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         int found = 0;
  85
  86         list_for_each_entry(jd, head, jd_list) {
  87                 if (jd->jd_jid == jid) {
  88                         found = 1;
  89                         break;
  90                 }
  91         }
  92
  93         if (!found)
  94                 jd = NULL;
  95
  96         return jd;
  97 }
  98
  99 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
 100 {
 101         struct gfs2_jdesc *jd;
 102
 103         spin_lock(&sdp->sd_jindex_spin);
 104         jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
 105         spin_unlock(&sdp->sd_jindex_spin);
 106
 107         return jd;
 108 }
 109
 110 int gfs2_jdesc_check(struct gfs2_jdesc *jd)
 111 {
 112         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 113         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 114         u64 size = i_size_read(jd->jd_inode);
 115
 116         if (gfs2_check_internal_file_size(jd->jd_inode, 8 << 20, BIT(30)))
 117                 return -EIO;
 118
 119         jd->jd_blocks = size >> sdp->sd_sb.sb_bsize_shift;
 120
 121         if (gfs2_write_alloc_required(ip, 0, size)) {
 122                 gfs2_consist_inode(ip);
 123                 return -EIO;
 124         }
 125
 126         return 0;
 127 }
 128
 129 /**
 130  * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
 131  * @sdp: the filesystem
 132  *
 133  * Returns: errno
 134  */
 135
 136 int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
 137 {
 138         struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
 139         struct gfs2_glock *j_gl = ip->i_gl;
 140         struct gfs2_log_header_host head;
 141         int error;
 142
 143         j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
 144         if (gfs2_withdrawn(sdp))
 145                 return -EIO;
 146
 147         error = gfs2_find_jhead(sdp->sd_jdesc, &head, false);
 148         if (error) {
 149                 gfs2_consist(sdp);
 150                 return error;
 151         }
 152
 153         if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 154                 gfs2_consist(sdp);
 155                 return -EIO;
 156         }
 157
 158         /*  Initialize some head of the log stuff  */
 159         sdp->sd_log_sequence = head.lh_sequence + 1;
 160         gfs2_log_pointers_init(sdp, head.lh_blkno);
 161
 162         error = gfs2_quota_init(sdp);
 163         if (!error && gfs2_withdrawn(sdp))
 164                 error = -EIO;
 165         if (!error)
 166                 set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 167         return error;
 168 }
 169
 170 void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
 171 {
 172         const struct gfs2_statfs_change *str = buf;
 173
 174         sc->sc_total = be64_to_cpu(str->sc_total);
 175         sc->sc_free = be64_to_cpu(str->sc_free);
 176         sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
 177 }
 178
 179 void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
 180 {
 181         struct gfs2_statfs_change *str = buf;
 182
 183         str->sc_total = cpu_to_be64(sc->sc_total);
 184         str->sc_free = cpu_to_be64(sc->sc_free);
 185         str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
 186 }
 187
 188 int gfs2_statfs_init(struct gfs2_sbd *sdp)
 189 {
 190         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 191         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 192         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 193         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 194         struct buffer_head *m_bh, *l_bh;
 195         struct gfs2_holder gh;
 196         int error;
 197
 198         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 199                                    &gh);
 200         if (error)
 201                 return error;
 202
 203         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 204         if (error)
 205                 goto out;
 206
 207         if (sdp->sd_args.ar_spectator) {
 208                 spin_lock(&sdp->sd_statfs_spin);
 209                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 210                                       sizeof(struct gfs2_dinode));
 211                 spin_unlock(&sdp->sd_statfs_spin);
 212         } else {
 213                 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 214                 if (error)
 215                         goto out_m_bh;
 216
 217                 spin_lock(&sdp->sd_statfs_spin);
 218                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 219                                       sizeof(struct gfs2_dinode));
 220                 gfs2_statfs_change_in(l_sc, l_bh->b_data +
 221                                       sizeof(struct gfs2_dinode));
 222                 spin_unlock(&sdp->sd_statfs_spin);
 223
 224                 brelse(l_bh);
 225         }
 226
 227 out_m_bh:
 228         brelse(m_bh);
 229 out:
 230         gfs2_glock_dq_uninit(&gh);
 231         return 0;
 232 }
 233
 234 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
 235                         s64 dinodes)
 236 {
 237         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 238         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 239         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 240         struct buffer_head *l_bh;
 241         s64 x, y;
 242         int need_sync = 0;
 243         int error;
 244
 245         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 246         if (error)
 247                 return;
 248
 249         gfs2_trans_add_meta(l_ip->i_gl, l_bh);
 250
 251         spin_lock(&sdp->sd_statfs_spin);
 252         l_sc->sc_total += total;
 253         l_sc->sc_free += free;
 254         l_sc->sc_dinodes += dinodes;
 255         gfs2_statfs_change_out(l_sc, l_bh->b_data + sizeof(struct gfs2_dinode));
 256         if (sdp->sd_args.ar_statfs_percent) {
 257                 x = 100 * l_sc->sc_free;
 258                 y = m_sc->sc_free * sdp->sd_args.ar_statfs_percent;
 259                 if (x >= y || x <= -y)
 260                         need_sync = 1;
 261         }
 262         spin_unlock(&sdp->sd_statfs_spin);
 263
 264         brelse(l_bh);
 265         if (need_sync)
 266                 gfs2_wake_up_statfs(sdp);
 267 }
 268
 269 void update_statfs(struct gfs2_sbd *sdp, struct buffer_head *m_bh,
 270                    struct buffer_head *l_bh)
 271 {
 272         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 273         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 274         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 275         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 276
 277         gfs2_trans_add_meta(l_ip->i_gl, l_bh);
 278         gfs2_trans_add_meta(m_ip->i_gl, m_bh);
 279
 280         spin_lock(&sdp->sd_statfs_spin);
 281         m_sc->sc_total += l_sc->sc_total;
 282         m_sc->sc_free += l_sc->sc_free;
 283         m_sc->sc_dinodes += l_sc->sc_dinodes;
 284         memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
 285         memset(l_bh->b_data + sizeof(struct gfs2_dinode),
 286                0, sizeof(struct gfs2_statfs_change));
 287         gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
 288         spin_unlock(&sdp->sd_statfs_spin);
 289 }
 290
 291 int gfs2_statfs_sync(struct super_block *sb, int type)
 292 {
 293         struct gfs2_sbd *sdp = sb->s_fs_info;
 294         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 295         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 296         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 297         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 298         struct gfs2_holder gh;
 299         struct buffer_head *m_bh, *l_bh;
 300         int error;
 301
 302         sb_start_write(sb);
 303         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 304                                    &gh);
 305         if (error)
 306                 goto out;
 307
 308         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 309         if (error)
 310                 goto out_unlock;
 311
 312         spin_lock(&sdp->sd_statfs_spin);
 313         gfs2_statfs_change_in(m_sc, m_bh->b_data +
 314                               sizeof(struct gfs2_dinode));
 315         if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
 316                 spin_unlock(&sdp->sd_statfs_spin);
 317                 goto out_bh;
 318         }
 319         spin_unlock(&sdp->sd_statfs_spin);
 320
 321         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 322         if (error)
 323                 goto out_bh;
 324
 325         error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
 326         if (error)
 327                 goto out_bh2;
 328
 329         update_statfs(sdp, m_bh, l_bh);
 330         sdp->sd_statfs_force_sync = 0;
 331
 332         gfs2_trans_end(sdp);
 333
 334 out_bh2:
 335         brelse(l_bh);
 336 out_bh:
 337         brelse(m_bh);
 338 out_unlock:
 339         gfs2_glock_dq_uninit(&gh);
 340 out:
 341         sb_end_write(sb);
 342         return error;
 343 }
 344
 345 struct lfcc {
 346         struct list_head list;
 347         struct gfs2_holder gh;
 348 };
 349
 350 /**
 351  * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
 352  *                            journals are clean
 353  * @sdp: the file system
 354  * @state: the state to put the transaction lock into
 355  * @t_gh: the hold on the transaction lock
 356  *
 357  * Returns: errno
 358  */
 359
 360 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp)
 361 {
 362         struct gfs2_inode *ip;
 363         struct gfs2_jdesc *jd;
 364         struct lfcc *lfcc;
 365         LIST_HEAD(list);
 366         struct gfs2_log_header_host lh;
 367         int error;
 368
 369         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 370                 lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
 371                 if (!lfcc) {
 372                         error = -ENOMEM;
 373                         goto out;
 374                 }
 375                 ip = GFS2_I(jd->jd_inode);
 376                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
 377                 if (error) {
 378                         kfree(lfcc);
 379                         goto out;
 380                 }
 381                 list_add(&lfcc->list, &list);
 382         }
 383
 384         error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_EXCLUSIVE,
 385                                    LM_FLAG_NOEXP, &sdp->sd_freeze_gh);
 386         if (error)
 387                 goto out;
 388
 389         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 390                 error = gfs2_jdesc_check(jd);
 391                 if (error)
 392                         break;
 393                 error = gfs2_find_jhead(jd, &lh, false);
 394                 if (error)
 395                         break;
 396                 if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 397                         error = -EBUSY;
 398                         break;
 399                 }
 400         }
 401
 402         if (error)
 403                 gfs2_freeze_unlock(&sdp->sd_freeze_gh);
 404
 405 out:
 406         while (!list_empty(&list)) {
 407                 lfcc = list_first_entry(&list, struct lfcc, list);
 408                 list_del(&lfcc->list);
 409                 gfs2_glock_dq_uninit(&lfcc->gh);
 410                 kfree(lfcc);
 411         }
 412         return error;
 413 }
 414
 415 void gfs2_dinode_out(const struct gfs2_inode *ip, void *buf)
 416 {
 417         struct gfs2_dinode *str = buf;
 418
 419         str->di_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 420         str->di_header.mh_type = cpu_to_be32(GFS2_METATYPE_DI);
 421         str->di_header.mh_format = cpu_to_be32(GFS2_FORMAT_DI);
 422         str->di_num.no_addr = cpu_to_be64(ip->i_no_addr);
 423         str->di_num.no_formal_ino = cpu_to_be64(ip->i_no_formal_ino);
 424         str->di_mode = cpu_to_be32(ip->i_inode.i_mode);
 425         str->di_uid = cpu_to_be32(i_uid_read(&ip->i_inode));
 426         str->di_gid = cpu_to_be32(i_gid_read(&ip->i_inode));
 427         str->di_nlink = cpu_to_be32(ip->i_inode.i_nlink);
 428         str->di_size = cpu_to_be64(i_size_read(&ip->i_inode));
 429         str->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(&ip->i_inode));
 430         str->di_atime = cpu_to_be64(ip->i_inode.i_atime.tv_sec);
 431         str->di_mtime = cpu_to_be64(ip->i_inode.i_mtime.tv_sec);
 432         str->di_ctime = cpu_to_be64(ip->i_inode.i_ctime.tv_sec);
 433
 434         str->di_goal_meta = cpu_to_be64(ip->i_goal);
 435         str->di_goal_data = cpu_to_be64(ip->i_goal);
 436         str->di_generation = cpu_to_be64(ip->i_generation);
 437
 438         str->di_flags = cpu_to_be32(ip->i_diskflags);
 439         str->di_height = cpu_to_be16(ip->i_height);
 440         str->di_payload_format = cpu_to_be32(S_ISDIR(ip->i_inode.i_mode) &&
 441                                              !(ip->i_diskflags & GFS2_DIF_EXHASH) ?
 442                                              GFS2_FORMAT_DE : 0);
 443         str->di_depth = cpu_to_be16(ip->i_depth);
 444         str->di_entries = cpu_to_be32(ip->i_entries);
 445
 446         str->di_eattr = cpu_to_be64(ip->i_eattr);
 447         str->di_atime_nsec = cpu_to_be32(ip->i_inode.i_atime.tv_nsec);
 448         str->di_mtime_nsec = cpu_to_be32(ip->i_inode.i_mtime.tv_nsec);
 449         str->di_ctime_nsec = cpu_to_be32(ip->i_inode.i_ctime.tv_nsec);
 450 }
 451
 452 /**
 453  * gfs2_write_inode - Make sure the inode is stable on the disk
 454  * @inode: The inode
 455  * @wbc: The writeback control structure
 456  *
 457  * Returns: errno
 458  */
 459
 460 static int gfs2_write_inode(struct inode *inode, struct writeback_control *wbc)
 461 {
 462         struct gfs2_inode *ip = GFS2_I(inode);
 463         struct gfs2_sbd *sdp = GFS2_SB(inode);
 464         struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl);
 465         struct backing_dev_info *bdi = inode_to_bdi(metamapping->host);
 466         int ret = 0;
 467         bool flush_all = (wbc->sync_mode == WB_SYNC_ALL || gfs2_is_jdata(ip));
 468
 469         if (flush_all)
 470                 gfs2_log_flush(GFS2_SB(inode), ip->i_gl,
 471                                GFS2_LOG_HEAD_FLUSH_NORMAL |
 472                                GFS2_LFC_WRITE_INODE);
 473         if (bdi->wb.dirty_exceeded)
 474                 gfs2_ail1_flush(sdp, wbc);
 475         else
 476                 filemap_fdatawrite(metamapping);
 477         if (flush_all)
 478                 ret = filemap_fdatawait(metamapping);
 479         if (ret)
 480                 mark_inode_dirty_sync(inode);
 481         else {
 482                 spin_lock(&inode->i_lock);
 483                 if (!(inode->i_flags & I_DIRTY))
 484                         gfs2_ordered_del_inode(ip);
 485                 spin_unlock(&inode->i_lock);
 486         }
 487         return ret;
 488 }
 489
 490 /**
 491  * gfs2_dirty_inode - check for atime updates
 492  * @inode: The inode in question
 493  * @flags: The type of dirty
 494  *
 495  * Unfortunately it can be called under any combination of inode
 496  * glock and transaction lock, so we have to check carefully.
 497  *
 498  * At the moment this deals only with atime - it should be possible
 499  * to expand that role in future, once a review of the locking has
 500  * been carried out.
 501  */
 502
 503 static void gfs2_dirty_inode(struct inode *inode, int flags)
 504 {
 505         struct gfs2_inode *ip = GFS2_I(inode);
 506         struct gfs2_sbd *sdp = GFS2_SB(inode);
 507         struct buffer_head *bh;
 508         struct gfs2_holder gh;
 509         int need_unlock = 0;
 510         int need_endtrans = 0;
 511         int ret;
 512
 513         if (!(flags & I_DIRTY_INODE))
 514                 return;
 515         if (unlikely(gfs2_withdrawn(sdp)))
 516                 return;
 517         if (!gfs2_glock_is_locked_by_me(ip->i_gl)) {
 518                 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
 519                 if (ret) {
 520                         fs_err(sdp, "dirty_inode: glock %d\n", ret);
 521                         gfs2_dump_glock(NULL, ip->i_gl, true);
 522                         return;
 523                 }
 524                 need_unlock = 1;
 525         } else if (WARN_ON_ONCE(ip->i_gl->gl_state != LM_ST_EXCLUSIVE))
 526                 return;
 527
 528         if (current->journal_info == NULL) {
 529                 ret = gfs2_trans_begin(sdp, RES_DINODE, 0);
 530                 if (ret) {
 531                         fs_err(sdp, "dirty_inode: gfs2_trans_begin %d\n", ret);
 532                         goto out;
 533                 }
 534                 need_endtrans = 1;
 535         }
 536
 537         ret = gfs2_meta_inode_buffer(ip, &bh);
 538         if (ret == 0) {
 539                 gfs2_trans_add_meta(ip->i_gl, bh);
 540                 gfs2_dinode_out(ip, bh->b_data);
 541                 brelse(bh);
 542         }
 543
 544         if (need_endtrans)
 545                 gfs2_trans_end(sdp);
 546 out:
 547         if (need_unlock)
 548                 gfs2_glock_dq_uninit(&gh);
 549 }
 550
 551 /**
 552  * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
 553  * @sdp: the filesystem
 554  *
 555  * Returns: errno
 556  */
 557
 558 int gfs2_make_fs_ro(struct gfs2_sbd *sdp)
 559 {
 560         int error = 0;
 561         int log_write_allowed = test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 562
 563         gfs2_flush_delete_work(sdp);
 564         if (!log_write_allowed && current == sdp->sd_quotad_process)
 565                 fs_warn(sdp, "The quotad daemon is withdrawing.\n");
 566         else if (sdp->sd_quotad_process)
 567                 kthread_stop(sdp->sd_quotad_process);
 568         sdp->sd_quotad_process = NULL;
 569
 570         if (!log_write_allowed && current == sdp->sd_logd_process)
 571                 fs_warn(sdp, "The logd daemon is withdrawing.\n");
 572         else if (sdp->sd_logd_process)
 573                 kthread_stop(sdp->sd_logd_process);
 574         sdp->sd_logd_process = NULL;
 575
 576         if (log_write_allowed) {
 577                 gfs2_quota_sync(sdp->sd_vfs, 0);
 578                 gfs2_statfs_sync(sdp->sd_vfs, 0);
 579
 580                 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SHUTDOWN |
 581                                GFS2_LFC_MAKE_FS_RO);
 582                 wait_event(sdp->sd_reserving_log_wait,
 583                            atomic_read(&sdp->sd_reserving_log) == 0);
 584                 gfs2_assert_warn(sdp, atomic_read(&sdp->sd_log_blks_free) ==
 585                                  sdp->sd_jdesc->jd_blocks);
 586         } else {
 587                 wait_event_timeout(sdp->sd_reserving_log_wait,
 588                                    atomic_read(&sdp->sd_reserving_log) == 0,
 589                                    HZ * 5);
 590         }
 591         gfs2_quota_cleanup(sdp);
 592
 593         if (!log_write_allowed)
 594                 sdp->sd_vfs->s_flags |= SB_RDONLY;
 595
 596         return error;
 597 }
 598
 599 /**
 600  * gfs2_put_super - Unmount the filesystem
 601  * @sb: The VFS superblock
 602  *
 603  */
 604
 605 static void gfs2_put_super(struct super_block *sb)
 606 {
 607         struct gfs2_sbd *sdp = sb->s_fs_info;
 608         int error;
 609         struct gfs2_jdesc *jd;
 610
 611         /* No more recovery requests */
 612         set_bit(SDF_NORECOVERY, &sdp->sd_flags);
 613         smp_mb();
 614
 615         /* Wait on outstanding recovery */
 616 restart:
 617         spin_lock(&sdp->sd_jindex_spin);
 618         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 619                 if (!test_bit(JDF_RECOVERY, &jd->jd_flags))
 620                         continue;
 621                 spin_unlock(&sdp->sd_jindex_spin);
 622                 wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
 623                             TASK_UNINTERRUPTIBLE);
 624                 goto restart;
 625         }
 626         spin_unlock(&sdp->sd_jindex_spin);
 627
 628         if (!sb_rdonly(sb)) {
 629                 error = gfs2_make_fs_ro(sdp);
 630                 if (error)
 631                         gfs2_io_error(sdp);
 632         }
 633         WARN_ON(gfs2_withdrawing(sdp));
 634
 635         /*  At this point, we're through modifying the disk  */
 636
 637         /*  Release stuff  */
 638
 639         iput(sdp->sd_jindex);
 640         iput(sdp->sd_statfs_inode);
 641         iput(sdp->sd_rindex);
 642         iput(sdp->sd_quota_inode);
 643
 644         gfs2_glock_put(sdp->sd_rename_gl);
 645         gfs2_glock_put(sdp->sd_freeze_gl);
 646
 647         if (!sdp->sd_args.ar_spectator) {
 648                 if (gfs2_holder_initialized(&sdp->sd_journal_gh))
 649                         gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
 650                 if (gfs2_holder_initialized(&sdp->sd_jinode_gh))
 651                         gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
 652                 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
 653                 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
 654                 free_local_statfs_inodes(sdp);
 655                 iput(sdp->sd_qc_inode);
 656         }
 657
 658         gfs2_glock_dq_uninit(&sdp->sd_live_gh);
 659         gfs2_clear_rgrpd(sdp);
 660         gfs2_jindex_free(sdp);
 661         /*  Take apart glock structures and buffer lists  */
 662         gfs2_gl_hash_clear(sdp);
 663         truncate_inode_pages_final(&sdp->sd_aspace);
 664         gfs2_delete_debugfs_file(sdp);
 665         /*  Unmount the locking protocol  */
 666         gfs2_lm_unmount(sdp);
 667
 668         /*  At this point, we're through participating in the lockspace  */
 669         gfs2_sys_fs_del(sdp);
 670         free_sbd(sdp);
 671 }
 672
 673 /**
 674  * gfs2_sync_fs - sync the filesystem
 675  * @sb: the superblock
 676  *
 677  * Flushes the log to disk.
 678  */
 679
 680 static int gfs2_sync_fs(struct super_block *sb, int wait)
 681 {
 682         struct gfs2_sbd *sdp = sb->s_fs_info;
 683
 684         gfs2_quota_sync(sb, -1);
 685         if (wait)
 686                 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
 687                                GFS2_LFC_SYNC_FS);
 688         return sdp->sd_log_error;
 689 }
 690
 691 void gfs2_freeze_func(struct work_struct *work)
 692 {
 693         int error;
 694         struct gfs2_holder freeze_gh;
 695         struct gfs2_sbd *sdp = container_of(work, struct gfs2_sbd, sd_freeze_work);
 696         struct super_block *sb = sdp->sd_vfs;
 697
 698         atomic_inc(&sb->s_active);
 699         error = gfs2_freeze_lock(sdp, &freeze_gh, 0);
 700         if (error) {
 701                 gfs2_assert_withdraw(sdp, 0);
 702         } else {
 703                 atomic_set(&sdp->sd_freeze_state, SFS_UNFROZEN);
 704                 error = thaw_super(sb);
 705                 if (error) {
 706                         fs_info(sdp, "GFS2: couldn't thaw filesystem: %d\n",
 707                                 error);
 708                         gfs2_assert_withdraw(sdp, 0);
 709                 }
 710                 gfs2_freeze_unlock(&freeze_gh);
 711         }
 712         deactivate_super(sb);
 713         clear_bit_unlock(SDF_FS_FROZEN, &sdp->sd_flags);
 714         wake_up_bit(&sdp->sd_flags, SDF_FS_FROZEN);
 715         return;
 716 }
 717
 718 /**
 719  * gfs2_freeze - prevent further writes to the filesystem
 720  * @sb: the VFS structure for the filesystem
 721  *
 722  */
 723
 724 static int gfs2_freeze(struct super_block *sb)
 725 {
 726         struct gfs2_sbd *sdp = sb->s_fs_info;
 727         int error;
 728
 729         mutex_lock(&sdp->sd_freeze_mutex);
 730         if (atomic_read(&sdp->sd_freeze_state) != SFS_UNFROZEN) {
 731                 error = -EBUSY;
 732                 goto out;
 733         }
 734
 735         for (;;) {
 736                 if (gfs2_withdrawn(sdp)) {
 737                         error = -EINVAL;
 738                         goto out;
 739                 }
 740
 741                 error = gfs2_lock_fs_check_clean(sdp);
 742                 if (!error)
 743                         break;
 744
 745                 if (error == -EBUSY)
 746                         fs_err(sdp, "waiting for recovery before freeze\n");
 747                 else if (error == -EIO) {
 748                         fs_err(sdp, "Fatal IO error: cannot freeze gfs2 due "
 749                                "to recovery error.\n");
 750                         goto out;
 751                 } else {
 752                         fs_err(sdp, "error freezing FS: %d\n", error);
 753                 }
 754                 fs_err(sdp, "retrying...\n");
 755                 msleep(1000);
 756         }
 757         set_bit(SDF_FS_FROZEN, &sdp->sd_flags);
 758 out:
 759         mutex_unlock(&sdp->sd_freeze_mutex);
 760         return error;
 761 }
 762
 763 /**
 764  * gfs2_unfreeze - reallow writes to the filesystem
 765  * @sb: the VFS structure for the filesystem
 766  *
 767  */
 768
 769 static int gfs2_unfreeze(struct super_block *sb)
 770 {
 771         struct gfs2_sbd *sdp = sb->s_fs_info;
 772
 773         mutex_lock(&sdp->sd_freeze_mutex);
 774         if (atomic_read(&sdp->sd_freeze_state) != SFS_FROZEN ||
 775             !gfs2_holder_initialized(&sdp->sd_freeze_gh)) {
 776                 mutex_unlock(&sdp->sd_freeze_mutex);
 777                 return -EINVAL;
 778         }
 779
 780         gfs2_freeze_unlock(&sdp->sd_freeze_gh);
 781         mutex_unlock(&sdp->sd_freeze_mutex);
 782         return wait_on_bit(&sdp->sd_flags, SDF_FS_FROZEN, TASK_INTERRUPTIBLE);
 783 }
 784
 785 /**
 786  * statfs_fill - fill in the sg for a given RG
 787  * @rgd: the RG
 788  * @sc: the sc structure
 789  *
 790  * Returns: 0 on success, -ESTALE if the LVB is invalid
 791  */
 792
 793 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
 794                             struct gfs2_statfs_change_host *sc)
 795 {
 796         gfs2_rgrp_verify(rgd);
 797         sc->sc_total += rgd->rd_data;
 798         sc->sc_free += rgd->rd_free;
 799         sc->sc_dinodes += rgd->rd_dinodes;
 800         return 0;
 801 }
 802
 803 /**
 804  * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
 805  * @sdp: the filesystem
 806  * @sc: the sc info that will be returned
 807  *
 808  * Any error (other than a signal) will cause this routine to fall back
 809  * to the synchronous version.
 810  *
 811  * FIXME: This really shouldn't busy wait like this.
 812  *
 813  * Returns: errno
 814  */
 815
 816 static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
 817 {
 818         struct gfs2_rgrpd *rgd_next;
 819         struct gfs2_holder *gha, *gh;
 820         unsigned int slots = 64;
 821         unsigned int x;
 822         int done;
 823         int error = 0, err;
 824
 825         memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
 826         gha = kmalloc_array(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
 827         if (!gha)
 828                 return -ENOMEM;
 829         for (x = 0; x < slots; x++)
 830                 gfs2_holder_mark_uninitialized(gha + x);
 831
 832         rgd_next = gfs2_rgrpd_get_first(sdp);
 833
 834         for (;;) {
 835                 done = 1;
 836
 837                 for (x = 0; x < slots; x++) {
 838                         gh = gha + x;
 839
 840                         if (gfs2_holder_initialized(gh) && gfs2_glock_poll(gh)) {
 841                                 err = gfs2_glock_wait(gh);
 842                                 if (err) {
 843                                         gfs2_holder_uninit(gh);
 844                                         error = err;
 845                                 } else {
 846                                         if (!error) {
 847                                                 struct gfs2_rgrpd *rgd =
 848                                                         gfs2_glock2rgrp(gh->gh_gl);
 849
 850                                                 error = statfs_slow_fill(rgd, sc);
 851                                         }
 852                                         gfs2_glock_dq_uninit(gh);
 853                                 }
 854                         }
 855
 856                         if (gfs2_holder_initialized(gh))
 857                                 done = 0;
 858                         else if (rgd_next && !error) {
 859                                 error = gfs2_glock_nq_init(rgd_next->rd_gl,
 860                                                            LM_ST_SHARED,
 861                                                            GL_ASYNC,
 862                                                            gh);
 863                                 rgd_next = gfs2_rgrpd_get_next(rgd_next);
 864                                 done = 0;
 865                         }
 866
 867                         if (signal_pending(current))
 868                                 error = -ERESTARTSYS;
 869                 }
 870
 871                 if (done)
 872                         break;
 873
 874                 yield();
 875         }
 876
 877         kfree(gha);
 878         return error;
 879 }
 880
 881 /**
 882  * gfs2_statfs_i - Do a statfs
 883  * @sdp: the filesystem
 884  * @sg: the sg structure
 885  *
 886  * Returns: errno
 887  */
 888
 889 static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
 890 {
 891         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 892         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 893
 894         spin_lock(&sdp->sd_statfs_spin);
 895
 896         *sc = *m_sc;
 897         sc->sc_total += l_sc->sc_total;
 898         sc->sc_free += l_sc->sc_free;
 899         sc->sc_dinodes += l_sc->sc_dinodes;
 900
 901         spin_unlock(&sdp->sd_statfs_spin);
 902
 903         if (sc->sc_free < 0)
 904                 sc->sc_free = 0;
 905         if (sc->sc_free > sc->sc_total)
 906                 sc->sc_free = sc->sc_total;
 907         if (sc->sc_dinodes < 0)
 908                 sc->sc_dinodes = 0;
 909
 910         return 0;
 911 }
 912
 913 /**
 914  * gfs2_statfs - Gather and return stats about the filesystem
 915  * @sb: The superblock
 916  * @statfsbuf: The buffer
 917  *
 918  * Returns: 0 on success or error code
 919  */
 920
 921 static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
 922 {
 923         struct super_block *sb = dentry->d_sb;
 924         struct gfs2_sbd *sdp = sb->s_fs_info;
 925         struct gfs2_statfs_change_host sc;
 926         int error;
 927
 928         error = gfs2_rindex_update(sdp);
 929         if (error)
 930                 return error;
 931
 932         if (gfs2_tune_get(sdp, gt_statfs_slow))
 933                 error = gfs2_statfs_slow(sdp, &sc);
 934         else
 935                 error = gfs2_statfs_i(sdp, &sc);
 936
 937         if (error)
 938                 return error;
 939
 940         buf->f_type = GFS2_MAGIC;
 941         buf->f_bsize = sdp->sd_sb.sb_bsize;
 942         buf->f_blocks = sc.sc_total;
 943         buf->f_bfree = sc.sc_free;
 944         buf->f_bavail = sc.sc_free;
 945         buf->f_files = sc.sc_dinodes + sc.sc_free;
 946         buf->f_ffree = sc.sc_free;
 947         buf->f_namelen = GFS2_FNAMESIZE;
 948
 949         return 0;
 950 }
 951
 952 /**
 953  * gfs2_drop_inode - Drop an inode (test for remote unlink)
 954  * @inode: The inode to drop
 955  *
 956  * If we've received a callback on an iopen lock then it's because a
 957  * remote node tried to deallocate the inode but failed due to this node
 958  * still having the inode open. Here we mark the link count zero
 959  * since we know that it must have reached zero if the GLF_DEMOTE flag
 960  * is set on the iopen glock. If we didn't do a disk read since the
 961  * remote node removed the final link then we might otherwise miss
 962  * this event. This check ensures that this node will deallocate the
 963  * inode's blocks, or alternatively pass the baton on to another
 964  * node for later deallocation.
 965  */
 966
 967 static int gfs2_drop_inode(struct inode *inode)
 968 {
 969         struct gfs2_inode *ip = GFS2_I(inode);
 970
 971         if (!test_bit(GIF_FREE_VFS_INODE, &ip->i_flags) &&
 972             inode->i_nlink &&
 973             gfs2_holder_initialized(&ip->i_iopen_gh)) {
 974                 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
 975                 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
 976                         clear_nlink(inode);
 977         }
 978
 979         /*
 980          * When under memory pressure when an inode's link count has dropped to
 981          * zero, defer deleting the inode to the delete workqueue.  This avoids
 982          * calling into DLM under memory pressure, which can deadlock.
 983          */
 984         if (!inode->i_nlink &&
 985             unlikely(current->flags & PF_MEMALLOC) &&
 986             gfs2_holder_initialized(&ip->i_iopen_gh)) {
 987                 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
 988
 989                 gfs2_glock_hold(gl);
 990                 if (!gfs2_queue_delete_work(gl, 0))
 991                         gfs2_glock_queue_put(gl);
 992                 return false;
 993         }
 994
 995         return generic_drop_inode(inode);
 996 }
 997
 998 static int is_ancestor(const struct dentry *d1, const struct dentry *d2)
 999 {
1000         do {
1001                 if (d1 == d2)
1002                         return 1;
1003                 d1 = d1->d_parent;
1004         } while (!IS_ROOT(d1));
1005         return 0;
1006 }
1007
1008 /**
1009  * gfs2_show_options - Show mount options for /proc/mounts
1010  * @s: seq_file structure
1011  * @root: root of this (sub)tree
1012  *
1013  * Returns: 0 on success or error code
1014  */
1015
1016 static int gfs2_show_options(struct seq_file *s, struct dentry *root)
1017 {
1018         struct gfs2_sbd *sdp = root->d_sb->s_fs_info;
1019         struct gfs2_args *args = &sdp->sd_args;
1020         unsigned int logd_secs, statfs_slow, statfs_quantum, quota_quantum;
1021
1022         spin_lock(&sdp->sd_tune.gt_spin);
1023         logd_secs = sdp->sd_tune.gt_logd_secs;
1024         quota_quantum = sdp->sd_tune.gt_quota_quantum;
1025         statfs_quantum = sdp->sd_tune.gt_statfs_quantum;
1026         statfs_slow = sdp->sd_tune.gt_statfs_slow;
1027         spin_unlock(&sdp->sd_tune.gt_spin);
1028
1029         if (is_ancestor(root, sdp->sd_master_dir))
1030                 seq_puts(s, ",meta");
1031         if (args->ar_lockproto[0])
1032                 seq_show_option(s, "lockproto", args->ar_lockproto);
1033         if (args->ar_locktable[0])
1034                 seq_show_option(s, "locktable", args->ar_locktable);
1035         if (args->ar_hostdata[0])
1036                 seq_show_option(s, "hostdata", args->ar_hostdata);
1037         if (args->ar_spectator)
1038                 seq_puts(s, ",spectator");
1039         if (args->ar_localflocks)
1040                 seq_puts(s, ",localflocks");
1041         if (args->ar_debug)
1042                 seq_puts(s, ",debug");
1043         if (args->ar_posix_acl)
1044                 seq_puts(s, ",acl");
1045         if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
1046                 char *state;
1047                 switch (args->ar_quota) {
1048                 case GFS2_QUOTA_OFF:
1049                         state = "off";
1050                         break;
1051                 case GFS2_QUOTA_ACCOUNT:
1052                         state = "account";
1053                         break;
1054                 case GFS2_QUOTA_ON:
1055                         state = "on";
1056                         break;
1057                 default:
1058                         state = "unknown";
1059                         break;
1060                 }
1061                 seq_printf(s, ",quota=%s", state);
1062         }
1063         if (args->ar_suiddir)
1064                 seq_puts(s, ",suiddir");
1065         if (args->ar_data != GFS2_DATA_DEFAULT) {
1066                 char *state;
1067                 switch (args->ar_data) {
1068                 case GFS2_DATA_WRITEBACK:
1069                         state = "writeback";
1070                         break;
1071                 case GFS2_DATA_ORDERED:
1072                         state = "ordered";
1073                         break;
1074                 default:
1075                         state = "unknown";
1076                         break;
1077                 }
1078                 seq_printf(s, ",data=%s", state);
1079         }
1080         if (args->ar_discard)
1081                 seq_puts(s, ",discard");
1082         if (logd_secs != 30)
1083                 seq_printf(s, ",commit=%d", logd_secs);
1084         if (statfs_quantum != 30)
1085                 seq_printf(s, ",statfs_quantum=%d", statfs_quantum);
1086         else if (statfs_slow)
1087                 seq_puts(s, ",statfs_quantum=0");
1088         if (quota_quantum != 60)
1089                 seq_printf(s, ",quota_quantum=%d", quota_quantum);
1090         if (args->ar_statfs_percent)
1091                 seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent);
1092         if (args->ar_errors != GFS2_ERRORS_DEFAULT) {
1093                 const char *state;
1094
1095                 switch (args->ar_errors) {
1096                 case GFS2_ERRORS_WITHDRAW:
1097                         state = "withdraw";
1098                         break;
1099                 case GFS2_ERRORS_PANIC:
1100                         state = "panic";
1101                         break;
1102                 default:
1103                         state = "unknown";
1104                         break;
1105                 }
1106                 seq_printf(s, ",errors=%s", state);
1107         }
1108         if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags))
1109                 seq_puts(s, ",nobarrier");
1110         if (test_bit(SDF_DEMOTE, &sdp->sd_flags))
1111                 seq_puts(s, ",demote_interface_used");
1112         if (args->ar_rgrplvb)
1113                 seq_puts(s, ",rgrplvb");
1114         if (args->ar_loccookie)
1115                 seq_puts(s, ",loccookie");
1116         return 0;
1117 }
1118
1119 static void gfs2_final_release_pages(struct gfs2_inode *ip)
1120 {
1121         struct inode *inode = &ip->i_inode;
1122         struct gfs2_glock *gl = ip->i_gl;
1123
1124         truncate_inode_pages(gfs2_glock2aspace(ip->i_gl), 0);
1125         truncate_inode_pages(&inode->i_data, 0);
1126
1127         if (atomic_read(&gl->gl_revokes) == 0) {
1128                 clear_bit(GLF_LFLUSH, &gl->gl_flags);
1129                 clear_bit(GLF_DIRTY, &gl->gl_flags);
1130         }
1131 }
1132
1133 static int gfs2_dinode_dealloc(struct gfs2_inode *ip)
1134 {
1135         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1136         struct gfs2_rgrpd *rgd;
1137         struct gfs2_holder gh;
1138         int error;
1139
1140         if (gfs2_get_inode_blocks(&ip->i_inode) != 1) {
1141                 gfs2_consist_inode(ip);
1142                 return -EIO;
1143         }
1144
1145         error = gfs2_rindex_update(sdp);
1146         if (error)
1147                 return error;
1148
1149         error = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
1150         if (error)
1151                 return error;
1152
1153         rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1154         if (!rgd) {
1155                 gfs2_consist_inode(ip);
1156                 error = -EIO;
1157                 goto out_qs;
1158         }
1159
1160         error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &gh);
1161         if (error)
1162                 goto out_qs;
1163
1164         error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA,
1165                                  sdp->sd_jdesc->jd_blocks);
1166         if (error)
1167                 goto out_rg_gunlock;
1168
1169         gfs2_free_di(rgd, ip);
1170
1171         gfs2_final_release_pages(ip);
1172
1173         gfs2_trans_end(sdp);
1174
1175 out_rg_gunlock:
1176         gfs2_glock_dq_uninit(&gh);
1177 out_qs:
1178         gfs2_quota_unhold(ip);
1179         return error;
1180 }
1181
1182 /**
1183  * gfs2_glock_put_eventually
1184  * @gl: The glock to put
1185  *
1186  * When under memory pressure, trigger a deferred glock put to make sure we
1187  * won't call into DLM and deadlock.  Otherwise, put the glock directly.
1188  */
1189
1190 static void gfs2_glock_put_eventually(struct gfs2_glock *gl)
1191 {
1192         if (current->flags & PF_MEMALLOC)
1193                 gfs2_glock_queue_put(gl);
1194         else
1195                 gfs2_glock_put(gl);
1196 }
1197
1198 static bool gfs2_upgrade_iopen_glock(struct inode *inode)
1199 {
1200         struct gfs2_inode *ip = GFS2_I(inode);
1201         struct gfs2_sbd *sdp = GFS2_SB(inode);
1202         struct gfs2_holder *gh = &ip->i_iopen_gh;
1203         long timeout = 5 * HZ;
1204         int error;
1205
1206         gh->gh_flags |= GL_NOCACHE;
1207         gfs2_glock_dq_wait(gh);
1208
1209         /*
1210          * If there are no other lock holders, we'll get the lock immediately.
1211          * Otherwise, the other nodes holding the lock will be notified about
1212          * our locking request.  If they don't have the inode open, they'll
1213          * evict the cached inode and release the lock.  Otherwise, if they
1214          * poke the inode glock, we'll take this as an indication that they
1215          * still need the iopen glock and that they'll take care of deleting
1216          * the inode when they're done.  As a last resort, if another node
1217          * keeps holding the iopen glock without showing any activity on the
1218          * inode glock, we'll eventually time out.
1219          *
1220          * Note that we're passing the LM_FLAG_TRY_1CB flag to the first
1221          * locking request as an optimization to notify lock holders as soon as
1222          * possible.  Without that flag, they'd be notified implicitly by the
1223          * second locking request.
1224          */
1225
1226         gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE, gh);
1227         error = gfs2_glock_nq(gh);
1228         if (error != GLR_TRYFAILED)
1229                 return !error;
1230
1231         gfs2_holder_reinit(LM_ST_EXCLUSIVE, GL_ASYNC | GL_NOCACHE, gh);
1232         error = gfs2_glock_nq(gh);
1233         if (error)
1234                 return false;
1235
1236         timeout = wait_event_interruptible_timeout(sdp->sd_async_glock_wait,
1237                 !test_bit(HIF_WAIT, &gh->gh_iflags) ||
1238                 test_bit(GLF_DEMOTE, &ip->i_gl->gl_flags),
1239                 timeout);
1240         if (!test_bit(HIF_HOLDER, &gh->gh_iflags)) {
1241                 gfs2_glock_dq(gh);
1242                 return false;
1243         }
1244         return true;
1245 }
1246
1247 /**
1248  * evict_should_delete - determine whether the inode is eligible for deletion
1249  * @inode: The inode to evict
1250  *
1251  * This function determines whether the evicted inode is eligible to be deleted
1252  * and locks the inode glock.
1253  *
1254  * Returns: the fate of the dinode
1255  */
1256 static enum dinode_demise evict_should_delete(struct inode *inode,
1257                                               struct gfs2_holder *gh)
1258 {
1259         struct gfs2_inode *ip = GFS2_I(inode);
1260         struct super_block *sb = inode->i_sb;
1261         struct gfs2_sbd *sdp = sb->s_fs_info;
1262         int ret;
1263
1264         if (test_bit(GIF_ALLOC_FAILED, &ip->i_flags)) {
1265                 BUG_ON(!gfs2_glock_is_locked_by_me(ip->i_gl));
1266                 goto should_delete;
1267         }
1268
1269         if (test_bit(GIF_DEFERRED_DELETE, &ip->i_flags))
1270                 return SHOULD_DEFER_EVICTION;
1271
1272         /* Deletes should never happen under memory pressure anymore.  */
1273         if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
1274                 return SHOULD_DEFER_EVICTION;
1275
1276         /* Must not read inode block until block type has been verified */
1277         ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, gh);
1278         if (unlikely(ret)) {
1279                 glock_clear_object(ip->i_iopen_gh.gh_gl, ip);
1280                 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1281                 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1282                 return SHOULD_DEFER_EVICTION;
1283         }
1284
1285         if (gfs2_inode_already_deleted(ip->i_gl, ip->i_no_formal_ino))
1286                 return SHOULD_NOT_DELETE_DINODE;
1287         ret = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED);
1288         if (ret)
1289                 return SHOULD_NOT_DELETE_DINODE;
1290
1291         if (test_bit(GIF_INVALID, &ip->i_flags)) {
1292                 ret = gfs2_inode_refresh(ip);
1293                 if (ret)
1294                         return SHOULD_NOT_DELETE_DINODE;
1295         }
1296
1297         /*
1298          * The inode may have been recreated in the meantime.
1299          */
1300         if (inode->i_nlink)
1301                 return SHOULD_NOT_DELETE_DINODE;
1302
1303 should_delete:
1304         if (gfs2_holder_initialized(&ip->i_iopen_gh) &&
1305             test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1306                 if (!gfs2_upgrade_iopen_glock(inode)) {
1307                         gfs2_holder_uninit(&ip->i_iopen_gh);
1308                         return SHOULD_NOT_DELETE_DINODE;
1309                 }
1310         }
1311         return SHOULD_DELETE_DINODE;
1312 }
1313
1314 /**
1315  * evict_unlinked_inode - delete the pieces of an unlinked evicted inode
1316  * @inode: The inode to evict
1317  */
1318 static int evict_unlinked_inode(struct inode *inode)
1319 {
1320         struct gfs2_inode *ip = GFS2_I(inode);
1321         int ret;
1322
1323         if (S_ISDIR(inode->i_mode) &&
1324             (ip->i_diskflags & GFS2_DIF_EXHASH)) {
1325                 ret = gfs2_dir_exhash_dealloc(ip);
1326                 if (ret)
1327                         goto out;
1328         }
1329
1330         if (ip->i_eattr) {
1331                 ret = gfs2_ea_dealloc(ip);
1332                 if (ret)
1333                         goto out;
1334         }
1335
1336         if (!gfs2_is_stuffed(ip)) {
1337                 ret = gfs2_file_dealloc(ip);
1338                 if (ret)
1339                         goto out;
1340         }
1341
1342         /* We're about to clear the bitmap for the dinode, but as soon as we
1343            do, gfs2_create_inode can create another inode at the same block
1344            location and try to set gl_object again. We clear gl_object here so
1345            that subsequent inode creates don't see an old gl_object. */
1346         glock_clear_object(ip->i_gl, ip);
1347         ret = gfs2_dinode_dealloc(ip);
1348         gfs2_inode_remember_delete(ip->i_gl, ip->i_no_formal_ino);
1349 out:
1350         return ret;
1351 }
1352
1353 /*
1354  * evict_linked_inode - evict an inode whose dinode has not been unlinked
1355  * @inode: The inode to evict
1356  */
1357 static int evict_linked_inode(struct inode *inode)
1358 {
1359         struct super_block *sb = inode->i_sb;
1360         struct gfs2_sbd *sdp = sb->s_fs_info;
1361         struct gfs2_inode *ip = GFS2_I(inode);
1362         struct address_space *metamapping;
1363         int ret;
1364
1365         gfs2_log_flush(sdp, ip->i_gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
1366                        GFS2_LFC_EVICT_INODE);
1367         metamapping = gfs2_glock2aspace(ip->i_gl);
1368         if (test_bit(GLF_DIRTY, &ip->i_gl->gl_flags)) {
1369                 filemap_fdatawrite(metamapping);
1370                 filemap_fdatawait(metamapping);
1371         }
1372         write_inode_now(inode, 1);
1373         gfs2_ail_flush(ip->i_gl, 0);
1374
1375         ret = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
1376         if (ret)
1377                 return ret;
1378
1379         /* Needs to be done before glock release & also in a transaction */
1380         truncate_inode_pages(&inode->i_data, 0);
1381         truncate_inode_pages(metamapping, 0);
1382         gfs2_trans_end(sdp);
1383         return 0;
1384 }
1385
1386 /**
1387  * gfs2_evict_inode - Remove an inode from cache
1388  * @inode: The inode to evict
1389  *
1390  * There are three cases to consider:
1391  * 1. i_nlink == 0, we are final opener (and must deallocate)
1392  * 2. i_nlink == 0, we are not the final opener (and cannot deallocate)
1393  * 3. i_nlink > 0
1394  *
1395  * If the fs is read only, then we have to treat all cases as per #3
1396  * since we are unable to do any deallocation. The inode will be
1397  * deallocated by the next read/write node to attempt an allocation
1398  * in the same resource group
1399  *
1400  * We have to (at the moment) hold the inodes main lock to cover
1401  * the gap between unlocking the shared lock on the iopen lock and
1402  * taking the exclusive lock. I'd rather do a shared -> exclusive
1403  * conversion on the iopen lock, but we can change that later. This
1404  * is safe, just less efficient.
1405  */
1406
1407 static void gfs2_evict_inode(struct inode *inode)
1408 {
1409         struct super_block *sb = inode->i_sb;
1410         struct gfs2_sbd *sdp = sb->s_fs_info;
1411         struct gfs2_inode *ip = GFS2_I(inode);
1412         struct gfs2_holder gh;
1413         int ret;
1414
1415         if (test_bit(GIF_FREE_VFS_INODE, &ip->i_flags)) {
1416                 clear_inode(inode);
1417                 return;
1418         }
1419
1420         if (inode->i_nlink || sb_rdonly(sb))
1421                 goto out;
1422
1423         /*
1424          * In case of an incomplete mount, gfs2_evict_inode() may be called for
1425          * system files without having an active journal to write to.  In that
1426          * case, skip the filesystem evict.
1427          */
1428         if (!sdp->sd_jdesc)
1429                 goto out;
1430
1431         gfs2_holder_mark_uninitialized(&gh);
1432         ret = evict_should_delete(inode, &gh);
1433         if (ret == SHOULD_DEFER_EVICTION)
1434                 goto out;
1435         if (ret == SHOULD_DELETE_DINODE)
1436                 ret = evict_unlinked_inode(inode);
1437         else
1438                 ret = evict_linked_inode(inode);
1439
1440         if (gfs2_rs_active(&ip->i_res))
1441                 gfs2_rs_deltree(&ip->i_res);
1442
1443         if (gfs2_holder_initialized(&gh)) {
1444                 glock_clear_object(ip->i_gl, ip);
1445                 gfs2_glock_dq_uninit(&gh);
1446         }
1447         if (ret && ret != GLR_TRYFAILED && ret != -EROFS)
1448                 fs_warn(sdp, "gfs2_evict_inode: %d\n", ret);
1449 out:
1450         truncate_inode_pages_final(&inode->i_data);
1451         if (ip->i_qadata)
1452                 gfs2_assert_warn(sdp, ip->i_qadata->qa_ref == 0);
1453         gfs2_rs_deltree(&ip->i_res);
1454         gfs2_ordered_del_inode(ip);
1455         clear_inode(inode);
1456         gfs2_dir_hash_inval(ip);
1457         if (gfs2_holder_initialized(&ip->i_iopen_gh)) {
1458                 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1459
1460                 glock_clear_object(gl, ip);
1461                 if (test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1462                         ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1463                         gfs2_glock_dq(&ip->i_iopen_gh);
1464                 }
1465                 gfs2_glock_hold(gl);
1466                 gfs2_holder_uninit(&ip->i_iopen_gh);
1467                 gfs2_glock_put_eventually(gl);
1468         }
1469         if (ip->i_gl) {
1470                 glock_clear_object(ip->i_gl, ip);
1471                 wait_on_bit_io(&ip->i_flags, GIF_GLOP_PENDING, TASK_UNINTERRUPTIBLE);
1472                 gfs2_glock_add_to_lru(ip->i_gl);
1473                 gfs2_glock_put_eventually(ip->i_gl);
1474                 rcu_assign_pointer(ip->i_gl, NULL);
1475         }
1476 }
1477
1478 static struct inode *gfs2_alloc_inode(struct super_block *sb)
1479 {
1480         struct gfs2_inode *ip;
1481
1482         ip = kmem_cache_alloc(gfs2_inode_cachep, GFP_KERNEL);
1483         if (!ip)
1484                 return NULL;
1485         ip->i_flags = 0;
1486         ip->i_gl = NULL;
1487         gfs2_holder_mark_uninitialized(&ip->i_iopen_gh);
1488         memset(&ip->i_res, 0, sizeof(ip->i_res));
1489         RB_CLEAR_NODE(&ip->i_res.rs_node);
1490         ip->i_rahead = 0;
1491         return &ip->i_inode;
1492 }
1493
1494 static void gfs2_free_inode(struct inode *inode)
1495 {
1496         kmem_cache_free(gfs2_inode_cachep, GFS2_I(inode));
1497 }
1498
1499 extern void free_local_statfs_inodes(struct gfs2_sbd *sdp)
1500 {
1501         struct local_statfs_inode *lsi, *safe;
1502
1503         /* Run through the statfs inodes list to iput and free memory */
1504         list_for_each_entry_safe(lsi, safe, &sdp->sd_sc_inodes_list, si_list) {
1505                 if (lsi->si_jid == sdp->sd_jdesc->jd_jid)
1506                         sdp->sd_sc_inode = NULL; /* belongs to this node */
1507                 if (lsi->si_sc_inode)
1508                         iput(lsi->si_sc_inode);
1509                 list_del(&lsi->si_list);
1510                 kfree(lsi);
1511         }
1512 }
1513
1514 extern struct inode *find_local_statfs_inode(struct gfs2_sbd *sdp,
1515                                              unsigned int index)
1516 {
1517         struct local_statfs_inode *lsi;
1518
1519         /* Return the local (per node) statfs inode in the
1520          * sdp->sd_sc_inodes_list corresponding to the 'index'. */
1521         list_for_each_entry(lsi, &sdp->sd_sc_inodes_list, si_list) {
1522                 if (lsi->si_jid == index)
1523                         return lsi->si_sc_inode;
1524         }
1525         return NULL;
1526 }
1527
1528 const struct super_operations gfs2_super_ops = {
1529         .alloc_inode            = gfs2_alloc_inode,
1530         .free_inode             = gfs2_free_inode,
1531         .write_inode            = gfs2_write_inode,
1532         .dirty_inode            = gfs2_dirty_inode,
1533         .evict_inode            = gfs2_evict_inode,
1534         .put_super              = gfs2_put_super,
1535         .sync_fs                = gfs2_sync_fs,
1536         .freeze_super           = gfs2_freeze,
1537         .thaw_super             = gfs2_unfreeze,
1538         .statfs                 = gfs2_statfs,
1539         .drop_inode             = gfs2_drop_inode,
1540         .show_options           = gfs2_show_options,
1541 };
1542