1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. 4 * Copyright (C) 2004-2007 Red Hat, Inc. All rights reserved. 5 */ 6 7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 8 9 #include <linux/bio.h> 10 #include <linux/sched/signal.h> 11 #include <linux/slab.h> 12 #include <linux/spinlock.h> 13 #include <linux/completion.h> 14 #include <linux/buffer_head.h> 15 #include <linux/statfs.h> 16 #include <linux/seq_file.h> 17 #include <linux/mount.h> 18 #include <linux/kthread.h> 19 #include <linux/delay.h> 20 #include <linux/gfs2_ondisk.h> 21 #include <linux/crc32.h> 22 #include <linux/time.h> 23 #include <linux/wait.h> 24 #include <linux/writeback.h> 25 #include <linux/backing-dev.h> 26 #include <linux/kernel.h> 27 28 #include "gfs2.h" 29 #include "incore.h" 30 #include "bmap.h" 31 #include "dir.h" 32 #include "glock.h" 33 #include "glops.h" 34 #include "inode.h" 35 #include "log.h" 36 #include "meta_io.h" 37 #include "quota.h" 38 #include "recovery.h" 39 #include "rgrp.h" 40 #include "super.h" 41 #include "trans.h" 42 #include "util.h" 43 #include "sys.h" 44 #include "xattr.h" 45 #include "lops.h" 46 47 enum dinode_demise { 48 SHOULD_DELETE_DINODE, 49 SHOULD_NOT_DELETE_DINODE, 50 SHOULD_DEFER_EVICTION, 51 }; 52 53 /** 54 * gfs2_jindex_free - Clear all the journal index information 55 * @sdp: The GFS2 superblock 56 * 57 */ 58 59 void gfs2_jindex_free(struct gfs2_sbd *sdp) 60 { 61 struct list_head list; 62 struct gfs2_jdesc *jd; 63 64 spin_lock(&sdp->sd_jindex_spin); 65 list_add(&list, &sdp->sd_jindex_list); 66 list_del_init(&sdp->sd_jindex_list); 67 sdp->sd_journals = 0; 68 spin_unlock(&sdp->sd_jindex_spin); 69 70 sdp->sd_jdesc = NULL; 71 while (!list_empty(&list)) { 72 jd = list_first_entry(&list, struct gfs2_jdesc, jd_list); 73 gfs2_free_journal_extents(jd); 74 list_del(&jd->jd_list); 75 iput(jd->jd_inode); 76 jd->jd_inode = NULL; 77 kfree(jd); 78 } 79 } 80 81 static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid) 82 { 83 struct gfs2_jdesc *jd; 84 int found = 0; 85 86 list_for_each_entry(jd, head, jd_list) { 87 if (jd->jd_jid == jid) { 88 found = 1; 89 break; 90 } 91 } 92 93 if (!found) 94 jd = NULL; 95 96 return jd; 97 } 98 99 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid) 100 { 101 struct gfs2_jdesc *jd; 102 103 spin_lock(&sdp->sd_jindex_spin); 104 jd = jdesc_find_i(&sdp->sd_jindex_list, jid); 105 spin_unlock(&sdp->sd_jindex_spin); 106 107 return jd; 108 } 109 110 int gfs2_jdesc_check(struct gfs2_jdesc *jd) 111 { 112 struct gfs2_inode *ip = GFS2_I(jd->jd_inode); 113 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode); 114 u64 size = i_size_read(jd->jd_inode); 115 116 if (gfs2_check_internal_file_size(jd->jd_inode, 8 << 20, BIT(30))) 117 return -EIO; 118 119 jd->jd_blocks = size >> sdp->sd_sb.sb_bsize_shift; 120 121 if (gfs2_write_alloc_required(ip, 0, size)) { 122 gfs2_consist_inode(ip); 123 return -EIO; 124 } 125 126 return 0; 127 } 128 129 static int init_threads(struct gfs2_sbd *sdp) 130 { 131 struct task_struct *p; 132 int error = 0; 133 134 p = kthread_run(gfs2_logd, sdp, "gfs2_logd"); 135 if (IS_ERR(p)) { 136 error = PTR_ERR(p); 137 fs_err(sdp, "can't start logd thread: %d\n", error); 138 return error; 139 } 140 sdp->sd_logd_process = p; 141 142 p = kthread_run(gfs2_quotad, sdp, "gfs2_quotad"); 143 if (IS_ERR(p)) { 144 error = PTR_ERR(p); 145 fs_err(sdp, "can't start quotad thread: %d\n", error); 146 goto fail; 147 } 148 sdp->sd_quotad_process = p; 149 return 0; 150 151 fail: 152 kthread_stop(sdp->sd_logd_process); 153 sdp->sd_logd_process = NULL; 154 return error; 155 } 156 157 /** 158 * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one 159 * @sdp: the filesystem 160 * 161 * Returns: errno 162 */ 163 164 int gfs2_make_fs_rw(struct gfs2_sbd *sdp) 165 { 166 struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode); 167 struct gfs2_glock *j_gl = ip->i_gl; 168 struct gfs2_holder freeze_gh; 169 struct gfs2_log_header_host head; 170 int error; 171 172 error = init_threads(sdp); 173 if (error) 174 return error; 175 176 error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED, 177 LM_FLAG_NOEXP | GL_EXACT, 178 &freeze_gh); 179 if (error) 180 goto fail_threads; 181 182 j_gl->gl_ops->go_inval(j_gl, DIO_METADATA); 183 if (gfs2_withdrawn(sdp)) { 184 error = -EIO; 185 goto fail; 186 } 187 188 error = gfs2_find_jhead(sdp->sd_jdesc, &head, false); 189 if (error || gfs2_withdrawn(sdp)) 190 goto fail; 191 192 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) { 193 gfs2_consist(sdp); 194 error = -EIO; 195 goto fail; 196 } 197 198 /* Initialize some head of the log stuff */ 199 sdp->sd_log_sequence = head.lh_sequence + 1; 200 gfs2_log_pointers_init(sdp, head.lh_blkno); 201 202 error = gfs2_quota_init(sdp); 203 if (error || gfs2_withdrawn(sdp)) 204 goto fail; 205 206 set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags); 207 208 gfs2_glock_dq_uninit(&freeze_gh); 209 210 return 0; 211 212 fail: 213 gfs2_glock_dq_uninit(&freeze_gh); 214 fail_threads: 215 if (sdp->sd_quotad_process) 216 kthread_stop(sdp->sd_quotad_process); 217 sdp->sd_quotad_process = NULL; 218 if (sdp->sd_logd_process) 219 kthread_stop(sdp->sd_logd_process); 220 sdp->sd_logd_process = NULL; 221 return error; 222 } 223 224 void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf) 225 { 226 const struct gfs2_statfs_change *str = buf; 227 228 sc->sc_total = be64_to_cpu(str->sc_total); 229 sc->sc_free = be64_to_cpu(str->sc_free); 230 sc->sc_dinodes = be64_to_cpu(str->sc_dinodes); 231 } 232 233 void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf) 234 { 235 struct gfs2_statfs_change *str = buf; 236 237 str->sc_total = cpu_to_be64(sc->sc_total); 238 str->sc_free = cpu_to_be64(sc->sc_free); 239 str->sc_dinodes = cpu_to_be64(sc->sc_dinodes); 240 } 241 242 int gfs2_statfs_init(struct gfs2_sbd *sdp) 243 { 244 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode); 245 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master; 246 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode); 247 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local; 248 struct buffer_head *m_bh, *l_bh; 249 struct gfs2_holder gh; 250 int error; 251 252 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE, 253 &gh); 254 if (error) 255 return error; 256 257 error = gfs2_meta_inode_buffer(m_ip, &m_bh); 258 if (error) 259 goto out; 260 261 if (sdp->sd_args.ar_spectator) { 262 spin_lock(&sdp->sd_statfs_spin); 263 gfs2_statfs_change_in(m_sc, m_bh->b_data + 264 sizeof(struct gfs2_dinode)); 265 spin_unlock(&sdp->sd_statfs_spin); 266 } else { 267 error = gfs2_meta_inode_buffer(l_ip, &l_bh); 268 if (error) 269 goto out_m_bh; 270 271 spin_lock(&sdp->sd_statfs_spin); 272 gfs2_statfs_change_in(m_sc, m_bh->b_data + 273 sizeof(struct gfs2_dinode)); 274 gfs2_statfs_change_in(l_sc, l_bh->b_data + 275 sizeof(struct gfs2_dinode)); 276 spin_unlock(&sdp->sd_statfs_spin); 277 278 brelse(l_bh); 279 } 280 281 out_m_bh: 282 brelse(m_bh); 283 out: 284 gfs2_glock_dq_uninit(&gh); 285 return 0; 286 } 287 288 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free, 289 s64 dinodes) 290 { 291 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode); 292 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local; 293 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master; 294 struct buffer_head *l_bh; 295 s64 x, y; 296 int need_sync = 0; 297 int error; 298 299 error = gfs2_meta_inode_buffer(l_ip, &l_bh); 300 if (error) 301 return; 302 303 gfs2_trans_add_meta(l_ip->i_gl, l_bh); 304 305 spin_lock(&sdp->sd_statfs_spin); 306 l_sc->sc_total += total; 307 l_sc->sc_free += free; 308 l_sc->sc_dinodes += dinodes; 309 gfs2_statfs_change_out(l_sc, l_bh->b_data + sizeof(struct gfs2_dinode)); 310 if (sdp->sd_args.ar_statfs_percent) { 311 x = 100 * l_sc->sc_free; 312 y = m_sc->sc_free * sdp->sd_args.ar_statfs_percent; 313 if (x >= y || x <= -y) 314 need_sync = 1; 315 } 316 spin_unlock(&sdp->sd_statfs_spin); 317 318 brelse(l_bh); 319 if (need_sync) 320 gfs2_wake_up_statfs(sdp); 321 } 322 323 void update_statfs(struct gfs2_sbd *sdp, struct buffer_head *m_bh, 324 struct buffer_head *l_bh) 325 { 326 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode); 327 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode); 328 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master; 329 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local; 330 331 gfs2_trans_add_meta(l_ip->i_gl, l_bh); 332 gfs2_trans_add_meta(m_ip->i_gl, m_bh); 333 334 spin_lock(&sdp->sd_statfs_spin); 335 m_sc->sc_total += l_sc->sc_total; 336 m_sc->sc_free += l_sc->sc_free; 337 m_sc->sc_dinodes += l_sc->sc_dinodes; 338 memset(l_sc, 0, sizeof(struct gfs2_statfs_change)); 339 memset(l_bh->b_data + sizeof(struct gfs2_dinode), 340 0, sizeof(struct gfs2_statfs_change)); 341 gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode)); 342 spin_unlock(&sdp->sd_statfs_spin); 343 } 344 345 int gfs2_statfs_sync(struct super_block *sb, int type) 346 { 347 struct gfs2_sbd *sdp = sb->s_fs_info; 348 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode); 349 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode); 350 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master; 351 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local; 352 struct gfs2_holder gh; 353 struct buffer_head *m_bh, *l_bh; 354 int error; 355 356 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE, 357 &gh); 358 if (error) 359 goto out; 360 361 error = gfs2_meta_inode_buffer(m_ip, &m_bh); 362 if (error) 363 goto out_unlock; 364 365 spin_lock(&sdp->sd_statfs_spin); 366 gfs2_statfs_change_in(m_sc, m_bh->b_data + 367 sizeof(struct gfs2_dinode)); 368 if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) { 369 spin_unlock(&sdp->sd_statfs_spin); 370 goto out_bh; 371 } 372 spin_unlock(&sdp->sd_statfs_spin); 373 374 error = gfs2_meta_inode_buffer(l_ip, &l_bh); 375 if (error) 376 goto out_bh; 377 378 error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0); 379 if (error) 380 goto out_bh2; 381 382 update_statfs(sdp, m_bh, l_bh); 383 sdp->sd_statfs_force_sync = 0; 384 385 gfs2_trans_end(sdp); 386 387 out_bh2: 388 brelse(l_bh); 389 out_bh: 390 brelse(m_bh); 391 out_unlock: 392 gfs2_glock_dq_uninit(&gh); 393 out: 394 return error; 395 } 396 397 struct lfcc { 398 struct list_head list; 399 struct gfs2_holder gh; 400 }; 401 402 /** 403 * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all 404 * journals are clean 405 * @sdp: the file system 406 * @state: the state to put the transaction lock into 407 * @t_gh: the hold on the transaction lock 408 * 409 * Returns: errno 410 */ 411 412 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp) 413 { 414 struct gfs2_inode *ip; 415 struct gfs2_jdesc *jd; 416 struct lfcc *lfcc; 417 LIST_HEAD(list); 418 struct gfs2_log_header_host lh; 419 int error; 420 421 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) { 422 lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL); 423 if (!lfcc) { 424 error = -ENOMEM; 425 goto out; 426 } 427 ip = GFS2_I(jd->jd_inode); 428 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh); 429 if (error) { 430 kfree(lfcc); 431 goto out; 432 } 433 list_add(&lfcc->list, &list); 434 } 435 436 error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_EXCLUSIVE, 437 LM_FLAG_NOEXP, &sdp->sd_freeze_gh); 438 if (error) 439 goto out; 440 441 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) { 442 error = gfs2_jdesc_check(jd); 443 if (error) 444 break; 445 error = gfs2_find_jhead(jd, &lh, false); 446 if (error) 447 break; 448 if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) { 449 error = -EBUSY; 450 break; 451 } 452 } 453 454 if (error) 455 gfs2_glock_dq_uninit(&sdp->sd_freeze_gh); 456 457 out: 458 while (!list_empty(&list)) { 459 lfcc = list_first_entry(&list, struct lfcc, list); 460 list_del(&lfcc->list); 461 gfs2_glock_dq_uninit(&lfcc->gh); 462 kfree(lfcc); 463 } 464 return error; 465 } 466 467 void gfs2_dinode_out(const struct gfs2_inode *ip, void *buf) 468 { 469 struct gfs2_dinode *str = buf; 470 471 str->di_header.mh_magic = cpu_to_be32(GFS2_MAGIC); 472 str->di_header.mh_type = cpu_to_be32(GFS2_METATYPE_DI); 473 str->di_header.mh_format = cpu_to_be32(GFS2_FORMAT_DI); 474 str->di_num.no_addr = cpu_to_be64(ip->i_no_addr); 475 str->di_num.no_formal_ino = cpu_to_be64(ip->i_no_formal_ino); 476 str->di_mode = cpu_to_be32(ip->i_inode.i_mode); 477 str->di_uid = cpu_to_be32(i_uid_read(&ip->i_inode)); 478 str->di_gid = cpu_to_be32(i_gid_read(&ip->i_inode)); 479 str->di_nlink = cpu_to_be32(ip->i_inode.i_nlink); 480 str->di_size = cpu_to_be64(i_size_read(&ip->i_inode)); 481 str->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(&ip->i_inode)); 482 str->di_atime = cpu_to_be64(ip->i_inode.i_atime.tv_sec); 483 str->di_mtime = cpu_to_be64(ip->i_inode.i_mtime.tv_sec); 484 str->di_ctime = cpu_to_be64(ip->i_inode.i_ctime.tv_sec); 485 486 str->di_goal_meta = cpu_to_be64(ip->i_goal); 487 str->di_goal_data = cpu_to_be64(ip->i_goal); 488 str->di_generation = cpu_to_be64(ip->i_generation); 489 490 str->di_flags = cpu_to_be32(ip->i_diskflags); 491 str->di_height = cpu_to_be16(ip->i_height); 492 str->di_payload_format = cpu_to_be32(S_ISDIR(ip->i_inode.i_mode) && 493 !(ip->i_diskflags & GFS2_DIF_EXHASH) ? 494 GFS2_FORMAT_DE : 0); 495 str->di_depth = cpu_to_be16(ip->i_depth); 496 str->di_entries = cpu_to_be32(ip->i_entries); 497 498 str->di_eattr = cpu_to_be64(ip->i_eattr); 499 str->di_atime_nsec = cpu_to_be32(ip->i_inode.i_atime.tv_nsec); 500 str->di_mtime_nsec = cpu_to_be32(ip->i_inode.i_mtime.tv_nsec); 501 str->di_ctime_nsec = cpu_to_be32(ip->i_inode.i_ctime.tv_nsec); 502 } 503 504 /** 505 * gfs2_write_inode - Make sure the inode is stable on the disk 506 * @inode: The inode 507 * @wbc: The writeback control structure 508 * 509 * Returns: errno 510 */ 511 512 static int gfs2_write_inode(struct inode *inode, struct writeback_control *wbc) 513 { 514 struct gfs2_inode *ip = GFS2_I(inode); 515 struct gfs2_sbd *sdp = GFS2_SB(inode); 516 struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl); 517 struct backing_dev_info *bdi = inode_to_bdi(metamapping->host); 518 int ret = 0; 519 bool flush_all = (wbc->sync_mode == WB_SYNC_ALL || gfs2_is_jdata(ip)); 520 521 if (flush_all) 522 gfs2_log_flush(GFS2_SB(inode), ip->i_gl, 523 GFS2_LOG_HEAD_FLUSH_NORMAL | 524 GFS2_LFC_WRITE_INODE); 525 if (bdi->wb.dirty_exceeded) 526 gfs2_ail1_flush(sdp, wbc); 527 else 528 filemap_fdatawrite(metamapping); 529 if (flush_all) 530 ret = filemap_fdatawait(metamapping); 531 if (ret) 532 mark_inode_dirty_sync(inode); 533 else { 534 spin_lock(&inode->i_lock); 535 if (!(inode->i_flags & I_DIRTY)) 536 gfs2_ordered_del_inode(ip); 537 spin_unlock(&inode->i_lock); 538 } 539 return ret; 540 } 541 542 /** 543 * gfs2_dirty_inode - check for atime updates 544 * @inode: The inode in question 545 * @flags: The type of dirty 546 * 547 * Unfortunately it can be called under any combination of inode 548 * glock and transaction lock, so we have to check carefully. 549 * 550 * At the moment this deals only with atime - it should be possible 551 * to expand that role in future, once a review of the locking has 552 * been carried out. 553 */ 554 555 static void gfs2_dirty_inode(struct inode *inode, int flags) 556 { 557 struct gfs2_inode *ip = GFS2_I(inode); 558 struct gfs2_sbd *sdp = GFS2_SB(inode); 559 struct buffer_head *bh; 560 struct gfs2_holder gh; 561 int need_unlock = 0; 562 int need_endtrans = 0; 563 int ret; 564 565 if (unlikely(gfs2_withdrawn(sdp))) 566 return; 567 if (!gfs2_glock_is_locked_by_me(ip->i_gl)) { 568 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh); 569 if (ret) { 570 fs_err(sdp, "dirty_inode: glock %d\n", ret); 571 gfs2_dump_glock(NULL, ip->i_gl, true); 572 return; 573 } 574 need_unlock = 1; 575 } else if (WARN_ON_ONCE(ip->i_gl->gl_state != LM_ST_EXCLUSIVE)) 576 return; 577 578 if (current->journal_info == NULL) { 579 ret = gfs2_trans_begin(sdp, RES_DINODE, 0); 580 if (ret) { 581 fs_err(sdp, "dirty_inode: gfs2_trans_begin %d\n", ret); 582 goto out; 583 } 584 need_endtrans = 1; 585 } 586 587 ret = gfs2_meta_inode_buffer(ip, &bh); 588 if (ret == 0) { 589 gfs2_trans_add_meta(ip->i_gl, bh); 590 gfs2_dinode_out(ip, bh->b_data); 591 brelse(bh); 592 } 593 594 if (need_endtrans) 595 gfs2_trans_end(sdp); 596 out: 597 if (need_unlock) 598 gfs2_glock_dq_uninit(&gh); 599 } 600 601 /** 602 * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one 603 * @sdp: the filesystem 604 * 605 * Returns: errno 606 */ 607 608 int gfs2_make_fs_ro(struct gfs2_sbd *sdp) 609 { 610 struct gfs2_holder freeze_gh; 611 int error = 0; 612 int log_write_allowed = test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags); 613 614 gfs2_holder_mark_uninitialized(&freeze_gh); 615 if (sdp->sd_freeze_gl && 616 !gfs2_glock_is_locked_by_me(sdp->sd_freeze_gl)) { 617 if (!log_write_allowed) { 618 error = gfs2_glock_nq_init(sdp->sd_freeze_gl, 619 LM_ST_SHARED, LM_FLAG_TRY | 620 LM_FLAG_NOEXP | GL_EXACT, 621 &freeze_gh); 622 if (error == GLR_TRYFAILED) 623 error = 0; 624 } else { 625 error = gfs2_glock_nq_init(sdp->sd_freeze_gl, 626 LM_ST_SHARED, 627 LM_FLAG_NOEXP | GL_EXACT, 628 &freeze_gh); 629 if (error && !gfs2_withdrawn(sdp)) 630 return error; 631 } 632 } 633 634 gfs2_flush_delete_work(sdp); 635 if (!log_write_allowed && current == sdp->sd_quotad_process) 636 fs_warn(sdp, "The quotad daemon is withdrawing.\n"); 637 else if (sdp->sd_quotad_process) 638 kthread_stop(sdp->sd_quotad_process); 639 sdp->sd_quotad_process = NULL; 640 641 if (!log_write_allowed && current == sdp->sd_logd_process) 642 fs_warn(sdp, "The logd daemon is withdrawing.\n"); 643 else if (sdp->sd_logd_process) 644 kthread_stop(sdp->sd_logd_process); 645 sdp->sd_logd_process = NULL; 646 647 if (log_write_allowed) { 648 gfs2_quota_sync(sdp->sd_vfs, 0); 649 gfs2_statfs_sync(sdp->sd_vfs, 0); 650 651 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SHUTDOWN | 652 GFS2_LFC_MAKE_FS_RO); 653 wait_event(sdp->sd_reserving_log_wait, 654 atomic_read(&sdp->sd_reserving_log) == 0); 655 gfs2_assert_warn(sdp, atomic_read(&sdp->sd_log_blks_free) == 656 sdp->sd_jdesc->jd_blocks); 657 } else { 658 wait_event_timeout(sdp->sd_reserving_log_wait, 659 atomic_read(&sdp->sd_reserving_log) == 0, 660 HZ * 5); 661 } 662 if (gfs2_holder_initialized(&freeze_gh)) 663 gfs2_glock_dq_uninit(&freeze_gh); 664 665 gfs2_quota_cleanup(sdp); 666 667 if (!log_write_allowed) 668 sdp->sd_vfs->s_flags |= SB_RDONLY; 669 670 return error; 671 } 672 673 /** 674 * gfs2_put_super - Unmount the filesystem 675 * @sb: The VFS superblock 676 * 677 */ 678 679 static void gfs2_put_super(struct super_block *sb) 680 { 681 struct gfs2_sbd *sdp = sb->s_fs_info; 682 int error; 683 struct gfs2_jdesc *jd; 684 685 /* No more recovery requests */ 686 set_bit(SDF_NORECOVERY, &sdp->sd_flags); 687 smp_mb(); 688 689 /* Wait on outstanding recovery */ 690 restart: 691 spin_lock(&sdp->sd_jindex_spin); 692 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) { 693 if (!test_bit(JDF_RECOVERY, &jd->jd_flags)) 694 continue; 695 spin_unlock(&sdp->sd_jindex_spin); 696 wait_on_bit(&jd->jd_flags, JDF_RECOVERY, 697 TASK_UNINTERRUPTIBLE); 698 goto restart; 699 } 700 spin_unlock(&sdp->sd_jindex_spin); 701 702 if (!sb_rdonly(sb)) { 703 error = gfs2_make_fs_ro(sdp); 704 if (error) 705 gfs2_io_error(sdp); 706 } 707 WARN_ON(gfs2_withdrawing(sdp)); 708 709 /* At this point, we're through modifying the disk */ 710 711 /* Release stuff */ 712 713 iput(sdp->sd_jindex); 714 iput(sdp->sd_statfs_inode); 715 iput(sdp->sd_rindex); 716 iput(sdp->sd_quota_inode); 717 718 gfs2_glock_put(sdp->sd_rename_gl); 719 gfs2_glock_put(sdp->sd_freeze_gl); 720 721 if (!sdp->sd_args.ar_spectator) { 722 if (gfs2_holder_initialized(&sdp->sd_journal_gh)) 723 gfs2_glock_dq_uninit(&sdp->sd_journal_gh); 724 if (gfs2_holder_initialized(&sdp->sd_jinode_gh)) 725 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh); 726 gfs2_glock_dq_uninit(&sdp->sd_sc_gh); 727 gfs2_glock_dq_uninit(&sdp->sd_qc_gh); 728 free_local_statfs_inodes(sdp); 729 iput(sdp->sd_qc_inode); 730 } 731 732 gfs2_glock_dq_uninit(&sdp->sd_live_gh); 733 gfs2_clear_rgrpd(sdp); 734 gfs2_jindex_free(sdp); 735 /* Take apart glock structures and buffer lists */ 736 gfs2_gl_hash_clear(sdp); 737 truncate_inode_pages_final(&sdp->sd_aspace); 738 gfs2_delete_debugfs_file(sdp); 739 /* Unmount the locking protocol */ 740 gfs2_lm_unmount(sdp); 741 742 /* At this point, we're through participating in the lockspace */ 743 gfs2_sys_fs_del(sdp); 744 free_sbd(sdp); 745 } 746 747 /** 748 * gfs2_sync_fs - sync the filesystem 749 * @sb: the superblock 750 * 751 * Flushes the log to disk. 752 */ 753 754 static int gfs2_sync_fs(struct super_block *sb, int wait) 755 { 756 struct gfs2_sbd *sdp = sb->s_fs_info; 757 758 gfs2_quota_sync(sb, -1); 759 if (wait) 760 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL | 761 GFS2_LFC_SYNC_FS); 762 return sdp->sd_log_error; 763 } 764 765 void gfs2_freeze_func(struct work_struct *work) 766 { 767 int error; 768 struct gfs2_holder freeze_gh; 769 struct gfs2_sbd *sdp = container_of(work, struct gfs2_sbd, sd_freeze_work); 770 struct super_block *sb = sdp->sd_vfs; 771 772 atomic_inc(&sb->s_active); 773 error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED, 774 LM_FLAG_NOEXP | GL_EXACT, &freeze_gh); 775 if (error) { 776 fs_info(sdp, "GFS2: couldn't get freeze lock : %d\n", error); 777 gfs2_assert_withdraw(sdp, 0); 778 } else { 779 atomic_set(&sdp->sd_freeze_state, SFS_UNFROZEN); 780 error = thaw_super(sb); 781 if (error) { 782 fs_info(sdp, "GFS2: couldn't thaw filesystem: %d\n", 783 error); 784 gfs2_assert_withdraw(sdp, 0); 785 } 786 gfs2_glock_dq_uninit(&freeze_gh); 787 } 788 deactivate_super(sb); 789 clear_bit_unlock(SDF_FS_FROZEN, &sdp->sd_flags); 790 wake_up_bit(&sdp->sd_flags, SDF_FS_FROZEN); 791 return; 792 } 793 794 /** 795 * gfs2_freeze - prevent further writes to the filesystem 796 * @sb: the VFS structure for the filesystem 797 * 798 */ 799 800 static int gfs2_freeze(struct super_block *sb) 801 { 802 struct gfs2_sbd *sdp = sb->s_fs_info; 803 int error = 0; 804 805 mutex_lock(&sdp->sd_freeze_mutex); 806 if (atomic_read(&sdp->sd_freeze_state) != SFS_UNFROZEN) 807 goto out; 808 809 for (;;) { 810 if (gfs2_withdrawn(sdp)) { 811 error = -EINVAL; 812 goto out; 813 } 814 815 error = gfs2_lock_fs_check_clean(sdp); 816 if (!error) 817 break; 818 819 if (error == -EBUSY) 820 fs_err(sdp, "waiting for recovery before freeze\n"); 821 else if (error == -EIO) { 822 fs_err(sdp, "Fatal IO error: cannot freeze gfs2 due " 823 "to recovery error.\n"); 824 goto out; 825 } else { 826 fs_err(sdp, "error freezing FS: %d\n", error); 827 } 828 fs_err(sdp, "retrying...\n"); 829 msleep(1000); 830 } 831 set_bit(SDF_FS_FROZEN, &sdp->sd_flags); 832 out: 833 mutex_unlock(&sdp->sd_freeze_mutex); 834 return error; 835 } 836 837 /** 838 * gfs2_unfreeze - reallow writes to the filesystem 839 * @sb: the VFS structure for the filesystem 840 * 841 */ 842 843 static int gfs2_unfreeze(struct super_block *sb) 844 { 845 struct gfs2_sbd *sdp = sb->s_fs_info; 846 847 mutex_lock(&sdp->sd_freeze_mutex); 848 if (atomic_read(&sdp->sd_freeze_state) != SFS_FROZEN || 849 !gfs2_holder_initialized(&sdp->sd_freeze_gh)) { 850 mutex_unlock(&sdp->sd_freeze_mutex); 851 return 0; 852 } 853 854 gfs2_glock_dq_uninit(&sdp->sd_freeze_gh); 855 mutex_unlock(&sdp->sd_freeze_mutex); 856 return wait_on_bit(&sdp->sd_flags, SDF_FS_FROZEN, TASK_INTERRUPTIBLE); 857 } 858 859 /** 860 * statfs_fill - fill in the sg for a given RG 861 * @rgd: the RG 862 * @sc: the sc structure 863 * 864 * Returns: 0 on success, -ESTALE if the LVB is invalid 865 */ 866 867 static int statfs_slow_fill(struct gfs2_rgrpd *rgd, 868 struct gfs2_statfs_change_host *sc) 869 { 870 gfs2_rgrp_verify(rgd); 871 sc->sc_total += rgd->rd_data; 872 sc->sc_free += rgd->rd_free; 873 sc->sc_dinodes += rgd->rd_dinodes; 874 return 0; 875 } 876 877 /** 878 * gfs2_statfs_slow - Stat a filesystem using asynchronous locking 879 * @sdp: the filesystem 880 * @sc: the sc info that will be returned 881 * 882 * Any error (other than a signal) will cause this routine to fall back 883 * to the synchronous version. 884 * 885 * FIXME: This really shouldn't busy wait like this. 886 * 887 * Returns: errno 888 */ 889 890 static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc) 891 { 892 struct gfs2_rgrpd *rgd_next; 893 struct gfs2_holder *gha, *gh; 894 unsigned int slots = 64; 895 unsigned int x; 896 int done; 897 int error = 0, err; 898 899 memset(sc, 0, sizeof(struct gfs2_statfs_change_host)); 900 gha = kmalloc_array(slots, sizeof(struct gfs2_holder), GFP_KERNEL); 901 if (!gha) 902 return -ENOMEM; 903 for (x = 0; x < slots; x++) 904 gfs2_holder_mark_uninitialized(gha + x); 905 906 rgd_next = gfs2_rgrpd_get_first(sdp); 907 908 for (;;) { 909 done = 1; 910 911 for (x = 0; x < slots; x++) { 912 gh = gha + x; 913 914 if (gfs2_holder_initialized(gh) && gfs2_glock_poll(gh)) { 915 err = gfs2_glock_wait(gh); 916 if (err) { 917 gfs2_holder_uninit(gh); 918 error = err; 919 } else { 920 if (!error) { 921 struct gfs2_rgrpd *rgd = 922 gfs2_glock2rgrp(gh->gh_gl); 923 924 error = statfs_slow_fill(rgd, sc); 925 } 926 gfs2_glock_dq_uninit(gh); 927 } 928 } 929 930 if (gfs2_holder_initialized(gh)) 931 done = 0; 932 else if (rgd_next && !error) { 933 error = gfs2_glock_nq_init(rgd_next->rd_gl, 934 LM_ST_SHARED, 935 GL_ASYNC, 936 gh); 937 rgd_next = gfs2_rgrpd_get_next(rgd_next); 938 done = 0; 939 } 940 941 if (signal_pending(current)) 942 error = -ERESTARTSYS; 943 } 944 945 if (done) 946 break; 947 948 yield(); 949 } 950 951 kfree(gha); 952 return error; 953 } 954 955 /** 956 * gfs2_statfs_i - Do a statfs 957 * @sdp: the filesystem 958 * @sg: the sg structure 959 * 960 * Returns: errno 961 */ 962 963 static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc) 964 { 965 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master; 966 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local; 967 968 spin_lock(&sdp->sd_statfs_spin); 969 970 *sc = *m_sc; 971 sc->sc_total += l_sc->sc_total; 972 sc->sc_free += l_sc->sc_free; 973 sc->sc_dinodes += l_sc->sc_dinodes; 974 975 spin_unlock(&sdp->sd_statfs_spin); 976 977 if (sc->sc_free < 0) 978 sc->sc_free = 0; 979 if (sc->sc_free > sc->sc_total) 980 sc->sc_free = sc->sc_total; 981 if (sc->sc_dinodes < 0) 982 sc->sc_dinodes = 0; 983 984 return 0; 985 } 986 987 /** 988 * gfs2_statfs - Gather and return stats about the filesystem 989 * @sb: The superblock 990 * @statfsbuf: The buffer 991 * 992 * Returns: 0 on success or error code 993 */ 994 995 static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf) 996 { 997 struct super_block *sb = dentry->d_sb; 998 struct gfs2_sbd *sdp = sb->s_fs_info; 999 struct gfs2_statfs_change_host sc; 1000 int error; 1001 1002 error = gfs2_rindex_update(sdp); 1003 if (error) 1004 return error; 1005 1006 if (gfs2_tune_get(sdp, gt_statfs_slow)) 1007 error = gfs2_statfs_slow(sdp, &sc); 1008 else 1009 error = gfs2_statfs_i(sdp, &sc); 1010 1011 if (error) 1012 return error; 1013 1014 buf->f_type = GFS2_MAGIC; 1015 buf->f_bsize = sdp->sd_sb.sb_bsize; 1016 buf->f_blocks = sc.sc_total; 1017 buf->f_bfree = sc.sc_free; 1018 buf->f_bavail = sc.sc_free; 1019 buf->f_files = sc.sc_dinodes + sc.sc_free; 1020 buf->f_ffree = sc.sc_free; 1021 buf->f_namelen = GFS2_FNAMESIZE; 1022 1023 return 0; 1024 } 1025 1026 /** 1027 * gfs2_drop_inode - Drop an inode (test for remote unlink) 1028 * @inode: The inode to drop 1029 * 1030 * If we've received a callback on an iopen lock then it's because a 1031 * remote node tried to deallocate the inode but failed due to this node 1032 * still having the inode open. Here we mark the link count zero 1033 * since we know that it must have reached zero if the GLF_DEMOTE flag 1034 * is set on the iopen glock. If we didn't do a disk read since the 1035 * remote node removed the final link then we might otherwise miss 1036 * this event. This check ensures that this node will deallocate the 1037 * inode's blocks, or alternatively pass the baton on to another 1038 * node for later deallocation. 1039 */ 1040 1041 static int gfs2_drop_inode(struct inode *inode) 1042 { 1043 struct gfs2_inode *ip = GFS2_I(inode); 1044 1045 if (!test_bit(GIF_FREE_VFS_INODE, &ip->i_flags) && 1046 inode->i_nlink && 1047 gfs2_holder_initialized(&ip->i_iopen_gh)) { 1048 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl; 1049 if (test_bit(GLF_DEMOTE, &gl->gl_flags)) 1050 clear_nlink(inode); 1051 } 1052 1053 /* 1054 * When under memory pressure when an inode's link count has dropped to 1055 * zero, defer deleting the inode to the delete workqueue. This avoids 1056 * calling into DLM under memory pressure, which can deadlock. 1057 */ 1058 if (!inode->i_nlink && 1059 unlikely(current->flags & PF_MEMALLOC) && 1060 gfs2_holder_initialized(&ip->i_iopen_gh)) { 1061 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl; 1062 1063 gfs2_glock_hold(gl); 1064 if (!gfs2_queue_delete_work(gl, 0)) 1065 gfs2_glock_queue_put(gl); 1066 return false; 1067 } 1068 1069 return generic_drop_inode(inode); 1070 } 1071 1072 static int is_ancestor(const struct dentry *d1, const struct dentry *d2) 1073 { 1074 do { 1075 if (d1 == d2) 1076 return 1; 1077 d1 = d1->d_parent; 1078 } while (!IS_ROOT(d1)); 1079 return 0; 1080 } 1081 1082 /** 1083 * gfs2_show_options - Show mount options for /proc/mounts 1084 * @s: seq_file structure 1085 * @root: root of this (sub)tree 1086 * 1087 * Returns: 0 on success or error code 1088 */ 1089 1090 static int gfs2_show_options(struct seq_file *s, struct dentry *root) 1091 { 1092 struct gfs2_sbd *sdp = root->d_sb->s_fs_info; 1093 struct gfs2_args *args = &sdp->sd_args; 1094 int val; 1095 1096 if (is_ancestor(root, sdp->sd_master_dir)) 1097 seq_puts(s, ",meta"); 1098 if (args->ar_lockproto[0]) 1099 seq_show_option(s, "lockproto", args->ar_lockproto); 1100 if (args->ar_locktable[0]) 1101 seq_show_option(s, "locktable", args->ar_locktable); 1102 if (args->ar_hostdata[0]) 1103 seq_show_option(s, "hostdata", args->ar_hostdata); 1104 if (args->ar_spectator) 1105 seq_puts(s, ",spectator"); 1106 if (args->ar_localflocks) 1107 seq_puts(s, ",localflocks"); 1108 if (args->ar_debug) 1109 seq_puts(s, ",debug"); 1110 if (args->ar_posix_acl) 1111 seq_puts(s, ",acl"); 1112 if (args->ar_quota != GFS2_QUOTA_DEFAULT) { 1113 char *state; 1114 switch (args->ar_quota) { 1115 case GFS2_QUOTA_OFF: 1116 state = "off"; 1117 break; 1118 case GFS2_QUOTA_ACCOUNT: 1119 state = "account"; 1120 break; 1121 case GFS2_QUOTA_ON: 1122 state = "on"; 1123 break; 1124 default: 1125 state = "unknown"; 1126 break; 1127 } 1128 seq_printf(s, ",quota=%s", state); 1129 } 1130 if (args->ar_suiddir) 1131 seq_puts(s, ",suiddir"); 1132 if (args->ar_data != GFS2_DATA_DEFAULT) { 1133 char *state; 1134 switch (args->ar_data) { 1135 case GFS2_DATA_WRITEBACK: 1136 state = "writeback"; 1137 break; 1138 case GFS2_DATA_ORDERED: 1139 state = "ordered"; 1140 break; 1141 default: 1142 state = "unknown"; 1143 break; 1144 } 1145 seq_printf(s, ",data=%s", state); 1146 } 1147 if (args->ar_discard) 1148 seq_puts(s, ",discard"); 1149 val = sdp->sd_tune.gt_logd_secs; 1150 if (val != 30) 1151 seq_printf(s, ",commit=%d", val); 1152 val = sdp->sd_tune.gt_statfs_quantum; 1153 if (val != 30) 1154 seq_printf(s, ",statfs_quantum=%d", val); 1155 else if (sdp->sd_tune.gt_statfs_slow) 1156 seq_puts(s, ",statfs_quantum=0"); 1157 val = sdp->sd_tune.gt_quota_quantum; 1158 if (val != 60) 1159 seq_printf(s, ",quota_quantum=%d", val); 1160 if (args->ar_statfs_percent) 1161 seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent); 1162 if (args->ar_errors != GFS2_ERRORS_DEFAULT) { 1163 const char *state; 1164 1165 switch (args->ar_errors) { 1166 case GFS2_ERRORS_WITHDRAW: 1167 state = "withdraw"; 1168 break; 1169 case GFS2_ERRORS_PANIC: 1170 state = "panic"; 1171 break; 1172 default: 1173 state = "unknown"; 1174 break; 1175 } 1176 seq_printf(s, ",errors=%s", state); 1177 } 1178 if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags)) 1179 seq_puts(s, ",nobarrier"); 1180 if (test_bit(SDF_DEMOTE, &sdp->sd_flags)) 1181 seq_puts(s, ",demote_interface_used"); 1182 if (args->ar_rgrplvb) 1183 seq_puts(s, ",rgrplvb"); 1184 if (args->ar_loccookie) 1185 seq_puts(s, ",loccookie"); 1186 return 0; 1187 } 1188 1189 static void gfs2_final_release_pages(struct gfs2_inode *ip) 1190 { 1191 struct inode *inode = &ip->i_inode; 1192 struct gfs2_glock *gl = ip->i_gl; 1193 1194 truncate_inode_pages(gfs2_glock2aspace(ip->i_gl), 0); 1195 truncate_inode_pages(&inode->i_data, 0); 1196 1197 if (atomic_read(&gl->gl_revokes) == 0) { 1198 clear_bit(GLF_LFLUSH, &gl->gl_flags); 1199 clear_bit(GLF_DIRTY, &gl->gl_flags); 1200 } 1201 } 1202 1203 static int gfs2_dinode_dealloc(struct gfs2_inode *ip) 1204 { 1205 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); 1206 struct gfs2_rgrpd *rgd; 1207 struct gfs2_holder gh; 1208 int error; 1209 1210 if (gfs2_get_inode_blocks(&ip->i_inode) != 1) { 1211 gfs2_consist_inode(ip); 1212 return -EIO; 1213 } 1214 1215 error = gfs2_rindex_update(sdp); 1216 if (error) 1217 return error; 1218 1219 error = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE); 1220 if (error) 1221 return error; 1222 1223 rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1); 1224 if (!rgd) { 1225 gfs2_consist_inode(ip); 1226 error = -EIO; 1227 goto out_qs; 1228 } 1229 1230 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &gh); 1231 if (error) 1232 goto out_qs; 1233 1234 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA, 1235 sdp->sd_jdesc->jd_blocks); 1236 if (error) 1237 goto out_rg_gunlock; 1238 1239 gfs2_free_di(rgd, ip); 1240 1241 gfs2_final_release_pages(ip); 1242 1243 gfs2_trans_end(sdp); 1244 1245 out_rg_gunlock: 1246 gfs2_glock_dq_uninit(&gh); 1247 out_qs: 1248 gfs2_quota_unhold(ip); 1249 return error; 1250 } 1251 1252 /** 1253 * gfs2_glock_put_eventually 1254 * @gl: The glock to put 1255 * 1256 * When under memory pressure, trigger a deferred glock put to make sure we 1257 * won't call into DLM and deadlock. Otherwise, put the glock directly. 1258 */ 1259 1260 static void gfs2_glock_put_eventually(struct gfs2_glock *gl) 1261 { 1262 if (current->flags & PF_MEMALLOC) 1263 gfs2_glock_queue_put(gl); 1264 else 1265 gfs2_glock_put(gl); 1266 } 1267 1268 static bool gfs2_upgrade_iopen_glock(struct inode *inode) 1269 { 1270 struct gfs2_inode *ip = GFS2_I(inode); 1271 struct gfs2_sbd *sdp = GFS2_SB(inode); 1272 struct gfs2_holder *gh = &ip->i_iopen_gh; 1273 long timeout = 5 * HZ; 1274 int error; 1275 1276 gh->gh_flags |= GL_NOCACHE; 1277 gfs2_glock_dq_wait(gh); 1278 1279 /* 1280 * If there are no other lock holders, we'll get the lock immediately. 1281 * Otherwise, the other nodes holding the lock will be notified about 1282 * our locking request. If they don't have the inode open, they'll 1283 * evict the cached inode and release the lock. Otherwise, if they 1284 * poke the inode glock, we'll take this as an indication that they 1285 * still need the iopen glock and that they'll take care of deleting 1286 * the inode when they're done. As a last resort, if another node 1287 * keeps holding the iopen glock without showing any activity on the 1288 * inode glock, we'll eventually time out. 1289 * 1290 * Note that we're passing the LM_FLAG_TRY_1CB flag to the first 1291 * locking request as an optimization to notify lock holders as soon as 1292 * possible. Without that flag, they'd be notified implicitly by the 1293 * second locking request. 1294 */ 1295 1296 gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE, gh); 1297 error = gfs2_glock_nq(gh); 1298 if (error != GLR_TRYFAILED) 1299 return !error; 1300 1301 gfs2_holder_reinit(LM_ST_EXCLUSIVE, GL_ASYNC | GL_NOCACHE, gh); 1302 error = gfs2_glock_nq(gh); 1303 if (error) 1304 return false; 1305 1306 timeout = wait_event_interruptible_timeout(sdp->sd_async_glock_wait, 1307 !test_bit(HIF_WAIT, &gh->gh_iflags) || 1308 test_bit(GLF_DEMOTE, &ip->i_gl->gl_flags), 1309 timeout); 1310 if (!test_bit(HIF_HOLDER, &gh->gh_iflags)) { 1311 gfs2_glock_dq(gh); 1312 return false; 1313 } 1314 return true; 1315 } 1316 1317 /** 1318 * evict_should_delete - determine whether the inode is eligible for deletion 1319 * @inode: The inode to evict 1320 * 1321 * This function determines whether the evicted inode is eligible to be deleted 1322 * and locks the inode glock. 1323 * 1324 * Returns: the fate of the dinode 1325 */ 1326 static enum dinode_demise evict_should_delete(struct inode *inode, 1327 struct gfs2_holder *gh) 1328 { 1329 struct gfs2_inode *ip = GFS2_I(inode); 1330 struct super_block *sb = inode->i_sb; 1331 struct gfs2_sbd *sdp = sb->s_fs_info; 1332 int ret; 1333 1334 if (test_bit(GIF_ALLOC_FAILED, &ip->i_flags)) { 1335 BUG_ON(!gfs2_glock_is_locked_by_me(ip->i_gl)); 1336 goto should_delete; 1337 } 1338 1339 if (test_bit(GIF_DEFERRED_DELETE, &ip->i_flags)) 1340 return SHOULD_DEFER_EVICTION; 1341 1342 /* Deletes should never happen under memory pressure anymore. */ 1343 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC)) 1344 return SHOULD_DEFER_EVICTION; 1345 1346 /* Must not read inode block until block type has been verified */ 1347 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, gh); 1348 if (unlikely(ret)) { 1349 glock_clear_object(ip->i_iopen_gh.gh_gl, ip); 1350 ip->i_iopen_gh.gh_flags |= GL_NOCACHE; 1351 gfs2_glock_dq_uninit(&ip->i_iopen_gh); 1352 return SHOULD_DEFER_EVICTION; 1353 } 1354 1355 if (gfs2_inode_already_deleted(ip->i_gl, ip->i_no_formal_ino)) 1356 return SHOULD_NOT_DELETE_DINODE; 1357 ret = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED); 1358 if (ret) 1359 return SHOULD_NOT_DELETE_DINODE; 1360 1361 if (test_bit(GIF_INVALID, &ip->i_flags)) { 1362 ret = gfs2_inode_refresh(ip); 1363 if (ret) 1364 return SHOULD_NOT_DELETE_DINODE; 1365 } 1366 1367 /* 1368 * The inode may have been recreated in the meantime. 1369 */ 1370 if (inode->i_nlink) 1371 return SHOULD_NOT_DELETE_DINODE; 1372 1373 should_delete: 1374 if (gfs2_holder_initialized(&ip->i_iopen_gh) && 1375 test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) { 1376 if (!gfs2_upgrade_iopen_glock(inode)) { 1377 gfs2_holder_uninit(&ip->i_iopen_gh); 1378 return SHOULD_NOT_DELETE_DINODE; 1379 } 1380 } 1381 return SHOULD_DELETE_DINODE; 1382 } 1383 1384 /** 1385 * evict_unlinked_inode - delete the pieces of an unlinked evicted inode 1386 * @inode: The inode to evict 1387 */ 1388 static int evict_unlinked_inode(struct inode *inode) 1389 { 1390 struct gfs2_inode *ip = GFS2_I(inode); 1391 int ret; 1392 1393 if (S_ISDIR(inode->i_mode) && 1394 (ip->i_diskflags & GFS2_DIF_EXHASH)) { 1395 ret = gfs2_dir_exhash_dealloc(ip); 1396 if (ret) 1397 goto out; 1398 } 1399 1400 if (ip->i_eattr) { 1401 ret = gfs2_ea_dealloc(ip); 1402 if (ret) 1403 goto out; 1404 } 1405 1406 if (!gfs2_is_stuffed(ip)) { 1407 ret = gfs2_file_dealloc(ip); 1408 if (ret) 1409 goto out; 1410 } 1411 1412 /* We're about to clear the bitmap for the dinode, but as soon as we 1413 do, gfs2_create_inode can create another inode at the same block 1414 location and try to set gl_object again. We clear gl_object here so 1415 that subsequent inode creates don't see an old gl_object. */ 1416 glock_clear_object(ip->i_gl, ip); 1417 ret = gfs2_dinode_dealloc(ip); 1418 gfs2_inode_remember_delete(ip->i_gl, ip->i_no_formal_ino); 1419 out: 1420 return ret; 1421 } 1422 1423 /* 1424 * evict_linked_inode - evict an inode whose dinode has not been unlinked 1425 * @inode: The inode to evict 1426 */ 1427 static int evict_linked_inode(struct inode *inode) 1428 { 1429 struct super_block *sb = inode->i_sb; 1430 struct gfs2_sbd *sdp = sb->s_fs_info; 1431 struct gfs2_inode *ip = GFS2_I(inode); 1432 struct address_space *metamapping; 1433 int ret; 1434 1435 gfs2_log_flush(sdp, ip->i_gl, GFS2_LOG_HEAD_FLUSH_NORMAL | 1436 GFS2_LFC_EVICT_INODE); 1437 metamapping = gfs2_glock2aspace(ip->i_gl); 1438 if (test_bit(GLF_DIRTY, &ip->i_gl->gl_flags)) { 1439 filemap_fdatawrite(metamapping); 1440 filemap_fdatawait(metamapping); 1441 } 1442 write_inode_now(inode, 1); 1443 gfs2_ail_flush(ip->i_gl, 0); 1444 1445 ret = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks); 1446 if (ret) 1447 return ret; 1448 1449 /* Needs to be done before glock release & also in a transaction */ 1450 truncate_inode_pages(&inode->i_data, 0); 1451 truncate_inode_pages(metamapping, 0); 1452 gfs2_trans_end(sdp); 1453 return 0; 1454 } 1455 1456 /** 1457 * gfs2_evict_inode - Remove an inode from cache 1458 * @inode: The inode to evict 1459 * 1460 * There are three cases to consider: 1461 * 1. i_nlink == 0, we are final opener (and must deallocate) 1462 * 2. i_nlink == 0, we are not the final opener (and cannot deallocate) 1463 * 3. i_nlink > 0 1464 * 1465 * If the fs is read only, then we have to treat all cases as per #3 1466 * since we are unable to do any deallocation. The inode will be 1467 * deallocated by the next read/write node to attempt an allocation 1468 * in the same resource group 1469 * 1470 * We have to (at the moment) hold the inodes main lock to cover 1471 * the gap between unlocking the shared lock on the iopen lock and 1472 * taking the exclusive lock. I'd rather do a shared -> exclusive 1473 * conversion on the iopen lock, but we can change that later. This 1474 * is safe, just less efficient. 1475 */ 1476 1477 static void gfs2_evict_inode(struct inode *inode) 1478 { 1479 struct super_block *sb = inode->i_sb; 1480 struct gfs2_sbd *sdp = sb->s_fs_info; 1481 struct gfs2_inode *ip = GFS2_I(inode); 1482 struct gfs2_holder gh; 1483 int ret; 1484 1485 if (test_bit(GIF_FREE_VFS_INODE, &ip->i_flags)) { 1486 clear_inode(inode); 1487 return; 1488 } 1489 1490 if (inode->i_nlink || sb_rdonly(sb)) 1491 goto out; 1492 1493 gfs2_holder_mark_uninitialized(&gh); 1494 ret = evict_should_delete(inode, &gh); 1495 if (ret == SHOULD_DEFER_EVICTION) 1496 goto out; 1497 if (ret == SHOULD_DELETE_DINODE) 1498 ret = evict_unlinked_inode(inode); 1499 else 1500 ret = evict_linked_inode(inode); 1501 1502 if (gfs2_rs_active(&ip->i_res)) 1503 gfs2_rs_deltree(&ip->i_res); 1504 1505 if (gfs2_holder_initialized(&gh)) { 1506 glock_clear_object(ip->i_gl, ip); 1507 gfs2_glock_dq_uninit(&gh); 1508 } 1509 if (ret && ret != GLR_TRYFAILED && ret != -EROFS) 1510 fs_warn(sdp, "gfs2_evict_inode: %d\n", ret); 1511 out: 1512 truncate_inode_pages_final(&inode->i_data); 1513 if (ip->i_qadata) 1514 gfs2_assert_warn(sdp, ip->i_qadata->qa_ref == 0); 1515 gfs2_rs_delete(ip, NULL); 1516 gfs2_ordered_del_inode(ip); 1517 clear_inode(inode); 1518 gfs2_dir_hash_inval(ip); 1519 if (ip->i_gl) { 1520 glock_clear_object(ip->i_gl, ip); 1521 wait_on_bit_io(&ip->i_flags, GIF_GLOP_PENDING, TASK_UNINTERRUPTIBLE); 1522 gfs2_glock_add_to_lru(ip->i_gl); 1523 gfs2_glock_put_eventually(ip->i_gl); 1524 ip->i_gl = NULL; 1525 } 1526 if (gfs2_holder_initialized(&ip->i_iopen_gh)) { 1527 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl; 1528 1529 glock_clear_object(gl, ip); 1530 if (test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) { 1531 ip->i_iopen_gh.gh_flags |= GL_NOCACHE; 1532 gfs2_glock_dq(&ip->i_iopen_gh); 1533 } 1534 gfs2_glock_hold(gl); 1535 gfs2_holder_uninit(&ip->i_iopen_gh); 1536 gfs2_glock_put_eventually(gl); 1537 } 1538 } 1539 1540 static struct inode *gfs2_alloc_inode(struct super_block *sb) 1541 { 1542 struct gfs2_inode *ip; 1543 1544 ip = kmem_cache_alloc(gfs2_inode_cachep, GFP_KERNEL); 1545 if (!ip) 1546 return NULL; 1547 ip->i_flags = 0; 1548 ip->i_gl = NULL; 1549 gfs2_holder_mark_uninitialized(&ip->i_iopen_gh); 1550 memset(&ip->i_res, 0, sizeof(ip->i_res)); 1551 RB_CLEAR_NODE(&ip->i_res.rs_node); 1552 ip->i_rahead = 0; 1553 return &ip->i_inode; 1554 } 1555 1556 static void gfs2_free_inode(struct inode *inode) 1557 { 1558 kmem_cache_free(gfs2_inode_cachep, GFS2_I(inode)); 1559 } 1560 1561 extern void free_local_statfs_inodes(struct gfs2_sbd *sdp) 1562 { 1563 struct local_statfs_inode *lsi, *safe; 1564 1565 /* Run through the statfs inodes list to iput and free memory */ 1566 list_for_each_entry_safe(lsi, safe, &sdp->sd_sc_inodes_list, si_list) { 1567 if (lsi->si_jid == sdp->sd_jdesc->jd_jid) 1568 sdp->sd_sc_inode = NULL; /* belongs to this node */ 1569 if (lsi->si_sc_inode) 1570 iput(lsi->si_sc_inode); 1571 list_del(&lsi->si_list); 1572 kfree(lsi); 1573 } 1574 } 1575 1576 extern struct inode *find_local_statfs_inode(struct gfs2_sbd *sdp, 1577 unsigned int index) 1578 { 1579 struct local_statfs_inode *lsi; 1580 1581 /* Return the local (per node) statfs inode in the 1582 * sdp->sd_sc_inodes_list corresponding to the 'index'. */ 1583 list_for_each_entry(lsi, &sdp->sd_sc_inodes_list, si_list) { 1584 if (lsi->si_jid == index) 1585 return lsi->si_sc_inode; 1586 } 1587 return NULL; 1588 } 1589 1590 const struct super_operations gfs2_super_ops = { 1591 .alloc_inode = gfs2_alloc_inode, 1592 .free_inode = gfs2_free_inode, 1593 .write_inode = gfs2_write_inode, 1594 .dirty_inode = gfs2_dirty_inode, 1595 .evict_inode = gfs2_evict_inode, 1596 .put_super = gfs2_put_super, 1597 .sync_fs = gfs2_sync_fs, 1598 .freeze_super = gfs2_freeze, 1599 .thaw_super = gfs2_unfreeze, 1600 .statfs = gfs2_statfs, 1601 .drop_inode = gfs2_drop_inode, 1602 .show_options = gfs2_show_options, 1603 }; 1604 1605