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         if (!log_write_allowed && current == sdp->sd_quotad_process)
 553                 fs_warn(sdp, "The quotad daemon is withdrawing.\n");
 554         else if (sdp->sd_quotad_process)
 555                 kthread_stop(sdp->sd_quotad_process);
 556         sdp->sd_quotad_process = NULL;
 557
 558         if (!log_write_allowed && current == sdp->sd_logd_process)
 559                 fs_warn(sdp, "The logd daemon is withdrawing.\n");
 560         else if (sdp->sd_logd_process)
 561                 kthread_stop(sdp->sd_logd_process);
 562         sdp->sd_logd_process = NULL;
 563
 564         if (log_write_allowed) {
 565                 gfs2_quota_sync(sdp->sd_vfs, 0);
 566                 gfs2_statfs_sync(sdp->sd_vfs, 0);
 567
 568                 /* We do two log flushes here. The first one commits dirty inodes
 569                  * and rgrps to the journal, but queues up revokes to the ail list.
 570                  * The second flush writes out and removes the revokes.
 571                  *
 572                  * The first must be done before the FLUSH_SHUTDOWN code
 573                  * clears the LIVE flag, otherwise it will not be able to start
 574                  * a transaction to write its revokes, and the error will cause
 575                  * a withdraw of the file system. */
 576                 gfs2_log_flush(sdp, NULL, GFS2_LFC_MAKE_FS_RO);
 577                 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SHUTDOWN |
 578                                GFS2_LFC_MAKE_FS_RO);
 579                 wait_event_timeout(sdp->sd_log_waitq,
 580                                    gfs2_log_is_empty(sdp),
 581                                    HZ * 5);
 582                 gfs2_assert_warn(sdp, gfs2_log_is_empty(sdp));
 583         }
 584         gfs2_quota_cleanup(sdp);
 585 }
 586
 587 /**
 588  * gfs2_put_super - Unmount the filesystem
 589  * @sb: The VFS superblock
 590  *
 591  */
 592
 593 static void gfs2_put_super(struct super_block *sb)
 594 {
 595         struct gfs2_sbd *sdp = sb->s_fs_info;
 596         struct gfs2_jdesc *jd;
 597
 598         /* No more recovery requests */
 599         set_bit(SDF_NORECOVERY, &sdp->sd_flags);
 600         smp_mb();
 601
 602         /* Wait on outstanding recovery */
 603 restart:
 604         spin_lock(&sdp->sd_jindex_spin);
 605         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 606                 if (!test_bit(JDF_RECOVERY, &jd->jd_flags))
 607                         continue;
 608                 spin_unlock(&sdp->sd_jindex_spin);
 609                 wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
 610                             TASK_UNINTERRUPTIBLE);
 611                 goto restart;
 612         }
 613         spin_unlock(&sdp->sd_jindex_spin);
 614
 615         if (!sb_rdonly(sb)) {
 616                 gfs2_make_fs_ro(sdp);
 617         }
 618         if (gfs2_withdrawn(sdp))
 619                 gfs2_quota_cleanup(sdp);
 620         WARN_ON(gfs2_withdrawing(sdp));
 621
 622         /*  At this point, we're through modifying the disk  */
 623
 624         /*  Release stuff  */
 625
 626         gfs2_freeze_unlock(&sdp->sd_freeze_gh);
 627
 628         iput(sdp->sd_jindex);
 629         iput(sdp->sd_statfs_inode);
 630         iput(sdp->sd_rindex);
 631         iput(sdp->sd_quota_inode);
 632
 633         gfs2_glock_put(sdp->sd_rename_gl);
 634         gfs2_glock_put(sdp->sd_freeze_gl);
 635
 636         if (!sdp->sd_args.ar_spectator) {
 637                 if (gfs2_holder_initialized(&sdp->sd_journal_gh))
 638                         gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
 639                 if (gfs2_holder_initialized(&sdp->sd_jinode_gh))
 640                         gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
 641                 brelse(sdp->sd_sc_bh);
 642                 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
 643                 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
 644                 free_local_statfs_inodes(sdp);
 645                 iput(sdp->sd_qc_inode);
 646         }
 647
 648         gfs2_glock_dq_uninit(&sdp->sd_live_gh);
 649         gfs2_clear_rgrpd(sdp);
 650         gfs2_jindex_free(sdp);
 651         /*  Take apart glock structures and buffer lists  */
 652         gfs2_gl_hash_clear(sdp);
 653         truncate_inode_pages_final(&sdp->sd_aspace);
 654         gfs2_delete_debugfs_file(sdp);
 655         /*  Unmount the locking protocol  */
 656         gfs2_lm_unmount(sdp);
 657
 658         /*  At this point, we're through participating in the lockspace  */
 659         gfs2_sys_fs_del(sdp);
 660         free_sbd(sdp);
 661 }
 662
 663 /**
 664  * gfs2_sync_fs - sync the filesystem
 665  * @sb: the superblock
 666  * @wait: true to wait for completion
 667  *
 668  * Flushes the log to disk.
 669  */
 670
 671 static int gfs2_sync_fs(struct super_block *sb, int wait)
 672 {
 673         struct gfs2_sbd *sdp = sb->s_fs_info;
 674
 675         gfs2_quota_sync(sb, -1);
 676         if (wait)
 677                 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
 678                                GFS2_LFC_SYNC_FS);
 679         return sdp->sd_log_error;
 680 }
 681
 682 static int gfs2_freeze_locally(struct gfs2_sbd *sdp)
 683 {
 684         struct super_block *sb = sdp->sd_vfs;
 685         int error;
 686
 687         error = freeze_super(sb);
 688         if (error)
 689                 return error;
 690
 691         if (test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
 692                 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_FREEZE |
 693                                GFS2_LFC_FREEZE_GO_SYNC);
 694                 if (gfs2_withdrawn(sdp)) {
 695                         thaw_super(sb);
 696                         return -EIO;
 697                 }
 698         }
 699         return 0;
 700 }
 701
 702 static int gfs2_do_thaw(struct gfs2_sbd *sdp)
 703 {
 704         struct super_block *sb = sdp->sd_vfs;
 705         int error;
 706
 707         error = gfs2_freeze_lock_shared(sdp);
 708         if (error)
 709                 goto fail;
 710         error = thaw_super(sb);
 711         if (!error)
 712                 return 0;
 713
 714 fail:
 715         fs_info(sdp, "GFS2: couldn't thaw filesystem: %d\n", error);
 716         gfs2_assert_withdraw(sdp, 0);
 717         return error;
 718 }
 719
 720 void gfs2_freeze_func(struct work_struct *work)
 721 {
 722         struct gfs2_sbd *sdp = container_of(work, struct gfs2_sbd, sd_freeze_work);
 723         struct super_block *sb = sdp->sd_vfs;
 724         int error;
 725
 726         mutex_lock(&sdp->sd_freeze_mutex);
 727         error = -EBUSY;
 728         if (test_bit(SDF_FROZEN, &sdp->sd_flags))
 729                 goto freeze_failed;
 730
 731         error = gfs2_freeze_locally(sdp);
 732         if (error)
 733                 goto freeze_failed;
 734
 735         gfs2_freeze_unlock(&sdp->sd_freeze_gh);
 736         set_bit(SDF_FROZEN, &sdp->sd_flags);
 737
 738         error = gfs2_do_thaw(sdp);
 739         if (error)
 740                 goto out;
 741
 742         clear_bit(SDF_FROZEN, &sdp->sd_flags);
 743         goto out;
 744
 745 freeze_failed:
 746         fs_info(sdp, "GFS2: couldn't freeze filesystem: %d\n", error);
 747
 748 out:
 749         mutex_unlock(&sdp->sd_freeze_mutex);
 750         deactivate_super(sb);
 751 }
 752
 753 /**
 754  * gfs2_freeze_super - prevent further writes to the filesystem
 755  * @sb: the VFS structure for the filesystem
 756  *
 757  */
 758
 759 static int gfs2_freeze_super(struct super_block *sb)
 760 {
 761         struct gfs2_sbd *sdp = sb->s_fs_info;
 762         int error;
 763
 764         if (!mutex_trylock(&sdp->sd_freeze_mutex))
 765                 return -EBUSY;
 766         error = -EBUSY;
 767         if (test_bit(SDF_FROZEN, &sdp->sd_flags))
 768                 goto out;
 769
 770         for (;;) {
 771                 error = gfs2_freeze_locally(sdp);
 772                 if (error) {
 773                         fs_info(sdp, "GFS2: couldn't freeze filesystem: %d\n",
 774                                 error);
 775                         goto out;
 776                 }
 777
 778                 error = gfs2_lock_fs_check_clean(sdp);
 779                 if (!error)
 780                         break;  /* success */
 781
 782                 error = gfs2_do_thaw(sdp);
 783                 if (error)
 784                         goto out;
 785
 786                 if (error == -EBUSY)
 787                         fs_err(sdp, "waiting for recovery before freeze\n");
 788                 else if (error == -EIO) {
 789                         fs_err(sdp, "Fatal IO error: cannot freeze gfs2 due "
 790                                "to recovery error.\n");
 791                         goto out;
 792                 } else {
 793                         fs_err(sdp, "error freezing FS: %d\n", error);
 794                 }
 795                 fs_err(sdp, "retrying...\n");
 796                 msleep(1000);
 797         }
 798
 799 out:
 800         if (!error) {
 801                 set_bit(SDF_FREEZE_INITIATOR, &sdp->sd_flags);
 802                 set_bit(SDF_FROZEN, &sdp->sd_flags);
 803         }
 804         mutex_unlock(&sdp->sd_freeze_mutex);
 805         return error;
 806 }
 807
 808 /**
 809  * gfs2_thaw_super - reallow writes to the filesystem
 810  * @sb: the VFS structure for the filesystem
 811  *
 812  */
 813
 814 static int gfs2_thaw_super(struct super_block *sb)
 815 {
 816         struct gfs2_sbd *sdp = sb->s_fs_info;
 817         int error;
 818
 819         if (!mutex_trylock(&sdp->sd_freeze_mutex))
 820                 return -EBUSY;
 821         error = -EINVAL;
 822         if (!test_bit(SDF_FREEZE_INITIATOR, &sdp->sd_flags))
 823                 goto out;
 824
 825         gfs2_freeze_unlock(&sdp->sd_freeze_gh);
 826
 827         error = gfs2_do_thaw(sdp);
 828
 829         if (!error) {
 830                 clear_bit(SDF_FREEZE_INITIATOR, &sdp->sd_flags);
 831                 clear_bit(SDF_FROZEN, &sdp->sd_flags);
 832         }
 833 out:
 834         mutex_unlock(&sdp->sd_freeze_mutex);
 835         return error;
 836 }
 837
 838 void gfs2_thaw_freeze_initiator(struct super_block *sb)
 839 {
 840         struct gfs2_sbd *sdp = sb->s_fs_info;
 841
 842         mutex_lock(&sdp->sd_freeze_mutex);
 843         if (!test_bit(SDF_FREEZE_INITIATOR, &sdp->sd_flags))
 844                 goto out;
 845
 846         gfs2_freeze_unlock(&sdp->sd_freeze_gh);
 847
 848 out:
 849         mutex_unlock(&sdp->sd_freeze_mutex);
 850 }
 851
 852 /**
 853  * statfs_slow_fill - fill in the sg for a given RG
 854  * @rgd: the RG
 855  * @sc: the sc structure
 856  *
 857  * Returns: 0 on success, -ESTALE if the LVB is invalid
 858  */
 859
 860 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
 861                             struct gfs2_statfs_change_host *sc)
 862 {
 863         gfs2_rgrp_verify(rgd);
 864         sc->sc_total += rgd->rd_data;
 865         sc->sc_free += rgd->rd_free;
 866         sc->sc_dinodes += rgd->rd_dinodes;
 867         return 0;
 868 }
 869
 870 /**
 871  * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
 872  * @sdp: the filesystem
 873  * @sc: the sc info that will be returned
 874  *
 875  * Any error (other than a signal) will cause this routine to fall back
 876  * to the synchronous version.
 877  *
 878  * FIXME: This really shouldn't busy wait like this.
 879  *
 880  * Returns: errno
 881  */
 882
 883 static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
 884 {
 885         struct gfs2_rgrpd *rgd_next;
 886         struct gfs2_holder *gha, *gh;
 887         unsigned int slots = 64;
 888         unsigned int x;
 889         int done;
 890         int error = 0, err;
 891
 892         memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
 893         gha = kmalloc_array(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
 894         if (!gha)
 895                 return -ENOMEM;
 896         for (x = 0; x < slots; x++)
 897                 gfs2_holder_mark_uninitialized(gha + x);
 898
 899         rgd_next = gfs2_rgrpd_get_first(sdp);
 900
 901         for (;;) {
 902                 done = 1;
 903
 904                 for (x = 0; x < slots; x++) {
 905                         gh = gha + x;
 906
 907                         if (gfs2_holder_initialized(gh) && gfs2_glock_poll(gh)) {
 908                                 err = gfs2_glock_wait(gh);
 909                                 if (err) {
 910                                         gfs2_holder_uninit(gh);
 911                                         error = err;
 912                                 } else {
 913                                         if (!error) {
 914                                                 struct gfs2_rgrpd *rgd =
 915                                                         gfs2_glock2rgrp(gh->gh_gl);
 916
 917                                                 error = statfs_slow_fill(rgd, sc);
 918                                         }
 919                                         gfs2_glock_dq_uninit(gh);
 920                                 }
 921                         }
 922
 923                         if (gfs2_holder_initialized(gh))
 924                                 done = 0;
 925                         else if (rgd_next && !error) {
 926                                 error = gfs2_glock_nq_init(rgd_next->rd_gl,
 927                                                            LM_ST_SHARED,
 928                                                            GL_ASYNC,
 929                                                            gh);
 930                                 rgd_next = gfs2_rgrpd_get_next(rgd_next);
 931                                 done = 0;
 932                         }
 933
 934                         if (signal_pending(current))
 935                                 error = -ERESTARTSYS;
 936                 }
 937
 938                 if (done)
 939                         break;
 940
 941                 yield();
 942         }
 943
 944         kfree(gha);
 945         return error;
 946 }
 947
 948 /**
 949  * gfs2_statfs_i - Do a statfs
 950  * @sdp: the filesystem
 951  * @sc: the sc structure
 952  *
 953  * Returns: errno
 954  */
 955
 956 static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
 957 {
 958         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 959         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 960
 961         spin_lock(&sdp->sd_statfs_spin);
 962
 963         *sc = *m_sc;
 964         sc->sc_total += l_sc->sc_total;
 965         sc->sc_free += l_sc->sc_free;
 966         sc->sc_dinodes += l_sc->sc_dinodes;
 967
 968         spin_unlock(&sdp->sd_statfs_spin);
 969
 970         if (sc->sc_free < 0)
 971                 sc->sc_free = 0;
 972         if (sc->sc_free > sc->sc_total)
 973                 sc->sc_free = sc->sc_total;
 974         if (sc->sc_dinodes < 0)
 975                 sc->sc_dinodes = 0;
 976
 977         return 0;
 978 }
 979
 980 /**
 981  * gfs2_statfs - Gather and return stats about the filesystem
 982  * @dentry: The name of the link
 983  * @buf: The buffer
 984  *
 985  * Returns: 0 on success or error code
 986  */
 987
 988 static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
 989 {
 990         struct super_block *sb = dentry->d_sb;
 991         struct gfs2_sbd *sdp = sb->s_fs_info;
 992         struct gfs2_statfs_change_host sc;
 993         int error;
 994
 995         error = gfs2_rindex_update(sdp);
 996         if (error)
 997                 return error;
 998
 999         if (gfs2_tune_get(sdp, gt_statfs_slow))
1000                 error = gfs2_statfs_slow(sdp, &sc);
1001         else
1002                 error = gfs2_statfs_i(sdp, &sc);
1003
1004         if (error)
1005                 return error;
1006
1007         buf->f_type = GFS2_MAGIC;
1008         buf->f_bsize = sdp->sd_sb.sb_bsize;
1009         buf->f_blocks = sc.sc_total;
1010         buf->f_bfree = sc.sc_free;
1011         buf->f_bavail = sc.sc_free;
1012         buf->f_files = sc.sc_dinodes + sc.sc_free;
1013         buf->f_ffree = sc.sc_free;
1014         buf->f_namelen = GFS2_FNAMESIZE;
1015
1016         return 0;
1017 }
1018
1019 /**
1020  * gfs2_drop_inode - Drop an inode (test for remote unlink)
1021  * @inode: The inode to drop
1022  *
1023  * If we've received a callback on an iopen lock then it's because a
1024  * remote node tried to deallocate the inode but failed due to this node
1025  * still having the inode open. Here we mark the link count zero
1026  * since we know that it must have reached zero if the GLF_DEMOTE flag
1027  * is set on the iopen glock. If we didn't do a disk read since the
1028  * remote node removed the final link then we might otherwise miss
1029  * this event. This check ensures that this node will deallocate the
1030  * inode's blocks, or alternatively pass the baton on to another
1031  * node for later deallocation.
1032  */
1033
1034 static int gfs2_drop_inode(struct inode *inode)
1035 {
1036         struct gfs2_inode *ip = GFS2_I(inode);
1037         struct gfs2_sbd *sdp = GFS2_SB(inode);
1038
1039         if (inode->i_nlink &&
1040             gfs2_holder_initialized(&ip->i_iopen_gh)) {
1041                 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1042                 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
1043                         clear_nlink(inode);
1044         }
1045
1046         /*
1047          * When under memory pressure when an inode's link count has dropped to
1048          * zero, defer deleting the inode to the delete workqueue.  This avoids
1049          * calling into DLM under memory pressure, which can deadlock.
1050          */
1051         if (!inode->i_nlink &&
1052             unlikely(current->flags & PF_MEMALLOC) &&
1053             gfs2_holder_initialized(&ip->i_iopen_gh)) {
1054                 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1055
1056                 gfs2_glock_hold(gl);
1057                 if (!gfs2_queue_try_to_evict(gl))
1058                         gfs2_glock_queue_put(gl);
1059                 return 0;
1060         }
1061
1062         /*
1063          * No longer cache inodes when trying to evict them all.
1064          */
1065         if (test_bit(SDF_EVICTING, &sdp->sd_flags))
1066                 return 1;
1067
1068         return generic_drop_inode(inode);
1069 }
1070
1071 static int is_ancestor(const struct dentry *d1, const struct dentry *d2)
1072 {
1073         do {
1074                 if (d1 == d2)
1075                         return 1;
1076                 d1 = d1->d_parent;
1077         } while (!IS_ROOT(d1));
1078         return 0;
1079 }
1080
1081 /**
1082  * gfs2_show_options - Show mount options for /proc/mounts
1083  * @s: seq_file structure
1084  * @root: root of this (sub)tree
1085  *
1086  * Returns: 0 on success or error code
1087  */
1088
1089 static int gfs2_show_options(struct seq_file *s, struct dentry *root)
1090 {
1091         struct gfs2_sbd *sdp = root->d_sb->s_fs_info;
1092         struct gfs2_args *args = &sdp->sd_args;
1093         unsigned int logd_secs, statfs_slow, statfs_quantum, quota_quantum;
1094
1095         spin_lock(&sdp->sd_tune.gt_spin);
1096         logd_secs = sdp->sd_tune.gt_logd_secs;
1097         quota_quantum = sdp->sd_tune.gt_quota_quantum;
1098         statfs_quantum = sdp->sd_tune.gt_statfs_quantum;
1099         statfs_slow = sdp->sd_tune.gt_statfs_slow;
1100         spin_unlock(&sdp->sd_tune.gt_spin);
1101
1102         if (is_ancestor(root, sdp->sd_master_dir))
1103                 seq_puts(s, ",meta");
1104         if (args->ar_lockproto[0])
1105                 seq_show_option(s, "lockproto", args->ar_lockproto);
1106         if (args->ar_locktable[0])
1107                 seq_show_option(s, "locktable", args->ar_locktable);
1108         if (args->ar_hostdata[0])
1109                 seq_show_option(s, "hostdata", args->ar_hostdata);
1110         if (args->ar_spectator)
1111                 seq_puts(s, ",spectator");
1112         if (args->ar_localflocks)
1113                 seq_puts(s, ",localflocks");
1114         if (args->ar_debug)
1115                 seq_puts(s, ",debug");
1116         if (args->ar_posix_acl)
1117                 seq_puts(s, ",acl");
1118         if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
1119                 char *state;
1120                 switch (args->ar_quota) {
1121                 case GFS2_QUOTA_OFF:
1122                         state = "off";
1123                         break;
1124                 case GFS2_QUOTA_ACCOUNT:
1125                         state = "account";
1126                         break;
1127                 case GFS2_QUOTA_ON:
1128                         state = "on";
1129                         break;
1130                 default:
1131                         state = "unknown";
1132                         break;
1133                 }
1134                 seq_printf(s, ",quota=%s", state);
1135         }
1136         if (args->ar_suiddir)
1137                 seq_puts(s, ",suiddir");
1138         if (args->ar_data != GFS2_DATA_DEFAULT) {
1139                 char *state;
1140                 switch (args->ar_data) {
1141                 case GFS2_DATA_WRITEBACK:
1142                         state = "writeback";
1143                         break;
1144                 case GFS2_DATA_ORDERED:
1145                         state = "ordered";
1146                         break;
1147                 default:
1148                         state = "unknown";
1149                         break;
1150                 }
1151                 seq_printf(s, ",data=%s", state);
1152         }
1153         if (args->ar_discard)
1154                 seq_puts(s, ",discard");
1155         if (logd_secs != 30)
1156                 seq_printf(s, ",commit=%d", logd_secs);
1157         if (statfs_quantum != 30)
1158                 seq_printf(s, ",statfs_quantum=%d", statfs_quantum);
1159         else if (statfs_slow)
1160                 seq_puts(s, ",statfs_quantum=0");
1161         if (quota_quantum != 60)
1162                 seq_printf(s, ",quota_quantum=%d", quota_quantum);
1163         if (args->ar_statfs_percent)
1164                 seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent);
1165         if (args->ar_errors != GFS2_ERRORS_DEFAULT) {
1166                 const char *state;
1167
1168                 switch (args->ar_errors) {
1169                 case GFS2_ERRORS_WITHDRAW:
1170                         state = "withdraw";
1171                         break;
1172                 case GFS2_ERRORS_PANIC:
1173                         state = "panic";
1174                         break;
1175                 default:
1176                         state = "unknown";
1177                         break;
1178                 }
1179                 seq_printf(s, ",errors=%s", state);
1180         }
1181         if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags))
1182                 seq_puts(s, ",nobarrier");
1183         if (test_bit(SDF_DEMOTE, &sdp->sd_flags))
1184                 seq_puts(s, ",demote_interface_used");
1185         if (args->ar_rgrplvb)
1186                 seq_puts(s, ",rgrplvb");
1187         if (args->ar_loccookie)
1188                 seq_puts(s, ",loccookie");
1189         return 0;
1190 }
1191
1192 static void gfs2_final_release_pages(struct gfs2_inode *ip)
1193 {
1194         struct inode *inode = &ip->i_inode;
1195         struct gfs2_glock *gl = ip->i_gl;
1196
1197         if (unlikely(!gl)) {
1198                 /* This can only happen during incomplete inode creation. */
1199                 BUG_ON(!test_bit(GIF_ALLOC_FAILED, &ip->i_flags));
1200                 return;
1201         }
1202
1203         truncate_inode_pages(gfs2_glock2aspace(gl), 0);
1204         truncate_inode_pages(&inode->i_data, 0);
1205
1206         if (atomic_read(&gl->gl_revokes) == 0) {
1207                 clear_bit(GLF_LFLUSH, &gl->gl_flags);
1208                 clear_bit(GLF_DIRTY, &gl->gl_flags);
1209         }
1210 }
1211
1212 static int gfs2_dinode_dealloc(struct gfs2_inode *ip)
1213 {
1214         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1215         struct gfs2_rgrpd *rgd;
1216         struct gfs2_holder gh;
1217         int error;
1218
1219         if (gfs2_get_inode_blocks(&ip->i_inode) != 1) {
1220                 gfs2_consist_inode(ip);
1221                 return -EIO;
1222         }
1223
1224         gfs2_rindex_update(sdp);
1225
1226         error = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
1227         if (error)
1228                 return error;
1229
1230         rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1231         if (!rgd) {
1232                 gfs2_consist_inode(ip);
1233                 error = -EIO;
1234                 goto out_qs;
1235         }
1236
1237         error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE,
1238                                    LM_FLAG_NODE_SCOPE, &gh);
1239         if (error)
1240                 goto out_qs;
1241
1242         error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA,
1243                                  sdp->sd_jdesc->jd_blocks);
1244         if (error)
1245                 goto out_rg_gunlock;
1246
1247         gfs2_free_di(rgd, ip);
1248
1249         gfs2_final_release_pages(ip);
1250
1251         gfs2_trans_end(sdp);
1252
1253 out_rg_gunlock:
1254         gfs2_glock_dq_uninit(&gh);
1255 out_qs:
1256         gfs2_quota_unhold(ip);
1257         return error;
1258 }
1259
1260 /**
1261  * gfs2_glock_put_eventually
1262  * @gl: The glock to put
1263  *
1264  * When under memory pressure, trigger a deferred glock put to make sure we
1265  * won't call into DLM and deadlock.  Otherwise, put the glock directly.
1266  */
1267
1268 static void gfs2_glock_put_eventually(struct gfs2_glock *gl)
1269 {
1270         if (current->flags & PF_MEMALLOC)
1271                 gfs2_glock_queue_put(gl);
1272         else
1273                 gfs2_glock_put(gl);
1274 }
1275
1276 static bool gfs2_upgrade_iopen_glock(struct inode *inode)
1277 {
1278         struct gfs2_inode *ip = GFS2_I(inode);
1279         struct gfs2_sbd *sdp = GFS2_SB(inode);
1280         struct gfs2_holder *gh = &ip->i_iopen_gh;
1281         long timeout = 5 * HZ;
1282         int error;
1283
1284         gh->gh_flags |= GL_NOCACHE;
1285         gfs2_glock_dq_wait(gh);
1286
1287         /*
1288          * If there are no other lock holders, we will immediately get
1289          * exclusive access to the iopen glock here.
1290          *
1291          * Otherwise, the other nodes holding the lock will be notified about
1292          * our locking request.  If they do not have the inode open, they are
1293          * expected to evict the cached inode and release the lock, allowing us
1294          * to proceed.
1295          *
1296          * Otherwise, if they cannot evict the inode, they are expected to poke
1297          * the inode glock (note: not the iopen glock).  We will notice that
1298          * and stop waiting for the iopen glock immediately.  The other node(s)
1299          * are then expected to take care of deleting the inode when they no
1300          * longer use it.
1301          *
1302          * As a last resort, if another node keeps holding the iopen glock
1303          * without showing any activity on the inode glock, we will eventually
1304          * time out and fail the iopen glock upgrade.
1305          *
1306          * Note that we're passing the LM_FLAG_TRY_1CB flag to the first
1307          * locking request as an optimization to notify lock holders as soon as
1308          * possible.  Without that flag, they'd be notified implicitly by the
1309          * second locking request.
1310          */
1311
1312         gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE, gh);
1313         error = gfs2_glock_nq(gh);
1314         if (error != GLR_TRYFAILED)
1315                 return !error;
1316
1317         gfs2_holder_reinit(LM_ST_EXCLUSIVE, GL_ASYNC | GL_NOCACHE, gh);
1318         error = gfs2_glock_nq(gh);
1319         if (error)
1320                 return false;
1321
1322         timeout = wait_event_interruptible_timeout(sdp->sd_async_glock_wait,
1323                 !test_bit(HIF_WAIT, &gh->gh_iflags) ||
1324                 test_bit(GLF_DEMOTE, &ip->i_gl->gl_flags),
1325                 timeout);
1326         if (!test_bit(HIF_HOLDER, &gh->gh_iflags)) {
1327                 gfs2_glock_dq(gh);
1328                 return false;
1329         }
1330         return gfs2_glock_holder_ready(gh) == 0;
1331 }
1332
1333 /**
1334  * evict_should_delete - determine whether the inode is eligible for deletion
1335  * @inode: The inode to evict
1336  * @gh: The glock holder structure
1337  *
1338  * This function determines whether the evicted inode is eligible to be deleted
1339  * and locks the inode glock.
1340  *
1341  * Returns: the fate of the dinode
1342  */
1343 static enum dinode_demise evict_should_delete(struct inode *inode,
1344                                               struct gfs2_holder *gh)
1345 {
1346         struct gfs2_inode *ip = GFS2_I(inode);
1347         struct super_block *sb = inode->i_sb;
1348         struct gfs2_sbd *sdp = sb->s_fs_info;
1349         int ret;
1350
1351         if (unlikely(test_bit(GIF_ALLOC_FAILED, &ip->i_flags)))
1352                 goto should_delete;
1353
1354         if (test_bit(GIF_DEFERRED_DELETE, &ip->i_flags))
1355                 return SHOULD_DEFER_EVICTION;
1356
1357         /* Deletes should never happen under memory pressure anymore.  */
1358         if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
1359                 return SHOULD_DEFER_EVICTION;
1360
1361         /* Must not read inode block until block type has been verified */
1362         ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, gh);
1363         if (unlikely(ret)) {
1364                 glock_clear_object(ip->i_iopen_gh.gh_gl, ip);
1365                 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1366                 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1367                 return SHOULD_DEFER_EVICTION;
1368         }
1369
1370         if (gfs2_inode_already_deleted(ip->i_gl, ip->i_no_formal_ino))
1371                 return SHOULD_NOT_DELETE_DINODE;
1372         ret = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED);
1373         if (ret)
1374                 return SHOULD_NOT_DELETE_DINODE;
1375
1376         ret = gfs2_instantiate(gh);
1377         if (ret)
1378                 return SHOULD_NOT_DELETE_DINODE;
1379
1380         /*
1381          * The inode may have been recreated in the meantime.
1382          */
1383         if (inode->i_nlink)
1384                 return SHOULD_NOT_DELETE_DINODE;
1385
1386 should_delete:
1387         if (gfs2_holder_initialized(&ip->i_iopen_gh) &&
1388             test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1389                 if (!gfs2_upgrade_iopen_glock(inode)) {
1390                         gfs2_holder_uninit(&ip->i_iopen_gh);
1391                         return SHOULD_NOT_DELETE_DINODE;
1392                 }
1393         }
1394         return SHOULD_DELETE_DINODE;
1395 }
1396
1397 /**
1398  * evict_unlinked_inode - delete the pieces of an unlinked evicted inode
1399  * @inode: The inode to evict
1400  */
1401 static int evict_unlinked_inode(struct inode *inode)
1402 {
1403         struct gfs2_inode *ip = GFS2_I(inode);
1404         int ret;
1405
1406         if (S_ISDIR(inode->i_mode) &&
1407             (ip->i_diskflags & GFS2_DIF_EXHASH)) {
1408                 ret = gfs2_dir_exhash_dealloc(ip);
1409                 if (ret)
1410                         goto out;
1411         }
1412
1413         if (ip->i_eattr) {
1414                 ret = gfs2_ea_dealloc(ip);
1415                 if (ret)
1416                         goto out;
1417         }
1418
1419         if (!gfs2_is_stuffed(ip)) {
1420                 ret = gfs2_file_dealloc(ip);
1421                 if (ret)
1422                         goto out;
1423         }
1424
1425         /*
1426          * As soon as we clear the bitmap for the dinode, gfs2_create_inode()
1427          * can get called to recreate it, or even gfs2_inode_lookup() if the
1428          * inode was recreated on another node in the meantime.
1429          *
1430          * However, inserting the new inode into the inode hash table will not
1431          * succeed until the old inode is removed, and that only happens after
1432          * ->evict_inode() returns.  The new inode is attached to its inode and
1433          *  iopen glocks after inserting it into the inode hash table, so at
1434          *  that point we can be sure that both glocks are unused.
1435          */
1436
1437         ret = gfs2_dinode_dealloc(ip);
1438         if (!ret && ip->i_gl)
1439                 gfs2_inode_remember_delete(ip->i_gl, ip->i_no_formal_ino);
1440
1441 out:
1442         return ret;
1443 }
1444
1445 /*
1446  * evict_linked_inode - evict an inode whose dinode has not been unlinked
1447  * @inode: The inode to evict
1448  */
1449 static int evict_linked_inode(struct inode *inode)
1450 {
1451         struct super_block *sb = inode->i_sb;
1452         struct gfs2_sbd *sdp = sb->s_fs_info;
1453         struct gfs2_inode *ip = GFS2_I(inode);
1454         struct address_space *metamapping;
1455         int ret;
1456
1457         gfs2_log_flush(sdp, ip->i_gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
1458                        GFS2_LFC_EVICT_INODE);
1459         metamapping = gfs2_glock2aspace(ip->i_gl);
1460         if (test_bit(GLF_DIRTY, &ip->i_gl->gl_flags)) {
1461                 filemap_fdatawrite(metamapping);
1462                 filemap_fdatawait(metamapping);
1463         }
1464         write_inode_now(inode, 1);
1465         gfs2_ail_flush(ip->i_gl, 0);
1466
1467         ret = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
1468         if (ret)
1469                 return ret;
1470
1471         /* Needs to be done before glock release & also in a transaction */
1472         truncate_inode_pages(&inode->i_data, 0);
1473         truncate_inode_pages(metamapping, 0);
1474         gfs2_trans_end(sdp);
1475         return 0;
1476 }
1477
1478 /**
1479  * gfs2_evict_inode - Remove an inode from cache
1480  * @inode: The inode to evict
1481  *
1482  * There are three cases to consider:
1483  * 1. i_nlink == 0, we are final opener (and must deallocate)
1484  * 2. i_nlink == 0, we are not the final opener (and cannot deallocate)
1485  * 3. i_nlink > 0
1486  *
1487  * If the fs is read only, then we have to treat all cases as per #3
1488  * since we are unable to do any deallocation. The inode will be
1489  * deallocated by the next read/write node to attempt an allocation
1490  * in the same resource group
1491  *
1492  * We have to (at the moment) hold the inodes main lock to cover
1493  * the gap between unlocking the shared lock on the iopen lock and
1494  * taking the exclusive lock. I'd rather do a shared -> exclusive
1495  * conversion on the iopen lock, but we can change that later. This
1496  * is safe, just less efficient.
1497  */
1498
1499 static void gfs2_evict_inode(struct inode *inode)
1500 {
1501         struct super_block *sb = inode->i_sb;
1502         struct gfs2_sbd *sdp = sb->s_fs_info;
1503         struct gfs2_inode *ip = GFS2_I(inode);
1504         struct gfs2_holder gh;
1505         int ret;
1506
1507         if (inode->i_nlink || sb_rdonly(sb) || !ip->i_no_addr)
1508                 goto out;
1509
1510         /*
1511          * In case of an incomplete mount, gfs2_evict_inode() may be called for
1512          * system files without having an active journal to write to.  In that
1513          * case, skip the filesystem evict.
1514          */
1515         if (!sdp->sd_jdesc)
1516                 goto out;
1517
1518         gfs2_holder_mark_uninitialized(&gh);
1519         ret = evict_should_delete(inode, &gh);
1520         if (ret == SHOULD_DEFER_EVICTION)
1521                 goto out;
1522         if (ret == SHOULD_DELETE_DINODE)
1523                 ret = evict_unlinked_inode(inode);
1524         else
1525                 ret = evict_linked_inode(inode);
1526
1527         if (gfs2_rs_active(&ip->i_res))
1528                 gfs2_rs_deltree(&ip->i_res);
1529
1530         if (gfs2_holder_initialized(&gh))
1531                 gfs2_glock_dq_uninit(&gh);
1532         if (ret && ret != GLR_TRYFAILED && ret != -EROFS)
1533                 fs_warn(sdp, "gfs2_evict_inode: %d\n", ret);
1534 out:
1535         truncate_inode_pages_final(&inode->i_data);
1536         if (ip->i_qadata)
1537                 gfs2_assert_warn(sdp, ip->i_qadata->qa_ref == 0);
1538         gfs2_rs_deltree(&ip->i_res);
1539         gfs2_ordered_del_inode(ip);
1540         clear_inode(inode);
1541         gfs2_dir_hash_inval(ip);
1542         if (gfs2_holder_initialized(&ip->i_iopen_gh)) {
1543                 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1544
1545                 glock_clear_object(gl, ip);
1546                 gfs2_glock_hold(gl);
1547                 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1548                 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1549                 gfs2_glock_put_eventually(gl);
1550         }
1551         if (ip->i_gl) {
1552                 glock_clear_object(ip->i_gl, ip);
1553                 wait_on_bit_io(&ip->i_flags, GIF_GLOP_PENDING, TASK_UNINTERRUPTIBLE);
1554                 gfs2_glock_add_to_lru(ip->i_gl);
1555                 gfs2_glock_put_eventually(ip->i_gl);
1556                 ip->i_gl = NULL;
1557         }
1558 }
1559
1560 static struct inode *gfs2_alloc_inode(struct super_block *sb)
1561 {
1562         struct gfs2_inode *ip;
1563
1564         ip = alloc_inode_sb(sb, gfs2_inode_cachep, GFP_KERNEL);
1565         if (!ip)
1566                 return NULL;
1567         ip->i_no_addr = 0;
1568         ip->i_flags = 0;
1569         ip->i_gl = NULL;
1570         gfs2_holder_mark_uninitialized(&ip->i_iopen_gh);
1571         memset(&ip->i_res, 0, sizeof(ip->i_res));
1572         RB_CLEAR_NODE(&ip->i_res.rs_node);
1573         ip->i_rahead = 0;
1574         return &ip->i_inode;
1575 }
1576
1577 static void gfs2_free_inode(struct inode *inode)
1578 {
1579         kmem_cache_free(gfs2_inode_cachep, GFS2_I(inode));
1580 }
1581
1582 extern void free_local_statfs_inodes(struct gfs2_sbd *sdp)
1583 {
1584         struct local_statfs_inode *lsi, *safe;
1585
1586         /* Run through the statfs inodes list to iput and free memory */
1587         list_for_each_entry_safe(lsi, safe, &sdp->sd_sc_inodes_list, si_list) {
1588                 if (lsi->si_jid == sdp->sd_jdesc->jd_jid)
1589                         sdp->sd_sc_inode = NULL; /* belongs to this node */
1590                 if (lsi->si_sc_inode)
1591                         iput(lsi->si_sc_inode);
1592                 list_del(&lsi->si_list);
1593                 kfree(lsi);
1594         }
1595 }
1596
1597 extern struct inode *find_local_statfs_inode(struct gfs2_sbd *sdp,
1598                                              unsigned int index)
1599 {
1600         struct local_statfs_inode *lsi;
1601
1602         /* Return the local (per node) statfs inode in the
1603          * sdp->sd_sc_inodes_list corresponding to the 'index'. */
1604         list_for_each_entry(lsi, &sdp->sd_sc_inodes_list, si_list) {
1605                 if (lsi->si_jid == index)
1606                         return lsi->si_sc_inode;
1607         }
1608         return NULL;
1609 }
1610
1611 const struct super_operations gfs2_super_ops = {
1612         .alloc_inode            = gfs2_alloc_inode,
1613         .free_inode             = gfs2_free_inode,
1614         .write_inode            = gfs2_write_inode,
1615         .dirty_inode            = gfs2_dirty_inode,
1616         .evict_inode            = gfs2_evict_inode,
1617         .put_super              = gfs2_put_super,
1618         .sync_fs                = gfs2_sync_fs,
1619         .freeze_super           = gfs2_freeze_super,
1620         .thaw_super             = gfs2_thaw_super,
1621         .statfs                 = gfs2_statfs,
1622         .drop_inode             = gfs2_drop_inode,
1623         .show_options           = gfs2_show_options,
1624 };
1625