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