xref: /openbmc/linux/fs/erofs/internal.h (revision 79f382b9)
1 /* SPDX-License-Identifier: GPL-2.0-only */
2 /*
3  * Copyright (C) 2017-2018 HUAWEI, Inc.
4  *             https://www.huawei.com/
5  * Created by Gao Xiang <gaoxiang25@huawei.com>
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 "erofs_fs.h"
20 
21 /* redefine pr_fmt "erofs: " */
22 #undef pr_fmt
23 #define pr_fmt(fmt) "erofs: " fmt
24 
25 __printf(3, 4) void _erofs_err(struct super_block *sb,
26 			       const char *function, const char *fmt, ...);
27 #define erofs_err(sb, fmt, ...)	\
28 	_erofs_err(sb, __func__, fmt "\n", ##__VA_ARGS__)
29 __printf(3, 4) void _erofs_info(struct super_block *sb,
30 			       const char *function, const char *fmt, ...);
31 #define erofs_info(sb, fmt, ...) \
32 	_erofs_info(sb, __func__, fmt "\n", ##__VA_ARGS__)
33 #ifdef CONFIG_EROFS_FS_DEBUG
34 #define erofs_dbg(x, ...)       pr_debug(x "\n", ##__VA_ARGS__)
35 #define DBG_BUGON               BUG_ON
36 #else
37 #define erofs_dbg(x, ...)       ((void)0)
38 #define DBG_BUGON(x)            ((void)(x))
39 #endif	/* !CONFIG_EROFS_FS_DEBUG */
40 
41 /* EROFS_SUPER_MAGIC_V1 to represent the whole file system */
42 #define EROFS_SUPER_MAGIC   EROFS_SUPER_MAGIC_V1
43 
44 typedef u64 erofs_nid_t;
45 typedef u64 erofs_off_t;
46 /* data type for filesystem-wide blocks number */
47 typedef u32 erofs_blk_t;
48 
49 struct erofs_fs_context {
50 #ifdef CONFIG_EROFS_FS_ZIP
51 	/* current strategy of how to use managed cache */
52 	unsigned char cache_strategy;
53 
54 	/* threshold for decompression synchronously */
55 	unsigned int max_sync_decompress_pages;
56 #endif
57 	unsigned int mount_opt;
58 };
59 
60 struct erofs_sb_info {
61 #ifdef CONFIG_EROFS_FS_ZIP
62 	/* list for all registered superblocks, mainly for shrinker */
63 	struct list_head list;
64 	struct mutex umount_mutex;
65 
66 	/* managed XArray arranged in physical block number */
67 	struct xarray managed_pslots;
68 
69 	unsigned int shrinker_run_no;
70 
71 	/* pseudo inode to manage cached pages */
72 	struct inode *managed_cache;
73 #endif	/* CONFIG_EROFS_FS_ZIP */
74 	u32 blocks;
75 	u32 meta_blkaddr;
76 #ifdef CONFIG_EROFS_FS_XATTR
77 	u32 xattr_blkaddr;
78 #endif
79 
80 	/* inode slot unit size in bit shift */
81 	unsigned char islotbits;
82 
83 	u32 build_time_nsec;
84 	u64 build_time;
85 
86 	/* what we really care is nid, rather than ino.. */
87 	erofs_nid_t root_nid;
88 	/* used for statfs, f_files - f_favail */
89 	u64 inos;
90 
91 	u8 uuid[16];                    /* 128-bit uuid for volume */
92 	u8 volume_name[16];             /* volume name */
93 	u32 feature_compat;
94 	u32 feature_incompat;
95 
96 	struct erofs_fs_context ctx;	/* options */
97 };
98 
99 #define EROFS_SB(sb) ((struct erofs_sb_info *)(sb)->s_fs_info)
100 #define EROFS_I_SB(inode) ((struct erofs_sb_info *)(inode)->i_sb->s_fs_info)
101 
102 /* Mount flags set via mount options or defaults */
103 #define EROFS_MOUNT_XATTR_USER		0x00000010
104 #define EROFS_MOUNT_POSIX_ACL		0x00000020
105 
106 #define clear_opt(ctx, option)	((ctx)->mount_opt &= ~EROFS_MOUNT_##option)
107 #define set_opt(ctx, option)	((ctx)->mount_opt |= EROFS_MOUNT_##option)
108 #define test_opt(ctx, option)	((ctx)->mount_opt & EROFS_MOUNT_##option)
109 
110 enum {
111 	EROFS_ZIP_CACHE_DISABLED,
112 	EROFS_ZIP_CACHE_READAHEAD,
113 	EROFS_ZIP_CACHE_READAROUND
114 };
115 
116 #ifdef CONFIG_EROFS_FS_ZIP
117 #define EROFS_LOCKED_MAGIC     (INT_MIN | 0xE0F510CCL)
118 
119 /* basic unit of the workstation of a super_block */
120 struct erofs_workgroup {
121 	/* the workgroup index in the workstation */
122 	pgoff_t index;
123 
124 	/* overall workgroup reference count */
125 	atomic_t refcount;
126 };
127 
128 #if defined(CONFIG_SMP)
129 static inline bool erofs_workgroup_try_to_freeze(struct erofs_workgroup *grp,
130 						 int val)
131 {
132 	preempt_disable();
133 	if (val != atomic_cmpxchg(&grp->refcount, val, EROFS_LOCKED_MAGIC)) {
134 		preempt_enable();
135 		return false;
136 	}
137 	return true;
138 }
139 
140 static inline void erofs_workgroup_unfreeze(struct erofs_workgroup *grp,
141 					    int orig_val)
142 {
143 	/*
144 	 * other observers should notice all modifications
145 	 * in the freezing period.
146 	 */
147 	smp_mb();
148 	atomic_set(&grp->refcount, orig_val);
149 	preempt_enable();
150 }
151 
152 static inline int erofs_wait_on_workgroup_freezed(struct erofs_workgroup *grp)
153 {
154 	return atomic_cond_read_relaxed(&grp->refcount,
155 					VAL != EROFS_LOCKED_MAGIC);
156 }
157 #else
158 static inline bool erofs_workgroup_try_to_freeze(struct erofs_workgroup *grp,
159 						 int val)
160 {
161 	preempt_disable();
162 	/* no need to spin on UP platforms, let's just disable preemption. */
163 	if (val != atomic_read(&grp->refcount)) {
164 		preempt_enable();
165 		return false;
166 	}
167 	return true;
168 }
169 
170 static inline void erofs_workgroup_unfreeze(struct erofs_workgroup *grp,
171 					    int orig_val)
172 {
173 	preempt_enable();
174 }
175 
176 static inline int erofs_wait_on_workgroup_freezed(struct erofs_workgroup *grp)
177 {
178 	int v = atomic_read(&grp->refcount);
179 
180 	/* workgroup is never freezed on uniprocessor systems */
181 	DBG_BUGON(v == EROFS_LOCKED_MAGIC);
182 	return v;
183 }
184 #endif	/* !CONFIG_SMP */
185 
186 /* hard limit of pages per compressed cluster */
187 #define Z_EROFS_CLUSTER_MAX_PAGES       (CONFIG_EROFS_FS_CLUSTER_PAGE_LIMIT)
188 #define EROFS_PCPUBUF_NR_PAGES          Z_EROFS_CLUSTER_MAX_PAGES
189 #else
190 #define EROFS_PCPUBUF_NR_PAGES          0
191 #endif	/* !CONFIG_EROFS_FS_ZIP */
192 
193 /* we strictly follow PAGE_SIZE and no buffer head yet */
194 #define LOG_BLOCK_SIZE		PAGE_SHIFT
195 
196 #undef LOG_SECTORS_PER_BLOCK
197 #define LOG_SECTORS_PER_BLOCK	(PAGE_SHIFT - 9)
198 
199 #undef SECTORS_PER_BLOCK
200 #define SECTORS_PER_BLOCK	(1 << SECTORS_PER_BLOCK)
201 
202 #define EROFS_BLKSIZ		(1 << LOG_BLOCK_SIZE)
203 
204 #if (EROFS_BLKSIZ % 4096 || !EROFS_BLKSIZ)
205 #error erofs cannot be used in this platform
206 #endif
207 
208 #define ROOT_NID(sb)		((sb)->root_nid)
209 
210 #define erofs_blknr(addr)       ((addr) / EROFS_BLKSIZ)
211 #define erofs_blkoff(addr)      ((addr) % EROFS_BLKSIZ)
212 #define blknr_to_addr(nr)       ((erofs_off_t)(nr) * EROFS_BLKSIZ)
213 
214 static inline erofs_off_t iloc(struct erofs_sb_info *sbi, erofs_nid_t nid)
215 {
216 	return blknr_to_addr(sbi->meta_blkaddr) + (nid << sbi->islotbits);
217 }
218 
219 /* atomic flag definitions */
220 #define EROFS_I_EA_INITED_BIT	0
221 #define EROFS_I_Z_INITED_BIT	1
222 
223 /* bitlock definitions (arranged in reverse order) */
224 #define EROFS_I_BL_XATTR_BIT	(BITS_PER_LONG - 1)
225 #define EROFS_I_BL_Z_BIT	(BITS_PER_LONG - 2)
226 
227 struct erofs_inode {
228 	erofs_nid_t nid;
229 
230 	/* atomic flags (including bitlocks) */
231 	unsigned long flags;
232 
233 	unsigned char datalayout;
234 	unsigned char inode_isize;
235 	unsigned short xattr_isize;
236 
237 	unsigned int xattr_shared_count;
238 	unsigned int *xattr_shared_xattrs;
239 
240 	union {
241 		erofs_blk_t raw_blkaddr;
242 #ifdef CONFIG_EROFS_FS_ZIP
243 		struct {
244 			unsigned short z_advise;
245 			unsigned char  z_algorithmtype[2];
246 			unsigned char  z_logical_clusterbits;
247 			unsigned char  z_physical_clusterbits[2];
248 		};
249 #endif	/* CONFIG_EROFS_FS_ZIP */
250 	};
251 	/* the corresponding vfs inode */
252 	struct inode vfs_inode;
253 };
254 
255 #define EROFS_I(ptr)	\
256 	container_of(ptr, struct erofs_inode, vfs_inode)
257 
258 static inline unsigned long erofs_inode_datablocks(struct inode *inode)
259 {
260 	/* since i_size cannot be changed */
261 	return DIV_ROUND_UP(inode->i_size, EROFS_BLKSIZ);
262 }
263 
264 static inline unsigned int erofs_bitrange(unsigned int value, unsigned int bit,
265 					  unsigned int bits)
266 {
267 
268 	return (value >> bit) & ((1 << bits) - 1);
269 }
270 
271 
272 static inline unsigned int erofs_inode_version(unsigned int value)
273 {
274 	return erofs_bitrange(value, EROFS_I_VERSION_BIT,
275 			      EROFS_I_VERSION_BITS);
276 }
277 
278 static inline unsigned int erofs_inode_datalayout(unsigned int value)
279 {
280 	return erofs_bitrange(value, EROFS_I_DATALAYOUT_BIT,
281 			      EROFS_I_DATALAYOUT_BITS);
282 }
283 
284 extern const struct super_operations erofs_sops;
285 
286 extern const struct address_space_operations erofs_raw_access_aops;
287 extern const struct address_space_operations z_erofs_aops;
288 
289 /*
290  * Logical to physical block mapping, used by erofs_map_blocks()
291  *
292  * Different with other file systems, it is used for 2 access modes:
293  *
294  * 1) RAW access mode:
295  *
296  * Users pass a valid (m_lblk, m_lofs -- usually 0) pair,
297  * and get the valid m_pblk, m_pofs and the longest m_len(in bytes).
298  *
299  * Note that m_lblk in the RAW access mode refers to the number of
300  * the compressed ondisk block rather than the uncompressed
301  * in-memory block for the compressed file.
302  *
303  * m_pofs equals to m_lofs except for the inline data page.
304  *
305  * 2) Normal access mode:
306  *
307  * If the inode is not compressed, it has no difference with
308  * the RAW access mode. However, if the inode is compressed,
309  * users should pass a valid (m_lblk, m_lofs) pair, and get
310  * the needed m_pblk, m_pofs, m_len to get the compressed data
311  * and the updated m_lblk, m_lofs which indicates the start
312  * of the corresponding uncompressed data in the file.
313  */
314 enum {
315 	BH_Zipped = BH_PrivateStart,
316 	BH_FullMapped,
317 };
318 
319 /* Has a disk mapping */
320 #define EROFS_MAP_MAPPED	(1 << BH_Mapped)
321 /* Located in metadata (could be copied from bd_inode) */
322 #define EROFS_MAP_META		(1 << BH_Meta)
323 /* The extent has been compressed */
324 #define EROFS_MAP_ZIPPED	(1 << BH_Zipped)
325 /* The length of extent is full */
326 #define EROFS_MAP_FULL_MAPPED	(1 << BH_FullMapped)
327 
328 struct erofs_map_blocks {
329 	erofs_off_t m_pa, m_la;
330 	u64 m_plen, m_llen;
331 
332 	unsigned int m_flags;
333 
334 	struct page *mpage;
335 };
336 
337 /* Flags used by erofs_map_blocks() */
338 #define EROFS_GET_BLOCKS_RAW    0x0001
339 
340 /* zmap.c */
341 #ifdef CONFIG_EROFS_FS_ZIP
342 int z_erofs_fill_inode(struct inode *inode);
343 int z_erofs_map_blocks_iter(struct inode *inode,
344 			    struct erofs_map_blocks *map,
345 			    int flags);
346 #else
347 static inline int z_erofs_fill_inode(struct inode *inode) { return -EOPNOTSUPP; }
348 static inline int z_erofs_map_blocks_iter(struct inode *inode,
349 					  struct erofs_map_blocks *map,
350 					  int flags)
351 {
352 	return -EOPNOTSUPP;
353 }
354 #endif	/* !CONFIG_EROFS_FS_ZIP */
355 
356 /* data.c */
357 struct page *erofs_get_meta_page(struct super_block *sb, erofs_blk_t blkaddr);
358 
359 int erofs_map_blocks(struct inode *, struct erofs_map_blocks *, int);
360 
361 /* inode.c */
362 static inline unsigned long erofs_inode_hash(erofs_nid_t nid)
363 {
364 #if BITS_PER_LONG == 32
365 	return (nid >> 32) ^ (nid & 0xffffffff);
366 #else
367 	return nid;
368 #endif
369 }
370 
371 extern const struct inode_operations erofs_generic_iops;
372 extern const struct inode_operations erofs_symlink_iops;
373 extern const struct inode_operations erofs_fast_symlink_iops;
374 
375 struct inode *erofs_iget(struct super_block *sb, erofs_nid_t nid, bool dir);
376 int erofs_getattr(struct user_namespace *mnt_userns, const struct path *path,
377 		  struct kstat *stat, u32 request_mask,
378 		  unsigned int query_flags);
379 
380 /* namei.c */
381 extern const struct inode_operations erofs_dir_iops;
382 
383 int erofs_namei(struct inode *dir, struct qstr *name,
384 		erofs_nid_t *nid, unsigned int *d_type);
385 
386 /* dir.c */
387 extern const struct file_operations erofs_dir_fops;
388 
389 /* utils.c / zdata.c */
390 struct page *erofs_allocpage(struct list_head *pool, gfp_t gfp);
391 
392 #if (EROFS_PCPUBUF_NR_PAGES > 0)
393 void *erofs_get_pcpubuf(unsigned int pagenr);
394 #define erofs_put_pcpubuf(buf) do { \
395 	(void)&(buf);	\
396 	preempt_enable();	\
397 } while (0)
398 #else
399 static inline void *erofs_get_pcpubuf(unsigned int pagenr)
400 {
401 	return ERR_PTR(-EOPNOTSUPP);
402 }
403 
404 #define erofs_put_pcpubuf(buf) do {} while (0)
405 #endif
406 
407 #ifdef CONFIG_EROFS_FS_ZIP
408 int erofs_workgroup_put(struct erofs_workgroup *grp);
409 struct erofs_workgroup *erofs_find_workgroup(struct super_block *sb,
410 					     pgoff_t index);
411 struct erofs_workgroup *erofs_insert_workgroup(struct super_block *sb,
412 					       struct erofs_workgroup *grp);
413 void erofs_workgroup_free_rcu(struct erofs_workgroup *grp);
414 void erofs_shrinker_register(struct super_block *sb);
415 void erofs_shrinker_unregister(struct super_block *sb);
416 int __init erofs_init_shrinker(void);
417 void erofs_exit_shrinker(void);
418 int __init z_erofs_init_zip_subsystem(void);
419 void z_erofs_exit_zip_subsystem(void);
420 int erofs_try_to_free_all_cached_pages(struct erofs_sb_info *sbi,
421 				       struct erofs_workgroup *egrp);
422 int erofs_try_to_free_cached_page(struct address_space *mapping,
423 				  struct page *page);
424 #else
425 static inline void erofs_shrinker_register(struct super_block *sb) {}
426 static inline void erofs_shrinker_unregister(struct super_block *sb) {}
427 static inline int erofs_init_shrinker(void) { return 0; }
428 static inline void erofs_exit_shrinker(void) {}
429 static inline int z_erofs_init_zip_subsystem(void) { return 0; }
430 static inline void z_erofs_exit_zip_subsystem(void) {}
431 #endif	/* !CONFIG_EROFS_FS_ZIP */
432 
433 #define EFSCORRUPTED    EUCLEAN         /* Filesystem is corrupted */
434 
435 #endif	/* __EROFS_INTERNAL_H */
436 
437