| 1 | // SPDX-License-Identifier: LGPL-2.1 |
| 2 | /* |
| 3 | * Copyright (c) 2008,2009 NEC Software Tohoku, Ltd. |
| 4 | * Written by Takashi Sato <t-sato@yk.jp.nec.com> |
| 5 | * Akira Fujita <a-fujita@rs.jp.nec.com> |
| 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 | /** |
| 17 | * get_ext_path() - Find an extent path for designated logical block number. |
| 18 | * @inode: inode to be searched |
| 19 | * @lblock: logical block number to find an extent path |
| 20 | * @path: pointer to an extent path |
| 21 | * |
| 22 | * ext4_find_extent wrapper. Return an extent path pointer on success, |
| 23 | * or an error pointer on failure. |
| 24 | */ |
| 25 | static inline struct ext4_ext_path * |
| 26 | get_ext_path(struct inode *inode, ext4_lblk_t lblock, |
| 27 | struct ext4_ext_path *path) |
| 28 | { |
| 29 | path = ext4_find_extent(inode, lblock, path, EXT4_EX_NOCACHE); |
| 30 | if (IS_ERR(path)) |
| 31 | return path; |
| 32 | if (path[ext_depth(inode)].p_ext == NULL) { |
| 33 | ext4_free_ext_path(path); |
| 34 | return ERR_PTR(-ENODATA); |
| 35 | } |
| 36 | return path; |
| 37 | } |
| 38 | |
| 39 | /** |
| 40 | * ext4_double_down_write_data_sem() - write lock two inodes's i_data_sem |
| 41 | * @first: inode to be locked |
| 42 | * @second: inode to be locked |
| 43 | * |
| 44 | * Acquire write lock of i_data_sem of the two inodes |
| 45 | */ |
| 46 | void |
| 47 | ext4_double_down_write_data_sem(struct inode *first, struct inode *second) |
| 48 | { |
| 49 | if (first < second) { |
| 50 | down_write(&EXT4_I(first)->i_data_sem); |
| 51 | down_write_nested(&EXT4_I(second)->i_data_sem, I_DATA_SEM_OTHER); |
| 52 | } else { |
| 53 | down_write(&EXT4_I(second)->i_data_sem); |
| 54 | down_write_nested(&EXT4_I(first)->i_data_sem, I_DATA_SEM_OTHER); |
| 55 | |
| 56 | } |
| 57 | } |
| 58 | |
| 59 | /** |
| 60 | * ext4_double_up_write_data_sem - Release two inodes' write lock of i_data_sem |
| 61 | * |
| 62 | * @orig_inode: original inode structure to be released its lock first |
| 63 | * @donor_inode: donor inode structure to be released its lock second |
| 64 | * Release write lock of i_data_sem of two inodes (orig and donor). |
| 65 | */ |
| 66 | void |
| 67 | ext4_double_up_write_data_sem(struct inode *orig_inode, |
| 68 | struct inode *donor_inode) |
| 69 | { |
| 70 | up_write(&EXT4_I(orig_inode)->i_data_sem); |
| 71 | up_write(&EXT4_I(donor_inode)->i_data_sem); |
| 72 | } |
| 73 | |
| 74 | /** |
| 75 | * mext_check_coverage - Check that all extents in range has the same type |
| 76 | * |
| 77 | * @inode: inode in question |
| 78 | * @from: block offset of inode |
| 79 | * @count: block count to be checked |
| 80 | * @unwritten: extents expected to be unwritten |
| 81 | * @err: pointer to save error value |
| 82 | * |
| 83 | * Return 1 if all extents in range has expected type, and zero otherwise. |
| 84 | */ |
| 85 | static int |
| 86 | mext_check_coverage(struct inode *inode, ext4_lblk_t from, ext4_lblk_t count, |
| 87 | int unwritten, int *err) |
| 88 | { |
| 89 | struct ext4_ext_path *path = NULL; |
| 90 | struct ext4_extent *ext; |
| 91 | int ret = 0; |
| 92 | ext4_lblk_t last = from + count; |
| 93 | while (from < last) { |
| 94 | path = get_ext_path(inode, from, path); |
| 95 | if (IS_ERR(path)) { |
| 96 | *err = PTR_ERR(path); |
| 97 | return ret; |
| 98 | } |
| 99 | ext = path[ext_depth(inode)].p_ext; |
| 100 | if (unwritten != ext4_ext_is_unwritten(ext)) |
| 101 | goto out; |
| 102 | from += ext4_ext_get_actual_len(ext); |
| 103 | } |
| 104 | ret = 1; |
| 105 | out: |
| 106 | ext4_free_ext_path(path); |
| 107 | return ret; |
| 108 | } |
| 109 | |
| 110 | /** |
| 111 | * mext_folio_double_lock - Grab and lock folio on both @inode1 and @inode2 |
| 112 | * |
| 113 | * @inode1: the inode structure |
| 114 | * @inode2: the inode structure |
| 115 | * @index1: folio index |
| 116 | * @index2: folio index |
| 117 | * @folio: result folio vector |
| 118 | * |
| 119 | * Grab two locked folio for inode's by inode order |
| 120 | */ |
| 121 | static int |
| 122 | mext_folio_double_lock(struct inode *inode1, struct inode *inode2, |
| 123 | pgoff_t index1, pgoff_t index2, struct folio *folio[2]) |
| 124 | { |
| 125 | struct address_space *mapping[2]; |
| 126 | unsigned int flags; |
| 127 | |
| 128 | BUG_ON(!inode1 || !inode2); |
| 129 | if (inode1 < inode2) { |
| 130 | mapping[0] = inode1->i_mapping; |
| 131 | mapping[1] = inode2->i_mapping; |
| 132 | } else { |
| 133 | swap(index1, index2); |
| 134 | mapping[0] = inode2->i_mapping; |
| 135 | mapping[1] = inode1->i_mapping; |
| 136 | } |
| 137 | |
| 138 | flags = memalloc_nofs_save(); |
| 139 | folio[0] = __filemap_get_folio(mapping[0], index1, FGP_WRITEBEGIN, |
| 140 | mapping_gfp_mask(mapping[0])); |
| 141 | if (IS_ERR(folio[0])) { |
| 142 | memalloc_nofs_restore(flags); |
| 143 | return PTR_ERR(folio[0]); |
| 144 | } |
| 145 | |
| 146 | folio[1] = __filemap_get_folio(mapping[1], index2, FGP_WRITEBEGIN, |
| 147 | mapping_gfp_mask(mapping[1])); |
| 148 | memalloc_nofs_restore(flags); |
| 149 | if (IS_ERR(folio[1])) { |
| 150 | folio_unlock(folio[0]); |
| 151 | folio_put(folio[0]); |
| 152 | return PTR_ERR(folio[1]); |
| 153 | } |
| 154 | /* |
| 155 | * __filemap_get_folio() may not wait on folio's writeback if |
| 156 | * BDI not demand that. But it is reasonable to be very conservative |
| 157 | * here and explicitly wait on folio's writeback |
| 158 | */ |
| 159 | folio_wait_writeback(folio[0]); |
| 160 | folio_wait_writeback(folio[1]); |
| 161 | if (inode1 > inode2) |
| 162 | swap(folio[0], folio[1]); |
| 163 | |
| 164 | return 0; |
| 165 | } |
| 166 | |
| 167 | /* Force folio buffers uptodate w/o dropping folio's lock */ |
| 168 | static int mext_page_mkuptodate(struct folio *folio, size_t from, size_t to) |
| 169 | { |
| 170 | struct inode *inode = folio->mapping->host; |
| 171 | sector_t block; |
| 172 | struct buffer_head *bh, *head; |
| 173 | unsigned int blocksize, block_start, block_end; |
| 174 | int nr = 0; |
| 175 | bool partial = false; |
| 176 | |
| 177 | BUG_ON(!folio_test_locked(folio)); |
| 178 | BUG_ON(folio_test_writeback(folio)); |
| 179 | |
| 180 | if (folio_test_uptodate(folio)) |
| 181 | return 0; |
| 182 | |
| 183 | blocksize = i_blocksize(inode); |
| 184 | head = folio_buffers(folio); |
| 185 | if (!head) |
| 186 | head = create_empty_buffers(folio, blocksize, 0); |
| 187 | |
| 188 | block = folio_pos(folio) >> inode->i_blkbits; |
| 189 | block_end = 0; |
| 190 | bh = head; |
| 191 | do { |
| 192 | block_start = block_end; |
| 193 | block_end = block_start + blocksize; |
| 194 | if (block_end <= from || block_start >= to) { |
| 195 | if (!buffer_uptodate(bh)) |
| 196 | partial = true; |
| 197 | continue; |
| 198 | } |
| 199 | if (buffer_uptodate(bh)) |
| 200 | continue; |
| 201 | if (!buffer_mapped(bh)) { |
| 202 | int err = ext4_get_block(inode, block, bh, 0); |
| 203 | if (err) |
| 204 | return err; |
| 205 | if (!buffer_mapped(bh)) { |
| 206 | folio_zero_range(folio, block_start, blocksize); |
| 207 | set_buffer_uptodate(bh); |
| 208 | continue; |
| 209 | } |
| 210 | } |
| 211 | lock_buffer(bh); |
| 212 | if (buffer_uptodate(bh)) { |
| 213 | unlock_buffer(bh); |
| 214 | continue; |
| 215 | } |
| 216 | ext4_read_bh_nowait(bh, 0, NULL, false); |
| 217 | nr++; |
| 218 | } while (block++, (bh = bh->b_this_page) != head); |
| 219 | |
| 220 | /* No io required */ |
| 221 | if (!nr) |
| 222 | goto out; |
| 223 | |
| 224 | bh = head; |
| 225 | do { |
| 226 | if (bh_offset(bh) + blocksize <= from) |
| 227 | continue; |
| 228 | if (bh_offset(bh) > to) |
| 229 | break; |
| 230 | wait_on_buffer(bh); |
| 231 | if (buffer_uptodate(bh)) |
| 232 | continue; |
| 233 | return -EIO; |
| 234 | } while ((bh = bh->b_this_page) != head); |
| 235 | out: |
| 236 | if (!partial) |
| 237 | folio_mark_uptodate(folio); |
| 238 | return 0; |
| 239 | } |
| 240 | |
| 241 | /** |
| 242 | * move_extent_per_page - Move extent data per page |
| 243 | * |
| 244 | * @o_filp: file structure of original file |
| 245 | * @donor_inode: donor inode |
| 246 | * @orig_page_offset: page index on original file |
| 247 | * @donor_page_offset: page index on donor file |
| 248 | * @data_offset_in_page: block index where data swapping starts |
| 249 | * @block_len_in_page: the number of blocks to be swapped |
| 250 | * @unwritten: orig extent is unwritten or not |
| 251 | * @err: pointer to save return value |
| 252 | * |
| 253 | * Save the data in original inode blocks and replace original inode extents |
| 254 | * with donor inode extents by calling ext4_swap_extents(). |
| 255 | * Finally, write out the saved data in new original inode blocks. Return |
| 256 | * replaced block count. |
| 257 | */ |
| 258 | static int |
| 259 | move_extent_per_page(struct file *o_filp, struct inode *donor_inode, |
| 260 | pgoff_t orig_page_offset, pgoff_t donor_page_offset, |
| 261 | int data_offset_in_page, |
| 262 | int block_len_in_page, int unwritten, int *err) |
| 263 | { |
| 264 | struct inode *orig_inode = file_inode(o_filp); |
| 265 | struct folio *folio[2] = {NULL, NULL}; |
| 266 | handle_t *handle; |
| 267 | ext4_lblk_t orig_blk_offset, donor_blk_offset; |
| 268 | unsigned long blocksize = orig_inode->i_sb->s_blocksize; |
| 269 | unsigned int tmp_data_size, data_size, replaced_size; |
| 270 | int i, err2, jblocks, retries = 0; |
| 271 | int replaced_count = 0; |
| 272 | int from; |
| 273 | int blocks_per_page = PAGE_SIZE >> orig_inode->i_blkbits; |
| 274 | struct super_block *sb = orig_inode->i_sb; |
| 275 | struct buffer_head *bh = NULL; |
| 276 | |
| 277 | /* |
| 278 | * It needs twice the amount of ordinary journal buffers because |
| 279 | * inode and donor_inode may change each different metadata blocks. |
| 280 | */ |
| 281 | again: |
| 282 | *err = 0; |
| 283 | jblocks = ext4_writepage_trans_blocks(orig_inode) * 2; |
| 284 | handle = ext4_journal_start(orig_inode, EXT4_HT_MOVE_EXTENTS, jblocks); |
| 285 | if (IS_ERR(handle)) { |
| 286 | *err = PTR_ERR(handle); |
| 287 | return 0; |
| 288 | } |
| 289 | |
| 290 | orig_blk_offset = orig_page_offset * blocks_per_page + |
| 291 | data_offset_in_page; |
| 292 | |
| 293 | donor_blk_offset = donor_page_offset * blocks_per_page + |
| 294 | data_offset_in_page; |
| 295 | |
| 296 | /* Calculate data_size */ |
| 297 | if ((orig_blk_offset + block_len_in_page - 1) == |
| 298 | ((orig_inode->i_size - 1) >> orig_inode->i_blkbits)) { |
| 299 | /* Replace the last block */ |
| 300 | tmp_data_size = orig_inode->i_size & (blocksize - 1); |
| 301 | /* |
| 302 | * If data_size equal zero, it shows data_size is multiples of |
| 303 | * blocksize. So we set appropriate value. |
| 304 | */ |
| 305 | if (tmp_data_size == 0) |
| 306 | tmp_data_size = blocksize; |
| 307 | |
| 308 | data_size = tmp_data_size + |
| 309 | ((block_len_in_page - 1) << orig_inode->i_blkbits); |
| 310 | } else |
| 311 | data_size = block_len_in_page << orig_inode->i_blkbits; |
| 312 | |
| 313 | replaced_size = data_size; |
| 314 | |
| 315 | *err = mext_folio_double_lock(orig_inode, donor_inode, orig_page_offset, |
| 316 | donor_page_offset, folio); |
| 317 | if (unlikely(*err < 0)) |
| 318 | goto stop_journal; |
| 319 | /* |
| 320 | * If orig extent was unwritten it can become initialized |
| 321 | * at any time after i_data_sem was dropped, in order to |
| 322 | * serialize with delalloc we have recheck extent while we |
| 323 | * hold page's lock, if it is still the case data copy is not |
| 324 | * necessary, just swap data blocks between orig and donor. |
| 325 | */ |
| 326 | if (unwritten) { |
| 327 | ext4_double_down_write_data_sem(orig_inode, donor_inode); |
| 328 | /* If any of extents in range became initialized we have to |
| 329 | * fallback to data copying */ |
| 330 | unwritten = mext_check_coverage(orig_inode, orig_blk_offset, |
| 331 | block_len_in_page, 1, err); |
| 332 | if (*err) |
| 333 | goto drop_data_sem; |
| 334 | |
| 335 | unwritten &= mext_check_coverage(donor_inode, donor_blk_offset, |
| 336 | block_len_in_page, 1, err); |
| 337 | if (*err) |
| 338 | goto drop_data_sem; |
| 339 | |
| 340 | if (!unwritten) { |
| 341 | ext4_double_up_write_data_sem(orig_inode, donor_inode); |
| 342 | goto data_copy; |
| 343 | } |
| 344 | if (!filemap_release_folio(folio[0], 0) || |
| 345 | !filemap_release_folio(folio[1], 0)) { |
| 346 | *err = -EBUSY; |
| 347 | goto drop_data_sem; |
| 348 | } |
| 349 | replaced_count = ext4_swap_extents(handle, orig_inode, |
| 350 | donor_inode, orig_blk_offset, |
| 351 | donor_blk_offset, |
| 352 | block_len_in_page, 1, err); |
| 353 | drop_data_sem: |
| 354 | ext4_double_up_write_data_sem(orig_inode, donor_inode); |
| 355 | goto unlock_folios; |
| 356 | } |
| 357 | data_copy: |
| 358 | from = offset_in_folio(folio[0], |
| 359 | orig_blk_offset << orig_inode->i_blkbits); |
| 360 | *err = mext_page_mkuptodate(folio[0], from, from + replaced_size); |
| 361 | if (*err) |
| 362 | goto unlock_folios; |
| 363 | |
| 364 | /* At this point all buffers in range are uptodate, old mapping layout |
| 365 | * is no longer required, try to drop it now. */ |
| 366 | if (!filemap_release_folio(folio[0], 0) || |
| 367 | !filemap_release_folio(folio[1], 0)) { |
| 368 | *err = -EBUSY; |
| 369 | goto unlock_folios; |
| 370 | } |
| 371 | ext4_double_down_write_data_sem(orig_inode, donor_inode); |
| 372 | replaced_count = ext4_swap_extents(handle, orig_inode, donor_inode, |
| 373 | orig_blk_offset, donor_blk_offset, |
| 374 | block_len_in_page, 1, err); |
| 375 | ext4_double_up_write_data_sem(orig_inode, donor_inode); |
| 376 | if (*err) { |
| 377 | if (replaced_count) { |
| 378 | block_len_in_page = replaced_count; |
| 379 | replaced_size = |
| 380 | block_len_in_page << orig_inode->i_blkbits; |
| 381 | } else |
| 382 | goto unlock_folios; |
| 383 | } |
| 384 | /* Perform all necessary steps similar write_begin()/write_end() |
| 385 | * but keeping in mind that i_size will not change */ |
| 386 | bh = folio_buffers(folio[0]); |
| 387 | if (!bh) |
| 388 | bh = create_empty_buffers(folio[0], |
| 389 | 1 << orig_inode->i_blkbits, 0); |
| 390 | for (i = 0; i < from >> orig_inode->i_blkbits; i++) |
| 391 | bh = bh->b_this_page; |
| 392 | for (i = 0; i < block_len_in_page; i++) { |
| 393 | *err = ext4_get_block(orig_inode, orig_blk_offset + i, bh, 0); |
| 394 | if (*err < 0) |
| 395 | goto repair_branches; |
| 396 | bh = bh->b_this_page; |
| 397 | } |
| 398 | |
| 399 | block_commit_write(folio[0], from, from + replaced_size); |
| 400 | |
| 401 | /* Even in case of data=writeback it is reasonable to pin |
| 402 | * inode to transaction, to prevent unexpected data loss */ |
| 403 | *err = ext4_jbd2_inode_add_write(handle, orig_inode, |
| 404 | (loff_t)orig_page_offset << PAGE_SHIFT, replaced_size); |
| 405 | |
| 406 | unlock_folios: |
| 407 | folio_unlock(folio[0]); |
| 408 | folio_put(folio[0]); |
| 409 | folio_unlock(folio[1]); |
| 410 | folio_put(folio[1]); |
| 411 | stop_journal: |
| 412 | ext4_journal_stop(handle); |
| 413 | if (*err == -ENOSPC && |
| 414 | ext4_should_retry_alloc(sb, &retries)) |
| 415 | goto again; |
| 416 | /* Buffer was busy because probably is pinned to journal transaction, |
| 417 | * force transaction commit may help to free it. */ |
| 418 | if (*err == -EBUSY && retries++ < 4 && EXT4_SB(sb)->s_journal && |
| 419 | jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal)) |
| 420 | goto again; |
| 421 | return replaced_count; |
| 422 | |
| 423 | repair_branches: |
| 424 | /* |
| 425 | * This should never ever happen! |
| 426 | * Extents are swapped already, but we are not able to copy data. |
| 427 | * Try to swap extents to it's original places |
| 428 | */ |
| 429 | ext4_double_down_write_data_sem(orig_inode, donor_inode); |
| 430 | replaced_count = ext4_swap_extents(handle, donor_inode, orig_inode, |
| 431 | orig_blk_offset, donor_blk_offset, |
| 432 | block_len_in_page, 0, &err2); |
| 433 | ext4_double_up_write_data_sem(orig_inode, donor_inode); |
| 434 | if (replaced_count != block_len_in_page) { |
| 435 | ext4_error_inode_block(orig_inode, (sector_t)(orig_blk_offset), |
| 436 | EIO, "Unable to copy data block," |
| 437 | " data will be lost."); |
| 438 | *err = -EIO; |
| 439 | } |
| 440 | replaced_count = 0; |
| 441 | goto unlock_folios; |
| 442 | } |
| 443 | |
| 444 | /** |
| 445 | * mext_check_arguments - Check whether move extent can be done |
| 446 | * |
| 447 | * @orig_inode: original inode |
| 448 | * @donor_inode: donor inode |
| 449 | * @orig_start: logical start offset in block for orig |
| 450 | * @donor_start: logical start offset in block for donor |
| 451 | * @len: the number of blocks to be moved |
| 452 | * |
| 453 | * Check the arguments of ext4_move_extents() whether the files can be |
| 454 | * exchanged with each other. |
| 455 | * Return 0 on success, or a negative error value on failure. |
| 456 | */ |
| 457 | static int |
| 458 | mext_check_arguments(struct inode *orig_inode, |
| 459 | struct inode *donor_inode, __u64 orig_start, |
| 460 | __u64 donor_start, __u64 *len) |
| 461 | { |
| 462 | __u64 orig_eof, donor_eof; |
| 463 | unsigned int blkbits = orig_inode->i_blkbits; |
| 464 | unsigned int blocksize = 1 << blkbits; |
| 465 | |
| 466 | orig_eof = (i_size_read(orig_inode) + blocksize - 1) >> blkbits; |
| 467 | donor_eof = (i_size_read(donor_inode) + blocksize - 1) >> blkbits; |
| 468 | |
| 469 | |
| 470 | if (donor_inode->i_mode & (S_ISUID|S_ISGID)) { |
| 471 | ext4_debug("ext4 move extent: suid or sgid is set" |
| 472 | " to donor file [ino:orig %lu, donor %lu]\n", |
| 473 | orig_inode->i_ino, donor_inode->i_ino); |
| 474 | return -EINVAL; |
| 475 | } |
| 476 | |
| 477 | if (IS_IMMUTABLE(donor_inode) || IS_APPEND(donor_inode)) |
| 478 | return -EPERM; |
| 479 | |
| 480 | /* Ext4 move extent does not support swap files */ |
| 481 | if (IS_SWAPFILE(orig_inode) || IS_SWAPFILE(donor_inode)) { |
| 482 | ext4_debug("ext4 move extent: The argument files should not be swap files [ino:orig %lu, donor %lu]\n", |
| 483 | orig_inode->i_ino, donor_inode->i_ino); |
| 484 | return -ETXTBSY; |
| 485 | } |
| 486 | |
| 487 | if (ext4_is_quota_file(orig_inode) && ext4_is_quota_file(donor_inode)) { |
| 488 | ext4_debug("ext4 move extent: The argument files should not be quota files [ino:orig %lu, donor %lu]\n", |
| 489 | orig_inode->i_ino, donor_inode->i_ino); |
| 490 | return -EOPNOTSUPP; |
| 491 | } |
| 492 | |
| 493 | /* Ext4 move extent supports only extent based file */ |
| 494 | if (!(ext4_test_inode_flag(orig_inode, EXT4_INODE_EXTENTS))) { |
| 495 | ext4_debug("ext4 move extent: orig file is not extents " |
| 496 | "based file [ino:orig %lu]\n", orig_inode->i_ino); |
| 497 | return -EOPNOTSUPP; |
| 498 | } else if (!(ext4_test_inode_flag(donor_inode, EXT4_INODE_EXTENTS))) { |
| 499 | ext4_debug("ext4 move extent: donor file is not extents " |
| 500 | "based file [ino:donor %lu]\n", 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 | /* Start offset should be same */ |
| 510 | if ((orig_start & ~(PAGE_MASK >> orig_inode->i_blkbits)) != |
| 511 | (donor_start & ~(PAGE_MASK >> orig_inode->i_blkbits))) { |
| 512 | ext4_debug("ext4 move extent: orig and donor's start " |
| 513 | "offsets are not aligned [ino:orig %lu, donor %lu]\n", |
| 514 | orig_inode->i_ino, donor_inode->i_ino); |
| 515 | return -EINVAL; |
| 516 | } |
| 517 | |
| 518 | if ((orig_start >= EXT_MAX_BLOCKS) || |
| 519 | (donor_start >= EXT_MAX_BLOCKS) || |
| 520 | (*len > EXT_MAX_BLOCKS) || |
| 521 | (donor_start + *len >= EXT_MAX_BLOCKS) || |
| 522 | (orig_start + *len >= EXT_MAX_BLOCKS)) { |
| 523 | ext4_debug("ext4 move extent: Can't handle over [%u] blocks " |
| 524 | "[ino:orig %lu, donor %lu]\n", EXT_MAX_BLOCKS, |
| 525 | orig_inode->i_ino, donor_inode->i_ino); |
| 526 | return -EINVAL; |
| 527 | } |
| 528 | if (orig_eof <= orig_start) |
| 529 | *len = 0; |
| 530 | else if (orig_eof < orig_start + *len - 1) |
| 531 | *len = orig_eof - orig_start; |
| 532 | if (donor_eof <= donor_start) |
| 533 | *len = 0; |
| 534 | else if (donor_eof < donor_start + *len - 1) |
| 535 | *len = donor_eof - donor_start; |
| 536 | if (!*len) { |
| 537 | ext4_debug("ext4 move extent: len should not be 0 " |
| 538 | "[ino:orig %lu, donor %lu]\n", orig_inode->i_ino, |
| 539 | donor_inode->i_ino); |
| 540 | return -EINVAL; |
| 541 | } |
| 542 | |
| 543 | return 0; |
| 544 | } |
| 545 | |
| 546 | /** |
| 547 | * ext4_move_extents - Exchange the specified range of a file |
| 548 | * |
| 549 | * @o_filp: file structure of the original file |
| 550 | * @d_filp: file structure of the donor file |
| 551 | * @orig_blk: start offset in block for orig |
| 552 | * @donor_blk: start offset in block for donor |
| 553 | * @len: the number of blocks to be moved |
| 554 | * @moved_len: moved block length |
| 555 | * |
| 556 | * This function returns 0 and moved block length is set in moved_len |
| 557 | * if succeed, otherwise returns error value. |
| 558 | * |
| 559 | */ |
| 560 | int |
| 561 | ext4_move_extents(struct file *o_filp, struct file *d_filp, __u64 orig_blk, |
| 562 | __u64 donor_blk, __u64 len, __u64 *moved_len) |
| 563 | { |
| 564 | struct inode *orig_inode = file_inode(o_filp); |
| 565 | struct inode *donor_inode = file_inode(d_filp); |
| 566 | struct ext4_ext_path *path = NULL; |
| 567 | int blocks_per_page = PAGE_SIZE >> orig_inode->i_blkbits; |
| 568 | ext4_lblk_t o_end, o_start = orig_blk; |
| 569 | ext4_lblk_t d_start = donor_blk; |
| 570 | int ret; |
| 571 | |
| 572 | if (orig_inode->i_sb != donor_inode->i_sb) { |
| 573 | ext4_debug("ext4 move extent: The argument files " |
| 574 | "should be in same FS [ino:orig %lu, donor %lu]\n", |
| 575 | orig_inode->i_ino, donor_inode->i_ino); |
| 576 | return -EINVAL; |
| 577 | } |
| 578 | |
| 579 | /* orig and donor should be different inodes */ |
| 580 | if (orig_inode == donor_inode) { |
| 581 | ext4_debug("ext4 move extent: The argument files should not " |
| 582 | "be same inode [ino:orig %lu, donor %lu]\n", |
| 583 | orig_inode->i_ino, donor_inode->i_ino); |
| 584 | return -EINVAL; |
| 585 | } |
| 586 | |
| 587 | /* Regular file check */ |
| 588 | if (!S_ISREG(orig_inode->i_mode) || !S_ISREG(donor_inode->i_mode)) { |
| 589 | ext4_debug("ext4 move extent: The argument files should be " |
| 590 | "regular file [ino:orig %lu, donor %lu]\n", |
| 591 | orig_inode->i_ino, donor_inode->i_ino); |
| 592 | return -EINVAL; |
| 593 | } |
| 594 | |
| 595 | /* TODO: it's not obvious how to swap blocks for inodes with full |
| 596 | journaling enabled */ |
| 597 | if (ext4_should_journal_data(orig_inode) || |
| 598 | ext4_should_journal_data(donor_inode)) { |
| 599 | ext4_msg(orig_inode->i_sb, KERN_ERR, |
| 600 | "Online defrag not supported with data journaling"); |
| 601 | return -EOPNOTSUPP; |
| 602 | } |
| 603 | |
| 604 | if (IS_ENCRYPTED(orig_inode) || IS_ENCRYPTED(donor_inode)) { |
| 605 | ext4_msg(orig_inode->i_sb, KERN_ERR, |
| 606 | "Online defrag not supported for encrypted files"); |
| 607 | return -EOPNOTSUPP; |
| 608 | } |
| 609 | |
| 610 | /* Protect orig and donor inodes against a truncate */ |
| 611 | lock_two_nondirectories(orig_inode, donor_inode); |
| 612 | |
| 613 | /* Wait for all existing dio workers */ |
| 614 | inode_dio_wait(orig_inode); |
| 615 | inode_dio_wait(donor_inode); |
| 616 | |
| 617 | /* Protect extent tree against block allocations via delalloc */ |
| 618 | ext4_double_down_write_data_sem(orig_inode, donor_inode); |
| 619 | /* Check the filesystem environment whether move_extent can be done */ |
| 620 | ret = mext_check_arguments(orig_inode, donor_inode, orig_blk, |
| 621 | donor_blk, &len); |
| 622 | if (ret) |
| 623 | goto out; |
| 624 | o_end = o_start + len; |
| 625 | |
| 626 | *moved_len = 0; |
| 627 | while (o_start < o_end) { |
| 628 | struct ext4_extent *ex; |
| 629 | ext4_lblk_t cur_blk, next_blk; |
| 630 | pgoff_t orig_page_index, donor_page_index; |
| 631 | int offset_in_page; |
| 632 | int unwritten, cur_len; |
| 633 | |
| 634 | path = get_ext_path(orig_inode, o_start, path); |
| 635 | if (IS_ERR(path)) { |
| 636 | ret = PTR_ERR(path); |
| 637 | goto out; |
| 638 | } |
| 639 | ex = path[path->p_depth].p_ext; |
| 640 | cur_blk = le32_to_cpu(ex->ee_block); |
| 641 | cur_len = ext4_ext_get_actual_len(ex); |
| 642 | /* Check hole before the start pos */ |
| 643 | if (cur_blk + cur_len - 1 < o_start) { |
| 644 | next_blk = ext4_ext_next_allocated_block(path); |
| 645 | if (next_blk == EXT_MAX_BLOCKS) { |
| 646 | ret = -ENODATA; |
| 647 | goto out; |
| 648 | } |
| 649 | d_start += next_blk - o_start; |
| 650 | o_start = next_blk; |
| 651 | continue; |
| 652 | /* Check hole after the start pos */ |
| 653 | } else if (cur_blk > o_start) { |
| 654 | /* Skip hole */ |
| 655 | d_start += cur_blk - o_start; |
| 656 | o_start = cur_blk; |
| 657 | /* Extent inside requested range ?*/ |
| 658 | if (cur_blk >= o_end) |
| 659 | goto out; |
| 660 | } else { /* in_range(o_start, o_blk, o_len) */ |
| 661 | cur_len += cur_blk - o_start; |
| 662 | } |
| 663 | unwritten = ext4_ext_is_unwritten(ex); |
| 664 | if (o_end - o_start < cur_len) |
| 665 | cur_len = o_end - o_start; |
| 666 | |
| 667 | orig_page_index = o_start >> (PAGE_SHIFT - |
| 668 | orig_inode->i_blkbits); |
| 669 | donor_page_index = d_start >> (PAGE_SHIFT - |
| 670 | donor_inode->i_blkbits); |
| 671 | offset_in_page = o_start % blocks_per_page; |
| 672 | if (cur_len > blocks_per_page - offset_in_page) |
| 673 | cur_len = blocks_per_page - offset_in_page; |
| 674 | /* |
| 675 | * Up semaphore to avoid following problems: |
| 676 | * a. transaction deadlock among ext4_journal_start, |
| 677 | * ->write_begin via pagefault, and jbd2_journal_commit |
| 678 | * b. racing with ->read_folio, ->write_begin, and |
| 679 | * ext4_get_block in move_extent_per_page |
| 680 | */ |
| 681 | ext4_double_up_write_data_sem(orig_inode, donor_inode); |
| 682 | /* Swap original branches with new branches */ |
| 683 | *moved_len += move_extent_per_page(o_filp, donor_inode, |
| 684 | orig_page_index, donor_page_index, |
| 685 | offset_in_page, cur_len, |
| 686 | unwritten, &ret); |
| 687 | ext4_double_down_write_data_sem(orig_inode, donor_inode); |
| 688 | if (ret < 0) |
| 689 | break; |
| 690 | o_start += cur_len; |
| 691 | d_start += cur_len; |
| 692 | } |
| 693 | |
| 694 | out: |
| 695 | if (*moved_len) { |
| 696 | ext4_discard_preallocations(orig_inode); |
| 697 | ext4_discard_preallocations(donor_inode); |
| 698 | } |
| 699 | |
| 700 | ext4_free_ext_path(path); |
| 701 | ext4_double_up_write_data_sem(orig_inode, donor_inode); |
| 702 | unlock_two_nondirectories(orig_inode, donor_inode); |
| 703 | |
| 704 | return ret; |
| 705 | } |