Alex Tomas | a86c618 | 2006-10-11 01:21:03 -0700 | [diff] [blame] | 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 | |
Christoph Hellwig | 3dcf545 | 2008-04-29 18:13:32 -0400 | [diff] [blame] | 19 | #ifndef _EXT4_EXTENTS |
| 20 | #define _EXT4_EXTENTS |
Alex Tomas | a86c618 | 2006-10-11 01:21:03 -0700 | [diff] [blame] | 21 | |
Christoph Hellwig | 3dcf545 | 2008-04-29 18:13:32 -0400 | [diff] [blame] | 22 | #include "ext4.h" |
Alex Tomas | a86c618 | 2006-10-11 01:21:03 -0700 | [diff] [blame] | 23 | |
| 24 | /* |
Robert P. J. Day | bbf2f9f | 2007-02-17 19:20:16 +0100 | [diff] [blame] | 25 | * With AGGRESSIVE_TEST defined, the capacity of index/leaf blocks |
Randy Dunlap | d0d856e | 2006-10-11 01:21:07 -0700 | [diff] [blame] | 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. |
Alex Tomas | a86c618 | 2006-10-11 01:21:03 -0700 | [diff] [blame] | 29 | */ |
Robert P. J. Day | bbf2f9f | 2007-02-17 19:20:16 +0100 | [diff] [blame] | 30 | #define AGGRESSIVE_TEST_ |
Alex Tomas | a86c618 | 2006-10-11 01:21:03 -0700 | [diff] [blame] | 31 | |
| 32 | /* |
Randy Dunlap | d0d856e | 2006-10-11 01:21:07 -0700 | [diff] [blame] | 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. |
Alex Tomas | a86c618 | 2006-10-11 01:21:03 -0700 | [diff] [blame] | 36 | */ |
| 37 | #define EXTENTS_STATS__ |
| 38 | |
| 39 | /* |
Randy Dunlap | d0d856e | 2006-10-11 01:21:07 -0700 | [diff] [blame] | 40 | * If CHECK_BINSEARCH is defined, then the results of the binary search |
| 41 | * will also be checked by linear search. |
Alex Tomas | a86c618 | 2006-10-11 01:21:03 -0700 | [diff] [blame] | 42 | */ |
| 43 | #define CHECK_BINSEARCH__ |
| 44 | |
| 45 | /* |
Mingming | 553f900 | 2009-09-18 13:34:55 -0400 | [diff] [blame] | 46 | * Turn on EXT_DEBUG to get lots of info about extents operations. |
Alex Tomas | a86c618 | 2006-10-11 01:21:03 -0700 | [diff] [blame] | 47 | */ |
| 48 | #define EXT_DEBUG__ |
| 49 | #ifdef EXT_DEBUG |
| 50 | #define ext_debug(a...) printk(a) |
| 51 | #else |
| 52 | #define ext_debug(a...) |
| 53 | #endif |
| 54 | |
| 55 | /* |
Randy Dunlap | d0d856e | 2006-10-11 01:21:07 -0700 | [diff] [blame] | 56 | * If EXT_STATS is defined then stats numbers are collected. |
| 57 | * These number will be displayed at umount time. |
Alex Tomas | a86c618 | 2006-10-11 01:21:03 -0700 | [diff] [blame] | 58 | */ |
| 59 | #define EXT_STATS_ |
| 60 | |
| 61 | |
| 62 | /* |
Randy Dunlap | d0d856e | 2006-10-11 01:21:07 -0700 | [diff] [blame] | 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. |
Alex Tomas | a86c618 | 2006-10-11 01:21:03 -0700 | [diff] [blame] | 66 | */ |
| 67 | |
| 68 | /* |
Randy Dunlap | d0d856e | 2006-10-11 01:21:07 -0700 | [diff] [blame] | 69 | * This is the extent on-disk structure. |
| 70 | * It's used at the bottom of the tree. |
Alex Tomas | a86c618 | 2006-10-11 01:21:03 -0700 | [diff] [blame] | 71 | */ |
| 72 | struct ext4_extent { |
| 73 | __le32 ee_block; /* first logical block extent covers */ |
| 74 | __le16 ee_len; /* number of blocks covered by extent */ |
| 75 | __le16 ee_start_hi; /* high 16 bits of physical block */ |
Aneesh Kumar K.V | b377611 | 2007-10-16 18:38:25 -0400 | [diff] [blame] | 76 | __le32 ee_start_lo; /* low 32 bits of physical block */ |
Alex Tomas | a86c618 | 2006-10-11 01:21:03 -0700 | [diff] [blame] | 77 | }; |
| 78 | |
| 79 | /* |
Randy Dunlap | d0d856e | 2006-10-11 01:21:07 -0700 | [diff] [blame] | 80 | * This is index on-disk structure. |
| 81 | * It's used at all the levels except the bottom. |
Alex Tomas | a86c618 | 2006-10-11 01:21:03 -0700 | [diff] [blame] | 82 | */ |
| 83 | struct ext4_extent_idx { |
| 84 | __le32 ei_block; /* index covers logical blocks from 'block' */ |
Aneesh Kumar K.V | d8dd0b4 | 2007-10-16 18:38:25 -0400 | [diff] [blame] | 85 | __le32 ei_leaf_lo; /* pointer to the physical block of the next * |
Randy Dunlap | d0d856e | 2006-10-11 01:21:07 -0700 | [diff] [blame] | 86 | * level. leaf or next index could be there */ |
Alex Tomas | a86c618 | 2006-10-11 01:21:03 -0700 | [diff] [blame] | 87 | __le16 ei_leaf_hi; /* high 16 bits of physical block */ |
| 88 | __u16 ei_unused; |
| 89 | }; |
| 90 | |
| 91 | /* |
Randy Dunlap | d0d856e | 2006-10-11 01:21:07 -0700 | [diff] [blame] | 92 | * Each block (leaves and indexes), even inode-stored has header. |
Alex Tomas | a86c618 | 2006-10-11 01:21:03 -0700 | [diff] [blame] | 93 | */ |
| 94 | struct ext4_extent_header { |
| 95 | __le16 eh_magic; /* probably will support different formats */ |
| 96 | __le16 eh_entries; /* number of valid entries */ |
| 97 | __le16 eh_max; /* capacity of store in entries */ |
Randy Dunlap | d0d856e | 2006-10-11 01:21:07 -0700 | [diff] [blame] | 98 | __le16 eh_depth; /* has tree real underlying blocks? */ |
Alex Tomas | a86c618 | 2006-10-11 01:21:03 -0700 | [diff] [blame] | 99 | __le32 eh_generation; /* generation of the tree */ |
| 100 | }; |
| 101 | |
| 102 | #define EXT4_EXT_MAGIC cpu_to_le16(0xf30a) |
| 103 | |
| 104 | /* |
Randy Dunlap | d0d856e | 2006-10-11 01:21:07 -0700 | [diff] [blame] | 105 | * Array of ext4_ext_path contains path to some extent. |
| 106 | * Creation/lookup routines use it for traversal/splitting/etc. |
| 107 | * Truncate uses it to simulate recursive walking. |
Alex Tomas | a86c618 | 2006-10-11 01:21:03 -0700 | [diff] [blame] | 108 | */ |
| 109 | struct ext4_ext_path { |
Alex Tomas | f65e6fb | 2006-10-11 01:21:05 -0700 | [diff] [blame] | 110 | ext4_fsblk_t p_block; |
Alex Tomas | a86c618 | 2006-10-11 01:21:03 -0700 | [diff] [blame] | 111 | __u16 p_depth; |
| 112 | struct ext4_extent *p_ext; |
| 113 | struct ext4_extent_idx *p_idx; |
| 114 | struct ext4_extent_header *p_hdr; |
| 115 | struct buffer_head *p_bh; |
| 116 | }; |
| 117 | |
| 118 | /* |
| 119 | * structure for external API |
| 120 | */ |
| 121 | |
| 122 | #define EXT4_EXT_CACHE_NO 0 |
| 123 | #define EXT4_EXT_CACHE_GAP 1 |
| 124 | #define EXT4_EXT_CACHE_EXTENT 2 |
| 125 | |
Eric Sandeen | 6873fa0 | 2008-10-07 00:46:36 -0400 | [diff] [blame] | 126 | /* |
| 127 | * to be called by ext4_ext_walk_space() |
| 128 | * negative retcode - error |
| 129 | * positive retcode - signal for ext4_ext_walk_space(), see below |
| 130 | * callback must return valid extent (passed or newly created) |
| 131 | */ |
| 132 | typedef int (*ext_prepare_callback)(struct inode *, struct ext4_ext_path *, |
| 133 | struct ext4_ext_cache *, |
| 134 | struct ext4_extent *, void *); |
| 135 | |
| 136 | #define EXT_CONTINUE 0 |
| 137 | #define EXT_BREAK 1 |
| 138 | #define EXT_REPEAT 2 |
Alex Tomas | a86c618 | 2006-10-11 01:21:03 -0700 | [diff] [blame] | 139 | |
Eric Sandeen | 0a80e98 | 2009-09-17 11:55:58 -0400 | [diff] [blame] | 140 | /* Maximum logical block in a file; ext4_extent's ee_block is __le32 */ |
Alex Tomas | a86c618 | 2006-10-11 01:21:03 -0700 | [diff] [blame] | 141 | #define EXT_MAX_BLOCK 0xffffffff |
| 142 | |
Amit Arora | 749269f | 2007-07-18 09:02:56 -0400 | [diff] [blame] | 143 | /* |
| 144 | * EXT_INIT_MAX_LEN is the maximum number of blocks we can have in an |
| 145 | * initialized extent. This is 2^15 and not (2^16 - 1), since we use the |
| 146 | * MSB of ee_len field in the extent datastructure to signify if this |
| 147 | * particular extent is an initialized extent or an uninitialized (i.e. |
| 148 | * preallocated). |
| 149 | * EXT_UNINIT_MAX_LEN is the maximum number of blocks we can have in an |
| 150 | * uninitialized extent. |
| 151 | * If ee_len is <= 0x8000, it is an initialized extent. Otherwise, it is an |
| 152 | * uninitialized one. In other words, if MSB of ee_len is set, it is an |
| 153 | * uninitialized extent with only one special scenario when ee_len = 0x8000. |
| 154 | * In this case we can not have an uninitialized extent of zero length and |
| 155 | * thus we make it as a special case of initialized extent with 0x8000 length. |
| 156 | * This way we get better extent-to-group alignment for initialized extents. |
| 157 | * Hence, the maximum number of blocks we can have in an *initialized* |
| 158 | * extent is 2^15 (32768) and in an *uninitialized* extent is 2^15-1 (32767). |
| 159 | */ |
| 160 | #define EXT_INIT_MAX_LEN (1UL << 15) |
| 161 | #define EXT_UNINIT_MAX_LEN (EXT_INIT_MAX_LEN - 1) |
Suparna Bhattacharya | 471d401 | 2006-10-11 01:21:06 -0700 | [diff] [blame] | 162 | |
Alex Tomas | a86c618 | 2006-10-11 01:21:03 -0700 | [diff] [blame] | 163 | |
| 164 | #define EXT_FIRST_EXTENT(__hdr__) \ |
| 165 | ((struct ext4_extent *) (((char *) (__hdr__)) + \ |
| 166 | sizeof(struct ext4_extent_header))) |
| 167 | #define EXT_FIRST_INDEX(__hdr__) \ |
| 168 | ((struct ext4_extent_idx *) (((char *) (__hdr__)) + \ |
| 169 | sizeof(struct ext4_extent_header))) |
| 170 | #define EXT_HAS_FREE_INDEX(__path__) \ |
Dave Kleikamp | 8c55e20 | 2007-05-24 13:04:54 -0400 | [diff] [blame] | 171 | (le16_to_cpu((__path__)->p_hdr->eh_entries) \ |
| 172 | < le16_to_cpu((__path__)->p_hdr->eh_max)) |
Alex Tomas | a86c618 | 2006-10-11 01:21:03 -0700 | [diff] [blame] | 173 | #define EXT_LAST_EXTENT(__hdr__) \ |
| 174 | (EXT_FIRST_EXTENT((__hdr__)) + le16_to_cpu((__hdr__)->eh_entries) - 1) |
| 175 | #define EXT_LAST_INDEX(__hdr__) \ |
| 176 | (EXT_FIRST_INDEX((__hdr__)) + le16_to_cpu((__hdr__)->eh_entries) - 1) |
| 177 | #define EXT_MAX_EXTENT(__hdr__) \ |
| 178 | (EXT_FIRST_EXTENT((__hdr__)) + le16_to_cpu((__hdr__)->eh_max) - 1) |
| 179 | #define EXT_MAX_INDEX(__hdr__) \ |
| 180 | (EXT_FIRST_INDEX((__hdr__)) + le16_to_cpu((__hdr__)->eh_max) - 1) |
| 181 | |
| 182 | static 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 | |
| 187 | static inline struct ext4_extent_header *ext_block_hdr(struct buffer_head *bh) |
| 188 | { |
| 189 | return (struct ext4_extent_header *) bh->b_data; |
| 190 | } |
| 191 | |
| 192 | static inline unsigned short ext_depth(struct inode *inode) |
| 193 | { |
| 194 | return le16_to_cpu(ext_inode_hdr(inode)->eh_depth); |
| 195 | } |
| 196 | |
Alex Tomas | a86c618 | 2006-10-11 01:21:03 -0700 | [diff] [blame] | 197 | static inline void |
| 198 | ext4_ext_invalidate_cache(struct inode *inode) |
| 199 | { |
| 200 | EXT4_I(inode)->i_cached_extent.ec_type = EXT4_EXT_CACHE_NO; |
| 201 | } |
| 202 | |
Amit Arora | a2df2a6 | 2007-07-17 21:42:41 -0400 | [diff] [blame] | 203 | static inline void ext4_ext_mark_uninitialized(struct ext4_extent *ext) |
| 204 | { |
Amit Arora | 749269f | 2007-07-18 09:02:56 -0400 | [diff] [blame] | 205 | /* We can not have an uninitialized extent of zero length! */ |
| 206 | BUG_ON((le16_to_cpu(ext->ee_len) & ~EXT_INIT_MAX_LEN) == 0); |
| 207 | ext->ee_len |= cpu_to_le16(EXT_INIT_MAX_LEN); |
Amit Arora | a2df2a6 | 2007-07-17 21:42:41 -0400 | [diff] [blame] | 208 | } |
| 209 | |
| 210 | static inline int ext4_ext_is_uninitialized(struct ext4_extent *ext) |
| 211 | { |
Amit Arora | 749269f | 2007-07-18 09:02:56 -0400 | [diff] [blame] | 212 | /* Extent with ee_len of 0x8000 is treated as an initialized extent */ |
| 213 | return (le16_to_cpu(ext->ee_len) > EXT_INIT_MAX_LEN); |
Amit Arora | a2df2a6 | 2007-07-17 21:42:41 -0400 | [diff] [blame] | 214 | } |
| 215 | |
| 216 | static inline int ext4_ext_get_actual_len(struct ext4_extent *ext) |
| 217 | { |
Amit Arora | 749269f | 2007-07-18 09:02:56 -0400 | [diff] [blame] | 218 | return (le16_to_cpu(ext->ee_len) <= EXT_INIT_MAX_LEN ? |
| 219 | le16_to_cpu(ext->ee_len) : |
| 220 | (le16_to_cpu(ext->ee_len) - EXT_INIT_MAX_LEN)); |
Amit Arora | a2df2a6 | 2007-07-17 21:42:41 -0400 | [diff] [blame] | 221 | } |
| 222 | |
Mingming Cao | 0031462 | 2009-09-28 15:49:08 -0400 | [diff] [blame] | 223 | static inline void ext4_ext_mark_initialized(struct ext4_extent *ext) |
| 224 | { |
| 225 | ext->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ext)); |
| 226 | } |
| 227 | |
Theodore Ts'o | bf89d16 | 2010-10-27 21:30:14 -0400 | [diff] [blame] | 228 | /* |
| 229 | * ext4_ext_pblock: |
| 230 | * combine low and high parts of physical block number into ext4_fsblk_t |
| 231 | */ |
| 232 | static inline ext4_fsblk_t ext4_ext_pblock(struct ext4_extent *ex) |
| 233 | { |
| 234 | ext4_fsblk_t block; |
| 235 | |
| 236 | block = le32_to_cpu(ex->ee_start_lo); |
| 237 | block |= ((ext4_fsblk_t) le16_to_cpu(ex->ee_start_hi) << 31) << 1; |
| 238 | return block; |
| 239 | } |
| 240 | |
| 241 | /* |
| 242 | * ext4_idx_pblock: |
| 243 | * combine low and high parts of a leaf physical block number into ext4_fsblk_t |
| 244 | */ |
| 245 | static inline ext4_fsblk_t ext4_idx_pblock(struct ext4_extent_idx *ix) |
| 246 | { |
| 247 | ext4_fsblk_t block; |
| 248 | |
| 249 | block = le32_to_cpu(ix->ei_leaf_lo); |
| 250 | block |= ((ext4_fsblk_t) le16_to_cpu(ix->ei_leaf_hi) << 31) << 1; |
| 251 | return block; |
| 252 | } |
| 253 | |
| 254 | /* |
| 255 | * ext4_ext_store_pblock: |
| 256 | * stores a large physical block number into an extent struct, |
| 257 | * breaking it into parts |
| 258 | */ |
| 259 | static inline void ext4_ext_store_pblock(struct ext4_extent *ex, |
| 260 | ext4_fsblk_t pb) |
| 261 | { |
| 262 | ex->ee_start_lo = cpu_to_le32((unsigned long) (pb & 0xffffffff)); |
| 263 | ex->ee_start_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) & |
| 264 | 0xffff); |
| 265 | } |
| 266 | |
| 267 | /* |
| 268 | * ext4_idx_store_pblock: |
| 269 | * stores a large physical block number into an index struct, |
| 270 | * breaking it into parts |
| 271 | */ |
| 272 | static inline void ext4_idx_store_pblock(struct ext4_extent_idx *ix, |
| 273 | ext4_fsblk_t pb) |
| 274 | { |
| 275 | ix->ei_leaf_lo = cpu_to_le32((unsigned long) (pb & 0xffffffff)); |
| 276 | ix->ei_leaf_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) & |
| 277 | 0xffff); |
| 278 | } |
| 279 | |
Theodore Ts'o | 9d0be50 | 2010-01-01 02:41:30 -0500 | [diff] [blame] | 280 | extern int ext4_ext_calc_metadata_amount(struct inode *inode, |
| 281 | sector_t lblocks); |
Alex Tomas | a86c618 | 2006-10-11 01:21:03 -0700 | [diff] [blame] | 282 | extern int ext4_extent_tree_init(handle_t *, struct inode *); |
Mingming Cao | ee12b63 | 2008-08-19 22:16:05 -0400 | [diff] [blame] | 283 | extern int ext4_ext_calc_credits_for_single_extent(struct inode *inode, |
| 284 | int num, |
| 285 | struct ext4_ext_path *path); |
Akira Fujita | 748de67 | 2009-06-17 19:24:03 -0400 | [diff] [blame] | 286 | extern int ext4_can_extents_be_merged(struct inode *inode, |
| 287 | struct ext4_extent *ex1, |
| 288 | struct ext4_extent *ex2); |
Mingming Cao | 0031462 | 2009-09-28 15:49:08 -0400 | [diff] [blame] | 289 | extern int ext4_ext_insert_extent(handle_t *, struct inode *, struct ext4_ext_path *, struct ext4_extent *, int); |
Aneesh Kumar K.V | 725d26d | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 290 | extern struct ext4_ext_path *ext4_ext_find_extent(struct inode *, ext4_lblk_t, |
| 291 | struct ext4_ext_path *); |
Aneesh Kumar K.V | b35905c | 2008-02-25 16:54:37 -0500 | [diff] [blame] | 292 | extern void ext4_ext_drop_refs(struct ext4_ext_path *); |
Aneesh Kumar K.V | 7a262f7 | 2009-03-27 16:39:58 -0400 | [diff] [blame] | 293 | extern int ext4_ext_check_inode(struct inode *inode); |
Christoph Hellwig | 3dcf545 | 2008-04-29 18:13:32 -0400 | [diff] [blame] | 294 | #endif /* _EXT4_EXTENTS */ |
Alex Tomas | a86c618 | 2006-10-11 01:21:03 -0700 | [diff] [blame] | 295 | |