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 evict_behavior { 48 EVICT_SHOULD_DELETE, 49 EVICT_SHOULD_SKIP_DELETE, 50 EVICT_SHOULD_DEFER_DELETE, 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 down_write(&sdp->sd_log_flush_lock); 71 sdp->sd_jdesc = NULL; 72 up_write(&sdp->sd_log_flush_lock); 73 74 while (!list_empty(&list)) { 75 jd = list_first_entry(&list, struct gfs2_jdesc, jd_list); 76 BUG_ON(jd->jd_log_bio); 77 gfs2_free_journal_extents(jd); 78 list_del(&jd->jd_list); 79 iput(jd->jd_inode); 80 jd->jd_inode = NULL; 81 kfree(jd); 82 } 83 } 84 85 static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid) 86 { 87 struct gfs2_jdesc *jd; 88 89 list_for_each_entry(jd, head, jd_list) { 90 if (jd->jd_jid == jid) 91 return jd; 92 } 93 return NULL; 94 } 95 96 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid) 97 { 98 struct gfs2_jdesc *jd; 99 100 spin_lock(&sdp->sd_jindex_spin); 101 jd = jdesc_find_i(&sdp->sd_jindex_list, jid); 102 spin_unlock(&sdp->sd_jindex_spin); 103 104 return jd; 105 } 106 107 int gfs2_jdesc_check(struct gfs2_jdesc *jd) 108 { 109 struct gfs2_inode *ip = GFS2_I(jd->jd_inode); 110 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode); 111 u64 size = i_size_read(jd->jd_inode); 112 113 if (gfs2_check_internal_file_size(jd->jd_inode, 8 << 20, BIT(30))) 114 return -EIO; 115 116 jd->jd_blocks = size >> sdp->sd_sb.sb_bsize_shift; 117 118 if (gfs2_write_alloc_required(ip, 0, size)) { 119 gfs2_consist_inode(ip); 120 return -EIO; 121 } 122 123 return 0; 124 } 125 126 /** 127 * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one 128 * @sdp: the filesystem 129 * 130 * Returns: errno 131 */ 132 133 int gfs2_make_fs_rw(struct gfs2_sbd *sdp) 134 { 135 struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode); 136 struct gfs2_glock *j_gl = ip->i_gl; 137 int error; 138 139 j_gl->gl_ops->go_inval(j_gl, DIO_METADATA); 140 if (gfs2_withdrawing_or_withdrawn(sdp)) 141 return -EIO; 142 143 if (sdp->sd_log_sequence == 0) { 144 fs_err(sdp, "unknown status of our own journal jid %d", 145 sdp->sd_lockstruct.ls_jid); 146 return -EIO; 147 } 148 149 error = gfs2_quota_init(sdp); 150 if (!error && gfs2_withdrawing_or_withdrawn(sdp)) 151 error = -EIO; 152 if (!error) 153 set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags); 154 return error; 155 } 156 157 void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf) 158 { 159 const struct gfs2_statfs_change *str = buf; 160 161 sc->sc_total = be64_to_cpu(str->sc_total); 162 sc->sc_free = be64_to_cpu(str->sc_free); 163 sc->sc_dinodes = be64_to_cpu(str->sc_dinodes); 164 } 165 166 void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf) 167 { 168 struct gfs2_statfs_change *str = buf; 169 170 str->sc_total = cpu_to_be64(sc->sc_total); 171 str->sc_free = cpu_to_be64(sc->sc_free); 172 str->sc_dinodes = cpu_to_be64(sc->sc_dinodes); 173 } 174 175 int gfs2_statfs_init(struct gfs2_sbd *sdp) 176 { 177 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode); 178 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master; 179 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local; 180 struct buffer_head *m_bh; 181 struct gfs2_holder gh; 182 int error; 183 184 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE, 185 &gh); 186 if (error) 187 return error; 188 189 error = gfs2_meta_inode_buffer(m_ip, &m_bh); 190 if (error) 191 goto out; 192 193 if (sdp->sd_args.ar_spectator) { 194 spin_lock(&sdp->sd_statfs_spin); 195 gfs2_statfs_change_in(m_sc, m_bh->b_data + 196 sizeof(struct gfs2_dinode)); 197 spin_unlock(&sdp->sd_statfs_spin); 198 } else { 199 spin_lock(&sdp->sd_statfs_spin); 200 gfs2_statfs_change_in(m_sc, m_bh->b_data + 201 sizeof(struct gfs2_dinode)); 202 gfs2_statfs_change_in(l_sc, sdp->sd_sc_bh->b_data + 203 sizeof(struct gfs2_dinode)); 204 spin_unlock(&sdp->sd_statfs_spin); 205 206 } 207 208 brelse(m_bh); 209 out: 210 gfs2_glock_dq_uninit(&gh); 211 return 0; 212 } 213 214 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free, 215 s64 dinodes) 216 { 217 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode); 218 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local; 219 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master; 220 s64 x, y; 221 int need_sync = 0; 222 223 gfs2_trans_add_meta(l_ip->i_gl, sdp->sd_sc_bh); 224 225 spin_lock(&sdp->sd_statfs_spin); 226 l_sc->sc_total += total; 227 l_sc->sc_free += free; 228 l_sc->sc_dinodes += dinodes; 229 gfs2_statfs_change_out(l_sc, sdp->sd_sc_bh->b_data + 230 sizeof(struct gfs2_dinode)); 231 if (sdp->sd_args.ar_statfs_percent) { 232 x = 100 * l_sc->sc_free; 233 y = m_sc->sc_free * sdp->sd_args.ar_statfs_percent; 234 if (x >= y || x <= -y) 235 need_sync = 1; 236 } 237 spin_unlock(&sdp->sd_statfs_spin); 238 239 if (need_sync) 240 gfs2_wake_up_statfs(sdp); 241 } 242 243 void update_statfs(struct gfs2_sbd *sdp, struct buffer_head *m_bh) 244 { 245 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode); 246 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode); 247 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master; 248 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local; 249 250 gfs2_trans_add_meta(l_ip->i_gl, sdp->sd_sc_bh); 251 gfs2_trans_add_meta(m_ip->i_gl, m_bh); 252 253 spin_lock(&sdp->sd_statfs_spin); 254 m_sc->sc_total += l_sc->sc_total; 255 m_sc->sc_free += l_sc->sc_free; 256 m_sc->sc_dinodes += l_sc->sc_dinodes; 257 memset(l_sc, 0, sizeof(struct gfs2_statfs_change)); 258 memset(sdp->sd_sc_bh->b_data + sizeof(struct gfs2_dinode), 259 0, sizeof(struct gfs2_statfs_change)); 260 gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode)); 261 spin_unlock(&sdp->sd_statfs_spin); 262 } 263 264 int gfs2_statfs_sync(struct super_block *sb, int type) 265 { 266 struct gfs2_sbd *sdp = sb->s_fs_info; 267 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode); 268 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master; 269 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local; 270 struct gfs2_holder gh; 271 struct buffer_head *m_bh; 272 int error; 273 274 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE, 275 &gh); 276 if (error) 277 goto out; 278 279 error = gfs2_meta_inode_buffer(m_ip, &m_bh); 280 if (error) 281 goto out_unlock; 282 283 spin_lock(&sdp->sd_statfs_spin); 284 gfs2_statfs_change_in(m_sc, m_bh->b_data + 285 sizeof(struct gfs2_dinode)); 286 if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) { 287 spin_unlock(&sdp->sd_statfs_spin); 288 goto out_bh; 289 } 290 spin_unlock(&sdp->sd_statfs_spin); 291 292 error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0); 293 if (error) 294 goto out_bh; 295 296 update_statfs(sdp, m_bh); 297 sdp->sd_statfs_force_sync = 0; 298 299 gfs2_trans_end(sdp); 300 301 out_bh: 302 brelse(m_bh); 303 out_unlock: 304 gfs2_glock_dq_uninit(&gh); 305 out: 306 return error; 307 } 308 309 struct lfcc { 310 struct list_head list; 311 struct gfs2_holder gh; 312 }; 313 314 /** 315 * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all 316 * journals are clean 317 * @sdp: the file system 318 * 319 * Returns: errno 320 */ 321 322 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp) 323 { 324 struct gfs2_inode *ip; 325 struct gfs2_jdesc *jd; 326 struct lfcc *lfcc; 327 LIST_HEAD(list); 328 struct gfs2_log_header_host lh; 329 int error, error2; 330 331 /* 332 * Grab all the journal glocks in SH mode. We are *probably* doing 333 * that to prevent recovery. 334 */ 335 336 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) { 337 lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL); 338 if (!lfcc) { 339 error = -ENOMEM; 340 goto out; 341 } 342 ip = GFS2_I(jd->jd_inode); 343 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh); 344 if (error) { 345 kfree(lfcc); 346 goto out; 347 } 348 list_add(&lfcc->list, &list); 349 } 350 351 gfs2_freeze_unlock(sdp); 352 353 error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_EXCLUSIVE, 354 LM_FLAG_NOEXP | GL_NOPID, 355 &sdp->sd_freeze_gh); 356 if (error) 357 goto relock_shared; 358 359 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) { 360 error = gfs2_jdesc_check(jd); 361 if (error) 362 break; 363 error = gfs2_find_jhead(jd, &lh); 364 if (error) 365 break; 366 if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) { 367 error = -EBUSY; 368 break; 369 } 370 } 371 372 if (!error) 373 goto out; /* success */ 374 375 gfs2_freeze_unlock(sdp); 376 377 relock_shared: 378 error2 = gfs2_freeze_lock_shared(sdp); 379 gfs2_assert_withdraw(sdp, !error2); 380 381 out: 382 while (!list_empty(&list)) { 383 lfcc = list_first_entry(&list, struct lfcc, list); 384 list_del(&lfcc->list); 385 gfs2_glock_dq_uninit(&lfcc->gh); 386 kfree(lfcc); 387 } 388 return error; 389 } 390 391 void gfs2_dinode_out(const struct gfs2_inode *ip, void *buf) 392 { 393 const struct inode *inode = &ip->i_inode; 394 struct gfs2_dinode *str = buf; 395 396 str->di_header.mh_magic = cpu_to_be32(GFS2_MAGIC); 397 str->di_header.mh_type = cpu_to_be32(GFS2_METATYPE_DI); 398 str->di_header.mh_format = cpu_to_be32(GFS2_FORMAT_DI); 399 str->di_num.no_addr = cpu_to_be64(ip->i_no_addr); 400 str->di_num.no_formal_ino = cpu_to_be64(ip->i_no_formal_ino); 401 str->di_mode = cpu_to_be32(inode->i_mode); 402 str->di_uid = cpu_to_be32(i_uid_read(inode)); 403 str->di_gid = cpu_to_be32(i_gid_read(inode)); 404 str->di_nlink = cpu_to_be32(inode->i_nlink); 405 str->di_size = cpu_to_be64(i_size_read(inode)); 406 str->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(inode)); 407 str->di_atime = cpu_to_be64(inode_get_atime_sec(inode)); 408 str->di_mtime = cpu_to_be64(inode_get_mtime_sec(inode)); 409 str->di_ctime = cpu_to_be64(inode_get_ctime_sec(inode)); 410 411 str->di_goal_meta = cpu_to_be64(ip->i_goal); 412 str->di_goal_data = cpu_to_be64(ip->i_goal); 413 str->di_generation = cpu_to_be64(ip->i_generation); 414 415 str->di_flags = cpu_to_be32(ip->i_diskflags); 416 str->di_height = cpu_to_be16(ip->i_height); 417 str->di_payload_format = cpu_to_be32(S_ISDIR(inode->i_mode) && 418 !(ip->i_diskflags & GFS2_DIF_EXHASH) ? 419 GFS2_FORMAT_DE : 0); 420 str->di_depth = cpu_to_be16(ip->i_depth); 421 str->di_entries = cpu_to_be32(ip->i_entries); 422 423 str->di_eattr = cpu_to_be64(ip->i_eattr); 424 str->di_atime_nsec = cpu_to_be32(inode_get_atime_nsec(inode)); 425 str->di_mtime_nsec = cpu_to_be32(inode_get_mtime_nsec(inode)); 426 str->di_ctime_nsec = cpu_to_be32(inode_get_ctime_nsec(inode)); 427 } 428 429 /** 430 * gfs2_write_inode - Make sure the inode is stable on the disk 431 * @inode: The inode 432 * @wbc: The writeback control structure 433 * 434 * Returns: errno 435 */ 436 437 static int gfs2_write_inode(struct inode *inode, struct writeback_control *wbc) 438 { 439 struct gfs2_inode *ip = GFS2_I(inode); 440 struct gfs2_sbd *sdp = GFS2_SB(inode); 441 struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl); 442 struct backing_dev_info *bdi = inode_to_bdi(metamapping->host); 443 int ret = 0; 444 bool flush_all = (wbc->sync_mode == WB_SYNC_ALL || gfs2_is_jdata(ip)); 445 446 if (flush_all) 447 gfs2_log_flush(GFS2_SB(inode), ip->i_gl, 448 GFS2_LOG_HEAD_FLUSH_NORMAL | 449 GFS2_LFC_WRITE_INODE); 450 if (bdi->wb.dirty_exceeded) 451 gfs2_ail1_flush(sdp, wbc); 452 else 453 filemap_fdatawrite(metamapping); 454 if (flush_all) 455 ret = filemap_fdatawait(metamapping); 456 if (ret) 457 mark_inode_dirty_sync(inode); 458 else { 459 spin_lock(&inode->i_lock); 460 if (!(inode->i_flags & I_DIRTY)) 461 gfs2_ordered_del_inode(ip); 462 spin_unlock(&inode->i_lock); 463 } 464 return ret; 465 } 466 467 /** 468 * gfs2_dirty_inode - check for atime updates 469 * @inode: The inode in question 470 * @flags: The type of dirty 471 * 472 * Unfortunately it can be called under any combination of inode 473 * glock and freeze glock, so we have to check carefully. 474 * 475 * At the moment this deals only with atime - it should be possible 476 * to expand that role in future, once a review of the locking has 477 * been carried out. 478 */ 479 480 static void gfs2_dirty_inode(struct inode *inode, int flags) 481 { 482 struct gfs2_inode *ip = GFS2_I(inode); 483 struct gfs2_sbd *sdp = GFS2_SB(inode); 484 struct buffer_head *bh; 485 struct gfs2_holder gh; 486 int need_unlock = 0; 487 int need_endtrans = 0; 488 int ret; 489 490 /* This can only happen during incomplete inode creation. */ 491 if (unlikely(!ip->i_gl)) 492 return; 493 494 if (gfs2_withdrawing_or_withdrawn(sdp)) 495 return; 496 if (!gfs2_glock_is_locked_by_me(ip->i_gl)) { 497 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh); 498 if (ret) { 499 fs_err(sdp, "dirty_inode: glock %d\n", ret); 500 gfs2_dump_glock(NULL, ip->i_gl, true); 501 return; 502 } 503 need_unlock = 1; 504 } else if (WARN_ON_ONCE(ip->i_gl->gl_state != LM_ST_EXCLUSIVE)) 505 return; 506 507 if (current->journal_info == NULL) { 508 ret = gfs2_trans_begin(sdp, RES_DINODE, 0); 509 if (ret) { 510 fs_err(sdp, "dirty_inode: gfs2_trans_begin %d\n", ret); 511 goto out; 512 } 513 need_endtrans = 1; 514 } 515 516 ret = gfs2_meta_inode_buffer(ip, &bh); 517 if (ret == 0) { 518 gfs2_trans_add_meta(ip->i_gl, bh); 519 gfs2_dinode_out(ip, bh->b_data); 520 brelse(bh); 521 } 522 523 if (need_endtrans) 524 gfs2_trans_end(sdp); 525 out: 526 if (need_unlock) 527 gfs2_glock_dq_uninit(&gh); 528 } 529 530 /** 531 * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one 532 * @sdp: the filesystem 533 * 534 * Returns: errno 535 */ 536 537 void gfs2_make_fs_ro(struct gfs2_sbd *sdp) 538 { 539 int log_write_allowed = test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags); 540 541 if (!test_bit(SDF_KILL, &sdp->sd_flags)) 542 gfs2_flush_delete_work(sdp); 543 544 gfs2_destroy_threads(sdp); 545 546 if (log_write_allowed) { 547 gfs2_quota_sync(sdp->sd_vfs, 0); 548 gfs2_statfs_sync(sdp->sd_vfs, 0); 549 550 /* We do two log flushes here. The first one commits dirty inodes 551 * and rgrps to the journal, but queues up revokes to the ail list. 552 * The second flush writes out and removes the revokes. 553 * 554 * The first must be done before the FLUSH_SHUTDOWN code 555 * clears the LIVE flag, otherwise it will not be able to start 556 * a transaction to write its revokes, and the error will cause 557 * a withdraw of the file system. */ 558 gfs2_log_flush(sdp, NULL, GFS2_LFC_MAKE_FS_RO); 559 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SHUTDOWN | 560 GFS2_LFC_MAKE_FS_RO); 561 wait_event_timeout(sdp->sd_log_waitq, 562 gfs2_log_is_empty(sdp), 563 HZ * 5); 564 gfs2_assert_warn(sdp, gfs2_log_is_empty(sdp)); 565 } 566 gfs2_quota_cleanup(sdp); 567 } 568 569 /** 570 * gfs2_put_super - Unmount the filesystem 571 * @sb: The VFS superblock 572 * 573 */ 574 575 static void gfs2_put_super(struct super_block *sb) 576 { 577 struct gfs2_sbd *sdp = sb->s_fs_info; 578 struct gfs2_jdesc *jd; 579 580 /* No more recovery requests */ 581 set_bit(SDF_NORECOVERY, &sdp->sd_flags); 582 smp_mb(); 583 584 /* Wait on outstanding recovery */ 585 restart: 586 spin_lock(&sdp->sd_jindex_spin); 587 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) { 588 if (!test_bit(JDF_RECOVERY, &jd->jd_flags)) 589 continue; 590 spin_unlock(&sdp->sd_jindex_spin); 591 wait_on_bit(&jd->jd_flags, JDF_RECOVERY, 592 TASK_UNINTERRUPTIBLE); 593 goto restart; 594 } 595 spin_unlock(&sdp->sd_jindex_spin); 596 597 if (!sb_rdonly(sb)) 598 gfs2_make_fs_ro(sdp); 599 else { 600 if (gfs2_withdrawing_or_withdrawn(sdp)) 601 gfs2_destroy_threads(sdp); 602 603 gfs2_quota_cleanup(sdp); 604 } 605 606 WARN_ON(gfs2_withdrawing(sdp)); 607 608 /* At this point, we're through modifying the disk */ 609 610 /* Release stuff */ 611 612 gfs2_freeze_unlock(sdp); 613 614 iput(sdp->sd_jindex); 615 iput(sdp->sd_statfs_inode); 616 iput(sdp->sd_rindex); 617 iput(sdp->sd_quota_inode); 618 619 gfs2_glock_put(sdp->sd_rename_gl); 620 gfs2_glock_put(sdp->sd_freeze_gl); 621 622 if (!sdp->sd_args.ar_spectator) { 623 if (gfs2_holder_initialized(&sdp->sd_journal_gh)) 624 gfs2_glock_dq_uninit(&sdp->sd_journal_gh); 625 if (gfs2_holder_initialized(&sdp->sd_jinode_gh)) 626 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh); 627 brelse(sdp->sd_sc_bh); 628 gfs2_glock_dq_uninit(&sdp->sd_sc_gh); 629 gfs2_glock_dq_uninit(&sdp->sd_qc_gh); 630 free_local_statfs_inodes(sdp); 631 iput(sdp->sd_qc_inode); 632 } 633 634 gfs2_glock_dq_uninit(&sdp->sd_live_gh); 635 gfs2_clear_rgrpd(sdp); 636 gfs2_jindex_free(sdp); 637 /* Take apart glock structures and buffer lists */ 638 gfs2_gl_hash_clear(sdp); 639 iput(sdp->sd_inode); 640 gfs2_delete_debugfs_file(sdp); 641 642 gfs2_sys_fs_del(sdp); 643 free_sbd(sdp); 644 } 645 646 /** 647 * gfs2_sync_fs - sync the filesystem 648 * @sb: the superblock 649 * @wait: true to wait for completion 650 * 651 * Flushes the log to disk. 652 */ 653 654 static int gfs2_sync_fs(struct super_block *sb, int wait) 655 { 656 struct gfs2_sbd *sdp = sb->s_fs_info; 657 658 gfs2_quota_sync(sb, -1); 659 if (wait) 660 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL | 661 GFS2_LFC_SYNC_FS); 662 return sdp->sd_log_error; 663 } 664 665 static int gfs2_do_thaw(struct gfs2_sbd *sdp, enum freeze_holder who, const void *freeze_owner) 666 { 667 struct super_block *sb = sdp->sd_vfs; 668 int error; 669 670 error = gfs2_freeze_lock_shared(sdp); 671 if (error) 672 goto fail; 673 error = thaw_super(sb, who, freeze_owner); 674 if (!error) 675 return 0; 676 677 fail: 678 fs_info(sdp, "GFS2: couldn't thaw filesystem: %d\n", error); 679 gfs2_assert_withdraw(sdp, 0); 680 return error; 681 } 682 683 void gfs2_freeze_func(struct work_struct *work) 684 { 685 struct gfs2_sbd *sdp = container_of(work, struct gfs2_sbd, sd_freeze_work); 686 struct super_block *sb = sdp->sd_vfs; 687 int error; 688 689 mutex_lock(&sdp->sd_freeze_mutex); 690 error = -EBUSY; 691 if (test_bit(SDF_FROZEN, &sdp->sd_flags)) 692 goto freeze_failed; 693 694 error = freeze_super(sb, FREEZE_HOLDER_USERSPACE, NULL); 695 if (error) 696 goto freeze_failed; 697 698 gfs2_freeze_unlock(sdp); 699 set_bit(SDF_FROZEN, &sdp->sd_flags); 700 701 error = gfs2_do_thaw(sdp, FREEZE_HOLDER_USERSPACE, NULL); 702 if (error) 703 goto out; 704 705 clear_bit(SDF_FROZEN, &sdp->sd_flags); 706 goto out; 707 708 freeze_failed: 709 fs_info(sdp, "GFS2: couldn't freeze filesystem: %d\n", error); 710 711 out: 712 mutex_unlock(&sdp->sd_freeze_mutex); 713 deactivate_super(sb); 714 } 715 716 /** 717 * gfs2_freeze_super - prevent further writes to the filesystem 718 * @sb: the VFS structure for the filesystem 719 * @who: freeze flags 720 * @freeze_owner: owner of the freeze 721 * 722 */ 723 724 static int gfs2_freeze_super(struct super_block *sb, enum freeze_holder who, 725 const void *freeze_owner) 726 { 727 struct gfs2_sbd *sdp = sb->s_fs_info; 728 int error; 729 730 if (!mutex_trylock(&sdp->sd_freeze_mutex)) 731 return -EBUSY; 732 if (test_bit(SDF_FROZEN, &sdp->sd_flags)) { 733 mutex_unlock(&sdp->sd_freeze_mutex); 734 return -EBUSY; 735 } 736 737 for (;;) { 738 error = freeze_super(sb, who, freeze_owner); 739 if (error) { 740 fs_info(sdp, "GFS2: couldn't freeze filesystem: %d\n", 741 error); 742 goto out; 743 } 744 745 error = gfs2_lock_fs_check_clean(sdp); 746 if (!error) { 747 set_bit(SDF_FREEZE_INITIATOR, &sdp->sd_flags); 748 set_bit(SDF_FROZEN, &sdp->sd_flags); 749 break; 750 } 751 752 error = gfs2_do_thaw(sdp, who, freeze_owner); 753 if (error) 754 goto out; 755 756 if (error == -EBUSY) 757 fs_err(sdp, "waiting for recovery before freeze\n"); 758 else if (error == -EIO) { 759 fs_err(sdp, "Fatal IO error: cannot freeze gfs2 due " 760 "to recovery error.\n"); 761 goto out; 762 } else { 763 fs_err(sdp, "error freezing FS: %d\n", error); 764 } 765 fs_err(sdp, "retrying...\n"); 766 msleep(1000); 767 } 768 769 out: 770 mutex_unlock(&sdp->sd_freeze_mutex); 771 return error; 772 } 773 774 static int gfs2_freeze_fs(struct super_block *sb) 775 { 776 struct gfs2_sbd *sdp = sb->s_fs_info; 777 778 if (test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) { 779 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_FREEZE | 780 GFS2_LFC_FREEZE_GO_SYNC); 781 if (gfs2_withdrawing_or_withdrawn(sdp)) 782 return -EIO; 783 } 784 return 0; 785 } 786 787 /** 788 * gfs2_thaw_super - reallow writes to the filesystem 789 * @sb: the VFS structure for the filesystem 790 * @who: freeze flags 791 * @freeze_owner: owner of the freeze 792 * 793 */ 794 795 static int gfs2_thaw_super(struct super_block *sb, enum freeze_holder who, 796 const void *freeze_owner) 797 { 798 struct gfs2_sbd *sdp = sb->s_fs_info; 799 int error; 800 801 if (!mutex_trylock(&sdp->sd_freeze_mutex)) 802 return -EBUSY; 803 if (!test_bit(SDF_FREEZE_INITIATOR, &sdp->sd_flags)) { 804 mutex_unlock(&sdp->sd_freeze_mutex); 805 return -EINVAL; 806 } 807 808 atomic_inc(&sb->s_active); 809 gfs2_freeze_unlock(sdp); 810 811 error = gfs2_do_thaw(sdp, who, freeze_owner); 812 813 if (!error) { 814 clear_bit(SDF_FREEZE_INITIATOR, &sdp->sd_flags); 815 clear_bit(SDF_FROZEN, &sdp->sd_flags); 816 } 817 mutex_unlock(&sdp->sd_freeze_mutex); 818 deactivate_super(sb); 819 return error; 820 } 821 822 void gfs2_thaw_freeze_initiator(struct super_block *sb) 823 { 824 struct gfs2_sbd *sdp = sb->s_fs_info; 825 826 mutex_lock(&sdp->sd_freeze_mutex); 827 if (!test_bit(SDF_FREEZE_INITIATOR, &sdp->sd_flags)) 828 goto out; 829 830 gfs2_freeze_unlock(sdp); 831 832 out: 833 mutex_unlock(&sdp->sd_freeze_mutex); 834 } 835 836 /** 837 * statfs_slow_fill - fill in the sg for a given RG 838 * @rgd: the RG 839 * @sc: the sc structure 840 * 841 * Returns: 0 on success, -ESTALE if the LVB is invalid 842 */ 843 844 static int statfs_slow_fill(struct gfs2_rgrpd *rgd, 845 struct gfs2_statfs_change_host *sc) 846 { 847 gfs2_rgrp_verify(rgd); 848 sc->sc_total += rgd->rd_data; 849 sc->sc_free += rgd->rd_free; 850 sc->sc_dinodes += rgd->rd_dinodes; 851 return 0; 852 } 853 854 /** 855 * gfs2_statfs_slow - Stat a filesystem using asynchronous locking 856 * @sdp: the filesystem 857 * @sc: the sc info that will be returned 858 * 859 * Any error (other than a signal) will cause this routine to fall back 860 * to the synchronous version. 861 * 862 * FIXME: This really shouldn't busy wait like this. 863 * 864 * Returns: errno 865 */ 866 867 static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc) 868 { 869 struct gfs2_rgrpd *rgd_next; 870 struct gfs2_holder *gha, *gh; 871 unsigned int slots = 64; 872 unsigned int x; 873 int done; 874 int error = 0, err; 875 876 memset(sc, 0, sizeof(struct gfs2_statfs_change_host)); 877 gha = kmalloc_array(slots, sizeof(struct gfs2_holder), GFP_KERNEL); 878 if (!gha) 879 return -ENOMEM; 880 for (x = 0; x < slots; x++) 881 gfs2_holder_mark_uninitialized(gha + x); 882 883 rgd_next = gfs2_rgrpd_get_first(sdp); 884 885 for (;;) { 886 done = 1; 887 888 for (x = 0; x < slots; x++) { 889 gh = gha + x; 890 891 if (gfs2_holder_initialized(gh) && gfs2_glock_poll(gh)) { 892 err = gfs2_glock_wait(gh); 893 if (err) { 894 gfs2_holder_uninit(gh); 895 error = err; 896 } else { 897 if (!error) { 898 struct gfs2_rgrpd *rgd = 899 gfs2_glock2rgrp(gh->gh_gl); 900 901 error = statfs_slow_fill(rgd, sc); 902 } 903 gfs2_glock_dq_uninit(gh); 904 } 905 } 906 907 if (gfs2_holder_initialized(gh)) 908 done = 0; 909 else if (rgd_next && !error) { 910 error = gfs2_glock_nq_init(rgd_next->rd_gl, 911 LM_ST_SHARED, 912 GL_ASYNC, 913 gh); 914 rgd_next = gfs2_rgrpd_get_next(rgd_next); 915 done = 0; 916 } 917 918 if (signal_pending(current)) 919 error = -ERESTARTSYS; 920 } 921 922 if (done) 923 break; 924 925 yield(); 926 } 927 928 kfree(gha); 929 return error; 930 } 931 932 /** 933 * gfs2_statfs_i - Do a statfs 934 * @sdp: the filesystem 935 * @sc: the sc structure 936 * 937 * Returns: errno 938 */ 939 940 static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc) 941 { 942 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master; 943 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local; 944 945 spin_lock(&sdp->sd_statfs_spin); 946 947 *sc = *m_sc; 948 sc->sc_total += l_sc->sc_total; 949 sc->sc_free += l_sc->sc_free; 950 sc->sc_dinodes += l_sc->sc_dinodes; 951 952 spin_unlock(&sdp->sd_statfs_spin); 953 954 if (sc->sc_free < 0) 955 sc->sc_free = 0; 956 if (sc->sc_free > sc->sc_total) 957 sc->sc_free = sc->sc_total; 958 if (sc->sc_dinodes < 0) 959 sc->sc_dinodes = 0; 960 961 return 0; 962 } 963 964 /** 965 * gfs2_statfs - Gather and return stats about the filesystem 966 * @dentry: The name of the link 967 * @buf: The buffer 968 * 969 * Returns: 0 on success or error code 970 */ 971 972 static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf) 973 { 974 struct super_block *sb = dentry->d_sb; 975 struct gfs2_sbd *sdp = sb->s_fs_info; 976 struct gfs2_statfs_change_host sc; 977 int error; 978 979 error = gfs2_rindex_update(sdp); 980 if (error) 981 return error; 982 983 if (gfs2_tune_get(sdp, gt_statfs_slow)) 984 error = gfs2_statfs_slow(sdp, &sc); 985 else 986 error = gfs2_statfs_i(sdp, &sc); 987 988 if (error) 989 return error; 990 991 buf->f_type = GFS2_MAGIC; 992 buf->f_bsize = sdp->sd_sb.sb_bsize; 993 buf->f_blocks = sc.sc_total; 994 buf->f_bfree = sc.sc_free; 995 buf->f_bavail = sc.sc_free; 996 buf->f_files = sc.sc_dinodes + sc.sc_free; 997 buf->f_ffree = sc.sc_free; 998 buf->f_namelen = GFS2_FNAMESIZE; 999 buf->f_fsid = uuid_to_fsid(sb->s_uuid.b); 1000 1001 return 0; 1002 } 1003 1004 /** 1005 * gfs2_drop_inode - Drop an inode (test for remote unlink) 1006 * @inode: The inode to drop 1007 * 1008 * If we've received a callback on an iopen lock then it's because a 1009 * remote node tried to deallocate the inode but failed due to this node 1010 * still having the inode open. Here we mark the link count zero 1011 * since we know that it must have reached zero if the GLF_DEMOTE flag 1012 * is set on the iopen glock. If we didn't do a disk read since the 1013 * remote node removed the final link then we might otherwise miss 1014 * this event. This check ensures that this node will deallocate the 1015 * inode's blocks, or alternatively pass the baton on to another 1016 * node for later deallocation. 1017 */ 1018 1019 static int gfs2_drop_inode(struct inode *inode) 1020 { 1021 struct gfs2_inode *ip = GFS2_I(inode); 1022 struct gfs2_sbd *sdp = GFS2_SB(inode); 1023 1024 if (inode->i_nlink && 1025 gfs2_holder_initialized(&ip->i_iopen_gh)) { 1026 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl; 1027 if (glock_needs_demote(gl)) 1028 clear_nlink(inode); 1029 } 1030 1031 /* 1032 * When under memory pressure when an inode's link count has dropped to 1033 * zero, defer deleting the inode to the delete workqueue. This avoids 1034 * calling into DLM under memory pressure, which can deadlock. 1035 */ 1036 if (!inode->i_nlink && 1037 unlikely(current->flags & PF_MEMALLOC) && 1038 gfs2_holder_initialized(&ip->i_iopen_gh)) { 1039 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl; 1040 1041 gfs2_glock_hold(gl); 1042 if (!gfs2_queue_verify_delete(gl, true)) 1043 gfs2_glock_put_async(gl); 1044 return 0; 1045 } 1046 1047 /* 1048 * No longer cache inodes when trying to evict them all. 1049 */ 1050 if (test_bit(SDF_EVICTING, &sdp->sd_flags)) 1051 return 1; 1052 1053 return generic_drop_inode(inode); 1054 } 1055 1056 /** 1057 * gfs2_show_options - Show mount options for /proc/mounts 1058 * @s: seq_file structure 1059 * @root: root of this (sub)tree 1060 * 1061 * Returns: 0 on success or error code 1062 */ 1063 1064 static int gfs2_show_options(struct seq_file *s, struct dentry *root) 1065 { 1066 struct gfs2_sbd *sdp = root->d_sb->s_fs_info; 1067 struct gfs2_args *args = &sdp->sd_args; 1068 unsigned int logd_secs, statfs_slow, statfs_quantum, quota_quantum; 1069 1070 spin_lock(&sdp->sd_tune.gt_spin); 1071 logd_secs = sdp->sd_tune.gt_logd_secs; 1072 quota_quantum = sdp->sd_tune.gt_quota_quantum; 1073 statfs_quantum = sdp->sd_tune.gt_statfs_quantum; 1074 statfs_slow = sdp->sd_tune.gt_statfs_slow; 1075 spin_unlock(&sdp->sd_tune.gt_spin); 1076 1077 if (is_subdir(root, sdp->sd_master_dir)) 1078 seq_puts(s, ",meta"); 1079 if (args->ar_lockproto[0]) 1080 seq_show_option(s, "lockproto", args->ar_lockproto); 1081 if (args->ar_locktable[0]) 1082 seq_show_option(s, "locktable", args->ar_locktable); 1083 if (args->ar_hostdata[0]) 1084 seq_show_option(s, "hostdata", args->ar_hostdata); 1085 if (args->ar_spectator) 1086 seq_puts(s, ",spectator"); 1087 if (args->ar_localflocks) 1088 seq_puts(s, ",localflocks"); 1089 if (args->ar_debug) 1090 seq_puts(s, ",debug"); 1091 if (args->ar_posix_acl) 1092 seq_puts(s, ",acl"); 1093 if (args->ar_quota != GFS2_QUOTA_DEFAULT) { 1094 char *state; 1095 switch (args->ar_quota) { 1096 case GFS2_QUOTA_OFF: 1097 state = "off"; 1098 break; 1099 case GFS2_QUOTA_ACCOUNT: 1100 state = "account"; 1101 break; 1102 case GFS2_QUOTA_ON: 1103 state = "on"; 1104 break; 1105 case GFS2_QUOTA_QUIET: 1106 state = "quiet"; 1107 break; 1108 default: 1109 state = "unknown"; 1110 break; 1111 } 1112 seq_printf(s, ",quota=%s", state); 1113 } 1114 if (args->ar_suiddir) 1115 seq_puts(s, ",suiddir"); 1116 if (args->ar_data != GFS2_DATA_DEFAULT) { 1117 char *state; 1118 switch (args->ar_data) { 1119 case GFS2_DATA_WRITEBACK: 1120 state = "writeback"; 1121 break; 1122 case GFS2_DATA_ORDERED: 1123 state = "ordered"; 1124 break; 1125 default: 1126 state = "unknown"; 1127 break; 1128 } 1129 seq_printf(s, ",data=%s", state); 1130 } 1131 if (args->ar_discard) 1132 seq_puts(s, ",discard"); 1133 if (logd_secs != 30) 1134 seq_printf(s, ",commit=%d", logd_secs); 1135 if (statfs_quantum != 30) 1136 seq_printf(s, ",statfs_quantum=%d", statfs_quantum); 1137 else if (statfs_slow) 1138 seq_puts(s, ",statfs_quantum=0"); 1139 if (quota_quantum != 60) 1140 seq_printf(s, ",quota_quantum=%d", quota_quantum); 1141 if (args->ar_statfs_percent) 1142 seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent); 1143 if (args->ar_errors != GFS2_ERRORS_DEFAULT) { 1144 const char *state; 1145 1146 switch (args->ar_errors) { 1147 case GFS2_ERRORS_WITHDRAW: 1148 state = "withdraw"; 1149 break; 1150 case GFS2_ERRORS_PANIC: 1151 state = "panic"; 1152 break; 1153 default: 1154 state = "unknown"; 1155 break; 1156 } 1157 seq_printf(s, ",errors=%s", state); 1158 } 1159 if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags)) 1160 seq_puts(s, ",nobarrier"); 1161 if (test_bit(SDF_DEMOTE, &sdp->sd_flags)) 1162 seq_puts(s, ",demote_interface_used"); 1163 if (args->ar_rgrplvb) 1164 seq_puts(s, ",rgrplvb"); 1165 if (args->ar_loccookie) 1166 seq_puts(s, ",loccookie"); 1167 return 0; 1168 } 1169 1170 /** 1171 * gfs2_glock_put_eventually 1172 * @gl: The glock to put 1173 * 1174 * When under memory pressure, trigger a deferred glock put to make sure we 1175 * won't call into DLM and deadlock. Otherwise, put the glock directly. 1176 */ 1177 1178 static void gfs2_glock_put_eventually(struct gfs2_glock *gl) 1179 { 1180 if (current->flags & PF_MEMALLOC) 1181 gfs2_glock_put_async(gl); 1182 else 1183 gfs2_glock_put(gl); 1184 } 1185 1186 static enum evict_behavior gfs2_upgrade_iopen_glock(struct inode *inode) 1187 { 1188 struct gfs2_inode *ip = GFS2_I(inode); 1189 struct gfs2_sbd *sdp = GFS2_SB(inode); 1190 struct gfs2_holder *gh = &ip->i_iopen_gh; 1191 int error; 1192 1193 gh->gh_flags |= GL_NOCACHE; 1194 gfs2_glock_dq_wait(gh); 1195 1196 /* 1197 * If there are no other lock holders, we will immediately get 1198 * exclusive access to the iopen glock here. 1199 * 1200 * Otherwise, the other nodes holding the lock will be notified about 1201 * our locking request (see iopen_go_callback()). If they do not have 1202 * the inode open, they are expected to evict the cached inode and 1203 * release the lock, allowing us to proceed. 1204 * 1205 * Otherwise, if they cannot evict the inode, they are expected to poke 1206 * the inode glock (note: not the iopen glock). We will notice that 1207 * and stop waiting for the iopen glock immediately. The other node(s) 1208 * are then expected to take care of deleting the inode when they no 1209 * longer use it. 1210 * 1211 * As a last resort, if another node keeps holding the iopen glock 1212 * without showing any activity on the inode glock, we will eventually 1213 * time out and fail the iopen glock upgrade. 1214 */ 1215 1216 gfs2_holder_reinit(LM_ST_EXCLUSIVE, GL_ASYNC | GL_NOCACHE, gh); 1217 error = gfs2_glock_nq(gh); 1218 if (error) 1219 return EVICT_SHOULD_SKIP_DELETE; 1220 1221 wait_event_interruptible_timeout(sdp->sd_async_glock_wait, 1222 !test_bit(HIF_WAIT, &gh->gh_iflags) || 1223 glock_needs_demote(ip->i_gl), 1224 5 * HZ); 1225 if (!test_bit(HIF_HOLDER, &gh->gh_iflags)) { 1226 gfs2_glock_dq(gh); 1227 if (glock_needs_demote(ip->i_gl)) 1228 return EVICT_SHOULD_SKIP_DELETE; 1229 return EVICT_SHOULD_DEFER_DELETE; 1230 } 1231 error = gfs2_glock_holder_ready(gh); 1232 if (error) 1233 return EVICT_SHOULD_SKIP_DELETE; 1234 return EVICT_SHOULD_DELETE; 1235 } 1236 1237 /** 1238 * evict_should_delete - determine whether the inode is eligible for deletion 1239 * @inode: The inode to evict 1240 * @gh: The glock holder structure 1241 * 1242 * This function determines whether the evicted inode is eligible to be deleted 1243 * and locks the inode glock. 1244 * 1245 * Returns: the fate of the dinode 1246 */ 1247 static enum evict_behavior evict_should_delete(struct inode *inode, 1248 struct gfs2_holder *gh) 1249 { 1250 struct gfs2_inode *ip = GFS2_I(inode); 1251 struct super_block *sb = inode->i_sb; 1252 struct gfs2_sbd *sdp = sb->s_fs_info; 1253 int ret; 1254 1255 if (gfs2_holder_initialized(&ip->i_iopen_gh) && 1256 test_bit(GLF_DEFER_DELETE, &ip->i_iopen_gh.gh_gl->gl_flags)) 1257 return EVICT_SHOULD_DEFER_DELETE; 1258 1259 /* Deletes should never happen under memory pressure anymore. */ 1260 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC)) 1261 return EVICT_SHOULD_DEFER_DELETE; 1262 1263 /* Must not read inode block until block type has been verified */ 1264 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, gh); 1265 if (unlikely(ret)) 1266 return EVICT_SHOULD_SKIP_DELETE; 1267 1268 if (gfs2_inode_already_deleted(ip->i_gl, ip->i_no_formal_ino)) 1269 return EVICT_SHOULD_SKIP_DELETE; 1270 ret = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED); 1271 if (ret) 1272 return EVICT_SHOULD_SKIP_DELETE; 1273 1274 ret = gfs2_instantiate(gh); 1275 if (ret) 1276 return EVICT_SHOULD_SKIP_DELETE; 1277 1278 /* 1279 * The inode may have been recreated in the meantime. 1280 */ 1281 if (inode->i_nlink) 1282 return EVICT_SHOULD_SKIP_DELETE; 1283 1284 if (gfs2_holder_initialized(&ip->i_iopen_gh) && 1285 test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) 1286 return gfs2_upgrade_iopen_glock(inode); 1287 return EVICT_SHOULD_DELETE; 1288 } 1289 1290 /** 1291 * evict_unlinked_inode - delete the pieces of an unlinked evicted inode 1292 * @inode: The inode to evict 1293 */ 1294 static int evict_unlinked_inode(struct inode *inode) 1295 { 1296 struct gfs2_inode *ip = GFS2_I(inode); 1297 int ret; 1298 1299 if (S_ISDIR(inode->i_mode) && 1300 (ip->i_diskflags & GFS2_DIF_EXHASH)) { 1301 ret = gfs2_dir_exhash_dealloc(ip); 1302 if (ret) 1303 goto out; 1304 } 1305 1306 if (ip->i_eattr) { 1307 ret = gfs2_ea_dealloc(ip, true); 1308 if (ret) 1309 goto out; 1310 } 1311 1312 if (!gfs2_is_stuffed(ip)) { 1313 ret = gfs2_file_dealloc(ip); 1314 if (ret) 1315 goto out; 1316 } 1317 1318 /* 1319 * As soon as we clear the bitmap for the dinode, gfs2_create_inode() 1320 * can get called to recreate it, or even gfs2_inode_lookup() if the 1321 * inode was recreated on another node in the meantime. 1322 * 1323 * However, inserting the new inode into the inode hash table will not 1324 * succeed until the old inode is removed, and that only happens after 1325 * ->evict_inode() returns. The new inode is attached to its inode and 1326 * iopen glocks after inserting it into the inode hash table, so at 1327 * that point we can be sure that both glocks are unused. 1328 */ 1329 1330 ret = gfs2_dinode_dealloc(ip); 1331 if (!ret && ip->i_gl) 1332 gfs2_inode_remember_delete(ip->i_gl, ip->i_no_formal_ino); 1333 1334 out: 1335 return ret; 1336 } 1337 1338 /* 1339 * evict_linked_inode - evict an inode whose dinode has not been unlinked 1340 * @inode: The inode to evict 1341 */ 1342 static int evict_linked_inode(struct inode *inode) 1343 { 1344 struct super_block *sb = inode->i_sb; 1345 struct gfs2_sbd *sdp = sb->s_fs_info; 1346 struct gfs2_inode *ip = GFS2_I(inode); 1347 struct address_space *metamapping; 1348 int ret; 1349 1350 gfs2_log_flush(sdp, ip->i_gl, GFS2_LOG_HEAD_FLUSH_NORMAL | 1351 GFS2_LFC_EVICT_INODE); 1352 metamapping = gfs2_glock2aspace(ip->i_gl); 1353 if (test_bit(GLF_DIRTY, &ip->i_gl->gl_flags)) { 1354 filemap_fdatawrite(metamapping); 1355 filemap_fdatawait(metamapping); 1356 } 1357 write_inode_now(inode, 1); 1358 gfs2_ail_flush(ip->i_gl, 0); 1359 1360 ret = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks); 1361 if (ret) 1362 return ret; 1363 1364 /* Needs to be done before glock release & also in a transaction */ 1365 truncate_inode_pages(&inode->i_data, 0); 1366 truncate_inode_pages(metamapping, 0); 1367 gfs2_trans_end(sdp); 1368 return 0; 1369 } 1370 1371 /** 1372 * gfs2_evict_inode - Remove an inode from cache 1373 * @inode: The inode to evict 1374 * 1375 * There are three cases to consider: 1376 * 1. i_nlink == 0, we are final opener (and must deallocate) 1377 * 2. i_nlink == 0, we are not the final opener (and cannot deallocate) 1378 * 3. i_nlink > 0 1379 * 1380 * If the fs is read only, then we have to treat all cases as per #3 1381 * since we are unable to do any deallocation. The inode will be 1382 * deallocated by the next read/write node to attempt an allocation 1383 * in the same resource group 1384 * 1385 * We have to (at the moment) hold the inodes main lock to cover 1386 * the gap between unlocking the shared lock on the iopen lock and 1387 * taking the exclusive lock. I'd rather do a shared -> exclusive 1388 * conversion on the iopen lock, but we can change that later. This 1389 * is safe, just less efficient. 1390 */ 1391 1392 static void gfs2_evict_inode(struct inode *inode) 1393 { 1394 struct super_block *sb = inode->i_sb; 1395 struct gfs2_sbd *sdp = sb->s_fs_info; 1396 struct gfs2_inode *ip = GFS2_I(inode); 1397 struct gfs2_holder gh; 1398 enum evict_behavior behavior; 1399 int ret; 1400 1401 gfs2_holder_mark_uninitialized(&gh); 1402 if (inode->i_nlink || sb_rdonly(sb) || !ip->i_no_addr) 1403 goto out; 1404 1405 /* 1406 * In case of an incomplete mount, gfs2_evict_inode() may be called for 1407 * system files without having an active journal to write to. In that 1408 * case, skip the filesystem evict. 1409 */ 1410 if (!sdp->sd_jdesc) 1411 goto out; 1412 1413 behavior = evict_should_delete(inode, &gh); 1414 if (behavior == EVICT_SHOULD_DEFER_DELETE && 1415 !test_bit(SDF_KILL, &sdp->sd_flags)) { 1416 struct gfs2_glock *io_gl = ip->i_iopen_gh.gh_gl; 1417 1418 if (io_gl) { 1419 gfs2_glock_hold(io_gl); 1420 if (!gfs2_queue_verify_delete(io_gl, true)) 1421 gfs2_glock_put(io_gl); 1422 goto out; 1423 } 1424 behavior = EVICT_SHOULD_SKIP_DELETE; 1425 } 1426 if (behavior == EVICT_SHOULD_DELETE) 1427 ret = evict_unlinked_inode(inode); 1428 else 1429 ret = evict_linked_inode(inode); 1430 1431 if (gfs2_rs_active(&ip->i_res)) 1432 gfs2_rs_deltree(&ip->i_res); 1433 1434 if (ret && ret != GLR_TRYFAILED && ret != -EROFS) 1435 fs_warn(sdp, "gfs2_evict_inode: %d\n", ret); 1436 out: 1437 if (gfs2_holder_initialized(&gh)) 1438 gfs2_glock_dq_uninit(&gh); 1439 truncate_inode_pages_final(&inode->i_data); 1440 if (ip->i_qadata) 1441 gfs2_assert_warn(sdp, ip->i_qadata->qa_ref == 0); 1442 gfs2_rs_deltree(&ip->i_res); 1443 gfs2_ordered_del_inode(ip); 1444 clear_inode(inode); 1445 gfs2_dir_hash_inval(ip); 1446 if (gfs2_holder_initialized(&ip->i_iopen_gh)) { 1447 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl; 1448 1449 glock_clear_object(gl, ip); 1450 gfs2_glock_hold(gl); 1451 ip->i_iopen_gh.gh_flags |= GL_NOCACHE; 1452 gfs2_glock_dq_uninit(&ip->i_iopen_gh); 1453 gfs2_glock_put_eventually(gl); 1454 } 1455 if (ip->i_gl) { 1456 glock_clear_object(ip->i_gl, ip); 1457 wait_on_bit_io(&ip->i_flags, GIF_GLOP_PENDING, TASK_UNINTERRUPTIBLE); 1458 gfs2_glock_put_eventually(ip->i_gl); 1459 rcu_assign_pointer(ip->i_gl, NULL); 1460 } 1461 } 1462 1463 static struct inode *gfs2_alloc_inode(struct super_block *sb) 1464 { 1465 struct gfs2_inode *ip; 1466 1467 ip = alloc_inode_sb(sb, gfs2_inode_cachep, GFP_KERNEL); 1468 if (!ip) 1469 return NULL; 1470 ip->i_no_addr = 0; 1471 ip->i_no_formal_ino = 0; 1472 ip->i_flags = 0; 1473 ip->i_gl = NULL; 1474 gfs2_holder_mark_uninitialized(&ip->i_iopen_gh); 1475 memset(&ip->i_res, 0, sizeof(ip->i_res)); 1476 RB_CLEAR_NODE(&ip->i_res.rs_node); 1477 ip->i_diskflags = 0; 1478 ip->i_rahead = 0; 1479 return &ip->i_inode; 1480 } 1481 1482 static void gfs2_free_inode(struct inode *inode) 1483 { 1484 kmem_cache_free(gfs2_inode_cachep, GFS2_I(inode)); 1485 } 1486 1487 void free_local_statfs_inodes(struct gfs2_sbd *sdp) 1488 { 1489 struct local_statfs_inode *lsi, *safe; 1490 1491 /* Run through the statfs inodes list to iput and free memory */ 1492 list_for_each_entry_safe(lsi, safe, &sdp->sd_sc_inodes_list, si_list) { 1493 if (lsi->si_jid == sdp->sd_jdesc->jd_jid) 1494 sdp->sd_sc_inode = NULL; /* belongs to this node */ 1495 if (lsi->si_sc_inode) 1496 iput(lsi->si_sc_inode); 1497 list_del(&lsi->si_list); 1498 kfree(lsi); 1499 } 1500 } 1501 1502 struct inode *find_local_statfs_inode(struct gfs2_sbd *sdp, 1503 unsigned int index) 1504 { 1505 struct local_statfs_inode *lsi; 1506 1507 /* Return the local (per node) statfs inode in the 1508 * sdp->sd_sc_inodes_list corresponding to the 'index'. */ 1509 list_for_each_entry(lsi, &sdp->sd_sc_inodes_list, si_list) { 1510 if (lsi->si_jid == index) 1511 return lsi->si_sc_inode; 1512 } 1513 return NULL; 1514 } 1515 1516 const struct super_operations gfs2_super_ops = { 1517 .alloc_inode = gfs2_alloc_inode, 1518 .free_inode = gfs2_free_inode, 1519 .write_inode = gfs2_write_inode, 1520 .dirty_inode = gfs2_dirty_inode, 1521 .evict_inode = gfs2_evict_inode, 1522 .put_super = gfs2_put_super, 1523 .sync_fs = gfs2_sync_fs, 1524 .freeze_super = gfs2_freeze_super, 1525 .freeze_fs = gfs2_freeze_fs, 1526 .thaw_super = gfs2_thaw_super, 1527 .statfs = gfs2_statfs, 1528 .drop_inode = gfs2_drop_inode, 1529 .show_options = gfs2_show_options, 1530 }; 1531 1532