| 1 |
// SPDX-License-Identifier: LGPL-2.1 |
| 2 |
/* |
| 3 |
* Copyright (c) 2008,2009 NEC Software Tohoku, Ltd. |
| 4 |
* Written by Takashi Sato <[email protected]> |
| 5 |
* Akira Fujita <[email protected]> |
| 6 |
*/ |
| 7 |
|
| 8 |
#include <linux/fs.h> |
| 9 |
#include <linux/quotaops.h> |
| 10 |
#include <linux/slab.h> |
| 11 |
#include <linux/sched/mm.h> |
| 12 |
#include "ext4_jbd2.h" |
| 13 |
#include "ext4.h" |
| 14 |
#include "ext4_extents.h" |
| 15 |
|
| 16 |
#include <trace/events/ext4.h> |
| 17 |
|
| 18 |
struct mext_data { |
| 19 |
struct inode *orig_inode; /* Origin file inode */ |
| 20 |
struct inode *donor_inode; /* Donor file inode */ |
| 21 |
struct ext4_map_blocks orig_map;/* Origin file's move mapping */ |
| 22 |
ext4_lblk_t donor_lblk; /* Start block of the donor file */ |
| 23 |
}; |
| 24 |
|
| 25 |
/** |
| 26 |
* ext4_double_down_write_data_sem() - write lock two inodes's i_data_sem |
| 27 |
* @first: inode to be locked |
| 28 |
* @second: inode to be locked |
| 29 |
* |
| 30 |
* Acquire write lock of i_data_sem of the two inodes |
| 31 |
*/ |
| 32 |
void |
| 33 |
ext4_double_down_write_data_sem(struct inode *first, struct inode *second) |
| 34 |
{ |
| 35 |
if (first < second) { |
| 36 |
down_write(&EXT4_I(first)->i_data_sem); |
| 37 |
down_write_nested(&EXT4_I(second)->i_data_sem, I_DATA_SEM_OTHER); |
| 38 |
} else { |
| 39 |
down_write(&EXT4_I(second)->i_data_sem); |
| 40 |
down_write_nested(&EXT4_I(first)->i_data_sem, I_DATA_SEM_OTHER); |
| 41 |
} |
| 42 |
} |
| 43 |
|
| 44 |
/** |
| 45 |
* ext4_double_up_write_data_sem - Release two inodes' write lock of i_data_sem |
| 46 |
* |
| 47 |
* @orig_inode: original inode structure to be released its lock first |
| 48 |
* @donor_inode: donor inode structure to be released its lock second |
| 49 |
* Release write lock of i_data_sem of two inodes (orig and donor). |
| 50 |
*/ |
| 51 |
void |
| 52 |
ext4_double_up_write_data_sem(struct inode *orig_inode, |
| 53 |
struct inode *donor_inode) |
| 54 |
{ |
| 55 |
up_write(&EXT4_I(orig_inode)->i_data_sem); |
| 56 |
up_write(&EXT4_I(donor_inode)->i_data_sem); |
| 57 |
} |
| 58 |
|
| 59 |
/* Grab and lock folio on both @inode1 and @inode2 by inode order. */ |
| 60 |
static int mext_folio_double_lock(struct inode *inode1, struct inode *inode2, |
| 61 |
pgoff_t index1, pgoff_t index2, size_t len, |
| 62 |
struct folio *folio[2]) |
| 63 |
{ |
| 64 |
struct address_space *mapping[2]; |
| 65 |
unsigned int flags; |
| 66 |
fgf_t fgp_flags = FGP_WRITEBEGIN; |
| 67 |
|
| 68 |
BUG_ON(!inode1 || !inode2); |
| 69 |
if (inode1 < inode2) { |
| 70 |
mapping[0] = inode1->i_mapping; |
| 71 |
mapping[1] = inode2->i_mapping; |
| 72 |
} else { |
| 73 |
swap(index1, index2); |
| 74 |
mapping[0] = inode2->i_mapping; |
| 75 |
mapping[1] = inode1->i_mapping; |
| 76 |
} |
| 77 |
|
| 78 |
flags = memalloc_nofs_save(); |
| 79 |
fgp_flags |= fgf_set_order(len); |
| 80 |
folio[0] = __filemap_get_folio(mapping[0], index1, fgp_flags, |
| 81 |
mapping_gfp_mask(mapping[0])); |
| 82 |
if (IS_ERR(folio[0])) { |
| 83 |
memalloc_nofs_restore(flags); |
| 84 |
return PTR_ERR(folio[0]); |
| 85 |
} |
| 86 |
|
| 87 |
folio[1] = __filemap_get_folio(mapping[1], index2, fgp_flags, |
| 88 |
mapping_gfp_mask(mapping[1])); |
| 89 |
memalloc_nofs_restore(flags); |
| 90 |
if (IS_ERR(folio[1])) { |
| 91 |
folio_unlock(folio[0]); |
| 92 |
folio_put(folio[0]); |
| 93 |
return PTR_ERR(folio[1]); |
| 94 |
} |
| 95 |
/* |
| 96 |
* __filemap_get_folio() may not wait on folio's writeback if |
| 97 |
* BDI not demand that. But it is reasonable to be very conservative |
| 98 |
* here and explicitly wait on folio's writeback |
| 99 |
*/ |
| 100 |
folio_wait_writeback(folio[0]); |
| 101 |
folio_wait_writeback(folio[1]); |
| 102 |
if (inode1 > inode2) |
| 103 |
swap(folio[0], folio[1]); |
| 104 |
|
| 105 |
return 0; |
| 106 |
} |
| 107 |
|
| 108 |
static void mext_folio_double_unlock(struct folio *folio[2]) |
| 109 |
{ |
| 110 |
folio_unlock(folio[0]); |
| 111 |
folio_put(folio[0]); |
| 112 |
folio_unlock(folio[1]); |
| 113 |
folio_put(folio[1]); |
| 114 |
} |
| 115 |
|
| 116 |
/* Force folio buffers uptodate w/o dropping folio's lock */ |
| 117 |
static int mext_folio_mkuptodate(struct folio *folio, size_t from, size_t to) |
| 118 |
{ |
| 119 |
struct inode *inode = folio->mapping->host; |
| 120 |
sector_t block; |
| 121 |
struct buffer_head *bh, *head; |
| 122 |
unsigned int blocksize, block_start, block_end; |
| 123 |
int nr = 0; |
| 124 |
bool partial = false; |
| 125 |
|
| 126 |
BUG_ON(!folio_test_locked(folio)); |
| 127 |
BUG_ON(folio_test_writeback(folio)); |
| 128 |
|
| 129 |
if (folio_test_uptodate(folio)) |
| 130 |
return 0; |
| 131 |
|
| 132 |
blocksize = i_blocksize(inode); |
| 133 |
head = folio_buffers(folio); |
| 134 |
if (!head) |
| 135 |
head = create_empty_buffers(folio, blocksize, 0); |
| 136 |
|
| 137 |
block = folio_pos(folio) >> inode->i_blkbits; |
| 138 |
block_end = 0; |
| 139 |
bh = head; |
| 140 |
do { |
| 141 |
block_start = block_end; |
| 142 |
block_end = block_start + blocksize; |
| 143 |
if (block_end <= from || block_start >= to) { |
| 144 |
if (!buffer_uptodate(bh)) |
| 145 |
partial = true; |
| 146 |
continue; |
| 147 |
} |
| 148 |
if (buffer_uptodate(bh)) |
| 149 |
continue; |
| 150 |
if (!buffer_mapped(bh)) { |
| 151 |
int err = ext4_get_block(inode, block, bh, 0); |
| 152 |
if (err) |
| 153 |
return err; |
| 154 |
if (!buffer_mapped(bh)) { |
| 155 |
folio_zero_range(folio, block_start, blocksize); |
| 156 |
set_buffer_uptodate(bh); |
| 157 |
continue; |
| 158 |
} |
| 159 |
} |
| 160 |
lock_buffer(bh); |
| 161 |
if (buffer_uptodate(bh)) { |
| 162 |
unlock_buffer(bh); |
| 163 |
continue; |
| 164 |
} |
| 165 |
ext4_read_bh_nowait(bh, 0, NULL, false); |
| 166 |
nr++; |
| 167 |
} while (block++, (bh = bh->b_this_page) != head); |
| 168 |
|
| 169 |
/* No io required */ |
| 170 |
if (!nr) |
| 171 |
goto out; |
| 172 |
|
| 173 |
bh = head; |
| 174 |
do { |
| 175 |
if (bh_offset(bh) + blocksize <= from) |
| 176 |
continue; |
| 177 |
if (bh_offset(bh) >= to) |
| 178 |
break; |
| 179 |
wait_on_buffer(bh); |
| 180 |
if (buffer_uptodate(bh)) |
| 181 |
continue; |
| 182 |
return -EIO; |
| 183 |
} while ((bh = bh->b_this_page) != head); |
| 184 |
out: |
| 185 |
if (!partial) |
| 186 |
folio_mark_uptodate(folio); |
| 187 |
return 0; |
| 188 |
} |
| 189 |
|
| 190 |
enum mext_move_type {MEXT_SKIP_EXTENT, MEXT_MOVE_EXTENT, MEXT_COPY_DATA}; |
| 191 |
|
| 192 |
/* |
| 193 |
* Start to move extent between the origin inode and the donor inode, |
| 194 |
* hold one folio for each inode and check the candidate moving extent |
| 195 |
* mapping status again. |
| 196 |
*/ |
| 197 |
static int mext_move_begin(struct mext_data *mext, struct folio *folio[2], |
| 198 |
enum mext_move_type *move_type) |
| 199 |
{ |
| 200 |
struct inode *orig_inode = mext->orig_inode; |
| 201 |
struct inode *donor_inode = mext->donor_inode; |
| 202 |
unsigned int blkbits = orig_inode->i_blkbits; |
| 203 |
struct ext4_map_blocks donor_map = {0}; |
| 204 |
loff_t orig_pos, donor_pos; |
| 205 |
size_t move_len; |
| 206 |
int ret; |
| 207 |
|
| 208 |
orig_pos = ((loff_t)mext->orig_map.m_lblk) << blkbits; |
| 209 |
donor_pos = ((loff_t)mext->donor_lblk) << blkbits; |
| 210 |
ret = mext_folio_double_lock(orig_inode, donor_inode, |
| 211 |
orig_pos >> PAGE_SHIFT, donor_pos >> PAGE_SHIFT, |
| 212 |
((size_t)mext->orig_map.m_len) << blkbits, folio); |
| 213 |
if (ret) |
| 214 |
return ret; |
| 215 |
|
| 216 |
/* |
| 217 |
* Check the origin inode's mapping information again under the |
| 218 |
* folio lock, as we do not hold the i_data_sem at all times, and |
| 219 |
* it may change during the concurrent write-back operation. |
| 220 |
*/ |
| 221 |
if (mext->orig_map.m_seq != READ_ONCE(EXT4_I(orig_inode)->i_es_seq)) { |
| 222 |
ret = -ESTALE; |
| 223 |
goto error; |
| 224 |
} |
| 225 |
|
| 226 |
/* Adjust the moving length according to the length of shorter folio. */ |
| 227 |
move_len = umin(folio_pos(folio[0]) + folio_size(folio[0]) - orig_pos, |
| 228 |
folio_pos(folio[1]) + folio_size(folio[1]) - donor_pos); |
| 229 |
move_len >>= blkbits; |
| 230 |
if (move_len < mext->orig_map.m_len) |
| 231 |
mext->orig_map.m_len = move_len; |
| 232 |
|
| 233 |
donor_map.m_lblk = mext->donor_lblk; |
| 234 |
donor_map.m_len = mext->orig_map.m_len; |
| 235 |
donor_map.m_flags = 0; |
| 236 |
ret = ext4_map_blocks(NULL, donor_inode, &donor_map, 0); |
| 237 |
if (ret < 0) |
| 238 |
goto error; |
| 239 |
|
| 240 |
/* Adjust the moving length according to the donor mapping length. */ |
| 241 |
mext->orig_map.m_len = donor_map.m_len; |
| 242 |
|
| 243 |
/* Skip moving if the donor range is a hole or a delalloc extent. */ |
| 244 |
if (!(donor_map.m_flags & (EXT4_MAP_MAPPED | EXT4_MAP_UNWRITTEN))) |
| 245 |
*move_type = MEXT_SKIP_EXTENT; |
| 246 |
/* If both mapping ranges are unwritten, no need to copy data. */ |
| 247 |
else if ((mext->orig_map.m_flags & EXT4_MAP_UNWRITTEN) && |
| 248 |
(donor_map.m_flags & EXT4_MAP_UNWRITTEN)) |
| 249 |
*move_type = MEXT_MOVE_EXTENT; |
| 250 |
else |
| 251 |
*move_type = MEXT_COPY_DATA; |
| 252 |
|
| 253 |
return 0; |
| 254 |
error: |
| 255 |
mext_folio_double_unlock(folio); |
| 256 |
return ret; |
| 257 |
} |
| 258 |
|
| 259 |
/* |
| 260 |
* Re-create the new moved mapping buffers of the original inode and commit |
| 261 |
* the entire written range. |
| 262 |
*/ |
| 263 |
static int mext_folio_mkwrite(struct inode *inode, struct folio *folio, |
| 264 |
size_t from, size_t to) |
| 265 |
{ |
| 266 |
unsigned int blocksize = i_blocksize(inode); |
| 267 |
struct buffer_head *bh, *head; |
| 268 |
size_t block_start, block_end; |
| 269 |
sector_t block; |
| 270 |
int ret; |
| 271 |
|
| 272 |
head = folio_buffers(folio); |
| 273 |
if (!head) |
| 274 |
head = create_empty_buffers(folio, blocksize, 0); |
| 275 |
|
| 276 |
block = folio_pos(folio) >> inode->i_blkbits; |
| 277 |
block_end = 0; |
| 278 |
bh = head; |
| 279 |
do { |
| 280 |
block_start = block_end; |
| 281 |
block_end = block_start + blocksize; |
| 282 |
if (block_end <= from || block_start >= to) |
| 283 |
continue; |
| 284 |
|
| 285 |
ret = ext4_get_block(inode, block, bh, 0); |
| 286 |
if (ret) |
| 287 |
return ret; |
| 288 |
} while (block++, (bh = bh->b_this_page) != head); |
| 289 |
|
| 290 |
block_commit_write(folio, from, to); |
| 291 |
return 0; |
| 292 |
} |
| 293 |
|
| 294 |
/* |
| 295 |
* Save the data in original inode extent blocks and replace one folio size |
| 296 |
* aligned original inode extent with one or one partial donor inode extent, |
| 297 |
* and then write out the saved data in new original inode blocks. Pass out |
| 298 |
* the replaced block count through m_len. Return 0 on success, and an error |
| 299 |
* code otherwise. |
| 300 |
*/ |
| 301 |
static int mext_move_extent(struct mext_data *mext, u64 *m_len) |
| 302 |
{ |
| 303 |
struct inode *orig_inode = mext->orig_inode; |
| 304 |
struct inode *donor_inode = mext->donor_inode; |
| 305 |
struct ext4_map_blocks *orig_map = &mext->orig_map; |
| 306 |
unsigned int blkbits = orig_inode->i_blkbits; |
| 307 |
struct folio *folio[2] = {NULL, NULL}; |
| 308 |
loff_t from, length; |
| 309 |
enum mext_move_type move_type = 0; |
| 310 |
handle_t *handle; |
| 311 |
u64 r_len = 0; |
| 312 |
unsigned int credits; |
| 313 |
int ret, ret2; |
| 314 |
|
| 315 |
*m_len = 0; |
| 316 |
trace_ext4_move_extent_enter(orig_inode, orig_map, donor_inode, |
| 317 |
mext->donor_lblk); |
| 318 |
credits = ext4_chunk_trans_extent(orig_inode, 0) * 2; |
| 319 |
handle = ext4_journal_start(orig_inode, EXT4_HT_MOVE_EXTENTS, credits); |
| 320 |
if (IS_ERR(handle)) { |
| 321 |
ret = PTR_ERR(handle); |
| 322 |
goto out; |
| 323 |
} |
| 324 |
ext4_fc_mark_ineligible(orig_inode->i_sb, EXT4_FC_REASON_MOVE_EXT, |
| 325 |
handle); |
| 326 |
|
| 327 |
ret = mext_move_begin(mext, folio, &move_type); |
| 328 |
if (ret) |
| 329 |
goto stop_handle; |
| 330 |
|
| 331 |
if (move_type == MEXT_SKIP_EXTENT) |
| 332 |
goto unlock; |
| 333 |
|
| 334 |
/* |
| 335 |
* Copy the data. First, read the original inode data into the page |
| 336 |
* cache. Then, release the existing mapping relationships and swap |
| 337 |
* the extent. Finally, re-establish the new mapping relationships |
| 338 |
* and dirty the page cache. |
| 339 |
*/ |
| 340 |
if (move_type == MEXT_COPY_DATA) { |
| 341 |
from = offset_in_folio(folio[0], |
| 342 |
((loff_t)orig_map->m_lblk) << blkbits); |
| 343 |
length = ((loff_t)orig_map->m_len) << blkbits; |
| 344 |
|
| 345 |
ret = mext_folio_mkuptodate(folio[0], from, from + length); |
| 346 |
if (ret) |
| 347 |
goto unlock; |
| 348 |
} |
| 349 |
|
| 350 |
if (!filemap_release_folio(folio[0], 0) || |
| 351 |
!filemap_release_folio(folio[1], 0)) { |
| 352 |
ret = -EBUSY; |
| 353 |
goto unlock; |
| 354 |
} |
| 355 |
|
| 356 |
/* Move extent */ |
| 357 |
ext4_double_down_write_data_sem(orig_inode, donor_inode); |
| 358 |
*m_len = ext4_swap_extents(handle, orig_inode, donor_inode, |
| 359 |
orig_map->m_lblk, mext->donor_lblk, |
| 360 |
orig_map->m_len, 1, &ret); |
| 361 |
ext4_double_up_write_data_sem(orig_inode, donor_inode); |
| 362 |
|
| 363 |
/* A short-length swap cannot occur after a successful swap extent. */ |
| 364 |
if (WARN_ON_ONCE(!ret && (*m_len != orig_map->m_len))) |
| 365 |
ret = -EIO; |
| 366 |
|
| 367 |
if (!(*m_len) || (move_type == MEXT_MOVE_EXTENT)) |
| 368 |
goto unlock; |
| 369 |
|
| 370 |
/* Copy data */ |
| 371 |
length = (*m_len) << blkbits; |
| 372 |
ret2 = mext_folio_mkwrite(orig_inode, folio[0], from, from + length); |
| 373 |
if (ret2) { |
| 374 |
if (!ret) |
| 375 |
ret = ret2; |
| 376 |
goto repair_branches; |
| 377 |
} |
| 378 |
/* |
| 379 |
* Even in case of data=writeback it is reasonable to pin |
| 380 |
* inode to transaction, to prevent unexpected data loss. |
| 381 |
*/ |
| 382 |
ret2 = ext4_jbd2_inode_add_write(handle, orig_inode, |
| 383 |
((loff_t)orig_map->m_lblk) << blkbits, length); |
| 384 |
if (!ret) |
| 385 |
ret = ret2; |
| 386 |
unlock: |
| 387 |
mext_folio_double_unlock(folio); |
| 388 |
stop_handle: |
| 389 |
ext4_journal_stop(handle); |
| 390 |
out: |
| 391 |
trace_ext4_move_extent_exit(orig_inode, orig_map->m_lblk, donor_inode, |
| 392 |
mext->donor_lblk, orig_map->m_len, *m_len, |
| 393 |
move_type, ret); |
| 394 |
return ret; |
| 395 |
|
| 396 |
repair_branches: |
| 397 |
ret2 = 0; |
| 398 |
ext4_double_down_write_data_sem(orig_inode, donor_inode); |
| 399 |
r_len = ext4_swap_extents(handle, donor_inode, orig_inode, |
| 400 |
mext->donor_lblk, orig_map->m_lblk, |
| 401 |
*m_len, 0, &ret2); |
| 402 |
ext4_double_up_write_data_sem(orig_inode, donor_inode); |
| 403 |
if (ret2 || r_len != *m_len) { |
| 404 |
ext4_error_inode_block(orig_inode, (sector_t)(orig_map->m_lblk), |
| 405 |
EIO, "Unable to copy data block, data will be lost!"); |
| 406 |
ret = -EIO; |
| 407 |
} |
| 408 |
*m_len = 0; |
| 409 |
goto unlock; |
| 410 |
} |
| 411 |
|
| 412 |
/* |
| 413 |
* Check the validity of the basic filesystem environment and the |
| 414 |
* inodes' support status. |
| 415 |
*/ |
| 416 |
static int mext_check_validity(struct inode *orig_inode, |
| 417 |
struct inode *donor_inode) |
| 418 |
{ |
| 419 |
struct super_block *sb = orig_inode->i_sb; |
| 420 |
|
| 421 |
/* origin and donor should be different inodes */ |
| 422 |
if (orig_inode == donor_inode) { |
| 423 |
ext4_debug("ext4 move extent: The argument files should not be same inode [ino:orig %lu, donor %lu]\n", |
| 424 |
orig_inode->i_ino, donor_inode->i_ino); |
| 425 |
return -EINVAL; |
| 426 |
} |
| 427 |
|
| 428 |
/* origin and donor should belone to the same filesystem */ |
| 429 |
if (orig_inode->i_sb != donor_inode->i_sb) { |
| 430 |
ext4_debug("ext4 move extent: The argument files should be in same FS [ino:orig %lu, donor %lu]\n", |
| 431 |
orig_inode->i_ino, donor_inode->i_ino); |
| 432 |
return -EINVAL; |
| 433 |
} |
| 434 |
|
| 435 |
/* Regular file check */ |
| 436 |
if (!S_ISREG(orig_inode->i_mode) || !S_ISREG(donor_inode->i_mode)) { |
| 437 |
ext4_debug("ext4 move extent: The argument files should be regular file [ino:orig %lu, donor %lu]\n", |
| 438 |
orig_inode->i_ino, donor_inode->i_ino); |
| 439 |
return -EINVAL; |
| 440 |
} |
| 441 |
|
| 442 |
if (ext4_has_feature_bigalloc(sb)) { |
| 443 |
ext4_msg(sb, KERN_ERR, |
| 444 |
"Online defrag not supported with bigalloc"); |
| 445 |
return -EOPNOTSUPP; |
| 446 |
} |
| 447 |
|
| 448 |
if (IS_DAX(orig_inode)) { |
| 449 |
ext4_msg(sb, KERN_ERR, |
| 450 |
"Online defrag not supported with DAX"); |
| 451 |
return -EOPNOTSUPP; |
| 452 |
} |
| 453 |
|
| 454 |
/* |
| 455 |
* TODO: it's not obvious how to swap blocks for inodes with full |
| 456 |
* journaling enabled. |
| 457 |
*/ |
| 458 |
if (ext4_should_journal_data(orig_inode) || |
| 459 |
ext4_should_journal_data(donor_inode)) { |
| 460 |
ext4_msg(sb, KERN_ERR, |
| 461 |
"Online defrag not supported with data journaling"); |
| 462 |
return -EOPNOTSUPP; |
| 463 |
} |
| 464 |
|
| 465 |
if (IS_ENCRYPTED(orig_inode) || IS_ENCRYPTED(donor_inode)) { |
| 466 |
ext4_msg(sb, KERN_ERR, |
| 467 |
"Online defrag not supported for encrypted files"); |
| 468 |
return -EOPNOTSUPP; |
| 469 |
} |
| 470 |
|
| 471 |
/* Ext4 move extent supports only extent based file */ |
| 472 |
if (!(ext4_test_inode_flag(orig_inode, EXT4_INODE_EXTENTS)) || |
| 473 |
!(ext4_test_inode_flag(donor_inode, EXT4_INODE_EXTENTS))) { |
| 474 |
ext4_msg(sb, KERN_ERR, |
| 475 |
"Online defrag not supported for non-extent files"); |
| 476 |
return -EOPNOTSUPP; |
| 477 |
} |
| 478 |
|
| 479 |
if (donor_inode->i_mode & (S_ISUID|S_ISGID)) { |
| 480 |
ext4_debug("ext4 move extent: suid or sgid is set to donor file [ino:orig %lu, donor %lu]\n", |
| 481 |
orig_inode->i_ino, donor_inode->i_ino); |
| 482 |
return -EINVAL; |
| 483 |
} |
| 484 |
|
| 485 |
if (IS_IMMUTABLE(donor_inode) || IS_APPEND(donor_inode)) { |
| 486 |
ext4_debug("ext4 move extent: donor should not be immutable or append file [ino:orig %lu, donor %lu]\n", |
| 487 |
orig_inode->i_ino, donor_inode->i_ino); |
| 488 |
return -EPERM; |
| 489 |
} |
| 490 |
|
| 491 |
/* Ext4 move extent does not support swap files */ |
| 492 |
if (IS_SWAPFILE(orig_inode) || IS_SWAPFILE(donor_inode)) { |
| 493 |
ext4_debug("ext4 move extent: The argument files should not be swap files [ino:orig %lu, donor %lu]\n", |
| 494 |
orig_inode->i_ino, donor_inode->i_ino); |
| 495 |
return -ETXTBSY; |
| 496 |
} |
| 497 |
|
| 498 |
if (ext4_is_quota_file(orig_inode) || ext4_is_quota_file(donor_inode)) { |
| 499 |
ext4_debug("ext4 move extent: The argument files should not be quota files [ino:orig %lu, donor %lu]\n", |
| 500 |
orig_inode->i_ino, donor_inode->i_ino); |
| 501 |
return -EOPNOTSUPP; |
| 502 |
} |
| 503 |
|
| 504 |
if ((!orig_inode->i_size) || (!donor_inode->i_size)) { |
| 505 |
ext4_debug("ext4 move extent: File size is 0 byte\n"); |
| 506 |
return -EINVAL; |
| 507 |
} |
| 508 |
|
| 509 |
return 0; |
| 510 |
} |
| 511 |
|
| 512 |
/* |
| 513 |
* Check the moving range of ext4_move_extents() whether the files can be |
| 514 |
* exchanged with each other, and adjust the length to fit within the file |
| 515 |
* size. Return 0 on success, or a negative error value on failure. |
| 516 |
*/ |
| 517 |
static int mext_check_adjust_range(struct inode *orig_inode, |
| 518 |
struct inode *donor_inode, __u64 orig_start, |
| 519 |
__u64 donor_start, __u64 *len) |
| 520 |
{ |
| 521 |
__u64 orig_eof, donor_eof; |
| 522 |
|
| 523 |
/* Start offset should be same */ |
| 524 |
if ((orig_start & ~(PAGE_MASK >> orig_inode->i_blkbits)) != |
| 525 |
(donor_start & ~(PAGE_MASK >> orig_inode->i_blkbits))) { |
| 526 |
ext4_debug("ext4 move extent: orig and donor's start offsets are not aligned [ino:orig %lu, donor %lu]\n", |
| 527 |
orig_inode->i_ino, donor_inode->i_ino); |
| 528 |
return -EINVAL; |
| 529 |
} |
| 530 |
|
| 531 |
if ((orig_start >= EXT_MAX_BLOCKS) || |
| 532 |
(donor_start >= EXT_MAX_BLOCKS) || |
| 533 |
(*len > EXT_MAX_BLOCKS) || |
| 534 |
(donor_start + *len >= EXT_MAX_BLOCKS) || |
| 535 |
(orig_start + *len >= EXT_MAX_BLOCKS)) { |
| 536 |
ext4_debug("ext4 move extent: Can't handle over [%u] blocks [ino:orig %lu, donor %lu]\n", |
| 537 |
EXT_MAX_BLOCKS, |
| 538 |
orig_inode->i_ino, donor_inode->i_ino); |
| 539 |
return -EINVAL; |
| 540 |
} |
| 541 |
|
| 542 |
orig_eof = EXT4_B_TO_LBLK(orig_inode, i_size_read(orig_inode)); |
| 543 |
donor_eof = EXT4_B_TO_LBLK(donor_inode, i_size_read(donor_inode)); |
| 544 |
if (orig_eof <= orig_start) |
| 545 |
*len = 0; |
| 546 |
else if (orig_eof < orig_start + *len - 1) |
| 547 |
*len = orig_eof - orig_start; |
| 548 |
if (donor_eof <= donor_start) |
| 549 |
*len = 0; |
| 550 |
else if (donor_eof < donor_start + *len - 1) |
| 551 |
*len = donor_eof - donor_start; |
| 552 |
if (!*len) { |
| 553 |
ext4_debug("ext4 move extent: len should not be 0 [ino:orig %lu, donor %lu]\n", |
| 554 |
orig_inode->i_ino, donor_inode->i_ino); |
| 555 |
return -EINVAL; |
| 556 |
} |
| 557 |
|
| 558 |
return 0; |
| 559 |
} |
| 560 |
|
| 561 |
/** |
| 562 |
* ext4_move_extents - Exchange the specified range of a file |
| 563 |
* |
| 564 |
* @o_filp: file structure of the original file |
| 565 |
* @d_filp: file structure of the donor file |
| 566 |
* @orig_blk: start offset in block for orig |
| 567 |
* @donor_blk: start offset in block for donor |
| 568 |
* @len: the number of blocks to be moved |
| 569 |
* @moved_len: moved block length |
| 570 |
* |
| 571 |
* This function returns 0 and moved block length is set in moved_len |
| 572 |
* if succeed, otherwise returns error value. |
| 573 |
*/ |
| 574 |
int ext4_move_extents(struct file *o_filp, struct file *d_filp, __u64 orig_blk, |
| 575 |
__u64 donor_blk, __u64 len, __u64 *moved_len) |
| 576 |
{ |
| 577 |
struct inode *orig_inode = file_inode(o_filp); |
| 578 |
struct inode *donor_inode = file_inode(d_filp); |
| 579 |
struct mext_data mext; |
| 580 |
struct super_block *sb = orig_inode->i_sb; |
| 581 |
struct ext4_sb_info *sbi = EXT4_SB(sb); |
| 582 |
int retries = 0; |
| 583 |
u64 m_len; |
| 584 |
int ret; |
| 585 |
|
| 586 |
*moved_len = 0; |
| 587 |
|
| 588 |
/* Protect orig and donor inodes against a truncate */ |
| 589 |
lock_two_nondirectories(orig_inode, donor_inode); |
| 590 |
|
| 591 |
ret = mext_check_validity(orig_inode, donor_inode); |
| 592 |
if (ret) |
| 593 |
goto out; |
| 594 |
|
| 595 |
/* Wait for all existing dio workers */ |
| 596 |
inode_dio_wait(orig_inode); |
| 597 |
inode_dio_wait(donor_inode); |
| 598 |
|
| 599 |
/* Check and adjust the specified move_extent range. */ |
| 600 |
ret = mext_check_adjust_range(orig_inode, donor_inode, orig_blk, |
| 601 |
donor_blk, &len); |
| 602 |
if (ret) |
| 603 |
goto out; |
| 604 |
|
| 605 |
mext.orig_inode = orig_inode; |
| 606 |
mext.donor_inode = donor_inode; |
| 607 |
while (len) { |
| 608 |
mext.orig_map.m_lblk = orig_blk; |
| 609 |
mext.orig_map.m_len = len; |
| 610 |
mext.orig_map.m_flags = 0; |
| 611 |
mext.donor_lblk = donor_blk; |
| 612 |
|
| 613 |
ret = ext4_map_blocks(NULL, orig_inode, &mext.orig_map, 0); |
| 614 |
if (ret < 0) |
| 615 |
goto out; |
| 616 |
|
| 617 |
/* Skip moving if it is a hole or a delalloc extent. */ |
| 618 |
if (mext.orig_map.m_flags & |
| 619 |
(EXT4_MAP_MAPPED | EXT4_MAP_UNWRITTEN)) { |
| 620 |
ret = mext_move_extent(&mext, &m_len); |
| 621 |
*moved_len += m_len; |
| 622 |
if (!ret) |
| 623 |
goto next; |
| 624 |
|
| 625 |
/* Move failed or partially failed. */ |
| 626 |
if (m_len) { |
| 627 |
orig_blk += m_len; |
| 628 |
donor_blk += m_len; |
| 629 |
len -= m_len; |
| 630 |
} |
| 631 |
if (ret == -ESTALE) |
| 632 |
continue; |
| 633 |
if (ret == -ENOSPC && |
| 634 |
ext4_should_retry_alloc(sb, &retries)) |
| 635 |
continue; |
| 636 |
if (ret == -EBUSY && |
| 637 |
sbi->s_journal && retries++ < 4 && |
| 638 |
jbd2_journal_force_commit_nested(sbi->s_journal)) |
| 639 |
continue; |
| 640 |
|
| 641 |
goto out; |
| 642 |
} |
| 643 |
next: |
| 644 |
orig_blk += mext.orig_map.m_len; |
| 645 |
donor_blk += mext.orig_map.m_len; |
| 646 |
len -= mext.orig_map.m_len; |
| 647 |
retries = 0; |
| 648 |
} |
| 649 |
|
| 650 |
out: |
| 651 |
if (*moved_len) { |
| 652 |
ext4_discard_preallocations(orig_inode); |
| 653 |
ext4_discard_preallocations(donor_inode); |
| 654 |
} |
| 655 |
|
| 656 |
unlock_two_nondirectories(orig_inode, donor_inode); |
| 657 |
return ret; |
| 658 |
} |