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 evict_behavior {
  48         EVICT_SHOULD_DELETE,
  49         EVICT_SHOULD_SKIP_DELETE,
  50         EVICT_SHOULD_DEFER_DELETE,
  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         down_write(&sdp->sd_log_flush_lock);
  71         sdp->sd_jdesc = NULL;
  72         up_write(&sdp->sd_log_flush_lock);
  73
  74         while (!list_empty(&list)) {
  75                 jd = list_first_entry(&list, struct gfs2_jdesc, jd_list);
  76                 BUG_ON(jd->jd_log_bio);
  77                 gfs2_free_journal_extents(jd);
  78                 list_del(&jd->jd_list);
  79                 iput(jd->jd_inode);
  80                 jd->jd_inode = NULL;
  81                 kfree(jd);
  82         }
  83 }
  84
  85 static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
  86 {
  87         struct gfs2_jdesc *jd;
  88
  89         list_for_each_entry(jd, head, jd_list) {
  90                 if (jd->jd_jid == jid)
  91                         return jd;
  92         }
  93         return NULL;
  94 }
  95
  96 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
  97 {
  98         struct gfs2_jdesc *jd;
  99
 100         spin_lock(&sdp->sd_jindex_spin);
 101         jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
 102         spin_unlock(&sdp->sd_jindex_spin);
 103
 104         return jd;
 105 }
 106
 107 int gfs2_jdesc_check(struct gfs2_jdesc *jd)
 108 {
 109         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 110         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 111         u64 size = i_size_read(jd->jd_inode);
 112
 113         if (gfs2_check_internal_file_size(jd->jd_inode, 8 << 20, BIT(30)))
 114                 return -EIO;
 115
 116         jd->jd_blocks = size >> sdp->sd_sb.sb_bsize_shift;
 117
 118         if (gfs2_write_alloc_required(ip, 0, size)) {
 119                 gfs2_consist_inode(ip);
 120                 return -EIO;
 121         }
 122
 123         return 0;
 124 }
 125
 126 /**
 127  * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
 128  * @sdp: the filesystem
 129  *
 130  * Returns: errno
 131  */
 132
 133 int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
 134 {
 135         struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
 136         struct gfs2_glock *j_gl = ip->i_gl;
 137         int error;
 138
 139         j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
 140         if (gfs2_withdrawing_or_withdrawn(sdp))
 141                 return -EIO;
 142
 143         if (sdp->sd_log_sequence == 0) {
 144                 fs_err(sdp, "unknown status of our own journal jid %d",
 145                        sdp->sd_lockstruct.ls_jid);
 146                 return -EIO;
 147         }
 148
 149         error = gfs2_quota_init(sdp);
 150         if (!error && gfs2_withdrawing_or_withdrawn(sdp))
 151                 error = -EIO;
 152         if (!error)
 153                 set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 154         return error;
 155 }
 156
 157 void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
 158 {
 159         const struct gfs2_statfs_change *str = buf;
 160
 161         sc->sc_total = be64_to_cpu(str->sc_total);
 162         sc->sc_free = be64_to_cpu(str->sc_free);
 163         sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
 164 }
 165
 166 void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
 167 {
 168         struct gfs2_statfs_change *str = buf;
 169
 170         str->sc_total = cpu_to_be64(sc->sc_total);
 171         str->sc_free = cpu_to_be64(sc->sc_free);
 172         str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
 173 }
 174
 175 int gfs2_statfs_init(struct gfs2_sbd *sdp)
 176 {
 177         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 178         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 179         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 180         struct buffer_head *m_bh;
 181         struct gfs2_holder gh;
 182         int error;
 183
 184         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 185                                    &gh);
 186         if (error)
 187                 return error;
 188
 189         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 190         if (error)
 191                 goto out;
 192
 193         if (sdp->sd_args.ar_spectator) {
 194                 spin_lock(&sdp->sd_statfs_spin);
 195                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 196                                       sizeof(struct gfs2_dinode));
 197                 spin_unlock(&sdp->sd_statfs_spin);
 198         } else {
 199                 spin_lock(&sdp->sd_statfs_spin);
 200                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 201                                       sizeof(struct gfs2_dinode));
 202                 gfs2_statfs_change_in(l_sc, sdp->sd_sc_bh->b_data +
 203                                       sizeof(struct gfs2_dinode));
 204                 spin_unlock(&sdp->sd_statfs_spin);
 205
 206         }
 207
 208         brelse(m_bh);
 209 out:
 210         gfs2_glock_dq_uninit(&gh);
 211         return 0;
 212 }
 213
 214 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
 215                         s64 dinodes)
 216 {
 217         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 218         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 219         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 220         s64 x, y;
 221         int need_sync = 0;
 222
 223         gfs2_trans_add_meta(l_ip->i_gl, sdp->sd_sc_bh);
 224
 225         spin_lock(&sdp->sd_statfs_spin);
 226         l_sc->sc_total += total;
 227         l_sc->sc_free += free;
 228         l_sc->sc_dinodes += dinodes;
 229         gfs2_statfs_change_out(l_sc, sdp->sd_sc_bh->b_data +
 230                                sizeof(struct gfs2_dinode));
 231         if (sdp->sd_args.ar_statfs_percent) {
 232                 x = 100 * l_sc->sc_free;
 233                 y = m_sc->sc_free * sdp->sd_args.ar_statfs_percent;
 234                 if (x >= y || x <= -y)
 235                         need_sync = 1;
 236         }
 237         spin_unlock(&sdp->sd_statfs_spin);
 238
 239         if (need_sync)
 240                 gfs2_wake_up_statfs(sdp);
 241 }
 242
 243 void update_statfs(struct gfs2_sbd *sdp, struct buffer_head *m_bh)
 244 {
 245         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 246         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 247         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 248         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 249
 250         gfs2_trans_add_meta(l_ip->i_gl, sdp->sd_sc_bh);
 251         gfs2_trans_add_meta(m_ip->i_gl, m_bh);
 252
 253         spin_lock(&sdp->sd_statfs_spin);
 254         m_sc->sc_total += l_sc->sc_total;
 255         m_sc->sc_free += l_sc->sc_free;
 256         m_sc->sc_dinodes += l_sc->sc_dinodes;
 257         memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
 258         memset(sdp->sd_sc_bh->b_data + sizeof(struct gfs2_dinode),
 259                0, sizeof(struct gfs2_statfs_change));
 260         gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
 261         spin_unlock(&sdp->sd_statfs_spin);
 262 }
 263
 264 int gfs2_statfs_sync(struct super_block *sb, int type)
 265 {
 266         struct gfs2_sbd *sdp = sb->s_fs_info;
 267         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 268         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 269         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 270         struct gfs2_holder gh;
 271         struct buffer_head *m_bh;
 272         int error;
 273
 274         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 275                                    &gh);
 276         if (error)
 277                 goto out;
 278
 279         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 280         if (error)
 281                 goto out_unlock;
 282
 283         spin_lock(&sdp->sd_statfs_spin);
 284         gfs2_statfs_change_in(m_sc, m_bh->b_data +
 285                               sizeof(struct gfs2_dinode));
 286         if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
 287                 spin_unlock(&sdp->sd_statfs_spin);
 288                 goto out_bh;
 289         }
 290         spin_unlock(&sdp->sd_statfs_spin);
 291
 292         error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
 293         if (error)
 294                 goto out_bh;
 295
 296         update_statfs(sdp, m_bh);
 297         sdp->sd_statfs_force_sync = 0;
 298
 299         gfs2_trans_end(sdp);
 300
 301 out_bh:
 302         brelse(m_bh);
 303 out_unlock:
 304         gfs2_glock_dq_uninit(&gh);
 305 out:
 306         return error;
 307 }
 308
 309 struct lfcc {
 310         struct list_head list;
 311         struct gfs2_holder gh;
 312 };
 313
 314 /**
 315  * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
 316  *                            journals are clean
 317  * @sdp: the file system
 318  *
 319  * Returns: errno
 320  */
 321
 322 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp)
 323 {
 324         struct gfs2_inode *ip;
 325         struct gfs2_jdesc *jd;
 326         struct lfcc *lfcc;
 327         LIST_HEAD(list);
 328         struct gfs2_log_header_host lh;
 329         int error, error2;
 330
 331         /*
 332          * Grab all the journal glocks in SH mode.  We are *probably* doing
 333          * that to prevent recovery.
 334          */
 335
 336         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 337                 lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
 338                 if (!lfcc) {
 339                         error = -ENOMEM;
 340                         goto out;
 341                 }
 342                 ip = GFS2_I(jd->jd_inode);
 343                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
 344                 if (error) {
 345                         kfree(lfcc);
 346                         goto out;
 347                 }
 348                 list_add(&lfcc->list, &list);
 349         }
 350
 351         gfs2_freeze_unlock(sdp);
 352
 353         error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_EXCLUSIVE,
 354                                    LM_FLAG_NOEXP | GL_NOPID,
 355                                    &sdp->sd_freeze_gh);
 356         if (error)
 357                 goto relock_shared;
 358
 359         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 360                 error = gfs2_jdesc_check(jd);
 361                 if (error)
 362                         break;
 363                 error = gfs2_find_jhead(jd, &lh);
 364                 if (error)
 365                         break;
 366                 if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 367                         error = -EBUSY;
 368                         break;
 369                 }
 370         }
 371
 372         if (!error)
 373                 goto out;  /* success */
 374
 375         gfs2_freeze_unlock(sdp);
 376
 377 relock_shared:
 378         error2 = gfs2_freeze_lock_shared(sdp);
 379         gfs2_assert_withdraw(sdp, !error2);
 380
 381 out:
 382         while (!list_empty(&list)) {
 383                 lfcc = list_first_entry(&list, struct lfcc, list);
 384                 list_del(&lfcc->list);
 385                 gfs2_glock_dq_uninit(&lfcc->gh);
 386                 kfree(lfcc);
 387         }
 388         return error;
 389 }
 390
 391 void gfs2_dinode_out(const struct gfs2_inode *ip, void *buf)
 392 {
 393         const struct inode *inode = &ip->i_inode;
 394         struct gfs2_dinode *str = buf;
 395
 396         str->di_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 397         str->di_header.mh_type = cpu_to_be32(GFS2_METATYPE_DI);
 398         str->di_header.mh_format = cpu_to_be32(GFS2_FORMAT_DI);
 399         str->di_num.no_addr = cpu_to_be64(ip->i_no_addr);
 400         str->di_num.no_formal_ino = cpu_to_be64(ip->i_no_formal_ino);
 401         str->di_mode = cpu_to_be32(inode->i_mode);
 402         str->di_uid = cpu_to_be32(i_uid_read(inode));
 403         str->di_gid = cpu_to_be32(i_gid_read(inode));
 404         str->di_nlink = cpu_to_be32(inode->i_nlink);
 405         str->di_size = cpu_to_be64(i_size_read(inode));
 406         str->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(inode));
 407         str->di_atime = cpu_to_be64(inode_get_atime_sec(inode));
 408         str->di_mtime = cpu_to_be64(inode_get_mtime_sec(inode));
 409         str->di_ctime = cpu_to_be64(inode_get_ctime_sec(inode));
 410
 411         str->di_goal_meta = cpu_to_be64(ip->i_goal);
 412         str->di_goal_data = cpu_to_be64(ip->i_goal);
 413         str->di_generation = cpu_to_be64(ip->i_generation);
 414
 415         str->di_flags = cpu_to_be32(ip->i_diskflags);
 416         str->di_height = cpu_to_be16(ip->i_height);
 417         str->di_payload_format = cpu_to_be32(S_ISDIR(inode->i_mode) &&
 418                                              !(ip->i_diskflags & GFS2_DIF_EXHASH) ?
 419                                              GFS2_FORMAT_DE : 0);
 420         str->di_depth = cpu_to_be16(ip->i_depth);
 421         str->di_entries = cpu_to_be32(ip->i_entries);
 422
 423         str->di_eattr = cpu_to_be64(ip->i_eattr);
 424         str->di_atime_nsec = cpu_to_be32(inode_get_atime_nsec(inode));
 425         str->di_mtime_nsec = cpu_to_be32(inode_get_mtime_nsec(inode));
 426         str->di_ctime_nsec = cpu_to_be32(inode_get_ctime_nsec(inode));
 427 }
 428
 429 /**
 430  * gfs2_write_inode - Make sure the inode is stable on the disk
 431  * @inode: The inode
 432  * @wbc: The writeback control structure
 433  *
 434  * Returns: errno
 435  */
 436
 437 static int gfs2_write_inode(struct inode *inode, struct writeback_control *wbc)
 438 {
 439         struct gfs2_inode *ip = GFS2_I(inode);
 440         struct gfs2_sbd *sdp = GFS2_SB(inode);
 441         struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl);
 442         struct backing_dev_info *bdi = inode_to_bdi(metamapping->host);
 443         int ret = 0;
 444         bool flush_all = (wbc->sync_mode == WB_SYNC_ALL || gfs2_is_jdata(ip));
 445
 446         if (flush_all)
 447                 gfs2_log_flush(GFS2_SB(inode), ip->i_gl,
 448                                GFS2_LOG_HEAD_FLUSH_NORMAL |
 449                                GFS2_LFC_WRITE_INODE);
 450         if (bdi->wb.dirty_exceeded)
 451                 gfs2_ail1_flush(sdp, wbc);
 452         else
 453                 filemap_fdatawrite(metamapping);
 454         if (flush_all)
 455                 ret = filemap_fdatawait(metamapping);
 456         if (ret)
 457                 mark_inode_dirty_sync(inode);
 458         else {
 459                 spin_lock(&inode->i_lock);
 460                 if (!(inode->i_flags & I_DIRTY))
 461                         gfs2_ordered_del_inode(ip);
 462                 spin_unlock(&inode->i_lock);
 463         }
 464         return ret;
 465 }
 466
 467 /**
 468  * gfs2_dirty_inode - check for atime updates
 469  * @inode: The inode in question
 470  * @flags: The type of dirty
 471  *
 472  * Unfortunately it can be called under any combination of inode
 473  * glock and freeze glock, so we have to check carefully.
 474  *
 475  * At the moment this deals only with atime - it should be possible
 476  * to expand that role in future, once a review of the locking has
 477  * been carried out.
 478  */
 479
 480 static void gfs2_dirty_inode(struct inode *inode, int flags)
 481 {
 482         struct gfs2_inode *ip = GFS2_I(inode);
 483         struct gfs2_sbd *sdp = GFS2_SB(inode);
 484         struct buffer_head *bh;
 485         struct gfs2_holder gh;
 486         int need_unlock = 0;
 487         int need_endtrans = 0;
 488         int ret;
 489
 490         if (unlikely(!ip->i_gl)) {
 491                 /* This can only happen during incomplete inode creation. */
 492                 BUG_ON(!test_bit(GIF_ALLOC_FAILED, &ip->i_flags));
 493                 return;
 494         }
 495
 496         if (gfs2_withdrawing_or_withdrawn(sdp))
 497                 return;
 498         if (!gfs2_glock_is_locked_by_me(ip->i_gl)) {
 499                 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
 500                 if (ret) {
 501                         fs_err(sdp, "dirty_inode: glock %d\n", ret);
 502                         gfs2_dump_glock(NULL, ip->i_gl, true);
 503                         return;
 504                 }
 505                 need_unlock = 1;
 506         } else if (WARN_ON_ONCE(ip->i_gl->gl_state != LM_ST_EXCLUSIVE))
 507                 return;
 508
 509         if (current->journal_info == NULL) {
 510                 ret = gfs2_trans_begin(sdp, RES_DINODE, 0);
 511                 if (ret) {
 512                         fs_err(sdp, "dirty_inode: gfs2_trans_begin %d\n", ret);
 513                         goto out;
 514                 }
 515                 need_endtrans = 1;
 516         }
 517
 518         ret = gfs2_meta_inode_buffer(ip, &bh);
 519         if (ret == 0) {
 520                 gfs2_trans_add_meta(ip->i_gl, bh);
 521                 gfs2_dinode_out(ip, bh->b_data);
 522                 brelse(bh);
 523         }
 524
 525         if (need_endtrans)
 526                 gfs2_trans_end(sdp);
 527 out:
 528         if (need_unlock)
 529                 gfs2_glock_dq_uninit(&gh);
 530 }
 531
 532 /**
 533  * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
 534  * @sdp: the filesystem
 535  *
 536  * Returns: errno
 537  */
 538
 539 void gfs2_make_fs_ro(struct gfs2_sbd *sdp)
 540 {
 541         int log_write_allowed = test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 542
 543         if (!test_bit(SDF_KILL, &sdp->sd_flags))
 544                 gfs2_flush_delete_work(sdp);
 545
 546         gfs2_destroy_threads(sdp);
 547
 548         if (log_write_allowed) {
 549                 gfs2_quota_sync(sdp->sd_vfs, 0);
 550                 gfs2_statfs_sync(sdp->sd_vfs, 0);
 551
 552                 /* We do two log flushes here. The first one commits dirty inodes
 553                  * and rgrps to the journal, but queues up revokes to the ail list.
 554                  * The second flush writes out and removes the revokes.
 555                  *
 556                  * The first must be done before the FLUSH_SHUTDOWN code
 557                  * clears the LIVE flag, otherwise it will not be able to start
 558                  * a transaction to write its revokes, and the error will cause
 559                  * a withdraw of the file system. */
 560                 gfs2_log_flush(sdp, NULL, GFS2_LFC_MAKE_FS_RO);
 561                 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SHUTDOWN |
 562                                GFS2_LFC_MAKE_FS_RO);
 563                 wait_event_timeout(sdp->sd_log_waitq,
 564                                    gfs2_log_is_empty(sdp),
 565                                    HZ * 5);
 566                 gfs2_assert_warn(sdp, gfs2_log_is_empty(sdp));
 567         }
 568         gfs2_quota_cleanup(sdp);
 569 }
 570
 571 /**
 572  * gfs2_put_super - Unmount the filesystem
 573  * @sb: The VFS superblock
 574  *
 575  */
 576
 577 static void gfs2_put_super(struct super_block *sb)
 578 {
 579         struct gfs2_sbd *sdp = sb->s_fs_info;
 580         struct gfs2_jdesc *jd;
 581
 582         /* No more recovery requests */
 583         set_bit(SDF_NORECOVERY, &sdp->sd_flags);
 584         smp_mb();
 585
 586         /* Wait on outstanding recovery */
 587 restart:
 588         spin_lock(&sdp->sd_jindex_spin);
 589         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 590                 if (!test_bit(JDF_RECOVERY, &jd->jd_flags))
 591                         continue;
 592                 spin_unlock(&sdp->sd_jindex_spin);
 593                 wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
 594                             TASK_UNINTERRUPTIBLE);
 595                 goto restart;
 596         }
 597         spin_unlock(&sdp->sd_jindex_spin);
 598
 599         if (!sb_rdonly(sb))
 600                 gfs2_make_fs_ro(sdp);
 601         else {
 602                 if (gfs2_withdrawing_or_withdrawn(sdp))
 603                         gfs2_destroy_threads(sdp);
 604
 605                 gfs2_quota_cleanup(sdp);
 606         }
 607
 608         WARN_ON(gfs2_withdrawing(sdp));
 609
 610         /*  At this point, we're through modifying the disk  */
 611
 612         /*  Release stuff  */
 613
 614         gfs2_freeze_unlock(sdp);
 615
 616         iput(sdp->sd_jindex);
 617         iput(sdp->sd_statfs_inode);
 618         iput(sdp->sd_rindex);
 619         iput(sdp->sd_quota_inode);
 620
 621         gfs2_glock_put(sdp->sd_rename_gl);
 622         gfs2_glock_put(sdp->sd_freeze_gl);
 623
 624         if (!sdp->sd_args.ar_spectator) {
 625                 if (gfs2_holder_initialized(&sdp->sd_journal_gh))
 626                         gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
 627                 if (gfs2_holder_initialized(&sdp->sd_jinode_gh))
 628                         gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
 629                 brelse(sdp->sd_sc_bh);
 630                 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
 631                 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
 632                 free_local_statfs_inodes(sdp);
 633                 iput(sdp->sd_qc_inode);
 634         }
 635
 636         gfs2_glock_dq_uninit(&sdp->sd_live_gh);
 637         gfs2_clear_rgrpd(sdp);
 638         gfs2_jindex_free(sdp);
 639         /*  Take apart glock structures and buffer lists  */
 640         gfs2_gl_hash_clear(sdp);
 641         iput(sdp->sd_inode);
 642         gfs2_delete_debugfs_file(sdp);
 643
 644         gfs2_sys_fs_del(sdp);
 645         free_sbd(sdp);
 646 }
 647
 648 /**
 649  * gfs2_sync_fs - sync the filesystem
 650  * @sb: the superblock
 651  * @wait: true to wait for completion
 652  *
 653  * Flushes the log to disk.
 654  */
 655
 656 static int gfs2_sync_fs(struct super_block *sb, int wait)
 657 {
 658         struct gfs2_sbd *sdp = sb->s_fs_info;
 659
 660         gfs2_quota_sync(sb, -1);
 661         if (wait)
 662                 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
 663                                GFS2_LFC_SYNC_FS);
 664         return sdp->sd_log_error;
 665 }
 666
 667 static int gfs2_do_thaw(struct gfs2_sbd *sdp, enum freeze_holder who, const void *freeze_owner)
 668 {
 669         struct super_block *sb = sdp->sd_vfs;
 670         int error;
 671
 672         error = gfs2_freeze_lock_shared(sdp);
 673         if (error)
 674                 goto fail;
 675         error = thaw_super(sb, who, freeze_owner);
 676         if (!error)
 677                 return 0;
 678
 679 fail:
 680         fs_info(sdp, "GFS2: couldn't thaw filesystem: %d\n", error);
 681         gfs2_assert_withdraw(sdp, 0);
 682         return error;
 683 }
 684
 685 void gfs2_freeze_func(struct work_struct *work)
 686 {
 687         struct gfs2_sbd *sdp = container_of(work, struct gfs2_sbd, sd_freeze_work);
 688         struct super_block *sb = sdp->sd_vfs;
 689         int error;
 690
 691         mutex_lock(&sdp->sd_freeze_mutex);
 692         error = -EBUSY;
 693         if (test_bit(SDF_FROZEN, &sdp->sd_flags))
 694                 goto freeze_failed;
 695
 696         error = freeze_super(sb, FREEZE_HOLDER_USERSPACE, NULL);
 697         if (error)
 698                 goto freeze_failed;
 699
 700         gfs2_freeze_unlock(sdp);
 701         set_bit(SDF_FROZEN, &sdp->sd_flags);
 702
 703         error = gfs2_do_thaw(sdp, FREEZE_HOLDER_USERSPACE, NULL);
 704         if (error)
 705                 goto out;
 706
 707         clear_bit(SDF_FROZEN, &sdp->sd_flags);
 708         goto out;
 709
 710 freeze_failed:
 711         fs_info(sdp, "GFS2: couldn't freeze filesystem: %d\n", error);
 712
 713 out:
 714         mutex_unlock(&sdp->sd_freeze_mutex);
 715         deactivate_super(sb);
 716 }
 717
 718 /**
 719  * gfs2_freeze_super - prevent further writes to the filesystem
 720  * @sb: the VFS structure for the filesystem
 721  * @who: freeze flags
 722  * @freeze_owner: owner of the freeze
 723  *
 724  */
 725
 726 static int gfs2_freeze_super(struct super_block *sb, enum freeze_holder who,
 727                              const void *freeze_owner)
 728 {
 729         struct gfs2_sbd *sdp = sb->s_fs_info;
 730         int error;
 731
 732         if (!mutex_trylock(&sdp->sd_freeze_mutex))
 733                 return -EBUSY;
 734         if (test_bit(SDF_FROZEN, &sdp->sd_flags)) {
 735                 mutex_unlock(&sdp->sd_freeze_mutex);
 736                 return -EBUSY;
 737         }
 738
 739         for (;;) {
 740                 error = freeze_super(sb, who, freeze_owner);
 741                 if (error) {
 742                         fs_info(sdp, "GFS2: couldn't freeze filesystem: %d\n",
 743                                 error);
 744                         goto out;
 745                 }
 746
 747                 error = gfs2_lock_fs_check_clean(sdp);
 748                 if (!error) {
 749                         set_bit(SDF_FREEZE_INITIATOR, &sdp->sd_flags);
 750                         set_bit(SDF_FROZEN, &sdp->sd_flags);
 751                         break;
 752                 }
 753
 754                 error = gfs2_do_thaw(sdp, who, freeze_owner);
 755                 if (error)
 756                         goto out;
 757
 758                 if (error == -EBUSY)
 759                         fs_err(sdp, "waiting for recovery before freeze\n");
 760                 else if (error == -EIO) {
 761                         fs_err(sdp, "Fatal IO error: cannot freeze gfs2 due "
 762                                "to recovery error.\n");
 763                         goto out;
 764                 } else {
 765                         fs_err(sdp, "error freezing FS: %d\n", error);
 766                 }
 767                 fs_err(sdp, "retrying...\n");
 768                 msleep(1000);
 769         }
 770
 771 out:
 772         mutex_unlock(&sdp->sd_freeze_mutex);
 773         return error;
 774 }
 775
 776 static int gfs2_freeze_fs(struct super_block *sb)
 777 {
 778         struct gfs2_sbd *sdp = sb->s_fs_info;
 779
 780         if (test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
 781                 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_FREEZE |
 782                                GFS2_LFC_FREEZE_GO_SYNC);
 783                 if (gfs2_withdrawing_or_withdrawn(sdp))
 784                         return -EIO;
 785         }
 786         return 0;
 787 }
 788
 789 /**
 790  * gfs2_thaw_super - reallow writes to the filesystem
 791  * @sb: the VFS structure for the filesystem
 792  * @who: freeze flags
 793  * @freeze_owner: owner of the freeze
 794  *
 795  */
 796
 797 static int gfs2_thaw_super(struct super_block *sb, enum freeze_holder who,
 798                            const void *freeze_owner)
 799 {
 800         struct gfs2_sbd *sdp = sb->s_fs_info;
 801         int error;
 802
 803         if (!mutex_trylock(&sdp->sd_freeze_mutex))
 804                 return -EBUSY;
 805         if (!test_bit(SDF_FREEZE_INITIATOR, &sdp->sd_flags)) {
 806                 mutex_unlock(&sdp->sd_freeze_mutex);
 807                 return -EINVAL;
 808         }
 809
 810         atomic_inc(&sb->s_active);
 811         gfs2_freeze_unlock(sdp);
 812
 813         error = gfs2_do_thaw(sdp, who, freeze_owner);
 814
 815         if (!error) {
 816                 clear_bit(SDF_FREEZE_INITIATOR, &sdp->sd_flags);
 817                 clear_bit(SDF_FROZEN, &sdp->sd_flags);
 818         }
 819         mutex_unlock(&sdp->sd_freeze_mutex);
 820         deactivate_super(sb);
 821         return error;
 822 }
 823
 824 void gfs2_thaw_freeze_initiator(struct super_block *sb)
 825 {
 826         struct gfs2_sbd *sdp = sb->s_fs_info;
 827
 828         mutex_lock(&sdp->sd_freeze_mutex);
 829         if (!test_bit(SDF_FREEZE_INITIATOR, &sdp->sd_flags))
 830                 goto out;
 831
 832         gfs2_freeze_unlock(sdp);
 833
 834 out:
 835         mutex_unlock(&sdp->sd_freeze_mutex);
 836 }
 837
 838 /**
 839  * statfs_slow_fill - fill in the sg for a given RG
 840  * @rgd: the RG
 841  * @sc: the sc structure
 842  *
 843  * Returns: 0 on success, -ESTALE if the LVB is invalid
 844  */
 845
 846 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
 847                             struct gfs2_statfs_change_host *sc)
 848 {
 849         gfs2_rgrp_verify(rgd);
 850         sc->sc_total += rgd->rd_data;
 851         sc->sc_free += rgd->rd_free;
 852         sc->sc_dinodes += rgd->rd_dinodes;
 853         return 0;
 854 }
 855
 856 /**
 857  * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
 858  * @sdp: the filesystem
 859  * @sc: the sc info that will be returned
 860  *
 861  * Any error (other than a signal) will cause this routine to fall back
 862  * to the synchronous version.
 863  *
 864  * FIXME: This really shouldn't busy wait like this.
 865  *
 866  * Returns: errno
 867  */
 868
 869 static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
 870 {
 871         struct gfs2_rgrpd *rgd_next;
 872         struct gfs2_holder *gha, *gh;
 873         unsigned int slots = 64;
 874         unsigned int x;
 875         int done;
 876         int error = 0, err;
 877
 878         memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
 879         gha = kmalloc_array(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
 880         if (!gha)
 881                 return -ENOMEM;
 882         for (x = 0; x < slots; x++)
 883                 gfs2_holder_mark_uninitialized(gha + x);
 884
 885         rgd_next = gfs2_rgrpd_get_first(sdp);
 886
 887         for (;;) {
 888                 done = 1;
 889
 890                 for (x = 0; x < slots; x++) {
 891                         gh = gha + x;
 892
 893                         if (gfs2_holder_initialized(gh) && gfs2_glock_poll(gh)) {
 894                                 err = gfs2_glock_wait(gh);
 895                                 if (err) {
 896                                         gfs2_holder_uninit(gh);
 897                                         error = err;
 898                                 } else {
 899                                         if (!error) {
 900                                                 struct gfs2_rgrpd *rgd =
 901                                                         gfs2_glock2rgrp(gh->gh_gl);
 902
 903                                                 error = statfs_slow_fill(rgd, sc);
 904                                         }
 905                                         gfs2_glock_dq_uninit(gh);
 906                                 }
 907                         }
 908
 909                         if (gfs2_holder_initialized(gh))
 910                                 done = 0;
 911                         else if (rgd_next && !error) {
 912                                 error = gfs2_glock_nq_init(rgd_next->rd_gl,
 913                                                            LM_ST_SHARED,
 914                                                            GL_ASYNC,
 915                                                            gh);
 916                                 rgd_next = gfs2_rgrpd_get_next(rgd_next);
 917                                 done = 0;
 918                         }
 919
 920                         if (signal_pending(current))
 921                                 error = -ERESTARTSYS;
 922                 }
 923
 924                 if (done)
 925                         break;
 926
 927                 yield();
 928         }
 929
 930         kfree(gha);
 931         return error;
 932 }
 933
 934 /**
 935  * gfs2_statfs_i - Do a statfs
 936  * @sdp: the filesystem
 937  * @sc: the sc structure
 938  *
 939  * Returns: errno
 940  */
 941
 942 static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
 943 {
 944         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 945         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 946
 947         spin_lock(&sdp->sd_statfs_spin);
 948
 949         *sc = *m_sc;
 950         sc->sc_total += l_sc->sc_total;
 951         sc->sc_free += l_sc->sc_free;
 952         sc->sc_dinodes += l_sc->sc_dinodes;
 953
 954         spin_unlock(&sdp->sd_statfs_spin);
 955
 956         if (sc->sc_free < 0)
 957                 sc->sc_free = 0;
 958         if (sc->sc_free > sc->sc_total)
 959                 sc->sc_free = sc->sc_total;
 960         if (sc->sc_dinodes < 0)
 961                 sc->sc_dinodes = 0;
 962
 963         return 0;
 964 }
 965
 966 /**
 967  * gfs2_statfs - Gather and return stats about the filesystem
 968  * @dentry: The name of the link
 969  * @buf: The buffer
 970  *
 971  * Returns: 0 on success or error code
 972  */
 973
 974 static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
 975 {
 976         struct super_block *sb = dentry->d_sb;
 977         struct gfs2_sbd *sdp = sb->s_fs_info;
 978         struct gfs2_statfs_change_host sc;
 979         int error;
 980
 981         error = gfs2_rindex_update(sdp);
 982         if (error)
 983                 return error;
 984
 985         if (gfs2_tune_get(sdp, gt_statfs_slow))
 986                 error = gfs2_statfs_slow(sdp, &sc);
 987         else
 988                 error = gfs2_statfs_i(sdp, &sc);
 989
 990         if (error)
 991                 return error;
 992
 993         buf->f_type = GFS2_MAGIC;
 994         buf->f_bsize = sdp->sd_sb.sb_bsize;
 995         buf->f_blocks = sc.sc_total;
 996         buf->f_bfree = sc.sc_free;
 997         buf->f_bavail = sc.sc_free;
 998         buf->f_files = sc.sc_dinodes + sc.sc_free;
 999         buf->f_ffree = sc.sc_free;
1000         buf->f_namelen = GFS2_FNAMESIZE;
1001         buf->f_fsid = uuid_to_fsid(sb->s_uuid.b);
1002
1003         return 0;
1004 }
1005
1006 /**
1007  * gfs2_drop_inode - Drop an inode (test for remote unlink)
1008  * @inode: The inode to drop
1009  *
1010  * If we've received a callback on an iopen lock then it's because a
1011  * remote node tried to deallocate the inode but failed due to this node
1012  * still having the inode open. Here we mark the link count zero
1013  * since we know that it must have reached zero if the GLF_DEMOTE flag
1014  * is set on the iopen glock. If we didn't do a disk read since the
1015  * remote node removed the final link then we might otherwise miss
1016  * this event. This check ensures that this node will deallocate the
1017  * inode's blocks, or alternatively pass the baton on to another
1018  * node for later deallocation.
1019  */
1020
1021 static int gfs2_drop_inode(struct inode *inode)
1022 {
1023         struct gfs2_inode *ip = GFS2_I(inode);
1024         struct gfs2_sbd *sdp = GFS2_SB(inode);
1025
1026         if (inode->i_nlink &&
1027             gfs2_holder_initialized(&ip->i_iopen_gh)) {
1028                 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1029                 if (glock_needs_demote(gl))
1030                         clear_nlink(inode);
1031         }
1032
1033         /*
1034          * When under memory pressure when an inode's link count has dropped to
1035          * zero, defer deleting the inode to the delete workqueue.  This avoids
1036          * calling into DLM under memory pressure, which can deadlock.
1037          */
1038         if (!inode->i_nlink &&
1039             unlikely(current->flags & PF_MEMALLOC) &&
1040             gfs2_holder_initialized(&ip->i_iopen_gh)) {
1041                 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1042
1043                 gfs2_glock_hold(gl);
1044                 if (!gfs2_queue_verify_delete(gl, true))
1045                         gfs2_glock_put_async(gl);
1046                 return 0;
1047         }
1048
1049         /*
1050          * No longer cache inodes when trying to evict them all.
1051          */
1052         if (test_bit(SDF_EVICTING, &sdp->sd_flags))
1053                 return 1;
1054
1055         return generic_drop_inode(inode);
1056 }
1057
1058 /**
1059  * gfs2_show_options - Show mount options for /proc/mounts
1060  * @s: seq_file structure
1061  * @root: root of this (sub)tree
1062  *
1063  * Returns: 0 on success or error code
1064  */
1065
1066 static int gfs2_show_options(struct seq_file *s, struct dentry *root)
1067 {
1068         struct gfs2_sbd *sdp = root->d_sb->s_fs_info;
1069         struct gfs2_args *args = &sdp->sd_args;
1070         unsigned int logd_secs, statfs_slow, statfs_quantum, quota_quantum;
1071
1072         spin_lock(&sdp->sd_tune.gt_spin);
1073         logd_secs = sdp->sd_tune.gt_logd_secs;
1074         quota_quantum = sdp->sd_tune.gt_quota_quantum;
1075         statfs_quantum = sdp->sd_tune.gt_statfs_quantum;
1076         statfs_slow = sdp->sd_tune.gt_statfs_slow;
1077         spin_unlock(&sdp->sd_tune.gt_spin);
1078
1079         if (is_subdir(root, sdp->sd_master_dir))
1080                 seq_puts(s, ",meta");
1081         if (args->ar_lockproto[0])
1082                 seq_show_option(s, "lockproto", args->ar_lockproto);
1083         if (args->ar_locktable[0])
1084                 seq_show_option(s, "locktable", args->ar_locktable);
1085         if (args->ar_hostdata[0])
1086                 seq_show_option(s, "hostdata", args->ar_hostdata);
1087         if (args->ar_spectator)
1088                 seq_puts(s, ",spectator");
1089         if (args->ar_localflocks)
1090                 seq_puts(s, ",localflocks");
1091         if (args->ar_debug)
1092                 seq_puts(s, ",debug");
1093         if (args->ar_posix_acl)
1094                 seq_puts(s, ",acl");
1095         if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
1096                 char *state;
1097                 switch (args->ar_quota) {
1098                 case GFS2_QUOTA_OFF:
1099                         state = "off";
1100                         break;
1101                 case GFS2_QUOTA_ACCOUNT:
1102                         state = "account";
1103                         break;
1104                 case GFS2_QUOTA_ON:
1105                         state = "on";
1106                         break;
1107                 case GFS2_QUOTA_QUIET:
1108                         state = "quiet";
1109                         break;
1110                 default:
1111                         state = "unknown";
1112                         break;
1113                 }
1114                 seq_printf(s, ",quota=%s", state);
1115         }
1116         if (args->ar_suiddir)
1117                 seq_puts(s, ",suiddir");
1118         if (args->ar_data != GFS2_DATA_DEFAULT) {
1119                 char *state;
1120                 switch (args->ar_data) {
1121                 case GFS2_DATA_WRITEBACK:
1122                         state = "writeback";
1123                         break;
1124                 case GFS2_DATA_ORDERED:
1125                         state = "ordered";
1126                         break;
1127                 default:
1128                         state = "unknown";
1129                         break;
1130                 }
1131                 seq_printf(s, ",data=%s", state);
1132         }
1133         if (args->ar_discard)
1134                 seq_puts(s, ",discard");
1135         if (logd_secs != 30)
1136                 seq_printf(s, ",commit=%d", logd_secs);
1137         if (statfs_quantum != 30)
1138                 seq_printf(s, ",statfs_quantum=%d", statfs_quantum);
1139         else if (statfs_slow)
1140                 seq_puts(s, ",statfs_quantum=0");
1141         if (quota_quantum != 60)
1142                 seq_printf(s, ",quota_quantum=%d", quota_quantum);
1143         if (args->ar_statfs_percent)
1144                 seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent);
1145         if (args->ar_errors != GFS2_ERRORS_DEFAULT) {
1146                 const char *state;
1147
1148                 switch (args->ar_errors) {
1149                 case GFS2_ERRORS_WITHDRAW:
1150                         state = "withdraw";
1151                         break;
1152                 case GFS2_ERRORS_PANIC:
1153                         state = "panic";
1154                         break;
1155                 default:
1156                         state = "unknown";
1157                         break;
1158                 }
1159                 seq_printf(s, ",errors=%s", state);
1160         }
1161         if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags))
1162                 seq_puts(s, ",nobarrier");
1163         if (test_bit(SDF_DEMOTE, &sdp->sd_flags))
1164                 seq_puts(s, ",demote_interface_used");
1165         if (args->ar_rgrplvb)
1166                 seq_puts(s, ",rgrplvb");
1167         if (args->ar_loccookie)
1168                 seq_puts(s, ",loccookie");
1169         return 0;
1170 }
1171
1172 /**
1173  * gfs2_glock_put_eventually
1174  * @gl: The glock to put
1175  *
1176  * When under memory pressure, trigger a deferred glock put to make sure we
1177  * won't call into DLM and deadlock.  Otherwise, put the glock directly.
1178  */
1179
1180 static void gfs2_glock_put_eventually(struct gfs2_glock *gl)
1181 {
1182         if (current->flags & PF_MEMALLOC)
1183                 gfs2_glock_put_async(gl);
1184         else
1185                 gfs2_glock_put(gl);
1186 }
1187
1188 static enum evict_behavior gfs2_upgrade_iopen_glock(struct inode *inode)
1189 {
1190         struct gfs2_inode *ip = GFS2_I(inode);
1191         struct gfs2_sbd *sdp = GFS2_SB(inode);
1192         struct gfs2_holder *gh = &ip->i_iopen_gh;
1193         int error;
1194
1195         gh->gh_flags |= GL_NOCACHE;
1196         gfs2_glock_dq_wait(gh);
1197
1198         /*
1199          * If there are no other lock holders, we will immediately get
1200          * exclusive access to the iopen glock here.
1201          *
1202          * Otherwise, the other nodes holding the lock will be notified about
1203          * our locking request (see iopen_go_callback()).  If they do not have
1204          * the inode open, they are expected to evict the cached inode and
1205          * release the lock, allowing us to proceed.
1206          *
1207          * Otherwise, if they cannot evict the inode, they are expected to poke
1208          * the inode glock (note: not the iopen glock).  We will notice that
1209          * and stop waiting for the iopen glock immediately.  The other node(s)
1210          * are then expected to take care of deleting the inode when they no
1211          * longer use it.
1212          *
1213          * As a last resort, if another node keeps holding the iopen glock
1214          * without showing any activity on the inode glock, we will eventually
1215          * time out and fail the iopen glock upgrade.
1216          */
1217
1218         gfs2_holder_reinit(LM_ST_EXCLUSIVE, GL_ASYNC | GL_NOCACHE, gh);
1219         error = gfs2_glock_nq(gh);
1220         if (error)
1221                 return EVICT_SHOULD_SKIP_DELETE;
1222
1223         wait_event_interruptible_timeout(sdp->sd_async_glock_wait,
1224                 !test_bit(HIF_WAIT, &gh->gh_iflags) ||
1225                 glock_needs_demote(ip->i_gl),
1226                 5 * HZ);
1227         if (!test_bit(HIF_HOLDER, &gh->gh_iflags)) {
1228                 gfs2_glock_dq(gh);
1229                 if (glock_needs_demote(ip->i_gl))
1230                         return EVICT_SHOULD_SKIP_DELETE;
1231                 return EVICT_SHOULD_DEFER_DELETE;
1232         }
1233         error = gfs2_glock_holder_ready(gh);
1234         if (error)
1235                 return EVICT_SHOULD_SKIP_DELETE;
1236         return EVICT_SHOULD_DELETE;
1237 }
1238
1239 /**
1240  * evict_should_delete - determine whether the inode is eligible for deletion
1241  * @inode: The inode to evict
1242  * @gh: The glock holder structure
1243  *
1244  * This function determines whether the evicted inode is eligible to be deleted
1245  * and locks the inode glock.
1246  *
1247  * Returns: the fate of the dinode
1248  */
1249 static enum evict_behavior evict_should_delete(struct inode *inode,
1250                                                struct gfs2_holder *gh)
1251 {
1252         struct gfs2_inode *ip = GFS2_I(inode);
1253         struct super_block *sb = inode->i_sb;
1254         struct gfs2_sbd *sdp = sb->s_fs_info;
1255         int ret;
1256
1257         if (gfs2_holder_initialized(&ip->i_iopen_gh) &&
1258             test_bit(GLF_DEFER_DELETE, &ip->i_iopen_gh.gh_gl->gl_flags))
1259                 return EVICT_SHOULD_DEFER_DELETE;
1260
1261         /* Deletes should never happen under memory pressure anymore.  */
1262         if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
1263                 return EVICT_SHOULD_DEFER_DELETE;
1264
1265         /* Must not read inode block until block type has been verified */
1266         ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, gh);
1267         if (unlikely(ret))
1268                 return EVICT_SHOULD_SKIP_DELETE;
1269
1270         if (gfs2_inode_already_deleted(ip->i_gl, ip->i_no_formal_ino))
1271                 return EVICT_SHOULD_SKIP_DELETE;
1272         ret = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED);
1273         if (ret)
1274                 return EVICT_SHOULD_SKIP_DELETE;
1275
1276         ret = gfs2_instantiate(gh);
1277         if (ret)
1278                 return EVICT_SHOULD_SKIP_DELETE;
1279
1280         /*
1281          * The inode may have been recreated in the meantime.
1282          */
1283         if (inode->i_nlink)
1284                 return EVICT_SHOULD_SKIP_DELETE;
1285
1286         if (gfs2_holder_initialized(&ip->i_iopen_gh) &&
1287             test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags))
1288                 return gfs2_upgrade_iopen_glock(inode);
1289         return EVICT_SHOULD_DELETE;
1290 }
1291
1292 /**
1293  * evict_unlinked_inode - delete the pieces of an unlinked evicted inode
1294  * @inode: The inode to evict
1295  */
1296 static int evict_unlinked_inode(struct inode *inode)
1297 {
1298         struct gfs2_inode *ip = GFS2_I(inode);
1299         int ret;
1300
1301         if (S_ISDIR(inode->i_mode) &&
1302             (ip->i_diskflags & GFS2_DIF_EXHASH)) {
1303                 ret = gfs2_dir_exhash_dealloc(ip);
1304                 if (ret)
1305                         goto out;
1306         }
1307
1308         if (ip->i_eattr) {
1309                 ret = gfs2_ea_dealloc(ip, true);
1310                 if (ret)
1311                         goto out;
1312         }
1313
1314         if (!gfs2_is_stuffed(ip)) {
1315                 ret = gfs2_file_dealloc(ip);
1316                 if (ret)
1317                         goto out;
1318         }
1319
1320         /*
1321          * As soon as we clear the bitmap for the dinode, gfs2_create_inode()
1322          * can get called to recreate it, or even gfs2_inode_lookup() if the
1323          * inode was recreated on another node in the meantime.
1324          *
1325          * However, inserting the new inode into the inode hash table will not
1326          * succeed until the old inode is removed, and that only happens after
1327          * ->evict_inode() returns.  The new inode is attached to its inode and
1328          *  iopen glocks after inserting it into the inode hash table, so at
1329          *  that point we can be sure that both glocks are unused.
1330          */
1331
1332         ret = gfs2_dinode_dealloc(ip);
1333         if (!ret && ip->i_gl)
1334                 gfs2_inode_remember_delete(ip->i_gl, ip->i_no_formal_ino);
1335
1336 out:
1337         return ret;
1338 }
1339
1340 /*
1341  * evict_linked_inode - evict an inode whose dinode has not been unlinked
1342  * @inode: The inode to evict
1343  */
1344 static int evict_linked_inode(struct inode *inode)
1345 {
1346         struct super_block *sb = inode->i_sb;
1347         struct gfs2_sbd *sdp = sb->s_fs_info;
1348         struct gfs2_inode *ip = GFS2_I(inode);
1349         struct address_space *metamapping;
1350         int ret;
1351
1352         gfs2_log_flush(sdp, ip->i_gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
1353                        GFS2_LFC_EVICT_INODE);
1354         metamapping = gfs2_glock2aspace(ip->i_gl);
1355         if (test_bit(GLF_DIRTY, &ip->i_gl->gl_flags)) {
1356                 filemap_fdatawrite(metamapping);
1357                 filemap_fdatawait(metamapping);
1358         }
1359         write_inode_now(inode, 1);
1360         gfs2_ail_flush(ip->i_gl, 0);
1361
1362         ret = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
1363         if (ret)
1364                 return ret;
1365
1366         /* Needs to be done before glock release & also in a transaction */
1367         truncate_inode_pages(&inode->i_data, 0);
1368         truncate_inode_pages(metamapping, 0);
1369         gfs2_trans_end(sdp);
1370         return 0;
1371 }
1372
1373 /**
1374  * gfs2_evict_inode - Remove an inode from cache
1375  * @inode: The inode to evict
1376  *
1377  * There are three cases to consider:
1378  * 1. i_nlink == 0, we are final opener (and must deallocate)
1379  * 2. i_nlink == 0, we are not the final opener (and cannot deallocate)
1380  * 3. i_nlink > 0
1381  *
1382  * If the fs is read only, then we have to treat all cases as per #3
1383  * since we are unable to do any deallocation. The inode will be
1384  * deallocated by the next read/write node to attempt an allocation
1385  * in the same resource group
1386  *
1387  * We have to (at the moment) hold the inodes main lock to cover
1388  * the gap between unlocking the shared lock on the iopen lock and
1389  * taking the exclusive lock. I'd rather do a shared -> exclusive
1390  * conversion on the iopen lock, but we can change that later. This
1391  * is safe, just less efficient.
1392  */
1393
1394 static void gfs2_evict_inode(struct inode *inode)
1395 {
1396         struct super_block *sb = inode->i_sb;
1397         struct gfs2_sbd *sdp = sb->s_fs_info;
1398         struct gfs2_inode *ip = GFS2_I(inode);
1399         struct gfs2_holder gh;
1400         enum evict_behavior behavior;
1401         int ret;
1402
1403         gfs2_holder_mark_uninitialized(&gh);
1404         if (inode->i_nlink || sb_rdonly(sb) || !ip->i_no_addr)
1405                 goto out;
1406
1407         /*
1408          * In case of an incomplete mount, gfs2_evict_inode() may be called for
1409          * system files without having an active journal to write to.  In that
1410          * case, skip the filesystem evict.
1411          */
1412         if (!sdp->sd_jdesc)
1413                 goto out;
1414
1415         behavior = evict_should_delete(inode, &gh);
1416         if (behavior == EVICT_SHOULD_DEFER_DELETE &&
1417             !test_bit(SDF_KILL, &sdp->sd_flags)) {
1418                 struct gfs2_glock *io_gl = ip->i_iopen_gh.gh_gl;
1419
1420                 if (io_gl) {
1421                         gfs2_glock_hold(io_gl);
1422                         if (!gfs2_queue_verify_delete(io_gl, true))
1423                                 gfs2_glock_put(io_gl);
1424                         goto out;
1425                 }
1426                 behavior = EVICT_SHOULD_SKIP_DELETE;
1427         }
1428         if (behavior == EVICT_SHOULD_DELETE)
1429                 ret = evict_unlinked_inode(inode);
1430         else
1431                 ret = evict_linked_inode(inode);
1432
1433         if (gfs2_rs_active(&ip->i_res))
1434                 gfs2_rs_deltree(&ip->i_res);
1435
1436         if (ret && ret != GLR_TRYFAILED && ret != -EROFS)
1437                 fs_warn(sdp, "gfs2_evict_inode: %d\n", ret);
1438 out:
1439         if (gfs2_holder_initialized(&gh))
1440                 gfs2_glock_dq_uninit(&gh);
1441         truncate_inode_pages_final(&inode->i_data);
1442         if (ip->i_qadata)
1443                 gfs2_assert_warn(sdp, ip->i_qadata->qa_ref == 0);
1444         gfs2_rs_deltree(&ip->i_res);
1445         gfs2_ordered_del_inode(ip);
1446         clear_inode(inode);
1447         gfs2_dir_hash_inval(ip);
1448         if (gfs2_holder_initialized(&ip->i_iopen_gh)) {
1449                 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1450
1451                 glock_clear_object(gl, ip);
1452                 gfs2_glock_hold(gl);
1453                 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1454                 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1455                 gfs2_glock_put_eventually(gl);
1456         }
1457         if (ip->i_gl) {
1458                 glock_clear_object(ip->i_gl, ip);
1459                 wait_on_bit_io(&ip->i_flags, GIF_GLOP_PENDING, TASK_UNINTERRUPTIBLE);
1460                 gfs2_glock_put_eventually(ip->i_gl);
1461                 rcu_assign_pointer(ip->i_gl, NULL);
1462         }
1463 }
1464
1465 static struct inode *gfs2_alloc_inode(struct super_block *sb)
1466 {
1467         struct gfs2_inode *ip;
1468
1469         ip = alloc_inode_sb(sb, gfs2_inode_cachep, GFP_KERNEL);
1470         if (!ip)
1471                 return NULL;
1472         ip->i_no_addr = 0;
1473         ip->i_no_formal_ino = 0;
1474         ip->i_flags = 0;
1475         ip->i_gl = NULL;
1476         gfs2_holder_mark_uninitialized(&ip->i_iopen_gh);
1477         memset(&ip->i_res, 0, sizeof(ip->i_res));
1478         RB_CLEAR_NODE(&ip->i_res.rs_node);
1479         ip->i_diskflags = 0;
1480         ip->i_rahead = 0;
1481         return &ip->i_inode;
1482 }
1483
1484 static void gfs2_free_inode(struct inode *inode)
1485 {
1486         kmem_cache_free(gfs2_inode_cachep, GFS2_I(inode));
1487 }
1488
1489 void free_local_statfs_inodes(struct gfs2_sbd *sdp)
1490 {
1491         struct local_statfs_inode *lsi, *safe;
1492
1493         /* Run through the statfs inodes list to iput and free memory */
1494         list_for_each_entry_safe(lsi, safe, &sdp->sd_sc_inodes_list, si_list) {
1495                 if (lsi->si_jid == sdp->sd_jdesc->jd_jid)
1496                         sdp->sd_sc_inode = NULL; /* belongs to this node */
1497                 if (lsi->si_sc_inode)
1498                         iput(lsi->si_sc_inode);
1499                 list_del(&lsi->si_list);
1500                 kfree(lsi);
1501         }
1502 }
1503
1504 struct inode *find_local_statfs_inode(struct gfs2_sbd *sdp,
1505                                       unsigned int index)
1506 {
1507         struct local_statfs_inode *lsi;
1508
1509         /* Return the local (per node) statfs inode in the
1510          * sdp->sd_sc_inodes_list corresponding to the 'index'. */
1511         list_for_each_entry(lsi, &sdp->sd_sc_inodes_list, si_list) {
1512                 if (lsi->si_jid == index)
1513                         return lsi->si_sc_inode;
1514         }
1515         return NULL;
1516 }
1517
1518 const struct super_operations gfs2_super_ops = {
1519         .alloc_inode            = gfs2_alloc_inode,
1520         .free_inode             = gfs2_free_inode,
1521         .write_inode            = gfs2_write_inode,
1522         .dirty_inode            = gfs2_dirty_inode,
1523         .evict_inode            = gfs2_evict_inode,
1524         .put_super              = gfs2_put_super,
1525         .sync_fs                = gfs2_sync_fs,
1526         .freeze_super           = gfs2_freeze_super,
1527         .freeze_fs              = gfs2_freeze_fs,
1528         .thaw_super             = gfs2_thaw_super,
1529         .statfs                 = gfs2_statfs,
1530         .drop_inode             = gfs2_drop_inode,
1531         .show_options           = gfs2_show_options,
1532 };
1533