1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. 4 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved. 5 */ 6 7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 8 9 #include <linux/sched.h> 10 #include <linux/slab.h> 11 #include <linux/spinlock.h> 12 #include <linux/completion.h> 13 #include <linux/buffer_head.h> 14 #include <linux/blkdev.h> 15 #include <linux/kthread.h> 16 #include <linux/export.h> 17 #include <linux/namei.h> 18 #include <linux/mount.h> 19 #include <linux/gfs2_ondisk.h> 20 #include <linux/quotaops.h> 21 #include <linux/lockdep.h> 22 #include <linux/module.h> 23 #include <linux/backing-dev.h> 24 #include <linux/fs_parser.h> 25 26 #include "gfs2.h" 27 #include "incore.h" 28 #include "bmap.h" 29 #include "glock.h" 30 #include "glops.h" 31 #include "inode.h" 32 #include "recovery.h" 33 #include "rgrp.h" 34 #include "super.h" 35 #include "sys.h" 36 #include "util.h" 37 #include "log.h" 38 #include "quota.h" 39 #include "dir.h" 40 #include "meta_io.h" 41 #include "trace_gfs2.h" 42 #include "lops.h" 43 44 #define DO 0 45 #define UNDO 1 46 47 /** 48 * gfs2_tune_init - Fill a gfs2_tune structure with default values 49 * @gt: tune 50 * 51 */ 52 53 static void gfs2_tune_init(struct gfs2_tune *gt) 54 { 55 spin_lock_init(>->gt_spin); 56 57 gt->gt_quota_warn_period = 10; 58 gt->gt_quota_scale_num = 1; 59 gt->gt_quota_scale_den = 1; 60 gt->gt_new_files_jdata = 0; 61 gt->gt_max_readahead = BIT(18); 62 gt->gt_complain_secs = 10; 63 } 64 65 void free_sbd(struct gfs2_sbd *sdp) 66 { 67 if (sdp->sd_lkstats) 68 free_percpu(sdp->sd_lkstats); 69 kfree(sdp); 70 } 71 72 static struct gfs2_sbd *init_sbd(struct super_block *sb) 73 { 74 struct gfs2_sbd *sdp; 75 struct address_space *mapping; 76 77 sdp = kzalloc(sizeof(struct gfs2_sbd), GFP_KERNEL); 78 if (!sdp) 79 return NULL; 80 81 sdp->sd_vfs = sb; 82 sdp->sd_lkstats = alloc_percpu(struct gfs2_pcpu_lkstats); 83 if (!sdp->sd_lkstats) 84 goto fail; 85 sb->s_fs_info = sdp; 86 87 set_bit(SDF_NOJOURNALID, &sdp->sd_flags); 88 gfs2_tune_init(&sdp->sd_tune); 89 90 init_waitqueue_head(&sdp->sd_kill_wait); 91 init_waitqueue_head(&sdp->sd_async_glock_wait); 92 atomic_set(&sdp->sd_glock_disposal, 0); 93 init_completion(&sdp->sd_locking_init); 94 init_completion(&sdp->sd_wdack); 95 spin_lock_init(&sdp->sd_statfs_spin); 96 97 spin_lock_init(&sdp->sd_rindex_spin); 98 sdp->sd_rindex_tree.rb_node = NULL; 99 100 INIT_LIST_HEAD(&sdp->sd_jindex_list); 101 spin_lock_init(&sdp->sd_jindex_spin); 102 mutex_init(&sdp->sd_jindex_mutex); 103 init_completion(&sdp->sd_journal_ready); 104 105 INIT_LIST_HEAD(&sdp->sd_quota_list); 106 mutex_init(&sdp->sd_quota_mutex); 107 mutex_init(&sdp->sd_quota_sync_mutex); 108 init_waitqueue_head(&sdp->sd_quota_wait); 109 spin_lock_init(&sdp->sd_bitmap_lock); 110 111 INIT_LIST_HEAD(&sdp->sd_sc_inodes_list); 112 113 mapping = &sdp->sd_aspace; 114 115 address_space_init_once(mapping); 116 mapping->a_ops = &gfs2_rgrp_aops; 117 mapping->host = sb->s_bdev->bd_inode; 118 mapping->flags = 0; 119 mapping_set_gfp_mask(mapping, GFP_NOFS); 120 mapping->private_data = NULL; 121 mapping->writeback_index = 0; 122 123 spin_lock_init(&sdp->sd_log_lock); 124 atomic_set(&sdp->sd_log_pinned, 0); 125 INIT_LIST_HEAD(&sdp->sd_log_revokes); 126 INIT_LIST_HEAD(&sdp->sd_log_ordered); 127 spin_lock_init(&sdp->sd_ordered_lock); 128 129 init_waitqueue_head(&sdp->sd_log_waitq); 130 init_waitqueue_head(&sdp->sd_logd_waitq); 131 spin_lock_init(&sdp->sd_ail_lock); 132 INIT_LIST_HEAD(&sdp->sd_ail1_list); 133 INIT_LIST_HEAD(&sdp->sd_ail2_list); 134 135 init_rwsem(&sdp->sd_log_flush_lock); 136 atomic_set(&sdp->sd_log_in_flight, 0); 137 init_waitqueue_head(&sdp->sd_log_flush_wait); 138 mutex_init(&sdp->sd_freeze_mutex); 139 INIT_LIST_HEAD(&sdp->sd_dead_glocks); 140 141 return sdp; 142 143 fail: 144 free_sbd(sdp); 145 return NULL; 146 } 147 148 /** 149 * gfs2_check_sb - Check superblock 150 * @sdp: the filesystem 151 * @silent: Don't print a message if the check fails 152 * 153 * Checks the version code of the FS is one that we understand how to 154 * read and that the sizes of the various on-disk structures have not 155 * changed. 156 */ 157 158 static int gfs2_check_sb(struct gfs2_sbd *sdp, int silent) 159 { 160 struct gfs2_sb_host *sb = &sdp->sd_sb; 161 162 if (sb->sb_magic != GFS2_MAGIC || 163 sb->sb_type != GFS2_METATYPE_SB) { 164 if (!silent) 165 pr_warn("not a GFS2 filesystem\n"); 166 return -EINVAL; 167 } 168 169 if (sb->sb_fs_format < GFS2_FS_FORMAT_MIN || 170 sb->sb_fs_format > GFS2_FS_FORMAT_MAX || 171 sb->sb_multihost_format != GFS2_FORMAT_MULTI) { 172 fs_warn(sdp, "Unknown on-disk format, unable to mount\n"); 173 return -EINVAL; 174 } 175 176 if (sb->sb_bsize < 512 || sb->sb_bsize > PAGE_SIZE || 177 (sb->sb_bsize & (sb->sb_bsize - 1))) { 178 pr_warn("Invalid block size\n"); 179 return -EINVAL; 180 } 181 if (sb->sb_bsize_shift != ffs(sb->sb_bsize) - 1) { 182 pr_warn("Invalid block size shift\n"); 183 return -EINVAL; 184 } 185 return 0; 186 } 187 188 static void end_bio_io_page(struct bio *bio) 189 { 190 struct page *page = bio->bi_private; 191 192 if (!bio->bi_status) 193 SetPageUptodate(page); 194 else 195 pr_warn("error %d reading superblock\n", bio->bi_status); 196 unlock_page(page); 197 } 198 199 static void gfs2_sb_in(struct gfs2_sbd *sdp, const void *buf) 200 { 201 struct gfs2_sb_host *sb = &sdp->sd_sb; 202 struct super_block *s = sdp->sd_vfs; 203 const struct gfs2_sb *str = buf; 204 205 sb->sb_magic = be32_to_cpu(str->sb_header.mh_magic); 206 sb->sb_type = be32_to_cpu(str->sb_header.mh_type); 207 sb->sb_fs_format = be32_to_cpu(str->sb_fs_format); 208 sb->sb_multihost_format = be32_to_cpu(str->sb_multihost_format); 209 sb->sb_bsize = be32_to_cpu(str->sb_bsize); 210 sb->sb_bsize_shift = be32_to_cpu(str->sb_bsize_shift); 211 sb->sb_master_dir.no_addr = be64_to_cpu(str->sb_master_dir.no_addr); 212 sb->sb_master_dir.no_formal_ino = be64_to_cpu(str->sb_master_dir.no_formal_ino); 213 sb->sb_root_dir.no_addr = be64_to_cpu(str->sb_root_dir.no_addr); 214 sb->sb_root_dir.no_formal_ino = be64_to_cpu(str->sb_root_dir.no_formal_ino); 215 216 memcpy(sb->sb_lockproto, str->sb_lockproto, GFS2_LOCKNAME_LEN); 217 memcpy(sb->sb_locktable, str->sb_locktable, GFS2_LOCKNAME_LEN); 218 memcpy(&s->s_uuid, str->sb_uuid, 16); 219 } 220 221 /** 222 * gfs2_read_super - Read the gfs2 super block from disk 223 * @sdp: The GFS2 super block 224 * @sector: The location of the super block 225 * @silent: Don't print a message if the check fails 226 * 227 * This uses the bio functions to read the super block from disk 228 * because we want to be 100% sure that we never read cached data. 229 * A super block is read twice only during each GFS2 mount and is 230 * never written to by the filesystem. The first time its read no 231 * locks are held, and the only details which are looked at are those 232 * relating to the locking protocol. Once locking is up and working, 233 * the sb is read again under the lock to establish the location of 234 * the master directory (contains pointers to journals etc) and the 235 * root directory. 236 * 237 * Returns: 0 on success or error 238 */ 239 240 static int gfs2_read_super(struct gfs2_sbd *sdp, sector_t sector, int silent) 241 { 242 struct super_block *sb = sdp->sd_vfs; 243 struct gfs2_sb *p; 244 struct page *page; 245 struct bio *bio; 246 247 page = alloc_page(GFP_NOFS); 248 if (unlikely(!page)) 249 return -ENOMEM; 250 251 ClearPageUptodate(page); 252 ClearPageDirty(page); 253 lock_page(page); 254 255 bio = bio_alloc(sb->s_bdev, 1, REQ_OP_READ | REQ_META, GFP_NOFS); 256 bio->bi_iter.bi_sector = sector * (sb->s_blocksize >> 9); 257 __bio_add_page(bio, page, PAGE_SIZE, 0); 258 259 bio->bi_end_io = end_bio_io_page; 260 bio->bi_private = page; 261 submit_bio(bio); 262 wait_on_page_locked(page); 263 bio_put(bio); 264 if (!PageUptodate(page)) { 265 __free_page(page); 266 return -EIO; 267 } 268 p = kmap(page); 269 gfs2_sb_in(sdp, p); 270 kunmap(page); 271 __free_page(page); 272 return gfs2_check_sb(sdp, silent); 273 } 274 275 /** 276 * gfs2_read_sb - Read super block 277 * @sdp: The GFS2 superblock 278 * @silent: Don't print message if mount fails 279 * 280 */ 281 282 static int gfs2_read_sb(struct gfs2_sbd *sdp, int silent) 283 { 284 u32 hash_blocks, ind_blocks, leaf_blocks; 285 u32 tmp_blocks; 286 unsigned int x; 287 int error; 288 289 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent); 290 if (error) { 291 if (!silent) 292 fs_err(sdp, "can't read superblock\n"); 293 return error; 294 } 295 296 sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift - 297 GFS2_BASIC_BLOCK_SHIFT; 298 sdp->sd_fsb2bb = BIT(sdp->sd_fsb2bb_shift); 299 sdp->sd_diptrs = (sdp->sd_sb.sb_bsize - 300 sizeof(struct gfs2_dinode)) / sizeof(u64); 301 sdp->sd_inptrs = (sdp->sd_sb.sb_bsize - 302 sizeof(struct gfs2_meta_header)) / sizeof(u64); 303 sdp->sd_ldptrs = (sdp->sd_sb.sb_bsize - 304 sizeof(struct gfs2_log_descriptor)) / sizeof(u64); 305 sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header); 306 sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2; 307 sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1; 308 sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64); 309 sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize - 310 sizeof(struct gfs2_meta_header)) / 311 sizeof(struct gfs2_quota_change); 312 sdp->sd_blocks_per_bitmap = (sdp->sd_sb.sb_bsize - 313 sizeof(struct gfs2_meta_header)) 314 * GFS2_NBBY; /* not the rgrp bitmap, subsequent bitmaps only */ 315 316 /* 317 * We always keep at least one block reserved for revokes in 318 * transactions. This greatly simplifies allocating additional 319 * revoke blocks. 320 */ 321 atomic_set(&sdp->sd_log_revokes_available, sdp->sd_ldptrs); 322 323 /* Compute maximum reservation required to add a entry to a directory */ 324 325 hash_blocks = DIV_ROUND_UP(sizeof(u64) * BIT(GFS2_DIR_MAX_DEPTH), 326 sdp->sd_jbsize); 327 328 ind_blocks = 0; 329 for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) { 330 tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs); 331 ind_blocks += tmp_blocks; 332 } 333 334 leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH; 335 336 sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks; 337 338 sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize - 339 sizeof(struct gfs2_dinode); 340 sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs; 341 for (x = 2;; x++) { 342 u64 space, d; 343 u32 m; 344 345 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs; 346 d = space; 347 m = do_div(d, sdp->sd_inptrs); 348 349 if (d != sdp->sd_heightsize[x - 1] || m) 350 break; 351 sdp->sd_heightsize[x] = space; 352 } 353 sdp->sd_max_height = x; 354 sdp->sd_heightsize[x] = ~0; 355 gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT); 356 357 sdp->sd_max_dents_per_leaf = (sdp->sd_sb.sb_bsize - 358 sizeof(struct gfs2_leaf)) / 359 GFS2_MIN_DIRENT_SIZE; 360 return 0; 361 } 362 363 static int init_names(struct gfs2_sbd *sdp, int silent) 364 { 365 char *proto, *table; 366 int error = 0; 367 368 proto = sdp->sd_args.ar_lockproto; 369 table = sdp->sd_args.ar_locktable; 370 371 /* Try to autodetect */ 372 373 if (!proto[0] || !table[0]) { 374 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent); 375 if (error) 376 return error; 377 378 if (!proto[0]) 379 proto = sdp->sd_sb.sb_lockproto; 380 if (!table[0]) 381 table = sdp->sd_sb.sb_locktable; 382 } 383 384 if (!table[0]) 385 table = sdp->sd_vfs->s_id; 386 387 BUILD_BUG_ON(GFS2_LOCKNAME_LEN > GFS2_FSNAME_LEN); 388 389 strscpy(sdp->sd_proto_name, proto, GFS2_LOCKNAME_LEN); 390 strscpy(sdp->sd_table_name, table, GFS2_LOCKNAME_LEN); 391 392 table = sdp->sd_table_name; 393 while ((table = strchr(table, '/'))) 394 *table = '_'; 395 396 return error; 397 } 398 399 static int init_locking(struct gfs2_sbd *sdp, struct gfs2_holder *mount_gh, 400 int undo) 401 { 402 int error = 0; 403 404 if (undo) 405 goto fail_trans; 406 407 error = gfs2_glock_nq_num(sdp, 408 GFS2_MOUNT_LOCK, &gfs2_nondisk_glops, 409 LM_ST_EXCLUSIVE, 410 LM_FLAG_NOEXP | GL_NOCACHE | GL_NOPID, 411 mount_gh); 412 if (error) { 413 fs_err(sdp, "can't acquire mount glock: %d\n", error); 414 goto fail; 415 } 416 417 error = gfs2_glock_nq_num(sdp, 418 GFS2_LIVE_LOCK, &gfs2_nondisk_glops, 419 LM_ST_SHARED, 420 LM_FLAG_NOEXP | GL_EXACT | GL_NOPID, 421 &sdp->sd_live_gh); 422 if (error) { 423 fs_err(sdp, "can't acquire live glock: %d\n", error); 424 goto fail_mount; 425 } 426 427 error = gfs2_glock_get(sdp, GFS2_RENAME_LOCK, &gfs2_nondisk_glops, 428 CREATE, &sdp->sd_rename_gl); 429 if (error) { 430 fs_err(sdp, "can't create rename glock: %d\n", error); 431 goto fail_live; 432 } 433 434 error = gfs2_glock_get(sdp, GFS2_FREEZE_LOCK, &gfs2_freeze_glops, 435 CREATE, &sdp->sd_freeze_gl); 436 if (error) { 437 fs_err(sdp, "can't create freeze glock: %d\n", error); 438 goto fail_rename; 439 } 440 441 return 0; 442 443 fail_trans: 444 gfs2_glock_put(sdp->sd_freeze_gl); 445 fail_rename: 446 gfs2_glock_put(sdp->sd_rename_gl); 447 fail_live: 448 gfs2_glock_dq_uninit(&sdp->sd_live_gh); 449 fail_mount: 450 gfs2_glock_dq_uninit(mount_gh); 451 fail: 452 return error; 453 } 454 455 static int gfs2_lookup_root(struct super_block *sb, struct dentry **dptr, 456 u64 no_addr, const char *name) 457 { 458 struct gfs2_sbd *sdp = sb->s_fs_info; 459 struct dentry *dentry; 460 struct inode *inode; 461 462 inode = gfs2_inode_lookup(sb, DT_DIR, no_addr, 0, 463 GFS2_BLKST_FREE /* ignore */); 464 if (IS_ERR(inode)) { 465 fs_err(sdp, "can't read in %s inode: %ld\n", name, PTR_ERR(inode)); 466 return PTR_ERR(inode); 467 } 468 dentry = d_make_root(inode); 469 if (!dentry) { 470 fs_err(sdp, "can't alloc %s dentry\n", name); 471 return -ENOMEM; 472 } 473 *dptr = dentry; 474 return 0; 475 } 476 477 static int init_sb(struct gfs2_sbd *sdp, int silent) 478 { 479 struct super_block *sb = sdp->sd_vfs; 480 struct gfs2_holder sb_gh; 481 u64 no_addr; 482 int ret; 483 484 ret = gfs2_glock_nq_num(sdp, GFS2_SB_LOCK, &gfs2_meta_glops, 485 LM_ST_SHARED, 0, &sb_gh); 486 if (ret) { 487 fs_err(sdp, "can't acquire superblock glock: %d\n", ret); 488 return ret; 489 } 490 491 ret = gfs2_read_sb(sdp, silent); 492 if (ret) { 493 fs_err(sdp, "can't read superblock: %d\n", ret); 494 goto out; 495 } 496 497 switch(sdp->sd_sb.sb_fs_format) { 498 case GFS2_FS_FORMAT_MAX: 499 sb->s_xattr = gfs2_xattr_handlers_max; 500 break; 501 502 case GFS2_FS_FORMAT_MIN: 503 sb->s_xattr = gfs2_xattr_handlers_min; 504 break; 505 506 default: 507 BUG(); 508 } 509 510 /* Set up the buffer cache and SB for real */ 511 if (sdp->sd_sb.sb_bsize < bdev_logical_block_size(sb->s_bdev)) { 512 ret = -EINVAL; 513 fs_err(sdp, "FS block size (%u) is too small for device " 514 "block size (%u)\n", 515 sdp->sd_sb.sb_bsize, bdev_logical_block_size(sb->s_bdev)); 516 goto out; 517 } 518 if (sdp->sd_sb.sb_bsize > PAGE_SIZE) { 519 ret = -EINVAL; 520 fs_err(sdp, "FS block size (%u) is too big for machine " 521 "page size (%u)\n", 522 sdp->sd_sb.sb_bsize, (unsigned int)PAGE_SIZE); 523 goto out; 524 } 525 sb_set_blocksize(sb, sdp->sd_sb.sb_bsize); 526 527 /* Get the root inode */ 528 no_addr = sdp->sd_sb.sb_root_dir.no_addr; 529 ret = gfs2_lookup_root(sb, &sdp->sd_root_dir, no_addr, "root"); 530 if (ret) 531 goto out; 532 533 /* Get the master inode */ 534 no_addr = sdp->sd_sb.sb_master_dir.no_addr; 535 ret = gfs2_lookup_root(sb, &sdp->sd_master_dir, no_addr, "master"); 536 if (ret) { 537 dput(sdp->sd_root_dir); 538 goto out; 539 } 540 sb->s_root = dget(sdp->sd_args.ar_meta ? sdp->sd_master_dir : sdp->sd_root_dir); 541 out: 542 gfs2_glock_dq_uninit(&sb_gh); 543 return ret; 544 } 545 546 static void gfs2_others_may_mount(struct gfs2_sbd *sdp) 547 { 548 char *message = "FIRSTMOUNT=Done"; 549 char *envp[] = { message, NULL }; 550 551 fs_info(sdp, "first mount done, others may mount\n"); 552 553 if (sdp->sd_lockstruct.ls_ops->lm_first_done) 554 sdp->sd_lockstruct.ls_ops->lm_first_done(sdp); 555 556 kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp); 557 } 558 559 /** 560 * gfs2_jindex_hold - Grab a lock on the jindex 561 * @sdp: The GFS2 superblock 562 * @ji_gh: the holder for the jindex glock 563 * 564 * Returns: errno 565 */ 566 567 static int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh) 568 { 569 struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex); 570 struct qstr name; 571 char buf[20]; 572 struct gfs2_jdesc *jd; 573 int error; 574 575 name.name = buf; 576 577 mutex_lock(&sdp->sd_jindex_mutex); 578 579 for (;;) { 580 struct gfs2_inode *jip; 581 582 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh); 583 if (error) 584 break; 585 586 name.len = sprintf(buf, "journal%u", sdp->sd_journals); 587 name.hash = gfs2_disk_hash(name.name, name.len); 588 589 error = gfs2_dir_check(sdp->sd_jindex, &name, NULL); 590 if (error == -ENOENT) { 591 error = 0; 592 break; 593 } 594 595 gfs2_glock_dq_uninit(ji_gh); 596 597 if (error) 598 break; 599 600 error = -ENOMEM; 601 jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL); 602 if (!jd) 603 break; 604 605 INIT_LIST_HEAD(&jd->extent_list); 606 INIT_LIST_HEAD(&jd->jd_revoke_list); 607 608 INIT_WORK(&jd->jd_work, gfs2_recover_func); 609 jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1); 610 if (IS_ERR_OR_NULL(jd->jd_inode)) { 611 if (!jd->jd_inode) 612 error = -ENOENT; 613 else 614 error = PTR_ERR(jd->jd_inode); 615 kfree(jd); 616 break; 617 } 618 619 d_mark_dontcache(jd->jd_inode); 620 spin_lock(&sdp->sd_jindex_spin); 621 jd->jd_jid = sdp->sd_journals++; 622 jip = GFS2_I(jd->jd_inode); 623 jd->jd_no_addr = jip->i_no_addr; 624 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list); 625 spin_unlock(&sdp->sd_jindex_spin); 626 } 627 628 mutex_unlock(&sdp->sd_jindex_mutex); 629 630 return error; 631 } 632 633 /** 634 * init_statfs - look up and initialize master and local (per node) statfs inodes 635 * @sdp: The GFS2 superblock 636 * 637 * This should be called after the jindex is initialized in init_journal() and 638 * before gfs2_journal_recovery() is called because we need to be able to write 639 * to these inodes during recovery. 640 * 641 * Returns: errno 642 */ 643 static int init_statfs(struct gfs2_sbd *sdp) 644 { 645 int error = 0; 646 struct inode *master = d_inode(sdp->sd_master_dir); 647 struct inode *pn = NULL; 648 char buf[30]; 649 struct gfs2_jdesc *jd; 650 struct gfs2_inode *ip; 651 652 sdp->sd_statfs_inode = gfs2_lookup_meta(master, "statfs"); 653 if (IS_ERR(sdp->sd_statfs_inode)) { 654 error = PTR_ERR(sdp->sd_statfs_inode); 655 fs_err(sdp, "can't read in statfs inode: %d\n", error); 656 goto out; 657 } 658 if (sdp->sd_args.ar_spectator) 659 goto out; 660 661 pn = gfs2_lookup_meta(master, "per_node"); 662 if (IS_ERR(pn)) { 663 error = PTR_ERR(pn); 664 fs_err(sdp, "can't find per_node directory: %d\n", error); 665 goto put_statfs; 666 } 667 668 /* For each jid, lookup the corresponding local statfs inode in the 669 * per_node metafs directory and save it in the sdp->sd_sc_inodes_list. */ 670 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) { 671 struct local_statfs_inode *lsi = 672 kmalloc(sizeof(struct local_statfs_inode), GFP_NOFS); 673 if (!lsi) { 674 error = -ENOMEM; 675 goto free_local; 676 } 677 sprintf(buf, "statfs_change%u", jd->jd_jid); 678 lsi->si_sc_inode = gfs2_lookup_meta(pn, buf); 679 if (IS_ERR(lsi->si_sc_inode)) { 680 error = PTR_ERR(lsi->si_sc_inode); 681 fs_err(sdp, "can't find local \"sc\" file#%u: %d\n", 682 jd->jd_jid, error); 683 kfree(lsi); 684 goto free_local; 685 } 686 lsi->si_jid = jd->jd_jid; 687 if (jd->jd_jid == sdp->sd_jdesc->jd_jid) 688 sdp->sd_sc_inode = lsi->si_sc_inode; 689 690 list_add_tail(&lsi->si_list, &sdp->sd_sc_inodes_list); 691 } 692 693 iput(pn); 694 pn = NULL; 695 ip = GFS2_I(sdp->sd_sc_inode); 696 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_NOPID, 697 &sdp->sd_sc_gh); 698 if (error) { 699 fs_err(sdp, "can't lock local \"sc\" file: %d\n", error); 700 goto free_local; 701 } 702 /* read in the local statfs buffer - other nodes don't change it. */ 703 error = gfs2_meta_inode_buffer(ip, &sdp->sd_sc_bh); 704 if (error) { 705 fs_err(sdp, "Cannot read in local statfs: %d\n", error); 706 goto unlock_sd_gh; 707 } 708 return 0; 709 710 unlock_sd_gh: 711 gfs2_glock_dq_uninit(&sdp->sd_sc_gh); 712 free_local: 713 free_local_statfs_inodes(sdp); 714 iput(pn); 715 put_statfs: 716 iput(sdp->sd_statfs_inode); 717 out: 718 return error; 719 } 720 721 /* Uninitialize and free up memory used by the list of statfs inodes */ 722 static void uninit_statfs(struct gfs2_sbd *sdp) 723 { 724 if (!sdp->sd_args.ar_spectator) { 725 brelse(sdp->sd_sc_bh); 726 gfs2_glock_dq_uninit(&sdp->sd_sc_gh); 727 free_local_statfs_inodes(sdp); 728 } 729 iput(sdp->sd_statfs_inode); 730 } 731 732 static int init_journal(struct gfs2_sbd *sdp, int undo) 733 { 734 struct inode *master = d_inode(sdp->sd_master_dir); 735 struct gfs2_holder ji_gh; 736 struct gfs2_inode *ip; 737 int error = 0; 738 739 gfs2_holder_mark_uninitialized(&ji_gh); 740 if (undo) 741 goto fail_statfs; 742 743 sdp->sd_jindex = gfs2_lookup_meta(master, "jindex"); 744 if (IS_ERR(sdp->sd_jindex)) { 745 fs_err(sdp, "can't lookup journal index: %d\n", error); 746 return PTR_ERR(sdp->sd_jindex); 747 } 748 749 /* Load in the journal index special file */ 750 751 error = gfs2_jindex_hold(sdp, &ji_gh); 752 if (error) { 753 fs_err(sdp, "can't read journal index: %d\n", error); 754 goto fail; 755 } 756 757 error = -EUSERS; 758 if (!gfs2_jindex_size(sdp)) { 759 fs_err(sdp, "no journals!\n"); 760 goto fail_jindex; 761 } 762 763 atomic_set(&sdp->sd_log_blks_needed, 0); 764 if (sdp->sd_args.ar_spectator) { 765 sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0); 766 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks); 767 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5); 768 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5); 769 } else { 770 if (sdp->sd_lockstruct.ls_jid >= gfs2_jindex_size(sdp)) { 771 fs_err(sdp, "can't mount journal #%u\n", 772 sdp->sd_lockstruct.ls_jid); 773 fs_err(sdp, "there are only %u journals (0 - %u)\n", 774 gfs2_jindex_size(sdp), 775 gfs2_jindex_size(sdp) - 1); 776 goto fail_jindex; 777 } 778 sdp->sd_jdesc = gfs2_jdesc_find(sdp, sdp->sd_lockstruct.ls_jid); 779 780 error = gfs2_glock_nq_num(sdp, sdp->sd_lockstruct.ls_jid, 781 &gfs2_journal_glops, 782 LM_ST_EXCLUSIVE, 783 LM_FLAG_NOEXP | GL_NOCACHE | GL_NOPID, 784 &sdp->sd_journal_gh); 785 if (error) { 786 fs_err(sdp, "can't acquire journal glock: %d\n", error); 787 goto fail_jindex; 788 } 789 790 ip = GFS2_I(sdp->sd_jdesc->jd_inode); 791 sdp->sd_jinode_gl = ip->i_gl; 792 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 793 LM_FLAG_NOEXP | GL_EXACT | 794 GL_NOCACHE | GL_NOPID, 795 &sdp->sd_jinode_gh); 796 if (error) { 797 fs_err(sdp, "can't acquire journal inode glock: %d\n", 798 error); 799 goto fail_journal_gh; 800 } 801 802 error = gfs2_jdesc_check(sdp->sd_jdesc); 803 if (error) { 804 fs_err(sdp, "my journal (%u) is bad: %d\n", 805 sdp->sd_jdesc->jd_jid, error); 806 goto fail_jinode_gh; 807 } 808 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks); 809 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5); 810 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5); 811 812 /* Map the extents for this journal's blocks */ 813 gfs2_map_journal_extents(sdp, sdp->sd_jdesc); 814 } 815 trace_gfs2_log_blocks(sdp, atomic_read(&sdp->sd_log_blks_free)); 816 817 /* Lookup statfs inodes here so journal recovery can use them. */ 818 error = init_statfs(sdp); 819 if (error) 820 goto fail_jinode_gh; 821 822 if (sdp->sd_lockstruct.ls_first) { 823 unsigned int x; 824 for (x = 0; x < sdp->sd_journals; x++) { 825 struct gfs2_jdesc *jd = gfs2_jdesc_find(sdp, x); 826 827 if (sdp->sd_args.ar_spectator) { 828 error = check_journal_clean(sdp, jd, true); 829 if (error) 830 goto fail_statfs; 831 continue; 832 } 833 error = gfs2_recover_journal(jd, true); 834 if (error) { 835 fs_err(sdp, "error recovering journal %u: %d\n", 836 x, error); 837 goto fail_statfs; 838 } 839 } 840 841 gfs2_others_may_mount(sdp); 842 } else if (!sdp->sd_args.ar_spectator) { 843 error = gfs2_recover_journal(sdp->sd_jdesc, true); 844 if (error) { 845 fs_err(sdp, "error recovering my journal: %d\n", error); 846 goto fail_statfs; 847 } 848 } 849 850 sdp->sd_log_idle = 1; 851 set_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags); 852 gfs2_glock_dq_uninit(&ji_gh); 853 INIT_WORK(&sdp->sd_freeze_work, gfs2_freeze_func); 854 return 0; 855 856 fail_statfs: 857 uninit_statfs(sdp); 858 fail_jinode_gh: 859 /* A withdraw may have done dq/uninit so now we need to check it */ 860 if (!sdp->sd_args.ar_spectator && 861 gfs2_holder_initialized(&sdp->sd_jinode_gh)) 862 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh); 863 fail_journal_gh: 864 if (!sdp->sd_args.ar_spectator && 865 gfs2_holder_initialized(&sdp->sd_journal_gh)) 866 gfs2_glock_dq_uninit(&sdp->sd_journal_gh); 867 fail_jindex: 868 gfs2_jindex_free(sdp); 869 if (gfs2_holder_initialized(&ji_gh)) 870 gfs2_glock_dq_uninit(&ji_gh); 871 fail: 872 iput(sdp->sd_jindex); 873 return error; 874 } 875 876 static struct lock_class_key gfs2_quota_imutex_key; 877 878 static int init_inodes(struct gfs2_sbd *sdp, int undo) 879 { 880 int error = 0; 881 struct inode *master = d_inode(sdp->sd_master_dir); 882 883 if (undo) 884 goto fail_qinode; 885 886 error = init_journal(sdp, undo); 887 complete_all(&sdp->sd_journal_ready); 888 if (error) 889 goto fail; 890 891 /* Read in the resource index inode */ 892 sdp->sd_rindex = gfs2_lookup_meta(master, "rindex"); 893 if (IS_ERR(sdp->sd_rindex)) { 894 error = PTR_ERR(sdp->sd_rindex); 895 fs_err(sdp, "can't get resource index inode: %d\n", error); 896 goto fail_journal; 897 } 898 sdp->sd_rindex_uptodate = 0; 899 900 /* Read in the quota inode */ 901 sdp->sd_quota_inode = gfs2_lookup_meta(master, "quota"); 902 if (IS_ERR(sdp->sd_quota_inode)) { 903 error = PTR_ERR(sdp->sd_quota_inode); 904 fs_err(sdp, "can't get quota file inode: %d\n", error); 905 goto fail_rindex; 906 } 907 /* 908 * i_rwsem on quota files is special. Since this inode is hidden system 909 * file, we are safe to define locking ourselves. 910 */ 911 lockdep_set_class(&sdp->sd_quota_inode->i_rwsem, 912 &gfs2_quota_imutex_key); 913 914 error = gfs2_rindex_update(sdp); 915 if (error) 916 goto fail_qinode; 917 918 return 0; 919 920 fail_qinode: 921 iput(sdp->sd_quota_inode); 922 fail_rindex: 923 gfs2_clear_rgrpd(sdp); 924 iput(sdp->sd_rindex); 925 fail_journal: 926 init_journal(sdp, UNDO); 927 fail: 928 return error; 929 } 930 931 static int init_per_node(struct gfs2_sbd *sdp, int undo) 932 { 933 struct inode *pn = NULL; 934 char buf[30]; 935 int error = 0; 936 struct gfs2_inode *ip; 937 struct inode *master = d_inode(sdp->sd_master_dir); 938 939 if (sdp->sd_args.ar_spectator) 940 return 0; 941 942 if (undo) 943 goto fail_qc_gh; 944 945 pn = gfs2_lookup_meta(master, "per_node"); 946 if (IS_ERR(pn)) { 947 error = PTR_ERR(pn); 948 fs_err(sdp, "can't find per_node directory: %d\n", error); 949 return error; 950 } 951 952 sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid); 953 sdp->sd_qc_inode = gfs2_lookup_meta(pn, buf); 954 if (IS_ERR(sdp->sd_qc_inode)) { 955 error = PTR_ERR(sdp->sd_qc_inode); 956 fs_err(sdp, "can't find local \"qc\" file: %d\n", error); 957 goto fail_ut_i; 958 } 959 960 iput(pn); 961 pn = NULL; 962 963 ip = GFS2_I(sdp->sd_qc_inode); 964 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_NOPID, 965 &sdp->sd_qc_gh); 966 if (error) { 967 fs_err(sdp, "can't lock local \"qc\" file: %d\n", error); 968 goto fail_qc_i; 969 } 970 971 return 0; 972 973 fail_qc_gh: 974 gfs2_glock_dq_uninit(&sdp->sd_qc_gh); 975 fail_qc_i: 976 iput(sdp->sd_qc_inode); 977 fail_ut_i: 978 iput(pn); 979 return error; 980 } 981 982 static const match_table_t nolock_tokens = { 983 { Opt_jid, "jid=%d", }, 984 { Opt_err, NULL }, 985 }; 986 987 static const struct lm_lockops nolock_ops = { 988 .lm_proto_name = "lock_nolock", 989 .lm_put_lock = gfs2_glock_free, 990 .lm_tokens = &nolock_tokens, 991 }; 992 993 /** 994 * gfs2_lm_mount - mount a locking protocol 995 * @sdp: the filesystem 996 * @silent: if 1, don't complain if the FS isn't a GFS2 fs 997 * 998 * Returns: errno 999 */ 1000 1001 static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent) 1002 { 1003 const struct lm_lockops *lm; 1004 struct lm_lockstruct *ls = &sdp->sd_lockstruct; 1005 struct gfs2_args *args = &sdp->sd_args; 1006 const char *proto = sdp->sd_proto_name; 1007 const char *table = sdp->sd_table_name; 1008 char *o, *options; 1009 int ret; 1010 1011 if (!strcmp("lock_nolock", proto)) { 1012 lm = &nolock_ops; 1013 sdp->sd_args.ar_localflocks = 1; 1014 #ifdef CONFIG_GFS2_FS_LOCKING_DLM 1015 } else if (!strcmp("lock_dlm", proto)) { 1016 lm = &gfs2_dlm_ops; 1017 #endif 1018 } else { 1019 pr_info("can't find protocol %s\n", proto); 1020 return -ENOENT; 1021 } 1022 1023 fs_info(sdp, "Trying to join cluster \"%s\", \"%s\"\n", proto, table); 1024 1025 ls->ls_ops = lm; 1026 ls->ls_first = 1; 1027 1028 for (options = args->ar_hostdata; (o = strsep(&options, ":")); ) { 1029 substring_t tmp[MAX_OPT_ARGS]; 1030 int token, option; 1031 1032 if (!o || !*o) 1033 continue; 1034 1035 token = match_token(o, *lm->lm_tokens, tmp); 1036 switch (token) { 1037 case Opt_jid: 1038 ret = match_int(&tmp[0], &option); 1039 if (ret || option < 0) 1040 goto hostdata_error; 1041 if (test_and_clear_bit(SDF_NOJOURNALID, &sdp->sd_flags)) 1042 ls->ls_jid = option; 1043 break; 1044 case Opt_id: 1045 case Opt_nodir: 1046 /* Obsolete, but left for backward compat purposes */ 1047 break; 1048 case Opt_first: 1049 ret = match_int(&tmp[0], &option); 1050 if (ret || (option != 0 && option != 1)) 1051 goto hostdata_error; 1052 ls->ls_first = option; 1053 break; 1054 case Opt_err: 1055 default: 1056 hostdata_error: 1057 fs_info(sdp, "unknown hostdata (%s)\n", o); 1058 return -EINVAL; 1059 } 1060 } 1061 1062 if (lm->lm_mount == NULL) { 1063 fs_info(sdp, "Now mounting FS (format %u)...\n", sdp->sd_sb.sb_fs_format); 1064 complete_all(&sdp->sd_locking_init); 1065 return 0; 1066 } 1067 ret = lm->lm_mount(sdp, table); 1068 if (ret == 0) 1069 fs_info(sdp, "Joined cluster. Now mounting FS (format %u)...\n", 1070 sdp->sd_sb.sb_fs_format); 1071 complete_all(&sdp->sd_locking_init); 1072 return ret; 1073 } 1074 1075 void gfs2_lm_unmount(struct gfs2_sbd *sdp) 1076 { 1077 const struct lm_lockops *lm = sdp->sd_lockstruct.ls_ops; 1078 if (!gfs2_withdrawing_or_withdrawn(sdp) && lm->lm_unmount) 1079 lm->lm_unmount(sdp); 1080 } 1081 1082 static int wait_on_journal(struct gfs2_sbd *sdp) 1083 { 1084 if (sdp->sd_lockstruct.ls_ops->lm_mount == NULL) 1085 return 0; 1086 1087 return wait_on_bit(&sdp->sd_flags, SDF_NOJOURNALID, TASK_INTERRUPTIBLE) 1088 ? -EINTR : 0; 1089 } 1090 1091 void gfs2_online_uevent(struct gfs2_sbd *sdp) 1092 { 1093 struct super_block *sb = sdp->sd_vfs; 1094 char ro[20]; 1095 char spectator[20]; 1096 char *envp[] = { ro, spectator, NULL }; 1097 sprintf(ro, "RDONLY=%d", sb_rdonly(sb)); 1098 sprintf(spectator, "SPECTATOR=%d", sdp->sd_args.ar_spectator ? 1 : 0); 1099 kobject_uevent_env(&sdp->sd_kobj, KOBJ_ONLINE, envp); 1100 } 1101 1102 static int init_threads(struct gfs2_sbd *sdp) 1103 { 1104 struct task_struct *p; 1105 int error = 0; 1106 1107 p = kthread_create(gfs2_logd, sdp, "gfs2_logd/%s", sdp->sd_fsname); 1108 if (IS_ERR(p)) { 1109 error = PTR_ERR(p); 1110 fs_err(sdp, "can't create logd thread: %d\n", error); 1111 return error; 1112 } 1113 get_task_struct(p); 1114 sdp->sd_logd_process = p; 1115 1116 p = kthread_create(gfs2_quotad, sdp, "gfs2_quotad/%s", sdp->sd_fsname); 1117 if (IS_ERR(p)) { 1118 error = PTR_ERR(p); 1119 fs_err(sdp, "can't create quotad thread: %d\n", error); 1120 goto fail; 1121 } 1122 get_task_struct(p); 1123 sdp->sd_quotad_process = p; 1124 1125 wake_up_process(sdp->sd_logd_process); 1126 wake_up_process(sdp->sd_quotad_process); 1127 return 0; 1128 1129 fail: 1130 kthread_stop(sdp->sd_logd_process); 1131 put_task_struct(sdp->sd_logd_process); 1132 sdp->sd_logd_process = NULL; 1133 return error; 1134 } 1135 1136 void gfs2_destroy_threads(struct gfs2_sbd *sdp) 1137 { 1138 if (sdp->sd_logd_process) { 1139 kthread_stop(sdp->sd_logd_process); 1140 put_task_struct(sdp->sd_logd_process); 1141 sdp->sd_logd_process = NULL; 1142 } 1143 if (sdp->sd_quotad_process) { 1144 kthread_stop(sdp->sd_quotad_process); 1145 put_task_struct(sdp->sd_quotad_process); 1146 sdp->sd_quotad_process = NULL; 1147 } 1148 } 1149 1150 /** 1151 * gfs2_fill_super - Read in superblock 1152 * @sb: The VFS superblock 1153 * @fc: Mount options and flags 1154 * 1155 * Returns: -errno 1156 */ 1157 static int gfs2_fill_super(struct super_block *sb, struct fs_context *fc) 1158 { 1159 struct gfs2_args *args = fc->fs_private; 1160 int silent = fc->sb_flags & SB_SILENT; 1161 struct gfs2_sbd *sdp; 1162 struct gfs2_holder mount_gh; 1163 int error; 1164 1165 sdp = init_sbd(sb); 1166 if (!sdp) { 1167 pr_warn("can't alloc struct gfs2_sbd\n"); 1168 return -ENOMEM; 1169 } 1170 sdp->sd_args = *args; 1171 1172 if (sdp->sd_args.ar_spectator) { 1173 sb->s_flags |= SB_RDONLY; 1174 set_bit(SDF_RORECOVERY, &sdp->sd_flags); 1175 } 1176 if (sdp->sd_args.ar_posix_acl) 1177 sb->s_flags |= SB_POSIXACL; 1178 if (sdp->sd_args.ar_nobarrier) 1179 set_bit(SDF_NOBARRIERS, &sdp->sd_flags); 1180 1181 sb->s_flags |= SB_NOSEC; 1182 sb->s_magic = GFS2_MAGIC; 1183 sb->s_op = &gfs2_super_ops; 1184 sb->s_d_op = &gfs2_dops; 1185 sb->s_export_op = &gfs2_export_ops; 1186 sb->s_qcop = &gfs2_quotactl_ops; 1187 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP; 1188 sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE; 1189 sb->s_time_gran = 1; 1190 sb->s_maxbytes = MAX_LFS_FILESIZE; 1191 1192 /* Set up the buffer cache and fill in some fake block size values 1193 to allow us to read-in the on-disk superblock. */ 1194 sdp->sd_sb.sb_bsize = sb_min_blocksize(sb, GFS2_BASIC_BLOCK); 1195 sdp->sd_sb.sb_bsize_shift = sb->s_blocksize_bits; 1196 sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift - 1197 GFS2_BASIC_BLOCK_SHIFT; 1198 sdp->sd_fsb2bb = BIT(sdp->sd_fsb2bb_shift); 1199 1200 sdp->sd_tune.gt_logd_secs = sdp->sd_args.ar_commit; 1201 sdp->sd_tune.gt_quota_quantum = sdp->sd_args.ar_quota_quantum; 1202 if (sdp->sd_args.ar_statfs_quantum) { 1203 sdp->sd_tune.gt_statfs_slow = 0; 1204 sdp->sd_tune.gt_statfs_quantum = sdp->sd_args.ar_statfs_quantum; 1205 } else { 1206 sdp->sd_tune.gt_statfs_slow = 1; 1207 sdp->sd_tune.gt_statfs_quantum = 30; 1208 } 1209 1210 error = init_names(sdp, silent); 1211 if (error) 1212 goto fail_free; 1213 1214 snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s", sdp->sd_table_name); 1215 1216 sdp->sd_delete_wq = alloc_workqueue("gfs2-delete/%s", 1217 WQ_MEM_RECLAIM | WQ_FREEZABLE, 0, sdp->sd_fsname); 1218 error = -ENOMEM; 1219 if (!sdp->sd_delete_wq) 1220 goto fail_free; 1221 1222 error = gfs2_sys_fs_add(sdp); 1223 if (error) 1224 goto fail_delete_wq; 1225 1226 gfs2_create_debugfs_file(sdp); 1227 1228 error = gfs2_lm_mount(sdp, silent); 1229 if (error) 1230 goto fail_debug; 1231 1232 error = init_locking(sdp, &mount_gh, DO); 1233 if (error) 1234 goto fail_lm; 1235 1236 error = init_sb(sdp, silent); 1237 if (error) 1238 goto fail_locking; 1239 1240 /* Turn rgrplvb on by default if fs format is recent enough */ 1241 if (!sdp->sd_args.ar_got_rgrplvb && sdp->sd_sb.sb_fs_format > 1801) 1242 sdp->sd_args.ar_rgrplvb = 1; 1243 1244 error = wait_on_journal(sdp); 1245 if (error) 1246 goto fail_sb; 1247 1248 /* 1249 * If user space has failed to join the cluster or some similar 1250 * failure has occurred, then the journal id will contain a 1251 * negative (error) number. This will then be returned to the 1252 * caller (of the mount syscall). We do this even for spectator 1253 * mounts (which just write a jid of 0 to indicate "ok" even though 1254 * the jid is unused in the spectator case) 1255 */ 1256 if (sdp->sd_lockstruct.ls_jid < 0) { 1257 error = sdp->sd_lockstruct.ls_jid; 1258 sdp->sd_lockstruct.ls_jid = 0; 1259 goto fail_sb; 1260 } 1261 1262 if (sdp->sd_args.ar_spectator) 1263 snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s.s", 1264 sdp->sd_table_name); 1265 else 1266 snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s.%u", 1267 sdp->sd_table_name, sdp->sd_lockstruct.ls_jid); 1268 1269 error = init_inodes(sdp, DO); 1270 if (error) 1271 goto fail_sb; 1272 1273 error = init_per_node(sdp, DO); 1274 if (error) 1275 goto fail_inodes; 1276 1277 error = gfs2_statfs_init(sdp); 1278 if (error) { 1279 fs_err(sdp, "can't initialize statfs subsystem: %d\n", error); 1280 goto fail_per_node; 1281 } 1282 1283 if (!sb_rdonly(sb)) { 1284 error = init_threads(sdp); 1285 if (error) 1286 goto fail_per_node; 1287 } 1288 1289 error = gfs2_freeze_lock_shared(sdp); 1290 if (error) 1291 goto fail_per_node; 1292 1293 if (!sb_rdonly(sb)) 1294 error = gfs2_make_fs_rw(sdp); 1295 1296 if (error) { 1297 gfs2_freeze_unlock(&sdp->sd_freeze_gh); 1298 gfs2_destroy_threads(sdp); 1299 fs_err(sdp, "can't make FS RW: %d\n", error); 1300 goto fail_per_node; 1301 } 1302 gfs2_glock_dq_uninit(&mount_gh); 1303 gfs2_online_uevent(sdp); 1304 return 0; 1305 1306 fail_per_node: 1307 init_per_node(sdp, UNDO); 1308 fail_inodes: 1309 init_inodes(sdp, UNDO); 1310 fail_sb: 1311 if (sdp->sd_root_dir) 1312 dput(sdp->sd_root_dir); 1313 if (sdp->sd_master_dir) 1314 dput(sdp->sd_master_dir); 1315 if (sb->s_root) 1316 dput(sb->s_root); 1317 sb->s_root = NULL; 1318 fail_locking: 1319 init_locking(sdp, &mount_gh, UNDO); 1320 fail_lm: 1321 complete_all(&sdp->sd_journal_ready); 1322 gfs2_gl_hash_clear(sdp); 1323 gfs2_lm_unmount(sdp); 1324 fail_debug: 1325 gfs2_delete_debugfs_file(sdp); 1326 gfs2_sys_fs_del(sdp); 1327 fail_delete_wq: 1328 destroy_workqueue(sdp->sd_delete_wq); 1329 fail_free: 1330 free_sbd(sdp); 1331 sb->s_fs_info = NULL; 1332 return error; 1333 } 1334 1335 /** 1336 * gfs2_get_tree - Get the GFS2 superblock and root directory 1337 * @fc: The filesystem context 1338 * 1339 * Returns: 0 or -errno on error 1340 */ 1341 static int gfs2_get_tree(struct fs_context *fc) 1342 { 1343 struct gfs2_args *args = fc->fs_private; 1344 struct gfs2_sbd *sdp; 1345 int error; 1346 1347 error = get_tree_bdev(fc, gfs2_fill_super); 1348 if (error) 1349 return error; 1350 1351 sdp = fc->root->d_sb->s_fs_info; 1352 dput(fc->root); 1353 if (args->ar_meta) 1354 fc->root = dget(sdp->sd_master_dir); 1355 else 1356 fc->root = dget(sdp->sd_root_dir); 1357 return 0; 1358 } 1359 1360 static void gfs2_fc_free(struct fs_context *fc) 1361 { 1362 struct gfs2_args *args = fc->fs_private; 1363 1364 kfree(args); 1365 } 1366 1367 enum gfs2_param { 1368 Opt_lockproto, 1369 Opt_locktable, 1370 Opt_hostdata, 1371 Opt_spectator, 1372 Opt_ignore_local_fs, 1373 Opt_localflocks, 1374 Opt_localcaching, 1375 Opt_debug, 1376 Opt_upgrade, 1377 Opt_acl, 1378 Opt_quota, 1379 Opt_quota_flag, 1380 Opt_suiddir, 1381 Opt_data, 1382 Opt_meta, 1383 Opt_discard, 1384 Opt_commit, 1385 Opt_errors, 1386 Opt_statfs_quantum, 1387 Opt_statfs_percent, 1388 Opt_quota_quantum, 1389 Opt_barrier, 1390 Opt_rgrplvb, 1391 Opt_loccookie, 1392 }; 1393 1394 static const struct constant_table gfs2_param_quota[] = { 1395 {"off", GFS2_QUOTA_OFF}, 1396 {"account", GFS2_QUOTA_ACCOUNT}, 1397 {"on", GFS2_QUOTA_ON}, 1398 {"quiet", GFS2_QUOTA_QUIET}, 1399 {} 1400 }; 1401 1402 enum opt_data { 1403 Opt_data_writeback = GFS2_DATA_WRITEBACK, 1404 Opt_data_ordered = GFS2_DATA_ORDERED, 1405 }; 1406 1407 static const struct constant_table gfs2_param_data[] = { 1408 {"writeback", Opt_data_writeback }, 1409 {"ordered", Opt_data_ordered }, 1410 {} 1411 }; 1412 1413 enum opt_errors { 1414 Opt_errors_withdraw = GFS2_ERRORS_WITHDRAW, 1415 Opt_errors_panic = GFS2_ERRORS_PANIC, 1416 }; 1417 1418 static const struct constant_table gfs2_param_errors[] = { 1419 {"withdraw", Opt_errors_withdraw }, 1420 {"panic", Opt_errors_panic }, 1421 {} 1422 }; 1423 1424 static const struct fs_parameter_spec gfs2_fs_parameters[] = { 1425 fsparam_string ("lockproto", Opt_lockproto), 1426 fsparam_string ("locktable", Opt_locktable), 1427 fsparam_string ("hostdata", Opt_hostdata), 1428 fsparam_flag ("spectator", Opt_spectator), 1429 fsparam_flag ("norecovery", Opt_spectator), 1430 fsparam_flag ("ignore_local_fs", Opt_ignore_local_fs), 1431 fsparam_flag ("localflocks", Opt_localflocks), 1432 fsparam_flag ("localcaching", Opt_localcaching), 1433 fsparam_flag_no("debug", Opt_debug), 1434 fsparam_flag ("upgrade", Opt_upgrade), 1435 fsparam_flag_no("acl", Opt_acl), 1436 fsparam_flag_no("suiddir", Opt_suiddir), 1437 fsparam_enum ("data", Opt_data, gfs2_param_data), 1438 fsparam_flag ("meta", Opt_meta), 1439 fsparam_flag_no("discard", Opt_discard), 1440 fsparam_s32 ("commit", Opt_commit), 1441 fsparam_enum ("errors", Opt_errors, gfs2_param_errors), 1442 fsparam_s32 ("statfs_quantum", Opt_statfs_quantum), 1443 fsparam_s32 ("statfs_percent", Opt_statfs_percent), 1444 fsparam_s32 ("quota_quantum", Opt_quota_quantum), 1445 fsparam_flag_no("barrier", Opt_barrier), 1446 fsparam_flag_no("rgrplvb", Opt_rgrplvb), 1447 fsparam_flag_no("loccookie", Opt_loccookie), 1448 /* quota can be a flag or an enum so it gets special treatment */ 1449 fsparam_flag_no("quota", Opt_quota_flag), 1450 fsparam_enum("quota", Opt_quota, gfs2_param_quota), 1451 {} 1452 }; 1453 1454 /* Parse a single mount parameter */ 1455 static int gfs2_parse_param(struct fs_context *fc, struct fs_parameter *param) 1456 { 1457 struct gfs2_args *args = fc->fs_private; 1458 struct fs_parse_result result; 1459 int o; 1460 1461 o = fs_parse(fc, gfs2_fs_parameters, param, &result); 1462 if (o < 0) 1463 return o; 1464 1465 switch (o) { 1466 case Opt_lockproto: 1467 strscpy(args->ar_lockproto, param->string, GFS2_LOCKNAME_LEN); 1468 break; 1469 case Opt_locktable: 1470 strscpy(args->ar_locktable, param->string, GFS2_LOCKNAME_LEN); 1471 break; 1472 case Opt_hostdata: 1473 strscpy(args->ar_hostdata, param->string, GFS2_LOCKNAME_LEN); 1474 break; 1475 case Opt_spectator: 1476 args->ar_spectator = 1; 1477 break; 1478 case Opt_ignore_local_fs: 1479 /* Retained for backwards compat only */ 1480 break; 1481 case Opt_localflocks: 1482 args->ar_localflocks = 1; 1483 break; 1484 case Opt_localcaching: 1485 /* Retained for backwards compat only */ 1486 break; 1487 case Opt_debug: 1488 if (result.boolean && args->ar_errors == GFS2_ERRORS_PANIC) 1489 return invalfc(fc, "-o debug and -o errors=panic are mutually exclusive"); 1490 args->ar_debug = result.boolean; 1491 break; 1492 case Opt_upgrade: 1493 /* Retained for backwards compat only */ 1494 break; 1495 case Opt_acl: 1496 args->ar_posix_acl = result.boolean; 1497 break; 1498 case Opt_quota_flag: 1499 args->ar_quota = result.negated ? GFS2_QUOTA_OFF : GFS2_QUOTA_ON; 1500 break; 1501 case Opt_quota: 1502 args->ar_quota = result.int_32; 1503 break; 1504 case Opt_suiddir: 1505 args->ar_suiddir = result.boolean; 1506 break; 1507 case Opt_data: 1508 /* The uint_32 result maps directly to GFS2_DATA_* */ 1509 args->ar_data = result.uint_32; 1510 break; 1511 case Opt_meta: 1512 args->ar_meta = 1; 1513 break; 1514 case Opt_discard: 1515 args->ar_discard = result.boolean; 1516 break; 1517 case Opt_commit: 1518 if (result.int_32 <= 0) 1519 return invalfc(fc, "commit mount option requires a positive numeric argument"); 1520 args->ar_commit = result.int_32; 1521 break; 1522 case Opt_statfs_quantum: 1523 if (result.int_32 < 0) 1524 return invalfc(fc, "statfs_quantum mount option requires a non-negative numeric argument"); 1525 args->ar_statfs_quantum = result.int_32; 1526 break; 1527 case Opt_quota_quantum: 1528 if (result.int_32 <= 0) 1529 return invalfc(fc, "quota_quantum mount option requires a positive numeric argument"); 1530 args->ar_quota_quantum = result.int_32; 1531 break; 1532 case Opt_statfs_percent: 1533 if (result.int_32 < 0 || result.int_32 > 100) 1534 return invalfc(fc, "statfs_percent mount option requires a numeric argument between 0 and 100"); 1535 args->ar_statfs_percent = result.int_32; 1536 break; 1537 case Opt_errors: 1538 if (args->ar_debug && result.uint_32 == GFS2_ERRORS_PANIC) 1539 return invalfc(fc, "-o debug and -o errors=panic are mutually exclusive"); 1540 args->ar_errors = result.uint_32; 1541 break; 1542 case Opt_barrier: 1543 args->ar_nobarrier = result.boolean; 1544 break; 1545 case Opt_rgrplvb: 1546 args->ar_rgrplvb = result.boolean; 1547 args->ar_got_rgrplvb = 1; 1548 break; 1549 case Opt_loccookie: 1550 args->ar_loccookie = result.boolean; 1551 break; 1552 default: 1553 return invalfc(fc, "invalid mount option: %s", param->key); 1554 } 1555 return 0; 1556 } 1557 1558 static int gfs2_reconfigure(struct fs_context *fc) 1559 { 1560 struct super_block *sb = fc->root->d_sb; 1561 struct gfs2_sbd *sdp = sb->s_fs_info; 1562 struct gfs2_args *oldargs = &sdp->sd_args; 1563 struct gfs2_args *newargs = fc->fs_private; 1564 struct gfs2_tune *gt = &sdp->sd_tune; 1565 int error = 0; 1566 1567 sync_filesystem(sb); 1568 1569 spin_lock(>->gt_spin); 1570 oldargs->ar_commit = gt->gt_logd_secs; 1571 oldargs->ar_quota_quantum = gt->gt_quota_quantum; 1572 if (gt->gt_statfs_slow) 1573 oldargs->ar_statfs_quantum = 0; 1574 else 1575 oldargs->ar_statfs_quantum = gt->gt_statfs_quantum; 1576 spin_unlock(>->gt_spin); 1577 1578 if (strcmp(newargs->ar_lockproto, oldargs->ar_lockproto)) { 1579 errorfc(fc, "reconfiguration of locking protocol not allowed"); 1580 return -EINVAL; 1581 } 1582 if (strcmp(newargs->ar_locktable, oldargs->ar_locktable)) { 1583 errorfc(fc, "reconfiguration of lock table not allowed"); 1584 return -EINVAL; 1585 } 1586 if (strcmp(newargs->ar_hostdata, oldargs->ar_hostdata)) { 1587 errorfc(fc, "reconfiguration of host data not allowed"); 1588 return -EINVAL; 1589 } 1590 if (newargs->ar_spectator != oldargs->ar_spectator) { 1591 errorfc(fc, "reconfiguration of spectator mode not allowed"); 1592 return -EINVAL; 1593 } 1594 if (newargs->ar_localflocks != oldargs->ar_localflocks) { 1595 errorfc(fc, "reconfiguration of localflocks not allowed"); 1596 return -EINVAL; 1597 } 1598 if (newargs->ar_meta != oldargs->ar_meta) { 1599 errorfc(fc, "switching between gfs2 and gfs2meta not allowed"); 1600 return -EINVAL; 1601 } 1602 if (oldargs->ar_spectator) 1603 fc->sb_flags |= SB_RDONLY; 1604 1605 if ((sb->s_flags ^ fc->sb_flags) & SB_RDONLY) { 1606 if (fc->sb_flags & SB_RDONLY) { 1607 gfs2_make_fs_ro(sdp); 1608 } else { 1609 error = gfs2_make_fs_rw(sdp); 1610 if (error) 1611 errorfc(fc, "unable to remount read-write"); 1612 } 1613 } 1614 sdp->sd_args = *newargs; 1615 1616 if (sdp->sd_args.ar_posix_acl) 1617 sb->s_flags |= SB_POSIXACL; 1618 else 1619 sb->s_flags &= ~SB_POSIXACL; 1620 if (sdp->sd_args.ar_nobarrier) 1621 set_bit(SDF_NOBARRIERS, &sdp->sd_flags); 1622 else 1623 clear_bit(SDF_NOBARRIERS, &sdp->sd_flags); 1624 spin_lock(>->gt_spin); 1625 gt->gt_logd_secs = newargs->ar_commit; 1626 gt->gt_quota_quantum = newargs->ar_quota_quantum; 1627 if (newargs->ar_statfs_quantum) { 1628 gt->gt_statfs_slow = 0; 1629 gt->gt_statfs_quantum = newargs->ar_statfs_quantum; 1630 } 1631 else { 1632 gt->gt_statfs_slow = 1; 1633 gt->gt_statfs_quantum = 30; 1634 } 1635 spin_unlock(>->gt_spin); 1636 1637 gfs2_online_uevent(sdp); 1638 return error; 1639 } 1640 1641 static const struct fs_context_operations gfs2_context_ops = { 1642 .free = gfs2_fc_free, 1643 .parse_param = gfs2_parse_param, 1644 .get_tree = gfs2_get_tree, 1645 .reconfigure = gfs2_reconfigure, 1646 }; 1647 1648 /* Set up the filesystem mount context */ 1649 static int gfs2_init_fs_context(struct fs_context *fc) 1650 { 1651 struct gfs2_args *args; 1652 1653 args = kmalloc(sizeof(*args), GFP_KERNEL); 1654 if (args == NULL) 1655 return -ENOMEM; 1656 1657 if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) { 1658 struct gfs2_sbd *sdp = fc->root->d_sb->s_fs_info; 1659 1660 *args = sdp->sd_args; 1661 } else { 1662 memset(args, 0, sizeof(*args)); 1663 args->ar_quota = GFS2_QUOTA_DEFAULT; 1664 args->ar_data = GFS2_DATA_DEFAULT; 1665 args->ar_commit = 30; 1666 args->ar_statfs_quantum = 30; 1667 args->ar_quota_quantum = 60; 1668 args->ar_errors = GFS2_ERRORS_DEFAULT; 1669 } 1670 fc->fs_private = args; 1671 fc->ops = &gfs2_context_ops; 1672 return 0; 1673 } 1674 1675 static int set_meta_super(struct super_block *s, struct fs_context *fc) 1676 { 1677 return -EINVAL; 1678 } 1679 1680 static int test_meta_super(struct super_block *s, struct fs_context *fc) 1681 { 1682 return (fc->sget_key == s->s_bdev); 1683 } 1684 1685 static int gfs2_meta_get_tree(struct fs_context *fc) 1686 { 1687 struct super_block *s; 1688 struct gfs2_sbd *sdp; 1689 struct path path; 1690 int error; 1691 1692 if (!fc->source || !*fc->source) 1693 return -EINVAL; 1694 1695 error = kern_path(fc->source, LOOKUP_FOLLOW, &path); 1696 if (error) { 1697 pr_warn("path_lookup on %s returned error %d\n", 1698 fc->source, error); 1699 return error; 1700 } 1701 fc->fs_type = &gfs2_fs_type; 1702 fc->sget_key = path.dentry->d_sb->s_bdev; 1703 s = sget_fc(fc, test_meta_super, set_meta_super); 1704 path_put(&path); 1705 if (IS_ERR(s)) { 1706 pr_warn("gfs2 mount does not exist\n"); 1707 return PTR_ERR(s); 1708 } 1709 if ((fc->sb_flags ^ s->s_flags) & SB_RDONLY) { 1710 deactivate_locked_super(s); 1711 return -EBUSY; 1712 } 1713 sdp = s->s_fs_info; 1714 fc->root = dget(sdp->sd_master_dir); 1715 return 0; 1716 } 1717 1718 static const struct fs_context_operations gfs2_meta_context_ops = { 1719 .free = gfs2_fc_free, 1720 .get_tree = gfs2_meta_get_tree, 1721 }; 1722 1723 static int gfs2_meta_init_fs_context(struct fs_context *fc) 1724 { 1725 int ret = gfs2_init_fs_context(fc); 1726 1727 if (ret) 1728 return ret; 1729 1730 fc->ops = &gfs2_meta_context_ops; 1731 return 0; 1732 } 1733 1734 /** 1735 * gfs2_evict_inodes - evict inodes cooperatively 1736 * @sb: the superblock 1737 * 1738 * When evicting an inode with a zero link count, we are trying to upgrade the 1739 * inode's iopen glock from SH to EX mode in order to determine if we can 1740 * delete the inode. The other nodes are supposed to evict the inode from 1741 * their caches if they can, and to poke the inode's inode glock if they cannot 1742 * do so. Either behavior allows gfs2_upgrade_iopen_glock() to proceed 1743 * quickly, but if the other nodes are not cooperating, the lock upgrading 1744 * attempt will time out. Since inodes are evicted sequentially, this can add 1745 * up quickly. 1746 * 1747 * Function evict_inodes() tries to keep the s_inode_list_lock list locked over 1748 * a long time, which prevents other inodes from being evicted concurrently. 1749 * This precludes the cooperative behavior we are looking for. This special 1750 * version of evict_inodes() avoids that. 1751 * 1752 * Modeled after drop_pagecache_sb(). 1753 */ 1754 static void gfs2_evict_inodes(struct super_block *sb) 1755 { 1756 struct inode *inode, *toput_inode = NULL; 1757 struct gfs2_sbd *sdp = sb->s_fs_info; 1758 1759 set_bit(SDF_EVICTING, &sdp->sd_flags); 1760 1761 spin_lock(&sb->s_inode_list_lock); 1762 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { 1763 spin_lock(&inode->i_lock); 1764 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) && 1765 !need_resched()) { 1766 spin_unlock(&inode->i_lock); 1767 continue; 1768 } 1769 atomic_inc(&inode->i_count); 1770 spin_unlock(&inode->i_lock); 1771 spin_unlock(&sb->s_inode_list_lock); 1772 1773 iput(toput_inode); 1774 toput_inode = inode; 1775 1776 cond_resched(); 1777 spin_lock(&sb->s_inode_list_lock); 1778 } 1779 spin_unlock(&sb->s_inode_list_lock); 1780 iput(toput_inode); 1781 } 1782 1783 static void gfs2_kill_sb(struct super_block *sb) 1784 { 1785 struct gfs2_sbd *sdp = sb->s_fs_info; 1786 1787 if (sdp == NULL) { 1788 kill_block_super(sb); 1789 return; 1790 } 1791 1792 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SYNC | GFS2_LFC_KILL_SB); 1793 dput(sdp->sd_root_dir); 1794 dput(sdp->sd_master_dir); 1795 sdp->sd_root_dir = NULL; 1796 sdp->sd_master_dir = NULL; 1797 shrink_dcache_sb(sb); 1798 1799 gfs2_evict_inodes(sb); 1800 1801 /* 1802 * Flush and then drain the delete workqueue here (via 1803 * destroy_workqueue()) to ensure that any delete work that 1804 * may be running will also see the SDF_KILL flag. 1805 */ 1806 set_bit(SDF_KILL, &sdp->sd_flags); 1807 gfs2_flush_delete_work(sdp); 1808 destroy_workqueue(sdp->sd_delete_wq); 1809 1810 kill_block_super(sb); 1811 } 1812 1813 struct file_system_type gfs2_fs_type = { 1814 .name = "gfs2", 1815 .fs_flags = FS_REQUIRES_DEV, 1816 .init_fs_context = gfs2_init_fs_context, 1817 .parameters = gfs2_fs_parameters, 1818 .kill_sb = gfs2_kill_sb, 1819 .owner = THIS_MODULE, 1820 }; 1821 MODULE_ALIAS_FS("gfs2"); 1822 1823 struct file_system_type gfs2meta_fs_type = { 1824 .name = "gfs2meta", 1825 .fs_flags = FS_REQUIRES_DEV, 1826 .init_fs_context = gfs2_meta_init_fs_context, 1827 .owner = THIS_MODULE, 1828 }; 1829 MODULE_ALIAS_FS("gfs2meta"); 1830