fs/gfs2/super.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   4  * Copyright (C) 2004-2007 Red Hat, Inc.  All rights reserved.
   5  */
   6
   7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
   8
   9 #include <linux/bio.h>
  10 #include <linux/sched/signal.h>
  11 #include <linux/slab.h>
  12 #include <linux/spinlock.h>
  13 #include <linux/completion.h>
  14 #include <linux/buffer_head.h>
  15 #include <linux/statfs.h>
  16 #include <linux/seq_file.h>
  17 #include <linux/mount.h>
  18 #include <linux/kthread.h>
  19 #include <linux/delay.h>
  20 #include <linux/gfs2_ondisk.h>
  21 #include <linux/crc32.h>
  22 #include <linux/time.h>
  23 #include <linux/wait.h>
  24 #include <linux/writeback.h>
  25 #include <linux/backing-dev.h>
  26 #include <linux/kernel.h>
  27
  28 #include "gfs2.h"
  29 #include "incore.h"
  30 #include "bmap.h"
  31 #include "dir.h"
  32 #include "glock.h"
  33 #include "glops.h"
  34 #include "inode.h"
  35 #include "log.h"
  36 #include "meta_io.h"
  37 #include "quota.h"
  38 #include "recovery.h"
  39 #include "rgrp.h"
  40 #include "super.h"
  41 #include "trans.h"
  42 #include "util.h"
  43 #include "sys.h"
  44 #include "xattr.h"
  45 #include "lops.h"
  46
  47 enum dinode_demise {
  48         SHOULD_DELETE_DINODE,
  49         SHOULD_NOT_DELETE_DINODE,
  50         SHOULD_DEFER_EVICTION,
  51 };
  52
  53 /**
  54  * gfs2_jindex_free - Clear all the journal index information
  55  * @sdp: The GFS2 superblock
  56  *
  57  */
  58
  59 void gfs2_jindex_free(struct gfs2_sbd *sdp)
  60 {
  61         struct list_head list;
  62         struct gfs2_jdesc *jd;
  63
  64         spin_lock(&sdp->sd_jindex_spin);
  65         list_add(&list, &sdp->sd_jindex_list);
  66         list_del_init(&sdp->sd_jindex_list);
  67         sdp->sd_journals = 0;
  68         spin_unlock(&sdp->sd_jindex_spin);
  69
  70         sdp->sd_jdesc = NULL;
  71         while (!list_empty(&list)) {
  72                 jd = list_first_entry(&list, struct gfs2_jdesc, jd_list);
  73                 gfs2_free_journal_extents(jd);
  74                 list_del(&jd->jd_list);
  75                 iput(jd->jd_inode);
  76                 jd->jd_inode = NULL;
  77                 kfree(jd);
  78         }
  79 }
  80
  81 static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
  82 {
  83         struct gfs2_jdesc *jd;
  84
  85         list_for_each_entry(jd, head, jd_list) {
  86                 if (jd->jd_jid == jid)
  87                         return jd;
  88         }
  89         return NULL;
  90 }
  91
  92 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
  93 {
  94         struct gfs2_jdesc *jd;
  95
  96         spin_lock(&sdp->sd_jindex_spin);
  97         jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
  98         spin_unlock(&sdp->sd_jindex_spin);
  99
 100         return jd;
 101 }
 102
 103 int gfs2_jdesc_check(struct gfs2_jdesc *jd)
 104 {
 105         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 106         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 107         u64 size = i_size_read(jd->jd_inode);
 108
 109         if (gfs2_check_internal_file_size(jd->jd_inode, 8 << 20, BIT(30)))
 110                 return -EIO;
 111
 112         jd->jd_blocks = size >> sdp->sd_sb.sb_bsize_shift;
 113
 114         if (gfs2_write_alloc_required(ip, 0, size)) {
 115                 gfs2_consist_inode(ip);
 116                 return -EIO;
 117         }
 118
 119         return 0;
 120 }
 121
 122 /**
 123  * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
 124  * @sdp: the filesystem
 125  *
 126  * Returns: errno
 127  */
 128
 129 int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
 130 {
 131         struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
 132         struct gfs2_glock *j_gl = ip->i_gl;
 133         struct gfs2_log_header_host head;
 134         int error;
 135
 136         j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
 137         if (gfs2_withdrawn(sdp))
 138                 return -EIO;
 139
 140         error = gfs2_find_jhead(sdp->sd_jdesc, &head, false);
 141         if (error) {
 142                 gfs2_consist(sdp);
 143                 return error;
 144         }
 145
 146         if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 147                 gfs2_consist(sdp);
 148                 return -EIO;
 149         }
 150
 151         /*  Initialize some head of the log stuff  */
 152         sdp->sd_log_sequence = head.lh_sequence + 1;
 153         gfs2_log_pointers_init(sdp, head.lh_blkno);
 154
 155         error = gfs2_quota_init(sdp);
 156         if (!error && gfs2_withdrawn(sdp))
 157                 error = -EIO;
 158         if (!error)
 159                 set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 160         return error;
 161 }
 162
 163 void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
 164 {
 165         const struct gfs2_statfs_change *str = buf;
 166
 167         sc->sc_total = be64_to_cpu(str->sc_total);
 168         sc->sc_free = be64_to_cpu(str->sc_free);
 169         sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
 170 }
 171
 172 void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
 173 {
 174         struct gfs2_statfs_change *str = buf;
 175
 176         str->sc_total = cpu_to_be64(sc->sc_total);
 177         str->sc_free = cpu_to_be64(sc->sc_free);
 178         str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
 179 }
 180
 181 int gfs2_statfs_init(struct gfs2_sbd *sdp)
 182 {
 183         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 184         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 185         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 186         struct buffer_head *m_bh;
 187         struct gfs2_holder gh;
 188         int error;
 189
 190         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 191                                    &gh);
 192         if (error)
 193                 return error;
 194
 195         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 196         if (error)
 197                 goto out;
 198
 199         if (sdp->sd_args.ar_spectator) {
 200                 spin_lock(&sdp->sd_statfs_spin);
 201                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 202                                       sizeof(struct gfs2_dinode));
 203                 spin_unlock(&sdp->sd_statfs_spin);
 204         } else {
 205                 spin_lock(&sdp->sd_statfs_spin);
 206                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 207                                       sizeof(struct gfs2_dinode));
 208                 gfs2_statfs_change_in(l_sc, sdp->sd_sc_bh->b_data +
 209                                       sizeof(struct gfs2_dinode));
 210                 spin_unlock(&sdp->sd_statfs_spin);
 211
 212         }
 213
 214         brelse(m_bh);
 215 out:
 216         gfs2_glock_dq_uninit(&gh);
 217         return 0;
 218 }
 219
 220 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
 221                         s64 dinodes)
 222 {
 223         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 224         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 225         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 226         s64 x, y;
 227         int need_sync = 0;
 228
 229         gfs2_trans_add_meta(l_ip->i_gl, sdp->sd_sc_bh);
 230
 231         spin_lock(&sdp->sd_statfs_spin);
 232         l_sc->sc_total += total;
 233         l_sc->sc_free += free;
 234         l_sc->sc_dinodes += dinodes;
 235         gfs2_statfs_change_out(l_sc, sdp->sd_sc_bh->b_data +
 236                                sizeof(struct gfs2_dinode));
 237         if (sdp->sd_args.ar_statfs_percent) {
 238                 x = 100 * l_sc->sc_free;
 239                 y = m_sc->sc_free * sdp->sd_args.ar_statfs_percent;
 240                 if (x >= y || x <= -y)
 241                         need_sync = 1;
 242         }
 243         spin_unlock(&sdp->sd_statfs_spin);
 244
 245         if (need_sync)
 246                 gfs2_wake_up_statfs(sdp);
 247 }
 248
 249 void update_statfs(struct gfs2_sbd *sdp, struct buffer_head *m_bh)
 250 {
 251         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 252         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 253         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 254         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 255
 256         gfs2_trans_add_meta(l_ip->i_gl, sdp->sd_sc_bh);
 257         gfs2_trans_add_meta(m_ip->i_gl, m_bh);
 258
 259         spin_lock(&sdp->sd_statfs_spin);
 260         m_sc->sc_total += l_sc->sc_total;
 261         m_sc->sc_free += l_sc->sc_free;
 262         m_sc->sc_dinodes += l_sc->sc_dinodes;
 263         memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
 264         memset(sdp->sd_sc_bh->b_data + sizeof(struct gfs2_dinode),
 265                0, sizeof(struct gfs2_statfs_change));
 266         gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
 267         spin_unlock(&sdp->sd_statfs_spin);
 268 }
 269
 270 int gfs2_statfs_sync(struct super_block *sb, int type)
 271 {
 272         struct gfs2_sbd *sdp = sb->s_fs_info;
 273         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 274         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 275         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 276         struct gfs2_holder gh;
 277         struct buffer_head *m_bh;
 278         int error;
 279
 280         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 281                                    &gh);
 282         if (error)
 283                 goto out;
 284
 285         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 286         if (error)
 287                 goto out_unlock;
 288
 289         spin_lock(&sdp->sd_statfs_spin);
 290         gfs2_statfs_change_in(m_sc, m_bh->b_data +
 291                               sizeof(struct gfs2_dinode));
 292         if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
 293                 spin_unlock(&sdp->sd_statfs_spin);
 294                 goto out_bh;
 295         }
 296         spin_unlock(&sdp->sd_statfs_spin);
 297
 298         error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
 299         if (error)
 300                 goto out_bh;
 301
 302         update_statfs(sdp, m_bh);
 303         sdp->sd_statfs_force_sync = 0;
 304
 305         gfs2_trans_end(sdp);
 306
 307 out_bh:
 308         brelse(m_bh);
 309 out_unlock:
 310         gfs2_glock_dq_uninit(&gh);
 311 out:
 312         return error;
 313 }
 314
 315 struct lfcc {
 316         struct list_head list;
 317         struct gfs2_holder gh;
 318 };
 319
 320 /**
 321  * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
 322  *                            journals are clean
 323  * @sdp: the file system
 324  *
 325  * Returns: errno
 326  */
 327
 328 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp)
 329 {
 330         struct gfs2_inode *ip;
 331         struct gfs2_jdesc *jd;
 332         struct lfcc *lfcc;
 333         LIST_HEAD(list);
 334         struct gfs2_log_header_host lh;
 335         int error, error2;
 336
 337         /*
 338          * Grab all the journal glocks in SH mode.  We are *probably* doing
 339          * that to prevent recovery.
 340          */
 341
 342         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 343                 lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
 344                 if (!lfcc) {
 345                         error = -ENOMEM;
 346                         goto out;
 347                 }
 348                 ip = GFS2_I(jd->jd_inode);
 349                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
 350                 if (error) {
 351                         kfree(lfcc);
 352                         goto out;
 353                 }
 354                 list_add(&lfcc->list, &list);
 355         }
 356
 357         gfs2_freeze_unlock(&sdp->sd_freeze_gh);
 358
 359         error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_EXCLUSIVE,
 360                                    LM_FLAG_NOEXP | GL_NOPID,
 361                                    &sdp->sd_freeze_gh);
 362         if (error)
 363                 goto relock_shared;
 364
 365         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 366                 error = gfs2_jdesc_check(jd);
 367                 if (error)
 368                         break;
 369                 error = gfs2_find_jhead(jd, &lh, false);
 370                 if (error)
 371                         break;
 372                 if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 373                         error = -EBUSY;
 374                         break;
 375                 }
 376         }
 377
 378         if (!error)
 379                 goto out;  /* success */
 380
 381         gfs2_freeze_unlock(&sdp->sd_freeze_gh);
 382
 383 relock_shared:
 384         error2 = gfs2_freeze_lock_shared(sdp);
 385         gfs2_assert_withdraw(sdp, !error2);
 386
 387 out:
 388         while (!list_empty(&list)) {
 389                 lfcc = list_first_entry(&list, struct lfcc, list);
 390                 list_del(&lfcc->list);
 391                 gfs2_glock_dq_uninit(&lfcc->gh);
 392                 kfree(lfcc);
 393         }
 394         return error;
 395 }
 396
 397 void gfs2_dinode_out(const struct gfs2_inode *ip, void *buf)
 398 {
 399         const struct inode *inode = &ip->i_inode;
 400         struct gfs2_dinode *str = buf;
 401
 402         str->di_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 403         str->di_header.mh_type = cpu_to_be32(GFS2_METATYPE_DI);
 404         str->di_header.mh_format = cpu_to_be32(GFS2_FORMAT_DI);
 405         str->di_num.no_addr = cpu_to_be64(ip->i_no_addr);
 406         str->di_num.no_formal_ino = cpu_to_be64(ip->i_no_formal_ino);
 407         str->di_mode = cpu_to_be32(inode->i_mode);
 408         str->di_uid = cpu_to_be32(i_uid_read(inode));
 409         str->di_gid = cpu_to_be32(i_gid_read(inode));
 410         str->di_nlink = cpu_to_be32(inode->i_nlink);
 411         str->di_size = cpu_to_be64(i_size_read(inode));
 412         str->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(inode));
 413         str->di_atime = cpu_to_be64(inode->i_atime.tv_sec);
 414         str->di_mtime = cpu_to_be64(inode->i_mtime.tv_sec);
 415         str->di_ctime = cpu_to_be64(inode->i_ctime.tv_sec);
 416
 417         str->di_goal_meta = cpu_to_be64(ip->i_goal);
 418         str->di_goal_data = cpu_to_be64(ip->i_goal);
 419         str->di_generation = cpu_to_be64(ip->i_generation);
 420
 421         str->di_flags = cpu_to_be32(ip->i_diskflags);
 422         str->di_height = cpu_to_be16(ip->i_height);
 423         str->di_payload_format = cpu_to_be32(S_ISDIR(inode->i_mode) &&
 424                                              !(ip->i_diskflags & GFS2_DIF_EXHASH) ?
 425                                              GFS2_FORMAT_DE : 0);
 426         str->di_depth = cpu_to_be16(ip->i_depth);
 427         str->di_entries = cpu_to_be32(ip->i_entries);
 428
 429         str->di_eattr = cpu_to_be64(ip->i_eattr);
 430         str->di_atime_nsec = cpu_to_be32(inode->i_atime.tv_nsec);
 431         str->di_mtime_nsec = cpu_to_be32(inode->i_mtime.tv_nsec);
 432         str->di_ctime_nsec = cpu_to_be32(inode->i_ctime.tv_nsec);
 433 }
 434
 435 /**
 436  * gfs2_write_inode - Make sure the inode is stable on the disk
 437  * @inode: The inode
 438  * @wbc: The writeback control structure
 439  *
 440  * Returns: errno
 441  */
 442
 443 static int gfs2_write_inode(struct inode *inode, struct writeback_control *wbc)
 444 {
 445         struct gfs2_inode *ip = GFS2_I(inode);
 446         struct gfs2_sbd *sdp = GFS2_SB(inode);
 447         struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl);
 448         struct backing_dev_info *bdi = inode_to_bdi(metamapping->host);
 449         int ret = 0;
 450         bool flush_all = (wbc->sync_mode == WB_SYNC_ALL || gfs2_is_jdata(ip));
 451
 452         if (flush_all)
 453                 gfs2_log_flush(GFS2_SB(inode), ip->i_gl,
 454                                GFS2_LOG_HEAD_FLUSH_NORMAL |
 455                                GFS2_LFC_WRITE_INODE);
 456         if (bdi->wb.dirty_exceeded)
 457                 gfs2_ail1_flush(sdp, wbc);
 458         else
 459                 filemap_fdatawrite(metamapping);
 460         if (flush_all)
 461                 ret = filemap_fdatawait(metamapping);
 462         if (ret)
 463                 mark_inode_dirty_sync(inode);
 464         else {
 465                 spin_lock(&inode->i_lock);
 466                 if (!(inode->i_flags & I_DIRTY))
 467                         gfs2_ordered_del_inode(ip);
 468                 spin_unlock(&inode->i_lock);
 469         }
 470         return ret;
 471 }
 472
 473 /**
 474  * gfs2_dirty_inode - check for atime updates
 475  * @inode: The inode in question
 476  * @flags: The type of dirty
 477  *
 478  * Unfortunately it can be called under any combination of inode
 479  * glock and freeze glock, so we have to check carefully.
 480  *
 481  * At the moment this deals only with atime - it should be possible
 482  * to expand that role in future, once a review of the locking has
 483  * been carried out.
 484  */
 485
 486 static void gfs2_dirty_inode(struct inode *inode, int flags)
 487 {
 488         struct gfs2_inode *ip = GFS2_I(inode);
 489         struct gfs2_sbd *sdp = GFS2_SB(inode);
 490         struct buffer_head *bh;
 491         struct gfs2_holder gh;
 492         int need_unlock = 0;
 493         int need_endtrans = 0;
 494         int ret;
 495
 496         if (unlikely(!ip->i_gl)) {
 497                 /* This can only happen during incomplete inode creation. */
 498                 BUG_ON(!test_bit(GIF_ALLOC_FAILED, &ip->i_flags));
 499                 return;
 500         }
 501
 502         if (unlikely(gfs2_withdrawn(sdp)))
 503                 return;
 504         if (!gfs2_glock_is_locked_by_me(ip->i_gl)) {
 505                 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
 506                 if (ret) {
 507                         fs_err(sdp, "dirty_inode: glock %d\n", ret);
 508                         gfs2_dump_glock(NULL, ip->i_gl, true);
 509                         return;
 510                 }
 511                 need_unlock = 1;
 512         } else if (WARN_ON_ONCE(ip->i_gl->gl_state != LM_ST_EXCLUSIVE))
 513                 return;
 514
 515         if (current->journal_info == NULL) {
 516                 ret = gfs2_trans_begin(sdp, RES_DINODE, 0);
 517                 if (ret) {
 518                         fs_err(sdp, "dirty_inode: gfs2_trans_begin %d\n", ret);
 519                         goto out;
 520                 }
 521                 need_endtrans = 1;
 522         }
 523
 524         ret = gfs2_meta_inode_buffer(ip, &bh);
 525         if (ret == 0) {
 526                 gfs2_trans_add_meta(ip->i_gl, bh);
 527                 gfs2_dinode_out(ip, bh->b_data);
 528                 brelse(bh);
 529         }
 530
 531         if (need_endtrans)
 532                 gfs2_trans_end(sdp);
 533 out:
 534         if (need_unlock)
 535                 gfs2_glock_dq_uninit(&gh);
 536 }
 537
 538 /**
 539  * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
 540  * @sdp: the filesystem
 541  *
 542  * Returns: errno
 543  */
 544
 545 void gfs2_make_fs_ro(struct gfs2_sbd *sdp)
 546 {
 547         int log_write_allowed = test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 548
 549         if (!test_bit(SDF_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         }
 608         if (gfs2_withdrawn(sdp)) {
 609                 gfs2_destroy_threads(sdp);
 610                 gfs2_quota_cleanup(sdp);
 611         }
 612         WARN_ON(gfs2_withdrawing(sdp));
 613
 614         /*  At this point, we're through modifying the disk  */
 615
 616         /*  Release stuff  */
 617
 618         gfs2_freeze_unlock(&sdp->sd_freeze_gh);
 619
 620         iput(sdp->sd_jindex);
 621         iput(sdp->sd_statfs_inode);
 622         iput(sdp->sd_rindex);
 623         iput(sdp->sd_quota_inode);
 624
 625         gfs2_glock_put(sdp->sd_rename_gl);
 626         gfs2_glock_put(sdp->sd_freeze_gl);
 627
 628         if (!sdp->sd_args.ar_spectator) {
 629                 if (gfs2_holder_initialized(&sdp->sd_journal_gh))
 630                         gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
 631                 if (gfs2_holder_initialized(&sdp->sd_jinode_gh))
 632                         gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
 633                 brelse(sdp->sd_sc_bh);
 634                 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
 635                 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
 636                 free_local_statfs_inodes(sdp);
 637                 iput(sdp->sd_qc_inode);
 638         }
 639
 640         gfs2_glock_dq_uninit(&sdp->sd_live_gh);
 641         gfs2_clear_rgrpd(sdp);
 642         gfs2_jindex_free(sdp);
 643         /*  Take apart glock structures and buffer lists  */
 644         gfs2_gl_hash_clear(sdp);
 645         truncate_inode_pages_final(&sdp->sd_aspace);
 646         gfs2_delete_debugfs_file(sdp);
 647         /*  Unmount the locking protocol  */
 648         gfs2_lm_unmount(sdp);
 649
 650         /*  At this point, we're through participating in the lockspace  */
 651         gfs2_sys_fs_del(sdp);
 652         free_sbd(sdp);
 653 }
 654
 655 /**
 656  * gfs2_sync_fs - sync the filesystem
 657  * @sb: the superblock
 658  * @wait: true to wait for completion
 659  *
 660  * Flushes the log to disk.
 661  */
 662
 663 static int gfs2_sync_fs(struct super_block *sb, int wait)
 664 {
 665         struct gfs2_sbd *sdp = sb->s_fs_info;
 666
 667         gfs2_quota_sync(sb, -1);
 668         if (wait)
 669                 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
 670                                GFS2_LFC_SYNC_FS);
 671         return sdp->sd_log_error;
 672 }
 673
 674 static int gfs2_freeze_locally(struct gfs2_sbd *sdp)
 675 {
 676         struct super_block *sb = sdp->sd_vfs;
 677         int error;
 678
 679         error = freeze_super(sb);
 680         if (error)
 681                 return error;
 682
 683         if (test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
 684                 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_FREEZE |
 685                                GFS2_LFC_FREEZE_GO_SYNC);
 686                 if (gfs2_withdrawn(sdp)) {
 687                         thaw_super(sb);
 688                         return -EIO;
 689                 }
 690         }
 691         return 0;
 692 }
 693
 694 static int gfs2_do_thaw(struct gfs2_sbd *sdp)
 695 {
 696         struct super_block *sb = sdp->sd_vfs;
 697         int error;
 698
 699         error = gfs2_freeze_lock_shared(sdp);
 700         if (error)
 701                 goto fail;
 702         error = thaw_super(sb);
 703         if (!error)
 704                 return 0;
 705
 706 fail:
 707         fs_info(sdp, "GFS2: couldn't thaw filesystem: %d\n", error);
 708         gfs2_assert_withdraw(sdp, 0);
 709         return error;
 710 }
 711
 712 void gfs2_freeze_func(struct work_struct *work)
 713 {
 714         struct gfs2_sbd *sdp = container_of(work, struct gfs2_sbd, sd_freeze_work);
 715         struct super_block *sb = sdp->sd_vfs;
 716         int error;
 717
 718         mutex_lock(&sdp->sd_freeze_mutex);
 719         error = -EBUSY;
 720         if (test_bit(SDF_FROZEN, &sdp->sd_flags))
 721                 goto freeze_failed;
 722
 723         error = gfs2_freeze_locally(sdp);
 724         if (error)
 725                 goto freeze_failed;
 726
 727         gfs2_freeze_unlock(&sdp->sd_freeze_gh);
 728         set_bit(SDF_FROZEN, &sdp->sd_flags);
 729
 730         error = gfs2_do_thaw(sdp);
 731         if (error)
 732                 goto out;
 733
 734         clear_bit(SDF_FROZEN, &sdp->sd_flags);
 735         goto out;
 736
 737 freeze_failed:
 738         fs_info(sdp, "GFS2: couldn't freeze filesystem: %d\n", error);
 739
 740 out:
 741         mutex_unlock(&sdp->sd_freeze_mutex);
 742         deactivate_super(sb);
 743 }
 744
 745 /**
 746  * gfs2_freeze_super - prevent further writes to the filesystem
 747  * @sb: the VFS structure for the filesystem
 748  *
 749  */
 750
 751 static int gfs2_freeze_super(struct super_block *sb)
 752 {
 753         struct gfs2_sbd *sdp = sb->s_fs_info;
 754         int error;
 755
 756         if (!mutex_trylock(&sdp->sd_freeze_mutex))
 757                 return -EBUSY;
 758         error = -EBUSY;
 759         if (test_bit(SDF_FROZEN, &sdp->sd_flags))
 760                 goto out;
 761
 762         for (;;) {
 763                 error = gfs2_freeze_locally(sdp);
 764                 if (error) {
 765                         fs_info(sdp, "GFS2: couldn't freeze filesystem: %d\n",
 766                                 error);
 767                         goto out;
 768                 }
 769
 770                 error = gfs2_lock_fs_check_clean(sdp);
 771                 if (!error)
 772                         break;  /* success */
 773
 774                 error = gfs2_do_thaw(sdp);
 775                 if (error)
 776                         goto out;
 777
 778                 if (error == -EBUSY)
 779                         fs_err(sdp, "waiting for recovery before freeze\n");
 780                 else if (error == -EIO) {
 781                         fs_err(sdp, "Fatal IO error: cannot freeze gfs2 due "
 782                                "to recovery error.\n");
 783                         goto out;
 784                 } else {
 785                         fs_err(sdp, "error freezing FS: %d\n", error);
 786                 }
 787                 fs_err(sdp, "retrying...\n");
 788                 msleep(1000);
 789         }
 790
 791 out:
 792         if (!error) {
 793                 set_bit(SDF_FREEZE_INITIATOR, &sdp->sd_flags);
 794                 set_bit(SDF_FROZEN, &sdp->sd_flags);
 795         }
 796         mutex_unlock(&sdp->sd_freeze_mutex);
 797         return error;
 798 }
 799
 800 /**
 801  * gfs2_thaw_super - reallow writes to the filesystem
 802  * @sb: the VFS structure for the filesystem
 803  *
 804  */
 805
 806 static int gfs2_thaw_super(struct super_block *sb)
 807 {
 808         struct gfs2_sbd *sdp = sb->s_fs_info;
 809         int error;
 810
 811         if (!mutex_trylock(&sdp->sd_freeze_mutex))
 812                 return -EBUSY;
 813         error = -EINVAL;
 814         if (!test_bit(SDF_FREEZE_INITIATOR, &sdp->sd_flags))
 815                 goto out;
 816
 817         gfs2_freeze_unlock(&sdp->sd_freeze_gh);
 818
 819         error = gfs2_do_thaw(sdp);
 820
 821         if (!error) {
 822                 clear_bit(SDF_FREEZE_INITIATOR, &sdp->sd_flags);
 823                 clear_bit(SDF_FROZEN, &sdp->sd_flags);
 824         }
 825 out:
 826         mutex_unlock(&sdp->sd_freeze_mutex);
 827         return error;
 828 }
 829
 830 void gfs2_thaw_freeze_initiator(struct super_block *sb)
 831 {
 832         struct gfs2_sbd *sdp = sb->s_fs_info;
 833
 834         mutex_lock(&sdp->sd_freeze_mutex);
 835         if (!test_bit(SDF_FREEZE_INITIATOR, &sdp->sd_flags))
 836                 goto out;
 837
 838         gfs2_freeze_unlock(&sdp->sd_freeze_gh);
 839
 840 out:
 841         mutex_unlock(&sdp->sd_freeze_mutex);
 842 }
 843
 844 /**
 845  * statfs_slow_fill - fill in the sg for a given RG
 846  * @rgd: the RG
 847  * @sc: the sc structure
 848  *
 849  * Returns: 0 on success, -ESTALE if the LVB is invalid
 850  */
 851
 852 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
 853                             struct gfs2_statfs_change_host *sc)
 854 {
 855         gfs2_rgrp_verify(rgd);
 856         sc->sc_total += rgd->rd_data;
 857         sc->sc_free += rgd->rd_free;
 858         sc->sc_dinodes += rgd->rd_dinodes;
 859         return 0;
 860 }
 861
 862 /**
 863  * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
 864  * @sdp: the filesystem
 865  * @sc: the sc info that will be returned
 866  *
 867  * Any error (other than a signal) will cause this routine to fall back
 868  * to the synchronous version.
 869  *
 870  * FIXME: This really shouldn't busy wait like this.
 871  *
 872  * Returns: errno
 873  */
 874
 875 static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
 876 {
 877         struct gfs2_rgrpd *rgd_next;
 878         struct gfs2_holder *gha, *gh;
 879         unsigned int slots = 64;
 880         unsigned int x;
 881         int done;
 882         int error = 0, err;
 883
 884         memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
 885         gha = kmalloc_array(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
 886         if (!gha)
 887                 return -ENOMEM;
 888         for (x = 0; x < slots; x++)
 889                 gfs2_holder_mark_uninitialized(gha + x);
 890
 891         rgd_next = gfs2_rgrpd_get_first(sdp);
 892
 893         for (;;) {
 894                 done = 1;
 895
 896                 for (x = 0; x < slots; x++) {
 897                         gh = gha + x;
 898
 899                         if (gfs2_holder_initialized(gh) && gfs2_glock_poll(gh)) {
 900                                 err = gfs2_glock_wait(gh);
 901                                 if (err) {
 902                                         gfs2_holder_uninit(gh);
 903                                         error = err;
 904                                 } else {
 905                                         if (!error) {
 906                                                 struct gfs2_rgrpd *rgd =
 907                                                         gfs2_glock2rgrp(gh->gh_gl);
 908
 909                                                 error = statfs_slow_fill(rgd, sc);
 910                                         }
 911                                         gfs2_glock_dq_uninit(gh);
 912                                 }
 913                         }
 914
 915                         if (gfs2_holder_initialized(gh))
 916                                 done = 0;
 917                         else if (rgd_next && !error) {
 918                                 error = gfs2_glock_nq_init(rgd_next->rd_gl,
 919                                                            LM_ST_SHARED,
 920                                                            GL_ASYNC,
 921                                                            gh);
 922                                 rgd_next = gfs2_rgrpd_get_next(rgd_next);
 923                                 done = 0;
 924                         }
 925
 926                         if (signal_pending(current))
 927                                 error = -ERESTARTSYS;
 928                 }
 929
 930                 if (done)
 931                         break;
 932
 933                 yield();
 934         }
 935
 936         kfree(gha);
 937         return error;
 938 }
 939
 940 /**
 941  * gfs2_statfs_i - Do a statfs
 942  * @sdp: the filesystem
 943  * @sc: the sc structure
 944  *
 945  * Returns: errno
 946  */
 947
 948 static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
 949 {
 950         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 951         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 952
 953         spin_lock(&sdp->sd_statfs_spin);
 954
 955         *sc = *m_sc;
 956         sc->sc_total += l_sc->sc_total;
 957         sc->sc_free += l_sc->sc_free;
 958         sc->sc_dinodes += l_sc->sc_dinodes;
 959
 960         spin_unlock(&sdp->sd_statfs_spin);
 961
 962         if (sc->sc_free < 0)
 963                 sc->sc_free = 0;
 964         if (sc->sc_free > sc->sc_total)
 965                 sc->sc_free = sc->sc_total;
 966         if (sc->sc_dinodes < 0)
 967                 sc->sc_dinodes = 0;
 968
 969         return 0;
 970 }
 971
 972 /**
 973  * gfs2_statfs - Gather and return stats about the filesystem
 974  * @dentry: The name of the link
 975  * @buf: The buffer
 976  *
 977  * Returns: 0 on success or error code
 978  */
 979
 980 static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
 981 {
 982         struct super_block *sb = dentry->d_sb;
 983         struct gfs2_sbd *sdp = sb->s_fs_info;
 984         struct gfs2_statfs_change_host sc;
 985         int error;
 986
 987         error = gfs2_rindex_update(sdp);
 988         if (error)
 989                 return error;
 990
 991         if (gfs2_tune_get(sdp, gt_statfs_slow))
 992                 error = gfs2_statfs_slow(sdp, &sc);
 993         else
 994                 error = gfs2_statfs_i(sdp, &sc);
 995
 996         if (error)
 997                 return error;
 998
 999         buf->f_type = GFS2_MAGIC;
1000         buf->f_bsize = sdp->sd_sb.sb_bsize;
1001         buf->f_blocks = sc.sc_total;
1002         buf->f_bfree = sc.sc_free;
1003         buf->f_bavail = sc.sc_free;
1004         buf->f_files = sc.sc_dinodes + sc.sc_free;
1005         buf->f_ffree = sc.sc_free;
1006         buf->f_namelen = GFS2_FNAMESIZE;
1007
1008         return 0;
1009 }
1010
1011 /**
1012  * gfs2_drop_inode - Drop an inode (test for remote unlink)
1013  * @inode: The inode to drop
1014  *
1015  * If we've received a callback on an iopen lock then it's because a
1016  * remote node tried to deallocate the inode but failed due to this node
1017  * still having the inode open. Here we mark the link count zero
1018  * since we know that it must have reached zero if the GLF_DEMOTE flag
1019  * is set on the iopen glock. If we didn't do a disk read since the
1020  * remote node removed the final link then we might otherwise miss
1021  * this event. This check ensures that this node will deallocate the
1022  * inode's blocks, or alternatively pass the baton on to another
1023  * node for later deallocation.
1024  */
1025
1026 static int gfs2_drop_inode(struct inode *inode)
1027 {
1028         struct gfs2_inode *ip = GFS2_I(inode);
1029         struct gfs2_sbd *sdp = GFS2_SB(inode);
1030
1031         if (inode->i_nlink &&
1032             gfs2_holder_initialized(&ip->i_iopen_gh)) {
1033                 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1034                 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
1035                         clear_nlink(inode);
1036         }
1037
1038         /*
1039          * When under memory pressure when an inode's link count has dropped to
1040          * zero, defer deleting the inode to the delete workqueue.  This avoids
1041          * calling into DLM under memory pressure, which can deadlock.
1042          */
1043         if (!inode->i_nlink &&
1044             unlikely(current->flags & PF_MEMALLOC) &&
1045             gfs2_holder_initialized(&ip->i_iopen_gh)) {
1046                 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1047
1048                 gfs2_glock_hold(gl);
1049                 if (!gfs2_queue_try_to_evict(gl))
1050                         gfs2_glock_queue_put(gl);
1051                 return 0;
1052         }
1053
1054         /*
1055          * No longer cache inodes when trying to evict them all.
1056          */
1057         if (test_bit(SDF_EVICTING, &sdp->sd_flags))
1058                 return 1;
1059
1060         return generic_drop_inode(inode);
1061 }
1062
1063 static int is_ancestor(const struct dentry *d1, const struct dentry *d2)
1064 {
1065         do {
1066                 if (d1 == d2)
1067                         return 1;
1068                 d1 = d1->d_parent;
1069         } while (!IS_ROOT(d1));
1070         return 0;
1071 }
1072
1073 /**
1074  * gfs2_show_options - Show mount options for /proc/mounts
1075  * @s: seq_file structure
1076  * @root: root of this (sub)tree
1077  *
1078  * Returns: 0 on success or error code
1079  */
1080
1081 static int gfs2_show_options(struct seq_file *s, struct dentry *root)
1082 {
1083         struct gfs2_sbd *sdp = root->d_sb->s_fs_info;
1084         struct gfs2_args *args = &sdp->sd_args;
1085         unsigned int logd_secs, statfs_slow, statfs_quantum, quota_quantum;
1086
1087         spin_lock(&sdp->sd_tune.gt_spin);
1088         logd_secs = sdp->sd_tune.gt_logd_secs;
1089         quota_quantum = sdp->sd_tune.gt_quota_quantum;
1090         statfs_quantum = sdp->sd_tune.gt_statfs_quantum;
1091         statfs_slow = sdp->sd_tune.gt_statfs_slow;
1092         spin_unlock(&sdp->sd_tune.gt_spin);
1093
1094         if (is_ancestor(root, sdp->sd_master_dir))
1095                 seq_puts(s, ",meta");
1096         if (args->ar_lockproto[0])
1097                 seq_show_option(s, "lockproto", args->ar_lockproto);
1098         if (args->ar_locktable[0])
1099                 seq_show_option(s, "locktable", args->ar_locktable);
1100         if (args->ar_hostdata[0])
1101                 seq_show_option(s, "hostdata", args->ar_hostdata);
1102         if (args->ar_spectator)
1103                 seq_puts(s, ",spectator");
1104         if (args->ar_localflocks)
1105                 seq_puts(s, ",localflocks");
1106         if (args->ar_debug)
1107                 seq_puts(s, ",debug");
1108         if (args->ar_posix_acl)
1109                 seq_puts(s, ",acl");
1110         if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
1111                 char *state;
1112                 switch (args->ar_quota) {
1113                 case GFS2_QUOTA_OFF:
1114                         state = "off";
1115                         break;
1116                 case GFS2_QUOTA_ACCOUNT:
1117                         state = "account";
1118                         break;
1119                 case GFS2_QUOTA_ON:
1120                         state = "on";
1121                         break;
1122                 default:
1123                         state = "unknown";
1124                         break;
1125                 }
1126                 seq_printf(s, ",quota=%s", state);
1127         }
1128         if (args->ar_suiddir)
1129                 seq_puts(s, ",suiddir");
1130         if (args->ar_data != GFS2_DATA_DEFAULT) {
1131                 char *state;
1132                 switch (args->ar_data) {
1133                 case GFS2_DATA_WRITEBACK:
1134                         state = "writeback";
1135                         break;
1136                 case GFS2_DATA_ORDERED:
1137                         state = "ordered";
1138                         break;
1139                 default:
1140                         state = "unknown";
1141                         break;
1142                 }
1143                 seq_printf(s, ",data=%s", state);
1144         }
1145         if (args->ar_discard)
1146                 seq_puts(s, ",discard");
1147         if (logd_secs != 30)
1148                 seq_printf(s, ",commit=%d", logd_secs);
1149         if (statfs_quantum != 30)
1150                 seq_printf(s, ",statfs_quantum=%d", statfs_quantum);
1151         else if (statfs_slow)
1152                 seq_puts(s, ",statfs_quantum=0");
1153         if (quota_quantum != 60)
1154                 seq_printf(s, ",quota_quantum=%d", quota_quantum);
1155         if (args->ar_statfs_percent)
1156                 seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent);
1157         if (args->ar_errors != GFS2_ERRORS_DEFAULT) {
1158                 const char *state;
1159
1160                 switch (args->ar_errors) {
1161                 case GFS2_ERRORS_WITHDRAW:
1162                         state = "withdraw";
1163                         break;
1164                 case GFS2_ERRORS_PANIC:
1165                         state = "panic";
1166                         break;
1167                 default:
1168                         state = "unknown";
1169                         break;
1170                 }
1171                 seq_printf(s, ",errors=%s", state);
1172         }
1173         if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags))
1174                 seq_puts(s, ",nobarrier");
1175         if (test_bit(SDF_DEMOTE, &sdp->sd_flags))
1176                 seq_puts(s, ",demote_interface_used");
1177         if (args->ar_rgrplvb)
1178                 seq_puts(s, ",rgrplvb");
1179         if (args->ar_loccookie)
1180                 seq_puts(s, ",loccookie");
1181         return 0;
1182 }
1183
1184 static void gfs2_final_release_pages(struct gfs2_inode *ip)
1185 {
1186         struct inode *inode = &ip->i_inode;
1187         struct gfs2_glock *gl = ip->i_gl;
1188
1189         if (unlikely(!gl)) {
1190                 /* This can only happen during incomplete inode creation. */
1191                 BUG_ON(!test_bit(GIF_ALLOC_FAILED, &ip->i_flags));
1192                 return;
1193         }
1194
1195         truncate_inode_pages(gfs2_glock2aspace(gl), 0);
1196         truncate_inode_pages(&inode->i_data, 0);
1197
1198         if (atomic_read(&gl->gl_revokes) == 0) {
1199                 clear_bit(GLF_LFLUSH, &gl->gl_flags);
1200                 clear_bit(GLF_DIRTY, &gl->gl_flags);
1201         }
1202 }
1203
1204 static int gfs2_dinode_dealloc(struct gfs2_inode *ip)
1205 {
1206         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1207         struct gfs2_rgrpd *rgd;
1208         struct gfs2_holder gh;
1209         int error;
1210
1211         if (gfs2_get_inode_blocks(&ip->i_inode) != 1) {
1212                 gfs2_consist_inode(ip);
1213                 return -EIO;
1214         }
1215
1216         gfs2_rindex_update(sdp);
1217
1218         error = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
1219         if (error)
1220                 return error;
1221
1222         rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1223         if (!rgd) {
1224                 gfs2_consist_inode(ip);
1225                 error = -EIO;
1226                 goto out_qs;
1227         }
1228
1229         error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE,
1230                                    LM_FLAG_NODE_SCOPE, &gh);
1231         if (error)
1232                 goto out_qs;
1233
1234         error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA,
1235                                  sdp->sd_jdesc->jd_blocks);
1236         if (error)
1237                 goto out_rg_gunlock;
1238
1239         gfs2_free_di(rgd, ip);
1240
1241         gfs2_final_release_pages(ip);
1242
1243         gfs2_trans_end(sdp);
1244
1245 out_rg_gunlock:
1246         gfs2_glock_dq_uninit(&gh);
1247 out_qs:
1248         gfs2_quota_unhold(ip);
1249         return error;
1250 }
1251
1252 /**
1253  * gfs2_glock_put_eventually
1254  * @gl: The glock to put
1255  *
1256  * When under memory pressure, trigger a deferred glock put to make sure we
1257  * won't call into DLM and deadlock.  Otherwise, put the glock directly.
1258  */
1259
1260 static void gfs2_glock_put_eventually(struct gfs2_glock *gl)
1261 {
1262         if (current->flags & PF_MEMALLOC)
1263                 gfs2_glock_queue_put(gl);
1264         else
1265                 gfs2_glock_put(gl);
1266 }
1267
1268 static bool gfs2_upgrade_iopen_glock(struct inode *inode)
1269 {
1270         struct gfs2_inode *ip = GFS2_I(inode);
1271         struct gfs2_sbd *sdp = GFS2_SB(inode);
1272         struct gfs2_holder *gh = &ip->i_iopen_gh;
1273         long timeout = 5 * HZ;
1274         int error;
1275
1276         gh->gh_flags |= GL_NOCACHE;
1277         gfs2_glock_dq_wait(gh);
1278
1279         /*
1280          * If there are no other lock holders, we will immediately get
1281          * exclusive access to the iopen glock here.
1282          *
1283          * Otherwise, the other nodes holding the lock will be notified about
1284          * our locking request.  If they do not have the inode open, they are
1285          * expected to evict the cached inode and release the lock, allowing us
1286          * to proceed.
1287          *
1288          * Otherwise, if they cannot evict the inode, they are expected to poke
1289          * the inode glock (note: not the iopen glock).  We will notice that
1290          * and stop waiting for the iopen glock immediately.  The other node(s)
1291          * are then expected to take care of deleting the inode when they no
1292          * longer use it.
1293          *
1294          * As a last resort, if another node keeps holding the iopen glock
1295          * without showing any activity on the inode glock, we will eventually
1296          * time out and fail the iopen glock upgrade.
1297          *
1298          * Note that we're passing the LM_FLAG_TRY_1CB flag to the first
1299          * locking request as an optimization to notify lock holders as soon as
1300          * possible.  Without that flag, they'd be notified implicitly by the
1301          * second locking request.
1302          */
1303
1304         gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE, gh);
1305         error = gfs2_glock_nq(gh);
1306         if (error != GLR_TRYFAILED)
1307                 return !error;
1308
1309         gfs2_holder_reinit(LM_ST_EXCLUSIVE, GL_ASYNC | GL_NOCACHE, gh);
1310         error = gfs2_glock_nq(gh);
1311         if (error)
1312                 return false;
1313
1314         timeout = wait_event_interruptible_timeout(sdp->sd_async_glock_wait,
1315                 !test_bit(HIF_WAIT, &gh->gh_iflags) ||
1316                 test_bit(GLF_DEMOTE, &ip->i_gl->gl_flags),
1317                 timeout);
1318         if (!test_bit(HIF_HOLDER, &gh->gh_iflags)) {
1319                 gfs2_glock_dq(gh);
1320                 return false;
1321         }
1322         return gfs2_glock_holder_ready(gh) == 0;
1323 }
1324
1325 /**
1326  * evict_should_delete - determine whether the inode is eligible for deletion
1327  * @inode: The inode to evict
1328  * @gh: The glock holder structure
1329  *
1330  * This function determines whether the evicted inode is eligible to be deleted
1331  * and locks the inode glock.
1332  *
1333  * Returns: the fate of the dinode
1334  */
1335 static enum dinode_demise evict_should_delete(struct inode *inode,
1336                                               struct gfs2_holder *gh)
1337 {
1338         struct gfs2_inode *ip = GFS2_I(inode);
1339         struct super_block *sb = inode->i_sb;
1340         struct gfs2_sbd *sdp = sb->s_fs_info;
1341         int ret;
1342
1343         if (unlikely(test_bit(GIF_ALLOC_FAILED, &ip->i_flags)))
1344                 goto should_delete;
1345
1346         if (test_bit(GIF_DEFERRED_DELETE, &ip->i_flags))
1347                 return SHOULD_DEFER_EVICTION;
1348
1349         /* Deletes should never happen under memory pressure anymore.  */
1350         if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
1351                 return SHOULD_DEFER_EVICTION;
1352
1353         /* Must not read inode block until block type has been verified */
1354         ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, gh);
1355         if (unlikely(ret)) {
1356                 glock_clear_object(ip->i_iopen_gh.gh_gl, ip);
1357                 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1358                 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1359                 return SHOULD_DEFER_EVICTION;
1360         }
1361
1362         if (gfs2_inode_already_deleted(ip->i_gl, ip->i_no_formal_ino))
1363                 return SHOULD_NOT_DELETE_DINODE;
1364         ret = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED);
1365         if (ret)
1366                 return SHOULD_NOT_DELETE_DINODE;
1367
1368         ret = gfs2_instantiate(gh);
1369         if (ret)
1370                 return SHOULD_NOT_DELETE_DINODE;
1371
1372         /*
1373          * The inode may have been recreated in the meantime.
1374          */
1375         if (inode->i_nlink)
1376                 return SHOULD_NOT_DELETE_DINODE;
1377
1378 should_delete:
1379         if (gfs2_holder_initialized(&ip->i_iopen_gh) &&
1380             test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1381                 if (!gfs2_upgrade_iopen_glock(inode)) {
1382                         gfs2_holder_uninit(&ip->i_iopen_gh);
1383                         return SHOULD_NOT_DELETE_DINODE;
1384                 }
1385         }
1386         return SHOULD_DELETE_DINODE;
1387 }
1388
1389 /**
1390  * evict_unlinked_inode - delete the pieces of an unlinked evicted inode
1391  * @inode: The inode to evict
1392  */
1393 static int evict_unlinked_inode(struct inode *inode)
1394 {
1395         struct gfs2_inode *ip = GFS2_I(inode);
1396         int ret;
1397
1398         if (S_ISDIR(inode->i_mode) &&
1399             (ip->i_diskflags & GFS2_DIF_EXHASH)) {
1400                 ret = gfs2_dir_exhash_dealloc(ip);
1401                 if (ret)
1402                         goto out;
1403         }
1404
1405         if (ip->i_eattr) {
1406                 ret = gfs2_ea_dealloc(ip);
1407                 if (ret)
1408                         goto out;
1409         }
1410
1411         if (!gfs2_is_stuffed(ip)) {
1412                 ret = gfs2_file_dealloc(ip);
1413                 if (ret)
1414                         goto out;
1415         }
1416
1417         /*
1418          * As soon as we clear the bitmap for the dinode, gfs2_create_inode()
1419          * can get called to recreate it, or even gfs2_inode_lookup() if the
1420          * inode was recreated on another node in the meantime.
1421          *
1422          * However, inserting the new inode into the inode hash table will not
1423          * succeed until the old inode is removed, and that only happens after
1424          * ->evict_inode() returns.  The new inode is attached to its inode and
1425          *  iopen glocks after inserting it into the inode hash table, so at
1426          *  that point we can be sure that both glocks are unused.
1427          */
1428
1429         ret = gfs2_dinode_dealloc(ip);
1430         if (!ret && ip->i_gl)
1431                 gfs2_inode_remember_delete(ip->i_gl, ip->i_no_formal_ino);
1432
1433 out:
1434         return ret;
1435 }
1436
1437 /*
1438  * evict_linked_inode - evict an inode whose dinode has not been unlinked
1439  * @inode: The inode to evict
1440  */
1441 static int evict_linked_inode(struct inode *inode)
1442 {
1443         struct super_block *sb = inode->i_sb;
1444         struct gfs2_sbd *sdp = sb->s_fs_info;
1445         struct gfs2_inode *ip = GFS2_I(inode);
1446         struct address_space *metamapping;
1447         int ret;
1448
1449         gfs2_log_flush(sdp, ip->i_gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
1450                        GFS2_LFC_EVICT_INODE);
1451         metamapping = gfs2_glock2aspace(ip->i_gl);
1452         if (test_bit(GLF_DIRTY, &ip->i_gl->gl_flags)) {
1453                 filemap_fdatawrite(metamapping);
1454                 filemap_fdatawait(metamapping);
1455         }
1456         write_inode_now(inode, 1);
1457         gfs2_ail_flush(ip->i_gl, 0);
1458
1459         ret = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
1460         if (ret)
1461                 return ret;
1462
1463         /* Needs to be done before glock release & also in a transaction */
1464         truncate_inode_pages(&inode->i_data, 0);
1465         truncate_inode_pages(metamapping, 0);
1466         gfs2_trans_end(sdp);
1467         return 0;
1468 }
1469
1470 /**
1471  * gfs2_evict_inode - Remove an inode from cache
1472  * @inode: The inode to evict
1473  *
1474  * There are three cases to consider:
1475  * 1. i_nlink == 0, we are final opener (and must deallocate)
1476  * 2. i_nlink == 0, we are not the final opener (and cannot deallocate)
1477  * 3. i_nlink > 0
1478  *
1479  * If the fs is read only, then we have to treat all cases as per #3
1480  * since we are unable to do any deallocation. The inode will be
1481  * deallocated by the next read/write node to attempt an allocation
1482  * in the same resource group
1483  *
1484  * We have to (at the moment) hold the inodes main lock to cover
1485  * the gap between unlocking the shared lock on the iopen lock and
1486  * taking the exclusive lock. I'd rather do a shared -> exclusive
1487  * conversion on the iopen lock, but we can change that later. This
1488  * is safe, just less efficient.
1489  */
1490
1491 static void gfs2_evict_inode(struct inode *inode)
1492 {
1493         struct super_block *sb = inode->i_sb;
1494         struct gfs2_sbd *sdp = sb->s_fs_info;
1495         struct gfs2_inode *ip = GFS2_I(inode);
1496         struct gfs2_holder gh;
1497         int ret;
1498
1499         if (inode->i_nlink || sb_rdonly(sb) || !ip->i_no_addr)
1500                 goto out;
1501
1502         /*
1503          * In case of an incomplete mount, gfs2_evict_inode() may be called for
1504          * system files without having an active journal to write to.  In that
1505          * case, skip the filesystem evict.
1506          */
1507         if (!sdp->sd_jdesc)
1508                 goto out;
1509
1510         gfs2_holder_mark_uninitialized(&gh);
1511         ret = evict_should_delete(inode, &gh);
1512         if (ret == SHOULD_DEFER_EVICTION)
1513                 goto out;
1514         if (ret == SHOULD_DELETE_DINODE)
1515                 ret = evict_unlinked_inode(inode);
1516         else
1517                 ret = evict_linked_inode(inode);
1518
1519         if (gfs2_rs_active(&ip->i_res))
1520                 gfs2_rs_deltree(&ip->i_res);
1521
1522         if (gfs2_holder_initialized(&gh))
1523                 gfs2_glock_dq_uninit(&gh);
1524         if (ret && ret != GLR_TRYFAILED && ret != -EROFS)
1525                 fs_warn(sdp, "gfs2_evict_inode: %d\n", ret);
1526 out:
1527         truncate_inode_pages_final(&inode->i_data);
1528         if (ip->i_qadata)
1529                 gfs2_assert_warn(sdp, ip->i_qadata->qa_ref == 0);
1530         gfs2_rs_deltree(&ip->i_res);
1531         gfs2_ordered_del_inode(ip);
1532         clear_inode(inode);
1533         gfs2_dir_hash_inval(ip);
1534         if (gfs2_holder_initialized(&ip->i_iopen_gh)) {
1535                 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1536
1537                 glock_clear_object(gl, ip);
1538                 gfs2_glock_hold(gl);
1539                 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1540                 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1541                 gfs2_glock_put_eventually(gl);
1542         }
1543         if (ip->i_gl) {
1544                 glock_clear_object(ip->i_gl, ip);
1545                 wait_on_bit_io(&ip->i_flags, GIF_GLOP_PENDING, TASK_UNINTERRUPTIBLE);
1546                 gfs2_glock_add_to_lru(ip->i_gl);
1547                 gfs2_glock_put_eventually(ip->i_gl);
1548                 ip->i_gl = NULL;
1549         }
1550 }
1551
1552 static struct inode *gfs2_alloc_inode(struct super_block *sb)
1553 {
1554         struct gfs2_inode *ip;
1555
1556         ip = alloc_inode_sb(sb, gfs2_inode_cachep, GFP_KERNEL);
1557         if (!ip)
1558                 return NULL;
1559         ip->i_no_addr = 0;
1560         ip->i_flags = 0;
1561         ip->i_gl = NULL;
1562         gfs2_holder_mark_uninitialized(&ip->i_iopen_gh);
1563         memset(&ip->i_res, 0, sizeof(ip->i_res));
1564         RB_CLEAR_NODE(&ip->i_res.rs_node);
1565         ip->i_rahead = 0;
1566         return &ip->i_inode;
1567 }
1568
1569 static void gfs2_free_inode(struct inode *inode)
1570 {
1571         kmem_cache_free(gfs2_inode_cachep, GFS2_I(inode));
1572 }
1573
1574 extern void free_local_statfs_inodes(struct gfs2_sbd *sdp)
1575 {
1576         struct local_statfs_inode *lsi, *safe;
1577
1578         /* Run through the statfs inodes list to iput and free memory */
1579         list_for_each_entry_safe(lsi, safe, &sdp->sd_sc_inodes_list, si_list) {
1580                 if (lsi->si_jid == sdp->sd_jdesc->jd_jid)
1581                         sdp->sd_sc_inode = NULL; /* belongs to this node */
1582                 if (lsi->si_sc_inode)
1583                         iput(lsi->si_sc_inode);
1584                 list_del(&lsi->si_list);
1585                 kfree(lsi);
1586         }
1587 }
1588
1589 extern struct inode *find_local_statfs_inode(struct gfs2_sbd *sdp,
1590                                              unsigned int index)
1591 {
1592         struct local_statfs_inode *lsi;
1593
1594         /* Return the local (per node) statfs inode in the
1595          * sdp->sd_sc_inodes_list corresponding to the 'index'. */
1596         list_for_each_entry(lsi, &sdp->sd_sc_inodes_list, si_list) {
1597                 if (lsi->si_jid == index)
1598                         return lsi->si_sc_inode;
1599         }
1600         return NULL;
1601 }
1602
1603 const struct super_operations gfs2_super_ops = {
1604         .alloc_inode            = gfs2_alloc_inode,
1605         .free_inode             = gfs2_free_inode,
1606         .write_inode            = gfs2_write_inode,
1607         .dirty_inode            = gfs2_dirty_inode,
1608         .evict_inode            = gfs2_evict_inode,
1609         .put_super              = gfs2_put_super,
1610         .sync_fs                = gfs2_sync_fs,
1611         .freeze_super           = gfs2_freeze_super,
1612         .thaw_super             = gfs2_thaw_super,
1613         .statfs                 = gfs2_statfs,
1614         .drop_inode             = gfs2_drop_inode,
1615         .show_options           = gfs2_show_options,
1616 };
1617