1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com
4 * Written by Alex Tomas <alex@clusterfs.com>
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.
18 #define AGGRESSIVE_TEST_
21 * With EXTENTS_STATS defined, the number of blocks and extents
22 * are collected in the truncate path. They'll be shown at
25 #define EXTENTS_STATS__
28 * If CHECK_BINSEARCH is defined, then the results of the binary search
29 * will also be checked by linear search.
31 #define CHECK_BINSEARCH__
34 * If EXT_STATS is defined then stats numbers are collected.
35 * These number will be displayed at umount time.
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
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.
55 struct ext4_extent_tail
{
56 __le32 et_checksum
; /* crc32c(uuid+inum+extent_block) */
60 * This is the extent on-disk structure.
61 * It's used at the bottom of the tree.
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 */
71 * This is index on-disk structure.
72 * It's used at all the levels except the bottom.
74 struct 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 */
83 * Each block (leaves and indexes), even inode-stored has header.
85 struct 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 */
93 #define EXT4_EXT_MAGIC cpu_to_le16(0xf30a)
95 #define EXT4_EXTENT_TAIL_OFFSET(hdr) \
96 (sizeof(struct ext4_extent_header) + \
97 (sizeof(struct ext4_extent) * le16_to_cpu((hdr)->eh_max)))
99 static inline struct ext4_extent_tail
*
100 find_ext4_extent_tail(struct ext4_extent_header
*eh
)
102 return (struct ext4_extent_tail
*)(((void *)eh
) +
103 EXT4_EXTENT_TAIL_OFFSET(eh
));
107 * Array of ext4_ext_path contains path to some extent.
108 * Creation/lookup routines use it for traversal/splitting/etc.
109 * Truncate uses it to simulate recursive walking.
111 struct ext4_ext_path
{
112 ext4_fsblk_t p_block
;
115 struct ext4_extent
*p_ext
;
116 struct ext4_extent_idx
*p_idx
;
117 struct ext4_extent_header
*p_hdr
;
118 struct buffer_head
*p_bh
;
122 * structure for external API
126 * EXT_INIT_MAX_LEN is the maximum number of blocks we can have in an
127 * initialized extent. This is 2^15 and not (2^16 - 1), since we use the
128 * MSB of ee_len field in the extent datastructure to signify if this
129 * particular extent is an initialized extent or an unwritten (i.e.
131 * EXT_UNWRITTEN_MAX_LEN is the maximum number of blocks we can have in an
133 * If ee_len is <= 0x8000, it is an initialized extent. Otherwise, it is an
134 * unwritten one. In other words, if MSB of ee_len is set, it is an
135 * unwritten extent with only one special scenario when ee_len = 0x8000.
136 * In this case we can not have an unwritten extent of zero length and
137 * thus we make it as a special case of initialized extent with 0x8000 length.
138 * This way we get better extent-to-group alignment for initialized extents.
139 * Hence, the maximum number of blocks we can have in an *initialized*
140 * extent is 2^15 (32768) and in an *unwritten* extent is 2^15-1 (32767).
142 #define EXT_INIT_MAX_LEN (1UL << 15)
143 #define EXT_UNWRITTEN_MAX_LEN (EXT_INIT_MAX_LEN - 1)
146 #define EXT_FIRST_EXTENT(__hdr__) \
147 ((struct ext4_extent *) (((char *) (__hdr__)) + \
148 sizeof(struct ext4_extent_header)))
149 #define EXT_FIRST_INDEX(__hdr__) \
150 ((struct ext4_extent_idx *) (((char *) (__hdr__)) + \
151 sizeof(struct ext4_extent_header)))
152 #define EXT_HAS_FREE_INDEX(__path__) \
153 (le16_to_cpu((__path__)->p_hdr->eh_entries) \
154 < le16_to_cpu((__path__)->p_hdr->eh_max))
155 #define EXT_LAST_EXTENT(__hdr__) \
156 (EXT_FIRST_EXTENT((__hdr__)) + le16_to_cpu((__hdr__)->eh_entries) - 1)
157 #define EXT_LAST_INDEX(__hdr__) \
158 (EXT_FIRST_INDEX((__hdr__)) + le16_to_cpu((__hdr__)->eh_entries) - 1)
159 #define EXT_MAX_EXTENT(__hdr__) \
160 (EXT_FIRST_EXTENT((__hdr__)) + le16_to_cpu((__hdr__)->eh_max) - 1)
161 #define EXT_MAX_INDEX(__hdr__) \
162 (EXT_FIRST_INDEX((__hdr__)) + le16_to_cpu((__hdr__)->eh_max) - 1)
164 static inline struct ext4_extent_header
*ext_inode_hdr(struct inode
*inode
)
166 return (struct ext4_extent_header
*) EXT4_I(inode
)->i_data
;
169 static inline struct ext4_extent_header
*ext_block_hdr(struct buffer_head
*bh
)
171 return (struct ext4_extent_header
*) bh
->b_data
;
174 static inline unsigned short ext_depth(struct inode
*inode
)
176 return le16_to_cpu(ext_inode_hdr(inode
)->eh_depth
);
179 static inline void ext4_ext_mark_unwritten(struct ext4_extent
*ext
)
181 /* We can not have an unwritten extent of zero length! */
182 BUG_ON((le16_to_cpu(ext
->ee_len
) & ~EXT_INIT_MAX_LEN
) == 0);
183 ext
->ee_len
|= cpu_to_le16(EXT_INIT_MAX_LEN
);
186 static inline int ext4_ext_is_unwritten(struct ext4_extent
*ext
)
188 /* Extent with ee_len of 0x8000 is treated as an initialized extent */
189 return (le16_to_cpu(ext
->ee_len
) > EXT_INIT_MAX_LEN
);
192 static inline int ext4_ext_get_actual_len(struct ext4_extent
*ext
)
194 return (le16_to_cpu(ext
->ee_len
) <= EXT_INIT_MAX_LEN
?
195 le16_to_cpu(ext
->ee_len
) :
196 (le16_to_cpu(ext
->ee_len
) - EXT_INIT_MAX_LEN
));
199 static inline void ext4_ext_mark_initialized(struct ext4_extent
*ext
)
201 ext
->ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ext
));
206 * combine low and high parts of physical block number into ext4_fsblk_t
208 static inline ext4_fsblk_t
ext4_ext_pblock(struct ext4_extent
*ex
)
212 block
= le32_to_cpu(ex
->ee_start_lo
);
213 block
|= ((ext4_fsblk_t
) le16_to_cpu(ex
->ee_start_hi
) << 31) << 1;
219 * combine low and high parts of a leaf physical block number into ext4_fsblk_t
221 static inline ext4_fsblk_t
ext4_idx_pblock(struct ext4_extent_idx
*ix
)
225 block
= le32_to_cpu(ix
->ei_leaf_lo
);
226 block
|= ((ext4_fsblk_t
) le16_to_cpu(ix
->ei_leaf_hi
) << 31) << 1;
231 * ext4_ext_store_pblock:
232 * stores a large physical block number into an extent struct,
233 * breaking it into parts
235 static inline void ext4_ext_store_pblock(struct ext4_extent
*ex
,
238 ex
->ee_start_lo
= cpu_to_le32((unsigned long) (pb
& 0xffffffff));
239 ex
->ee_start_hi
= cpu_to_le16((unsigned long) ((pb
>> 31) >> 1) &
244 * ext4_idx_store_pblock:
245 * stores a large physical block number into an index struct,
246 * breaking it into parts
248 static inline void ext4_idx_store_pblock(struct ext4_extent_idx
*ix
,
251 ix
->ei_leaf_lo
= cpu_to_le32((unsigned long) (pb
& 0xffffffff));
252 ix
->ei_leaf_hi
= cpu_to_le16((unsigned long) ((pb
>> 31) >> 1) &
256 #define ext4_ext_dirty(handle, inode, path) \
257 __ext4_ext_dirty(__func__, __LINE__, (handle), (inode), (path))
258 int __ext4_ext_dirty(const char *where
, unsigned int line
, handle_t
*handle
,
259 struct inode
*inode
, struct ext4_ext_path
*path
);
261 #endif /* _EXT4_EXTENTS */