drivers/staging/erofs/internal.h

   1 /* SPDX-License-Identifier: GPL-2.0
   2  *
   3  * linux/drivers/staging/erofs/internal.h
   4  *
   5  * Copyright (C) 2017-2018 HUAWEI, Inc.
   6  *             http://www.huawei.com/
   7  * Created by Gao Xiang <gaoxiang25@huawei.com>
   8  *
   9  * This file is subject to the terms and conditions of the GNU General Public
  10  * License.  See the file COPYING in the main directory of the Linux
  11  * distribution for more details.
  12  */
  13 #ifndef __INTERNAL_H
  14 #define __INTERNAL_H
  15
  16 #include <linux/fs.h>
  17 #include <linux/dcache.h>
  18 #include <linux/mm.h>
  19 #include <linux/pagemap.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/cleancache.h>
  23 #include <linux/slab.h>
  24 #include <linux/vmalloc.h>
  25 #include "erofs_fs.h"
  26
  27 /* redefine pr_fmt "erofs: " */
  28 #undef pr_fmt
  29 #define pr_fmt(fmt) "erofs: " fmt
  30
  31 #define errln(x, ...)   pr_err(x "\n", ##__VA_ARGS__)
  32 #define infoln(x, ...)  pr_info(x "\n", ##__VA_ARGS__)
  33 #ifdef CONFIG_EROFS_FS_DEBUG
  34 #define debugln(x, ...) pr_debug(x "\n", ##__VA_ARGS__)
  35
  36 #define dbg_might_sleep         might_sleep
  37 #define DBG_BUGON               BUG_ON
  38 #else
  39 #define debugln(x, ...)         ((void)0)
  40
  41 #define dbg_might_sleep()       ((void)0)
  42 #define DBG_BUGON(x)            ((void)(x))
  43 #endif
  44
  45 enum {
  46         FAULT_KMALLOC,
  47         FAULT_MAX,
  48 };
  49
  50 #ifdef CONFIG_EROFS_FAULT_INJECTION
  51 extern char *erofs_fault_name[FAULT_MAX];
  52 #define IS_FAULT_SET(fi, type) ((fi)->inject_type & (1 << (type)))
  53
  54 struct erofs_fault_info {
  55         atomic_t inject_ops;
  56         unsigned int inject_rate;
  57         unsigned int inject_type;
  58 };
  59 #endif
  60
  61 #ifdef CONFIG_EROFS_FS_ZIP_CACHE_BIPOLAR
  62 #define EROFS_FS_ZIP_CACHE_LVL  (2)
  63 #elif defined(EROFS_FS_ZIP_CACHE_UNIPOLAR)
  64 #define EROFS_FS_ZIP_CACHE_LVL  (1)
  65 #else
  66 #define EROFS_FS_ZIP_CACHE_LVL  (0)
  67 #endif
  68
  69 #if (!defined(EROFS_FS_HAS_MANAGED_CACHE) && (EROFS_FS_ZIP_CACHE_LVL > 0))
  70 #define EROFS_FS_HAS_MANAGED_CACHE
  71 #endif
  72
  73 /* EROFS_SUPER_MAGIC_V1 to represent the whole file system */
  74 #define EROFS_SUPER_MAGIC   EROFS_SUPER_MAGIC_V1
  75
  76 typedef u64 erofs_nid_t;
  77
  78 struct erofs_sb_info {
  79         /* list for all registered superblocks, mainly for shrinker */
  80         struct list_head list;
  81         struct mutex umount_mutex;
  82
  83         u32 blocks;
  84         u32 meta_blkaddr;
  85 #ifdef CONFIG_EROFS_FS_XATTR
  86         u32 xattr_blkaddr;
  87 #endif
  88
  89         /* inode slot unit size in bit shift */
  90         unsigned char islotbits;
  91 #ifdef CONFIG_EROFS_FS_ZIP
  92         /* cluster size in bit shift */
  93         unsigned char clusterbits;
  94
  95         /* the dedicated workstation for compression */
  96         struct radix_tree_root workstn_tree;
  97
  98         /* threshold for decompression synchronously */
  99         unsigned int max_sync_decompress_pages;
 100
 101 #ifdef EROFS_FS_HAS_MANAGED_CACHE
 102         struct inode *managed_cache;
 103 #endif
 104
 105 #endif
 106
 107         u32 build_time_nsec;
 108         u64 build_time;
 109
 110         /* what we really care is nid, rather than ino.. */
 111         erofs_nid_t root_nid;
 112         /* used for statfs, f_files - f_favail */
 113         u64 inos;
 114
 115         u8 uuid[16];                    /* 128-bit uuid for volume */
 116         u8 volume_name[16];             /* volume name */
 117         char *dev_name;
 118
 119         unsigned int mount_opt;
 120         unsigned int shrinker_run_no;
 121
 122 #ifdef CONFIG_EROFS_FAULT_INJECTION
 123         struct erofs_fault_info fault_info;     /* For fault injection */
 124 #endif
 125 };
 126
 127 #ifdef CONFIG_EROFS_FAULT_INJECTION
 128 #define erofs_show_injection_info(type)                                 \
 129         infoln("inject %s in %s of %pS", erofs_fault_name[type],        \
 130                 __func__, __builtin_return_address(0))
 131
 132 static inline bool time_to_inject(struct erofs_sb_info *sbi, int type)
 133 {
 134         struct erofs_fault_info *ffi = &sbi->fault_info;
 135
 136         if (!ffi->inject_rate)
 137                 return false;
 138
 139         if (!IS_FAULT_SET(ffi, type))
 140                 return false;
 141
 142         atomic_inc(&ffi->inject_ops);
 143         if (atomic_read(&ffi->inject_ops) >= ffi->inject_rate) {
 144                 atomic_set(&ffi->inject_ops, 0);
 145                 return true;
 146         }
 147         return false;
 148 }
 149 #else
 150 static inline bool time_to_inject(struct erofs_sb_info *sbi, int type)
 151 {
 152         return false;
 153 }
 154
 155 static inline void erofs_show_injection_info(int type)
 156 {
 157 }
 158 #endif
 159
 160 static inline void *erofs_kmalloc(struct erofs_sb_info *sbi,
 161                                         size_t size, gfp_t flags)
 162 {
 163         if (time_to_inject(sbi, FAULT_KMALLOC)) {
 164                 erofs_show_injection_info(FAULT_KMALLOC);
 165                 return NULL;
 166         }
 167         return kmalloc(size, flags);
 168 }
 169
 170 #define EROFS_SB(sb) ((struct erofs_sb_info *)(sb)->s_fs_info)
 171 #define EROFS_I_SB(inode) ((struct erofs_sb_info *)(inode)->i_sb->s_fs_info)
 172
 173 /* Mount flags set via mount options or defaults */
 174 #define EROFS_MOUNT_XATTR_USER          0x00000010
 175 #define EROFS_MOUNT_POSIX_ACL           0x00000020
 176 #define EROFS_MOUNT_FAULT_INJECTION     0x00000040
 177
 178 #define clear_opt(sbi, option)  ((sbi)->mount_opt &= ~EROFS_MOUNT_##option)
 179 #define set_opt(sbi, option)    ((sbi)->mount_opt |= EROFS_MOUNT_##option)
 180 #define test_opt(sbi, option)   ((sbi)->mount_opt & EROFS_MOUNT_##option)
 181
 182 #ifdef CONFIG_EROFS_FS_ZIP
 183 #define erofs_workstn_lock(sbi)         xa_lock(&(sbi)->workstn_tree)
 184 #define erofs_workstn_unlock(sbi)       xa_unlock(&(sbi)->workstn_tree)
 185
 186 /* basic unit of the workstation of a super_block */
 187 struct erofs_workgroup {
 188         /* the workgroup index in the workstation */
 189         pgoff_t index;
 190
 191         /* overall workgroup reference count */
 192         atomic_t refcount;
 193 };
 194
 195 #define EROFS_LOCKED_MAGIC     (INT_MIN | 0xE0F510CCL)
 196
 197 #if defined(CONFIG_SMP)
 198 static inline bool erofs_workgroup_try_to_freeze(struct erofs_workgroup *grp,
 199                                                  int val)
 200 {
 201         preempt_disable();
 202         if (val != atomic_cmpxchg(&grp->refcount, val, EROFS_LOCKED_MAGIC)) {
 203                 preempt_enable();
 204                 return false;
 205         }
 206         return true;
 207 }
 208
 209 static inline void erofs_workgroup_unfreeze(struct erofs_workgroup *grp,
 210                                             int orig_val)
 211 {
 212         /*
 213          * other observers should notice all modifications
 214          * in the freezing period.
 215          */
 216         smp_mb();
 217         atomic_set(&grp->refcount, orig_val);
 218         preempt_enable();
 219 }
 220
 221 static inline int erofs_wait_on_workgroup_freezed(struct erofs_workgroup *grp)
 222 {
 223         return atomic_cond_read_relaxed(&grp->refcount,
 224                                         VAL != EROFS_LOCKED_MAGIC);
 225 }
 226 #else
 227 static inline bool erofs_workgroup_try_to_freeze(struct erofs_workgroup *grp,
 228                                                  int val)
 229 {
 230         preempt_disable();
 231         /* no need to spin on UP platforms, let's just disable preemption. */
 232         if (val != atomic_read(&grp->refcount)) {
 233                 preempt_enable();
 234                 return false;
 235         }
 236         return true;
 237 }
 238
 239 static inline void erofs_workgroup_unfreeze(struct erofs_workgroup *grp,
 240                                             int orig_val)
 241 {
 242         preempt_enable();
 243 }
 244
 245 static inline int erofs_wait_on_workgroup_freezed(struct erofs_workgroup *grp)
 246 {
 247         int v = atomic_read(&grp->refcount);
 248
 249         /* workgroup is never freezed on uniprocessor systems */
 250         DBG_BUGON(v == EROFS_LOCKED_MAGIC);
 251         return v;
 252 }
 253 #endif
 254
 255 extern int erofs_workgroup_put(struct erofs_workgroup *grp);
 256 extern struct erofs_workgroup *erofs_find_workgroup(
 257         struct super_block *sb, pgoff_t index, bool *tag);
 258
 259 extern int erofs_register_workgroup(struct super_block *sb,
 260         struct erofs_workgroup *grp, bool tag);
 261
 262 extern unsigned long erofs_shrink_workstation(struct erofs_sb_info *sbi,
 263         unsigned long nr_shrink, bool cleanup);
 264
 265 extern void erofs_workgroup_free_rcu(struct erofs_workgroup *grp);
 266
 267 #ifdef EROFS_FS_HAS_MANAGED_CACHE
 268 extern int erofs_try_to_free_all_cached_pages(struct erofs_sb_info *sbi,
 269         struct erofs_workgroup *egrp);
 270 extern int erofs_try_to_free_cached_page(struct address_space *mapping,
 271         struct page *page);
 272
 273 #define MNGD_MAPPING(sbi)       ((sbi)->managed_cache->i_mapping)
 274 #else
 275 #define MNGD_MAPPING(sbi)       (NULL)
 276 #endif
 277
 278 #define DEFAULT_MAX_SYNC_DECOMPRESS_PAGES       3
 279
 280 static inline bool __should_decompress_synchronously(struct erofs_sb_info *sbi,
 281                                                      unsigned int nr)
 282 {
 283         return nr <= sbi->max_sync_decompress_pages;
 284 }
 285
 286 int __init z_erofs_init_zip_subsystem(void);
 287 void z_erofs_exit_zip_subsystem(void);
 288 #else
 289 /* dummy initializer/finalizer for the decompression subsystem */
 290 static inline int z_erofs_init_zip_subsystem(void) { return 0; }
 291 static inline void z_erofs_exit_zip_subsystem(void) {}
 292 #endif
 293
 294 /* we strictly follow PAGE_SIZE and no buffer head yet */
 295 #define LOG_BLOCK_SIZE          PAGE_SHIFT
 296
 297 #undef LOG_SECTORS_PER_BLOCK
 298 #define LOG_SECTORS_PER_BLOCK   (PAGE_SHIFT - 9)
 299
 300 #undef SECTORS_PER_BLOCK
 301 #define SECTORS_PER_BLOCK       (1 << SECTORS_PER_BLOCK)
 302
 303 #define EROFS_BLKSIZ            (1 << LOG_BLOCK_SIZE)
 304
 305 #if (EROFS_BLKSIZ % 4096 || !EROFS_BLKSIZ)
 306 #error erofs cannot be used in this platform
 307 #endif
 308
 309 #define ROOT_NID(sb)            ((sb)->root_nid)
 310
 311 #ifdef CONFIG_EROFS_FS_ZIP
 312 /* hard limit of pages per compressed cluster */
 313 #define Z_EROFS_CLUSTER_MAX_PAGES       (CONFIG_EROFS_FS_CLUSTER_PAGE_LIMIT)
 314
 315 /* page count of a compressed cluster */
 316 #define erofs_clusterpages(sbi)         ((1 << (sbi)->clusterbits) / PAGE_SIZE)
 317 #endif
 318
 319 typedef u64 erofs_off_t;
 320
 321 /* data type for filesystem-wide blocks number */
 322 typedef u32 erofs_blk_t;
 323
 324 #define erofs_blknr(addr)       ((addr) / EROFS_BLKSIZ)
 325 #define erofs_blkoff(addr)      ((addr) % EROFS_BLKSIZ)
 326 #define blknr_to_addr(nr)       ((erofs_off_t)(nr) * EROFS_BLKSIZ)
 327
 328 static inline erofs_off_t iloc(struct erofs_sb_info *sbi, erofs_nid_t nid)
 329 {
 330         return blknr_to_addr(sbi->meta_blkaddr) + (nid << sbi->islotbits);
 331 }
 332
 333 #define inode_set_inited_xattr(inode)   (EROFS_V(inode)->flags |= 1)
 334 #define inode_has_inited_xattr(inode)   (EROFS_V(inode)->flags & 1)
 335
 336 struct erofs_vnode {
 337         erofs_nid_t nid;
 338         unsigned int flags;
 339
 340         unsigned char data_mapping_mode;
 341         /* inline size in bytes */
 342         unsigned char inode_isize;
 343         unsigned short xattr_isize;
 344
 345         unsigned xattr_shared_count;
 346         unsigned *xattr_shared_xattrs;
 347
 348         erofs_blk_t raw_blkaddr;
 349
 350         /* the corresponding vfs inode */
 351         struct inode vfs_inode;
 352 };
 353
 354 #define EROFS_V(ptr)    \
 355         container_of(ptr, struct erofs_vnode, vfs_inode)
 356
 357 #define __inode_advise(x, bit, bits) \
 358         (((x) >> (bit)) & ((1 << (bits)) - 1))
 359
 360 #define __inode_version(advise) \
 361         __inode_advise(advise, EROFS_I_VERSION_BIT,     \
 362                 EROFS_I_VERSION_BITS)
 363
 364 #define __inode_data_mapping(advise)    \
 365         __inode_advise(advise, EROFS_I_DATA_MAPPING_BIT,\
 366                 EROFS_I_DATA_MAPPING_BITS)
 367
 368 static inline unsigned long inode_datablocks(struct inode *inode)
 369 {
 370         /* since i_size cannot be changed */
 371         return DIV_ROUND_UP(inode->i_size, EROFS_BLKSIZ);
 372 }
 373
 374 static inline bool is_inode_layout_plain(struct inode *inode)
 375 {
 376         return EROFS_V(inode)->data_mapping_mode == EROFS_INODE_LAYOUT_PLAIN;
 377 }
 378
 379 static inline bool is_inode_layout_compression(struct inode *inode)
 380 {
 381         return EROFS_V(inode)->data_mapping_mode ==
 382                                         EROFS_INODE_LAYOUT_COMPRESSION;
 383 }
 384
 385 static inline bool is_inode_layout_inline(struct inode *inode)
 386 {
 387         return EROFS_V(inode)->data_mapping_mode == EROFS_INODE_LAYOUT_INLINE;
 388 }
 389
 390 extern const struct super_operations erofs_sops;
 391
 392 extern const struct address_space_operations erofs_raw_access_aops;
 393 #ifdef CONFIG_EROFS_FS_ZIP
 394 extern const struct address_space_operations z_erofs_vle_normalaccess_aops;
 395 #endif
 396
 397 /*
 398  * Logical to physical block mapping, used by erofs_map_blocks()
 399  *
 400  * Different with other file systems, it is used for 2 access modes:
 401  *
 402  * 1) RAW access mode:
 403  *
 404  * Users pass a valid (m_lblk, m_lofs -- usually 0) pair,
 405  * and get the valid m_pblk, m_pofs and the longest m_len(in bytes).
 406  *
 407  * Note that m_lblk in the RAW access mode refers to the number of
 408  * the compressed ondisk block rather than the uncompressed
 409  * in-memory block for the compressed file.
 410  *
 411  * m_pofs equals to m_lofs except for the inline data page.
 412  *
 413  * 2) Normal access mode:
 414  *
 415  * If the inode is not compressed, it has no difference with
 416  * the RAW access mode. However, if the inode is compressed,
 417  * users should pass a valid (m_lblk, m_lofs) pair, and get
 418  * the needed m_pblk, m_pofs, m_len to get the compressed data
 419  * and the updated m_lblk, m_lofs which indicates the start
 420  * of the corresponding uncompressed data in the file.
 421  */
 422 enum {
 423         BH_Zipped = BH_PrivateStart,
 424 };
 425
 426 /* Has a disk mapping */
 427 #define EROFS_MAP_MAPPED        (1 << BH_Mapped)
 428 /* Located in metadata (could be copied from bd_inode) */
 429 #define EROFS_MAP_META          (1 << BH_Meta)
 430 /* The extent has been compressed */
 431 #define EROFS_MAP_ZIPPED        (1 << BH_Zipped)
 432
 433 struct erofs_map_blocks {
 434         erofs_off_t m_pa, m_la;
 435         u64 m_plen, m_llen;
 436
 437         unsigned int m_flags;
 438
 439         struct page *mpage;
 440 };
 441
 442 /* Flags used by erofs_map_blocks() */
 443 #define EROFS_GET_BLOCKS_RAW    0x0001
 444
 445 #ifdef CONFIG_EROFS_FS_ZIP
 446 int z_erofs_map_blocks_iter(struct inode *inode,
 447                             struct erofs_map_blocks *map,
 448                             int flags);
 449 #else
 450 static inline int z_erofs_map_blocks_iter(struct inode *inode,
 451                                           struct erofs_map_blocks *map,
 452                                           int flags)
 453 {
 454         return -ENOTSUPP;
 455 }
 456 #endif
 457
 458 /* data.c */
 459 static inline struct bio *
 460 erofs_grab_bio(struct super_block *sb,
 461                erofs_blk_t blkaddr, unsigned int nr_pages,
 462                bio_end_io_t endio, bool nofail)
 463 {
 464         const gfp_t gfp = GFP_NOIO;
 465         struct bio *bio;
 466
 467         do {
 468                 if (nr_pages == 1) {
 469                         bio = bio_alloc(gfp | (nofail ? __GFP_NOFAIL : 0), 1);
 470                         if (unlikely(bio == NULL)) {
 471                                 DBG_BUGON(nofail);
 472                                 return ERR_PTR(-ENOMEM);
 473                         }
 474                         break;
 475                 }
 476                 bio = bio_alloc(gfp, nr_pages);
 477                 nr_pages /= 2;
 478         } while (unlikely(bio == NULL));
 479
 480         bio->bi_end_io = endio;
 481         bio_set_dev(bio, sb->s_bdev);
 482         bio->bi_iter.bi_sector = (sector_t)blkaddr << LOG_SECTORS_PER_BLOCK;
 483         return bio;
 484 }
 485
 486 static inline void __submit_bio(struct bio *bio, unsigned op, unsigned op_flags)
 487 {
 488         bio_set_op_attrs(bio, op, op_flags);
 489         submit_bio(bio);
 490 }
 491
 492 #ifndef CONFIG_EROFS_FS_IO_MAX_RETRIES
 493 #define EROFS_IO_MAX_RETRIES_NOFAIL     0
 494 #else
 495 #define EROFS_IO_MAX_RETRIES_NOFAIL     CONFIG_EROFS_FS_IO_MAX_RETRIES
 496 #endif
 497
 498 extern struct page *__erofs_get_meta_page(struct super_block *sb,
 499         erofs_blk_t blkaddr, bool prio, bool nofail);
 500
 501 static inline struct page *erofs_get_meta_page(struct super_block *sb,
 502         erofs_blk_t blkaddr, bool prio)
 503 {
 504         return __erofs_get_meta_page(sb, blkaddr, prio, false);
 505 }
 506
 507 static inline struct page *erofs_get_meta_page_nofail(struct super_block *sb,
 508         erofs_blk_t blkaddr, bool prio)
 509 {
 510         return __erofs_get_meta_page(sb, blkaddr, prio, true);
 511 }
 512
 513 extern int erofs_map_blocks(struct inode *, struct erofs_map_blocks *, int);
 514
 515 static inline struct page *
 516 erofs_get_inline_page(struct inode *inode,
 517                       erofs_blk_t blkaddr)
 518 {
 519         return erofs_get_meta_page(inode->i_sb,
 520                 blkaddr, S_ISDIR(inode->i_mode));
 521 }
 522
 523 /* inode.c */
 524 static inline unsigned long erofs_inode_hash(erofs_nid_t nid)
 525 {
 526 #if BITS_PER_LONG == 32
 527         return (nid >> 32) ^ (nid & 0xffffffff);
 528 #else
 529         return nid;
 530 #endif
 531 }
 532
 533 extern struct inode *erofs_iget(struct super_block *sb,
 534         erofs_nid_t nid, bool dir);
 535
 536 extern const struct inode_operations erofs_generic_iops;
 537 extern const struct inode_operations erofs_symlink_iops;
 538 extern const struct inode_operations erofs_fast_symlink_iops;
 539
 540 static inline void set_inode_fast_symlink(struct inode *inode)
 541 {
 542         inode->i_op = &erofs_fast_symlink_iops;
 543 }
 544
 545 static inline bool is_inode_fast_symlink(struct inode *inode)
 546 {
 547         return inode->i_op == &erofs_fast_symlink_iops;
 548 }
 549
 550 /* namei.c */
 551 extern const struct inode_operations erofs_dir_iops;
 552
 553 int erofs_namei(struct inode *dir, struct qstr *name,
 554                 erofs_nid_t *nid, unsigned int *d_type);
 555
 556 /* dir.c */
 557 extern const struct file_operations erofs_dir_fops;
 558
 559 static inline void *erofs_vmap(struct page **pages, unsigned int count)
 560 {
 561 #ifdef CONFIG_EROFS_FS_USE_VM_MAP_RAM
 562         int i = 0;
 563
 564         while (1) {
 565                 void *addr = vm_map_ram(pages, count, -1, PAGE_KERNEL);
 566                 /* retry two more times (totally 3 times) */
 567                 if (addr != NULL || ++i >= 3)
 568                         return addr;
 569                 vm_unmap_aliases();
 570         }
 571         return NULL;
 572 #else
 573         return vmap(pages, count, VM_MAP, PAGE_KERNEL);
 574 #endif
 575 }
 576
 577 static inline void erofs_vunmap(const void *mem, unsigned int count)
 578 {
 579 #ifdef CONFIG_EROFS_FS_USE_VM_MAP_RAM
 580         vm_unmap_ram(mem, count);
 581 #else
 582         vunmap(mem);
 583 #endif
 584 }
 585
 586 /* utils.c */
 587 extern struct page *erofs_allocpage(struct list_head *pool, gfp_t gfp);
 588
 589 extern void erofs_register_super(struct super_block *sb);
 590 extern void erofs_unregister_super(struct super_block *sb);
 591
 592 extern struct shrinker erofs_shrinker_info;
 593
 594 #ifndef lru_to_page
 595 #define lru_to_page(head) (list_entry((head)->prev, struct page, lru))
 596 #endif
 597
 598 #endif
 599