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 /**
  48  * gfs2_jindex_free - Clear all the journal index information
  49  * @sdp: The GFS2 superblock
  50  *
  51  */
  52
  53 void gfs2_jindex_free(struct gfs2_sbd *sdp)
  54 {
  55         struct list_head list;
  56         struct gfs2_jdesc *jd;
  57
  58         spin_lock(&sdp->sd_jindex_spin);
  59         list_add(&list, &sdp->sd_jindex_list);
  60         list_del_init(&sdp->sd_jindex_list);
  61         sdp->sd_journals = 0;
  62         spin_unlock(&sdp->sd_jindex_spin);
  63
  64         while (!list_empty(&list)) {
  65                 jd = list_entry(list.next, struct gfs2_jdesc, jd_list);
  66                 gfs2_free_journal_extents(jd);
  67                 list_del(&jd->jd_list);
  68                 iput(jd->jd_inode);
  69                 kfree(jd);
  70         }
  71 }
  72
  73 static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
  74 {
  75         struct gfs2_jdesc *jd;
  76         int found = 0;
  77
  78         list_for_each_entry(jd, head, jd_list) {
  79                 if (jd->jd_jid == jid) {
  80                         found = 1;
  81                         break;
  82                 }
  83         }
  84
  85         if (!found)
  86                 jd = NULL;
  87
  88         return jd;
  89 }
  90
  91 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
  92 {
  93         struct gfs2_jdesc *jd;
  94
  95         spin_lock(&sdp->sd_jindex_spin);
  96         jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
  97         spin_unlock(&sdp->sd_jindex_spin);
  98
  99         return jd;
 100 }
 101
 102 int gfs2_jdesc_check(struct gfs2_jdesc *jd)
 103 {
 104         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 105         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 106         u64 size = i_size_read(jd->jd_inode);
 107
 108         if (gfs2_check_internal_file_size(jd->jd_inode, 8 << 20, BIT(30)))
 109                 return -EIO;
 110
 111         jd->jd_blocks = size >> sdp->sd_sb.sb_bsize_shift;
 112
 113         if (gfs2_write_alloc_required(ip, 0, size)) {
 114                 gfs2_consist_inode(ip);
 115                 return -EIO;
 116         }
 117
 118         return 0;
 119 }
 120
 121 static int init_threads(struct gfs2_sbd *sdp)
 122 {
 123         struct task_struct *p;
 124         int error = 0;
 125
 126         p = kthread_run(gfs2_logd, sdp, "gfs2_logd");
 127         if (IS_ERR(p)) {
 128                 error = PTR_ERR(p);
 129                 fs_err(sdp, "can't start logd thread: %d\n", error);
 130                 return error;
 131         }
 132         sdp->sd_logd_process = p;
 133
 134         p = kthread_run(gfs2_quotad, sdp, "gfs2_quotad");
 135         if (IS_ERR(p)) {
 136                 error = PTR_ERR(p);
 137                 fs_err(sdp, "can't start quotad thread: %d\n", error);
 138                 goto fail;
 139         }
 140         sdp->sd_quotad_process = p;
 141         return 0;
 142
 143 fail:
 144         kthread_stop(sdp->sd_logd_process);
 145         sdp->sd_logd_process = NULL;
 146         return error;
 147 }
 148
 149 /**
 150  * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
 151  * @sdp: the filesystem
 152  *
 153  * Returns: errno
 154  */
 155
 156 int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
 157 {
 158         struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
 159         struct gfs2_glock *j_gl = ip->i_gl;
 160         struct gfs2_holder freeze_gh;
 161         struct gfs2_log_header_host head;
 162         int error;
 163
 164         error = init_threads(sdp);
 165         if (error)
 166                 return error;
 167
 168         error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED, 0,
 169                                    &freeze_gh);
 170         if (error)
 171                 goto fail_threads;
 172
 173         j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
 174
 175         error = gfs2_find_jhead(sdp->sd_jdesc, &head, false);
 176         if (error)
 177                 goto fail;
 178
 179         if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 180                 gfs2_consist(sdp);
 181                 error = -EIO;
 182                 goto fail;
 183         }
 184
 185         /*  Initialize some head of the log stuff  */
 186         sdp->sd_log_sequence = head.lh_sequence + 1;
 187         gfs2_log_pointers_init(sdp, head.lh_blkno);
 188
 189         error = gfs2_quota_init(sdp);
 190         if (error)
 191                 goto fail;
 192
 193         set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 194
 195         gfs2_glock_dq_uninit(&freeze_gh);
 196
 197         return 0;
 198
 199 fail:
 200         freeze_gh.gh_flags |= GL_NOCACHE;
 201         gfs2_glock_dq_uninit(&freeze_gh);
 202 fail_threads:
 203         if (sdp->sd_quotad_process)
 204                 kthread_stop(sdp->sd_quotad_process);
 205         sdp->sd_quotad_process = NULL;
 206         if (sdp->sd_logd_process)
 207                 kthread_stop(sdp->sd_logd_process);
 208         sdp->sd_logd_process = NULL;
 209         return error;
 210 }
 211
 212 void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
 213 {
 214         const struct gfs2_statfs_change *str = buf;
 215
 216         sc->sc_total = be64_to_cpu(str->sc_total);
 217         sc->sc_free = be64_to_cpu(str->sc_free);
 218         sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
 219 }
 220
 221 static void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
 222 {
 223         struct gfs2_statfs_change *str = buf;
 224
 225         str->sc_total = cpu_to_be64(sc->sc_total);
 226         str->sc_free = cpu_to_be64(sc->sc_free);
 227         str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
 228 }
 229
 230 int gfs2_statfs_init(struct gfs2_sbd *sdp)
 231 {
 232         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 233         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 234         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 235         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 236         struct buffer_head *m_bh, *l_bh;
 237         struct gfs2_holder gh;
 238         int error;
 239
 240         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 241                                    &gh);
 242         if (error)
 243                 return error;
 244
 245         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 246         if (error)
 247                 goto out;
 248
 249         if (sdp->sd_args.ar_spectator) {
 250                 spin_lock(&sdp->sd_statfs_spin);
 251                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 252                                       sizeof(struct gfs2_dinode));
 253                 spin_unlock(&sdp->sd_statfs_spin);
 254         } else {
 255                 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 256                 if (error)
 257                         goto out_m_bh;
 258
 259                 spin_lock(&sdp->sd_statfs_spin);
 260                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 261                                       sizeof(struct gfs2_dinode));
 262                 gfs2_statfs_change_in(l_sc, l_bh->b_data +
 263                                       sizeof(struct gfs2_dinode));
 264                 spin_unlock(&sdp->sd_statfs_spin);
 265
 266                 brelse(l_bh);
 267         }
 268
 269 out_m_bh:
 270         brelse(m_bh);
 271 out:
 272         gfs2_glock_dq_uninit(&gh);
 273         return 0;
 274 }
 275
 276 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
 277                         s64 dinodes)
 278 {
 279         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 280         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 281         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 282         struct buffer_head *l_bh;
 283         s64 x, y;
 284         int need_sync = 0;
 285         int error;
 286
 287         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 288         if (error)
 289                 return;
 290
 291         gfs2_trans_add_meta(l_ip->i_gl, l_bh);
 292
 293         spin_lock(&sdp->sd_statfs_spin);
 294         l_sc->sc_total += total;
 295         l_sc->sc_free += free;
 296         l_sc->sc_dinodes += dinodes;
 297         gfs2_statfs_change_out(l_sc, l_bh->b_data + sizeof(struct gfs2_dinode));
 298         if (sdp->sd_args.ar_statfs_percent) {
 299                 x = 100 * l_sc->sc_free;
 300                 y = m_sc->sc_free * sdp->sd_args.ar_statfs_percent;
 301                 if (x >= y || x <= -y)
 302                         need_sync = 1;
 303         }
 304         spin_unlock(&sdp->sd_statfs_spin);
 305
 306         brelse(l_bh);
 307         if (need_sync)
 308                 gfs2_wake_up_statfs(sdp);
 309 }
 310
 311 void update_statfs(struct gfs2_sbd *sdp, struct buffer_head *m_bh,
 312                    struct buffer_head *l_bh)
 313 {
 314         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 315         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 316         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 317         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 318
 319         gfs2_trans_add_meta(l_ip->i_gl, l_bh);
 320         gfs2_trans_add_meta(m_ip->i_gl, m_bh);
 321
 322         spin_lock(&sdp->sd_statfs_spin);
 323         m_sc->sc_total += l_sc->sc_total;
 324         m_sc->sc_free += l_sc->sc_free;
 325         m_sc->sc_dinodes += l_sc->sc_dinodes;
 326         memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
 327         memset(l_bh->b_data + sizeof(struct gfs2_dinode),
 328                0, sizeof(struct gfs2_statfs_change));
 329         gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
 330         spin_unlock(&sdp->sd_statfs_spin);
 331 }
 332
 333 int gfs2_statfs_sync(struct super_block *sb, int type)
 334 {
 335         struct gfs2_sbd *sdp = sb->s_fs_info;
 336         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 337         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 338         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 339         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 340         struct gfs2_holder gh;
 341         struct buffer_head *m_bh, *l_bh;
 342         int error;
 343
 344         sb_start_write(sb);
 345         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 346                                    &gh);
 347         if (error)
 348                 goto out;
 349
 350         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 351         if (error)
 352                 goto out_unlock;
 353
 354         spin_lock(&sdp->sd_statfs_spin);
 355         gfs2_statfs_change_in(m_sc, m_bh->b_data +
 356                               sizeof(struct gfs2_dinode));
 357         if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
 358                 spin_unlock(&sdp->sd_statfs_spin);
 359                 goto out_bh;
 360         }
 361         spin_unlock(&sdp->sd_statfs_spin);
 362
 363         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 364         if (error)
 365                 goto out_bh;
 366
 367         error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
 368         if (error)
 369                 goto out_bh2;
 370
 371         update_statfs(sdp, m_bh, l_bh);
 372         sdp->sd_statfs_force_sync = 0;
 373
 374         gfs2_trans_end(sdp);
 375
 376 out_bh2:
 377         brelse(l_bh);
 378 out_bh:
 379         brelse(m_bh);
 380 out_unlock:
 381         gfs2_glock_dq_uninit(&gh);
 382 out:
 383         sb_end_write(sb);
 384         return error;
 385 }
 386
 387 struct lfcc {
 388         struct list_head list;
 389         struct gfs2_holder gh;
 390 };
 391
 392 /**
 393  * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
 394  *                            journals are clean
 395  * @sdp: the file system
 396  * @state: the state to put the transaction lock into
 397  * @t_gh: the hold on the transaction lock
 398  *
 399  * Returns: errno
 400  */
 401
 402 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp,
 403                                     struct gfs2_holder *freeze_gh)
 404 {
 405         struct gfs2_inode *ip;
 406         struct gfs2_jdesc *jd;
 407         struct lfcc *lfcc;
 408         LIST_HEAD(list);
 409         struct gfs2_log_header_host lh;
 410         int error;
 411
 412         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 413                 lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
 414                 if (!lfcc) {
 415                         error = -ENOMEM;
 416                         goto out;
 417                 }
 418                 ip = GFS2_I(jd->jd_inode);
 419                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
 420                 if (error) {
 421                         kfree(lfcc);
 422                         goto out;
 423                 }
 424                 list_add(&lfcc->list, &list);
 425         }
 426
 427         error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_EXCLUSIVE,
 428                                    GL_NOCACHE, freeze_gh);
 429
 430         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 431                 error = gfs2_jdesc_check(jd);
 432                 if (error)
 433                         break;
 434                 error = gfs2_find_jhead(jd, &lh, false);
 435                 if (error)
 436                         break;
 437                 if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 438                         error = -EBUSY;
 439                         break;
 440                 }
 441         }
 442
 443         if (error)
 444                 gfs2_glock_dq_uninit(freeze_gh);
 445
 446 out:
 447         while (!list_empty(&list)) {
 448                 lfcc = list_entry(list.next, struct lfcc, list);
 449                 list_del(&lfcc->list);
 450                 gfs2_glock_dq_uninit(&lfcc->gh);
 451                 kfree(lfcc);
 452         }
 453         return error;
 454 }
 455
 456 void gfs2_dinode_out(const struct gfs2_inode *ip, void *buf)
 457 {
 458         struct gfs2_dinode *str = buf;
 459
 460         str->di_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 461         str->di_header.mh_type = cpu_to_be32(GFS2_METATYPE_DI);
 462         str->di_header.mh_format = cpu_to_be32(GFS2_FORMAT_DI);
 463         str->di_num.no_addr = cpu_to_be64(ip->i_no_addr);
 464         str->di_num.no_formal_ino = cpu_to_be64(ip->i_no_formal_ino);
 465         str->di_mode = cpu_to_be32(ip->i_inode.i_mode);
 466         str->di_uid = cpu_to_be32(i_uid_read(&ip->i_inode));
 467         str->di_gid = cpu_to_be32(i_gid_read(&ip->i_inode));
 468         str->di_nlink = cpu_to_be32(ip->i_inode.i_nlink);
 469         str->di_size = cpu_to_be64(i_size_read(&ip->i_inode));
 470         str->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(&ip->i_inode));
 471         str->di_atime = cpu_to_be64(ip->i_inode.i_atime.tv_sec);
 472         str->di_mtime = cpu_to_be64(ip->i_inode.i_mtime.tv_sec);
 473         str->di_ctime = cpu_to_be64(ip->i_inode.i_ctime.tv_sec);
 474
 475         str->di_goal_meta = cpu_to_be64(ip->i_goal);
 476         str->di_goal_data = cpu_to_be64(ip->i_goal);
 477         str->di_generation = cpu_to_be64(ip->i_generation);
 478
 479         str->di_flags = cpu_to_be32(ip->i_diskflags);
 480         str->di_height = cpu_to_be16(ip->i_height);
 481         str->di_payload_format = cpu_to_be32(S_ISDIR(ip->i_inode.i_mode) &&
 482                                              !(ip->i_diskflags & GFS2_DIF_EXHASH) ?
 483                                              GFS2_FORMAT_DE : 0);
 484         str->di_depth = cpu_to_be16(ip->i_depth);
 485         str->di_entries = cpu_to_be32(ip->i_entries);
 486
 487         str->di_eattr = cpu_to_be64(ip->i_eattr);
 488         str->di_atime_nsec = cpu_to_be32(ip->i_inode.i_atime.tv_nsec);
 489         str->di_mtime_nsec = cpu_to_be32(ip->i_inode.i_mtime.tv_nsec);
 490         str->di_ctime_nsec = cpu_to_be32(ip->i_inode.i_ctime.tv_nsec);
 491 }
 492
 493 /**
 494  * gfs2_write_inode - Make sure the inode is stable on the disk
 495  * @inode: The inode
 496  * @wbc: The writeback control structure
 497  *
 498  * Returns: errno
 499  */
 500
 501 static int gfs2_write_inode(struct inode *inode, struct writeback_control *wbc)
 502 {
 503         struct gfs2_inode *ip = GFS2_I(inode);
 504         struct gfs2_sbd *sdp = GFS2_SB(inode);
 505         struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl);
 506         struct backing_dev_info *bdi = inode_to_bdi(metamapping->host);
 507         int ret = 0;
 508         bool flush_all = (wbc->sync_mode == WB_SYNC_ALL || gfs2_is_jdata(ip));
 509
 510         if (flush_all)
 511                 gfs2_log_flush(GFS2_SB(inode), ip->i_gl,
 512                                GFS2_LOG_HEAD_FLUSH_NORMAL |
 513                                GFS2_LFC_WRITE_INODE);
 514         if (bdi->wb.dirty_exceeded)
 515                 gfs2_ail1_flush(sdp, wbc);
 516         else
 517                 filemap_fdatawrite(metamapping);
 518         if (flush_all)
 519                 ret = filemap_fdatawait(metamapping);
 520         if (ret)
 521                 mark_inode_dirty_sync(inode);
 522         else {
 523                 spin_lock(&inode->i_lock);
 524                 if (!(inode->i_flags & I_DIRTY))
 525                         gfs2_ordered_del_inode(ip);
 526                 spin_unlock(&inode->i_lock);
 527         }
 528         return ret;
 529 }
 530
 531 /**
 532  * gfs2_dirty_inode - check for atime updates
 533  * @inode: The inode in question
 534  * @flags: The type of dirty
 535  *
 536  * Unfortunately it can be called under any combination of inode
 537  * glock and transaction lock, so we have to check carefully.
 538  *
 539  * At the moment this deals only with atime - it should be possible
 540  * to expand that role in future, once a review of the locking has
 541  * been carried out.
 542  */
 543
 544 static void gfs2_dirty_inode(struct inode *inode, int flags)
 545 {
 546         struct gfs2_inode *ip = GFS2_I(inode);
 547         struct gfs2_sbd *sdp = GFS2_SB(inode);
 548         struct buffer_head *bh;
 549         struct gfs2_holder gh;
 550         int need_unlock = 0;
 551         int need_endtrans = 0;
 552         int ret;
 553
 554         if (!(flags & I_DIRTY_INODE))
 555                 return;
 556         if (unlikely(test_bit(SDF_WITHDRAWN, &sdp->sd_flags)))
 557                 return;
 558         if (!gfs2_glock_is_locked_by_me(ip->i_gl)) {
 559                 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
 560                 if (ret) {
 561                         fs_err(sdp, "dirty_inode: glock %d\n", ret);
 562                         return;
 563                 }
 564                 need_unlock = 1;
 565         } else if (WARN_ON_ONCE(ip->i_gl->gl_state != LM_ST_EXCLUSIVE))
 566                 return;
 567
 568         if (current->journal_info == NULL) {
 569                 ret = gfs2_trans_begin(sdp, RES_DINODE, 0);
 570                 if (ret) {
 571                         fs_err(sdp, "dirty_inode: gfs2_trans_begin %d\n", ret);
 572                         goto out;
 573                 }
 574                 need_endtrans = 1;
 575         }
 576
 577         ret = gfs2_meta_inode_buffer(ip, &bh);
 578         if (ret == 0) {
 579                 gfs2_trans_add_meta(ip->i_gl, bh);
 580                 gfs2_dinode_out(ip, bh->b_data);
 581                 brelse(bh);
 582         }
 583
 584         if (need_endtrans)
 585                 gfs2_trans_end(sdp);
 586 out:
 587         if (need_unlock)
 588                 gfs2_glock_dq_uninit(&gh);
 589 }
 590
 591 /**
 592  * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
 593  * @sdp: the filesystem
 594  *
 595  * Returns: errno
 596  */
 597
 598 int gfs2_make_fs_ro(struct gfs2_sbd *sdp)
 599 {
 600         struct gfs2_holder freeze_gh;
 601         int error;
 602
 603         error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED, GL_NOCACHE,
 604                                    &freeze_gh);
 605         if (error && !test_bit(SDF_WITHDRAWN, &sdp->sd_flags))
 606                 return error;
 607
 608         flush_workqueue(gfs2_delete_workqueue);
 609         if (sdp->sd_quotad_process)
 610                 kthread_stop(sdp->sd_quotad_process);
 611         sdp->sd_quotad_process = NULL;
 612         if (sdp->sd_logd_process)
 613                 kthread_stop(sdp->sd_logd_process);
 614         sdp->sd_logd_process = NULL;
 615
 616         gfs2_quota_sync(sdp->sd_vfs, 0);
 617         gfs2_statfs_sync(sdp->sd_vfs, 0);
 618
 619         gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SHUTDOWN |
 620                        GFS2_LFC_MAKE_FS_RO);
 621         wait_event(sdp->sd_reserving_log_wait, atomic_read(&sdp->sd_reserving_log) == 0);
 622         gfs2_assert_warn(sdp, atomic_read(&sdp->sd_log_blks_free) == sdp->sd_jdesc->jd_blocks);
 623
 624         if (gfs2_holder_initialized(&freeze_gh))
 625                 gfs2_glock_dq_uninit(&freeze_gh);
 626
 627         gfs2_quota_cleanup(sdp);
 628
 629         return error;
 630 }
 631
 632 /**
 633  * gfs2_put_super - Unmount the filesystem
 634  * @sb: The VFS superblock
 635  *
 636  */
 637
 638 static void gfs2_put_super(struct super_block *sb)
 639 {
 640         struct gfs2_sbd *sdp = sb->s_fs_info;
 641         int error;
 642         struct gfs2_jdesc *jd;
 643
 644         /* No more recovery requests */
 645         set_bit(SDF_NORECOVERY, &sdp->sd_flags);
 646         smp_mb();
 647
 648         /* Wait on outstanding recovery */
 649 restart:
 650         spin_lock(&sdp->sd_jindex_spin);
 651         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 652                 if (!test_bit(JDF_RECOVERY, &jd->jd_flags))
 653                         continue;
 654                 spin_unlock(&sdp->sd_jindex_spin);
 655                 wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
 656                             TASK_UNINTERRUPTIBLE);
 657                 goto restart;
 658         }
 659         spin_unlock(&sdp->sd_jindex_spin);
 660
 661         if (!sb_rdonly(sb)) {
 662                 error = gfs2_make_fs_ro(sdp);
 663                 if (error)
 664                         gfs2_io_error(sdp);
 665         }
 666         /*  At this point, we're through modifying the disk  */
 667
 668         /*  Release stuff  */
 669
 670         iput(sdp->sd_jindex);
 671         iput(sdp->sd_statfs_inode);
 672         iput(sdp->sd_rindex);
 673         iput(sdp->sd_quota_inode);
 674
 675         gfs2_glock_put(sdp->sd_rename_gl);
 676         gfs2_glock_put(sdp->sd_freeze_gl);
 677
 678         if (!sdp->sd_args.ar_spectator) {
 679                 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
 680                 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
 681                 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
 682                 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
 683                 iput(sdp->sd_sc_inode);
 684                 iput(sdp->sd_qc_inode);
 685         }
 686
 687         gfs2_glock_dq_uninit(&sdp->sd_live_gh);
 688         gfs2_clear_rgrpd(sdp);
 689         gfs2_jindex_free(sdp);
 690         /*  Take apart glock structures and buffer lists  */
 691         gfs2_gl_hash_clear(sdp);
 692         truncate_inode_pages_final(&sdp->sd_aspace);
 693         gfs2_delete_debugfs_file(sdp);
 694         /*  Unmount the locking protocol  */
 695         gfs2_lm_unmount(sdp);
 696
 697         /*  At this point, we're through participating in the lockspace  */
 698         gfs2_sys_fs_del(sdp);
 699         free_sbd(sdp);
 700 }
 701
 702 /**
 703  * gfs2_sync_fs - sync the filesystem
 704  * @sb: the superblock
 705  *
 706  * Flushes the log to disk.
 707  */
 708
 709 static int gfs2_sync_fs(struct super_block *sb, int wait)
 710 {
 711         struct gfs2_sbd *sdp = sb->s_fs_info;
 712
 713         gfs2_quota_sync(sb, -1);
 714         if (wait)
 715                 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
 716                                GFS2_LFC_SYNC_FS);
 717         return sdp->sd_log_error;
 718 }
 719
 720 void gfs2_freeze_func(struct work_struct *work)
 721 {
 722         int error;
 723         struct gfs2_holder freeze_gh;
 724         struct gfs2_sbd *sdp = container_of(work, struct gfs2_sbd, sd_freeze_work);
 725         struct super_block *sb = sdp->sd_vfs;
 726
 727         atomic_inc(&sb->s_active);
 728         error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED, 0,
 729                                    &freeze_gh);
 730         if (error) {
 731                 fs_info(sdp, "GFS2: couldn't get freeze lock : %d\n", error);
 732                 gfs2_assert_withdraw(sdp, 0);
 733         } else {
 734                 atomic_set(&sdp->sd_freeze_state, SFS_UNFROZEN);
 735                 error = thaw_super(sb);
 736                 if (error) {
 737                         fs_info(sdp, "GFS2: couldn't thaw filesystem: %d\n",
 738                                 error);
 739                         gfs2_assert_withdraw(sdp, 0);
 740                 }
 741                 if (!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))
 742                         freeze_gh.gh_flags |= GL_NOCACHE;
 743                 gfs2_glock_dq_uninit(&freeze_gh);
 744         }
 745         deactivate_super(sb);
 746         clear_bit_unlock(SDF_FS_FROZEN, &sdp->sd_flags);
 747         wake_up_bit(&sdp->sd_flags, SDF_FS_FROZEN);
 748         return;
 749 }
 750
 751 /**
 752  * gfs2_freeze - prevent further writes to the filesystem
 753  * @sb: the VFS structure for the filesystem
 754  *
 755  */
 756
 757 static int gfs2_freeze(struct super_block *sb)
 758 {
 759         struct gfs2_sbd *sdp = sb->s_fs_info;
 760         int error;
 761
 762         mutex_lock(&sdp->sd_freeze_mutex);
 763         if (atomic_read(&sdp->sd_freeze_state) != SFS_UNFROZEN) {
 764                 error = -EBUSY;
 765                 goto out;
 766         }
 767
 768         if (test_bit(SDF_WITHDRAWN, &sdp->sd_flags)) {
 769                 error = -EINVAL;
 770                 goto out;
 771         }
 772
 773         for (;;) {
 774                 error = gfs2_lock_fs_check_clean(sdp, &sdp->sd_freeze_gh);
 775                 if (!error)
 776                         break;
 777
 778                 if (error == -EBUSY)
 779                         fs_err(sdp, "waiting for recovery before freeze\n");
 780                 else
 781                         fs_err(sdp, "error freezing FS: %d\n", error);
 782
 783                 fs_err(sdp, "retrying...\n");
 784                 msleep(1000);
 785         }
 786         set_bit(SDF_FS_FROZEN, &sdp->sd_flags);
 787 out:
 788         mutex_unlock(&sdp->sd_freeze_mutex);
 789         return error;
 790 }
 791
 792 /**
 793  * gfs2_unfreeze - reallow writes to the filesystem
 794  * @sb: the VFS structure for the filesystem
 795  *
 796  */
 797
 798 static int gfs2_unfreeze(struct super_block *sb)
 799 {
 800         struct gfs2_sbd *sdp = sb->s_fs_info;
 801
 802         mutex_lock(&sdp->sd_freeze_mutex);
 803         if (atomic_read(&sdp->sd_freeze_state) != SFS_FROZEN ||
 804             !gfs2_holder_initialized(&sdp->sd_freeze_gh)) {
 805                 mutex_unlock(&sdp->sd_freeze_mutex);
 806                 return -EINVAL;
 807         }
 808
 809         gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
 810         mutex_unlock(&sdp->sd_freeze_mutex);
 811         return wait_on_bit(&sdp->sd_flags, SDF_FS_FROZEN, TASK_INTERRUPTIBLE);
 812 }
 813
 814 /**
 815  * statfs_fill - fill in the sg for a given RG
 816  * @rgd: the RG
 817  * @sc: the sc structure
 818  *
 819  * Returns: 0 on success, -ESTALE if the LVB is invalid
 820  */
 821
 822 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
 823                             struct gfs2_statfs_change_host *sc)
 824 {
 825         gfs2_rgrp_verify(rgd);
 826         sc->sc_total += rgd->rd_data;
 827         sc->sc_free += rgd->rd_free;
 828         sc->sc_dinodes += rgd->rd_dinodes;
 829         return 0;
 830 }
 831
 832 /**
 833  * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
 834  * @sdp: the filesystem
 835  * @sc: the sc info that will be returned
 836  *
 837  * Any error (other than a signal) will cause this routine to fall back
 838  * to the synchronous version.
 839  *
 840  * FIXME: This really shouldn't busy wait like this.
 841  *
 842  * Returns: errno
 843  */
 844
 845 static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
 846 {
 847         struct gfs2_rgrpd *rgd_next;
 848         struct gfs2_holder *gha, *gh;
 849         unsigned int slots = 64;
 850         unsigned int x;
 851         int done;
 852         int error = 0, err;
 853
 854         memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
 855         gha = kmalloc_array(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
 856         if (!gha)
 857                 return -ENOMEM;
 858         for (x = 0; x < slots; x++)
 859                 gfs2_holder_mark_uninitialized(gha + x);
 860
 861         rgd_next = gfs2_rgrpd_get_first(sdp);
 862
 863         for (;;) {
 864                 done = 1;
 865
 866                 for (x = 0; x < slots; x++) {
 867                         gh = gha + x;
 868
 869                         if (gfs2_holder_initialized(gh) && gfs2_glock_poll(gh)) {
 870                                 err = gfs2_glock_wait(gh);
 871                                 if (err) {
 872                                         gfs2_holder_uninit(gh);
 873                                         error = err;
 874                                 } else {
 875                                         if (!error) {
 876                                                 struct gfs2_rgrpd *rgd =
 877                                                         gfs2_glock2rgrp(gh->gh_gl);
 878
 879                                                 error = statfs_slow_fill(rgd, sc);
 880                                         }
 881                                         gfs2_glock_dq_uninit(gh);
 882                                 }
 883                         }
 884
 885                         if (gfs2_holder_initialized(gh))
 886                                 done = 0;
 887                         else if (rgd_next && !error) {
 888                                 error = gfs2_glock_nq_init(rgd_next->rd_gl,
 889                                                            LM_ST_SHARED,
 890                                                            GL_ASYNC,
 891                                                            gh);
 892                                 rgd_next = gfs2_rgrpd_get_next(rgd_next);
 893                                 done = 0;
 894                         }
 895
 896                         if (signal_pending(current))
 897                                 error = -ERESTARTSYS;
 898                 }
 899
 900                 if (done)
 901                         break;
 902
 903                 yield();
 904         }
 905
 906         kfree(gha);
 907         return error;
 908 }
 909
 910 /**
 911  * gfs2_statfs_i - Do a statfs
 912  * @sdp: the filesystem
 913  * @sg: the sg structure
 914  *
 915  * Returns: errno
 916  */
 917
 918 static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
 919 {
 920         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 921         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 922
 923         spin_lock(&sdp->sd_statfs_spin);
 924
 925         *sc = *m_sc;
 926         sc->sc_total += l_sc->sc_total;
 927         sc->sc_free += l_sc->sc_free;
 928         sc->sc_dinodes += l_sc->sc_dinodes;
 929
 930         spin_unlock(&sdp->sd_statfs_spin);
 931
 932         if (sc->sc_free < 0)
 933                 sc->sc_free = 0;
 934         if (sc->sc_free > sc->sc_total)
 935                 sc->sc_free = sc->sc_total;
 936         if (sc->sc_dinodes < 0)
 937                 sc->sc_dinodes = 0;
 938
 939         return 0;
 940 }
 941
 942 /**
 943  * gfs2_statfs - Gather and return stats about the filesystem
 944  * @sb: The superblock
 945  * @statfsbuf: The buffer
 946  *
 947  * Returns: 0 on success or error code
 948  */
 949
 950 static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
 951 {
 952         struct super_block *sb = dentry->d_sb;
 953         struct gfs2_sbd *sdp = sb->s_fs_info;
 954         struct gfs2_statfs_change_host sc;
 955         int error;
 956
 957         error = gfs2_rindex_update(sdp);
 958         if (error)
 959                 return error;
 960
 961         if (gfs2_tune_get(sdp, gt_statfs_slow))
 962                 error = gfs2_statfs_slow(sdp, &sc);
 963         else
 964                 error = gfs2_statfs_i(sdp, &sc);
 965
 966         if (error)
 967                 return error;
 968
 969         buf->f_type = GFS2_MAGIC;
 970         buf->f_bsize = sdp->sd_sb.sb_bsize;
 971         buf->f_blocks = sc.sc_total;
 972         buf->f_bfree = sc.sc_free;
 973         buf->f_bavail = sc.sc_free;
 974         buf->f_files = sc.sc_dinodes + sc.sc_free;
 975         buf->f_ffree = sc.sc_free;
 976         buf->f_namelen = GFS2_FNAMESIZE;
 977
 978         return 0;
 979 }
 980
 981 /**
 982  * gfs2_drop_inode - Drop an inode (test for remote unlink)
 983  * @inode: The inode to drop
 984  *
 985  * If we've received a callback on an iopen lock then it's because a
 986  * remote node tried to deallocate the inode but failed due to this node
 987  * still having the inode open. Here we mark the link count zero
 988  * since we know that it must have reached zero if the GLF_DEMOTE flag
 989  * is set on the iopen glock. If we didn't do a disk read since the
 990  * remote node removed the final link then we might otherwise miss
 991  * this event. This check ensures that this node will deallocate the
 992  * inode's blocks, or alternatively pass the baton on to another
 993  * node for later deallocation.
 994  */
 995
 996 static int gfs2_drop_inode(struct inode *inode)
 997 {
 998         struct gfs2_inode *ip = GFS2_I(inode);
 999
1000         if (!test_bit(GIF_FREE_VFS_INODE, &ip->i_flags) &&
1001             inode->i_nlink &&
1002             gfs2_holder_initialized(&ip->i_iopen_gh)) {
1003                 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1004                 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
1005                         clear_nlink(inode);
1006         }
1007
1008         /*
1009          * When under memory pressure when an inode's link count has dropped to
1010          * zero, defer deleting the inode to the delete workqueue.  This avoids
1011          * calling into DLM under memory pressure, which can deadlock.
1012          */
1013         if (!inode->i_nlink &&
1014             unlikely(current->flags & PF_MEMALLOC) &&
1015             gfs2_holder_initialized(&ip->i_iopen_gh)) {
1016                 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1017
1018                 gfs2_glock_hold(gl);
1019                 if (queue_work(gfs2_delete_workqueue, &gl->gl_delete) == 0)
1020                         gfs2_glock_queue_put(gl);
1021                 return false;
1022         }
1023
1024         return generic_drop_inode(inode);
1025 }
1026
1027 static int is_ancestor(const struct dentry *d1, const struct dentry *d2)
1028 {
1029         do {
1030                 if (d1 == d2)
1031                         return 1;
1032                 d1 = d1->d_parent;
1033         } while (!IS_ROOT(d1));
1034         return 0;
1035 }
1036
1037 /**
1038  * gfs2_show_options - Show mount options for /proc/mounts
1039  * @s: seq_file structure
1040  * @root: root of this (sub)tree
1041  *
1042  * Returns: 0 on success or error code
1043  */
1044
1045 static int gfs2_show_options(struct seq_file *s, struct dentry *root)
1046 {
1047         struct gfs2_sbd *sdp = root->d_sb->s_fs_info;
1048         struct gfs2_args *args = &sdp->sd_args;
1049         unsigned int logd_secs, statfs_slow, statfs_quantum, quota_quantum;
1050
1051         spin_lock(&sdp->sd_tune.gt_spin);
1052         logd_secs = sdp->sd_tune.gt_logd_secs;
1053         quota_quantum = sdp->sd_tune.gt_quota_quantum;
1054         statfs_quantum = sdp->sd_tune.gt_statfs_quantum;
1055         statfs_slow = sdp->sd_tune.gt_statfs_slow;
1056         spin_unlock(&sdp->sd_tune.gt_spin);
1057
1058         if (is_ancestor(root, sdp->sd_master_dir))
1059                 seq_puts(s, ",meta");
1060         if (args->ar_lockproto[0])
1061                 seq_show_option(s, "lockproto", args->ar_lockproto);
1062         if (args->ar_locktable[0])
1063                 seq_show_option(s, "locktable", args->ar_locktable);
1064         if (args->ar_hostdata[0])
1065                 seq_show_option(s, "hostdata", args->ar_hostdata);
1066         if (args->ar_spectator)
1067                 seq_puts(s, ",spectator");
1068         if (args->ar_localflocks)
1069                 seq_puts(s, ",localflocks");
1070         if (args->ar_debug)
1071                 seq_puts(s, ",debug");
1072         if (args->ar_posix_acl)
1073                 seq_puts(s, ",acl");
1074         if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
1075                 char *state;
1076                 switch (args->ar_quota) {
1077                 case GFS2_QUOTA_OFF:
1078                         state = "off";
1079                         break;
1080                 case GFS2_QUOTA_ACCOUNT:
1081                         state = "account";
1082                         break;
1083                 case GFS2_QUOTA_ON:
1084                         state = "on";
1085                         break;
1086                 default:
1087                         state = "unknown";
1088                         break;
1089                 }
1090                 seq_printf(s, ",quota=%s", state);
1091         }
1092         if (args->ar_suiddir)
1093                 seq_puts(s, ",suiddir");
1094         if (args->ar_data != GFS2_DATA_DEFAULT) {
1095                 char *state;
1096                 switch (args->ar_data) {
1097                 case GFS2_DATA_WRITEBACK:
1098                         state = "writeback";
1099                         break;
1100                 case GFS2_DATA_ORDERED:
1101                         state = "ordered";
1102                         break;
1103                 default:
1104                         state = "unknown";
1105                         break;
1106                 }
1107                 seq_printf(s, ",data=%s", state);
1108         }
1109         if (args->ar_discard)
1110                 seq_puts(s, ",discard");
1111         if (logd_secs != 30)
1112                 seq_printf(s, ",commit=%d", logd_secs);
1113         if (statfs_quantum != 30)
1114                 seq_printf(s, ",statfs_quantum=%d", statfs_quantum);
1115         else if (statfs_slow)
1116                 seq_puts(s, ",statfs_quantum=0");
1117         if (quota_quantum != 60)
1118                 seq_printf(s, ",quota_quantum=%d", quota_quantum);
1119         if (args->ar_statfs_percent)
1120                 seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent);
1121         if (args->ar_errors != GFS2_ERRORS_DEFAULT) {
1122                 const char *state;
1123
1124                 switch (args->ar_errors) {
1125                 case GFS2_ERRORS_WITHDRAW:
1126                         state = "withdraw";
1127                         break;
1128                 case GFS2_ERRORS_PANIC:
1129                         state = "panic";
1130                         break;
1131                 default:
1132                         state = "unknown";
1133                         break;
1134                 }
1135                 seq_printf(s, ",errors=%s", state);
1136         }
1137         if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags))
1138                 seq_puts(s, ",nobarrier");
1139         if (test_bit(SDF_DEMOTE, &sdp->sd_flags))
1140                 seq_puts(s, ",demote_interface_used");
1141         if (args->ar_rgrplvb)
1142                 seq_puts(s, ",rgrplvb");
1143         if (args->ar_loccookie)
1144                 seq_puts(s, ",loccookie");
1145         return 0;
1146 }
1147
1148 static void gfs2_final_release_pages(struct gfs2_inode *ip)
1149 {
1150         struct inode *inode = &ip->i_inode;
1151         struct gfs2_glock *gl = ip->i_gl;
1152
1153         truncate_inode_pages(gfs2_glock2aspace(ip->i_gl), 0);
1154         truncate_inode_pages(&inode->i_data, 0);
1155
1156         if (atomic_read(&gl->gl_revokes) == 0) {
1157                 clear_bit(GLF_LFLUSH, &gl->gl_flags);
1158                 clear_bit(GLF_DIRTY, &gl->gl_flags);
1159         }
1160 }
1161
1162 static int gfs2_dinode_dealloc(struct gfs2_inode *ip)
1163 {
1164         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1165         struct gfs2_rgrpd *rgd;
1166         struct gfs2_holder gh;
1167         int error;
1168
1169         if (gfs2_get_inode_blocks(&ip->i_inode) != 1) {
1170                 gfs2_consist_inode(ip);
1171                 return -EIO;
1172         }
1173
1174         error = gfs2_rindex_update(sdp);
1175         if (error)
1176                 return error;
1177
1178         error = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
1179         if (error)
1180                 return error;
1181
1182         rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1183         if (!rgd) {
1184                 gfs2_consist_inode(ip);
1185                 error = -EIO;
1186                 goto out_qs;
1187         }
1188
1189         error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &gh);
1190         if (error)
1191                 goto out_qs;
1192
1193         error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA,
1194                                  sdp->sd_jdesc->jd_blocks);
1195         if (error)
1196                 goto out_rg_gunlock;
1197
1198         gfs2_free_di(rgd, ip);
1199
1200         gfs2_final_release_pages(ip);
1201
1202         gfs2_trans_end(sdp);
1203
1204 out_rg_gunlock:
1205         gfs2_glock_dq_uninit(&gh);
1206 out_qs:
1207         gfs2_quota_unhold(ip);
1208         return error;
1209 }
1210
1211 /**
1212  * gfs2_glock_put_eventually
1213  * @gl: The glock to put
1214  *
1215  * When under memory pressure, trigger a deferred glock put to make sure we
1216  * won't call into DLM and deadlock.  Otherwise, put the glock directly.
1217  */
1218
1219 static void gfs2_glock_put_eventually(struct gfs2_glock *gl)
1220 {
1221         if (current->flags & PF_MEMALLOC)
1222                 gfs2_glock_queue_put(gl);
1223         else
1224                 gfs2_glock_put(gl);
1225 }
1226
1227 /**
1228  * gfs2_evict_inode - Remove an inode from cache
1229  * @inode: The inode to evict
1230  *
1231  * There are three cases to consider:
1232  * 1. i_nlink == 0, we are final opener (and must deallocate)
1233  * 2. i_nlink == 0, we are not the final opener (and cannot deallocate)
1234  * 3. i_nlink > 0
1235  *
1236  * If the fs is read only, then we have to treat all cases as per #3
1237  * since we are unable to do any deallocation. The inode will be
1238  * deallocated by the next read/write node to attempt an allocation
1239  * in the same resource group
1240  *
1241  * We have to (at the moment) hold the inodes main lock to cover
1242  * the gap between unlocking the shared lock on the iopen lock and
1243  * taking the exclusive lock. I'd rather do a shared -> exclusive
1244  * conversion on the iopen lock, but we can change that later. This
1245  * is safe, just less efficient.
1246  */
1247
1248 static void gfs2_evict_inode(struct inode *inode)
1249 {
1250         struct super_block *sb = inode->i_sb;
1251         struct gfs2_sbd *sdp = sb->s_fs_info;
1252         struct gfs2_inode *ip = GFS2_I(inode);
1253         struct gfs2_holder gh;
1254         struct address_space *metamapping;
1255         int error;
1256
1257         if (test_bit(GIF_FREE_VFS_INODE, &ip->i_flags)) {
1258                 clear_inode(inode);
1259                 return;
1260         }
1261
1262         if (inode->i_nlink || sb_rdonly(sb))
1263                 goto out;
1264
1265         /*
1266          * In case of an incomplete mount, gfs2_evict_inode() may be called for
1267          * system files without having an active journal to write to.  In that
1268          * case, skip the filesystem evict.
1269          */
1270         if (!sdp->sd_jdesc)
1271                 goto out;
1272
1273         if (test_bit(GIF_ALLOC_FAILED, &ip->i_flags)) {
1274                 BUG_ON(!gfs2_glock_is_locked_by_me(ip->i_gl));
1275                 gfs2_holder_mark_uninitialized(&gh);
1276                 goto alloc_failed;
1277         }
1278
1279         /* Deletes should never happen under memory pressure anymore.  */
1280         if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
1281                 goto out;
1282
1283         /* Must not read inode block until block type has been verified */
1284         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, &gh);
1285         if (unlikely(error)) {
1286                 glock_clear_object(ip->i_iopen_gh.gh_gl, ip);
1287                 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1288                 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1289                 goto out;
1290         }
1291
1292         error = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED);
1293         if (error)
1294                 goto out_truncate;
1295
1296         if (test_bit(GIF_INVALID, &ip->i_flags)) {
1297                 error = gfs2_inode_refresh(ip);
1298                 if (error)
1299                         goto out_truncate;
1300         }
1301
1302         /*
1303          * The inode may have been recreated in the meantime.
1304          */
1305         if (inode->i_nlink)
1306                 goto out_truncate;
1307
1308 alloc_failed:
1309         if (gfs2_holder_initialized(&ip->i_iopen_gh) &&
1310             test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1311                 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1312                 gfs2_glock_dq_wait(&ip->i_iopen_gh);
1313                 gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE,
1314                                    &ip->i_iopen_gh);
1315                 error = gfs2_glock_nq(&ip->i_iopen_gh);
1316                 if (error)
1317                         goto out_truncate;
1318         }
1319
1320         if (S_ISDIR(inode->i_mode) &&
1321             (ip->i_diskflags & GFS2_DIF_EXHASH)) {
1322                 error = gfs2_dir_exhash_dealloc(ip);
1323                 if (error)
1324                         goto out_unlock;
1325         }
1326
1327         if (ip->i_eattr) {
1328                 error = gfs2_ea_dealloc(ip);
1329                 if (error)
1330                         goto out_unlock;
1331         }
1332
1333         if (!gfs2_is_stuffed(ip)) {
1334                 error = gfs2_file_dealloc(ip);
1335                 if (error)
1336                         goto out_unlock;
1337         }
1338
1339         /* We're about to clear the bitmap for the dinode, but as soon as we
1340            do, gfs2_create_inode can create another inode at the same block
1341            location and try to set gl_object again. We clear gl_object here so
1342            that subsequent inode creates don't see an old gl_object. */
1343         glock_clear_object(ip->i_gl, ip);
1344         error = gfs2_dinode_dealloc(ip);
1345         goto out_unlock;
1346
1347 out_truncate:
1348         gfs2_log_flush(sdp, ip->i_gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
1349                        GFS2_LFC_EVICT_INODE);
1350         metamapping = gfs2_glock2aspace(ip->i_gl);
1351         if (test_bit(GLF_DIRTY, &ip->i_gl->gl_flags)) {
1352                 filemap_fdatawrite(metamapping);
1353                 filemap_fdatawait(metamapping);
1354         }
1355         write_inode_now(inode, 1);
1356         gfs2_ail_flush(ip->i_gl, 0);
1357
1358         error = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
1359         if (error)
1360                 goto out_unlock;
1361         /* Needs to be done before glock release & also in a transaction */
1362         truncate_inode_pages(&inode->i_data, 0);
1363         truncate_inode_pages(metamapping, 0);
1364         gfs2_trans_end(sdp);
1365
1366 out_unlock:
1367         if (gfs2_rs_active(&ip->i_res))
1368                 gfs2_rs_deltree(&ip->i_res);
1369
1370         if (gfs2_holder_initialized(&ip->i_iopen_gh)) {
1371                 glock_clear_object(ip->i_iopen_gh.gh_gl, ip);
1372                 if (test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1373                         ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1374                         gfs2_glock_dq(&ip->i_iopen_gh);
1375                 }
1376                 gfs2_holder_uninit(&ip->i_iopen_gh);
1377         }
1378         if (gfs2_holder_initialized(&gh)) {
1379                 glock_clear_object(ip->i_gl, ip);
1380                 gfs2_glock_dq_uninit(&gh);
1381         }
1382         if (error && error != GLR_TRYFAILED && error != -EROFS)
1383                 fs_warn(sdp, "gfs2_evict_inode: %d\n", error);
1384 out:
1385         truncate_inode_pages_final(&inode->i_data);
1386         gfs2_rsqa_delete(ip, NULL);
1387         gfs2_ordered_del_inode(ip);
1388         clear_inode(inode);
1389         gfs2_dir_hash_inval(ip);
1390         glock_clear_object(ip->i_gl, ip);
1391         wait_on_bit_io(&ip->i_flags, GIF_GLOP_PENDING, TASK_UNINTERRUPTIBLE);
1392         gfs2_glock_add_to_lru(ip->i_gl);
1393         gfs2_glock_put_eventually(ip->i_gl);
1394         ip->i_gl = NULL;
1395         if (gfs2_holder_initialized(&ip->i_iopen_gh)) {
1396                 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1397
1398                 glock_clear_object(gl, ip);
1399                 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1400                 gfs2_glock_hold(gl);
1401                 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1402                 gfs2_glock_put_eventually(gl);
1403         }
1404 }
1405
1406 static struct inode *gfs2_alloc_inode(struct super_block *sb)
1407 {
1408         struct gfs2_inode *ip;
1409
1410         ip = kmem_cache_alloc(gfs2_inode_cachep, GFP_KERNEL);
1411         if (!ip)
1412                 return NULL;
1413         ip->i_flags = 0;
1414         ip->i_gl = NULL;
1415         memset(&ip->i_res, 0, sizeof(ip->i_res));
1416         RB_CLEAR_NODE(&ip->i_res.rs_node);
1417         ip->i_rahead = 0;
1418         return &ip->i_inode;
1419 }
1420
1421 static void gfs2_free_inode(struct inode *inode)
1422 {
1423         kmem_cache_free(gfs2_inode_cachep, GFS2_I(inode));
1424 }
1425
1426 const struct super_operations gfs2_super_ops = {
1427         .alloc_inode            = gfs2_alloc_inode,
1428         .free_inode             = gfs2_free_inode,
1429         .write_inode            = gfs2_write_inode,
1430         .dirty_inode            = gfs2_dirty_inode,
1431         .evict_inode            = gfs2_evict_inode,
1432         .put_super              = gfs2_put_super,
1433         .sync_fs                = gfs2_sync_fs,
1434         .freeze_super           = gfs2_freeze,
1435         .thaw_super             = gfs2_unfreeze,
1436         .statfs                 = gfs2_statfs,
1437         .drop_inode             = gfs2_drop_inode,
1438         .show_options           = gfs2_show_options,
1439 };
1440