| 1 |
// SPDX-License-Identifier: LGPL-2.1 |
| 2 |
/* |
| 3 |
* Copyright (c) 2012 Taobao. |
| 4 |
* Written by Tao Ma <[email protected]> |
| 5 |
*/ |
| 6 |
|
| 7 |
#include <linux/iomap.h> |
| 8 |
#include <linux/fiemap.h> |
| 9 |
#include <linux/namei.h> |
| 10 |
#include <linux/iversion.h> |
| 11 |
#include <linux/sched/mm.h> |
| 12 |
|
| 13 |
#include "ext4_jbd2.h" |
| 14 |
#include "ext4.h" |
| 15 |
#include "xattr.h" |
| 16 |
#include "truncate.h" |
| 17 |
|
| 18 |
#define EXT4_XATTR_SYSTEM_DATA "data" |
| 19 |
#define EXT4_MIN_INLINE_DATA_SIZE ((sizeof(__le32) * EXT4_N_BLOCKS)) |
| 20 |
#define EXT4_INLINE_DOTDOT_OFFSET 2 |
| 21 |
#define EXT4_INLINE_DOTDOT_SIZE 4 |
| 22 |
|
| 23 |
|
| 24 |
static int ext4_da_convert_inline_data_to_extent(struct address_space *mapping, |
| 25 |
struct inode *inode, |
| 26 |
void **fsdata); |
| 27 |
|
| 28 |
static int ext4_get_inline_size(struct inode *inode) |
| 29 |
{ |
| 30 |
if (EXT4_I(inode)->i_inline_off) |
| 31 |
return EXT4_I(inode)->i_inline_size; |
| 32 |
|
| 33 |
return 0; |
| 34 |
} |
| 35 |
|
| 36 |
static int get_max_inline_xattr_value_size(struct inode *inode, |
| 37 |
struct ext4_iloc *iloc) |
| 38 |
{ |
| 39 |
struct ext4_xattr_ibody_header *header; |
| 40 |
struct ext4_xattr_entry *entry; |
| 41 |
struct ext4_inode *raw_inode; |
| 42 |
void *end; |
| 43 |
int free, min_offs; |
| 44 |
|
| 45 |
if (!EXT4_INODE_HAS_XATTR_SPACE(inode)) |
| 46 |
return 0; |
| 47 |
|
| 48 |
min_offs = EXT4_SB(inode->i_sb)->s_inode_size - |
| 49 |
EXT4_GOOD_OLD_INODE_SIZE - |
| 50 |
EXT4_I(inode)->i_extra_isize - |
| 51 |
sizeof(struct ext4_xattr_ibody_header); |
| 52 |
|
| 53 |
/* |
| 54 |
* We need to subtract another sizeof(__u32) since an in-inode xattr |
| 55 |
* needs an empty 4 bytes to indicate the gap between the xattr entry |
| 56 |
* and the name/value pair. |
| 57 |
*/ |
| 58 |
if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR)) |
| 59 |
return EXT4_XATTR_SIZE(min_offs - |
| 60 |
EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA)) - |
| 61 |
EXT4_XATTR_ROUND - sizeof(__u32)); |
| 62 |
|
| 63 |
raw_inode = ext4_raw_inode(iloc); |
| 64 |
header = IHDR(inode, raw_inode); |
| 65 |
entry = IFIRST(header); |
| 66 |
end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size; |
| 67 |
|
| 68 |
/* Compute min_offs. */ |
| 69 |
while (!IS_LAST_ENTRY(entry)) { |
| 70 |
void *next = EXT4_XATTR_NEXT(entry); |
| 71 |
|
| 72 |
if (next >= end) { |
| 73 |
EXT4_ERROR_INODE(inode, |
| 74 |
"corrupt xattr in inline inode"); |
| 75 |
return 0; |
| 76 |
} |
| 77 |
if (!entry->e_value_inum && entry->e_value_size) { |
| 78 |
size_t offs = le16_to_cpu(entry->e_value_offs); |
| 79 |
if (offs < min_offs) |
| 80 |
min_offs = offs; |
| 81 |
} |
| 82 |
entry = next; |
| 83 |
} |
| 84 |
free = min_offs - |
| 85 |
((void *)entry - (void *)IFIRST(header)) - sizeof(__u32); |
| 86 |
|
| 87 |
if (EXT4_I(inode)->i_inline_off) { |
| 88 |
entry = (struct ext4_xattr_entry *) |
| 89 |
((void *)raw_inode + EXT4_I(inode)->i_inline_off); |
| 90 |
|
| 91 |
free += EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size)); |
| 92 |
goto out; |
| 93 |
} |
| 94 |
|
| 95 |
free -= EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA)); |
| 96 |
|
| 97 |
if (free > EXT4_XATTR_ROUND) |
| 98 |
free = EXT4_XATTR_SIZE(free - EXT4_XATTR_ROUND); |
| 99 |
else |
| 100 |
free = 0; |
| 101 |
|
| 102 |
out: |
| 103 |
return free; |
| 104 |
} |
| 105 |
|
| 106 |
/* |
| 107 |
* Get the maximum size we now can store in an inode. |
| 108 |
* If we can't find the space for a xattr entry, don't use the space |
| 109 |
* of the extents since we have no space to indicate the inline data. |
| 110 |
*/ |
| 111 |
int ext4_get_max_inline_size(struct inode *inode) |
| 112 |
{ |
| 113 |
int error, max_inline_size; |
| 114 |
struct ext4_iloc iloc; |
| 115 |
|
| 116 |
if (EXT4_I(inode)->i_extra_isize == 0) |
| 117 |
return 0; |
| 118 |
|
| 119 |
error = ext4_get_inode_loc(inode, &iloc); |
| 120 |
if (error) { |
| 121 |
ext4_error_inode_err(inode, __func__, __LINE__, 0, -error, |
| 122 |
"can't get inode location %lu", |
| 123 |
inode->i_ino); |
| 124 |
return 0; |
| 125 |
} |
| 126 |
|
| 127 |
down_read(&EXT4_I(inode)->xattr_sem); |
| 128 |
max_inline_size = get_max_inline_xattr_value_size(inode, &iloc); |
| 129 |
up_read(&EXT4_I(inode)->xattr_sem); |
| 130 |
|
| 131 |
brelse(iloc.bh); |
| 132 |
|
| 133 |
if (!max_inline_size) |
| 134 |
return 0; |
| 135 |
|
| 136 |
return max_inline_size + EXT4_MIN_INLINE_DATA_SIZE; |
| 137 |
} |
| 138 |
|
| 139 |
/* |
| 140 |
* this function does not take xattr_sem, which is OK because it is |
| 141 |
* currently only used in a code path coming form ext4_iget, before |
| 142 |
* the new inode has been unlocked |
| 143 |
*/ |
| 144 |
int ext4_find_inline_data_nolock(struct inode *inode) |
| 145 |
{ |
| 146 |
struct ext4_xattr_ibody_find is = { |
| 147 |
.s = { .not_found = -ENODATA, }, |
| 148 |
}; |
| 149 |
struct ext4_xattr_info i = { |
| 150 |
.name_index = EXT4_XATTR_INDEX_SYSTEM, |
| 151 |
.name = EXT4_XATTR_SYSTEM_DATA, |
| 152 |
}; |
| 153 |
int error; |
| 154 |
|
| 155 |
if (EXT4_I(inode)->i_extra_isize == 0) |
| 156 |
return 0; |
| 157 |
|
| 158 |
error = ext4_get_inode_loc(inode, &is.iloc); |
| 159 |
if (error) |
| 160 |
return error; |
| 161 |
|
| 162 |
error = ext4_xattr_ibody_find(inode, &i, &is); |
| 163 |
if (error) |
| 164 |
goto out; |
| 165 |
|
| 166 |
if (!is.s.not_found) { |
| 167 |
if (is.s.here->e_value_inum) { |
| 168 |
EXT4_ERROR_INODE(inode, "inline data xattr refers " |
| 169 |
"to an external xattr inode"); |
| 170 |
error = -EFSCORRUPTED; |
| 171 |
goto out; |
| 172 |
} |
| 173 |
EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here - |
| 174 |
(void *)ext4_raw_inode(&is.iloc)); |
| 175 |
EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE + |
| 176 |
le32_to_cpu(is.s.here->e_value_size); |
| 177 |
} |
| 178 |
out: |
| 179 |
brelse(is.iloc.bh); |
| 180 |
return error; |
| 181 |
} |
| 182 |
|
| 183 |
static int ext4_read_inline_data(struct inode *inode, void *buffer, |
| 184 |
unsigned int len, |
| 185 |
struct ext4_iloc *iloc) |
| 186 |
{ |
| 187 |
struct ext4_xattr_entry *entry; |
| 188 |
struct ext4_xattr_ibody_header *header; |
| 189 |
int cp_len = 0; |
| 190 |
struct ext4_inode *raw_inode; |
| 191 |
|
| 192 |
if (!len) |
| 193 |
return 0; |
| 194 |
|
| 195 |
BUG_ON(len > EXT4_I(inode)->i_inline_size); |
| 196 |
|
| 197 |
cp_len = min_t(unsigned int, len, EXT4_MIN_INLINE_DATA_SIZE); |
| 198 |
|
| 199 |
raw_inode = ext4_raw_inode(iloc); |
| 200 |
memcpy(buffer, (void *)(raw_inode->i_block), cp_len); |
| 201 |
|
| 202 |
len -= cp_len; |
| 203 |
buffer += cp_len; |
| 204 |
|
| 205 |
if (!len) |
| 206 |
goto out; |
| 207 |
|
| 208 |
header = IHDR(inode, raw_inode); |
| 209 |
entry = (struct ext4_xattr_entry *)((void *)raw_inode + |
| 210 |
EXT4_I(inode)->i_inline_off); |
| 211 |
len = min_t(unsigned int, len, |
| 212 |
(unsigned int)le32_to_cpu(entry->e_value_size)); |
| 213 |
|
| 214 |
memcpy(buffer, |
| 215 |
(void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs), len); |
| 216 |
cp_len += len; |
| 217 |
|
| 218 |
out: |
| 219 |
return cp_len; |
| 220 |
} |
| 221 |
|
| 222 |
/* |
| 223 |
* write the buffer to the inline inode. |
| 224 |
* If 'create' is set, we don't need to do the extra copy in the xattr |
| 225 |
* value since it is already handled by ext4_xattr_ibody_set. |
| 226 |
* That saves us one memcpy. |
| 227 |
*/ |
| 228 |
static void ext4_write_inline_data(struct inode *inode, struct ext4_iloc *iloc, |
| 229 |
void *buffer, loff_t pos, unsigned int len) |
| 230 |
{ |
| 231 |
struct ext4_xattr_entry *entry; |
| 232 |
struct ext4_xattr_ibody_header *header; |
| 233 |
struct ext4_inode *raw_inode; |
| 234 |
int cp_len = 0; |
| 235 |
|
| 236 |
if (unlikely(ext4_emergency_state(inode->i_sb))) |
| 237 |
return; |
| 238 |
|
| 239 |
BUG_ON(!EXT4_I(inode)->i_inline_off); |
| 240 |
BUG_ON(pos + len > EXT4_I(inode)->i_inline_size); |
| 241 |
|
| 242 |
raw_inode = ext4_raw_inode(iloc); |
| 243 |
buffer += pos; |
| 244 |
|
| 245 |
if (pos < EXT4_MIN_INLINE_DATA_SIZE) { |
| 246 |
cp_len = pos + len > EXT4_MIN_INLINE_DATA_SIZE ? |
| 247 |
EXT4_MIN_INLINE_DATA_SIZE - pos : len; |
| 248 |
memcpy((void *)raw_inode->i_block + pos, buffer, cp_len); |
| 249 |
|
| 250 |
len -= cp_len; |
| 251 |
buffer += cp_len; |
| 252 |
pos += cp_len; |
| 253 |
} |
| 254 |
|
| 255 |
if (!len) |
| 256 |
return; |
| 257 |
|
| 258 |
pos -= EXT4_MIN_INLINE_DATA_SIZE; |
| 259 |
header = IHDR(inode, raw_inode); |
| 260 |
entry = (struct ext4_xattr_entry *)((void *)raw_inode + |
| 261 |
EXT4_I(inode)->i_inline_off); |
| 262 |
|
| 263 |
memcpy((void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs) + pos, |
| 264 |
buffer, len); |
| 265 |
} |
| 266 |
|
| 267 |
static int ext4_create_inline_data(handle_t *handle, |
| 268 |
struct inode *inode, unsigned len) |
| 269 |
{ |
| 270 |
int error; |
| 271 |
void *value = NULL; |
| 272 |
struct ext4_xattr_ibody_find is = { |
| 273 |
.s = { .not_found = -ENODATA, }, |
| 274 |
}; |
| 275 |
struct ext4_xattr_info i = { |
| 276 |
.name_index = EXT4_XATTR_INDEX_SYSTEM, |
| 277 |
.name = EXT4_XATTR_SYSTEM_DATA, |
| 278 |
}; |
| 279 |
|
| 280 |
error = ext4_get_inode_loc(inode, &is.iloc); |
| 281 |
if (error) |
| 282 |
return error; |
| 283 |
|
| 284 |
BUFFER_TRACE(is.iloc.bh, "get_write_access"); |
| 285 |
error = ext4_journal_get_write_access(handle, inode->i_sb, is.iloc.bh, |
| 286 |
EXT4_JTR_NONE); |
| 287 |
if (error) |
| 288 |
goto out; |
| 289 |
|
| 290 |
if (len > EXT4_MIN_INLINE_DATA_SIZE) { |
| 291 |
value = EXT4_ZERO_XATTR_VALUE; |
| 292 |
len -= EXT4_MIN_INLINE_DATA_SIZE; |
| 293 |
} else { |
| 294 |
value = ""; |
| 295 |
len = 0; |
| 296 |
} |
| 297 |
|
| 298 |
/* Insert the xttr entry. */ |
| 299 |
i.value = value; |
| 300 |
i.value_len = len; |
| 301 |
|
| 302 |
error = ext4_xattr_ibody_find(inode, &i, &is); |
| 303 |
if (error) |
| 304 |
goto out; |
| 305 |
|
| 306 |
if (!is.s.not_found) { |
| 307 |
EXT4_ERROR_INODE(inode, "unexpected inline data xattr"); |
| 308 |
error = -EFSCORRUPTED; |
| 309 |
goto out; |
| 310 |
} |
| 311 |
|
| 312 |
error = ext4_xattr_ibody_set(handle, inode, &i, &is); |
| 313 |
if (error) { |
| 314 |
if (error == -ENOSPC) |
| 315 |
ext4_clear_inode_state(inode, |
| 316 |
EXT4_STATE_MAY_INLINE_DATA); |
| 317 |
goto out; |
| 318 |
} |
| 319 |
|
| 320 |
memset((void *)ext4_raw_inode(&is.iloc)->i_block, |
| 321 |
0, EXT4_MIN_INLINE_DATA_SIZE); |
| 322 |
|
| 323 |
EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here - |
| 324 |
(void *)ext4_raw_inode(&is.iloc)); |
| 325 |
EXT4_I(inode)->i_inline_size = len + EXT4_MIN_INLINE_DATA_SIZE; |
| 326 |
ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS); |
| 327 |
ext4_set_inode_flag(inode, EXT4_INODE_INLINE_DATA); |
| 328 |
get_bh(is.iloc.bh); |
| 329 |
error = ext4_mark_iloc_dirty(handle, inode, &is.iloc); |
| 330 |
|
| 331 |
out: |
| 332 |
brelse(is.iloc.bh); |
| 333 |
return error; |
| 334 |
} |
| 335 |
|
| 336 |
static int ext4_update_inline_data(handle_t *handle, struct inode *inode, |
| 337 |
unsigned int len) |
| 338 |
{ |
| 339 |
int error; |
| 340 |
void *value = NULL; |
| 341 |
struct ext4_xattr_ibody_find is = { |
| 342 |
.s = { .not_found = -ENODATA, }, |
| 343 |
}; |
| 344 |
struct ext4_xattr_info i = { |
| 345 |
.name_index = EXT4_XATTR_INDEX_SYSTEM, |
| 346 |
.name = EXT4_XATTR_SYSTEM_DATA, |
| 347 |
}; |
| 348 |
|
| 349 |
/* If the old space is ok, write the data directly. */ |
| 350 |
if (len <= EXT4_I(inode)->i_inline_size) |
| 351 |
return 0; |
| 352 |
|
| 353 |
error = ext4_get_inode_loc(inode, &is.iloc); |
| 354 |
if (error) |
| 355 |
return error; |
| 356 |
|
| 357 |
error = ext4_xattr_ibody_find(inode, &i, &is); |
| 358 |
if (error) |
| 359 |
goto out; |
| 360 |
|
| 361 |
if (is.s.not_found) { |
| 362 |
EXT4_ERROR_INODE(inode, "missing inline data xattr"); |
| 363 |
error = -EFSCORRUPTED; |
| 364 |
goto out; |
| 365 |
} |
| 366 |
|
| 367 |
len -= EXT4_MIN_INLINE_DATA_SIZE; |
| 368 |
value = kzalloc(len, GFP_NOFS); |
| 369 |
if (!value) { |
| 370 |
error = -ENOMEM; |
| 371 |
goto out; |
| 372 |
} |
| 373 |
|
| 374 |
error = ext4_xattr_ibody_get(inode, i.name_index, i.name, |
| 375 |
value, len); |
| 376 |
if (error < 0) |
| 377 |
goto out; |
| 378 |
|
| 379 |
BUFFER_TRACE(is.iloc.bh, "get_write_access"); |
| 380 |
error = ext4_journal_get_write_access(handle, inode->i_sb, is.iloc.bh, |
| 381 |
EXT4_JTR_NONE); |
| 382 |
if (error) |
| 383 |
goto out; |
| 384 |
|
| 385 |
/* Update the xattr entry. */ |
| 386 |
i.value = value; |
| 387 |
i.value_len = len; |
| 388 |
|
| 389 |
error = ext4_xattr_ibody_set(handle, inode, &i, &is); |
| 390 |
if (error) |
| 391 |
goto out; |
| 392 |
|
| 393 |
EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here - |
| 394 |
(void *)ext4_raw_inode(&is.iloc)); |
| 395 |
EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE + |
| 396 |
le32_to_cpu(is.s.here->e_value_size); |
| 397 |
ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA); |
| 398 |
get_bh(is.iloc.bh); |
| 399 |
error = ext4_mark_iloc_dirty(handle, inode, &is.iloc); |
| 400 |
|
| 401 |
out: |
| 402 |
kfree(value); |
| 403 |
brelse(is.iloc.bh); |
| 404 |
return error; |
| 405 |
} |
| 406 |
|
| 407 |
static int ext4_prepare_inline_data(handle_t *handle, struct inode *inode, |
| 408 |
loff_t len) |
| 409 |
{ |
| 410 |
int ret, size, no_expand; |
| 411 |
struct ext4_inode_info *ei = EXT4_I(inode); |
| 412 |
|
| 413 |
if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) |
| 414 |
return -ENOSPC; |
| 415 |
|
| 416 |
size = ext4_get_max_inline_size(inode); |
| 417 |
if (size < len) |
| 418 |
return -ENOSPC; |
| 419 |
|
| 420 |
ext4_write_lock_xattr(inode, &no_expand); |
| 421 |
/* |
| 422 |
* ei->i_inline_size may have changed since the initial check |
| 423 |
* if other xattrs were added. Recalculate to ensure |
| 424 |
* ext4_update_inline_data() validates against current capacity. |
| 425 |
*/ |
| 426 |
(void) ext4_find_inline_data_nolock(inode); |
| 427 |
if (ei->i_inline_off) |
| 428 |
ret = ext4_update_inline_data(handle, inode, len); |
| 429 |
else |
| 430 |
ret = ext4_create_inline_data(handle, inode, len); |
| 431 |
|
| 432 |
ext4_write_unlock_xattr(inode, &no_expand); |
| 433 |
return ret; |
| 434 |
} |
| 435 |
|
| 436 |
static int ext4_destroy_inline_data_nolock(handle_t *handle, |
| 437 |
struct inode *inode) |
| 438 |
{ |
| 439 |
struct ext4_inode_info *ei = EXT4_I(inode); |
| 440 |
struct ext4_xattr_ibody_find is = { |
| 441 |
.s = { .not_found = 0, }, |
| 442 |
}; |
| 443 |
struct ext4_xattr_info i = { |
| 444 |
.name_index = EXT4_XATTR_INDEX_SYSTEM, |
| 445 |
.name = EXT4_XATTR_SYSTEM_DATA, |
| 446 |
.value = NULL, |
| 447 |
.value_len = 0, |
| 448 |
}; |
| 449 |
int error; |
| 450 |
|
| 451 |
if (!ei->i_inline_off) |
| 452 |
return 0; |
| 453 |
|
| 454 |
down_write(&ei->i_data_sem); |
| 455 |
|
| 456 |
error = ext4_get_inode_loc(inode, &is.iloc); |
| 457 |
if (error) { |
| 458 |
up_write(&ei->i_data_sem); |
| 459 |
return error; |
| 460 |
} |
| 461 |
|
| 462 |
error = ext4_xattr_ibody_find(inode, &i, &is); |
| 463 |
if (error) |
| 464 |
goto out; |
| 465 |
|
| 466 |
BUFFER_TRACE(is.iloc.bh, "get_write_access"); |
| 467 |
error = ext4_journal_get_write_access(handle, inode->i_sb, is.iloc.bh, |
| 468 |
EXT4_JTR_NONE); |
| 469 |
if (error) |
| 470 |
goto out; |
| 471 |
|
| 472 |
error = ext4_xattr_ibody_set(handle, inode, &i, &is); |
| 473 |
if (error) |
| 474 |
goto out; |
| 475 |
|
| 476 |
memset((void *)ext4_raw_inode(&is.iloc)->i_block, |
| 477 |
0, EXT4_MIN_INLINE_DATA_SIZE); |
| 478 |
memset(ei->i_data, 0, EXT4_MIN_INLINE_DATA_SIZE); |
| 479 |
|
| 480 |
if (ext4_has_feature_extents(inode->i_sb)) { |
| 481 |
if (S_ISDIR(inode->i_mode) || |
| 482 |
S_ISREG(inode->i_mode) || S_ISLNK(inode->i_mode)) { |
| 483 |
ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS); |
| 484 |
ext4_ext_tree_init(handle, inode); |
| 485 |
} |
| 486 |
} |
| 487 |
ext4_clear_inode_flag(inode, EXT4_INODE_INLINE_DATA); |
| 488 |
|
| 489 |
get_bh(is.iloc.bh); |
| 490 |
error = ext4_mark_iloc_dirty(handle, inode, &is.iloc); |
| 491 |
|
| 492 |
EXT4_I(inode)->i_inline_off = 0; |
| 493 |
EXT4_I(inode)->i_inline_size = 0; |
| 494 |
ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA); |
| 495 |
out: |
| 496 |
brelse(is.iloc.bh); |
| 497 |
if (error == -ENODATA) |
| 498 |
error = 0; |
| 499 |
up_write(&ei->i_data_sem); |
| 500 |
return error; |
| 501 |
} |
| 502 |
|
| 503 |
static int ext4_read_inline_folio(struct inode *inode, struct folio *folio) |
| 504 |
{ |
| 505 |
void *kaddr; |
| 506 |
int ret = 0; |
| 507 |
size_t len; |
| 508 |
struct ext4_iloc iloc; |
| 509 |
|
| 510 |
BUG_ON(!folio_test_locked(folio)); |
| 511 |
BUG_ON(!ext4_has_inline_data(inode)); |
| 512 |
BUG_ON(folio->index); |
| 513 |
|
| 514 |
if (!EXT4_I(inode)->i_inline_off) { |
| 515 |
ext4_warning(inode->i_sb, "inode %lu doesn't have inline data.", |
| 516 |
inode->i_ino); |
| 517 |
goto out; |
| 518 |
} |
| 519 |
|
| 520 |
ret = ext4_get_inode_loc(inode, &iloc); |
| 521 |
if (ret) |
| 522 |
goto out; |
| 523 |
|
| 524 |
len = min_t(size_t, ext4_get_inline_size(inode), i_size_read(inode)); |
| 525 |
|
| 526 |
if (len > PAGE_SIZE) { |
| 527 |
ext4_error_inode(inode, __func__, __LINE__, 0, |
| 528 |
"inline size %zu exceeds PAGE_SIZE", len); |
| 529 |
ret = -EFSCORRUPTED; |
| 530 |
brelse(iloc.bh); |
| 531 |
goto out; |
| 532 |
} |
| 533 |
|
| 534 |
kaddr = kmap_local_folio(folio, 0); |
| 535 |
ret = ext4_read_inline_data(inode, kaddr, len, &iloc); |
| 536 |
kaddr = folio_zero_tail(folio, len, kaddr + len); |
| 537 |
kunmap_local(kaddr); |
| 538 |
folio_mark_uptodate(folio); |
| 539 |
brelse(iloc.bh); |
| 540 |
|
| 541 |
out: |
| 542 |
return ret; |
| 543 |
} |
| 544 |
|
| 545 |
int ext4_readpage_inline(struct inode *inode, struct folio *folio) |
| 546 |
{ |
| 547 |
int ret = 0; |
| 548 |
|
| 549 |
down_read(&EXT4_I(inode)->xattr_sem); |
| 550 |
if (!ext4_has_inline_data(inode)) { |
| 551 |
up_read(&EXT4_I(inode)->xattr_sem); |
| 552 |
return -EAGAIN; |
| 553 |
} |
| 554 |
|
| 555 |
/* |
| 556 |
* Current inline data can only exist in the 1st page, |
| 557 |
* So for all the other pages, just set them uptodate. |
| 558 |
*/ |
| 559 |
if (!folio->index) |
| 560 |
ret = ext4_read_inline_folio(inode, folio); |
| 561 |
else if (!folio_test_uptodate(folio)) { |
| 562 |
folio_zero_segment(folio, 0, folio_size(folio)); |
| 563 |
folio_mark_uptodate(folio); |
| 564 |
} |
| 565 |
|
| 566 |
up_read(&EXT4_I(inode)->xattr_sem); |
| 567 |
|
| 568 |
folio_unlock(folio); |
| 569 |
return ret >= 0 ? 0 : ret; |
| 570 |
} |
| 571 |
|
| 572 |
static int ext4_convert_inline_data_to_extent(struct address_space *mapping, |
| 573 |
struct inode *inode) |
| 574 |
{ |
| 575 |
int ret, needed_blocks, no_expand; |
| 576 |
handle_t *handle = NULL; |
| 577 |
int retries = 0, sem_held = 0; |
| 578 |
struct folio *folio = NULL; |
| 579 |
unsigned from, to; |
| 580 |
struct ext4_iloc iloc; |
| 581 |
|
| 582 |
if (!ext4_has_inline_data(inode)) { |
| 583 |
/* |
| 584 |
* clear the flag so that no new write |
| 585 |
* will trap here again. |
| 586 |
*/ |
| 587 |
ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA); |
| 588 |
return 0; |
| 589 |
} |
| 590 |
|
| 591 |
needed_blocks = ext4_chunk_trans_extent(inode, 1); |
| 592 |
|
| 593 |
ret = ext4_get_inode_loc(inode, &iloc); |
| 594 |
if (ret) |
| 595 |
return ret; |
| 596 |
|
| 597 |
retry: |
| 598 |
handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks); |
| 599 |
if (IS_ERR(handle)) { |
| 600 |
ret = PTR_ERR(handle); |
| 601 |
handle = NULL; |
| 602 |
goto out; |
| 603 |
} |
| 604 |
|
| 605 |
/* We cannot recurse into the filesystem as the transaction is already |
| 606 |
* started */ |
| 607 |
folio = __filemap_get_folio(mapping, 0, FGP_WRITEBEGIN | FGP_NOFS, |
| 608 |
mapping_gfp_mask(mapping)); |
| 609 |
if (IS_ERR(folio)) { |
| 610 |
ret = PTR_ERR(folio); |
| 611 |
goto out_nofolio; |
| 612 |
} |
| 613 |
|
| 614 |
ext4_write_lock_xattr(inode, &no_expand); |
| 615 |
sem_held = 1; |
| 616 |
/* If some one has already done this for us, just exit. */ |
| 617 |
if (!ext4_has_inline_data(inode)) { |
| 618 |
ret = 0; |
| 619 |
goto out; |
| 620 |
} |
| 621 |
|
| 622 |
from = 0; |
| 623 |
to = ext4_get_inline_size(inode); |
| 624 |
if (!folio_test_uptodate(folio)) { |
| 625 |
ret = ext4_read_inline_folio(inode, folio); |
| 626 |
if (ret < 0) |
| 627 |
goto out; |
| 628 |
} |
| 629 |
|
| 630 |
ext4_fc_track_inode(handle, inode); |
| 631 |
ret = ext4_destroy_inline_data_nolock(handle, inode); |
| 632 |
if (ret) |
| 633 |
goto out; |
| 634 |
|
| 635 |
if (ext4_should_dioread_nolock(inode)) { |
| 636 |
ret = ext4_block_write_begin(handle, folio, from, to, |
| 637 |
ext4_get_block_unwritten); |
| 638 |
} else |
| 639 |
ret = ext4_block_write_begin(handle, folio, from, to, |
| 640 |
ext4_get_block); |
| 641 |
clear_buffer_new(folio_buffers(folio)); |
| 642 |
|
| 643 |
if (!ret && ext4_should_journal_data(inode)) { |
| 644 |
ret = ext4_walk_page_buffers(handle, inode, |
| 645 |
folio_buffers(folio), from, to, |
| 646 |
NULL, do_journal_get_write_access); |
| 647 |
} |
| 648 |
|
| 649 |
if (ret) { |
| 650 |
folio_unlock(folio); |
| 651 |
folio_put(folio); |
| 652 |
folio = NULL; |
| 653 |
ext4_orphan_add(handle, inode); |
| 654 |
ext4_write_unlock_xattr(inode, &no_expand); |
| 655 |
sem_held = 0; |
| 656 |
ext4_journal_stop(handle); |
| 657 |
handle = NULL; |
| 658 |
ext4_truncate_failed_write(inode); |
| 659 |
/* |
| 660 |
* If truncate failed early the inode might |
| 661 |
* still be on the orphan list; we need to |
| 662 |
* make sure the inode is removed from the |
| 663 |
* orphan list in that case. |
| 664 |
*/ |
| 665 |
if (inode->i_nlink) |
| 666 |
ext4_orphan_del(NULL, inode); |
| 667 |
} |
| 668 |
|
| 669 |
if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) |
| 670 |
goto retry; |
| 671 |
|
| 672 |
if (folio) |
| 673 |
block_commit_write(folio, from, to); |
| 674 |
out: |
| 675 |
if (folio) { |
| 676 |
folio_unlock(folio); |
| 677 |
folio_put(folio); |
| 678 |
} |
| 679 |
out_nofolio: |
| 680 |
if (sem_held) |
| 681 |
ext4_write_unlock_xattr(inode, &no_expand); |
| 682 |
if (handle) |
| 683 |
ext4_journal_stop(handle); |
| 684 |
brelse(iloc.bh); |
| 685 |
return ret; |
| 686 |
} |
| 687 |
|
| 688 |
/* |
| 689 |
* Prepare the write for the inline data. |
| 690 |
* If the data can be written into the inode, we just read |
| 691 |
* the page and make it uptodate, and start the journal. |
| 692 |
* Otherwise read the page, makes it dirty so that it can be |
| 693 |
* handle in writepages(the i_disksize update is left to the |
| 694 |
* normal ext4_da_write_end). |
| 695 |
*/ |
| 696 |
int ext4_generic_write_inline_data(struct address_space *mapping, |
| 697 |
struct inode *inode, |
| 698 |
loff_t pos, unsigned len, |
| 699 |
struct folio **foliop, |
| 700 |
void **fsdata, bool da) |
| 701 |
{ |
| 702 |
int ret; |
| 703 |
handle_t *handle; |
| 704 |
struct folio *folio; |
| 705 |
struct ext4_iloc iloc; |
| 706 |
int retries = 0; |
| 707 |
|
| 708 |
ret = ext4_get_inode_loc(inode, &iloc); |
| 709 |
if (ret) |
| 710 |
return ret; |
| 711 |
|
| 712 |
retry_journal: |
| 713 |
handle = ext4_journal_start(inode, EXT4_HT_INODE, 1); |
| 714 |
if (IS_ERR(handle)) { |
| 715 |
ret = PTR_ERR(handle); |
| 716 |
goto out_release_bh; |
| 717 |
} |
| 718 |
|
| 719 |
ret = ext4_prepare_inline_data(handle, inode, pos + len); |
| 720 |
if (ret && ret != -ENOSPC) |
| 721 |
goto out_stop_journal; |
| 722 |
|
| 723 |
if (ret == -ENOSPC) { |
| 724 |
ext4_journal_stop(handle); |
| 725 |
if (!da) { |
| 726 |
brelse(iloc.bh); |
| 727 |
/* Retry inside */ |
| 728 |
return ext4_convert_inline_data_to_extent(mapping, inode); |
| 729 |
} |
| 730 |
|
| 731 |
ret = ext4_da_convert_inline_data_to_extent(mapping, inode, fsdata); |
| 732 |
if (ret == -ENOSPC && |
| 733 |
ext4_should_retry_alloc(inode->i_sb, &retries)) |
| 734 |
goto retry_journal; |
| 735 |
goto out_release_bh; |
| 736 |
} |
| 737 |
|
| 738 |
folio = __filemap_get_folio(mapping, 0, FGP_WRITEBEGIN | FGP_NOFS, |
| 739 |
mapping_gfp_mask(mapping)); |
| 740 |
if (IS_ERR(folio)) { |
| 741 |
ret = PTR_ERR(folio); |
| 742 |
goto out_stop_journal; |
| 743 |
} |
| 744 |
|
| 745 |
down_read(&EXT4_I(inode)->xattr_sem); |
| 746 |
/* Someone else had converted it to extent */ |
| 747 |
if (!ext4_has_inline_data(inode)) { |
| 748 |
ret = 0; |
| 749 |
goto out_release_folio; |
| 750 |
} |
| 751 |
|
| 752 |
if (!folio_test_uptodate(folio)) { |
| 753 |
ret = ext4_read_inline_folio(inode, folio); |
| 754 |
if (ret < 0) |
| 755 |
goto out_release_folio; |
| 756 |
} |
| 757 |
|
| 758 |
ret = ext4_journal_get_write_access(handle, inode->i_sb, iloc.bh, EXT4_JTR_NONE); |
| 759 |
if (ret) |
| 760 |
goto out_release_folio; |
| 761 |
*foliop = folio; |
| 762 |
up_read(&EXT4_I(inode)->xattr_sem); |
| 763 |
brelse(iloc.bh); |
| 764 |
return 1; |
| 765 |
|
| 766 |
out_release_folio: |
| 767 |
up_read(&EXT4_I(inode)->xattr_sem); |
| 768 |
folio_unlock(folio); |
| 769 |
folio_put(folio); |
| 770 |
out_stop_journal: |
| 771 |
ext4_journal_stop(handle); |
| 772 |
out_release_bh: |
| 773 |
brelse(iloc.bh); |
| 774 |
return ret; |
| 775 |
} |
| 776 |
|
| 777 |
/* |
| 778 |
* Try to write data in the inode. |
| 779 |
* If the inode has inline data, check whether the new write can be |
| 780 |
* in the inode also. If not, create the page the handle, move the data |
| 781 |
* to the page make it update and let the later codes create extent for it. |
| 782 |
*/ |
| 783 |
int ext4_try_to_write_inline_data(struct address_space *mapping, |
| 784 |
struct inode *inode, |
| 785 |
loff_t pos, unsigned len, |
| 786 |
struct folio **foliop) |
| 787 |
{ |
| 788 |
if (pos + len > ext4_get_max_inline_size(inode)) |
| 789 |
return ext4_convert_inline_data_to_extent(mapping, inode); |
| 790 |
return ext4_generic_write_inline_data(mapping, inode, pos, len, |
| 791 |
foliop, NULL, false); |
| 792 |
} |
| 793 |
|
| 794 |
int ext4_write_inline_data_end(struct inode *inode, loff_t pos, unsigned len, |
| 795 |
unsigned copied, struct folio *folio) |
| 796 |
{ |
| 797 |
handle_t *handle = ext4_journal_current_handle(); |
| 798 |
int no_expand; |
| 799 |
void *kaddr; |
| 800 |
struct ext4_iloc iloc; |
| 801 |
int ret = 0, ret2; |
| 802 |
|
| 803 |
if (unlikely(copied < len) && !folio_test_uptodate(folio)) |
| 804 |
copied = 0; |
| 805 |
|
| 806 |
if (likely(copied)) { |
| 807 |
ret = ext4_get_inode_loc(inode, &iloc); |
| 808 |
if (ret) { |
| 809 |
folio_unlock(folio); |
| 810 |
folio_put(folio); |
| 811 |
ext4_std_error(inode->i_sb, ret); |
| 812 |
goto out; |
| 813 |
} |
| 814 |
ext4_write_lock_xattr(inode, &no_expand); |
| 815 |
BUG_ON(!ext4_has_inline_data(inode)); |
| 816 |
|
| 817 |
/* |
| 818 |
* ei->i_inline_off may have changed since |
| 819 |
* ext4_write_begin() called |
| 820 |
* ext4_try_to_write_inline_data() |
| 821 |
*/ |
| 822 |
(void) ext4_find_inline_data_nolock(inode); |
| 823 |
|
| 824 |
kaddr = kmap_local_folio(folio, 0); |
| 825 |
ext4_write_inline_data(inode, &iloc, kaddr, pos, copied); |
| 826 |
kunmap_local(kaddr); |
| 827 |
folio_mark_uptodate(folio); |
| 828 |
/* clear dirty flag so that writepages wouldn't work for us. */ |
| 829 |
folio_clear_dirty(folio); |
| 830 |
|
| 831 |
ext4_write_unlock_xattr(inode, &no_expand); |
| 832 |
brelse(iloc.bh); |
| 833 |
|
| 834 |
/* |
| 835 |
* It's important to update i_size while still holding folio |
| 836 |
* lock: page writeout could otherwise come in and zero |
| 837 |
* beyond i_size. |
| 838 |
*/ |
| 839 |
ext4_update_inode_size(inode, pos + copied); |
| 840 |
} |
| 841 |
folio_unlock(folio); |
| 842 |
folio_put(folio); |
| 843 |
|
| 844 |
/* |
| 845 |
* Don't mark the inode dirty under folio lock. First, it unnecessarily |
| 846 |
* makes the holding time of folio lock longer. Second, it forces lock |
| 847 |
* ordering of folio lock and transaction start for journaling |
| 848 |
* filesystems. |
| 849 |
*/ |
| 850 |
if (likely(copied)) |
| 851 |
mark_inode_dirty(inode); |
| 852 |
out: |
| 853 |
/* |
| 854 |
* If we didn't copy as much data as expected, we need to trim back |
| 855 |
* size of xattr containing inline data. |
| 856 |
*/ |
| 857 |
if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
| 858 |
ext4_orphan_add(handle, inode); |
| 859 |
|
| 860 |
ret2 = ext4_journal_stop(handle); |
| 861 |
if (!ret) |
| 862 |
ret = ret2; |
| 863 |
if (pos + len > inode->i_size) { |
| 864 |
ext4_truncate_failed_write(inode); |
| 865 |
/* |
| 866 |
* If truncate failed early the inode might still be |
| 867 |
* on the orphan list; we need to make sure the inode |
| 868 |
* is removed from the orphan list in that case. |
| 869 |
*/ |
| 870 |
if (inode->i_nlink) |
| 871 |
ext4_orphan_del(NULL, inode); |
| 872 |
} |
| 873 |
return ret ? ret : copied; |
| 874 |
} |
| 875 |
|
| 876 |
/* |
| 877 |
* Try to make the page cache and handle ready for the inline data case. |
| 878 |
* We can call this function in 2 cases: |
| 879 |
* 1. The inode is created and the first write exceeds inline size. We can |
| 880 |
* clear the inode state safely. |
| 881 |
* 2. The inode has inline data, then we need to read the data, make it |
| 882 |
* update and dirty so that ext4_da_writepages can handle it. We don't |
| 883 |
* need to start the journal since the file's metadata isn't changed now. |
| 884 |
*/ |
| 885 |
static int ext4_da_convert_inline_data_to_extent(struct address_space *mapping, |
| 886 |
struct inode *inode, |
| 887 |
void **fsdata) |
| 888 |
{ |
| 889 |
int ret = 0, inline_size; |
| 890 |
struct folio *folio; |
| 891 |
|
| 892 |
folio = __filemap_get_folio(mapping, 0, FGP_WRITEBEGIN, |
| 893 |
mapping_gfp_mask(mapping)); |
| 894 |
if (IS_ERR(folio)) |
| 895 |
return PTR_ERR(folio); |
| 896 |
|
| 897 |
down_read(&EXT4_I(inode)->xattr_sem); |
| 898 |
if (!ext4_has_inline_data(inode)) { |
| 899 |
ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA); |
| 900 |
goto out; |
| 901 |
} |
| 902 |
|
| 903 |
inline_size = ext4_get_inline_size(inode); |
| 904 |
|
| 905 |
if (!folio_test_uptodate(folio)) { |
| 906 |
ret = ext4_read_inline_folio(inode, folio); |
| 907 |
if (ret < 0) |
| 908 |
goto out; |
| 909 |
} |
| 910 |
|
| 911 |
ret = ext4_block_write_begin(NULL, folio, 0, inline_size, |
| 912 |
ext4_da_get_block_prep); |
| 913 |
if (ret) { |
| 914 |
up_read(&EXT4_I(inode)->xattr_sem); |
| 915 |
folio_unlock(folio); |
| 916 |
folio_put(folio); |
| 917 |
ext4_truncate_failed_write(inode); |
| 918 |
return ret; |
| 919 |
} |
| 920 |
|
| 921 |
clear_buffer_new(folio_buffers(folio)); |
| 922 |
folio_mark_dirty(folio); |
| 923 |
folio_mark_uptodate(folio); |
| 924 |
ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA); |
| 925 |
*fsdata = (void *)CONVERT_INLINE_DATA; |
| 926 |
|
| 927 |
out: |
| 928 |
up_read(&EXT4_I(inode)->xattr_sem); |
| 929 |
if (folio) { |
| 930 |
folio_unlock(folio); |
| 931 |
folio_put(folio); |
| 932 |
} |
| 933 |
return ret; |
| 934 |
} |
| 935 |
|
| 936 |
#ifdef INLINE_DIR_DEBUG |
| 937 |
void ext4_show_inline_dir(struct inode *dir, struct buffer_head *bh, |
| 938 |
void *inline_start, int inline_size) |
| 939 |
{ |
| 940 |
int offset; |
| 941 |
unsigned short de_len; |
| 942 |
struct ext4_dir_entry_2 *de = inline_start; |
| 943 |
void *dlimit = inline_start + inline_size; |
| 944 |
|
| 945 |
trace_printk("inode %lu\n", dir->i_ino); |
| 946 |
offset = 0; |
| 947 |
while ((void *)de < dlimit) { |
| 948 |
de_len = ext4_rec_len_from_disk(de->rec_len, inline_size); |
| 949 |
trace_printk("de: off %u rlen %u name %.*s nlen %u ino %u\n", |
| 950 |
offset, de_len, de->name_len, de->name, |
| 951 |
de->name_len, le32_to_cpu(de->inode)); |
| 952 |
if (ext4_check_dir_entry(dir, NULL, de, bh, |
| 953 |
inline_start, inline_size, offset)) |
| 954 |
BUG(); |
| 955 |
|
| 956 |
offset += de_len; |
| 957 |
de = (struct ext4_dir_entry_2 *) ((char *) de + de_len); |
| 958 |
} |
| 959 |
} |
| 960 |
#else |
| 961 |
#define ext4_show_inline_dir(dir, bh, inline_start, inline_size) |
| 962 |
#endif |
| 963 |
|
| 964 |
/* |
| 965 |
* Add a new entry into a inline dir. |
| 966 |
* It will return -ENOSPC if no space is available, and -EIO |
| 967 |
* and -EEXIST if directory entry already exists. |
| 968 |
*/ |
| 969 |
static int ext4_add_dirent_to_inline(handle_t *handle, |
| 970 |
struct ext4_filename *fname, |
| 971 |
struct inode *dir, |
| 972 |
struct inode *inode, |
| 973 |
struct ext4_iloc *iloc, |
| 974 |
void *inline_start, int inline_size) |
| 975 |
{ |
| 976 |
int err; |
| 977 |
struct ext4_dir_entry_2 *de; |
| 978 |
|
| 979 |
err = ext4_find_dest_de(dir, iloc->bh, inline_start, |
| 980 |
inline_size, fname, &de); |
| 981 |
if (err) |
| 982 |
return err; |
| 983 |
|
| 984 |
BUFFER_TRACE(iloc->bh, "get_write_access"); |
| 985 |
err = ext4_journal_get_write_access(handle, dir->i_sb, iloc->bh, |
| 986 |
EXT4_JTR_NONE); |
| 987 |
if (err) |
| 988 |
return err; |
| 989 |
ext4_insert_dentry(dir, inode, de, inline_size, fname); |
| 990 |
|
| 991 |
ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size); |
| 992 |
|
| 993 |
/* |
| 994 |
* XXX shouldn't update any times until successful |
| 995 |
* completion of syscall, but too many callers depend |
| 996 |
* on this. |
| 997 |
* |
| 998 |
* XXX similarly, too many callers depend on |
| 999 |
* ext4_new_inode() setting the times, but error |
| 1000 |
* recovery deletes the inode, so the worst that can |
| 1001 |
* happen is that the times are slightly out of date |
| 1002 |
* and/or different from the directory change time. |
| 1003 |
*/ |
| 1004 |
inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir)); |
| 1005 |
ext4_update_dx_flag(dir); |
| 1006 |
inode_inc_iversion(dir); |
| 1007 |
return 1; |
| 1008 |
} |
| 1009 |
|
| 1010 |
static void *ext4_get_inline_xattr_pos(struct inode *inode, |
| 1011 |
struct ext4_iloc *iloc) |
| 1012 |
{ |
| 1013 |
struct ext4_xattr_entry *entry; |
| 1014 |
struct ext4_xattr_ibody_header *header; |
| 1015 |
|
| 1016 |
BUG_ON(!EXT4_I(inode)->i_inline_off); |
| 1017 |
|
| 1018 |
header = IHDR(inode, ext4_raw_inode(iloc)); |
| 1019 |
entry = (struct ext4_xattr_entry *)((void *)ext4_raw_inode(iloc) + |
| 1020 |
EXT4_I(inode)->i_inline_off); |
| 1021 |
|
| 1022 |
return (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs); |
| 1023 |
} |
| 1024 |
|
| 1025 |
/* Set the final de to cover the whole block. */ |
| 1026 |
void ext4_update_final_de(void *de_buf, int old_size, int new_size) |
| 1027 |
{ |
| 1028 |
struct ext4_dir_entry_2 *de, *prev_de; |
| 1029 |
void *limit; |
| 1030 |
int de_len; |
| 1031 |
|
| 1032 |
de = de_buf; |
| 1033 |
if (old_size) { |
| 1034 |
limit = de_buf + old_size; |
| 1035 |
do { |
| 1036 |
prev_de = de; |
| 1037 |
de_len = ext4_rec_len_from_disk(de->rec_len, old_size); |
| 1038 |
de_buf += de_len; |
| 1039 |
de = de_buf; |
| 1040 |
} while (de_buf < limit); |
| 1041 |
|
| 1042 |
prev_de->rec_len = ext4_rec_len_to_disk(de_len + new_size - |
| 1043 |
old_size, new_size); |
| 1044 |
} else { |
| 1045 |
/* this is just created, so create an empty entry. */ |
| 1046 |
de->inode = 0; |
| 1047 |
de->rec_len = ext4_rec_len_to_disk(new_size, new_size); |
| 1048 |
} |
| 1049 |
} |
| 1050 |
|
| 1051 |
static int ext4_update_inline_dir(handle_t *handle, struct inode *dir, |
| 1052 |
struct ext4_iloc *iloc) |
| 1053 |
{ |
| 1054 |
int ret; |
| 1055 |
int old_size = EXT4_I(dir)->i_inline_size - EXT4_MIN_INLINE_DATA_SIZE; |
| 1056 |
int new_size = get_max_inline_xattr_value_size(dir, iloc); |
| 1057 |
|
| 1058 |
if (new_size - old_size <= ext4_dir_rec_len(1, NULL)) |
| 1059 |
return -ENOSPC; |
| 1060 |
|
| 1061 |
ret = ext4_update_inline_data(handle, dir, |
| 1062 |
new_size + EXT4_MIN_INLINE_DATA_SIZE); |
| 1063 |
if (ret) |
| 1064 |
return ret; |
| 1065 |
|
| 1066 |
ext4_update_final_de(ext4_get_inline_xattr_pos(dir, iloc), old_size, |
| 1067 |
EXT4_I(dir)->i_inline_size - |
| 1068 |
EXT4_MIN_INLINE_DATA_SIZE); |
| 1069 |
dir->i_size = EXT4_I(dir)->i_disksize = EXT4_I(dir)->i_inline_size; |
| 1070 |
return 0; |
| 1071 |
} |
| 1072 |
|
| 1073 |
static void ext4_restore_inline_data(handle_t *handle, struct inode *inode, |
| 1074 |
struct ext4_iloc *iloc, |
| 1075 |
void *buf, int inline_size) |
| 1076 |
{ |
| 1077 |
int ret; |
| 1078 |
|
| 1079 |
ret = ext4_create_inline_data(handle, inode, inline_size); |
| 1080 |
if (ret) { |
| 1081 |
ext4_msg(inode->i_sb, KERN_EMERG, |
| 1082 |
"error restoring inline_data for inode -- potential data loss! (inode %lu, error %d)", |
| 1083 |
inode->i_ino, ret); |
| 1084 |
return; |
| 1085 |
} |
| 1086 |
ext4_write_inline_data(inode, iloc, buf, 0, inline_size); |
| 1087 |
ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA); |
| 1088 |
} |
| 1089 |
|
| 1090 |
static int ext4_convert_inline_data_nolock(handle_t *handle, |
| 1091 |
struct inode *inode, |
| 1092 |
struct ext4_iloc *iloc) |
| 1093 |
{ |
| 1094 |
int error; |
| 1095 |
void *buf = NULL; |
| 1096 |
struct buffer_head *data_bh = NULL; |
| 1097 |
struct ext4_map_blocks map; |
| 1098 |
int inline_size; |
| 1099 |
|
| 1100 |
inline_size = ext4_get_inline_size(inode); |
| 1101 |
buf = kmalloc(inline_size, GFP_NOFS); |
| 1102 |
if (!buf) { |
| 1103 |
error = -ENOMEM; |
| 1104 |
goto out; |
| 1105 |
} |
| 1106 |
|
| 1107 |
error = ext4_read_inline_data(inode, buf, inline_size, iloc); |
| 1108 |
if (error < 0) |
| 1109 |
goto out; |
| 1110 |
|
| 1111 |
/* |
| 1112 |
* Make sure the inline directory entries pass checks before we try to |
| 1113 |
* convert them, so that we avoid touching stuff that needs fsck. |
| 1114 |
*/ |
| 1115 |
if (S_ISDIR(inode->i_mode)) { |
| 1116 |
error = ext4_check_all_de(inode, iloc->bh, |
| 1117 |
buf + EXT4_INLINE_DOTDOT_SIZE, |
| 1118 |
inline_size - EXT4_INLINE_DOTDOT_SIZE); |
| 1119 |
if (error) |
| 1120 |
goto out; |
| 1121 |
} |
| 1122 |
|
| 1123 |
error = ext4_destroy_inline_data_nolock(handle, inode); |
| 1124 |
if (error) |
| 1125 |
goto out; |
| 1126 |
|
| 1127 |
map.m_lblk = 0; |
| 1128 |
map.m_len = 1; |
| 1129 |
map.m_flags = 0; |
| 1130 |
error = ext4_map_blocks(handle, inode, &map, EXT4_GET_BLOCKS_CREATE); |
| 1131 |
if (error < 0) |
| 1132 |
goto out_restore; |
| 1133 |
if (!(map.m_flags & EXT4_MAP_MAPPED)) { |
| 1134 |
error = -EIO; |
| 1135 |
goto out_restore; |
| 1136 |
} |
| 1137 |
|
| 1138 |
data_bh = sb_getblk(inode->i_sb, map.m_pblk); |
| 1139 |
if (!data_bh) { |
| 1140 |
error = -ENOMEM; |
| 1141 |
goto out_restore; |
| 1142 |
} |
| 1143 |
|
| 1144 |
lock_buffer(data_bh); |
| 1145 |
error = ext4_journal_get_create_access(handle, inode->i_sb, data_bh, |
| 1146 |
EXT4_JTR_NONE); |
| 1147 |
if (error) { |
| 1148 |
unlock_buffer(data_bh); |
| 1149 |
error = -EIO; |
| 1150 |
goto out_restore; |
| 1151 |
} |
| 1152 |
memset(data_bh->b_data, 0, inode->i_sb->s_blocksize); |
| 1153 |
|
| 1154 |
if (!S_ISDIR(inode->i_mode)) { |
| 1155 |
memcpy(data_bh->b_data, buf, inline_size); |
| 1156 |
set_buffer_uptodate(data_bh); |
| 1157 |
unlock_buffer(data_bh); |
| 1158 |
error = ext4_handle_dirty_metadata(handle, |
| 1159 |
inode, data_bh); |
| 1160 |
} else { |
| 1161 |
unlock_buffer(data_bh); |
| 1162 |
inode->i_size = inode->i_sb->s_blocksize; |
| 1163 |
i_size_write(inode, inode->i_sb->s_blocksize); |
| 1164 |
EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize; |
| 1165 |
|
| 1166 |
error = ext4_init_dirblock(handle, inode, data_bh, |
| 1167 |
le32_to_cpu(((struct ext4_dir_entry_2 *)buf)->inode), |
| 1168 |
buf + EXT4_INLINE_DOTDOT_SIZE, |
| 1169 |
inline_size - EXT4_INLINE_DOTDOT_SIZE); |
| 1170 |
if (!error) |
| 1171 |
error = ext4_mark_inode_dirty(handle, inode); |
| 1172 |
} |
| 1173 |
|
| 1174 |
out_restore: |
| 1175 |
if (error) |
| 1176 |
ext4_restore_inline_data(handle, inode, iloc, buf, inline_size); |
| 1177 |
|
| 1178 |
out: |
| 1179 |
brelse(data_bh); |
| 1180 |
kfree(buf); |
| 1181 |
return error; |
| 1182 |
} |
| 1183 |
|
| 1184 |
/* |
| 1185 |
* Try to add the new entry to the inline data. |
| 1186 |
* If succeeds, return 0. If not, extended the inline dir and copied data to |
| 1187 |
* the new created block. |
| 1188 |
*/ |
| 1189 |
int ext4_try_add_inline_entry(handle_t *handle, struct ext4_filename *fname, |
| 1190 |
struct inode *dir, struct inode *inode) |
| 1191 |
{ |
| 1192 |
int ret, ret2, inline_size, no_expand; |
| 1193 |
void *inline_start; |
| 1194 |
struct ext4_iloc iloc; |
| 1195 |
|
| 1196 |
ret = ext4_get_inode_loc(dir, &iloc); |
| 1197 |
if (ret) |
| 1198 |
return ret; |
| 1199 |
|
| 1200 |
ext4_write_lock_xattr(dir, &no_expand); |
| 1201 |
if (!ext4_has_inline_data(dir)) |
| 1202 |
goto out; |
| 1203 |
|
| 1204 |
inline_start = (void *)ext4_raw_inode(&iloc)->i_block + |
| 1205 |
EXT4_INLINE_DOTDOT_SIZE; |
| 1206 |
inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE; |
| 1207 |
|
| 1208 |
ret = ext4_add_dirent_to_inline(handle, fname, dir, inode, &iloc, |
| 1209 |
inline_start, inline_size); |
| 1210 |
if (ret != -ENOSPC) |
| 1211 |
goto out; |
| 1212 |
|
| 1213 |
/* check whether it can be inserted to inline xattr space. */ |
| 1214 |
inline_size = EXT4_I(dir)->i_inline_size - |
| 1215 |
EXT4_MIN_INLINE_DATA_SIZE; |
| 1216 |
if (!inline_size) { |
| 1217 |
/* Try to use the xattr space.*/ |
| 1218 |
ret = ext4_update_inline_dir(handle, dir, &iloc); |
| 1219 |
if (ret && ret != -ENOSPC) |
| 1220 |
goto out; |
| 1221 |
|
| 1222 |
inline_size = EXT4_I(dir)->i_inline_size - |
| 1223 |
EXT4_MIN_INLINE_DATA_SIZE; |
| 1224 |
} |
| 1225 |
|
| 1226 |
if (inline_size) { |
| 1227 |
inline_start = ext4_get_inline_xattr_pos(dir, &iloc); |
| 1228 |
|
| 1229 |
ret = ext4_add_dirent_to_inline(handle, fname, dir, |
| 1230 |
inode, &iloc, inline_start, |
| 1231 |
inline_size); |
| 1232 |
|
| 1233 |
if (ret != -ENOSPC) |
| 1234 |
goto out; |
| 1235 |
} |
| 1236 |
|
| 1237 |
/* |
| 1238 |
* The inline space is filled up, so create a new block for it. |
| 1239 |
* As the extent tree will be created, we have to save the inline |
| 1240 |
* dir first. |
| 1241 |
*/ |
| 1242 |
ret = ext4_convert_inline_data_nolock(handle, dir, &iloc); |
| 1243 |
|
| 1244 |
out: |
| 1245 |
ext4_write_unlock_xattr(dir, &no_expand); |
| 1246 |
ret2 = ext4_mark_inode_dirty(handle, dir); |
| 1247 |
if (unlikely(ret2 && !ret)) |
| 1248 |
ret = ret2; |
| 1249 |
brelse(iloc.bh); |
| 1250 |
return ret; |
| 1251 |
} |
| 1252 |
|
| 1253 |
/* |
| 1254 |
* This function fills a red-black tree with information from an |
| 1255 |
* inlined dir. It returns the number directory entries loaded |
| 1256 |
* into the tree. If there is an error it is returned in err. |
| 1257 |
*/ |
| 1258 |
int ext4_inlinedir_to_tree(struct file *dir_file, |
| 1259 |
struct inode *dir, ext4_lblk_t block, |
| 1260 |
struct dx_hash_info *hinfo, |
| 1261 |
__u32 start_hash, __u32 start_minor_hash, |
| 1262 |
int *has_inline_data) |
| 1263 |
{ |
| 1264 |
int err = 0, count = 0; |
| 1265 |
unsigned int parent_ino; |
| 1266 |
int pos; |
| 1267 |
struct ext4_dir_entry_2 *de; |
| 1268 |
struct inode *inode = file_inode(dir_file); |
| 1269 |
int ret, inline_size = 0; |
| 1270 |
struct ext4_iloc iloc; |
| 1271 |
void *dir_buf = NULL; |
| 1272 |
struct ext4_dir_entry_2 fake; |
| 1273 |
struct fscrypt_str tmp_str; |
| 1274 |
|
| 1275 |
ret = ext4_get_inode_loc(inode, &iloc); |
| 1276 |
if (ret) |
| 1277 |
return ret; |
| 1278 |
|
| 1279 |
down_read(&EXT4_I(inode)->xattr_sem); |
| 1280 |
if (!ext4_has_inline_data(inode)) { |
| 1281 |
up_read(&EXT4_I(inode)->xattr_sem); |
| 1282 |
*has_inline_data = 0; |
| 1283 |
goto out; |
| 1284 |
} |
| 1285 |
|
| 1286 |
inline_size = ext4_get_inline_size(inode); |
| 1287 |
dir_buf = kmalloc(inline_size, GFP_NOFS); |
| 1288 |
if (!dir_buf) { |
| 1289 |
ret = -ENOMEM; |
| 1290 |
up_read(&EXT4_I(inode)->xattr_sem); |
| 1291 |
goto out; |
| 1292 |
} |
| 1293 |
|
| 1294 |
ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc); |
| 1295 |
up_read(&EXT4_I(inode)->xattr_sem); |
| 1296 |
if (ret < 0) |
| 1297 |
goto out; |
| 1298 |
|
| 1299 |
pos = 0; |
| 1300 |
parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode); |
| 1301 |
while (pos < inline_size) { |
| 1302 |
/* |
| 1303 |
* As inlined dir doesn't store any information about '.' and |
| 1304 |
* only the inode number of '..' is stored, we have to handle |
| 1305 |
* them differently. |
| 1306 |
*/ |
| 1307 |
if (pos == 0) { |
| 1308 |
fake.inode = cpu_to_le32(inode->i_ino); |
| 1309 |
fake.name_len = 1; |
| 1310 |
memcpy(fake.name, ".", 2); |
| 1311 |
fake.rec_len = ext4_rec_len_to_disk( |
| 1312 |
ext4_dir_rec_len(fake.name_len, NULL), |
| 1313 |
inline_size); |
| 1314 |
ext4_set_de_type(inode->i_sb, &fake, S_IFDIR); |
| 1315 |
de = &fake; |
| 1316 |
pos = EXT4_INLINE_DOTDOT_OFFSET; |
| 1317 |
} else if (pos == EXT4_INLINE_DOTDOT_OFFSET) { |
| 1318 |
fake.inode = cpu_to_le32(parent_ino); |
| 1319 |
fake.name_len = 2; |
| 1320 |
memcpy(fake.name, "..", 3); |
| 1321 |
fake.rec_len = ext4_rec_len_to_disk( |
| 1322 |
ext4_dir_rec_len(fake.name_len, NULL), |
| 1323 |
inline_size); |
| 1324 |
ext4_set_de_type(inode->i_sb, &fake, S_IFDIR); |
| 1325 |
de = &fake; |
| 1326 |
pos = EXT4_INLINE_DOTDOT_SIZE; |
| 1327 |
} else { |
| 1328 |
de = (struct ext4_dir_entry_2 *)(dir_buf + pos); |
| 1329 |
pos += ext4_rec_len_from_disk(de->rec_len, inline_size); |
| 1330 |
if (ext4_check_dir_entry(inode, dir_file, de, |
| 1331 |
iloc.bh, dir_buf, |
| 1332 |
inline_size, pos)) { |
| 1333 |
ret = count; |
| 1334 |
goto out; |
| 1335 |
} |
| 1336 |
} |
| 1337 |
|
| 1338 |
if (ext4_hash_in_dirent(dir)) { |
| 1339 |
hinfo->hash = EXT4_DIRENT_HASH(de); |
| 1340 |
hinfo->minor_hash = EXT4_DIRENT_MINOR_HASH(de); |
| 1341 |
} else { |
| 1342 |
err = ext4fs_dirhash(dir, de->name, de->name_len, hinfo); |
| 1343 |
if (err) { |
| 1344 |
ret = err; |
| 1345 |
goto out; |
| 1346 |
} |
| 1347 |
} |
| 1348 |
if ((hinfo->hash < start_hash) || |
| 1349 |
((hinfo->hash == start_hash) && |
| 1350 |
(hinfo->minor_hash < start_minor_hash))) |
| 1351 |
continue; |
| 1352 |
if (de->inode == 0) |
| 1353 |
continue; |
| 1354 |
tmp_str.name = de->name; |
| 1355 |
tmp_str.len = de->name_len; |
| 1356 |
err = ext4_htree_store_dirent(dir_file, hinfo->hash, |
| 1357 |
hinfo->minor_hash, de, &tmp_str); |
| 1358 |
if (err) { |
| 1359 |
ret = err; |
| 1360 |
goto out; |
| 1361 |
} |
| 1362 |
count++; |
| 1363 |
} |
| 1364 |
ret = count; |
| 1365 |
out: |
| 1366 |
kfree(dir_buf); |
| 1367 |
brelse(iloc.bh); |
| 1368 |
return ret; |
| 1369 |
} |
| 1370 |
|
| 1371 |
/* |
| 1372 |
* So this function is called when the volume is mkfsed with |
| 1373 |
* dir_index disabled. In order to keep f_pos persistent |
| 1374 |
* after we convert from an inlined dir to a blocked based, |
| 1375 |
* we just pretend that we are a normal dir and return the |
| 1376 |
* offset as if '.' and '..' really take place. |
| 1377 |
* |
| 1378 |
*/ |
| 1379 |
int ext4_read_inline_dir(struct file *file, |
| 1380 |
struct dir_context *ctx, |
| 1381 |
int *has_inline_data) |
| 1382 |
{ |
| 1383 |
unsigned int offset, parent_ino; |
| 1384 |
int i; |
| 1385 |
struct ext4_dir_entry_2 *de; |
| 1386 |
struct super_block *sb; |
| 1387 |
struct inode *inode = file_inode(file); |
| 1388 |
int ret, inline_size = 0; |
| 1389 |
struct ext4_iloc iloc; |
| 1390 |
void *dir_buf = NULL; |
| 1391 |
int dotdot_offset, dotdot_size, extra_offset, extra_size; |
| 1392 |
struct dir_private_info *info = file->private_data; |
| 1393 |
|
| 1394 |
ret = ext4_get_inode_loc(inode, &iloc); |
| 1395 |
if (ret) |
| 1396 |
return ret; |
| 1397 |
|
| 1398 |
down_read(&EXT4_I(inode)->xattr_sem); |
| 1399 |
if (!ext4_has_inline_data(inode)) { |
| 1400 |
up_read(&EXT4_I(inode)->xattr_sem); |
| 1401 |
*has_inline_data = 0; |
| 1402 |
goto out; |
| 1403 |
} |
| 1404 |
|
| 1405 |
inline_size = ext4_get_inline_size(inode); |
| 1406 |
dir_buf = kmalloc(inline_size, GFP_NOFS); |
| 1407 |
if (!dir_buf) { |
| 1408 |
ret = -ENOMEM; |
| 1409 |
up_read(&EXT4_I(inode)->xattr_sem); |
| 1410 |
goto out; |
| 1411 |
} |
| 1412 |
|
| 1413 |
ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc); |
| 1414 |
up_read(&EXT4_I(inode)->xattr_sem); |
| 1415 |
if (ret < 0) |
| 1416 |
goto out; |
| 1417 |
|
| 1418 |
ret = 0; |
| 1419 |
sb = inode->i_sb; |
| 1420 |
parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode); |
| 1421 |
offset = ctx->pos; |
| 1422 |
|
| 1423 |
/* |
| 1424 |
* dotdot_offset and dotdot_size is the real offset and |
| 1425 |
* size for ".." and "." if the dir is block based while |
| 1426 |
* the real size for them are only EXT4_INLINE_DOTDOT_SIZE. |
| 1427 |
* So we will use extra_offset and extra_size to indicate them |
| 1428 |
* during the inline dir iteration. |
| 1429 |
*/ |
| 1430 |
dotdot_offset = ext4_dir_rec_len(1, NULL); |
| 1431 |
dotdot_size = dotdot_offset + ext4_dir_rec_len(2, NULL); |
| 1432 |
extra_offset = dotdot_size - EXT4_INLINE_DOTDOT_SIZE; |
| 1433 |
extra_size = extra_offset + inline_size; |
| 1434 |
|
| 1435 |
/* |
| 1436 |
* If the cookie has changed since the last call to |
| 1437 |
* readdir(2), then we might be pointing to an invalid |
| 1438 |
* dirent right now. Scan from the start of the inline |
| 1439 |
* dir to make sure. |
| 1440 |
*/ |
| 1441 |
if (!inode_eq_iversion(inode, info->cookie)) { |
| 1442 |
for (i = 0; i < extra_size && i < offset;) { |
| 1443 |
/* |
| 1444 |
* "." is with offset 0 and |
| 1445 |
* ".." is dotdot_offset. |
| 1446 |
*/ |
| 1447 |
if (!i) { |
| 1448 |
i = dotdot_offset; |
| 1449 |
continue; |
| 1450 |
} else if (i == dotdot_offset) { |
| 1451 |
i = dotdot_size; |
| 1452 |
continue; |
| 1453 |
} |
| 1454 |
/* for other entry, the real offset in |
| 1455 |
* the buf has to be tuned accordingly. |
| 1456 |
*/ |
| 1457 |
de = (struct ext4_dir_entry_2 *) |
| 1458 |
(dir_buf + i - extra_offset); |
| 1459 |
/* It's too expensive to do a full |
| 1460 |
* dirent test each time round this |
| 1461 |
* loop, but we do have to test at |
| 1462 |
* least that it is non-zero. A |
| 1463 |
* failure will be detected in the |
| 1464 |
* dirent test below. */ |
| 1465 |
if (ext4_rec_len_from_disk(de->rec_len, extra_size) |
| 1466 |
< ext4_dir_rec_len(1, NULL)) |
| 1467 |
break; |
| 1468 |
i += ext4_rec_len_from_disk(de->rec_len, |
| 1469 |
extra_size); |
| 1470 |
} |
| 1471 |
offset = i; |
| 1472 |
ctx->pos = offset; |
| 1473 |
info->cookie = inode_query_iversion(inode); |
| 1474 |
} |
| 1475 |
|
| 1476 |
while (ctx->pos < extra_size) { |
| 1477 |
if (ctx->pos == 0) { |
| 1478 |
if (!dir_emit(ctx, ".", 1, inode->i_ino, DT_DIR)) |
| 1479 |
goto out; |
| 1480 |
ctx->pos = dotdot_offset; |
| 1481 |
continue; |
| 1482 |
} |
| 1483 |
|
| 1484 |
if (ctx->pos == dotdot_offset) { |
| 1485 |
if (!dir_emit(ctx, "..", 2, parent_ino, DT_DIR)) |
| 1486 |
goto out; |
| 1487 |
ctx->pos = dotdot_size; |
| 1488 |
continue; |
| 1489 |
} |
| 1490 |
|
| 1491 |
de = (struct ext4_dir_entry_2 *) |
| 1492 |
(dir_buf + ctx->pos - extra_offset); |
| 1493 |
if (ext4_check_dir_entry(inode, file, de, iloc.bh, dir_buf, |
| 1494 |
extra_size, ctx->pos)) |
| 1495 |
goto out; |
| 1496 |
if (le32_to_cpu(de->inode)) { |
| 1497 |
if (!dir_emit(ctx, de->name, de->name_len, |
| 1498 |
le32_to_cpu(de->inode), |
| 1499 |
get_dtype(sb, de->file_type))) |
| 1500 |
goto out; |
| 1501 |
} |
| 1502 |
ctx->pos += ext4_rec_len_from_disk(de->rec_len, extra_size); |
| 1503 |
} |
| 1504 |
out: |
| 1505 |
kfree(dir_buf); |
| 1506 |
brelse(iloc.bh); |
| 1507 |
return ret; |
| 1508 |
} |
| 1509 |
|
| 1510 |
void *ext4_read_inline_link(struct inode *inode) |
| 1511 |
{ |
| 1512 |
struct ext4_iloc iloc; |
| 1513 |
int ret, inline_size; |
| 1514 |
void *link; |
| 1515 |
|
| 1516 |
ret = ext4_get_inode_loc(inode, &iloc); |
| 1517 |
if (ret) |
| 1518 |
return ERR_PTR(ret); |
| 1519 |
|
| 1520 |
ret = -ENOMEM; |
| 1521 |
inline_size = ext4_get_inline_size(inode); |
| 1522 |
link = kmalloc(inline_size + 1, GFP_NOFS); |
| 1523 |
if (!link) |
| 1524 |
goto out; |
| 1525 |
|
| 1526 |
ret = ext4_read_inline_data(inode, link, inline_size, &iloc); |
| 1527 |
if (ret < 0) { |
| 1528 |
kfree(link); |
| 1529 |
goto out; |
| 1530 |
} |
| 1531 |
nd_terminate_link(link, inode->i_size, ret); |
| 1532 |
out: |
| 1533 |
if (ret < 0) |
| 1534 |
link = ERR_PTR(ret); |
| 1535 |
brelse(iloc.bh); |
| 1536 |
return link; |
| 1537 |
} |
| 1538 |
|
| 1539 |
struct buffer_head *ext4_get_first_inline_block(struct inode *inode, |
| 1540 |
struct ext4_dir_entry_2 **parent_de, |
| 1541 |
int *retval) |
| 1542 |
{ |
| 1543 |
struct ext4_iloc iloc; |
| 1544 |
|
| 1545 |
*retval = ext4_get_inode_loc(inode, &iloc); |
| 1546 |
if (*retval) |
| 1547 |
return NULL; |
| 1548 |
|
| 1549 |
*parent_de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block; |
| 1550 |
|
| 1551 |
return iloc.bh; |
| 1552 |
} |
| 1553 |
|
| 1554 |
/* |
| 1555 |
* Try to create the inline data for the new dir. |
| 1556 |
* If it succeeds, return 0, otherwise return the error. |
| 1557 |
* In case of ENOSPC, the caller should create the normal disk layout dir. |
| 1558 |
*/ |
| 1559 |
int ext4_try_create_inline_dir(handle_t *handle, struct inode *parent, |
| 1560 |
struct inode *inode) |
| 1561 |
{ |
| 1562 |
int ret, inline_size = EXT4_MIN_INLINE_DATA_SIZE; |
| 1563 |
struct ext4_iloc iloc; |
| 1564 |
struct ext4_dir_entry_2 *de; |
| 1565 |
|
| 1566 |
ret = ext4_get_inode_loc(inode, &iloc); |
| 1567 |
if (ret) |
| 1568 |
return ret; |
| 1569 |
|
| 1570 |
ret = ext4_prepare_inline_data(handle, inode, inline_size); |
| 1571 |
if (ret) |
| 1572 |
goto out; |
| 1573 |
|
| 1574 |
/* |
| 1575 |
* For inline dir, we only save the inode information for the ".." |
| 1576 |
* and create a fake dentry to cover the left space. |
| 1577 |
*/ |
| 1578 |
de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block; |
| 1579 |
de->inode = cpu_to_le32(parent->i_ino); |
| 1580 |
de = (struct ext4_dir_entry_2 *)((void *)de + EXT4_INLINE_DOTDOT_SIZE); |
| 1581 |
de->inode = 0; |
| 1582 |
de->rec_len = ext4_rec_len_to_disk( |
| 1583 |
inline_size - EXT4_INLINE_DOTDOT_SIZE, |
| 1584 |
inline_size); |
| 1585 |
set_nlink(inode, 2); |
| 1586 |
inode->i_size = EXT4_I(inode)->i_disksize = inline_size; |
| 1587 |
out: |
| 1588 |
brelse(iloc.bh); |
| 1589 |
return ret; |
| 1590 |
} |
| 1591 |
|
| 1592 |
struct buffer_head *ext4_find_inline_entry(struct inode *dir, |
| 1593 |
struct ext4_filename *fname, |
| 1594 |
struct ext4_dir_entry_2 **res_dir, |
| 1595 |
int *has_inline_data) |
| 1596 |
{ |
| 1597 |
struct ext4_xattr_ibody_find is = { |
| 1598 |
.s = { .not_found = -ENODATA, }, |
| 1599 |
}; |
| 1600 |
struct ext4_xattr_info i = { |
| 1601 |
.name_index = EXT4_XATTR_INDEX_SYSTEM, |
| 1602 |
.name = EXT4_XATTR_SYSTEM_DATA, |
| 1603 |
}; |
| 1604 |
int ret; |
| 1605 |
void *inline_start; |
| 1606 |
int inline_size; |
| 1607 |
|
| 1608 |
ret = ext4_get_inode_loc(dir, &is.iloc); |
| 1609 |
if (ret) |
| 1610 |
return ERR_PTR(ret); |
| 1611 |
|
| 1612 |
down_read(&EXT4_I(dir)->xattr_sem); |
| 1613 |
|
| 1614 |
ret = ext4_xattr_ibody_find(dir, &i, &is); |
| 1615 |
if (ret) |
| 1616 |
goto out; |
| 1617 |
|
| 1618 |
if (!ext4_has_inline_data(dir)) { |
| 1619 |
*has_inline_data = 0; |
| 1620 |
goto out; |
| 1621 |
} |
| 1622 |
|
| 1623 |
inline_start = (void *)ext4_raw_inode(&is.iloc)->i_block + |
| 1624 |
EXT4_INLINE_DOTDOT_SIZE; |
| 1625 |
inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE; |
| 1626 |
ret = ext4_search_dir(is.iloc.bh, inline_start, inline_size, |
| 1627 |
dir, fname, 0, res_dir); |
| 1628 |
if (ret == 1) |
| 1629 |
goto out_find; |
| 1630 |
if (ret < 0) |
| 1631 |
goto out; |
| 1632 |
|
| 1633 |
if (ext4_get_inline_size(dir) == EXT4_MIN_INLINE_DATA_SIZE) |
| 1634 |
goto out; |
| 1635 |
|
| 1636 |
inline_start = ext4_get_inline_xattr_pos(dir, &is.iloc); |
| 1637 |
inline_size = ext4_get_inline_size(dir) - EXT4_MIN_INLINE_DATA_SIZE; |
| 1638 |
|
| 1639 |
ret = ext4_search_dir(is.iloc.bh, inline_start, inline_size, |
| 1640 |
dir, fname, 0, res_dir); |
| 1641 |
if (ret == 1) |
| 1642 |
goto out_find; |
| 1643 |
|
| 1644 |
out: |
| 1645 |
brelse(is.iloc.bh); |
| 1646 |
if (ret < 0) |
| 1647 |
is.iloc.bh = ERR_PTR(ret); |
| 1648 |
else |
| 1649 |
is.iloc.bh = NULL; |
| 1650 |
out_find: |
| 1651 |
up_read(&EXT4_I(dir)->xattr_sem); |
| 1652 |
return is.iloc.bh; |
| 1653 |
} |
| 1654 |
|
| 1655 |
int ext4_delete_inline_entry(handle_t *handle, |
| 1656 |
struct inode *dir, |
| 1657 |
struct ext4_dir_entry_2 *de_del, |
| 1658 |
struct buffer_head *bh, |
| 1659 |
int *has_inline_data) |
| 1660 |
{ |
| 1661 |
int err, inline_size, no_expand; |
| 1662 |
struct ext4_iloc iloc; |
| 1663 |
void *inline_start; |
| 1664 |
|
| 1665 |
err = ext4_get_inode_loc(dir, &iloc); |
| 1666 |
if (err) |
| 1667 |
return err; |
| 1668 |
|
| 1669 |
ext4_write_lock_xattr(dir, &no_expand); |
| 1670 |
if (!ext4_has_inline_data(dir)) { |
| 1671 |
*has_inline_data = 0; |
| 1672 |
goto out; |
| 1673 |
} |
| 1674 |
|
| 1675 |
if ((void *)de_del - ((void *)ext4_raw_inode(&iloc)->i_block) < |
| 1676 |
EXT4_MIN_INLINE_DATA_SIZE) { |
| 1677 |
inline_start = (void *)ext4_raw_inode(&iloc)->i_block + |
| 1678 |
EXT4_INLINE_DOTDOT_SIZE; |
| 1679 |
inline_size = EXT4_MIN_INLINE_DATA_SIZE - |
| 1680 |
EXT4_INLINE_DOTDOT_SIZE; |
| 1681 |
} else { |
| 1682 |
inline_start = ext4_get_inline_xattr_pos(dir, &iloc); |
| 1683 |
inline_size = ext4_get_inline_size(dir) - |
| 1684 |
EXT4_MIN_INLINE_DATA_SIZE; |
| 1685 |
} |
| 1686 |
|
| 1687 |
BUFFER_TRACE(bh, "get_write_access"); |
| 1688 |
err = ext4_journal_get_write_access(handle, dir->i_sb, bh, |
| 1689 |
EXT4_JTR_NONE); |
| 1690 |
if (err) |
| 1691 |
goto out; |
| 1692 |
|
| 1693 |
err = ext4_generic_delete_entry(dir, de_del, bh, |
| 1694 |
inline_start, inline_size, 0); |
| 1695 |
if (err) |
| 1696 |
goto out; |
| 1697 |
|
| 1698 |
ext4_show_inline_dir(dir, iloc.bh, inline_start, inline_size); |
| 1699 |
out: |
| 1700 |
ext4_write_unlock_xattr(dir, &no_expand); |
| 1701 |
if (likely(err == 0)) |
| 1702 |
err = ext4_mark_inode_dirty(handle, dir); |
| 1703 |
brelse(iloc.bh); |
| 1704 |
if (err != -ENOENT) |
| 1705 |
ext4_std_error(dir->i_sb, err); |
| 1706 |
return err; |
| 1707 |
} |
| 1708 |
|
| 1709 |
/* |
| 1710 |
* Get the inline dentry at offset. |
| 1711 |
*/ |
| 1712 |
static inline struct ext4_dir_entry_2 * |
| 1713 |
ext4_get_inline_entry(struct inode *inode, |
| 1714 |
struct ext4_iloc *iloc, |
| 1715 |
unsigned int offset, |
| 1716 |
void **inline_start, |
| 1717 |
int *inline_size) |
| 1718 |
{ |
| 1719 |
void *inline_pos; |
| 1720 |
|
| 1721 |
BUG_ON(offset > ext4_get_inline_size(inode)); |
| 1722 |
|
| 1723 |
if (offset < EXT4_MIN_INLINE_DATA_SIZE) { |
| 1724 |
inline_pos = (void *)ext4_raw_inode(iloc)->i_block; |
| 1725 |
*inline_size = EXT4_MIN_INLINE_DATA_SIZE; |
| 1726 |
} else { |
| 1727 |
inline_pos = ext4_get_inline_xattr_pos(inode, iloc); |
| 1728 |
offset -= EXT4_MIN_INLINE_DATA_SIZE; |
| 1729 |
*inline_size = ext4_get_inline_size(inode) - |
| 1730 |
EXT4_MIN_INLINE_DATA_SIZE; |
| 1731 |
} |
| 1732 |
|
| 1733 |
if (inline_start) |
| 1734 |
*inline_start = inline_pos; |
| 1735 |
return (struct ext4_dir_entry_2 *)(inline_pos + offset); |
| 1736 |
} |
| 1737 |
|
| 1738 |
bool empty_inline_dir(struct inode *dir, int *has_inline_data) |
| 1739 |
{ |
| 1740 |
int err, inline_size; |
| 1741 |
struct ext4_iloc iloc; |
| 1742 |
size_t inline_len; |
| 1743 |
void *inline_pos; |
| 1744 |
unsigned int offset; |
| 1745 |
struct ext4_dir_entry_2 *de; |
| 1746 |
bool ret = false; |
| 1747 |
|
| 1748 |
err = ext4_get_inode_loc(dir, &iloc); |
| 1749 |
if (err) { |
| 1750 |
EXT4_ERROR_INODE_ERR(dir, -err, |
| 1751 |
"error %d getting inode %lu block", |
| 1752 |
err, dir->i_ino); |
| 1753 |
return false; |
| 1754 |
} |
| 1755 |
|
| 1756 |
down_read(&EXT4_I(dir)->xattr_sem); |
| 1757 |
if (!ext4_has_inline_data(dir)) { |
| 1758 |
*has_inline_data = 0; |
| 1759 |
ret = true; |
| 1760 |
goto out; |
| 1761 |
} |
| 1762 |
|
| 1763 |
de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block; |
| 1764 |
if (!le32_to_cpu(de->inode)) { |
| 1765 |
ext4_warning(dir->i_sb, |
| 1766 |
"bad inline directory (dir #%lu) - no `..'", |
| 1767 |
dir->i_ino); |
| 1768 |
goto out; |
| 1769 |
} |
| 1770 |
|
| 1771 |
inline_len = ext4_get_inline_size(dir); |
| 1772 |
offset = EXT4_INLINE_DOTDOT_SIZE; |
| 1773 |
while (offset < inline_len) { |
| 1774 |
de = ext4_get_inline_entry(dir, &iloc, offset, |
| 1775 |
&inline_pos, &inline_size); |
| 1776 |
if (ext4_check_dir_entry(dir, NULL, de, |
| 1777 |
iloc.bh, inline_pos, |
| 1778 |
inline_size, offset)) { |
| 1779 |
ext4_warning(dir->i_sb, |
| 1780 |
"bad inline directory (dir #%lu) - " |
| 1781 |
"inode %u, rec_len %u, name_len %d" |
| 1782 |
"inline size %d", |
| 1783 |
dir->i_ino, le32_to_cpu(de->inode), |
| 1784 |
le16_to_cpu(de->rec_len), de->name_len, |
| 1785 |
inline_size); |
| 1786 |
goto out; |
| 1787 |
} |
| 1788 |
if (le32_to_cpu(de->inode)) { |
| 1789 |
goto out; |
| 1790 |
} |
| 1791 |
offset += ext4_rec_len_from_disk(de->rec_len, inline_size); |
| 1792 |
} |
| 1793 |
|
| 1794 |
ret = true; |
| 1795 |
out: |
| 1796 |
up_read(&EXT4_I(dir)->xattr_sem); |
| 1797 |
brelse(iloc.bh); |
| 1798 |
return ret; |
| 1799 |
} |
| 1800 |
|
| 1801 |
int ext4_destroy_inline_data(handle_t *handle, struct inode *inode) |
| 1802 |
{ |
| 1803 |
int ret, no_expand; |
| 1804 |
|
| 1805 |
ext4_write_lock_xattr(inode, &no_expand); |
| 1806 |
ret = ext4_destroy_inline_data_nolock(handle, inode); |
| 1807 |
ext4_write_unlock_xattr(inode, &no_expand); |
| 1808 |
|
| 1809 |
return ret; |
| 1810 |
} |
| 1811 |
|
| 1812 |
int ext4_inline_data_iomap(struct inode *inode, struct iomap *iomap) |
| 1813 |
{ |
| 1814 |
__u64 addr; |
| 1815 |
int error = -EAGAIN; |
| 1816 |
struct ext4_iloc iloc; |
| 1817 |
|
| 1818 |
down_read(&EXT4_I(inode)->xattr_sem); |
| 1819 |
if (!ext4_has_inline_data(inode)) |
| 1820 |
goto out; |
| 1821 |
|
| 1822 |
error = ext4_get_inode_loc(inode, &iloc); |
| 1823 |
if (error) |
| 1824 |
goto out; |
| 1825 |
|
| 1826 |
addr = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits; |
| 1827 |
addr += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data; |
| 1828 |
addr += offsetof(struct ext4_inode, i_block); |
| 1829 |
|
| 1830 |
brelse(iloc.bh); |
| 1831 |
|
| 1832 |
iomap->addr = addr; |
| 1833 |
iomap->offset = 0; |
| 1834 |
iomap->length = min_t(loff_t, ext4_get_inline_size(inode), |
| 1835 |
i_size_read(inode)); |
| 1836 |
iomap->type = IOMAP_INLINE; |
| 1837 |
iomap->flags = 0; |
| 1838 |
|
| 1839 |
out: |
| 1840 |
up_read(&EXT4_I(inode)->xattr_sem); |
| 1841 |
return error; |
| 1842 |
} |
| 1843 |
|
| 1844 |
int ext4_inline_data_truncate(struct inode *inode, int *has_inline) |
| 1845 |
{ |
| 1846 |
handle_t *handle; |
| 1847 |
int inline_size, value_len, needed_blocks, no_expand, err = 0; |
| 1848 |
size_t i_size; |
| 1849 |
void *value = NULL; |
| 1850 |
struct ext4_xattr_ibody_find is = { |
| 1851 |
.s = { .not_found = -ENODATA, }, |
| 1852 |
}; |
| 1853 |
struct ext4_xattr_info i = { |
| 1854 |
.name_index = EXT4_XATTR_INDEX_SYSTEM, |
| 1855 |
.name = EXT4_XATTR_SYSTEM_DATA, |
| 1856 |
}; |
| 1857 |
|
| 1858 |
|
| 1859 |
needed_blocks = ext4_chunk_trans_extent(inode, 1); |
| 1860 |
handle = ext4_journal_start(inode, EXT4_HT_INODE, needed_blocks); |
| 1861 |
if (IS_ERR(handle)) |
| 1862 |
return PTR_ERR(handle); |
| 1863 |
|
| 1864 |
ext4_write_lock_xattr(inode, &no_expand); |
| 1865 |
if (!ext4_has_inline_data(inode)) { |
| 1866 |
ext4_write_unlock_xattr(inode, &no_expand); |
| 1867 |
*has_inline = 0; |
| 1868 |
ext4_journal_stop(handle); |
| 1869 |
return 0; |
| 1870 |
} |
| 1871 |
|
| 1872 |
if ((err = ext4_orphan_add(handle, inode)) != 0) |
| 1873 |
goto out; |
| 1874 |
|
| 1875 |
if ((err = ext4_get_inode_loc(inode, &is.iloc)) != 0) |
| 1876 |
goto out; |
| 1877 |
|
| 1878 |
down_write(&EXT4_I(inode)->i_data_sem); |
| 1879 |
i_size = inode->i_size; |
| 1880 |
inline_size = ext4_get_inline_size(inode); |
| 1881 |
EXT4_I(inode)->i_disksize = i_size; |
| 1882 |
|
| 1883 |
if (i_size < inline_size) { |
| 1884 |
/* |
| 1885 |
* if there's inline data to truncate and this file was |
| 1886 |
* converted to extents after that inline data was written, |
| 1887 |
* the extent status cache must be cleared to avoid leaving |
| 1888 |
* behind stale delayed allocated extent entries |
| 1889 |
*/ |
| 1890 |
if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) |
| 1891 |
ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS); |
| 1892 |
|
| 1893 |
/* Clear the content in the xattr space. */ |
| 1894 |
if (inline_size > EXT4_MIN_INLINE_DATA_SIZE) { |
| 1895 |
if ((err = ext4_xattr_ibody_find(inode, &i, &is)) != 0) |
| 1896 |
goto out_error; |
| 1897 |
|
| 1898 |
if (is.s.not_found) { |
| 1899 |
EXT4_ERROR_INODE(inode, |
| 1900 |
"missing inline data xattr"); |
| 1901 |
err = -EFSCORRUPTED; |
| 1902 |
goto out_error; |
| 1903 |
} |
| 1904 |
|
| 1905 |
value_len = le32_to_cpu(is.s.here->e_value_size); |
| 1906 |
value = kmalloc(value_len, GFP_NOFS); |
| 1907 |
if (!value) { |
| 1908 |
err = -ENOMEM; |
| 1909 |
goto out_error; |
| 1910 |
} |
| 1911 |
|
| 1912 |
err = ext4_xattr_ibody_get(inode, i.name_index, |
| 1913 |
i.name, value, value_len); |
| 1914 |
if (err <= 0) |
| 1915 |
goto out_error; |
| 1916 |
|
| 1917 |
i.value = value; |
| 1918 |
i.value_len = i_size > EXT4_MIN_INLINE_DATA_SIZE ? |
| 1919 |
i_size - EXT4_MIN_INLINE_DATA_SIZE : 0; |
| 1920 |
err = ext4_xattr_ibody_set(handle, inode, &i, &is); |
| 1921 |
if (err) |
| 1922 |
goto out_error; |
| 1923 |
} |
| 1924 |
|
| 1925 |
/* Clear the content within i_blocks. */ |
| 1926 |
if (i_size < EXT4_MIN_INLINE_DATA_SIZE) { |
| 1927 |
void *p = (void *) ext4_raw_inode(&is.iloc)->i_block; |
| 1928 |
memset(p + i_size, 0, |
| 1929 |
EXT4_MIN_INLINE_DATA_SIZE - i_size); |
| 1930 |
} |
| 1931 |
|
| 1932 |
EXT4_I(inode)->i_inline_size = i_size < |
| 1933 |
EXT4_MIN_INLINE_DATA_SIZE ? |
| 1934 |
EXT4_MIN_INLINE_DATA_SIZE : i_size; |
| 1935 |
} |
| 1936 |
|
| 1937 |
out_error: |
| 1938 |
up_write(&EXT4_I(inode)->i_data_sem); |
| 1939 |
out: |
| 1940 |
brelse(is.iloc.bh); |
| 1941 |
ext4_write_unlock_xattr(inode, &no_expand); |
| 1942 |
kfree(value); |
| 1943 |
if (inode->i_nlink) |
| 1944 |
ext4_orphan_del(handle, inode); |
| 1945 |
|
| 1946 |
if (err == 0) { |
| 1947 |
inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode)); |
| 1948 |
err = ext4_mark_inode_dirty(handle, inode); |
| 1949 |
if (IS_SYNC(inode)) |
| 1950 |
ext4_handle_sync(handle); |
| 1951 |
} |
| 1952 |
ext4_journal_stop(handle); |
| 1953 |
return err; |
| 1954 |
} |
| 1955 |
|
| 1956 |
int ext4_convert_inline_data(struct inode *inode) |
| 1957 |
{ |
| 1958 |
int error, needed_blocks, no_expand; |
| 1959 |
handle_t *handle; |
| 1960 |
struct ext4_iloc iloc; |
| 1961 |
|
| 1962 |
if (!ext4_has_inline_data(inode)) { |
| 1963 |
ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA); |
| 1964 |
return 0; |
| 1965 |
} else if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { |
| 1966 |
/* |
| 1967 |
* Inode has inline data but EXT4_STATE_MAY_INLINE_DATA is |
| 1968 |
* cleared. This means we are in the middle of moving of |
| 1969 |
* inline data to delay allocated block. Just force writeout |
| 1970 |
* here to finish conversion. |
| 1971 |
*/ |
| 1972 |
error = filemap_flush(inode->i_mapping); |
| 1973 |
if (error) |
| 1974 |
return error; |
| 1975 |
if (!ext4_has_inline_data(inode)) |
| 1976 |
return 0; |
| 1977 |
} |
| 1978 |
|
| 1979 |
needed_blocks = ext4_chunk_trans_extent(inode, 1); |
| 1980 |
|
| 1981 |
iloc.bh = NULL; |
| 1982 |
error = ext4_get_inode_loc(inode, &iloc); |
| 1983 |
if (error) |
| 1984 |
return error; |
| 1985 |
|
| 1986 |
handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks); |
| 1987 |
if (IS_ERR(handle)) { |
| 1988 |
error = PTR_ERR(handle); |
| 1989 |
goto out_free; |
| 1990 |
} |
| 1991 |
|
| 1992 |
ext4_write_lock_xattr(inode, &no_expand); |
| 1993 |
if (ext4_has_inline_data(inode)) |
| 1994 |
error = ext4_convert_inline_data_nolock(handle, inode, &iloc); |
| 1995 |
ext4_write_unlock_xattr(inode, &no_expand); |
| 1996 |
ext4_journal_stop(handle); |
| 1997 |
out_free: |
| 1998 |
brelse(iloc.bh); |
| 1999 |
return error; |
| 2000 |
} |