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 static int init_threads(struct gfs2_sbd *sdp)
 130 {
 131         struct task_struct *p;
 132         int error = 0;
 133
 134         p = kthread_run(gfs2_logd, sdp, "gfs2_logd");
 135         if (IS_ERR(p)) {
 136                 error = PTR_ERR(p);
 137                 fs_err(sdp, "can't start logd thread: %d\n", error);
 138                 return error;
 139         }
 140         sdp->sd_logd_process = p;
 141
 142         p = kthread_run(gfs2_quotad, sdp, "gfs2_quotad");
 143         if (IS_ERR(p)) {
 144                 error = PTR_ERR(p);
 145                 fs_err(sdp, "can't start quotad thread: %d\n", error);
 146                 goto fail;
 147         }
 148         sdp->sd_quotad_process = p;
 149         return 0;
 150
 151 fail:
 152         kthread_stop(sdp->sd_logd_process);
 153         sdp->sd_logd_process = NULL;
 154         return error;
 155 }
 156
 157 /**
 158  * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
 159  * @sdp: the filesystem
 160  *
 161  * Returns: errno
 162  */
 163
 164 int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
 165 {
 166         struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
 167         struct gfs2_glock *j_gl = ip->i_gl;
 168         struct gfs2_holder freeze_gh;
 169         struct gfs2_log_header_host head;
 170         int error;
 171
 172         error = init_threads(sdp);
 173         if (error)
 174                 return error;
 175
 176         error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED,
 177                                    LM_FLAG_NOEXP | GL_EXACT,
 178                                    &freeze_gh);
 179         if (error)
 180                 goto fail_threads;
 181
 182         j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
 183         if (gfs2_withdrawn(sdp)) {
 184                 error = -EIO;
 185                 goto fail;
 186         }
 187
 188         error = gfs2_find_jhead(sdp->sd_jdesc, &head, false);
 189         if (error || gfs2_withdrawn(sdp))
 190                 goto fail;
 191
 192         if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 193                 gfs2_consist(sdp);
 194                 error = -EIO;
 195                 goto fail;
 196         }
 197
 198         /*  Initialize some head of the log stuff  */
 199         sdp->sd_log_sequence = head.lh_sequence + 1;
 200         gfs2_log_pointers_init(sdp, head.lh_blkno);
 201
 202         error = gfs2_quota_init(sdp);
 203         if (error || gfs2_withdrawn(sdp))
 204                 goto fail;
 205
 206         set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 207
 208         gfs2_glock_dq_uninit(&freeze_gh);
 209
 210         return 0;
 211
 212 fail:
 213         gfs2_glock_dq_uninit(&freeze_gh);
 214 fail_threads:
 215         if (sdp->sd_quotad_process)
 216                 kthread_stop(sdp->sd_quotad_process);
 217         sdp->sd_quotad_process = NULL;
 218         if (sdp->sd_logd_process)
 219                 kthread_stop(sdp->sd_logd_process);
 220         sdp->sd_logd_process = NULL;
 221         return error;
 222 }
 223
 224 void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
 225 {
 226         const struct gfs2_statfs_change *str = buf;
 227
 228         sc->sc_total = be64_to_cpu(str->sc_total);
 229         sc->sc_free = be64_to_cpu(str->sc_free);
 230         sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
 231 }
 232
 233 void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
 234 {
 235         struct gfs2_statfs_change *str = buf;
 236
 237         str->sc_total = cpu_to_be64(sc->sc_total);
 238         str->sc_free = cpu_to_be64(sc->sc_free);
 239         str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
 240 }
 241
 242 int gfs2_statfs_init(struct gfs2_sbd *sdp)
 243 {
 244         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 245         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 246         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 247         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 248         struct buffer_head *m_bh, *l_bh;
 249         struct gfs2_holder gh;
 250         int error;
 251
 252         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 253                                    &gh);
 254         if (error)
 255                 return error;
 256
 257         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 258         if (error)
 259                 goto out;
 260
 261         if (sdp->sd_args.ar_spectator) {
 262                 spin_lock(&sdp->sd_statfs_spin);
 263                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 264                                       sizeof(struct gfs2_dinode));
 265                 spin_unlock(&sdp->sd_statfs_spin);
 266         } else {
 267                 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 268                 if (error)
 269                         goto out_m_bh;
 270
 271                 spin_lock(&sdp->sd_statfs_spin);
 272                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 273                                       sizeof(struct gfs2_dinode));
 274                 gfs2_statfs_change_in(l_sc, l_bh->b_data +
 275                                       sizeof(struct gfs2_dinode));
 276                 spin_unlock(&sdp->sd_statfs_spin);
 277
 278                 brelse(l_bh);
 279         }
 280
 281 out_m_bh:
 282         brelse(m_bh);
 283 out:
 284         gfs2_glock_dq_uninit(&gh);
 285         return 0;
 286 }
 287
 288 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
 289                         s64 dinodes)
 290 {
 291         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 292         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 293         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 294         struct buffer_head *l_bh;
 295         s64 x, y;
 296         int need_sync = 0;
 297         int error;
 298
 299         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 300         if (error)
 301                 return;
 302
 303         gfs2_trans_add_meta(l_ip->i_gl, l_bh);
 304
 305         spin_lock(&sdp->sd_statfs_spin);
 306         l_sc->sc_total += total;
 307         l_sc->sc_free += free;
 308         l_sc->sc_dinodes += dinodes;
 309         gfs2_statfs_change_out(l_sc, l_bh->b_data + sizeof(struct gfs2_dinode));
 310         if (sdp->sd_args.ar_statfs_percent) {
 311                 x = 100 * l_sc->sc_free;
 312                 y = m_sc->sc_free * sdp->sd_args.ar_statfs_percent;
 313                 if (x >= y || x <= -y)
 314                         need_sync = 1;
 315         }
 316         spin_unlock(&sdp->sd_statfs_spin);
 317
 318         brelse(l_bh);
 319         if (need_sync)
 320                 gfs2_wake_up_statfs(sdp);
 321 }
 322
 323 void update_statfs(struct gfs2_sbd *sdp, struct buffer_head *m_bh,
 324                    struct buffer_head *l_bh)
 325 {
 326         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 327         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 328         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 329         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 330
 331         gfs2_trans_add_meta(l_ip->i_gl, l_bh);
 332         gfs2_trans_add_meta(m_ip->i_gl, m_bh);
 333
 334         spin_lock(&sdp->sd_statfs_spin);
 335         m_sc->sc_total += l_sc->sc_total;
 336         m_sc->sc_free += l_sc->sc_free;
 337         m_sc->sc_dinodes += l_sc->sc_dinodes;
 338         memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
 339         memset(l_bh->b_data + sizeof(struct gfs2_dinode),
 340                0, sizeof(struct gfs2_statfs_change));
 341         gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
 342         spin_unlock(&sdp->sd_statfs_spin);
 343 }
 344
 345 int gfs2_statfs_sync(struct super_block *sb, int type)
 346 {
 347         struct gfs2_sbd *sdp = sb->s_fs_info;
 348         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 349         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 350         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 351         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 352         struct gfs2_holder gh;
 353         struct buffer_head *m_bh, *l_bh;
 354         int error;
 355
 356         sb_start_write(sb);
 357         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 358                                    &gh);
 359         if (error)
 360                 goto out;
 361
 362         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 363         if (error)
 364                 goto out_unlock;
 365
 366         spin_lock(&sdp->sd_statfs_spin);
 367         gfs2_statfs_change_in(m_sc, m_bh->b_data +
 368                               sizeof(struct gfs2_dinode));
 369         if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
 370                 spin_unlock(&sdp->sd_statfs_spin);
 371                 goto out_bh;
 372         }
 373         spin_unlock(&sdp->sd_statfs_spin);
 374
 375         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 376         if (error)
 377                 goto out_bh;
 378
 379         error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
 380         if (error)
 381                 goto out_bh2;
 382
 383         update_statfs(sdp, m_bh, l_bh);
 384         sdp->sd_statfs_force_sync = 0;
 385
 386         gfs2_trans_end(sdp);
 387
 388 out_bh2:
 389         brelse(l_bh);
 390 out_bh:
 391         brelse(m_bh);
 392 out_unlock:
 393         gfs2_glock_dq_uninit(&gh);
 394 out:
 395         sb_end_write(sb);
 396         return error;
 397 }
 398
 399 struct lfcc {
 400         struct list_head list;
 401         struct gfs2_holder gh;
 402 };
 403
 404 /**
 405  * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
 406  *                            journals are clean
 407  * @sdp: the file system
 408  * @state: the state to put the transaction lock into
 409  * @t_gh: the hold on the transaction lock
 410  *
 411  * Returns: errno
 412  */
 413
 414 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp)
 415 {
 416         struct gfs2_inode *ip;
 417         struct gfs2_jdesc *jd;
 418         struct lfcc *lfcc;
 419         LIST_HEAD(list);
 420         struct gfs2_log_header_host lh;
 421         int error;
 422
 423         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 424                 lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
 425                 if (!lfcc) {
 426                         error = -ENOMEM;
 427                         goto out;
 428                 }
 429                 ip = GFS2_I(jd->jd_inode);
 430                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
 431                 if (error) {
 432                         kfree(lfcc);
 433                         goto out;
 434                 }
 435                 list_add(&lfcc->list, &list);
 436         }
 437
 438         error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_EXCLUSIVE,
 439                                    LM_FLAG_NOEXP, &sdp->sd_freeze_gh);
 440         if (error)
 441                 goto out;
 442
 443         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 444                 error = gfs2_jdesc_check(jd);
 445                 if (error)
 446                         break;
 447                 error = gfs2_find_jhead(jd, &lh, false);
 448                 if (error)
 449                         break;
 450                 if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 451                         error = -EBUSY;
 452                         break;
 453                 }
 454         }
 455
 456         if (error)
 457                 gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
 458
 459 out:
 460         while (!list_empty(&list)) {
 461                 lfcc = list_first_entry(&list, struct lfcc, list);
 462                 list_del(&lfcc->list);
 463                 gfs2_glock_dq_uninit(&lfcc->gh);
 464                 kfree(lfcc);
 465         }
 466         return error;
 467 }
 468
 469 void gfs2_dinode_out(const struct gfs2_inode *ip, void *buf)
 470 {
 471         struct gfs2_dinode *str = buf;
 472
 473         str->di_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 474         str->di_header.mh_type = cpu_to_be32(GFS2_METATYPE_DI);
 475         str->di_header.mh_format = cpu_to_be32(GFS2_FORMAT_DI);
 476         str->di_num.no_addr = cpu_to_be64(ip->i_no_addr);
 477         str->di_num.no_formal_ino = cpu_to_be64(ip->i_no_formal_ino);
 478         str->di_mode = cpu_to_be32(ip->i_inode.i_mode);
 479         str->di_uid = cpu_to_be32(i_uid_read(&ip->i_inode));
 480         str->di_gid = cpu_to_be32(i_gid_read(&ip->i_inode));
 481         str->di_nlink = cpu_to_be32(ip->i_inode.i_nlink);
 482         str->di_size = cpu_to_be64(i_size_read(&ip->i_inode));
 483         str->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(&ip->i_inode));
 484         str->di_atime = cpu_to_be64(ip->i_inode.i_atime.tv_sec);
 485         str->di_mtime = cpu_to_be64(ip->i_inode.i_mtime.tv_sec);
 486         str->di_ctime = cpu_to_be64(ip->i_inode.i_ctime.tv_sec);
 487
 488         str->di_goal_meta = cpu_to_be64(ip->i_goal);
 489         str->di_goal_data = cpu_to_be64(ip->i_goal);
 490         str->di_generation = cpu_to_be64(ip->i_generation);
 491
 492         str->di_flags = cpu_to_be32(ip->i_diskflags);
 493         str->di_height = cpu_to_be16(ip->i_height);
 494         str->di_payload_format = cpu_to_be32(S_ISDIR(ip->i_inode.i_mode) &&
 495                                              !(ip->i_diskflags & GFS2_DIF_EXHASH) ?
 496                                              GFS2_FORMAT_DE : 0);
 497         str->di_depth = cpu_to_be16(ip->i_depth);
 498         str->di_entries = cpu_to_be32(ip->i_entries);
 499
 500         str->di_eattr = cpu_to_be64(ip->i_eattr);
 501         str->di_atime_nsec = cpu_to_be32(ip->i_inode.i_atime.tv_nsec);
 502         str->di_mtime_nsec = cpu_to_be32(ip->i_inode.i_mtime.tv_nsec);
 503         str->di_ctime_nsec = cpu_to_be32(ip->i_inode.i_ctime.tv_nsec);
 504 }
 505
 506 /**
 507  * gfs2_write_inode - Make sure the inode is stable on the disk
 508  * @inode: The inode
 509  * @wbc: The writeback control structure
 510  *
 511  * Returns: errno
 512  */
 513
 514 static int gfs2_write_inode(struct inode *inode, struct writeback_control *wbc)
 515 {
 516         struct gfs2_inode *ip = GFS2_I(inode);
 517         struct gfs2_sbd *sdp = GFS2_SB(inode);
 518         struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl);
 519         struct backing_dev_info *bdi = inode_to_bdi(metamapping->host);
 520         int ret = 0;
 521         bool flush_all = (wbc->sync_mode == WB_SYNC_ALL || gfs2_is_jdata(ip));
 522
 523         if (flush_all)
 524                 gfs2_log_flush(GFS2_SB(inode), ip->i_gl,
 525                                GFS2_LOG_HEAD_FLUSH_NORMAL |
 526                                GFS2_LFC_WRITE_INODE);
 527         if (bdi->wb.dirty_exceeded)
 528                 gfs2_ail1_flush(sdp, wbc);
 529         else
 530                 filemap_fdatawrite(metamapping);
 531         if (flush_all)
 532                 ret = filemap_fdatawait(metamapping);
 533         if (ret)
 534                 mark_inode_dirty_sync(inode);
 535         else {
 536                 spin_lock(&inode->i_lock);
 537                 if (!(inode->i_flags & I_DIRTY))
 538                         gfs2_ordered_del_inode(ip);
 539                 spin_unlock(&inode->i_lock);
 540         }
 541         return ret;
 542 }
 543
 544 /**
 545  * gfs2_dirty_inode - check for atime updates
 546  * @inode: The inode in question
 547  * @flags: The type of dirty
 548  *
 549  * Unfortunately it can be called under any combination of inode
 550  * glock and transaction lock, so we have to check carefully.
 551  *
 552  * At the moment this deals only with atime - it should be possible
 553  * to expand that role in future, once a review of the locking has
 554  * been carried out.
 555  */
 556
 557 static void gfs2_dirty_inode(struct inode *inode, int flags)
 558 {
 559         struct gfs2_inode *ip = GFS2_I(inode);
 560         struct gfs2_sbd *sdp = GFS2_SB(inode);
 561         struct buffer_head *bh;
 562         struct gfs2_holder gh;
 563         int need_unlock = 0;
 564         int need_endtrans = 0;
 565         int ret;
 566
 567         if (!(flags & I_DIRTY_INODE))
 568                 return;
 569         if (unlikely(gfs2_withdrawn(sdp)))
 570                 return;
 571         if (!gfs2_glock_is_locked_by_me(ip->i_gl)) {
 572                 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
 573                 if (ret) {
 574                         fs_err(sdp, "dirty_inode: glock %d\n", ret);
 575                         gfs2_dump_glock(NULL, ip->i_gl, true);
 576                         return;
 577                 }
 578                 need_unlock = 1;
 579         } else if (WARN_ON_ONCE(ip->i_gl->gl_state != LM_ST_EXCLUSIVE))
 580                 return;
 581
 582         if (current->journal_info == NULL) {
 583                 ret = gfs2_trans_begin(sdp, RES_DINODE, 0);
 584                 if (ret) {
 585                         fs_err(sdp, "dirty_inode: gfs2_trans_begin %d\n", ret);
 586                         goto out;
 587                 }
 588                 need_endtrans = 1;
 589         }
 590
 591         ret = gfs2_meta_inode_buffer(ip, &bh);
 592         if (ret == 0) {
 593                 gfs2_trans_add_meta(ip->i_gl, bh);
 594                 gfs2_dinode_out(ip, bh->b_data);
 595                 brelse(bh);
 596         }
 597
 598         if (need_endtrans)
 599                 gfs2_trans_end(sdp);
 600 out:
 601         if (need_unlock)
 602                 gfs2_glock_dq_uninit(&gh);
 603 }
 604
 605 /**
 606  * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
 607  * @sdp: the filesystem
 608  *
 609  * Returns: errno
 610  */
 611
 612 int gfs2_make_fs_ro(struct gfs2_sbd *sdp)
 613 {
 614         struct gfs2_holder freeze_gh;
 615         int error = 0;
 616         int log_write_allowed = test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 617
 618         gfs2_holder_mark_uninitialized(&freeze_gh);
 619         if (sdp->sd_freeze_gl &&
 620             !gfs2_glock_is_locked_by_me(sdp->sd_freeze_gl)) {
 621                 if (!log_write_allowed) {
 622                         error = gfs2_glock_nq_init(sdp->sd_freeze_gl,
 623                                                    LM_ST_SHARED, LM_FLAG_TRY |
 624                                                    LM_FLAG_NOEXP | GL_EXACT,
 625                                                    &freeze_gh);
 626                         if (error == GLR_TRYFAILED)
 627                                 error = 0;
 628                 } else {
 629                         error = gfs2_glock_nq_init(sdp->sd_freeze_gl,
 630                                                    LM_ST_SHARED,
 631                                                    LM_FLAG_NOEXP | GL_EXACT,
 632                                                    &freeze_gh);
 633                         if (error && !gfs2_withdrawn(sdp))
 634                                 return error;
 635                 }
 636         }
 637
 638         gfs2_flush_delete_work(sdp);
 639         if (!log_write_allowed && current == sdp->sd_quotad_process)
 640                 fs_warn(sdp, "The quotad daemon is withdrawing.\n");
 641         else if (sdp->sd_quotad_process)
 642                 kthread_stop(sdp->sd_quotad_process);
 643         sdp->sd_quotad_process = NULL;
 644
 645         if (!log_write_allowed && current == sdp->sd_logd_process)
 646                 fs_warn(sdp, "The logd daemon is withdrawing.\n");
 647         else if (sdp->sd_logd_process)
 648                 kthread_stop(sdp->sd_logd_process);
 649         sdp->sd_logd_process = NULL;
 650
 651         if (log_write_allowed) {
 652                 gfs2_quota_sync(sdp->sd_vfs, 0);
 653                 gfs2_statfs_sync(sdp->sd_vfs, 0);
 654
 655                 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SHUTDOWN |
 656                                GFS2_LFC_MAKE_FS_RO);
 657                 wait_event(sdp->sd_reserving_log_wait,
 658                            atomic_read(&sdp->sd_reserving_log) == 0);
 659                 gfs2_assert_warn(sdp, atomic_read(&sdp->sd_log_blks_free) ==
 660                                  sdp->sd_jdesc->jd_blocks);
 661         } else {
 662                 wait_event_timeout(sdp->sd_reserving_log_wait,
 663                                    atomic_read(&sdp->sd_reserving_log) == 0,
 664                                    HZ * 5);
 665         }
 666         if (gfs2_holder_initialized(&freeze_gh))
 667                 gfs2_glock_dq_uninit(&freeze_gh);
 668
 669         gfs2_quota_cleanup(sdp);
 670
 671         if (!log_write_allowed)
 672                 sdp->sd_vfs->s_flags |= SB_RDONLY;
 673
 674         return error;
 675 }
 676
 677 /**
 678  * gfs2_put_super - Unmount the filesystem
 679  * @sb: The VFS superblock
 680  *
 681  */
 682
 683 static void gfs2_put_super(struct super_block *sb)
 684 {
 685         struct gfs2_sbd *sdp = sb->s_fs_info;
 686         int error;
 687         struct gfs2_jdesc *jd;
 688
 689         /* No more recovery requests */
 690         set_bit(SDF_NORECOVERY, &sdp->sd_flags);
 691         smp_mb();
 692
 693         /* Wait on outstanding recovery */
 694 restart:
 695         spin_lock(&sdp->sd_jindex_spin);
 696         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 697                 if (!test_bit(JDF_RECOVERY, &jd->jd_flags))
 698                         continue;
 699                 spin_unlock(&sdp->sd_jindex_spin);
 700                 wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
 701                             TASK_UNINTERRUPTIBLE);
 702                 goto restart;
 703         }
 704         spin_unlock(&sdp->sd_jindex_spin);
 705
 706         if (!sb_rdonly(sb)) {
 707                 error = gfs2_make_fs_ro(sdp);
 708                 if (error)
 709                         gfs2_io_error(sdp);
 710         }
 711         WARN_ON(gfs2_withdrawing(sdp));
 712
 713         /*  At this point, we're through modifying the disk  */
 714
 715         /*  Release stuff  */
 716
 717         iput(sdp->sd_jindex);
 718         iput(sdp->sd_statfs_inode);
 719         iput(sdp->sd_rindex);
 720         iput(sdp->sd_quota_inode);
 721
 722         gfs2_glock_put(sdp->sd_rename_gl);
 723         gfs2_glock_put(sdp->sd_freeze_gl);
 724
 725         if (!sdp->sd_args.ar_spectator) {
 726                 if (gfs2_holder_initialized(&sdp->sd_journal_gh))
 727                         gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
 728                 if (gfs2_holder_initialized(&sdp->sd_jinode_gh))
 729                         gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
 730                 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
 731                 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
 732                 free_local_statfs_inodes(sdp);
 733                 iput(sdp->sd_qc_inode);
 734         }
 735
 736         gfs2_glock_dq_uninit(&sdp->sd_live_gh);
 737         gfs2_clear_rgrpd(sdp);
 738         gfs2_jindex_free(sdp);
 739         /*  Take apart glock structures and buffer lists  */
 740         gfs2_gl_hash_clear(sdp);
 741         truncate_inode_pages_final(&sdp->sd_aspace);
 742         gfs2_delete_debugfs_file(sdp);
 743         /*  Unmount the locking protocol  */
 744         gfs2_lm_unmount(sdp);
 745
 746         /*  At this point, we're through participating in the lockspace  */
 747         gfs2_sys_fs_del(sdp);
 748         free_sbd(sdp);
 749 }
 750
 751 /**
 752  * gfs2_sync_fs - sync the filesystem
 753  * @sb: the superblock
 754  *
 755  * Flushes the log to disk.
 756  */
 757
 758 static int gfs2_sync_fs(struct super_block *sb, int wait)
 759 {
 760         struct gfs2_sbd *sdp = sb->s_fs_info;
 761
 762         gfs2_quota_sync(sb, -1);
 763         if (wait)
 764                 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
 765                                GFS2_LFC_SYNC_FS);
 766         return sdp->sd_log_error;
 767 }
 768
 769 void gfs2_freeze_func(struct work_struct *work)
 770 {
 771         int error;
 772         struct gfs2_holder freeze_gh;
 773         struct gfs2_sbd *sdp = container_of(work, struct gfs2_sbd, sd_freeze_work);
 774         struct super_block *sb = sdp->sd_vfs;
 775
 776         atomic_inc(&sb->s_active);
 777         error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED,
 778                                    LM_FLAG_NOEXP | GL_EXACT, &freeze_gh);
 779         if (error) {
 780                 fs_info(sdp, "GFS2: couldn't get freeze lock : %d\n", error);
 781                 gfs2_assert_withdraw(sdp, 0);
 782         } else {
 783                 atomic_set(&sdp->sd_freeze_state, SFS_UNFROZEN);
 784                 error = thaw_super(sb);
 785                 if (error) {
 786                         fs_info(sdp, "GFS2: couldn't thaw filesystem: %d\n",
 787                                 error);
 788                         gfs2_assert_withdraw(sdp, 0);
 789                 }
 790                 gfs2_glock_dq_uninit(&freeze_gh);
 791         }
 792         deactivate_super(sb);
 793         clear_bit_unlock(SDF_FS_FROZEN, &sdp->sd_flags);
 794         wake_up_bit(&sdp->sd_flags, SDF_FS_FROZEN);
 795         return;
 796 }
 797
 798 /**
 799  * gfs2_freeze - prevent further writes to the filesystem
 800  * @sb: the VFS structure for the filesystem
 801  *
 802  */
 803
 804 static int gfs2_freeze(struct super_block *sb)
 805 {
 806         struct gfs2_sbd *sdp = sb->s_fs_info;
 807         int error = 0;
 808
 809         mutex_lock(&sdp->sd_freeze_mutex);
 810         if (atomic_read(&sdp->sd_freeze_state) != SFS_UNFROZEN)
 811                 goto out;
 812
 813         for (;;) {
 814                 if (gfs2_withdrawn(sdp)) {
 815                         error = -EINVAL;
 816                         goto out;
 817                 }
 818
 819                 error = gfs2_lock_fs_check_clean(sdp);
 820                 if (!error)
 821                         break;
 822
 823                 if (error == -EBUSY)
 824                         fs_err(sdp, "waiting for recovery before freeze\n");
 825                 else if (error == -EIO) {
 826                         fs_err(sdp, "Fatal IO error: cannot freeze gfs2 due "
 827                                "to recovery error.\n");
 828                         goto out;
 829                 } else {
 830                         fs_err(sdp, "error freezing FS: %d\n", error);
 831                 }
 832                 fs_err(sdp, "retrying...\n");
 833                 msleep(1000);
 834         }
 835         set_bit(SDF_FS_FROZEN, &sdp->sd_flags);
 836 out:
 837         mutex_unlock(&sdp->sd_freeze_mutex);
 838         return error;
 839 }
 840
 841 /**
 842  * gfs2_unfreeze - reallow writes to the filesystem
 843  * @sb: the VFS structure for the filesystem
 844  *
 845  */
 846
 847 static int gfs2_unfreeze(struct super_block *sb)
 848 {
 849         struct gfs2_sbd *sdp = sb->s_fs_info;
 850
 851         mutex_lock(&sdp->sd_freeze_mutex);
 852         if (atomic_read(&sdp->sd_freeze_state) != SFS_FROZEN ||
 853             !gfs2_holder_initialized(&sdp->sd_freeze_gh)) {
 854                 mutex_unlock(&sdp->sd_freeze_mutex);
 855                 return 0;
 856         }
 857
 858         gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
 859         mutex_unlock(&sdp->sd_freeze_mutex);
 860         return wait_on_bit(&sdp->sd_flags, SDF_FS_FROZEN, TASK_INTERRUPTIBLE);
 861 }
 862
 863 /**
 864  * statfs_fill - fill in the sg for a given RG
 865  * @rgd: the RG
 866  * @sc: the sc structure
 867  *
 868  * Returns: 0 on success, -ESTALE if the LVB is invalid
 869  */
 870
 871 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
 872                             struct gfs2_statfs_change_host *sc)
 873 {
 874         gfs2_rgrp_verify(rgd);
 875         sc->sc_total += rgd->rd_data;
 876         sc->sc_free += rgd->rd_free;
 877         sc->sc_dinodes += rgd->rd_dinodes;
 878         return 0;
 879 }
 880
 881 /**
 882  * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
 883  * @sdp: the filesystem
 884  * @sc: the sc info that will be returned
 885  *
 886  * Any error (other than a signal) will cause this routine to fall back
 887  * to the synchronous version.
 888  *
 889  * FIXME: This really shouldn't busy wait like this.
 890  *
 891  * Returns: errno
 892  */
 893
 894 static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
 895 {
 896         struct gfs2_rgrpd *rgd_next;
 897         struct gfs2_holder *gha, *gh;
 898         unsigned int slots = 64;
 899         unsigned int x;
 900         int done;
 901         int error = 0, err;
 902
 903         memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
 904         gha = kmalloc_array(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
 905         if (!gha)
 906                 return -ENOMEM;
 907         for (x = 0; x < slots; x++)
 908                 gfs2_holder_mark_uninitialized(gha + x);
 909
 910         rgd_next = gfs2_rgrpd_get_first(sdp);
 911
 912         for (;;) {
 913                 done = 1;
 914
 915                 for (x = 0; x < slots; x++) {
 916                         gh = gha + x;
 917
 918                         if (gfs2_holder_initialized(gh) && gfs2_glock_poll(gh)) {
 919                                 err = gfs2_glock_wait(gh);
 920                                 if (err) {
 921                                         gfs2_holder_uninit(gh);
 922                                         error = err;
 923                                 } else {
 924                                         if (!error) {
 925                                                 struct gfs2_rgrpd *rgd =
 926                                                         gfs2_glock2rgrp(gh->gh_gl);
 927
 928                                                 error = statfs_slow_fill(rgd, sc);
 929                                         }
 930                                         gfs2_glock_dq_uninit(gh);
 931                                 }
 932                         }
 933
 934                         if (gfs2_holder_initialized(gh))
 935                                 done = 0;
 936                         else if (rgd_next && !error) {
 937                                 error = gfs2_glock_nq_init(rgd_next->rd_gl,
 938                                                            LM_ST_SHARED,
 939                                                            GL_ASYNC,
 940                                                            gh);
 941                                 rgd_next = gfs2_rgrpd_get_next(rgd_next);
 942                                 done = 0;
 943                         }
 944
 945                         if (signal_pending(current))
 946                                 error = -ERESTARTSYS;
 947                 }
 948
 949                 if (done)
 950                         break;
 951
 952                 yield();
 953         }
 954
 955         kfree(gha);
 956         return error;
 957 }
 958
 959 /**
 960  * gfs2_statfs_i - Do a statfs
 961  * @sdp: the filesystem
 962  * @sg: the sg structure
 963  *
 964  * Returns: errno
 965  */
 966
 967 static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
 968 {
 969         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 970         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 971
 972         spin_lock(&sdp->sd_statfs_spin);
 973
 974         *sc = *m_sc;
 975         sc->sc_total += l_sc->sc_total;
 976         sc->sc_free += l_sc->sc_free;
 977         sc->sc_dinodes += l_sc->sc_dinodes;
 978
 979         spin_unlock(&sdp->sd_statfs_spin);
 980
 981         if (sc->sc_free < 0)
 982                 sc->sc_free = 0;
 983         if (sc->sc_free > sc->sc_total)
 984                 sc->sc_free = sc->sc_total;
 985         if (sc->sc_dinodes < 0)
 986                 sc->sc_dinodes = 0;
 987
 988         return 0;
 989 }
 990
 991 /**
 992  * gfs2_statfs - Gather and return stats about the filesystem
 993  * @sb: The superblock
 994  * @statfsbuf: The buffer
 995  *
 996  * Returns: 0 on success or error code
 997  */
 998
 999 static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
1000 {
1001         struct super_block *sb = dentry->d_sb;
1002         struct gfs2_sbd *sdp = sb->s_fs_info;
1003         struct gfs2_statfs_change_host sc;
1004         int error;
1005
1006         error = gfs2_rindex_update(sdp);
1007         if (error)
1008                 return error;
1009
1010         if (gfs2_tune_get(sdp, gt_statfs_slow))
1011                 error = gfs2_statfs_slow(sdp, &sc);
1012         else
1013                 error = gfs2_statfs_i(sdp, &sc);
1014
1015         if (error)
1016                 return error;
1017
1018         buf->f_type = GFS2_MAGIC;
1019         buf->f_bsize = sdp->sd_sb.sb_bsize;
1020         buf->f_blocks = sc.sc_total;
1021         buf->f_bfree = sc.sc_free;
1022         buf->f_bavail = sc.sc_free;
1023         buf->f_files = sc.sc_dinodes + sc.sc_free;
1024         buf->f_ffree = sc.sc_free;
1025         buf->f_namelen = GFS2_FNAMESIZE;
1026
1027         return 0;
1028 }
1029
1030 /**
1031  * gfs2_drop_inode - Drop an inode (test for remote unlink)
1032  * @inode: The inode to drop
1033  *
1034  * If we've received a callback on an iopen lock then it's because a
1035  * remote node tried to deallocate the inode but failed due to this node
1036  * still having the inode open. Here we mark the link count zero
1037  * since we know that it must have reached zero if the GLF_DEMOTE flag
1038  * is set on the iopen glock. If we didn't do a disk read since the
1039  * remote node removed the final link then we might otherwise miss
1040  * this event. This check ensures that this node will deallocate the
1041  * inode's blocks, or alternatively pass the baton on to another
1042  * node for later deallocation.
1043  */
1044
1045 static int gfs2_drop_inode(struct inode *inode)
1046 {
1047         struct gfs2_inode *ip = GFS2_I(inode);
1048
1049         if (!test_bit(GIF_FREE_VFS_INODE, &ip->i_flags) &&
1050             inode->i_nlink &&
1051             gfs2_holder_initialized(&ip->i_iopen_gh)) {
1052                 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1053                 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
1054                         clear_nlink(inode);
1055         }
1056
1057         /*
1058          * When under memory pressure when an inode's link count has dropped to
1059          * zero, defer deleting the inode to the delete workqueue.  This avoids
1060          * calling into DLM under memory pressure, which can deadlock.
1061          */
1062         if (!inode->i_nlink &&
1063             unlikely(current->flags & PF_MEMALLOC) &&
1064             gfs2_holder_initialized(&ip->i_iopen_gh)) {
1065                 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1066
1067                 gfs2_glock_hold(gl);
1068                 if (!gfs2_queue_delete_work(gl, 0))
1069                         gfs2_glock_queue_put(gl);
1070                 return false;
1071         }
1072
1073         return generic_drop_inode(inode);
1074 }
1075
1076 static int is_ancestor(const struct dentry *d1, const struct dentry *d2)
1077 {
1078         do {
1079                 if (d1 == d2)
1080                         return 1;
1081                 d1 = d1->d_parent;
1082         } while (!IS_ROOT(d1));
1083         return 0;
1084 }
1085
1086 /**
1087  * gfs2_show_options - Show mount options for /proc/mounts
1088  * @s: seq_file structure
1089  * @root: root of this (sub)tree
1090  *
1091  * Returns: 0 on success or error code
1092  */
1093
1094 static int gfs2_show_options(struct seq_file *s, struct dentry *root)
1095 {
1096         struct gfs2_sbd *sdp = root->d_sb->s_fs_info;
1097         struct gfs2_args *args = &sdp->sd_args;
1098         int val;
1099
1100         if (is_ancestor(root, sdp->sd_master_dir))
1101                 seq_puts(s, ",meta");
1102         if (args->ar_lockproto[0])
1103                 seq_show_option(s, "lockproto", args->ar_lockproto);
1104         if (args->ar_locktable[0])
1105                 seq_show_option(s, "locktable", args->ar_locktable);
1106         if (args->ar_hostdata[0])
1107                 seq_show_option(s, "hostdata", args->ar_hostdata);
1108         if (args->ar_spectator)
1109                 seq_puts(s, ",spectator");
1110         if (args->ar_localflocks)
1111                 seq_puts(s, ",localflocks");
1112         if (args->ar_debug)
1113                 seq_puts(s, ",debug");
1114         if (args->ar_posix_acl)
1115                 seq_puts(s, ",acl");
1116         if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
1117                 char *state;
1118                 switch (args->ar_quota) {
1119                 case GFS2_QUOTA_OFF:
1120                         state = "off";
1121                         break;
1122                 case GFS2_QUOTA_ACCOUNT:
1123                         state = "account";
1124                         break;
1125                 case GFS2_QUOTA_ON:
1126                         state = "on";
1127                         break;
1128                 default:
1129                         state = "unknown";
1130                         break;
1131                 }
1132                 seq_printf(s, ",quota=%s", state);
1133         }
1134         if (args->ar_suiddir)
1135                 seq_puts(s, ",suiddir");
1136         if (args->ar_data != GFS2_DATA_DEFAULT) {
1137                 char *state;
1138                 switch (args->ar_data) {
1139                 case GFS2_DATA_WRITEBACK:
1140                         state = "writeback";
1141                         break;
1142                 case GFS2_DATA_ORDERED:
1143                         state = "ordered";
1144                         break;
1145                 default:
1146                         state = "unknown";
1147                         break;
1148                 }
1149                 seq_printf(s, ",data=%s", state);
1150         }
1151         if (args->ar_discard)
1152                 seq_puts(s, ",discard");
1153         val = sdp->sd_tune.gt_logd_secs;
1154         if (val != 30)
1155                 seq_printf(s, ",commit=%d", val);
1156         val = sdp->sd_tune.gt_statfs_quantum;
1157         if (val != 30)
1158                 seq_printf(s, ",statfs_quantum=%d", val);
1159         else if (sdp->sd_tune.gt_statfs_slow)
1160                 seq_puts(s, ",statfs_quantum=0");
1161         val = sdp->sd_tune.gt_quota_quantum;
1162         if (val != 60)
1163                 seq_printf(s, ",quota_quantum=%d", val);
1164         if (args->ar_statfs_percent)
1165                 seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent);
1166         if (args->ar_errors != GFS2_ERRORS_DEFAULT) {
1167                 const char *state;
1168
1169                 switch (args->ar_errors) {
1170                 case GFS2_ERRORS_WITHDRAW:
1171                         state = "withdraw";
1172                         break;
1173                 case GFS2_ERRORS_PANIC:
1174                         state = "panic";
1175                         break;
1176                 default:
1177                         state = "unknown";
1178                         break;
1179                 }
1180                 seq_printf(s, ",errors=%s", state);
1181         }
1182         if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags))
1183                 seq_puts(s, ",nobarrier");
1184         if (test_bit(SDF_DEMOTE, &sdp->sd_flags))
1185                 seq_puts(s, ",demote_interface_used");
1186         if (args->ar_rgrplvb)
1187                 seq_puts(s, ",rgrplvb");
1188         if (args->ar_loccookie)
1189                 seq_puts(s, ",loccookie");
1190         return 0;
1191 }
1192
1193 static void gfs2_final_release_pages(struct gfs2_inode *ip)
1194 {
1195         struct inode *inode = &ip->i_inode;
1196         struct gfs2_glock *gl = ip->i_gl;
1197
1198         truncate_inode_pages(gfs2_glock2aspace(ip->i_gl), 0);
1199         truncate_inode_pages(&inode->i_data, 0);
1200
1201         if (atomic_read(&gl->gl_revokes) == 0) {
1202                 clear_bit(GLF_LFLUSH, &gl->gl_flags);
1203                 clear_bit(GLF_DIRTY, &gl->gl_flags);
1204         }
1205 }
1206
1207 static int gfs2_dinode_dealloc(struct gfs2_inode *ip)
1208 {
1209         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1210         struct gfs2_rgrpd *rgd;
1211         struct gfs2_holder gh;
1212         int error;
1213
1214         if (gfs2_get_inode_blocks(&ip->i_inode) != 1) {
1215                 gfs2_consist_inode(ip);
1216                 return -EIO;
1217         }
1218
1219         error = gfs2_rindex_update(sdp);
1220         if (error)
1221                 return error;
1222
1223         error = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
1224         if (error)
1225                 return error;
1226
1227         rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1228         if (!rgd) {
1229                 gfs2_consist_inode(ip);
1230                 error = -EIO;
1231                 goto out_qs;
1232         }
1233
1234         error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &gh);
1235         if (error)
1236                 goto out_qs;
1237
1238         error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA,
1239                                  sdp->sd_jdesc->jd_blocks);
1240         if (error)
1241                 goto out_rg_gunlock;
1242
1243         gfs2_free_di(rgd, ip);
1244
1245         gfs2_final_release_pages(ip);
1246
1247         gfs2_trans_end(sdp);
1248
1249 out_rg_gunlock:
1250         gfs2_glock_dq_uninit(&gh);
1251 out_qs:
1252         gfs2_quota_unhold(ip);
1253         return error;
1254 }
1255
1256 /**
1257  * gfs2_glock_put_eventually
1258  * @gl: The glock to put
1259  *
1260  * When under memory pressure, trigger a deferred glock put to make sure we
1261  * won't call into DLM and deadlock.  Otherwise, put the glock directly.
1262  */
1263
1264 static void gfs2_glock_put_eventually(struct gfs2_glock *gl)
1265 {
1266         if (current->flags & PF_MEMALLOC)
1267                 gfs2_glock_queue_put(gl);
1268         else
1269                 gfs2_glock_put(gl);
1270 }
1271
1272 static bool gfs2_upgrade_iopen_glock(struct inode *inode)
1273 {
1274         struct gfs2_inode *ip = GFS2_I(inode);
1275         struct gfs2_sbd *sdp = GFS2_SB(inode);
1276         struct gfs2_holder *gh = &ip->i_iopen_gh;
1277         long timeout = 5 * HZ;
1278         int error;
1279
1280         gh->gh_flags |= GL_NOCACHE;
1281         gfs2_glock_dq_wait(gh);
1282
1283         /*
1284          * If there are no other lock holders, we'll get the lock immediately.
1285          * Otherwise, the other nodes holding the lock will be notified about
1286          * our locking request.  If they don't have the inode open, they'll
1287          * evict the cached inode and release the lock.  Otherwise, if they
1288          * poke the inode glock, we'll take this as an indication that they
1289          * still need the iopen glock and that they'll take care of deleting
1290          * the inode when they're done.  As a last resort, if another node
1291          * keeps holding the iopen glock without showing any activity on the
1292          * inode glock, we'll eventually time out.
1293          *
1294          * Note that we're passing the LM_FLAG_TRY_1CB flag to the first
1295          * locking request as an optimization to notify lock holders as soon as
1296          * possible.  Without that flag, they'd be notified implicitly by the
1297          * second locking request.
1298          */
1299
1300         gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE, gh);
1301         error = gfs2_glock_nq(gh);
1302         if (error != GLR_TRYFAILED)
1303                 return !error;
1304
1305         gfs2_holder_reinit(LM_ST_EXCLUSIVE, GL_ASYNC | GL_NOCACHE, gh);
1306         error = gfs2_glock_nq(gh);
1307         if (error)
1308                 return false;
1309
1310         timeout = wait_event_interruptible_timeout(sdp->sd_async_glock_wait,
1311                 !test_bit(HIF_WAIT, &gh->gh_iflags) ||
1312                 test_bit(GLF_DEMOTE, &ip->i_gl->gl_flags),
1313                 timeout);
1314         if (!test_bit(HIF_HOLDER, &gh->gh_iflags)) {
1315                 gfs2_glock_dq(gh);
1316                 return false;
1317         }
1318         return true;
1319 }
1320
1321 /**
1322  * evict_should_delete - determine whether the inode is eligible for deletion
1323  * @inode: The inode to evict
1324  *
1325  * This function determines whether the evicted inode is eligible to be deleted
1326  * and locks the inode glock.
1327  *
1328  * Returns: the fate of the dinode
1329  */
1330 static enum dinode_demise evict_should_delete(struct inode *inode,
1331                                               struct gfs2_holder *gh)
1332 {
1333         struct gfs2_inode *ip = GFS2_I(inode);
1334         struct super_block *sb = inode->i_sb;
1335         struct gfs2_sbd *sdp = sb->s_fs_info;
1336         int ret;
1337
1338         if (test_bit(GIF_ALLOC_FAILED, &ip->i_flags)) {
1339                 BUG_ON(!gfs2_glock_is_locked_by_me(ip->i_gl));
1340                 goto should_delete;
1341         }
1342
1343         if (test_bit(GIF_DEFERRED_DELETE, &ip->i_flags))
1344                 return SHOULD_DEFER_EVICTION;
1345
1346         /* Deletes should never happen under memory pressure anymore.  */
1347         if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
1348                 return SHOULD_DEFER_EVICTION;
1349
1350         /* Must not read inode block until block type has been verified */
1351         ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, gh);
1352         if (unlikely(ret)) {
1353                 glock_clear_object(ip->i_iopen_gh.gh_gl, ip);
1354                 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1355                 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1356                 return SHOULD_DEFER_EVICTION;
1357         }
1358
1359         if (gfs2_inode_already_deleted(ip->i_gl, ip->i_no_formal_ino))
1360                 return SHOULD_NOT_DELETE_DINODE;
1361         ret = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED);
1362         if (ret)
1363                 return SHOULD_NOT_DELETE_DINODE;
1364
1365         if (test_bit(GIF_INVALID, &ip->i_flags)) {
1366                 ret = gfs2_inode_refresh(ip);
1367                 if (ret)
1368                         return SHOULD_NOT_DELETE_DINODE;
1369         }
1370
1371         /*
1372          * The inode may have been recreated in the meantime.
1373          */
1374         if (inode->i_nlink)
1375                 return SHOULD_NOT_DELETE_DINODE;
1376
1377 should_delete:
1378         if (gfs2_holder_initialized(&ip->i_iopen_gh) &&
1379             test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1380                 if (!gfs2_upgrade_iopen_glock(inode)) {
1381                         gfs2_holder_uninit(&ip->i_iopen_gh);
1382                         return SHOULD_NOT_DELETE_DINODE;
1383                 }
1384         }
1385         return SHOULD_DELETE_DINODE;
1386 }
1387
1388 /**
1389  * evict_unlinked_inode - delete the pieces of an unlinked evicted inode
1390  * @inode: The inode to evict
1391  */
1392 static int evict_unlinked_inode(struct inode *inode)
1393 {
1394         struct gfs2_inode *ip = GFS2_I(inode);
1395         int ret;
1396
1397         if (S_ISDIR(inode->i_mode) &&
1398             (ip->i_diskflags & GFS2_DIF_EXHASH)) {
1399                 ret = gfs2_dir_exhash_dealloc(ip);
1400                 if (ret)
1401                         goto out;
1402         }
1403
1404         if (ip->i_eattr) {
1405                 ret = gfs2_ea_dealloc(ip);
1406                 if (ret)
1407                         goto out;
1408         }
1409
1410         if (!gfs2_is_stuffed(ip)) {
1411                 ret = gfs2_file_dealloc(ip);
1412                 if (ret)
1413                         goto out;
1414         }
1415
1416         /* We're about to clear the bitmap for the dinode, but as soon as we
1417            do, gfs2_create_inode can create another inode at the same block
1418            location and try to set gl_object again. We clear gl_object here so
1419            that subsequent inode creates don't see an old gl_object. */
1420         glock_clear_object(ip->i_gl, ip);
1421         ret = gfs2_dinode_dealloc(ip);
1422         gfs2_inode_remember_delete(ip->i_gl, ip->i_no_formal_ino);
1423 out:
1424         return ret;
1425 }
1426
1427 /*
1428  * evict_linked_inode - evict an inode whose dinode has not been unlinked
1429  * @inode: The inode to evict
1430  */
1431 static int evict_linked_inode(struct inode *inode)
1432 {
1433         struct super_block *sb = inode->i_sb;
1434         struct gfs2_sbd *sdp = sb->s_fs_info;
1435         struct gfs2_inode *ip = GFS2_I(inode);
1436         struct address_space *metamapping;
1437         int ret;
1438
1439         gfs2_log_flush(sdp, ip->i_gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
1440                        GFS2_LFC_EVICT_INODE);
1441         metamapping = gfs2_glock2aspace(ip->i_gl);
1442         if (test_bit(GLF_DIRTY, &ip->i_gl->gl_flags)) {
1443                 filemap_fdatawrite(metamapping);
1444                 filemap_fdatawait(metamapping);
1445         }
1446         write_inode_now(inode, 1);
1447         gfs2_ail_flush(ip->i_gl, 0);
1448
1449         ret = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
1450         if (ret)
1451                 return ret;
1452
1453         /* Needs to be done before glock release & also in a transaction */
1454         truncate_inode_pages(&inode->i_data, 0);
1455         truncate_inode_pages(metamapping, 0);
1456         gfs2_trans_end(sdp);
1457         return 0;
1458 }
1459
1460 /**
1461  * gfs2_evict_inode - Remove an inode from cache
1462  * @inode: The inode to evict
1463  *
1464  * There are three cases to consider:
1465  * 1. i_nlink == 0, we are final opener (and must deallocate)
1466  * 2. i_nlink == 0, we are not the final opener (and cannot deallocate)
1467  * 3. i_nlink > 0
1468  *
1469  * If the fs is read only, then we have to treat all cases as per #3
1470  * since we are unable to do any deallocation. The inode will be
1471  * deallocated by the next read/write node to attempt an allocation
1472  * in the same resource group
1473  *
1474  * We have to (at the moment) hold the inodes main lock to cover
1475  * the gap between unlocking the shared lock on the iopen lock and
1476  * taking the exclusive lock. I'd rather do a shared -> exclusive
1477  * conversion on the iopen lock, but we can change that later. This
1478  * is safe, just less efficient.
1479  */
1480
1481 static void gfs2_evict_inode(struct inode *inode)
1482 {
1483         struct super_block *sb = inode->i_sb;
1484         struct gfs2_sbd *sdp = sb->s_fs_info;
1485         struct gfs2_inode *ip = GFS2_I(inode);
1486         struct gfs2_holder gh;
1487         int ret;
1488
1489         if (test_bit(GIF_FREE_VFS_INODE, &ip->i_flags)) {
1490                 clear_inode(inode);
1491                 return;
1492         }
1493
1494         if (inode->i_nlink || sb_rdonly(sb))
1495                 goto out;
1496
1497         gfs2_holder_mark_uninitialized(&gh);
1498         ret = evict_should_delete(inode, &gh);
1499         if (ret == SHOULD_DEFER_EVICTION)
1500                 goto out;
1501         if (ret == SHOULD_DELETE_DINODE)
1502                 ret = evict_unlinked_inode(inode);
1503         else
1504                 ret = evict_linked_inode(inode);
1505
1506         if (gfs2_rs_active(&ip->i_res))
1507                 gfs2_rs_deltree(&ip->i_res);
1508
1509         if (gfs2_holder_initialized(&gh)) {
1510                 glock_clear_object(ip->i_gl, ip);
1511                 gfs2_glock_dq_uninit(&gh);
1512         }
1513         if (ret && ret != GLR_TRYFAILED && ret != -EROFS)
1514                 fs_warn(sdp, "gfs2_evict_inode: %d\n", ret);
1515 out:
1516         truncate_inode_pages_final(&inode->i_data);
1517         if (ip->i_qadata)
1518                 gfs2_assert_warn(sdp, ip->i_qadata->qa_ref == 0);
1519         gfs2_rs_delete(ip, NULL);
1520         gfs2_ordered_del_inode(ip);
1521         clear_inode(inode);
1522         gfs2_dir_hash_inval(ip);
1523         if (ip->i_gl) {
1524                 glock_clear_object(ip->i_gl, ip);
1525                 wait_on_bit_io(&ip->i_flags, GIF_GLOP_PENDING, TASK_UNINTERRUPTIBLE);
1526                 gfs2_glock_add_to_lru(ip->i_gl);
1527                 gfs2_glock_put_eventually(ip->i_gl);
1528                 ip->i_gl = NULL;
1529         }
1530         if (gfs2_holder_initialized(&ip->i_iopen_gh)) {
1531                 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1532
1533                 glock_clear_object(gl, ip);
1534                 if (test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1535                         ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1536                         gfs2_glock_dq(&ip->i_iopen_gh);
1537                 }
1538                 gfs2_glock_hold(gl);
1539                 gfs2_holder_uninit(&ip->i_iopen_gh);
1540                 gfs2_glock_put_eventually(gl);
1541         }
1542 }
1543
1544 static struct inode *gfs2_alloc_inode(struct super_block *sb)
1545 {
1546         struct gfs2_inode *ip;
1547
1548         ip = kmem_cache_alloc(gfs2_inode_cachep, GFP_KERNEL);
1549         if (!ip)
1550                 return NULL;
1551         ip->i_flags = 0;
1552         ip->i_gl = NULL;
1553         gfs2_holder_mark_uninitialized(&ip->i_iopen_gh);
1554         memset(&ip->i_res, 0, sizeof(ip->i_res));
1555         RB_CLEAR_NODE(&ip->i_res.rs_node);
1556         ip->i_rahead = 0;
1557         return &ip->i_inode;
1558 }
1559
1560 static void gfs2_free_inode(struct inode *inode)
1561 {
1562         kmem_cache_free(gfs2_inode_cachep, GFS2_I(inode));
1563 }
1564
1565 extern void free_local_statfs_inodes(struct gfs2_sbd *sdp)
1566 {
1567         struct local_statfs_inode *lsi, *safe;
1568
1569         /* Run through the statfs inodes list to iput and free memory */
1570         list_for_each_entry_safe(lsi, safe, &sdp->sd_sc_inodes_list, si_list) {
1571                 if (lsi->si_jid == sdp->sd_jdesc->jd_jid)
1572                         sdp->sd_sc_inode = NULL; /* belongs to this node */
1573                 if (lsi->si_sc_inode)
1574                         iput(lsi->si_sc_inode);
1575                 list_del(&lsi->si_list);
1576                 kfree(lsi);
1577         }
1578 }
1579
1580 extern struct inode *find_local_statfs_inode(struct gfs2_sbd *sdp,
1581                                              unsigned int index)
1582 {
1583         struct local_statfs_inode *lsi;
1584
1585         /* Return the local (per node) statfs inode in the
1586          * sdp->sd_sc_inodes_list corresponding to the 'index'. */
1587         list_for_each_entry(lsi, &sdp->sd_sc_inodes_list, si_list) {
1588                 if (lsi->si_jid == index)
1589                         return lsi->si_sc_inode;
1590         }
1591         return NULL;
1592 }
1593
1594 const struct super_operations gfs2_super_ops = {
1595         .alloc_inode            = gfs2_alloc_inode,
1596         .free_inode             = gfs2_free_inode,
1597         .write_inode            = gfs2_write_inode,
1598         .dirty_inode            = gfs2_dirty_inode,
1599         .evict_inode            = gfs2_evict_inode,
1600         .put_super              = gfs2_put_super,
1601         .sync_fs                = gfs2_sync_fs,
1602         .freeze_super           = gfs2_freeze,
1603         .thaw_super             = gfs2_unfreeze,
1604         .statfs                 = gfs2_statfs,
1605         .drop_inode             = gfs2_drop_inode,
1606         .show_options           = gfs2_show_options,
1607 };
1608