYFAILED && ret != -EROFS)
1412                 fs_warn(sdp, "gfs2_evict_inode: %d\n", ret);
1413 out:
1414         truncate_inode_pages_final(&inode->i_data);
1415         if (ip->i_qadata)
1416                 gfs2_assert_warn(sdp, ip->i_qadata->qa_ref == 0);
1417         gfs2_rs_deltree(&ip->i_res);
1418         gfs2_ordered_del_inode(ip);
1419         clear_inode(inode);
1420         gfs2_dir_hash_inval(ip);
1421         if (gfs2_holder_initialized(&ip->i_iopen_gh)) {
1422                 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1423
1424                 glock_clear_object(gl, ip);
1425                 if (test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1426                         ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1427                         gfs2_glock_dq(&ip->i_iopen_gh);
1428                 }
1429                 gfs2_glock_hold(gl);
1430                 gfs2_holder_uninit(&ip->i_iopen_gh);
1431                 gfs2_glock_put_eventually(gl);
1432         }
1433         if (ip->i_gl) {
1434                 glock_clear_object(ip->i_gl, ip);
1435                 wait_on_bit_io(&ip->i_flags, GIF_GLOP_PENDING, TASK_UNINTERRUPTIBLE);
1436                 gfs2_glock_add_to_lru(ip->i_gl);
1437                 gfs2_glock_put_eventually(ip->i_gl);
1438                 rcu_assign_pointer(ip->i_gl, NULL);
1439         }
1440 }
1441
1442 static struct inode *gfs2_alloc_inode(struct super_block *sb)
1443 {
1444         struct gfs2_inode *ip;
1445
1446         ip = alloc_inode_sb(sb, gfs2_inode_cachep, GFP_KERNEL);
1447         if (!ip)
1448                 return NULL;
1449         ip->i_flags = 0;
1450         ip->i_gl = NULL;
1451         gfs2_holder_mark_uninitialized(&ip->i_iopen_gh);
1452         memset(&ip->i_res, 0, sizeof(ip->i_res));
1453         RB_CLEAR_NODE(&ip->i_res.rs_node);
1454         ip->i_rahead = 0;
1455         return &ip->i_inode;
1456 }
1457
1458 static void gfs2_free_inode(struct inode *inode)
1459 {
1460         kmem_cache_free(gfs2_inode_cachep, GFS2_I(inode));
1461 }
1462
1463 extern void free_local_statfs_inodes(struct gfs2_sbd *sdp)
1464 {
1465         struct local_statfs_inode *lsi, *safe;
1466
1467         /* Run through the statfs inodes list to iput and free memory */
1468         list_for_each_entry_safe(lsi, safe, &sdp->sd_sc_inodes_list, si_list) {
1469                 if (lsi->si_jid == sdp->sd_jdesc->jd_jid)
1470                         sdp->sd_sc_inode = NULL; /* belongs to this node */
1471                 if (lsi->si_sc_inode)
1472                         iput(lsi->si_sc_inode);
1473                 list_del(&lsi->si_list);
1474                 kfree(lsi);
1475         }
1476 }
1477
1478 extern struct inode *find_local_statfs_inode(struct gfs2_sbd *sdp,
1479                                              unsigned int index)
1480 {
1481         struct local_statfs_inode *lsi;
1482
1483         /* Return the local (per node) statfs inode in the
1484          * sdp->sd_sc_inodes_list corresponding to the 'index'. */
1485         list_for_each_entry(lsi, &sdp->sd_sc_inodes_list, si_list) {
1486                 if (lsi->si_jid == index)
1487                         return lsi->si_sc_inode;
1488         }
1489         return NULL;
1490 }
1491
1492 const struct super_operations gfs2_super_ops = {
1493         .alloc_inode            = gfs2_alloc_inode,
1494         .free_inode             = gfs2_free_inode,
1495         .write_inode            = gfs2_write_inode,
1496         .dirty_inode            = gfs2_dirty_inode,
1497         .evict_inode            = gfs2_evict_inode,
1498         .put_super              = gfs2_put_super,
1499         .sync_fs                = gfs2_sync_fs,
1500         .freeze_super           = gfs2_freeze,
1501         .thaw_super             = gfs2_unfreeze,
1502         .statfs                 = gfs2_statfs,
1503         .drop_inode             = gfs2_drop_inode,
1504         .show_options           = gfs2_show_options,
1505 };
1506