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 * ext4_inode has i_block array (60 bytes total).
35 * The first 12 bytes store ext4_extent_header;
36 * the remainder stores an array of ext4_extent.
37 * For non-inode extent blocks, ext4_extent_tail
38 * follows the array.
39 */
40
41 /*
42 * This is the extent tail on-disk structure.
43 * All other extent structures are 12 bytes long. It turns out that
44 * block_size % 12 >= 4 for at least all powers of 2 greater than 512, which
45 * covers all valid ext4 block sizes. Therefore, this tail structure can be
46 * crammed into the end of the block without having to rebalance the tree.
47 */
48 struct ext4_extent_tail {
49 __le32 et_checksum; /* crc32c(uuid+inum+extent_block) */
50 };
51
52 /*
53 * This is the extent on-disk structure.
54 * It's used at the bottom of the tree.
55 */
56 struct ext4_extent {
57 __le32 ee_block; /* first logical block extent covers */
58 __le16 ee_len; /* number of blocks covered by extent */
59 __le16 ee_start_hi; /* high 16 bits of physical block */
60 __le32 ee_start_lo; /* low 32 bits of physical block */
61 };
62
63 /*
64 * This is index on-disk structure.
65 * It's used at all the levels except the bottom.
66 */
67 struct ext4_extent_idx {
68 __le32 ei_block; /* index covers logical blocks from 'block' */
69 __le32 ei_leaf_lo; /* pointer to the physical block of the next *
70 * level. leaf or next index could be there */
71 __le16 ei_leaf_hi; /* high 16 bits of physical block */
72 __u16 ei_unused;
73 };
74
75 /*
76 * Each block (leaves and indexes), even inode-stored has header.
77 */
78 struct ext4_extent_header {
79 __le16 eh_magic; /* probably will support different formats */
80 __le16 eh_entries; /* number of valid entries */
81 __le16 eh_max; /* capacity of store in entries */
82 __le16 eh_depth; /* has tree real underlying blocks? */
83 __le32 eh_generation; /* generation of the tree */
84 };
85
86 #define EXT4_EXT_MAGIC cpu_to_le16(0xf30a)
87 #define EXT4_MAX_EXTENT_DEPTH 5
88
89 #define EXT4_EXTENT_TAIL_OFFSET(hdr) \
90 (sizeof(struct ext4_extent_header) + \
91 (sizeof(struct ext4_extent) * le16_to_cpu((hdr)->eh_max)))
92
93 static inline struct ext4_extent_tail *
find_ext4_extent_tail(struct ext4_extent_header * eh)94 find_ext4_extent_tail(struct ext4_extent_header *eh)
95 {
96 return (struct ext4_extent_tail *)(((void *)eh) +
97 EXT4_EXTENT_TAIL_OFFSET(eh));
98 }
99
100 /*
101 * Array of ext4_ext_path contains path to some extent.
102 * Creation/lookup routines use it for traversal/splitting/etc.
103 * Truncate uses it to simulate recursive walking.
104 */
105 struct ext4_ext_path {
106 ext4_fsblk_t p_block;
107 __u16 p_depth;
108 __u16 p_maxdepth;
109 struct ext4_extent *p_ext;
110 struct ext4_extent_idx *p_idx;
111 struct ext4_extent_header *p_hdr;
112 struct buffer_head *p_bh;
113 };
114
115 /*
116 * Used to record a portion of a cluster found at the beginning or end
117 * of an extent while traversing the extent tree during space removal.
118 * A partial cluster may be removed if it does not contain blocks shared
119 * with extents that aren't being deleted (tofree state). Otherwise,
120 * it cannot be removed (nofree state).
121 */
122 struct partial_cluster {
123 ext4_fsblk_t pclu; /* physical cluster number */
124 ext4_lblk_t lblk; /* logical block number within logical cluster */
125 enum {initial, tofree, nofree} state;
126 };
127
128 /*
129 * structure for external API
130 */
131
132 /*
133 * EXT_INIT_MAX_LEN is the maximum number of blocks we can have in an
134 * initialized extent. This is 2^15 and not (2^16 - 1), since we use the
135 * MSB of ee_len field in the extent datastructure to signify if this
136 * particular extent is an initialized extent or an unwritten (i.e.
137 * preallocated).
138 * EXT_UNWRITTEN_MAX_LEN is the maximum number of blocks we can have in an
139 * unwritten extent.
140 * If ee_len is <= 0x8000, it is an initialized extent. Otherwise, it is an
141 * unwritten one. In other words, if MSB of ee_len is set, it is an
142 * unwritten extent with only one special scenario when ee_len = 0x8000.
143 * In this case we can not have an unwritten extent of zero length and
144 * thus we make it as a special case of initialized extent with 0x8000 length.
145 * This way we get better extent-to-group alignment for initialized extents.
146 * Hence, the maximum number of blocks we can have in an *initialized*
147 * extent is 2^15 (32768) and in an *unwritten* extent is 2^15-1 (32767).
148 */
149 #define EXT_INIT_MAX_LEN (1UL << 15)
150 #define EXT_UNWRITTEN_MAX_LEN (EXT_INIT_MAX_LEN - 1)
151
152
153 #define EXT_FIRST_EXTENT(__hdr__) \
154 ((struct ext4_extent *) (((char *) (__hdr__)) + \
155 sizeof(struct ext4_extent_header)))
156 #define EXT_FIRST_INDEX(__hdr__) \
157 ((struct ext4_extent_idx *) (((char *) (__hdr__)) + \
158 sizeof(struct ext4_extent_header)))
159 #define EXT_HAS_FREE_INDEX(__path__) \
160 (le16_to_cpu((__path__)->p_hdr->eh_entries) \
161 < le16_to_cpu((__path__)->p_hdr->eh_max))
162 #define EXT_LAST_EXTENT(__hdr__) \
163 (EXT_FIRST_EXTENT((__hdr__)) + le16_to_cpu((__hdr__)->eh_entries) - 1)
164 #define EXT_LAST_INDEX(__hdr__) \
165 (EXT_FIRST_INDEX((__hdr__)) + le16_to_cpu((__hdr__)->eh_entries) - 1)
166 #define EXT_MAX_EXTENT(__hdr__) \
167 ((le16_to_cpu((__hdr__)->eh_max)) ? \
168 ((EXT_FIRST_EXTENT((__hdr__)) + le16_to_cpu((__hdr__)->eh_max) - 1)) \
169 : NULL)
170 #define EXT_MAX_INDEX(__hdr__) \
171 ((le16_to_cpu((__hdr__)->eh_max)) ? \
172 ((EXT_FIRST_INDEX((__hdr__)) + le16_to_cpu((__hdr__)->eh_max) - 1)) \
173 : NULL)
174
ext_inode_hdr(struct inode * inode)175 static inline struct ext4_extent_header *ext_inode_hdr(struct inode *inode)
176 {
177 return (struct ext4_extent_header *) EXT4_I(inode)->i_data;
178 }
179
ext_block_hdr(struct buffer_head * bh)180 static inline struct ext4_extent_header *ext_block_hdr(struct buffer_head *bh)
181 {
182 return (struct ext4_extent_header *) bh->b_data;
183 }
184
ext_depth(struct inode * inode)185 static inline unsigned short ext_depth(struct inode *inode)
186 {
187 return le16_to_cpu(ext_inode_hdr(inode)->eh_depth);
188 }
189
ext4_ext_mark_unwritten(struct ext4_extent * ext)190 static inline void ext4_ext_mark_unwritten(struct ext4_extent *ext)
191 {
192 /* We can not have an unwritten extent of zero length! */
193 BUG_ON((le16_to_cpu(ext->ee_len) & ~EXT_INIT_MAX_LEN) == 0);
194 ext->ee_len |= cpu_to_le16(EXT_INIT_MAX_LEN);
195 }
196
ext4_ext_is_unwritten(struct ext4_extent * ext)197 static inline int ext4_ext_is_unwritten(struct ext4_extent *ext)
198 {
199 /* Extent with ee_len of 0x8000 is treated as an initialized extent */
200 return (le16_to_cpu(ext->ee_len) > EXT_INIT_MAX_LEN);
201 }
202
ext4_ext_get_actual_len(struct ext4_extent * ext)203 static inline int ext4_ext_get_actual_len(struct ext4_extent *ext)
204 {
205 return (le16_to_cpu(ext->ee_len) <= EXT_INIT_MAX_LEN ?
206 le16_to_cpu(ext->ee_len) :
207 (le16_to_cpu(ext->ee_len) - EXT_INIT_MAX_LEN));
208 }
209
ext4_ext_mark_initialized(struct ext4_extent * ext)210 static inline void ext4_ext_mark_initialized(struct ext4_extent *ext)
211 {
212 ext->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ext));
213 }
214
215 /*
216 * ext4_ext_pblock:
217 * combine low and high parts of physical block number into ext4_fsblk_t
218 */
ext4_ext_pblock(struct ext4_extent * ex)219 static inline ext4_fsblk_t ext4_ext_pblock(struct ext4_extent *ex)
220 {
221 ext4_fsblk_t block;
222
223 block = le32_to_cpu(ex->ee_start_lo);
224 block |= ((ext4_fsblk_t) le16_to_cpu(ex->ee_start_hi) << 31) << 1;
225 return block;
226 }
227
228 /*
229 * ext4_idx_pblock:
230 * combine low and high parts of a leaf physical block number into ext4_fsblk_t
231 */
ext4_idx_pblock(struct ext4_extent_idx * ix)232 static inline ext4_fsblk_t ext4_idx_pblock(struct ext4_extent_idx *ix)
233 {
234 ext4_fsblk_t block;
235
236 block = le32_to_cpu(ix->ei_leaf_lo);
237 block |= ((ext4_fsblk_t) le16_to_cpu(ix->ei_leaf_hi) << 31) << 1;
238 return block;
239 }
240
241 /*
242 * ext4_ext_store_pblock:
243 * stores a large physical block number into an extent struct,
244 * breaking it into parts
245 */
ext4_ext_store_pblock(struct ext4_extent * ex,ext4_fsblk_t pb)246 static inline void ext4_ext_store_pblock(struct ext4_extent *ex,
247 ext4_fsblk_t pb)
248 {
249 ex->ee_start_lo = cpu_to_le32((unsigned long) (pb & 0xffffffff));
250 ex->ee_start_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) &
251 0xffff);
252 }
253
254 /*
255 * ext4_idx_store_pblock:
256 * stores a large physical block number into an index struct,
257 * breaking it into parts
258 */
ext4_idx_store_pblock(struct ext4_extent_idx * ix,ext4_fsblk_t pb)259 static inline void ext4_idx_store_pblock(struct ext4_extent_idx *ix,
260 ext4_fsblk_t pb)
261 {
262 ix->ei_leaf_lo = cpu_to_le32((unsigned long) (pb & 0xffffffff));
263 ix->ei_leaf_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) &
264 0xffff);
265 }
266
267 #endif /* _EXT4_EXTENTS */
268
269