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 || gfs2_withdrawn(sdp))
 142                 return error;
 143
 144         if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 145                 gfs2_consist(sdp);
 146                 return -EIO;
 147         }
 148
 149         /*  Initialize some head of the log stuff  */
 150         sdp->sd_log_sequence = head.lh_sequence + 1;
 151         gfs2_log_pointers_init(sdp, head.lh_blkno);
 152
 153         error = gfs2_quota_init(sdp);
 154         if (!error && !gfs2_withdrawn(sdp))
 155                 set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 156         return error;
 157 }
 158
 159 void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
 160 {
 161         const struct gfs2_statfs_change *str = buf;
 162
 163         sc->sc_total = be64_to_cpu(str->sc_total);
 164         sc->sc_free = be64_to_cpu(str->sc_free);
 165         sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
 166 }
 167
 168 void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
 169 {
 170         struct gfs2_statfs_change *str = buf;
 171
 172         str->sc_total = cpu_to_be64(sc->sc_total);
 173         str->sc_free = cpu_to_be64(sc->sc_free);
 174         str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
 175 }
 176
 177 int gfs2_statfs_init(struct gfs2_sbd *sdp)
 178 {
 179         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 180         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 181         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 182         struct buffer_head *m_bh;
 183         struct gfs2_holder gh;
 184         int error;
 185
 186         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 187                                    &gh);
 188         if (error)
 189                 return error;
 190
 191         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 192         if (error)
 193                 goto out;
 194
 195         if (sdp->sd_args.ar_spectator) {
 196                 spin_lock(&sdp->sd_statfs_spin);
 197                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 198                                       sizeof(struct gfs2_dinode));
 199                 spin_unlock(&sdp->sd_statfs_spin);
 200         } else {
 201                 spin_lock(&sdp->sd_statfs_spin);
 202                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 203                                       sizeof(struct gfs2_dinode));
 204                 gfs2_statfs_change_in(l_sc, sdp->sd_sc_bh->b_data +
 205                                       sizeof(struct gfs2_dinode));
 206                 spin_unlock(&sdp->sd_statfs_spin);
 207
 208         }
 209
 210         brelse(m_bh);
 211 out:
 212         gfs2_glock_dq_uninit(&gh);
 213         return 0;
 214 }
 215
 216 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
 217                         s64 dinodes)
 218 {
 219         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 220         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 221         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 222         s64 x, y;
 223         int need_sync = 0;
 224
 225         gfs2_trans_add_meta(l_ip->i_gl, sdp->sd_sc_bh);
 226
 227         spin_lock(&sdp->sd_statfs_spin);
 228         l_sc->sc_total += total;
 229         l_sc->sc_free += free;
 230         l_sc->sc_dinodes += dinodes;
 231         gfs2_statfs_change_out(l_sc, sdp->sd_sc_bh->b_data +
 232                                sizeof(struct gfs2_dinode));
 233         if (sdp->sd_args.ar_statfs_percent) {
 234                 x = 100 * l_sc->sc_free;
 235                 y = m_sc->sc_free * sdp->sd_args.ar_statfs_percent;
 236                 if (x >= y || x <= -y)
 237                         need_sync = 1;
 238         }
 239         spin_unlock(&sdp->sd_statfs_spin);
 240
 241         if (need_sync)
 242                 gfs2_wake_up_statfs(sdp);
 243 }
 244
 245 void update_statfs(struct gfs2_sbd *sdp, struct buffer_head *m_bh)
 246 {
 247         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 248         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 249         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 250         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 251
 252         gfs2_trans_add_meta(l_ip->i_gl, sdp->sd_sc_bh);
 253         gfs2_trans_add_meta(m_ip->i_gl, m_bh);
 254
 255         spin_lock(&sdp->sd_statfs_spin);
 256         m_sc->sc_total += l_sc->sc_total;
 257         m_sc->sc_free += l_sc->sc_free;
 258         m_sc->sc_dinodes += l_sc->sc_dinodes;
 259         memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
 260         memset(sdp->sd_sc_bh->b_data + sizeof(struct gfs2_dinode),
 261                0, sizeof(struct gfs2_statfs_change));
 262         gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
 263         spin_unlock(&sdp->sd_statfs_spin);
 264 }
 265
 266 int gfs2_statfs_sync(struct super_block *sb, int type)
 267 {
 268         struct gfs2_sbd *sdp = sb->s_fs_info;
 269         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 270         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 271         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 272         struct gfs2_holder gh;
 273         struct buffer_head *m_bh;
 274         int error;
 275
 276         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 277                                    &gh);
 278         if (error)
 279                 goto out;
 280
 281         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 282         if (error)
 283                 goto out_unlock;
 284
 285         spin_lock(&sdp->sd_statfs_spin);
 286         gfs2_statfs_change_in(m_sc, m_bh->b_data +
 287                               sizeof(struct gfs2_dinode));
 288         if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
 289                 spin_unlock(&sdp->sd_statfs_spin);
 290                 goto out_bh;
 291         }
 292         spin_unlock(&sdp->sd_statfs_spin);
 293
 294         error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
 295         if (error)
 296                 goto out_bh;
 297
 298         update_statfs(sdp, m_bh);
 299         sdp->sd_statfs_force_sync = 0;
 300
 301         gfs2_trans_end(sdp);
 302
 303 out_bh:
 304         brelse(m_bh);
 305 out_unlock:
 306         gfs2_glock_dq_uninit(&gh);
 307 out:
 308         return error;
 309 }
 310
 311 struct lfcc {
 312         struct list_head list;
 313         struct gfs2_holder gh;
 314 };
 315
 316 /**
 317  * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
 318  *                            journals are clean
 319  * @sdp: the file system
 320  *
 321  * Returns: errno
 322  */
 323
 324 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp)
 325 {
 326         struct gfs2_inode *ip;
 327         struct gfs2_jdesc *jd;
 328         struct lfcc *lfcc;
 329         LIST_HEAD(list);
 330         struct gfs2_log_header_host lh;
 331         int error;
 332
 333         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 334                 lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
 335                 if (!lfcc) {
 336                         error = -ENOMEM;
 337                         goto out;
 338                 }
 339                 ip = GFS2_I(jd->jd_inode);
 340                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
 341                 if (error) {
 342                         kfree(lfcc);
 343                         goto out;
 344                 }
 345                 list_add(&lfcc->list, &list);
 346         }
 347
 348         error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_EXCLUSIVE,
 349                                    LM_FLAG_NOEXP | GL_NOPID,
 350                                    &sdp->sd_freeze_gh);
 351         if (error)
 352                 goto out;
 353
 354         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 355                 error = gfs2_jdesc_check(jd);
 356                 if (error)
 357                         break;
 358                 error = gfs2_find_jhead(jd, &lh, false);
 359                 if (error)
 360                         break;
 361                 if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 362                         error = -EBUSY;
 363                         break;
 364                 }
 365         }
 366
 367         if (error)
 368                 gfs2_freeze_unlock(&sdp->sd_freeze_gh);
 369
 370 out:
 371         while (!list_empty(&list)) {
 372                 lfcc = list_first_entry(&list, struct lfcc, list);
 373                 list_del(&lfcc->list);
 374                 gfs2_glock_dq_uninit(&lfcc->gh);
 375                 kfree(lfcc);
 376         }
 377         return error;
 378 }
 379
 380 void gfs2_dinode_out(const struct gfs2_inode *ip, void *buf)
 381 {
 382         const struct inode *inode = &ip->i_inode;
 383         struct gfs2_dinode *str = buf;
 384
 385         str->di_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 386         str->di_header.mh_type = cpu_to_be32(GFS2_METATYPE_DI);
 387         str->di_header.mh_format = cpu_to_be32(GFS2_FORMAT_DI);
 388         str->di_num.no_addr = cpu_to_be64(ip->i_no_addr);
 389         str->di_num.no_formal_ino = cpu_to_be64(ip->i_no_formal_ino);
 390         str->di_mode = cpu_to_be32(inode->i_mode);
 391         str->di_uid = cpu_to_be32(i_uid_read(inode));
 392         str->di_gid = cpu_to_be32(i_gid_read(inode));
 393         str->di_nlink = cpu_to_be32(inode->i_nlink);
 394         str->di_size = cpu_to_be64(i_size_read(inode));
 395         str->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(inode));
 396         str->di_atime = cpu_to_be64(inode->i_atime.tv_sec);
 397         str->di_mtime = cpu_to_be64(inode->i_mtime.tv_sec);
 398         str->di_ctime = cpu_to_be64(inode->i_ctime.tv_sec);
 399
 400         str->di_goal_meta = cpu_to_be64(ip->i_goal);
 401         str->di_goal_data = cpu_to_be64(ip->i_goal);
 402         str->di_generation = cpu_to_be64(ip->i_generation);
 403
 404         str->di_flags = cpu_to_be32(ip->i_diskflags);
 405         str->di_height = cpu_to_be16(ip->i_height);
 406         str->di_payload_format = cpu_to_be32(S_ISDIR(inode->i_mode) &&
 407                                              !(ip->i_diskflags & GFS2_DIF_EXHASH) ?
 408                                              GFS2_FORMAT_DE : 0);
 409         str->di_depth = cpu_to_be16(ip->i_depth);
 410         str->di_entries = cpu_to_be32(ip->i_entries);
 411
 412         str->di_eattr = cpu_to_be64(ip->i_eattr);
 413         str->di_atime_nsec = cpu_to_be32(inode->i_atime.tv_nsec);
 414         str->di_mtime_nsec = cpu_to_be32(inode->i_mtime.tv_nsec);
 415         str->di_ctime_nsec = cpu_to_be32(inode->i_ctime.tv_nsec);
 416 }
 417
 418 /**
 419  * gfs2_write_inode - Make sure the inode is stable on the disk
 420  * @inode: The inode
 421  * @wbc: The writeback control structure
 422  *
 423  * Returns: errno
 424  */
 425
 426 static int gfs2_write_inode(struct inode *inode, struct writeback_control *wbc)
 427 {
 428         struct gfs2_inode *ip = GFS2_I(inode);
 429         struct gfs2_sbd *sdp = GFS2_SB(inode);
 430         struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl);
 431         struct backing_dev_info *bdi = inode_to_bdi(metamapping->host);
 432         int ret = 0;
 433         bool flush_all = (wbc->sync_mode == WB_SYNC_ALL || gfs2_is_jdata(ip));
 434
 435         if (flush_all)
 436                 gfs2_log_flush(GFS2_SB(inode), ip->i_gl,
 437                                GFS2_LOG_HEAD_FLUSH_NORMAL |
 438                                GFS2_LFC_WRITE_INODE);
 439         if (bdi->wb.dirty_exceeded)
 440                 gfs2_ail1_flush(sdp, wbc);
 441         else
 442                 filemap_fdatawrite(metamapping);
 443         if (flush_all)
 444                 ret = filemap_fdatawait(metamapping);
 445         if (ret)
 446                 mark_inode_dirty_sync(inode);
 447         else {
 448                 spin_lock(&inode->i_lock);
 449                 if (!(inode->i_flags & I_DIRTY))
 450                         gfs2_ordered_del_inode(ip);
 451                 spin_unlock(&inode->i_lock);
 452         }
 453         return ret;
 454 }
 455
 456 /**
 457  * gfs2_dirty_inode - check for atime updates
 458  * @inode: The inode in question
 459  * @flags: The type of dirty
 460  *
 461  * Unfortunately it can be called under any combination of inode
 462  * glock and transaction lock, so we have to check carefully.
 463  *
 464  * At the moment this deals only with atime - it should be possible
 465  * to expand that role in future, once a review of the locking has
 466  * been carried out.
 467  */
 468
 469 static void gfs2_dirty_inode(struct inode *inode, int flags)
 470 {
 471         struct gfs2_inode *ip = GFS2_I(inode);
 472         struct gfs2_sbd *sdp = GFS2_SB(inode);
 473         struct buffer_head *bh;
 474         struct gfs2_holder gh;
 475         int need_unlock = 0;
 476         int need_endtrans = 0;
 477         int ret;
 478
 479         if (unlikely(!ip->i_gl)) {
 480                 /* This can only happen during incomplete inode creation. */
 481                 BUG_ON(!test_bit(GIF_ALLOC_FAILED, &ip->i_flags));
 482                 return;
 483         }
 484
 485         if (unlikely(gfs2_withdrawn(sdp)))
 486                 return;
 487         if (!gfs2_glock_is_locked_by_me(ip->i_gl)) {
 488                 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
 489                 if (ret) {
 490                         fs_err(sdp, "dirty_inode: glock %d\n", ret);
 491                         gfs2_dump_glock(NULL, ip->i_gl, true);
 492                         return;
 493                 }
 494                 need_unlock = 1;
 495         } else if (WARN_ON_ONCE(ip->i_gl->gl_state != LM_ST_EXCLUSIVE))
 496                 return;
 497
 498         if (current->journal_info == NULL) {
 499                 ret = gfs2_trans_begin(sdp, RES_DINODE, 0);
 500                 if (ret) {
 501                         fs_err(sdp, "dirty_inode: gfs2_trans_begin %d\n", ret);
 502                         goto out;
 503                 }
 504                 need_endtrans = 1;
 505         }
 506
 507         ret = gfs2_meta_inode_buffer(ip, &bh);
 508         if (ret == 0) {
 509                 gfs2_trans_add_meta(ip->i_gl, bh);
 510                 gfs2_dinode_out(ip, bh->b_data);
 511                 brelse(bh);
 512         }
 513
 514         if (need_endtrans)
 515                 gfs2_trans_end(sdp);
 516 out:
 517         if (need_unlock)
 518                 gfs2_glock_dq_uninit(&gh);
 519 }
 520
 521 /**
 522  * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
 523  * @sdp: the filesystem
 524  *
 525  * Returns: errno
 526  */
 527
 528 void gfs2_make_fs_ro(struct gfs2_sbd *sdp)
 529 {
 530         int log_write_allowed = test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 531
 532         gfs2_flush_delete_work(sdp);
 533         if (!log_write_allowed && current == sdp->sd_quotad_process)
 534                 fs_warn(sdp, "The quotad daemon is withdrawing.\n");
 535         else if (sdp->sd_quotad_process)
 536                 kthread_stop(sdp->sd_quotad_process);
 537         sdp->sd_quotad_process = NULL;
 538
 539         if (!log_write_allowed && current == sdp->sd_logd_process)
 540                 fs_warn(sdp, "The logd daemon is withdrawing.\n");
 541         else if (sdp->sd_logd_process)
 542                 kthread_stop(sdp->sd_logd_process);
 543         sdp->sd_logd_process = NULL;
 544
 545         if (log_write_allowed) {
 546                 gfs2_quota_sync(sdp->sd_vfs, 0);
 547                 gfs2_statfs_sync(sdp->sd_vfs, 0);
 548
 549                 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SHUTDOWN |
 550                                GFS2_LFC_MAKE_FS_RO);
 551                 wait_event_timeout(sdp->sd_log_waitq,
 552                                    gfs2_log_is_empty(sdp),
 553                                    HZ * 5);
 554                 gfs2_assert_warn(sdp, gfs2_log_is_empty(sdp));
 555         } else {
 556                 wait_event_timeout(sdp->sd_log_waitq,
 557                                    gfs2_log_is_empty(sdp),
 558                                    HZ * 5);
 559         }
 560         gfs2_quota_cleanup(sdp);
 561
 562         if (!log_write_allowed)
 563                 sdp->sd_vfs->s_flags |= SB_RDONLY;
 564 }
 565
 566 /**
 567  * gfs2_put_super - Unmount the filesystem
 568  * @sb: The VFS superblock
 569  *
 570  */
 571
 572 static void gfs2_put_super(struct super_block *sb)
 573 {
 574         struct gfs2_sbd *sdp = sb->s_fs_info;
 575         struct gfs2_jdesc *jd;
 576
 577         /* No more recovery requests */
 578         set_bit(SDF_NORECOVERY, &sdp->sd_flags);
 579         smp_mb();
 580
 581         /* Wait on outstanding recovery */
 582 restart:
 583         spin_lock(&sdp->sd_jindex_spin);
 584         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 585                 if (!test_bit(JDF_RECOVERY, &jd->jd_flags))
 586                         continue;
 587                 spin_unlock(&sdp->sd_jindex_spin);
 588                 wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
 589                             TASK_UNINTERRUPTIBLE);
 590                 goto restart;
 591         }
 592         spin_unlock(&sdp->sd_jindex_spin);
 593
 594         if (!sb_rdonly(sb)) {
 595                 gfs2_make_fs_ro(sdp);
 596         }
 597         WARN_ON(gfs2_withdrawing(sdp));
 598
 599         /*  At this point, we're through modifying the disk  */
 600
 601         /*  Release stuff  */
 602
 603         iput(sdp->sd_jindex);
 604         iput(sdp->sd_statfs_inode);
 605         iput(sdp->sd_rindex);
 606         iput(sdp->sd_quota_inode);
 607
 608         gfs2_glock_put(sdp->sd_rename_gl);
 609         gfs2_glock_put(sdp->sd_freeze_gl);
 610
 611         if (!sdp->sd_args.ar_spectator) {
 612                 if (gfs2_holder_initialized(&sdp->sd_journal_gh))
 613                         gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
 614                 if (gfs2_holder_initialized(&sdp->sd_jinode_gh))
 615                         gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
 616                 brelse(sdp->sd_sc_bh);
 617                 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
 618                 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
 619                 free_local_statfs_inodes(sdp);
 620                 iput(sdp->sd_qc_inode);
 621         }
 622
 623         gfs2_glock_dq_uninit(&sdp->sd_live_gh);
 624         gfs2_clear_rgrpd(sdp);
 625         gfs2_jindex_free(sdp);
 626         /*  Take apart glock structures and buffer lists  */
 627         gfs2_gl_hash_clear(sdp);
 628         truncate_inode_pages_final(&sdp->sd_aspace);
 629         gfs2_delete_debugfs_file(sdp);
 630         /*  Unmount the locking protocol  */
 631         gfs2_lm_unmount(sdp);
 632
 633         /*  At this point, we're through participating in the lockspace  */
 634         gfs2_sys_fs_del(sdp);
 635         free_sbd(sdp);
 636 }
 637
 638 /**
 639  * gfs2_sync_fs - sync the filesystem
 640  * @sb: the superblock
 641  * @wait: true to wait for completion
 642  *
 643  * Flushes the log to disk.
 644  */
 645
 646 static int gfs2_sync_fs(struct super_block *sb, int wait)
 647 {
 648         struct gfs2_sbd *sdp = sb->s_fs_info;
 649
 650         gfs2_quota_sync(sb, -1);
 651         if (wait)
 652                 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
 653                                GFS2_LFC_SYNC_FS);
 654         return sdp->sd_log_error;
 655 }
 656
 657 void gfs2_freeze_func(struct work_struct *work)
 658 {
 659         int error;
 660         struct gfs2_holder freeze_gh;
 661         struct gfs2_sbd *sdp = container_of(work, struct gfs2_sbd, sd_freeze_work);
 662         struct super_block *sb = sdp->sd_vfs;
 663
 664         atomic_inc(&sb->s_active);
 665         error = gfs2_freeze_lock(sdp, &freeze_gh, 0);
 666         if (error) {
 667                 gfs2_assert_withdraw(sdp, 0);
 668         } else {
 669                 atomic_set(&sdp->sd_freeze_state, SFS_UNFROZEN);
 670                 error = thaw_super(sb);
 671                 if (error) {
 672                         fs_info(sdp, "GFS2: couldn't thaw filesystem: %d\n",
 673                                 error);
 674                         gfs2_assert_withdraw(sdp, 0);
 675                 }
 676                 gfs2_freeze_unlock(&freeze_gh);
 677         }
 678         deactivate_super(sb);
 679         clear_bit_unlock(SDF_FS_FROZEN, &sdp->sd_flags);
 680         wake_up_bit(&sdp->sd_flags, SDF_FS_FROZEN);
 681         return;
 682 }
 683
 684 /**
 685  * gfs2_freeze - prevent further writes to the filesystem
 686  * @sb: the VFS structure for the filesystem
 687  *
 688  */
 689
 690 static int gfs2_freeze(struct super_block *sb)
 691 {
 692         struct gfs2_sbd *sdp = sb->s_fs_info;
 693         int error;
 694
 695         mutex_lock(&sdp->sd_freeze_mutex);
 696         if (atomic_read(&sdp->sd_freeze_state) != SFS_UNFROZEN) {
 697                 error = -EBUSY;
 698                 goto out;
 699         }
 700
 701         for (;;) {
 702                 if (gfs2_withdrawn(sdp)) {
 703                         error = -EINVAL;
 704                         goto out;
 705                 }
 706
 707                 error = gfs2_lock_fs_check_clean(sdp);
 708                 if (!error)
 709                         break;
 710
 711                 if (error == -EBUSY)
 712                         fs_err(sdp, "waiting for recovery before freeze\n");
 713                 else if (error == -EIO) {
 714                         fs_err(sdp, "Fatal IO error: cannot freeze gfs2 due "
 715                                "to recovery error.\n");
 716                         goto out;
 717                 } else {
 718                         fs_err(sdp, "error freezing FS: %d\n", error);
 719                 }
 720                 fs_err(sdp, "retrying...\n");
 721                 msleep(1000);
 722         }
 723         set_bit(SDF_FS_FROZEN, &sdp->sd_flags);
 724 out:
 725         mutex_unlock(&sdp->sd_freeze_mutex);
 726         return error;
 727 }
 728
 729 /**
 730  * gfs2_unfreeze - reallow writes to the filesystem
 731  * @sb: the VFS structure for the filesystem
 732  *
 733  */
 734
 735 static int gfs2_unfreeze(struct super_block *sb)
 736 {
 737         struct gfs2_sbd *sdp = sb->s_fs_info;
 738
 739         mutex_lock(&sdp->sd_freeze_mutex);
 740         if (atomic_read(&sdp->sd_freeze_state) != SFS_FROZEN ||
 741             !gfs2_holder_initialized(&sdp->sd_freeze_gh)) {
 742                 mutex_unlock(&sdp->sd_freeze_mutex);
 743                 return -EINVAL;
 744         }
 745
 746         gfs2_freeze_unlock(&sdp->sd_freeze_gh);
 747         mutex_unlock(&sdp->sd_freeze_mutex);
 748         return wait_on_bit(&sdp->sd_flags, SDF_FS_FROZEN, TASK_INTERRUPTIBLE);
 749 }
 750
 751 /**
 752  * statfs_slow_fill - fill in the sg for a given RG
 753  * @rgd: the RG
 754  * @sc: the sc structure
 755  *
 756  * Returns: 0 on success, -ESTALE if the LVB is invalid
 757  */
 758
 759 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
 760                             struct gfs2_statfs_change_host *sc)
 761 {
 762         gfs2_rgrp_verify(rgd);
 763         sc->sc_total += rgd->rd_data;
 764         sc->sc_free += rgd->rd_free;
 765         sc->sc_dinodes += rgd->rd_dinodes;
 766         return 0;
 767 }
 768
 769 /**
 770  * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
 771  * @sdp: the filesystem
 772  * @sc: the sc info that will be returned
 773  *
 774  * Any error (other than a signal) will cause this routine to fall back
 775  * to the synchronous version.
 776  *
 777  * FIXME: This really shouldn't busy wait like this.
 778  *
 779  * Returns: errno
 780  */
 781
 782 static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
 783 {
 784         struct gfs2_rgrpd *rgd_next;
 785         struct gfs2_holder *gha, *gh;
 786         unsigned int slots = 64;
 787         unsigned int x;
 788         int done;
 789         int error = 0, err;
 790
 791         memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
 792         gha = kmalloc_array(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
 793         if (!gha)
 794                 return -ENOMEM;
 795         for (x = 0; x < slots; x++)
 796                 gfs2_holder_mark_uninitialized(gha + x);
 797
 798         rgd_next = gfs2_rgrpd_get_first(sdp);
 799
 800         for (;;) {
 801                 done = 1;
 802
 803                 for (x = 0; x < slots; x++) {
 804                         gh = gha + x;
 805
 806                         if (gfs2_holder_initialized(gh) && gfs2_glock_poll(gh)) {
 807                                 err = gfs2_glock_wait(gh);
 808                                 if (err) {
 809                                         gfs2_holder_uninit(gh);
 810                                         error = err;
 811                                 } else {
 812                                         if (!error) {
 813                                                 struct gfs2_rgrpd *rgd =
 814                                                         gfs2_glock2rgrp(gh->gh_gl);
 815
 816                                                 error = statfs_slow_fill(rgd, sc);
 817                                         }
 818                                         gfs2_glock_dq_uninit(gh);
 819                                 }
 820                         }
 821
 822                         if (gfs2_holder_initialized(gh))
 823                                 done = 0;
 824                         else if (rgd_next && !error) {
 825                                 error = gfs2_glock_nq_init(rgd_next->rd_gl,
 826                                                            LM_ST_SHARED,
 827                                                            GL_ASYNC,
 828                                                            gh);
 829                                 rgd_next = gfs2_rgrpd_get_next(rgd_next);
 830                                 done = 0;
 831                         }
 832
 833                         if (signal_pending(current))
 834                                 error = -ERESTARTSYS;
 835                 }
 836
 837                 if (done)
 838                         break;
 839
 840                 yield();
 841         }
 842
 843         kfree(gha);
 844         return error;
 845 }
 846
 847 /**
 848  * gfs2_statfs_i - Do a statfs
 849  * @sdp: the filesystem
 850  * @sc: the sc structure
 851  *
 852  * Returns: errno
 853  */
 854
 855 static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
 856 {
 857         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 858         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 859
 860         spin_lock(&sdp->sd_statfs_spin);
 861
 862         *sc = *m_sc;
 863         sc->sc_total += l_sc->sc_total;
 864         sc->sc_free += l_sc->sc_free;
 865         sc->sc_dinodes += l_sc->sc_dinodes;
 866
 867         spin_unlock(&sdp->sd_statfs_spin);
 868
 869         if (sc->sc_free < 0)
 870                 sc->sc_free = 0;
 871         if (sc->sc_free > sc->sc_total)
 872                 sc->sc_free = sc->sc_total;
 873         if (sc->sc_dinodes < 0)
 874                 sc->sc_dinodes = 0;
 875
 876         return 0;
 877 }
 878
 879 /**
 880  * gfs2_statfs - Gather and return stats about the filesystem
 881  * @dentry: The name of the link
 882  * @buf: The buffer
 883  *
 884  * Returns: 0 on success or error code
 885  */
 886
 887 static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
 888 {
 889         struct super_block *sb = dentry->d_sb;
 890         struct gfs2_sbd *sdp = sb->s_fs_info;
 891         struct gfs2_statfs_change_host sc;
 892         int error;
 893
 894         error = gfs2_rindex_update(sdp);
 895         if (error)
 896                 return error;
 897
 898         if (gfs2_tune_get(sdp, gt_statfs_slow))
 899                 error = gfs2_statfs_slow(sdp, &sc);
 900         else
 901                 error = gfs2_statfs_i(sdp, &sc);
 902
 903         if (error)
 904                 return error;
 905
 906         buf->f_type = GFS2_MAGIC;
 907         buf->f_bsize = sdp->sd_sb.sb_bsize;
 908         buf->f_blocks = sc.sc_total;
 909         buf->f_bfree = sc.sc_free;
 910         buf->f_bavail = sc.sc_free;
 911         buf->f_files = sc.sc_dinodes + sc.sc_free;
 912         buf->f_ffree = sc.sc_free;
 913         buf->f_namelen = GFS2_FNAMESIZE;
 914
 915         return 0;
 916 }
 917
 918 /**
 919  * gfs2_drop_inode - Drop an inode (test for remote unlink)
 920  * @inode: The inode to drop
 921  *
 922  * If we've received a callback on an iopen lock then it's because a
 923  * remote node tried to deallocate the inode but failed due to this node
 924  * still having the inode open. Here we mark the link count zero
 925  * since we know that it must have reached zero if the GLF_DEMOTE flag
 926  * is set on the iopen glock. If we didn't do a disk read since the
 927  * remote node removed the final link then we might otherwise miss
 928  * this event. This check ensures that this node will deallocate the
 929  * inode's blocks, or alternatively pass the baton on to another
 930  * node for later deallocation.
 931  */
 932
 933 static int gfs2_drop_inode(struct inode *inode)
 934 {
 935         struct gfs2_inode *ip = GFS2_I(inode);
 936
 937         if (inode->i_nlink &&
 938             gfs2_holder_initialized(&ip->i_iopen_gh)) {
 939                 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
 940                 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
 941                         clear_nlink(inode);
 942         }
 943
 944         /*
 945          * When under memory pressure when an inode's link count has dropped to
 946          * zero, defer deleting the inode to the delete workqueue.  This avoids
 947          * calling into DLM under memory pressure, which can deadlock.
 948          */
 949         if (!inode->i_nlink &&
 950             unlikely(current->flags & PF_MEMALLOC) &&
 951             gfs2_holder_initialized(&ip->i_iopen_gh)) {
 952                 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
 953
 954                 gfs2_glock_hold(gl);
 955                 if (!gfs2_queue_delete_work(gl, 0))
 956                         gfs2_glock_queue_put(gl);
 957                 return 0;
 958         }
 959
 960         return generic_drop_inode(inode);
 961 }
 962
 963 static int is_ancestor(const struct dentry *d1, const struct dentry *d2)
 964 {
 965         do {
 966                 if (d1 == d2)
 967                         return 1;
 968                 d1 = d1->d_parent;
 969         } while (!IS_ROOT(d1));
 970         return 0;
 971 }
 972
 973 /**
 974  * gfs2_show_options - Show mount options for /proc/mounts
 975  * @s: seq_file structure
 976  * @root: root of this (sub)tree
 977  *
 978  * Returns: 0 on success or error code
 979  */
 980
 981 static int gfs2_show_options(struct seq_file *s, struct dentry *root)
 982 {
 983         struct gfs2_sbd *sdp = root->d_sb->s_fs_info;
 984         struct gfs2_args *args = &sdp->sd_args;
 985         int val;
 986
 987         if (is_ancestor(root, sdp->sd_master_dir))
 988                 seq_puts(s, ",meta");
 989         if (args->ar_lockproto[0])
 990                 seq_show_option(s, "lockproto", args->ar_lockproto);
 991         if (args->ar_locktable[0])
 992                 seq_show_option(s, "locktable", args->ar_locktable);
 993         if (args->ar_hostdata[0])
 994                 seq_show_option(s, "hostdata", args->ar_hostdata);
 995         if (args->ar_spectator)
 996                 seq_puts(s, ",spectator");
 997         if (args->ar_localflocks)
 998                 seq_puts(s, ",localflocks");
 999         if (args->ar_debug)
1000                 seq_puts(s, ",debug");
1001         if (args->ar_posix_acl)
1002                 seq_puts(s, ",acl");
1003         if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
1004                 char *state;
1005                 switch (args->ar_quota) {
1006                 case GFS2_QUOTA_OFF:
1007                         state = "off";
1008                         break;
1009                 case GFS2_QUOTA_ACCOUNT:
1010                         state = "account";
1011                         break;
1012                 case GFS2_QUOTA_ON:
1013                         state = "on";
1014                         break;
1015                 default:
1016                         state = "unknown";
1017                         break;
1018                 }
1019                 seq_printf(s, ",quota=%s", state);
1020         }
1021         if (args->ar_suiddir)
1022                 seq_puts(s, ",suiddir");
1023         if (args->ar_data != GFS2_DATA_DEFAULT) {
1024                 char *state;
1025                 switch (args->ar_data) {
1026                 case GFS2_DATA_WRITEBACK:
1027                         state = "writeback";
1028                         break;
1029                 case GFS2_DATA_ORDERED:
1030                         state = "ordered";
1031                         break;
1032                 default:
1033                         state = "unknown";
1034                         break;
1035                 }
1036                 seq_printf(s, ",data=%s", state);
1037         }
1038         if (args->ar_discard)
1039                 seq_puts(s, ",discard");
1040         val = sdp->sd_tune.gt_logd_secs;
1041         if (val != 30)
1042                 seq_printf(s, ",commit=%d", val);
1043         val = sdp->sd_tune.gt_statfs_quantum;
1044         if (val != 30)
1045                 seq_printf(s, ",statfs_quantum=%d", val);
1046         else if (sdp->sd_tune.gt_statfs_slow)
1047                 seq_puts(s, ",statfs_quantum=0");
1048         val = sdp->sd_tune.gt_quota_quantum;
1049         if (val != 60)
1050                 seq_printf(s, ",quota_quantum=%d", val);
1051         if (args->ar_statfs_percent)
1052                 seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent);
1053         if (args->ar_errors != GFS2_ERRORS_DEFAULT) {
1054                 const char *state;
1055
1056                 switch (args->ar_errors) {
1057                 case GFS2_ERRORS_WITHDRAW:
1058                         state = "withdraw";
1059                         break;
1060                 case GFS2_ERRORS_PANIC:
1061                         state = "panic";
1062                         break;
1063                 default:
1064                         state = "unknown";
1065                         break;
1066                 }
1067                 seq_printf(s, ",errors=%s", state);
1068         }
1069         if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags))
1070                 seq_puts(s, ",nobarrier");
1071         if (test_bit(SDF_DEMOTE, &sdp->sd_flags))
1072                 seq_puts(s, ",demote_interface_used");
1073         if (args->ar_rgrplvb)
1074                 seq_puts(s, ",rgrplvb");
1075         if (args->ar_loccookie)
1076                 seq_puts(s, ",loccookie");
1077         return 0;
1078 }
1079
1080 static void gfs2_final_release_pages(struct gfs2_inode *ip)
1081 {
1082         struct inode *inode = &ip->i_inode;
1083         struct gfs2_glock *gl = ip->i_gl;
1084
1085         if (unlikely(!gl)) {
1086                 /* This can only happen during incomplete inode creation. */
1087                 BUG_ON(!test_bit(GIF_ALLOC_FAILED, &ip->i_flags));
1088                 return;
1089         }
1090
1091         truncate_inode_pages(gfs2_glock2aspace(gl), 0);
1092         truncate_inode_pages(&inode->i_data, 0);
1093
1094         if (atomic_read(&gl->gl_revokes) == 0) {
1095                 clear_bit(GLF_LFLUSH, &gl->gl_flags);
1096                 clear_bit(GLF_DIRTY, &gl->gl_flags);
1097         }
1098 }
1099
1100 static int gfs2_dinode_dealloc(struct gfs2_inode *ip)
1101 {
1102         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1103         struct gfs2_rgrpd *rgd;
1104         struct gfs2_holder gh;
1105         int error;
1106
1107         if (gfs2_get_inode_blocks(&ip->i_inode) != 1) {
1108                 gfs2_consist_inode(ip);
1109                 return -EIO;
1110         }
1111
1112         error = gfs2_rindex_update(sdp);
1113         if (error)
1114                 return error;
1115
1116         error = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
1117         if (error)
1118                 return error;
1119
1120         rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1121         if (!rgd) {
1122                 gfs2_consist_inode(ip);
1123                 error = -EIO;
1124                 goto out_qs;
1125         }
1126
1127         error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE,
1128                                    LM_FLAG_NODE_SCOPE, &gh);
1129         if (error)
1130                 goto out_qs;
1131
1132         error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA,
1133                                  sdp->sd_jdesc->jd_blocks);
1134         if (error)
1135                 goto out_rg_gunlock;
1136
1137         gfs2_free_di(rgd, ip);
1138
1139         gfs2_final_release_pages(ip);
1140
1141         gfs2_trans_end(sdp);
1142
1143 out_rg_gunlock:
1144         gfs2_glock_dq_uninit(&gh);
1145 out_qs:
1146         gfs2_quota_unhold(ip);
1147         return error;
1148 }
1149
1150 /**
1151  * gfs2_glock_put_eventually
1152  * @gl: The glock to put
1153  *
1154  * When under memory pressure, trigger a deferred glock put to make sure we
1155  * won't call into DLM and deadlock.  Otherwise, put the glock directly.
1156  */
1157
1158 static void gfs2_glock_put_eventually(struct gfs2_glock *gl)
1159 {
1160         if (current->flags & PF_MEMALLOC)
1161                 gfs2_glock_queue_put(gl);
1162         else
1163                 gfs2_glock_put(gl);
1164 }
1165
1166 static bool gfs2_upgrade_iopen_glock(struct inode *inode)
1167 {
1168         struct gfs2_inode *ip = GFS2_I(inode);
1169         struct gfs2_sbd *sdp = GFS2_SB(inode);
1170         struct gfs2_holder *gh = &ip->i_iopen_gh;
1171         long timeout = 5 * HZ;
1172         int error;
1173
1174         gh->gh_flags |= GL_NOCACHE;
1175         gfs2_glock_dq_wait(gh);
1176
1177         /*
1178          * If there are no other lock holders, we'll get the lock immediately.
1179          * Otherwise, the other nodes holding the lock will be notified about
1180          * our locking request.  If they don't have the inode open, they'll
1181          * evict the cached inode and release the lock.  Otherwise, if they
1182          * poke the inode glock, we'll take this as an indication that they
1183          * still need the iopen glock and that they'll take care of deleting
1184          * the inode when they're done.  As a last resort, if another node
1185          * keeps holding the iopen glock without showing any activity on the
1186          * inode glock, we'll eventually time out.
1187          *
1188          * Note that we're passing the LM_FLAG_TRY_1CB flag to the first
1189          * locking request as an optimization to notify lock holders as soon as
1190          * possible.  Without that flag, they'd be notified implicitly by the
1191          * second locking request.
1192          */
1193
1194         gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE, gh);
1195         error = gfs2_glock_nq(gh);
1196         if (error != GLR_TRYFAILED)
1197                 return !error;
1198
1199         gfs2_holder_reinit(LM_ST_EXCLUSIVE, GL_ASYNC | GL_NOCACHE, gh);
1200         error = gfs2_glock_nq(gh);
1201         if (error)
1202                 return false;
1203
1204         timeout = wait_event_interruptible_timeout(sdp->sd_async_glock_wait,
1205                 !test_bit(HIF_WAIT, &gh->gh_iflags) ||
1206                 test_bit(GLF_DEMOTE, &ip->i_gl->gl_flags),
1207                 timeout);
1208         if (!test_bit(HIF_HOLDER, &gh->gh_iflags)) {
1209                 gfs2_glock_dq(gh);
1210                 return false;
1211         }
1212         return gfs2_glock_holder_ready(gh) == 0;
1213 }
1214
1215 /**
1216  * evict_should_delete - determine whether the inode is eligible for deletion
1217  * @inode: The inode to evict
1218  * @gh: The glock holder structure
1219  *
1220  * This function determines whether the evicted inode is eligible to be deleted
1221  * and locks the inode glock.
1222  *
1223  * Returns: the fate of the dinode
1224  */
1225 static enum dinode_demise evict_should_delete(struct inode *inode,
1226                                               struct gfs2_holder *gh)
1227 {
1228         struct gfs2_inode *ip = GFS2_I(inode);
1229         struct super_block *sb = inode->i_sb;
1230         struct gfs2_sbd *sdp = sb->s_fs_info;
1231         int ret;
1232
1233         if (unlikely(test_bit(GIF_ALLOC_FAILED, &ip->i_flags)))
1234                 goto should_delete;
1235
1236         if (test_bit(GIF_DEFERRED_DELETE, &ip->i_flags))
1237                 return SHOULD_DEFER_EVICTION;
1238
1239         /* Deletes should never happen under memory pressure anymore.  */
1240         if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
1241                 return SHOULD_DEFER_EVICTION;
1242
1243         /* Must not read inode block until block type has been verified */
1244         ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, gh);
1245         if (unlikely(ret)) {
1246                 glock_clear_object(ip->i_iopen_gh.gh_gl, ip);
1247                 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1248                 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1249                 return SHOULD_DEFER_EVICTION;
1250         }
1251
1252         if (gfs2_inode_already_deleted(ip->i_gl, ip->i_no_formal_ino))
1253                 return SHOULD_NOT_DELETE_DINODE;
1254         ret = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED);
1255         if (ret)
1256                 return SHOULD_NOT_DELETE_DINODE;
1257
1258         ret = gfs2_instantiate(gh);
1259         if (ret)
1260                 return SHOULD_NOT_DELETE_DINODE;
1261
1262         /*
1263          * The inode may have been recreated in the meantime.
1264          */
1265         if (inode->i_nlink)
1266                 return SHOULD_NOT_DELETE_DINODE;
1267
1268 should_delete:
1269         if (gfs2_holder_initialized(&ip->i_iopen_gh) &&
1270             test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1271                 if (!gfs2_upgrade_iopen_glock(inode)) {
1272                         gfs2_holder_uninit(&ip->i_iopen_gh);
1273                         return SHOULD_NOT_DELETE_DINODE;
1274                 }
1275         }
1276         return SHOULD_DELETE_DINODE;
1277 }
1278
1279 /**
1280  * evict_unlinked_inode - delete the pieces of an unlinked evicted inode
1281  * @inode: The inode to evict
1282  */
1283 static int evict_unlinked_inode(struct inode *inode)
1284 {
1285         struct gfs2_inode *ip = GFS2_I(inode);
1286         int ret;
1287
1288         if (S_ISDIR(inode->i_mode) &&
1289             (ip->i_diskflags & GFS2_DIF_EXHASH)) {
1290                 ret = gfs2_dir_exhash_dealloc(ip);
1291                 if (ret)
1292                         goto out;
1293         }
1294
1295         if (ip->i_eattr) {
1296                 ret = gfs2_ea_dealloc(ip);
1297                 if (ret)
1298                         goto out;
1299         }
1300
1301         if (!gfs2_is_stuffed(ip)) {
1302                 ret = gfs2_file_dealloc(ip);
1303                 if (ret)
1304                         goto out;
1305         }
1306
1307         if (ip->i_gl)
1308                 gfs2_inode_remember_delete(ip->i_gl, ip->i_no_formal_ino);
1309
1310         /*
1311          * As soon as we clear the bitmap for the dinode, gfs2_create_inode()
1312          * can get called to recreate it, or even gfs2_inode_lookup() if the
1313          * inode was recreated on another node in the meantime.
1314          *
1315          * However, inserting the new inode into the inode hash table will not
1316          * succeed until the old inode is removed, and that only happens after
1317          * ->evict_inode() returns.  The new inode is attached to its inode and
1318          *  iopen glocks after inserting it into the inode hash table, so at
1319          *  that point we can be sure that both glocks are unused.
1320          */
1321
1322         ret = gfs2_dinode_dealloc(ip);
1323 out:
1324         return ret;
1325 }
1326
1327 /*
1328  * evict_linked_inode - evict an inode whose dinode has not been unlinked
1329  * @inode: The inode to evict
1330  */
1331 static int evict_linked_inode(struct inode *inode)
1332 {
1333         struct super_block *sb = inode->i_sb;
1334         struct gfs2_sbd *sdp = sb->s_fs_info;
1335         struct gfs2_inode *ip = GFS2_I(inode);
1336         struct address_space *metamapping;
1337         int ret;
1338
1339         gfs2_log_flush(sdp, ip->i_gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
1340                        GFS2_LFC_EVICT_INODE);
1341         metamapping = gfs2_glock2aspace(ip->i_gl);
1342         if (test_bit(GLF_DIRTY, &ip->i_gl->gl_flags)) {
1343                 filemap_fdatawrite(metamapping);
1344                 filemap_fdatawait(metamapping);
1345         }
1346         write_inode_now(inode, 1);
1347         gfs2_ail_flush(ip->i_gl, 0);
1348
1349         ret = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
1350         if (ret)
1351                 return ret;
1352
1353         /* Needs to be done before glock release & also in a transaction */
1354         truncate_inode_pages(&inode->i_data, 0);
1355         truncate_inode_pages(metamapping, 0);
1356         gfs2_trans_end(sdp);
1357         return 0;
1358 }
1359
1360 /**
1361  * gfs2_evict_inode - Remove an inode from cache
1362  * @inode: The inode to evict
1363  *
1364  * There are three cases to consider:
1365  * 1. i_nlink == 0, we are final opener (and must deallocate)
1366  * 2. i_nlink == 0, we are not the final opener (and cannot deallocate)
1367  * 3. i_nlink > 0
1368  *
1369  * If the fs is read only, then we have to treat all cases as per #3
1370  * since we are unable to do any deallocation. The inode will be
1371  * deallocated by the next read/write node to attempt an allocation
1372  * in the same resource group
1373  *
1374  * We have to (at the moment) hold the inodes main lock to cover
1375  * the gap between unlocking the shared lock on the iopen lock and
1376  * taking the exclusive lock. I'd rather do a shared -> exclusive
1377  * conversion on the iopen lock, but we can change that later. This
1378  * is safe, just less efficient.
1379  */
1380
1381 static void gfs2_evict_inode(struct inode *inode)
1382 {
1383         struct super_block *sb = inode->i_sb;
1384         struct gfs2_sbd *sdp = sb->s_fs_info;
1385         struct gfs2_inode *ip = GFS2_I(inode);
1386         struct gfs2_holder gh;
1387         int ret;
1388
1389         if (inode->i_nlink || sb_rdonly(sb) || !ip->i_no_addr)
1390                 goto out;
1391
1392         gfs2_holder_mark_uninitialized(&gh);
1393         ret = evict_should_delete(inode, &gh);
1394         if (ret == SHOULD_DEFER_EVICTION)
1395                 goto out;
1396         if (ret == SHOULD_DELETE_DINODE)
1397                 ret = evict_unlinked_inode(inode);
1398         else
1399                 ret = evict_linked_inode(inode);
1400
1401         if (gfs2_rs_active(&ip->i_res))
1402                 gfs2_rs_deltree(&ip->i_res);
1403
1404         if (gfs2_holder_initialized(&gh)) {
1405                 glock_clear_object(ip->i_gl, ip);
1406                 gfs2_glock_dq_uninit(&gh);
1407         }
1408         if (ret && ret != GLR_TRYFAILED && ret != -EROFS)
1409                 fs_warn(sdp, "gfs2_evict_inode: %d\n", ret);
1410 out:
1411         truncate_inode_pages_final(&inode->i_data);
1412         if (ip->i_qadata)
1413                 gfs2_assert_warn(sdp, ip->i_qadata->qa_ref == 0);
1414         gfs2_rs_deltree(&ip->i_res);
1415         gfs2_ordered_del_inode(ip);
1416         clear_inode(inode);
1417         gfs2_dir_hash_inval(ip);
1418         if (gfs2_holder_initialized(&ip->i_iopen_gh)) {
1419                 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1420
1421                 glock_clear_object(gl, ip);
1422                 gfs2_glock_hold(gl);
1423                 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1424                 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1425                 gfs2_glock_put_eventually(gl);
1426         }
1427         if (ip->i_gl) {
1428                 glock_clear_object(ip->i_gl, ip);
1429                 wait_on_bit_io(&ip->i_flags, GIF_GLOP_PENDING, TASK_UNINTERRUPTIBLE);
1430                 gfs2_glock_add_to_lru(ip->i_gl);
1431                 gfs2_glock_put_eventually(ip->i_gl);
1432                 ip->i_gl = NULL;
1433         }
1434 }
1435
1436 static struct inode *gfs2_alloc_inode(struct super_block *sb)
1437 {
1438         struct gfs2_inode *ip;
1439
1440         ip = alloc_inode_sb(sb, gfs2_inode_cachep, GFP_KERNEL);
1441         if (!ip)
1442                 return NULL;
1443         ip->i_no_addr = 0;
1444         ip->i_flags = 0;
1445         ip->i_gl = NULL;
1446         gfs2_holder_mark_uninitialized(&ip->i_iopen_gh);
1447         memset(&ip->i_res, 0, sizeof(ip->i_res));
1448         RB_CLEAR_NODE(&ip->i_res.rs_node);
1449         ip->i_rahead = 0;
1450         return &ip->i_inode;
1451 }
1452
1453 static void gfs2_free_inode(struct inode *inode)
1454 {
1455         kmem_cache_free(gfs2_inode_cachep, GFS2_I(inode));
1456 }
1457
1458 extern void free_local_statfs_inodes(struct gfs2_sbd *sdp)
1459 {
1460         struct local_statfs_inode *lsi, *safe;
1461
1462         /* Run through the statfs inodes list to iput and free memory */
1463         list_for_each_entry_safe(lsi, safe, &sdp->sd_sc_inodes_list, si_list) {
1464                 if (lsi->si_jid == sdp->sd_jdesc->jd_jid)
1465                         sdp->sd_sc_inode = NULL; /* belongs to this node */
1466                 if (lsi->si_sc_inode)
1467                         iput(lsi->si_sc_inode);
1468                 list_del(&lsi->si_list);
1469                 kfree(lsi);
1470         }
1471 }
1472
1473 extern struct inode *find_local_statfs_inode(struct gfs2_sbd *sdp,
1474                                              unsigned int index)
1475 {
1476         struct local_statfs_inode *lsi;
1477
1478         /* Return the local (per node) statfs inode in the
1479          * sdp->sd_sc_inodes_list corresponding to the 'index'. */
1480         list_for_each_entry(lsi, &sdp->sd_sc_inodes_list, si_list) {
1481                 if (lsi->si_jid == index)
1482                         return lsi->si_sc_inode;
1483         }
1484         return NULL;
1485 }
1486
1487 const struct super_operations gfs2_super_ops = {
1488         .alloc_inode            = gfs2_alloc_inode,
1489         .free_inode             = gfs2_free_inode,
1490         .write_inode            = gfs2_write_inode,
1491         .dirty_inode            = gfs2_dirty_inode,
1492         .evict_inode            = gfs2_evict_inode,
1493         .put_super              = gfs2_put_super,
1494         .sync_fs                = gfs2_sync_fs,
1495         .freeze_super           = gfs2_freeze,
1496         .thaw_super             = gfs2_unfreeze,
1497         .statfs                 = gfs2_statfs,
1498         .drop_inode             = gfs2_drop_inode,
1499         .show_options           = gfs2_show_options,
1500 };
1501