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_glock_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 INIT_LIST_HEAD(&sdp->sd_trunc_list); 110 spin_lock_init(&sdp->sd_trunc_lock); 111 spin_lock_init(&sdp->sd_bitmap_lock); 112 113 INIT_LIST_HEAD(&sdp->sd_sc_inodes_list); 114 115 mapping = &sdp->sd_aspace; 116 117 address_space_init_once(mapping); 118 mapping->a_ops = &gfs2_rgrp_aops; 119 mapping->host = sb->s_bdev->bd_inode; 120 mapping->flags = 0; 121 mapping_set_gfp_mask(mapping, GFP_NOFS); 122 mapping->private_data = NULL; 123 mapping->writeback_index = 0; 124 125 spin_lock_init(&sdp->sd_log_lock); 126 atomic_set(&sdp->sd_log_pinned, 0); 127 INIT_LIST_HEAD(&sdp->sd_log_revokes); 128 INIT_LIST_HEAD(&sdp->sd_log_ordered); 129 spin_lock_init(&sdp->sd_ordered_lock); 130 131 init_waitqueue_head(&sdp->sd_log_waitq); 132 init_waitqueue_head(&sdp->sd_logd_waitq); 133 spin_lock_init(&sdp->sd_ail_lock); 134 INIT_LIST_HEAD(&sdp->sd_ail1_list); 135 INIT_LIST_HEAD(&sdp->sd_ail2_list); 136 137 init_rwsem(&sdp->sd_log_flush_lock); 138 atomic_set(&sdp->sd_log_in_flight, 0); 139 init_waitqueue_head(&sdp->sd_log_flush_wait); 140 atomic_set(&sdp->sd_freeze_state, SFS_UNFROZEN); 141 mutex_init(&sdp->sd_freeze_mutex); 142 143 return sdp; 144 145 fail: 146 free_sbd(sdp); 147 return NULL; 148 } 149 150 /** 151 * gfs2_check_sb - Check superblock 152 * @sdp: the filesystem 153 * @silent: Don't print a message if the check fails 154 * 155 * Checks the version code of the FS is one that we understand how to 156 * read and that the sizes of the various on-disk structures have not 157 * changed. 158 */ 159 160 static int gfs2_check_sb(struct gfs2_sbd *sdp, int silent) 161 { 162 struct gfs2_sb_host *sb = &sdp->sd_sb; 163 164 if (sb->sb_magic != GFS2_MAGIC || 165 sb->sb_type != GFS2_METATYPE_SB) { 166 if (!silent) 167 pr_warn("not a GFS2 filesystem\n"); 168 return -EINVAL; 169 } 170 171 if (sb->sb_fs_format < GFS2_FS_FORMAT_MIN || 172 sb->sb_fs_format > GFS2_FS_FORMAT_MAX || 173 sb->sb_multihost_format != GFS2_FORMAT_MULTI) { 174 fs_warn(sdp, "Unknown on-disk format, unable to mount\n"); 175 return -EINVAL; 176 } 177 178 if (sb->sb_bsize < 512 || sb->sb_bsize > PAGE_SIZE || 179 (sb->sb_bsize & (sb->sb_bsize - 1))) { 180 pr_warn("Invalid block size\n"); 181 return -EINVAL; 182 } 183 184 return 0; 185 } 186 187 static void end_bio_io_page(struct bio *bio) 188 { 189 struct page *page = bio->bi_private; 190 191 if (!bio->bi_status) 192 SetPageUptodate(page); 193 else 194 pr_warn("error %d reading superblock\n", bio->bi_status); 195 unlock_page(page); 196 } 197 198 static void gfs2_sb_in(struct gfs2_sbd *sdp, const void *buf) 199 { 200 struct gfs2_sb_host *sb = &sdp->sd_sb; 201 struct super_block *s = sdp->sd_vfs; 202 const struct gfs2_sb *str = buf; 203 204 sb->sb_magic = be32_to_cpu(str->sb_header.mh_magic); 205 sb->sb_type = be32_to_cpu(str->sb_header.mh_type); 206 sb->sb_fs_format = be32_to_cpu(str->sb_fs_format); 207 sb->sb_multihost_format = be32_to_cpu(str->sb_multihost_format); 208 sb->sb_bsize = be32_to_cpu(str->sb_bsize); 209 sb->sb_bsize_shift = be32_to_cpu(str->sb_bsize_shift); 210 sb->sb_master_dir.no_addr = be64_to_cpu(str->sb_master_dir.no_addr); 211 sb->sb_master_dir.no_formal_ino = be64_to_cpu(str->sb_master_dir.no_formal_ino); 212 sb->sb_root_dir.no_addr = be64_to_cpu(str->sb_root_dir.no_addr); 213 sb->sb_root_dir.no_formal_ino = be64_to_cpu(str->sb_root_dir.no_formal_ino); 214 215 memcpy(sb->sb_lockproto, str->sb_lockproto, GFS2_LOCKNAME_LEN); 216 memcpy(sb->sb_locktable, str->sb_locktable, GFS2_LOCKNAME_LEN); 217 memcpy(&s->s_uuid, str->sb_uuid, 16); 218 } 219 220 /** 221 * gfs2_read_super - Read the gfs2 super block from disk 222 * @sdp: The GFS2 super block 223 * @sector: The location of the super block 224 * @silent: Don't print a message if the check fails 225 * 226 * This uses the bio functions to read the super block from disk 227 * because we want to be 100% sure that we never read cached data. 228 * A super block is read twice only during each GFS2 mount and is 229 * never written to by the filesystem. The first time its read no 230 * locks are held, and the only details which are looked at are those 231 * relating to the locking protocol. Once locking is up and working, 232 * the sb is read again under the lock to establish the location of 233 * the master directory (contains pointers to journals etc) and the 234 * root directory. 235 * 236 * Returns: 0 on success or error 237 */ 238 239 static int gfs2_read_super(struct gfs2_sbd *sdp, sector_t sector, int silent) 240 { 241 struct super_block *sb = sdp->sd_vfs; 242 struct gfs2_sb *p; 243 struct page *page; 244 struct bio *bio; 245 246 page = alloc_page(GFP_NOFS); 247 if (unlikely(!page)) 248 return -ENOMEM; 249 250 ClearPageUptodate(page); 251 ClearPageDirty(page); 252 lock_page(page); 253 254 bio = bio_alloc(GFP_NOFS, 1); 255 bio->bi_iter.bi_sector = sector * (sb->s_blocksize >> 9); 256 bio_set_dev(bio, sb->s_bdev); 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 bio_set_op_attrs(bio, REQ_OP_READ, REQ_META); 262 submit_bio(bio); 263 wait_on_page_locked(page); 264 bio_put(bio); 265 if (!PageUptodate(page)) { 266 __free_page(page); 267 return -EIO; 268 } 269 p = kmap(page); 270 gfs2_sb_in(sdp, p); 271 kunmap(page); 272 __free_page(page); 273 return gfs2_check_sb(sdp, silent); 274 } 275 276 /** 277 * gfs2_read_sb - Read super block 278 * @sdp: The GFS2 superblock 279 * @silent: Don't print message if mount fails 280 * 281 */ 282 283 static int gfs2_read_sb(struct gfs2_sbd *sdp, int silent) 284 { 285 u32 hash_blocks, ind_blocks, leaf_blocks; 286 u32 tmp_blocks; 287 unsigned int x; 288 int error; 289 290 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent); 291 if (error) { 292 if (!silent) 293 fs_err(sdp, "can't read superblock\n"); 294 return error; 295 } 296 297 sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift - 298 GFS2_BASIC_BLOCK_SHIFT; 299 sdp->sd_fsb2bb = BIT(sdp->sd_fsb2bb_shift); 300 sdp->sd_diptrs = (sdp->sd_sb.sb_bsize - 301 sizeof(struct gfs2_dinode)) / sizeof(u64); 302 sdp->sd_inptrs = (sdp->sd_sb.sb_bsize - 303 sizeof(struct gfs2_meta_header)) / sizeof(u64); 304 sdp->sd_ldptrs = (sdp->sd_sb.sb_bsize - 305 sizeof(struct gfs2_log_descriptor)) / sizeof(u64); 306 sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header); 307 sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2; 308 sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1; 309 sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64); 310 sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize - 311 sizeof(struct gfs2_meta_header)) / 312 sizeof(struct gfs2_quota_change); 313 sdp->sd_blocks_per_bitmap = (sdp->sd_sb.sb_bsize - 314 sizeof(struct gfs2_meta_header)) 315 * GFS2_NBBY; /* not the rgrp bitmap, subsequent bitmaps only */ 316 317 /* 318 * We always keep at least one block reserved for revokes in 319 * transactions. This greatly simplifies allocating additional 320 * revoke blocks. 321 */ 322 atomic_set(&sdp->sd_log_revokes_available, sdp->sd_ldptrs); 323 324 /* Compute maximum reservation required to add a entry to a directory */ 325 326 hash_blocks = DIV_ROUND_UP(sizeof(u64) * BIT(GFS2_DIR_MAX_DEPTH), 327 sdp->sd_jbsize); 328 329 ind_blocks = 0; 330 for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) { 331 tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs); 332 ind_blocks += tmp_blocks; 333 } 334 335 leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH; 336 337 sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks; 338 339 sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize - 340 sizeof(struct gfs2_dinode); 341 sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs; 342 for (x = 2;; x++) { 343 u64 space, d; 344 u32 m; 345 346 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs; 347 d = space; 348 m = do_div(d, sdp->sd_inptrs); 349 350 if (d != sdp->sd_heightsize[x - 1] || m) 351 break; 352 sdp->sd_heightsize[x] = space; 353 } 354 sdp->sd_max_height = x; 355 sdp->sd_heightsize[x] = ~0; 356 gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT); 357 358 sdp->sd_max_dents_per_leaf = (sdp->sd_sb.sb_bsize - 359 sizeof(struct gfs2_leaf)) / 360 GFS2_MIN_DIRENT_SIZE; 361 return 0; 362 } 363 364 static int init_names(struct gfs2_sbd *sdp, int silent) 365 { 366 char *proto, *table; 367 int error = 0; 368 369 proto = sdp->sd_args.ar_lockproto; 370 table = sdp->sd_args.ar_locktable; 371 372 /* Try to autodetect */ 373 374 if (!proto[0] || !table[0]) { 375 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent); 376 if (error) 377 return error; 378 379 if (!proto[0]) 380 proto = sdp->sd_sb.sb_lockproto; 381 if (!table[0]) 382 table = sdp->sd_sb.sb_locktable; 383 } 384 385 if (!table[0]) 386 table = sdp->sd_vfs->s_id; 387 388 strlcpy(sdp->sd_proto_name, proto, GFS2_FSNAME_LEN); 389 strlcpy(sdp->sd_table_name, table, GFS2_FSNAME_LEN); 390 391 table = sdp->sd_table_name; 392 while ((table = strchr(table, '/'))) 393 *table = '_'; 394 395 return error; 396 } 397 398 static int init_locking(struct gfs2_sbd *sdp, struct gfs2_holder *mount_gh, 399 int undo) 400 { 401 int error = 0; 402 403 if (undo) 404 goto fail_trans; 405 406 error = gfs2_glock_nq_num(sdp, 407 GFS2_MOUNT_LOCK, &gfs2_nondisk_glops, 408 LM_ST_EXCLUSIVE, LM_FLAG_NOEXP | GL_NOCACHE, 409 mount_gh); 410 if (error) { 411 fs_err(sdp, "can't acquire mount glock: %d\n", error); 412 goto fail; 413 } 414 415 error = gfs2_glock_nq_num(sdp, 416 GFS2_LIVE_LOCK, &gfs2_nondisk_glops, 417 LM_ST_SHARED, 418 LM_FLAG_NOEXP | GL_EXACT, 419 &sdp->sd_live_gh); 420 if (error) { 421 fs_err(sdp, "can't acquire live glock: %d\n", error); 422 goto fail_mount; 423 } 424 425 error = gfs2_glock_get(sdp, GFS2_RENAME_LOCK, &gfs2_nondisk_glops, 426 CREATE, &sdp->sd_rename_gl); 427 if (error) { 428 fs_err(sdp, "can't create rename glock: %d\n", error); 429 goto fail_live; 430 } 431 432 error = gfs2_glock_get(sdp, GFS2_FREEZE_LOCK, &gfs2_freeze_glops, 433 CREATE, &sdp->sd_freeze_gl); 434 if (error) { 435 fs_err(sdp, "can't create transaction glock: %d\n", error); 436 goto fail_rename; 437 } 438 439 return 0; 440 441 fail_trans: 442 gfs2_glock_put(sdp->sd_freeze_gl); 443 fail_rename: 444 gfs2_glock_put(sdp->sd_rename_gl); 445 fail_live: 446 gfs2_glock_dq_uninit(&sdp->sd_live_gh); 447 fail_mount: 448 gfs2_glock_dq_uninit(mount_gh); 449 fail: 450 return error; 451 } 452 453 static int gfs2_lookup_root(struct super_block *sb, struct dentry **dptr, 454 u64 no_addr, const char *name) 455 { 456 struct gfs2_sbd *sdp = sb->s_fs_info; 457 struct dentry *dentry; 458 struct inode *inode; 459 460 inode = gfs2_inode_lookup(sb, DT_DIR, no_addr, 0, 461 GFS2_BLKST_FREE /* ignore */); 462 if (IS_ERR(inode)) { 463 fs_err(sdp, "can't read in %s inode: %ld\n", name, PTR_ERR(inode)); 464 return PTR_ERR(inode); 465 } 466 dentry = d_make_root(inode); 467 if (!dentry) { 468 fs_err(sdp, "can't alloc %s dentry\n", name); 469 return -ENOMEM; 470 } 471 *dptr = dentry; 472 return 0; 473 } 474 475 static int init_sb(struct gfs2_sbd *sdp, int silent) 476 { 477 struct super_block *sb = sdp->sd_vfs; 478 struct gfs2_holder sb_gh; 479 u64 no_addr; 480 int ret; 481 482 ret = gfs2_glock_nq_num(sdp, GFS2_SB_LOCK, &gfs2_meta_glops, 483 LM_ST_SHARED, 0, &sb_gh); 484 if (ret) { 485 fs_err(sdp, "can't acquire superblock glock: %d\n", ret); 486 return ret; 487 } 488 489 ret = gfs2_read_sb(sdp, silent); 490 if (ret) { 491 fs_err(sdp, "can't read superblock: %d\n", ret); 492 goto out; 493 } 494 495 switch(sdp->sd_sb.sb_fs_format) { 496 case GFS2_FS_FORMAT_MAX: 497 sb->s_xattr = gfs2_xattr_handlers_max; 498 break; 499 500 case GFS2_FS_FORMAT_MIN: 501 sb->s_xattr = gfs2_xattr_handlers_min; 502 break; 503 504 default: 505 BUG(); 506 } 507 508 /* Set up the buffer cache and SB for real */ 509 if (sdp->sd_sb.sb_bsize < bdev_logical_block_size(sb->s_bdev)) { 510 ret = -EINVAL; 511 fs_err(sdp, "FS block size (%u) is too small for device " 512 "block size (%u)\n", 513 sdp->sd_sb.sb_bsize, bdev_logical_block_size(sb->s_bdev)); 514 goto out; 515 } 516 if (sdp->sd_sb.sb_bsize > PAGE_SIZE) { 517 ret = -EINVAL; 518 fs_err(sdp, "FS block size (%u) is too big for machine " 519 "page size (%u)\n", 520 sdp->sd_sb.sb_bsize, (unsigned int)PAGE_SIZE); 521 goto out; 522 } 523 sb_set_blocksize(sb, sdp->sd_sb.sb_bsize); 524 525 /* Get the root inode */ 526 no_addr = sdp->sd_sb.sb_root_dir.no_addr; 527 ret = gfs2_lookup_root(sb, &sdp->sd_root_dir, no_addr, "root"); 528 if (ret) 529 goto out; 530 531 /* Get the master inode */ 532 no_addr = sdp->sd_sb.sb_master_dir.no_addr; 533 ret = gfs2_lookup_root(sb, &sdp->sd_master_dir, no_addr, "master"); 534 if (ret) { 535 dput(sdp->sd_root_dir); 536 goto out; 537 } 538 sb->s_root = dget(sdp->sd_args.ar_meta ? sdp->sd_master_dir : sdp->sd_root_dir); 539 out: 540 gfs2_glock_dq_uninit(&sb_gh); 541 return ret; 542 } 543 544 static void gfs2_others_may_mount(struct gfs2_sbd *sdp) 545 { 546 char *message = "FIRSTMOUNT=Done"; 547 char *envp[] = { message, NULL }; 548 549 fs_info(sdp, "first mount done, others may mount\n"); 550 551 if (sdp->sd_lockstruct.ls_ops->lm_first_done) 552 sdp->sd_lockstruct.ls_ops->lm_first_done(sdp); 553 554 kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp); 555 } 556 557 /** 558 * gfs2_jindex_hold - Grab a lock on the jindex 559 * @sdp: The GFS2 superblock 560 * @ji_gh: the holder for the jindex glock 561 * 562 * Returns: errno 563 */ 564 565 static int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh) 566 { 567 struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex); 568 struct qstr name; 569 char buf[20]; 570 struct gfs2_jdesc *jd; 571 int error; 572 573 name.name = buf; 574 575 mutex_lock(&sdp->sd_jindex_mutex); 576 577 for (;;) { 578 struct gfs2_inode *jip; 579 580 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh); 581 if (error) 582 break; 583 584 name.len = sprintf(buf, "journal%u", sdp->sd_journals); 585 name.hash = gfs2_disk_hash(name.name, name.len); 586 587 error = gfs2_dir_check(sdp->sd_jindex, &name, NULL); 588 if (error == -ENOENT) { 589 error = 0; 590 break; 591 } 592 593 gfs2_glock_dq_uninit(ji_gh); 594 595 if (error) 596 break; 597 598 error = -ENOMEM; 599 jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL); 600 if (!jd) 601 break; 602 603 INIT_LIST_HEAD(&jd->extent_list); 604 INIT_LIST_HEAD(&jd->jd_revoke_list); 605 606 INIT_WORK(&jd->jd_work, gfs2_recover_func); 607 jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1); 608 if (IS_ERR_OR_NULL(jd->jd_inode)) { 609 if (!jd->jd_inode) 610 error = -ENOENT; 611 else 612 error = PTR_ERR(jd->jd_inode); 613 kfree(jd); 614 break; 615 } 616 617 spin_lock(&sdp->sd_jindex_spin); 618 jd->jd_jid = sdp->sd_journals++; 619 jip = GFS2_I(jd->jd_inode); 620 jd->jd_no_addr = jip->i_no_addr; 621 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list); 622 spin_unlock(&sdp->sd_jindex_spin); 623 } 624 625 mutex_unlock(&sdp->sd_jindex_mutex); 626 627 return error; 628 } 629 630 /** 631 * init_statfs - look up and initialize master and local (per node) statfs inodes 632 * @sdp: The GFS2 superblock 633 * 634 * This should be called after the jindex is initialized in init_journal() and 635 * before gfs2_journal_recovery() is called because we need to be able to write 636 * to these inodes during recovery. 637 * 638 * Returns: errno 639 */ 640 static int init_statfs(struct gfs2_sbd *sdp) 641 { 642 int error = 0; 643 struct inode *master = d_inode(sdp->sd_master_dir); 644 struct inode *pn = NULL; 645 char buf[30]; 646 struct gfs2_jdesc *jd; 647 struct gfs2_inode *ip; 648 649 sdp->sd_statfs_inode = gfs2_lookup_simple(master, "statfs"); 650 if (IS_ERR(sdp->sd_statfs_inode)) { 651 error = PTR_ERR(sdp->sd_statfs_inode); 652 fs_err(sdp, "can't read in statfs inode: %d\n", error); 653 goto out; 654 } 655 if (sdp->sd_args.ar_spectator) 656 goto out; 657 658 pn = gfs2_lookup_simple(master, "per_node"); 659 if (IS_ERR(pn)) { 660 error = PTR_ERR(pn); 661 fs_err(sdp, "can't find per_node directory: %d\n", error); 662 goto put_statfs; 663 } 664 665 /* For each jid, lookup the corresponding local statfs inode in the 666 * per_node metafs directory and save it in the sdp->sd_sc_inodes_list. */ 667 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) { 668 struct local_statfs_inode *lsi = 669 kmalloc(sizeof(struct local_statfs_inode), GFP_NOFS); 670 if (!lsi) { 671 error = -ENOMEM; 672 goto free_local; 673 } 674 sprintf(buf, "statfs_change%u", jd->jd_jid); 675 lsi->si_sc_inode = gfs2_lookup_simple(pn, buf); 676 if (IS_ERR(lsi->si_sc_inode)) { 677 error = PTR_ERR(lsi->si_sc_inode); 678 fs_err(sdp, "can't find local \"sc\" file#%u: %d\n", 679 jd->jd_jid, error); 680 goto free_local; 681 } 682 lsi->si_jid = jd->jd_jid; 683 if (jd->jd_jid == sdp->sd_jdesc->jd_jid) 684 sdp->sd_sc_inode = lsi->si_sc_inode; 685 686 list_add_tail(&lsi->si_list, &sdp->sd_sc_inodes_list); 687 } 688 689 iput(pn); 690 pn = NULL; 691 ip = GFS2_I(sdp->sd_sc_inode); 692 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, 693 &sdp->sd_sc_gh); 694 if (error) { 695 fs_err(sdp, "can't lock local \"sc\" file: %d\n", error); 696 goto free_local; 697 } 698 return 0; 699 700 free_local: 701 free_local_statfs_inodes(sdp); 702 iput(pn); 703 put_statfs: 704 iput(sdp->sd_statfs_inode); 705 out: 706 return error; 707 } 708 709 /* Uninitialize and free up memory used by the list of statfs inodes */ 710 static void uninit_statfs(struct gfs2_sbd *sdp) 711 { 712 if (!sdp->sd_args.ar_spectator) { 713 gfs2_glock_dq_uninit(&sdp->sd_sc_gh); 714 free_local_statfs_inodes(sdp); 715 } 716 iput(sdp->sd_statfs_inode); 717 } 718 719 static int init_journal(struct gfs2_sbd *sdp, int undo) 720 { 721 struct inode *master = d_inode(sdp->sd_master_dir); 722 struct gfs2_holder ji_gh; 723 struct gfs2_inode *ip; 724 int jindex = 1; 725 int error = 0; 726 727 if (undo) { 728 jindex = 0; 729 goto fail_statfs; 730 } 731 732 sdp->sd_jindex = gfs2_lookup_simple(master, "jindex"); 733 if (IS_ERR(sdp->sd_jindex)) { 734 fs_err(sdp, "can't lookup journal index: %d\n", error); 735 return PTR_ERR(sdp->sd_jindex); 736 } 737 738 /* Load in the journal index special file */ 739 740 error = gfs2_jindex_hold(sdp, &ji_gh); 741 if (error) { 742 fs_err(sdp, "can't read journal index: %d\n", error); 743 goto fail; 744 } 745 746 error = -EUSERS; 747 if (!gfs2_jindex_size(sdp)) { 748 fs_err(sdp, "no journals!\n"); 749 goto fail_jindex; 750 } 751 752 atomic_set(&sdp->sd_log_blks_needed, 0); 753 if (sdp->sd_args.ar_spectator) { 754 sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0); 755 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks); 756 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5); 757 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5); 758 } else { 759 if (sdp->sd_lockstruct.ls_jid >= gfs2_jindex_size(sdp)) { 760 fs_err(sdp, "can't mount journal #%u\n", 761 sdp->sd_lockstruct.ls_jid); 762 fs_err(sdp, "there are only %u journals (0 - %u)\n", 763 gfs2_jindex_size(sdp), 764 gfs2_jindex_size(sdp) - 1); 765 goto fail_jindex; 766 } 767 sdp->sd_jdesc = gfs2_jdesc_find(sdp, sdp->sd_lockstruct.ls_jid); 768 769 error = gfs2_glock_nq_num(sdp, sdp->sd_lockstruct.ls_jid, 770 &gfs2_journal_glops, 771 LM_ST_EXCLUSIVE, 772 LM_FLAG_NOEXP | GL_NOCACHE, 773 &sdp->sd_journal_gh); 774 if (error) { 775 fs_err(sdp, "can't acquire journal glock: %d\n", error); 776 goto fail_jindex; 777 } 778 779 ip = GFS2_I(sdp->sd_jdesc->jd_inode); 780 sdp->sd_jinode_gl = ip->i_gl; 781 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 782 LM_FLAG_NOEXP | GL_EXACT | GL_NOCACHE, 783 &sdp->sd_jinode_gh); 784 if (error) { 785 fs_err(sdp, "can't acquire journal inode glock: %d\n", 786 error); 787 goto fail_journal_gh; 788 } 789 790 error = gfs2_jdesc_check(sdp->sd_jdesc); 791 if (error) { 792 fs_err(sdp, "my journal (%u) is bad: %d\n", 793 sdp->sd_jdesc->jd_jid, error); 794 goto fail_jinode_gh; 795 } 796 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks); 797 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5); 798 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5); 799 800 /* Map the extents for this journal's blocks */ 801 gfs2_map_journal_extents(sdp, sdp->sd_jdesc); 802 } 803 trace_gfs2_log_blocks(sdp, atomic_read(&sdp->sd_log_blks_free)); 804 805 /* Lookup statfs inodes here so journal recovery can use them. */ 806 error = init_statfs(sdp); 807 if (error) 808 goto fail_jinode_gh; 809 810 if (sdp->sd_lockstruct.ls_first) { 811 unsigned int x; 812 for (x = 0; x < sdp->sd_journals; x++) { 813 struct gfs2_jdesc *jd = gfs2_jdesc_find(sdp, x); 814 815 if (sdp->sd_args.ar_spectator) { 816 error = check_journal_clean(sdp, jd, true); 817 if (error) 818 goto fail_statfs; 819 continue; 820 } 821 error = gfs2_recover_journal(jd, true); 822 if (error) { 823 fs_err(sdp, "error recovering journal %u: %d\n", 824 x, error); 825 goto fail_statfs; 826 } 827 } 828 829 gfs2_others_may_mount(sdp); 830 } else if (!sdp->sd_args.ar_spectator) { 831 error = gfs2_recover_journal(sdp->sd_jdesc, true); 832 if (error) { 833 fs_err(sdp, "error recovering my journal: %d\n", error); 834 goto fail_statfs; 835 } 836 } 837 838 sdp->sd_log_idle = 1; 839 set_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags); 840 gfs2_glock_dq_uninit(&ji_gh); 841 jindex = 0; 842 INIT_WORK(&sdp->sd_freeze_work, gfs2_freeze_func); 843 return 0; 844 845 fail_statfs: 846 uninit_statfs(sdp); 847 fail_jinode_gh: 848 /* A withdraw may have done dq/uninit so now we need to check it */ 849 if (!sdp->sd_args.ar_spectator && 850 gfs2_holder_initialized(&sdp->sd_jinode_gh)) 851 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh); 852 fail_journal_gh: 853 if (!sdp->sd_args.ar_spectator && 854 gfs2_holder_initialized(&sdp->sd_journal_gh)) 855 gfs2_glock_dq_uninit(&sdp->sd_journal_gh); 856 fail_jindex: 857 gfs2_jindex_free(sdp); 858 if (jindex) 859 gfs2_glock_dq_uninit(&ji_gh); 860 fail: 861 iput(sdp->sd_jindex); 862 return error; 863 } 864 865 static struct lock_class_key gfs2_quota_imutex_key; 866 867 static int init_inodes(struct gfs2_sbd *sdp, int undo) 868 { 869 int error = 0; 870 struct inode *master = d_inode(sdp->sd_master_dir); 871 872 if (undo) 873 goto fail_qinode; 874 875 error = init_journal(sdp, undo); 876 complete_all(&sdp->sd_journal_ready); 877 if (error) 878 goto fail; 879 880 /* Read in the resource index inode */ 881 sdp->sd_rindex = gfs2_lookup_simple(master, "rindex"); 882 if (IS_ERR(sdp->sd_rindex)) { 883 error = PTR_ERR(sdp->sd_rindex); 884 fs_err(sdp, "can't get resource index inode: %d\n", error); 885 goto fail_journal; 886 } 887 sdp->sd_rindex_uptodate = 0; 888 889 /* Read in the quota inode */ 890 sdp->sd_quota_inode = gfs2_lookup_simple(master, "quota"); 891 if (IS_ERR(sdp->sd_quota_inode)) { 892 error = PTR_ERR(sdp->sd_quota_inode); 893 fs_err(sdp, "can't get quota file inode: %d\n", error); 894 goto fail_rindex; 895 } 896 /* 897 * i_rwsem on quota files is special. Since this inode is hidden system 898 * file, we are safe to define locking ourselves. 899 */ 900 lockdep_set_class(&sdp->sd_quota_inode->i_rwsem, 901 &gfs2_quota_imutex_key); 902 903 error = gfs2_rindex_update(sdp); 904 if (error) 905 goto fail_qinode; 906 907 return 0; 908 909 fail_qinode: 910 iput(sdp->sd_quota_inode); 911 fail_rindex: 912 gfs2_clear_rgrpd(sdp); 913 iput(sdp->sd_rindex); 914 fail_journal: 915 init_journal(sdp, UNDO); 916 fail: 917 return error; 918 } 919 920 static int init_per_node(struct gfs2_sbd *sdp, int undo) 921 { 922 struct inode *pn = NULL; 923 char buf[30]; 924 int error = 0; 925 struct gfs2_inode *ip; 926 struct inode *master = d_inode(sdp->sd_master_dir); 927 928 if (sdp->sd_args.ar_spectator) 929 return 0; 930 931 if (undo) 932 goto fail_qc_gh; 933 934 pn = gfs2_lookup_simple(master, "per_node"); 935 if (IS_ERR(pn)) { 936 error = PTR_ERR(pn); 937 fs_err(sdp, "can't find per_node directory: %d\n", error); 938 return error; 939 } 940 941 sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid); 942 sdp->sd_qc_inode = gfs2_lookup_simple(pn, buf); 943 if (IS_ERR(sdp->sd_qc_inode)) { 944 error = PTR_ERR(sdp->sd_qc_inode); 945 fs_err(sdp, "can't find local \"qc\" file: %d\n", error); 946 goto fail_ut_i; 947 } 948 949 iput(pn); 950 pn = NULL; 951 952 ip = GFS2_I(sdp->sd_qc_inode); 953 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, 954 &sdp->sd_qc_gh); 955 if (error) { 956 fs_err(sdp, "can't lock local \"qc\" file: %d\n", error); 957 goto fail_qc_i; 958 } 959 960 return 0; 961 962 fail_qc_gh: 963 gfs2_glock_dq_uninit(&sdp->sd_qc_gh); 964 fail_qc_i: 965 iput(sdp->sd_qc_inode); 966 fail_ut_i: 967 iput(pn); 968 return error; 969 } 970 971 static const match_table_t nolock_tokens = { 972 { Opt_jid, "jid=%d", }, 973 { Opt_err, NULL }, 974 }; 975 976 static const struct lm_lockops nolock_ops = { 977 .lm_proto_name = "lock_nolock", 978 .lm_put_lock = gfs2_glock_free, 979 .lm_tokens = &nolock_tokens, 980 }; 981 982 /** 983 * gfs2_lm_mount - mount a locking protocol 984 * @sdp: the filesystem 985 * @silent: if 1, don't complain if the FS isn't a GFS2 fs 986 * 987 * Returns: errno 988 */ 989 990 static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent) 991 { 992 const struct lm_lockops *lm; 993 struct lm_lockstruct *ls = &sdp->sd_lockstruct; 994 struct gfs2_args *args = &sdp->sd_args; 995 const char *proto = sdp->sd_proto_name; 996 const char *table = sdp->sd_table_name; 997 char *o, *options; 998 int ret; 999 1000 if (!strcmp("lock_nolock", proto)) { 1001 lm = &nolock_ops; 1002 sdp->sd_args.ar_localflocks = 1; 1003 #ifdef CONFIG_GFS2_FS_LOCKING_DLM 1004 } else if (!strcmp("lock_dlm", proto)) { 1005 lm = &gfs2_dlm_ops; 1006 #endif 1007 } else { 1008 pr_info("can't find protocol %s\n", proto); 1009 return -ENOENT; 1010 } 1011 1012 fs_info(sdp, "Trying to join cluster \"%s\", \"%s\"\n", proto, table); 1013 1014 ls->ls_ops = lm; 1015 ls->ls_first = 1; 1016 1017 for (options = args->ar_hostdata; (o = strsep(&options, ":")); ) { 1018 substring_t tmp[MAX_OPT_ARGS]; 1019 int token, option; 1020 1021 if (!o || !*o) 1022 continue; 1023 1024 token = match_token(o, *lm->lm_tokens, tmp); 1025 switch (token) { 1026 case Opt_jid: 1027 ret = match_int(&tmp[0], &option); 1028 if (ret || option < 0) 1029 goto hostdata_error; 1030 if (test_and_clear_bit(SDF_NOJOURNALID, &sdp->sd_flags)) 1031 ls->ls_jid = option; 1032 break; 1033 case Opt_id: 1034 case Opt_nodir: 1035 /* Obsolete, but left for backward compat purposes */ 1036 break; 1037 case Opt_first: 1038 ret = match_int(&tmp[0], &option); 1039 if (ret || (option != 0 && option != 1)) 1040 goto hostdata_error; 1041 ls->ls_first = option; 1042 break; 1043 case Opt_err: 1044 default: 1045 hostdata_error: 1046 fs_info(sdp, "unknown hostdata (%s)\n", o); 1047 return -EINVAL; 1048 } 1049 } 1050 1051 if (lm->lm_mount == NULL) { 1052 fs_info(sdp, "Now mounting FS (format %u)...\n", sdp->sd_sb.sb_fs_format); 1053 complete_all(&sdp->sd_locking_init); 1054 return 0; 1055 } 1056 ret = lm->lm_mount(sdp, table); 1057 if (ret == 0) 1058 fs_info(sdp, "Joined cluster. Now mounting FS (format %u)...\n", 1059 sdp->sd_sb.sb_fs_format); 1060 complete_all(&sdp->sd_locking_init); 1061 return ret; 1062 } 1063 1064 void gfs2_lm_unmount(struct gfs2_sbd *sdp) 1065 { 1066 const struct lm_lockops *lm = sdp->sd_lockstruct.ls_ops; 1067 if (likely(!gfs2_withdrawn(sdp)) && lm->lm_unmount) 1068 lm->lm_unmount(sdp); 1069 } 1070 1071 static int wait_on_journal(struct gfs2_sbd *sdp) 1072 { 1073 if (sdp->sd_lockstruct.ls_ops->lm_mount == NULL) 1074 return 0; 1075 1076 return wait_on_bit(&sdp->sd_flags, SDF_NOJOURNALID, TASK_INTERRUPTIBLE) 1077 ? -EINTR : 0; 1078 } 1079 1080 void gfs2_online_uevent(struct gfs2_sbd *sdp) 1081 { 1082 struct super_block *sb = sdp->sd_vfs; 1083 char ro[20]; 1084 char spectator[20]; 1085 char *envp[] = { ro, spectator, NULL }; 1086 sprintf(ro, "RDONLY=%d", sb_rdonly(sb)); 1087 sprintf(spectator, "SPECTATOR=%d", sdp->sd_args.ar_spectator ? 1 : 0); 1088 kobject_uevent_env(&sdp->sd_kobj, KOBJ_ONLINE, envp); 1089 } 1090 1091 /** 1092 * gfs2_fill_super - Read in superblock 1093 * @sb: The VFS superblock 1094 * @fc: Mount options and flags 1095 * 1096 * Returns: -errno 1097 */ 1098 static int gfs2_fill_super(struct super_block *sb, struct fs_context *fc) 1099 { 1100 struct gfs2_args *args = fc->fs_private; 1101 int silent = fc->sb_flags & SB_SILENT; 1102 struct gfs2_sbd *sdp; 1103 struct gfs2_holder mount_gh; 1104 struct gfs2_holder freeze_gh; 1105 int error; 1106 1107 sdp = init_sbd(sb); 1108 if (!sdp) { 1109 pr_warn("can't alloc struct gfs2_sbd\n"); 1110 return -ENOMEM; 1111 } 1112 sdp->sd_args = *args; 1113 1114 if (sdp->sd_args.ar_spectator) { 1115 sb->s_flags |= SB_RDONLY; 1116 set_bit(SDF_RORECOVERY, &sdp->sd_flags); 1117 } 1118 if (sdp->sd_args.ar_posix_acl) 1119 sb->s_flags |= SB_POSIXACL; 1120 if (sdp->sd_args.ar_nobarrier) 1121 set_bit(SDF_NOBARRIERS, &sdp->sd_flags); 1122 1123 sb->s_flags |= SB_NOSEC; 1124 sb->s_magic = GFS2_MAGIC; 1125 sb->s_op = &gfs2_super_ops; 1126 sb->s_d_op = &gfs2_dops; 1127 sb->s_export_op = &gfs2_export_ops; 1128 sb->s_qcop = &gfs2_quotactl_ops; 1129 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP; 1130 sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE; 1131 sb->s_time_gran = 1; 1132 sb->s_maxbytes = MAX_LFS_FILESIZE; 1133 1134 /* Set up the buffer cache and fill in some fake block size values 1135 to allow us to read-in the on-disk superblock. */ 1136 sdp->sd_sb.sb_bsize = sb_min_blocksize(sb, GFS2_BASIC_BLOCK); 1137 sdp->sd_sb.sb_bsize_shift = sb->s_blocksize_bits; 1138 sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift - 1139 GFS2_BASIC_BLOCK_SHIFT; 1140 sdp->sd_fsb2bb = BIT(sdp->sd_fsb2bb_shift); 1141 1142 sdp->sd_tune.gt_logd_secs = sdp->sd_args.ar_commit; 1143 sdp->sd_tune.gt_quota_quantum = sdp->sd_args.ar_quota_quantum; 1144 if (sdp->sd_args.ar_statfs_quantum) { 1145 sdp->sd_tune.gt_statfs_slow = 0; 1146 sdp->sd_tune.gt_statfs_quantum = sdp->sd_args.ar_statfs_quantum; 1147 } else { 1148 sdp->sd_tune.gt_statfs_slow = 1; 1149 sdp->sd_tune.gt_statfs_quantum = 30; 1150 } 1151 1152 error = init_names(sdp, silent); 1153 if (error) 1154 goto fail_free; 1155 1156 snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s", sdp->sd_table_name); 1157 1158 error = gfs2_sys_fs_add(sdp); 1159 if (error) 1160 goto fail_free; 1161 1162 gfs2_create_debugfs_file(sdp); 1163 1164 error = gfs2_lm_mount(sdp, silent); 1165 if (error) 1166 goto fail_debug; 1167 1168 error = init_locking(sdp, &mount_gh, DO); 1169 if (error) 1170 goto fail_lm; 1171 1172 error = init_sb(sdp, silent); 1173 if (error) 1174 goto fail_locking; 1175 1176 /* Turn rgrplvb on by default if fs format is recent enough */ 1177 if (!sdp->sd_args.ar_got_rgrplvb && sdp->sd_sb.sb_fs_format > 1801) 1178 sdp->sd_args.ar_rgrplvb = 1; 1179 1180 error = wait_on_journal(sdp); 1181 if (error) 1182 goto fail_sb; 1183 1184 /* 1185 * If user space has failed to join the cluster or some similar 1186 * failure has occurred, then the journal id will contain a 1187 * negative (error) number. This will then be returned to the 1188 * caller (of the mount syscall). We do this even for spectator 1189 * mounts (which just write a jid of 0 to indicate "ok" even though 1190 * the jid is unused in the spectator case) 1191 */ 1192 if (sdp->sd_lockstruct.ls_jid < 0) { 1193 error = sdp->sd_lockstruct.ls_jid; 1194 sdp->sd_lockstruct.ls_jid = 0; 1195 goto fail_sb; 1196 } 1197 1198 if (sdp->sd_args.ar_spectator) 1199 snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s.s", 1200 sdp->sd_table_name); 1201 else 1202 snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s.%u", 1203 sdp->sd_table_name, sdp->sd_lockstruct.ls_jid); 1204 1205 error = init_inodes(sdp, DO); 1206 if (error) 1207 goto fail_sb; 1208 1209 error = init_per_node(sdp, DO); 1210 if (error) 1211 goto fail_inodes; 1212 1213 error = gfs2_statfs_init(sdp); 1214 if (error) { 1215 fs_err(sdp, "can't initialize statfs subsystem: %d\n", error); 1216 goto fail_per_node; 1217 } 1218 1219 error = gfs2_freeze_lock(sdp, &freeze_gh, 0); 1220 if (error) 1221 goto fail_per_node; 1222 1223 if (!sb_rdonly(sb)) 1224 error = gfs2_make_fs_rw(sdp); 1225 1226 gfs2_freeze_unlock(&freeze_gh); 1227 if (error) { 1228 fs_err(sdp, "can't make FS RW: %d\n", error); 1229 goto fail_per_node; 1230 } 1231 gfs2_glock_dq_uninit(&mount_gh); 1232 gfs2_online_uevent(sdp); 1233 return 0; 1234 1235 fail_per_node: 1236 init_per_node(sdp, UNDO); 1237 fail_inodes: 1238 init_inodes(sdp, UNDO); 1239 fail_sb: 1240 if (sdp->sd_root_dir) 1241 dput(sdp->sd_root_dir); 1242 if (sdp->sd_master_dir) 1243 dput(sdp->sd_master_dir); 1244 if (sb->s_root) 1245 dput(sb->s_root); 1246 sb->s_root = NULL; 1247 fail_locking: 1248 init_locking(sdp, &mount_gh, UNDO); 1249 fail_lm: 1250 complete_all(&sdp->sd_journal_ready); 1251 gfs2_gl_hash_clear(sdp); 1252 gfs2_lm_unmount(sdp); 1253 fail_debug: 1254 gfs2_delete_debugfs_file(sdp); 1255 gfs2_sys_fs_del(sdp); 1256 fail_free: 1257 free_sbd(sdp); 1258 sb->s_fs_info = NULL; 1259 return error; 1260 } 1261 1262 /** 1263 * gfs2_get_tree - Get the GFS2 superblock and root directory 1264 * @fc: The filesystem context 1265 * 1266 * Returns: 0 or -errno on error 1267 */ 1268 static int gfs2_get_tree(struct fs_context *fc) 1269 { 1270 struct gfs2_args *args = fc->fs_private; 1271 struct gfs2_sbd *sdp; 1272 int error; 1273 1274 error = get_tree_bdev(fc, gfs2_fill_super); 1275 if (error) 1276 return error; 1277 1278 sdp = fc->root->d_sb->s_fs_info; 1279 dput(fc->root); 1280 if (args->ar_meta) 1281 fc->root = dget(sdp->sd_master_dir); 1282 else 1283 fc->root = dget(sdp->sd_root_dir); 1284 return 0; 1285 } 1286 1287 static void gfs2_fc_free(struct fs_context *fc) 1288 { 1289 struct gfs2_args *args = fc->fs_private; 1290 1291 kfree(args); 1292 } 1293 1294 enum gfs2_param { 1295 Opt_lockproto, 1296 Opt_locktable, 1297 Opt_hostdata, 1298 Opt_spectator, 1299 Opt_ignore_local_fs, 1300 Opt_localflocks, 1301 Opt_localcaching, 1302 Opt_debug, 1303 Opt_upgrade, 1304 Opt_acl, 1305 Opt_quota, 1306 Opt_quota_flag, 1307 Opt_suiddir, 1308 Opt_data, 1309 Opt_meta, 1310 Opt_discard, 1311 Opt_commit, 1312 Opt_errors, 1313 Opt_statfs_quantum, 1314 Opt_statfs_percent, 1315 Opt_quota_quantum, 1316 Opt_barrier, 1317 Opt_rgrplvb, 1318 Opt_loccookie, 1319 }; 1320 1321 static const struct constant_table gfs2_param_quota[] = { 1322 {"off", GFS2_QUOTA_OFF}, 1323 {"account", GFS2_QUOTA_ACCOUNT}, 1324 {"on", GFS2_QUOTA_ON}, 1325 {} 1326 }; 1327 1328 enum opt_data { 1329 Opt_data_writeback = GFS2_DATA_WRITEBACK, 1330 Opt_data_ordered = GFS2_DATA_ORDERED, 1331 }; 1332 1333 static const struct constant_table gfs2_param_data[] = { 1334 {"writeback", Opt_data_writeback }, 1335 {"ordered", Opt_data_ordered }, 1336 {} 1337 }; 1338 1339 enum opt_errors { 1340 Opt_errors_withdraw = GFS2_ERRORS_WITHDRAW, 1341 Opt_errors_panic = GFS2_ERRORS_PANIC, 1342 }; 1343 1344 static const struct constant_table gfs2_param_errors[] = { 1345 {"withdraw", Opt_errors_withdraw }, 1346 {"panic", Opt_errors_panic }, 1347 {} 1348 }; 1349 1350 static const struct fs_parameter_spec gfs2_fs_parameters[] = { 1351 fsparam_string ("lockproto", Opt_lockproto), 1352 fsparam_string ("locktable", Opt_locktable), 1353 fsparam_string ("hostdata", Opt_hostdata), 1354 fsparam_flag ("spectator", Opt_spectator), 1355 fsparam_flag ("norecovery", Opt_spectator), 1356 fsparam_flag ("ignore_local_fs", Opt_ignore_local_fs), 1357 fsparam_flag ("localflocks", Opt_localflocks), 1358 fsparam_flag ("localcaching", Opt_localcaching), 1359 fsparam_flag_no("debug", Opt_debug), 1360 fsparam_flag ("upgrade", Opt_upgrade), 1361 fsparam_flag_no("acl", Opt_acl), 1362 fsparam_flag_no("suiddir", Opt_suiddir), 1363 fsparam_enum ("data", Opt_data, gfs2_param_data), 1364 fsparam_flag ("meta", Opt_meta), 1365 fsparam_flag_no("discard", Opt_discard), 1366 fsparam_s32 ("commit", Opt_commit), 1367 fsparam_enum ("errors", Opt_errors, gfs2_param_errors), 1368 fsparam_s32 ("statfs_quantum", Opt_statfs_quantum), 1369 fsparam_s32 ("statfs_percent", Opt_statfs_percent), 1370 fsparam_s32 ("quota_quantum", Opt_quota_quantum), 1371 fsparam_flag_no("barrier", Opt_barrier), 1372 fsparam_flag_no("rgrplvb", Opt_rgrplvb), 1373 fsparam_flag_no("loccookie", Opt_loccookie), 1374 /* quota can be a flag or an enum so it gets special treatment */ 1375 fsparam_flag_no("quota", Opt_quota_flag), 1376 fsparam_enum("quota", Opt_quota, gfs2_param_quota), 1377 {} 1378 }; 1379 1380 /* Parse a single mount parameter */ 1381 static int gfs2_parse_param(struct fs_context *fc, struct fs_parameter *param) 1382 { 1383 struct gfs2_args *args = fc->fs_private; 1384 struct fs_parse_result result; 1385 int o; 1386 1387 o = fs_parse(fc, gfs2_fs_parameters, param, &result); 1388 if (o < 0) 1389 return o; 1390 1391 switch (o) { 1392 case Opt_lockproto: 1393 strlcpy(args->ar_lockproto, param->string, GFS2_LOCKNAME_LEN); 1394 break; 1395 case Opt_locktable: 1396 strlcpy(args->ar_locktable, param->string, GFS2_LOCKNAME_LEN); 1397 break; 1398 case Opt_hostdata: 1399 strlcpy(args->ar_hostdata, param->string, GFS2_LOCKNAME_LEN); 1400 break; 1401 case Opt_spectator: 1402 args->ar_spectator = 1; 1403 break; 1404 case Opt_ignore_local_fs: 1405 /* Retained for backwards compat only */ 1406 break; 1407 case Opt_localflocks: 1408 args->ar_localflocks = 1; 1409 break; 1410 case Opt_localcaching: 1411 /* Retained for backwards compat only */ 1412 break; 1413 case Opt_debug: 1414 if (result.boolean && args->ar_errors == GFS2_ERRORS_PANIC) 1415 return invalfc(fc, "-o debug and -o errors=panic are mutually exclusive"); 1416 args->ar_debug = result.boolean; 1417 break; 1418 case Opt_upgrade: 1419 /* Retained for backwards compat only */ 1420 break; 1421 case Opt_acl: 1422 args->ar_posix_acl = result.boolean; 1423 break; 1424 case Opt_quota_flag: 1425 args->ar_quota = result.negated ? GFS2_QUOTA_OFF : GFS2_QUOTA_ON; 1426 break; 1427 case Opt_quota: 1428 args->ar_quota = result.int_32; 1429 break; 1430 case Opt_suiddir: 1431 args->ar_suiddir = result.boolean; 1432 break; 1433 case Opt_data: 1434 /* The uint_32 result maps directly to GFS2_DATA_* */ 1435 args->ar_data = result.uint_32; 1436 break; 1437 case Opt_meta: 1438 args->ar_meta = 1; 1439 break; 1440 case Opt_discard: 1441 args->ar_discard = result.boolean; 1442 break; 1443 case Opt_commit: 1444 if (result.int_32 <= 0) 1445 return invalfc(fc, "commit mount option requires a positive numeric argument"); 1446 args->ar_commit = result.int_32; 1447 break; 1448 case Opt_statfs_quantum: 1449 if (result.int_32 < 0) 1450 return invalfc(fc, "statfs_quantum mount option requires a non-negative numeric argument"); 1451 args->ar_statfs_quantum = result.int_32; 1452 break; 1453 case Opt_quota_quantum: 1454 if (result.int_32 <= 0) 1455 return invalfc(fc, "quota_quantum mount option requires a positive numeric argument"); 1456 args->ar_quota_quantum = result.int_32; 1457 break; 1458 case Opt_statfs_percent: 1459 if (result.int_32 < 0 || result.int_32 > 100) 1460 return invalfc(fc, "statfs_percent mount option requires a numeric argument between 0 and 100"); 1461 args->ar_statfs_percent = result.int_32; 1462 break; 1463 case Opt_errors: 1464 if (args->ar_debug && result.uint_32 == GFS2_ERRORS_PANIC) 1465 return invalfc(fc, "-o debug and -o errors=panic are mutually exclusive"); 1466 args->ar_errors = result.uint_32; 1467 break; 1468 case Opt_barrier: 1469 args->ar_nobarrier = result.boolean; 1470 break; 1471 case Opt_rgrplvb: 1472 args->ar_rgrplvb = result.boolean; 1473 args->ar_got_rgrplvb = 1; 1474 break; 1475 case Opt_loccookie: 1476 args->ar_loccookie = result.boolean; 1477 break; 1478 default: 1479 return invalfc(fc, "invalid mount option: %s", param->key); 1480 } 1481 return 0; 1482 } 1483 1484 static int gfs2_reconfigure(struct fs_context *fc) 1485 { 1486 struct super_block *sb = fc->root->d_sb; 1487 struct gfs2_sbd *sdp = sb->s_fs_info; 1488 struct gfs2_args *oldargs = &sdp->sd_args; 1489 struct gfs2_args *newargs = fc->fs_private; 1490 struct gfs2_tune *gt = &sdp->sd_tune; 1491 int error = 0; 1492 1493 sync_filesystem(sb); 1494 1495 spin_lock(>->gt_spin); 1496 oldargs->ar_commit = gt->gt_logd_secs; 1497 oldargs->ar_quota_quantum = gt->gt_quota_quantum; 1498 if (gt->gt_statfs_slow) 1499 oldargs->ar_statfs_quantum = 0; 1500 else 1501 oldargs->ar_statfs_quantum = gt->gt_statfs_quantum; 1502 spin_unlock(>->gt_spin); 1503 1504 if (strcmp(newargs->ar_lockproto, oldargs->ar_lockproto)) { 1505 errorfc(fc, "reconfiguration of locking protocol not allowed"); 1506 return -EINVAL; 1507 } 1508 if (strcmp(newargs->ar_locktable, oldargs->ar_locktable)) { 1509 errorfc(fc, "reconfiguration of lock table not allowed"); 1510 return -EINVAL; 1511 } 1512 if (strcmp(newargs->ar_hostdata, oldargs->ar_hostdata)) { 1513 errorfc(fc, "reconfiguration of host data not allowed"); 1514 return -EINVAL; 1515 } 1516 if (newargs->ar_spectator != oldargs->ar_spectator) { 1517 errorfc(fc, "reconfiguration of spectator mode not allowed"); 1518 return -EINVAL; 1519 } 1520 if (newargs->ar_localflocks != oldargs->ar_localflocks) { 1521 errorfc(fc, "reconfiguration of localflocks not allowed"); 1522 return -EINVAL; 1523 } 1524 if (newargs->ar_meta != oldargs->ar_meta) { 1525 errorfc(fc, "switching between gfs2 and gfs2meta not allowed"); 1526 return -EINVAL; 1527 } 1528 if (oldargs->ar_spectator) 1529 fc->sb_flags |= SB_RDONLY; 1530 1531 if ((sb->s_flags ^ fc->sb_flags) & SB_RDONLY) { 1532 struct gfs2_holder freeze_gh; 1533 1534 error = gfs2_freeze_lock(sdp, &freeze_gh, 0); 1535 if (error) 1536 return -EINVAL; 1537 1538 if (fc->sb_flags & SB_RDONLY) { 1539 gfs2_make_fs_ro(sdp); 1540 } else { 1541 error = gfs2_make_fs_rw(sdp); 1542 if (error) 1543 errorfc(fc, "unable to remount read-write"); 1544 } 1545 gfs2_freeze_unlock(&freeze_gh); 1546 } 1547 sdp->sd_args = *newargs; 1548 1549 if (sdp->sd_args.ar_posix_acl) 1550 sb->s_flags |= SB_POSIXACL; 1551 else 1552 sb->s_flags &= ~SB_POSIXACL; 1553 if (sdp->sd_args.ar_nobarrier) 1554 set_bit(SDF_NOBARRIERS, &sdp->sd_flags); 1555 else 1556 clear_bit(SDF_NOBARRIERS, &sdp->sd_flags); 1557 spin_lock(>->gt_spin); 1558 gt->gt_logd_secs = newargs->ar_commit; 1559 gt->gt_quota_quantum = newargs->ar_quota_quantum; 1560 if (newargs->ar_statfs_quantum) { 1561 gt->gt_statfs_slow = 0; 1562 gt->gt_statfs_quantum = newargs->ar_statfs_quantum; 1563 } 1564 else { 1565 gt->gt_statfs_slow = 1; 1566 gt->gt_statfs_quantum = 30; 1567 } 1568 spin_unlock(>->gt_spin); 1569 1570 gfs2_online_uevent(sdp); 1571 return error; 1572 } 1573 1574 static const struct fs_context_operations gfs2_context_ops = { 1575 .free = gfs2_fc_free, 1576 .parse_param = gfs2_parse_param, 1577 .get_tree = gfs2_get_tree, 1578 .reconfigure = gfs2_reconfigure, 1579 }; 1580 1581 /* Set up the filesystem mount context */ 1582 static int gfs2_init_fs_context(struct fs_context *fc) 1583 { 1584 struct gfs2_args *args; 1585 1586 args = kmalloc(sizeof(*args), GFP_KERNEL); 1587 if (args == NULL) 1588 return -ENOMEM; 1589 1590 if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) { 1591 struct gfs2_sbd *sdp = fc->root->d_sb->s_fs_info; 1592 1593 *args = sdp->sd_args; 1594 } else { 1595 memset(args, 0, sizeof(*args)); 1596 args->ar_quota = GFS2_QUOTA_DEFAULT; 1597 args->ar_data = GFS2_DATA_DEFAULT; 1598 args->ar_commit = 30; 1599 args->ar_statfs_quantum = 30; 1600 args->ar_quota_quantum = 60; 1601 args->ar_errors = GFS2_ERRORS_DEFAULT; 1602 } 1603 fc->fs_private = args; 1604 fc->ops = &gfs2_context_ops; 1605 return 0; 1606 } 1607 1608 static int set_meta_super(struct super_block *s, struct fs_context *fc) 1609 { 1610 return -EINVAL; 1611 } 1612 1613 static int test_meta_super(struct super_block *s, struct fs_context *fc) 1614 { 1615 return (fc->sget_key == s->s_bdev); 1616 } 1617 1618 static int gfs2_meta_get_tree(struct fs_context *fc) 1619 { 1620 struct super_block *s; 1621 struct gfs2_sbd *sdp; 1622 struct path path; 1623 int error; 1624 1625 if (!fc->source || !*fc->source) 1626 return -EINVAL; 1627 1628 error = kern_path(fc->source, LOOKUP_FOLLOW, &path); 1629 if (error) { 1630 pr_warn("path_lookup on %s returned error %d\n", 1631 fc->source, error); 1632 return error; 1633 } 1634 fc->fs_type = &gfs2_fs_type; 1635 fc->sget_key = path.dentry->d_sb->s_bdev; 1636 s = sget_fc(fc, test_meta_super, set_meta_super); 1637 path_put(&path); 1638 if (IS_ERR(s)) { 1639 pr_warn("gfs2 mount does not exist\n"); 1640 return PTR_ERR(s); 1641 } 1642 if ((fc->sb_flags ^ s->s_flags) & SB_RDONLY) { 1643 deactivate_locked_super(s); 1644 return -EBUSY; 1645 } 1646 sdp = s->s_fs_info; 1647 fc->root = dget(sdp->sd_master_dir); 1648 return 0; 1649 } 1650 1651 static const struct fs_context_operations gfs2_meta_context_ops = { 1652 .free = gfs2_fc_free, 1653 .get_tree = gfs2_meta_get_tree, 1654 }; 1655 1656 static int gfs2_meta_init_fs_context(struct fs_context *fc) 1657 { 1658 int ret = gfs2_init_fs_context(fc); 1659 1660 if (ret) 1661 return ret; 1662 1663 fc->ops = &gfs2_meta_context_ops; 1664 return 0; 1665 } 1666 1667 static void gfs2_kill_sb(struct super_block *sb) 1668 { 1669 struct gfs2_sbd *sdp = sb->s_fs_info; 1670 1671 if (sdp == NULL) { 1672 kill_block_super(sb); 1673 return; 1674 } 1675 1676 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SYNC | GFS2_LFC_KILL_SB); 1677 dput(sdp->sd_root_dir); 1678 dput(sdp->sd_master_dir); 1679 sdp->sd_root_dir = NULL; 1680 sdp->sd_master_dir = NULL; 1681 shrink_dcache_sb(sb); 1682 kill_block_super(sb); 1683 } 1684 1685 struct file_system_type gfs2_fs_type = { 1686 .name = "gfs2", 1687 .fs_flags = FS_REQUIRES_DEV, 1688 .init_fs_context = gfs2_init_fs_context, 1689 .parameters = gfs2_fs_parameters, 1690 .kill_sb = gfs2_kill_sb, 1691 .owner = THIS_MODULE, 1692 }; 1693 MODULE_ALIAS_FS("gfs2"); 1694 1695 struct file_system_type gfs2meta_fs_type = { 1696 .name = "gfs2meta", 1697 .fs_flags = FS_REQUIRES_DEV, 1698 .init_fs_context = gfs2_meta_init_fs_context, 1699 .owner = THIS_MODULE, 1700 }; 1701 MODULE_ALIAS_FS("gfs2meta"); 1702