Loading...
1/*
2 * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com
3 * Written by Alex Tomas <alex@clusterfs.com>
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License version 2 as
7 * published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public Licens
15 * along with this program; if not, write to the Free Software
16 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-
17 */
18
19#ifndef _EXT4_EXTENTS
20#define _EXT4_EXTENTS
21
22#include "ext4.h"
23
24/*
25 * With AGGRESSIVE_TEST defined, the capacity of index/leaf blocks
26 * becomes very small, so index split, in-depth growing and
27 * other hard changes happen much more often.
28 * This is for debug purposes only.
29 */
30#define AGGRESSIVE_TEST_
31
32/*
33 * With EXTENTS_STATS defined, the number of blocks and extents
34 * are collected in the truncate path. They'll be shown at
35 * umount time.
36 */
37#define EXTENTS_STATS__
38
39/*
40 * If CHECK_BINSEARCH is defined, then the results of the binary search
41 * will also be checked by linear search.
42 */
43#define CHECK_BINSEARCH__
44
45/*
46 * Turn on EXT_DEBUG to get lots of info about extents operations.
47 */
48#define EXT_DEBUG__
49#ifdef EXT_DEBUG
50#define ext_debug(fmt, ...) printk(fmt, ##__VA_ARGS__)
51#else
52#define ext_debug(fmt, ...) no_printk(fmt, ##__VA_ARGS__)
53#endif
54
55/*
56 * If EXT_STATS is defined then stats numbers are collected.
57 * These number will be displayed at umount time.
58 */
59#define EXT_STATS_
60
61
62/*
63 * ext4_inode has i_block array (60 bytes total).
64 * The first 12 bytes store ext4_extent_header;
65 * the remainder stores an array of ext4_extent.
66 * For non-inode extent blocks, ext4_extent_tail
67 * follows the array.
68 */
69
70/*
71 * This is the extent tail on-disk structure.
72 * All other extent structures are 12 bytes long. It turns out that
73 * block_size % 12 >= 4 for at least all powers of 2 greater than 512, which
74 * covers all valid ext4 block sizes. Therefore, this tail structure can be
75 * crammed into the end of the block without having to rebalance the tree.
76 */
77struct ext4_extent_tail {
78 __le32 et_checksum; /* crc32c(uuid+inum+extent_block) */
79};
80
81/*
82 * This is the extent on-disk structure.
83 * It's used at the bottom of the tree.
84 */
85struct ext4_extent {
86 __le32 ee_block; /* first logical block extent covers */
87 __le16 ee_len; /* number of blocks covered by extent */
88 __le16 ee_start_hi; /* high 16 bits of physical block */
89 __le32 ee_start_lo; /* low 32 bits of physical block */
90};
91
92/*
93 * This is index on-disk structure.
94 * It's used at all the levels except the bottom.
95 */
96struct ext4_extent_idx {
97 __le32 ei_block; /* index covers logical blocks from 'block' */
98 __le32 ei_leaf_lo; /* pointer to the physical block of the next *
99 * level. leaf or next index could be there */
100 __le16 ei_leaf_hi; /* high 16 bits of physical block */
101 __u16 ei_unused;
102};
103
104/*
105 * Each block (leaves and indexes), even inode-stored has header.
106 */
107struct ext4_extent_header {
108 __le16 eh_magic; /* probably will support different formats */
109 __le16 eh_entries; /* number of valid entries */
110 __le16 eh_max; /* capacity of store in entries */
111 __le16 eh_depth; /* has tree real underlying blocks? */
112 __le32 eh_generation; /* generation of the tree */
113};
114
115#define EXT4_EXT_MAGIC cpu_to_le16(0xf30a)
116
117#define EXT4_EXTENT_TAIL_OFFSET(hdr) \
118 (sizeof(struct ext4_extent_header) + \
119 (sizeof(struct ext4_extent) * le16_to_cpu((hdr)->eh_max)))
120
121static inline struct ext4_extent_tail *
122find_ext4_extent_tail(struct ext4_extent_header *eh)
123{
124 return (struct ext4_extent_tail *)(((void *)eh) +
125 EXT4_EXTENT_TAIL_OFFSET(eh));
126}
127
128/*
129 * Array of ext4_ext_path contains path to some extent.
130 * Creation/lookup routines use it for traversal/splitting/etc.
131 * Truncate uses it to simulate recursive walking.
132 */
133struct ext4_ext_path {
134 ext4_fsblk_t p_block;
135 __u16 p_depth;
136 struct ext4_extent *p_ext;
137 struct ext4_extent_idx *p_idx;
138 struct ext4_extent_header *p_hdr;
139 struct buffer_head *p_bh;
140};
141
142/*
143 * structure for external API
144 */
145
146/*
147 * to be called by ext4_ext_walk_space()
148 * negative retcode - error
149 * positive retcode - signal for ext4_ext_walk_space(), see below
150 * callback must return valid extent (passed or newly created)
151 */
152typedef int (*ext_prepare_callback)(struct inode *, ext4_lblk_t,
153 struct ext4_ext_cache *,
154 struct ext4_extent *, void *);
155
156#define EXT_CONTINUE 0
157#define EXT_BREAK 1
158#define EXT_REPEAT 2
159
160/*
161 * Maximum number of logical blocks in a file; ext4_extent's ee_block is
162 * __le32.
163 */
164#define EXT_MAX_BLOCKS 0xffffffff
165
166/*
167 * EXT_INIT_MAX_LEN is the maximum number of blocks we can have in an
168 * initialized extent. This is 2^15 and not (2^16 - 1), since we use the
169 * MSB of ee_len field in the extent datastructure to signify if this
170 * particular extent is an initialized extent or an uninitialized (i.e.
171 * preallocated).
172 * EXT_UNINIT_MAX_LEN is the maximum number of blocks we can have in an
173 * uninitialized extent.
174 * If ee_len is <= 0x8000, it is an initialized extent. Otherwise, it is an
175 * uninitialized one. In other words, if MSB of ee_len is set, it is an
176 * uninitialized extent with only one special scenario when ee_len = 0x8000.
177 * In this case we can not have an uninitialized extent of zero length and
178 * thus we make it as a special case of initialized extent with 0x8000 length.
179 * This way we get better extent-to-group alignment for initialized extents.
180 * Hence, the maximum number of blocks we can have in an *initialized*
181 * extent is 2^15 (32768) and in an *uninitialized* extent is 2^15-1 (32767).
182 */
183#define EXT_INIT_MAX_LEN (1UL << 15)
184#define EXT_UNINIT_MAX_LEN (EXT_INIT_MAX_LEN - 1)
185
186
187#define EXT_FIRST_EXTENT(__hdr__) \
188 ((struct ext4_extent *) (((char *) (__hdr__)) + \
189 sizeof(struct ext4_extent_header)))
190#define EXT_FIRST_INDEX(__hdr__) \
191 ((struct ext4_extent_idx *) (((char *) (__hdr__)) + \
192 sizeof(struct ext4_extent_header)))
193#define EXT_HAS_FREE_INDEX(__path__) \
194 (le16_to_cpu((__path__)->p_hdr->eh_entries) \
195 < le16_to_cpu((__path__)->p_hdr->eh_max))
196#define EXT_LAST_EXTENT(__hdr__) \
197 (EXT_FIRST_EXTENT((__hdr__)) + le16_to_cpu((__hdr__)->eh_entries) - 1)
198#define EXT_LAST_INDEX(__hdr__) \
199 (EXT_FIRST_INDEX((__hdr__)) + le16_to_cpu((__hdr__)->eh_entries) - 1)
200#define EXT_MAX_EXTENT(__hdr__) \
201 (EXT_FIRST_EXTENT((__hdr__)) + le16_to_cpu((__hdr__)->eh_max) - 1)
202#define EXT_MAX_INDEX(__hdr__) \
203 (EXT_FIRST_INDEX((__hdr__)) + le16_to_cpu((__hdr__)->eh_max) - 1)
204
205static inline struct ext4_extent_header *ext_inode_hdr(struct inode *inode)
206{
207 return (struct ext4_extent_header *) EXT4_I(inode)->i_data;
208}
209
210static inline struct ext4_extent_header *ext_block_hdr(struct buffer_head *bh)
211{
212 return (struct ext4_extent_header *) bh->b_data;
213}
214
215static inline unsigned short ext_depth(struct inode *inode)
216{
217 return le16_to_cpu(ext_inode_hdr(inode)->eh_depth);
218}
219
220static inline void
221ext4_ext_invalidate_cache(struct inode *inode)
222{
223 EXT4_I(inode)->i_cached_extent.ec_len = 0;
224}
225
226static inline void ext4_ext_mark_uninitialized(struct ext4_extent *ext)
227{
228 /* We can not have an uninitialized extent of zero length! */
229 BUG_ON((le16_to_cpu(ext->ee_len) & ~EXT_INIT_MAX_LEN) == 0);
230 ext->ee_len |= cpu_to_le16(EXT_INIT_MAX_LEN);
231}
232
233static inline int ext4_ext_is_uninitialized(struct ext4_extent *ext)
234{
235 /* Extent with ee_len of 0x8000 is treated as an initialized extent */
236 return (le16_to_cpu(ext->ee_len) > EXT_INIT_MAX_LEN);
237}
238
239static inline int ext4_ext_get_actual_len(struct ext4_extent *ext)
240{
241 return (le16_to_cpu(ext->ee_len) <= EXT_INIT_MAX_LEN ?
242 le16_to_cpu(ext->ee_len) :
243 (le16_to_cpu(ext->ee_len) - EXT_INIT_MAX_LEN));
244}
245
246static inline void ext4_ext_mark_initialized(struct ext4_extent *ext)
247{
248 ext->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ext));
249}
250
251/*
252 * ext4_ext_pblock:
253 * combine low and high parts of physical block number into ext4_fsblk_t
254 */
255static inline ext4_fsblk_t ext4_ext_pblock(struct ext4_extent *ex)
256{
257 ext4_fsblk_t block;
258
259 block = le32_to_cpu(ex->ee_start_lo);
260 block |= ((ext4_fsblk_t) le16_to_cpu(ex->ee_start_hi) << 31) << 1;
261 return block;
262}
263
264/*
265 * ext4_idx_pblock:
266 * combine low and high parts of a leaf physical block number into ext4_fsblk_t
267 */
268static inline ext4_fsblk_t ext4_idx_pblock(struct ext4_extent_idx *ix)
269{
270 ext4_fsblk_t block;
271
272 block = le32_to_cpu(ix->ei_leaf_lo);
273 block |= ((ext4_fsblk_t) le16_to_cpu(ix->ei_leaf_hi) << 31) << 1;
274 return block;
275}
276
277/*
278 * ext4_ext_store_pblock:
279 * stores a large physical block number into an extent struct,
280 * breaking it into parts
281 */
282static inline void ext4_ext_store_pblock(struct ext4_extent *ex,
283 ext4_fsblk_t pb)
284{
285 ex->ee_start_lo = cpu_to_le32((unsigned long) (pb & 0xffffffff));
286 ex->ee_start_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) &
287 0xffff);
288}
289
290/*
291 * ext4_idx_store_pblock:
292 * stores a large physical block number into an index struct,
293 * breaking it into parts
294 */
295static inline void ext4_idx_store_pblock(struct ext4_extent_idx *ix,
296 ext4_fsblk_t pb)
297{
298 ix->ei_leaf_lo = cpu_to_le32((unsigned long) (pb & 0xffffffff));
299 ix->ei_leaf_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) &
300 0xffff);
301}
302
303extern int ext4_ext_calc_metadata_amount(struct inode *inode,
304 ext4_lblk_t lblocks);
305extern int ext4_extent_tree_init(handle_t *, struct inode *);
306extern int ext4_ext_calc_credits_for_single_extent(struct inode *inode,
307 int num,
308 struct ext4_ext_path *path);
309extern int ext4_can_extents_be_merged(struct inode *inode,
310 struct ext4_extent *ex1,
311 struct ext4_extent *ex2);
312extern int ext4_ext_insert_extent(handle_t *, struct inode *, struct ext4_ext_path *, struct ext4_extent *, int);
313extern struct ext4_ext_path *ext4_ext_find_extent(struct inode *, ext4_lblk_t,
314 struct ext4_ext_path *);
315extern void ext4_ext_drop_refs(struct ext4_ext_path *);
316extern int ext4_ext_check_inode(struct inode *inode);
317extern int ext4_find_delalloc_cluster(struct inode *inode, ext4_lblk_t lblk,
318 int search_hint_reverse);
319#endif /* _EXT4_EXTENTS */
320
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com
4 * Written by Alex Tomas <alex@clusterfs.com>
5 */
6
7#ifndef _EXT4_EXTENTS
8#define _EXT4_EXTENTS
9
10#include "ext4.h"
11
12/*
13 * With AGGRESSIVE_TEST defined, the capacity of index/leaf blocks
14 * becomes very small, so index split, in-depth growing and
15 * other hard changes happen much more often.
16 * This is for debug purposes only.
17 */
18#define AGGRESSIVE_TEST_
19
20/*
21 * With EXTENTS_STATS defined, the number of blocks and extents
22 * are collected in the truncate path. They'll be shown at
23 * umount time.
24 */
25#define EXTENTS_STATS__
26
27/*
28 * If CHECK_BINSEARCH is defined, then the results of the binary search
29 * will also be checked by linear search.
30 */
31#define CHECK_BINSEARCH__
32
33/*
34 * If EXT_STATS is defined then stats numbers are collected.
35 * These number will be displayed at umount time.
36 */
37#define EXT_STATS_
38
39
40/*
41 * ext4_inode has i_block array (60 bytes total).
42 * The first 12 bytes store ext4_extent_header;
43 * the remainder stores an array of ext4_extent.
44 * For non-inode extent blocks, ext4_extent_tail
45 * follows the array.
46 */
47
48/*
49 * This is the extent tail on-disk structure.
50 * All other extent structures are 12 bytes long. It turns out that
51 * block_size % 12 >= 4 for at least all powers of 2 greater than 512, which
52 * covers all valid ext4 block sizes. Therefore, this tail structure can be
53 * crammed into the end of the block without having to rebalance the tree.
54 */
55struct ext4_extent_tail {
56 __le32 et_checksum; /* crc32c(uuid+inum+extent_block) */
57};
58
59/*
60 * This is the extent on-disk structure.
61 * It's used at the bottom of the tree.
62 */
63struct ext4_extent {
64 __le32 ee_block; /* first logical block extent covers */
65 __le16 ee_len; /* number of blocks covered by extent */
66 __le16 ee_start_hi; /* high 16 bits of physical block */
67 __le32 ee_start_lo; /* low 32 bits of physical block */
68};
69
70/*
71 * This is index on-disk structure.
72 * It's used at all the levels except the bottom.
73 */
74struct ext4_extent_idx {
75 __le32 ei_block; /* index covers logical blocks from 'block' */
76 __le32 ei_leaf_lo; /* pointer to the physical block of the next *
77 * level. leaf or next index could be there */
78 __le16 ei_leaf_hi; /* high 16 bits of physical block */
79 __u16 ei_unused;
80};
81
82/*
83 * Each block (leaves and indexes), even inode-stored has header.
84 */
85struct ext4_extent_header {
86 __le16 eh_magic; /* probably will support different formats */
87 __le16 eh_entries; /* number of valid entries */
88 __le16 eh_max; /* capacity of store in entries */
89 __le16 eh_depth; /* has tree real underlying blocks? */
90 __le32 eh_generation; /* generation of the tree */
91};
92
93#define EXT4_EXT_MAGIC cpu_to_le16(0xf30a)
94#define EXT4_MAX_EXTENT_DEPTH 5
95
96#define EXT4_EXTENT_TAIL_OFFSET(hdr) \
97 (sizeof(struct ext4_extent_header) + \
98 (sizeof(struct ext4_extent) * le16_to_cpu((hdr)->eh_max)))
99
100static inline struct ext4_extent_tail *
101find_ext4_extent_tail(struct ext4_extent_header *eh)
102{
103 return (struct ext4_extent_tail *)(((void *)eh) +
104 EXT4_EXTENT_TAIL_OFFSET(eh));
105}
106
107/*
108 * Array of ext4_ext_path contains path to some extent.
109 * Creation/lookup routines use it for traversal/splitting/etc.
110 * Truncate uses it to simulate recursive walking.
111 */
112struct ext4_ext_path {
113 ext4_fsblk_t p_block;
114 __u16 p_depth;
115 __u16 p_maxdepth;
116 struct ext4_extent *p_ext;
117 struct ext4_extent_idx *p_idx;
118 struct ext4_extent_header *p_hdr;
119 struct buffer_head *p_bh;
120};
121
122/*
123 * Used to record a portion of a cluster found at the beginning or end
124 * of an extent while traversing the extent tree during space removal.
125 * A partial cluster may be removed if it does not contain blocks shared
126 * with extents that aren't being deleted (tofree state). Otherwise,
127 * it cannot be removed (nofree state).
128 */
129struct partial_cluster {
130 ext4_fsblk_t pclu; /* physical cluster number */
131 ext4_lblk_t lblk; /* logical block number within logical cluster */
132 enum {initial, tofree, nofree} state;
133};
134
135/*
136 * structure for external API
137 */
138
139/*
140 * EXT_INIT_MAX_LEN is the maximum number of blocks we can have in an
141 * initialized extent. This is 2^15 and not (2^16 - 1), since we use the
142 * MSB of ee_len field in the extent datastructure to signify if this
143 * particular extent is an initialized extent or an unwritten (i.e.
144 * preallocated).
145 * EXT_UNWRITTEN_MAX_LEN is the maximum number of blocks we can have in an
146 * unwritten extent.
147 * If ee_len is <= 0x8000, it is an initialized extent. Otherwise, it is an
148 * unwritten one. In other words, if MSB of ee_len is set, it is an
149 * unwritten extent with only one special scenario when ee_len = 0x8000.
150 * In this case we can not have an unwritten extent of zero length and
151 * thus we make it as a special case of initialized extent with 0x8000 length.
152 * This way we get better extent-to-group alignment for initialized extents.
153 * Hence, the maximum number of blocks we can have in an *initialized*
154 * extent is 2^15 (32768) and in an *unwritten* extent is 2^15-1 (32767).
155 */
156#define EXT_INIT_MAX_LEN (1UL << 15)
157#define EXT_UNWRITTEN_MAX_LEN (EXT_INIT_MAX_LEN - 1)
158
159
160#define EXT_FIRST_EXTENT(__hdr__) \
161 ((struct ext4_extent *) (((char *) (__hdr__)) + \
162 sizeof(struct ext4_extent_header)))
163#define EXT_FIRST_INDEX(__hdr__) \
164 ((struct ext4_extent_idx *) (((char *) (__hdr__)) + \
165 sizeof(struct ext4_extent_header)))
166#define EXT_HAS_FREE_INDEX(__path__) \
167 (le16_to_cpu((__path__)->p_hdr->eh_entries) \
168 < le16_to_cpu((__path__)->p_hdr->eh_max))
169#define EXT_LAST_EXTENT(__hdr__) \
170 (EXT_FIRST_EXTENT((__hdr__)) + le16_to_cpu((__hdr__)->eh_entries) - 1)
171#define EXT_LAST_INDEX(__hdr__) \
172 (EXT_FIRST_INDEX((__hdr__)) + le16_to_cpu((__hdr__)->eh_entries) - 1)
173#define EXT_MAX_EXTENT(__hdr__) \
174 ((le16_to_cpu((__hdr__)->eh_max)) ? \
175 ((EXT_FIRST_EXTENT((__hdr__)) + le16_to_cpu((__hdr__)->eh_max) - 1)) \
176 : NULL)
177#define EXT_MAX_INDEX(__hdr__) \
178 ((le16_to_cpu((__hdr__)->eh_max)) ? \
179 ((EXT_FIRST_INDEX((__hdr__)) + le16_to_cpu((__hdr__)->eh_max) - 1)) \
180 : NULL)
181
182static inline struct ext4_extent_header *ext_inode_hdr(struct inode *inode)
183{
184 return (struct ext4_extent_header *) EXT4_I(inode)->i_data;
185}
186
187static inline struct ext4_extent_header *ext_block_hdr(struct buffer_head *bh)
188{
189 return (struct ext4_extent_header *) bh->b_data;
190}
191
192static inline unsigned short ext_depth(struct inode *inode)
193{
194 return le16_to_cpu(ext_inode_hdr(inode)->eh_depth);
195}
196
197static inline void ext4_ext_mark_unwritten(struct ext4_extent *ext)
198{
199 /* We can not have an unwritten extent of zero length! */
200 BUG_ON((le16_to_cpu(ext->ee_len) & ~EXT_INIT_MAX_LEN) == 0);
201 ext->ee_len |= cpu_to_le16(EXT_INIT_MAX_LEN);
202}
203
204static inline int ext4_ext_is_unwritten(struct ext4_extent *ext)
205{
206 /* Extent with ee_len of 0x8000 is treated as an initialized extent */
207 return (le16_to_cpu(ext->ee_len) > EXT_INIT_MAX_LEN);
208}
209
210static inline int ext4_ext_get_actual_len(struct ext4_extent *ext)
211{
212 return (le16_to_cpu(ext->ee_len) <= EXT_INIT_MAX_LEN ?
213 le16_to_cpu(ext->ee_len) :
214 (le16_to_cpu(ext->ee_len) - EXT_INIT_MAX_LEN));
215}
216
217static inline void ext4_ext_mark_initialized(struct ext4_extent *ext)
218{
219 ext->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ext));
220}
221
222/*
223 * ext4_ext_pblock:
224 * combine low and high parts of physical block number into ext4_fsblk_t
225 */
226static inline ext4_fsblk_t ext4_ext_pblock(struct ext4_extent *ex)
227{
228 ext4_fsblk_t block;
229
230 block = le32_to_cpu(ex->ee_start_lo);
231 block |= ((ext4_fsblk_t) le16_to_cpu(ex->ee_start_hi) << 31) << 1;
232 return block;
233}
234
235/*
236 * ext4_idx_pblock:
237 * combine low and high parts of a leaf physical block number into ext4_fsblk_t
238 */
239static inline ext4_fsblk_t ext4_idx_pblock(struct ext4_extent_idx *ix)
240{
241 ext4_fsblk_t block;
242
243 block = le32_to_cpu(ix->ei_leaf_lo);
244 block |= ((ext4_fsblk_t) le16_to_cpu(ix->ei_leaf_hi) << 31) << 1;
245 return block;
246}
247
248/*
249 * ext4_ext_store_pblock:
250 * stores a large physical block number into an extent struct,
251 * breaking it into parts
252 */
253static inline void ext4_ext_store_pblock(struct ext4_extent *ex,
254 ext4_fsblk_t pb)
255{
256 ex->ee_start_lo = cpu_to_le32((unsigned long) (pb & 0xffffffff));
257 ex->ee_start_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) &
258 0xffff);
259}
260
261/*
262 * ext4_idx_store_pblock:
263 * stores a large physical block number into an index struct,
264 * breaking it into parts
265 */
266static inline void ext4_idx_store_pblock(struct ext4_extent_idx *ix,
267 ext4_fsblk_t pb)
268{
269 ix->ei_leaf_lo = cpu_to_le32((unsigned long) (pb & 0xffffffff));
270 ix->ei_leaf_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) &
271 0xffff);
272}
273
274#endif /* _EXT4_EXTENTS */
275