xref: /linux/fs/f2fs/node.h (revision 995832b2cebe6969d1b42635db698803ee31294d)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 /*
3  * fs/f2fs/node.h
4  *
5  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
6  *             http://www.samsung.com/
7  */
8 /* start node id of a node block dedicated to the given node id */
9 #define	START_NID(nid) (((nid) / NAT_ENTRY_PER_BLOCK) * NAT_ENTRY_PER_BLOCK)
10 
11 /* node block offset on the NAT area dedicated to the given start node id */
12 #define	NAT_BLOCK_OFFSET(start_nid) ((start_nid) / NAT_ENTRY_PER_BLOCK)
13 
14 /* # of pages to perform synchronous readahead before building free nids */
15 #define FREE_NID_PAGES	8
16 #define MAX_FREE_NIDS	(NAT_ENTRY_PER_BLOCK * FREE_NID_PAGES)
17 
18 /* size of free nid batch when shrinking */
19 #define SHRINK_NID_BATCH_SIZE	8
20 
21 #define DEF_RA_NID_PAGES	0	/* # of nid pages to be readaheaded */
22 
23 /* maximum readahead size for node during getting data blocks */
24 #define MAX_RA_NODE		128
25 
26 /* control the memory footprint threshold (10MB per 1GB ram) */
27 #define DEF_RAM_THRESHOLD	1
28 
29 /* control dirty nats ratio threshold (default: 10% over max nid count) */
30 #define DEF_DIRTY_NAT_RATIO_THRESHOLD		10
31 /* control total # of nats */
32 #define DEF_NAT_CACHE_THRESHOLD			100000
33 
34 /* control total # of node writes used for roll-forward recovery */
35 #define DEF_RF_NODE_BLOCKS			0
36 
37 /* vector size for gang look-up from nat cache that consists of radix tree */
38 #define NAT_VEC_SIZE	32
39 
40 /* return value for read_node_page */
41 #define LOCKED_PAGE	1
42 
43 /* check pinned file's alignment status of physical blocks */
44 #define FILE_NOT_ALIGNED	1
45 
46 /* For flag in struct node_info */
47 enum {
48 	IS_CHECKPOINTED,	/* is it checkpointed before? */
49 	HAS_FSYNCED_INODE,	/* is the inode fsynced before? */
50 	HAS_LAST_FSYNC,		/* has the latest node fsync mark? */
51 	IS_DIRTY,		/* this nat entry is dirty? */
52 	IS_PREALLOC,		/* nat entry is preallocated */
53 };
54 
55 /*
56  * For node information
57  */
58 struct node_info {
59 	nid_t nid;		/* node id */
60 	nid_t ino;		/* inode number of the node's owner */
61 	block_t	blk_addr;	/* block address of the node */
62 	unsigned char version;	/* version of the node */
63 	unsigned char flag;	/* for node information bits */
64 };
65 
66 struct nat_entry {
67 	struct list_head list;	/* for clean or dirty nat list */
68 	struct node_info ni;	/* in-memory node information */
69 };
70 
71 #define nat_get_nid(nat)		((nat)->ni.nid)
72 #define nat_set_nid(nat, n)		((nat)->ni.nid = (n))
73 #define nat_get_blkaddr(nat)		((nat)->ni.blk_addr)
74 #define nat_set_blkaddr(nat, b)		((nat)->ni.blk_addr = (b))
75 #define nat_get_ino(nat)		((nat)->ni.ino)
76 #define nat_set_ino(nat, i)		((nat)->ni.ino = (i))
77 #define nat_get_version(nat)		((nat)->ni.version)
78 #define nat_set_version(nat, v)		((nat)->ni.version = (v))
79 
80 #define inc_node_version(version)	(++(version))
81 
82 static inline void copy_node_info(struct node_info *dst,
83 						struct node_info *src)
84 {
85 	dst->nid = src->nid;
86 	dst->ino = src->ino;
87 	dst->blk_addr = src->blk_addr;
88 	dst->version = src->version;
89 	/* should not copy flag here */
90 }
91 
92 static inline void set_nat_flag(struct nat_entry *ne,
93 				unsigned int type, bool set)
94 {
95 	if (set)
96 		ne->ni.flag |= BIT(type);
97 	else
98 		ne->ni.flag &= ~BIT(type);
99 }
100 
101 static inline bool get_nat_flag(struct nat_entry *ne, unsigned int type)
102 {
103 	return ne->ni.flag & BIT(type);
104 }
105 
106 static inline void nat_reset_flag(struct nat_entry *ne)
107 {
108 	/* these states can be set only after checkpoint was done */
109 	set_nat_flag(ne, IS_CHECKPOINTED, true);
110 	set_nat_flag(ne, HAS_FSYNCED_INODE, false);
111 	set_nat_flag(ne, HAS_LAST_FSYNC, true);
112 }
113 
114 static inline void node_info_from_raw_nat(struct node_info *ni,
115 						struct f2fs_nat_entry *raw_ne)
116 {
117 	ni->ino = le32_to_cpu(raw_ne->ino);
118 	ni->blk_addr = le32_to_cpu(raw_ne->block_addr);
119 	ni->version = raw_ne->version;
120 }
121 
122 static inline void raw_nat_from_node_info(struct f2fs_nat_entry *raw_ne,
123 						struct node_info *ni)
124 {
125 	raw_ne->ino = cpu_to_le32(ni->ino);
126 	raw_ne->block_addr = cpu_to_le32(ni->blk_addr);
127 	raw_ne->version = ni->version;
128 }
129 
130 static inline bool excess_dirty_nats(struct f2fs_sb_info *sbi)
131 {
132 	/* nat_cnt[] is heuristic accounting sampled locklessly here. */
133 	return data_race(READ_ONCE(NM_I(sbi)->nat_cnt[DIRTY_NAT])) >=
134 					NM_I(sbi)->max_nid *
135 					NM_I(sbi)->dirty_nats_ratio / 100;
136 }
137 
138 static inline bool excess_cached_nats(struct f2fs_sb_info *sbi)
139 {
140 	/* nat_cnt[] is heuristic accounting sampled locklessly here. */
141 	return data_race(READ_ONCE(NM_I(sbi)->nat_cnt[TOTAL_NAT])) >=
142 					DEF_NAT_CACHE_THRESHOLD;
143 }
144 
145 enum mem_type {
146 	FREE_NIDS,	/* indicates the free nid list */
147 	NAT_ENTRIES,	/* indicates the cached nat entry */
148 	DIRTY_DENTS,	/* indicates dirty dentry pages */
149 	INO_ENTRIES,	/* indicates inode entries */
150 	READ_EXTENT_CACHE,	/* indicates read extent cache */
151 	AGE_EXTENT_CACHE,	/* indicates age extent cache */
152 	DISCARD_CACHE,	/* indicates memory of cached discard cmds */
153 	COMPRESS_PAGE,	/* indicates memory of cached compressed pages */
154 	BASE_CHECK,	/* check kernel status */
155 };
156 
157 struct nat_entry_set {
158 	struct list_head set_list;	/* link with other nat sets */
159 	struct list_head entry_list;	/* link with dirty nat entries */
160 	nid_t set;			/* set number*/
161 	unsigned int entry_cnt;		/* the # of nat entries in set */
162 };
163 
164 struct free_nid {
165 	struct list_head list;	/* for free node id list */
166 	nid_t nid;		/* node id */
167 	int state;		/* in use or not: FREE_NID or PREALLOC_NID */
168 };
169 
170 static inline void next_free_nid(struct f2fs_sb_info *sbi, nid_t *nid)
171 {
172 	struct f2fs_nm_info *nm_i = NM_I(sbi);
173 	struct free_nid *fnid;
174 
175 	spin_lock(&nm_i->nid_list_lock);
176 	if (nm_i->nid_cnt[FREE_NID] <= 0) {
177 		spin_unlock(&nm_i->nid_list_lock);
178 		return;
179 	}
180 	fnid = list_first_entry(&nm_i->free_nid_list, struct free_nid, list);
181 	*nid = fnid->nid;
182 	spin_unlock(&nm_i->nid_list_lock);
183 }
184 
185 /*
186  * inline functions
187  */
188 static inline void get_nat_bitmap(struct f2fs_sb_info *sbi, void *addr)
189 {
190 	struct f2fs_nm_info *nm_i = NM_I(sbi);
191 
192 #ifdef CONFIG_F2FS_CHECK_FS
193 	if (memcmp(nm_i->nat_bitmap, nm_i->nat_bitmap_mir,
194 						nm_i->bitmap_size))
195 		f2fs_bug_on(sbi, 1);
196 #endif
197 	memcpy(addr, nm_i->nat_bitmap, nm_i->bitmap_size);
198 }
199 
200 static inline pgoff_t current_nat_addr(struct f2fs_sb_info *sbi, nid_t start)
201 {
202 	struct f2fs_nm_info *nm_i = NM_I(sbi);
203 	pgoff_t block_off;
204 	pgoff_t block_addr;
205 
206 	/*
207 	 * block_off = segment_off * 512 + off_in_segment
208 	 * OLD = (segment_off * 512) * 2 + off_in_segment
209 	 * NEW = 2 * (segment_off * 512 + off_in_segment) - off_in_segment
210 	 */
211 	block_off = NAT_BLOCK_OFFSET(start);
212 
213 	block_addr = (pgoff_t)(nm_i->nat_blkaddr +
214 		(block_off << 1) -
215 		(block_off & (BLKS_PER_SEG(sbi) - 1)));
216 
217 	if (f2fs_test_bit(block_off, nm_i->nat_bitmap))
218 		block_addr += BLKS_PER_SEG(sbi);
219 
220 	return block_addr;
221 }
222 
223 static inline pgoff_t next_nat_addr(struct f2fs_sb_info *sbi,
224 						pgoff_t block_addr)
225 {
226 	struct f2fs_nm_info *nm_i = NM_I(sbi);
227 
228 	block_addr -= nm_i->nat_blkaddr;
229 	block_addr ^= BIT(sbi->log_blocks_per_seg);
230 	return block_addr + nm_i->nat_blkaddr;
231 }
232 
233 static inline void set_to_next_nat(struct f2fs_nm_info *nm_i, nid_t start_nid)
234 {
235 	unsigned int block_off = NAT_BLOCK_OFFSET(start_nid);
236 
237 	f2fs_change_bit(block_off, nm_i->nat_bitmap);
238 #ifdef CONFIG_F2FS_CHECK_FS
239 	f2fs_change_bit(block_off, nm_i->nat_bitmap_mir);
240 #endif
241 }
242 
243 static inline nid_t ino_of_node(const struct folio *node_folio)
244 {
245 	struct f2fs_node *rn = F2FS_NODE(node_folio);
246 	return le32_to_cpu(rn->footer.ino);
247 }
248 
249 static inline nid_t nid_of_node(const struct folio *node_folio)
250 {
251 	struct f2fs_node *rn = F2FS_NODE(node_folio);
252 	return le32_to_cpu(rn->footer.nid);
253 }
254 
255 static inline unsigned int ofs_of_node(const struct folio *node_folio)
256 {
257 	struct f2fs_node *rn = F2FS_NODE(node_folio);
258 	unsigned flag = le32_to_cpu(rn->footer.flag);
259 	return flag >> OFFSET_BIT_SHIFT;
260 }
261 
262 static inline __u64 cpver_of_node(const struct folio *node_folio)
263 {
264 	struct f2fs_node *rn = F2FS_NODE(node_folio);
265 	return le64_to_cpu(rn->footer.cp_ver);
266 }
267 
268 static inline block_t next_blkaddr_of_node(const struct folio *node_folio)
269 {
270 	struct f2fs_node *rn = F2FS_NODE(node_folio);
271 	return le32_to_cpu(rn->footer.next_blkaddr);
272 }
273 
274 static inline void fill_node_footer(const struct folio *folio, nid_t nid,
275 				nid_t ino, unsigned int ofs, bool reset)
276 {
277 	struct f2fs_node *rn = F2FS_NODE(folio);
278 	unsigned int old_flag = 0;
279 
280 	if (reset)
281 		memset(rn, 0, sizeof(*rn));
282 	else
283 		old_flag = le32_to_cpu(rn->footer.flag);
284 
285 	rn->footer.nid = cpu_to_le32(nid);
286 	rn->footer.ino = cpu_to_le32(ino);
287 
288 	/* should remain old flag bits such as COLD_BIT_SHIFT */
289 	rn->footer.flag = cpu_to_le32((ofs << OFFSET_BIT_SHIFT) |
290 					(old_flag & OFFSET_BIT_MASK));
291 }
292 
293 static inline void copy_node_footer(const struct folio *dst,
294 		const struct folio *src)
295 {
296 	struct f2fs_node *src_rn = F2FS_NODE(src);
297 	struct f2fs_node *dst_rn = F2FS_NODE(dst);
298 	memcpy(&dst_rn->footer, &src_rn->footer, sizeof(struct node_footer));
299 }
300 
301 static inline void fill_node_footer_blkaddr(struct folio *folio, block_t blkaddr)
302 {
303 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(F2FS_F_SB(folio));
304 	struct f2fs_node *rn = F2FS_NODE(folio);
305 	__u64 cp_ver = cur_cp_version(ckpt);
306 
307 	if (__is_set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG))
308 		cp_ver |= (cur_cp_crc(ckpt) << 32);
309 
310 	rn->footer.cp_ver = cpu_to_le64(cp_ver);
311 	rn->footer.next_blkaddr = cpu_to_le32(blkaddr);
312 }
313 
314 static inline bool is_recoverable_dnode(const struct folio *folio)
315 {
316 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(F2FS_F_SB(folio));
317 	__u64 cp_ver = cur_cp_version(ckpt);
318 
319 	/* Don't care crc part, if fsck.f2fs sets it. */
320 	if (__is_set_ckpt_flags(ckpt, CP_NOCRC_RECOVERY_FLAG))
321 		return (cp_ver << 32) == (cpver_of_node(folio) << 32);
322 
323 	if (__is_set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG))
324 		cp_ver |= (cur_cp_crc(ckpt) << 32);
325 
326 	return cp_ver == cpver_of_node(folio);
327 }
328 
329 /*
330  * f2fs assigns the following node offsets described as (num).
331  * N = NIDS_PER_BLOCK
332  *
333  *  Inode block (0)
334  *    |- direct node (1)
335  *    |- direct node (2)
336  *    |- indirect node (3)
337  *    |            `- direct node (4 => 4 + N - 1)
338  *    |- indirect node (4 + N)
339  *    |            `- direct node (5 + N => 5 + 2N - 1)
340  *    `- double indirect node (5 + 2N)
341  *                 `- indirect node (6 + 2N)
342  *                       `- direct node
343  *                 ......
344  *                 `- indirect node ((6 + 2N) + x(N + 1))
345  *                       `- direct node
346  *                 ......
347  *                 `- indirect node ((6 + 2N) + (N - 1)(N + 1))
348  *                       `- direct node
349  */
350 static inline bool IS_DNODE(const struct folio *node_folio)
351 {
352 	unsigned int ofs = ofs_of_node(node_folio);
353 
354 	if (f2fs_has_xattr_block(ofs))
355 		return true;
356 
357 	if (ofs == 3 || ofs == 4 + NIDS_PER_BLOCK ||
358 			ofs == 5 + 2 * NIDS_PER_BLOCK)
359 		return false;
360 	if (ofs >= 6 + 2 * NIDS_PER_BLOCK) {
361 		ofs -= 6 + 2 * NIDS_PER_BLOCK;
362 		if (!((long int)ofs % (NIDS_PER_BLOCK + 1)))
363 			return false;
364 	}
365 	return true;
366 }
367 
368 static inline int set_nid(struct folio *folio, int off, nid_t nid, bool i)
369 {
370 	struct f2fs_node *rn = F2FS_NODE(folio);
371 
372 	f2fs_folio_wait_writeback(folio, NODE, true, true);
373 
374 	if (i)
375 		rn->i.i_nid[off - NODE_DIR1_BLOCK] = cpu_to_le32(nid);
376 	else
377 		rn->in.nid[off] = cpu_to_le32(nid);
378 	return folio_mark_dirty(folio);
379 }
380 
381 static inline nid_t get_nid(const struct folio *folio, int off, bool i)
382 {
383 	struct f2fs_node *rn = F2FS_NODE(folio);
384 
385 	if (i)
386 		return le32_to_cpu(rn->i.i_nid[off - NODE_DIR1_BLOCK]);
387 	return le32_to_cpu(rn->in.nid[off]);
388 }
389 
390 /*
391  * Coldness identification:
392  *  - Mark cold files in f2fs_inode_info
393  *  - Mark cold node blocks in their node footer
394  *  - Mark cold data pages in page cache
395  */
396 
397 static inline int is_node(const struct folio *folio, int type)
398 {
399 	struct f2fs_node *rn = F2FS_NODE(folio);
400 	return le32_to_cpu(rn->footer.flag) & BIT(type);
401 }
402 
403 #define is_cold_node(folio)	is_node(folio, COLD_BIT_SHIFT)
404 #define is_fsync_dnode(folio)	is_node(folio, FSYNC_BIT_SHIFT)
405 #define is_dent_dnode(folio)	is_node(folio, DENT_BIT_SHIFT)
406 
407 static inline void __set_mark(const struct folio *folio, bool mark, int type)
408 {
409 	struct f2fs_node *rn = F2FS_NODE(folio);
410 	unsigned int flag = le32_to_cpu(rn->footer.flag);
411 
412 	if (mark)
413 		flag |= BIT(type);
414 	else
415 		flag &= ~BIT(type);
416 	rn->footer.flag = cpu_to_le32(flag);
417 }
418 
419 static inline void set_cold_node(const struct folio *folio, bool is_dir)
420 {
421 	__set_mark(folio, !is_dir, COLD_BIT_SHIFT);
422 }
423 
424 static inline void set_mark(struct folio *folio, bool mark, int type)
425 {
426 	__set_mark(folio, mark, type);
427 
428 #ifdef CONFIG_F2FS_CHECK_FS
429 	f2fs_inode_chksum_set(F2FS_F_SB(folio), folio);
430 #endif
431 }
432 #define set_dentry_mark(folio, mark)	set_mark(folio, mark, DENT_BIT_SHIFT)
433 #define set_fsync_mark(folio, mark)	set_mark(folio, mark, FSYNC_BIT_SHIFT)
434