1 /* SPDX-License-Identifier: GPL-2.0-only */ 2 /* 3 * Copyright (C) 2017-2018 HUAWEI, Inc. 4 * https://www.huawei.com/ 5 * Copyright (C) 2021, Alibaba Cloud 6 */ 7 #ifndef __EROFS_INTERNAL_H 8 #define __EROFS_INTERNAL_H 9 10 #include <linux/fs.h> 11 #include <linux/dcache.h> 12 #include <linux/mm.h> 13 #include <linux/pagemap.h> 14 #include <linux/bio.h> 15 #include <linux/buffer_head.h> 16 #include <linux/magic.h> 17 #include <linux/slab.h> 18 #include <linux/vmalloc.h> 19 #include <linux/iomap.h> 20 #include "erofs_fs.h" 21 22 /* redefine pr_fmt "erofs: " */ 23 #undef pr_fmt 24 #define pr_fmt(fmt) "erofs: " fmt 25 26 __printf(3, 4) void _erofs_err(struct super_block *sb, 27 const char *function, const char *fmt, ...); 28 #define erofs_err(sb, fmt, ...) \ 29 _erofs_err(sb, __func__, fmt "\n", ##__VA_ARGS__) 30 __printf(3, 4) void _erofs_info(struct super_block *sb, 31 const char *function, const char *fmt, ...); 32 #define erofs_info(sb, fmt, ...) \ 33 _erofs_info(sb, __func__, fmt "\n", ##__VA_ARGS__) 34 #ifdef CONFIG_EROFS_FS_DEBUG 35 #define erofs_dbg(x, ...) pr_debug(x "\n", ##__VA_ARGS__) 36 #define DBG_BUGON BUG_ON 37 #else 38 #define erofs_dbg(x, ...) ((void)0) 39 #define DBG_BUGON(x) ((void)(x)) 40 #endif /* !CONFIG_EROFS_FS_DEBUG */ 41 42 /* EROFS_SUPER_MAGIC_V1 to represent the whole file system */ 43 #define EROFS_SUPER_MAGIC EROFS_SUPER_MAGIC_V1 44 45 typedef u64 erofs_nid_t; 46 typedef u64 erofs_off_t; 47 /* data type for filesystem-wide blocks number */ 48 typedef u32 erofs_blk_t; 49 50 struct erofs_device_info { 51 char *path; 52 struct block_device *bdev; 53 struct dax_device *dax_dev; 54 u64 dax_part_off; 55 56 u32 blocks; 57 u32 mapped_blkaddr; 58 }; 59 60 enum { 61 EROFS_SYNC_DECOMPRESS_AUTO, 62 EROFS_SYNC_DECOMPRESS_FORCE_ON, 63 EROFS_SYNC_DECOMPRESS_FORCE_OFF 64 }; 65 66 struct erofs_mount_opts { 67 #ifdef CONFIG_EROFS_FS_ZIP 68 /* current strategy of how to use managed cache */ 69 unsigned char cache_strategy; 70 /* strategy of sync decompression (0 - auto, 1 - force on, 2 - force off) */ 71 unsigned int sync_decompress; 72 73 /* threshold for decompression synchronously */ 74 unsigned int max_sync_decompress_pages; 75 #endif 76 unsigned int mount_opt; 77 }; 78 79 struct erofs_dev_context { 80 struct idr tree; 81 struct rw_semaphore rwsem; 82 83 unsigned int extra_devices; 84 }; 85 86 struct erofs_fs_context { 87 struct erofs_mount_opts opt; 88 struct erofs_dev_context *devs; 89 }; 90 91 /* all filesystem-wide lz4 configurations */ 92 struct erofs_sb_lz4_info { 93 /* # of pages needed for EROFS lz4 rolling decompression */ 94 u16 max_distance_pages; 95 /* maximum possible blocks for pclusters in the filesystem */ 96 u16 max_pclusterblks; 97 }; 98 99 struct erofs_sb_info { 100 struct erofs_mount_opts opt; /* options */ 101 #ifdef CONFIG_EROFS_FS_ZIP 102 /* list for all registered superblocks, mainly for shrinker */ 103 struct list_head list; 104 struct mutex umount_mutex; 105 106 /* managed XArray arranged in physical block number */ 107 struct xarray managed_pslots; 108 109 unsigned int shrinker_run_no; 110 u16 available_compr_algs; 111 112 /* pseudo inode to manage cached pages */ 113 struct inode *managed_cache; 114 115 struct erofs_sb_lz4_info lz4; 116 #endif /* CONFIG_EROFS_FS_ZIP */ 117 struct erofs_dev_context *devs; 118 struct dax_device *dax_dev; 119 u64 dax_part_off; 120 u64 total_blocks; 121 u32 primarydevice_blocks; 122 123 u32 meta_blkaddr; 124 #ifdef CONFIG_EROFS_FS_XATTR 125 u32 xattr_blkaddr; 126 #endif 127 u16 device_id_mask; /* valid bits of device id to be used */ 128 129 /* inode slot unit size in bit shift */ 130 unsigned char islotbits; 131 132 u32 sb_size; /* total superblock size */ 133 u32 build_time_nsec; 134 u64 build_time; 135 136 /* what we really care is nid, rather than ino.. */ 137 erofs_nid_t root_nid; 138 /* used for statfs, f_files - f_favail */ 139 u64 inos; 140 141 u8 uuid[16]; /* 128-bit uuid for volume */ 142 u8 volume_name[16]; /* volume name */ 143 u32 feature_compat; 144 u32 feature_incompat; 145 146 /* sysfs support */ 147 struct kobject s_kobj; /* /sys/fs/erofs/<devname> */ 148 struct completion s_kobj_unregister; 149 }; 150 151 #define EROFS_SB(sb) ((struct erofs_sb_info *)(sb)->s_fs_info) 152 #define EROFS_I_SB(inode) ((struct erofs_sb_info *)(inode)->i_sb->s_fs_info) 153 154 /* Mount flags set via mount options or defaults */ 155 #define EROFS_MOUNT_XATTR_USER 0x00000010 156 #define EROFS_MOUNT_POSIX_ACL 0x00000020 157 #define EROFS_MOUNT_DAX_ALWAYS 0x00000040 158 #define EROFS_MOUNT_DAX_NEVER 0x00000080 159 160 #define clear_opt(opt, option) ((opt)->mount_opt &= ~EROFS_MOUNT_##option) 161 #define set_opt(opt, option) ((opt)->mount_opt |= EROFS_MOUNT_##option) 162 #define test_opt(opt, option) ((opt)->mount_opt & EROFS_MOUNT_##option) 163 164 enum { 165 EROFS_ZIP_CACHE_DISABLED, 166 EROFS_ZIP_CACHE_READAHEAD, 167 EROFS_ZIP_CACHE_READAROUND 168 }; 169 170 #ifdef CONFIG_EROFS_FS_ZIP 171 #define EROFS_LOCKED_MAGIC (INT_MIN | 0xE0F510CCL) 172 173 /* basic unit of the workstation of a super_block */ 174 struct erofs_workgroup { 175 /* the workgroup index in the workstation */ 176 pgoff_t index; 177 178 /* overall workgroup reference count */ 179 atomic_t refcount; 180 }; 181 182 #if defined(CONFIG_SMP) 183 static inline bool erofs_workgroup_try_to_freeze(struct erofs_workgroup *grp, 184 int val) 185 { 186 preempt_disable(); 187 if (val != atomic_cmpxchg(&grp->refcount, val, EROFS_LOCKED_MAGIC)) { 188 preempt_enable(); 189 return false; 190 } 191 return true; 192 } 193 194 static inline void erofs_workgroup_unfreeze(struct erofs_workgroup *grp, 195 int orig_val) 196 { 197 /* 198 * other observers should notice all modifications 199 * in the freezing period. 200 */ 201 smp_mb(); 202 atomic_set(&grp->refcount, orig_val); 203 preempt_enable(); 204 } 205 206 static inline int erofs_wait_on_workgroup_freezed(struct erofs_workgroup *grp) 207 { 208 return atomic_cond_read_relaxed(&grp->refcount, 209 VAL != EROFS_LOCKED_MAGIC); 210 } 211 #else 212 static inline bool erofs_workgroup_try_to_freeze(struct erofs_workgroup *grp, 213 int val) 214 { 215 preempt_disable(); 216 /* no need to spin on UP platforms, let's just disable preemption. */ 217 if (val != atomic_read(&grp->refcount)) { 218 preempt_enable(); 219 return false; 220 } 221 return true; 222 } 223 224 static inline void erofs_workgroup_unfreeze(struct erofs_workgroup *grp, 225 int orig_val) 226 { 227 preempt_enable(); 228 } 229 230 static inline int erofs_wait_on_workgroup_freezed(struct erofs_workgroup *grp) 231 { 232 int v = atomic_read(&grp->refcount); 233 234 /* workgroup is never freezed on uniprocessor systems */ 235 DBG_BUGON(v == EROFS_LOCKED_MAGIC); 236 return v; 237 } 238 #endif /* !CONFIG_SMP */ 239 #endif /* !CONFIG_EROFS_FS_ZIP */ 240 241 /* we strictly follow PAGE_SIZE and no buffer head yet */ 242 #define LOG_BLOCK_SIZE PAGE_SHIFT 243 244 #undef LOG_SECTORS_PER_BLOCK 245 #define LOG_SECTORS_PER_BLOCK (PAGE_SHIFT - 9) 246 247 #undef SECTORS_PER_BLOCK 248 #define SECTORS_PER_BLOCK (1 << SECTORS_PER_BLOCK) 249 250 #define EROFS_BLKSIZ (1 << LOG_BLOCK_SIZE) 251 252 #if (EROFS_BLKSIZ % 4096 || !EROFS_BLKSIZ) 253 #error erofs cannot be used in this platform 254 #endif 255 256 enum erofs_kmap_type { 257 EROFS_NO_KMAP, /* don't map the buffer */ 258 EROFS_KMAP, /* use kmap() to map the buffer */ 259 EROFS_KMAP_ATOMIC, /* use kmap_atomic() to map the buffer */ 260 }; 261 262 struct erofs_buf { 263 struct page *page; 264 void *base; 265 enum erofs_kmap_type kmap_type; 266 }; 267 #define __EROFS_BUF_INITIALIZER ((struct erofs_buf){ .page = NULL }) 268 269 #define ROOT_NID(sb) ((sb)->root_nid) 270 271 #define erofs_blknr(addr) ((addr) / EROFS_BLKSIZ) 272 #define erofs_blkoff(addr) ((addr) % EROFS_BLKSIZ) 273 #define blknr_to_addr(nr) ((erofs_off_t)(nr) * EROFS_BLKSIZ) 274 275 static inline erofs_off_t iloc(struct erofs_sb_info *sbi, erofs_nid_t nid) 276 { 277 return blknr_to_addr(sbi->meta_blkaddr) + (nid << sbi->islotbits); 278 } 279 280 #define EROFS_FEATURE_FUNCS(name, compat, feature) \ 281 static inline bool erofs_sb_has_##name(struct erofs_sb_info *sbi) \ 282 { \ 283 return sbi->feature_##compat & EROFS_FEATURE_##feature; \ 284 } 285 286 EROFS_FEATURE_FUNCS(zero_padding, incompat, INCOMPAT_ZERO_PADDING) 287 EROFS_FEATURE_FUNCS(compr_cfgs, incompat, INCOMPAT_COMPR_CFGS) 288 EROFS_FEATURE_FUNCS(big_pcluster, incompat, INCOMPAT_BIG_PCLUSTER) 289 EROFS_FEATURE_FUNCS(chunked_file, incompat, INCOMPAT_CHUNKED_FILE) 290 EROFS_FEATURE_FUNCS(device_table, incompat, INCOMPAT_DEVICE_TABLE) 291 EROFS_FEATURE_FUNCS(compr_head2, incompat, INCOMPAT_COMPR_HEAD2) 292 EROFS_FEATURE_FUNCS(ztailpacking, incompat, INCOMPAT_ZTAILPACKING) 293 EROFS_FEATURE_FUNCS(sb_chksum, compat, COMPAT_SB_CHKSUM) 294 295 /* atomic flag definitions */ 296 #define EROFS_I_EA_INITED_BIT 0 297 #define EROFS_I_Z_INITED_BIT 1 298 299 /* bitlock definitions (arranged in reverse order) */ 300 #define EROFS_I_BL_XATTR_BIT (BITS_PER_LONG - 1) 301 #define EROFS_I_BL_Z_BIT (BITS_PER_LONG - 2) 302 303 struct erofs_inode { 304 erofs_nid_t nid; 305 306 /* atomic flags (including bitlocks) */ 307 unsigned long flags; 308 309 unsigned char datalayout; 310 unsigned char inode_isize; 311 unsigned short xattr_isize; 312 313 unsigned int xattr_shared_count; 314 unsigned int *xattr_shared_xattrs; 315 316 union { 317 erofs_blk_t raw_blkaddr; 318 struct { 319 unsigned short chunkformat; 320 unsigned char chunkbits; 321 }; 322 #ifdef CONFIG_EROFS_FS_ZIP 323 struct { 324 unsigned short z_advise; 325 unsigned char z_algorithmtype[2]; 326 unsigned char z_logical_clusterbits; 327 unsigned long z_tailextent_headlcn; 328 unsigned int z_idataoff; 329 unsigned short z_idata_size; 330 }; 331 #endif /* CONFIG_EROFS_FS_ZIP */ 332 }; 333 /* the corresponding vfs inode */ 334 struct inode vfs_inode; 335 }; 336 337 #define EROFS_I(ptr) \ 338 container_of(ptr, struct erofs_inode, vfs_inode) 339 340 static inline unsigned long erofs_inode_datablocks(struct inode *inode) 341 { 342 /* since i_size cannot be changed */ 343 return DIV_ROUND_UP(inode->i_size, EROFS_BLKSIZ); 344 } 345 346 static inline unsigned int erofs_bitrange(unsigned int value, unsigned int bit, 347 unsigned int bits) 348 { 349 350 return (value >> bit) & ((1 << bits) - 1); 351 } 352 353 354 static inline unsigned int erofs_inode_version(unsigned int value) 355 { 356 return erofs_bitrange(value, EROFS_I_VERSION_BIT, 357 EROFS_I_VERSION_BITS); 358 } 359 360 static inline unsigned int erofs_inode_datalayout(unsigned int value) 361 { 362 return erofs_bitrange(value, EROFS_I_DATALAYOUT_BIT, 363 EROFS_I_DATALAYOUT_BITS); 364 } 365 366 /* 367 * Different from grab_cache_page_nowait(), reclaiming is never triggered 368 * when allocating new pages. 369 */ 370 static inline 371 struct page *erofs_grab_cache_page_nowait(struct address_space *mapping, 372 pgoff_t index) 373 { 374 return pagecache_get_page(mapping, index, 375 FGP_LOCK|FGP_CREAT|FGP_NOFS|FGP_NOWAIT, 376 readahead_gfp_mask(mapping) & ~__GFP_RECLAIM); 377 } 378 379 extern const struct super_operations erofs_sops; 380 381 extern const struct address_space_operations erofs_raw_access_aops; 382 extern const struct address_space_operations z_erofs_aops; 383 384 /* 385 * Logical to physical block mapping 386 * 387 * Different with other file systems, it is used for 2 access modes: 388 * 389 * 1) RAW access mode: 390 * 391 * Users pass a valid (m_lblk, m_lofs -- usually 0) pair, 392 * and get the valid m_pblk, m_pofs and the longest m_len(in bytes). 393 * 394 * Note that m_lblk in the RAW access mode refers to the number of 395 * the compressed ondisk block rather than the uncompressed 396 * in-memory block for the compressed file. 397 * 398 * m_pofs equals to m_lofs except for the inline data page. 399 * 400 * 2) Normal access mode: 401 * 402 * If the inode is not compressed, it has no difference with 403 * the RAW access mode. However, if the inode is compressed, 404 * users should pass a valid (m_lblk, m_lofs) pair, and get 405 * the needed m_pblk, m_pofs, m_len to get the compressed data 406 * and the updated m_lblk, m_lofs which indicates the start 407 * of the corresponding uncompressed data in the file. 408 */ 409 enum { 410 BH_Encoded = BH_PrivateStart, 411 BH_FullMapped, 412 }; 413 414 /* Has a disk mapping */ 415 #define EROFS_MAP_MAPPED (1 << BH_Mapped) 416 /* Located in metadata (could be copied from bd_inode) */ 417 #define EROFS_MAP_META (1 << BH_Meta) 418 /* The extent is encoded */ 419 #define EROFS_MAP_ENCODED (1 << BH_Encoded) 420 /* The length of extent is full */ 421 #define EROFS_MAP_FULL_MAPPED (1 << BH_FullMapped) 422 423 struct erofs_map_blocks { 424 struct erofs_buf buf; 425 426 erofs_off_t m_pa, m_la; 427 u64 m_plen, m_llen; 428 429 unsigned short m_deviceid; 430 char m_algorithmformat; 431 unsigned int m_flags; 432 }; 433 434 /* Flags used by erofs_map_blocks_flatmode() */ 435 #define EROFS_GET_BLOCKS_RAW 0x0001 436 /* 437 * Used to get the exact decompressed length, e.g. fiemap (consider lookback 438 * approach instead if possible since it's more metadata lightweight.) 439 */ 440 #define EROFS_GET_BLOCKS_FIEMAP 0x0002 441 /* Used to map the whole extent if non-negligible data is requested for LZMA */ 442 #define EROFS_GET_BLOCKS_READMORE 0x0004 443 /* Used to map tail extent for tailpacking inline pcluster */ 444 #define EROFS_GET_BLOCKS_FINDTAIL 0x0008 445 446 enum { 447 Z_EROFS_COMPRESSION_SHIFTED = Z_EROFS_COMPRESSION_MAX, 448 Z_EROFS_COMPRESSION_RUNTIME_MAX 449 }; 450 451 /* zmap.c */ 452 extern const struct iomap_ops z_erofs_iomap_report_ops; 453 454 #ifdef CONFIG_EROFS_FS_ZIP 455 int z_erofs_fill_inode(struct inode *inode); 456 int z_erofs_map_blocks_iter(struct inode *inode, 457 struct erofs_map_blocks *map, 458 int flags); 459 #else 460 static inline int z_erofs_fill_inode(struct inode *inode) { return -EOPNOTSUPP; } 461 static inline int z_erofs_map_blocks_iter(struct inode *inode, 462 struct erofs_map_blocks *map, 463 int flags) 464 { 465 return -EOPNOTSUPP; 466 } 467 #endif /* !CONFIG_EROFS_FS_ZIP */ 468 469 struct erofs_map_dev { 470 struct block_device *m_bdev; 471 struct dax_device *m_daxdev; 472 u64 m_dax_part_off; 473 474 erofs_off_t m_pa; 475 unsigned int m_deviceid; 476 }; 477 478 /* data.c */ 479 extern const struct file_operations erofs_file_fops; 480 void erofs_unmap_metabuf(struct erofs_buf *buf); 481 void erofs_put_metabuf(struct erofs_buf *buf); 482 void *erofs_read_metabuf(struct erofs_buf *buf, struct super_block *sb, 483 erofs_blk_t blkaddr, enum erofs_kmap_type type); 484 int erofs_map_dev(struct super_block *sb, struct erofs_map_dev *dev); 485 int erofs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, 486 u64 start, u64 len); 487 488 /* inode.c */ 489 static inline unsigned long erofs_inode_hash(erofs_nid_t nid) 490 { 491 #if BITS_PER_LONG == 32 492 return (nid >> 32) ^ (nid & 0xffffffff); 493 #else 494 return nid; 495 #endif 496 } 497 498 extern const struct inode_operations erofs_generic_iops; 499 extern const struct inode_operations erofs_symlink_iops; 500 extern const struct inode_operations erofs_fast_symlink_iops; 501 502 struct inode *erofs_iget(struct super_block *sb, erofs_nid_t nid, bool dir); 503 int erofs_getattr(struct user_namespace *mnt_userns, const struct path *path, 504 struct kstat *stat, u32 request_mask, 505 unsigned int query_flags); 506 507 /* namei.c */ 508 extern const struct inode_operations erofs_dir_iops; 509 510 int erofs_namei(struct inode *dir, struct qstr *name, 511 erofs_nid_t *nid, unsigned int *d_type); 512 513 /* dir.c */ 514 extern const struct file_operations erofs_dir_fops; 515 516 static inline void *erofs_vm_map_ram(struct page **pages, unsigned int count) 517 { 518 int retried = 0; 519 520 while (1) { 521 void *p = vm_map_ram(pages, count, -1); 522 523 /* retry two more times (totally 3 times) */ 524 if (p || ++retried >= 3) 525 return p; 526 vm_unmap_aliases(); 527 } 528 return NULL; 529 } 530 531 /* pcpubuf.c */ 532 void *erofs_get_pcpubuf(unsigned int requiredpages); 533 void erofs_put_pcpubuf(void *ptr); 534 int erofs_pcpubuf_growsize(unsigned int nrpages); 535 void erofs_pcpubuf_init(void); 536 void erofs_pcpubuf_exit(void); 537 538 /* sysfs.c */ 539 int erofs_register_sysfs(struct super_block *sb); 540 void erofs_unregister_sysfs(struct super_block *sb); 541 int __init erofs_init_sysfs(void); 542 void erofs_exit_sysfs(void); 543 544 /* utils.c / zdata.c */ 545 struct page *erofs_allocpage(struct page **pagepool, gfp_t gfp); 546 static inline void erofs_pagepool_add(struct page **pagepool, 547 struct page *page) 548 { 549 set_page_private(page, (unsigned long)*pagepool); 550 *pagepool = page; 551 } 552 void erofs_release_pages(struct page **pagepool); 553 554 #ifdef CONFIG_EROFS_FS_ZIP 555 int erofs_workgroup_put(struct erofs_workgroup *grp); 556 struct erofs_workgroup *erofs_find_workgroup(struct super_block *sb, 557 pgoff_t index); 558 struct erofs_workgroup *erofs_insert_workgroup(struct super_block *sb, 559 struct erofs_workgroup *grp); 560 void erofs_workgroup_free_rcu(struct erofs_workgroup *grp); 561 void erofs_shrinker_register(struct super_block *sb); 562 void erofs_shrinker_unregister(struct super_block *sb); 563 int __init erofs_init_shrinker(void); 564 void erofs_exit_shrinker(void); 565 int __init z_erofs_init_zip_subsystem(void); 566 void z_erofs_exit_zip_subsystem(void); 567 int erofs_try_to_free_all_cached_pages(struct erofs_sb_info *sbi, 568 struct erofs_workgroup *egrp); 569 int erofs_try_to_free_cached_page(struct page *page); 570 int z_erofs_load_lz4_config(struct super_block *sb, 571 struct erofs_super_block *dsb, 572 struct z_erofs_lz4_cfgs *lz4, int len); 573 #else 574 static inline void erofs_shrinker_register(struct super_block *sb) {} 575 static inline void erofs_shrinker_unregister(struct super_block *sb) {} 576 static inline int erofs_init_shrinker(void) { return 0; } 577 static inline void erofs_exit_shrinker(void) {} 578 static inline int z_erofs_init_zip_subsystem(void) { return 0; } 579 static inline void z_erofs_exit_zip_subsystem(void) {} 580 static inline int z_erofs_load_lz4_config(struct super_block *sb, 581 struct erofs_super_block *dsb, 582 struct z_erofs_lz4_cfgs *lz4, int len) 583 { 584 if (lz4 || dsb->u1.lz4_max_distance) { 585 erofs_err(sb, "lz4 algorithm isn't enabled"); 586 return -EINVAL; 587 } 588 return 0; 589 } 590 #endif /* !CONFIG_EROFS_FS_ZIP */ 591 592 #ifdef CONFIG_EROFS_FS_ZIP_LZMA 593 int z_erofs_lzma_init(void); 594 void z_erofs_lzma_exit(void); 595 int z_erofs_load_lzma_config(struct super_block *sb, 596 struct erofs_super_block *dsb, 597 struct z_erofs_lzma_cfgs *lzma, int size); 598 #else 599 static inline int z_erofs_lzma_init(void) { return 0; } 600 static inline int z_erofs_lzma_exit(void) { return 0; } 601 static inline int z_erofs_load_lzma_config(struct super_block *sb, 602 struct erofs_super_block *dsb, 603 struct z_erofs_lzma_cfgs *lzma, int size) { 604 if (lzma) { 605 erofs_err(sb, "lzma algorithm isn't enabled"); 606 return -EINVAL; 607 } 608 return 0; 609 } 610 #endif /* !CONFIG_EROFS_FS_ZIP */ 611 612 #define EFSCORRUPTED EUCLEAN /* Filesystem is corrupted */ 613 614 #endif /* __EROFS_INTERNAL_H */ 615