fs/gfs2/super.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   4  * Copyright (C) 2004-2007 Red Hat, Inc.  All rights reserved.
   5  */
   6
   7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
   8
   9 #include <linux/bio.h>
  10 #include <linux/sched/signal.h>
  11 #include <linux/slab.h>
  12 #include <linux/spinlock.h>
  13 #include <linux/completion.h>
  14 #include <linux/buffer_head.h>
  15 #include <linux/statfs.h>
  16 #include <linux/seq_file.h>
  17 #include <linux/mount.h>
  18 #include <linux/kthread.h>
  19 #include <linux/delay.h>
  20 #include <linux/gfs2_ondisk.h>
  21 #include <linux/crc32.h>
  22 #include <linux/time.h>
  23 #include <linux/wait.h>
  24 #include <linux/writeback.h>
  25 #include <linux/backing-dev.h>
  26 #include <linux/kernel.h>
  27
  28 #include "gfs2.h"
  29 #include "incore.h"
  30 #include "bmap.h"
  31 #include "dir.h"
  32 #include "glock.h"
  33 #include "glops.h"
  34 #include "inode.h"
  35 #include "log.h"
  36 #include "meta_io.h"
  37 #include "quota.h"
  38 #include "recovery.h"
  39 #include "rgrp.h"
  40 #include "super.h"
  41 #include "trans.h"
  42 #include "util.h"
  43 #include "sys.h"
  44 #include "xattr.h"
  45 #include "lops.h"
  46
  47 enum dinode_demise {
  48         SHOULD_DELETE_DINODE,
  49         SHOULD_NOT_DELETE_DINODE,
  50         SHOULD_DEFER_EVICTION,
  51 };
  52
  53 /**
  54  * gfs2_jindex_free - Clear all the journal index information
  55  * @sdp: The GFS2 superblock
  56  *
  57  */
  58
  59 void gfs2_jindex_free(struct gfs2_sbd *sdp)
  60 {
  61         struct list_head list;
  62         struct gfs2_jdesc *jd;
  63
  64         spin_lock(&sdp->sd_jindex_spin);
  65         list_add(&list, &sdp->sd_jindex_list);
  66         list_del_init(&sdp->sd_jindex_list);
  67         sdp->sd_journals = 0;
  68         spin_unlock(&sdp->sd_jindex_spin);
  69
  70         sdp->sd_jdesc = NULL;
  71         while (!list_empty(&list)) {
  72                 jd = list_first_entry(&list, struct gfs2_jdesc, jd_list);
  73                 gfs2_free_journal_extents(jd);
  74                 list_del(&jd->jd_list);
  75                 iput(jd->jd_inode);
  76                 jd->jd_inode = NULL;
  77                 kfree(jd);
  78         }
  79 }
  80
  81 static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
  82 {
  83         struct gfs2_jdesc *jd;
  84
  85         list_for_each_entry(jd, head, jd_list) {
  86                 if (jd->jd_jid == jid)
  87                         return jd;
  88         }
  89         return NULL;
  90 }
  91
  92 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
  93 {
  94         struct gfs2_jdesc *jd;
  95
  96         spin_lock(&sdp->sd_jindex_spin);
  97         jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
  98         spin_unlock(&sdp->sd_jindex_spin);
  99
 100         return jd;
 101 }
 102
 103 int gfs2_jdesc_check(struct gfs2_jdesc *jd)
 104 {
 105         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 106         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 107         u64 size = i_size_read(jd->jd_inode);
 108
 109         if (gfs2_check_internal_file_size(jd->jd_inode, 8 << 20, BIT(30)))
 110                 return -EIO;
 111
 112         jd->jd_blocks = size >> sdp->sd_sb.sb_bsize_shift;
 113
 114         if (gfs2_write_alloc_required(ip, 0, size)) {
 115                 gfs2_consist_inode(ip);
 116                 return -EIO;
 117         }
 118
 119         return 0;
 120 }
 121
 122 /**
 123  * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
 124  * @sdp: the filesystem
 125  *
 126  * Returns: errno
 127  */
 128
 129 int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
 130 {
 131         struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
 132         struct gfs2_glock *j_gl = ip->i_gl;
 133         struct gfs2_log_header_host head;
 134         int error;
 135
 136         j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
 137         if (gfs2_withdrawn(sdp))
 138                 return -EIO;
 139
 140         error = gfs2_find_jhead(sdp->sd_jdesc, &head, false);
 141         if (error) {
 142                 gfs2_consist(sdp);
 143                 return error;
 144         }
 145
 146         if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 147                 gfs2_consist(sdp);
 148                 return -EIO;
 149         }
 150
 151         /*  Initialize some head of the log stuff  */
 152         sdp->sd_log_sequence = head.lh_sequence + 1;
 153         gfs2_log_pointers_init(sdp, head.lh_blkno);
 154
 155         error = gfs2_quota_init(sdp);
 156         if (!error && gfs2_withdrawn(sdp))
 157                 error = -EIO;
 158         if (!error)
 159                 set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 160         return error;
 161 }
 162
 163 void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
 164 {
 165         const struct gfs2_statfs_change *str = buf;
 166
 167         sc->sc_total = be64_to_cpu(str->sc_total);
 168         sc->sc_free = be64_to_cpu(str->sc_free);
 169         sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
 170 }
 171
 172 void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
 173 {
 174         struct gfs2_statfs_change *str = buf;
 175
 176         str->sc_total = cpu_to_be64(sc->sc_total);
 177         str->sc_free = cpu_to_be64(sc->sc_free);
 178         str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
 179 }
 180
 181 int gfs2_statfs_init(struct gfs2_sbd *sdp)
 182 {
 183         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 184         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 185         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 186         struct buffer_head *m_bh;
 187         struct gfs2_holder gh;
 188         int error;
 189
 190         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 191                                    &gh);
 192         if (error)
 193                 return error;
 194
 195         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 196         if (error)
 197                 goto out;
 198
 199         if (sdp->sd_args.ar_spectator) {
 200                 spin_lock(&sdp->sd_statfs_spin);
 201                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 202                                       sizeof(struct gfs2_dinode));
 203                 spin_unlock(&sdp->sd_statfs_spin);
 204         } else {
 205                 spin_lock(&sdp->sd_statfs_spin);
 206                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 207                                       sizeof(struct gfs2_dinode));
 208                 gfs2_statfs_change_in(l_sc, sdp->sd_sc_bh->b_data +
 209                                       sizeof(struct gfs2_dinode));
 210                 spin_unlock(&sdp->sd_statfs_spin);
 211
 212         }
 213
 214         brelse(m_bh);
 215 out:
 216         gfs2_glock_dq_uninit(&gh);
 217         return 0;
 218 }
 219
 220 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
 221                         s64 dinodes)
 222 {
 223         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 224         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 225         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 226         s64 x, y;
 227         int need_sync = 0;
 228
 229         gfs2_trans_add_meta(l_ip->i_gl, sdp->sd_sc_bh);
 230
 231         spin_lock(&sdp->sd_statfs_spin);
 232         l_sc->sc_total += total;
 233         l_sc->sc_free += free;
 234         l_sc->sc_dinodes += dinodes;
 235         gfs2_statfs_change_out(l_sc, sdp->sd_sc_bh->b_data +
 236                                sizeof(struct gfs2_dinode));
 237         if (sdp->sd_args.ar_statfs_percent) {
 238                 x = 100 * l_sc->sc_free;
 239                 y = m_sc->sc_free * sdp->sd_args.ar_statfs_percent;
 240                 if (x >= y || x <= -y)
 241                         need_sync = 1;
 242         }
 243         spin_unlock(&sdp->sd_statfs_spin);
 244
 245         if (need_sync)
 246                 gfs2_wake_up_statfs(sdp);
 247 }
 248
 249 void update_statfs(struct gfs2_sbd *sdp, struct buffer_head *m_bh)
 250 {
 251         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 252         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 253         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 254         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 255
 256         gfs2_trans_add_meta(l_ip->i_gl, sdp->sd_sc_bh);
 257         gfs2_trans_add_meta(m_ip->i_gl, m_bh);
 258
 259         spin_lock(&sdp->sd_statfs_spin);
 260         m_sc->sc_total += l_sc->sc_total;
 261         m_sc->sc_free += l_sc->sc_free;
 262         m_sc->sc_dinodes += l_sc->sc_dinodes;
 263         memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
 264         memset(sdp->sd_sc_bh->b_data + sizeof(struct gfs2_dinode),
 265                0, sizeof(struct gfs2_statfs_change));
 266         gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
 267         spin_unlock(&sdp->sd_statfs_spin);
 268 }
 269
 270 int gfs2_statfs_sync(struct super_block *sb, int type)
 271 {
 272         struct gfs2_sbd *sdp = sb->s_fs_info;
 273         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 274         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 275         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 276         struct gfs2_holder gh;
 277         struct buffer_head *m_bh;
 278         int error;
 279
 280         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 281                                    &gh);
 282         if (error)
 283                 goto out;
 284
 285         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 286         if (error)
 287                 goto out_unlock;
 288
 289         spin_lock(&sdp->sd_statfs_spin);
 290         gfs2_statfs_change_in(m_sc, m_bh->b_data +
 291                               sizeof(struct gfs2_dinode));
 292         if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
 293                 spin_unlock(&sdp->sd_statfs_spin);
 294                 goto out_bh;
 295         }
 296         spin_unlock(&sdp->sd_statfs_spin);
 297
 298         error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
 299         if (error)
 300                 goto out_bh;
 301
 302         update_statfs(sdp, m_bh);
 303         sdp->sd_statfs_force_sync = 0;
 304
 305         gfs2_trans_end(sdp);
 306
 307 out_bh:
 308         brelse(m_bh);
 309 out_unlock:
 310         gfs2_glock_dq_uninit(&gh);
 311 out:
 312         return error;
 313 }
 314
 315 struct lfcc {
 316         struct list_head list;
 317         struct gfs2_holder gh;
 318 };
 319
 320 /**
 321  * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
 322  *                            journals are clean
 323  * @sdp: the file system
 324  *
 325  * Returns: errno
 326  */
 327
 328 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp)
 329 {
 330         struct gfs2_inode *ip;
 331         struct gfs2_jdesc *jd;
 332         struct lfcc *lfcc;
 333         LIST_HEAD(list);
 334         struct gfs2_log_header_host lh;
 335         int error;
 336
 337         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 338                 lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
 339                 if (!lfcc) {
 340                         error = -ENOMEM;
 341                         goto out;
 342                 }
 343                 ip = GFS2_I(jd->jd_inode);
 344                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
 345                 if (error) {
 346                         kfree(lfcc);
 347                         goto out;
 348                 }
 349                 list_add(&lfcc->list, &list);
 350         }
 351
 352         error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_EXCLUSIVE,
 353                                    LM_FLAG_NOEXP | GL_NOPID,
 354                                    &sdp->sd_freeze_gh);
 355         if (error)
 356                 goto out;
 357
 358         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 359                 error = gfs2_jdesc_check(jd);
 360                 if (error)
 361                         break;
 362                 error = gfs2_find_jhead(jd, &lh, false);
 363                 if (error)
 364                         break;
 365                 if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 366                         error = -EBUSY;
 367                         break;
 368                 }
 369         }
 370
 371         if (error)
 372                 gfs2_freeze_unlock(&sdp->sd_freeze_gh);
 373
 374 out:
 375         while (!list_empty(&list)) {
 376                 lfcc = list_first_entry(&list, struct lfcc, list);
 377                 list_del(&lfcc->list);
 378                 gfs2_glock_dq_uninit(&lfcc->gh);
 379                 kfree(lfcc);
 380         }
 381         return error;
 382 }
 383
 384 void gfs2_dinode_out(const struct gfs2_inode *ip, void *buf)
 385 {
 386         const struct inode *inode = &ip->i_inode;
 387         struct gfs2_dinode *str = buf;
 388
 389         str->di_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 390         str->di_header.mh_type = cpu_to_be32(GFS2_METATYPE_DI);
 391         str->di_header.mh_format = cpu_to_be32(GFS2_FORMAT_DI);
 392         str->di_num.no_addr = cpu_to_be64(ip->i_no_addr);
 393         str->di_num.no_formal_ino = cpu_to_be64(ip->i_no_formal_ino);
 394         str->di_mode = cpu_to_be32(inode->i_mode);
 395         str->di_uid = cpu_to_be32(i_uid_read(inode));
 396         str->di_gid = cpu_to_be32(i_gid_read(inode));
 397         str->di_nlink = cpu_to_be32(inode->i_nlink);
 398         str->di_size = cpu_to_be64(i_size_read(inode));
 399         str->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(inode));
 400         str->di_atime = cpu_to_be64(inode->i_atime.tv_sec);
 401         str->di_mtime = cpu_to_be64(inode->i_mtime.tv_sec);
 402         str->di_ctime = cpu_to_be64(inode->i_ctime.tv_sec);
 403
 404         str->di_goal_meta = cpu_to_be64(ip->i_goal);
 405         str->di_goal_data = cpu_to_be64(ip->i_goal);
 406         str->di_generation = cpu_to_be64(ip->i_generation);
 407
 408         str->di_flags = cpu_to_be32(ip->i_diskflags);
 409         str->di_height = cpu_to_be16(ip->i_height);
 410         str->di_payload_format = cpu_to_be32(S_ISDIR(inode->i_mode) &&
 411                                              !(ip->i_diskflags & GFS2_DIF_EXHASH) ?
 412                                              GFS2_FORMAT_DE : 0);
 413         str->di_depth = cpu_to_be16(ip->i_depth);
 414         str->di_entries = cpu_to_be32(ip->i_entries);
 415
 416         str->di_eattr = cpu_to_be64(ip->i_eattr);
 417         str->di_atime_nsec = cpu_to_be32(inode->i_atime.tv_nsec);
 418         str->di_mtime_nsec = cpu_to_be32(inode->i_mtime.tv_nsec);
 419         str->di_ctime_nsec = cpu_to_be32(inode->i_ctime.tv_nsec);
 420 }
 421
 422 /**
 423  * gfs2_write_inode - Make sure the inode is stable on the disk
 424  * @inode: The inode
 425  * @wbc: The writeback control structure
 426  *
 427  * Returns: errno
 428  */
 429
 430 static int gfs2_write_inode(struct inode *inode, struct writeback_control *wbc)
 431 {
 432         struct gfs2_inode *ip = GFS2_I(inode);
 433         struct gfs2_sbd *sdp = GFS2_SB(inode);
 434         struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl);
 435         struct backing_dev_info *bdi = inode_to_bdi(metamapping->host);
 436         int ret = 0;
 437         bool flush_all = (wbc->sync_mode == WB_SYNC_ALL || gfs2_is_jdata(ip));
 438
 439         if (flush_all)
 440                 gfs2_log_flush(GFS2_SB(inode), ip->i_gl,
 441                                GFS2_LOG_HEAD_FLUSH_NORMAL |
 442                                GFS2_LFC_WRITE_INODE);
 443         if (bdi->wb.dirty_exceeded)
 444                 gfs2_ail1_flush(sdp, wbc);
 445         else
 446                 filemap_fdatawrite(metamapping);
 447         if (flush_all)
 448                 ret = filemap_fdatawait(metamapping);
 449         if (ret)
 450                 mark_inode_dirty_sync(inode);
 451         else {
 452                 spin_lock(&inode->i_lock);
 453                 if (!(inode->i_flags & I_DIRTY))
 454                         gfs2_ordered_del_inode(ip);
 455                 spin_unlock(&inode->i_lock);
 456         }
 457         return ret;
 458 }
 459
 460 /**
 461  * gfs2_dirty_inode - check for atime updates
 462  * @inode: The inode in question
 463  * @flags: The type of dirty
 464  *
 465  * Unfortunately it can be called under any combination of inode
 466  * glock and transaction lock, so we have to check carefully.
 467  *
 468  * At the moment this deals only with atime - it should be possible
 469  * to expand that role in future, once a review of the locking has
 470  * been carried out.
 471  */
 472
 473 static void gfs2_dirty_inode(struct inode *inode, int flags)
 474 {
 475         struct gfs2_inode *ip = GFS2_I(inode);
 476         struct gfs2_sbd *sdp = GFS2_SB(inode);
 477         struct buffer_head *bh;
 478         struct gfs2_holder gh;
 479         int need_unlock = 0;
 480         int need_endtrans = 0;
 481         int ret;
 482
 483         if (unlikely(!ip->i_gl)) {
 484                 /* This can only happen during incomplete inode creation. */
 485                 BUG_ON(!test_bit(GIF_ALLOC_FAILED, &ip->i_flags));
 486                 return;
 487         }
 488
 489         if (unlikely(gfs2_withdrawn(sdp)))
 490                 return;
 491         if (!gfs2_glock_is_locked_by_me(ip->i_gl)) {
 492                 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
 493                 if (ret) {
 494                         fs_err(sdp, "dirty_inode: glock %d\n", ret);
 495                         gfs2_dump_glock(NULL, ip->i_gl, true);
 496                         return;
 497                 }
 498                 need_unlock = 1;
 499         } else if (WARN_ON_ONCE(ip->i_gl->gl_state != LM_ST_EXCLUSIVE))
 500                 return;
 501
 502         if (current->journal_info == NULL) {
 503                 ret = gfs2_trans_begin(sdp, RES_DINODE, 0);
 504                 if (ret) {
 505                         fs_err(sdp, "dirty_inode: gfs2_trans_begin %d\n", ret);
 506                         goto out;
 507                 }
 508                 need_endtrans = 1;
 509         }
 510
 511         ret = gfs2_meta_inode_buffer(ip, &bh);
 512         if (ret == 0) {
 513                 gfs2_trans_add_meta(ip->i_gl, bh);
 514                 gfs2_dinode_out(ip, bh->b_data);
 515                 brelse(bh);
 516         }
 517
 518         if (need_endtrans)
 519                 gfs2_trans_end(sdp);
 520 out:
 521         if (need_unlock)
 522                 gfs2_glock_dq_uninit(&gh);
 523 }
 524
 525 /**
 526  * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
 527  * @sdp: the filesystem
 528  *
 529  * Returns: errno
 530  */
 531
 532 void gfs2_make_fs_ro(struct gfs2_sbd *sdp)
 533 {
 534         int log_write_allowed = test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 535
 536         if (!test_bit(SDF_DEACTIVATING, &sdp->sd_flags))
 537                 gfs2_flush_delete_work(sdp);
 538
 539         if (!log_write_allowed && current == sdp->sd_quotad_process)
 540                 fs_warn(sdp, "The quotad daemon is withdrawing.\n");
 541         else if (sdp->sd_quotad_process)
 542                 kthread_stop(sdp->sd_quotad_process);
 543         sdp->sd_quotad_process = NULL;
 544
 545         if (!log_write_allowed && current == sdp->sd_logd_process)
 546                 fs_warn(sdp, "The logd daemon is withdrawing.\n");
 547         else if (sdp->sd_logd_process)
 548                 kthread_stop(sdp->sd_logd_process);
 549         sdp->sd_logd_process = NULL;
 550
 551         if (log_write_allowed) {
 552                 gfs2_quota_sync(sdp->sd_vfs, 0);
 553                 gfs2_statfs_sync(sdp->sd_vfs, 0);
 554
 555                 /* We do two log flushes here. The first one commits dirty inodes
 556                  * and rgrps to the journal, but queues up revokes to the ail list.
 557                  * The second flush writes out and removes the revokes.
 558                  *
 559                  * The first must be done before the FLUSH_SHUTDOWN code
 560                  * clears the LIVE flag, otherwise it will not be able to start
 561                  * a transaction to write its revokes, and the error will cause
 562                  * a withdraw of the file system. */
 563                 gfs2_log_flush(sdp, NULL, GFS2_LFC_MAKE_FS_RO);
 564                 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SHUTDOWN |
 565                                GFS2_LFC_MAKE_FS_RO);
 566                 wait_event_timeout(sdp->sd_log_waitq,
 567                                    gfs2_log_is_empty(sdp),
 568                                    HZ * 5);
 569                 gfs2_assert_warn(sdp, gfs2_log_is_empty(sdp));
 570         } else {
 571                 wait_event_timeout(sdp->sd_log_waitq,
 572                                    gfs2_log_is_empty(sdp),
 573                                    HZ * 5);
 574         }
 575         gfs2_quota_cleanup(sdp);
 576
 577         if (!log_write_allowed)
 578                 sdp->sd_vfs->s_flags |= SB_RDONLY;
 579 }
 580
 581 /**
 582  * gfs2_put_super - Unmount the filesystem
 583  * @sb: The VFS superblock
 584  *
 585  */
 586
 587 static void gfs2_put_super(struct super_block *sb)
 588 {
 589         struct gfs2_sbd *sdp = sb->s_fs_info;
 590         struct gfs2_jdesc *jd;
 591
 592         /* No more recovery requests */
 593         set_bit(SDF_NORECOVERY, &sdp->sd_flags);
 594         smp_mb();
 595
 596         /* Wait on outstanding recovery */
 597 restart:
 598         spin_lock(&sdp->sd_jindex_spin);
 599         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 600                 if (!test_bit(JDF_RECOVERY, &jd->jd_flags))
 601                         continue;
 602                 spin_unlock(&sdp->sd_jindex_spin);
 603                 wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
 604                             TASK_UNINTERRUPTIBLE);
 605                 goto restart;
 606         }
 607         spin_unlock(&sdp->sd_jindex_spin);
 608
 609         if (!sb_rdonly(sb)) {
 610                 gfs2_make_fs_ro(sdp);
 611         }
 612         WARN_ON(gfs2_withdrawing(sdp));
 613
 614         /*  At this point, we're through modifying the disk  */
 615
 616         /*  Release stuff  */
 617
 618         iput(sdp->sd_jindex);
 619         iput(sdp->sd_statfs_inode);
 620         iput(sdp->sd_rindex);
 621         iput(sdp->sd_quota_inode);
 622
 623         gfs2_glock_put(sdp->sd_rename_gl);
 624         gfs2_glock_put(sdp->sd_freeze_gl);
 625
 626         if (!sdp->sd_args.ar_spectator) {
 627                 if (gfs2_holder_initialized(&sdp->sd_journal_gh))
 628                         gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
 629                 if (gfs2_holder_initialized(&sdp->sd_jinode_gh))
 630                         gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
 631                 brelse(sdp->sd_sc_bh);
 632                 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
 633                 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
 634                 free_local_statfs_inodes(sdp);
 635                 iput(sdp->sd_qc_inode);
 636         }
 637
 638         gfs2_glock_dq_uninit(&sdp->sd_live_gh);
 639         gfs2_clear_rgrpd(sdp);
 640         gfs2_jindex_free(sdp);
 641         /*  Take apart glock structures and buffer lists  */
 642         gfs2_gl_hash_clear(sdp);
 643         truncate_inode_pages_final(&sdp->sd_aspace);
 644         gfs2_delete_debugfs_file(sdp);
 645         /*  Unmount the locking protocol  */
 646         gfs2_lm_unmount(sdp);
 647
 648         /*  At this point, we're through participating in the lockspace  */
 649         gfs2_sys_fs_del(sdp);
 650         free_sbd(sdp);
 651 }
 652
 653 /**
 654  * gfs2_sync_fs - sync the filesystem
 655  * @sb: the superblock
 656  * @wait: true to wait for completion
 657  *
 658  * Flushes the log to disk.
 659  */
 660
 661 static int gfs2_sync_fs(struct super_block *sb, int wait)
 662 {
 663         struct gfs2_sbd *sdp = sb->s_fs_info;
 664
 665         gfs2_quota_sync(sb, -1);
 666         if (wait)
 667                 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
 668                                GFS2_LFC_SYNC_FS);
 669         return sdp->sd_log_error;
 670 }
 671
 672 void gfs2_freeze_func(struct work_struct *work)
 673 {
 674         int error;
 675         struct gfs2_holder freeze_gh;
 676         struct gfs2_sbd *sdp = container_of(work, struct gfs2_sbd, sd_freeze_work);
 677         struct super_block *sb = sdp->sd_vfs;
 678
 679         atomic_inc(&sb->s_active);
 680         error = gfs2_freeze_lock(sdp, &freeze_gh, 0);
 681         if (error) {
 682                 gfs2_assert_withdraw(sdp, 0);
 683         } else {
 684                 atomic_set(&sdp->sd_freeze_state, SFS_UNFROZEN);
 685                 error = thaw_super(sb);
 686                 if (error) {
 687                         fs_info(sdp, "GFS2: couldn't thaw filesystem: %d\n",
 688                                 error);
 689                         gfs2_assert_withdraw(sdp, 0);
 690                 }
 691                 gfs2_freeze_unlock(&freeze_gh);
 692         }
 693         deactivate_super(sb);
 694         clear_bit_unlock(SDF_FS_FROZEN, &sdp->sd_flags);
 695         wake_up_bit(&sdp->sd_flags, SDF_FS_FROZEN);
 696         return;
 697 }
 698
 699 /**
 700  * gfs2_freeze - prevent further writes to the filesystem
 701  * @sb: the VFS structure for the filesystem
 702  *
 703  */
 704
 705 static int gfs2_freeze(struct super_block *sb)
 706 {
 707         struct gfs2_sbd *sdp = sb->s_fs_info;
 708         int error;
 709
 710         mutex_lock(&sdp->sd_freeze_mutex);
 711         if (atomic_read(&sdp->sd_freeze_state) != SFS_UNFROZEN) {
 712                 error = -EBUSY;
 713                 goto out;
 714         }
 715
 716         for (;;) {
 717                 if (gfs2_withdrawn(sdp)) {
 718                         error = -EINVAL;
 719                         goto out;
 720                 }
 721
 722                 error = gfs2_lock_fs_check_clean(sdp);
 723                 if (!error)
 724                         break;
 725
 726                 if (error == -EBUSY)
 727                         fs_err(sdp, "waiting for recovery before freeze\n");
 728                 else if (error == -EIO) {
 729                         fs_err(sdp, "Fatal IO error: cannot freeze gfs2 due "
 730                                "to recovery error.\n");
 731                         goto out;
 732                 } else {
 733                         fs_err(sdp, "error freezing FS: %d\n", error);
 734                 }
 735                 fs_err(sdp, "retrying...\n");
 736                 msleep(1000);
 737         }
 738         set_bit(SDF_FS_FROZEN, &sdp->sd_flags);
 739 out:
 740         mutex_unlock(&sdp->sd_freeze_mutex);
 741         return error;
 742 }
 743
 744 /**
 745  * gfs2_unfreeze - reallow writes to the filesystem
 746  * @sb: the VFS structure for the filesystem
 747  *
 748  */
 749
 750 static int gfs2_unfreeze(struct super_block *sb)
 751 {
 752         struct gfs2_sbd *sdp = sb->s_fs_info;
 753
 754         mutex_lock(&sdp->sd_freeze_mutex);
 755         if (atomic_read(&sdp->sd_freeze_state) != SFS_FROZEN ||
 756             !gfs2_holder_initialized(&sdp->sd_freeze_gh)) {
 757                 mutex_unlock(&sdp->sd_freeze_mutex);
 758                 return -EINVAL;
 759         }
 760
 761         gfs2_freeze_unlock(&sdp->sd_freeze_gh);
 762         mutex_unlock(&sdp->sd_freeze_mutex);
 763         return wait_on_bit(&sdp->sd_flags, SDF_FS_FROZEN, TASK_INTERRUPTIBLE);
 764 }
 765
 766 /**
 767  * statfs_slow_fill - fill in the sg for a given RG
 768  * @rgd: the RG
 769  * @sc: the sc structure
 770  *
 771  * Returns: 0 on success, -ESTALE if the LVB is invalid
 772  */
 773
 774 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
 775                             struct gfs2_statfs_change_host *sc)
 776 {
 777         gfs2_rgrp_verify(rgd);
 778         sc->sc_total += rgd->rd_data;
 779         sc->sc_free += rgd->rd_free;
 780         sc->sc_dinodes += rgd->rd_dinodes;
 781         return 0;
 782 }
 783
 784 /**
 785  * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
 786  * @sdp: the filesystem
 787  * @sc: the sc info that will be returned
 788  *
 789  * Any error (other than a signal) will cause this routine to fall back
 790  * to the synchronous version.
 791  *
 792  * FIXME: This really shouldn't busy wait like this.
 793  *
 794  * Returns: errno
 795  */
 796
 797 static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
 798 {
 799         struct gfs2_rgrpd *rgd_next;
 800         struct gfs2_holder *gha, *gh;
 801         unsigned int slots = 64;
 802         unsigned int x;
 803         int done;
 804         int error = 0, err;
 805
 806         memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
 807         gha = kmalloc_array(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
 808         if (!gha)
 809                 return -ENOMEM;
 810         for (x = 0; x < slots; x++)
 811                 gfs2_holder_mark_uninitialized(gha + x);
 812
 813         rgd_next = gfs2_rgrpd_get_first(sdp);
 814
 815         for (;;) {
 816                 done = 1;
 817
 818                 for (x = 0; x < slots; x++) {
 819                         gh = gha + x;
 820
 821                         if (gfs2_holder_initialized(gh) && gfs2_glock_poll(gh)) {
 822                                 err = gfs2_glock_wait(gh);
 823                                 if (err) {
 824                                         gfs2_holder_uninit(gh);
 825                                         error = err;
 826                                 } else {
 827                                         if (!error) {
 828                                                 struct gfs2_rgrpd *rgd =
 829                                                         gfs2_glock2rgrp(gh->gh_gl);
 830
 831                                                 error = statfs_slow_fill(rgd, sc);
 832                                         }
 833                                         gfs2_glock_dq_uninit(gh);
 834                                 }
 835                         }
 836
 837                         if (gfs2_holder_initialized(gh))
 838                                 done = 0;
 839                         else if (rgd_next && !error) {
 840                                 error = gfs2_glock_nq_init(rgd_next->rd_gl,
 841                                                            LM_ST_SHARED,
 842                                                            GL_ASYNC,
 843                                                            gh);
 844                                 rgd_next = gfs2_rgrpd_get_next(rgd_next);
 845                                 done = 0;
 846                         }
 847
 848                         if (signal_pending(current))
 849                                 error = -ERESTARTSYS;
 850                 }
 851
 852                 if (done)
 853                         break;
 854
 855                 yield();
 856         }
 857
 858         kfree(gha);
 859         return error;
 860 }
 861
 862 /**
 863  * gfs2_statfs_i - Do a statfs
 864  * @sdp: the filesystem
 865  * @sc: the sc structure
 866  *
 867  * Returns: errno
 868  */
 869
 870 static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
 871 {
 872         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 873         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 874
 875         spin_lock(&sdp->sd_statfs_spin);
 876
 877         *sc = *m_sc;
 878         sc->sc_total += l_sc->sc_total;
 879         sc->sc_free += l_sc->sc_free;
 880         sc->sc_dinodes += l_sc->sc_dinodes;
 881
 882         spin_unlock(&sdp->sd_statfs_spin);
 883
 884         if (sc->sc_free < 0)
 885                 sc->sc_free = 0;
 886         if (sc->sc_free > sc->sc_total)
 887                 sc->sc_free = sc->sc_total;
 888         if (sc->sc_dinodes < 0)
 889                 sc->sc_dinodes = 0;
 890
 891         return 0;
 892 }
 893
 894 /**
 895  * gfs2_statfs - Gather and return stats about the filesystem
 896  * @dentry: The name of the link
 897  * @buf: The buffer
 898  *
 899  * Returns: 0 on success or error code
 900  */
 901
 902 static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
 903 {
 904         struct super_block *sb = dentry->d_sb;
 905         struct gfs2_sbd *sdp = sb->s_fs_info;
 906         struct gfs2_statfs_change_host sc;
 907         int error;
 908
 909         error = gfs2_rindex_update(sdp);
 910         if (error)
 911                 return error;
 912
 913         if (gfs2_tune_get(sdp, gt_statfs_slow))
 914                 error = gfs2_statfs_slow(sdp, &sc);
 915         else
 916                 error = gfs2_statfs_i(sdp, &sc);
 917
 918         if (error)
 919                 return error;
 920
 921         buf->f_type = GFS2_MAGIC;
 922         buf->f_bsize = sdp->sd_sb.sb_bsize;
 923         buf->f_blocks = sc.sc_total;
 924         buf->f_bfree = sc.sc_free;
 925         buf->f_bavail = sc.sc_free;
 926         buf->f_files = sc.sc_dinodes + sc.sc_free;
 927         buf->f_ffree = sc.sc_free;
 928         buf->f_namelen = GFS2_FNAMESIZE;
 929
 930         return 0;
 931 }
 932
 933 /**
 934  * gfs2_drop_inode - Drop an inode (test for remote unlink)
 935  * @inode: The inode to drop
 936  *
 937  * If we've received a callback on an iopen lock then it's because a
 938  * remote node tried to deallocate the inode but failed due to this node
 939  * still having the inode open. Here we mark the link count zero
 940  * since we know that it must have reached zero if the GLF_DEMOTE flag
 941  * is set on the iopen glock. If we didn't do a disk read since the
 942  * remote node removed the final link then we might otherwise miss
 943  * this event. This check ensures that this node will deallocate the
 944  * inode's blocks, or alternatively pass the baton on to another
 945  * node for later deallocation.
 946  */
 947
 948 static int gfs2_drop_inode(struct inode *inode)
 949 {
 950         struct gfs2_inode *ip = GFS2_I(inode);
 951         struct gfs2_sbd *sdp = GFS2_SB(inode);
 952
 953         if (inode->i_nlink &&
 954             gfs2_holder_initialized(&ip->i_iopen_gh)) {
 955                 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
 956                 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
 957                         clear_nlink(inode);
 958         }
 959
 960         /*
 961          * When under memory pressure when an inode's link count has dropped to
 962          * zero, defer deleting the inode to the delete workqueue.  This avoids
 963          * calling into DLM under memory pressure, which can deadlock.
 964          */
 965         if (!inode->i_nlink &&
 966             unlikely(current->flags & PF_MEMALLOC) &&
 967             gfs2_holder_initialized(&ip->i_iopen_gh)) {
 968                 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
 969
 970                 gfs2_glock_hold(gl);
 971                 if (!gfs2_queue_try_to_evict(gl))
 972                         gfs2_glock_queue_put(gl);
 973                 return 0;
 974         }
 975
 976         /*
 977          * No longer cache inodes when trying to evict them all.
 978          */
 979         if (test_bit(SDF_EVICTING, &sdp->sd_flags))
 980                 return 1;
 981
 982         return generic_drop_inode(inode);
 983 }
 984
 985 static int is_ancestor(const struct dentry *d1, const struct dentry *d2)
 986 {
 987         do {
 988                 if (d1 == d2)
 989                         return 1;
 990                 d1 = d1->d_parent;
 991         } while (!IS_ROOT(d1));
 992         return 0;
 993 }
 994
 995 /**
 996  * gfs2_show_options - Show mount options for /proc/mounts
 997  * @s: seq_file structure
 998  * @root: root of this (sub)tree
 999  *
1000  * Returns: 0 on success or error code
1001  */
1002
1003 static int gfs2_show_options(struct seq_file *s, struct dentry *root)
1004 {
1005         struct gfs2_sbd *sdp = root->d_sb->s_fs_info;
1006         struct gfs2_args *args = &sdp->sd_args;
1007         int val;
1008
1009         if (is_ancestor(root, sdp->sd_master_dir))
1010                 seq_puts(s, ",meta");
1011         if (args->ar_lockproto[0])
1012                 seq_show_option(s, "lockproto", args->ar_lockproto);
1013         if (args->ar_locktable[0])
1014                 seq_show_option(s, "locktable", args->ar_locktable);
1015         if (args->ar_hostdata[0])
1016                 seq_show_option(s, "hostdata", args->ar_hostdata);
1017         if (args->ar_spectator)
1018                 seq_puts(s, ",spectator");
1019         if (args->ar_localflocks)
1020                 seq_puts(s, ",localflocks");
1021         if (args->ar_debug)
1022                 seq_puts(s, ",debug");
1023         if (args->ar_posix_acl)
1024                 seq_puts(s, ",acl");
1025         if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
1026                 char *state;
1027                 switch (args->ar_quota) {
1028                 case GFS2_QUOTA_OFF:
1029                         state = "off";
1030                         break;
1031                 case GFS2_QUOTA_ACCOUNT:
1032                         state = "account";
1033                         break;
1034                 case GFS2_QUOTA_ON:
1035                         state = "on";
1036                         break;
1037                 default:
1038                         state = "unknown";
1039                         break;
1040                 }
1041                 seq_printf(s, ",quota=%s", state);
1042         }
1043         if (args->ar_suiddir)
1044                 seq_puts(s, ",suiddir");
1045         if (args->ar_data != GFS2_DATA_DEFAULT) {
1046                 char *state;
1047                 switch (args->ar_data) {
1048                 case GFS2_DATA_WRITEBACK:
1049                         state = "writeback";
1050                         break;
1051                 case GFS2_DATA_ORDERED:
1052                         state = "ordered";
1053                         break;
1054                 default:
1055                         state = "unknown";
1056                         break;
1057                 }
1058                 seq_printf(s, ",data=%s", state);
1059         }
1060         if (args->ar_discard)
1061                 seq_puts(s, ",discard");
1062         val = sdp->sd_tune.gt_logd_secs;
1063         if (val != 30)
1064                 seq_printf(s, ",commit=%d", val);
1065         val = sdp->sd_tune.gt_statfs_quantum;
1066         if (val != 30)
1067                 seq_printf(s, ",statfs_quantum=%d", val);
1068         else if (sdp->sd_tune.gt_statfs_slow)
1069                 seq_puts(s, ",statfs_quantum=0");
1070         val = sdp->sd_tune.gt_quota_quantum;
1071         if (val != 60)
1072                 seq_printf(s, ",quota_quantum=%d", val);
1073         if (args->ar_statfs_percent)
1074                 seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent);
1075         if (args->ar_errors != GFS2_ERRORS_DEFAULT) {
1076                 const char *state;
1077
1078                 switch (args->ar_errors) {
1079                 case GFS2_ERRORS_WITHDRAW:
1080                         state = "withdraw";
1081                         break;
1082                 case GFS2_ERRORS_PANIC:
1083                         state = "panic";
1084                         break;
1085                 default:
1086                         state = "unknown";
1087                         break;
1088                 }
1089                 seq_printf(s, ",errors=%s", state);
1090         }
1091         if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags))
1092                 seq_puts(s, ",nobarrier");
1093         if (test_bit(SDF_DEMOTE, &sdp->sd_flags))
1094                 seq_puts(s, ",demote_interface_used");
1095         if (args->ar_rgrplvb)
1096                 seq_puts(s, ",rgrplvb");
1097         if (args->ar_loccookie)
1098                 seq_puts(s, ",loccookie");
1099         return 0;
1100 }
1101
1102 static void gfs2_final_release_pages(struct gfs2_inode *ip)
1103 {
1104         struct inode *inode = &ip->i_inode;
1105         struct gfs2_glock *gl = ip->i_gl;
1106
1107         if (unlikely(!gl)) {
1108                 /* This can only happen during incomplete inode creation. */
1109                 BUG_ON(!test_bit(GIF_ALLOC_FAILED, &ip->i_flags));
1110                 return;
1111         }
1112
1113         truncate_inode_pages(gfs2_glock2aspace(gl), 0);
1114         truncate_inode_pages(&inode->i_data, 0);
1115
1116         if (atomic_read(&gl->gl_revokes) == 0) {
1117                 clear_bit(GLF_LFLUSH, &gl->gl_flags);
1118                 clear_bit(GLF_DIRTY, &gl->gl_flags);
1119         }
1120 }
1121
1122 static int gfs2_dinode_dealloc(struct gfs2_inode *ip)
1123 {
1124         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1125         struct gfs2_rgrpd *rgd;
1126         struct gfs2_holder gh;
1127         int error;
1128
1129         if (gfs2_get_inode_blocks(&ip->i_inode) != 1) {
1130                 gfs2_consist_inode(ip);
1131                 return -EIO;
1132         }
1133
1134         error = gfs2_rindex_update(sdp);
1135         if (error)
1136                 return error;
1137
1138         error = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
1139         if (error)
1140                 return error;
1141
1142         rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1143         if (!rgd) {
1144                 gfs2_consist_inode(ip);
1145                 error = -EIO;
1146                 goto out_qs;
1147         }
1148
1149         error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE,
1150                                    LM_FLAG_NODE_SCOPE, &gh);
1151         if (error)
1152                 goto out_qs;
1153
1154         error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA,
1155                                  sdp->sd_jdesc->jd_blocks);
1156         if (error)
1157                 goto out_rg_gunlock;
1158
1159         gfs2_free_di(rgd, ip);
1160
1161         gfs2_final_release_pages(ip);
1162
1163         gfs2_trans_end(sdp);
1164
1165 out_rg_gunlock:
1166         gfs2_glock_dq_uninit(&gh);
1167 out_qs:
1168         gfs2_quota_unhold(ip);
1169         return error;
1170 }
1171
1172 /**
1173  * gfs2_glock_put_eventually
1174  * @gl: The glock to put
1175  *
1176  * When under memory pressure, trigger a deferred glock put to make sure we
1177  * won't call into DLM and deadlock.  Otherwise, put the glock directly.
1178  */
1179
1180 static void gfs2_glock_put_eventually(struct gfs2_glock *gl)
1181 {
1182         if (current->flags & PF_MEMALLOC)
1183                 gfs2_glock_queue_put(gl);
1184         else
1185                 gfs2_glock_put(gl);
1186 }
1187
1188 static bool gfs2_upgrade_iopen_glock(struct inode *inode)
1189 {
1190         struct gfs2_inode *ip = GFS2_I(inode);
1191         struct gfs2_sbd *sdp = GFS2_SB(inode);
1192         struct gfs2_holder *gh = &ip->i_iopen_gh;
1193         long timeout = 5 * HZ;
1194         int error;
1195
1196         gh->gh_flags |= GL_NOCACHE;
1197         gfs2_glock_dq_wait(gh);
1198
1199         /*
1200          * If there are no other lock holders, we will immediately get
1201          * exclusive access to the iopen glock here.
1202          *
1203          * Otherwise, the other nodes holding the lock will be notified about
1204          * our locking request.  If they do not have the inode open, they are
1205          * expected to evict the cached inode and release the lock, allowing us
1206          * to proceed.
1207          *
1208          * Otherwise, if they cannot evict the inode, they are expected to poke
1209          * the inode glock (note: not the iopen glock).  We will notice that
1210          * and stop waiting for the iopen glock immediately.  The other node(s)
1211          * are then expected to take care of deleting the inode when they no
1212          * longer use it.
1213          *
1214          * As a last resort, if another node keeps holding the iopen glock
1215          * without showing any activity on the inode glock, we will eventually
1216          * time out and fail the iopen glock upgrade.
1217          *
1218          * Note that we're passing the LM_FLAG_TRY_1CB flag to the first
1219          * locking request as an optimization to notify lock holders as soon as
1220          * possible.  Without that flag, they'd be notified implicitly by the
1221          * second locking request.
1222          */
1223
1224         gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE, gh);
1225         error = gfs2_glock_nq(gh);
1226         if (error != GLR_TRYFAILED)
1227                 return !error;
1228
1229         gfs2_holder_reinit(LM_ST_EXCLUSIVE, GL_ASYNC | GL_NOCACHE, gh);
1230         error = gfs2_glock_nq(gh);
1231         if (error)
1232                 return false;
1233
1234         timeout = wait_event_interruptible_timeout(sdp->sd_async_glock_wait,
1235                 !test_bit(HIF_WAIT, &gh->gh_iflags) ||
1236                 test_bit(GLF_DEMOTE, &ip->i_gl->gl_flags),
1237                 timeout);
1238         if (!test_bit(HIF_HOLDER, &gh->gh_iflags)) {
1239                 gfs2_glock_dq(gh);
1240                 return false;
1241         }
1242         return gfs2_glock_holder_ready(gh) == 0;
1243 }
1244
1245 /**
1246  * evict_should_delete - determine whether the inode is eligible for deletion
1247  * @inode: The inode to evict
1248  * @gh: The glock holder structure
1249  *
1250  * This function determines whether the evicted inode is eligible to be deleted
1251  * and locks the inode glock.
1252  *
1253  * Returns: the fate of the dinode
1254  */
1255 static enum dinode_demise evict_should_delete(struct inode *inode,
1256                                               struct gfs2_holder *gh)
1257 {
1258         struct gfs2_inode *ip = GFS2_I(inode);
1259         struct super_block *sb = inode->i_sb;
1260         struct gfs2_sbd *sdp = sb->s_fs_info;
1261         int ret;
1262
1263         if (unlikely(test_bit(GIF_ALLOC_FAILED, &ip->i_flags)))
1264                 goto should_delete;
1265
1266         if (test_bit(GIF_DEFERRED_DELETE, &ip->i_flags))
1267                 return SHOULD_DEFER_EVICTION;
1268
1269         /* Deletes should never happen under memory pressure anymore.  */
1270         if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
1271                 return SHOULD_DEFER_EVICTION;
1272
1273         /* Must not read inode block until block type has been verified */
1274         ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, gh);
1275         if (unlikely(ret)) {
1276                 glock_clear_object(ip->i_iopen_gh.gh_gl, ip);
1277                 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1278                 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1279                 return SHOULD_DEFER_EVICTION;
1280         }
1281
1282         if (gfs2_inode_already_deleted(ip->i_gl, ip->i_no_formal_ino))
1283                 return SHOULD_NOT_DELETE_DINODE;
1284         ret = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED);
1285         if (ret)
1286                 return SHOULD_NOT_DELETE_DINODE;
1287
1288         ret = gfs2_instantiate(gh);
1289         if (ret)
1290                 return SHOULD_NOT_DELETE_DINODE;
1291
1292         /*
1293          * The inode may have been recreated in the meantime.
1294          */
1295         if (inode->i_nlink)
1296                 return SHOULD_NOT_DELETE_DINODE;
1297
1298 should_delete:
1299         if (gfs2_holder_initialized(&ip->i_iopen_gh) &&
1300             test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1301                 if (!gfs2_upgrade_iopen_glock(inode)) {
1302                         gfs2_holder_uninit(&ip->i_iopen_gh);
1303                         return SHOULD_NOT_DELETE_DINODE;
1304                 }
1305         }
1306         return SHOULD_DELETE_DINODE;
1307 }
1308
1309 /**
1310  * evict_unlinked_inode - delete the pieces of an unlinked evicted inode
1311  * @inode: The inode to evict
1312  */
1313 static int evict_unlinked_inode(struct inode *inode)
1314 {
1315         struct gfs2_inode *ip = GFS2_I(inode);
1316         int ret;
1317
1318         if (S_ISDIR(inode->i_mode) &&
1319             (ip->i_diskflags & GFS2_DIF_EXHASH)) {
1320                 ret = gfs2_dir_exhash_dealloc(ip);
1321                 if (ret)
1322                         goto out;
1323         }
1324
1325         if (ip->i_eattr) {
1326                 ret = gfs2_ea_dealloc(ip);
1327                 if (ret)
1328                         goto out;
1329         }
1330
1331         if (!gfs2_is_stuffed(ip)) {
1332                 ret = gfs2_file_dealloc(ip);
1333                 if (ret)
1334                         goto out;
1335         }
1336
1337         if (ip->i_gl)
1338                 gfs2_inode_remember_delete(ip->i_gl, ip->i_no_formal_ino);
1339
1340         /*
1341          * As soon as we clear the bitmap for the dinode, gfs2_create_inode()
1342          * can get called to recreate it, or even gfs2_inode_lookup() if the
1343          * inode was recreated on another node in the meantime.
1344          *
1345          * However, inserting the new inode into the inode hash table will not
1346          * succeed until the old inode is removed, and that only happens after
1347          * ->evict_inode() returns.  The new inode is attached to its inode and
1348          *  iopen glocks after inserting it into the inode hash table, so at
1349          *  that point we can be sure that both glocks are unused.
1350          */
1351
1352         ret = gfs2_dinode_dealloc(ip);
1353 out:
1354         return ret;
1355 }
1356
1357 /*
1358  * evict_linked_inode - evict an inode whose dinode has not been unlinked
1359  * @inode: The inode to evict
1360  */
1361 static int evict_linked_inode(struct inode *inode)
1362 {
1363         struct super_block *sb = inode->i_sb;
1364         struct gfs2_sbd *sdp = sb->s_fs_info;
1365         struct gfs2_inode *ip = GFS2_I(inode);
1366         struct address_space *metamapping;
1367         int ret;
1368
1369         gfs2_log_flush(sdp, ip->i_gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
1370                        GFS2_LFC_EVICT_INODE);
1371         metamapping = gfs2_glock2aspace(ip->i_gl);
1372         if (test_bit(GLF_DIRTY, &ip->i_gl->gl_flags)) {
1373                 filemap_fdatawrite(metamapping);
1374                 filemap_fdatawait(metamapping);
1375         }
1376         write_inode_now(inode, 1);
1377         gfs2_ail_flush(ip->i_gl, 0);
1378
1379         ret = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
1380         if (ret)
1381                 return ret;
1382
1383         /* Needs to be done before glock release & also in a transaction */
1384         truncate_inode_pages(&inode->i_data, 0);
1385         truncate_inode_pages(metamapping, 0);
1386         gfs2_trans_end(sdp);
1387         return 0;
1388 }
1389
1390 /**
1391  * gfs2_evict_inode - Remove an inode from cache
1392  * @inode: The inode to evict
1393  *
1394  * There are three cases to consider:
1395  * 1. i_nlink == 0, we are final opener (and must deallocate)
1396  * 2. i_nlink == 0, we are not the final opener (and cannot deallocate)
1397  * 3. i_nlink > 0
1398  *
1399  * If the fs is read only, then we have to treat all cases as per #3
1400  * since we are unable to do any deallocation. The inode will be
1401  * deallocated by the next read/write node to attempt an allocation
1402  * in the same resource group
1403  *
1404  * We have to (at the moment) hold the inodes main lock to cover
1405  * the gap between unlocking the shared lock on the iopen lock and
1406  * taking the exclusive lock. I'd rather do a shared -> exclusive
1407  * conversion on the iopen lock, but we can change that later. This
1408  * is safe, just less efficient.
1409  */
1410
1411 static void gfs2_evict_inode(struct inode *inode)
1412 {
1413         struct super_block *sb = inode->i_sb;
1414         struct gfs2_sbd *sdp = sb->s_fs_info;
1415         struct gfs2_inode *ip = GFS2_I(inode);
1416         struct gfs2_holder gh;
1417         int ret;
1418
1419         if (inode->i_nlink || sb_rdonly(sb) || !ip->i_no_addr)
1420                 goto out;
1421
1422         /*
1423          * In case of an incomplete mount, gfs2_evict_inode() may be called for
1424          * system files without having an active journal to write to.  In that
1425          * case, skip the filesystem evict.
1426          */
1427         if (!sdp->sd_jdesc)
1428                 goto out;
1429
1430         gfs2_holder_mark_uninitialized(&gh);
1431         ret = evict_should_delete(inode, &gh);
1432         if (ret == SHOULD_DEFER_EVICTION)
1433                 goto out;
1434         if (ret == SHOULD_DELETE_DINODE)
1435                 ret = evict_unlinked_inode(inode);
1436         else
1437                 ret = evict_linked_inode(inode);
1438
1439         if (gfs2_rs_active(&ip->i_res))
1440                 gfs2_rs_deltree(&ip->i_res);
1441
1442         if (gfs2_holder_initialized(&gh))
1443                 gfs2_glock_dq_uninit(&gh);
1444         if (ret && ret != GLR_TRYFAILED && ret != -EROFS)
1445                 fs_warn(sdp, "gfs2_evict_inode: %d\n", ret);
1446 out:
1447         truncate_inode_pages_final(&inode->i_data);
1448         if (ip->i_qadata)
1449                 gfs2_assert_warn(sdp, ip->i_qadata->qa_ref == 0);
1450         gfs2_rs_deltree(&ip->i_res);
1451         gfs2_ordered_del_inode(ip);
1452         clear_inode(inode);
1453         gfs2_dir_hash_inval(ip);
1454         if (gfs2_holder_initialized(&ip->i_iopen_gh)) {
1455                 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1456
1457                 glock_clear_object(gl, ip);
1458                 gfs2_glock_hold(gl);
1459                 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1460                 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1461                 gfs2_glock_put_eventually(gl);
1462         }
1463         if (ip->i_gl) {
1464                 glock_clear_object(ip->i_gl, ip);
1465                 wait_on_bit_io(&ip->i_flags, GIF_GLOP_PENDING, TASK_UNINTERRUPTIBLE);
1466                 gfs2_glock_add_to_lru(ip->i_gl);
1467                 gfs2_glock_put_eventually(ip->i_gl);
1468                 ip->i_gl = NULL;
1469         }
1470 }
1471
1472 static struct inode *gfs2_alloc_inode(struct super_block *sb)
1473 {
1474         struct gfs2_inode *ip;
1475
1476         ip = alloc_inode_sb(sb, gfs2_inode_cachep, GFP_KERNEL);
1477         if (!ip)
1478                 return NULL;
1479         ip->i_no_addr = 0;
1480         ip->i_flags = 0;
1481         ip->i_gl = NULL;
1482         gfs2_holder_mark_uninitialized(&ip->i_iopen_gh);
1483         memset(&ip->i_res, 0, sizeof(ip->i_res));
1484         RB_CLEAR_NODE(&ip->i_res.rs_node);
1485         ip->i_rahead = 0;
1486         return &ip->i_inode;
1487 }
1488
1489 static void gfs2_free_inode(struct inode *inode)
1490 {
1491         kmem_cache_free(gfs2_inode_cachep, GFS2_I(inode));
1492 }
1493
1494 extern void free_local_statfs_inodes(struct gfs2_sbd *sdp)
1495 {
1496         struct local_statfs_inode *lsi, *safe;
1497
1498         /* Run through the statfs inodes list to iput and free memory */
1499         list_for_each_entry_safe(lsi, safe, &sdp->sd_sc_inodes_list, si_list) {
1500                 if (lsi->si_jid == sdp->sd_jdesc->jd_jid)
1501                         sdp->sd_sc_inode = NULL; /* belongs to this node */
1502                 if (lsi->si_sc_inode)
1503                         iput(lsi->si_sc_inode);
1504                 list_del(&lsi->si_list);
1505                 kfree(lsi);
1506         }
1507 }
1508
1509 extern struct inode *find_local_statfs_inode(struct gfs2_sbd *sdp,
1510                                              unsigned int index)
1511 {
1512         struct local_statfs_inode *lsi;
1513
1514         /* Return the local (per node) statfs inode in the
1515          * sdp->sd_sc_inodes_list corresponding to the 'index'. */
1516         list_for_each_entry(lsi, &sdp->sd_sc_inodes_list, si_list) {
1517                 if (lsi->si_jid == index)
1518                         return lsi->si_sc_inode;
1519         }
1520         return NULL;
1521 }
1522
1523 const struct super_operations gfs2_super_ops = {
1524         .alloc_inode            = gfs2_alloc_inode,
1525         .free_inode             = gfs2_free_inode,
1526         .write_inode            = gfs2_write_inode,
1527         .dirty_inode            = gfs2_dirty_inode,
1528         .evict_inode            = gfs2_evict_inode,
1529         .put_super              = gfs2_put_super,
1530         .sync_fs                = gfs2_sync_fs,
1531         .freeze_super           = gfs2_freeze,
1532         .thaw_super             = gfs2_unfreeze,
1533         .statfs                 = gfs2_statfs,
1534         .drop_inode             = gfs2_drop_inode,
1535         .show_options           = gfs2_show_options,
1536 };
1537