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         sdp->sd_jdesc = NULL;
  65         while (!list_empty(&list)) {
  66                 jd = list_first_entry(&list, struct gfs2_jdesc, jd_list);
  67                 gfs2_free_journal_extents(jd);
  68                 list_del(&jd->jd_list);
  69                 iput(jd->jd_inode);
  70                 jd->jd_inode = NULL;
  71                 kfree(jd);
  72         }
  73 }
  74
  75 static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
  76 {
  77         struct gfs2_jdesc *jd;
  78         int found = 0;
  79
  80         list_for_each_entry(jd, head, jd_list) {
  81                 if (jd->jd_jid == jid) {
  82                         found = 1;
  83                         break;
  84                 }
  85         }
  86
  87         if (!found)
  88                 jd = NULL;
  89
  90         return jd;
  91 }
  92
  93 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
  94 {
  95         struct gfs2_jdesc *jd;
  96
  97         spin_lock(&sdp->sd_jindex_spin);
  98         jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
  99         spin_unlock(&sdp->sd_jindex_spin);
 100
 101         return jd;
 102 }
 103
 104 int gfs2_jdesc_check(struct gfs2_jdesc *jd)
 105 {
 106         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 107         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 108         u64 size = i_size_read(jd->jd_inode);
 109
 110         if (gfs2_check_internal_file_size(jd->jd_inode, 8 << 20, BIT(30)))
 111                 return -EIO;
 112
 113         jd->jd_blocks = size >> sdp->sd_sb.sb_bsize_shift;
 114
 115         if (gfs2_write_alloc_required(ip, 0, size)) {
 116                 gfs2_consist_inode(ip);
 117                 return -EIO;
 118         }
 119
 120         return 0;
 121 }
 122
 123 static int init_threads(struct gfs2_sbd *sdp)
 124 {
 125         struct task_struct *p;
 126         int error = 0;
 127
 128         p = kthread_run(gfs2_logd, sdp, "gfs2_logd");
 129         if (IS_ERR(p)) {
 130                 error = PTR_ERR(p);
 131                 fs_err(sdp, "can't start logd thread: %d\n", error);
 132                 return error;
 133         }
 134         sdp->sd_logd_process = p;
 135
 136         p = kthread_run(gfs2_quotad, sdp, "gfs2_quotad");
 137         if (IS_ERR(p)) {
 138                 error = PTR_ERR(p);
 139                 fs_err(sdp, "can't start quotad thread: %d\n", error);
 140                 goto fail;
 141         }
 142         sdp->sd_quotad_process = p;
 143         return 0;
 144
 145 fail:
 146         kthread_stop(sdp->sd_logd_process);
 147         sdp->sd_logd_process = NULL;
 148         return error;
 149 }
 150
 151 /**
 152  * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
 153  * @sdp: the filesystem
 154  *
 155  * Returns: errno
 156  */
 157
 158 int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
 159 {
 160         struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
 161         struct gfs2_glock *j_gl = ip->i_gl;
 162         struct gfs2_holder freeze_gh;
 163         struct gfs2_log_header_host head;
 164         int error;
 165
 166         error = init_threads(sdp);
 167         if (error)
 168                 return error;
 169
 170         error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED, 0,
 171                                    &freeze_gh);
 172         if (error)
 173                 goto fail_threads;
 174
 175         j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
 176         if (gfs2_withdrawn(sdp)) {
 177                 error = -EIO;
 178                 goto fail;
 179         }
 180
 181         error = gfs2_find_jhead(sdp->sd_jdesc, &head, false);
 182         if (error || gfs2_withdrawn(sdp))
 183                 goto fail;
 184
 185         if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 186                 gfs2_consist(sdp);
 187                 error = -EIO;
 188                 goto fail;
 189         }
 190
 191         /*  Initialize some head of the log stuff  */
 192         sdp->sd_log_sequence = head.lh_sequence + 1;
 193         gfs2_log_pointers_init(sdp, head.lh_blkno);
 194
 195         error = gfs2_quota_init(sdp);
 196         if (error || gfs2_withdrawn(sdp))
 197                 goto fail;
 198
 199         set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 200
 201         gfs2_glock_dq_uninit(&freeze_gh);
 202
 203         return 0;
 204
 205 fail:
 206         freeze_gh.gh_flags |= GL_NOCACHE;
 207         gfs2_glock_dq_uninit(&freeze_gh);
 208 fail_threads:
 209         if (sdp->sd_quotad_process)
 210                 kthread_stop(sdp->sd_quotad_process);
 211         sdp->sd_quotad_process = NULL;
 212         if (sdp->sd_logd_process)
 213                 kthread_stop(sdp->sd_logd_process);
 214         sdp->sd_logd_process = NULL;
 215         return error;
 216 }
 217
 218 void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
 219 {
 220         const struct gfs2_statfs_change *str = buf;
 221
 222         sc->sc_total = be64_to_cpu(str->sc_total);
 223         sc->sc_free = be64_to_cpu(str->sc_free);
 224         sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
 225 }
 226
 227 static void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
 228 {
 229         struct gfs2_statfs_change *str = buf;
 230
 231         str->sc_total = cpu_to_be64(sc->sc_total);
 232         str->sc_free = cpu_to_be64(sc->sc_free);
 233         str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
 234 }
 235
 236 int gfs2_statfs_init(struct gfs2_sbd *sdp)
 237 {
 238         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 239         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 240         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 241         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 242         struct buffer_head *m_bh, *l_bh;
 243         struct gfs2_holder gh;
 244         int error;
 245
 246         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 247                                    &gh);
 248         if (error)
 249                 return error;
 250
 251         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 252         if (error)
 253                 goto out;
 254
 255         if (sdp->sd_args.ar_spectator) {
 256                 spin_lock(&sdp->sd_statfs_spin);
 257                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 258                                       sizeof(struct gfs2_dinode));
 259                 spin_unlock(&sdp->sd_statfs_spin);
 260         } else {
 261                 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 262                 if (error)
 263                         goto out_m_bh;
 264
 265                 spin_lock(&sdp->sd_statfs_spin);
 266                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 267                                       sizeof(struct gfs2_dinode));
 268                 gfs2_statfs_change_in(l_sc, l_bh->b_data +
 269                                       sizeof(struct gfs2_dinode));
 270                 spin_unlock(&sdp->sd_statfs_spin);
 271
 272                 brelse(l_bh);
 273         }
 274
 275 out_m_bh:
 276         brelse(m_bh);
 277 out:
 278         gfs2_glock_dq_uninit(&gh);
 279         return 0;
 280 }
 281
 282 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
 283                         s64 dinodes)
 284 {
 285         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 286         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 287         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 288         struct buffer_head *l_bh;
 289         s64 x, y;
 290         int need_sync = 0;
 291         int error;
 292
 293         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 294         if (error)
 295                 return;
 296
 297         gfs2_trans_add_meta(l_ip->i_gl, l_bh);
 298
 299         spin_lock(&sdp->sd_statfs_spin);
 300         l_sc->sc_total += total;
 301         l_sc->sc_free += free;
 302         l_sc->sc_dinodes += dinodes;
 303         gfs2_statfs_change_out(l_sc, l_bh->b_data + sizeof(struct gfs2_dinode));
 304         if (sdp->sd_args.ar_statfs_percent) {
 305                 x = 100 * l_sc->sc_free;
 306                 y = m_sc->sc_free * sdp->sd_args.ar_statfs_percent;
 307                 if (x >= y || x <= -y)
 308                         need_sync = 1;
 309         }
 310         spin_unlock(&sdp->sd_statfs_spin);
 311
 312         brelse(l_bh);
 313         if (need_sync)
 314                 gfs2_wake_up_statfs(sdp);
 315 }
 316
 317 void update_statfs(struct gfs2_sbd *sdp, struct buffer_head *m_bh,
 318                    struct buffer_head *l_bh)
 319 {
 320         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 321         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 322         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 323         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 324
 325         gfs2_trans_add_meta(l_ip->i_gl, l_bh);
 326         gfs2_trans_add_meta(m_ip->i_gl, m_bh);
 327
 328         spin_lock(&sdp->sd_statfs_spin);
 329         m_sc->sc_total += l_sc->sc_total;
 330         m_sc->sc_free += l_sc->sc_free;
 331         m_sc->sc_dinodes += l_sc->sc_dinodes;
 332         memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
 333         memset(l_bh->b_data + sizeof(struct gfs2_dinode),
 334                0, sizeof(struct gfs2_statfs_change));
 335         gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
 336         spin_unlock(&sdp->sd_statfs_spin);
 337 }
 338
 339 int gfs2_statfs_sync(struct super_block *sb, int type)
 340 {
 341         struct gfs2_sbd *sdp = sb->s_fs_info;
 342         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 343         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 344         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 345         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 346         struct gfs2_holder gh;
 347         struct buffer_head *m_bh, *l_bh;
 348         int error;
 349
 350         sb_start_write(sb);
 351         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 352                                    &gh);
 353         if (error)
 354                 goto out;
 355
 356         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 357         if (error)
 358                 goto out_unlock;
 359
 360         spin_lock(&sdp->sd_statfs_spin);
 361         gfs2_statfs_change_in(m_sc, m_bh->b_data +
 362                               sizeof(struct gfs2_dinode));
 363         if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
 364                 spin_unlock(&sdp->sd_statfs_spin);
 365                 goto out_bh;
 366         }
 367         spin_unlock(&sdp->sd_statfs_spin);
 368
 369         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 370         if (error)
 371                 goto out_bh;
 372
 373         error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
 374         if (error)
 375                 goto out_bh2;
 376
 377         update_statfs(sdp, m_bh, l_bh);
 378         sdp->sd_statfs_force_sync = 0;
 379
 380         gfs2_trans_end(sdp);
 381
 382 out_bh2:
 383         brelse(l_bh);
 384 out_bh:
 385         brelse(m_bh);
 386 out_unlock:
 387         gfs2_glock_dq_uninit(&gh);
 388 out:
 389         sb_end_write(sb);
 390         return error;
 391 }
 392
 393 struct lfcc {
 394         struct list_head list;
 395         struct gfs2_holder gh;
 396 };
 397
 398 /**
 399  * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
 400  *                            journals are clean
 401  * @sdp: the file system
 402  * @state: the state to put the transaction lock into
 403  * @t_gh: the hold on the transaction lock
 404  *
 405  * Returns: errno
 406  */
 407
 408 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp)
 409 {
 410         struct gfs2_inode *ip;
 411         struct gfs2_jdesc *jd;
 412         struct lfcc *lfcc;
 413         LIST_HEAD(list);
 414         struct gfs2_log_header_host lh;
 415         int error;
 416
 417         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 418                 lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
 419                 if (!lfcc) {
 420                         error = -ENOMEM;
 421                         goto out;
 422                 }
 423                 ip = GFS2_I(jd->jd_inode);
 424                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
 425                 if (error) {
 426                         kfree(lfcc);
 427                         goto out;
 428                 }
 429                 list_add(&lfcc->list, &list);
 430         }
 431
 432         error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_EXCLUSIVE,
 433                                    GL_NOCACHE, &sdp->sd_freeze_gh);
 434         if (error)
 435                 goto out;
 436
 437         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 438                 error = gfs2_jdesc_check(jd);
 439                 if (error)
 440                         break;
 441                 error = gfs2_find_jhead(jd, &lh, false);
 442                 if (error)
 443                         break;
 444                 if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 445                         error = -EBUSY;
 446                         break;
 447                 }
 448         }
 449
 450         if (error)
 451                 gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
 452
 453 out:
 454         while (!list_empty(&list)) {
 455                 lfcc = list_first_entry(&list, struct lfcc, list);
 456                 list_del(&lfcc->list);
 457                 gfs2_glock_dq_uninit(&lfcc->gh);
 458                 kfree(lfcc);
 459         }
 460         return error;
 461 }
 462
 463 void gfs2_dinode_out(const struct gfs2_inode *ip, void *buf)
 464 {
 465         struct gfs2_dinode *str = buf;
 466
 467         str->di_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 468         str->di_header.mh_type = cpu_to_be32(GFS2_METATYPE_DI);
 469         str->di_header.mh_format = cpu_to_be32(GFS2_FORMAT_DI);
 470         str->di_num.no_addr = cpu_to_be64(ip->i_no_addr);
 471         str->di_num.no_formal_ino = cpu_to_be64(ip->i_no_formal_ino);
 472         str->di_mode = cpu_to_be32(ip->i_inode.i_mode);
 473         str->di_uid = cpu_to_be32(i_uid_read(&ip->i_inode));
 474         str->di_gid = cpu_to_be32(i_gid_read(&ip->i_inode));
 475         str->di_nlink = cpu_to_be32(ip->i_inode.i_nlink);
 476         str->di_size = cpu_to_be64(i_size_read(&ip->i_inode));
 477         str->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(&ip->i_inode));
 478         str->di_atime = cpu_to_be64(ip->i_inode.i_atime.tv_sec);
 479         str->di_mtime = cpu_to_be64(ip->i_inode.i_mtime.tv_sec);
 480         str->di_ctime = cpu_to_be64(ip->i_inode.i_ctime.tv_sec);
 481
 482         str->di_goal_meta = cpu_to_be64(ip->i_goal);
 483         str->di_goal_data = cpu_to_be64(ip->i_goal);
 484         str->di_generation = cpu_to_be64(ip->i_generation);
 485
 486         str->di_flags = cpu_to_be32(ip->i_diskflags);
 487         str->di_height = cpu_to_be16(ip->i_height);
 488         str->di_payload_format = cpu_to_be32(S_ISDIR(ip->i_inode.i_mode) &&
 489                                              !(ip->i_diskflags & GFS2_DIF_EXHASH) ?
 490                                              GFS2_FORMAT_DE : 0);
 491         str->di_depth = cpu_to_be16(ip->i_depth);
 492         str->di_entries = cpu_to_be32(ip->i_entries);
 493
 494         str->di_eattr = cpu_to_be64(ip->i_eattr);
 495         str->di_atime_nsec = cpu_to_be32(ip->i_inode.i_atime.tv_nsec);
 496         str->di_mtime_nsec = cpu_to_be32(ip->i_inode.i_mtime.tv_nsec);
 497         str->di_ctime_nsec = cpu_to_be32(ip->i_inode.i_ctime.tv_nsec);
 498 }
 499
 500 /**
 501  * gfs2_write_inode - Make sure the inode is stable on the disk
 502  * @inode: The inode
 503  * @wbc: The writeback control structure
 504  *
 505  * Returns: errno
 506  */
 507
 508 static int gfs2_write_inode(struct inode *inode, struct writeback_control *wbc)
 509 {
 510         struct gfs2_inode *ip = GFS2_I(inode);
 511         struct gfs2_sbd *sdp = GFS2_SB(inode);
 512         struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl);
 513         struct backing_dev_info *bdi = inode_to_bdi(metamapping->host);
 514         int ret = 0;
 515         bool flush_all = (wbc->sync_mode == WB_SYNC_ALL || gfs2_is_jdata(ip));
 516
 517         if (flush_all)
 518                 gfs2_log_flush(GFS2_SB(inode), ip->i_gl,
 519                                GFS2_LOG_HEAD_FLUSH_NORMAL |
 520                                GFS2_LFC_WRITE_INODE);
 521         if (bdi->wb.dirty_exceeded)
 522                 gfs2_ail1_flush(sdp, wbc);
 523         else
 524                 filemap_fdatawrite(metamapping);
 525         if (flush_all)
 526                 ret = filemap_fdatawait(metamapping);
 527         if (ret)
 528                 mark_inode_dirty_sync(inode);
 529         else {
 530                 spin_lock(&inode->i_lock);
 531                 if (!(inode->i_flags & I_DIRTY))
 532                         gfs2_ordered_del_inode(ip);
 533                 spin_unlock(&inode->i_lock);
 534         }
 535         return ret;
 536 }
 537
 538 /**
 539  * gfs2_dirty_inode - check for atime updates
 540  * @inode: The inode in question
 541  * @flags: The type of dirty
 542  *
 543  * Unfortunately it can be called under any combination of inode
 544  * glock and transaction lock, so we have to check carefully.
 545  *
 546  * At the moment this deals only with atime - it should be possible
 547  * to expand that role in future, once a review of the locking has
 548  * been carried out.
 549  */
 550
 551 static void gfs2_dirty_inode(struct inode *inode, int flags)
 552 {
 553         struct gfs2_inode *ip = GFS2_I(inode);
 554         struct gfs2_sbd *sdp = GFS2_SB(inode);
 555         struct buffer_head *bh;
 556         struct gfs2_holder gh;
 557         int need_unlock = 0;
 558         int need_endtrans = 0;
 559         int ret;
 560
 561         if (!(flags & I_DIRTY_INODE))
 562                 return;
 563         if (unlikely(gfs2_withdrawn(sdp)))
 564                 return;
 565         if (!gfs2_glock_is_locked_by_me(ip->i_gl)) {
 566                 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
 567                 if (ret) {
 568                         fs_err(sdp, "dirty_inode: glock %d\n", ret);
 569                         return;
 570                 }
 571                 need_unlock = 1;
 572         } else if (WARN_ON_ONCE(ip->i_gl->gl_state != LM_ST_EXCLUSIVE))
 573                 return;
 574
 575         if (current->journal_info == NULL) {
 576                 ret = gfs2_trans_begin(sdp, RES_DINODE, 0);
 577                 if (ret) {
 578                         fs_err(sdp, "dirty_inode: gfs2_trans_begin %d\n", ret);
 579                         goto out;
 580                 }
 581                 need_endtrans = 1;
 582         }
 583
 584         ret = gfs2_meta_inode_buffer(ip, &bh);
 585         if (ret == 0) {
 586                 gfs2_trans_add_meta(ip->i_gl, bh);
 587                 gfs2_dinode_out(ip, bh->b_data);
 588                 brelse(bh);
 589         }
 590
 591         if (need_endtrans)
 592                 gfs2_trans_end(sdp);
 593 out:
 594         if (need_unlock)
 595                 gfs2_glock_dq_uninit(&gh);
 596 }
 597
 598 /**
 599  * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
 600  * @sdp: the filesystem
 601  *
 602  * Returns: errno
 603  */
 604
 605 int gfs2_make_fs_ro(struct gfs2_sbd *sdp)
 606 {
 607         struct gfs2_holder freeze_gh;
 608         int error = 0;
 609         int log_write_allowed = test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 610
 611         gfs2_holder_mark_uninitialized(&freeze_gh);
 612         if (sdp->sd_freeze_gl &&
 613             !gfs2_glock_is_locked_by_me(sdp->sd_freeze_gl)) {
 614                 if (!log_write_allowed) {
 615                         error = gfs2_glock_nq_init(sdp->sd_freeze_gl,
 616                                                    LM_ST_SHARED, GL_NOCACHE |
 617                                                    LM_FLAG_TRY, &freeze_gh);
 618                         if (error == GLR_TRYFAILED)
 619                                 error = 0;
 620                 } else {
 621                         error = gfs2_glock_nq_init(sdp->sd_freeze_gl,
 622                                                    LM_ST_SHARED, GL_NOCACHE,
 623                                                    &freeze_gh);
 624                         if (error && !gfs2_withdrawn(sdp))
 625                                 return error;
 626                 }
 627         }
 628
 629         flush_workqueue(gfs2_delete_workqueue);
 630         if (!log_write_allowed && current == sdp->sd_quotad_process)
 631                 fs_warn(sdp, "The quotad daemon is withdrawing.\n");
 632         else if (sdp->sd_quotad_process)
 633                 kthread_stop(sdp->sd_quotad_process);
 634         sdp->sd_quotad_process = NULL;
 635
 636         if (!log_write_allowed && current == sdp->sd_logd_process)
 637                 fs_warn(sdp, "The logd daemon is withdrawing.\n");
 638         else if (sdp->sd_logd_process)
 639                 kthread_stop(sdp->sd_logd_process);
 640         sdp->sd_logd_process = NULL;
 641
 642         if (log_write_allowed) {
 643                 gfs2_quota_sync(sdp->sd_vfs, 0);
 644                 gfs2_statfs_sync(sdp->sd_vfs, 0);
 645
 646                 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SHUTDOWN |
 647                                GFS2_LFC_MAKE_FS_RO);
 648                 wait_event(sdp->sd_reserving_log_wait,
 649                            atomic_read(&sdp->sd_reserving_log) == 0);
 650                 gfs2_assert_warn(sdp, atomic_read(&sdp->sd_log_blks_free) ==
 651                                  sdp->sd_jdesc->jd_blocks);
 652         } else {
 653                 wait_event_timeout(sdp->sd_reserving_log_wait,
 654                                    atomic_read(&sdp->sd_reserving_log) == 0,
 655                                    HZ * 5);
 656         }
 657         if (gfs2_holder_initialized(&freeze_gh))
 658                 gfs2_glock_dq_uninit(&freeze_gh);
 659
 660         gfs2_quota_cleanup(sdp);
 661
 662         if (!log_write_allowed)
 663                 sdp->sd_vfs->s_flags |= SB_RDONLY;
 664
 665         return error;
 666 }
 667
 668 /**
 669  * gfs2_put_super - Unmount the filesystem
 670  * @sb: The VFS superblock
 671  *
 672  */
 673
 674 static void gfs2_put_super(struct super_block *sb)
 675 {
 676         struct gfs2_sbd *sdp = sb->s_fs_info;
 677         int error;
 678         struct gfs2_jdesc *jd;
 679
 680         /* No more recovery requests */
 681         set_bit(SDF_NORECOVERY, &sdp->sd_flags);
 682         smp_mb();
 683
 684         /* Wait on outstanding recovery */
 685 restart:
 686         spin_lock(&sdp->sd_jindex_spin);
 687         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 688                 if (!test_bit(JDF_RECOVERY, &jd->jd_flags))
 689                         continue;
 690                 spin_unlock(&sdp->sd_jindex_spin);
 691                 wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
 692                             TASK_UNINTERRUPTIBLE);
 693                 goto restart;
 694         }
 695         spin_unlock(&sdp->sd_jindex_spin);
 696
 697         if (!sb_rdonly(sb)) {
 698                 error = gfs2_make_fs_ro(sdp);
 699                 if (error)
 700                         gfs2_io_error(sdp);
 701         }
 702         /*  At this point, we're through modifying the disk  */
 703
 704         /*  Release stuff  */
 705
 706         iput(sdp->sd_jindex);
 707         iput(sdp->sd_statfs_inode);
 708         iput(sdp->sd_rindex);
 709         iput(sdp->sd_quota_inode);
 710
 711         gfs2_glock_put(sdp->sd_rename_gl);
 712         gfs2_glock_put(sdp->sd_freeze_gl);
 713
 714         if (!sdp->sd_args.ar_spectator) {
 715                 if (gfs2_holder_initialized(&sdp->sd_journal_gh))
 716                         gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
 717                 if (gfs2_holder_initialized(&sdp->sd_jinode_gh))
 718                         gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
 719                 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
 720                 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
 721                 iput(sdp->sd_sc_inode);
 722                 iput(sdp->sd_qc_inode);
 723         }
 724
 725         gfs2_glock_dq_uninit(&sdp->sd_live_gh);
 726         gfs2_clear_rgrpd(sdp);
 727         gfs2_jindex_free(sdp);
 728         /*  Take apart glock structures and buffer lists  */
 729         gfs2_gl_hash_clear(sdp);
 730         gfs2_delete_debugfs_file(sdp);
 731         /*  Unmount the locking protocol  */
 732         gfs2_lm_unmount(sdp);
 733
 734         /*  At this point, we're through participating in the lockspace  */
 735         gfs2_sys_fs_del(sdp);
 736 }
 737
 738 /**
 739  * gfs2_sync_fs - sync the filesystem
 740  * @sb: the superblock
 741  *
 742  * Flushes the log to disk.
 743  */
 744
 745 static int gfs2_sync_fs(struct super_block *sb, int wait)
 746 {
 747         struct gfs2_sbd *sdp = sb->s_fs_info;
 748
 749         gfs2_quota_sync(sb, -1);
 750         if (wait)
 751                 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
 752                                GFS2_LFC_SYNC_FS);
 753         return sdp->sd_log_error;
 754 }
 755
 756 void gfs2_freeze_func(struct work_struct *work)
 757 {
 758         int error;
 759         struct gfs2_holder freeze_gh;
 760         struct gfs2_sbd *sdp = container_of(work, struct gfs2_sbd, sd_freeze_work);
 761         struct super_block *sb = sdp->sd_vfs;
 762
 763         atomic_inc(&sb->s_active);
 764         error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED, 0,
 765                                    &freeze_gh);
 766         if (error) {
 767                 fs_info(sdp, "GFS2: couldn't get freeze lock : %d\n", error);
 768                 gfs2_assert_withdraw(sdp, 0);
 769         } else {
 770                 atomic_set(&sdp->sd_freeze_state, SFS_UNFROZEN);
 771                 error = thaw_super(sb);
 772                 if (error) {
 773                         fs_info(sdp, "GFS2: couldn't thaw filesystem: %d\n",
 774                                 error);
 775                         gfs2_assert_withdraw(sdp, 0);
 776                 }
 777                 if (!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))
 778                         freeze_gh.gh_flags |= GL_NOCACHE;
 779                 gfs2_glock_dq_uninit(&freeze_gh);
 780         }
 781         deactivate_super(sb);
 782         clear_bit_unlock(SDF_FS_FROZEN, &sdp->sd_flags);
 783         wake_up_bit(&sdp->sd_flags, SDF_FS_FROZEN);
 784         return;
 785 }
 786
 787 /**
 788  * gfs2_freeze - prevent further writes to the filesystem
 789  * @sb: the VFS structure for the filesystem
 790  *
 791  */
 792
 793 static int gfs2_freeze(struct super_block *sb)
 794 {
 795         struct gfs2_sbd *sdp = sb->s_fs_info;
 796         int error = 0;
 797
 798         mutex_lock(&sdp->sd_freeze_mutex);
 799         if (atomic_read(&sdp->sd_freeze_state) != SFS_UNFROZEN)
 800                 goto out;
 801
 802         for (;;) {
 803                 if (gfs2_withdrawn(sdp)) {
 804                         error = -EINVAL;
 805                         goto out;
 806                 }
 807
 808                 error = gfs2_lock_fs_check_clean(sdp);
 809                 if (!error)
 810                         break;
 811
 812                 if (error == -EBUSY)
 813                         fs_err(sdp, "waiting for recovery before freeze\n");
 814                 else if (error == -EIO) {
 815                         fs_err(sdp, "Fatal IO error: cannot freeze gfs2 due "
 816                                "to recovery error.\n");
 817                         goto out;
 818                 } else {
 819                         fs_err(sdp, "error freezing FS: %d\n", error);
 820                 }
 821                 fs_err(sdp, "retrying...\n");
 822                 msleep(1000);
 823         }
 824         set_bit(SDF_FS_FROZEN, &sdp->sd_flags);
 825 out:
 826         mutex_unlock(&sdp->sd_freeze_mutex);
 827         return error;
 828 }
 829
 830 /**
 831  * gfs2_unfreeze - reallow writes to the filesystem
 832  * @sb: the VFS structure for the filesystem
 833  *
 834  */
 835
 836 static int gfs2_unfreeze(struct super_block *sb)
 837 {
 838         struct gfs2_sbd *sdp = sb->s_fs_info;
 839
 840         mutex_lock(&sdp->sd_freeze_mutex);
 841         if (atomic_read(&sdp->sd_freeze_state) != SFS_FROZEN ||
 842             !gfs2_holder_initialized(&sdp->sd_freeze_gh)) {
 843                 mutex_unlock(&sdp->sd_freeze_mutex);
 844                 return 0;
 845         }
 846
 847         gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
 848         mutex_unlock(&sdp->sd_freeze_mutex);
 849         return wait_on_bit(&sdp->sd_flags, SDF_FS_FROZEN, TASK_INTERRUPTIBLE);
 850 }
 851
 852 /**
 853  * statfs_fill - fill in the sg for a given RG
 854  * @rgd: the RG
 855  * @sc: the sc structure
 856  *
 857  * Returns: 0 on success, -ESTALE if the LVB is invalid
 858  */
 859
 860 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
 861                             struct gfs2_statfs_change_host *sc)
 862 {
 863         gfs2_rgrp_verify(rgd);
 864         sc->sc_total += rgd->rd_data;
 865         sc->sc_free += rgd->rd_free;
 866         sc->sc_dinodes += rgd->rd_dinodes;
 867         return 0;
 868 }
 869
 870 /**
 871  * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
 872  * @sdp: the filesystem
 873  * @sc: the sc info that will be returned
 874  *
 875  * Any error (other than a signal) will cause this routine to fall back
 876  * to the synchronous version.
 877  *
 878  * FIXME: This really shouldn't busy wait like this.
 879  *
 880  * Returns: errno
 881  */
 882
 883 static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
 884 {
 885         struct gfs2_rgrpd *rgd_next;
 886         struct gfs2_holder *gha, *gh;
 887         unsigned int slots = 64;
 888         unsigned int x;
 889         int done;
 890         int error = 0, err;
 891
 892         memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
 893         gha = kmalloc_array(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
 894         if (!gha)
 895                 return -ENOMEM;
 896         for (x = 0; x < slots; x++)
 897                 gfs2_holder_mark_uninitialized(gha + x);
 898
 899         rgd_next = gfs2_rgrpd_get_first(sdp);
 900
 901         for (;;) {
 902                 done = 1;
 903
 904                 for (x = 0; x < slots; x++) {
 905                         gh = gha + x;
 906
 907                         if (gfs2_holder_initialized(gh) && gfs2_glock_poll(gh)) {
 908                                 err = gfs2_glock_wait(gh);
 909                                 if (err) {
 910                                         gfs2_holder_uninit(gh);
 911                                         error = err;
 912                                 } else {
 913                                         if (!error) {
 914                                                 struct gfs2_rgrpd *rgd =
 915                                                         gfs2_glock2rgrp(gh->gh_gl);
 916
 917                                                 error = statfs_slow_fill(rgd, sc);
 918                                         }
 919                                         gfs2_glock_dq_uninit(gh);
 920                                 }
 921                         }
 922
 923                         if (gfs2_holder_initialized(gh))
 924                                 done = 0;
 925                         else if (rgd_next && !error) {
 926                                 error = gfs2_glock_nq_init(rgd_next->rd_gl,
 927                                                            LM_ST_SHARED,
 928                                                            GL_ASYNC,
 929                                                            gh);
 930                                 rgd_next = gfs2_rgrpd_get_next(rgd_next);
 931                                 done = 0;
 932                         }
 933
 934                         if (signal_pending(current))
 935                                 error = -ERESTARTSYS;
 936                 }
 937
 938                 if (done)
 939                         break;
 940
 941                 yield();
 942         }
 943
 944         kfree(gha);
 945         return error;
 946 }
 947
 948 /**
 949  * gfs2_statfs_i - Do a statfs
 950  * @sdp: the filesystem
 951  * @sg: the sg structure
 952  *
 953  * Returns: errno
 954  */
 955
 956 static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
 957 {
 958         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 959         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 960
 961         spin_lock(&sdp->sd_statfs_spin);
 962
 963         *sc = *m_sc;
 964         sc->sc_total += l_sc->sc_total;
 965         sc->sc_free += l_sc->sc_free;
 966         sc->sc_dinodes += l_sc->sc_dinodes;
 967
 968         spin_unlock(&sdp->sd_statfs_spin);
 969
 970         if (sc->sc_free < 0)
 971                 sc->sc_free = 0;
 972         if (sc->sc_free > sc->sc_total)
 973                 sc->sc_free = sc->sc_total;
 974         if (sc->sc_dinodes < 0)
 975                 sc->sc_dinodes = 0;
 976
 977         return 0;
 978 }
 979
 980 /**
 981  * gfs2_statfs - Gather and return stats about the filesystem
 982  * @sb: The superblock
 983  * @statfsbuf: The buffer
 984  *
 985  * Returns: 0 on success or error code
 986  */
 987
 988 static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
 989 {
 990         struct super_block *sb = dentry->d_sb;
 991         struct gfs2_sbd *sdp = sb->s_fs_info;
 992         struct gfs2_statfs_change_host sc;
 993         int error;
 994
 995         error = gfs2_rindex_update(sdp);
 996         if (error)
 997                 return error;
 998
 999         if (gfs2_tune_get(sdp, gt_statfs_slow))
1000                 error = gfs2_statfs_slow(sdp, &sc);
1001         else
1002                 error = gfs2_statfs_i(sdp, &sc);
1003
1004         if (error)
1005                 return error;
1006
1007         buf->f_type = GFS2_MAGIC;
1008         buf->f_bsize = sdp->sd_sb.sb_bsize;
1009         buf->f_blocks = sc.sc_total;
1010         buf->f_bfree = sc.sc_free;
1011         buf->f_bavail = sc.sc_free;
1012         buf->f_files = sc.sc_dinodes + sc.sc_free;
1013         buf->f_ffree = sc.sc_free;
1014         buf->f_namelen = GFS2_FNAMESIZE;
1015
1016         return 0;
1017 }
1018
1019 /**
1020  * gfs2_drop_inode - Drop an inode (test for remote unlink)
1021  * @inode: The inode to drop
1022  *
1023  * If we've received a callback on an iopen lock then it's because a
1024  * remote node tried to deallocate the inode but failed due to this node
1025  * still having the inode open. Here we mark the link count zero
1026  * since we know that it must have reached zero if the GLF_DEMOTE flag
1027  * is set on the iopen glock. If we didn't do a disk read since the
1028  * remote node removed the final link then we might otherwise miss
1029  * this event. This check ensures that this node will deallocate the
1030  * inode's blocks, or alternatively pass the baton on to another
1031  * node for later deallocation.
1032  */
1033
1034 static int gfs2_drop_inode(struct inode *inode)
1035 {
1036         struct gfs2_inode *ip = GFS2_I(inode);
1037
1038         if (!test_bit(GIF_FREE_VFS_INODE, &ip->i_flags) &&
1039             inode->i_nlink &&
1040             gfs2_holder_initialized(&ip->i_iopen_gh)) {
1041                 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1042                 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
1043                         clear_nlink(inode);
1044         }
1045
1046         /*
1047          * When under memory pressure when an inode's link count has dropped to
1048          * zero, defer deleting the inode to the delete workqueue.  This avoids
1049          * calling into DLM under memory pressure, which can deadlock.
1050          */
1051         if (!inode->i_nlink &&
1052             unlikely(current->flags & PF_MEMALLOC) &&
1053             gfs2_holder_initialized(&ip->i_iopen_gh)) {
1054                 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1055
1056                 gfs2_glock_hold(gl);
1057                 if (queue_work(gfs2_delete_workqueue, &gl->gl_delete) == 0)
1058                         gfs2_glock_queue_put(gl);
1059                 return false;
1060         }
1061
1062         return generic_drop_inode(inode);
1063 }
1064
1065 static int is_ancestor(const struct dentry *d1, const struct dentry *d2)
1066 {
1067         do {
1068                 if (d1 == d2)
1069                         return 1;
1070                 d1 = d1->d_parent;
1071         } while (!IS_ROOT(d1));
1072         return 0;
1073 }
1074
1075 /**
1076  * gfs2_show_options - Show mount options for /proc/mounts
1077  * @s: seq_file structure
1078  * @root: root of this (sub)tree
1079  *
1080  * Returns: 0 on success or error code
1081  */
1082
1083 static int gfs2_show_options(struct seq_file *s, struct dentry *root)
1084 {
1085         struct gfs2_sbd *sdp = root->d_sb->s_fs_info;
1086         struct gfs2_args *args = &sdp->sd_args;
1087         int val;
1088
1089         if (is_ancestor(root, sdp->sd_master_dir))
1090                 seq_puts(s, ",meta");
1091         if (args->ar_lockproto[0])
1092                 seq_show_option(s, "lockproto", args->ar_lockproto);
1093         if (args->ar_locktable[0])
1094                 seq_show_option(s, "locktable", args->ar_locktable);
1095         if (args->ar_hostdata[0])
1096                 seq_show_option(s, "hostdata", args->ar_hostdata);
1097         if (args->ar_spectator)
1098                 seq_puts(s, ",spectator");
1099         if (args->ar_localflocks)
1100                 seq_puts(s, ",localflocks");
1101         if (args->ar_debug)
1102                 seq_puts(s, ",debug");
1103         if (args->ar_posix_acl)
1104                 seq_puts(s, ",acl");
1105         if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
1106                 char *state;
1107                 switch (args->ar_quota) {
1108                 case GFS2_QUOTA_OFF:
1109                         state = "off";
1110                         break;
1111                 case GFS2_QUOTA_ACCOUNT:
1112                         state = "account";
1113                         break;
1114                 case GFS2_QUOTA_ON:
1115                         state = "on";
1116                         break;
1117                 default:
1118                         state = "unknown";
1119                         break;
1120                 }
1121                 seq_printf(s, ",quota=%s", state);
1122         }
1123         if (args->ar_suiddir)
1124                 seq_puts(s, ",suiddir");
1125         if (args->ar_data != GFS2_DATA_DEFAULT) {
1126                 char *state;
1127                 switch (args->ar_data) {
1128                 case GFS2_DATA_WRITEBACK:
1129                         state = "writeback";
1130                         break;
1131                 case GFS2_DATA_ORDERED:
1132                         state = "ordered";
1133                         break;
1134                 default:
1135                         state = "unknown";
1136                         break;
1137                 }
1138                 seq_printf(s, ",data=%s", state);
1139         }
1140         if (args->ar_discard)
1141                 seq_puts(s, ",discard");
1142         val = sdp->sd_tune.gt_logd_secs;
1143         if (val != 30)
1144                 seq_printf(s, ",commit=%d", val);
1145         val = sdp->sd_tune.gt_statfs_quantum;
1146         if (val != 30)
1147                 seq_printf(s, ",statfs_quantum=%d", val);
1148         else if (sdp->sd_tune.gt_statfs_slow)
1149                 seq_puts(s, ",statfs_quantum=0");
1150         val = sdp->sd_tune.gt_quota_quantum;
1151         if (val != 60)
1152                 seq_printf(s, ",quota_quantum=%d", val);
1153         if (args->ar_statfs_percent)
1154                 seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent);
1155         if (args->ar_errors != GFS2_ERRORS_DEFAULT) {
1156                 const char *state;
1157
1158                 switch (args->ar_errors) {
1159                 case GFS2_ERRORS_WITHDRAW:
1160                         state = "withdraw";
1161                         break;
1162                 case GFS2_ERRORS_PANIC:
1163                         state = "panic";
1164                         break;
1165                 default:
1166                         state = "unknown";
1167                         break;
1168                 }
1169                 seq_printf(s, ",errors=%s", state);
1170         }
1171         if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags))
1172                 seq_puts(s, ",nobarrier");
1173         if (test_bit(SDF_DEMOTE, &sdp->sd_flags))
1174                 seq_puts(s, ",demote_interface_used");
1175         if (args->ar_rgrplvb)
1176                 seq_puts(s, ",rgrplvb");
1177         if (args->ar_loccookie)
1178                 seq_puts(s, ",loccookie");
1179         return 0;
1180 }
1181
1182 static void gfs2_final_release_pages(struct gfs2_inode *ip)
1183 {
1184         struct inode *inode = &ip->i_inode;
1185         struct gfs2_glock *gl = ip->i_gl;
1186
1187         truncate_inode_pages(gfs2_glock2aspace(ip->i_gl), 0);
1188         truncate_inode_pages(&inode->i_data, 0);
1189
1190         if (atomic_read(&gl->gl_revokes) == 0) {
1191                 clear_bit(GLF_LFLUSH, &gl->gl_flags);
1192                 clear_bit(GLF_DIRTY, &gl->gl_flags);
1193         }
1194 }
1195
1196 static int gfs2_dinode_dealloc(struct gfs2_inode *ip)
1197 {
1198         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1199         struct gfs2_rgrpd *rgd;
1200         struct gfs2_holder gh;
1201         int error;
1202
1203         if (gfs2_get_inode_blocks(&ip->i_inode) != 1) {
1204                 gfs2_consist_inode(ip);
1205                 return -EIO;
1206         }
1207
1208         error = gfs2_rindex_update(sdp);
1209         if (error)
1210                 return error;
1211
1212         error = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
1213         if (error)
1214                 return error;
1215
1216         rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1217         if (!rgd) {
1218                 gfs2_consist_inode(ip);
1219                 error = -EIO;
1220                 goto out_qs;
1221         }
1222
1223         error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &gh);
1224         if (error)
1225                 goto out_qs;
1226
1227         error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA,
1228                                  sdp->sd_jdesc->jd_blocks);
1229         if (error)
1230                 goto out_rg_gunlock;
1231
1232         gfs2_free_di(rgd, ip);
1233
1234         gfs2_final_release_pages(ip);
1235
1236         gfs2_trans_end(sdp);
1237
1238 out_rg_gunlock:
1239         gfs2_glock_dq_uninit(&gh);
1240 out_qs:
1241         gfs2_quota_unhold(ip);
1242         return error;
1243 }
1244
1245 /**
1246  * gfs2_glock_put_eventually
1247  * @gl: The glock to put
1248  *
1249  * When under memory pressure, trigger a deferred glock put to make sure we
1250  * won't call into DLM and deadlock.  Otherwise, put the glock directly.
1251  */
1252
1253 static void gfs2_glock_put_eventually(struct gfs2_glock *gl)
1254 {
1255         if (current->flags & PF_MEMALLOC)
1256                 gfs2_glock_queue_put(gl);
1257         else
1258                 gfs2_glock_put(gl);
1259 }
1260
1261 /**
1262  * gfs2_evict_inode - Remove an inode from cache
1263  * @inode: The inode to evict
1264  *
1265  * There are three cases to consider:
1266  * 1. i_nlink == 0, we are final opener (and must deallocate)
1267  * 2. i_nlink == 0, we are not the final opener (and cannot deallocate)
1268  * 3. i_nlink > 0
1269  *
1270  * If the fs is read only, then we have to treat all cases as per #3
1271  * since we are unable to do any deallocation. The inode will be
1272  * deallocated by the next read/write node to attempt an allocation
1273  * in the same resource group
1274  *
1275  * We have to (at the moment) hold the inodes main lock to cover
1276  * the gap between unlocking the shared lock on the iopen lock and
1277  * taking the exclusive lock. I'd rather do a shared -> exclusive
1278  * conversion on the iopen lock, but we can change that later. This
1279  * is safe, just less efficient.
1280  */
1281
1282 static void gfs2_evict_inode(struct inode *inode)
1283 {
1284         struct super_block *sb = inode->i_sb;
1285         struct gfs2_sbd *sdp = sb->s_fs_info;
1286         struct gfs2_inode *ip = GFS2_I(inode);
1287         struct gfs2_holder gh;
1288         struct address_space *metamapping;
1289         int error;
1290
1291         if (test_bit(GIF_FREE_VFS_INODE, &ip->i_flags)) {
1292                 clear_inode(inode);
1293                 return;
1294         }
1295
1296         if (inode->i_nlink || sb_rdonly(sb))
1297                 goto out;
1298
1299         if (test_bit(GIF_ALLOC_FAILED, &ip->i_flags)) {
1300                 BUG_ON(!gfs2_glock_is_locked_by_me(ip->i_gl));
1301                 gfs2_holder_mark_uninitialized(&gh);
1302                 goto alloc_failed;
1303         }
1304
1305         /* Deletes should never happen under memory pressure anymore.  */
1306         if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
1307                 goto out;
1308
1309         /* Must not read inode block until block type has been verified */
1310         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, &gh);
1311         if (unlikely(error)) {
1312                 glock_clear_object(ip->i_iopen_gh.gh_gl, ip);
1313                 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1314                 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1315                 goto out;
1316         }
1317
1318         error = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED);
1319         if (error)
1320                 goto out_truncate;
1321
1322         if (test_bit(GIF_INVALID, &ip->i_flags)) {
1323                 error = gfs2_inode_refresh(ip);
1324                 if (error)
1325                         goto out_truncate;
1326         }
1327
1328         /*
1329          * The inode may have been recreated in the meantime.
1330          */
1331         if (inode->i_nlink)
1332                 goto out_truncate;
1333
1334 alloc_failed:
1335         if (gfs2_holder_initialized(&ip->i_iopen_gh) &&
1336             test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1337                 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1338                 gfs2_glock_dq_wait(&ip->i_iopen_gh);
1339                 gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE,
1340                                    &ip->i_iopen_gh);
1341                 error = gfs2_glock_nq(&ip->i_iopen_gh);
1342                 if (error)
1343                         goto out_truncate;
1344         }
1345
1346         if (S_ISDIR(inode->i_mode) &&
1347             (ip->i_diskflags & GFS2_DIF_EXHASH)) {
1348                 error = gfs2_dir_exhash_dealloc(ip);
1349                 if (error)
1350                         goto out_unlock;
1351         }
1352
1353         if (ip->i_eattr) {
1354                 error = gfs2_ea_dealloc(ip);
1355                 if (error)
1356                         goto out_unlock;
1357         }
1358
1359         if (!gfs2_is_stuffed(ip)) {
1360                 error = gfs2_file_dealloc(ip);
1361                 if (error)
1362                         goto out_unlock;
1363         }
1364
1365         /* We're about to clear the bitmap for the dinode, but as soon as we
1366            do, gfs2_create_inode can create another inode at the same block
1367            location and try to set gl_object again. We clear gl_object here so
1368            that subsequent inode creates don't see an old gl_object. */
1369         glock_clear_object(ip->i_gl, ip);
1370         error = gfs2_dinode_dealloc(ip);
1371         goto out_unlock;
1372
1373 out_truncate:
1374         gfs2_log_flush(sdp, ip->i_gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
1375                        GFS2_LFC_EVICT_INODE);
1376         metamapping = gfs2_glock2aspace(ip->i_gl);
1377         if (test_bit(GLF_DIRTY, &ip->i_gl->gl_flags)) {
1378                 filemap_fdatawrite(metamapping);
1379                 filemap_fdatawait(metamapping);
1380         }
1381         write_inode_now(inode, 1);
1382         gfs2_ail_flush(ip->i_gl, 0);
1383
1384         error = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
1385         if (error)
1386                 goto out_unlock;
1387         /* Needs to be done before glock release & also in a transaction */
1388         truncate_inode_pages(&inode->i_data, 0);
1389         truncate_inode_pages(metamapping, 0);
1390         gfs2_trans_end(sdp);
1391
1392 out_unlock:
1393         if (gfs2_rs_active(&ip->i_res))
1394                 gfs2_rs_deltree(&ip->i_res);
1395
1396         if (gfs2_holder_initialized(&gh)) {
1397                 glock_clear_object(ip->i_gl, ip);
1398                 gfs2_glock_dq_uninit(&gh);
1399         }
1400         if (error && error != GLR_TRYFAILED && error != -EROFS)
1401                 fs_warn(sdp, "gfs2_evict_inode: %d\n", error);
1402 out:
1403         truncate_inode_pages_final(&inode->i_data);
1404         if (ip->i_qadata)
1405                 gfs2_assert_warn(sdp, ip->i_qadata->qa_ref == 0);
1406         gfs2_rs_delete(ip, NULL);
1407         gfs2_ordered_del_inode(ip);
1408         clear_inode(inode);
1409         gfs2_dir_hash_inval(ip);
1410         if (ip->i_gl) {
1411                 glock_clear_object(ip->i_gl, ip);
1412                 wait_on_bit_io(&ip->i_flags, GIF_GLOP_PENDING, TASK_UNINTERRUPTIBLE);
1413                 gfs2_glock_add_to_lru(ip->i_gl);
1414                 gfs2_glock_put_eventually(ip->i_gl);
1415                 ip->i_gl = NULL;
1416         }
1417         if (gfs2_holder_initialized(&ip->i_iopen_gh)) {
1418                 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1419
1420                 glock_clear_object(gl, ip);
1421                 if (test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1422                         ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1423                         gfs2_glock_dq(&ip->i_iopen_gh);
1424                 }
1425                 gfs2_glock_hold(gl);
1426                 gfs2_holder_uninit(&ip->i_iopen_gh);
1427                 gfs2_glock_put_eventually(gl);
1428         }
1429 }
1430
1431 static struct inode *gfs2_alloc_inode(struct super_block *sb)
1432 {
1433         struct gfs2_inode *ip;
1434
1435         ip = kmem_cache_alloc(gfs2_inode_cachep, GFP_KERNEL);
1436         if (!ip)
1437                 return NULL;
1438         ip->i_flags = 0;
1439         ip->i_gl = NULL;
1440         gfs2_holder_mark_uninitialized(&ip->i_iopen_gh);
1441         memset(&ip->i_res, 0, sizeof(ip->i_res));
1442         RB_CLEAR_NODE(&ip->i_res.rs_node);
1443         ip->i_rahead = 0;
1444         return &ip->i_inode;
1445 }
1446
1447 static void gfs2_free_inode(struct inode *inode)
1448 {
1449         kmem_cache_free(gfs2_inode_cachep, GFS2_I(inode));
1450 }
1451
1452 const struct super_operations gfs2_super_ops = {
1453         .alloc_inode            = gfs2_alloc_inode,
1454         .free_inode             = gfs2_free_inode,
1455         .write_inode            = gfs2_write_inode,
1456         .dirty_inode            = gfs2_dirty_inode,
1457         .evict_inode            = gfs2_evict_inode,
1458         .put_super              = gfs2_put_super,
1459         .sync_fs                = gfs2_sync_fs,
1460         .freeze_super           = gfs2_freeze,
1461         .thaw_super             = gfs2_unfreeze,
1462         .statfs                 = gfs2_statfs,
1463         .drop_inode             = gfs2_drop_inode,
1464         .show_options           = gfs2_show_options,
1465 };
1466