1/* 2 * fs/logfs/super.c 3 * 4 * As should be obvious for Linux kernel code, license is GPLv2 5 * 6 * Copyright (c) 2005-2008 Joern Engel <joern@logfs.org> 7 * 8 * Generally contains mount/umount code and also serves as a dump area for 9 * any functions that don't fit elsewhere and neither justify a file of their 10 * own. 11 */ 12#include "logfs.h" 13#include <linux/bio.h> 14#include <linux/slab.h> 15#include <linux/blkdev.h> 16#include <linux/module.h> 17#include <linux/mtd/mtd.h> 18#include <linux/statfs.h> 19#include <linux/buffer_head.h> 20 21static DEFINE_MUTEX(emergency_mutex); 22static struct page *emergency_page; 23 24struct page *emergency_read_begin(struct address_space *mapping, pgoff_t index) 25{ 26 filler_t *filler = (filler_t *)mapping->a_ops->readpage; 27 struct page *page; 28 int err; 29 30 page = read_cache_page(mapping, index, filler, NULL); 31 if (page) 32 return page; 33 34 /* No more pages available, switch to emergency page */ 35 printk(KERN_INFO"Logfs: Using emergency page\n"); 36 mutex_lock(&emergency_mutex); 37 err = filler(NULL, emergency_page); 38 if (err) { 39 mutex_unlock(&emergency_mutex); 40 printk(KERN_EMERG"Logfs: Error reading emergency page\n"); 41 return ERR_PTR(err); 42 } 43 return emergency_page; 44} 45 46void emergency_read_end(struct page *page) 47{ 48 if (page == emergency_page) 49 mutex_unlock(&emergency_mutex); 50 else 51 page_cache_release(page); 52} 53 54static void dump_segfile(struct super_block *sb) 55{ 56 struct logfs_super *super = logfs_super(sb); 57 struct logfs_segment_entry se; 58 u32 segno; 59 60 for (segno = 0; segno < super->s_no_segs; segno++) { 61 logfs_get_segment_entry(sb, segno, &se); 62 printk("%3x: %6x %8x", segno, be32_to_cpu(se.ec_level), 63 be32_to_cpu(se.valid)); 64 if (++segno < super->s_no_segs) { 65 logfs_get_segment_entry(sb, segno, &se); 66 printk(" %6x %8x", be32_to_cpu(se.ec_level), 67 be32_to_cpu(se.valid)); 68 } 69 if (++segno < super->s_no_segs) { 70 logfs_get_segment_entry(sb, segno, &se); 71 printk(" %6x %8x", be32_to_cpu(se.ec_level), 72 be32_to_cpu(se.valid)); 73 } 74 if (++segno < super->s_no_segs) { 75 logfs_get_segment_entry(sb, segno, &se); 76 printk(" %6x %8x", be32_to_cpu(se.ec_level), 77 be32_to_cpu(se.valid)); 78 } 79 printk("\n"); 80 } 81} 82 83/* 84 * logfs_crash_dump - dump debug information to device 85 * 86 * The LogFS superblock only occupies part of a segment. This function will 87 * write as much debug information as it can gather into the spare space. 88 */ 89void logfs_crash_dump(struct super_block *sb) 90{ 91 dump_segfile(sb); 92} 93 94/* 95 * FIXME: There should be a reserve for root, similar to ext2. 96 */ 97int logfs_statfs(struct dentry *dentry, struct kstatfs *stats) 98{ 99 struct super_block *sb = dentry->d_sb; 100 struct logfs_super *super = logfs_super(sb); 101 102 stats->f_type = LOGFS_MAGIC_U32; 103 stats->f_bsize = sb->s_blocksize; 104 stats->f_blocks = super->s_size >> LOGFS_BLOCK_BITS >> 3; 105 stats->f_bfree = super->s_free_bytes >> sb->s_blocksize_bits; 106 stats->f_bavail = super->s_free_bytes >> sb->s_blocksize_bits; 107 stats->f_files = 0; 108 stats->f_ffree = 0; 109 stats->f_namelen = LOGFS_MAX_NAMELEN; 110 return 0; 111} 112 113static int logfs_sb_set(struct super_block *sb, void *_super) 114{ 115 struct logfs_super *super = _super; 116 117 sb->s_fs_info = super; 118 sb->s_mtd = super->s_mtd; 119 sb->s_bdev = super->s_bdev; 120#ifdef CONFIG_BLOCK 121 if (sb->s_bdev) 122 sb->s_bdi = &bdev_get_queue(sb->s_bdev)->backing_dev_info; 123#endif 124#ifdef CONFIG_MTD 125 if (sb->s_mtd) 126 sb->s_bdi = sb->s_mtd->backing_dev_info; 127#endif 128 return 0; 129} 130 131static int logfs_sb_test(struct super_block *sb, void *_super) 132{ 133 struct logfs_super *super = _super; 134 struct mtd_info *mtd = super->s_mtd; 135 136 if (mtd && sb->s_mtd == mtd) 137 return 1; 138 if (super->s_bdev && sb->s_bdev == super->s_bdev) 139 return 1; 140 return 0; 141} 142 143static void set_segment_header(struct logfs_segment_header *sh, u8 type, 144 u8 level, u32 segno, u32 ec) 145{ 146 sh->pad = 0; 147 sh->type = type; 148 sh->level = level; 149 sh->segno = cpu_to_be32(segno); 150 sh->ec = cpu_to_be32(ec); 151 sh->gec = cpu_to_be64(segno); 152 sh->crc = logfs_crc32(sh, LOGFS_SEGMENT_HEADERSIZE, 4); 153} 154 155static void logfs_write_ds(struct super_block *sb, struct logfs_disk_super *ds, 156 u32 segno, u32 ec) 157{ 158 struct logfs_super *super = logfs_super(sb); 159 struct logfs_segment_header *sh = &ds->ds_sh; 160 int i; 161 162 memset(ds, 0, sizeof(*ds)); 163 set_segment_header(sh, SEG_SUPER, 0, segno, ec); 164 165 ds->ds_ifile_levels = super->s_ifile_levels; 166 ds->ds_iblock_levels = super->s_iblock_levels; 167 ds->ds_data_levels = super->s_data_levels; /* XXX: Remove */ 168 ds->ds_segment_shift = super->s_segshift; 169 ds->ds_block_shift = sb->s_blocksize_bits; 170 ds->ds_write_shift = super->s_writeshift; 171 ds->ds_filesystem_size = cpu_to_be64(super->s_size); 172 ds->ds_segment_size = cpu_to_be32(super->s_segsize); 173 ds->ds_bad_seg_reserve = cpu_to_be32(super->s_bad_seg_reserve); 174 ds->ds_feature_incompat = cpu_to_be64(super->s_feature_incompat); 175 ds->ds_feature_ro_compat= cpu_to_be64(super->s_feature_ro_compat); 176 ds->ds_feature_compat = cpu_to_be64(super->s_feature_compat); 177 ds->ds_feature_flags = cpu_to_be64(super->s_feature_flags); 178 ds->ds_root_reserve = cpu_to_be64(super->s_root_reserve); 179 ds->ds_speed_reserve = cpu_to_be64(super->s_speed_reserve); 180 journal_for_each(i) 181 ds->ds_journal_seg[i] = cpu_to_be32(super->s_journal_seg[i]); 182 ds->ds_magic = cpu_to_be64(LOGFS_MAGIC); 183 ds->ds_crc = logfs_crc32(ds, sizeof(*ds), 184 LOGFS_SEGMENT_HEADERSIZE + 12); 185} 186 187static int write_one_sb(struct super_block *sb, 188 struct page *(*find_sb)(struct super_block *sb, u64 *ofs)) 189{ 190 struct logfs_super *super = logfs_super(sb); 191 struct logfs_disk_super *ds; 192 struct logfs_segment_entry se; 193 struct page *page; 194 u64 ofs; 195 u32 ec, segno; 196 int err; 197 198 page = find_sb(sb, &ofs); 199 if (!page) 200 return -EIO; 201 ds = page_address(page); 202 segno = seg_no(sb, ofs); 203 logfs_get_segment_entry(sb, segno, &se); 204 ec = be32_to_cpu(se.ec_level) >> 4; 205 ec++; 206 logfs_set_segment_erased(sb, segno, ec, 0); 207 logfs_write_ds(sb, ds, segno, ec); 208 err = super->s_devops->write_sb(sb, page); 209 page_cache_release(page); 210 return err; 211} 212 213int logfs_write_sb(struct super_block *sb) 214{ 215 struct logfs_super *super = logfs_super(sb); 216 int err; 217 218 /* First superblock */ 219 err = write_one_sb(sb, super->s_devops->find_first_sb); 220 if (err) 221 return err; 222 223 /* Last superblock */ 224 err = write_one_sb(sb, super->s_devops->find_last_sb); 225 if (err) 226 return err; 227 return 0; 228} 229 230static int ds_cmp(const void *ds0, const void *ds1) 231{ 232 size_t len = sizeof(struct logfs_disk_super); 233 234 /* We know the segment headers differ, so ignore them */ 235 len -= LOGFS_SEGMENT_HEADERSIZE; 236 ds0 += LOGFS_SEGMENT_HEADERSIZE; 237 ds1 += LOGFS_SEGMENT_HEADERSIZE; 238 return memcmp(ds0, ds1, len); 239} 240 241static int logfs_recover_sb(struct super_block *sb) 242{ 243 struct logfs_super *super = logfs_super(sb); 244 struct logfs_disk_super _ds0, *ds0 = &_ds0; 245 struct logfs_disk_super _ds1, *ds1 = &_ds1; 246 int err, valid0, valid1; 247 248 /* read first superblock */ 249 err = wbuf_read(sb, super->s_sb_ofs[0], sizeof(*ds0), ds0); 250 if (err) 251 return err; 252 /* read last superblock */ 253 err = wbuf_read(sb, super->s_sb_ofs[1], sizeof(*ds1), ds1); 254 if (err) 255 return err; 256 valid0 = logfs_check_ds(ds0) == 0; 257 valid1 = logfs_check_ds(ds1) == 0; 258 259 if (!valid0 && valid1) { 260 printk(KERN_INFO"First superblock is invalid - fixing.\n"); 261 return write_one_sb(sb, super->s_devops->find_first_sb); 262 } 263 if (valid0 && !valid1) { 264 printk(KERN_INFO"Last superblock is invalid - fixing.\n"); 265 return write_one_sb(sb, super->s_devops->find_last_sb); 266 } 267 if (valid0 && valid1 && ds_cmp(ds0, ds1)) { 268 printk(KERN_INFO"Superblocks don't match - fixing.\n"); 269 return logfs_write_sb(sb); 270 } 271 /* If neither is valid now, something's wrong. Didn't we properly 272 * check them before?!? */ 273 BUG_ON(!valid0 && !valid1); 274 return 0; 275} 276 277static int logfs_make_writeable(struct super_block *sb) 278{ 279 int err; 280 281 err = logfs_open_segfile(sb); 282 if (err) 283 return err; 284 285 /* Repair any broken superblock copies */ 286 err = logfs_recover_sb(sb); 287 if (err) 288 return err; 289 290 /* Check areas for trailing unaccounted data */ 291 err = logfs_check_areas(sb); 292 if (err) 293 return err; 294 295 /* Do one GC pass before any data gets dirtied */ 296 logfs_gc_pass(sb); 297 298 /* after all initializations are done, replay the journal 299 * for rw-mounts, if necessary */ 300 err = logfs_replay_journal(sb); 301 if (err) 302 return err; 303 304 return 0; 305} 306 307static int logfs_get_sb_final(struct super_block *sb) 308{ 309 struct logfs_super *super = logfs_super(sb); 310 struct inode *rootdir; 311 int err; 312 313 /* root dir */ 314 rootdir = logfs_iget(sb, LOGFS_INO_ROOT); 315 if (IS_ERR(rootdir)) 316 goto fail; 317 318 sb->s_root = d_make_root(rootdir); 319 if (!sb->s_root) 320 goto fail; 321 322 /* at that point we know that ->put_super() will be called */ 323 super->s_erase_page = alloc_pages(GFP_KERNEL, 0); 324 if (!super->s_erase_page) 325 return -ENOMEM; 326 memset(page_address(super->s_erase_page), 0xFF, PAGE_SIZE); 327 328 /* FIXME: check for read-only mounts */ 329 err = logfs_make_writeable(sb); 330 if (err) { 331 __free_page(super->s_erase_page); 332 return err; 333 } 334 335 log_super("LogFS: Finished mounting\n"); 336 return 0; 337 338fail: 339 iput(super->s_master_inode); 340 iput(super->s_segfile_inode); 341 iput(super->s_mapping_inode); 342 return -EIO; 343} 344 345int logfs_check_ds(struct logfs_disk_super *ds) 346{ 347 struct logfs_segment_header *sh = &ds->ds_sh; 348 349 if (ds->ds_magic != cpu_to_be64(LOGFS_MAGIC)) 350 return -EINVAL; 351 if (sh->crc != logfs_crc32(sh, LOGFS_SEGMENT_HEADERSIZE, 4)) 352 return -EINVAL; 353 if (ds->ds_crc != logfs_crc32(ds, sizeof(*ds), 354 LOGFS_SEGMENT_HEADERSIZE + 12)) 355 return -EINVAL; 356 return 0; 357} 358 359static struct page *find_super_block(struct super_block *sb) 360{ 361 struct logfs_super *super = logfs_super(sb); 362 struct page *first, *last; 363 364 first = super->s_devops->find_first_sb(sb, &super->s_sb_ofs[0]); 365 if (!first || IS_ERR(first)) 366 return NULL; 367 last = super->s_devops->find_last_sb(sb, &super->s_sb_ofs[1]); 368 if (!last || IS_ERR(last)) { 369 page_cache_release(first); 370 return NULL; 371 } 372 373 if (!logfs_check_ds(page_address(first))) { 374 page_cache_release(last); 375 return first; 376 } 377 378 /* First one didn't work, try the second superblock */ 379 if (!logfs_check_ds(page_address(last))) { 380 page_cache_release(first); 381 return last; 382 } 383 384 /* Neither worked, sorry folks */ 385 page_cache_release(first); 386 page_cache_release(last); 387 return NULL; 388} 389 390static int __logfs_read_sb(struct super_block *sb) 391{ 392 struct logfs_super *super = logfs_super(sb); 393 struct page *page; 394 struct logfs_disk_super *ds; 395 int i; 396 397 page = find_super_block(sb); 398 if (!page) 399 return -EINVAL; 400 401 ds = page_address(page); 402 super->s_size = be64_to_cpu(ds->ds_filesystem_size); 403 super->s_root_reserve = be64_to_cpu(ds->ds_root_reserve); 404 super->s_speed_reserve = be64_to_cpu(ds->ds_speed_reserve); 405 super->s_bad_seg_reserve = be32_to_cpu(ds->ds_bad_seg_reserve); 406 super->s_segsize = 1 << ds->ds_segment_shift; 407 super->s_segmask = (1 << ds->ds_segment_shift) - 1; 408 super->s_segshift = ds->ds_segment_shift; 409 sb->s_blocksize = 1 << ds->ds_block_shift; 410 sb->s_blocksize_bits = ds->ds_block_shift; 411 super->s_writesize = 1 << ds->ds_write_shift; 412 super->s_writeshift = ds->ds_write_shift; 413 super->s_no_segs = super->s_size >> super->s_segshift; 414 super->s_no_blocks = super->s_segsize >> sb->s_blocksize_bits; 415 super->s_feature_incompat = be64_to_cpu(ds->ds_feature_incompat); 416 super->s_feature_ro_compat = be64_to_cpu(ds->ds_feature_ro_compat); 417 super->s_feature_compat = be64_to_cpu(ds->ds_feature_compat); 418 super->s_feature_flags = be64_to_cpu(ds->ds_feature_flags); 419 420 journal_for_each(i) 421 super->s_journal_seg[i] = be32_to_cpu(ds->ds_journal_seg[i]); 422 423 super->s_ifile_levels = ds->ds_ifile_levels; 424 super->s_iblock_levels = ds->ds_iblock_levels; 425 super->s_data_levels = ds->ds_data_levels; 426 super->s_total_levels = super->s_ifile_levels + super->s_iblock_levels 427 + super->s_data_levels; 428 page_cache_release(page); 429 return 0; 430} 431 432static int logfs_read_sb(struct super_block *sb, int read_only) 433{ 434 struct logfs_super *super = logfs_super(sb); 435 int ret; 436 437 super->s_btree_pool = mempool_create(32, btree_alloc, btree_free, NULL); 438 if (!super->s_btree_pool) 439 return -ENOMEM; 440 441 btree_init_mempool64(&super->s_shadow_tree.new, super->s_btree_pool); 442 btree_init_mempool64(&super->s_shadow_tree.old, super->s_btree_pool); 443 btree_init_mempool32(&super->s_shadow_tree.segment_map, 444 super->s_btree_pool); 445 446 ret = logfs_init_mapping(sb); 447 if (ret) 448 return ret; 449 450 ret = __logfs_read_sb(sb); 451 if (ret) 452 return ret; 453 454 if (super->s_feature_incompat & ~LOGFS_FEATURES_INCOMPAT) 455 return -EIO; 456 if ((super->s_feature_ro_compat & ~LOGFS_FEATURES_RO_COMPAT) && 457 !read_only) 458 return -EIO; 459 460 ret = logfs_init_rw(sb); 461 if (ret) 462 return ret; 463 464 ret = logfs_init_areas(sb); 465 if (ret) 466 return ret; 467 468 ret = logfs_init_gc(sb); 469 if (ret) 470 return ret; 471 472 ret = logfs_init_journal(sb); 473 if (ret) 474 return ret; 475 476 return 0; 477} 478 479static void logfs_kill_sb(struct super_block *sb) 480{ 481 struct logfs_super *super = logfs_super(sb); 482 483 log_super("LogFS: Start unmounting\n"); 484 /* Alias entries slow down mount, so evict as many as possible */ 485 sync_filesystem(sb); 486 logfs_write_anchor(sb); 487 free_areas(sb); 488 489 /* 490 * From this point on alias entries are simply dropped - and any 491 * writes to the object store are considered bugs. 492 */ 493 log_super("LogFS: Now in shutdown\n"); 494 generic_shutdown_super(sb); 495 super->s_flags |= LOGFS_SB_FLAG_SHUTDOWN; 496 497 BUG_ON(super->s_dirty_used_bytes || super->s_dirty_free_bytes); 498 499 logfs_cleanup_gc(sb); 500 logfs_cleanup_journal(sb); 501 logfs_cleanup_areas(sb); 502 logfs_cleanup_rw(sb); 503 if (super->s_erase_page) 504 __free_page(super->s_erase_page); 505 super->s_devops->put_device(super); 506 logfs_mempool_destroy(super->s_btree_pool); 507 logfs_mempool_destroy(super->s_alias_pool); 508 kfree(super); 509 log_super("LogFS: Finished unmounting\n"); 510} 511 512static struct dentry *logfs_get_sb_device(struct logfs_super *super, 513 struct file_system_type *type, int flags) 514{ 515 struct super_block *sb; 516 int err = -ENOMEM; 517 static int mount_count; 518 519 log_super("LogFS: Start mount %x\n", mount_count++); 520 521 err = -EINVAL; 522 sb = sget(type, logfs_sb_test, logfs_sb_set, flags | MS_NOATIME, super); 523 if (IS_ERR(sb)) { 524 super->s_devops->put_device(super); 525 kfree(super); 526 return ERR_CAST(sb); 527 } 528 529 if (sb->s_root) { 530 /* Device is already in use */ 531 super->s_devops->put_device(super); 532 kfree(super); 533 return dget(sb->s_root); 534 } 535 536 /* 537 * sb->s_maxbytes is limited to 8TB. On 32bit systems, the page cache 538 * only covers 16TB and the upper 8TB are used for indirect blocks. 539 * On 64bit system we could bump up the limit, but that would make 540 * the filesystem incompatible with 32bit systems. 541 */ 542 sb->s_maxbytes = (1ull << 43) - 1; 543 sb->s_max_links = LOGFS_LINK_MAX; 544 sb->s_op = &logfs_super_operations; 545 546 err = logfs_read_sb(sb, sb->s_flags & MS_RDONLY); 547 if (err) 548 goto err1; 549 550 sb->s_flags |= MS_ACTIVE; 551 err = logfs_get_sb_final(sb); 552 if (err) { 553 deactivate_locked_super(sb); 554 return ERR_PTR(err); 555 } 556 return dget(sb->s_root); 557 558err1: 559 /* no ->s_root, no ->put_super() */ 560 iput(super->s_master_inode); 561 iput(super->s_segfile_inode); 562 iput(super->s_mapping_inode); 563 deactivate_locked_super(sb); 564 return ERR_PTR(err); 565} 566 567static struct dentry *logfs_mount(struct file_system_type *type, int flags, 568 const char *devname, void *data) 569{ 570 ulong mtdnr; 571 struct logfs_super *super; 572 int err; 573 574 super = kzalloc(sizeof(*super), GFP_KERNEL); 575 if (!super) 576 return ERR_PTR(-ENOMEM); 577 578 mutex_init(&super->s_dirop_mutex); 579 mutex_init(&super->s_object_alias_mutex); 580 INIT_LIST_HEAD(&super->s_freeing_list); 581 582 if (!devname) 583 err = logfs_get_sb_bdev(super, type, devname); 584 else if (strncmp(devname, "mtd", 3)) 585 err = logfs_get_sb_bdev(super, type, devname); 586 else { 587 char *garbage; 588 mtdnr = simple_strtoul(devname+3, &garbage, 0); 589 if (*garbage) 590 err = -EINVAL; 591 else 592 err = logfs_get_sb_mtd(super, mtdnr); 593 } 594 595 if (err) { 596 kfree(super); 597 return ERR_PTR(err); 598 } 599 600 return logfs_get_sb_device(super, type, flags); 601} 602 603static struct file_system_type logfs_fs_type = { 604 .owner = THIS_MODULE, 605 .name = "logfs", 606 .mount = logfs_mount, 607 .kill_sb = logfs_kill_sb, 608 .fs_flags = FS_REQUIRES_DEV, 609 610}; 611MODULE_ALIAS_FS("logfs"); 612 613static int __init logfs_init(void) 614{ 615 int ret; 616 617 emergency_page = alloc_pages(GFP_KERNEL, 0); 618 if (!emergency_page) 619 return -ENOMEM; 620 621 ret = logfs_compr_init(); 622 if (ret) 623 goto out1; 624 625 ret = logfs_init_inode_cache(); 626 if (ret) 627 goto out2; 628 629 ret = register_filesystem(&logfs_fs_type); 630 if (!ret) 631 return 0; 632 logfs_destroy_inode_cache(); 633out2: 634 logfs_compr_exit(); 635out1: 636 __free_pages(emergency_page, 0); 637 return ret; 638} 639 640static void __exit logfs_exit(void) 641{ 642 unregister_filesystem(&logfs_fs_type); 643 logfs_destroy_inode_cache(); 644 logfs_compr_exit(); 645 __free_pages(emergency_page, 0); 646} 647 648module_init(logfs_init); 649module_exit(logfs_exit); 650 651MODULE_LICENSE("GPL v2"); 652MODULE_AUTHOR("Joern Engel <joern@logfs.org>"); 653MODULE_DESCRIPTION("scalable flash filesystem"); 654