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 #ifdef CONFIG_EROFS_FAULT_INJECTION
  46 enum {
  47         FAULT_KMALLOC,
  48         FAULT_MAX,
  49 };
  50
  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 #ifdef EROFS_FS_HAS_MANAGED_CACHE
  99         struct inode *managed_cache;
 100 #endif
 101
 102 #endif
 103
 104         u32 build_time_nsec;
 105         u64 build_time;
 106
 107         /* what we really care is nid, rather than ino.. */
 108         erofs_nid_t root_nid;
 109         /* used for statfs, f_files - f_favail */
 110         u64 inos;
 111
 112         u8 uuid[16];                    /* 128-bit uuid for volume */
 113         u8 volume_name[16];             /* volume name */
 114         u32 requirements;
 115
 116         char *dev_name;
 117
 118         unsigned int mount_opt;
 119         unsigned int shrinker_run_no;
 120
 121 #ifdef CONFIG_EROFS_FAULT_INJECTION
 122         struct erofs_fault_info fault_info;     /* For fault injection */
 123 #endif
 124 };
 125
 126 #ifdef CONFIG_EROFS_FAULT_INJECTION
 127 #define erofs_show_injection_info(type)                                 \
 128         infoln("inject %s in %s of %pS", erofs_fault_name[type],        \
 129                 __func__, __builtin_return_address(0))
 130
 131 static inline bool time_to_inject(struct erofs_sb_info *sbi, int type)
 132 {
 133         struct erofs_fault_info *ffi = &sbi->fault_info;
 134
 135         if (!ffi->inject_rate)
 136                 return false;
 137
 138         if (!IS_FAULT_SET(ffi, type))
 139                 return false;
 140
 141         atomic_inc(&ffi->inject_ops);
 142         if (atomic_read(&ffi->inject_ops) >= ffi->inject_rate) {
 143                 atomic_set(&ffi->inject_ops, 0);
 144                 return true;
 145         }
 146         return false;
 147 }
 148 #endif
 149
 150 static inline void *erofs_kmalloc(struct erofs_sb_info *sbi,
 151                                         size_t size, gfp_t flags)
 152 {
 153 #ifdef CONFIG_EROFS_FAULT_INJECTION
 154         if (time_to_inject(sbi, FAULT_KMALLOC)) {
 155                 erofs_show_injection_info(FAULT_KMALLOC);
 156                 return NULL;
 157         }
 158 #endif
 159         return kmalloc(size, flags);
 160 }
 161
 162 #define EROFS_SB(sb) ((struct erofs_sb_info *)(sb)->s_fs_info)
 163 #define EROFS_I_SB(inode) ((struct erofs_sb_info *)(inode)->i_sb->s_fs_info)
 164
 165 /* Mount flags set via mount options or defaults */
 166 #define EROFS_MOUNT_XATTR_USER          0x00000010
 167 #define EROFS_MOUNT_POSIX_ACL           0x00000020
 168 #define EROFS_MOUNT_FAULT_INJECTION     0x00000040
 169
 170 #define clear_opt(sbi, option)  ((sbi)->mount_opt &= ~EROFS_MOUNT_##option)
 171 #define set_opt(sbi, option)    ((sbi)->mount_opt |= EROFS_MOUNT_##option)
 172 #define test_opt(sbi, option)   ((sbi)->mount_opt & EROFS_MOUNT_##option)
 173
 174 #ifdef CONFIG_EROFS_FS_ZIP
 175 #define erofs_workstn_lock(sbi)         xa_lock(&(sbi)->workstn_tree)
 176 #define erofs_workstn_unlock(sbi)       xa_unlock(&(sbi)->workstn_tree)
 177
 178 /* basic unit of the workstation of a super_block */
 179 struct erofs_workgroup {
 180         /* the workgroup index in the workstation */
 181         pgoff_t index;
 182
 183         /* overall workgroup reference count */
 184         atomic_t refcount;
 185 };
 186
 187 #define EROFS_LOCKED_MAGIC     (INT_MIN | 0xE0F510CCL)
 188
 189 #if defined(CONFIG_SMP)
 190 static inline bool erofs_workgroup_try_to_freeze(struct erofs_workgroup *grp,
 191                                                  int val)
 192 {
 193         preempt_disable();
 194         if (val != atomic_cmpxchg(&grp->refcount, val, EROFS_LOCKED_MAGIC)) {
 195                 preempt_enable();
 196                 return false;
 197         }
 198         return true;
 199 }
 200
 201 static inline void erofs_workgroup_unfreeze(struct erofs_workgroup *grp,
 202                                             int orig_val)
 203 {
 204         /*
 205          * other observers should notice all modifications
 206          * in the freezing period.
 207          */
 208         smp_mb();
 209         atomic_set(&grp->refcount, orig_val);
 210         preempt_enable();
 211 }
 212
 213 static inline int erofs_wait_on_workgroup_freezed(struct erofs_workgroup *grp)
 214 {
 215         return atomic_cond_read_relaxed(&grp->refcount,
 216                                         VAL != EROFS_LOCKED_MAGIC);
 217 }
 218 #else
 219 static inline bool erofs_workgroup_try_to_freeze(struct erofs_workgroup *grp,
 220                                                  int val)
 221 {
 222         preempt_disable();
 223         /* no need to spin on UP platforms, let's just disable preemption. */
 224         if (val != atomic_read(&grp->refcount)) {
 225                 preempt_enable();
 226                 return false;
 227         }
 228         return true;
 229 }
 230
 231 static inline void erofs_workgroup_unfreeze(struct erofs_workgroup *grp,
 232                                             int orig_val)
 233 {
 234         preempt_enable();
 235 }
 236
 237 static inline int erofs_wait_on_workgroup_freezed(struct erofs_workgroup *grp)
 238 {
 239         int v = atomic_read(&grp->refcount);
 240
 241         /* workgroup is never freezed on uniprocessor systems */
 242         DBG_BUGON(v == EROFS_LOCKED_MAGIC);
 243         return v;
 244 }
 245 #endif
 246
 247 static inline bool erofs_workgroup_get(struct erofs_workgroup *grp, int *ocnt)
 248 {
 249         int o;
 250
 251 repeat:
 252         o = erofs_wait_on_workgroup_freezed(grp);
 253
 254         if (unlikely(o <= 0))
 255                 return -1;
 256
 257         if (unlikely(atomic_cmpxchg(&grp->refcount, o, o + 1) != o))
 258                 goto repeat;
 259
 260         *ocnt = o;
 261         return 0;
 262 }
 263
 264 #define __erofs_workgroup_get(grp)      atomic_inc(&(grp)->refcount)
 265 #define __erofs_workgroup_put(grp)      atomic_dec(&(grp)->refcount)
 266
 267 extern int erofs_workgroup_put(struct erofs_workgroup *grp);
 268
 269 extern struct erofs_workgroup *erofs_find_workgroup(
 270         struct super_block *sb, pgoff_t index, bool *tag);
 271
 272 extern int erofs_register_workgroup(struct super_block *sb,
 273         struct erofs_workgroup *grp, bool tag);
 274
 275 extern unsigned long erofs_shrink_workstation(struct erofs_sb_info *sbi,
 276         unsigned long nr_shrink, bool cleanup);
 277
 278 static inline void erofs_workstation_cleanup_all(struct super_block *sb)
 279 {
 280         erofs_shrink_workstation(EROFS_SB(sb), ~0UL, true);
 281 }
 282
 283 #ifdef EROFS_FS_HAS_MANAGED_CACHE
 284 #define EROFS_UNALLOCATED_CACHED_PAGE   ((void *)0x5F0EF00D)
 285
 286 extern int erofs_try_to_free_all_cached_pages(struct erofs_sb_info *sbi,
 287         struct erofs_workgroup *egrp);
 288 extern int erofs_try_to_free_cached_page(struct address_space *mapping,
 289         struct page *page);
 290 #endif
 291
 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 /* atomic flag definitions */
 334 #define EROFS_V_EA_INITED_BIT   0
 335
 336 /* bitlock definitions (arranged in reverse order) */
 337 #define EROFS_V_BL_XATTR_BIT    (BITS_PER_LONG - 1)
 338
 339 struct erofs_vnode {
 340         erofs_nid_t nid;
 341
 342         /* atomic flags (including bitlocks) */
 343         unsigned long flags;
 344
 345         unsigned char data_mapping_mode;
 346         /* inline size in bytes */
 347         unsigned char inode_isize;
 348         unsigned short xattr_isize;
 349
 350         unsigned xattr_shared_count;
 351         unsigned *xattr_shared_xattrs;
 352
 353         erofs_blk_t raw_blkaddr;
 354
 355         /* the corresponding vfs inode */
 356         struct inode vfs_inode;
 357 };
 358
 359 #define EROFS_V(ptr)    \
 360         container_of(ptr, struct erofs_vnode, vfs_inode)
 361
 362 #define __inode_advise(x, bit, bits) \
 363         (((x) >> (bit)) & ((1 << (bits)) - 1))
 364
 365 #define __inode_version(advise) \
 366         __inode_advise(advise, EROFS_I_VERSION_BIT,     \
 367                 EROFS_I_VERSION_BITS)
 368
 369 #define __inode_data_mapping(advise)    \
 370         __inode_advise(advise, EROFS_I_DATA_MAPPING_BIT,\
 371                 EROFS_I_DATA_MAPPING_BITS)
 372
 373 static inline unsigned long inode_datablocks(struct inode *inode)
 374 {
 375         /* since i_size cannot be changed */
 376         return DIV_ROUND_UP(inode->i_size, EROFS_BLKSIZ);
 377 }
 378
 379 static inline bool is_inode_layout_plain(struct inode *inode)
 380 {
 381         return EROFS_V(inode)->data_mapping_mode == EROFS_INODE_LAYOUT_PLAIN;
 382 }
 383
 384 static inline bool is_inode_layout_compression(struct inode *inode)
 385 {
 386         return EROFS_V(inode)->data_mapping_mode ==
 387                                         EROFS_INODE_LAYOUT_COMPRESSION;
 388 }
 389
 390 static inline bool is_inode_layout_inline(struct inode *inode)
 391 {
 392         return EROFS_V(inode)->data_mapping_mode == EROFS_INODE_LAYOUT_INLINE;
 393 }
 394
 395 extern const struct super_operations erofs_sops;
 396 extern const struct inode_operations erofs_dir_iops;
 397 extern const struct file_operations erofs_dir_fops;
 398
 399 extern const struct address_space_operations erofs_raw_access_aops;
 400 #ifdef CONFIG_EROFS_FS_ZIP
 401 extern const struct address_space_operations z_erofs_vle_normalaccess_aops;
 402 #endif
 403
 404 /*
 405  * Logical to physical block mapping, used by erofs_map_blocks()
 406  *
 407  * Different with other file systems, it is used for 2 access modes:
 408  *
 409  * 1) RAW access mode:
 410  *
 411  * Users pass a valid (m_lblk, m_lofs -- usually 0) pair,
 412  * and get the valid m_pblk, m_pofs and the longest m_len(in bytes).
 413  *
 414  * Note that m_lblk in the RAW access mode refers to the number of
 415  * the compressed ondisk block rather than the uncompressed
 416  * in-memory block for the compressed file.
 417  *
 418  * m_pofs equals to m_lofs except for the inline data page.
 419  *
 420  * 2) Normal access mode:
 421  *
 422  * If the inode is not compressed, it has no difference with
 423  * the RAW access mode. However, if the inode is compressed,
 424  * users should pass a valid (m_lblk, m_lofs) pair, and get
 425  * the needed m_pblk, m_pofs, m_len to get the compressed data
 426  * and the updated m_lblk, m_lofs which indicates the start
 427  * of the corresponding uncompressed data in the file.
 428  */
 429 enum {
 430         BH_Zipped = BH_PrivateStart,
 431 };
 432
 433 /* Has a disk mapping */
 434 #define EROFS_MAP_MAPPED        (1 << BH_Mapped)
 435 /* Located in metadata (could be copied from bd_inode) */
 436 #define EROFS_MAP_META          (1 << BH_Meta)
 437 /* The extent has been compressed */
 438 #define EROFS_MAP_ZIPPED        (1 << BH_Zipped)
 439
 440 struct erofs_map_blocks {
 441         erofs_off_t m_pa, m_la;
 442         u64 m_plen, m_llen;
 443
 444         unsigned int m_flags;
 445 };
 446
 447 /* Flags used by erofs_map_blocks() */
 448 #define EROFS_GET_BLOCKS_RAW    0x0001
 449
 450 /* data.c */
 451 static inline struct bio *prepare_bio(
 452         struct super_block *sb,
 453         erofs_blk_t blkaddr, unsigned nr_pages,
 454         bio_end_io_t endio)
 455 {
 456         gfp_t gfp = GFP_NOIO;
 457         struct bio *bio = bio_alloc(gfp, nr_pages);
 458
 459         if (unlikely(bio == NULL) &&
 460                 (current->flags & PF_MEMALLOC)) {
 461                 do {
 462                         nr_pages /= 2;
 463                         if (unlikely(!nr_pages)) {
 464                                 bio = bio_alloc(gfp | __GFP_NOFAIL, 1);
 465                                 BUG_ON(bio == NULL);
 466                                 break;
 467                         }
 468                         bio = bio_alloc(gfp, nr_pages);
 469                 } while (bio == NULL);
 470         }
 471
 472         bio->bi_end_io = endio;
 473         bio_set_dev(bio, sb->s_bdev);
 474         bio->bi_iter.bi_sector = blkaddr << LOG_SECTORS_PER_BLOCK;
 475         return bio;
 476 }
 477
 478 static inline void __submit_bio(struct bio *bio, unsigned op, unsigned op_flags)
 479 {
 480         bio_set_op_attrs(bio, op, op_flags);
 481         submit_bio(bio);
 482 }
 483
 484 extern struct page *erofs_get_meta_page(struct super_block *sb,
 485         erofs_blk_t blkaddr, bool prio);
 486 extern int erofs_map_blocks(struct inode *, struct erofs_map_blocks *, int);
 487 extern int erofs_map_blocks_iter(struct inode *, struct erofs_map_blocks *,
 488         struct page **, int);
 489
 490 struct erofs_map_blocks_iter {
 491         struct erofs_map_blocks map;
 492         struct page *mpage;
 493 };
 494
 495
 496 static inline struct page *
 497 erofs_get_inline_page(struct inode *inode,
 498                       erofs_blk_t blkaddr)
 499 {
 500         return erofs_get_meta_page(inode->i_sb,
 501                 blkaddr, S_ISDIR(inode->i_mode));
 502 }
 503
 504 /* inode.c */
 505 extern struct inode *erofs_iget(struct super_block *sb,
 506         erofs_nid_t nid, bool dir);
 507
 508 /* dir.c */
 509 int erofs_namei(struct inode *dir, struct qstr *name,
 510         erofs_nid_t *nid, unsigned *d_type);
 511
 512 /* xattr.c */
 513 #ifdef CONFIG_EROFS_FS_XATTR
 514 extern const struct xattr_handler *erofs_xattr_handlers[];
 515 #endif
 516
 517 /* symlink */
 518 #ifdef CONFIG_EROFS_FS_XATTR
 519 extern const struct inode_operations erofs_symlink_xattr_iops;
 520 extern const struct inode_operations erofs_fast_symlink_xattr_iops;
 521 extern const struct inode_operations erofs_special_inode_operations;
 522 #endif
 523
 524 static inline void set_inode_fast_symlink(struct inode *inode)
 525 {
 526 #ifdef CONFIG_EROFS_FS_XATTR
 527         inode->i_op = &erofs_fast_symlink_xattr_iops;
 528 #else
 529         inode->i_op = &simple_symlink_inode_operations;
 530 #endif
 531 }
 532
 533 static inline bool is_inode_fast_symlink(struct inode *inode)
 534 {
 535 #ifdef CONFIG_EROFS_FS_XATTR
 536         return inode->i_op == &erofs_fast_symlink_xattr_iops;
 537 #else
 538         return inode->i_op == &simple_symlink_inode_operations;
 539 #endif
 540 }
 541
 542 static inline void *erofs_vmap(struct page **pages, unsigned int count)
 543 {
 544 #ifdef CONFIG_EROFS_FS_USE_VM_MAP_RAM
 545         int i = 0;
 546
 547         while (1) {
 548                 void *addr = vm_map_ram(pages, count, -1, PAGE_KERNEL);
 549                 /* retry two more times (totally 3 times) */
 550                 if (addr != NULL || ++i >= 3)
 551                         return addr;
 552                 vm_unmap_aliases();
 553         }
 554         return NULL;
 555 #else
 556         return vmap(pages, count, VM_MAP, PAGE_KERNEL);
 557 #endif
 558 }
 559
 560 static inline void erofs_vunmap(const void *mem, unsigned int count)
 561 {
 562 #ifdef CONFIG_EROFS_FS_USE_VM_MAP_RAM
 563         vm_unmap_ram(mem, count);
 564 #else
 565         vunmap(mem);
 566 #endif
 567 }
 568
 569 /* utils.c */
 570 extern struct page *erofs_allocpage(struct list_head *pool, gfp_t gfp);
 571
 572 extern void erofs_register_super(struct super_block *sb);
 573 extern void erofs_unregister_super(struct super_block *sb);
 574
 575 extern unsigned long erofs_shrink_count(struct shrinker *shrink,
 576         struct shrink_control *sc);
 577 extern unsigned long erofs_shrink_scan(struct shrinker *shrink,
 578         struct shrink_control *sc);
 579
 580 #ifndef lru_to_page
 581 #define lru_to_page(head) (list_entry((head)->prev, struct page, lru))
 582 #endif
 583
 584 #endif
 585