xref: /openbmc/linux/fs/erofs/internal.h (revision ee8ec048)
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 
55 	u32 blocks;
56 	u32 mapped_blkaddr;
57 };
58 
59 struct erofs_mount_opts {
60 #ifdef CONFIG_EROFS_FS_ZIP
61 	/* current strategy of how to use managed cache */
62 	unsigned char cache_strategy;
63 	/* strategy of sync decompression (false - auto, true - force on) */
64 	bool readahead_sync_decompress;
65 
66 	/* threshold for decompression synchronously */
67 	unsigned int max_sync_decompress_pages;
68 #endif
69 	unsigned int mount_opt;
70 };
71 
72 struct erofs_dev_context {
73 	struct idr tree;
74 	struct rw_semaphore rwsem;
75 
76 	unsigned int extra_devices;
77 };
78 
79 struct erofs_fs_context {
80 	struct erofs_mount_opts opt;
81 	struct erofs_dev_context *devs;
82 };
83 
84 /* all filesystem-wide lz4 configurations */
85 struct erofs_sb_lz4_info {
86 	/* # of pages needed for EROFS lz4 rolling decompression */
87 	u16 max_distance_pages;
88 	/* maximum possible blocks for pclusters in the filesystem */
89 	u16 max_pclusterblks;
90 };
91 
92 struct erofs_sb_info {
93 	struct erofs_mount_opts opt;	/* options */
94 #ifdef CONFIG_EROFS_FS_ZIP
95 	/* list for all registered superblocks, mainly for shrinker */
96 	struct list_head list;
97 	struct mutex umount_mutex;
98 
99 	/* managed XArray arranged in physical block number */
100 	struct xarray managed_pslots;
101 
102 	unsigned int shrinker_run_no;
103 	u16 available_compr_algs;
104 
105 	/* pseudo inode to manage cached pages */
106 	struct inode *managed_cache;
107 
108 	struct erofs_sb_lz4_info lz4;
109 #endif	/* CONFIG_EROFS_FS_ZIP */
110 	struct erofs_dev_context *devs;
111 	struct dax_device *dax_dev;
112 	u64 total_blocks;
113 	u32 primarydevice_blocks;
114 
115 	u32 meta_blkaddr;
116 #ifdef CONFIG_EROFS_FS_XATTR
117 	u32 xattr_blkaddr;
118 #endif
119 	u16 device_id_mask;	/* valid bits of device id to be used */
120 
121 	/* inode slot unit size in bit shift */
122 	unsigned char islotbits;
123 
124 	u32 sb_size;			/* total superblock size */
125 	u32 build_time_nsec;
126 	u64 build_time;
127 
128 	/* what we really care is nid, rather than ino.. */
129 	erofs_nid_t root_nid;
130 	/* used for statfs, f_files - f_favail */
131 	u64 inos;
132 
133 	u8 uuid[16];                    /* 128-bit uuid for volume */
134 	u8 volume_name[16];             /* volume name */
135 	u32 feature_compat;
136 	u32 feature_incompat;
137 };
138 
139 #define EROFS_SB(sb) ((struct erofs_sb_info *)(sb)->s_fs_info)
140 #define EROFS_I_SB(inode) ((struct erofs_sb_info *)(inode)->i_sb->s_fs_info)
141 
142 /* Mount flags set via mount options or defaults */
143 #define EROFS_MOUNT_XATTR_USER		0x00000010
144 #define EROFS_MOUNT_POSIX_ACL		0x00000020
145 #define EROFS_MOUNT_DAX_ALWAYS		0x00000040
146 #define EROFS_MOUNT_DAX_NEVER		0x00000080
147 
148 #define clear_opt(opt, option)	((opt)->mount_opt &= ~EROFS_MOUNT_##option)
149 #define set_opt(opt, option)	((opt)->mount_opt |= EROFS_MOUNT_##option)
150 #define test_opt(opt, option)	((opt)->mount_opt & EROFS_MOUNT_##option)
151 
152 enum {
153 	EROFS_ZIP_CACHE_DISABLED,
154 	EROFS_ZIP_CACHE_READAHEAD,
155 	EROFS_ZIP_CACHE_READAROUND
156 };
157 
158 #ifdef CONFIG_EROFS_FS_ZIP
159 #define EROFS_LOCKED_MAGIC     (INT_MIN | 0xE0F510CCL)
160 
161 /* basic unit of the workstation of a super_block */
162 struct erofs_workgroup {
163 	/* the workgroup index in the workstation */
164 	pgoff_t index;
165 
166 	/* overall workgroup reference count */
167 	atomic_t refcount;
168 };
169 
170 #if defined(CONFIG_SMP)
171 static inline bool erofs_workgroup_try_to_freeze(struct erofs_workgroup *grp,
172 						 int val)
173 {
174 	preempt_disable();
175 	if (val != atomic_cmpxchg(&grp->refcount, val, EROFS_LOCKED_MAGIC)) {
176 		preempt_enable();
177 		return false;
178 	}
179 	return true;
180 }
181 
182 static inline void erofs_workgroup_unfreeze(struct erofs_workgroup *grp,
183 					    int orig_val)
184 {
185 	/*
186 	 * other observers should notice all modifications
187 	 * in the freezing period.
188 	 */
189 	smp_mb();
190 	atomic_set(&grp->refcount, orig_val);
191 	preempt_enable();
192 }
193 
194 static inline int erofs_wait_on_workgroup_freezed(struct erofs_workgroup *grp)
195 {
196 	return atomic_cond_read_relaxed(&grp->refcount,
197 					VAL != EROFS_LOCKED_MAGIC);
198 }
199 #else
200 static inline bool erofs_workgroup_try_to_freeze(struct erofs_workgroup *grp,
201 						 int val)
202 {
203 	preempt_disable();
204 	/* no need to spin on UP platforms, let's just disable preemption. */
205 	if (val != atomic_read(&grp->refcount)) {
206 		preempt_enable();
207 		return false;
208 	}
209 	return true;
210 }
211 
212 static inline void erofs_workgroup_unfreeze(struct erofs_workgroup *grp,
213 					    int orig_val)
214 {
215 	preempt_enable();
216 }
217 
218 static inline int erofs_wait_on_workgroup_freezed(struct erofs_workgroup *grp)
219 {
220 	int v = atomic_read(&grp->refcount);
221 
222 	/* workgroup is never freezed on uniprocessor systems */
223 	DBG_BUGON(v == EROFS_LOCKED_MAGIC);
224 	return v;
225 }
226 #endif	/* !CONFIG_SMP */
227 #endif	/* !CONFIG_EROFS_FS_ZIP */
228 
229 /* we strictly follow PAGE_SIZE and no buffer head yet */
230 #define LOG_BLOCK_SIZE		PAGE_SHIFT
231 
232 #undef LOG_SECTORS_PER_BLOCK
233 #define LOG_SECTORS_PER_BLOCK	(PAGE_SHIFT - 9)
234 
235 #undef SECTORS_PER_BLOCK
236 #define SECTORS_PER_BLOCK	(1 << SECTORS_PER_BLOCK)
237 
238 #define EROFS_BLKSIZ		(1 << LOG_BLOCK_SIZE)
239 
240 #if (EROFS_BLKSIZ % 4096 || !EROFS_BLKSIZ)
241 #error erofs cannot be used in this platform
242 #endif
243 
244 #define ROOT_NID(sb)		((sb)->root_nid)
245 
246 #define erofs_blknr(addr)       ((addr) / EROFS_BLKSIZ)
247 #define erofs_blkoff(addr)      ((addr) % EROFS_BLKSIZ)
248 #define blknr_to_addr(nr)       ((erofs_off_t)(nr) * EROFS_BLKSIZ)
249 
250 static inline erofs_off_t iloc(struct erofs_sb_info *sbi, erofs_nid_t nid)
251 {
252 	return blknr_to_addr(sbi->meta_blkaddr) + (nid << sbi->islotbits);
253 }
254 
255 #define EROFS_FEATURE_FUNCS(name, compat, feature) \
256 static inline bool erofs_sb_has_##name(struct erofs_sb_info *sbi) \
257 { \
258 	return sbi->feature_##compat & EROFS_FEATURE_##feature; \
259 }
260 
261 EROFS_FEATURE_FUNCS(lz4_0padding, incompat, INCOMPAT_LZ4_0PADDING)
262 EROFS_FEATURE_FUNCS(compr_cfgs, incompat, INCOMPAT_COMPR_CFGS)
263 EROFS_FEATURE_FUNCS(big_pcluster, incompat, INCOMPAT_BIG_PCLUSTER)
264 EROFS_FEATURE_FUNCS(device_table, incompat, INCOMPAT_DEVICE_TABLE)
265 EROFS_FEATURE_FUNCS(sb_chksum, compat, COMPAT_SB_CHKSUM)
266 
267 /* atomic flag definitions */
268 #define EROFS_I_EA_INITED_BIT	0
269 #define EROFS_I_Z_INITED_BIT	1
270 
271 /* bitlock definitions (arranged in reverse order) */
272 #define EROFS_I_BL_XATTR_BIT	(BITS_PER_LONG - 1)
273 #define EROFS_I_BL_Z_BIT	(BITS_PER_LONG - 2)
274 
275 struct erofs_inode {
276 	erofs_nid_t nid;
277 
278 	/* atomic flags (including bitlocks) */
279 	unsigned long flags;
280 
281 	unsigned char datalayout;
282 	unsigned char inode_isize;
283 	unsigned short xattr_isize;
284 
285 	unsigned int xattr_shared_count;
286 	unsigned int *xattr_shared_xattrs;
287 
288 	union {
289 		erofs_blk_t raw_blkaddr;
290 		struct {
291 			unsigned short	chunkformat;
292 			unsigned char	chunkbits;
293 		};
294 #ifdef CONFIG_EROFS_FS_ZIP
295 		struct {
296 			unsigned short z_advise;
297 			unsigned char  z_algorithmtype[2];
298 			unsigned char  z_logical_clusterbits;
299 		};
300 #endif	/* CONFIG_EROFS_FS_ZIP */
301 	};
302 	/* the corresponding vfs inode */
303 	struct inode vfs_inode;
304 };
305 
306 #define EROFS_I(ptr)	\
307 	container_of(ptr, struct erofs_inode, vfs_inode)
308 
309 static inline unsigned long erofs_inode_datablocks(struct inode *inode)
310 {
311 	/* since i_size cannot be changed */
312 	return DIV_ROUND_UP(inode->i_size, EROFS_BLKSIZ);
313 }
314 
315 static inline unsigned int erofs_bitrange(unsigned int value, unsigned int bit,
316 					  unsigned int bits)
317 {
318 
319 	return (value >> bit) & ((1 << bits) - 1);
320 }
321 
322 
323 static inline unsigned int erofs_inode_version(unsigned int value)
324 {
325 	return erofs_bitrange(value, EROFS_I_VERSION_BIT,
326 			      EROFS_I_VERSION_BITS);
327 }
328 
329 static inline unsigned int erofs_inode_datalayout(unsigned int value)
330 {
331 	return erofs_bitrange(value, EROFS_I_DATALAYOUT_BIT,
332 			      EROFS_I_DATALAYOUT_BITS);
333 }
334 
335 /*
336  * Different from grab_cache_page_nowait(), reclaiming is never triggered
337  * when allocating new pages.
338  */
339 static inline
340 struct page *erofs_grab_cache_page_nowait(struct address_space *mapping,
341 					  pgoff_t index)
342 {
343 	return pagecache_get_page(mapping, index,
344 			FGP_LOCK|FGP_CREAT|FGP_NOFS|FGP_NOWAIT,
345 			readahead_gfp_mask(mapping) & ~__GFP_RECLAIM);
346 }
347 
348 extern const struct super_operations erofs_sops;
349 
350 extern const struct address_space_operations erofs_raw_access_aops;
351 extern const struct address_space_operations z_erofs_aops;
352 
353 /*
354  * Logical to physical block mapping
355  *
356  * Different with other file systems, it is used for 2 access modes:
357  *
358  * 1) RAW access mode:
359  *
360  * Users pass a valid (m_lblk, m_lofs -- usually 0) pair,
361  * and get the valid m_pblk, m_pofs and the longest m_len(in bytes).
362  *
363  * Note that m_lblk in the RAW access mode refers to the number of
364  * the compressed ondisk block rather than the uncompressed
365  * in-memory block for the compressed file.
366  *
367  * m_pofs equals to m_lofs except for the inline data page.
368  *
369  * 2) Normal access mode:
370  *
371  * If the inode is not compressed, it has no difference with
372  * the RAW access mode. However, if the inode is compressed,
373  * users should pass a valid (m_lblk, m_lofs) pair, and get
374  * the needed m_pblk, m_pofs, m_len to get the compressed data
375  * and the updated m_lblk, m_lofs which indicates the start
376  * of the corresponding uncompressed data in the file.
377  */
378 enum {
379 	BH_Encoded = BH_PrivateStart,
380 	BH_FullMapped,
381 };
382 
383 /* Has a disk mapping */
384 #define EROFS_MAP_MAPPED	(1 << BH_Mapped)
385 /* Located in metadata (could be copied from bd_inode) */
386 #define EROFS_MAP_META		(1 << BH_Meta)
387 /* The extent is encoded */
388 #define EROFS_MAP_ENCODED	(1 << BH_Encoded)
389 /* The length of extent is full */
390 #define EROFS_MAP_FULL_MAPPED	(1 << BH_FullMapped)
391 
392 struct erofs_map_blocks {
393 	erofs_off_t m_pa, m_la;
394 	u64 m_plen, m_llen;
395 
396 	unsigned short m_deviceid;
397 	char m_algorithmformat;
398 	unsigned int m_flags;
399 
400 	struct page *mpage;
401 };
402 
403 /* Flags used by erofs_map_blocks_flatmode() */
404 #define EROFS_GET_BLOCKS_RAW    0x0001
405 /*
406  * Used to get the exact decompressed length, e.g. fiemap (consider lookback
407  * approach instead if possible since it's more metadata lightweight.)
408  */
409 #define EROFS_GET_BLOCKS_FIEMAP	0x0002
410 /* Used to map the whole extent if non-negligible data is requested for LZMA */
411 #define EROFS_GET_BLOCKS_READMORE	0x0004
412 
413 enum {
414 	Z_EROFS_COMPRESSION_SHIFTED = Z_EROFS_COMPRESSION_MAX,
415 	Z_EROFS_COMPRESSION_RUNTIME_MAX
416 };
417 
418 /* zmap.c */
419 extern const struct iomap_ops z_erofs_iomap_report_ops;
420 
421 #ifdef CONFIG_EROFS_FS_ZIP
422 int z_erofs_fill_inode(struct inode *inode);
423 int z_erofs_map_blocks_iter(struct inode *inode,
424 			    struct erofs_map_blocks *map,
425 			    int flags);
426 #else
427 static inline int z_erofs_fill_inode(struct inode *inode) { return -EOPNOTSUPP; }
428 static inline int z_erofs_map_blocks_iter(struct inode *inode,
429 					  struct erofs_map_blocks *map,
430 					  int flags)
431 {
432 	return -EOPNOTSUPP;
433 }
434 #endif	/* !CONFIG_EROFS_FS_ZIP */
435 
436 struct erofs_map_dev {
437 	struct block_device *m_bdev;
438 	struct dax_device *m_daxdev;
439 
440 	erofs_off_t m_pa;
441 	unsigned int m_deviceid;
442 };
443 
444 /* data.c */
445 extern const struct file_operations erofs_file_fops;
446 struct page *erofs_get_meta_page(struct super_block *sb, erofs_blk_t blkaddr);
447 int erofs_map_dev(struct super_block *sb, struct erofs_map_dev *dev);
448 int erofs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
449 		 u64 start, u64 len);
450 
451 /* inode.c */
452 static inline unsigned long erofs_inode_hash(erofs_nid_t nid)
453 {
454 #if BITS_PER_LONG == 32
455 	return (nid >> 32) ^ (nid & 0xffffffff);
456 #else
457 	return nid;
458 #endif
459 }
460 
461 extern const struct inode_operations erofs_generic_iops;
462 extern const struct inode_operations erofs_symlink_iops;
463 extern const struct inode_operations erofs_fast_symlink_iops;
464 
465 struct inode *erofs_iget(struct super_block *sb, erofs_nid_t nid, bool dir);
466 int erofs_getattr(struct user_namespace *mnt_userns, const struct path *path,
467 		  struct kstat *stat, u32 request_mask,
468 		  unsigned int query_flags);
469 
470 /* namei.c */
471 extern const struct inode_operations erofs_dir_iops;
472 
473 int erofs_namei(struct inode *dir, struct qstr *name,
474 		erofs_nid_t *nid, unsigned int *d_type);
475 
476 /* dir.c */
477 extern const struct file_operations erofs_dir_fops;
478 
479 static inline void *erofs_vm_map_ram(struct page **pages, unsigned int count)
480 {
481 	int retried = 0;
482 
483 	while (1) {
484 		void *p = vm_map_ram(pages, count, -1);
485 
486 		/* retry two more times (totally 3 times) */
487 		if (p || ++retried >= 3)
488 			return p;
489 		vm_unmap_aliases();
490 	}
491 	return NULL;
492 }
493 
494 /* pcpubuf.c */
495 void *erofs_get_pcpubuf(unsigned int requiredpages);
496 void erofs_put_pcpubuf(void *ptr);
497 int erofs_pcpubuf_growsize(unsigned int nrpages);
498 void erofs_pcpubuf_init(void);
499 void erofs_pcpubuf_exit(void);
500 
501 /* utils.c / zdata.c */
502 struct page *erofs_allocpage(struct page **pagepool, gfp_t gfp);
503 static inline void erofs_pagepool_add(struct page **pagepool,
504 		struct page *page)
505 {
506 	set_page_private(page, (unsigned long)*pagepool);
507 	*pagepool = page;
508 }
509 void erofs_release_pages(struct page **pagepool);
510 
511 #ifdef CONFIG_EROFS_FS_ZIP
512 int erofs_workgroup_put(struct erofs_workgroup *grp);
513 struct erofs_workgroup *erofs_find_workgroup(struct super_block *sb,
514 					     pgoff_t index);
515 struct erofs_workgroup *erofs_insert_workgroup(struct super_block *sb,
516 					       struct erofs_workgroup *grp);
517 void erofs_workgroup_free_rcu(struct erofs_workgroup *grp);
518 void erofs_shrinker_register(struct super_block *sb);
519 void erofs_shrinker_unregister(struct super_block *sb);
520 int __init erofs_init_shrinker(void);
521 void erofs_exit_shrinker(void);
522 int __init z_erofs_init_zip_subsystem(void);
523 void z_erofs_exit_zip_subsystem(void);
524 int erofs_try_to_free_all_cached_pages(struct erofs_sb_info *sbi,
525 				       struct erofs_workgroup *egrp);
526 int erofs_try_to_free_cached_page(struct page *page);
527 int z_erofs_load_lz4_config(struct super_block *sb,
528 			    struct erofs_super_block *dsb,
529 			    struct z_erofs_lz4_cfgs *lz4, int len);
530 #else
531 static inline void erofs_shrinker_register(struct super_block *sb) {}
532 static inline void erofs_shrinker_unregister(struct super_block *sb) {}
533 static inline int erofs_init_shrinker(void) { return 0; }
534 static inline void erofs_exit_shrinker(void) {}
535 static inline int z_erofs_init_zip_subsystem(void) { return 0; }
536 static inline void z_erofs_exit_zip_subsystem(void) {}
537 static inline int z_erofs_load_lz4_config(struct super_block *sb,
538 				  struct erofs_super_block *dsb,
539 				  struct z_erofs_lz4_cfgs *lz4, int len)
540 {
541 	if (lz4 || dsb->u1.lz4_max_distance) {
542 		erofs_err(sb, "lz4 algorithm isn't enabled");
543 		return -EINVAL;
544 	}
545 	return 0;
546 }
547 #endif	/* !CONFIG_EROFS_FS_ZIP */
548 
549 #ifdef CONFIG_EROFS_FS_ZIP_LZMA
550 int z_erofs_lzma_init(void);
551 void z_erofs_lzma_exit(void);
552 int z_erofs_load_lzma_config(struct super_block *sb,
553 			     struct erofs_super_block *dsb,
554 			     struct z_erofs_lzma_cfgs *lzma, int size);
555 #else
556 static inline int z_erofs_lzma_init(void) { return 0; }
557 static inline int z_erofs_lzma_exit(void) { return 0; }
558 static inline int z_erofs_load_lzma_config(struct super_block *sb,
559 			     struct erofs_super_block *dsb,
560 			     struct z_erofs_lzma_cfgs *lzma, int size) {
561 	if (lzma) {
562 		erofs_err(sb, "lzma algorithm isn't enabled");
563 		return -EINVAL;
564 	}
565 	return 0;
566 }
567 #endif	/* !CONFIG_EROFS_FS_ZIP */
568 
569 #define EFSCORRUPTED    EUCLEAN         /* Filesystem is corrupted */
570 
571 #endif	/* __EROFS_INTERNAL_H */
572