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 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 atomic_set(&sdp->sd_freeze_state, SFS_UNFROZEN); 139 mutex_init(&sdp->sd_freeze_mutex); 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 transaction 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_simple(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_simple(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_simple(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 jindex = 1; 738 int error = 0; 739 740 if (undo) { 741 jindex = 0; 742 goto fail_statfs; 743 } 744 745 sdp->sd_jindex = gfs2_lookup_simple(master, "jindex"); 746 if (IS_ERR(sdp->sd_jindex)) { 747 fs_err(sdp, "can't lookup journal index: %d\n", error); 748 return PTR_ERR(sdp->sd_jindex); 749 } 750 751 /* Load in the journal index special file */ 752 753 error = gfs2_jindex_hold(sdp, &ji_gh); 754 if (error) { 755 fs_err(sdp, "can't read journal index: %d\n", error); 756 goto fail; 757 } 758 759 error = -EUSERS; 760 if (!gfs2_jindex_size(sdp)) { 761 fs_err(sdp, "no journals!\n"); 762 goto fail_jindex; 763 } 764 765 atomic_set(&sdp->sd_log_blks_needed, 0); 766 if (sdp->sd_args.ar_spectator) { 767 sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0); 768 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks); 769 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5); 770 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5); 771 } else { 772 if (sdp->sd_lockstruct.ls_jid >= gfs2_jindex_size(sdp)) { 773 fs_err(sdp, "can't mount journal #%u\n", 774 sdp->sd_lockstruct.ls_jid); 775 fs_err(sdp, "there are only %u journals (0 - %u)\n", 776 gfs2_jindex_size(sdp), 777 gfs2_jindex_size(sdp) - 1); 778 goto fail_jindex; 779 } 780 sdp->sd_jdesc = gfs2_jdesc_find(sdp, sdp->sd_lockstruct.ls_jid); 781 782 error = gfs2_glock_nq_num(sdp, sdp->sd_lockstruct.ls_jid, 783 &gfs2_journal_glops, 784 LM_ST_EXCLUSIVE, 785 LM_FLAG_NOEXP | GL_NOCACHE | GL_NOPID, 786 &sdp->sd_journal_gh); 787 if (error) { 788 fs_err(sdp, "can't acquire journal glock: %d\n", error); 789 goto fail_jindex; 790 } 791 792 ip = GFS2_I(sdp->sd_jdesc->jd_inode); 793 sdp->sd_jinode_gl = ip->i_gl; 794 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 795 LM_FLAG_NOEXP | GL_EXACT | 796 GL_NOCACHE | GL_NOPID, 797 &sdp->sd_jinode_gh); 798 if (error) { 799 fs_err(sdp, "can't acquire journal inode glock: %d\n", 800 error); 801 goto fail_journal_gh; 802 } 803 804 error = gfs2_jdesc_check(sdp->sd_jdesc); 805 if (error) { 806 fs_err(sdp, "my journal (%u) is bad: %d\n", 807 sdp->sd_jdesc->jd_jid, error); 808 goto fail_jinode_gh; 809 } 810 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks); 811 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5); 812 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5); 813 814 /* Map the extents for this journal's blocks */ 815 gfs2_map_journal_extents(sdp, sdp->sd_jdesc); 816 } 817 trace_gfs2_log_blocks(sdp, atomic_read(&sdp->sd_log_blks_free)); 818 819 /* Lookup statfs inodes here so journal recovery can use them. */ 820 error = init_statfs(sdp); 821 if (error) 822 goto fail_jinode_gh; 823 824 if (sdp->sd_lockstruct.ls_first) { 825 unsigned int x; 826 for (x = 0; x < sdp->sd_journals; x++) { 827 struct gfs2_jdesc *jd = gfs2_jdesc_find(sdp, x); 828 829 if (sdp->sd_args.ar_spectator) { 830 error = check_journal_clean(sdp, jd, true); 831 if (error) 832 goto fail_statfs; 833 continue; 834 } 835 error = gfs2_recover_journal(jd, true); 836 if (error) { 837 fs_err(sdp, "error recovering journal %u: %d\n", 838 x, error); 839 goto fail_statfs; 840 } 841 } 842 843 gfs2_others_may_mount(sdp); 844 } else if (!sdp->sd_args.ar_spectator) { 845 error = gfs2_recover_journal(sdp->sd_jdesc, true); 846 if (error) { 847 fs_err(sdp, "error recovering my journal: %d\n", error); 848 goto fail_statfs; 849 } 850 } 851 852 sdp->sd_log_idle = 1; 853 set_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags); 854 gfs2_glock_dq_uninit(&ji_gh); 855 jindex = 0; 856 INIT_WORK(&sdp->sd_freeze_work, gfs2_freeze_func); 857 return 0; 858 859 fail_statfs: 860 uninit_statfs(sdp); 861 fail_jinode_gh: 862 /* A withdraw may have done dq/uninit so now we need to check it */ 863 if (!sdp->sd_args.ar_spectator && 864 gfs2_holder_initialized(&sdp->sd_jinode_gh)) 865 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh); 866 fail_journal_gh: 867 if (!sdp->sd_args.ar_spectator && 868 gfs2_holder_initialized(&sdp->sd_journal_gh)) 869 gfs2_glock_dq_uninit(&sdp->sd_journal_gh); 870 fail_jindex: 871 gfs2_jindex_free(sdp); 872 if (jindex) 873 gfs2_glock_dq_uninit(&ji_gh); 874 fail: 875 iput(sdp->sd_jindex); 876 return error; 877 } 878 879 static struct lock_class_key gfs2_quota_imutex_key; 880 881 static int init_inodes(struct gfs2_sbd *sdp, int undo) 882 { 883 int error = 0; 884 struct inode *master = d_inode(sdp->sd_master_dir); 885 886 if (undo) 887 goto fail_qinode; 888 889 error = init_journal(sdp, undo); 890 complete_all(&sdp->sd_journal_ready); 891 if (error) 892 goto fail; 893 894 /* Read in the resource index inode */ 895 sdp->sd_rindex = gfs2_lookup_simple(master, "rindex"); 896 if (IS_ERR(sdp->sd_rindex)) { 897 error = PTR_ERR(sdp->sd_rindex); 898 fs_err(sdp, "can't get resource index inode: %d\n", error); 899 goto fail_journal; 900 } 901 sdp->sd_rindex_uptodate = 0; 902 903 /* Read in the quota inode */ 904 sdp->sd_quota_inode = gfs2_lookup_simple(master, "quota"); 905 if (IS_ERR(sdp->sd_quota_inode)) { 906 error = PTR_ERR(sdp->sd_quota_inode); 907 fs_err(sdp, "can't get quota file inode: %d\n", error); 908 goto fail_rindex; 909 } 910 /* 911 * i_rwsem on quota files is special. Since this inode is hidden system 912 * file, we are safe to define locking ourselves. 913 */ 914 lockdep_set_class(&sdp->sd_quota_inode->i_rwsem, 915 &gfs2_quota_imutex_key); 916 917 error = gfs2_rindex_update(sdp); 918 if (error) 919 goto fail_qinode; 920 921 return 0; 922 923 fail_qinode: 924 iput(sdp->sd_quota_inode); 925 fail_rindex: 926 gfs2_clear_rgrpd(sdp); 927 iput(sdp->sd_rindex); 928 fail_journal: 929 init_journal(sdp, UNDO); 930 fail: 931 return error; 932 } 933 934 static int init_per_node(struct gfs2_sbd *sdp, int undo) 935 { 936 struct inode *pn = NULL; 937 char buf[30]; 938 int error = 0; 939 struct gfs2_inode *ip; 940 struct inode *master = d_inode(sdp->sd_master_dir); 941 942 if (sdp->sd_args.ar_spectator) 943 return 0; 944 945 if (undo) 946 goto fail_qc_gh; 947 948 pn = gfs2_lookup_simple(master, "per_node"); 949 if (IS_ERR(pn)) { 950 error = PTR_ERR(pn); 951 fs_err(sdp, "can't find per_node directory: %d\n", error); 952 return error; 953 } 954 955 sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid); 956 sdp->sd_qc_inode = gfs2_lookup_simple(pn, buf); 957 if (IS_ERR(sdp->sd_qc_inode)) { 958 error = PTR_ERR(sdp->sd_qc_inode); 959 fs_err(sdp, "can't find local \"qc\" file: %d\n", error); 960 goto fail_ut_i; 961 } 962 963 iput(pn); 964 pn = NULL; 965 966 ip = GFS2_I(sdp->sd_qc_inode); 967 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_NOPID, 968 &sdp->sd_qc_gh); 969 if (error) { 970 fs_err(sdp, "can't lock local \"qc\" file: %d\n", error); 971 goto fail_qc_i; 972 } 973 974 return 0; 975 976 fail_qc_gh: 977 gfs2_glock_dq_uninit(&sdp->sd_qc_gh); 978 fail_qc_i: 979 iput(sdp->sd_qc_inode); 980 fail_ut_i: 981 iput(pn); 982 return error; 983 } 984 985 static const match_table_t nolock_tokens = { 986 { Opt_jid, "jid=%d", }, 987 { Opt_err, NULL }, 988 }; 989 990 static const struct lm_lockops nolock_ops = { 991 .lm_proto_name = "lock_nolock", 992 .lm_put_lock = gfs2_glock_free, 993 .lm_tokens = &nolock_tokens, 994 }; 995 996 /** 997 * gfs2_lm_mount - mount a locking protocol 998 * @sdp: the filesystem 999 * @silent: if 1, don't complain if the FS isn't a GFS2 fs 1000 * 1001 * Returns: errno 1002 */ 1003 1004 static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent) 1005 { 1006 const struct lm_lockops *lm; 1007 struct lm_lockstruct *ls = &sdp->sd_lockstruct; 1008 struct gfs2_args *args = &sdp->sd_args; 1009 const char *proto = sdp->sd_proto_name; 1010 const char *table = sdp->sd_table_name; 1011 char *o, *options; 1012 int ret; 1013 1014 if (!strcmp("lock_nolock", proto)) { 1015 lm = &nolock_ops; 1016 sdp->sd_args.ar_localflocks = 1; 1017 #ifdef CONFIG_GFS2_FS_LOCKING_DLM 1018 } else if (!strcmp("lock_dlm", proto)) { 1019 lm = &gfs2_dlm_ops; 1020 #endif 1021 } else { 1022 pr_info("can't find protocol %s\n", proto); 1023 return -ENOENT; 1024 } 1025 1026 fs_info(sdp, "Trying to join cluster \"%s\", \"%s\"\n", proto, table); 1027 1028 ls->ls_ops = lm; 1029 ls->ls_first = 1; 1030 1031 for (options = args->ar_hostdata; (o = strsep(&options, ":")); ) { 1032 substring_t tmp[MAX_OPT_ARGS]; 1033 int token, option; 1034 1035 if (!o || !*o) 1036 continue; 1037 1038 token = match_token(o, *lm->lm_tokens, tmp); 1039 switch (token) { 1040 case Opt_jid: 1041 ret = match_int(&tmp[0], &option); 1042 if (ret || option < 0) 1043 goto hostdata_error; 1044 if (test_and_clear_bit(SDF_NOJOURNALID, &sdp->sd_flags)) 1045 ls->ls_jid = option; 1046 break; 1047 case Opt_id: 1048 case Opt_nodir: 1049 /* Obsolete, but left for backward compat purposes */ 1050 break; 1051 case Opt_first: 1052 ret = match_int(&tmp[0], &option); 1053 if (ret || (option != 0 && option != 1)) 1054 goto hostdata_error; 1055 ls->ls_first = option; 1056 break; 1057 case Opt_err: 1058 default: 1059 hostdata_error: 1060 fs_info(sdp, "unknown hostdata (%s)\n", o); 1061 return -EINVAL; 1062 } 1063 } 1064 1065 if (lm->lm_mount == NULL) { 1066 fs_info(sdp, "Now mounting FS (format %u)...\n", sdp->sd_sb.sb_fs_format); 1067 complete_all(&sdp->sd_locking_init); 1068 return 0; 1069 } 1070 ret = lm->lm_mount(sdp, table); 1071 if (ret == 0) 1072 fs_info(sdp, "Joined cluster. Now mounting FS (format %u)...\n", 1073 sdp->sd_sb.sb_fs_format); 1074 complete_all(&sdp->sd_locking_init); 1075 return ret; 1076 } 1077 1078 void gfs2_lm_unmount(struct gfs2_sbd *sdp) 1079 { 1080 const struct lm_lockops *lm = sdp->sd_lockstruct.ls_ops; 1081 if (likely(!gfs2_withdrawn(sdp)) && lm->lm_unmount) 1082 lm->lm_unmount(sdp); 1083 } 1084 1085 static int wait_on_journal(struct gfs2_sbd *sdp) 1086 { 1087 if (sdp->sd_lockstruct.ls_ops->lm_mount == NULL) 1088 return 0; 1089 1090 return wait_on_bit(&sdp->sd_flags, SDF_NOJOURNALID, TASK_INTERRUPTIBLE) 1091 ? -EINTR : 0; 1092 } 1093 1094 void gfs2_online_uevent(struct gfs2_sbd *sdp) 1095 { 1096 struct super_block *sb = sdp->sd_vfs; 1097 char ro[20]; 1098 char spectator[20]; 1099 char *envp[] = { ro, spectator, NULL }; 1100 sprintf(ro, "RDONLY=%d", sb_rdonly(sb)); 1101 sprintf(spectator, "SPECTATOR=%d", sdp->sd_args.ar_spectator ? 1 : 0); 1102 kobject_uevent_env(&sdp->sd_kobj, KOBJ_ONLINE, envp); 1103 } 1104 1105 static int init_threads(struct gfs2_sbd *sdp) 1106 { 1107 struct task_struct *p; 1108 int error = 0; 1109 1110 p = kthread_run(gfs2_logd, sdp, "gfs2_logd"); 1111 if (IS_ERR(p)) { 1112 error = PTR_ERR(p); 1113 fs_err(sdp, "can't start logd thread: %d\n", error); 1114 return error; 1115 } 1116 sdp->sd_logd_process = p; 1117 1118 p = kthread_run(gfs2_quotad, sdp, "gfs2_quotad"); 1119 if (IS_ERR(p)) { 1120 error = PTR_ERR(p); 1121 fs_err(sdp, "can't start quotad thread: %d\n", error); 1122 goto fail; 1123 } 1124 sdp->sd_quotad_process = p; 1125 return 0; 1126 1127 fail: 1128 kthread_stop(sdp->sd_logd_process); 1129 sdp->sd_logd_process = NULL; 1130 return error; 1131 } 1132 1133 /** 1134 * gfs2_fill_super - Read in superblock 1135 * @sb: The VFS superblock 1136 * @fc: Mount options and flags 1137 * 1138 * Returns: -errno 1139 */ 1140 static int gfs2_fill_super(struct super_block *sb, struct fs_context *fc) 1141 { 1142 struct gfs2_args *args = fc->fs_private; 1143 int silent = fc->sb_flags & SB_SILENT; 1144 struct gfs2_sbd *sdp; 1145 struct gfs2_holder mount_gh; 1146 struct gfs2_holder freeze_gh; 1147 int error; 1148 1149 sdp = init_sbd(sb); 1150 if (!sdp) { 1151 pr_warn("can't alloc struct gfs2_sbd\n"); 1152 return -ENOMEM; 1153 } 1154 sdp->sd_args = *args; 1155 1156 if (sdp->sd_args.ar_spectator) { 1157 sb->s_flags |= SB_RDONLY; 1158 set_bit(SDF_RORECOVERY, &sdp->sd_flags); 1159 } 1160 if (sdp->sd_args.ar_posix_acl) 1161 sb->s_flags |= SB_POSIXACL; 1162 if (sdp->sd_args.ar_nobarrier) 1163 set_bit(SDF_NOBARRIERS, &sdp->sd_flags); 1164 1165 sb->s_flags |= SB_NOSEC; 1166 sb->s_magic = GFS2_MAGIC; 1167 sb->s_op = &gfs2_super_ops; 1168 sb->s_d_op = &gfs2_dops; 1169 sb->s_export_op = &gfs2_export_ops; 1170 sb->s_qcop = &gfs2_quotactl_ops; 1171 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP; 1172 sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE; 1173 sb->s_time_gran = 1; 1174 sb->s_maxbytes = MAX_LFS_FILESIZE; 1175 1176 /* Set up the buffer cache and fill in some fake block size values 1177 to allow us to read-in the on-disk superblock. */ 1178 sdp->sd_sb.sb_bsize = sb_min_blocksize(sb, GFS2_BASIC_BLOCK); 1179 sdp->sd_sb.sb_bsize_shift = sb->s_blocksize_bits; 1180 sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift - 1181 GFS2_BASIC_BLOCK_SHIFT; 1182 sdp->sd_fsb2bb = BIT(sdp->sd_fsb2bb_shift); 1183 1184 sdp->sd_tune.gt_logd_secs = sdp->sd_args.ar_commit; 1185 sdp->sd_tune.gt_quota_quantum = sdp->sd_args.ar_quota_quantum; 1186 if (sdp->sd_args.ar_statfs_quantum) { 1187 sdp->sd_tune.gt_statfs_slow = 0; 1188 sdp->sd_tune.gt_statfs_quantum = sdp->sd_args.ar_statfs_quantum; 1189 } else { 1190 sdp->sd_tune.gt_statfs_slow = 1; 1191 sdp->sd_tune.gt_statfs_quantum = 30; 1192 } 1193 1194 error = init_names(sdp, silent); 1195 if (error) 1196 goto fail_free; 1197 1198 snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s", sdp->sd_table_name); 1199 1200 sdp->sd_delete_wq = alloc_workqueue("gfs2-delete/%s", 1201 WQ_MEM_RECLAIM | WQ_FREEZABLE, 0, sdp->sd_fsname); 1202 error = -ENOMEM; 1203 if (!sdp->sd_delete_wq) 1204 goto fail_free; 1205 1206 error = gfs2_sys_fs_add(sdp); 1207 if (error) 1208 goto fail_delete_wq; 1209 1210 gfs2_create_debugfs_file(sdp); 1211 1212 error = gfs2_lm_mount(sdp, silent); 1213 if (error) 1214 goto fail_debug; 1215 1216 error = init_locking(sdp, &mount_gh, DO); 1217 if (error) 1218 goto fail_lm; 1219 1220 error = init_sb(sdp, silent); 1221 if (error) 1222 goto fail_locking; 1223 1224 /* Turn rgrplvb on by default if fs format is recent enough */ 1225 if (!sdp->sd_args.ar_got_rgrplvb && sdp->sd_sb.sb_fs_format > 1801) 1226 sdp->sd_args.ar_rgrplvb = 1; 1227 1228 error = wait_on_journal(sdp); 1229 if (error) 1230 goto fail_sb; 1231 1232 /* 1233 * If user space has failed to join the cluster or some similar 1234 * failure has occurred, then the journal id will contain a 1235 * negative (error) number. This will then be returned to the 1236 * caller (of the mount syscall). We do this even for spectator 1237 * mounts (which just write a jid of 0 to indicate "ok" even though 1238 * the jid is unused in the spectator case) 1239 */ 1240 if (sdp->sd_lockstruct.ls_jid < 0) { 1241 error = sdp->sd_lockstruct.ls_jid; 1242 sdp->sd_lockstruct.ls_jid = 0; 1243 goto fail_sb; 1244 } 1245 1246 if (sdp->sd_args.ar_spectator) 1247 snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s.s", 1248 sdp->sd_table_name); 1249 else 1250 snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s.%u", 1251 sdp->sd_table_name, sdp->sd_lockstruct.ls_jid); 1252 1253 error = init_inodes(sdp, DO); 1254 if (error) 1255 goto fail_sb; 1256 1257 error = init_per_node(sdp, DO); 1258 if (error) 1259 goto fail_inodes; 1260 1261 error = gfs2_statfs_init(sdp); 1262 if (error) { 1263 fs_err(sdp, "can't initialize statfs subsystem: %d\n", error); 1264 goto fail_per_node; 1265 } 1266 1267 if (!sb_rdonly(sb)) { 1268 error = init_threads(sdp); 1269 if (error) { 1270 gfs2_withdraw_delayed(sdp); 1271 goto fail_per_node; 1272 } 1273 } 1274 1275 error = gfs2_freeze_lock(sdp, &freeze_gh, 0); 1276 if (error) 1277 goto fail_per_node; 1278 1279 if (!sb_rdonly(sb)) 1280 error = gfs2_make_fs_rw(sdp); 1281 1282 gfs2_freeze_unlock(&freeze_gh); 1283 if (error) { 1284 if (sdp->sd_quotad_process) 1285 kthread_stop(sdp->sd_quotad_process); 1286 sdp->sd_quotad_process = NULL; 1287 if (sdp->sd_logd_process) 1288 kthread_stop(sdp->sd_logd_process); 1289 sdp->sd_logd_process = NULL; 1290 fs_err(sdp, "can't make FS RW: %d\n", error); 1291 goto fail_per_node; 1292 } 1293 gfs2_glock_dq_uninit(&mount_gh); 1294 gfs2_online_uevent(sdp); 1295 return 0; 1296 1297 fail_per_node: 1298 init_per_node(sdp, UNDO); 1299 fail_inodes: 1300 init_inodes(sdp, UNDO); 1301 fail_sb: 1302 if (sdp->sd_root_dir) 1303 dput(sdp->sd_root_dir); 1304 if (sdp->sd_master_dir) 1305 dput(sdp->sd_master_dir); 1306 if (sb->s_root) 1307 dput(sb->s_root); 1308 sb->s_root = NULL; 1309 fail_locking: 1310 init_locking(sdp, &mount_gh, UNDO); 1311 fail_lm: 1312 complete_all(&sdp->sd_journal_ready); 1313 gfs2_gl_hash_clear(sdp); 1314 gfs2_lm_unmount(sdp); 1315 fail_debug: 1316 gfs2_delete_debugfs_file(sdp); 1317 gfs2_sys_fs_del(sdp); 1318 fail_delete_wq: 1319 destroy_workqueue(sdp->sd_delete_wq); 1320 fail_free: 1321 free_sbd(sdp); 1322 sb->s_fs_info = NULL; 1323 return error; 1324 } 1325 1326 /** 1327 * gfs2_get_tree - Get the GFS2 superblock and root directory 1328 * @fc: The filesystem context 1329 * 1330 * Returns: 0 or -errno on error 1331 */ 1332 static int gfs2_get_tree(struct fs_context *fc) 1333 { 1334 struct gfs2_args *args = fc->fs_private; 1335 struct gfs2_sbd *sdp; 1336 int error; 1337 1338 error = get_tree_bdev(fc, gfs2_fill_super); 1339 if (error) 1340 return error; 1341 1342 sdp = fc->root->d_sb->s_fs_info; 1343 dput(fc->root); 1344 if (args->ar_meta) 1345 fc->root = dget(sdp->sd_master_dir); 1346 else 1347 fc->root = dget(sdp->sd_root_dir); 1348 return 0; 1349 } 1350 1351 static void gfs2_fc_free(struct fs_context *fc) 1352 { 1353 struct gfs2_args *args = fc->fs_private; 1354 1355 kfree(args); 1356 } 1357 1358 enum gfs2_param { 1359 Opt_lockproto, 1360 Opt_locktable, 1361 Opt_hostdata, 1362 Opt_spectator, 1363 Opt_ignore_local_fs, 1364 Opt_localflocks, 1365 Opt_localcaching, 1366 Opt_debug, 1367 Opt_upgrade, 1368 Opt_acl, 1369 Opt_quota, 1370 Opt_quota_flag, 1371 Opt_suiddir, 1372 Opt_data, 1373 Opt_meta, 1374 Opt_discard, 1375 Opt_commit, 1376 Opt_errors, 1377 Opt_statfs_quantum, 1378 Opt_statfs_percent, 1379 Opt_quota_quantum, 1380 Opt_barrier, 1381 Opt_rgrplvb, 1382 Opt_loccookie, 1383 }; 1384 1385 static const struct constant_table gfs2_param_quota[] = { 1386 {"off", GFS2_QUOTA_OFF}, 1387 {"account", GFS2_QUOTA_ACCOUNT}, 1388 {"on", GFS2_QUOTA_ON}, 1389 {} 1390 }; 1391 1392 enum opt_data { 1393 Opt_data_writeback = GFS2_DATA_WRITEBACK, 1394 Opt_data_ordered = GFS2_DATA_ORDERED, 1395 }; 1396 1397 static const struct constant_table gfs2_param_data[] = { 1398 {"writeback", Opt_data_writeback }, 1399 {"ordered", Opt_data_ordered }, 1400 {} 1401 }; 1402 1403 enum opt_errors { 1404 Opt_errors_withdraw = GFS2_ERRORS_WITHDRAW, 1405 Opt_errors_panic = GFS2_ERRORS_PANIC, 1406 }; 1407 1408 static const struct constant_table gfs2_param_errors[] = { 1409 {"withdraw", Opt_errors_withdraw }, 1410 {"panic", Opt_errors_panic }, 1411 {} 1412 }; 1413 1414 static const struct fs_parameter_spec gfs2_fs_parameters[] = { 1415 fsparam_string ("lockproto", Opt_lockproto), 1416 fsparam_string ("locktable", Opt_locktable), 1417 fsparam_string ("hostdata", Opt_hostdata), 1418 fsparam_flag ("spectator", Opt_spectator), 1419 fsparam_flag ("norecovery", Opt_spectator), 1420 fsparam_flag ("ignore_local_fs", Opt_ignore_local_fs), 1421 fsparam_flag ("localflocks", Opt_localflocks), 1422 fsparam_flag ("localcaching", Opt_localcaching), 1423 fsparam_flag_no("debug", Opt_debug), 1424 fsparam_flag ("upgrade", Opt_upgrade), 1425 fsparam_flag_no("acl", Opt_acl), 1426 fsparam_flag_no("suiddir", Opt_suiddir), 1427 fsparam_enum ("data", Opt_data, gfs2_param_data), 1428 fsparam_flag ("meta", Opt_meta), 1429 fsparam_flag_no("discard", Opt_discard), 1430 fsparam_s32 ("commit", Opt_commit), 1431 fsparam_enum ("errors", Opt_errors, gfs2_param_errors), 1432 fsparam_s32 ("statfs_quantum", Opt_statfs_quantum), 1433 fsparam_s32 ("statfs_percent", Opt_statfs_percent), 1434 fsparam_s32 ("quota_quantum", Opt_quota_quantum), 1435 fsparam_flag_no("barrier", Opt_barrier), 1436 fsparam_flag_no("rgrplvb", Opt_rgrplvb), 1437 fsparam_flag_no("loccookie", Opt_loccookie), 1438 /* quota can be a flag or an enum so it gets special treatment */ 1439 fsparam_flag_no("quota", Opt_quota_flag), 1440 fsparam_enum("quota", Opt_quota, gfs2_param_quota), 1441 {} 1442 }; 1443 1444 /* Parse a single mount parameter */ 1445 static int gfs2_parse_param(struct fs_context *fc, struct fs_parameter *param) 1446 { 1447 struct gfs2_args *args = fc->fs_private; 1448 struct fs_parse_result result; 1449 int o; 1450 1451 o = fs_parse(fc, gfs2_fs_parameters, param, &result); 1452 if (o < 0) 1453 return o; 1454 1455 switch (o) { 1456 case Opt_lockproto: 1457 strscpy(args->ar_lockproto, param->string, GFS2_LOCKNAME_LEN); 1458 break; 1459 case Opt_locktable: 1460 strscpy(args->ar_locktable, param->string, GFS2_LOCKNAME_LEN); 1461 break; 1462 case Opt_hostdata: 1463 strscpy(args->ar_hostdata, param->string, GFS2_LOCKNAME_LEN); 1464 break; 1465 case Opt_spectator: 1466 args->ar_spectator = 1; 1467 break; 1468 case Opt_ignore_local_fs: 1469 /* Retained for backwards compat only */ 1470 break; 1471 case Opt_localflocks: 1472 args->ar_localflocks = 1; 1473 break; 1474 case Opt_localcaching: 1475 /* Retained for backwards compat only */ 1476 break; 1477 case Opt_debug: 1478 if (result.boolean && args->ar_errors == GFS2_ERRORS_PANIC) 1479 return invalfc(fc, "-o debug and -o errors=panic are mutually exclusive"); 1480 args->ar_debug = result.boolean; 1481 break; 1482 case Opt_upgrade: 1483 /* Retained for backwards compat only */ 1484 break; 1485 case Opt_acl: 1486 args->ar_posix_acl = result.boolean; 1487 break; 1488 case Opt_quota_flag: 1489 args->ar_quota = result.negated ? GFS2_QUOTA_OFF : GFS2_QUOTA_ON; 1490 break; 1491 case Opt_quota: 1492 args->ar_quota = result.int_32; 1493 break; 1494 case Opt_suiddir: 1495 args->ar_suiddir = result.boolean; 1496 break; 1497 case Opt_data: 1498 /* The uint_32 result maps directly to GFS2_DATA_* */ 1499 args->ar_data = result.uint_32; 1500 break; 1501 case Opt_meta: 1502 args->ar_meta = 1; 1503 break; 1504 case Opt_discard: 1505 args->ar_discard = result.boolean; 1506 break; 1507 case Opt_commit: 1508 if (result.int_32 <= 0) 1509 return invalfc(fc, "commit mount option requires a positive numeric argument"); 1510 args->ar_commit = result.int_32; 1511 break; 1512 case Opt_statfs_quantum: 1513 if (result.int_32 < 0) 1514 return invalfc(fc, "statfs_quantum mount option requires a non-negative numeric argument"); 1515 args->ar_statfs_quantum = result.int_32; 1516 break; 1517 case Opt_quota_quantum: 1518 if (result.int_32 <= 0) 1519 return invalfc(fc, "quota_quantum mount option requires a positive numeric argument"); 1520 args->ar_quota_quantum = result.int_32; 1521 break; 1522 case Opt_statfs_percent: 1523 if (result.int_32 < 0 || result.int_32 > 100) 1524 return invalfc(fc, "statfs_percent mount option requires a numeric argument between 0 and 100"); 1525 args->ar_statfs_percent = result.int_32; 1526 break; 1527 case Opt_errors: 1528 if (args->ar_debug && result.uint_32 == GFS2_ERRORS_PANIC) 1529 return invalfc(fc, "-o debug and -o errors=panic are mutually exclusive"); 1530 args->ar_errors = result.uint_32; 1531 break; 1532 case Opt_barrier: 1533 args->ar_nobarrier = result.boolean; 1534 break; 1535 case Opt_rgrplvb: 1536 args->ar_rgrplvb = result.boolean; 1537 args->ar_got_rgrplvb = 1; 1538 break; 1539 case Opt_loccookie: 1540 args->ar_loccookie = result.boolean; 1541 break; 1542 default: 1543 return invalfc(fc, "invalid mount option: %s", param->key); 1544 } 1545 return 0; 1546 } 1547 1548 static int gfs2_reconfigure(struct fs_context *fc) 1549 { 1550 struct super_block *sb = fc->root->d_sb; 1551 struct gfs2_sbd *sdp = sb->s_fs_info; 1552 struct gfs2_args *oldargs = &sdp->sd_args; 1553 struct gfs2_args *newargs = fc->fs_private; 1554 struct gfs2_tune *gt = &sdp->sd_tune; 1555 int error = 0; 1556 1557 sync_filesystem(sb); 1558 1559 spin_lock(>->gt_spin); 1560 oldargs->ar_commit = gt->gt_logd_secs; 1561 oldargs->ar_quota_quantum = gt->gt_quota_quantum; 1562 if (gt->gt_statfs_slow) 1563 oldargs->ar_statfs_quantum = 0; 1564 else 1565 oldargs->ar_statfs_quantum = gt->gt_statfs_quantum; 1566 spin_unlock(>->gt_spin); 1567 1568 if (strcmp(newargs->ar_lockproto, oldargs->ar_lockproto)) { 1569 errorfc(fc, "reconfiguration of locking protocol not allowed"); 1570 return -EINVAL; 1571 } 1572 if (strcmp(newargs->ar_locktable, oldargs->ar_locktable)) { 1573 errorfc(fc, "reconfiguration of lock table not allowed"); 1574 return -EINVAL; 1575 } 1576 if (strcmp(newargs->ar_hostdata, oldargs->ar_hostdata)) { 1577 errorfc(fc, "reconfiguration of host data not allowed"); 1578 return -EINVAL; 1579 } 1580 if (newargs->ar_spectator != oldargs->ar_spectator) { 1581 errorfc(fc, "reconfiguration of spectator mode not allowed"); 1582 return -EINVAL; 1583 } 1584 if (newargs->ar_localflocks != oldargs->ar_localflocks) { 1585 errorfc(fc, "reconfiguration of localflocks not allowed"); 1586 return -EINVAL; 1587 } 1588 if (newargs->ar_meta != oldargs->ar_meta) { 1589 errorfc(fc, "switching between gfs2 and gfs2meta not allowed"); 1590 return -EINVAL; 1591 } 1592 if (oldargs->ar_spectator) 1593 fc->sb_flags |= SB_RDONLY; 1594 1595 if ((sb->s_flags ^ fc->sb_flags) & SB_RDONLY) { 1596 struct gfs2_holder freeze_gh; 1597 1598 error = gfs2_freeze_lock(sdp, &freeze_gh, 0); 1599 if (error) 1600 return -EINVAL; 1601 1602 if (fc->sb_flags & SB_RDONLY) { 1603 gfs2_make_fs_ro(sdp); 1604 } else { 1605 error = gfs2_make_fs_rw(sdp); 1606 if (error) 1607 errorfc(fc, "unable to remount read-write"); 1608 } 1609 gfs2_freeze_unlock(&freeze_gh); 1610 } 1611 sdp->sd_args = *newargs; 1612 1613 if (sdp->sd_args.ar_posix_acl) 1614 sb->s_flags |= SB_POSIXACL; 1615 else 1616 sb->s_flags &= ~SB_POSIXACL; 1617 if (sdp->sd_args.ar_nobarrier) 1618 set_bit(SDF_NOBARRIERS, &sdp->sd_flags); 1619 else 1620 clear_bit(SDF_NOBARRIERS, &sdp->sd_flags); 1621 spin_lock(>->gt_spin); 1622 gt->gt_logd_secs = newargs->ar_commit; 1623 gt->gt_quota_quantum = newargs->ar_quota_quantum; 1624 if (newargs->ar_statfs_quantum) { 1625 gt->gt_statfs_slow = 0; 1626 gt->gt_statfs_quantum = newargs->ar_statfs_quantum; 1627 } 1628 else { 1629 gt->gt_statfs_slow = 1; 1630 gt->gt_statfs_quantum = 30; 1631 } 1632 spin_unlock(>->gt_spin); 1633 1634 gfs2_online_uevent(sdp); 1635 return error; 1636 } 1637 1638 static const struct fs_context_operations gfs2_context_ops = { 1639 .free = gfs2_fc_free, 1640 .parse_param = gfs2_parse_param, 1641 .get_tree = gfs2_get_tree, 1642 .reconfigure = gfs2_reconfigure, 1643 }; 1644 1645 /* Set up the filesystem mount context */ 1646 static int gfs2_init_fs_context(struct fs_context *fc) 1647 { 1648 struct gfs2_args *args; 1649 1650 args = kmalloc(sizeof(*args), GFP_KERNEL); 1651 if (args == NULL) 1652 return -ENOMEM; 1653 1654 if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) { 1655 struct gfs2_sbd *sdp = fc->root->d_sb->s_fs_info; 1656 1657 *args = sdp->sd_args; 1658 } else { 1659 memset(args, 0, sizeof(*args)); 1660 args->ar_quota = GFS2_QUOTA_DEFAULT; 1661 args->ar_data = GFS2_DATA_DEFAULT; 1662 args->ar_commit = 30; 1663 args->ar_statfs_quantum = 30; 1664 args->ar_quota_quantum = 60; 1665 args->ar_errors = GFS2_ERRORS_DEFAULT; 1666 } 1667 fc->fs_private = args; 1668 fc->ops = &gfs2_context_ops; 1669 return 0; 1670 } 1671 1672 static int set_meta_super(struct super_block *s, struct fs_context *fc) 1673 { 1674 return -EINVAL; 1675 } 1676 1677 static int test_meta_super(struct super_block *s, struct fs_context *fc) 1678 { 1679 return (fc->sget_key == s->s_bdev); 1680 } 1681 1682 static int gfs2_meta_get_tree(struct fs_context *fc) 1683 { 1684 struct super_block *s; 1685 struct gfs2_sbd *sdp; 1686 struct path path; 1687 int error; 1688 1689 if (!fc->source || !*fc->source) 1690 return -EINVAL; 1691 1692 error = kern_path(fc->source, LOOKUP_FOLLOW, &path); 1693 if (error) { 1694 pr_warn("path_lookup on %s returned error %d\n", 1695 fc->source, error); 1696 return error; 1697 } 1698 fc->fs_type = &gfs2_fs_type; 1699 fc->sget_key = path.dentry->d_sb->s_bdev; 1700 s = sget_fc(fc, test_meta_super, set_meta_super); 1701 path_put(&path); 1702 if (IS_ERR(s)) { 1703 pr_warn("gfs2 mount does not exist\n"); 1704 return PTR_ERR(s); 1705 } 1706 if ((fc->sb_flags ^ s->s_flags) & SB_RDONLY) { 1707 deactivate_locked_super(s); 1708 return -EBUSY; 1709 } 1710 sdp = s->s_fs_info; 1711 fc->root = dget(sdp->sd_master_dir); 1712 return 0; 1713 } 1714 1715 static const struct fs_context_operations gfs2_meta_context_ops = { 1716 .free = gfs2_fc_free, 1717 .get_tree = gfs2_meta_get_tree, 1718 }; 1719 1720 static int gfs2_meta_init_fs_context(struct fs_context *fc) 1721 { 1722 int ret = gfs2_init_fs_context(fc); 1723 1724 if (ret) 1725 return ret; 1726 1727 fc->ops = &gfs2_meta_context_ops; 1728 return 0; 1729 } 1730 1731 /** 1732 * gfs2_evict_inodes - evict inodes cooperatively 1733 * @sb: the superblock 1734 * 1735 * When evicting an inode with a zero link count, we are trying to upgrade the 1736 * inode's iopen glock from SH to EX mode in order to determine if we can 1737 * delete the inode. The other nodes are supposed to evict the inode from 1738 * their caches if they can, and to poke the inode's inode glock if they cannot 1739 * do so. Either behavior allows gfs2_upgrade_iopen_glock() to proceed 1740 * quickly, but if the other nodes are not cooperating, the lock upgrading 1741 * attempt will time out. Since inodes are evicted sequentially, this can add 1742 * up quickly. 1743 * 1744 * Function evict_inodes() tries to keep the s_inode_list_lock list locked over 1745 * a long time, which prevents other inodes from being evicted concurrently. 1746 * This precludes the cooperative behavior we are looking for. This special 1747 * version of evict_inodes() avoids that. 1748 * 1749 * Modeled after drop_pagecache_sb(). 1750 */ 1751 static void gfs2_evict_inodes(struct super_block *sb) 1752 { 1753 struct inode *inode, *toput_inode = NULL; 1754 struct gfs2_sbd *sdp = sb->s_fs_info; 1755 1756 set_bit(SDF_EVICTING, &sdp->sd_flags); 1757 1758 spin_lock(&sb->s_inode_list_lock); 1759 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { 1760 spin_lock(&inode->i_lock); 1761 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) && 1762 !need_resched()) { 1763 spin_unlock(&inode->i_lock); 1764 continue; 1765 } 1766 atomic_inc(&inode->i_count); 1767 spin_unlock(&inode->i_lock); 1768 spin_unlock(&sb->s_inode_list_lock); 1769 1770 iput(toput_inode); 1771 toput_inode = inode; 1772 1773 cond_resched(); 1774 spin_lock(&sb->s_inode_list_lock); 1775 } 1776 spin_unlock(&sb->s_inode_list_lock); 1777 iput(toput_inode); 1778 } 1779 1780 static void gfs2_kill_sb(struct super_block *sb) 1781 { 1782 struct gfs2_sbd *sdp = sb->s_fs_info; 1783 1784 if (sdp == NULL) { 1785 kill_block_super(sb); 1786 return; 1787 } 1788 1789 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SYNC | GFS2_LFC_KILL_SB); 1790 dput(sdp->sd_root_dir); 1791 dput(sdp->sd_master_dir); 1792 sdp->sd_root_dir = NULL; 1793 sdp->sd_master_dir = NULL; 1794 shrink_dcache_sb(sb); 1795 1796 gfs2_evict_inodes(sb); 1797 1798 /* 1799 * Flush and then drain the delete workqueue here (via 1800 * destroy_workqueue()) to ensure that any delete work that 1801 * may be running will also see the SDF_DEACTIVATING flag. 1802 */ 1803 set_bit(SDF_DEACTIVATING, &sdp->sd_flags); 1804 gfs2_flush_delete_work(sdp); 1805 destroy_workqueue(sdp->sd_delete_wq); 1806 1807 kill_block_super(sb); 1808 } 1809 1810 struct file_system_type gfs2_fs_type = { 1811 .name = "gfs2", 1812 .fs_flags = FS_REQUIRES_DEV, 1813 .init_fs_context = gfs2_init_fs_context, 1814 .parameters = gfs2_fs_parameters, 1815 .kill_sb = gfs2_kill_sb, 1816 .owner = THIS_MODULE, 1817 }; 1818 MODULE_ALIAS_FS("gfs2"); 1819 1820 struct file_system_type gfs2meta_fs_type = { 1821 .name = "gfs2meta", 1822 .fs_flags = FS_REQUIRES_DEV, 1823 .init_fs_context = gfs2_meta_init_fs_context, 1824 .owner = THIS_MODULE, 1825 }; 1826 MODULE_ALIAS_FS("gfs2meta"); 1827