1 /* 2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. 3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved. 4 * 5 * This copyrighted material is made available to anyone wishing to use, 6 * modify, copy, or redistribute it subject to the terms and conditions 7 * of the GNU General Public License version 2. 8 */ 9 10 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 11 12 #include <linux/sched.h> 13 #include <linux/slab.h> 14 #include <linux/spinlock.h> 15 #include <linux/completion.h> 16 #include <linux/buffer_head.h> 17 #include <linux/blkdev.h> 18 #include <linux/kthread.h> 19 #include <linux/export.h> 20 #include <linux/namei.h> 21 #include <linux/mount.h> 22 #include <linux/gfs2_ondisk.h> 23 #include <linux/quotaops.h> 24 #include <linux/lockdep.h> 25 #include <linux/module.h> 26 27 #include "gfs2.h" 28 #include "incore.h" 29 #include "bmap.h" 30 #include "glock.h" 31 #include "glops.h" 32 #include "inode.h" 33 #include "recovery.h" 34 #include "rgrp.h" 35 #include "super.h" 36 #include "sys.h" 37 #include "util.h" 38 #include "log.h" 39 #include "quota.h" 40 #include "dir.h" 41 #include "meta_io.h" 42 #include "trace_gfs2.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 = 1 << 18; 62 gt->gt_complain_secs = 10; 63 } 64 65 static struct gfs2_sbd *init_sbd(struct super_block *sb) 66 { 67 struct gfs2_sbd *sdp; 68 struct address_space *mapping; 69 70 sdp = kzalloc(sizeof(struct gfs2_sbd), GFP_KERNEL); 71 if (!sdp) 72 return NULL; 73 74 sb->s_fs_info = sdp; 75 sdp->sd_vfs = sb; 76 sdp->sd_lkstats = alloc_percpu(struct gfs2_pcpu_lkstats); 77 if (!sdp->sd_lkstats) { 78 kfree(sdp); 79 return NULL; 80 } 81 82 set_bit(SDF_NOJOURNALID, &sdp->sd_flags); 83 gfs2_tune_init(&sdp->sd_tune); 84 85 init_waitqueue_head(&sdp->sd_glock_wait); 86 atomic_set(&sdp->sd_glock_disposal, 0); 87 init_completion(&sdp->sd_locking_init); 88 init_completion(&sdp->sd_wdack); 89 spin_lock_init(&sdp->sd_statfs_spin); 90 91 spin_lock_init(&sdp->sd_rindex_spin); 92 sdp->sd_rindex_tree.rb_node = NULL; 93 94 INIT_LIST_HEAD(&sdp->sd_jindex_list); 95 spin_lock_init(&sdp->sd_jindex_spin); 96 mutex_init(&sdp->sd_jindex_mutex); 97 init_completion(&sdp->sd_journal_ready); 98 99 INIT_LIST_HEAD(&sdp->sd_quota_list); 100 mutex_init(&sdp->sd_quota_mutex); 101 mutex_init(&sdp->sd_quota_sync_mutex); 102 init_waitqueue_head(&sdp->sd_quota_wait); 103 INIT_LIST_HEAD(&sdp->sd_trunc_list); 104 spin_lock_init(&sdp->sd_trunc_lock); 105 spin_lock_init(&sdp->sd_bitmap_lock); 106 107 mapping = &sdp->sd_aspace; 108 109 address_space_init_once(mapping); 110 mapping->a_ops = &gfs2_rgrp_aops; 111 mapping->host = sb->s_bdev->bd_inode; 112 mapping->flags = 0; 113 mapping_set_gfp_mask(mapping, GFP_NOFS); 114 mapping->private_data = NULL; 115 mapping->backing_dev_info = sb->s_bdi; 116 mapping->writeback_index = 0; 117 118 spin_lock_init(&sdp->sd_log_lock); 119 atomic_set(&sdp->sd_log_pinned, 0); 120 INIT_LIST_HEAD(&sdp->sd_log_le_revoke); 121 INIT_LIST_HEAD(&sdp->sd_log_le_ordered); 122 spin_lock_init(&sdp->sd_ordered_lock); 123 124 init_waitqueue_head(&sdp->sd_log_waitq); 125 init_waitqueue_head(&sdp->sd_logd_waitq); 126 spin_lock_init(&sdp->sd_ail_lock); 127 INIT_LIST_HEAD(&sdp->sd_ail1_list); 128 INIT_LIST_HEAD(&sdp->sd_ail2_list); 129 130 init_rwsem(&sdp->sd_log_flush_lock); 131 atomic_set(&sdp->sd_log_in_flight, 0); 132 atomic_set(&sdp->sd_reserving_log, 0); 133 init_waitqueue_head(&sdp->sd_reserving_log_wait); 134 init_waitqueue_head(&sdp->sd_log_flush_wait); 135 atomic_set(&sdp->sd_freeze_state, SFS_UNFROZEN); 136 mutex_init(&sdp->sd_freeze_mutex); 137 138 return sdp; 139 } 140 141 142 /** 143 * gfs2_check_sb - Check superblock 144 * @sdp: the filesystem 145 * @sb: The superblock 146 * @silent: Don't print a message if the check fails 147 * 148 * Checks the version code of the FS is one that we understand how to 149 * read and that the sizes of the various on-disk structures have not 150 * changed. 151 */ 152 153 static int gfs2_check_sb(struct gfs2_sbd *sdp, int silent) 154 { 155 struct gfs2_sb_host *sb = &sdp->sd_sb; 156 157 if (sb->sb_magic != GFS2_MAGIC || 158 sb->sb_type != GFS2_METATYPE_SB) { 159 if (!silent) 160 pr_warn("not a GFS2 filesystem\n"); 161 return -EINVAL; 162 } 163 164 /* If format numbers match exactly, we're done. */ 165 166 if (sb->sb_fs_format == GFS2_FORMAT_FS && 167 sb->sb_multihost_format == GFS2_FORMAT_MULTI) 168 return 0; 169 170 fs_warn(sdp, "Unknown on-disk format, unable to mount\n"); 171 172 return -EINVAL; 173 } 174 175 static void end_bio_io_page(struct bio *bio, int error) 176 { 177 struct page *page = bio->bi_private; 178 179 if (!error) 180 SetPageUptodate(page); 181 else 182 pr_warn("error %d reading superblock\n", error); 183 unlock_page(page); 184 } 185 186 static void gfs2_sb_in(struct gfs2_sbd *sdp, const void *buf) 187 { 188 struct gfs2_sb_host *sb = &sdp->sd_sb; 189 struct super_block *s = sdp->sd_vfs; 190 const struct gfs2_sb *str = buf; 191 192 sb->sb_magic = be32_to_cpu(str->sb_header.mh_magic); 193 sb->sb_type = be32_to_cpu(str->sb_header.mh_type); 194 sb->sb_format = be32_to_cpu(str->sb_header.mh_format); 195 sb->sb_fs_format = be32_to_cpu(str->sb_fs_format); 196 sb->sb_multihost_format = be32_to_cpu(str->sb_multihost_format); 197 sb->sb_bsize = be32_to_cpu(str->sb_bsize); 198 sb->sb_bsize_shift = be32_to_cpu(str->sb_bsize_shift); 199 sb->sb_master_dir.no_addr = be64_to_cpu(str->sb_master_dir.no_addr); 200 sb->sb_master_dir.no_formal_ino = be64_to_cpu(str->sb_master_dir.no_formal_ino); 201 sb->sb_root_dir.no_addr = be64_to_cpu(str->sb_root_dir.no_addr); 202 sb->sb_root_dir.no_formal_ino = be64_to_cpu(str->sb_root_dir.no_formal_ino); 203 204 memcpy(sb->sb_lockproto, str->sb_lockproto, GFS2_LOCKNAME_LEN); 205 memcpy(sb->sb_locktable, str->sb_locktable, GFS2_LOCKNAME_LEN); 206 memcpy(s->s_uuid, str->sb_uuid, 16); 207 } 208 209 /** 210 * gfs2_read_super - Read the gfs2 super block from disk 211 * @sdp: The GFS2 super block 212 * @sector: The location of the super block 213 * @error: The error code to return 214 * 215 * This uses the bio functions to read the super block from disk 216 * because we want to be 100% sure that we never read cached data. 217 * A super block is read twice only during each GFS2 mount and is 218 * never written to by the filesystem. The first time its read no 219 * locks are held, and the only details which are looked at are those 220 * relating to the locking protocol. Once locking is up and working, 221 * the sb is read again under the lock to establish the location of 222 * the master directory (contains pointers to journals etc) and the 223 * root directory. 224 * 225 * Returns: 0 on success or error 226 */ 227 228 static int gfs2_read_super(struct gfs2_sbd *sdp, sector_t sector, int silent) 229 { 230 struct super_block *sb = sdp->sd_vfs; 231 struct gfs2_sb *p; 232 struct page *page; 233 struct bio *bio; 234 235 page = alloc_page(GFP_NOFS); 236 if (unlikely(!page)) 237 return -ENOMEM; 238 239 ClearPageUptodate(page); 240 ClearPageDirty(page); 241 lock_page(page); 242 243 bio = bio_alloc(GFP_NOFS, 1); 244 bio->bi_iter.bi_sector = sector * (sb->s_blocksize >> 9); 245 bio->bi_bdev = sb->s_bdev; 246 bio_add_page(bio, page, PAGE_SIZE, 0); 247 248 bio->bi_end_io = end_bio_io_page; 249 bio->bi_private = page; 250 submit_bio(READ_SYNC | REQ_META, bio); 251 wait_on_page_locked(page); 252 bio_put(bio); 253 if (!PageUptodate(page)) { 254 __free_page(page); 255 return -EIO; 256 } 257 p = kmap(page); 258 gfs2_sb_in(sdp, p); 259 kunmap(page); 260 __free_page(page); 261 return gfs2_check_sb(sdp, silent); 262 } 263 264 /** 265 * gfs2_read_sb - Read super block 266 * @sdp: The GFS2 superblock 267 * @silent: Don't print message if mount fails 268 * 269 */ 270 271 static int gfs2_read_sb(struct gfs2_sbd *sdp, int silent) 272 { 273 u32 hash_blocks, ind_blocks, leaf_blocks; 274 u32 tmp_blocks; 275 unsigned int x; 276 int error; 277 278 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent); 279 if (error) { 280 if (!silent) 281 fs_err(sdp, "can't read superblock\n"); 282 return error; 283 } 284 285 sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift - 286 GFS2_BASIC_BLOCK_SHIFT; 287 sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift; 288 sdp->sd_diptrs = (sdp->sd_sb.sb_bsize - 289 sizeof(struct gfs2_dinode)) / sizeof(u64); 290 sdp->sd_inptrs = (sdp->sd_sb.sb_bsize - 291 sizeof(struct gfs2_meta_header)) / sizeof(u64); 292 sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header); 293 sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2; 294 sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1; 295 sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64); 296 sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize - 297 sizeof(struct gfs2_meta_header)) / 298 sizeof(struct gfs2_quota_change); 299 sdp->sd_blocks_per_bitmap = (sdp->sd_sb.sb_bsize - 300 sizeof(struct gfs2_meta_header)) 301 * GFS2_NBBY; /* not the rgrp bitmap, subsequent bitmaps only */ 302 303 /* Compute maximum reservation required to add a entry to a directory */ 304 305 hash_blocks = DIV_ROUND_UP(sizeof(u64) * (1 << GFS2_DIR_MAX_DEPTH), 306 sdp->sd_jbsize); 307 308 ind_blocks = 0; 309 for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) { 310 tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs); 311 ind_blocks += tmp_blocks; 312 } 313 314 leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH; 315 316 sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks; 317 318 sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize - 319 sizeof(struct gfs2_dinode); 320 sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs; 321 for (x = 2;; x++) { 322 u64 space, d; 323 u32 m; 324 325 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs; 326 d = space; 327 m = do_div(d, sdp->sd_inptrs); 328 329 if (d != sdp->sd_heightsize[x - 1] || m) 330 break; 331 sdp->sd_heightsize[x] = space; 332 } 333 sdp->sd_max_height = x; 334 sdp->sd_heightsize[x] = ~0; 335 gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT); 336 337 sdp->sd_jheightsize[0] = sdp->sd_sb.sb_bsize - 338 sizeof(struct gfs2_dinode); 339 sdp->sd_jheightsize[1] = sdp->sd_jbsize * sdp->sd_diptrs; 340 for (x = 2;; x++) { 341 u64 space, d; 342 u32 m; 343 344 space = sdp->sd_jheightsize[x - 1] * sdp->sd_inptrs; 345 d = space; 346 m = do_div(d, sdp->sd_inptrs); 347 348 if (d != sdp->sd_jheightsize[x - 1] || m) 349 break; 350 sdp->sd_jheightsize[x] = space; 351 } 352 sdp->sd_max_jheight = x; 353 sdp->sd_jheightsize[x] = ~0; 354 gfs2_assert(sdp, sdp->sd_max_jheight <= GFS2_MAX_META_HEIGHT); 355 356 return 0; 357 } 358 359 static int init_names(struct gfs2_sbd *sdp, int silent) 360 { 361 char *proto, *table; 362 int error = 0; 363 364 proto = sdp->sd_args.ar_lockproto; 365 table = sdp->sd_args.ar_locktable; 366 367 /* Try to autodetect */ 368 369 if (!proto[0] || !table[0]) { 370 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent); 371 if (error) 372 return error; 373 374 if (!proto[0]) 375 proto = sdp->sd_sb.sb_lockproto; 376 if (!table[0]) 377 table = sdp->sd_sb.sb_locktable; 378 } 379 380 if (!table[0]) 381 table = sdp->sd_vfs->s_id; 382 383 strlcpy(sdp->sd_proto_name, proto, GFS2_FSNAME_LEN); 384 strlcpy(sdp->sd_table_name, table, GFS2_FSNAME_LEN); 385 386 table = sdp->sd_table_name; 387 while ((table = strchr(table, '/'))) 388 *table = '_'; 389 390 return error; 391 } 392 393 static int init_locking(struct gfs2_sbd *sdp, struct gfs2_holder *mount_gh, 394 int undo) 395 { 396 int error = 0; 397 398 if (undo) 399 goto fail_trans; 400 401 error = gfs2_glock_nq_num(sdp, 402 GFS2_MOUNT_LOCK, &gfs2_nondisk_glops, 403 LM_ST_EXCLUSIVE, LM_FLAG_NOEXP | GL_NOCACHE, 404 mount_gh); 405 if (error) { 406 fs_err(sdp, "can't acquire mount glock: %d\n", error); 407 goto fail; 408 } 409 410 error = gfs2_glock_nq_num(sdp, 411 GFS2_LIVE_LOCK, &gfs2_nondisk_glops, 412 LM_ST_SHARED, 413 LM_FLAG_NOEXP | GL_EXACT, 414 &sdp->sd_live_gh); 415 if (error) { 416 fs_err(sdp, "can't acquire live glock: %d\n", error); 417 goto fail_mount; 418 } 419 420 error = gfs2_glock_get(sdp, GFS2_RENAME_LOCK, &gfs2_nondisk_glops, 421 CREATE, &sdp->sd_rename_gl); 422 if (error) { 423 fs_err(sdp, "can't create rename glock: %d\n", error); 424 goto fail_live; 425 } 426 427 error = gfs2_glock_get(sdp, GFS2_FREEZE_LOCK, &gfs2_freeze_glops, 428 CREATE, &sdp->sd_freeze_gl); 429 if (error) { 430 fs_err(sdp, "can't create transaction glock: %d\n", error); 431 goto fail_rename; 432 } 433 434 return 0; 435 436 fail_trans: 437 gfs2_glock_put(sdp->sd_freeze_gl); 438 fail_rename: 439 gfs2_glock_put(sdp->sd_rename_gl); 440 fail_live: 441 gfs2_glock_dq_uninit(&sdp->sd_live_gh); 442 fail_mount: 443 gfs2_glock_dq_uninit(mount_gh); 444 fail: 445 return error; 446 } 447 448 static int gfs2_lookup_root(struct super_block *sb, struct dentry **dptr, 449 u64 no_addr, const char *name) 450 { 451 struct gfs2_sbd *sdp = sb->s_fs_info; 452 struct dentry *dentry; 453 struct inode *inode; 454 455 inode = gfs2_inode_lookup(sb, DT_DIR, no_addr, 0, 0); 456 if (IS_ERR(inode)) { 457 fs_err(sdp, "can't read in %s inode: %ld\n", name, PTR_ERR(inode)); 458 return PTR_ERR(inode); 459 } 460 dentry = d_make_root(inode); 461 if (!dentry) { 462 fs_err(sdp, "can't alloc %s dentry\n", name); 463 return -ENOMEM; 464 } 465 *dptr = dentry; 466 return 0; 467 } 468 469 static int init_sb(struct gfs2_sbd *sdp, int silent) 470 { 471 struct super_block *sb = sdp->sd_vfs; 472 struct gfs2_holder sb_gh; 473 u64 no_addr; 474 int ret; 475 476 ret = gfs2_glock_nq_num(sdp, GFS2_SB_LOCK, &gfs2_meta_glops, 477 LM_ST_SHARED, 0, &sb_gh); 478 if (ret) { 479 fs_err(sdp, "can't acquire superblock glock: %d\n", ret); 480 return ret; 481 } 482 483 ret = gfs2_read_sb(sdp, silent); 484 if (ret) { 485 fs_err(sdp, "can't read superblock: %d\n", ret); 486 goto out; 487 } 488 489 /* Set up the buffer cache and SB for real */ 490 if (sdp->sd_sb.sb_bsize < bdev_logical_block_size(sb->s_bdev)) { 491 ret = -EINVAL; 492 fs_err(sdp, "FS block size (%u) is too small for device " 493 "block size (%u)\n", 494 sdp->sd_sb.sb_bsize, bdev_logical_block_size(sb->s_bdev)); 495 goto out; 496 } 497 if (sdp->sd_sb.sb_bsize > PAGE_SIZE) { 498 ret = -EINVAL; 499 fs_err(sdp, "FS block size (%u) is too big for machine " 500 "page size (%u)\n", 501 sdp->sd_sb.sb_bsize, (unsigned int)PAGE_SIZE); 502 goto out; 503 } 504 sb_set_blocksize(sb, sdp->sd_sb.sb_bsize); 505 506 /* Get the root inode */ 507 no_addr = sdp->sd_sb.sb_root_dir.no_addr; 508 ret = gfs2_lookup_root(sb, &sdp->sd_root_dir, no_addr, "root"); 509 if (ret) 510 goto out; 511 512 /* Get the master inode */ 513 no_addr = sdp->sd_sb.sb_master_dir.no_addr; 514 ret = gfs2_lookup_root(sb, &sdp->sd_master_dir, no_addr, "master"); 515 if (ret) { 516 dput(sdp->sd_root_dir); 517 goto out; 518 } 519 sb->s_root = dget(sdp->sd_args.ar_meta ? sdp->sd_master_dir : sdp->sd_root_dir); 520 out: 521 gfs2_glock_dq_uninit(&sb_gh); 522 return ret; 523 } 524 525 static void gfs2_others_may_mount(struct gfs2_sbd *sdp) 526 { 527 char *message = "FIRSTMOUNT=Done"; 528 char *envp[] = { message, NULL }; 529 530 fs_info(sdp, "first mount done, others may mount\n"); 531 532 if (sdp->sd_lockstruct.ls_ops->lm_first_done) 533 sdp->sd_lockstruct.ls_ops->lm_first_done(sdp); 534 535 kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp); 536 } 537 538 /** 539 * gfs2_jindex_hold - Grab a lock on the jindex 540 * @sdp: The GFS2 superblock 541 * @ji_gh: the holder for the jindex glock 542 * 543 * Returns: errno 544 */ 545 546 static int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh) 547 { 548 struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex); 549 struct qstr name; 550 char buf[20]; 551 struct gfs2_jdesc *jd; 552 int error; 553 554 name.name = buf; 555 556 mutex_lock(&sdp->sd_jindex_mutex); 557 558 for (;;) { 559 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh); 560 if (error) 561 break; 562 563 name.len = sprintf(buf, "journal%u", sdp->sd_journals); 564 name.hash = gfs2_disk_hash(name.name, name.len); 565 566 error = gfs2_dir_check(sdp->sd_jindex, &name, NULL); 567 if (error == -ENOENT) { 568 error = 0; 569 break; 570 } 571 572 gfs2_glock_dq_uninit(ji_gh); 573 574 if (error) 575 break; 576 577 error = -ENOMEM; 578 jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL); 579 if (!jd) 580 break; 581 582 INIT_LIST_HEAD(&jd->extent_list); 583 INIT_LIST_HEAD(&jd->jd_revoke_list); 584 585 INIT_WORK(&jd->jd_work, gfs2_recover_func); 586 jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1); 587 if (!jd->jd_inode || IS_ERR(jd->jd_inode)) { 588 if (!jd->jd_inode) 589 error = -ENOENT; 590 else 591 error = PTR_ERR(jd->jd_inode); 592 kfree(jd); 593 break; 594 } 595 596 spin_lock(&sdp->sd_jindex_spin); 597 jd->jd_jid = sdp->sd_journals++; 598 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list); 599 spin_unlock(&sdp->sd_jindex_spin); 600 } 601 602 mutex_unlock(&sdp->sd_jindex_mutex); 603 604 return error; 605 } 606 607 /** 608 * check_journal_clean - Make sure a journal is clean for a spectator mount 609 * @sdp: The GFS2 superblock 610 * @jd: The journal descriptor 611 * 612 * Returns: 0 if the journal is clean or locked, else an error 613 */ 614 static int check_journal_clean(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd) 615 { 616 int error; 617 struct gfs2_holder j_gh; 618 struct gfs2_log_header_host head; 619 struct gfs2_inode *ip; 620 621 ip = GFS2_I(jd->jd_inode); 622 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_NOEXP | 623 GL_EXACT | GL_NOCACHE, &j_gh); 624 if (error) { 625 fs_err(sdp, "Error locking journal for spectator mount.\n"); 626 return -EPERM; 627 } 628 error = gfs2_jdesc_check(jd); 629 if (error) { 630 fs_err(sdp, "Error checking journal for spectator mount.\n"); 631 goto out_unlock; 632 } 633 error = gfs2_find_jhead(jd, &head); 634 if (error) { 635 fs_err(sdp, "Error parsing journal for spectator mount.\n"); 636 goto out_unlock; 637 } 638 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) { 639 error = -EPERM; 640 fs_err(sdp, "jid=%u: Journal is dirty, so the first mounter " 641 "must not be a spectator.\n", jd->jd_jid); 642 } 643 644 out_unlock: 645 gfs2_glock_dq_uninit(&j_gh); 646 return error; 647 } 648 649 static int init_journal(struct gfs2_sbd *sdp, int undo) 650 { 651 struct inode *master = sdp->sd_master_dir->d_inode; 652 struct gfs2_holder ji_gh; 653 struct gfs2_inode *ip; 654 int jindex = 1; 655 int error = 0; 656 657 if (undo) { 658 jindex = 0; 659 goto fail_jinode_gh; 660 } 661 662 sdp->sd_jindex = gfs2_lookup_simple(master, "jindex"); 663 if (IS_ERR(sdp->sd_jindex)) { 664 fs_err(sdp, "can't lookup journal index: %d\n", error); 665 return PTR_ERR(sdp->sd_jindex); 666 } 667 668 /* Load in the journal index special file */ 669 670 error = gfs2_jindex_hold(sdp, &ji_gh); 671 if (error) { 672 fs_err(sdp, "can't read journal index: %d\n", error); 673 goto fail; 674 } 675 676 error = -EUSERS; 677 if (!gfs2_jindex_size(sdp)) { 678 fs_err(sdp, "no journals!\n"); 679 goto fail_jindex; 680 } 681 682 if (sdp->sd_args.ar_spectator) { 683 sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0); 684 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks); 685 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5); 686 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5); 687 } else { 688 if (sdp->sd_lockstruct.ls_jid >= gfs2_jindex_size(sdp)) { 689 fs_err(sdp, "can't mount journal #%u\n", 690 sdp->sd_lockstruct.ls_jid); 691 fs_err(sdp, "there are only %u journals (0 - %u)\n", 692 gfs2_jindex_size(sdp), 693 gfs2_jindex_size(sdp) - 1); 694 goto fail_jindex; 695 } 696 sdp->sd_jdesc = gfs2_jdesc_find(sdp, sdp->sd_lockstruct.ls_jid); 697 698 error = gfs2_glock_nq_num(sdp, sdp->sd_lockstruct.ls_jid, 699 &gfs2_journal_glops, 700 LM_ST_EXCLUSIVE, LM_FLAG_NOEXP, 701 &sdp->sd_journal_gh); 702 if (error) { 703 fs_err(sdp, "can't acquire journal glock: %d\n", error); 704 goto fail_jindex; 705 } 706 707 ip = GFS2_I(sdp->sd_jdesc->jd_inode); 708 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 709 LM_FLAG_NOEXP | GL_EXACT | GL_NOCACHE, 710 &sdp->sd_jinode_gh); 711 if (error) { 712 fs_err(sdp, "can't acquire journal inode glock: %d\n", 713 error); 714 goto fail_journal_gh; 715 } 716 717 error = gfs2_jdesc_check(sdp->sd_jdesc); 718 if (error) { 719 fs_err(sdp, "my journal (%u) is bad: %d\n", 720 sdp->sd_jdesc->jd_jid, error); 721 goto fail_jinode_gh; 722 } 723 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks); 724 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5); 725 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5); 726 727 /* Map the extents for this journal's blocks */ 728 gfs2_map_journal_extents(sdp, sdp->sd_jdesc); 729 } 730 trace_gfs2_log_blocks(sdp, atomic_read(&sdp->sd_log_blks_free)); 731 732 if (sdp->sd_lockstruct.ls_first) { 733 unsigned int x; 734 for (x = 0; x < sdp->sd_journals; x++) { 735 struct gfs2_jdesc *jd = gfs2_jdesc_find(sdp, x); 736 737 if (sdp->sd_args.ar_spectator) { 738 error = check_journal_clean(sdp, jd); 739 if (error) 740 goto fail_jinode_gh; 741 continue; 742 } 743 error = gfs2_recover_journal(jd, true); 744 if (error) { 745 fs_err(sdp, "error recovering journal %u: %d\n", 746 x, error); 747 goto fail_jinode_gh; 748 } 749 } 750 751 gfs2_others_may_mount(sdp); 752 } else if (!sdp->sd_args.ar_spectator) { 753 error = gfs2_recover_journal(sdp->sd_jdesc, true); 754 if (error) { 755 fs_err(sdp, "error recovering my journal: %d\n", error); 756 goto fail_jinode_gh; 757 } 758 } 759 760 set_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags); 761 gfs2_glock_dq_uninit(&ji_gh); 762 jindex = 0; 763 INIT_WORK(&sdp->sd_freeze_work, gfs2_freeze_func); 764 return 0; 765 766 fail_jinode_gh: 767 if (!sdp->sd_args.ar_spectator) 768 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh); 769 fail_journal_gh: 770 if (!sdp->sd_args.ar_spectator) 771 gfs2_glock_dq_uninit(&sdp->sd_journal_gh); 772 fail_jindex: 773 gfs2_jindex_free(sdp); 774 if (jindex) 775 gfs2_glock_dq_uninit(&ji_gh); 776 fail: 777 iput(sdp->sd_jindex); 778 return error; 779 } 780 781 static struct lock_class_key gfs2_quota_imutex_key; 782 783 static int init_inodes(struct gfs2_sbd *sdp, int undo) 784 { 785 int error = 0; 786 struct inode *master = sdp->sd_master_dir->d_inode; 787 788 if (undo) 789 goto fail_qinode; 790 791 error = init_journal(sdp, undo); 792 complete_all(&sdp->sd_journal_ready); 793 if (error) 794 goto fail; 795 796 /* Read in the master statfs inode */ 797 sdp->sd_statfs_inode = gfs2_lookup_simple(master, "statfs"); 798 if (IS_ERR(sdp->sd_statfs_inode)) { 799 error = PTR_ERR(sdp->sd_statfs_inode); 800 fs_err(sdp, "can't read in statfs inode: %d\n", error); 801 goto fail_journal; 802 } 803 804 /* Read in the resource index inode */ 805 sdp->sd_rindex = gfs2_lookup_simple(master, "rindex"); 806 if (IS_ERR(sdp->sd_rindex)) { 807 error = PTR_ERR(sdp->sd_rindex); 808 fs_err(sdp, "can't get resource index inode: %d\n", error); 809 goto fail_statfs; 810 } 811 sdp->sd_rindex_uptodate = 0; 812 813 /* Read in the quota inode */ 814 sdp->sd_quota_inode = gfs2_lookup_simple(master, "quota"); 815 if (IS_ERR(sdp->sd_quota_inode)) { 816 error = PTR_ERR(sdp->sd_quota_inode); 817 fs_err(sdp, "can't get quota file inode: %d\n", error); 818 goto fail_rindex; 819 } 820 /* 821 * i_mutex on quota files is special. Since this inode is hidden system 822 * file, we are safe to define locking ourselves. 823 */ 824 lockdep_set_class(&sdp->sd_quota_inode->i_mutex, 825 &gfs2_quota_imutex_key); 826 827 error = gfs2_rindex_update(sdp); 828 if (error) 829 goto fail_qinode; 830 831 return 0; 832 833 fail_qinode: 834 iput(sdp->sd_quota_inode); 835 fail_rindex: 836 gfs2_clear_rgrpd(sdp); 837 iput(sdp->sd_rindex); 838 fail_statfs: 839 iput(sdp->sd_statfs_inode); 840 fail_journal: 841 init_journal(sdp, UNDO); 842 fail: 843 return error; 844 } 845 846 static int init_per_node(struct gfs2_sbd *sdp, int undo) 847 { 848 struct inode *pn = NULL; 849 char buf[30]; 850 int error = 0; 851 struct gfs2_inode *ip; 852 struct inode *master = sdp->sd_master_dir->d_inode; 853 854 if (sdp->sd_args.ar_spectator) 855 return 0; 856 857 if (undo) 858 goto fail_qc_gh; 859 860 pn = gfs2_lookup_simple(master, "per_node"); 861 if (IS_ERR(pn)) { 862 error = PTR_ERR(pn); 863 fs_err(sdp, "can't find per_node directory: %d\n", error); 864 return error; 865 } 866 867 sprintf(buf, "statfs_change%u", sdp->sd_jdesc->jd_jid); 868 sdp->sd_sc_inode = gfs2_lookup_simple(pn, buf); 869 if (IS_ERR(sdp->sd_sc_inode)) { 870 error = PTR_ERR(sdp->sd_sc_inode); 871 fs_err(sdp, "can't find local \"sc\" file: %d\n", error); 872 goto fail; 873 } 874 875 sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid); 876 sdp->sd_qc_inode = gfs2_lookup_simple(pn, buf); 877 if (IS_ERR(sdp->sd_qc_inode)) { 878 error = PTR_ERR(sdp->sd_qc_inode); 879 fs_err(sdp, "can't find local \"qc\" file: %d\n", error); 880 goto fail_ut_i; 881 } 882 883 iput(pn); 884 pn = NULL; 885 886 ip = GFS2_I(sdp->sd_sc_inode); 887 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, 888 &sdp->sd_sc_gh); 889 if (error) { 890 fs_err(sdp, "can't lock local \"sc\" file: %d\n", error); 891 goto fail_qc_i; 892 } 893 894 ip = GFS2_I(sdp->sd_qc_inode); 895 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, 896 &sdp->sd_qc_gh); 897 if (error) { 898 fs_err(sdp, "can't lock local \"qc\" file: %d\n", error); 899 goto fail_ut_gh; 900 } 901 902 return 0; 903 904 fail_qc_gh: 905 gfs2_glock_dq_uninit(&sdp->sd_qc_gh); 906 fail_ut_gh: 907 gfs2_glock_dq_uninit(&sdp->sd_sc_gh); 908 fail_qc_i: 909 iput(sdp->sd_qc_inode); 910 fail_ut_i: 911 iput(sdp->sd_sc_inode); 912 fail: 913 if (pn) 914 iput(pn); 915 return error; 916 } 917 918 static const match_table_t nolock_tokens = { 919 { Opt_jid, "jid=%d\n", }, 920 { Opt_err, NULL }, 921 }; 922 923 static const struct lm_lockops nolock_ops = { 924 .lm_proto_name = "lock_nolock", 925 .lm_put_lock = gfs2_glock_free, 926 .lm_tokens = &nolock_tokens, 927 }; 928 929 /** 930 * gfs2_lm_mount - mount a locking protocol 931 * @sdp: the filesystem 932 * @args: mount arguments 933 * @silent: if 1, don't complain if the FS isn't a GFS2 fs 934 * 935 * Returns: errno 936 */ 937 938 static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent) 939 { 940 const struct lm_lockops *lm; 941 struct lm_lockstruct *ls = &sdp->sd_lockstruct; 942 struct gfs2_args *args = &sdp->sd_args; 943 const char *proto = sdp->sd_proto_name; 944 const char *table = sdp->sd_table_name; 945 char *o, *options; 946 int ret; 947 948 if (!strcmp("lock_nolock", proto)) { 949 lm = &nolock_ops; 950 sdp->sd_args.ar_localflocks = 1; 951 #ifdef CONFIG_GFS2_FS_LOCKING_DLM 952 } else if (!strcmp("lock_dlm", proto)) { 953 lm = &gfs2_dlm_ops; 954 #endif 955 } else { 956 pr_info("can't find protocol %s\n", proto); 957 return -ENOENT; 958 } 959 960 fs_info(sdp, "Trying to join cluster \"%s\", \"%s\"\n", proto, table); 961 962 ls->ls_ops = lm; 963 ls->ls_first = 1; 964 965 for (options = args->ar_hostdata; (o = strsep(&options, ":")); ) { 966 substring_t tmp[MAX_OPT_ARGS]; 967 int token, option; 968 969 if (!o || !*o) 970 continue; 971 972 token = match_token(o, *lm->lm_tokens, tmp); 973 switch (token) { 974 case Opt_jid: 975 ret = match_int(&tmp[0], &option); 976 if (ret || option < 0) 977 goto hostdata_error; 978 if (test_and_clear_bit(SDF_NOJOURNALID, &sdp->sd_flags)) 979 ls->ls_jid = option; 980 break; 981 case Opt_id: 982 case Opt_nodir: 983 /* Obsolete, but left for backward compat purposes */ 984 break; 985 case Opt_first: 986 ret = match_int(&tmp[0], &option); 987 if (ret || (option != 0 && option != 1)) 988 goto hostdata_error; 989 ls->ls_first = option; 990 break; 991 case Opt_err: 992 default: 993 hostdata_error: 994 fs_info(sdp, "unknown hostdata (%s)\n", o); 995 return -EINVAL; 996 } 997 } 998 999 if (lm->lm_mount == NULL) { 1000 fs_info(sdp, "Now mounting FS...\n"); 1001 complete_all(&sdp->sd_locking_init); 1002 return 0; 1003 } 1004 ret = lm->lm_mount(sdp, table); 1005 if (ret == 0) 1006 fs_info(sdp, "Joined cluster. Now mounting FS...\n"); 1007 complete_all(&sdp->sd_locking_init); 1008 return ret; 1009 } 1010 1011 void gfs2_lm_unmount(struct gfs2_sbd *sdp) 1012 { 1013 const struct lm_lockops *lm = sdp->sd_lockstruct.ls_ops; 1014 if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) && 1015 lm->lm_unmount) 1016 lm->lm_unmount(sdp); 1017 } 1018 1019 static int wait_on_journal(struct gfs2_sbd *sdp) 1020 { 1021 if (sdp->sd_lockstruct.ls_ops->lm_mount == NULL) 1022 return 0; 1023 1024 return wait_on_bit(&sdp->sd_flags, SDF_NOJOURNALID, TASK_INTERRUPTIBLE) 1025 ? -EINTR : 0; 1026 } 1027 1028 void gfs2_online_uevent(struct gfs2_sbd *sdp) 1029 { 1030 struct super_block *sb = sdp->sd_vfs; 1031 char ro[20]; 1032 char spectator[20]; 1033 char *envp[] = { ro, spectator, NULL }; 1034 sprintf(ro, "RDONLY=%d", (sb->s_flags & MS_RDONLY) ? 1 : 0); 1035 sprintf(spectator, "SPECTATOR=%d", sdp->sd_args.ar_spectator ? 1 : 0); 1036 kobject_uevent_env(&sdp->sd_kobj, KOBJ_ONLINE, envp); 1037 } 1038 1039 /** 1040 * fill_super - Read in superblock 1041 * @sb: The VFS superblock 1042 * @data: Mount options 1043 * @silent: Don't complain if it's not a GFS2 filesystem 1044 * 1045 * Returns: errno 1046 */ 1047 1048 static int fill_super(struct super_block *sb, struct gfs2_args *args, int silent) 1049 { 1050 struct gfs2_sbd *sdp; 1051 struct gfs2_holder mount_gh; 1052 int error; 1053 1054 sdp = init_sbd(sb); 1055 if (!sdp) { 1056 pr_warn("can't alloc struct gfs2_sbd\n"); 1057 return -ENOMEM; 1058 } 1059 sdp->sd_args = *args; 1060 1061 if (sdp->sd_args.ar_spectator) { 1062 sb->s_flags |= MS_RDONLY; 1063 set_bit(SDF_RORECOVERY, &sdp->sd_flags); 1064 } 1065 if (sdp->sd_args.ar_posix_acl) 1066 sb->s_flags |= MS_POSIXACL; 1067 if (sdp->sd_args.ar_nobarrier) 1068 set_bit(SDF_NOBARRIERS, &sdp->sd_flags); 1069 1070 sb->s_flags |= MS_NOSEC; 1071 sb->s_magic = GFS2_MAGIC; 1072 sb->s_op = &gfs2_super_ops; 1073 sb->s_d_op = &gfs2_dops; 1074 sb->s_export_op = &gfs2_export_ops; 1075 sb->s_xattr = gfs2_xattr_handlers; 1076 sb->s_qcop = &gfs2_quotactl_ops; 1077 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP; 1078 sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE; 1079 sb->s_time_gran = 1; 1080 sb->s_maxbytes = MAX_LFS_FILESIZE; 1081 1082 /* Set up the buffer cache and fill in some fake block size values 1083 to allow us to read-in the on-disk superblock. */ 1084 sdp->sd_sb.sb_bsize = sb_min_blocksize(sb, GFS2_BASIC_BLOCK); 1085 sdp->sd_sb.sb_bsize_shift = sb->s_blocksize_bits; 1086 sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift - 1087 GFS2_BASIC_BLOCK_SHIFT; 1088 sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift; 1089 1090 sdp->sd_tune.gt_logd_secs = sdp->sd_args.ar_commit; 1091 sdp->sd_tune.gt_quota_quantum = sdp->sd_args.ar_quota_quantum; 1092 if (sdp->sd_args.ar_statfs_quantum) { 1093 sdp->sd_tune.gt_statfs_slow = 0; 1094 sdp->sd_tune.gt_statfs_quantum = sdp->sd_args.ar_statfs_quantum; 1095 } else { 1096 sdp->sd_tune.gt_statfs_slow = 1; 1097 sdp->sd_tune.gt_statfs_quantum = 30; 1098 } 1099 1100 error = init_names(sdp, silent); 1101 if (error) { 1102 /* In this case, we haven't initialized sysfs, so we have to 1103 manually free the sdp. */ 1104 free_percpu(sdp->sd_lkstats); 1105 kfree(sdp); 1106 sb->s_fs_info = NULL; 1107 return error; 1108 } 1109 1110 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s", sdp->sd_table_name); 1111 1112 error = gfs2_sys_fs_add(sdp); 1113 /* 1114 * If we hit an error here, gfs2_sys_fs_add will have called function 1115 * kobject_put which causes the sysfs usage count to go to zero, which 1116 * causes sysfs to call function gfs2_sbd_release, which frees sdp. 1117 * Subsequent error paths here will call gfs2_sys_fs_del, which also 1118 * kobject_put to free sdp. 1119 */ 1120 if (error) 1121 return error; 1122 1123 gfs2_create_debugfs_file(sdp); 1124 1125 error = gfs2_lm_mount(sdp, silent); 1126 if (error) 1127 goto fail_debug; 1128 1129 error = init_locking(sdp, &mount_gh, DO); 1130 if (error) 1131 goto fail_lm; 1132 1133 error = init_sb(sdp, silent); 1134 if (error) 1135 goto fail_locking; 1136 1137 error = wait_on_journal(sdp); 1138 if (error) 1139 goto fail_sb; 1140 1141 /* 1142 * If user space has failed to join the cluster or some similar 1143 * failure has occurred, then the journal id will contain a 1144 * negative (error) number. This will then be returned to the 1145 * caller (of the mount syscall). We do this even for spectator 1146 * mounts (which just write a jid of 0 to indicate "ok" even though 1147 * the jid is unused in the spectator case) 1148 */ 1149 if (sdp->sd_lockstruct.ls_jid < 0) { 1150 error = sdp->sd_lockstruct.ls_jid; 1151 sdp->sd_lockstruct.ls_jid = 0; 1152 goto fail_sb; 1153 } 1154 1155 if (sdp->sd_args.ar_spectator) 1156 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s.s", 1157 sdp->sd_table_name); 1158 else 1159 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s.%u", 1160 sdp->sd_table_name, sdp->sd_lockstruct.ls_jid); 1161 1162 error = init_inodes(sdp, DO); 1163 if (error) 1164 goto fail_sb; 1165 1166 error = init_per_node(sdp, DO); 1167 if (error) 1168 goto fail_inodes; 1169 1170 error = gfs2_statfs_init(sdp); 1171 if (error) { 1172 fs_err(sdp, "can't initialize statfs subsystem: %d\n", error); 1173 goto fail_per_node; 1174 } 1175 1176 if (!(sb->s_flags & MS_RDONLY)) { 1177 error = gfs2_make_fs_rw(sdp); 1178 if (error) { 1179 fs_err(sdp, "can't make FS RW: %d\n", error); 1180 goto fail_per_node; 1181 } 1182 } 1183 1184 gfs2_glock_dq_uninit(&mount_gh); 1185 gfs2_online_uevent(sdp); 1186 return 0; 1187 1188 fail_per_node: 1189 init_per_node(sdp, UNDO); 1190 fail_inodes: 1191 init_inodes(sdp, UNDO); 1192 fail_sb: 1193 if (sdp->sd_root_dir) 1194 dput(sdp->sd_root_dir); 1195 if (sdp->sd_master_dir) 1196 dput(sdp->sd_master_dir); 1197 if (sb->s_root) 1198 dput(sb->s_root); 1199 sb->s_root = NULL; 1200 fail_locking: 1201 init_locking(sdp, &mount_gh, UNDO); 1202 fail_lm: 1203 complete_all(&sdp->sd_journal_ready); 1204 gfs2_gl_hash_clear(sdp); 1205 gfs2_lm_unmount(sdp); 1206 fail_debug: 1207 gfs2_delete_debugfs_file(sdp); 1208 free_percpu(sdp->sd_lkstats); 1209 /* gfs2_sys_fs_del must be the last thing we do, since it causes 1210 * sysfs to call function gfs2_sbd_release, which frees sdp. */ 1211 gfs2_sys_fs_del(sdp); 1212 sb->s_fs_info = NULL; 1213 return error; 1214 } 1215 1216 static int set_gfs2_super(struct super_block *s, void *data) 1217 { 1218 s->s_bdev = data; 1219 s->s_dev = s->s_bdev->bd_dev; 1220 1221 /* 1222 * We set the bdi here to the queue backing, file systems can 1223 * overwrite this in ->fill_super() 1224 */ 1225 s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info; 1226 return 0; 1227 } 1228 1229 static int test_gfs2_super(struct super_block *s, void *ptr) 1230 { 1231 struct block_device *bdev = ptr; 1232 return (bdev == s->s_bdev); 1233 } 1234 1235 /** 1236 * gfs2_mount - Get the GFS2 superblock 1237 * @fs_type: The GFS2 filesystem type 1238 * @flags: Mount flags 1239 * @dev_name: The name of the device 1240 * @data: The mount arguments 1241 * 1242 * Q. Why not use get_sb_bdev() ? 1243 * A. We need to select one of two root directories to mount, independent 1244 * of whether this is the initial, or subsequent, mount of this sb 1245 * 1246 * Returns: 0 or -ve on error 1247 */ 1248 1249 static struct dentry *gfs2_mount(struct file_system_type *fs_type, int flags, 1250 const char *dev_name, void *data) 1251 { 1252 struct block_device *bdev; 1253 struct super_block *s; 1254 fmode_t mode = FMODE_READ | FMODE_EXCL; 1255 int error; 1256 struct gfs2_args args; 1257 struct gfs2_sbd *sdp; 1258 1259 if (!(flags & MS_RDONLY)) 1260 mode |= FMODE_WRITE; 1261 1262 bdev = blkdev_get_by_path(dev_name, mode, fs_type); 1263 if (IS_ERR(bdev)) 1264 return ERR_CAST(bdev); 1265 1266 /* 1267 * once the super is inserted into the list by sget, s_umount 1268 * will protect the lockfs code from trying to start a snapshot 1269 * while we are mounting 1270 */ 1271 mutex_lock(&bdev->bd_fsfreeze_mutex); 1272 if (bdev->bd_fsfreeze_count > 0) { 1273 mutex_unlock(&bdev->bd_fsfreeze_mutex); 1274 error = -EBUSY; 1275 goto error_bdev; 1276 } 1277 s = sget(fs_type, test_gfs2_super, set_gfs2_super, flags, bdev); 1278 mutex_unlock(&bdev->bd_fsfreeze_mutex); 1279 error = PTR_ERR(s); 1280 if (IS_ERR(s)) 1281 goto error_bdev; 1282 1283 if (s->s_root) { 1284 /* 1285 * s_umount nests inside bd_mutex during 1286 * __invalidate_device(). blkdev_put() acquires 1287 * bd_mutex and can't be called under s_umount. Drop 1288 * s_umount temporarily. This is safe as we're 1289 * holding an active reference. 1290 */ 1291 up_write(&s->s_umount); 1292 blkdev_put(bdev, mode); 1293 down_write(&s->s_umount); 1294 } 1295 1296 memset(&args, 0, sizeof(args)); 1297 args.ar_quota = GFS2_QUOTA_DEFAULT; 1298 args.ar_data = GFS2_DATA_DEFAULT; 1299 args.ar_commit = 30; 1300 args.ar_statfs_quantum = 30; 1301 args.ar_quota_quantum = 60; 1302 args.ar_errors = GFS2_ERRORS_DEFAULT; 1303 1304 error = gfs2_mount_args(&args, data); 1305 if (error) { 1306 pr_warn("can't parse mount arguments\n"); 1307 goto error_super; 1308 } 1309 1310 if (s->s_root) { 1311 error = -EBUSY; 1312 if ((flags ^ s->s_flags) & MS_RDONLY) 1313 goto error_super; 1314 } else { 1315 char b[BDEVNAME_SIZE]; 1316 1317 s->s_mode = mode; 1318 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id)); 1319 sb_set_blocksize(s, block_size(bdev)); 1320 error = fill_super(s, &args, flags & MS_SILENT ? 1 : 0); 1321 if (error) 1322 goto error_super; 1323 s->s_flags |= MS_ACTIVE; 1324 bdev->bd_super = s; 1325 } 1326 1327 sdp = s->s_fs_info; 1328 if (args.ar_meta) 1329 return dget(sdp->sd_master_dir); 1330 else 1331 return dget(sdp->sd_root_dir); 1332 1333 error_super: 1334 deactivate_locked_super(s); 1335 return ERR_PTR(error); 1336 error_bdev: 1337 blkdev_put(bdev, mode); 1338 return ERR_PTR(error); 1339 } 1340 1341 static int set_meta_super(struct super_block *s, void *ptr) 1342 { 1343 return -EINVAL; 1344 } 1345 1346 static struct dentry *gfs2_mount_meta(struct file_system_type *fs_type, 1347 int flags, const char *dev_name, void *data) 1348 { 1349 struct super_block *s; 1350 struct gfs2_sbd *sdp; 1351 struct path path; 1352 int error; 1353 1354 error = kern_path(dev_name, LOOKUP_FOLLOW, &path); 1355 if (error) { 1356 pr_warn("path_lookup on %s returned error %d\n", 1357 dev_name, error); 1358 return ERR_PTR(error); 1359 } 1360 s = sget(&gfs2_fs_type, test_gfs2_super, set_meta_super, flags, 1361 path.dentry->d_inode->i_sb->s_bdev); 1362 path_put(&path); 1363 if (IS_ERR(s)) { 1364 pr_warn("gfs2 mount does not exist\n"); 1365 return ERR_CAST(s); 1366 } 1367 if ((flags ^ s->s_flags) & MS_RDONLY) { 1368 deactivate_locked_super(s); 1369 return ERR_PTR(-EBUSY); 1370 } 1371 sdp = s->s_fs_info; 1372 return dget(sdp->sd_master_dir); 1373 } 1374 1375 static void gfs2_kill_sb(struct super_block *sb) 1376 { 1377 struct gfs2_sbd *sdp = sb->s_fs_info; 1378 1379 if (sdp == NULL) { 1380 kill_block_super(sb); 1381 return; 1382 } 1383 1384 gfs2_log_flush(sdp, NULL, SYNC_FLUSH); 1385 dput(sdp->sd_root_dir); 1386 dput(sdp->sd_master_dir); 1387 sdp->sd_root_dir = NULL; 1388 sdp->sd_master_dir = NULL; 1389 shrink_dcache_sb(sb); 1390 gfs2_delete_debugfs_file(sdp); 1391 free_percpu(sdp->sd_lkstats); 1392 kill_block_super(sb); 1393 } 1394 1395 struct file_system_type gfs2_fs_type = { 1396 .name = "gfs2", 1397 .fs_flags = FS_REQUIRES_DEV, 1398 .mount = gfs2_mount, 1399 .kill_sb = gfs2_kill_sb, 1400 .owner = THIS_MODULE, 1401 }; 1402 MODULE_ALIAS_FS("gfs2"); 1403 1404 struct file_system_type gfs2meta_fs_type = { 1405 .name = "gfs2meta", 1406 .fs_flags = FS_REQUIRES_DEV, 1407 .mount = gfs2_mount_meta, 1408 .owner = THIS_MODULE, 1409 }; 1410 MODULE_ALIAS_FS("gfs2meta"); 1411