xref: /openbmc/linux/fs/erofs/internal.h (revision 30048cda)
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 	/* strategy of sync decompression (false - auto, true - force on) */
54 	bool readahead_sync_decompress;
55 
56 	/* threshold for decompression synchronously */
57 	unsigned int max_sync_decompress_pages;
58 #endif
59 	unsigned int mount_opt;
60 };
61 
62 struct erofs_sb_info {
63 #ifdef CONFIG_EROFS_FS_ZIP
64 	/* list for all registered superblocks, mainly for shrinker */
65 	struct list_head list;
66 	struct mutex umount_mutex;
67 
68 	/* managed XArray arranged in physical block number */
69 	struct xarray managed_pslots;
70 
71 	unsigned int shrinker_run_no;
72 
73 	/* pseudo inode to manage cached pages */
74 	struct inode *managed_cache;
75 #endif	/* CONFIG_EROFS_FS_ZIP */
76 	u32 blocks;
77 	u32 meta_blkaddr;
78 #ifdef CONFIG_EROFS_FS_XATTR
79 	u32 xattr_blkaddr;
80 #endif
81 
82 	/* inode slot unit size in bit shift */
83 	unsigned char islotbits;
84 
85 	u32 build_time_nsec;
86 	u64 build_time;
87 
88 	/* what we really care is nid, rather than ino.. */
89 	erofs_nid_t root_nid;
90 	/* used for statfs, f_files - f_favail */
91 	u64 inos;
92 
93 	u8 uuid[16];                    /* 128-bit uuid for volume */
94 	u8 volume_name[16];             /* volume name */
95 	u32 feature_compat;
96 	u32 feature_incompat;
97 
98 	struct erofs_fs_context ctx;	/* options */
99 };
100 
101 #define EROFS_SB(sb) ((struct erofs_sb_info *)(sb)->s_fs_info)
102 #define EROFS_I_SB(inode) ((struct erofs_sb_info *)(inode)->i_sb->s_fs_info)
103 
104 /* Mount flags set via mount options or defaults */
105 #define EROFS_MOUNT_XATTR_USER		0x00000010
106 #define EROFS_MOUNT_POSIX_ACL		0x00000020
107 
108 #define clear_opt(ctx, option)	((ctx)->mount_opt &= ~EROFS_MOUNT_##option)
109 #define set_opt(ctx, option)	((ctx)->mount_opt |= EROFS_MOUNT_##option)
110 #define test_opt(ctx, option)	((ctx)->mount_opt & EROFS_MOUNT_##option)
111 
112 enum {
113 	EROFS_ZIP_CACHE_DISABLED,
114 	EROFS_ZIP_CACHE_READAHEAD,
115 	EROFS_ZIP_CACHE_READAROUND
116 };
117 
118 #ifdef CONFIG_EROFS_FS_ZIP
119 #define EROFS_LOCKED_MAGIC     (INT_MIN | 0xE0F510CCL)
120 
121 /* basic unit of the workstation of a super_block */
122 struct erofs_workgroup {
123 	/* the workgroup index in the workstation */
124 	pgoff_t index;
125 
126 	/* overall workgroup reference count */
127 	atomic_t refcount;
128 };
129 
130 #if defined(CONFIG_SMP)
131 static inline bool erofs_workgroup_try_to_freeze(struct erofs_workgroup *grp,
132 						 int val)
133 {
134 	preempt_disable();
135 	if (val != atomic_cmpxchg(&grp->refcount, val, EROFS_LOCKED_MAGIC)) {
136 		preempt_enable();
137 		return false;
138 	}
139 	return true;
140 }
141 
142 static inline void erofs_workgroup_unfreeze(struct erofs_workgroup *grp,
143 					    int orig_val)
144 {
145 	/*
146 	 * other observers should notice all modifications
147 	 * in the freezing period.
148 	 */
149 	smp_mb();
150 	atomic_set(&grp->refcount, orig_val);
151 	preempt_enable();
152 }
153 
154 static inline int erofs_wait_on_workgroup_freezed(struct erofs_workgroup *grp)
155 {
156 	return atomic_cond_read_relaxed(&grp->refcount,
157 					VAL != EROFS_LOCKED_MAGIC);
158 }
159 #else
160 static inline bool erofs_workgroup_try_to_freeze(struct erofs_workgroup *grp,
161 						 int val)
162 {
163 	preempt_disable();
164 	/* no need to spin on UP platforms, let's just disable preemption. */
165 	if (val != atomic_read(&grp->refcount)) {
166 		preempt_enable();
167 		return false;
168 	}
169 	return true;
170 }
171 
172 static inline void erofs_workgroup_unfreeze(struct erofs_workgroup *grp,
173 					    int orig_val)
174 {
175 	preempt_enable();
176 }
177 
178 static inline int erofs_wait_on_workgroup_freezed(struct erofs_workgroup *grp)
179 {
180 	int v = atomic_read(&grp->refcount);
181 
182 	/* workgroup is never freezed on uniprocessor systems */
183 	DBG_BUGON(v == EROFS_LOCKED_MAGIC);
184 	return v;
185 }
186 #endif	/* !CONFIG_SMP */
187 
188 /* hard limit of pages per compressed cluster */
189 #define Z_EROFS_CLUSTER_MAX_PAGES       (CONFIG_EROFS_FS_CLUSTER_PAGE_LIMIT)
190 #define EROFS_PCPUBUF_NR_PAGES          Z_EROFS_CLUSTER_MAX_PAGES
191 #else
192 #define EROFS_PCPUBUF_NR_PAGES          0
193 #endif	/* !CONFIG_EROFS_FS_ZIP */
194 
195 /* we strictly follow PAGE_SIZE and no buffer head yet */
196 #define LOG_BLOCK_SIZE		PAGE_SHIFT
197 
198 #undef LOG_SECTORS_PER_BLOCK
199 #define LOG_SECTORS_PER_BLOCK	(PAGE_SHIFT - 9)
200 
201 #undef SECTORS_PER_BLOCK
202 #define SECTORS_PER_BLOCK	(1 << SECTORS_PER_BLOCK)
203 
204 #define EROFS_BLKSIZ		(1 << LOG_BLOCK_SIZE)
205 
206 #if (EROFS_BLKSIZ % 4096 || !EROFS_BLKSIZ)
207 #error erofs cannot be used in this platform
208 #endif
209 
210 #define ROOT_NID(sb)		((sb)->root_nid)
211 
212 #define erofs_blknr(addr)       ((addr) / EROFS_BLKSIZ)
213 #define erofs_blkoff(addr)      ((addr) % EROFS_BLKSIZ)
214 #define blknr_to_addr(nr)       ((erofs_off_t)(nr) * EROFS_BLKSIZ)
215 
216 static inline erofs_off_t iloc(struct erofs_sb_info *sbi, erofs_nid_t nid)
217 {
218 	return blknr_to_addr(sbi->meta_blkaddr) + (nid << sbi->islotbits);
219 }
220 
221 /* atomic flag definitions */
222 #define EROFS_I_EA_INITED_BIT	0
223 #define EROFS_I_Z_INITED_BIT	1
224 
225 /* bitlock definitions (arranged in reverse order) */
226 #define EROFS_I_BL_XATTR_BIT	(BITS_PER_LONG - 1)
227 #define EROFS_I_BL_Z_BIT	(BITS_PER_LONG - 2)
228 
229 struct erofs_inode {
230 	erofs_nid_t nid;
231 
232 	/* atomic flags (including bitlocks) */
233 	unsigned long flags;
234 
235 	unsigned char datalayout;
236 	unsigned char inode_isize;
237 	unsigned short xattr_isize;
238 
239 	unsigned int xattr_shared_count;
240 	unsigned int *xattr_shared_xattrs;
241 
242 	union {
243 		erofs_blk_t raw_blkaddr;
244 #ifdef CONFIG_EROFS_FS_ZIP
245 		struct {
246 			unsigned short z_advise;
247 			unsigned char  z_algorithmtype[2];
248 			unsigned char  z_logical_clusterbits;
249 			unsigned char  z_physical_clusterbits[2];
250 		};
251 #endif	/* CONFIG_EROFS_FS_ZIP */
252 	};
253 	/* the corresponding vfs inode */
254 	struct inode vfs_inode;
255 };
256 
257 #define EROFS_I(ptr)	\
258 	container_of(ptr, struct erofs_inode, vfs_inode)
259 
260 static inline unsigned long erofs_inode_datablocks(struct inode *inode)
261 {
262 	/* since i_size cannot be changed */
263 	return DIV_ROUND_UP(inode->i_size, EROFS_BLKSIZ);
264 }
265 
266 static inline unsigned int erofs_bitrange(unsigned int value, unsigned int bit,
267 					  unsigned int bits)
268 {
269 
270 	return (value >> bit) & ((1 << bits) - 1);
271 }
272 
273 
274 static inline unsigned int erofs_inode_version(unsigned int value)
275 {
276 	return erofs_bitrange(value, EROFS_I_VERSION_BIT,
277 			      EROFS_I_VERSION_BITS);
278 }
279 
280 static inline unsigned int erofs_inode_datalayout(unsigned int value)
281 {
282 	return erofs_bitrange(value, EROFS_I_DATALAYOUT_BIT,
283 			      EROFS_I_DATALAYOUT_BITS);
284 }
285 
286 extern const struct super_operations erofs_sops;
287 
288 extern const struct address_space_operations erofs_raw_access_aops;
289 extern const struct address_space_operations z_erofs_aops;
290 
291 /*
292  * Logical to physical block mapping, used by erofs_map_blocks()
293  *
294  * Different with other file systems, it is used for 2 access modes:
295  *
296  * 1) RAW access mode:
297  *
298  * Users pass a valid (m_lblk, m_lofs -- usually 0) pair,
299  * and get the valid m_pblk, m_pofs and the longest m_len(in bytes).
300  *
301  * Note that m_lblk in the RAW access mode refers to the number of
302  * the compressed ondisk block rather than the uncompressed
303  * in-memory block for the compressed file.
304  *
305  * m_pofs equals to m_lofs except for the inline data page.
306  *
307  * 2) Normal access mode:
308  *
309  * If the inode is not compressed, it has no difference with
310  * the RAW access mode. However, if the inode is compressed,
311  * users should pass a valid (m_lblk, m_lofs) pair, and get
312  * the needed m_pblk, m_pofs, m_len to get the compressed data
313  * and the updated m_lblk, m_lofs which indicates the start
314  * of the corresponding uncompressed data in the file.
315  */
316 enum {
317 	BH_Zipped = BH_PrivateStart,
318 	BH_FullMapped,
319 };
320 
321 /* Has a disk mapping */
322 #define EROFS_MAP_MAPPED	(1 << BH_Mapped)
323 /* Located in metadata (could be copied from bd_inode) */
324 #define EROFS_MAP_META		(1 << BH_Meta)
325 /* The extent has been compressed */
326 #define EROFS_MAP_ZIPPED	(1 << BH_Zipped)
327 /* The length of extent is full */
328 #define EROFS_MAP_FULL_MAPPED	(1 << BH_FullMapped)
329 
330 struct erofs_map_blocks {
331 	erofs_off_t m_pa, m_la;
332 	u64 m_plen, m_llen;
333 
334 	unsigned int m_flags;
335 
336 	struct page *mpage;
337 };
338 
339 /* Flags used by erofs_map_blocks() */
340 #define EROFS_GET_BLOCKS_RAW    0x0001
341 
342 /* zmap.c */
343 #ifdef CONFIG_EROFS_FS_ZIP
344 int z_erofs_fill_inode(struct inode *inode);
345 int z_erofs_map_blocks_iter(struct inode *inode,
346 			    struct erofs_map_blocks *map,
347 			    int flags);
348 #else
349 static inline int z_erofs_fill_inode(struct inode *inode) { return -EOPNOTSUPP; }
350 static inline int z_erofs_map_blocks_iter(struct inode *inode,
351 					  struct erofs_map_blocks *map,
352 					  int flags)
353 {
354 	return -EOPNOTSUPP;
355 }
356 #endif	/* !CONFIG_EROFS_FS_ZIP */
357 
358 /* data.c */
359 struct page *erofs_get_meta_page(struct super_block *sb, erofs_blk_t blkaddr);
360 
361 int erofs_map_blocks(struct inode *, struct erofs_map_blocks *, int);
362 
363 /* inode.c */
364 static inline unsigned long erofs_inode_hash(erofs_nid_t nid)
365 {
366 #if BITS_PER_LONG == 32
367 	return (nid >> 32) ^ (nid & 0xffffffff);
368 #else
369 	return nid;
370 #endif
371 }
372 
373 extern const struct inode_operations erofs_generic_iops;
374 extern const struct inode_operations erofs_symlink_iops;
375 extern const struct inode_operations erofs_fast_symlink_iops;
376 
377 struct inode *erofs_iget(struct super_block *sb, erofs_nid_t nid, bool dir);
378 int erofs_getattr(struct user_namespace *mnt_userns, const struct path *path,
379 		  struct kstat *stat, u32 request_mask,
380 		  unsigned int query_flags);
381 
382 /* namei.c */
383 extern const struct inode_operations erofs_dir_iops;
384 
385 int erofs_namei(struct inode *dir, struct qstr *name,
386 		erofs_nid_t *nid, unsigned int *d_type);
387 
388 /* dir.c */
389 extern const struct file_operations erofs_dir_fops;
390 
391 /* utils.c / zdata.c */
392 struct page *erofs_allocpage(struct list_head *pool, gfp_t gfp);
393 
394 #if (EROFS_PCPUBUF_NR_PAGES > 0)
395 void *erofs_get_pcpubuf(unsigned int pagenr);
396 #define erofs_put_pcpubuf(buf) do { \
397 	(void)&(buf);	\
398 	preempt_enable();	\
399 } while (0)
400 #else
401 static inline void *erofs_get_pcpubuf(unsigned int pagenr)
402 {
403 	return ERR_PTR(-EOPNOTSUPP);
404 }
405 
406 #define erofs_put_pcpubuf(buf) do {} while (0)
407 #endif
408 
409 #ifdef CONFIG_EROFS_FS_ZIP
410 int erofs_workgroup_put(struct erofs_workgroup *grp);
411 struct erofs_workgroup *erofs_find_workgroup(struct super_block *sb,
412 					     pgoff_t index);
413 struct erofs_workgroup *erofs_insert_workgroup(struct super_block *sb,
414 					       struct erofs_workgroup *grp);
415 void erofs_workgroup_free_rcu(struct erofs_workgroup *grp);
416 void erofs_shrinker_register(struct super_block *sb);
417 void erofs_shrinker_unregister(struct super_block *sb);
418 int __init erofs_init_shrinker(void);
419 void erofs_exit_shrinker(void);
420 int __init z_erofs_init_zip_subsystem(void);
421 void z_erofs_exit_zip_subsystem(void);
422 int erofs_try_to_free_all_cached_pages(struct erofs_sb_info *sbi,
423 				       struct erofs_workgroup *egrp);
424 int erofs_try_to_free_cached_page(struct address_space *mapping,
425 				  struct page *page);
426 #else
427 static inline void erofs_shrinker_register(struct super_block *sb) {}
428 static inline void erofs_shrinker_unregister(struct super_block *sb) {}
429 static inline int erofs_init_shrinker(void) { return 0; }
430 static inline void erofs_exit_shrinker(void) {}
431 static inline int z_erofs_init_zip_subsystem(void) { return 0; }
432 static inline void z_erofs_exit_zip_subsystem(void) {}
433 #endif	/* !CONFIG_EROFS_FS_ZIP */
434 
435 #define EFSCORRUPTED    EUCLEAN         /* Filesystem is corrupted */
436 
437 #endif	/* __EROFS_INTERNAL_H */
438 
439