| 1 |
// SPDX-License-Identifier: GPL-2.0 |
| 2 |
/* |
| 3 |
* Copyright (c) 2003-2006, Cluster File Systems, Inc, [email protected] |
| 4 |
* Written by Alex Tomas <[email protected]> |
| 5 |
* |
| 6 |
* Architecture independence: |
| 7 |
* Copyright (c) 2005, Bull S.A. |
| 8 |
* Written by Pierre Peiffer <[email protected]> |
| 9 |
*/ |
| 10 |
|
| 11 |
/* |
| 12 |
* Extents support for EXT4 |
| 13 |
* |
| 14 |
* TODO: |
| 15 |
* - ext4*_error() should be used in some situations |
| 16 |
* - analyze all BUG()/BUG_ON(), use -EIO where appropriate |
| 17 |
* - smart tree reduction |
| 18 |
*/ |
| 19 |
|
| 20 |
#include <linux/fs.h> |
| 21 |
#include <linux/time.h> |
| 22 |
#include <linux/jbd2.h> |
| 23 |
#include <linux/highuid.h> |
| 24 |
#include <linux/pagemap.h> |
| 25 |
#include <linux/quotaops.h> |
| 26 |
#include <linux/string.h> |
| 27 |
#include <linux/slab.h> |
| 28 |
#include <linux/uaccess.h> |
| 29 |
#include <linux/fiemap.h> |
| 30 |
#include <linux/iomap.h> |
| 31 |
#include <linux/sched/mm.h> |
| 32 |
#include "ext4_jbd2.h" |
| 33 |
#include "ext4_extents.h" |
| 34 |
#include "xattr.h" |
| 35 |
#include <kunit/static_stub.h> |
| 36 |
|
| 37 |
#include <trace/events/ext4.h> |
| 38 |
|
| 39 |
/* |
| 40 |
* used by extent splitting. |
| 41 |
*/ |
| 42 |
#define EXT4_EXT_MAY_ZEROOUT 0x1 /* safe to zeroout if split fails \ |
| 43 |
due to ENOSPC */ |
| 44 |
static struct ext4_ext_path *ext4_split_convert_extents( |
| 45 |
handle_t *handle, struct inode *inode, struct ext4_map_blocks *map, |
| 46 |
struct ext4_ext_path *path, int flags, unsigned int *allocated); |
| 47 |
|
| 48 |
static __le32 ext4_extent_block_csum(struct inode *inode, |
| 49 |
struct ext4_extent_header *eh) |
| 50 |
{ |
| 51 |
struct ext4_inode_info *ei = EXT4_I(inode); |
| 52 |
__u32 csum; |
| 53 |
|
| 54 |
csum = ext4_chksum(ei->i_csum_seed, (__u8 *)eh, |
| 55 |
EXT4_EXTENT_TAIL_OFFSET(eh)); |
| 56 |
return cpu_to_le32(csum); |
| 57 |
} |
| 58 |
|
| 59 |
static int ext4_extent_block_csum_verify(struct inode *inode, |
| 60 |
struct ext4_extent_header *eh) |
| 61 |
{ |
| 62 |
struct ext4_extent_tail *et; |
| 63 |
|
| 64 |
if (!ext4_has_feature_metadata_csum(inode->i_sb)) |
| 65 |
return 1; |
| 66 |
|
| 67 |
et = find_ext4_extent_tail(eh); |
| 68 |
if (et->et_checksum != ext4_extent_block_csum(inode, eh)) |
| 69 |
return 0; |
| 70 |
return 1; |
| 71 |
} |
| 72 |
|
| 73 |
static void ext4_extent_block_csum_set(struct inode *inode, |
| 74 |
struct ext4_extent_header *eh) |
| 75 |
{ |
| 76 |
struct ext4_extent_tail *et; |
| 77 |
|
| 78 |
if (!ext4_has_feature_metadata_csum(inode->i_sb)) |
| 79 |
return; |
| 80 |
|
| 81 |
et = find_ext4_extent_tail(eh); |
| 82 |
et->et_checksum = ext4_extent_block_csum(inode, eh); |
| 83 |
} |
| 84 |
|
| 85 |
static struct ext4_ext_path *ext4_split_extent_at(handle_t *handle, |
| 86 |
struct inode *inode, |
| 87 |
struct ext4_ext_path *path, |
| 88 |
ext4_lblk_t split, int flags); |
| 89 |
|
| 90 |
static int ext4_ext_trunc_restart_fn(struct inode *inode, int *dropped) |
| 91 |
{ |
| 92 |
/* |
| 93 |
* Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this |
| 94 |
* moment, get_block can be called only for blocks inside i_size since |
| 95 |
* page cache has been already dropped and writes are blocked by |
| 96 |
* i_rwsem. So we can safely drop the i_data_sem here. |
| 97 |
*/ |
| 98 |
BUG_ON(EXT4_JOURNAL(inode) == NULL); |
| 99 |
ext4_discard_preallocations(inode); |
| 100 |
up_write(&EXT4_I(inode)->i_data_sem); |
| 101 |
*dropped = 1; |
| 102 |
return 0; |
| 103 |
} |
| 104 |
|
| 105 |
static inline void ext4_ext_path_brelse(struct ext4_ext_path *path) |
| 106 |
{ |
| 107 |
brelse(path->p_bh); |
| 108 |
path->p_bh = NULL; |
| 109 |
} |
| 110 |
|
| 111 |
static void ext4_ext_drop_refs(struct ext4_ext_path *path) |
| 112 |
{ |
| 113 |
int depth, i; |
| 114 |
|
| 115 |
if (IS_ERR_OR_NULL(path)) |
| 116 |
return; |
| 117 |
depth = path->p_depth; |
| 118 |
for (i = 0; i <= depth; i++, path++) |
| 119 |
ext4_ext_path_brelse(path); |
| 120 |
} |
| 121 |
|
| 122 |
void ext4_free_ext_path(struct ext4_ext_path *path) |
| 123 |
{ |
| 124 |
if (IS_ERR_OR_NULL(path)) |
| 125 |
return; |
| 126 |
ext4_ext_drop_refs(path); |
| 127 |
kfree(path); |
| 128 |
} |
| 129 |
|
| 130 |
/* |
| 131 |
* Make sure 'handle' has at least 'check_cred' credits. If not, restart |
| 132 |
* transaction with 'restart_cred' credits. The function drops i_data_sem |
| 133 |
* when restarting transaction and gets it after transaction is restarted. |
| 134 |
* |
| 135 |
* The function returns 0 on success, 1 if transaction had to be restarted, |
| 136 |
* and < 0 in case of fatal error. |
| 137 |
*/ |
| 138 |
int ext4_datasem_ensure_credits(handle_t *handle, struct inode *inode, |
| 139 |
int check_cred, int restart_cred, |
| 140 |
int revoke_cred) |
| 141 |
{ |
| 142 |
int ret; |
| 143 |
int dropped = 0; |
| 144 |
|
| 145 |
ret = ext4_journal_ensure_credits_fn(handle, check_cred, restart_cred, |
| 146 |
revoke_cred, ext4_ext_trunc_restart_fn(inode, &dropped)); |
| 147 |
if (dropped) |
| 148 |
down_write(&EXT4_I(inode)->i_data_sem); |
| 149 |
return ret; |
| 150 |
} |
| 151 |
|
| 152 |
/* |
| 153 |
* could return: |
| 154 |
* - EROFS |
| 155 |
* - ENOMEM |
| 156 |
*/ |
| 157 |
static int ext4_ext_get_access(handle_t *handle, struct inode *inode, |
| 158 |
struct ext4_ext_path *path) |
| 159 |
{ |
| 160 |
int err = 0; |
| 161 |
|
| 162 |
if (path->p_bh) { |
| 163 |
/* path points to block */ |
| 164 |
BUFFER_TRACE(path->p_bh, "get_write_access"); |
| 165 |
err = ext4_journal_get_write_access(handle, inode->i_sb, |
| 166 |
path->p_bh, EXT4_JTR_NONE); |
| 167 |
/* |
| 168 |
* The extent buffer's verified bit will be set again in |
| 169 |
* __ext4_ext_dirty(). We could leave an inconsistent |
| 170 |
* buffer if the extents updating procudure break off du |
| 171 |
* to some error happens, force to check it again. |
| 172 |
*/ |
| 173 |
if (!err) |
| 174 |
clear_buffer_verified(path->p_bh); |
| 175 |
} |
| 176 |
/* path points to leaf/index in inode body */ |
| 177 |
/* we use in-core data, no need to protect them */ |
| 178 |
return err; |
| 179 |
} |
| 180 |
|
| 181 |
/* |
| 182 |
* could return: |
| 183 |
* - EROFS |
| 184 |
* - ENOMEM |
| 185 |
* - EIO |
| 186 |
*/ |
| 187 |
int __ext4_ext_dirty(const char *where, unsigned int line, |
| 188 |
handle_t *handle, struct inode *inode, |
| 189 |
struct ext4_ext_path *path) |
| 190 |
{ |
| 191 |
int err; |
| 192 |
|
| 193 |
KUNIT_STATIC_STUB_REDIRECT(__ext4_ext_dirty, where, line, handle, inode, |
| 194 |
path); |
| 195 |
|
| 196 |
WARN_ON(!rwsem_is_locked(&EXT4_I(inode)->i_data_sem)); |
| 197 |
if (path->p_bh) { |
| 198 |
ext4_extent_block_csum_set(inode, ext_block_hdr(path->p_bh)); |
| 199 |
/* path points to block */ |
| 200 |
err = __ext4_handle_dirty_metadata(where, line, handle, |
| 201 |
inode, path->p_bh); |
| 202 |
/* Extents updating done, re-set verified flag */ |
| 203 |
if (!err) |
| 204 |
set_buffer_verified(path->p_bh); |
| 205 |
} else { |
| 206 |
/* path points to leaf/index in inode body */ |
| 207 |
err = ext4_mark_inode_dirty(handle, inode); |
| 208 |
} |
| 209 |
return err; |
| 210 |
} |
| 211 |
|
| 212 |
#define ext4_ext_dirty(handle, inode, path) \ |
| 213 |
__ext4_ext_dirty(__func__, __LINE__, (handle), (inode), (path)) |
| 214 |
|
| 215 |
static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode, |
| 216 |
struct ext4_ext_path *path, |
| 217 |
ext4_lblk_t block) |
| 218 |
{ |
| 219 |
if (path) { |
| 220 |
int depth = path->p_depth; |
| 221 |
struct ext4_extent *ex; |
| 222 |
|
| 223 |
/* |
| 224 |
* Try to predict block placement assuming that we are |
| 225 |
* filling in a file which will eventually be |
| 226 |
* non-sparse --- i.e., in the case of libbfd writing |
| 227 |
* an ELF object sections out-of-order but in a way |
| 228 |
* the eventually results in a contiguous object or |
| 229 |
* executable file, or some database extending a table |
| 230 |
* space file. However, this is actually somewhat |
| 231 |
* non-ideal if we are writing a sparse file such as |
| 232 |
* qemu or KVM writing a raw image file that is going |
| 233 |
* to stay fairly sparse, since it will end up |
| 234 |
* fragmenting the file system's free space. Maybe we |
| 235 |
* should have some hueristics or some way to allow |
| 236 |
* userspace to pass a hint to file system, |
| 237 |
* especially if the latter case turns out to be |
| 238 |
* common. |
| 239 |
*/ |
| 240 |
ex = path[depth].p_ext; |
| 241 |
if (ex) { |
| 242 |
ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex); |
| 243 |
ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block); |
| 244 |
|
| 245 |
if (block > ext_block) |
| 246 |
return ext_pblk + (block - ext_block); |
| 247 |
else |
| 248 |
return ext_pblk - (ext_block - block); |
| 249 |
} |
| 250 |
|
| 251 |
/* it looks like index is empty; |
| 252 |
* try to find starting block from index itself */ |
| 253 |
if (path[depth].p_bh) |
| 254 |
return path[depth].p_bh->b_blocknr; |
| 255 |
} |
| 256 |
|
| 257 |
/* OK. use inode's group */ |
| 258 |
return ext4_inode_to_goal_block(inode); |
| 259 |
} |
| 260 |
|
| 261 |
/* |
| 262 |
* Allocation for a meta data block |
| 263 |
*/ |
| 264 |
static ext4_fsblk_t |
| 265 |
ext4_ext_new_meta_block(handle_t *handle, struct inode *inode, |
| 266 |
struct ext4_ext_path *path, |
| 267 |
struct ext4_extent *ex, int *err, unsigned int flags) |
| 268 |
{ |
| 269 |
ext4_fsblk_t goal, newblock; |
| 270 |
|
| 271 |
goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block)); |
| 272 |
newblock = ext4_new_meta_blocks(handle, inode, goal, flags, |
| 273 |
NULL, err); |
| 274 |
return newblock; |
| 275 |
} |
| 276 |
|
| 277 |
static inline int ext4_ext_space_block(struct inode *inode, int check) |
| 278 |
{ |
| 279 |
int size; |
| 280 |
|
| 281 |
size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header)) |
| 282 |
/ sizeof(struct ext4_extent); |
| 283 |
#ifdef AGGRESSIVE_TEST |
| 284 |
if (!check && size > 6) |
| 285 |
size = 6; |
| 286 |
#endif |
| 287 |
return size; |
| 288 |
} |
| 289 |
|
| 290 |
static inline int ext4_ext_space_block_idx(struct inode *inode, int check) |
| 291 |
{ |
| 292 |
int size; |
| 293 |
|
| 294 |
size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header)) |
| 295 |
/ sizeof(struct ext4_extent_idx); |
| 296 |
#ifdef AGGRESSIVE_TEST |
| 297 |
if (!check && size > 5) |
| 298 |
size = 5; |
| 299 |
#endif |
| 300 |
return size; |
| 301 |
} |
| 302 |
|
| 303 |
static inline int ext4_ext_space_root(struct inode *inode, int check) |
| 304 |
{ |
| 305 |
int size; |
| 306 |
|
| 307 |
size = sizeof(EXT4_I(inode)->i_data); |
| 308 |
size -= sizeof(struct ext4_extent_header); |
| 309 |
size /= sizeof(struct ext4_extent); |
| 310 |
#ifdef AGGRESSIVE_TEST |
| 311 |
if (!check && size > 3) |
| 312 |
size = 3; |
| 313 |
#endif |
| 314 |
return size; |
| 315 |
} |
| 316 |
|
| 317 |
static inline int ext4_ext_space_root_idx(struct inode *inode, int check) |
| 318 |
{ |
| 319 |
int size; |
| 320 |
|
| 321 |
size = sizeof(EXT4_I(inode)->i_data); |
| 322 |
size -= sizeof(struct ext4_extent_header); |
| 323 |
size /= sizeof(struct ext4_extent_idx); |
| 324 |
#ifdef AGGRESSIVE_TEST |
| 325 |
if (!check && size > 4) |
| 326 |
size = 4; |
| 327 |
#endif |
| 328 |
return size; |
| 329 |
} |
| 330 |
|
| 331 |
static inline struct ext4_ext_path * |
| 332 |
ext4_force_split_extent_at(handle_t *handle, struct inode *inode, |
| 333 |
struct ext4_ext_path *path, ext4_lblk_t lblk, |
| 334 |
int nofail) |
| 335 |
{ |
| 336 |
int flags = EXT4_EX_NOCACHE | EXT4_GET_BLOCKS_SPLIT_NOMERGE; |
| 337 |
|
| 338 |
if (nofail) |
| 339 |
flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL | EXT4_EX_NOFAIL; |
| 340 |
|
| 341 |
return ext4_split_extent_at(handle, inode, path, lblk, flags); |
| 342 |
} |
| 343 |
|
| 344 |
static int |
| 345 |
ext4_ext_max_entries(struct inode *inode, int depth) |
| 346 |
{ |
| 347 |
int max; |
| 348 |
|
| 349 |
if (depth == ext_depth(inode)) { |
| 350 |
if (depth == 0) |
| 351 |
max = ext4_ext_space_root(inode, 1); |
| 352 |
else |
| 353 |
max = ext4_ext_space_root_idx(inode, 1); |
| 354 |
} else { |
| 355 |
if (depth == 0) |
| 356 |
max = ext4_ext_space_block(inode, 1); |
| 357 |
else |
| 358 |
max = ext4_ext_space_block_idx(inode, 1); |
| 359 |
} |
| 360 |
|
| 361 |
return max; |
| 362 |
} |
| 363 |
|
| 364 |
static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext) |
| 365 |
{ |
| 366 |
ext4_fsblk_t block = ext4_ext_pblock(ext); |
| 367 |
int len = ext4_ext_get_actual_len(ext); |
| 368 |
ext4_lblk_t lblock = le32_to_cpu(ext->ee_block); |
| 369 |
|
| 370 |
/* |
| 371 |
* We allow neither: |
| 372 |
* - zero length |
| 373 |
* - overflow/wrap-around |
| 374 |
*/ |
| 375 |
if (lblock + len <= lblock) |
| 376 |
return 0; |
| 377 |
return ext4_inode_block_valid(inode, block, len); |
| 378 |
} |
| 379 |
|
| 380 |
static int ext4_valid_extent_idx(struct inode *inode, |
| 381 |
struct ext4_extent_idx *ext_idx) |
| 382 |
{ |
| 383 |
ext4_fsblk_t block = ext4_idx_pblock(ext_idx); |
| 384 |
|
| 385 |
return ext4_inode_block_valid(inode, block, 1); |
| 386 |
} |
| 387 |
|
| 388 |
static int ext4_valid_extent_entries(struct inode *inode, |
| 389 |
struct ext4_extent_header *eh, |
| 390 |
ext4_lblk_t lblk, ext4_fsblk_t *pblk, |
| 391 |
int depth) |
| 392 |
{ |
| 393 |
unsigned short entries; |
| 394 |
ext4_lblk_t lblock = 0; |
| 395 |
ext4_lblk_t cur = 0; |
| 396 |
|
| 397 |
if (eh->eh_entries == 0) |
| 398 |
return 1; |
| 399 |
|
| 400 |
entries = le16_to_cpu(eh->eh_entries); |
| 401 |
|
| 402 |
if (depth == 0) { |
| 403 |
/* leaf entries */ |
| 404 |
struct ext4_extent *ext = EXT_FIRST_EXTENT(eh); |
| 405 |
|
| 406 |
/* |
| 407 |
* The logical block in the first entry should equal to |
| 408 |
* the number in the index block. |
| 409 |
*/ |
| 410 |
if (depth != ext_depth(inode) && |
| 411 |
lblk != le32_to_cpu(ext->ee_block)) |
| 412 |
return 0; |
| 413 |
while (entries) { |
| 414 |
if (!ext4_valid_extent(inode, ext)) |
| 415 |
return 0; |
| 416 |
|
| 417 |
/* Check for overlapping extents */ |
| 418 |
lblock = le32_to_cpu(ext->ee_block); |
| 419 |
if (lblock < cur) { |
| 420 |
*pblk = ext4_ext_pblock(ext); |
| 421 |
return 0; |
| 422 |
} |
| 423 |
cur = lblock + ext4_ext_get_actual_len(ext); |
| 424 |
ext++; |
| 425 |
entries--; |
| 426 |
} |
| 427 |
} else { |
| 428 |
struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh); |
| 429 |
|
| 430 |
/* |
| 431 |
* The logical block in the first entry should equal to |
| 432 |
* the number in the parent index block. |
| 433 |
*/ |
| 434 |
if (depth != ext_depth(inode) && |
| 435 |
lblk != le32_to_cpu(ext_idx->ei_block)) |
| 436 |
return 0; |
| 437 |
while (entries) { |
| 438 |
if (!ext4_valid_extent_idx(inode, ext_idx)) |
| 439 |
return 0; |
| 440 |
|
| 441 |
/* Check for overlapping index extents */ |
| 442 |
lblock = le32_to_cpu(ext_idx->ei_block); |
| 443 |
if (lblock < cur) { |
| 444 |
*pblk = ext4_idx_pblock(ext_idx); |
| 445 |
return 0; |
| 446 |
} |
| 447 |
ext_idx++; |
| 448 |
entries--; |
| 449 |
cur = lblock + 1; |
| 450 |
} |
| 451 |
} |
| 452 |
return 1; |
| 453 |
} |
| 454 |
|
| 455 |
static int __ext4_ext_check(const char *function, unsigned int line, |
| 456 |
struct inode *inode, struct ext4_extent_header *eh, |
| 457 |
int depth, ext4_fsblk_t pblk, ext4_lblk_t lblk) |
| 458 |
{ |
| 459 |
const char *error_msg; |
| 460 |
int max = 0, err = -EFSCORRUPTED; |
| 461 |
|
| 462 |
if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) { |
| 463 |
error_msg = "invalid magic"; |
| 464 |
goto corrupted; |
| 465 |
} |
| 466 |
if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) { |
| 467 |
error_msg = "unexpected eh_depth"; |
| 468 |
goto corrupted; |
| 469 |
} |
| 470 |
if (unlikely(eh->eh_max == 0)) { |
| 471 |
error_msg = "invalid eh_max"; |
| 472 |
goto corrupted; |
| 473 |
} |
| 474 |
max = ext4_ext_max_entries(inode, depth); |
| 475 |
if (unlikely(le16_to_cpu(eh->eh_max) > max)) { |
| 476 |
error_msg = "too large eh_max"; |
| 477 |
goto corrupted; |
| 478 |
} |
| 479 |
if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) { |
| 480 |
error_msg = "invalid eh_entries"; |
| 481 |
goto corrupted; |
| 482 |
} |
| 483 |
if (unlikely((eh->eh_entries == 0) && (depth > 0))) { |
| 484 |
error_msg = "eh_entries is 0 but eh_depth is > 0"; |
| 485 |
goto corrupted; |
| 486 |
} |
| 487 |
if (!ext4_valid_extent_entries(inode, eh, lblk, &pblk, depth)) { |
| 488 |
error_msg = "invalid extent entries"; |
| 489 |
goto corrupted; |
| 490 |
} |
| 491 |
if (unlikely(depth > 32)) { |
| 492 |
error_msg = "too large eh_depth"; |
| 493 |
goto corrupted; |
| 494 |
} |
| 495 |
/* Verify checksum on non-root extent tree nodes */ |
| 496 |
if (ext_depth(inode) != depth && |
| 497 |
!ext4_extent_block_csum_verify(inode, eh)) { |
| 498 |
error_msg = "extent tree corrupted"; |
| 499 |
err = -EFSBADCRC; |
| 500 |
goto corrupted; |
| 501 |
} |
| 502 |
return 0; |
| 503 |
|
| 504 |
corrupted: |
| 505 |
ext4_error_inode_err(inode, function, line, 0, -err, |
| 506 |
"pblk %llu bad header/extent: %s - magic %x, " |
| 507 |
"entries %u, max %u(%u), depth %u(%u)", |
| 508 |
(unsigned long long) pblk, error_msg, |
| 509 |
le16_to_cpu(eh->eh_magic), |
| 510 |
le16_to_cpu(eh->eh_entries), |
| 511 |
le16_to_cpu(eh->eh_max), |
| 512 |
max, le16_to_cpu(eh->eh_depth), depth); |
| 513 |
return err; |
| 514 |
} |
| 515 |
|
| 516 |
#define ext4_ext_check(inode, eh, depth, pblk) \ |
| 517 |
__ext4_ext_check(__func__, __LINE__, (inode), (eh), (depth), (pblk), 0) |
| 518 |
|
| 519 |
int ext4_ext_check_inode(struct inode *inode) |
| 520 |
{ |
| 521 |
return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode), 0); |
| 522 |
} |
| 523 |
|
| 524 |
static void ext4_cache_extents(struct inode *inode, |
| 525 |
struct ext4_extent_header *eh) |
| 526 |
{ |
| 527 |
struct ext4_extent *ex = EXT_FIRST_EXTENT(eh); |
| 528 |
ext4_lblk_t prev = 0; |
| 529 |
int i; |
| 530 |
|
| 531 |
KUNIT_STATIC_STUB_REDIRECT(ext4_cache_extents, inode, eh); |
| 532 |
|
| 533 |
for (i = le16_to_cpu(eh->eh_entries); i > 0; i--, ex++) { |
| 534 |
unsigned int status = EXTENT_STATUS_WRITTEN; |
| 535 |
ext4_lblk_t lblk = le32_to_cpu(ex->ee_block); |
| 536 |
int len = ext4_ext_get_actual_len(ex); |
| 537 |
|
| 538 |
if (prev && (prev != lblk)) |
| 539 |
ext4_es_cache_extent(inode, prev, lblk - prev, ~0, |
| 540 |
EXTENT_STATUS_HOLE); |
| 541 |
|
| 542 |
if (ext4_ext_is_unwritten(ex)) |
| 543 |
status = EXTENT_STATUS_UNWRITTEN; |
| 544 |
ext4_es_cache_extent(inode, lblk, len, |
| 545 |
ext4_ext_pblock(ex), status); |
| 546 |
prev = lblk + len; |
| 547 |
} |
| 548 |
} |
| 549 |
|
| 550 |
static struct buffer_head * |
| 551 |
__read_extent_tree_block(const char *function, unsigned int line, |
| 552 |
struct inode *inode, struct ext4_extent_idx *idx, |
| 553 |
int depth, int flags) |
| 554 |
{ |
| 555 |
struct buffer_head *bh; |
| 556 |
int err; |
| 557 |
gfp_t gfp_flags = __GFP_MOVABLE | GFP_NOFS; |
| 558 |
ext4_fsblk_t pblk; |
| 559 |
|
| 560 |
if (flags & EXT4_EX_NOFAIL) |
| 561 |
gfp_flags |= __GFP_NOFAIL; |
| 562 |
|
| 563 |
pblk = ext4_idx_pblock(idx); |
| 564 |
bh = sb_getblk_gfp(inode->i_sb, pblk, gfp_flags); |
| 565 |
if (unlikely(!bh)) |
| 566 |
return ERR_PTR(-ENOMEM); |
| 567 |
|
| 568 |
if (!bh_uptodate_or_lock(bh)) { |
| 569 |
trace_ext4_ext_load_extent(inode, pblk, _RET_IP_); |
| 570 |
err = ext4_read_bh(bh, 0, NULL, false); |
| 571 |
if (err < 0) |
| 572 |
goto errout; |
| 573 |
} |
| 574 |
if (buffer_verified(bh) && !(flags & EXT4_EX_FORCE_CACHE)) |
| 575 |
return bh; |
| 576 |
err = __ext4_ext_check(function, line, inode, ext_block_hdr(bh), |
| 577 |
depth, pblk, le32_to_cpu(idx->ei_block)); |
| 578 |
if (err) |
| 579 |
goto errout; |
| 580 |
set_buffer_verified(bh); |
| 581 |
/* |
| 582 |
* If this is a leaf block, cache all of its entries |
| 583 |
*/ |
| 584 |
if (!(flags & EXT4_EX_NOCACHE) && depth == 0) { |
| 585 |
struct ext4_extent_header *eh = ext_block_hdr(bh); |
| 586 |
ext4_cache_extents(inode, eh); |
| 587 |
} |
| 588 |
return bh; |
| 589 |
errout: |
| 590 |
put_bh(bh); |
| 591 |
return ERR_PTR(err); |
| 592 |
|
| 593 |
} |
| 594 |
|
| 595 |
#define read_extent_tree_block(inode, idx, depth, flags) \ |
| 596 |
__read_extent_tree_block(__func__, __LINE__, (inode), (idx), \ |
| 597 |
(depth), (flags)) |
| 598 |
|
| 599 |
/* |
| 600 |
* This function is called to cache a file's extent information in the |
| 601 |
* extent status tree |
| 602 |
*/ |
| 603 |
int ext4_ext_precache(struct inode *inode) |
| 604 |
{ |
| 605 |
struct ext4_inode_info *ei = EXT4_I(inode); |
| 606 |
struct ext4_ext_path *path = NULL; |
| 607 |
struct buffer_head *bh; |
| 608 |
int i = 0, depth, ret = 0; |
| 609 |
|
| 610 |
if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
| 611 |
return 0; /* not an extent-mapped inode */ |
| 612 |
|
| 613 |
ext4_check_map_extents_env(inode); |
| 614 |
|
| 615 |
down_read(&ei->i_data_sem); |
| 616 |
depth = ext_depth(inode); |
| 617 |
|
| 618 |
/* Don't cache anything if there are no external extent blocks */ |
| 619 |
if (!depth) { |
| 620 |
up_read(&ei->i_data_sem); |
| 621 |
return ret; |
| 622 |
} |
| 623 |
|
| 624 |
path = kzalloc_objs(struct ext4_ext_path, depth + 1, GFP_NOFS); |
| 625 |
if (path == NULL) { |
| 626 |
up_read(&ei->i_data_sem); |
| 627 |
return -ENOMEM; |
| 628 |
} |
| 629 |
|
| 630 |
path[0].p_hdr = ext_inode_hdr(inode); |
| 631 |
ret = ext4_ext_check(inode, path[0].p_hdr, depth, 0); |
| 632 |
if (ret) |
| 633 |
goto out; |
| 634 |
path[0].p_idx = EXT_FIRST_INDEX(path[0].p_hdr); |
| 635 |
while (i >= 0) { |
| 636 |
/* |
| 637 |
* If this is a leaf block or we've reached the end of |
| 638 |
* the index block, go up |
| 639 |
*/ |
| 640 |
if ((i == depth) || |
| 641 |
path[i].p_idx > EXT_LAST_INDEX(path[i].p_hdr)) { |
| 642 |
ext4_ext_path_brelse(path + i); |
| 643 |
i--; |
| 644 |
continue; |
| 645 |
} |
| 646 |
bh = read_extent_tree_block(inode, path[i].p_idx++, |
| 647 |
depth - i - 1, |
| 648 |
EXT4_EX_FORCE_CACHE); |
| 649 |
if (IS_ERR(bh)) { |
| 650 |
ret = PTR_ERR(bh); |
| 651 |
break; |
| 652 |
} |
| 653 |
i++; |
| 654 |
path[i].p_bh = bh; |
| 655 |
path[i].p_hdr = ext_block_hdr(bh); |
| 656 |
path[i].p_idx = EXT_FIRST_INDEX(path[i].p_hdr); |
| 657 |
} |
| 658 |
ext4_set_inode_state(inode, EXT4_STATE_EXT_PRECACHED); |
| 659 |
out: |
| 660 |
up_read(&ei->i_data_sem); |
| 661 |
ext4_free_ext_path(path); |
| 662 |
return ret; |
| 663 |
} |
| 664 |
|
| 665 |
#ifdef EXT_DEBUG |
| 666 |
static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path) |
| 667 |
{ |
| 668 |
int k, l = path->p_depth; |
| 669 |
|
| 670 |
ext_debug(inode, "path:"); |
| 671 |
for (k = 0; k <= l; k++, path++) { |
| 672 |
if (path->p_idx) { |
| 673 |
ext_debug(inode, " %d->%llu", |
| 674 |
le32_to_cpu(path->p_idx->ei_block), |
| 675 |
ext4_idx_pblock(path->p_idx)); |
| 676 |
} else if (path->p_ext) { |
| 677 |
ext_debug(inode, " %d:[%d]%d:%llu ", |
| 678 |
le32_to_cpu(path->p_ext->ee_block), |
| 679 |
ext4_ext_is_unwritten(path->p_ext), |
| 680 |
ext4_ext_get_actual_len(path->p_ext), |
| 681 |
ext4_ext_pblock(path->p_ext)); |
| 682 |
} else |
| 683 |
ext_debug(inode, " []"); |
| 684 |
} |
| 685 |
ext_debug(inode, "\n"); |
| 686 |
} |
| 687 |
|
| 688 |
static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path) |
| 689 |
{ |
| 690 |
int depth = ext_depth(inode); |
| 691 |
struct ext4_extent_header *eh; |
| 692 |
struct ext4_extent *ex; |
| 693 |
int i; |
| 694 |
|
| 695 |
if (IS_ERR_OR_NULL(path)) |
| 696 |
return; |
| 697 |
|
| 698 |
eh = path[depth].p_hdr; |
| 699 |
ex = EXT_FIRST_EXTENT(eh); |
| 700 |
|
| 701 |
ext_debug(inode, "Displaying leaf extents\n"); |
| 702 |
|
| 703 |
for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) { |
| 704 |
ext_debug(inode, "%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block), |
| 705 |
ext4_ext_is_unwritten(ex), |
| 706 |
ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex)); |
| 707 |
} |
| 708 |
ext_debug(inode, "\n"); |
| 709 |
} |
| 710 |
|
| 711 |
static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path, |
| 712 |
ext4_fsblk_t newblock, int level) |
| 713 |
{ |
| 714 |
int depth = ext_depth(inode); |
| 715 |
struct ext4_extent *ex; |
| 716 |
|
| 717 |
if (depth != level) { |
| 718 |
struct ext4_extent_idx *idx; |
| 719 |
idx = path[level].p_idx; |
| 720 |
while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) { |
| 721 |
ext_debug(inode, "%d: move %d:%llu in new index %llu\n", |
| 722 |
level, le32_to_cpu(idx->ei_block), |
| 723 |
ext4_idx_pblock(idx), newblock); |
| 724 |
idx++; |
| 725 |
} |
| 726 |
|
| 727 |
return; |
| 728 |
} |
| 729 |
|
| 730 |
ex = path[depth].p_ext; |
| 731 |
while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) { |
| 732 |
ext_debug(inode, "move %d:%llu:[%d]%d in new leaf %llu\n", |
| 733 |
le32_to_cpu(ex->ee_block), |
| 734 |
ext4_ext_pblock(ex), |
| 735 |
ext4_ext_is_unwritten(ex), |
| 736 |
ext4_ext_get_actual_len(ex), |
| 737 |
newblock); |
| 738 |
ex++; |
| 739 |
} |
| 740 |
} |
| 741 |
|
| 742 |
#else |
| 743 |
#define ext4_ext_show_path(inode, path) |
| 744 |
#define ext4_ext_show_leaf(inode, path) |
| 745 |
#define ext4_ext_show_move(inode, path, newblock, level) |
| 746 |
#endif |
| 747 |
|
| 748 |
/* |
| 749 |
* ext4_ext_binsearch_idx: |
| 750 |
* binary search for the closest index of the given block |
| 751 |
* the header must be checked before calling this |
| 752 |
*/ |
| 753 |
static void |
| 754 |
ext4_ext_binsearch_idx(struct inode *inode, |
| 755 |
struct ext4_ext_path *path, ext4_lblk_t block) |
| 756 |
{ |
| 757 |
struct ext4_extent_header *eh = path->p_hdr; |
| 758 |
struct ext4_extent_idx *r, *l, *m; |
| 759 |
|
| 760 |
|
| 761 |
ext_debug(inode, "binsearch for %u(idx): ", block); |
| 762 |
|
| 763 |
l = EXT_FIRST_INDEX(eh) + 1; |
| 764 |
r = EXT_LAST_INDEX(eh); |
| 765 |
while (l <= r) { |
| 766 |
m = l + (r - l) / 2; |
| 767 |
ext_debug(inode, "%p(%u):%p(%u):%p(%u) ", l, |
| 768 |
le32_to_cpu(l->ei_block), m, le32_to_cpu(m->ei_block), |
| 769 |
r, le32_to_cpu(r->ei_block)); |
| 770 |
|
| 771 |
if (block < le32_to_cpu(m->ei_block)) |
| 772 |
r = m - 1; |
| 773 |
else |
| 774 |
l = m + 1; |
| 775 |
} |
| 776 |
|
| 777 |
path->p_idx = l - 1; |
| 778 |
ext_debug(inode, " -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block), |
| 779 |
ext4_idx_pblock(path->p_idx)); |
| 780 |
|
| 781 |
#ifdef CHECK_BINSEARCH |
| 782 |
{ |
| 783 |
struct ext4_extent_idx *chix, *ix; |
| 784 |
int k; |
| 785 |
|
| 786 |
chix = ix = EXT_FIRST_INDEX(eh); |
| 787 |
for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) { |
| 788 |
if (k != 0 && le32_to_cpu(ix->ei_block) <= |
| 789 |
le32_to_cpu(ix[-1].ei_block)) { |
| 790 |
printk(KERN_DEBUG "k=%d, ix=0x%p, " |
| 791 |
"first=0x%p\n", k, |
| 792 |
ix, EXT_FIRST_INDEX(eh)); |
| 793 |
printk(KERN_DEBUG "%u <= %u\n", |
| 794 |
le32_to_cpu(ix->ei_block), |
| 795 |
le32_to_cpu(ix[-1].ei_block)); |
| 796 |
} |
| 797 |
BUG_ON(k && le32_to_cpu(ix->ei_block) |
| 798 |
<= le32_to_cpu(ix[-1].ei_block)); |
| 799 |
if (block < le32_to_cpu(ix->ei_block)) |
| 800 |
break; |
| 801 |
chix = ix; |
| 802 |
} |
| 803 |
BUG_ON(chix != path->p_idx); |
| 804 |
} |
| 805 |
#endif |
| 806 |
|
| 807 |
} |
| 808 |
|
| 809 |
/* |
| 810 |
* ext4_ext_binsearch: |
| 811 |
* binary search for closest extent of the given block |
| 812 |
* the header must be checked before calling this |
| 813 |
*/ |
| 814 |
static void |
| 815 |
ext4_ext_binsearch(struct inode *inode, |
| 816 |
struct ext4_ext_path *path, ext4_lblk_t block) |
| 817 |
{ |
| 818 |
struct ext4_extent_header *eh = path->p_hdr; |
| 819 |
struct ext4_extent *r, *l, *m; |
| 820 |
|
| 821 |
if (eh->eh_entries == 0) { |
| 822 |
/* |
| 823 |
* this leaf is empty: |
| 824 |
* we get such a leaf in split/add case |
| 825 |
*/ |
| 826 |
return; |
| 827 |
} |
| 828 |
|
| 829 |
ext_debug(inode, "binsearch for %u: ", block); |
| 830 |
|
| 831 |
l = EXT_FIRST_EXTENT(eh) + 1; |
| 832 |
r = EXT_LAST_EXTENT(eh); |
| 833 |
|
| 834 |
while (l <= r) { |
| 835 |
m = l + (r - l) / 2; |
| 836 |
ext_debug(inode, "%p(%u):%p(%u):%p(%u) ", l, |
| 837 |
le32_to_cpu(l->ee_block), m, le32_to_cpu(m->ee_block), |
| 838 |
r, le32_to_cpu(r->ee_block)); |
| 839 |
|
| 840 |
if (block < le32_to_cpu(m->ee_block)) |
| 841 |
r = m - 1; |
| 842 |
else |
| 843 |
l = m + 1; |
| 844 |
} |
| 845 |
|
| 846 |
path->p_ext = l - 1; |
| 847 |
ext_debug(inode, " -> %d:%llu:[%d]%d ", |
| 848 |
le32_to_cpu(path->p_ext->ee_block), |
| 849 |
ext4_ext_pblock(path->p_ext), |
| 850 |
ext4_ext_is_unwritten(path->p_ext), |
| 851 |
ext4_ext_get_actual_len(path->p_ext)); |
| 852 |
|
| 853 |
#ifdef CHECK_BINSEARCH |
| 854 |
{ |
| 855 |
struct ext4_extent *chex, *ex; |
| 856 |
int k; |
| 857 |
|
| 858 |
chex = ex = EXT_FIRST_EXTENT(eh); |
| 859 |
for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) { |
| 860 |
BUG_ON(k && le32_to_cpu(ex->ee_block) |
| 861 |
<= le32_to_cpu(ex[-1].ee_block)); |
| 862 |
if (block < le32_to_cpu(ex->ee_block)) |
| 863 |
break; |
| 864 |
chex = ex; |
| 865 |
} |
| 866 |
BUG_ON(chex != path->p_ext); |
| 867 |
} |
| 868 |
#endif |
| 869 |
|
| 870 |
} |
| 871 |
|
| 872 |
void ext4_ext_tree_init(handle_t *handle, struct inode *inode) |
| 873 |
{ |
| 874 |
struct ext4_extent_header *eh; |
| 875 |
|
| 876 |
eh = ext_inode_hdr(inode); |
| 877 |
eh->eh_depth = 0; |
| 878 |
eh->eh_entries = 0; |
| 879 |
eh->eh_magic = EXT4_EXT_MAGIC; |
| 880 |
eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0)); |
| 881 |
eh->eh_generation = 0; |
| 882 |
ext4_mark_inode_dirty(handle, inode); |
| 883 |
} |
| 884 |
|
| 885 |
struct ext4_ext_path * |
| 886 |
ext4_find_extent(struct inode *inode, ext4_lblk_t block, |
| 887 |
struct ext4_ext_path *path, int flags) |
| 888 |
{ |
| 889 |
struct ext4_extent_header *eh; |
| 890 |
struct buffer_head *bh; |
| 891 |
short int depth, i, ppos = 0; |
| 892 |
int ret; |
| 893 |
gfp_t gfp_flags = GFP_NOFS; |
| 894 |
|
| 895 |
KUNIT_STATIC_STUB_REDIRECT(ext4_find_extent, inode, block, path, flags); |
| 896 |
|
| 897 |
if (flags & EXT4_EX_NOFAIL) |
| 898 |
gfp_flags |= __GFP_NOFAIL; |
| 899 |
|
| 900 |
eh = ext_inode_hdr(inode); |
| 901 |
depth = ext_depth(inode); |
| 902 |
if (depth < 0 || depth > EXT4_MAX_EXTENT_DEPTH) { |
| 903 |
EXT4_ERROR_INODE(inode, "inode has invalid extent depth: %d", |
| 904 |
depth); |
| 905 |
ret = -EFSCORRUPTED; |
| 906 |
goto err; |
| 907 |
} |
| 908 |
|
| 909 |
if (path) { |
| 910 |
ext4_ext_drop_refs(path); |
| 911 |
if (depth > path[0].p_maxdepth) { |
| 912 |
kfree(path); |
| 913 |
path = NULL; |
| 914 |
} |
| 915 |
} |
| 916 |
if (!path) { |
| 917 |
/* account possible depth increase */ |
| 918 |
path = kzalloc_objs(struct ext4_ext_path, depth + 2, gfp_flags); |
| 919 |
if (unlikely(!path)) |
| 920 |
return ERR_PTR(-ENOMEM); |
| 921 |
path[0].p_maxdepth = depth + 1; |
| 922 |
} |
| 923 |
path[0].p_hdr = eh; |
| 924 |
path[0].p_bh = NULL; |
| 925 |
|
| 926 |
i = depth; |
| 927 |
if (!(flags & EXT4_EX_NOCACHE) && depth == 0) |
| 928 |
ext4_cache_extents(inode, eh); |
| 929 |
/* walk through the tree */ |
| 930 |
while (i) { |
| 931 |
ext_debug(inode, "depth %d: num %d, max %d\n", |
| 932 |
ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max)); |
| 933 |
|
| 934 |
ext4_ext_binsearch_idx(inode, path + ppos, block); |
| 935 |
path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx); |
| 936 |
path[ppos].p_depth = i; |
| 937 |
path[ppos].p_ext = NULL; |
| 938 |
|
| 939 |
bh = read_extent_tree_block(inode, path[ppos].p_idx, --i, flags); |
| 940 |
if (IS_ERR(bh)) { |
| 941 |
ret = PTR_ERR(bh); |
| 942 |
goto err; |
| 943 |
} |
| 944 |
|
| 945 |
eh = ext_block_hdr(bh); |
| 946 |
ppos++; |
| 947 |
path[ppos].p_bh = bh; |
| 948 |
path[ppos].p_hdr = eh; |
| 949 |
} |
| 950 |
|
| 951 |
path[ppos].p_depth = i; |
| 952 |
path[ppos].p_ext = NULL; |
| 953 |
path[ppos].p_idx = NULL; |
| 954 |
|
| 955 |
/* find extent */ |
| 956 |
ext4_ext_binsearch(inode, path + ppos, block); |
| 957 |
/* if not an empty leaf */ |
| 958 |
if (path[ppos].p_ext) |
| 959 |
path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext); |
| 960 |
|
| 961 |
ext4_ext_show_path(inode, path); |
| 962 |
|
| 963 |
return path; |
| 964 |
|
| 965 |
err: |
| 966 |
ext4_free_ext_path(path); |
| 967 |
return ERR_PTR(ret); |
| 968 |
} |
| 969 |
|
| 970 |
/* |
| 971 |
* ext4_ext_insert_index: |
| 972 |
* insert new index [@logical;@ptr] into the block at @curp; |
| 973 |
* check where to insert: before @curp or after @curp |
| 974 |
*/ |
| 975 |
static int ext4_ext_insert_index(handle_t *handle, struct inode *inode, |
| 976 |
struct ext4_ext_path *curp, |
| 977 |
int logical, ext4_fsblk_t ptr) |
| 978 |
{ |
| 979 |
struct ext4_extent_idx *ix; |
| 980 |
int len, err; |
| 981 |
|
| 982 |
err = ext4_ext_get_access(handle, inode, curp); |
| 983 |
if (err) |
| 984 |
return err; |
| 985 |
|
| 986 |
if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) { |
| 987 |
EXT4_ERROR_INODE(inode, |
| 988 |
"logical %d == ei_block %d!", |
| 989 |
logical, le32_to_cpu(curp->p_idx->ei_block)); |
| 990 |
return -EFSCORRUPTED; |
| 991 |
} |
| 992 |
|
| 993 |
if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries) |
| 994 |
>= le16_to_cpu(curp->p_hdr->eh_max))) { |
| 995 |
EXT4_ERROR_INODE(inode, |
| 996 |
"eh_entries %d >= eh_max %d!", |
| 997 |
le16_to_cpu(curp->p_hdr->eh_entries), |
| 998 |
le16_to_cpu(curp->p_hdr->eh_max)); |
| 999 |
return -EFSCORRUPTED; |
| 1000 |
} |
| 1001 |
|
| 1002 |
if (logical > le32_to_cpu(curp->p_idx->ei_block)) { |
| 1003 |
/* insert after */ |
| 1004 |
ext_debug(inode, "insert new index %d after: %llu\n", |
| 1005 |
logical, ptr); |
| 1006 |
ix = curp->p_idx + 1; |
| 1007 |
} else { |
| 1008 |
/* insert before */ |
| 1009 |
ext_debug(inode, "insert new index %d before: %llu\n", |
| 1010 |
logical, ptr); |
| 1011 |
ix = curp->p_idx; |
| 1012 |
} |
| 1013 |
|
| 1014 |
if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) { |
| 1015 |
EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!"); |
| 1016 |
return -EFSCORRUPTED; |
| 1017 |
} |
| 1018 |
|
| 1019 |
len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1; |
| 1020 |
BUG_ON(len < 0); |
| 1021 |
if (len > 0) { |
| 1022 |
ext_debug(inode, "insert new index %d: " |
| 1023 |
"move %d indices from 0x%p to 0x%p\n", |
| 1024 |
logical, len, ix, ix + 1); |
| 1025 |
memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx)); |
| 1026 |
} |
| 1027 |
|
| 1028 |
ix->ei_block = cpu_to_le32(logical); |
| 1029 |
ext4_idx_store_pblock(ix, ptr); |
| 1030 |
le16_add_cpu(&curp->p_hdr->eh_entries, 1); |
| 1031 |
|
| 1032 |
if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) { |
| 1033 |
EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!"); |
| 1034 |
return -EFSCORRUPTED; |
| 1035 |
} |
| 1036 |
|
| 1037 |
err = ext4_ext_dirty(handle, inode, curp); |
| 1038 |
ext4_std_error(inode->i_sb, err); |
| 1039 |
|
| 1040 |
return err; |
| 1041 |
} |
| 1042 |
|
| 1043 |
/* |
| 1044 |
* ext4_ext_split: |
| 1045 |
* inserts new subtree into the path, using free index entry |
| 1046 |
* at depth @at: |
| 1047 |
* - allocates all needed blocks (new leaf and all intermediate index blocks) |
| 1048 |
* - makes decision where to split |
| 1049 |
* - moves remaining extents and index entries (right to the split point) |
| 1050 |
* into the newly allocated blocks |
| 1051 |
* - initializes subtree |
| 1052 |
*/ |
| 1053 |
static int ext4_ext_split(handle_t *handle, struct inode *inode, |
| 1054 |
unsigned int flags, |
| 1055 |
struct ext4_ext_path *path, |
| 1056 |
struct ext4_extent *newext, int at) |
| 1057 |
{ |
| 1058 |
struct buffer_head *bh = NULL; |
| 1059 |
int depth = ext_depth(inode); |
| 1060 |
struct ext4_extent_header *neh; |
| 1061 |
struct ext4_extent_idx *fidx; |
| 1062 |
int i = at, k, m, a; |
| 1063 |
ext4_fsblk_t newblock, oldblock; |
| 1064 |
__le32 border; |
| 1065 |
ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */ |
| 1066 |
gfp_t gfp_flags = GFP_NOFS; |
| 1067 |
int err = 0; |
| 1068 |
size_t ext_size = 0; |
| 1069 |
|
| 1070 |
if (flags & EXT4_EX_NOFAIL) |
| 1071 |
gfp_flags |= __GFP_NOFAIL; |
| 1072 |
|
| 1073 |
/* make decision: where to split? */ |
| 1074 |
/* FIXME: now decision is simplest: at current extent */ |
| 1075 |
|
| 1076 |
/* if current leaf will be split, then we should use |
| 1077 |
* border from split point */ |
| 1078 |
if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) { |
| 1079 |
EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!"); |
| 1080 |
return -EFSCORRUPTED; |
| 1081 |
} |
| 1082 |
if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) { |
| 1083 |
border = path[depth].p_ext[1].ee_block; |
| 1084 |
ext_debug(inode, "leaf will be split." |
| 1085 |
" next leaf starts at %d\n", |
| 1086 |
le32_to_cpu(border)); |
| 1087 |
} else { |
| 1088 |
border = newext->ee_block; |
| 1089 |
ext_debug(inode, "leaf will be added." |
| 1090 |
" next leaf starts at %d\n", |
| 1091 |
le32_to_cpu(border)); |
| 1092 |
} |
| 1093 |
|
| 1094 |
/* |
| 1095 |
* If error occurs, then we break processing |
| 1096 |
* and mark filesystem read-only. index won't |
| 1097 |
* be inserted and tree will be in consistent |
| 1098 |
* state. Next mount will repair buffers too. |
| 1099 |
*/ |
| 1100 |
|
| 1101 |
/* |
| 1102 |
* Get array to track all allocated blocks. |
| 1103 |
* We need this to handle errors and free blocks |
| 1104 |
* upon them. |
| 1105 |
*/ |
| 1106 |
ablocks = kzalloc_objs(ext4_fsblk_t, depth, gfp_flags); |
| 1107 |
if (!ablocks) |
| 1108 |
return -ENOMEM; |
| 1109 |
|
| 1110 |
/* allocate all needed blocks */ |
| 1111 |
ext_debug(inode, "allocate %d blocks for indexes/leaf\n", depth - at); |
| 1112 |
for (a = 0; a < depth - at; a++) { |
| 1113 |
newblock = ext4_ext_new_meta_block(handle, inode, path, |
| 1114 |
newext, &err, flags); |
| 1115 |
if (newblock == 0) |
| 1116 |
goto cleanup; |
| 1117 |
ablocks[a] = newblock; |
| 1118 |
} |
| 1119 |
|
| 1120 |
/* initialize new leaf */ |
| 1121 |
newblock = ablocks[--a]; |
| 1122 |
if (unlikely(newblock == 0)) { |
| 1123 |
EXT4_ERROR_INODE(inode, "newblock == 0!"); |
| 1124 |
err = -EFSCORRUPTED; |
| 1125 |
goto cleanup; |
| 1126 |
} |
| 1127 |
bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS); |
| 1128 |
if (unlikely(!bh)) { |
| 1129 |
err = -ENOMEM; |
| 1130 |
goto cleanup; |
| 1131 |
} |
| 1132 |
lock_buffer(bh); |
| 1133 |
|
| 1134 |
err = ext4_journal_get_create_access(handle, inode->i_sb, bh, |
| 1135 |
EXT4_JTR_NONE); |
| 1136 |
if (err) |
| 1137 |
goto cleanup; |
| 1138 |
|
| 1139 |
neh = ext_block_hdr(bh); |
| 1140 |
neh->eh_entries = 0; |
| 1141 |
neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0)); |
| 1142 |
neh->eh_magic = EXT4_EXT_MAGIC; |
| 1143 |
neh->eh_depth = 0; |
| 1144 |
neh->eh_generation = 0; |
| 1145 |
|
| 1146 |
/* move remainder of path[depth] to the new leaf */ |
| 1147 |
if (unlikely(path[depth].p_hdr->eh_entries != |
| 1148 |
path[depth].p_hdr->eh_max)) { |
| 1149 |
EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!", |
| 1150 |
path[depth].p_hdr->eh_entries, |
| 1151 |
path[depth].p_hdr->eh_max); |
| 1152 |
err = -EFSCORRUPTED; |
| 1153 |
goto cleanup; |
| 1154 |
} |
| 1155 |
/* start copy from next extent */ |
| 1156 |
m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++; |
| 1157 |
ext4_ext_show_move(inode, path, newblock, depth); |
| 1158 |
if (m) { |
| 1159 |
struct ext4_extent *ex; |
| 1160 |
ex = EXT_FIRST_EXTENT(neh); |
| 1161 |
memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m); |
| 1162 |
le16_add_cpu(&neh->eh_entries, m); |
| 1163 |
} |
| 1164 |
|
| 1165 |
/* zero out unused area in the extent block */ |
| 1166 |
ext_size = sizeof(struct ext4_extent_header) + |
| 1167 |
sizeof(struct ext4_extent) * le16_to_cpu(neh->eh_entries); |
| 1168 |
memset(bh->b_data + ext_size, 0, inode->i_sb->s_blocksize - ext_size); |
| 1169 |
ext4_extent_block_csum_set(inode, neh); |
| 1170 |
set_buffer_uptodate(bh); |
| 1171 |
unlock_buffer(bh); |
| 1172 |
|
| 1173 |
err = ext4_handle_dirty_metadata(handle, inode, bh); |
| 1174 |
if (err) |
| 1175 |
goto cleanup; |
| 1176 |
brelse(bh); |
| 1177 |
bh = NULL; |
| 1178 |
|
| 1179 |
/* correct old leaf */ |
| 1180 |
if (m) { |
| 1181 |
err = ext4_ext_get_access(handle, inode, path + depth); |
| 1182 |
if (err) |
| 1183 |
goto cleanup; |
| 1184 |
le16_add_cpu(&path[depth].p_hdr->eh_entries, -m); |
| 1185 |
err = ext4_ext_dirty(handle, inode, path + depth); |
| 1186 |
if (err) |
| 1187 |
goto cleanup; |
| 1188 |
|
| 1189 |
} |
| 1190 |
|
| 1191 |
/* create intermediate indexes */ |
| 1192 |
k = depth - at - 1; |
| 1193 |
if (unlikely(k < 0)) { |
| 1194 |
EXT4_ERROR_INODE(inode, "k %d < 0!", k); |
| 1195 |
err = -EFSCORRUPTED; |
| 1196 |
goto cleanup; |
| 1197 |
} |
| 1198 |
if (k) |
| 1199 |
ext_debug(inode, "create %d intermediate indices\n", k); |
| 1200 |
/* insert new index into current index block */ |
| 1201 |
/* current depth stored in i var */ |
| 1202 |
i = depth - 1; |
| 1203 |
while (k--) { |
| 1204 |
oldblock = newblock; |
| 1205 |
newblock = ablocks[--a]; |
| 1206 |
bh = sb_getblk(inode->i_sb, newblock); |
| 1207 |
if (unlikely(!bh)) { |
| 1208 |
err = -ENOMEM; |
| 1209 |
goto cleanup; |
| 1210 |
} |
| 1211 |
lock_buffer(bh); |
| 1212 |
|
| 1213 |
err = ext4_journal_get_create_access(handle, inode->i_sb, bh, |
| 1214 |
EXT4_JTR_NONE); |
| 1215 |
if (err) |
| 1216 |
goto cleanup; |
| 1217 |
|
| 1218 |
neh = ext_block_hdr(bh); |
| 1219 |
neh->eh_entries = cpu_to_le16(1); |
| 1220 |
neh->eh_magic = EXT4_EXT_MAGIC; |
| 1221 |
neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0)); |
| 1222 |
neh->eh_depth = cpu_to_le16(depth - i); |
| 1223 |
neh->eh_generation = 0; |
| 1224 |
fidx = EXT_FIRST_INDEX(neh); |
| 1225 |
fidx->ei_block = border; |
| 1226 |
ext4_idx_store_pblock(fidx, oldblock); |
| 1227 |
|
| 1228 |
ext_debug(inode, "int.index at %d (block %llu): %u -> %llu\n", |
| 1229 |
i, newblock, le32_to_cpu(border), oldblock); |
| 1230 |
|
| 1231 |
/* move remainder of path[i] to the new index block */ |
| 1232 |
if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) != |
| 1233 |
EXT_LAST_INDEX(path[i].p_hdr))) { |
| 1234 |
EXT4_ERROR_INODE(inode, |
| 1235 |
"EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!", |
| 1236 |
le32_to_cpu(path[i].p_ext->ee_block)); |
| 1237 |
err = -EFSCORRUPTED; |
| 1238 |
goto cleanup; |
| 1239 |
} |
| 1240 |
/* start copy indexes */ |
| 1241 |
m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++; |
| 1242 |
ext_debug(inode, "cur 0x%p, last 0x%p\n", path[i].p_idx, |
| 1243 |
EXT_MAX_INDEX(path[i].p_hdr)); |
| 1244 |
ext4_ext_show_move(inode, path, newblock, i); |
| 1245 |
if (m) { |
| 1246 |
memmove(++fidx, path[i].p_idx, |
| 1247 |
sizeof(struct ext4_extent_idx) * m); |
| 1248 |
le16_add_cpu(&neh->eh_entries, m); |
| 1249 |
} |
| 1250 |
/* zero out unused area in the extent block */ |
| 1251 |
ext_size = sizeof(struct ext4_extent_header) + |
| 1252 |
(sizeof(struct ext4_extent) * le16_to_cpu(neh->eh_entries)); |
| 1253 |
memset(bh->b_data + ext_size, 0, |
| 1254 |
inode->i_sb->s_blocksize - ext_size); |
| 1255 |
ext4_extent_block_csum_set(inode, neh); |
| 1256 |
set_buffer_uptodate(bh); |
| 1257 |
unlock_buffer(bh); |
| 1258 |
|
| 1259 |
err = ext4_handle_dirty_metadata(handle, inode, bh); |
| 1260 |
if (err) |
| 1261 |
goto cleanup; |
| 1262 |
brelse(bh); |
| 1263 |
bh = NULL; |
| 1264 |
|
| 1265 |
/* correct old index */ |
| 1266 |
if (m) { |
| 1267 |
err = ext4_ext_get_access(handle, inode, path + i); |
| 1268 |
if (err) |
| 1269 |
goto cleanup; |
| 1270 |
le16_add_cpu(&path[i].p_hdr->eh_entries, -m); |
| 1271 |
err = ext4_ext_dirty(handle, inode, path + i); |
| 1272 |
if (err) |
| 1273 |
goto cleanup; |
| 1274 |
} |
| 1275 |
|
| 1276 |
i--; |
| 1277 |
} |
| 1278 |
|
| 1279 |
/* insert new index */ |
| 1280 |
err = ext4_ext_insert_index(handle, inode, path + at, |
| 1281 |
le32_to_cpu(border), newblock); |
| 1282 |
|
| 1283 |
cleanup: |
| 1284 |
if (bh) { |
| 1285 |
if (buffer_locked(bh)) |
| 1286 |
unlock_buffer(bh); |
| 1287 |
brelse(bh); |
| 1288 |
} |
| 1289 |
|
| 1290 |
if (err) { |
| 1291 |
/* free all allocated blocks in error case */ |
| 1292 |
for (i = 0; i < depth; i++) { |
| 1293 |
if (!ablocks[i]) |
| 1294 |
continue; |
| 1295 |
ext4_free_blocks(handle, inode, NULL, ablocks[i], 1, |
| 1296 |
EXT4_FREE_BLOCKS_METADATA); |
| 1297 |
} |
| 1298 |
} |
| 1299 |
kfree(ablocks); |
| 1300 |
|
| 1301 |
return err; |
| 1302 |
} |
| 1303 |
|
| 1304 |
/* |
| 1305 |
* ext4_ext_grow_indepth: |
| 1306 |
* implements tree growing procedure: |
| 1307 |
* - allocates new block |
| 1308 |
* - moves top-level data (index block or leaf) into the new block |
| 1309 |
* - initializes new top-level, creating index that points to the |
| 1310 |
* just created block |
| 1311 |
*/ |
| 1312 |
static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode, |
| 1313 |
unsigned int flags) |
| 1314 |
{ |
| 1315 |
struct ext4_extent_header *neh; |
| 1316 |
struct buffer_head *bh; |
| 1317 |
ext4_fsblk_t newblock, goal = 0; |
| 1318 |
struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es; |
| 1319 |
int err = 0; |
| 1320 |
size_t ext_size = 0; |
| 1321 |
|
| 1322 |
/* Try to prepend new index to old one */ |
| 1323 |
if (ext_depth(inode)) |
| 1324 |
goal = ext4_idx_pblock(EXT_FIRST_INDEX(ext_inode_hdr(inode))); |
| 1325 |
if (goal > le32_to_cpu(es->s_first_data_block)) { |
| 1326 |
flags |= EXT4_MB_HINT_TRY_GOAL; |
| 1327 |
goal--; |
| 1328 |
} else |
| 1329 |
goal = ext4_inode_to_goal_block(inode); |
| 1330 |
newblock = ext4_new_meta_blocks(handle, inode, goal, flags, |
| 1331 |
NULL, &err); |
| 1332 |
if (newblock == 0) |
| 1333 |
return err; |
| 1334 |
|
| 1335 |
bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS); |
| 1336 |
if (unlikely(!bh)) |
| 1337 |
return -ENOMEM; |
| 1338 |
lock_buffer(bh); |
| 1339 |
|
| 1340 |
err = ext4_journal_get_create_access(handle, inode->i_sb, bh, |
| 1341 |
EXT4_JTR_NONE); |
| 1342 |
if (err) { |
| 1343 |
unlock_buffer(bh); |
| 1344 |
goto out; |
| 1345 |
} |
| 1346 |
|
| 1347 |
ext_size = sizeof(EXT4_I(inode)->i_data); |
| 1348 |
/* move top-level index/leaf into new block */ |
| 1349 |
memmove(bh->b_data, EXT4_I(inode)->i_data, ext_size); |
| 1350 |
/* zero out unused area in the extent block */ |
| 1351 |
memset(bh->b_data + ext_size, 0, inode->i_sb->s_blocksize - ext_size); |
| 1352 |
|
| 1353 |
/* set size of new block */ |
| 1354 |
neh = ext_block_hdr(bh); |
| 1355 |
/* old root could have indexes or leaves |
| 1356 |
* so calculate e_max right way */ |
| 1357 |
if (ext_depth(inode)) |
| 1358 |
neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0)); |
| 1359 |
else |
| 1360 |
neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0)); |
| 1361 |
neh->eh_magic = EXT4_EXT_MAGIC; |
| 1362 |
ext4_extent_block_csum_set(inode, neh); |
| 1363 |
set_buffer_uptodate(bh); |
| 1364 |
set_buffer_verified(bh); |
| 1365 |
unlock_buffer(bh); |
| 1366 |
|
| 1367 |
err = ext4_handle_dirty_metadata(handle, inode, bh); |
| 1368 |
if (err) |
| 1369 |
goto out; |
| 1370 |
|
| 1371 |
/* Update top-level index: num,max,pointer */ |
| 1372 |
neh = ext_inode_hdr(inode); |
| 1373 |
neh->eh_entries = cpu_to_le16(1); |
| 1374 |
ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock); |
| 1375 |
if (neh->eh_depth == 0) { |
| 1376 |
/* Root extent block becomes index block */ |
| 1377 |
neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0)); |
| 1378 |
EXT_FIRST_INDEX(neh)->ei_block = |
| 1379 |
EXT_FIRST_EXTENT(neh)->ee_block; |
| 1380 |
} |
| 1381 |
ext_debug(inode, "new root: num %d(%d), lblock %d, ptr %llu\n", |
| 1382 |
le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max), |
| 1383 |
le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block), |
| 1384 |
ext4_idx_pblock(EXT_FIRST_INDEX(neh))); |
| 1385 |
|
| 1386 |
le16_add_cpu(&neh->eh_depth, 1); |
| 1387 |
err = ext4_mark_inode_dirty(handle, inode); |
| 1388 |
out: |
| 1389 |
brelse(bh); |
| 1390 |
|
| 1391 |
return err; |
| 1392 |
} |
| 1393 |
|
| 1394 |
/* |
| 1395 |
* ext4_ext_create_new_leaf: |
| 1396 |
* finds empty index and adds new leaf. |
| 1397 |
* if no free index is found, then it requests in-depth growing. |
| 1398 |
*/ |
| 1399 |
static struct ext4_ext_path * |
| 1400 |
ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode, |
| 1401 |
unsigned int mb_flags, unsigned int gb_flags, |
| 1402 |
struct ext4_ext_path *path, |
| 1403 |
struct ext4_extent *newext) |
| 1404 |
{ |
| 1405 |
struct ext4_ext_path *curp; |
| 1406 |
int depth, i, err = 0; |
| 1407 |
ext4_lblk_t ee_block = le32_to_cpu(newext->ee_block); |
| 1408 |
|
| 1409 |
repeat: |
| 1410 |
i = depth = ext_depth(inode); |
| 1411 |
|
| 1412 |
/* walk up to the tree and look for free index entry */ |
| 1413 |
curp = path + depth; |
| 1414 |
while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) { |
| 1415 |
i--; |
| 1416 |
curp--; |
| 1417 |
} |
| 1418 |
|
| 1419 |
/* we use already allocated block for index block, |
| 1420 |
* so subsequent data blocks should be contiguous */ |
| 1421 |
if (EXT_HAS_FREE_INDEX(curp)) { |
| 1422 |
/* if we found index with free entry, then use that |
| 1423 |
* entry: create all needed subtree and add new leaf */ |
| 1424 |
err = ext4_ext_split(handle, inode, mb_flags, path, newext, i); |
| 1425 |
if (err) |
| 1426 |
goto errout; |
| 1427 |
|
| 1428 |
/* refill path */ |
| 1429 |
path = ext4_find_extent(inode, ee_block, path, gb_flags); |
| 1430 |
return path; |
| 1431 |
} |
| 1432 |
|
| 1433 |
/* tree is full, time to grow in depth */ |
| 1434 |
err = ext4_ext_grow_indepth(handle, inode, mb_flags); |
| 1435 |
if (err) |
| 1436 |
goto errout; |
| 1437 |
|
| 1438 |
/* refill path */ |
| 1439 |
path = ext4_find_extent(inode, ee_block, path, gb_flags); |
| 1440 |
if (IS_ERR(path)) |
| 1441 |
return path; |
| 1442 |
|
| 1443 |
/* |
| 1444 |
* only first (depth 0 -> 1) produces free space; |
| 1445 |
* in all other cases we have to split the grown tree |
| 1446 |
*/ |
| 1447 |
depth = ext_depth(inode); |
| 1448 |
if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) { |
| 1449 |
/* now we need to split */ |
| 1450 |
goto repeat; |
| 1451 |
} |
| 1452 |
|
| 1453 |
return path; |
| 1454 |
|
| 1455 |
errout: |
| 1456 |
ext4_free_ext_path(path); |
| 1457 |
return ERR_PTR(err); |
| 1458 |
} |
| 1459 |
|
| 1460 |
/* |
| 1461 |
* search the closest allocated block to the left for *logical |
| 1462 |
* and returns it at @logical + it's physical address at @phys |
| 1463 |
* if *logical is the smallest allocated block, the function |
| 1464 |
* returns 0 at @phys |
| 1465 |
* return value contains 0 (success) or error code |
| 1466 |
*/ |
| 1467 |
static int ext4_ext_search_left(struct inode *inode, |
| 1468 |
struct ext4_ext_path *path, |
| 1469 |
ext4_lblk_t *logical, ext4_fsblk_t *phys) |
| 1470 |
{ |
| 1471 |
struct ext4_extent_idx *ix; |
| 1472 |
struct ext4_extent *ex; |
| 1473 |
int depth, ee_len; |
| 1474 |
|
| 1475 |
if (unlikely(path == NULL)) { |
| 1476 |
EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical); |
| 1477 |
return -EFSCORRUPTED; |
| 1478 |
} |
| 1479 |
depth = path->p_depth; |
| 1480 |
*phys = 0; |
| 1481 |
|
| 1482 |
if (depth == 0 && path->p_ext == NULL) |
| 1483 |
return 0; |
| 1484 |
|
| 1485 |
/* usually extent in the path covers blocks smaller |
| 1486 |
* then *logical, but it can be that extent is the |
| 1487 |
* first one in the file */ |
| 1488 |
|
| 1489 |
ex = path[depth].p_ext; |
| 1490 |
ee_len = ext4_ext_get_actual_len(ex); |
| 1491 |
if (*logical < le32_to_cpu(ex->ee_block)) { |
| 1492 |
if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) { |
| 1493 |
EXT4_ERROR_INODE(inode, |
| 1494 |
"EXT_FIRST_EXTENT != ex *logical %d ee_block %d!", |
| 1495 |
*logical, le32_to_cpu(ex->ee_block)); |
| 1496 |
return -EFSCORRUPTED; |
| 1497 |
} |
| 1498 |
while (--depth >= 0) { |
| 1499 |
ix = path[depth].p_idx; |
| 1500 |
if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) { |
| 1501 |
EXT4_ERROR_INODE(inode, |
| 1502 |
"ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!", |
| 1503 |
ix != NULL ? le32_to_cpu(ix->ei_block) : 0, |
| 1504 |
le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block), |
| 1505 |
depth); |
| 1506 |
return -EFSCORRUPTED; |
| 1507 |
} |
| 1508 |
} |
| 1509 |
return 0; |
| 1510 |
} |
| 1511 |
|
| 1512 |
if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) { |
| 1513 |
EXT4_ERROR_INODE(inode, |
| 1514 |
"logical %d < ee_block %d + ee_len %d!", |
| 1515 |
*logical, le32_to_cpu(ex->ee_block), ee_len); |
| 1516 |
return -EFSCORRUPTED; |
| 1517 |
} |
| 1518 |
|
| 1519 |
*logical = le32_to_cpu(ex->ee_block) + ee_len - 1; |
| 1520 |
*phys = ext4_ext_pblock(ex) + ee_len - 1; |
| 1521 |
return 0; |
| 1522 |
} |
| 1523 |
|
| 1524 |
/* |
| 1525 |
* Search the closest allocated block to the right for *logical |
| 1526 |
* and returns it at @logical + it's physical address at @phys. |
| 1527 |
* If not exists, return 0 and @phys is set to 0. We will return |
| 1528 |
* 1 which means we found an allocated block and ret_ex is valid. |
| 1529 |
* Or return a (< 0) error code. |
| 1530 |
*/ |
| 1531 |
static int ext4_ext_search_right(struct inode *inode, |
| 1532 |
struct ext4_ext_path *path, |
| 1533 |
ext4_lblk_t *logical, ext4_fsblk_t *phys, |
| 1534 |
struct ext4_extent *ret_ex, int flags) |
| 1535 |
{ |
| 1536 |
struct buffer_head *bh = NULL; |
| 1537 |
struct ext4_extent_header *eh; |
| 1538 |
struct ext4_extent_idx *ix; |
| 1539 |
struct ext4_extent *ex; |
| 1540 |
int depth; /* Note, NOT eh_depth; depth from top of tree */ |
| 1541 |
int ee_len; |
| 1542 |
|
| 1543 |
if (unlikely(path == NULL)) { |
| 1544 |
EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical); |
| 1545 |
return -EFSCORRUPTED; |
| 1546 |
} |
| 1547 |
depth = path->p_depth; |
| 1548 |
*phys = 0; |
| 1549 |
|
| 1550 |
if (depth == 0 && path->p_ext == NULL) |
| 1551 |
return 0; |
| 1552 |
|
| 1553 |
/* usually extent in the path covers blocks smaller |
| 1554 |
* then *logical, but it can be that extent is the |
| 1555 |
* first one in the file */ |
| 1556 |
|
| 1557 |
ex = path[depth].p_ext; |
| 1558 |
ee_len = ext4_ext_get_actual_len(ex); |
| 1559 |
if (*logical < le32_to_cpu(ex->ee_block)) { |
| 1560 |
if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) { |
| 1561 |
EXT4_ERROR_INODE(inode, |
| 1562 |
"first_extent(path[%d].p_hdr) != ex", |
| 1563 |
depth); |
| 1564 |
return -EFSCORRUPTED; |
| 1565 |
} |
| 1566 |
while (--depth >= 0) { |
| 1567 |
ix = path[depth].p_idx; |
| 1568 |
if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) { |
| 1569 |
EXT4_ERROR_INODE(inode, |
| 1570 |
"ix != EXT_FIRST_INDEX *logical %d!", |
| 1571 |
*logical); |
| 1572 |
return -EFSCORRUPTED; |
| 1573 |
} |
| 1574 |
} |
| 1575 |
goto found_extent; |
| 1576 |
} |
| 1577 |
|
| 1578 |
if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) { |
| 1579 |
EXT4_ERROR_INODE(inode, |
| 1580 |
"logical %d < ee_block %d + ee_len %d!", |
| 1581 |
*logical, le32_to_cpu(ex->ee_block), ee_len); |
| 1582 |
return -EFSCORRUPTED; |
| 1583 |
} |
| 1584 |
|
| 1585 |
if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) { |
| 1586 |
/* next allocated block in this leaf */ |
| 1587 |
ex++; |
| 1588 |
goto found_extent; |
| 1589 |
} |
| 1590 |
|
| 1591 |
/* go up and search for index to the right */ |
| 1592 |
while (--depth >= 0) { |
| 1593 |
ix = path[depth].p_idx; |
| 1594 |
if (ix != EXT_LAST_INDEX(path[depth].p_hdr)) |
| 1595 |
goto got_index; |
| 1596 |
} |
| 1597 |
|
| 1598 |
/* we've gone up to the root and found no index to the right */ |
| 1599 |
return 0; |
| 1600 |
|
| 1601 |
got_index: |
| 1602 |
/* we've found index to the right, let's |
| 1603 |
* follow it and find the closest allocated |
| 1604 |
* block to the right */ |
| 1605 |
ix++; |
| 1606 |
while (++depth < path->p_depth) { |
| 1607 |
/* subtract from p_depth to get proper eh_depth */ |
| 1608 |
bh = read_extent_tree_block(inode, ix, path->p_depth - depth, |
| 1609 |
flags); |
| 1610 |
if (IS_ERR(bh)) |
| 1611 |
return PTR_ERR(bh); |
| 1612 |
eh = ext_block_hdr(bh); |
| 1613 |
ix = EXT_FIRST_INDEX(eh); |
| 1614 |
put_bh(bh); |
| 1615 |
} |
| 1616 |
|
| 1617 |
bh = read_extent_tree_block(inode, ix, path->p_depth - depth, flags); |
| 1618 |
if (IS_ERR(bh)) |
| 1619 |
return PTR_ERR(bh); |
| 1620 |
eh = ext_block_hdr(bh); |
| 1621 |
ex = EXT_FIRST_EXTENT(eh); |
| 1622 |
found_extent: |
| 1623 |
*logical = le32_to_cpu(ex->ee_block); |
| 1624 |
*phys = ext4_ext_pblock(ex); |
| 1625 |
if (ret_ex) |
| 1626 |
*ret_ex = *ex; |
| 1627 |
if (bh) |
| 1628 |
put_bh(bh); |
| 1629 |
return 1; |
| 1630 |
} |
| 1631 |
|
| 1632 |
/* |
| 1633 |
* ext4_ext_next_allocated_block: |
| 1634 |
* returns allocated block in subsequent extent or EXT_MAX_BLOCKS. |
| 1635 |
* NOTE: it considers block number from index entry as |
| 1636 |
* allocated block. Thus, index entries have to be consistent |
| 1637 |
* with leaves. |
| 1638 |
*/ |
| 1639 |
ext4_lblk_t |
| 1640 |
ext4_ext_next_allocated_block(struct ext4_ext_path *path) |
| 1641 |
{ |
| 1642 |
int depth; |
| 1643 |
|
| 1644 |
BUG_ON(path == NULL); |
| 1645 |
depth = path->p_depth; |
| 1646 |
|
| 1647 |
if (depth == 0 && path->p_ext == NULL) |
| 1648 |
return EXT_MAX_BLOCKS; |
| 1649 |
|
| 1650 |
while (depth >= 0) { |
| 1651 |
struct ext4_ext_path *p = &path[depth]; |
| 1652 |
|
| 1653 |
if (depth == path->p_depth) { |
| 1654 |
/* leaf */ |
| 1655 |
if (p->p_ext && p->p_ext != EXT_LAST_EXTENT(p->p_hdr)) |
| 1656 |
return le32_to_cpu(p->p_ext[1].ee_block); |
| 1657 |
} else { |
| 1658 |
/* index */ |
| 1659 |
if (p->p_idx != EXT_LAST_INDEX(p->p_hdr)) |
| 1660 |
return le32_to_cpu(p->p_idx[1].ei_block); |
| 1661 |
} |
| 1662 |
depth--; |
| 1663 |
} |
| 1664 |
|
| 1665 |
return EXT_MAX_BLOCKS; |
| 1666 |
} |
| 1667 |
|
| 1668 |
/* |
| 1669 |
* ext4_ext_next_leaf_block: |
| 1670 |
* returns first allocated block from next leaf or EXT_MAX_BLOCKS |
| 1671 |
*/ |
| 1672 |
static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path) |
| 1673 |
{ |
| 1674 |
int depth; |
| 1675 |
|
| 1676 |
BUG_ON(path == NULL); |
| 1677 |
depth = path->p_depth; |
| 1678 |
|
| 1679 |
/* zero-tree has no leaf blocks at all */ |
| 1680 |
if (depth == 0) |
| 1681 |
return EXT_MAX_BLOCKS; |
| 1682 |
|
| 1683 |
/* go to index block */ |
| 1684 |
depth--; |
| 1685 |
|
| 1686 |
while (depth >= 0) { |
| 1687 |
if (path[depth].p_idx != |
| 1688 |
EXT_LAST_INDEX(path[depth].p_hdr)) |
| 1689 |
return (ext4_lblk_t) |
| 1690 |
le32_to_cpu(path[depth].p_idx[1].ei_block); |
| 1691 |
depth--; |
| 1692 |
} |
| 1693 |
|
| 1694 |
return EXT_MAX_BLOCKS; |
| 1695 |
} |
| 1696 |
|
| 1697 |
/* |
| 1698 |
* ext4_ext_correct_indexes: |
| 1699 |
* if leaf gets modified and modified extent is first in the leaf, |
| 1700 |
* then we have to correct all indexes above. |
| 1701 |
* TODO: do we need to correct tree in all cases? |
| 1702 |
*/ |
| 1703 |
static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode, |
| 1704 |
struct ext4_ext_path *path) |
| 1705 |
{ |
| 1706 |
struct ext4_extent_header *eh; |
| 1707 |
int depth = ext_depth(inode); |
| 1708 |
struct ext4_extent *ex; |
| 1709 |
__le32 border; |
| 1710 |
int k, err = 0; |
| 1711 |
|
| 1712 |
eh = path[depth].p_hdr; |
| 1713 |
ex = path[depth].p_ext; |
| 1714 |
|
| 1715 |
if (unlikely(ex == NULL || eh == NULL)) { |
| 1716 |
EXT4_ERROR_INODE(inode, |
| 1717 |
"ex %p == NULL or eh %p == NULL", ex, eh); |
| 1718 |
return -EFSCORRUPTED; |
| 1719 |
} |
| 1720 |
|
| 1721 |
if (depth == 0) { |
| 1722 |
/* there is no tree at all */ |
| 1723 |
return 0; |
| 1724 |
} |
| 1725 |
|
| 1726 |
if (ex != EXT_FIRST_EXTENT(eh)) { |
| 1727 |
/* we correct tree if first leaf got modified only */ |
| 1728 |
return 0; |
| 1729 |
} |
| 1730 |
|
| 1731 |
/* |
| 1732 |
* TODO: we need correction if border is smaller than current one |
| 1733 |
*/ |
| 1734 |
k = depth - 1; |
| 1735 |
border = path[depth].p_ext->ee_block; |
| 1736 |
err = ext4_ext_get_access(handle, inode, path + k); |
| 1737 |
if (err) |
| 1738 |
return err; |
| 1739 |
if (unlikely(path[k].p_idx > EXT_LAST_INDEX(path[k].p_hdr))) { |
| 1740 |
EXT4_ERROR_INODE(inode, |
| 1741 |
"path[%d].p_idx %p > EXT_LAST_INDEX %p", |
| 1742 |
k, path[k].p_idx, |
| 1743 |
EXT_LAST_INDEX(path[k].p_hdr)); |
| 1744 |
return -EFSCORRUPTED; |
| 1745 |
} |
| 1746 |
path[k].p_idx->ei_block = border; |
| 1747 |
err = ext4_ext_dirty(handle, inode, path + k); |
| 1748 |
if (err) |
| 1749 |
return err; |
| 1750 |
|
| 1751 |
while (k--) { |
| 1752 |
/* change all left-side indexes */ |
| 1753 |
if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr)) |
| 1754 |
break; |
| 1755 |
err = ext4_ext_get_access(handle, inode, path + k); |
| 1756 |
if (err) |
| 1757 |
goto clean; |
| 1758 |
if (unlikely(path[k].p_idx > EXT_LAST_INDEX(path[k].p_hdr))) { |
| 1759 |
EXT4_ERROR_INODE(inode, |
| 1760 |
"path[%d].p_idx %p > EXT_LAST_INDEX %p", |
| 1761 |
k, path[k].p_idx, |
| 1762 |
EXT_LAST_INDEX(path[k].p_hdr)); |
| 1763 |
err = -EFSCORRUPTED; |
| 1764 |
goto clean; |
| 1765 |
} |
| 1766 |
path[k].p_idx->ei_block = border; |
| 1767 |
err = ext4_ext_dirty(handle, inode, path + k); |
| 1768 |
if (err) |
| 1769 |
goto clean; |
| 1770 |
} |
| 1771 |
return 0; |
| 1772 |
|
| 1773 |
clean: |
| 1774 |
/* |
| 1775 |
* The path[k].p_bh is either unmodified or with no verified bit |
| 1776 |
* set (see ext4_ext_get_access()). So just clear the verified bit |
| 1777 |
* of the successfully modified extents buffers, which will force |
| 1778 |
* these extents to be checked to avoid using inconsistent data. |
| 1779 |
*/ |
| 1780 |
while (++k < depth) |
| 1781 |
clear_buffer_verified(path[k].p_bh); |
| 1782 |
|
| 1783 |
return err; |
| 1784 |
} |
| 1785 |
|
| 1786 |
static int ext4_can_extents_be_merged(struct inode *inode, |
| 1787 |
struct ext4_extent *ex1, |
| 1788 |
struct ext4_extent *ex2) |
| 1789 |
{ |
| 1790 |
unsigned short ext1_ee_len, ext2_ee_len; |
| 1791 |
|
| 1792 |
if (ext4_ext_is_unwritten(ex1) != ext4_ext_is_unwritten(ex2)) |
| 1793 |
return 0; |
| 1794 |
|
| 1795 |
ext1_ee_len = ext4_ext_get_actual_len(ex1); |
| 1796 |
ext2_ee_len = ext4_ext_get_actual_len(ex2); |
| 1797 |
|
| 1798 |
if (le32_to_cpu(ex1->ee_block) + ext1_ee_len != |
| 1799 |
le32_to_cpu(ex2->ee_block)) |
| 1800 |
return 0; |
| 1801 |
|
| 1802 |
if (ext1_ee_len + ext2_ee_len > EXT_INIT_MAX_LEN) |
| 1803 |
return 0; |
| 1804 |
|
| 1805 |
if (ext4_ext_is_unwritten(ex1) && |
| 1806 |
ext1_ee_len + ext2_ee_len > EXT_UNWRITTEN_MAX_LEN) |
| 1807 |
return 0; |
| 1808 |
#ifdef AGGRESSIVE_TEST |
| 1809 |
if (ext1_ee_len >= 4) |
| 1810 |
return 0; |
| 1811 |
#endif |
| 1812 |
|
| 1813 |
if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2)) |
| 1814 |
return 1; |
| 1815 |
return 0; |
| 1816 |
} |
| 1817 |
|
| 1818 |
/* |
| 1819 |
* This function tries to merge the "ex" extent to the next extent in the tree. |
| 1820 |
* It always tries to merge towards right. If you want to merge towards |
| 1821 |
* left, pass "ex - 1" as argument instead of "ex". |
| 1822 |
* Returns 0 if the extents (ex and ex+1) were _not_ merged and returns |
| 1823 |
* 1 if they got merged. |
| 1824 |
*/ |
| 1825 |
static int ext4_ext_try_to_merge_right(struct inode *inode, |
| 1826 |
struct ext4_ext_path *path, |
| 1827 |
struct ext4_extent *ex) |
| 1828 |
{ |
| 1829 |
struct ext4_extent_header *eh; |
| 1830 |
unsigned int depth, len; |
| 1831 |
int merge_done = 0, unwritten; |
| 1832 |
|
| 1833 |
depth = ext_depth(inode); |
| 1834 |
BUG_ON(path[depth].p_hdr == NULL); |
| 1835 |
eh = path[depth].p_hdr; |
| 1836 |
|
| 1837 |
while (ex < EXT_LAST_EXTENT(eh)) { |
| 1838 |
if (!ext4_can_extents_be_merged(inode, ex, ex + 1)) |
| 1839 |
break; |
| 1840 |
/* merge with next extent! */ |
| 1841 |
unwritten = ext4_ext_is_unwritten(ex); |
| 1842 |
ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex) |
| 1843 |
+ ext4_ext_get_actual_len(ex + 1)); |
| 1844 |
if (unwritten) |
| 1845 |
ext4_ext_mark_unwritten(ex); |
| 1846 |
|
| 1847 |
if (ex + 1 < EXT_LAST_EXTENT(eh)) { |
| 1848 |
len = (EXT_LAST_EXTENT(eh) - ex - 1) |
| 1849 |
* sizeof(struct ext4_extent); |
| 1850 |
memmove(ex + 1, ex + 2, len); |
| 1851 |
} |
| 1852 |
le16_add_cpu(&eh->eh_entries, -1); |
| 1853 |
merge_done = 1; |
| 1854 |
WARN_ON(eh->eh_entries == 0); |
| 1855 |
if (!eh->eh_entries) |
| 1856 |
EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!"); |
| 1857 |
} |
| 1858 |
|
| 1859 |
return merge_done; |
| 1860 |
} |
| 1861 |
|
| 1862 |
/* |
| 1863 |
* This function does a very simple check to see if we can collapse |
| 1864 |
* an extent tree with a single extent tree leaf block into the inode. |
| 1865 |
*/ |
| 1866 |
static void ext4_ext_try_to_merge_up(handle_t *handle, |
| 1867 |
struct inode *inode, |
| 1868 |
struct ext4_ext_path *path) |
| 1869 |
{ |
| 1870 |
size_t s; |
| 1871 |
unsigned max_root = ext4_ext_space_root(inode, 0); |
| 1872 |
ext4_fsblk_t blk; |
| 1873 |
|
| 1874 |
if ((path[0].p_depth != 1) || |
| 1875 |
(le16_to_cpu(path[0].p_hdr->eh_entries) != 1) || |
| 1876 |
(le16_to_cpu(path[1].p_hdr->eh_entries) > max_root)) |
| 1877 |
return; |
| 1878 |
|
| 1879 |
/* |
| 1880 |
* We need to modify the block allocation bitmap and the block |
| 1881 |
* group descriptor to release the extent tree block. If we |
| 1882 |
* can't get the journal credits, give up. |
| 1883 |
*/ |
| 1884 |
if (ext4_journal_extend(handle, 2, |
| 1885 |
ext4_free_metadata_revoke_credits(inode->i_sb, 1))) |
| 1886 |
return; |
| 1887 |
|
| 1888 |
/* |
| 1889 |
* Copy the extent data up to the inode |
| 1890 |
*/ |
| 1891 |
blk = ext4_idx_pblock(path[0].p_idx); |
| 1892 |
s = le16_to_cpu(path[1].p_hdr->eh_entries) * |
| 1893 |
sizeof(struct ext4_extent_idx); |
| 1894 |
s += sizeof(struct ext4_extent_header); |
| 1895 |
|
| 1896 |
path[1].p_maxdepth = path[0].p_maxdepth; |
| 1897 |
memcpy(path[0].p_hdr, path[1].p_hdr, s); |
| 1898 |
path[0].p_depth = 0; |
| 1899 |
path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) + |
| 1900 |
(path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr)); |
| 1901 |
path[0].p_hdr->eh_max = cpu_to_le16(max_root); |
| 1902 |
|
| 1903 |
ext4_ext_path_brelse(path + 1); |
| 1904 |
ext4_free_blocks(handle, inode, NULL, blk, 1, |
| 1905 |
EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET); |
| 1906 |
} |
| 1907 |
|
| 1908 |
/* |
| 1909 |
* This function tries to merge the @ex extent to neighbours in the tree, then |
| 1910 |
* tries to collapse the extent tree into the inode. |
| 1911 |
*/ |
| 1912 |
static void ext4_ext_try_to_merge(handle_t *handle, |
| 1913 |
struct inode *inode, |
| 1914 |
struct ext4_ext_path *path, |
| 1915 |
struct ext4_extent *ex) |
| 1916 |
{ |
| 1917 |
struct ext4_extent_header *eh; |
| 1918 |
unsigned int depth; |
| 1919 |
int merge_done = 0; |
| 1920 |
|
| 1921 |
depth = ext_depth(inode); |
| 1922 |
BUG_ON(path[depth].p_hdr == NULL); |
| 1923 |
eh = path[depth].p_hdr; |
| 1924 |
|
| 1925 |
if (ex > EXT_FIRST_EXTENT(eh)) |
| 1926 |
merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1); |
| 1927 |
|
| 1928 |
if (!merge_done) |
| 1929 |
(void) ext4_ext_try_to_merge_right(inode, path, ex); |
| 1930 |
|
| 1931 |
ext4_ext_try_to_merge_up(handle, inode, path); |
| 1932 |
} |
| 1933 |
|
| 1934 |
/* |
| 1935 |
* check if a portion of the "newext" extent overlaps with an |
| 1936 |
* existing extent. |
| 1937 |
* |
| 1938 |
* If there is an overlap discovered, it updates the length of the newext |
| 1939 |
* such that there will be no overlap, and then returns 1. |
| 1940 |
* If there is no overlap found, it returns 0. |
| 1941 |
*/ |
| 1942 |
static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi, |
| 1943 |
struct inode *inode, |
| 1944 |
struct ext4_extent *newext, |
| 1945 |
struct ext4_ext_path *path) |
| 1946 |
{ |
| 1947 |
ext4_lblk_t b1, b2; |
| 1948 |
unsigned int depth, len1; |
| 1949 |
unsigned int ret = 0; |
| 1950 |
|
| 1951 |
b1 = le32_to_cpu(newext->ee_block); |
| 1952 |
len1 = ext4_ext_get_actual_len(newext); |
| 1953 |
depth = ext_depth(inode); |
| 1954 |
if (!path[depth].p_ext) |
| 1955 |
goto out; |
| 1956 |
b2 = EXT4_LBLK_CMASK(sbi, le32_to_cpu(path[depth].p_ext->ee_block)); |
| 1957 |
|
| 1958 |
/* |
| 1959 |
* get the next allocated block if the extent in the path |
| 1960 |
* is before the requested block(s) |
| 1961 |
*/ |
| 1962 |
if (b2 < b1) { |
| 1963 |
b2 = ext4_ext_next_allocated_block(path); |
| 1964 |
if (b2 == EXT_MAX_BLOCKS) |
| 1965 |
goto out; |
| 1966 |
b2 = EXT4_LBLK_CMASK(sbi, b2); |
| 1967 |
} |
| 1968 |
|
| 1969 |
/* check for wrap through zero on extent logical start block*/ |
| 1970 |
if (b1 + len1 < b1) { |
| 1971 |
len1 = EXT_MAX_BLOCKS - b1; |
| 1972 |
newext->ee_len = cpu_to_le16(len1); |
| 1973 |
ret = 1; |
| 1974 |
} |
| 1975 |
|
| 1976 |
/* check for overlap */ |
| 1977 |
if (b1 + len1 > b2) { |
| 1978 |
newext->ee_len = cpu_to_le16(b2 - b1); |
| 1979 |
ret = 1; |
| 1980 |
} |
| 1981 |
out: |
| 1982 |
return ret; |
| 1983 |
} |
| 1984 |
|
| 1985 |
/* |
| 1986 |
* ext4_ext_insert_extent: |
| 1987 |
* tries to merge requested extent into the existing extent or |
| 1988 |
* inserts requested extent as new one into the tree, |
| 1989 |
* creating new leaf in the no-space case. |
| 1990 |
*/ |
| 1991 |
struct ext4_ext_path * |
| 1992 |
ext4_ext_insert_extent(handle_t *handle, struct inode *inode, |
| 1993 |
struct ext4_ext_path *path, |
| 1994 |
struct ext4_extent *newext, int gb_flags) |
| 1995 |
{ |
| 1996 |
struct ext4_extent_header *eh; |
| 1997 |
struct ext4_extent *ex, *fex; |
| 1998 |
struct ext4_extent *nearex; /* nearest extent */ |
| 1999 |
int depth, len, err = 0; |
| 2000 |
ext4_lblk_t next; |
| 2001 |
int mb_flags = 0, unwritten; |
| 2002 |
|
| 2003 |
KUNIT_STATIC_STUB_REDIRECT(ext4_ext_insert_extent, handle, inode, path, |
| 2004 |
newext, gb_flags); |
| 2005 |
|
| 2006 |
if (gb_flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) |
| 2007 |
mb_flags |= EXT4_MB_DELALLOC_RESERVED; |
| 2008 |
if (unlikely(ext4_ext_get_actual_len(newext) == 0)) { |
| 2009 |
EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0"); |
| 2010 |
err = -EFSCORRUPTED; |
| 2011 |
goto errout; |
| 2012 |
} |
| 2013 |
depth = ext_depth(inode); |
| 2014 |
ex = path[depth].p_ext; |
| 2015 |
eh = path[depth].p_hdr; |
| 2016 |
if (unlikely(path[depth].p_hdr == NULL)) { |
| 2017 |
EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth); |
| 2018 |
err = -EFSCORRUPTED; |
| 2019 |
goto errout; |
| 2020 |
} |
| 2021 |
|
| 2022 |
/* try to insert block into found extent and return */ |
| 2023 |
if (ex && !(gb_flags & EXT4_GET_BLOCKS_SPLIT_NOMERGE)) { |
| 2024 |
|
| 2025 |
/* |
| 2026 |
* Try to see whether we should rather test the extent on |
| 2027 |
* right from ex, or from the left of ex. This is because |
| 2028 |
* ext4_find_extent() can return either extent on the |
| 2029 |
* left, or on the right from the searched position. This |
| 2030 |
* will make merging more effective. |
| 2031 |
*/ |
| 2032 |
if (ex < EXT_LAST_EXTENT(eh) && |
| 2033 |
(le32_to_cpu(ex->ee_block) + |
| 2034 |
ext4_ext_get_actual_len(ex) < |
| 2035 |
le32_to_cpu(newext->ee_block))) { |
| 2036 |
ex += 1; |
| 2037 |
goto prepend; |
| 2038 |
} else if ((ex > EXT_FIRST_EXTENT(eh)) && |
| 2039 |
(le32_to_cpu(newext->ee_block) + |
| 2040 |
ext4_ext_get_actual_len(newext) < |
| 2041 |
le32_to_cpu(ex->ee_block))) |
| 2042 |
ex -= 1; |
| 2043 |
|
| 2044 |
/* Try to append newex to the ex */ |
| 2045 |
if (ext4_can_extents_be_merged(inode, ex, newext)) { |
| 2046 |
ext_debug(inode, "append [%d]%d block to %u:[%d]%d" |
| 2047 |
"(from %llu)\n", |
| 2048 |
ext4_ext_is_unwritten(newext), |
| 2049 |
ext4_ext_get_actual_len(newext), |
| 2050 |
le32_to_cpu(ex->ee_block), |
| 2051 |
ext4_ext_is_unwritten(ex), |
| 2052 |
ext4_ext_get_actual_len(ex), |
| 2053 |
ext4_ext_pblock(ex)); |
| 2054 |
err = ext4_ext_get_access(handle, inode, |
| 2055 |
path + depth); |
| 2056 |
if (err) |
| 2057 |
goto errout; |
| 2058 |
unwritten = ext4_ext_is_unwritten(ex); |
| 2059 |
ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex) |
| 2060 |
+ ext4_ext_get_actual_len(newext)); |
| 2061 |
if (unwritten) |
| 2062 |
ext4_ext_mark_unwritten(ex); |
| 2063 |
nearex = ex; |
| 2064 |
goto merge; |
| 2065 |
} |
| 2066 |
|
| 2067 |
prepend: |
| 2068 |
/* Try to prepend newex to the ex */ |
| 2069 |
if (ext4_can_extents_be_merged(inode, newext, ex)) { |
| 2070 |
ext_debug(inode, "prepend %u[%d]%d block to %u:[%d]%d" |
| 2071 |
"(from %llu)\n", |
| 2072 |
le32_to_cpu(newext->ee_block), |
| 2073 |
ext4_ext_is_unwritten(newext), |
| 2074 |
ext4_ext_get_actual_len(newext), |
| 2075 |
le32_to_cpu(ex->ee_block), |
| 2076 |
ext4_ext_is_unwritten(ex), |
| 2077 |
ext4_ext_get_actual_len(ex), |
| 2078 |
ext4_ext_pblock(ex)); |
| 2079 |
err = ext4_ext_get_access(handle, inode, |
| 2080 |
path + depth); |
| 2081 |
if (err) |
| 2082 |
goto errout; |
| 2083 |
|
| 2084 |
unwritten = ext4_ext_is_unwritten(ex); |
| 2085 |
ex->ee_block = newext->ee_block; |
| 2086 |
ext4_ext_store_pblock(ex, ext4_ext_pblock(newext)); |
| 2087 |
ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex) |
| 2088 |
+ ext4_ext_get_actual_len(newext)); |
| 2089 |
if (unwritten) |
| 2090 |
ext4_ext_mark_unwritten(ex); |
| 2091 |
nearex = ex; |
| 2092 |
goto merge; |
| 2093 |
} |
| 2094 |
} |
| 2095 |
|
| 2096 |
depth = ext_depth(inode); |
| 2097 |
eh = path[depth].p_hdr; |
| 2098 |
if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) |
| 2099 |
goto has_space; |
| 2100 |
|
| 2101 |
/* probably next leaf has space for us? */ |
| 2102 |
fex = EXT_LAST_EXTENT(eh); |
| 2103 |
next = EXT_MAX_BLOCKS; |
| 2104 |
if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block)) |
| 2105 |
next = ext4_ext_next_leaf_block(path); |
| 2106 |
if (next != EXT_MAX_BLOCKS) { |
| 2107 |
struct ext4_ext_path *npath; |
| 2108 |
|
| 2109 |
ext_debug(inode, "next leaf block - %u\n", next); |
| 2110 |
npath = ext4_find_extent(inode, next, NULL, gb_flags); |
| 2111 |
if (IS_ERR(npath)) { |
| 2112 |
err = PTR_ERR(npath); |
| 2113 |
goto errout; |
| 2114 |
} |
| 2115 |
BUG_ON(npath->p_depth != path->p_depth); |
| 2116 |
eh = npath[depth].p_hdr; |
| 2117 |
if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) { |
| 2118 |
ext_debug(inode, "next leaf isn't full(%d)\n", |
| 2119 |
le16_to_cpu(eh->eh_entries)); |
| 2120 |
ext4_free_ext_path(path); |
| 2121 |
path = npath; |
| 2122 |
goto has_space; |
| 2123 |
} |
| 2124 |
ext_debug(inode, "next leaf has no free space(%d,%d)\n", |
| 2125 |
le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max)); |
| 2126 |
ext4_free_ext_path(npath); |
| 2127 |
} |
| 2128 |
|
| 2129 |
/* |
| 2130 |
* There is no free space in the found leaf. |
| 2131 |
* We're gonna add a new leaf in the tree. |
| 2132 |
*/ |
| 2133 |
if (gb_flags & EXT4_GET_BLOCKS_METADATA_NOFAIL) |
| 2134 |
mb_flags |= EXT4_MB_USE_RESERVED; |
| 2135 |
path = ext4_ext_create_new_leaf(handle, inode, mb_flags, gb_flags, |
| 2136 |
path, newext); |
| 2137 |
if (IS_ERR(path)) |
| 2138 |
return path; |
| 2139 |
depth = ext_depth(inode); |
| 2140 |
eh = path[depth].p_hdr; |
| 2141 |
|
| 2142 |
has_space: |
| 2143 |
nearex = path[depth].p_ext; |
| 2144 |
|
| 2145 |
err = ext4_ext_get_access(handle, inode, path + depth); |
| 2146 |
if (err) |
| 2147 |
goto errout; |
| 2148 |
|
| 2149 |
if (!nearex) { |
| 2150 |
/* there is no extent in this leaf, create first one */ |
| 2151 |
ext_debug(inode, "first extent in the leaf: %u:%llu:[%d]%d\n", |
| 2152 |
le32_to_cpu(newext->ee_block), |
| 2153 |
ext4_ext_pblock(newext), |
| 2154 |
ext4_ext_is_unwritten(newext), |
| 2155 |
ext4_ext_get_actual_len(newext)); |
| 2156 |
nearex = EXT_FIRST_EXTENT(eh); |
| 2157 |
} else { |
| 2158 |
if (le32_to_cpu(newext->ee_block) |
| 2159 |
> le32_to_cpu(nearex->ee_block)) { |
| 2160 |
/* Insert after */ |
| 2161 |
ext_debug(inode, "insert %u:%llu:[%d]%d before: " |
| 2162 |
"nearest %p\n", |
| 2163 |
le32_to_cpu(newext->ee_block), |
| 2164 |
ext4_ext_pblock(newext), |
| 2165 |
ext4_ext_is_unwritten(newext), |
| 2166 |
ext4_ext_get_actual_len(newext), |
| 2167 |
nearex); |
| 2168 |
nearex++; |
| 2169 |
} else { |
| 2170 |
/* Insert before */ |
| 2171 |
BUG_ON(newext->ee_block == nearex->ee_block); |
| 2172 |
ext_debug(inode, "insert %u:%llu:[%d]%d after: " |
| 2173 |
"nearest %p\n", |
| 2174 |
le32_to_cpu(newext->ee_block), |
| 2175 |
ext4_ext_pblock(newext), |
| 2176 |
ext4_ext_is_unwritten(newext), |
| 2177 |
ext4_ext_get_actual_len(newext), |
| 2178 |
nearex); |
| 2179 |
} |
| 2180 |
len = EXT_LAST_EXTENT(eh) - nearex + 1; |
| 2181 |
if (len > 0) { |
| 2182 |
ext_debug(inode, "insert %u:%llu:[%d]%d: " |
| 2183 |
"move %d extents from 0x%p to 0x%p\n", |
| 2184 |
le32_to_cpu(newext->ee_block), |
| 2185 |
ext4_ext_pblock(newext), |
| 2186 |
ext4_ext_is_unwritten(newext), |
| 2187 |
ext4_ext_get_actual_len(newext), |
| 2188 |
len, nearex, nearex + 1); |
| 2189 |
memmove(nearex + 1, nearex, |
| 2190 |
len * sizeof(struct ext4_extent)); |
| 2191 |
} |
| 2192 |
} |
| 2193 |
|
| 2194 |
le16_add_cpu(&eh->eh_entries, 1); |
| 2195 |
path[depth].p_ext = nearex; |
| 2196 |
nearex->ee_block = newext->ee_block; |
| 2197 |
ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext)); |
| 2198 |
nearex->ee_len = newext->ee_len; |
| 2199 |
|
| 2200 |
merge: |
| 2201 |
/* try to merge extents */ |
| 2202 |
if (!(gb_flags & EXT4_GET_BLOCKS_SPLIT_NOMERGE)) |
| 2203 |
ext4_ext_try_to_merge(handle, inode, path, nearex); |
| 2204 |
|
| 2205 |
/* time to correct all indexes above */ |
| 2206 |
err = ext4_ext_correct_indexes(handle, inode, path); |
| 2207 |
if (err) |
| 2208 |
goto errout; |
| 2209 |
|
| 2210 |
err = ext4_ext_dirty(handle, inode, path + path->p_depth); |
| 2211 |
if (err) |
| 2212 |
goto errout; |
| 2213 |
|
| 2214 |
return path; |
| 2215 |
|
| 2216 |
errout: |
| 2217 |
ext4_free_ext_path(path); |
| 2218 |
return ERR_PTR(err); |
| 2219 |
} |
| 2220 |
|
| 2221 |
static int ext4_fill_es_cache_info(struct inode *inode, |
| 2222 |
ext4_lblk_t block, ext4_lblk_t num, |
| 2223 |
struct fiemap_extent_info *fieinfo) |
| 2224 |
{ |
| 2225 |
ext4_lblk_t next, end = block + num - 1; |
| 2226 |
struct extent_status es; |
| 2227 |
unsigned char blksize_bits = inode->i_sb->s_blocksize_bits; |
| 2228 |
unsigned int flags; |
| 2229 |
int err; |
| 2230 |
|
| 2231 |
while (block <= end) { |
| 2232 |
next = 0; |
| 2233 |
flags = 0; |
| 2234 |
if (!ext4_es_lookup_extent(inode, block, &next, &es, NULL)) |
| 2235 |
break; |
| 2236 |
if (ext4_es_is_unwritten(&es)) |
| 2237 |
flags |= FIEMAP_EXTENT_UNWRITTEN; |
| 2238 |
if (ext4_es_is_delayed(&es)) |
| 2239 |
flags |= (FIEMAP_EXTENT_DELALLOC | |
| 2240 |
FIEMAP_EXTENT_UNKNOWN); |
| 2241 |
if (ext4_es_is_hole(&es)) |
| 2242 |
flags |= EXT4_FIEMAP_EXTENT_HOLE; |
| 2243 |
if (next == 0) |
| 2244 |
flags |= FIEMAP_EXTENT_LAST; |
| 2245 |
if (flags & (FIEMAP_EXTENT_DELALLOC| |
| 2246 |
EXT4_FIEMAP_EXTENT_HOLE)) |
| 2247 |
es.es_pblk = 0; |
| 2248 |
else |
| 2249 |
es.es_pblk = ext4_es_pblock(&es); |
| 2250 |
err = fiemap_fill_next_extent(fieinfo, |
| 2251 |
(__u64)es.es_lblk << blksize_bits, |
| 2252 |
(__u64)es.es_pblk << blksize_bits, |
| 2253 |
(__u64)es.es_len << blksize_bits, |
| 2254 |
flags); |
| 2255 |
if (next == 0) |
| 2256 |
break; |
| 2257 |
block = next; |
| 2258 |
if (err < 0) |
| 2259 |
return err; |
| 2260 |
if (err == 1) |
| 2261 |
return 0; |
| 2262 |
} |
| 2263 |
return 0; |
| 2264 |
} |
| 2265 |
|
| 2266 |
|
| 2267 |
/* |
| 2268 |
* ext4_ext_find_hole - find hole around given block according to the given path |
| 2269 |
* @inode: inode we lookup in |
| 2270 |
* @path: path in extent tree to @lblk |
| 2271 |
* @lblk: pointer to logical block around which we want to determine hole |
| 2272 |
* |
| 2273 |
* Determine hole length (and start if easily possible) around given logical |
| 2274 |
* block. We don't try too hard to find the beginning of the hole but @path |
| 2275 |
* actually points to extent before @lblk, we provide it. |
| 2276 |
* |
| 2277 |
* The function returns the length of a hole starting at @lblk. We update @lblk |
| 2278 |
* to the beginning of the hole if we managed to find it. |
| 2279 |
*/ |
| 2280 |
static ext4_lblk_t ext4_ext_find_hole(struct inode *inode, |
| 2281 |
struct ext4_ext_path *path, |
| 2282 |
ext4_lblk_t *lblk) |
| 2283 |
{ |
| 2284 |
int depth = ext_depth(inode); |
| 2285 |
struct ext4_extent *ex; |
| 2286 |
ext4_lblk_t len; |
| 2287 |
|
| 2288 |
ex = path[depth].p_ext; |
| 2289 |
if (ex == NULL) { |
| 2290 |
/* there is no extent yet, so gap is [0;-] */ |
| 2291 |
*lblk = 0; |
| 2292 |
len = EXT_MAX_BLOCKS; |
| 2293 |
} else if (*lblk < le32_to_cpu(ex->ee_block)) { |
| 2294 |
len = le32_to_cpu(ex->ee_block) - *lblk; |
| 2295 |
} else if (*lblk >= le32_to_cpu(ex->ee_block) |
| 2296 |
+ ext4_ext_get_actual_len(ex)) { |
| 2297 |
ext4_lblk_t next; |
| 2298 |
|
| 2299 |
*lblk = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex); |
| 2300 |
next = ext4_ext_next_allocated_block(path); |
| 2301 |
BUG_ON(next == *lblk); |
| 2302 |
len = next - *lblk; |
| 2303 |
} else { |
| 2304 |
BUG(); |
| 2305 |
} |
| 2306 |
return len; |
| 2307 |
} |
| 2308 |
|
| 2309 |
/* |
| 2310 |
* ext4_ext_rm_idx: |
| 2311 |
* removes index from the index block. |
| 2312 |
*/ |
| 2313 |
static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode, |
| 2314 |
struct ext4_ext_path *path, int depth) |
| 2315 |
{ |
| 2316 |
int err; |
| 2317 |
ext4_fsblk_t leaf; |
| 2318 |
int k = depth - 1; |
| 2319 |
|
| 2320 |
/* free index block */ |
| 2321 |
leaf = ext4_idx_pblock(path[k].p_idx); |
| 2322 |
if (unlikely(path[k].p_hdr->eh_entries == 0)) { |
| 2323 |
EXT4_ERROR_INODE(inode, "path[%d].p_hdr->eh_entries == 0", k); |
| 2324 |
return -EFSCORRUPTED; |
| 2325 |
} |
| 2326 |
err = ext4_ext_get_access(handle, inode, path + k); |
| 2327 |
if (err) |
| 2328 |
return err; |
| 2329 |
|
| 2330 |
if (path[k].p_idx != EXT_LAST_INDEX(path[k].p_hdr)) { |
| 2331 |
int len = EXT_LAST_INDEX(path[k].p_hdr) - path[k].p_idx; |
| 2332 |
len *= sizeof(struct ext4_extent_idx); |
| 2333 |
memmove(path[k].p_idx, path[k].p_idx + 1, len); |
| 2334 |
} |
| 2335 |
|
| 2336 |
le16_add_cpu(&path[k].p_hdr->eh_entries, -1); |
| 2337 |
err = ext4_ext_dirty(handle, inode, path + k); |
| 2338 |
if (err) |
| 2339 |
return err; |
| 2340 |
ext_debug(inode, "index is empty, remove it, free block %llu\n", leaf); |
| 2341 |
trace_ext4_ext_rm_idx(inode, leaf); |
| 2342 |
|
| 2343 |
ext4_free_blocks(handle, inode, NULL, leaf, 1, |
| 2344 |
EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET); |
| 2345 |
|
| 2346 |
while (--k >= 0) { |
| 2347 |
if (path[k + 1].p_idx != EXT_FIRST_INDEX(path[k + 1].p_hdr)) |
| 2348 |
break; |
| 2349 |
err = ext4_ext_get_access(handle, inode, path + k); |
| 2350 |
if (err) |
| 2351 |
goto clean; |
| 2352 |
path[k].p_idx->ei_block = path[k + 1].p_idx->ei_block; |
| 2353 |
err = ext4_ext_dirty(handle, inode, path + k); |
| 2354 |
if (err) |
| 2355 |
goto clean; |
| 2356 |
} |
| 2357 |
return 0; |
| 2358 |
|
| 2359 |
clean: |
| 2360 |
/* |
| 2361 |
* The path[k].p_bh is either unmodified or with no verified bit |
| 2362 |
* set (see ext4_ext_get_access()). So just clear the verified bit |
| 2363 |
* of the successfully modified extents buffers, which will force |
| 2364 |
* these extents to be checked to avoid using inconsistent data. |
| 2365 |
*/ |
| 2366 |
while (++k < depth) |
| 2367 |
clear_buffer_verified(path[k].p_bh); |
| 2368 |
|
| 2369 |
return err; |
| 2370 |
} |
| 2371 |
|
| 2372 |
/* |
| 2373 |
* ext4_ext_calc_credits_for_single_extent: |
| 2374 |
* This routine returns max. credits that needed to insert an extent |
| 2375 |
* to the extent tree. |
| 2376 |
* When pass the actual path, the caller should calculate credits |
| 2377 |
* under i_data_sem. |
| 2378 |
*/ |
| 2379 |
int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks, |
| 2380 |
struct ext4_ext_path *path) |
| 2381 |
{ |
| 2382 |
if (path) { |
| 2383 |
int depth = ext_depth(inode); |
| 2384 |
int ret = 0; |
| 2385 |
|
| 2386 |
/* probably there is space in leaf? */ |
| 2387 |
if (le16_to_cpu(path[depth].p_hdr->eh_entries) |
| 2388 |
< le16_to_cpu(path[depth].p_hdr->eh_max)) { |
| 2389 |
|
| 2390 |
/* |
| 2391 |
* There are some space in the leaf tree, no |
| 2392 |
* need to account for leaf block credit |
| 2393 |
* |
| 2394 |
* bitmaps and block group descriptor blocks |
| 2395 |
* and other metadata blocks still need to be |
| 2396 |
* accounted. |
| 2397 |
*/ |
| 2398 |
/* 1 bitmap, 1 block group descriptor */ |
| 2399 |
ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb); |
| 2400 |
return ret; |
| 2401 |
} |
| 2402 |
} |
| 2403 |
|
| 2404 |
return ext4_chunk_trans_blocks(inode, nrblocks); |
| 2405 |
} |
| 2406 |
|
| 2407 |
/* |
| 2408 |
* How many index/leaf blocks need to change/allocate to add @extents extents? |
| 2409 |
* |
| 2410 |
* If we add a single extent, then in the worse case, each tree level |
| 2411 |
* index/leaf need to be changed in case of the tree split. |
| 2412 |
* |
| 2413 |
* If more extents are inserted, they could cause the whole tree split more |
| 2414 |
* than once, but this is really rare. |
| 2415 |
*/ |
| 2416 |
int ext4_ext_index_trans_blocks(struct inode *inode, int extents) |
| 2417 |
{ |
| 2418 |
int index; |
| 2419 |
|
| 2420 |
/* If we are converting the inline data, only one is needed here. */ |
| 2421 |
if (ext4_has_inline_data(inode)) |
| 2422 |
return 1; |
| 2423 |
|
| 2424 |
/* |
| 2425 |
* Extent tree can change between the time we estimate credits and |
| 2426 |
* the time we actually modify the tree. Assume the worst case. |
| 2427 |
*/ |
| 2428 |
if (extents <= 1) |
| 2429 |
index = (EXT4_MAX_EXTENT_DEPTH * 2) + extents; |
| 2430 |
else |
| 2431 |
index = (EXT4_MAX_EXTENT_DEPTH * 3) + |
| 2432 |
DIV_ROUND_UP(extents, ext4_ext_space_block(inode, 0)); |
| 2433 |
|
| 2434 |
return index; |
| 2435 |
} |
| 2436 |
|
| 2437 |
static inline int get_default_free_blocks_flags(struct inode *inode) |
| 2438 |
{ |
| 2439 |
if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode) || |
| 2440 |
ext4_test_inode_flag(inode, EXT4_INODE_EA_INODE)) |
| 2441 |
return EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET; |
| 2442 |
else if (ext4_should_journal_data(inode)) |
| 2443 |
return EXT4_FREE_BLOCKS_FORGET; |
| 2444 |
return 0; |
| 2445 |
} |
| 2446 |
|
| 2447 |
/* |
| 2448 |
* ext4_rereserve_cluster - increment the reserved cluster count when |
| 2449 |
* freeing a cluster with a pending reservation |
| 2450 |
* |
| 2451 |
* @inode - file containing the cluster |
| 2452 |
* @lblk - logical block in cluster to be reserved |
| 2453 |
* |
| 2454 |
* Increments the reserved cluster count and adjusts quota in a bigalloc |
| 2455 |
* file system when freeing a partial cluster containing at least one |
| 2456 |
* delayed and unwritten block. A partial cluster meeting that |
| 2457 |
* requirement will have a pending reservation. If so, the |
| 2458 |
* RERESERVE_CLUSTER flag is used when calling ext4_free_blocks() to |
| 2459 |
* defer reserved and allocated space accounting to a subsequent call |
| 2460 |
* to this function. |
| 2461 |
*/ |
| 2462 |
static void ext4_rereserve_cluster(struct inode *inode, ext4_lblk_t lblk) |
| 2463 |
{ |
| 2464 |
struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
| 2465 |
struct ext4_inode_info *ei = EXT4_I(inode); |
| 2466 |
|
| 2467 |
dquot_reclaim_block(inode, EXT4_C2B(sbi, 1)); |
| 2468 |
|
| 2469 |
spin_lock(&ei->i_block_reservation_lock); |
| 2470 |
ei->i_reserved_data_blocks++; |
| 2471 |
percpu_counter_add(&sbi->s_dirtyclusters_counter, 1); |
| 2472 |
spin_unlock(&ei->i_block_reservation_lock); |
| 2473 |
|
| 2474 |
percpu_counter_add(&sbi->s_freeclusters_counter, 1); |
| 2475 |
ext4_remove_pending(inode, lblk); |
| 2476 |
} |
| 2477 |
|
| 2478 |
static int ext4_remove_blocks(handle_t *handle, struct inode *inode, |
| 2479 |
struct ext4_extent *ex, |
| 2480 |
struct partial_cluster *partial, |
| 2481 |
ext4_lblk_t from, ext4_lblk_t to) |
| 2482 |
{ |
| 2483 |
struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
| 2484 |
unsigned short ee_len = ext4_ext_get_actual_len(ex); |
| 2485 |
ext4_fsblk_t last_pblk, pblk; |
| 2486 |
ext4_lblk_t num; |
| 2487 |
int flags; |
| 2488 |
|
| 2489 |
/* only extent tail removal is allowed */ |
| 2490 |
if (from < le32_to_cpu(ex->ee_block) || |
| 2491 |
to != le32_to_cpu(ex->ee_block) + ee_len - 1) { |
| 2492 |
ext4_error(sbi->s_sb, |
| 2493 |
"strange request: removal(2) %u-%u from %u:%u", |
| 2494 |
from, to, le32_to_cpu(ex->ee_block), ee_len); |
| 2495 |
return 0; |
| 2496 |
} |
| 2497 |
|
| 2498 |
#ifdef EXTENTS_STATS |
| 2499 |
spin_lock(&sbi->s_ext_stats_lock); |
| 2500 |
sbi->s_ext_blocks += ee_len; |
| 2501 |
sbi->s_ext_extents++; |
| 2502 |
if (ee_len < sbi->s_ext_min) |
| 2503 |
sbi->s_ext_min = ee_len; |
| 2504 |
if (ee_len > sbi->s_ext_max) |
| 2505 |
sbi->s_ext_max = ee_len; |
| 2506 |
if (ext_depth(inode) > sbi->s_depth_max) |
| 2507 |
sbi->s_depth_max = ext_depth(inode); |
| 2508 |
spin_unlock(&sbi->s_ext_stats_lock); |
| 2509 |
#endif |
| 2510 |
|
| 2511 |
trace_ext4_remove_blocks(inode, ex, from, to, partial); |
| 2512 |
|
| 2513 |
/* |
| 2514 |
* if we have a partial cluster, and it's different from the |
| 2515 |
* cluster of the last block in the extent, we free it |
| 2516 |
*/ |
| 2517 |
last_pblk = ext4_ext_pblock(ex) + ee_len - 1; |
| 2518 |
|
| 2519 |
if (partial->state != initial && |
| 2520 |
partial->pclu != EXT4_B2C(sbi, last_pblk)) { |
| 2521 |
if (partial->state == tofree) { |
| 2522 |
flags = get_default_free_blocks_flags(inode); |
| 2523 |
if (ext4_is_pending(inode, partial->lblk)) |
| 2524 |
flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER; |
| 2525 |
ext4_free_blocks(handle, inode, NULL, |
| 2526 |
EXT4_C2B(sbi, partial->pclu), |
| 2527 |
sbi->s_cluster_ratio, flags); |
| 2528 |
if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER) |
| 2529 |
ext4_rereserve_cluster(inode, partial->lblk); |
| 2530 |
} |
| 2531 |
partial->state = initial; |
| 2532 |
} |
| 2533 |
|
| 2534 |
num = le32_to_cpu(ex->ee_block) + ee_len - from; |
| 2535 |
pblk = ext4_ext_pblock(ex) + ee_len - num; |
| 2536 |
|
| 2537 |
/* |
| 2538 |
* We free the partial cluster at the end of the extent (if any), |
| 2539 |
* unless the cluster is used by another extent (partial_cluster |
| 2540 |
* state is nofree). If a partial cluster exists here, it must be |
| 2541 |
* shared with the last block in the extent. |
| 2542 |
*/ |
| 2543 |
flags = get_default_free_blocks_flags(inode); |
| 2544 |
|
| 2545 |
/* partial, left end cluster aligned, right end unaligned */ |
| 2546 |
if ((EXT4_LBLK_COFF(sbi, to) != sbi->s_cluster_ratio - 1) && |
| 2547 |
(EXT4_LBLK_CMASK(sbi, to) >= from) && |
| 2548 |
(partial->state != nofree)) { |
| 2549 |
if (ext4_is_pending(inode, to)) |
| 2550 |
flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER; |
| 2551 |
ext4_free_blocks(handle, inode, NULL, |
| 2552 |
EXT4_PBLK_CMASK(sbi, last_pblk), |
| 2553 |
sbi->s_cluster_ratio, flags); |
| 2554 |
if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER) |
| 2555 |
ext4_rereserve_cluster(inode, to); |
| 2556 |
partial->state = initial; |
| 2557 |
flags = get_default_free_blocks_flags(inode); |
| 2558 |
} |
| 2559 |
|
| 2560 |
flags |= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER; |
| 2561 |
|
| 2562 |
/* |
| 2563 |
* For bigalloc file systems, we never free a partial cluster |
| 2564 |
* at the beginning of the extent. Instead, we check to see if we |
| 2565 |
* need to free it on a subsequent call to ext4_remove_blocks, |
| 2566 |
* or at the end of ext4_ext_rm_leaf or ext4_ext_remove_space. |
| 2567 |
*/ |
| 2568 |
flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER; |
| 2569 |
ext4_free_blocks(handle, inode, NULL, pblk, num, flags); |
| 2570 |
|
| 2571 |
/* reset the partial cluster if we've freed past it */ |
| 2572 |
if (partial->state != initial && partial->pclu != EXT4_B2C(sbi, pblk)) |
| 2573 |
partial->state = initial; |
| 2574 |
|
| 2575 |
/* |
| 2576 |
* If we've freed the entire extent but the beginning is not left |
| 2577 |
* cluster aligned and is not marked as ineligible for freeing we |
| 2578 |
* record the partial cluster at the beginning of the extent. It |
| 2579 |
* wasn't freed by the preceding ext4_free_blocks() call, and we |
| 2580 |
* need to look farther to the left to determine if it's to be freed |
| 2581 |
* (not shared with another extent). Else, reset the partial |
| 2582 |
* cluster - we're either done freeing or the beginning of the |
| 2583 |
* extent is left cluster aligned. |
| 2584 |
*/ |
| 2585 |
if (EXT4_LBLK_COFF(sbi, from) && num == ee_len) { |
| 2586 |
if (partial->state == initial) { |
| 2587 |
partial->pclu = EXT4_B2C(sbi, pblk); |
| 2588 |
partial->lblk = from; |
| 2589 |
partial->state = tofree; |
| 2590 |
} |
| 2591 |
} else { |
| 2592 |
partial->state = initial; |
| 2593 |
} |
| 2594 |
|
| 2595 |
return 0; |
| 2596 |
} |
| 2597 |
|
| 2598 |
/* |
| 2599 |
* ext4_ext_rm_leaf() Removes the extents associated with the |
| 2600 |
* blocks appearing between "start" and "end". Both "start" |
| 2601 |
* and "end" must appear in the same extent or EIO is returned. |
| 2602 |
* |
| 2603 |
* @handle: The journal handle |
| 2604 |
* @inode: The files inode |
| 2605 |
* @path: The path to the leaf |
| 2606 |
* @partial_cluster: The cluster which we'll have to free if all extents |
| 2607 |
* has been released from it. However, if this value is |
| 2608 |
* negative, it's a cluster just to the right of the |
| 2609 |
* punched region and it must not be freed. |
| 2610 |
* @start: The first block to remove |
| 2611 |
* @end: The last block to remove |
| 2612 |
*/ |
| 2613 |
static int |
| 2614 |
ext4_ext_rm_leaf(handle_t *handle, struct inode *inode, |
| 2615 |
struct ext4_ext_path *path, |
| 2616 |
struct partial_cluster *partial, |
| 2617 |
ext4_lblk_t start, ext4_lblk_t end) |
| 2618 |
{ |
| 2619 |
struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
| 2620 |
int err = 0, correct_index = 0; |
| 2621 |
int depth = ext_depth(inode), credits, revoke_credits; |
| 2622 |
struct ext4_extent_header *eh; |
| 2623 |
ext4_lblk_t a, b; |
| 2624 |
unsigned num; |
| 2625 |
ext4_lblk_t ex_ee_block; |
| 2626 |
unsigned short ex_ee_len; |
| 2627 |
unsigned unwritten = 0; |
| 2628 |
struct ext4_extent *ex; |
| 2629 |
ext4_fsblk_t pblk; |
| 2630 |
|
| 2631 |
/* the header must be checked already in ext4_ext_remove_space() */ |
| 2632 |
ext_debug(inode, "truncate since %u in leaf to %u\n", start, end); |
| 2633 |
if (!path[depth].p_hdr) |
| 2634 |
path[depth].p_hdr = ext_block_hdr(path[depth].p_bh); |
| 2635 |
eh = path[depth].p_hdr; |
| 2636 |
if (unlikely(path[depth].p_hdr == NULL)) { |
| 2637 |
EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth); |
| 2638 |
return -EFSCORRUPTED; |
| 2639 |
} |
| 2640 |
/* find where to start removing */ |
| 2641 |
ex = path[depth].p_ext; |
| 2642 |
if (!ex) |
| 2643 |
ex = EXT_LAST_EXTENT(eh); |
| 2644 |
|
| 2645 |
ex_ee_block = le32_to_cpu(ex->ee_block); |
| 2646 |
ex_ee_len = ext4_ext_get_actual_len(ex); |
| 2647 |
|
| 2648 |
trace_ext4_ext_rm_leaf(inode, start, ex, partial); |
| 2649 |
|
| 2650 |
while (ex >= EXT_FIRST_EXTENT(eh) && |
| 2651 |
ex_ee_block + ex_ee_len > start) { |
| 2652 |
|
| 2653 |
if (ext4_ext_is_unwritten(ex)) |
| 2654 |
unwritten = 1; |
| 2655 |
else |
| 2656 |
unwritten = 0; |
| 2657 |
|
| 2658 |
ext_debug(inode, "remove ext %u:[%d]%d\n", ex_ee_block, |
| 2659 |
unwritten, ex_ee_len); |
| 2660 |
path[depth].p_ext = ex; |
| 2661 |
|
| 2662 |
a = max(ex_ee_block, start); |
| 2663 |
b = min(ex_ee_block + ex_ee_len - 1, end); |
| 2664 |
|
| 2665 |
ext_debug(inode, " border %u:%u\n", a, b); |
| 2666 |
|
| 2667 |
/* If this extent is beyond the end of the hole, skip it */ |
| 2668 |
if (end < ex_ee_block) { |
| 2669 |
/* |
| 2670 |
* We're going to skip this extent and move to another, |
| 2671 |
* so note that its first cluster is in use to avoid |
| 2672 |
* freeing it when removing blocks. Eventually, the |
| 2673 |
* right edge of the truncated/punched region will |
| 2674 |
* be just to the left. |
| 2675 |
*/ |
| 2676 |
if (sbi->s_cluster_ratio > 1) { |
| 2677 |
pblk = ext4_ext_pblock(ex); |
| 2678 |
partial->pclu = EXT4_B2C(sbi, pblk); |
| 2679 |
partial->state = nofree; |
| 2680 |
} |
| 2681 |
ex--; |
| 2682 |
ex_ee_block = le32_to_cpu(ex->ee_block); |
| 2683 |
ex_ee_len = ext4_ext_get_actual_len(ex); |
| 2684 |
continue; |
| 2685 |
} else if (b != ex_ee_block + ex_ee_len - 1) { |
| 2686 |
EXT4_ERROR_INODE(inode, |
| 2687 |
"can not handle truncate %u:%u " |
| 2688 |
"on extent %u:%u", |
| 2689 |
start, end, ex_ee_block, |
| 2690 |
ex_ee_block + ex_ee_len - 1); |
| 2691 |
err = -EFSCORRUPTED; |
| 2692 |
goto out; |
| 2693 |
} else if (a != ex_ee_block) { |
| 2694 |
/* remove tail of the extent */ |
| 2695 |
num = a - ex_ee_block; |
| 2696 |
} else { |
| 2697 |
/* remove whole extent: excellent! */ |
| 2698 |
num = 0; |
| 2699 |
} |
| 2700 |
/* |
| 2701 |
* 3 for leaf, sb, and inode plus 2 (bmap and group |
| 2702 |
* descriptor) for each block group; assume two block |
| 2703 |
* groups plus ex_ee_len/blocks_per_block_group for |
| 2704 |
* the worst case |
| 2705 |
*/ |
| 2706 |
credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb)); |
| 2707 |
if (ex == EXT_FIRST_EXTENT(eh)) { |
| 2708 |
correct_index = 1; |
| 2709 |
credits += (ext_depth(inode)) + 1; |
| 2710 |
} |
| 2711 |
credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb); |
| 2712 |
/* |
| 2713 |
* We may end up freeing some index blocks and data from the |
| 2714 |
* punched range. Note that partial clusters are accounted for |
| 2715 |
* by ext4_free_data_revoke_credits(). |
| 2716 |
*/ |
| 2717 |
revoke_credits = |
| 2718 |
ext4_free_metadata_revoke_credits(inode->i_sb, |
| 2719 |
ext_depth(inode)) + |
| 2720 |
ext4_free_data_revoke_credits(inode, b - a + 1); |
| 2721 |
|
| 2722 |
err = ext4_datasem_ensure_credits(handle, inode, credits, |
| 2723 |
credits, revoke_credits); |
| 2724 |
if (err) { |
| 2725 |
if (err > 0) |
| 2726 |
err = -EAGAIN; |
| 2727 |
goto out; |
| 2728 |
} |
| 2729 |
|
| 2730 |
err = ext4_ext_get_access(handle, inode, path + depth); |
| 2731 |
if (err) |
| 2732 |
goto out; |
| 2733 |
|
| 2734 |
err = ext4_remove_blocks(handle, inode, ex, partial, a, b); |
| 2735 |
if (err) |
| 2736 |
goto out; |
| 2737 |
|
| 2738 |
if (num == 0) |
| 2739 |
/* this extent is removed; mark slot entirely unused */ |
| 2740 |
ext4_ext_store_pblock(ex, 0); |
| 2741 |
|
| 2742 |
ex->ee_len = cpu_to_le16(num); |
| 2743 |
/* |
| 2744 |
* Do not mark unwritten if all the blocks in the |
| 2745 |
* extent have been removed. |
| 2746 |
*/ |
| 2747 |
if (unwritten && num) |
| 2748 |
ext4_ext_mark_unwritten(ex); |
| 2749 |
/* |
| 2750 |
* If the extent was completely released, |
| 2751 |
* we need to remove it from the leaf |
| 2752 |
*/ |
| 2753 |
if (num == 0) { |
| 2754 |
if (end != EXT_MAX_BLOCKS - 1) { |
| 2755 |
/* |
| 2756 |
* For hole punching, we need to scoot all the |
| 2757 |
* extents up when an extent is removed so that |
| 2758 |
* we dont have blank extents in the middle |
| 2759 |
*/ |
| 2760 |
memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) * |
| 2761 |
sizeof(struct ext4_extent)); |
| 2762 |
|
| 2763 |
/* Now get rid of the one at the end */ |
| 2764 |
memset(EXT_LAST_EXTENT(eh), 0, |
| 2765 |
sizeof(struct ext4_extent)); |
| 2766 |
} |
| 2767 |
le16_add_cpu(&eh->eh_entries, -1); |
| 2768 |
} |
| 2769 |
|
| 2770 |
err = ext4_ext_dirty(handle, inode, path + depth); |
| 2771 |
if (err) |
| 2772 |
goto out; |
| 2773 |
|
| 2774 |
ext_debug(inode, "new extent: %u:%u:%llu\n", ex_ee_block, num, |
| 2775 |
ext4_ext_pblock(ex)); |
| 2776 |
ex--; |
| 2777 |
ex_ee_block = le32_to_cpu(ex->ee_block); |
| 2778 |
ex_ee_len = ext4_ext_get_actual_len(ex); |
| 2779 |
} |
| 2780 |
|
| 2781 |
if (correct_index && eh->eh_entries) |
| 2782 |
err = ext4_ext_correct_indexes(handle, inode, path); |
| 2783 |
|
| 2784 |
/* |
| 2785 |
* If there's a partial cluster and at least one extent remains in |
| 2786 |
* the leaf, free the partial cluster if it isn't shared with the |
| 2787 |
* current extent. If it is shared with the current extent |
| 2788 |
* we reset the partial cluster because we've reached the start of the |
| 2789 |
* truncated/punched region and we're done removing blocks. |
| 2790 |
*/ |
| 2791 |
if (partial->state == tofree && ex >= EXT_FIRST_EXTENT(eh)) { |
| 2792 |
pblk = ext4_ext_pblock(ex) + ex_ee_len - 1; |
| 2793 |
if (partial->pclu != EXT4_B2C(sbi, pblk)) { |
| 2794 |
int flags = get_default_free_blocks_flags(inode); |
| 2795 |
|
| 2796 |
if (ext4_is_pending(inode, partial->lblk)) |
| 2797 |
flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER; |
| 2798 |
ext4_free_blocks(handle, inode, NULL, |
| 2799 |
EXT4_C2B(sbi, partial->pclu), |
| 2800 |
sbi->s_cluster_ratio, flags); |
| 2801 |
if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER) |
| 2802 |
ext4_rereserve_cluster(inode, partial->lblk); |
| 2803 |
} |
| 2804 |
partial->state = initial; |
| 2805 |
} |
| 2806 |
|
| 2807 |
/* if this leaf is free, then we should |
| 2808 |
* remove it from index block above */ |
| 2809 |
if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL) |
| 2810 |
err = ext4_ext_rm_idx(handle, inode, path, depth); |
| 2811 |
|
| 2812 |
out: |
| 2813 |
return err; |
| 2814 |
} |
| 2815 |
|
| 2816 |
/* |
| 2817 |
* ext4_ext_more_to_rm: |
| 2818 |
* returns 1 if current index has to be freed (even partial) |
| 2819 |
*/ |
| 2820 |
static int |
| 2821 |
ext4_ext_more_to_rm(struct ext4_ext_path *path) |
| 2822 |
{ |
| 2823 |
BUG_ON(path->p_idx == NULL); |
| 2824 |
|
| 2825 |
if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr)) |
| 2826 |
return 0; |
| 2827 |
|
| 2828 |
/* |
| 2829 |
* if truncate on deeper level happened, it wasn't partial, |
| 2830 |
* so we have to consider current index for truncation |
| 2831 |
*/ |
| 2832 |
if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block) |
| 2833 |
return 0; |
| 2834 |
return 1; |
| 2835 |
} |
| 2836 |
|
| 2837 |
int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start, |
| 2838 |
ext4_lblk_t end) |
| 2839 |
{ |
| 2840 |
struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
| 2841 |
int depth = ext_depth(inode); |
| 2842 |
struct ext4_ext_path *path = NULL; |
| 2843 |
struct partial_cluster partial; |
| 2844 |
handle_t *handle; |
| 2845 |
int i = 0, err = 0; |
| 2846 |
int flags = EXT4_EX_NOCACHE | EXT4_EX_NOFAIL; |
| 2847 |
|
| 2848 |
partial.pclu = 0; |
| 2849 |
partial.lblk = 0; |
| 2850 |
partial.state = initial; |
| 2851 |
|
| 2852 |
ext_debug(inode, "truncate since %u to %u\n", start, end); |
| 2853 |
|
| 2854 |
/* probably first extent we're gonna free will be last in block */ |
| 2855 |
handle = ext4_journal_start_with_revoke(inode, EXT4_HT_TRUNCATE, |
| 2856 |
depth + 1, |
| 2857 |
ext4_free_metadata_revoke_credits(inode->i_sb, depth)); |
| 2858 |
if (IS_ERR(handle)) |
| 2859 |
return PTR_ERR(handle); |
| 2860 |
|
| 2861 |
again: |
| 2862 |
trace_ext4_ext_remove_space(inode, start, end, depth); |
| 2863 |
|
| 2864 |
/* |
| 2865 |
* Check if we are removing extents inside the extent tree. If that |
| 2866 |
* is the case, we are going to punch a hole inside the extent tree |
| 2867 |
* so we have to check whether we need to split the extent covering |
| 2868 |
* the last block to remove so we can easily remove the part of it |
| 2869 |
* in ext4_ext_rm_leaf(). |
| 2870 |
*/ |
| 2871 |
if (end < EXT_MAX_BLOCKS - 1) { |
| 2872 |
struct ext4_extent *ex; |
| 2873 |
ext4_lblk_t ee_block, ex_end, lblk; |
| 2874 |
ext4_fsblk_t pblk; |
| 2875 |
|
| 2876 |
/* find extent for or closest extent to this block */ |
| 2877 |
path = ext4_find_extent(inode, end, NULL, flags); |
| 2878 |
if (IS_ERR(path)) { |
| 2879 |
ext4_journal_stop(handle); |
| 2880 |
return PTR_ERR(path); |
| 2881 |
} |
| 2882 |
depth = ext_depth(inode); |
| 2883 |
/* Leaf not may not exist only if inode has no blocks at all */ |
| 2884 |
ex = path[depth].p_ext; |
| 2885 |
if (!ex) { |
| 2886 |
if (depth) { |
| 2887 |
EXT4_ERROR_INODE(inode, |
| 2888 |
"path[%d].p_hdr == NULL", |
| 2889 |
depth); |
| 2890 |
err = -EFSCORRUPTED; |
| 2891 |
} |
| 2892 |
goto out; |
| 2893 |
} |
| 2894 |
|
| 2895 |
ee_block = le32_to_cpu(ex->ee_block); |
| 2896 |
ex_end = ee_block + ext4_ext_get_actual_len(ex) - 1; |
| 2897 |
|
| 2898 |
/* |
| 2899 |
* See if the last block is inside the extent, if so split |
| 2900 |
* the extent at 'end' block so we can easily remove the |
| 2901 |
* tail of the first part of the split extent in |
| 2902 |
* ext4_ext_rm_leaf(). |
| 2903 |
*/ |
| 2904 |
if (end >= ee_block && end < ex_end) { |
| 2905 |
|
| 2906 |
/* |
| 2907 |
* If we're going to split the extent, note that |
| 2908 |
* the cluster containing the block after 'end' is |
| 2909 |
* in use to avoid freeing it when removing blocks. |
| 2910 |
*/ |
| 2911 |
if (sbi->s_cluster_ratio > 1) { |
| 2912 |
pblk = ext4_ext_pblock(ex) + end - ee_block + 1; |
| 2913 |
partial.pclu = EXT4_B2C(sbi, pblk); |
| 2914 |
partial.state = nofree; |
| 2915 |
} |
| 2916 |
|
| 2917 |
/* |
| 2918 |
* Split the extent in two so that 'end' is the last |
| 2919 |
* block in the first new extent. Also we should not |
| 2920 |
* fail removing space due to ENOSPC so try to use |
| 2921 |
* reserved block if that happens. |
| 2922 |
*/ |
| 2923 |
path = ext4_force_split_extent_at(handle, inode, path, |
| 2924 |
end + 1, 1); |
| 2925 |
if (IS_ERR(path)) { |
| 2926 |
err = PTR_ERR(path); |
| 2927 |
goto out; |
| 2928 |
} |
| 2929 |
} else if (sbi->s_cluster_ratio > 1 && end >= ex_end && |
| 2930 |
partial.state == initial) { |
| 2931 |
/* |
| 2932 |
* If we're punching, there's an extent to the right. |
| 2933 |
* If the partial cluster hasn't been set, set it to |
| 2934 |
* that extent's first cluster and its state to nofree |
| 2935 |
* so it won't be freed should it contain blocks to be |
| 2936 |
* removed. If it's already set (tofree/nofree), we're |
| 2937 |
* retrying and keep the original partial cluster info |
| 2938 |
* so a cluster marked tofree as a result of earlier |
| 2939 |
* extent removal is not lost. |
| 2940 |
*/ |
| 2941 |
lblk = ex_end + 1; |
| 2942 |
err = ext4_ext_search_right(inode, path, &lblk, &pblk, |
| 2943 |
NULL, flags); |
| 2944 |
if (err < 0) |
| 2945 |
goto out; |
| 2946 |
if (pblk) { |
| 2947 |
partial.pclu = EXT4_B2C(sbi, pblk); |
| 2948 |
partial.state = nofree; |
| 2949 |
} |
| 2950 |
} |
| 2951 |
} |
| 2952 |
/* |
| 2953 |
* We start scanning from right side, freeing all the blocks |
| 2954 |
* after i_size and walking into the tree depth-wise. |
| 2955 |
*/ |
| 2956 |
depth = ext_depth(inode); |
| 2957 |
if (path) { |
| 2958 |
int k = i = depth; |
| 2959 |
while (--k > 0) |
| 2960 |
path[k].p_block = |
| 2961 |
le16_to_cpu(path[k].p_hdr->eh_entries)+1; |
| 2962 |
} else { |
| 2963 |
path = kzalloc_objs(struct ext4_ext_path, depth + 1, |
| 2964 |
GFP_NOFS | __GFP_NOFAIL); |
| 2965 |
path[0].p_maxdepth = path[0].p_depth = depth; |
| 2966 |
path[0].p_hdr = ext_inode_hdr(inode); |
| 2967 |
i = 0; |
| 2968 |
|
| 2969 |
if (ext4_ext_check(inode, path[0].p_hdr, depth, 0)) { |
| 2970 |
err = -EFSCORRUPTED; |
| 2971 |
goto out; |
| 2972 |
} |
| 2973 |
} |
| 2974 |
err = 0; |
| 2975 |
|
| 2976 |
while (i >= 0 && err == 0) { |
| 2977 |
if (i == depth) { |
| 2978 |
/* this is leaf block */ |
| 2979 |
err = ext4_ext_rm_leaf(handle, inode, path, |
| 2980 |
&partial, start, end); |
| 2981 |
/* root level has p_bh == NULL, brelse() eats this */ |
| 2982 |
ext4_ext_path_brelse(path + i); |
| 2983 |
i--; |
| 2984 |
continue; |
| 2985 |
} |
| 2986 |
|
| 2987 |
/* this is index block */ |
| 2988 |
if (!path[i].p_hdr) { |
| 2989 |
ext_debug(inode, "initialize header\n"); |
| 2990 |
path[i].p_hdr = ext_block_hdr(path[i].p_bh); |
| 2991 |
} |
| 2992 |
|
| 2993 |
if (!path[i].p_idx) { |
| 2994 |
/* this level hasn't been touched yet */ |
| 2995 |
path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr); |
| 2996 |
path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1; |
| 2997 |
ext_debug(inode, "init index ptr: hdr 0x%p, num %d\n", |
| 2998 |
path[i].p_hdr, |
| 2999 |
le16_to_cpu(path[i].p_hdr->eh_entries)); |
| 3000 |
} else { |
| 3001 |
/* we were already here, see at next index */ |
| 3002 |
path[i].p_idx--; |
| 3003 |
} |
| 3004 |
|
| 3005 |
ext_debug(inode, "level %d - index, first 0x%p, cur 0x%p\n", |
| 3006 |
i, EXT_FIRST_INDEX(path[i].p_hdr), |
| 3007 |
path[i].p_idx); |
| 3008 |
if (ext4_ext_more_to_rm(path + i)) { |
| 3009 |
struct buffer_head *bh; |
| 3010 |
/* go to the next level */ |
| 3011 |
ext_debug(inode, "move to level %d (block %llu)\n", |
| 3012 |
i + 1, ext4_idx_pblock(path[i].p_idx)); |
| 3013 |
memset(path + i + 1, 0, sizeof(*path)); |
| 3014 |
bh = read_extent_tree_block(inode, path[i].p_idx, |
| 3015 |
depth - i - 1, flags); |
| 3016 |
if (IS_ERR(bh)) { |
| 3017 |
/* should we reset i_size? */ |
| 3018 |
err = PTR_ERR(bh); |
| 3019 |
break; |
| 3020 |
} |
| 3021 |
/* Yield here to deal with large extent trees. |
| 3022 |
* Should be a no-op if we did IO above. */ |
| 3023 |
cond_resched(); |
| 3024 |
if (WARN_ON(i + 1 > depth)) { |
| 3025 |
err = -EFSCORRUPTED; |
| 3026 |
break; |
| 3027 |
} |
| 3028 |
path[i + 1].p_bh = bh; |
| 3029 |
|
| 3030 |
/* save actual number of indexes since this |
| 3031 |
* number is changed at the next iteration */ |
| 3032 |
path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries); |
| 3033 |
i++; |
| 3034 |
} else { |
| 3035 |
/* we finished processing this index, go up */ |
| 3036 |
if (path[i].p_hdr->eh_entries == 0 && i > 0) { |
| 3037 |
/* index is empty, remove it; |
| 3038 |
* handle must be already prepared by the |
| 3039 |
* truncatei_leaf() */ |
| 3040 |
err = ext4_ext_rm_idx(handle, inode, path, i); |
| 3041 |
} |
| 3042 |
/* root level has p_bh == NULL, brelse() eats this */ |
| 3043 |
ext4_ext_path_brelse(path + i); |
| 3044 |
i--; |
| 3045 |
ext_debug(inode, "return to level %d\n", i); |
| 3046 |
} |
| 3047 |
} |
| 3048 |
|
| 3049 |
trace_ext4_ext_remove_space_done(inode, start, end, depth, &partial, |
| 3050 |
path->p_hdr->eh_entries); |
| 3051 |
|
| 3052 |
/* |
| 3053 |
* if there's a partial cluster and we have removed the first extent |
| 3054 |
* in the file, then we also free the partial cluster, if any |
| 3055 |
*/ |
| 3056 |
if (partial.state == tofree && err == 0) { |
| 3057 |
int flags = get_default_free_blocks_flags(inode); |
| 3058 |
|
| 3059 |
if (ext4_is_pending(inode, partial.lblk)) |
| 3060 |
flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER; |
| 3061 |
ext4_free_blocks(handle, inode, NULL, |
| 3062 |
EXT4_C2B(sbi, partial.pclu), |
| 3063 |
sbi->s_cluster_ratio, flags); |
| 3064 |
if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER) |
| 3065 |
ext4_rereserve_cluster(inode, partial.lblk); |
| 3066 |
partial.state = initial; |
| 3067 |
} |
| 3068 |
|
| 3069 |
/* TODO: flexible tree reduction should be here */ |
| 3070 |
if (path->p_hdr->eh_entries == 0) { |
| 3071 |
/* |
| 3072 |
* truncate to zero freed all the tree, |
| 3073 |
* so we need to correct eh_depth |
| 3074 |
*/ |
| 3075 |
err = ext4_ext_get_access(handle, inode, path); |
| 3076 |
if (err == 0) { |
| 3077 |
ext_inode_hdr(inode)->eh_depth = 0; |
| 3078 |
ext_inode_hdr(inode)->eh_max = |
| 3079 |
cpu_to_le16(ext4_ext_space_root(inode, 0)); |
| 3080 |
err = ext4_ext_dirty(handle, inode, path); |
| 3081 |
} |
| 3082 |
} |
| 3083 |
out: |
| 3084 |
ext4_free_ext_path(path); |
| 3085 |
path = NULL; |
| 3086 |
if (err == -EAGAIN) |
| 3087 |
goto again; |
| 3088 |
ext4_journal_stop(handle); |
| 3089 |
|
| 3090 |
return err; |
| 3091 |
} |
| 3092 |
|
| 3093 |
/* |
| 3094 |
* called at mount time |
| 3095 |
*/ |
| 3096 |
void ext4_ext_init(struct super_block *sb) |
| 3097 |
{ |
| 3098 |
/* |
| 3099 |
* possible initialization would be here |
| 3100 |
*/ |
| 3101 |
|
| 3102 |
if (ext4_has_feature_extents(sb)) { |
| 3103 |
#if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS) |
| 3104 |
printk(KERN_INFO "EXT4-fs: file extents enabled" |
| 3105 |
#ifdef AGGRESSIVE_TEST |
| 3106 |
", aggressive tests" |
| 3107 |
#endif |
| 3108 |
#ifdef CHECK_BINSEARCH |
| 3109 |
", check binsearch" |
| 3110 |
#endif |
| 3111 |
#ifdef EXTENTS_STATS |
| 3112 |
", stats" |
| 3113 |
#endif |
| 3114 |
"\n"); |
| 3115 |
#endif |
| 3116 |
#ifdef EXTENTS_STATS |
| 3117 |
spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock); |
| 3118 |
EXT4_SB(sb)->s_ext_min = 1 << 30; |
| 3119 |
EXT4_SB(sb)->s_ext_max = 0; |
| 3120 |
#endif |
| 3121 |
} |
| 3122 |
} |
| 3123 |
|
| 3124 |
/* |
| 3125 |
* called at umount time |
| 3126 |
*/ |
| 3127 |
void ext4_ext_release(struct super_block *sb) |
| 3128 |
{ |
| 3129 |
if (!ext4_has_feature_extents(sb)) |
| 3130 |
return; |
| 3131 |
|
| 3132 |
#ifdef EXTENTS_STATS |
| 3133 |
if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) { |
| 3134 |
struct ext4_sb_info *sbi = EXT4_SB(sb); |
| 3135 |
printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n", |
| 3136 |
sbi->s_ext_blocks, sbi->s_ext_extents, |
| 3137 |
sbi->s_ext_blocks / sbi->s_ext_extents); |
| 3138 |
printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n", |
| 3139 |
sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max); |
| 3140 |
} |
| 3141 |
#endif |
| 3142 |
} |
| 3143 |
|
| 3144 |
static void ext4_zeroout_es(struct inode *inode, struct ext4_extent *ex) |
| 3145 |
{ |
| 3146 |
ext4_lblk_t ee_block; |
| 3147 |
ext4_fsblk_t ee_pblock; |
| 3148 |
unsigned int ee_len; |
| 3149 |
|
| 3150 |
ee_block = le32_to_cpu(ex->ee_block); |
| 3151 |
ee_len = ext4_ext_get_actual_len(ex); |
| 3152 |
ee_pblock = ext4_ext_pblock(ex); |
| 3153 |
|
| 3154 |
if (ee_len == 0) |
| 3155 |
return; |
| 3156 |
|
| 3157 |
ext4_es_insert_extent(inode, ee_block, ee_len, ee_pblock, |
| 3158 |
EXTENT_STATUS_WRITTEN, false); |
| 3159 |
} |
| 3160 |
|
| 3161 |
/* FIXME!! we need to try to merge to left or right after zero-out */ |
| 3162 |
int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex) |
| 3163 |
{ |
| 3164 |
ext4_fsblk_t ee_pblock; |
| 3165 |
unsigned int ee_len; |
| 3166 |
|
| 3167 |
KUNIT_STATIC_STUB_REDIRECT(ext4_ext_zeroout, inode, ex); |
| 3168 |
|
| 3169 |
ee_len = ext4_ext_get_actual_len(ex); |
| 3170 |
ee_pblock = ext4_ext_pblock(ex); |
| 3171 |
return ext4_issue_zeroout(inode, le32_to_cpu(ex->ee_block), ee_pblock, |
| 3172 |
ee_len); |
| 3173 |
} |
| 3174 |
|
| 3175 |
/* |
| 3176 |
* ext4_split_extent_at() splits an extent at given block. |
| 3177 |
* |
| 3178 |
* @handle: the journal handle |
| 3179 |
* @inode: the file inode |
| 3180 |
* @path: the path to the extent |
| 3181 |
* @split: the logical block where the extent is splitted. |
| 3182 |
* @flags: flags used to insert new extent to extent tree. |
| 3183 |
* |
| 3184 |
* |
| 3185 |
* Splits extent [a, b] into two extents [a, @split) and [@split, b], states |
| 3186 |
* of which are same as the original extent. No conversion is performed. |
| 3187 |
* |
| 3188 |
* Return an extent path pointer on success, or an error pointer on failure. On |
| 3189 |
* failure, the extent is restored to original state. |
| 3190 |
*/ |
| 3191 |
static struct ext4_ext_path *ext4_split_extent_at(handle_t *handle, |
| 3192 |
struct inode *inode, |
| 3193 |
struct ext4_ext_path *path, |
| 3194 |
ext4_lblk_t split, |
| 3195 |
int flags) |
| 3196 |
{ |
| 3197 |
ext4_fsblk_t newblock; |
| 3198 |
ext4_lblk_t ee_block; |
| 3199 |
struct ext4_extent *ex, newex, orig_ex; |
| 3200 |
struct ext4_extent *ex2 = NULL; |
| 3201 |
unsigned int ee_len, depth; |
| 3202 |
int err = 0, insert_err = 0, is_unwrit = 0; |
| 3203 |
|
| 3204 |
/* Do not cache extents that are in the process of being modified. */ |
| 3205 |
flags |= EXT4_EX_NOCACHE; |
| 3206 |
|
| 3207 |
ext_debug(inode, "logical block %llu\n", (unsigned long long)split); |
| 3208 |
|
| 3209 |
ext4_ext_show_leaf(inode, path); |
| 3210 |
|
| 3211 |
depth = ext_depth(inode); |
| 3212 |
ex = path[depth].p_ext; |
| 3213 |
ee_block = le32_to_cpu(ex->ee_block); |
| 3214 |
ee_len = ext4_ext_get_actual_len(ex); |
| 3215 |
newblock = split - ee_block + ext4_ext_pblock(ex); |
| 3216 |
is_unwrit = ext4_ext_is_unwritten(ex); |
| 3217 |
|
| 3218 |
BUG_ON(split < ee_block || split >= (ee_block + ee_len)); |
| 3219 |
|
| 3220 |
/* |
| 3221 |
* No split needed |
| 3222 |
*/ |
| 3223 |
if (split == ee_block) |
| 3224 |
goto out; |
| 3225 |
|
| 3226 |
err = ext4_ext_get_access(handle, inode, path + depth); |
| 3227 |
if (err) |
| 3228 |
goto out; |
| 3229 |
|
| 3230 |
/* case a */ |
| 3231 |
memcpy(&orig_ex, ex, sizeof(orig_ex)); |
| 3232 |
ex->ee_len = cpu_to_le16(split - ee_block); |
| 3233 |
if (is_unwrit) |
| 3234 |
ext4_ext_mark_unwritten(ex); |
| 3235 |
|
| 3236 |
/* |
| 3237 |
* path may lead to new leaf, not to original leaf any more |
| 3238 |
* after ext4_ext_insert_extent() returns, |
| 3239 |
*/ |
| 3240 |
err = ext4_ext_dirty(handle, inode, path + depth); |
| 3241 |
if (err) |
| 3242 |
goto fix_extent_len; |
| 3243 |
|
| 3244 |
ex2 = &newex; |
| 3245 |
ex2->ee_block = cpu_to_le32(split); |
| 3246 |
ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block)); |
| 3247 |
ext4_ext_store_pblock(ex2, newblock); |
| 3248 |
if (is_unwrit) |
| 3249 |
ext4_ext_mark_unwritten(ex2); |
| 3250 |
|
| 3251 |
path = ext4_ext_insert_extent(handle, inode, path, &newex, flags); |
| 3252 |
if (!IS_ERR(path)) |
| 3253 |
return path; |
| 3254 |
|
| 3255 |
insert_err = PTR_ERR(path); |
| 3256 |
err = 0; |
| 3257 |
if (insert_err != -ENOSPC && insert_err != -EDQUOT && |
| 3258 |
insert_err != -ENOMEM) |
| 3259 |
goto out_path; |
| 3260 |
|
| 3261 |
/* |
| 3262 |
* Get a new path to try to zeroout or fix the extent length. |
| 3263 |
* Using EXT4_EX_NOFAIL guarantees that ext4_find_extent() |
| 3264 |
* will not return -ENOMEM, otherwise -ENOMEM will cause a |
| 3265 |
* retry in do_writepages(), and a WARN_ON may be triggered |
| 3266 |
* in ext4_da_update_reserve_space() due to an incorrect |
| 3267 |
* ee_len causing the i_reserved_data_blocks exception. |
| 3268 |
*/ |
| 3269 |
path = ext4_find_extent(inode, ee_block, NULL, flags | EXT4_EX_NOFAIL); |
| 3270 |
if (IS_ERR(path)) { |
| 3271 |
EXT4_ERROR_INODE(inode, "Failed split extent on %u, err %ld", |
| 3272 |
split, PTR_ERR(path)); |
| 3273 |
goto out_path; |
| 3274 |
} |
| 3275 |
|
| 3276 |
depth = ext_depth(inode); |
| 3277 |
ex = path[depth].p_ext; |
| 3278 |
if (!ex) { |
| 3279 |
EXT4_ERROR_INODE(inode, |
| 3280 |
"bad extent address lblock: %lu, depth: %d pblock %llu", |
| 3281 |
(unsigned long)ee_block, depth, path[depth].p_block); |
| 3282 |
err = -EFSCORRUPTED; |
| 3283 |
goto out; |
| 3284 |
} |
| 3285 |
|
| 3286 |
err = ext4_ext_get_access(handle, inode, path + depth); |
| 3287 |
if (err) |
| 3288 |
goto out; |
| 3289 |
|
| 3290 |
fix_extent_len: |
| 3291 |
ex->ee_len = orig_ex.ee_len; |
| 3292 |
err = ext4_ext_dirty(handle, inode, path + path->p_depth); |
| 3293 |
out: |
| 3294 |
if (err || insert_err) { |
| 3295 |
ext4_free_ext_path(path); |
| 3296 |
path = err ? ERR_PTR(err) : ERR_PTR(insert_err); |
| 3297 |
} |
| 3298 |
out_path: |
| 3299 |
if (IS_ERR(path)) |
| 3300 |
/* Remove all remaining potentially stale extents. */ |
| 3301 |
ext4_es_remove_extent(inode, ee_block, ee_len); |
| 3302 |
ext4_ext_show_leaf(inode, path); |
| 3303 |
return path; |
| 3304 |
} |
| 3305 |
|
| 3306 |
static int ext4_split_extent_zeroout(handle_t *handle, struct inode *inode, |
| 3307 |
struct ext4_ext_path *path, |
| 3308 |
struct ext4_map_blocks *map, int flags) |
| 3309 |
{ |
| 3310 |
struct ext4_extent *ex; |
| 3311 |
unsigned int ee_len, depth; |
| 3312 |
ext4_lblk_t ee_block; |
| 3313 |
uint64_t lblk, pblk, len; |
| 3314 |
int is_unwrit; |
| 3315 |
int err = 0; |
| 3316 |
|
| 3317 |
depth = ext_depth(inode); |
| 3318 |
ex = path[depth].p_ext; |
| 3319 |
ee_block = le32_to_cpu(ex->ee_block); |
| 3320 |
ee_len = ext4_ext_get_actual_len(ex); |
| 3321 |
is_unwrit = ext4_ext_is_unwritten(ex); |
| 3322 |
|
| 3323 |
if (flags & EXT4_GET_BLOCKS_CONVERT) { |
| 3324 |
/* |
| 3325 |
* EXT4_GET_BLOCKS_CONVERT: Caller wants the range specified by |
| 3326 |
* map to be initialized. Zeroout everything except the map |
| 3327 |
* range. |
| 3328 |
*/ |
| 3329 |
|
| 3330 |
loff_t map_end = (loff_t) map->m_lblk + map->m_len; |
| 3331 |
loff_t ex_end = (loff_t) ee_block + ee_len; |
| 3332 |
|
| 3333 |
if (!is_unwrit) |
| 3334 |
/* Shouldn't happen. Just exit */ |
| 3335 |
return -EINVAL; |
| 3336 |
|
| 3337 |
/* zeroout left */ |
| 3338 |
if (map->m_lblk > ee_block) { |
| 3339 |
lblk = ee_block; |
| 3340 |
len = map->m_lblk - ee_block; |
| 3341 |
pblk = ext4_ext_pblock(ex); |
| 3342 |
err = ext4_issue_zeroout(inode, lblk, pblk, len); |
| 3343 |
if (err) |
| 3344 |
/* ZEROOUT failed, just return original error */ |
| 3345 |
return err; |
| 3346 |
} |
| 3347 |
|
| 3348 |
/* zeroout right */ |
| 3349 |
if (map_end < ex_end) { |
| 3350 |
lblk = map_end; |
| 3351 |
len = ex_end - map_end; |
| 3352 |
pblk = ext4_ext_pblock(ex) + (map_end - ee_block); |
| 3353 |
err = ext4_issue_zeroout(inode, lblk, pblk, len); |
| 3354 |
if (err) |
| 3355 |
/* ZEROOUT failed, just return original error */ |
| 3356 |
return err; |
| 3357 |
} |
| 3358 |
} else if (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN) { |
| 3359 |
/* |
| 3360 |
* EXT4_GET_BLOCKS_CONVERT_UNWRITTEN: Caller wants the |
| 3361 |
* range specified by map to be marked unwritten. |
| 3362 |
* Zeroout the map range leaving rest as it is. |
| 3363 |
*/ |
| 3364 |
|
| 3365 |
if (is_unwrit) |
| 3366 |
/* Shouldn't happen. Just exit */ |
| 3367 |
return -EINVAL; |
| 3368 |
|
| 3369 |
lblk = map->m_lblk; |
| 3370 |
len = map->m_len; |
| 3371 |
pblk = ext4_ext_pblock(ex) + (map->m_lblk - ee_block); |
| 3372 |
err = ext4_issue_zeroout(inode, lblk, pblk, len); |
| 3373 |
if (err) |
| 3374 |
/* ZEROOUT failed, just return original error */ |
| 3375 |
return err; |
| 3376 |
} else { |
| 3377 |
/* |
| 3378 |
* We no longer perform unwritten to unwritten splits in IO paths. |
| 3379 |
* Hence this should not happen. |
| 3380 |
*/ |
| 3381 |
WARN_ON_ONCE(true); |
| 3382 |
return -EINVAL; |
| 3383 |
} |
| 3384 |
|
| 3385 |
err = ext4_ext_get_access(handle, inode, path + depth); |
| 3386 |
if (err) |
| 3387 |
return err; |
| 3388 |
|
| 3389 |
ext4_ext_mark_initialized(ex); |
| 3390 |
|
| 3391 |
err = ext4_ext_dirty(handle, inode, path + depth); |
| 3392 |
if (err) |
| 3393 |
return err; |
| 3394 |
|
| 3395 |
return 0; |
| 3396 |
} |
| 3397 |
|
| 3398 |
/* |
| 3399 |
* ext4_split_extent() splits an extent and mark extent which is covered |
| 3400 |
* by @map as split_flags indicates |
| 3401 |
* |
| 3402 |
* It may result in splitting the extent into multiple extents (up to three) |
| 3403 |
* There are three possibilities: |
| 3404 |
* a> There is no split required |
| 3405 |
* b> Splits in two extents: Split is happening at either end of the extent |
| 3406 |
* c> Splits in three extents: Somone is splitting in middle of the extent |
| 3407 |
* |
| 3408 |
*/ |
| 3409 |
static struct ext4_ext_path *ext4_split_extent(handle_t *handle, |
| 3410 |
struct inode *inode, |
| 3411 |
struct ext4_ext_path *path, |
| 3412 |
struct ext4_map_blocks *map, |
| 3413 |
int split_flag, int flags, |
| 3414 |
unsigned int *allocated, bool *did_zeroout) |
| 3415 |
{ |
| 3416 |
ext4_lblk_t ee_block, orig_ee_block; |
| 3417 |
struct ext4_extent *ex; |
| 3418 |
unsigned int ee_len, orig_ee_len, depth; |
| 3419 |
int unwritten, orig_unwritten; |
| 3420 |
int orig_err = 0; |
| 3421 |
|
| 3422 |
depth = ext_depth(inode); |
| 3423 |
ex = path[depth].p_ext; |
| 3424 |
ee_block = le32_to_cpu(ex->ee_block); |
| 3425 |
ee_len = ext4_ext_get_actual_len(ex); |
| 3426 |
unwritten = ext4_ext_is_unwritten(ex); |
| 3427 |
|
| 3428 |
orig_ee_block = ee_block; |
| 3429 |
orig_ee_len = ee_len; |
| 3430 |
orig_unwritten = unwritten; |
| 3431 |
|
| 3432 |
/* Do not cache extents that are in the process of being modified. */ |
| 3433 |
flags |= EXT4_EX_NOCACHE; |
| 3434 |
|
| 3435 |
if (map->m_lblk + map->m_len < ee_block + ee_len) { |
| 3436 |
path = ext4_split_extent_at(handle, inode, path, |
| 3437 |
map->m_lblk + map->m_len, flags); |
| 3438 |
if (IS_ERR(path)) |
| 3439 |
goto try_zeroout; |
| 3440 |
|
| 3441 |
/* |
| 3442 |
* Update path is required because previous ext4_split_extent_at |
| 3443 |
* may result in split of original leaf or extent zeroout. |
| 3444 |
*/ |
| 3445 |
path = ext4_find_extent(inode, map->m_lblk, path, flags); |
| 3446 |
if (IS_ERR(path)) |
| 3447 |
goto try_zeroout; |
| 3448 |
|
| 3449 |
depth = ext_depth(inode); |
| 3450 |
ex = path[depth].p_ext; |
| 3451 |
if (!ex) { |
| 3452 |
EXT4_ERROR_INODE(inode, "unexpected hole at %lu", |
| 3453 |
(unsigned long) map->m_lblk); |
| 3454 |
ext4_free_ext_path(path); |
| 3455 |
return ERR_PTR(-EFSCORRUPTED); |
| 3456 |
} |
| 3457 |
|
| 3458 |
/* extent would have changed so update original values */ |
| 3459 |
orig_ee_block = le32_to_cpu(ex->ee_block); |
| 3460 |
orig_ee_len = ext4_ext_get_actual_len(ex); |
| 3461 |
orig_unwritten = ext4_ext_is_unwritten(ex); |
| 3462 |
} |
| 3463 |
|
| 3464 |
if (map->m_lblk >= ee_block) { |
| 3465 |
path = ext4_split_extent_at(handle, inode, path, map->m_lblk, |
| 3466 |
flags); |
| 3467 |
if (IS_ERR(path)) |
| 3468 |
goto try_zeroout; |
| 3469 |
} |
| 3470 |
|
| 3471 |
goto success; |
| 3472 |
|
| 3473 |
try_zeroout: |
| 3474 |
/* |
| 3475 |
* There was an error in splitting the extent. So instead, just zeroout |
| 3476 |
* unwritten portions and convert it to initialized as a last resort. If |
| 3477 |
* there is any failure here we just return the original error |
| 3478 |
*/ |
| 3479 |
|
| 3480 |
orig_err = PTR_ERR(path); |
| 3481 |
if (orig_err != -ENOSPC && orig_err != -EDQUOT && orig_err != -ENOMEM) |
| 3482 |
goto out_orig_err; |
| 3483 |
|
| 3484 |
/* we can't zeroout? just return the original err */ |
| 3485 |
if (!(split_flag & EXT4_EXT_MAY_ZEROOUT)) |
| 3486 |
goto out_orig_err; |
| 3487 |
|
| 3488 |
if (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN) { |
| 3489 |
int max_zeroout_blks = |
| 3490 |
EXT4_SB(inode->i_sb)->s_extent_max_zeroout_kb >> |
| 3491 |
(inode->i_sb->s_blocksize_bits - 10); |
| 3492 |
|
| 3493 |
if (map->m_len > max_zeroout_blks) |
| 3494 |
goto out_orig_err; |
| 3495 |
} |
| 3496 |
|
| 3497 |
path = ext4_find_extent(inode, map->m_lblk, NULL, flags); |
| 3498 |
if (IS_ERR(path)) |
| 3499 |
goto out_orig_err; |
| 3500 |
|
| 3501 |
depth = ext_depth(inode); |
| 3502 |
ex = path[depth].p_ext; |
| 3503 |
ee_block = le32_to_cpu(ex->ee_block); |
| 3504 |
ee_len = ext4_ext_get_actual_len(ex); |
| 3505 |
unwritten = ext4_ext_is_unwritten(ex); |
| 3506 |
|
| 3507 |
/* extent to zeroout should have been unchanged but its not */ |
| 3508 |
if (WARN_ON(ee_block != orig_ee_block || ee_len != orig_ee_len || |
| 3509 |
unwritten != orig_unwritten)) |
| 3510 |
goto out_free_path; |
| 3511 |
|
| 3512 |
if (ext4_split_extent_zeroout(handle, inode, path, map, flags)) |
| 3513 |
goto out_free_path; |
| 3514 |
|
| 3515 |
/* zeroout succeeded */ |
| 3516 |
if (did_zeroout) |
| 3517 |
*did_zeroout = true; |
| 3518 |
|
| 3519 |
success: |
| 3520 |
if (allocated) { |
| 3521 |
if (map->m_lblk + map->m_len > ee_block + ee_len) |
| 3522 |
*allocated = ee_len - (map->m_lblk - ee_block); |
| 3523 |
else |
| 3524 |
*allocated = map->m_len; |
| 3525 |
} |
| 3526 |
ext4_ext_show_leaf(inode, path); |
| 3527 |
return path; |
| 3528 |
|
| 3529 |
out_free_path: |
| 3530 |
ext4_free_ext_path(path); |
| 3531 |
out_orig_err: |
| 3532 |
return ERR_PTR(orig_err); |
| 3533 |
|
| 3534 |
} |
| 3535 |
|
| 3536 |
/* |
| 3537 |
* This function is called by ext4_ext_map_blocks() if someone tries to write |
| 3538 |
* to an unwritten extent. It may result in splitting the unwritten |
| 3539 |
* extent into multiple extents (up to three - one initialized and two |
| 3540 |
* unwritten). |
| 3541 |
* There are three possibilities: |
| 3542 |
* a> There is no split required: Entire extent should be initialized |
| 3543 |
* b> Splits in two extents: Write is happening at either end of the extent |
| 3544 |
* c> Splits in three extents: Somone is writing in middle of the extent |
| 3545 |
* |
| 3546 |
* Pre-conditions: |
| 3547 |
* - The extent pointed to by 'path' is unwritten. |
| 3548 |
* - The extent pointed to by 'path' contains a superset |
| 3549 |
* of the logical span [map->m_lblk, map->m_lblk + map->m_len). |
| 3550 |
* |
| 3551 |
* Post-conditions on success: |
| 3552 |
* - the returned value is the number of blocks beyond map->l_lblk |
| 3553 |
* that are allocated and initialized. |
| 3554 |
* It is guaranteed to be >= map->m_len. |
| 3555 |
*/ |
| 3556 |
static struct ext4_ext_path * |
| 3557 |
ext4_ext_convert_to_initialized(handle_t *handle, struct inode *inode, |
| 3558 |
struct ext4_map_blocks *map, struct ext4_ext_path *path, |
| 3559 |
int flags, unsigned int *allocated) |
| 3560 |
{ |
| 3561 |
struct ext4_sb_info *sbi; |
| 3562 |
struct ext4_extent_header *eh; |
| 3563 |
struct ext4_map_blocks split_map; |
| 3564 |
struct ext4_extent zero_ex1, zero_ex2; |
| 3565 |
struct ext4_extent *ex, *abut_ex; |
| 3566 |
ext4_lblk_t ee_block, eof_block; |
| 3567 |
unsigned int ee_len, depth, map_len = map->m_len; |
| 3568 |
int err = 0; |
| 3569 |
unsigned int max_zeroout = 0; |
| 3570 |
|
| 3571 |
ext_debug(inode, "logical block %llu, max_blocks %u\n", |
| 3572 |
(unsigned long long)map->m_lblk, map_len); |
| 3573 |
|
| 3574 |
sbi = EXT4_SB(inode->i_sb); |
| 3575 |
eof_block = (EXT4_I(inode)->i_disksize + inode->i_sb->s_blocksize - 1) |
| 3576 |
>> inode->i_sb->s_blocksize_bits; |
| 3577 |
if (eof_block < map->m_lblk + map_len) |
| 3578 |
eof_block = map->m_lblk + map_len; |
| 3579 |
|
| 3580 |
depth = ext_depth(inode); |
| 3581 |
eh = path[depth].p_hdr; |
| 3582 |
ex = path[depth].p_ext; |
| 3583 |
ee_block = le32_to_cpu(ex->ee_block); |
| 3584 |
ee_len = ext4_ext_get_actual_len(ex); |
| 3585 |
zero_ex1.ee_len = 0; |
| 3586 |
zero_ex2.ee_len = 0; |
| 3587 |
|
| 3588 |
trace_ext4_ext_convert_to_initialized_enter(inode, map, ex); |
| 3589 |
|
| 3590 |
/* Pre-conditions */ |
| 3591 |
BUG_ON(!ext4_ext_is_unwritten(ex)); |
| 3592 |
BUG_ON(!in_range(map->m_lblk, ee_block, ee_len)); |
| 3593 |
|
| 3594 |
/* |
| 3595 |
* Attempt to transfer newly initialized blocks from the currently |
| 3596 |
* unwritten extent to its neighbor. This is much cheaper |
| 3597 |
* than an insertion followed by a merge as those involve costly |
| 3598 |
* memmove() calls. Transferring to the left is the common case in |
| 3599 |
* steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE) |
| 3600 |
* followed by append writes. |
| 3601 |
* |
| 3602 |
* Limitations of the current logic: |
| 3603 |
* - L1: we do not deal with writes covering the whole extent. |
| 3604 |
* This would require removing the extent if the transfer |
| 3605 |
* is possible. |
| 3606 |
* - L2: we only attempt to merge with an extent stored in the |
| 3607 |
* same extent tree node. |
| 3608 |
*/ |
| 3609 |
*allocated = 0; |
| 3610 |
if ((map->m_lblk == ee_block) && |
| 3611 |
/* See if we can merge left */ |
| 3612 |
(map_len < ee_len) && /*L1*/ |
| 3613 |
(ex > EXT_FIRST_EXTENT(eh))) { /*L2*/ |
| 3614 |
ext4_lblk_t prev_lblk; |
| 3615 |
ext4_fsblk_t prev_pblk, ee_pblk; |
| 3616 |
unsigned int prev_len; |
| 3617 |
|
| 3618 |
abut_ex = ex - 1; |
| 3619 |
prev_lblk = le32_to_cpu(abut_ex->ee_block); |
| 3620 |
prev_len = ext4_ext_get_actual_len(abut_ex); |
| 3621 |
prev_pblk = ext4_ext_pblock(abut_ex); |
| 3622 |
ee_pblk = ext4_ext_pblock(ex); |
| 3623 |
|
| 3624 |
/* |
| 3625 |
* A transfer of blocks from 'ex' to 'abut_ex' is allowed |
| 3626 |
* upon those conditions: |
| 3627 |
* - C1: abut_ex is initialized, |
| 3628 |
* - C2: abut_ex is logically abutting ex, |
| 3629 |
* - C3: abut_ex is physically abutting ex, |
| 3630 |
* - C4: abut_ex can receive the additional blocks without |
| 3631 |
* overflowing the (initialized) length limit. |
| 3632 |
*/ |
| 3633 |
if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/ |
| 3634 |
((prev_lblk + prev_len) == ee_block) && /*C2*/ |
| 3635 |
((prev_pblk + prev_len) == ee_pblk) && /*C3*/ |
| 3636 |
(prev_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/ |
| 3637 |
err = ext4_ext_get_access(handle, inode, path + depth); |
| 3638 |
if (err) |
| 3639 |
goto errout; |
| 3640 |
|
| 3641 |
trace_ext4_ext_convert_to_initialized_fastpath(inode, |
| 3642 |
map, ex, abut_ex); |
| 3643 |
|
| 3644 |
/* Shift the start of ex by 'map_len' blocks */ |
| 3645 |
ex->ee_block = cpu_to_le32(ee_block + map_len); |
| 3646 |
ext4_ext_store_pblock(ex, ee_pblk + map_len); |
| 3647 |
ex->ee_len = cpu_to_le16(ee_len - map_len); |
| 3648 |
ext4_ext_mark_unwritten(ex); /* Restore the flag */ |
| 3649 |
|
| 3650 |
/* Extend abut_ex by 'map_len' blocks */ |
| 3651 |
abut_ex->ee_len = cpu_to_le16(prev_len + map_len); |
| 3652 |
|
| 3653 |
/* Result: number of initialized blocks past m_lblk */ |
| 3654 |
*allocated = map_len; |
| 3655 |
} |
| 3656 |
} else if (((map->m_lblk + map_len) == (ee_block + ee_len)) && |
| 3657 |
(map_len < ee_len) && /*L1*/ |
| 3658 |
ex < EXT_LAST_EXTENT(eh)) { /*L2*/ |
| 3659 |
/* See if we can merge right */ |
| 3660 |
ext4_lblk_t next_lblk; |
| 3661 |
ext4_fsblk_t next_pblk, ee_pblk; |
| 3662 |
unsigned int next_len; |
| 3663 |
|
| 3664 |
abut_ex = ex + 1; |
| 3665 |
next_lblk = le32_to_cpu(abut_ex->ee_block); |
| 3666 |
next_len = ext4_ext_get_actual_len(abut_ex); |
| 3667 |
next_pblk = ext4_ext_pblock(abut_ex); |
| 3668 |
ee_pblk = ext4_ext_pblock(ex); |
| 3669 |
|
| 3670 |
/* |
| 3671 |
* A transfer of blocks from 'ex' to 'abut_ex' is allowed |
| 3672 |
* upon those conditions: |
| 3673 |
* - C1: abut_ex is initialized, |
| 3674 |
* - C2: abut_ex is logically abutting ex, |
| 3675 |
* - C3: abut_ex is physically abutting ex, |
| 3676 |
* - C4: abut_ex can receive the additional blocks without |
| 3677 |
* overflowing the (initialized) length limit. |
| 3678 |
*/ |
| 3679 |
if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/ |
| 3680 |
((map->m_lblk + map_len) == next_lblk) && /*C2*/ |
| 3681 |
((ee_pblk + ee_len) == next_pblk) && /*C3*/ |
| 3682 |
(next_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/ |
| 3683 |
err = ext4_ext_get_access(handle, inode, path + depth); |
| 3684 |
if (err) |
| 3685 |
goto errout; |
| 3686 |
|
| 3687 |
trace_ext4_ext_convert_to_initialized_fastpath(inode, |
| 3688 |
map, ex, abut_ex); |
| 3689 |
|
| 3690 |
/* Shift the start of abut_ex by 'map_len' blocks */ |
| 3691 |
abut_ex->ee_block = cpu_to_le32(next_lblk - map_len); |
| 3692 |
ext4_ext_store_pblock(abut_ex, next_pblk - map_len); |
| 3693 |
ex->ee_len = cpu_to_le16(ee_len - map_len); |
| 3694 |
ext4_ext_mark_unwritten(ex); /* Restore the flag */ |
| 3695 |
|
| 3696 |
/* Extend abut_ex by 'map_len' blocks */ |
| 3697 |
abut_ex->ee_len = cpu_to_le16(next_len + map_len); |
| 3698 |
|
| 3699 |
/* Result: number of initialized blocks past m_lblk */ |
| 3700 |
*allocated = map_len; |
| 3701 |
} |
| 3702 |
} |
| 3703 |
if (*allocated) { |
| 3704 |
/* Mark the block containing both extents as dirty */ |
| 3705 |
err = ext4_ext_dirty(handle, inode, path + depth); |
| 3706 |
|
| 3707 |
/* Update path to point to the right extent */ |
| 3708 |
path[depth].p_ext = abut_ex; |
| 3709 |
if (err) |
| 3710 |
goto errout; |
| 3711 |
goto out; |
| 3712 |
} else |
| 3713 |
*allocated = ee_len - (map->m_lblk - ee_block); |
| 3714 |
|
| 3715 |
WARN_ON(map->m_lblk < ee_block); |
| 3716 |
/* |
| 3717 |
* It is safe to convert extent to initialized via explicit |
| 3718 |
* zeroout only if extent is fully inside i_size or new_size. |
| 3719 |
*/ |
| 3720 |
if (ee_block + ee_len <= eof_block) |
| 3721 |
max_zeroout = sbi->s_extent_max_zeroout_kb >> |
| 3722 |
(inode->i_sb->s_blocksize_bits - 10); |
| 3723 |
|
| 3724 |
/* |
| 3725 |
* five cases: |
| 3726 |
* 1. split the extent into three extents. |
| 3727 |
* 2. split the extent into two extents, zeroout the head of the first |
| 3728 |
* extent. |
| 3729 |
* 3. split the extent into two extents, zeroout the tail of the second |
| 3730 |
* extent. |
| 3731 |
* 4. split the extent into two extents with out zeroout. |
| 3732 |
* 5. no splitting needed, just possibly zeroout the head and / or the |
| 3733 |
* tail of the extent. |
| 3734 |
*/ |
| 3735 |
split_map.m_lblk = map->m_lblk; |
| 3736 |
split_map.m_len = map->m_len; |
| 3737 |
|
| 3738 |
if (max_zeroout && (*allocated > split_map.m_len)) { |
| 3739 |
if (*allocated <= max_zeroout) { |
| 3740 |
/* case 3 or 5 */ |
| 3741 |
zero_ex1.ee_block = |
| 3742 |
cpu_to_le32(split_map.m_lblk + |
| 3743 |
split_map.m_len); |
| 3744 |
zero_ex1.ee_len = |
| 3745 |
cpu_to_le16(*allocated - split_map.m_len); |
| 3746 |
ext4_ext_store_pblock(&zero_ex1, |
| 3747 |
ext4_ext_pblock(ex) + split_map.m_lblk + |
| 3748 |
split_map.m_len - ee_block); |
| 3749 |
err = ext4_ext_zeroout(inode, &zero_ex1); |
| 3750 |
if (err) |
| 3751 |
goto fallback; |
| 3752 |
split_map.m_len = *allocated; |
| 3753 |
} |
| 3754 |
if (split_map.m_lblk - ee_block + split_map.m_len < |
| 3755 |
max_zeroout) { |
| 3756 |
/* case 2 or 5 */ |
| 3757 |
if (split_map.m_lblk != ee_block) { |
| 3758 |
zero_ex2.ee_block = ex->ee_block; |
| 3759 |
zero_ex2.ee_len = cpu_to_le16(split_map.m_lblk - |
| 3760 |
ee_block); |
| 3761 |
ext4_ext_store_pblock(&zero_ex2, |
| 3762 |
ext4_ext_pblock(ex)); |
| 3763 |
err = ext4_ext_zeroout(inode, &zero_ex2); |
| 3764 |
if (err) |
| 3765 |
goto fallback; |
| 3766 |
} |
| 3767 |
|
| 3768 |
split_map.m_len += split_map.m_lblk - ee_block; |
| 3769 |
split_map.m_lblk = ee_block; |
| 3770 |
*allocated = map->m_len; |
| 3771 |
} |
| 3772 |
} |
| 3773 |
|
| 3774 |
fallback: |
| 3775 |
path = ext4_split_convert_extents(handle, inode, &split_map, path, |
| 3776 |
flags | EXT4_GET_BLOCKS_CONVERT, NULL); |
| 3777 |
if (IS_ERR(path)) |
| 3778 |
return path; |
| 3779 |
out: |
| 3780 |
/* If we have gotten a failure, don't zero out status tree */ |
| 3781 |
ext4_zeroout_es(inode, &zero_ex1); |
| 3782 |
ext4_zeroout_es(inode, &zero_ex2); |
| 3783 |
return path; |
| 3784 |
|
| 3785 |
errout: |
| 3786 |
ext4_free_ext_path(path); |
| 3787 |
return ERR_PTR(err); |
| 3788 |
} |
| 3789 |
|
| 3790 |
/* |
| 3791 |
* This function is called by ext4_ext_map_blocks() from |
| 3792 |
* ext4_get_blocks_dio_write() when DIO to write |
| 3793 |
* to an unwritten extent. |
| 3794 |
* |
| 3795 |
* Writing to an unwritten extent may result in splitting the unwritten |
| 3796 |
* extent into multiple initialized/unwritten extents (up to three) |
| 3797 |
* There are three possibilities: |
| 3798 |
* a> There is no split required: Entire extent should be unwritten |
| 3799 |
* b> Splits in two extents: Write is happening at either end of the extent |
| 3800 |
* c> Splits in three extents: Somone is writing in middle of the extent |
| 3801 |
* |
| 3802 |
* This works the same way in the case of initialized -> unwritten conversion. |
| 3803 |
* |
| 3804 |
* One of more index blocks maybe needed if the extent tree grow after |
| 3805 |
* the unwritten extent split. To prevent ENOSPC occur at the IO |
| 3806 |
* complete, we need to split the unwritten extent before DIO submit |
| 3807 |
* the IO. The unwritten extent called at this time will be split |
| 3808 |
* into three unwritten extent(at most). After IO complete, the part |
| 3809 |
* being filled will be convert to initialized by the end_io callback function |
| 3810 |
* via ext4_convert_unwritten_extents(). |
| 3811 |
* |
| 3812 |
* The size of unwritten extent to be written is passed to the caller via the |
| 3813 |
* allocated pointer. Return an extent path pointer on success, or an error |
| 3814 |
* pointer on failure. |
| 3815 |
*/ |
| 3816 |
static struct ext4_ext_path *ext4_split_convert_extents(handle_t *handle, |
| 3817 |
struct inode *inode, |
| 3818 |
struct ext4_map_blocks *map, |
| 3819 |
struct ext4_ext_path *path, |
| 3820 |
int flags, unsigned int *allocated) |
| 3821 |
{ |
| 3822 |
ext4_lblk_t eof_block; |
| 3823 |
ext4_lblk_t ee_block; |
| 3824 |
struct ext4_extent *ex; |
| 3825 |
unsigned int ee_len; |
| 3826 |
int split_flag = 0, depth, err = 0; |
| 3827 |
bool did_zeroout = false; |
| 3828 |
|
| 3829 |
ext_debug(inode, "logical block %llu, max_blocks %u\n", |
| 3830 |
(unsigned long long)map->m_lblk, map->m_len); |
| 3831 |
|
| 3832 |
eof_block = (EXT4_I(inode)->i_disksize + inode->i_sb->s_blocksize - 1) |
| 3833 |
>> inode->i_sb->s_blocksize_bits; |
| 3834 |
if (eof_block < map->m_lblk + map->m_len) |
| 3835 |
eof_block = map->m_lblk + map->m_len; |
| 3836 |
depth = ext_depth(inode); |
| 3837 |
ex = path[depth].p_ext; |
| 3838 |
ee_block = le32_to_cpu(ex->ee_block); |
| 3839 |
ee_len = ext4_ext_get_actual_len(ex); |
| 3840 |
|
| 3841 |
/* No split needed */ |
| 3842 |
if (ee_block == map->m_lblk && ee_len == map->m_len) |
| 3843 |
goto convert; |
| 3844 |
|
| 3845 |
/* |
| 3846 |
* It is only safe to convert extent to initialized via explicit |
| 3847 |
* zeroout only if extent is fully inside i_size or new_size. |
| 3848 |
*/ |
| 3849 |
split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0; |
| 3850 |
|
| 3851 |
/* |
| 3852 |
* pass SPLIT_NOMERGE explicitly so we don't end up merging extents we |
| 3853 |
* just split. |
| 3854 |
*/ |
| 3855 |
path = ext4_split_extent(handle, inode, path, map, split_flag, |
| 3856 |
flags | EXT4_GET_BLOCKS_SPLIT_NOMERGE, |
| 3857 |
allocated, &did_zeroout); |
| 3858 |
if (IS_ERR(path)) |
| 3859 |
return path; |
| 3860 |
|
| 3861 |
convert: |
| 3862 |
path = ext4_find_extent(inode, map->m_lblk, path, flags); |
| 3863 |
if (IS_ERR(path)) |
| 3864 |
return path; |
| 3865 |
|
| 3866 |
depth = ext_depth(inode); |
| 3867 |
ex = path[depth].p_ext; |
| 3868 |
|
| 3869 |
/* |
| 3870 |
* Conversion is already handled in case of zeroout |
| 3871 |
*/ |
| 3872 |
if (!did_zeroout) { |
| 3873 |
err = ext4_ext_get_access(handle, inode, path + depth); |
| 3874 |
if (err) |
| 3875 |
goto err; |
| 3876 |
|
| 3877 |
if (flags & EXT4_GET_BLOCKS_CONVERT) |
| 3878 |
ext4_ext_mark_initialized(ex); |
| 3879 |
else if (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN) |
| 3880 |
ext4_ext_mark_unwritten(ex); |
| 3881 |
|
| 3882 |
if (!(flags & EXT4_GET_BLOCKS_SPLIT_NOMERGE)) |
| 3883 |
/* |
| 3884 |
* note: ext4_ext_correct_indexes() isn't needed here because |
| 3885 |
* borders are not changed |
| 3886 |
*/ |
| 3887 |
ext4_ext_try_to_merge(handle, inode, path, ex); |
| 3888 |
|
| 3889 |
err = ext4_ext_dirty(handle, inode, path + depth); |
| 3890 |
if (err) |
| 3891 |
goto err; |
| 3892 |
} |
| 3893 |
|
| 3894 |
/* Lets update the extent status tree after conversion */ |
| 3895 |
if (!(flags & EXT4_EX_NOCACHE)) |
| 3896 |
ext4_es_insert_extent(inode, le32_to_cpu(ex->ee_block), |
| 3897 |
ext4_ext_get_actual_len(ex), |
| 3898 |
ext4_ext_pblock(ex), |
| 3899 |
ext4_ext_is_unwritten(ex) ? |
| 3900 |
EXTENT_STATUS_UNWRITTEN : |
| 3901 |
EXTENT_STATUS_WRITTEN, |
| 3902 |
false); |
| 3903 |
|
| 3904 |
err: |
| 3905 |
if (err) { |
| 3906 |
ext4_free_ext_path(path); |
| 3907 |
return ERR_PTR(err); |
| 3908 |
} |
| 3909 |
|
| 3910 |
return path; |
| 3911 |
} |
| 3912 |
|
| 3913 |
static struct ext4_ext_path * |
| 3914 |
ext4_convert_unwritten_extents_endio(handle_t *handle, struct inode *inode, |
| 3915 |
struct ext4_map_blocks *map, |
| 3916 |
struct ext4_ext_path *path, int flags) |
| 3917 |
{ |
| 3918 |
struct ext4_extent *ex; |
| 3919 |
ext4_lblk_t ee_block; |
| 3920 |
unsigned int ee_len; |
| 3921 |
int depth; |
| 3922 |
|
| 3923 |
depth = ext_depth(inode); |
| 3924 |
ex = path[depth].p_ext; |
| 3925 |
ee_block = le32_to_cpu(ex->ee_block); |
| 3926 |
ee_len = ext4_ext_get_actual_len(ex); |
| 3927 |
|
| 3928 |
ext_debug(inode, "logical block %llu, max_blocks %u\n", |
| 3929 |
(unsigned long long)ee_block, ee_len); |
| 3930 |
|
| 3931 |
return ext4_split_convert_extents(handle, inode, map, path, flags, |
| 3932 |
NULL); |
| 3933 |
} |
| 3934 |
|
| 3935 |
static struct ext4_ext_path * |
| 3936 |
convert_initialized_extent(handle_t *handle, struct inode *inode, |
| 3937 |
struct ext4_map_blocks *map, |
| 3938 |
struct ext4_ext_path *path, |
| 3939 |
int flags, |
| 3940 |
unsigned int *allocated) |
| 3941 |
{ |
| 3942 |
struct ext4_extent *ex; |
| 3943 |
ext4_lblk_t ee_block; |
| 3944 |
unsigned int ee_len; |
| 3945 |
int depth; |
| 3946 |
|
| 3947 |
/* |
| 3948 |
* Make sure that the extent is no bigger than we support with |
| 3949 |
* unwritten extent |
| 3950 |
*/ |
| 3951 |
if (map->m_len > EXT_UNWRITTEN_MAX_LEN) |
| 3952 |
map->m_len = EXT_UNWRITTEN_MAX_LEN / 2; |
| 3953 |
|
| 3954 |
depth = ext_depth(inode); |
| 3955 |
ex = path[depth].p_ext; |
| 3956 |
ee_block = le32_to_cpu(ex->ee_block); |
| 3957 |
ee_len = ext4_ext_get_actual_len(ex); |
| 3958 |
|
| 3959 |
ext_debug(inode, "logical block %llu, max_blocks %u\n", |
| 3960 |
(unsigned long long)ee_block, ee_len); |
| 3961 |
|
| 3962 |
path = ext4_split_convert_extents(handle, inode, map, path, flags, |
| 3963 |
NULL); |
| 3964 |
if (IS_ERR(path)) |
| 3965 |
return path; |
| 3966 |
|
| 3967 |
ext4_ext_show_leaf(inode, path); |
| 3968 |
|
| 3969 |
ext4_update_inode_fsync_trans(handle, inode, 1); |
| 3970 |
|
| 3971 |
/* |
| 3972 |
* The extent might be initialized in case of zeroout. |
| 3973 |
*/ |
| 3974 |
path = ext4_find_extent(inode, map->m_lblk, path, flags); |
| 3975 |
if (IS_ERR(path)) |
| 3976 |
return path; |
| 3977 |
|
| 3978 |
depth = ext_depth(inode); |
| 3979 |
ex = path[depth].p_ext; |
| 3980 |
|
| 3981 |
if (ext4_ext_is_unwritten(ex)) |
| 3982 |
map->m_flags |= EXT4_MAP_UNWRITTEN; |
| 3983 |
else |
| 3984 |
map->m_flags |= EXT4_MAP_MAPPED; |
| 3985 |
if (*allocated > map->m_len) |
| 3986 |
*allocated = map->m_len; |
| 3987 |
map->m_len = *allocated; |
| 3988 |
return path; |
| 3989 |
} |
| 3990 |
|
| 3991 |
static struct ext4_ext_path * |
| 3992 |
ext4_ext_handle_unwritten_extents(handle_t *handle, struct inode *inode, |
| 3993 |
struct ext4_map_blocks *map, |
| 3994 |
struct ext4_ext_path *path, int flags, |
| 3995 |
unsigned int *allocated, ext4_fsblk_t newblock) |
| 3996 |
{ |
| 3997 |
int err = 0; |
| 3998 |
|
| 3999 |
ext_debug(inode, "logical block %llu, max_blocks %u, flags 0x%x, allocated %u\n", |
| 4000 |
(unsigned long long)map->m_lblk, map->m_len, flags, |
| 4001 |
*allocated); |
| 4002 |
ext4_ext_show_leaf(inode, path); |
| 4003 |
|
| 4004 |
/* |
| 4005 |
* When writing into unwritten space, we should not fail to |
| 4006 |
* allocate metadata blocks for the new extent block if needed. |
| 4007 |
*/ |
| 4008 |
flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL; |
| 4009 |
|
| 4010 |
trace_ext4_ext_handle_unwritten_extents(inode, map, flags, |
| 4011 |
*allocated, newblock); |
| 4012 |
|
| 4013 |
/* IO end_io complete, convert the filled extent to written */ |
| 4014 |
if (flags & EXT4_GET_BLOCKS_CONVERT) { |
| 4015 |
path = ext4_convert_unwritten_extents_endio(handle, inode, |
| 4016 |
map, path, flags); |
| 4017 |
if (IS_ERR(path)) |
| 4018 |
return path; |
| 4019 |
ext4_update_inode_fsync_trans(handle, inode, 1); |
| 4020 |
goto map_out; |
| 4021 |
} |
| 4022 |
/* buffered IO cases */ |
| 4023 |
/* |
| 4024 |
* repeat fallocate creation request |
| 4025 |
* we already have an unwritten extent |
| 4026 |
*/ |
| 4027 |
if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) { |
| 4028 |
map->m_flags |= EXT4_MAP_UNWRITTEN; |
| 4029 |
goto map_out; |
| 4030 |
} |
| 4031 |
|
| 4032 |
/* buffered READ or buffered write_begin() lookup */ |
| 4033 |
if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) { |
| 4034 |
/* |
| 4035 |
* We have blocks reserved already. We |
| 4036 |
* return allocated blocks so that delalloc |
| 4037 |
* won't do block reservation for us. But |
| 4038 |
* the buffer head will be unmapped so that |
| 4039 |
* a read from the block returns 0s. |
| 4040 |
*/ |
| 4041 |
map->m_flags |= EXT4_MAP_UNWRITTEN; |
| 4042 |
goto out1; |
| 4043 |
} |
| 4044 |
|
| 4045 |
/* |
| 4046 |
* Default case when (flags & EXT4_GET_BLOCKS_CREATE) == 1. |
| 4047 |
* For buffered writes, at writepage time, etc. Convert a |
| 4048 |
* discovered unwritten extent to written. |
| 4049 |
*/ |
| 4050 |
path = ext4_ext_convert_to_initialized(handle, inode, map, path, |
| 4051 |
flags, allocated); |
| 4052 |
if (IS_ERR(path)) |
| 4053 |
return path; |
| 4054 |
ext4_update_inode_fsync_trans(handle, inode, 1); |
| 4055 |
/* |
| 4056 |
* shouldn't get a 0 allocated when converting an unwritten extent |
| 4057 |
* unless m_len is 0 (bug) or extent has been corrupted |
| 4058 |
*/ |
| 4059 |
if (unlikely(*allocated == 0)) { |
| 4060 |
EXT4_ERROR_INODE(inode, "unexpected allocated == 0, m_len = %u", |
| 4061 |
map->m_len); |
| 4062 |
err = -EFSCORRUPTED; |
| 4063 |
goto errout; |
| 4064 |
} |
| 4065 |
|
| 4066 |
map->m_flags |= EXT4_MAP_NEW; |
| 4067 |
map_out: |
| 4068 |
map->m_flags |= EXT4_MAP_MAPPED; |
| 4069 |
out1: |
| 4070 |
map->m_pblk = newblock; |
| 4071 |
if (*allocated > map->m_len) |
| 4072 |
*allocated = map->m_len; |
| 4073 |
map->m_len = *allocated; |
| 4074 |
ext4_ext_show_leaf(inode, path); |
| 4075 |
return path; |
| 4076 |
|
| 4077 |
errout: |
| 4078 |
ext4_free_ext_path(path); |
| 4079 |
return ERR_PTR(err); |
| 4080 |
} |
| 4081 |
|
| 4082 |
/* |
| 4083 |
* get_implied_cluster_alloc - check to see if the requested |
| 4084 |
* allocation (in the map structure) overlaps with a cluster already |
| 4085 |
* allocated in an extent. |
| 4086 |
* @sb The filesystem superblock structure |
| 4087 |
* @map The requested lblk->pblk mapping |
| 4088 |
* @ex The extent structure which might contain an implied |
| 4089 |
* cluster allocation |
| 4090 |
* |
| 4091 |
* This function is called by ext4_ext_map_blocks() after we failed to |
| 4092 |
* find blocks that were already in the inode's extent tree. Hence, |
| 4093 |
* we know that the beginning of the requested region cannot overlap |
| 4094 |
* the extent from the inode's extent tree. There are three cases we |
| 4095 |
* want to catch. The first is this case: |
| 4096 |
* |
| 4097 |
* |--- cluster # N--| |
| 4098 |
* |--- extent ---| |---- requested region ---| |
| 4099 |
* |==========| |
| 4100 |
* |
| 4101 |
* The second case that we need to test for is this one: |
| 4102 |
* |
| 4103 |
* |--------- cluster # N ----------------| |
| 4104 |
* |--- requested region --| |------- extent ----| |
| 4105 |
* |=======================| |
| 4106 |
* |
| 4107 |
* The third case is when the requested region lies between two extents |
| 4108 |
* within the same cluster: |
| 4109 |
* |------------- cluster # N-------------| |
| 4110 |
* |----- ex -----| |---- ex_right ----| |
| 4111 |
* |------ requested region ------| |
| 4112 |
* |================| |
| 4113 |
* |
| 4114 |
* In each of the above cases, we need to set the map->m_pblk and |
| 4115 |
* map->m_len so it corresponds to the return the extent labelled as |
| 4116 |
* "|====|" from cluster #N, since it is already in use for data in |
| 4117 |
* cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to |
| 4118 |
* signal to ext4_ext_map_blocks() that map->m_pblk should be treated |
| 4119 |
* as a new "allocated" block region. Otherwise, we will return 0 and |
| 4120 |
* ext4_ext_map_blocks() will then allocate one or more new clusters |
| 4121 |
* by calling ext4_mb_new_blocks(). |
| 4122 |
*/ |
| 4123 |
static int get_implied_cluster_alloc(struct super_block *sb, |
| 4124 |
struct ext4_map_blocks *map, |
| 4125 |
struct ext4_extent *ex, |
| 4126 |
struct ext4_ext_path *path) |
| 4127 |
{ |
| 4128 |
struct ext4_sb_info *sbi = EXT4_SB(sb); |
| 4129 |
ext4_lblk_t c_offset = EXT4_LBLK_COFF(sbi, map->m_lblk); |
| 4130 |
ext4_lblk_t ex_cluster_start, ex_cluster_end; |
| 4131 |
ext4_lblk_t rr_cluster_start; |
| 4132 |
ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block); |
| 4133 |
ext4_fsblk_t ee_start = ext4_ext_pblock(ex); |
| 4134 |
unsigned short ee_len = ext4_ext_get_actual_len(ex); |
| 4135 |
|
| 4136 |
/* The extent passed in that we are trying to match */ |
| 4137 |
ex_cluster_start = EXT4_B2C(sbi, ee_block); |
| 4138 |
ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1); |
| 4139 |
|
| 4140 |
/* The requested region passed into ext4_map_blocks() */ |
| 4141 |
rr_cluster_start = EXT4_B2C(sbi, map->m_lblk); |
| 4142 |
|
| 4143 |
if ((rr_cluster_start == ex_cluster_end) || |
| 4144 |
(rr_cluster_start == ex_cluster_start)) { |
| 4145 |
if (rr_cluster_start == ex_cluster_end) |
| 4146 |
ee_start += ee_len - 1; |
| 4147 |
map->m_pblk = EXT4_PBLK_CMASK(sbi, ee_start) + c_offset; |
| 4148 |
map->m_len = min(map->m_len, |
| 4149 |
(unsigned) sbi->s_cluster_ratio - c_offset); |
| 4150 |
/* |
| 4151 |
* Check for and handle this case: |
| 4152 |
* |
| 4153 |
* |--------- cluster # N-------------| |
| 4154 |
* |------- extent ----| |
| 4155 |
* |--- requested region ---| |
| 4156 |
* |===========| |
| 4157 |
*/ |
| 4158 |
|
| 4159 |
if (map->m_lblk < ee_block) |
| 4160 |
map->m_len = min(map->m_len, ee_block - map->m_lblk); |
| 4161 |
|
| 4162 |
/* |
| 4163 |
* Check for the case where there is already another allocated |
| 4164 |
* block to the right of 'ex' but before the end of the cluster. |
| 4165 |
* |
| 4166 |
* |------------- cluster # N-------------| |
| 4167 |
* |----- ex -----| |---- ex_right ----| |
| 4168 |
* |------ requested region ------| |
| 4169 |
* |================| |
| 4170 |
*/ |
| 4171 |
if (map->m_lblk > ee_block) { |
| 4172 |
ext4_lblk_t next = ext4_ext_next_allocated_block(path); |
| 4173 |
map->m_len = min(map->m_len, next - map->m_lblk); |
| 4174 |
} |
| 4175 |
|
| 4176 |
trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1); |
| 4177 |
return 1; |
| 4178 |
} |
| 4179 |
|
| 4180 |
trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0); |
| 4181 |
return 0; |
| 4182 |
} |
| 4183 |
|
| 4184 |
/* |
| 4185 |
* Determine hole length around the given logical block, first try to |
| 4186 |
* locate and expand the hole from the given @path, and then adjust it |
| 4187 |
* if it's partially or completely converted to delayed extents, insert |
| 4188 |
* it into the extent cache tree if it's indeed a hole, finally return |
| 4189 |
* the length of the determined extent. |
| 4190 |
*/ |
| 4191 |
static ext4_lblk_t ext4_ext_determine_insert_hole(struct inode *inode, |
| 4192 |
struct ext4_ext_path *path, |
| 4193 |
ext4_lblk_t lblk) |
| 4194 |
{ |
| 4195 |
ext4_lblk_t hole_start, len; |
| 4196 |
struct extent_status es; |
| 4197 |
|
| 4198 |
hole_start = lblk; |
| 4199 |
len = ext4_ext_find_hole(inode, path, &hole_start); |
| 4200 |
again: |
| 4201 |
ext4_es_find_extent_range(inode, &ext4_es_is_delayed, hole_start, |
| 4202 |
hole_start + len - 1, &es); |
| 4203 |
if (!es.es_len) |
| 4204 |
goto insert_hole; |
| 4205 |
|
| 4206 |
/* |
| 4207 |
* There's a delalloc extent in the hole, handle it if the delalloc |
| 4208 |
* extent is in front of, behind and straddle the queried range. |
| 4209 |
*/ |
| 4210 |
if (lblk >= es.es_lblk + es.es_len) { |
| 4211 |
/* |
| 4212 |
* The delalloc extent is in front of the queried range, |
| 4213 |
* find again from the queried start block. |
| 4214 |
*/ |
| 4215 |
len -= lblk - hole_start; |
| 4216 |
hole_start = lblk; |
| 4217 |
goto again; |
| 4218 |
} else if (in_range(lblk, es.es_lblk, es.es_len)) { |
| 4219 |
/* |
| 4220 |
* The delalloc extent containing lblk, it must have been |
| 4221 |
* added after ext4_map_blocks() checked the extent status |
| 4222 |
* tree so we are not holding i_rwsem and delalloc info is |
| 4223 |
* only stabilized by i_data_sem we are going to release |
| 4224 |
* soon. Don't modify the extent status tree and report |
| 4225 |
* extent as a hole, just adjust the length to the delalloc |
| 4226 |
* extent's after lblk. |
| 4227 |
*/ |
| 4228 |
len = es.es_lblk + es.es_len - lblk; |
| 4229 |
return len; |
| 4230 |
} else { |
| 4231 |
/* |
| 4232 |
* The delalloc extent is partially or completely behind |
| 4233 |
* the queried range, update hole length until the |
| 4234 |
* beginning of the delalloc extent. |
| 4235 |
*/ |
| 4236 |
len = min(es.es_lblk - hole_start, len); |
| 4237 |
} |
| 4238 |
|
| 4239 |
insert_hole: |
| 4240 |
/* Put just found gap into cache to speed up subsequent requests */ |
| 4241 |
ext_debug(inode, " -> %u:%u\n", hole_start, len); |
| 4242 |
ext4_es_cache_extent(inode, hole_start, len, ~0, EXTENT_STATUS_HOLE); |
| 4243 |
|
| 4244 |
/* Update hole_len to reflect hole size after lblk */ |
| 4245 |
if (hole_start != lblk) |
| 4246 |
len -= lblk - hole_start; |
| 4247 |
|
| 4248 |
return len; |
| 4249 |
} |
| 4250 |
|
| 4251 |
/* |
| 4252 |
* Block allocation/map/preallocation routine for extents based files |
| 4253 |
* |
| 4254 |
* |
| 4255 |
* Need to be called with |
| 4256 |
* down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block |
| 4257 |
* (ie, flags is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem) |
| 4258 |
* |
| 4259 |
* return > 0, number of blocks already mapped/allocated |
| 4260 |
* if flags doesn't contain EXT4_GET_BLOCKS_CREATE and these are pre-allocated blocks |
| 4261 |
* buffer head is unmapped |
| 4262 |
* otherwise blocks are mapped |
| 4263 |
* |
| 4264 |
* return = 0, if plain look up failed (blocks have not been allocated) |
| 4265 |
* buffer head is unmapped |
| 4266 |
* |
| 4267 |
* return < 0, error case. |
| 4268 |
*/ |
| 4269 |
int ext4_ext_map_blocks(handle_t *handle, struct inode *inode, |
| 4270 |
struct ext4_map_blocks *map, int flags) |
| 4271 |
{ |
| 4272 |
struct ext4_ext_path *path = NULL; |
| 4273 |
struct ext4_extent newex, *ex, ex2; |
| 4274 |
struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
| 4275 |
ext4_fsblk_t newblock = 0, pblk; |
| 4276 |
int err = 0, depth; |
| 4277 |
unsigned int allocated = 0, offset = 0; |
| 4278 |
unsigned int allocated_clusters = 0; |
| 4279 |
struct ext4_allocation_request ar; |
| 4280 |
ext4_lblk_t cluster_offset; |
| 4281 |
|
| 4282 |
ext_debug(inode, "blocks %u/%u requested\n", map->m_lblk, map->m_len); |
| 4283 |
trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags); |
| 4284 |
|
| 4285 |
/* find extent for this block */ |
| 4286 |
path = ext4_find_extent(inode, map->m_lblk, NULL, flags); |
| 4287 |
if (IS_ERR(path)) { |
| 4288 |
err = PTR_ERR(path); |
| 4289 |
goto out; |
| 4290 |
} |
| 4291 |
|
| 4292 |
depth = ext_depth(inode); |
| 4293 |
|
| 4294 |
/* |
| 4295 |
* consistent leaf must not be empty; |
| 4296 |
* this situation is possible, though, _during_ tree modification; |
| 4297 |
* this is why assert can't be put in ext4_find_extent() |
| 4298 |
*/ |
| 4299 |
if (unlikely(path[depth].p_ext == NULL && depth != 0)) { |
| 4300 |
EXT4_ERROR_INODE(inode, "bad extent address " |
| 4301 |
"lblock: %lu, depth: %d pblock %lld", |
| 4302 |
(unsigned long) map->m_lblk, depth, |
| 4303 |
path[depth].p_block); |
| 4304 |
err = -EFSCORRUPTED; |
| 4305 |
goto out; |
| 4306 |
} |
| 4307 |
|
| 4308 |
ex = path[depth].p_ext; |
| 4309 |
if (ex) { |
| 4310 |
ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block); |
| 4311 |
ext4_fsblk_t ee_start = ext4_ext_pblock(ex); |
| 4312 |
unsigned short ee_len; |
| 4313 |
|
| 4314 |
|
| 4315 |
/* |
| 4316 |
* unwritten extents are treated as holes, except that |
| 4317 |
* we split out initialized portions during a write. |
| 4318 |
*/ |
| 4319 |
ee_len = ext4_ext_get_actual_len(ex); |
| 4320 |
|
| 4321 |
trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len); |
| 4322 |
|
| 4323 |
/* if found extent covers block, simply return it */ |
| 4324 |
if (in_range(map->m_lblk, ee_block, ee_len)) { |
| 4325 |
newblock = map->m_lblk - ee_block + ee_start; |
| 4326 |
/* number of remaining blocks in the extent */ |
| 4327 |
allocated = ee_len - (map->m_lblk - ee_block); |
| 4328 |
ext_debug(inode, "%u fit into %u:%d -> %llu\n", |
| 4329 |
map->m_lblk, ee_block, ee_len, newblock); |
| 4330 |
|
| 4331 |
/* |
| 4332 |
* If the extent is initialized check whether the |
| 4333 |
* caller wants to convert it to unwritten. |
| 4334 |
*/ |
| 4335 |
if ((!ext4_ext_is_unwritten(ex)) && |
| 4336 |
(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) { |
| 4337 |
path = convert_initialized_extent(handle, |
| 4338 |
inode, map, path, flags, &allocated); |
| 4339 |
if (IS_ERR(path)) |
| 4340 |
err = PTR_ERR(path); |
| 4341 |
goto out; |
| 4342 |
} else if (!ext4_ext_is_unwritten(ex)) { |
| 4343 |
map->m_flags |= EXT4_MAP_MAPPED; |
| 4344 |
map->m_pblk = newblock; |
| 4345 |
if (allocated > map->m_len) |
| 4346 |
allocated = map->m_len; |
| 4347 |
map->m_len = allocated; |
| 4348 |
ext4_ext_show_leaf(inode, path); |
| 4349 |
goto out; |
| 4350 |
} |
| 4351 |
|
| 4352 |
path = ext4_ext_handle_unwritten_extents( |
| 4353 |
handle, inode, map, path, flags, |
| 4354 |
&allocated, newblock); |
| 4355 |
if (IS_ERR(path)) |
| 4356 |
err = PTR_ERR(path); |
| 4357 |
goto out; |
| 4358 |
} |
| 4359 |
} |
| 4360 |
|
| 4361 |
/* |
| 4362 |
* requested block isn't allocated yet; |
| 4363 |
* we couldn't try to create block if flags doesn't contain EXT4_GET_BLOCKS_CREATE |
| 4364 |
*/ |
| 4365 |
if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) { |
| 4366 |
ext4_lblk_t len; |
| 4367 |
|
| 4368 |
len = ext4_ext_determine_insert_hole(inode, path, map->m_lblk); |
| 4369 |
|
| 4370 |
map->m_pblk = 0; |
| 4371 |
map->m_len = min_t(unsigned int, map->m_len, len); |
| 4372 |
goto out; |
| 4373 |
} |
| 4374 |
|
| 4375 |
/* |
| 4376 |
* Okay, we need to do block allocation. |
| 4377 |
*/ |
| 4378 |
newex.ee_block = cpu_to_le32(map->m_lblk); |
| 4379 |
cluster_offset = EXT4_LBLK_COFF(sbi, map->m_lblk); |
| 4380 |
|
| 4381 |
/* |
| 4382 |
* If we are doing bigalloc, check to see if the extent returned |
| 4383 |
* by ext4_find_extent() implies a cluster we can use. |
| 4384 |
*/ |
| 4385 |
if (cluster_offset && ex && |
| 4386 |
get_implied_cluster_alloc(inode->i_sb, map, ex, path)) { |
| 4387 |
ar.len = allocated = map->m_len; |
| 4388 |
newblock = map->m_pblk; |
| 4389 |
goto got_allocated_blocks; |
| 4390 |
} |
| 4391 |
|
| 4392 |
/* find neighbour allocated blocks */ |
| 4393 |
ar.lleft = map->m_lblk; |
| 4394 |
err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft); |
| 4395 |
if (err) |
| 4396 |
goto out; |
| 4397 |
ar.lright = map->m_lblk; |
| 4398 |
err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, |
| 4399 |
&ex2, flags); |
| 4400 |
if (err < 0) |
| 4401 |
goto out; |
| 4402 |
|
| 4403 |
/* Check if the extent after searching to the right implies a |
| 4404 |
* cluster we can use. */ |
| 4405 |
if ((sbi->s_cluster_ratio > 1) && err && |
| 4406 |
get_implied_cluster_alloc(inode->i_sb, map, &ex2, path)) { |
| 4407 |
ar.len = allocated = map->m_len; |
| 4408 |
newblock = map->m_pblk; |
| 4409 |
err = 0; |
| 4410 |
goto got_allocated_blocks; |
| 4411 |
} |
| 4412 |
|
| 4413 |
/* |
| 4414 |
* See if request is beyond maximum number of blocks we can have in |
| 4415 |
* a single extent. For an initialized extent this limit is |
| 4416 |
* EXT_INIT_MAX_LEN and for an unwritten extent this limit is |
| 4417 |
* EXT_UNWRITTEN_MAX_LEN. |
| 4418 |
*/ |
| 4419 |
if (map->m_len > EXT_INIT_MAX_LEN && |
| 4420 |
!(flags & EXT4_GET_BLOCKS_UNWRIT_EXT)) |
| 4421 |
map->m_len = EXT_INIT_MAX_LEN; |
| 4422 |
else if (map->m_len > EXT_UNWRITTEN_MAX_LEN && |
| 4423 |
(flags & EXT4_GET_BLOCKS_UNWRIT_EXT)) |
| 4424 |
map->m_len = EXT_UNWRITTEN_MAX_LEN; |
| 4425 |
|
| 4426 |
/* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */ |
| 4427 |
newex.ee_len = cpu_to_le16(map->m_len); |
| 4428 |
err = ext4_ext_check_overlap(sbi, inode, &newex, path); |
| 4429 |
if (err) |
| 4430 |
allocated = ext4_ext_get_actual_len(&newex); |
| 4431 |
else |
| 4432 |
allocated = map->m_len; |
| 4433 |
|
| 4434 |
/* allocate new block */ |
| 4435 |
ar.inode = inode; |
| 4436 |
ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk); |
| 4437 |
ar.logical = map->m_lblk; |
| 4438 |
/* |
| 4439 |
* We calculate the offset from the beginning of the cluster |
| 4440 |
* for the logical block number, since when we allocate a |
| 4441 |
* physical cluster, the physical block should start at the |
| 4442 |
* same offset from the beginning of the cluster. This is |
| 4443 |
* needed so that future calls to get_implied_cluster_alloc() |
| 4444 |
* work correctly. |
| 4445 |
*/ |
| 4446 |
offset = EXT4_LBLK_COFF(sbi, map->m_lblk); |
| 4447 |
ar.len = EXT4_NUM_B2C(sbi, offset+allocated); |
| 4448 |
ar.goal -= offset; |
| 4449 |
ar.logical -= offset; |
| 4450 |
if (S_ISREG(inode->i_mode)) |
| 4451 |
ar.flags = EXT4_MB_HINT_DATA; |
| 4452 |
else |
| 4453 |
/* disable in-core preallocation for non-regular files */ |
| 4454 |
ar.flags = 0; |
| 4455 |
if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE) |
| 4456 |
ar.flags |= EXT4_MB_HINT_NOPREALLOC; |
| 4457 |
if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) |
| 4458 |
ar.flags |= EXT4_MB_DELALLOC_RESERVED; |
| 4459 |
if (flags & EXT4_GET_BLOCKS_METADATA_NOFAIL) |
| 4460 |
ar.flags |= EXT4_MB_USE_RESERVED; |
| 4461 |
newblock = ext4_mb_new_blocks(handle, &ar, &err); |
| 4462 |
if (!newblock) |
| 4463 |
goto out; |
| 4464 |
allocated_clusters = ar.len; |
| 4465 |
ar.len = EXT4_C2B(sbi, ar.len) - offset; |
| 4466 |
ext_debug(inode, "allocate new block: goal %llu, found %llu/%u, requested %u\n", |
| 4467 |
ar.goal, newblock, ar.len, allocated); |
| 4468 |
if (ar.len > allocated) |
| 4469 |
ar.len = allocated; |
| 4470 |
|
| 4471 |
got_allocated_blocks: |
| 4472 |
/* try to insert new extent into found leaf and return */ |
| 4473 |
pblk = newblock + offset; |
| 4474 |
ext4_ext_store_pblock(&newex, pblk); |
| 4475 |
newex.ee_len = cpu_to_le16(ar.len); |
| 4476 |
/* Mark unwritten */ |
| 4477 |
if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) { |
| 4478 |
ext4_ext_mark_unwritten(&newex); |
| 4479 |
map->m_flags |= EXT4_MAP_UNWRITTEN; |
| 4480 |
} |
| 4481 |
|
| 4482 |
path = ext4_ext_insert_extent(handle, inode, path, &newex, flags); |
| 4483 |
if (IS_ERR(path)) { |
| 4484 |
err = PTR_ERR(path); |
| 4485 |
/* |
| 4486 |
* Gracefully handle out of space conditions. If the filesystem |
| 4487 |
* is inconsistent, we'll just leak allocated blocks to avoid |
| 4488 |
* causing even more damage. |
| 4489 |
*/ |
| 4490 |
if (allocated_clusters && (err == -EDQUOT || err == -ENOSPC)) { |
| 4491 |
int fb_flags = 0; |
| 4492 |
/* |
| 4493 |
* free data blocks we just allocated. |
| 4494 |
* not a good idea to call discard here directly, |
| 4495 |
* but otherwise we'd need to call it every free(). |
| 4496 |
*/ |
| 4497 |
ext4_discard_preallocations(inode); |
| 4498 |
if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) |
| 4499 |
fb_flags = EXT4_FREE_BLOCKS_NO_QUOT_UPDATE; |
| 4500 |
ext4_free_blocks(handle, inode, NULL, newblock, |
| 4501 |
EXT4_C2B(sbi, allocated_clusters), |
| 4502 |
fb_flags); |
| 4503 |
} |
| 4504 |
goto out; |
| 4505 |
} |
| 4506 |
|
| 4507 |
/* |
| 4508 |
* Cache the extent and update transaction to commit on fdatasync only |
| 4509 |
* when it is _not_ an unwritten extent. |
| 4510 |
*/ |
| 4511 |
if ((flags & EXT4_GET_BLOCKS_UNWRIT_EXT) == 0) |
| 4512 |
ext4_update_inode_fsync_trans(handle, inode, 1); |
| 4513 |
else |
| 4514 |
ext4_update_inode_fsync_trans(handle, inode, 0); |
| 4515 |
|
| 4516 |
map->m_flags |= (EXT4_MAP_NEW | EXT4_MAP_MAPPED); |
| 4517 |
map->m_pblk = pblk; |
| 4518 |
map->m_len = ar.len; |
| 4519 |
allocated = map->m_len; |
| 4520 |
ext4_ext_show_leaf(inode, path); |
| 4521 |
out: |
| 4522 |
/* |
| 4523 |
* We never use EXT4_GET_BLOCKS_QUERY_LAST_IN_LEAF with CREATE flag. |
| 4524 |
* So we know that the depth used here is correct, since there was no |
| 4525 |
* block allocation done if EXT4_GET_BLOCKS_QUERY_LAST_IN_LEAF is set. |
| 4526 |
* If tomorrow we start using this QUERY flag with CREATE, then we will |
| 4527 |
* need to re-calculate the depth as it might have changed due to block |
| 4528 |
* allocation. |
| 4529 |
*/ |
| 4530 |
if (flags & EXT4_GET_BLOCKS_QUERY_LAST_IN_LEAF) { |
| 4531 |
WARN_ON_ONCE(flags & EXT4_GET_BLOCKS_CREATE); |
| 4532 |
if (!err && ex && (ex == EXT_LAST_EXTENT(path[depth].p_hdr))) |
| 4533 |
map->m_flags |= EXT4_MAP_QUERY_LAST_IN_LEAF; |
| 4534 |
} |
| 4535 |
|
| 4536 |
ext4_free_ext_path(path); |
| 4537 |
|
| 4538 |
trace_ext4_ext_map_blocks_exit(inode, flags, map, |
| 4539 |
err ? err : allocated); |
| 4540 |
return err ? err : allocated; |
| 4541 |
} |
| 4542 |
|
| 4543 |
int ext4_ext_truncate(handle_t *handle, struct inode *inode) |
| 4544 |
{ |
| 4545 |
struct super_block *sb = inode->i_sb; |
| 4546 |
ext4_lblk_t last_block; |
| 4547 |
int err = 0; |
| 4548 |
|
| 4549 |
/* |
| 4550 |
* TODO: optimization is possible here. |
| 4551 |
* Probably we need not scan at all, |
| 4552 |
* because page truncation is enough. |
| 4553 |
*/ |
| 4554 |
|
| 4555 |
/* we have to know where to truncate from in crash case */ |
| 4556 |
EXT4_I(inode)->i_disksize = inode->i_size; |
| 4557 |
err = ext4_mark_inode_dirty(handle, inode); |
| 4558 |
if (err) |
| 4559 |
return err; |
| 4560 |
|
| 4561 |
last_block = (inode->i_size + sb->s_blocksize - 1) |
| 4562 |
>> EXT4_BLOCK_SIZE_BITS(sb); |
| 4563 |
ext4_es_remove_extent(inode, last_block, EXT_MAX_BLOCKS - last_block); |
| 4564 |
|
| 4565 |
retry_remove_space: |
| 4566 |
err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1); |
| 4567 |
if (err == -ENOMEM) { |
| 4568 |
memalloc_retry_wait(GFP_ATOMIC); |
| 4569 |
goto retry_remove_space; |
| 4570 |
} |
| 4571 |
return err; |
| 4572 |
} |
| 4573 |
|
| 4574 |
static int ext4_alloc_file_blocks(struct file *file, ext4_lblk_t offset, |
| 4575 |
ext4_lblk_t len, loff_t new_size, |
| 4576 |
int flags) |
| 4577 |
{ |
| 4578 |
struct inode *inode = file_inode(file); |
| 4579 |
handle_t *handle; |
| 4580 |
int ret = 0, ret2 = 0, ret3 = 0; |
| 4581 |
int retries = 0; |
| 4582 |
int depth = 0; |
| 4583 |
struct ext4_map_blocks map; |
| 4584 |
unsigned int credits; |
| 4585 |
loff_t epos, old_size = i_size_read(inode); |
| 4586 |
unsigned int blkbits = inode->i_blkbits; |
| 4587 |
bool alloc_zero = false; |
| 4588 |
|
| 4589 |
BUG_ON(!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)); |
| 4590 |
map.m_lblk = offset; |
| 4591 |
map.m_len = len; |
| 4592 |
/* |
| 4593 |
* Don't normalize the request if it can fit in one extent so |
| 4594 |
* that it doesn't get unnecessarily split into multiple |
| 4595 |
* extents. |
| 4596 |
*/ |
| 4597 |
if (len <= EXT_UNWRITTEN_MAX_LEN) |
| 4598 |
flags |= EXT4_GET_BLOCKS_NO_NORMALIZE; |
| 4599 |
|
| 4600 |
/* |
| 4601 |
* Do the actual write zero during a running journal transaction |
| 4602 |
* costs a lot. First allocate an unwritten extent and then |
| 4603 |
* convert it to written after zeroing it out. |
| 4604 |
*/ |
| 4605 |
if (flags & EXT4_GET_BLOCKS_ZERO) { |
| 4606 |
flags &= ~EXT4_GET_BLOCKS_ZERO; |
| 4607 |
flags |= EXT4_GET_BLOCKS_UNWRIT_EXT; |
| 4608 |
alloc_zero = true; |
| 4609 |
} |
| 4610 |
|
| 4611 |
/* |
| 4612 |
* credits to insert 1 extent into extent tree |
| 4613 |
*/ |
| 4614 |
credits = ext4_chunk_trans_blocks(inode, len); |
| 4615 |
depth = ext_depth(inode); |
| 4616 |
|
| 4617 |
retry: |
| 4618 |
while (len) { |
| 4619 |
/* |
| 4620 |
* Recalculate credits when extent tree depth changes. |
| 4621 |
*/ |
| 4622 |
if (depth != ext_depth(inode)) { |
| 4623 |
credits = ext4_chunk_trans_blocks(inode, len); |
| 4624 |
depth = ext_depth(inode); |
| 4625 |
} |
| 4626 |
|
| 4627 |
handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, |
| 4628 |
credits); |
| 4629 |
if (IS_ERR(handle)) { |
| 4630 |
ret = PTR_ERR(handle); |
| 4631 |
break; |
| 4632 |
} |
| 4633 |
ret = ext4_map_blocks(handle, inode, &map, flags); |
| 4634 |
if (ret <= 0) { |
| 4635 |
ext4_debug("inode #%lu: block %u: len %u: " |
| 4636 |
"ext4_ext_map_blocks returned %d", |
| 4637 |
inode->i_ino, map.m_lblk, |
| 4638 |
map.m_len, ret); |
| 4639 |
ext4_mark_inode_dirty(handle, inode); |
| 4640 |
ext4_journal_stop(handle); |
| 4641 |
break; |
| 4642 |
} |
| 4643 |
/* |
| 4644 |
* allow a full retry cycle for any remaining allocations |
| 4645 |
*/ |
| 4646 |
retries = 0; |
| 4647 |
epos = EXT4_LBLK_TO_B(inode, map.m_lblk + ret); |
| 4648 |
inode_set_ctime_current(inode); |
| 4649 |
if (new_size) { |
| 4650 |
if (epos > new_size) |
| 4651 |
epos = new_size; |
| 4652 |
if (ext4_update_inode_size(inode, epos) & 0x1) |
| 4653 |
inode_set_mtime_to_ts(inode, |
| 4654 |
inode_get_ctime(inode)); |
| 4655 |
if (epos > old_size) { |
| 4656 |
pagecache_isize_extended(inode, old_size, epos); |
| 4657 |
ext4_zero_partial_blocks(handle, inode, |
| 4658 |
old_size, epos - old_size); |
| 4659 |
} |
| 4660 |
} |
| 4661 |
ret2 = ext4_mark_inode_dirty(handle, inode); |
| 4662 |
ext4_update_inode_fsync_trans(handle, inode, 1); |
| 4663 |
ret3 = ext4_journal_stop(handle); |
| 4664 |
ret2 = ret3 ? ret3 : ret2; |
| 4665 |
if (unlikely(ret2)) |
| 4666 |
break; |
| 4667 |
|
| 4668 |
if (alloc_zero && |
| 4669 |
(map.m_flags & (EXT4_MAP_MAPPED | EXT4_MAP_UNWRITTEN))) { |
| 4670 |
ret2 = ext4_issue_zeroout(inode, map.m_lblk, map.m_pblk, |
| 4671 |
map.m_len); |
| 4672 |
if (likely(!ret2)) |
| 4673 |
ret2 = ext4_convert_unwritten_extents(NULL, |
| 4674 |
inode, (loff_t)map.m_lblk << blkbits, |
| 4675 |
(loff_t)map.m_len << blkbits); |
| 4676 |
if (ret2) |
| 4677 |
break; |
| 4678 |
} |
| 4679 |
|
| 4680 |
map.m_lblk += ret; |
| 4681 |
map.m_len = len = len - ret; |
| 4682 |
} |
| 4683 |
if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) |
| 4684 |
goto retry; |
| 4685 |
|
| 4686 |
return ret > 0 ? ret2 : ret; |
| 4687 |
} |
| 4688 |
|
| 4689 |
static int ext4_collapse_range(struct file *file, loff_t offset, loff_t len); |
| 4690 |
|
| 4691 |
static int ext4_insert_range(struct file *file, loff_t offset, loff_t len); |
| 4692 |
|
| 4693 |
static long ext4_zero_range(struct file *file, loff_t offset, |
| 4694 |
loff_t len, int mode) |
| 4695 |
{ |
| 4696 |
struct inode *inode = file_inode(file); |
| 4697 |
handle_t *handle = NULL; |
| 4698 |
loff_t new_size = 0; |
| 4699 |
loff_t end = offset + len; |
| 4700 |
ext4_lblk_t start_lblk, end_lblk; |
| 4701 |
unsigned int blocksize = i_blocksize(inode); |
| 4702 |
unsigned int blkbits = inode->i_blkbits; |
| 4703 |
int ret, flags, credits; |
| 4704 |
|
| 4705 |
trace_ext4_zero_range(inode, offset, len, mode); |
| 4706 |
WARN_ON_ONCE(!inode_is_locked(inode)); |
| 4707 |
|
| 4708 |
/* Indirect files do not support unwritten extents */ |
| 4709 |
if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) |
| 4710 |
return -EOPNOTSUPP; |
| 4711 |
|
| 4712 |
if (!(mode & FALLOC_FL_KEEP_SIZE) && |
| 4713 |
(end > inode->i_size || end > EXT4_I(inode)->i_disksize)) { |
| 4714 |
new_size = end; |
| 4715 |
ret = inode_newsize_ok(inode, new_size); |
| 4716 |
if (ret) |
| 4717 |
return ret; |
| 4718 |
} |
| 4719 |
|
| 4720 |
flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT; |
| 4721 |
/* Preallocate the range including the unaligned edges */ |
| 4722 |
if (!IS_ALIGNED(offset | end, blocksize)) { |
| 4723 |
ext4_lblk_t alloc_lblk = offset >> blkbits; |
| 4724 |
ext4_lblk_t len_lblk = EXT4_MAX_BLOCKS(len, offset, blkbits); |
| 4725 |
|
| 4726 |
ret = ext4_alloc_file_blocks(file, alloc_lblk, len_lblk, |
| 4727 |
new_size, flags); |
| 4728 |
if (ret) |
| 4729 |
return ret; |
| 4730 |
} |
| 4731 |
|
| 4732 |
ret = ext4_update_disksize_before_punch(inode, offset, len); |
| 4733 |
if (ret) |
| 4734 |
return ret; |
| 4735 |
|
| 4736 |
/* Now release the pages and zero block aligned part of pages */ |
| 4737 |
ret = ext4_truncate_page_cache_block_range(inode, offset, end); |
| 4738 |
if (ret) |
| 4739 |
return ret; |
| 4740 |
|
| 4741 |
/* Zero range excluding the unaligned edges */ |
| 4742 |
start_lblk = EXT4_B_TO_LBLK(inode, offset); |
| 4743 |
end_lblk = end >> blkbits; |
| 4744 |
if (end_lblk > start_lblk) { |
| 4745 |
ext4_lblk_t zero_blks = end_lblk - start_lblk; |
| 4746 |
|
| 4747 |
if (mode & FALLOC_FL_WRITE_ZEROES) |
| 4748 |
flags = EXT4_GET_BLOCKS_CREATE_ZERO | EXT4_EX_NOCACHE; |
| 4749 |
else |
| 4750 |
flags |= (EXT4_GET_BLOCKS_CONVERT_UNWRITTEN | |
| 4751 |
EXT4_EX_NOCACHE); |
| 4752 |
ret = ext4_alloc_file_blocks(file, start_lblk, zero_blks, |
| 4753 |
new_size, flags); |
| 4754 |
if (ret) |
| 4755 |
return ret; |
| 4756 |
} |
| 4757 |
/* Finish zeroing out if it doesn't contain partial block */ |
| 4758 |
if (IS_ALIGNED(offset | end, blocksize)) |
| 4759 |
return ret; |
| 4760 |
|
| 4761 |
/* |
| 4762 |
* In worst case we have to writeout two nonadjacent unwritten |
| 4763 |
* blocks and update the inode |
| 4764 |
*/ |
| 4765 |
credits = (2 * ext4_ext_index_trans_blocks(inode, 2)) + 1; |
| 4766 |
if (ext4_should_journal_data(inode)) |
| 4767 |
credits += 2; |
| 4768 |
handle = ext4_journal_start(inode, EXT4_HT_MISC, credits); |
| 4769 |
if (IS_ERR(handle)) { |
| 4770 |
ret = PTR_ERR(handle); |
| 4771 |
ext4_std_error(inode->i_sb, ret); |
| 4772 |
return ret; |
| 4773 |
} |
| 4774 |
|
| 4775 |
/* Zero out partial block at the edges of the range */ |
| 4776 |
ret = ext4_zero_partial_blocks(handle, inode, offset, len); |
| 4777 |
if (ret) |
| 4778 |
goto out_handle; |
| 4779 |
|
| 4780 |
if (new_size) |
| 4781 |
ext4_update_inode_size(inode, new_size); |
| 4782 |
ret = ext4_mark_inode_dirty(handle, inode); |
| 4783 |
if (unlikely(ret)) |
| 4784 |
goto out_handle; |
| 4785 |
|
| 4786 |
ext4_update_inode_fsync_trans(handle, inode, 1); |
| 4787 |
if (file->f_flags & O_SYNC) |
| 4788 |
ext4_handle_sync(handle); |
| 4789 |
|
| 4790 |
out_handle: |
| 4791 |
ext4_journal_stop(handle); |
| 4792 |
return ret; |
| 4793 |
} |
| 4794 |
|
| 4795 |
static long ext4_do_fallocate(struct file *file, loff_t offset, |
| 4796 |
loff_t len, int mode) |
| 4797 |
{ |
| 4798 |
struct inode *inode = file_inode(file); |
| 4799 |
loff_t end = offset + len; |
| 4800 |
loff_t new_size = 0; |
| 4801 |
ext4_lblk_t start_lblk, len_lblk; |
| 4802 |
int ret; |
| 4803 |
|
| 4804 |
trace_ext4_fallocate_enter(inode, offset, len, mode); |
| 4805 |
WARN_ON_ONCE(!inode_is_locked(inode)); |
| 4806 |
|
| 4807 |
start_lblk = offset >> inode->i_blkbits; |
| 4808 |
len_lblk = EXT4_MAX_BLOCKS(len, offset, inode->i_blkbits); |
| 4809 |
|
| 4810 |
/* We only support preallocation for extent-based files only. */ |
| 4811 |
if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) { |
| 4812 |
ret = -EOPNOTSUPP; |
| 4813 |
goto out; |
| 4814 |
} |
| 4815 |
|
| 4816 |
if (!(mode & FALLOC_FL_KEEP_SIZE) && |
| 4817 |
(end > inode->i_size || end > EXT4_I(inode)->i_disksize)) { |
| 4818 |
new_size = end; |
| 4819 |
ret = inode_newsize_ok(inode, new_size); |
| 4820 |
if (ret) |
| 4821 |
goto out; |
| 4822 |
} |
| 4823 |
|
| 4824 |
ret = ext4_alloc_file_blocks(file, start_lblk, len_lblk, new_size, |
| 4825 |
EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT); |
| 4826 |
if (ret) |
| 4827 |
goto out; |
| 4828 |
|
| 4829 |
if (file->f_flags & O_SYNC && EXT4_SB(inode->i_sb)->s_journal) { |
| 4830 |
ret = ext4_fc_commit(EXT4_SB(inode->i_sb)->s_journal, |
| 4831 |
EXT4_I(inode)->i_sync_tid); |
| 4832 |
} |
| 4833 |
out: |
| 4834 |
trace_ext4_fallocate_exit(inode, offset, len_lblk, ret); |
| 4835 |
return ret; |
| 4836 |
} |
| 4837 |
|
| 4838 |
/* |
| 4839 |
* preallocate space for a file. This implements ext4's fallocate file |
| 4840 |
* operation, which gets called from sys_fallocate system call. |
| 4841 |
* For block-mapped files, posix_fallocate should fall back to the method |
| 4842 |
* of writing zeroes to the required new blocks (the same behavior which is |
| 4843 |
* expected for file systems which do not support fallocate() system call). |
| 4844 |
*/ |
| 4845 |
long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len) |
| 4846 |
{ |
| 4847 |
struct inode *inode = file_inode(file); |
| 4848 |
struct address_space *mapping = file->f_mapping; |
| 4849 |
int ret; |
| 4850 |
|
| 4851 |
/* |
| 4852 |
* Encrypted inodes can't handle collapse range or insert |
| 4853 |
* range since we would need to re-encrypt blocks with a |
| 4854 |
* different IV or XTS tweak (which are based on the logical |
| 4855 |
* block number). |
| 4856 |
*/ |
| 4857 |
if (IS_ENCRYPTED(inode) && |
| 4858 |
(mode & (FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_INSERT_RANGE))) |
| 4859 |
return -EOPNOTSUPP; |
| 4860 |
/* |
| 4861 |
* Don't allow writing zeroes if the underlying device does not |
| 4862 |
* enable the unmap write zeroes operation. |
| 4863 |
*/ |
| 4864 |
if ((mode & FALLOC_FL_WRITE_ZEROES) && |
| 4865 |
!bdev_write_zeroes_unmap_sectors(inode->i_sb->s_bdev)) |
| 4866 |
return -EOPNOTSUPP; |
| 4867 |
|
| 4868 |
/* Return error if mode is not supported */ |
| 4869 |
if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | |
| 4870 |
FALLOC_FL_ZERO_RANGE | FALLOC_FL_COLLAPSE_RANGE | |
| 4871 |
FALLOC_FL_INSERT_RANGE | FALLOC_FL_WRITE_ZEROES)) |
| 4872 |
return -EOPNOTSUPP; |
| 4873 |
|
| 4874 |
inode_lock(inode); |
| 4875 |
ret = ext4_convert_inline_data(inode); |
| 4876 |
if (ret) |
| 4877 |
goto out_inode_lock; |
| 4878 |
|
| 4879 |
/* Wait all existing dio workers, newcomers will block on i_rwsem */ |
| 4880 |
inode_dio_wait(inode); |
| 4881 |
|
| 4882 |
ret = file_modified(file); |
| 4883 |
if (ret) |
| 4884 |
goto out_inode_lock; |
| 4885 |
|
| 4886 |
if ((mode & FALLOC_FL_MODE_MASK) == FALLOC_FL_ALLOCATE_RANGE) { |
| 4887 |
ret = ext4_do_fallocate(file, offset, len, mode); |
| 4888 |
goto out_inode_lock; |
| 4889 |
} |
| 4890 |
|
| 4891 |
/* |
| 4892 |
* Follow-up operations will drop page cache, hold invalidate lock |
| 4893 |
* to prevent page faults from reinstantiating pages we have |
| 4894 |
* released from page cache. |
| 4895 |
*/ |
| 4896 |
filemap_invalidate_lock(mapping); |
| 4897 |
|
| 4898 |
ret = ext4_break_layouts(inode); |
| 4899 |
if (ret) |
| 4900 |
goto out_invalidate_lock; |
| 4901 |
|
| 4902 |
switch (mode & FALLOC_FL_MODE_MASK) { |
| 4903 |
case FALLOC_FL_PUNCH_HOLE: |
| 4904 |
ret = ext4_punch_hole(file, offset, len); |
| 4905 |
break; |
| 4906 |
case FALLOC_FL_COLLAPSE_RANGE: |
| 4907 |
ret = ext4_collapse_range(file, offset, len); |
| 4908 |
break; |
| 4909 |
case FALLOC_FL_INSERT_RANGE: |
| 4910 |
ret = ext4_insert_range(file, offset, len); |
| 4911 |
break; |
| 4912 |
case FALLOC_FL_ZERO_RANGE: |
| 4913 |
case FALLOC_FL_WRITE_ZEROES: |
| 4914 |
ret = ext4_zero_range(file, offset, len, mode); |
| 4915 |
break; |
| 4916 |
default: |
| 4917 |
ret = -EOPNOTSUPP; |
| 4918 |
} |
| 4919 |
|
| 4920 |
out_invalidate_lock: |
| 4921 |
filemap_invalidate_unlock(mapping); |
| 4922 |
out_inode_lock: |
| 4923 |
inode_unlock(inode); |
| 4924 |
return ret; |
| 4925 |
} |
| 4926 |
|
| 4927 |
/* |
| 4928 |
* This function converts a range of blocks to written extents. The caller of |
| 4929 |
* this function will pass the start offset and the size. all unwritten extents |
| 4930 |
* within this range will be converted to written extents. |
| 4931 |
* |
| 4932 |
* This function is called from the direct IO end io call back function for |
| 4933 |
* atomic writes, to convert the unwritten extents after IO is completed. |
| 4934 |
* |
| 4935 |
* Note that the requirement for atomic writes is that all conversion should |
| 4936 |
* happen atomically in a single fs journal transaction. We mainly only allocate |
| 4937 |
* unwritten extents either on a hole on a pre-exiting unwritten extent range in |
| 4938 |
* ext4_map_blocks_atomic_write(). The only case where we can have multiple |
| 4939 |
* unwritten extents in a range [offset, offset+len) is when there is a split |
| 4940 |
* unwritten extent between two leaf nodes which was cached in extent status |
| 4941 |
* cache during ext4_iomap_alloc() time. That will allow |
| 4942 |
* ext4_map_blocks_atomic_write() to return the unwritten extent range w/o going |
| 4943 |
* into the slow path. That means we might need a loop for conversion of this |
| 4944 |
* unwritten extent split across leaf block within a single journal transaction. |
| 4945 |
* Split extents across leaf nodes is a rare case, but let's still handle that |
| 4946 |
* to meet the requirements of multi-fsblock atomic writes. |
| 4947 |
* |
| 4948 |
* Returns 0 on success. |
| 4949 |
*/ |
| 4950 |
int ext4_convert_unwritten_extents_atomic(handle_t *handle, struct inode *inode, |
| 4951 |
loff_t offset, ssize_t len) |
| 4952 |
{ |
| 4953 |
unsigned int max_blocks; |
| 4954 |
int ret = 0, ret2 = 0, ret3 = 0; |
| 4955 |
struct ext4_map_blocks map; |
| 4956 |
unsigned int blkbits = inode->i_blkbits; |
| 4957 |
unsigned int credits = 0; |
| 4958 |
int flags = EXT4_GET_BLOCKS_IO_CONVERT_EXT | EXT4_EX_NOCACHE; |
| 4959 |
|
| 4960 |
map.m_lblk = offset >> blkbits; |
| 4961 |
max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits); |
| 4962 |
|
| 4963 |
if (!handle) { |
| 4964 |
/* |
| 4965 |
* TODO: An optimization can be added later by having an extent |
| 4966 |
* status flag e.g. EXTENT_STATUS_SPLIT_LEAF. If we query that |
| 4967 |
* it can tell if the extent in the cache is a split extent. |
| 4968 |
* But for now let's assume pextents as 2 always. |
| 4969 |
*/ |
| 4970 |
credits = ext4_meta_trans_blocks(inode, max_blocks, 2); |
| 4971 |
} |
| 4972 |
|
| 4973 |
if (credits) { |
| 4974 |
handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, credits); |
| 4975 |
if (IS_ERR(handle)) { |
| 4976 |
ret = PTR_ERR(handle); |
| 4977 |
return ret; |
| 4978 |
} |
| 4979 |
} |
| 4980 |
|
| 4981 |
while (ret >= 0 && ret < max_blocks) { |
| 4982 |
map.m_lblk += ret; |
| 4983 |
map.m_len = (max_blocks -= ret); |
| 4984 |
ret = ext4_map_blocks(handle, inode, &map, flags); |
| 4985 |
if (ret != max_blocks) |
| 4986 |
ext4_msg(inode->i_sb, KERN_INFO, |
| 4987 |
"inode #%lu: block %u: len %u: " |
| 4988 |
"split block mapping found for atomic write, " |
| 4989 |
"ret = %d", |
| 4990 |
inode->i_ino, map.m_lblk, |
| 4991 |
map.m_len, ret); |
| 4992 |
if (ret <= 0) |
| 4993 |
break; |
| 4994 |
} |
| 4995 |
|
| 4996 |
ret2 = ext4_mark_inode_dirty(handle, inode); |
| 4997 |
|
| 4998 |
if (credits) { |
| 4999 |
ret3 = ext4_journal_stop(handle); |
| 5000 |
if (unlikely(ret3)) |
| 5001 |
ret2 = ret3; |
| 5002 |
} |
| 5003 |
|
| 5004 |
if (ret <= 0 || ret2) |
| 5005 |
ext4_warning(inode->i_sb, |
| 5006 |
"inode #%lu: block %u: len %u: " |
| 5007 |
"returned %d or %d", |
| 5008 |
inode->i_ino, map.m_lblk, |
| 5009 |
map.m_len, ret, ret2); |
| 5010 |
|
| 5011 |
return ret > 0 ? ret2 : ret; |
| 5012 |
} |
| 5013 |
|
| 5014 |
/* |
| 5015 |
* This function convert a range of blocks to written extents |
| 5016 |
* The caller of this function will pass the start offset and the size. |
| 5017 |
* all unwritten extents within this range will be converted to |
| 5018 |
* written extents. |
| 5019 |
* |
| 5020 |
* This function is called from the direct IO end io call back |
| 5021 |
* function, to convert the fallocated extents after IO is completed. |
| 5022 |
* Returns 0 on success. |
| 5023 |
*/ |
| 5024 |
int ext4_convert_unwritten_extents(handle_t *handle, struct inode *inode, |
| 5025 |
loff_t offset, ssize_t len) |
| 5026 |
{ |
| 5027 |
unsigned int max_blocks; |
| 5028 |
int ret = 0, ret2 = 0, ret3 = 0; |
| 5029 |
struct ext4_map_blocks map; |
| 5030 |
unsigned int blkbits = inode->i_blkbits; |
| 5031 |
unsigned int credits = 0; |
| 5032 |
|
| 5033 |
map.m_lblk = offset >> blkbits; |
| 5034 |
max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits); |
| 5035 |
|
| 5036 |
if (!handle) { |
| 5037 |
/* |
| 5038 |
* credits to insert 1 extent into extent tree |
| 5039 |
*/ |
| 5040 |
credits = ext4_chunk_trans_blocks(inode, max_blocks); |
| 5041 |
} |
| 5042 |
while (ret >= 0 && ret < max_blocks) { |
| 5043 |
map.m_lblk += ret; |
| 5044 |
map.m_len = (max_blocks -= ret); |
| 5045 |
if (credits) { |
| 5046 |
handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, |
| 5047 |
credits); |
| 5048 |
if (IS_ERR(handle)) { |
| 5049 |
ret = PTR_ERR(handle); |
| 5050 |
break; |
| 5051 |
} |
| 5052 |
} |
| 5053 |
/* |
| 5054 |
* Do not cache any unrelated extents, as it does not hold the |
| 5055 |
* i_rwsem or invalidate_lock, which could corrupt the extent |
| 5056 |
* status tree. |
| 5057 |
*/ |
| 5058 |
ret = ext4_map_blocks(handle, inode, &map, |
| 5059 |
EXT4_GET_BLOCKS_IO_CONVERT_EXT | |
| 5060 |
EXT4_EX_NOCACHE); |
| 5061 |
if (ret <= 0) |
| 5062 |
ext4_warning(inode->i_sb, |
| 5063 |
"inode #%lu: block %u: len %u: " |
| 5064 |
"ext4_ext_map_blocks returned %d", |
| 5065 |
inode->i_ino, map.m_lblk, |
| 5066 |
map.m_len, ret); |
| 5067 |
ret2 = ext4_mark_inode_dirty(handle, inode); |
| 5068 |
if (credits) { |
| 5069 |
ret3 = ext4_journal_stop(handle); |
| 5070 |
if (unlikely(ret3)) |
| 5071 |
ret2 = ret3; |
| 5072 |
} |
| 5073 |
|
| 5074 |
if (ret <= 0 || ret2) |
| 5075 |
break; |
| 5076 |
} |
| 5077 |
return ret > 0 ? ret2 : ret; |
| 5078 |
} |
| 5079 |
|
| 5080 |
int ext4_convert_unwritten_io_end_vec(handle_t *handle, ext4_io_end_t *io_end) |
| 5081 |
{ |
| 5082 |
int ret = 0, err = 0; |
| 5083 |
struct ext4_io_end_vec *io_end_vec; |
| 5084 |
|
| 5085 |
/* |
| 5086 |
* This is somewhat ugly but the idea is clear: When transaction is |
| 5087 |
* reserved, everything goes into it. Otherwise we rather start several |
| 5088 |
* smaller transactions for conversion of each extent separately. |
| 5089 |
*/ |
| 5090 |
if (handle) { |
| 5091 |
handle = ext4_journal_start_reserved(handle, |
| 5092 |
EXT4_HT_EXT_CONVERT); |
| 5093 |
if (IS_ERR(handle)) |
| 5094 |
return PTR_ERR(handle); |
| 5095 |
} |
| 5096 |
|
| 5097 |
list_for_each_entry(io_end_vec, &io_end->list_vec, list) { |
| 5098 |
ret = ext4_convert_unwritten_extents(handle, io_end->inode, |
| 5099 |
io_end_vec->offset, |
| 5100 |
io_end_vec->size); |
| 5101 |
if (ret) |
| 5102 |
break; |
| 5103 |
} |
| 5104 |
|
| 5105 |
if (handle) |
| 5106 |
err = ext4_journal_stop(handle); |
| 5107 |
|
| 5108 |
return ret < 0 ? ret : err; |
| 5109 |
} |
| 5110 |
|
| 5111 |
static int ext4_iomap_xattr_fiemap(struct inode *inode, struct iomap *iomap) |
| 5112 |
{ |
| 5113 |
__u64 physical = 0; |
| 5114 |
__u64 length = 0; |
| 5115 |
int blockbits = inode->i_sb->s_blocksize_bits; |
| 5116 |
int error = 0; |
| 5117 |
u16 iomap_type; |
| 5118 |
|
| 5119 |
/* in-inode? */ |
| 5120 |
if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) { |
| 5121 |
struct ext4_iloc iloc; |
| 5122 |
int offset; /* offset of xattr in inode */ |
| 5123 |
|
| 5124 |
error = ext4_get_inode_loc(inode, &iloc); |
| 5125 |
if (error) |
| 5126 |
return error; |
| 5127 |
physical = (__u64)iloc.bh->b_blocknr << blockbits; |
| 5128 |
offset = EXT4_GOOD_OLD_INODE_SIZE + |
| 5129 |
EXT4_I(inode)->i_extra_isize; |
| 5130 |
physical += offset; |
| 5131 |
length = EXT4_SB(inode->i_sb)->s_inode_size - offset; |
| 5132 |
brelse(iloc.bh); |
| 5133 |
iomap_type = IOMAP_INLINE; |
| 5134 |
} else if (EXT4_I(inode)->i_file_acl) { /* external block */ |
| 5135 |
physical = (__u64)EXT4_I(inode)->i_file_acl << blockbits; |
| 5136 |
length = inode->i_sb->s_blocksize; |
| 5137 |
iomap_type = IOMAP_MAPPED; |
| 5138 |
} else { |
| 5139 |
/* no in-inode or external block for xattr, so return -ENOENT */ |
| 5140 |
error = -ENOENT; |
| 5141 |
goto out; |
| 5142 |
} |
| 5143 |
|
| 5144 |
iomap->addr = physical; |
| 5145 |
iomap->offset = 0; |
| 5146 |
iomap->length = length; |
| 5147 |
iomap->type = iomap_type; |
| 5148 |
iomap->flags = 0; |
| 5149 |
out: |
| 5150 |
return error; |
| 5151 |
} |
| 5152 |
|
| 5153 |
static int ext4_iomap_xattr_begin(struct inode *inode, loff_t offset, |
| 5154 |
loff_t length, unsigned flags, |
| 5155 |
struct iomap *iomap, struct iomap *srcmap) |
| 5156 |
{ |
| 5157 |
int error; |
| 5158 |
|
| 5159 |
error = ext4_iomap_xattr_fiemap(inode, iomap); |
| 5160 |
if (error == 0 && (offset >= iomap->length)) |
| 5161 |
error = -ENOENT; |
| 5162 |
return error; |
| 5163 |
} |
| 5164 |
|
| 5165 |
static const struct iomap_ops ext4_iomap_xattr_ops = { |
| 5166 |
.iomap_begin = ext4_iomap_xattr_begin, |
| 5167 |
}; |
| 5168 |
|
| 5169 |
static int ext4_fiemap_check_ranges(struct inode *inode, u64 start, u64 *len) |
| 5170 |
{ |
| 5171 |
u64 maxbytes = ext4_get_maxbytes(inode); |
| 5172 |
|
| 5173 |
if (*len == 0) |
| 5174 |
return -EINVAL; |
| 5175 |
if (start > maxbytes) |
| 5176 |
return -EFBIG; |
| 5177 |
|
| 5178 |
/* |
| 5179 |
* Shrink request scope to what the fs can actually handle. |
| 5180 |
*/ |
| 5181 |
if (*len > maxbytes || (maxbytes - *len) < start) |
| 5182 |
*len = maxbytes - start; |
| 5183 |
return 0; |
| 5184 |
} |
| 5185 |
|
| 5186 |
int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
| 5187 |
u64 start, u64 len) |
| 5188 |
{ |
| 5189 |
int error = 0; |
| 5190 |
|
| 5191 |
inode_lock_shared(inode); |
| 5192 |
if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) { |
| 5193 |
error = ext4_ext_precache(inode); |
| 5194 |
if (error) |
| 5195 |
goto unlock; |
| 5196 |
fieinfo->fi_flags &= ~FIEMAP_FLAG_CACHE; |
| 5197 |
} |
| 5198 |
|
| 5199 |
/* |
| 5200 |
* For bitmap files the maximum size limit could be smaller than |
| 5201 |
* s_maxbytes, so check len here manually instead of just relying on the |
| 5202 |
* generic check. |
| 5203 |
*/ |
| 5204 |
error = ext4_fiemap_check_ranges(inode, start, &len); |
| 5205 |
if (error) |
| 5206 |
goto unlock; |
| 5207 |
|
| 5208 |
if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) { |
| 5209 |
fieinfo->fi_flags &= ~FIEMAP_FLAG_XATTR; |
| 5210 |
error = iomap_fiemap(inode, fieinfo, start, len, |
| 5211 |
&ext4_iomap_xattr_ops); |
| 5212 |
} else { |
| 5213 |
error = iomap_fiemap(inode, fieinfo, start, len, |
| 5214 |
&ext4_iomap_report_ops); |
| 5215 |
} |
| 5216 |
unlock: |
| 5217 |
inode_unlock_shared(inode); |
| 5218 |
return error; |
| 5219 |
} |
| 5220 |
|
| 5221 |
int ext4_get_es_cache(struct inode *inode, struct fiemap_extent_info *fieinfo, |
| 5222 |
__u64 start, __u64 len) |
| 5223 |
{ |
| 5224 |
ext4_lblk_t start_blk, len_blks; |
| 5225 |
__u64 last_blk; |
| 5226 |
int error = 0; |
| 5227 |
|
| 5228 |
if (ext4_has_inline_data(inode)) { |
| 5229 |
int has_inline; |
| 5230 |
|
| 5231 |
down_read(&EXT4_I(inode)->xattr_sem); |
| 5232 |
has_inline = ext4_has_inline_data(inode); |
| 5233 |
up_read(&EXT4_I(inode)->xattr_sem); |
| 5234 |
if (has_inline) |
| 5235 |
return 0; |
| 5236 |
} |
| 5237 |
|
| 5238 |
if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) { |
| 5239 |
inode_lock_shared(inode); |
| 5240 |
error = ext4_ext_precache(inode); |
| 5241 |
inode_unlock_shared(inode); |
| 5242 |
if (error) |
| 5243 |
return error; |
| 5244 |
fieinfo->fi_flags &= ~FIEMAP_FLAG_CACHE; |
| 5245 |
} |
| 5246 |
|
| 5247 |
error = fiemap_prep(inode, fieinfo, start, &len, 0); |
| 5248 |
if (error) |
| 5249 |
return error; |
| 5250 |
|
| 5251 |
error = ext4_fiemap_check_ranges(inode, start, &len); |
| 5252 |
if (error) |
| 5253 |
return error; |
| 5254 |
|
| 5255 |
start_blk = start >> inode->i_sb->s_blocksize_bits; |
| 5256 |
last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits; |
| 5257 |
if (last_blk >= EXT_MAX_BLOCKS) |
| 5258 |
last_blk = EXT_MAX_BLOCKS-1; |
| 5259 |
len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1; |
| 5260 |
|
| 5261 |
/* |
| 5262 |
* Walk the extent tree gathering extent information |
| 5263 |
* and pushing extents back to the user. |
| 5264 |
*/ |
| 5265 |
return ext4_fill_es_cache_info(inode, start_blk, len_blks, fieinfo); |
| 5266 |
} |
| 5267 |
|
| 5268 |
/* |
| 5269 |
* ext4_ext_shift_path_extents: |
| 5270 |
* Shift the extents of a path structure lying between path[depth].p_ext |
| 5271 |
* and EXT_LAST_EXTENT(path[depth].p_hdr), by @shift blocks. @SHIFT tells |
| 5272 |
* if it is right shift or left shift operation. |
| 5273 |
*/ |
| 5274 |
static int |
| 5275 |
ext4_ext_shift_path_extents(struct ext4_ext_path *path, ext4_lblk_t shift, |
| 5276 |
struct inode *inode, handle_t *handle, |
| 5277 |
enum SHIFT_DIRECTION SHIFT) |
| 5278 |
{ |
| 5279 |
int depth, err = 0; |
| 5280 |
struct ext4_extent *ex_start, *ex_last; |
| 5281 |
bool update = false; |
| 5282 |
int credits, restart_credits; |
| 5283 |
depth = path->p_depth; |
| 5284 |
|
| 5285 |
while (depth >= 0) { |
| 5286 |
if (depth == path->p_depth) { |
| 5287 |
ex_start = path[depth].p_ext; |
| 5288 |
if (!ex_start) |
| 5289 |
return -EFSCORRUPTED; |
| 5290 |
|
| 5291 |
ex_last = EXT_LAST_EXTENT(path[depth].p_hdr); |
| 5292 |
/* leaf + sb + inode */ |
| 5293 |
credits = 3; |
| 5294 |
if (ex_start == EXT_FIRST_EXTENT(path[depth].p_hdr)) { |
| 5295 |
update = true; |
| 5296 |
/* extent tree + sb + inode */ |
| 5297 |
credits = depth + 2; |
| 5298 |
} |
| 5299 |
|
| 5300 |
restart_credits = ext4_chunk_trans_extent(inode, 0); |
| 5301 |
err = ext4_datasem_ensure_credits(handle, inode, credits, |
| 5302 |
restart_credits, 0); |
| 5303 |
if (err) { |
| 5304 |
if (err > 0) |
| 5305 |
err = -EAGAIN; |
| 5306 |
goto out; |
| 5307 |
} |
| 5308 |
|
| 5309 |
err = ext4_ext_get_access(handle, inode, path + depth); |
| 5310 |
if (err) |
| 5311 |
goto out; |
| 5312 |
|
| 5313 |
while (ex_start <= ex_last) { |
| 5314 |
if (SHIFT == SHIFT_LEFT) { |
| 5315 |
le32_add_cpu(&ex_start->ee_block, |
| 5316 |
-shift); |
| 5317 |
/* Try to merge to the left. */ |
| 5318 |
if ((ex_start > |
| 5319 |
EXT_FIRST_EXTENT(path[depth].p_hdr)) |
| 5320 |
&& |
| 5321 |
ext4_ext_try_to_merge_right(inode, |
| 5322 |
path, ex_start - 1)) |
| 5323 |
ex_last--; |
| 5324 |
else |
| 5325 |
ex_start++; |
| 5326 |
} else { |
| 5327 |
le32_add_cpu(&ex_last->ee_block, shift); |
| 5328 |
ext4_ext_try_to_merge_right(inode, path, |
| 5329 |
ex_last); |
| 5330 |
ex_last--; |
| 5331 |
} |
| 5332 |
} |
| 5333 |
err = ext4_ext_dirty(handle, inode, path + depth); |
| 5334 |
if (err) |
| 5335 |
goto out; |
| 5336 |
|
| 5337 |
if (--depth < 0 || !update) |
| 5338 |
break; |
| 5339 |
} |
| 5340 |
|
| 5341 |
/* Update index too */ |
| 5342 |
err = ext4_ext_get_access(handle, inode, path + depth); |
| 5343 |
if (err) |
| 5344 |
goto out; |
| 5345 |
|
| 5346 |
if (SHIFT == SHIFT_LEFT) |
| 5347 |
le32_add_cpu(&path[depth].p_idx->ei_block, -shift); |
| 5348 |
else |
| 5349 |
le32_add_cpu(&path[depth].p_idx->ei_block, shift); |
| 5350 |
err = ext4_ext_dirty(handle, inode, path + depth); |
| 5351 |
if (err) |
| 5352 |
goto out; |
| 5353 |
|
| 5354 |
/* we are done if current index is not a starting index */ |
| 5355 |
if (path[depth].p_idx != EXT_FIRST_INDEX(path[depth].p_hdr)) |
| 5356 |
break; |
| 5357 |
|
| 5358 |
depth--; |
| 5359 |
} |
| 5360 |
|
| 5361 |
out: |
| 5362 |
return err; |
| 5363 |
} |
| 5364 |
|
| 5365 |
/* |
| 5366 |
* ext4_ext_shift_extents: |
| 5367 |
* All the extents which lies in the range from @start to the last allocated |
| 5368 |
* block for the @inode are shifted either towards left or right (depending |
| 5369 |
* upon @SHIFT) by @shift blocks. |
| 5370 |
* On success, 0 is returned, error otherwise. |
| 5371 |
*/ |
| 5372 |
static int |
| 5373 |
ext4_ext_shift_extents(struct inode *inode, handle_t *handle, |
| 5374 |
ext4_lblk_t start, ext4_lblk_t shift, |
| 5375 |
enum SHIFT_DIRECTION SHIFT) |
| 5376 |
{ |
| 5377 |
struct ext4_ext_path *path; |
| 5378 |
int ret = 0, depth; |
| 5379 |
struct ext4_extent *extent; |
| 5380 |
ext4_lblk_t stop, *iterator, ex_start, ex_end; |
| 5381 |
ext4_lblk_t tmp = EXT_MAX_BLOCKS; |
| 5382 |
|
| 5383 |
/* Let path point to the last extent */ |
| 5384 |
path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL, |
| 5385 |
EXT4_EX_NOCACHE); |
| 5386 |
if (IS_ERR(path)) |
| 5387 |
return PTR_ERR(path); |
| 5388 |
|
| 5389 |
depth = path->p_depth; |
| 5390 |
extent = path[depth].p_ext; |
| 5391 |
if (!extent) |
| 5392 |
goto out; |
| 5393 |
|
| 5394 |
stop = le32_to_cpu(extent->ee_block); |
| 5395 |
|
| 5396 |
/* |
| 5397 |
* For left shifts, make sure the hole on the left is big enough to |
| 5398 |
* accommodate the shift. For right shifts, make sure the last extent |
| 5399 |
* won't be shifted beyond EXT_MAX_BLOCKS. |
| 5400 |
*/ |
| 5401 |
if (SHIFT == SHIFT_LEFT) { |
| 5402 |
path = ext4_find_extent(inode, start - 1, path, |
| 5403 |
EXT4_EX_NOCACHE); |
| 5404 |
if (IS_ERR(path)) |
| 5405 |
return PTR_ERR(path); |
| 5406 |
depth = path->p_depth; |
| 5407 |
extent = path[depth].p_ext; |
| 5408 |
if (extent) { |
| 5409 |
ex_start = le32_to_cpu(extent->ee_block); |
| 5410 |
ex_end = le32_to_cpu(extent->ee_block) + |
| 5411 |
ext4_ext_get_actual_len(extent); |
| 5412 |
} else { |
| 5413 |
ex_start = 0; |
| 5414 |
ex_end = 0; |
| 5415 |
} |
| 5416 |
|
| 5417 |
if ((start == ex_start && shift > ex_start) || |
| 5418 |
(shift > start - ex_end)) { |
| 5419 |
ret = -EINVAL; |
| 5420 |
goto out; |
| 5421 |
} |
| 5422 |
} else { |
| 5423 |
if (shift > EXT_MAX_BLOCKS - |
| 5424 |
(stop + ext4_ext_get_actual_len(extent))) { |
| 5425 |
ret = -EINVAL; |
| 5426 |
goto out; |
| 5427 |
} |
| 5428 |
} |
| 5429 |
|
| 5430 |
/* |
| 5431 |
* In case of left shift, iterator points to start and it is increased |
| 5432 |
* till we reach stop. In case of right shift, iterator points to stop |
| 5433 |
* and it is decreased till we reach start. |
| 5434 |
*/ |
| 5435 |
again: |
| 5436 |
ret = 0; |
| 5437 |
if (SHIFT == SHIFT_LEFT) |
| 5438 |
iterator = &start; |
| 5439 |
else |
| 5440 |
iterator = &stop; |
| 5441 |
|
| 5442 |
if (tmp != EXT_MAX_BLOCKS) |
| 5443 |
*iterator = tmp; |
| 5444 |
|
| 5445 |
/* |
| 5446 |
* Its safe to start updating extents. Start and stop are unsigned, so |
| 5447 |
* in case of right shift if extent with 0 block is reached, iterator |
| 5448 |
* becomes NULL to indicate the end of the loop. |
| 5449 |
*/ |
| 5450 |
while (iterator && start <= stop) { |
| 5451 |
path = ext4_find_extent(inode, *iterator, path, |
| 5452 |
EXT4_EX_NOCACHE); |
| 5453 |
if (IS_ERR(path)) |
| 5454 |
return PTR_ERR(path); |
| 5455 |
depth = path->p_depth; |
| 5456 |
extent = path[depth].p_ext; |
| 5457 |
if (!extent) { |
| 5458 |
EXT4_ERROR_INODE(inode, "unexpected hole at %lu", |
| 5459 |
(unsigned long) *iterator); |
| 5460 |
ret = -EFSCORRUPTED; |
| 5461 |
goto out; |
| 5462 |
} |
| 5463 |
if (SHIFT == SHIFT_LEFT && *iterator > |
| 5464 |
le32_to_cpu(extent->ee_block)) { |
| 5465 |
/* Hole, move to the next extent */ |
| 5466 |
if (extent < EXT_LAST_EXTENT(path[depth].p_hdr)) { |
| 5467 |
path[depth].p_ext++; |
| 5468 |
} else { |
| 5469 |
*iterator = ext4_ext_next_allocated_block(path); |
| 5470 |
continue; |
| 5471 |
} |
| 5472 |
} |
| 5473 |
|
| 5474 |
tmp = *iterator; |
| 5475 |
if (SHIFT == SHIFT_LEFT) { |
| 5476 |
extent = EXT_LAST_EXTENT(path[depth].p_hdr); |
| 5477 |
*iterator = le32_to_cpu(extent->ee_block) + |
| 5478 |
ext4_ext_get_actual_len(extent); |
| 5479 |
} else { |
| 5480 |
extent = EXT_FIRST_EXTENT(path[depth].p_hdr); |
| 5481 |
if (le32_to_cpu(extent->ee_block) > start) |
| 5482 |
*iterator = le32_to_cpu(extent->ee_block) - 1; |
| 5483 |
else if (le32_to_cpu(extent->ee_block) == start) |
| 5484 |
iterator = NULL; |
| 5485 |
else { |
| 5486 |
extent = EXT_LAST_EXTENT(path[depth].p_hdr); |
| 5487 |
while (le32_to_cpu(extent->ee_block) >= start) |
| 5488 |
extent--; |
| 5489 |
|
| 5490 |
if (extent == EXT_LAST_EXTENT(path[depth].p_hdr)) |
| 5491 |
break; |
| 5492 |
|
| 5493 |
extent++; |
| 5494 |
iterator = NULL; |
| 5495 |
} |
| 5496 |
path[depth].p_ext = extent; |
| 5497 |
} |
| 5498 |
ret = ext4_ext_shift_path_extents(path, shift, inode, |
| 5499 |
handle, SHIFT); |
| 5500 |
/* iterator can be NULL which means we should break */ |
| 5501 |
if (ret == -EAGAIN) |
| 5502 |
goto again; |
| 5503 |
if (ret) |
| 5504 |
break; |
| 5505 |
} |
| 5506 |
out: |
| 5507 |
ext4_free_ext_path(path); |
| 5508 |
return ret; |
| 5509 |
} |
| 5510 |
|
| 5511 |
/* |
| 5512 |
* ext4_collapse_range: |
| 5513 |
* This implements the fallocate's collapse range functionality for ext4 |
| 5514 |
* Returns: 0 and non-zero on error. |
| 5515 |
*/ |
| 5516 |
static int ext4_collapse_range(struct file *file, loff_t offset, loff_t len) |
| 5517 |
{ |
| 5518 |
struct inode *inode = file_inode(file); |
| 5519 |
struct super_block *sb = inode->i_sb; |
| 5520 |
struct address_space *mapping = inode->i_mapping; |
| 5521 |
loff_t end = offset + len; |
| 5522 |
ext4_lblk_t start_lblk, end_lblk; |
| 5523 |
handle_t *handle; |
| 5524 |
unsigned int credits; |
| 5525 |
loff_t start, new_size; |
| 5526 |
int ret; |
| 5527 |
|
| 5528 |
trace_ext4_collapse_range(inode, offset, len); |
| 5529 |
WARN_ON_ONCE(!inode_is_locked(inode)); |
| 5530 |
|
| 5531 |
/* Currently just for extent based files */ |
| 5532 |
if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
| 5533 |
return -EOPNOTSUPP; |
| 5534 |
/* Collapse range works only on fs cluster size aligned regions. */ |
| 5535 |
if (!IS_ALIGNED(offset | len, EXT4_CLUSTER_SIZE(sb))) |
| 5536 |
return -EINVAL; |
| 5537 |
/* |
| 5538 |
* There is no need to overlap collapse range with EOF, in which case |
| 5539 |
* it is effectively a truncate operation |
| 5540 |
*/ |
| 5541 |
if (end >= inode->i_size) |
| 5542 |
return -EINVAL; |
| 5543 |
|
| 5544 |
/* |
| 5545 |
* Write tail of the last page before removed range and data that |
| 5546 |
* will be shifted since they will get removed from the page cache |
| 5547 |
* below. We are also protected from pages becoming dirty by |
| 5548 |
* i_rwsem and invalidate_lock. |
| 5549 |
* Need to round down offset to be aligned with page size boundary |
| 5550 |
* for page size > block size. |
| 5551 |
*/ |
| 5552 |
start = round_down(offset, PAGE_SIZE); |
| 5553 |
ret = filemap_write_and_wait_range(mapping, start, offset); |
| 5554 |
if (!ret) |
| 5555 |
ret = filemap_write_and_wait_range(mapping, end, LLONG_MAX); |
| 5556 |
if (ret) |
| 5557 |
return ret; |
| 5558 |
|
| 5559 |
truncate_pagecache(inode, start); |
| 5560 |
|
| 5561 |
credits = ext4_chunk_trans_extent(inode, 0); |
| 5562 |
handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); |
| 5563 |
if (IS_ERR(handle)) |
| 5564 |
return PTR_ERR(handle); |
| 5565 |
|
| 5566 |
ext4_fc_mark_ineligible(sb, EXT4_FC_REASON_FALLOC_RANGE, handle); |
| 5567 |
|
| 5568 |
start_lblk = offset >> inode->i_blkbits; |
| 5569 |
end_lblk = (offset + len) >> inode->i_blkbits; |
| 5570 |
|
| 5571 |
ext4_check_map_extents_env(inode); |
| 5572 |
|
| 5573 |
down_write(&EXT4_I(inode)->i_data_sem); |
| 5574 |
ext4_discard_preallocations(inode); |
| 5575 |
ext4_es_remove_extent(inode, start_lblk, EXT_MAX_BLOCKS - start_lblk); |
| 5576 |
|
| 5577 |
ret = ext4_ext_remove_space(inode, start_lblk, end_lblk - 1); |
| 5578 |
if (ret) { |
| 5579 |
up_write(&EXT4_I(inode)->i_data_sem); |
| 5580 |
goto out_handle; |
| 5581 |
} |
| 5582 |
ext4_discard_preallocations(inode); |
| 5583 |
|
| 5584 |
ret = ext4_ext_shift_extents(inode, handle, end_lblk, |
| 5585 |
end_lblk - start_lblk, SHIFT_LEFT); |
| 5586 |
if (ret) { |
| 5587 |
up_write(&EXT4_I(inode)->i_data_sem); |
| 5588 |
goto out_handle; |
| 5589 |
} |
| 5590 |
|
| 5591 |
new_size = inode->i_size - len; |
| 5592 |
i_size_write(inode, new_size); |
| 5593 |
EXT4_I(inode)->i_disksize = new_size; |
| 5594 |
|
| 5595 |
up_write(&EXT4_I(inode)->i_data_sem); |
| 5596 |
ret = ext4_mark_inode_dirty(handle, inode); |
| 5597 |
if (ret) |
| 5598 |
goto out_handle; |
| 5599 |
|
| 5600 |
ext4_update_inode_fsync_trans(handle, inode, 1); |
| 5601 |
if (IS_SYNC(inode)) |
| 5602 |
ext4_handle_sync(handle); |
| 5603 |
|
| 5604 |
out_handle: |
| 5605 |
ext4_journal_stop(handle); |
| 5606 |
return ret; |
| 5607 |
} |
| 5608 |
|
| 5609 |
/* |
| 5610 |
* ext4_insert_range: |
| 5611 |
* This function implements the FALLOC_FL_INSERT_RANGE flag of fallocate. |
| 5612 |
* The data blocks starting from @offset to the EOF are shifted by @len |
| 5613 |
* towards right to create a hole in the @inode. Inode size is increased |
| 5614 |
* by len bytes. |
| 5615 |
* Returns 0 on success, error otherwise. |
| 5616 |
*/ |
| 5617 |
static int ext4_insert_range(struct file *file, loff_t offset, loff_t len) |
| 5618 |
{ |
| 5619 |
struct inode *inode = file_inode(file); |
| 5620 |
struct super_block *sb = inode->i_sb; |
| 5621 |
struct address_space *mapping = inode->i_mapping; |
| 5622 |
handle_t *handle; |
| 5623 |
struct ext4_ext_path *path; |
| 5624 |
struct ext4_extent *extent; |
| 5625 |
ext4_lblk_t start_lblk, len_lblk, ee_start_lblk = 0; |
| 5626 |
unsigned int credits, ee_len; |
| 5627 |
int ret, depth; |
| 5628 |
loff_t start; |
| 5629 |
|
| 5630 |
trace_ext4_insert_range(inode, offset, len); |
| 5631 |
WARN_ON_ONCE(!inode_is_locked(inode)); |
| 5632 |
|
| 5633 |
/* Currently just for extent based files */ |
| 5634 |
if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
| 5635 |
return -EOPNOTSUPP; |
| 5636 |
/* Insert range works only on fs cluster size aligned regions. */ |
| 5637 |
if (!IS_ALIGNED(offset | len, EXT4_CLUSTER_SIZE(sb))) |
| 5638 |
return -EINVAL; |
| 5639 |
/* Offset must be less than i_size */ |
| 5640 |
if (offset >= inode->i_size) |
| 5641 |
return -EINVAL; |
| 5642 |
/* Check whether the maximum file size would be exceeded */ |
| 5643 |
if (len > inode->i_sb->s_maxbytes - inode->i_size) |
| 5644 |
return -EFBIG; |
| 5645 |
|
| 5646 |
/* |
| 5647 |
* Write out all dirty pages. Need to round down to align start offset |
| 5648 |
* to page size boundary for page size > block size. |
| 5649 |
*/ |
| 5650 |
start = round_down(offset, PAGE_SIZE); |
| 5651 |
ret = filemap_write_and_wait_range(mapping, start, LLONG_MAX); |
| 5652 |
if (ret) |
| 5653 |
return ret; |
| 5654 |
|
| 5655 |
truncate_pagecache(inode, start); |
| 5656 |
|
| 5657 |
credits = ext4_chunk_trans_extent(inode, 0); |
| 5658 |
handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); |
| 5659 |
if (IS_ERR(handle)) |
| 5660 |
return PTR_ERR(handle); |
| 5661 |
|
| 5662 |
ext4_fc_mark_ineligible(sb, EXT4_FC_REASON_FALLOC_RANGE, handle); |
| 5663 |
|
| 5664 |
/* Expand file to avoid data loss if there is error while shifting */ |
| 5665 |
inode->i_size += len; |
| 5666 |
EXT4_I(inode)->i_disksize += len; |
| 5667 |
ret = ext4_mark_inode_dirty(handle, inode); |
| 5668 |
if (ret) |
| 5669 |
goto out_handle; |
| 5670 |
|
| 5671 |
start_lblk = offset >> inode->i_blkbits; |
| 5672 |
len_lblk = len >> inode->i_blkbits; |
| 5673 |
|
| 5674 |
ext4_check_map_extents_env(inode); |
| 5675 |
|
| 5676 |
down_write(&EXT4_I(inode)->i_data_sem); |
| 5677 |
ext4_discard_preallocations(inode); |
| 5678 |
|
| 5679 |
path = ext4_find_extent(inode, start_lblk, NULL, 0); |
| 5680 |
if (IS_ERR(path)) { |
| 5681 |
up_write(&EXT4_I(inode)->i_data_sem); |
| 5682 |
ret = PTR_ERR(path); |
| 5683 |
goto out_handle; |
| 5684 |
} |
| 5685 |
|
| 5686 |
depth = ext_depth(inode); |
| 5687 |
extent = path[depth].p_ext; |
| 5688 |
if (extent) { |
| 5689 |
ee_start_lblk = le32_to_cpu(extent->ee_block); |
| 5690 |
ee_len = ext4_ext_get_actual_len(extent); |
| 5691 |
|
| 5692 |
/* |
| 5693 |
* If start_lblk is not the starting block of extent, split |
| 5694 |
* the extent @start_lblk |
| 5695 |
*/ |
| 5696 |
if ((start_lblk > ee_start_lblk) && |
| 5697 |
(start_lblk < (ee_start_lblk + ee_len))) { |
| 5698 |
path = ext4_split_extent_at(handle, inode, path, |
| 5699 |
start_lblk, EXT4_EX_NOCACHE | |
| 5700 |
EXT4_GET_BLOCKS_SPLIT_NOMERGE | |
| 5701 |
EXT4_GET_BLOCKS_METADATA_NOFAIL); |
| 5702 |
} |
| 5703 |
|
| 5704 |
if (IS_ERR(path)) { |
| 5705 |
up_write(&EXT4_I(inode)->i_data_sem); |
| 5706 |
ret = PTR_ERR(path); |
| 5707 |
goto out_handle; |
| 5708 |
} |
| 5709 |
} |
| 5710 |
|
| 5711 |
ext4_free_ext_path(path); |
| 5712 |
ext4_es_remove_extent(inode, start_lblk, EXT_MAX_BLOCKS - start_lblk); |
| 5713 |
|
| 5714 |
/* |
| 5715 |
* if start_lblk lies in a hole which is at start of file, use |
| 5716 |
* ee_start_lblk to shift extents |
| 5717 |
*/ |
| 5718 |
ret = ext4_ext_shift_extents(inode, handle, |
| 5719 |
max(ee_start_lblk, start_lblk), len_lblk, SHIFT_RIGHT); |
| 5720 |
up_write(&EXT4_I(inode)->i_data_sem); |
| 5721 |
if (ret) |
| 5722 |
goto out_handle; |
| 5723 |
|
| 5724 |
ext4_update_inode_fsync_trans(handle, inode, 1); |
| 5725 |
if (IS_SYNC(inode)) |
| 5726 |
ext4_handle_sync(handle); |
| 5727 |
|
| 5728 |
out_handle: |
| 5729 |
ext4_journal_stop(handle); |
| 5730 |
return ret; |
| 5731 |
} |
| 5732 |
|
| 5733 |
/** |
| 5734 |
* ext4_swap_extents() - Swap extents between two inodes |
| 5735 |
* @handle: handle for this transaction |
| 5736 |
* @inode1: First inode |
| 5737 |
* @inode2: Second inode |
| 5738 |
* @lblk1: Start block for first inode |
| 5739 |
* @lblk2: Start block for second inode |
| 5740 |
* @count: Number of blocks to swap |
| 5741 |
* @unwritten: Mark second inode's extents as unwritten after swap |
| 5742 |
* @erp: Pointer to save error value |
| 5743 |
* |
| 5744 |
* This helper routine does exactly what is promise "swap extents". All other |
| 5745 |
* stuff such as page-cache locking consistency, bh mapping consistency or |
| 5746 |
* extent's data copying must be performed by caller. |
| 5747 |
* Locking: |
| 5748 |
* i_rwsem is held for both inodes |
| 5749 |
* i_data_sem is locked for write for both inodes |
| 5750 |
* Assumptions: |
| 5751 |
* All pages from requested range are locked for both inodes |
| 5752 |
*/ |
| 5753 |
int |
| 5754 |
ext4_swap_extents(handle_t *handle, struct inode *inode1, |
| 5755 |
struct inode *inode2, ext4_lblk_t lblk1, ext4_lblk_t lblk2, |
| 5756 |
ext4_lblk_t count, int unwritten, int *erp) |
| 5757 |
{ |
| 5758 |
struct ext4_ext_path *path1 = NULL; |
| 5759 |
struct ext4_ext_path *path2 = NULL; |
| 5760 |
int replaced_count = 0; |
| 5761 |
|
| 5762 |
BUG_ON(!rwsem_is_locked(&EXT4_I(inode1)->i_data_sem)); |
| 5763 |
BUG_ON(!rwsem_is_locked(&EXT4_I(inode2)->i_data_sem)); |
| 5764 |
BUG_ON(!inode_is_locked(inode1)); |
| 5765 |
BUG_ON(!inode_is_locked(inode2)); |
| 5766 |
|
| 5767 |
ext4_es_remove_extent(inode1, lblk1, count); |
| 5768 |
ext4_es_remove_extent(inode2, lblk2, count); |
| 5769 |
|
| 5770 |
while (count) { |
| 5771 |
struct ext4_extent *ex1, *ex2, tmp_ex; |
| 5772 |
ext4_lblk_t e1_blk, e2_blk; |
| 5773 |
int e1_len, e2_len, len; |
| 5774 |
int split = 0; |
| 5775 |
|
| 5776 |
path1 = ext4_find_extent(inode1, lblk1, path1, EXT4_EX_NOCACHE); |
| 5777 |
if (IS_ERR(path1)) { |
| 5778 |
*erp = PTR_ERR(path1); |
| 5779 |
goto errout; |
| 5780 |
} |
| 5781 |
path2 = ext4_find_extent(inode2, lblk2, path2, EXT4_EX_NOCACHE); |
| 5782 |
if (IS_ERR(path2)) { |
| 5783 |
*erp = PTR_ERR(path2); |
| 5784 |
goto errout; |
| 5785 |
} |
| 5786 |
ex1 = path1[path1->p_depth].p_ext; |
| 5787 |
ex2 = path2[path2->p_depth].p_ext; |
| 5788 |
/* Do we have something to swap ? */ |
| 5789 |
if (unlikely(!ex2 || !ex1)) |
| 5790 |
goto errout; |
| 5791 |
|
| 5792 |
e1_blk = le32_to_cpu(ex1->ee_block); |
| 5793 |
e2_blk = le32_to_cpu(ex2->ee_block); |
| 5794 |
e1_len = ext4_ext_get_actual_len(ex1); |
| 5795 |
e2_len = ext4_ext_get_actual_len(ex2); |
| 5796 |
|
| 5797 |
/* Hole handling */ |
| 5798 |
if (!in_range(lblk1, e1_blk, e1_len) || |
| 5799 |
!in_range(lblk2, e2_blk, e2_len)) { |
| 5800 |
ext4_lblk_t next1, next2; |
| 5801 |
|
| 5802 |
/* if hole after extent, then go to next extent */ |
| 5803 |
next1 = ext4_ext_next_allocated_block(path1); |
| 5804 |
next2 = ext4_ext_next_allocated_block(path2); |
| 5805 |
/* If hole before extent, then shift to that extent */ |
| 5806 |
if (e1_blk > lblk1) |
| 5807 |
next1 = e1_blk; |
| 5808 |
if (e2_blk > lblk2) |
| 5809 |
next2 = e2_blk; |
| 5810 |
/* Do we have something to swap */ |
| 5811 |
if (next1 == EXT_MAX_BLOCKS || next2 == EXT_MAX_BLOCKS) |
| 5812 |
goto errout; |
| 5813 |
/* Move to the rightest boundary */ |
| 5814 |
len = next1 - lblk1; |
| 5815 |
if (len < next2 - lblk2) |
| 5816 |
len = next2 - lblk2; |
| 5817 |
if (len > count) |
| 5818 |
len = count; |
| 5819 |
lblk1 += len; |
| 5820 |
lblk2 += len; |
| 5821 |
count -= len; |
| 5822 |
continue; |
| 5823 |
} |
| 5824 |
|
| 5825 |
/* Prepare left boundary */ |
| 5826 |
if (e1_blk < lblk1) { |
| 5827 |
split = 1; |
| 5828 |
path1 = ext4_force_split_extent_at(handle, inode1, |
| 5829 |
path1, lblk1, 0); |
| 5830 |
if (IS_ERR(path1)) { |
| 5831 |
*erp = PTR_ERR(path1); |
| 5832 |
goto errout; |
| 5833 |
} |
| 5834 |
} |
| 5835 |
if (e2_blk < lblk2) { |
| 5836 |
split = 1; |
| 5837 |
path2 = ext4_force_split_extent_at(handle, inode2, |
| 5838 |
path2, lblk2, 0); |
| 5839 |
if (IS_ERR(path2)) { |
| 5840 |
*erp = PTR_ERR(path2); |
| 5841 |
goto errout; |
| 5842 |
} |
| 5843 |
} |
| 5844 |
/* ext4_split_extent_at() may result in leaf extent split, |
| 5845 |
* path must to be revalidated. */ |
| 5846 |
if (split) |
| 5847 |
continue; |
| 5848 |
|
| 5849 |
/* Prepare right boundary */ |
| 5850 |
len = count; |
| 5851 |
if (len > e1_blk + e1_len - lblk1) |
| 5852 |
len = e1_blk + e1_len - lblk1; |
| 5853 |
if (len > e2_blk + e2_len - lblk2) |
| 5854 |
len = e2_blk + e2_len - lblk2; |
| 5855 |
|
| 5856 |
if (len != e1_len) { |
| 5857 |
split = 1; |
| 5858 |
path1 = ext4_force_split_extent_at(handle, inode1, |
| 5859 |
path1, lblk1 + len, 0); |
| 5860 |
if (IS_ERR(path1)) { |
| 5861 |
*erp = PTR_ERR(path1); |
| 5862 |
goto errout; |
| 5863 |
} |
| 5864 |
} |
| 5865 |
if (len != e2_len) { |
| 5866 |
split = 1; |
| 5867 |
path2 = ext4_force_split_extent_at(handle, inode2, |
| 5868 |
path2, lblk2 + len, 0); |
| 5869 |
if (IS_ERR(path2)) { |
| 5870 |
*erp = PTR_ERR(path2); |
| 5871 |
goto errout; |
| 5872 |
} |
| 5873 |
} |
| 5874 |
/* ext4_split_extent_at() may result in leaf extent split, |
| 5875 |
* path must to be revalidated. */ |
| 5876 |
if (split) |
| 5877 |
continue; |
| 5878 |
|
| 5879 |
BUG_ON(e2_len != e1_len); |
| 5880 |
*erp = ext4_ext_get_access(handle, inode1, path1 + path1->p_depth); |
| 5881 |
if (unlikely(*erp)) |
| 5882 |
goto errout; |
| 5883 |
*erp = ext4_ext_get_access(handle, inode2, path2 + path2->p_depth); |
| 5884 |
if (unlikely(*erp)) |
| 5885 |
goto errout; |
| 5886 |
|
| 5887 |
/* Both extents are fully inside boundaries. Swap it now */ |
| 5888 |
tmp_ex = *ex1; |
| 5889 |
ext4_ext_store_pblock(ex1, ext4_ext_pblock(ex2)); |
| 5890 |
ext4_ext_store_pblock(ex2, ext4_ext_pblock(&tmp_ex)); |
| 5891 |
ex1->ee_len = cpu_to_le16(e2_len); |
| 5892 |
ex2->ee_len = cpu_to_le16(e1_len); |
| 5893 |
if (unwritten) |
| 5894 |
ext4_ext_mark_unwritten(ex2); |
| 5895 |
if (ext4_ext_is_unwritten(&tmp_ex)) |
| 5896 |
ext4_ext_mark_unwritten(ex1); |
| 5897 |
|
| 5898 |
ext4_ext_try_to_merge(handle, inode2, path2, ex2); |
| 5899 |
ext4_ext_try_to_merge(handle, inode1, path1, ex1); |
| 5900 |
*erp = ext4_ext_dirty(handle, inode2, path2 + |
| 5901 |
path2->p_depth); |
| 5902 |
if (unlikely(*erp)) |
| 5903 |
goto errout; |
| 5904 |
*erp = ext4_ext_dirty(handle, inode1, path1 + |
| 5905 |
path1->p_depth); |
| 5906 |
/* |
| 5907 |
* Looks scarry ah..? second inode already points to new blocks, |
| 5908 |
* and it was successfully dirtied. But luckily error may happen |
| 5909 |
* only due to journal error, so full transaction will be |
| 5910 |
* aborted anyway. |
| 5911 |
*/ |
| 5912 |
if (unlikely(*erp)) |
| 5913 |
goto errout; |
| 5914 |
|
| 5915 |
lblk1 += len; |
| 5916 |
lblk2 += len; |
| 5917 |
replaced_count += len; |
| 5918 |
count -= len; |
| 5919 |
} |
| 5920 |
|
| 5921 |
errout: |
| 5922 |
ext4_free_ext_path(path1); |
| 5923 |
ext4_free_ext_path(path2); |
| 5924 |
return replaced_count; |
| 5925 |
} |
| 5926 |
|
| 5927 |
/* |
| 5928 |
* ext4_clu_mapped - determine whether any block in a logical cluster has |
| 5929 |
* been mapped to a physical cluster |
| 5930 |
* |
| 5931 |
* @inode - file containing the logical cluster |
| 5932 |
* @lclu - logical cluster of interest |
| 5933 |
* |
| 5934 |
* Returns 1 if any block in the logical cluster is mapped, signifying |
| 5935 |
* that a physical cluster has been allocated for it. Otherwise, |
| 5936 |
* returns 0. Can also return negative error codes. Derived from |
| 5937 |
* ext4_ext_map_blocks(). |
| 5938 |
*/ |
| 5939 |
int ext4_clu_mapped(struct inode *inode, ext4_lblk_t lclu) |
| 5940 |
{ |
| 5941 |
struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
| 5942 |
struct ext4_ext_path *path; |
| 5943 |
int depth, mapped = 0, err = 0; |
| 5944 |
struct ext4_extent *extent; |
| 5945 |
ext4_lblk_t first_lblk, first_lclu, last_lclu; |
| 5946 |
|
| 5947 |
/* |
| 5948 |
* if data can be stored inline, the logical cluster isn't |
| 5949 |
* mapped - no physical clusters have been allocated, and the |
| 5950 |
* file has no extents |
| 5951 |
*/ |
| 5952 |
if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) || |
| 5953 |
ext4_has_inline_data(inode)) |
| 5954 |
return 0; |
| 5955 |
|
| 5956 |
/* search for the extent closest to the first block in the cluster */ |
| 5957 |
path = ext4_find_extent(inode, EXT4_C2B(sbi, lclu), NULL, 0); |
| 5958 |
if (IS_ERR(path)) |
| 5959 |
return PTR_ERR(path); |
| 5960 |
|
| 5961 |
depth = ext_depth(inode); |
| 5962 |
|
| 5963 |
/* |
| 5964 |
* A consistent leaf must not be empty. This situation is possible, |
| 5965 |
* though, _during_ tree modification, and it's why an assert can't |
| 5966 |
* be put in ext4_find_extent(). |
| 5967 |
*/ |
| 5968 |
if (unlikely(path[depth].p_ext == NULL && depth != 0)) { |
| 5969 |
EXT4_ERROR_INODE(inode, |
| 5970 |
"bad extent address - lblock: %lu, depth: %d, pblock: %lld", |
| 5971 |
(unsigned long) EXT4_C2B(sbi, lclu), |
| 5972 |
depth, path[depth].p_block); |
| 5973 |
err = -EFSCORRUPTED; |
| 5974 |
goto out; |
| 5975 |
} |
| 5976 |
|
| 5977 |
extent = path[depth].p_ext; |
| 5978 |
|
| 5979 |
/* can't be mapped if the extent tree is empty */ |
| 5980 |
if (extent == NULL) |
| 5981 |
goto out; |
| 5982 |
|
| 5983 |
first_lblk = le32_to_cpu(extent->ee_block); |
| 5984 |
first_lclu = EXT4_B2C(sbi, first_lblk); |
| 5985 |
|
| 5986 |
/* |
| 5987 |
* Three possible outcomes at this point - found extent spanning |
| 5988 |
* the target cluster, to the left of the target cluster, or to the |
| 5989 |
* right of the target cluster. The first two cases are handled here. |
| 5990 |
* The last case indicates the target cluster is not mapped. |
| 5991 |
*/ |
| 5992 |
if (lclu >= first_lclu) { |
| 5993 |
last_lclu = EXT4_B2C(sbi, first_lblk + |
| 5994 |
ext4_ext_get_actual_len(extent) - 1); |
| 5995 |
if (lclu <= last_lclu) { |
| 5996 |
mapped = 1; |
| 5997 |
} else { |
| 5998 |
first_lblk = ext4_ext_next_allocated_block(path); |
| 5999 |
first_lclu = EXT4_B2C(sbi, first_lblk); |
| 6000 |
if (lclu == first_lclu) |
| 6001 |
mapped = 1; |
| 6002 |
} |
| 6003 |
} |
| 6004 |
|
| 6005 |
out: |
| 6006 |
ext4_free_ext_path(path); |
| 6007 |
|
| 6008 |
return err ? err : mapped; |
| 6009 |
} |
| 6010 |
|
| 6011 |
/* |
| 6012 |
* Updates physical block address and unwritten status of extent |
| 6013 |
* starting at lblk start and of len. If such an extent doesn't exist, |
| 6014 |
* this function splits the extent tree appropriately to create an |
| 6015 |
* extent like this. This function is called in the fast commit |
| 6016 |
* replay path. Returns 0 on success and error on failure. |
| 6017 |
*/ |
| 6018 |
int ext4_ext_replay_update_ex(struct inode *inode, ext4_lblk_t start, |
| 6019 |
int len, int unwritten, ext4_fsblk_t pblk) |
| 6020 |
{ |
| 6021 |
struct ext4_ext_path *path; |
| 6022 |
struct ext4_extent *ex; |
| 6023 |
int ret; |
| 6024 |
|
| 6025 |
path = ext4_find_extent(inode, start, NULL, 0); |
| 6026 |
if (IS_ERR(path)) |
| 6027 |
return PTR_ERR(path); |
| 6028 |
ex = path[path->p_depth].p_ext; |
| 6029 |
if (!ex) { |
| 6030 |
ret = -EFSCORRUPTED; |
| 6031 |
goto out; |
| 6032 |
} |
| 6033 |
|
| 6034 |
if (le32_to_cpu(ex->ee_block) != start || |
| 6035 |
ext4_ext_get_actual_len(ex) != len) { |
| 6036 |
/* We need to split this extent to match our extent first */ |
| 6037 |
down_write(&EXT4_I(inode)->i_data_sem); |
| 6038 |
path = ext4_force_split_extent_at(NULL, inode, path, start, 1); |
| 6039 |
up_write(&EXT4_I(inode)->i_data_sem); |
| 6040 |
if (IS_ERR(path)) { |
| 6041 |
ret = PTR_ERR(path); |
| 6042 |
goto out; |
| 6043 |
} |
| 6044 |
|
| 6045 |
path = ext4_find_extent(inode, start, path, 0); |
| 6046 |
if (IS_ERR(path)) |
| 6047 |
return PTR_ERR(path); |
| 6048 |
|
| 6049 |
ex = path[path->p_depth].p_ext; |
| 6050 |
WARN_ON(le32_to_cpu(ex->ee_block) != start); |
| 6051 |
|
| 6052 |
if (ext4_ext_get_actual_len(ex) != len) { |
| 6053 |
down_write(&EXT4_I(inode)->i_data_sem); |
| 6054 |
path = ext4_force_split_extent_at(NULL, inode, path, |
| 6055 |
start + len, 1); |
| 6056 |
up_write(&EXT4_I(inode)->i_data_sem); |
| 6057 |
if (IS_ERR(path)) { |
| 6058 |
ret = PTR_ERR(path); |
| 6059 |
goto out; |
| 6060 |
} |
| 6061 |
|
| 6062 |
path = ext4_find_extent(inode, start, path, 0); |
| 6063 |
if (IS_ERR(path)) |
| 6064 |
return PTR_ERR(path); |
| 6065 |
ex = path[path->p_depth].p_ext; |
| 6066 |
} |
| 6067 |
} |
| 6068 |
if (unwritten) |
| 6069 |
ext4_ext_mark_unwritten(ex); |
| 6070 |
else |
| 6071 |
ext4_ext_mark_initialized(ex); |
| 6072 |
ext4_ext_store_pblock(ex, pblk); |
| 6073 |
down_write(&EXT4_I(inode)->i_data_sem); |
| 6074 |
ret = ext4_ext_dirty(NULL, inode, &path[path->p_depth]); |
| 6075 |
up_write(&EXT4_I(inode)->i_data_sem); |
| 6076 |
out: |
| 6077 |
ext4_free_ext_path(path); |
| 6078 |
ext4_mark_inode_dirty(NULL, inode); |
| 6079 |
return ret; |
| 6080 |
} |
| 6081 |
|
| 6082 |
/* Try to shrink the extent tree */ |
| 6083 |
void ext4_ext_replay_shrink_inode(struct inode *inode, ext4_lblk_t end) |
| 6084 |
{ |
| 6085 |
struct ext4_ext_path *path = NULL; |
| 6086 |
struct ext4_extent *ex; |
| 6087 |
ext4_lblk_t old_cur, cur = 0; |
| 6088 |
|
| 6089 |
while (cur < end) { |
| 6090 |
path = ext4_find_extent(inode, cur, NULL, 0); |
| 6091 |
if (IS_ERR(path)) |
| 6092 |
return; |
| 6093 |
ex = path[path->p_depth].p_ext; |
| 6094 |
if (!ex) { |
| 6095 |
ext4_free_ext_path(path); |
| 6096 |
ext4_mark_inode_dirty(NULL, inode); |
| 6097 |
return; |
| 6098 |
} |
| 6099 |
old_cur = cur; |
| 6100 |
cur = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex); |
| 6101 |
if (cur <= old_cur) |
| 6102 |
cur = old_cur + 1; |
| 6103 |
ext4_ext_try_to_merge(NULL, inode, path, ex); |
| 6104 |
down_write(&EXT4_I(inode)->i_data_sem); |
| 6105 |
ext4_ext_dirty(NULL, inode, &path[path->p_depth]); |
| 6106 |
up_write(&EXT4_I(inode)->i_data_sem); |
| 6107 |
ext4_mark_inode_dirty(NULL, inode); |
| 6108 |
ext4_free_ext_path(path); |
| 6109 |
} |
| 6110 |
} |
| 6111 |
|
| 6112 |
/* Check if *cur is a hole and if it is, skip it */ |
| 6113 |
static int skip_hole(struct inode *inode, ext4_lblk_t *cur) |
| 6114 |
{ |
| 6115 |
int ret; |
| 6116 |
struct ext4_map_blocks map; |
| 6117 |
|
| 6118 |
map.m_lblk = *cur; |
| 6119 |
map.m_len = ((inode->i_size) >> inode->i_sb->s_blocksize_bits) - *cur; |
| 6120 |
|
| 6121 |
ret = ext4_map_blocks(NULL, inode, &map, 0); |
| 6122 |
if (ret < 0) |
| 6123 |
return ret; |
| 6124 |
if (ret != 0) |
| 6125 |
return 0; |
| 6126 |
*cur = *cur + map.m_len; |
| 6127 |
return 0; |
| 6128 |
} |
| 6129 |
|
| 6130 |
/* Count number of blocks used by this inode and update i_blocks */ |
| 6131 |
int ext4_ext_replay_set_iblocks(struct inode *inode) |
| 6132 |
{ |
| 6133 |
struct ext4_ext_path *path = NULL, *path2 = NULL; |
| 6134 |
struct ext4_extent *ex; |
| 6135 |
ext4_lblk_t cur = 0, end; |
| 6136 |
int numblks = 0, i, ret = 0; |
| 6137 |
ext4_fsblk_t cmp1, cmp2; |
| 6138 |
struct ext4_map_blocks map; |
| 6139 |
|
| 6140 |
/* Determin the size of the file first */ |
| 6141 |
path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL, |
| 6142 |
EXT4_EX_NOCACHE); |
| 6143 |
if (IS_ERR(path)) |
| 6144 |
return PTR_ERR(path); |
| 6145 |
ex = path[path->p_depth].p_ext; |
| 6146 |
if (!ex) |
| 6147 |
goto out; |
| 6148 |
end = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex); |
| 6149 |
|
| 6150 |
/* Count the number of data blocks */ |
| 6151 |
cur = 0; |
| 6152 |
while (cur < end) { |
| 6153 |
map.m_lblk = cur; |
| 6154 |
map.m_len = end - cur; |
| 6155 |
ret = ext4_map_blocks(NULL, inode, &map, 0); |
| 6156 |
if (ret < 0) |
| 6157 |
break; |
| 6158 |
if (ret > 0) |
| 6159 |
numblks += ret; |
| 6160 |
cur = cur + map.m_len; |
| 6161 |
} |
| 6162 |
|
| 6163 |
/* |
| 6164 |
* Count the number of extent tree blocks. We do it by looking up |
| 6165 |
* two successive extents and determining the difference between |
| 6166 |
* their paths. When path is different for 2 successive extents |
| 6167 |
* we compare the blocks in the path at each level and increment |
| 6168 |
* iblocks by total number of differences found. |
| 6169 |
*/ |
| 6170 |
cur = 0; |
| 6171 |
ret = skip_hole(inode, &cur); |
| 6172 |
if (ret < 0) |
| 6173 |
goto out; |
| 6174 |
path = ext4_find_extent(inode, cur, path, 0); |
| 6175 |
if (IS_ERR(path)) |
| 6176 |
goto out; |
| 6177 |
numblks += path->p_depth; |
| 6178 |
while (cur < end) { |
| 6179 |
path = ext4_find_extent(inode, cur, path, 0); |
| 6180 |
if (IS_ERR(path)) |
| 6181 |
break; |
| 6182 |
ex = path[path->p_depth].p_ext; |
| 6183 |
if (!ex) |
| 6184 |
goto cleanup; |
| 6185 |
|
| 6186 |
cur = max(cur + 1, le32_to_cpu(ex->ee_block) + |
| 6187 |
ext4_ext_get_actual_len(ex)); |
| 6188 |
ret = skip_hole(inode, &cur); |
| 6189 |
if (ret < 0) |
| 6190 |
break; |
| 6191 |
|
| 6192 |
path2 = ext4_find_extent(inode, cur, path2, 0); |
| 6193 |
if (IS_ERR(path2)) |
| 6194 |
break; |
| 6195 |
|
| 6196 |
for (i = 0; i <= max(path->p_depth, path2->p_depth); i++) { |
| 6197 |
cmp1 = cmp2 = 0; |
| 6198 |
if (i <= path->p_depth) |
| 6199 |
cmp1 = path[i].p_bh ? |
| 6200 |
path[i].p_bh->b_blocknr : 0; |
| 6201 |
if (i <= path2->p_depth) |
| 6202 |
cmp2 = path2[i].p_bh ? |
| 6203 |
path2[i].p_bh->b_blocknr : 0; |
| 6204 |
if (cmp1 != cmp2 && cmp2 != 0) |
| 6205 |
numblks++; |
| 6206 |
} |
| 6207 |
} |
| 6208 |
|
| 6209 |
out: |
| 6210 |
inode->i_blocks = numblks << (inode->i_sb->s_blocksize_bits - 9); |
| 6211 |
ext4_mark_inode_dirty(NULL, inode); |
| 6212 |
cleanup: |
| 6213 |
ext4_free_ext_path(path); |
| 6214 |
ext4_free_ext_path(path2); |
| 6215 |
return 0; |
| 6216 |
} |
| 6217 |
|
| 6218 |
int ext4_ext_clear_bb(struct inode *inode) |
| 6219 |
{ |
| 6220 |
struct ext4_ext_path *path = NULL; |
| 6221 |
struct ext4_extent *ex; |
| 6222 |
ext4_lblk_t cur = 0, end; |
| 6223 |
int j, ret = 0; |
| 6224 |
struct ext4_map_blocks map; |
| 6225 |
|
| 6226 |
if (ext4_test_inode_flag(inode, EXT4_INODE_INLINE_DATA)) |
| 6227 |
return 0; |
| 6228 |
|
| 6229 |
/* Determin the size of the file first */ |
| 6230 |
path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL, |
| 6231 |
EXT4_EX_NOCACHE); |
| 6232 |
if (IS_ERR(path)) |
| 6233 |
return PTR_ERR(path); |
| 6234 |
ex = path[path->p_depth].p_ext; |
| 6235 |
if (!ex) |
| 6236 |
goto out; |
| 6237 |
end = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex); |
| 6238 |
|
| 6239 |
cur = 0; |
| 6240 |
while (cur < end) { |
| 6241 |
map.m_lblk = cur; |
| 6242 |
map.m_len = end - cur; |
| 6243 |
ret = ext4_map_blocks(NULL, inode, &map, 0); |
| 6244 |
if (ret < 0) |
| 6245 |
break; |
| 6246 |
if (ret > 0) { |
| 6247 |
path = ext4_find_extent(inode, map.m_lblk, path, 0); |
| 6248 |
if (!IS_ERR(path)) { |
| 6249 |
for (j = 0; j < path->p_depth; j++) { |
| 6250 |
ext4_mb_mark_bb(inode->i_sb, |
| 6251 |
path[j].p_block, 1, false); |
| 6252 |
ext4_fc_record_regions(inode->i_sb, inode->i_ino, |
| 6253 |
0, path[j].p_block, 1, 1); |
| 6254 |
} |
| 6255 |
} else { |
| 6256 |
path = NULL; |
| 6257 |
} |
| 6258 |
ext4_mb_mark_bb(inode->i_sb, map.m_pblk, map.m_len, false); |
| 6259 |
ext4_fc_record_regions(inode->i_sb, inode->i_ino, |
| 6260 |
map.m_lblk, map.m_pblk, map.m_len, 1); |
| 6261 |
} |
| 6262 |
cur = cur + map.m_len; |
| 6263 |
} |
| 6264 |
|
| 6265 |
out: |
| 6266 |
ext4_free_ext_path(path); |
| 6267 |
return 0; |
| 6268 |
} |
| 6269 |
|
| 6270 |
#if IS_ENABLED(CONFIG_EXT4_KUNIT_TESTS) |
| 6271 |
int ext4_ext_space_root_idx_test(struct inode *inode, int check) |
| 6272 |
{ |
| 6273 |
return ext4_ext_space_root_idx(inode, check); |
| 6274 |
} |
| 6275 |
EXPORT_SYMBOL_FOR_EXT4_TEST(ext4_ext_space_root_idx_test); |
| 6276 |
|
| 6277 |
struct ext4_ext_path *ext4_split_convert_extents_test(handle_t *handle, |
| 6278 |
struct inode *inode, struct ext4_map_blocks *map, |
| 6279 |
struct ext4_ext_path *path, int flags, |
| 6280 |
unsigned int *allocated) |
| 6281 |
{ |
| 6282 |
return ext4_split_convert_extents(handle, inode, map, path, |
| 6283 |
flags, allocated); |
| 6284 |
} |
| 6285 |
EXPORT_SYMBOL_FOR_EXT4_TEST(ext4_split_convert_extents_test); |
| 6286 |
|
| 6287 |
EXPORT_SYMBOL_FOR_EXT4_TEST(__ext4_ext_dirty); |
| 6288 |
EXPORT_SYMBOL_FOR_EXT4_TEST(ext4_ext_zeroout); |
| 6289 |
EXPORT_SYMBOL_FOR_EXT4_TEST(ext4_es_register_shrinker); |
| 6290 |
EXPORT_SYMBOL_FOR_EXT4_TEST(ext4_es_unregister_shrinker); |
| 6291 |
EXPORT_SYMBOL_FOR_EXT4_TEST(ext4_map_create_blocks); |
| 6292 |
EXPORT_SYMBOL_FOR_EXT4_TEST(ext4_es_init_tree); |
| 6293 |
EXPORT_SYMBOL_FOR_EXT4_TEST(ext4_es_lookup_extent); |
| 6294 |
EXPORT_SYMBOL_FOR_EXT4_TEST(ext4_es_insert_extent); |
| 6295 |
EXPORT_SYMBOL_FOR_EXT4_TEST(ext4_ext_insert_extent); |
| 6296 |
EXPORT_SYMBOL_FOR_EXT4_TEST(ext4_find_extent); |
| 6297 |
EXPORT_SYMBOL_FOR_EXT4_TEST(ext4_issue_zeroout); |
| 6298 |
EXPORT_SYMBOL_FOR_EXT4_TEST(ext4_map_query_blocks); |
| 6299 |
#endif |