1 /* 2 * Copyright (c) 2008,2009 NEC Software Tohoku, Ltd. 3 * Written by Takashi Sato <t-sato@yk.jp.nec.com> 4 * Akira Fujita <a-fujita@rs.jp.nec.com> 5 * 6 * This program is free software; you can redistribute it and/or modify it 7 * under the terms of version 2.1 of the GNU Lesser General Public License 8 * as published by the Free Software Foundation. 9 * 10 * This program is distributed in the hope that it will be useful, 11 * but WITHOUT ANY WARRANTY; without even the implied warranty of 12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 13 * GNU General Public License for more details. 14 */ 15 16 #include <linux/fs.h> 17 #include <linux/quotaops.h> 18 #include <linux/slab.h> 19 #include "ext4_jbd2.h" 20 #include "ext4.h" 21 #include "ext4_extents.h" 22 23 /** 24 * get_ext_path - Find an extent path for designated logical block number. 25 * 26 * @inode: an inode which is searched 27 * @lblock: logical block number to find an extent path 28 * @path: pointer to an extent path pointer (for output) 29 * 30 * ext4_ext_find_extent wrapper. Return 0 on success, or a negative error value 31 * on failure. 32 */ 33 static inline int 34 get_ext_path(struct inode *inode, ext4_lblk_t lblock, 35 struct ext4_ext_path **orig_path) 36 { 37 int ret = 0; 38 struct ext4_ext_path *path; 39 40 path = ext4_ext_find_extent(inode, lblock, *orig_path); 41 if (IS_ERR(path)) 42 ret = PTR_ERR(path); 43 else if (path[ext_depth(inode)].p_ext == NULL) 44 ret = -ENODATA; 45 else 46 *orig_path = path; 47 48 return ret; 49 } 50 51 /** 52 * copy_extent_status - Copy the extent's initialization status 53 * 54 * @src: an extent for getting initialize status 55 * @dest: an extent to be set the status 56 */ 57 static void 58 copy_extent_status(struct ext4_extent *src, struct ext4_extent *dest) 59 { 60 if (ext4_ext_is_uninitialized(src)) 61 ext4_ext_mark_uninitialized(dest); 62 else 63 dest->ee_len = cpu_to_le16(ext4_ext_get_actual_len(dest)); 64 } 65 66 /** 67 * mext_next_extent - Search for the next extent and set it to "extent" 68 * 69 * @inode: inode which is searched 70 * @path: this will obtain data for the next extent 71 * @extent: pointer to the next extent we have just gotten 72 * 73 * Search the next extent in the array of ext4_ext_path structure (@path) 74 * and set it to ext4_extent structure (@extent). In addition, the member of 75 * @path (->p_ext) also points the next extent. Return 0 on success, 1 if 76 * ext4_ext_path structure refers to the last extent, or a negative error 77 * value on failure. 78 */ 79 static int 80 mext_next_extent(struct inode *inode, struct ext4_ext_path *path, 81 struct ext4_extent **extent) 82 { 83 struct ext4_extent_header *eh; 84 int ppos, leaf_ppos = path->p_depth; 85 86 ppos = leaf_ppos; 87 if (EXT_LAST_EXTENT(path[ppos].p_hdr) > path[ppos].p_ext) { 88 /* leaf block */ 89 *extent = ++path[ppos].p_ext; 90 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext); 91 return 0; 92 } 93 94 while (--ppos >= 0) { 95 if (EXT_LAST_INDEX(path[ppos].p_hdr) > 96 path[ppos].p_idx) { 97 int cur_ppos = ppos; 98 99 /* index block */ 100 path[ppos].p_idx++; 101 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx); 102 if (path[ppos+1].p_bh) 103 brelse(path[ppos+1].p_bh); 104 path[ppos+1].p_bh = 105 sb_bread(inode->i_sb, path[ppos].p_block); 106 if (!path[ppos+1].p_bh) 107 return -EIO; 108 path[ppos+1].p_hdr = 109 ext_block_hdr(path[ppos+1].p_bh); 110 111 /* Halfway index block */ 112 while (++cur_ppos < leaf_ppos) { 113 path[cur_ppos].p_idx = 114 EXT_FIRST_INDEX(path[cur_ppos].p_hdr); 115 path[cur_ppos].p_block = 116 ext4_idx_pblock(path[cur_ppos].p_idx); 117 if (path[cur_ppos+1].p_bh) 118 brelse(path[cur_ppos+1].p_bh); 119 path[cur_ppos+1].p_bh = sb_bread(inode->i_sb, 120 path[cur_ppos].p_block); 121 if (!path[cur_ppos+1].p_bh) 122 return -EIO; 123 path[cur_ppos+1].p_hdr = 124 ext_block_hdr(path[cur_ppos+1].p_bh); 125 } 126 127 path[leaf_ppos].p_ext = *extent = NULL; 128 129 eh = path[leaf_ppos].p_hdr; 130 if (le16_to_cpu(eh->eh_entries) == 0) 131 /* empty leaf is found */ 132 return -ENODATA; 133 134 /* leaf block */ 135 path[leaf_ppos].p_ext = *extent = 136 EXT_FIRST_EXTENT(path[leaf_ppos].p_hdr); 137 path[leaf_ppos].p_block = 138 ext4_ext_pblock(path[leaf_ppos].p_ext); 139 return 0; 140 } 141 } 142 /* We found the last extent */ 143 return 1; 144 } 145 146 /** 147 * ext4_double_down_write_data_sem - Acquire two inodes' write lock 148 * of i_data_sem 149 * 150 * Acquire write lock of i_data_sem of the two inodes 151 */ 152 void 153 ext4_double_down_write_data_sem(struct inode *first, struct inode *second) 154 { 155 if (first < second) { 156 down_write(&EXT4_I(first)->i_data_sem); 157 down_write_nested(&EXT4_I(second)->i_data_sem, SINGLE_DEPTH_NESTING); 158 } else { 159 down_write(&EXT4_I(second)->i_data_sem); 160 down_write_nested(&EXT4_I(first)->i_data_sem, SINGLE_DEPTH_NESTING); 161 162 } 163 } 164 165 /** 166 * ext4_double_up_write_data_sem - Release two inodes' write lock of i_data_sem 167 * 168 * @orig_inode: original inode structure to be released its lock first 169 * @donor_inode: donor inode structure to be released its lock second 170 * Release write lock of i_data_sem of two inodes (orig and donor). 171 */ 172 void 173 ext4_double_up_write_data_sem(struct inode *orig_inode, 174 struct inode *donor_inode) 175 { 176 up_write(&EXT4_I(orig_inode)->i_data_sem); 177 up_write(&EXT4_I(donor_inode)->i_data_sem); 178 } 179 180 /** 181 * mext_insert_across_blocks - Insert extents across leaf block 182 * 183 * @handle: journal handle 184 * @orig_inode: original inode 185 * @o_start: first original extent to be changed 186 * @o_end: last original extent to be changed 187 * @start_ext: first new extent to be inserted 188 * @new_ext: middle of new extent to be inserted 189 * @end_ext: last new extent to be inserted 190 * 191 * Allocate a new leaf block and insert extents into it. Return 0 on success, 192 * or a negative error value on failure. 193 */ 194 static int 195 mext_insert_across_blocks(handle_t *handle, struct inode *orig_inode, 196 struct ext4_extent *o_start, struct ext4_extent *o_end, 197 struct ext4_extent *start_ext, struct ext4_extent *new_ext, 198 struct ext4_extent *end_ext) 199 { 200 struct ext4_ext_path *orig_path = NULL; 201 ext4_lblk_t eblock = 0; 202 int new_flag = 0; 203 int end_flag = 0; 204 int err = 0; 205 206 if (start_ext->ee_len && new_ext->ee_len && end_ext->ee_len) { 207 if (o_start == o_end) { 208 209 /* start_ext new_ext end_ext 210 * donor |---------|-----------|--------| 211 * orig |------------------------------| 212 */ 213 end_flag = 1; 214 } else { 215 216 /* start_ext new_ext end_ext 217 * donor |---------|----------|---------| 218 * orig |---------------|--------------| 219 */ 220 o_end->ee_block = end_ext->ee_block; 221 o_end->ee_len = end_ext->ee_len; 222 ext4_ext_store_pblock(o_end, ext4_ext_pblock(end_ext)); 223 } 224 225 o_start->ee_len = start_ext->ee_len; 226 eblock = le32_to_cpu(start_ext->ee_block); 227 new_flag = 1; 228 229 } else if (start_ext->ee_len && new_ext->ee_len && 230 !end_ext->ee_len && o_start == o_end) { 231 232 /* start_ext new_ext 233 * donor |--------------|---------------| 234 * orig |------------------------------| 235 */ 236 o_start->ee_len = start_ext->ee_len; 237 eblock = le32_to_cpu(start_ext->ee_block); 238 new_flag = 1; 239 240 } else if (!start_ext->ee_len && new_ext->ee_len && 241 end_ext->ee_len && o_start == o_end) { 242 243 /* new_ext end_ext 244 * donor |--------------|---------------| 245 * orig |------------------------------| 246 */ 247 o_end->ee_block = end_ext->ee_block; 248 o_end->ee_len = end_ext->ee_len; 249 ext4_ext_store_pblock(o_end, ext4_ext_pblock(end_ext)); 250 251 /* 252 * Set 0 to the extent block if new_ext was 253 * the first block. 254 */ 255 if (new_ext->ee_block) 256 eblock = le32_to_cpu(new_ext->ee_block); 257 258 new_flag = 1; 259 } else { 260 ext4_debug("ext4 move extent: Unexpected insert case\n"); 261 return -EIO; 262 } 263 264 if (new_flag) { 265 err = get_ext_path(orig_inode, eblock, &orig_path); 266 if (err) 267 goto out; 268 269 if (ext4_ext_insert_extent(handle, orig_inode, 270 orig_path, new_ext, 0)) 271 goto out; 272 } 273 274 if (end_flag) { 275 err = get_ext_path(orig_inode, 276 le32_to_cpu(end_ext->ee_block) - 1, &orig_path); 277 if (err) 278 goto out; 279 280 if (ext4_ext_insert_extent(handle, orig_inode, 281 orig_path, end_ext, 0)) 282 goto out; 283 } 284 out: 285 if (orig_path) { 286 ext4_ext_drop_refs(orig_path); 287 kfree(orig_path); 288 } 289 290 return err; 291 292 } 293 294 /** 295 * mext_insert_inside_block - Insert new extent to the extent block 296 * 297 * @o_start: first original extent to be moved 298 * @o_end: last original extent to be moved 299 * @start_ext: first new extent to be inserted 300 * @new_ext: middle of new extent to be inserted 301 * @end_ext: last new extent to be inserted 302 * @eh: extent header of target leaf block 303 * @range_to_move: used to decide how to insert extent 304 * 305 * Insert extents into the leaf block. The extent (@o_start) is overwritten 306 * by inserted extents. 307 */ 308 static void 309 mext_insert_inside_block(struct ext4_extent *o_start, 310 struct ext4_extent *o_end, 311 struct ext4_extent *start_ext, 312 struct ext4_extent *new_ext, 313 struct ext4_extent *end_ext, 314 struct ext4_extent_header *eh, 315 int range_to_move) 316 { 317 int i = 0; 318 unsigned long len; 319 320 /* Move the existing extents */ 321 if (range_to_move && o_end < EXT_LAST_EXTENT(eh)) { 322 len = (unsigned long)(EXT_LAST_EXTENT(eh) + 1) - 323 (unsigned long)(o_end + 1); 324 memmove(o_end + 1 + range_to_move, o_end + 1, len); 325 } 326 327 /* Insert start entry */ 328 if (start_ext->ee_len) 329 o_start[i++].ee_len = start_ext->ee_len; 330 331 /* Insert new entry */ 332 if (new_ext->ee_len) { 333 o_start[i] = *new_ext; 334 ext4_ext_store_pblock(&o_start[i++], ext4_ext_pblock(new_ext)); 335 } 336 337 /* Insert end entry */ 338 if (end_ext->ee_len) 339 o_start[i] = *end_ext; 340 341 /* Increment the total entries counter on the extent block */ 342 le16_add_cpu(&eh->eh_entries, range_to_move); 343 } 344 345 /** 346 * mext_insert_extents - Insert new extent 347 * 348 * @handle: journal handle 349 * @orig_inode: original inode 350 * @orig_path: path indicates first extent to be changed 351 * @o_start: first original extent to be changed 352 * @o_end: last original extent to be changed 353 * @start_ext: first new extent to be inserted 354 * @new_ext: middle of new extent to be inserted 355 * @end_ext: last new extent to be inserted 356 * 357 * Call the function to insert extents. If we cannot add more extents into 358 * the leaf block, we call mext_insert_across_blocks() to create a 359 * new leaf block. Otherwise call mext_insert_inside_block(). Return 0 360 * on success, or a negative error value on failure. 361 */ 362 static int 363 mext_insert_extents(handle_t *handle, struct inode *orig_inode, 364 struct ext4_ext_path *orig_path, 365 struct ext4_extent *o_start, 366 struct ext4_extent *o_end, 367 struct ext4_extent *start_ext, 368 struct ext4_extent *new_ext, 369 struct ext4_extent *end_ext) 370 { 371 struct ext4_extent_header *eh; 372 unsigned long need_slots, slots_range; 373 int range_to_move, depth, ret; 374 375 /* 376 * The extents need to be inserted 377 * start_extent + new_extent + end_extent. 378 */ 379 need_slots = (start_ext->ee_len ? 1 : 0) + (end_ext->ee_len ? 1 : 0) + 380 (new_ext->ee_len ? 1 : 0); 381 382 /* The number of slots between start and end */ 383 slots_range = ((unsigned long)(o_end + 1) - (unsigned long)o_start + 1) 384 / sizeof(struct ext4_extent); 385 386 /* Range to move the end of extent */ 387 range_to_move = need_slots - slots_range; 388 depth = orig_path->p_depth; 389 orig_path += depth; 390 eh = orig_path->p_hdr; 391 392 if (depth) { 393 /* Register to journal */ 394 ret = ext4_journal_get_write_access(handle, orig_path->p_bh); 395 if (ret) 396 return ret; 397 } 398 399 /* Expansion */ 400 if (range_to_move > 0 && 401 (range_to_move > le16_to_cpu(eh->eh_max) 402 - le16_to_cpu(eh->eh_entries))) { 403 404 ret = mext_insert_across_blocks(handle, orig_inode, o_start, 405 o_end, start_ext, new_ext, end_ext); 406 if (ret < 0) 407 return ret; 408 } else 409 mext_insert_inside_block(o_start, o_end, start_ext, new_ext, 410 end_ext, eh, range_to_move); 411 412 return ext4_ext_dirty(handle, orig_inode, orig_path); 413 } 414 415 /** 416 * mext_leaf_block - Move one leaf extent block into the inode. 417 * 418 * @handle: journal handle 419 * @orig_inode: original inode 420 * @orig_path: path indicates first extent to be changed 421 * @dext: donor extent 422 * @from: start offset on the target file 423 * 424 * In order to insert extents into the leaf block, we must divide the extent 425 * in the leaf block into three extents. The one is located to be inserted 426 * extents, and the others are located around it. 427 * 428 * Therefore, this function creates structures to save extents of the leaf 429 * block, and inserts extents by calling mext_insert_extents() with 430 * created extents. Return 0 on success, or a negative error value on failure. 431 */ 432 static int 433 mext_leaf_block(handle_t *handle, struct inode *orig_inode, 434 struct ext4_ext_path *orig_path, struct ext4_extent *dext, 435 ext4_lblk_t *from) 436 { 437 struct ext4_extent *oext, *o_start, *o_end, *prev_ext; 438 struct ext4_extent new_ext, start_ext, end_ext; 439 ext4_lblk_t new_ext_end; 440 int oext_alen, new_ext_alen, end_ext_alen; 441 int depth = ext_depth(orig_inode); 442 int ret; 443 444 start_ext.ee_block = end_ext.ee_block = 0; 445 o_start = o_end = oext = orig_path[depth].p_ext; 446 oext_alen = ext4_ext_get_actual_len(oext); 447 start_ext.ee_len = end_ext.ee_len = 0; 448 449 new_ext.ee_block = cpu_to_le32(*from); 450 ext4_ext_store_pblock(&new_ext, ext4_ext_pblock(dext)); 451 new_ext.ee_len = dext->ee_len; 452 new_ext_alen = ext4_ext_get_actual_len(&new_ext); 453 new_ext_end = le32_to_cpu(new_ext.ee_block) + new_ext_alen - 1; 454 455 /* 456 * Case: original extent is first 457 * oext |--------| 458 * new_ext |--| 459 * start_ext |--| 460 */ 461 if (le32_to_cpu(oext->ee_block) < le32_to_cpu(new_ext.ee_block) && 462 le32_to_cpu(new_ext.ee_block) < 463 le32_to_cpu(oext->ee_block) + oext_alen) { 464 start_ext.ee_len = cpu_to_le16(le32_to_cpu(new_ext.ee_block) - 465 le32_to_cpu(oext->ee_block)); 466 start_ext.ee_block = oext->ee_block; 467 copy_extent_status(oext, &start_ext); 468 } else if (oext > EXT_FIRST_EXTENT(orig_path[depth].p_hdr)) { 469 prev_ext = oext - 1; 470 /* 471 * We can merge new_ext into previous extent, 472 * if these are contiguous and same extent type. 473 */ 474 if (ext4_can_extents_be_merged(orig_inode, prev_ext, 475 &new_ext)) { 476 o_start = prev_ext; 477 start_ext.ee_len = cpu_to_le16( 478 ext4_ext_get_actual_len(prev_ext) + 479 new_ext_alen); 480 start_ext.ee_block = oext->ee_block; 481 copy_extent_status(prev_ext, &start_ext); 482 new_ext.ee_len = 0; 483 } 484 } 485 486 /* 487 * Case: new_ext_end must be less than oext 488 * oext |-----------| 489 * new_ext |-------| 490 */ 491 if (le32_to_cpu(oext->ee_block) + oext_alen - 1 < new_ext_end) { 492 EXT4_ERROR_INODE(orig_inode, 493 "new_ext_end(%u) should be less than or equal to " 494 "oext->ee_block(%u) + oext_alen(%d) - 1", 495 new_ext_end, le32_to_cpu(oext->ee_block), 496 oext_alen); 497 ret = -EIO; 498 goto out; 499 } 500 501 /* 502 * Case: new_ext is smaller than original extent 503 * oext |---------------| 504 * new_ext |-----------| 505 * end_ext |---| 506 */ 507 if (le32_to_cpu(oext->ee_block) <= new_ext_end && 508 new_ext_end < le32_to_cpu(oext->ee_block) + oext_alen - 1) { 509 end_ext.ee_len = 510 cpu_to_le16(le32_to_cpu(oext->ee_block) + 511 oext_alen - 1 - new_ext_end); 512 copy_extent_status(oext, &end_ext); 513 end_ext_alen = ext4_ext_get_actual_len(&end_ext); 514 ext4_ext_store_pblock(&end_ext, 515 (ext4_ext_pblock(o_end) + oext_alen - end_ext_alen)); 516 end_ext.ee_block = 517 cpu_to_le32(le32_to_cpu(o_end->ee_block) + 518 oext_alen - end_ext_alen); 519 } 520 521 ret = mext_insert_extents(handle, orig_inode, orig_path, o_start, 522 o_end, &start_ext, &new_ext, &end_ext); 523 out: 524 return ret; 525 } 526 527 /** 528 * mext_calc_swap_extents - Calculate extents for extent swapping. 529 * 530 * @tmp_dext: the extent that will belong to the original inode 531 * @tmp_oext: the extent that will belong to the donor inode 532 * @orig_off: block offset of original inode 533 * @donor_off: block offset of donor inode 534 * @max_count: the maximum length of extents 535 * 536 * Return 0 on success, or a negative error value on failure. 537 */ 538 static int 539 mext_calc_swap_extents(struct ext4_extent *tmp_dext, 540 struct ext4_extent *tmp_oext, 541 ext4_lblk_t orig_off, ext4_lblk_t donor_off, 542 ext4_lblk_t max_count) 543 { 544 ext4_lblk_t diff, orig_diff; 545 struct ext4_extent dext_old, oext_old; 546 547 BUG_ON(orig_off != donor_off); 548 549 /* original and donor extents have to cover the same block offset */ 550 if (orig_off < le32_to_cpu(tmp_oext->ee_block) || 551 le32_to_cpu(tmp_oext->ee_block) + 552 ext4_ext_get_actual_len(tmp_oext) - 1 < orig_off) 553 return -ENODATA; 554 555 if (orig_off < le32_to_cpu(tmp_dext->ee_block) || 556 le32_to_cpu(tmp_dext->ee_block) + 557 ext4_ext_get_actual_len(tmp_dext) - 1 < orig_off) 558 return -ENODATA; 559 560 dext_old = *tmp_dext; 561 oext_old = *tmp_oext; 562 563 /* When tmp_dext is too large, pick up the target range. */ 564 diff = donor_off - le32_to_cpu(tmp_dext->ee_block); 565 566 ext4_ext_store_pblock(tmp_dext, ext4_ext_pblock(tmp_dext) + diff); 567 le32_add_cpu(&tmp_dext->ee_block, diff); 568 le16_add_cpu(&tmp_dext->ee_len, -diff); 569 570 if (max_count < ext4_ext_get_actual_len(tmp_dext)) 571 tmp_dext->ee_len = cpu_to_le16(max_count); 572 573 orig_diff = orig_off - le32_to_cpu(tmp_oext->ee_block); 574 ext4_ext_store_pblock(tmp_oext, ext4_ext_pblock(tmp_oext) + orig_diff); 575 576 /* Adjust extent length if donor extent is larger than orig */ 577 if (ext4_ext_get_actual_len(tmp_dext) > 578 ext4_ext_get_actual_len(tmp_oext) - orig_diff) 579 tmp_dext->ee_len = cpu_to_le16(le16_to_cpu(tmp_oext->ee_len) - 580 orig_diff); 581 582 tmp_oext->ee_len = cpu_to_le16(ext4_ext_get_actual_len(tmp_dext)); 583 584 copy_extent_status(&oext_old, tmp_dext); 585 copy_extent_status(&dext_old, tmp_oext); 586 587 return 0; 588 } 589 590 /** 591 * mext_check_coverage - Check that all extents in range has the same type 592 * 593 * @inode: inode in question 594 * @from: block offset of inode 595 * @count: block count to be checked 596 * @uninit: extents expected to be uninitialized 597 * @err: pointer to save error value 598 * 599 * Return 1 if all extents in range has expected type, and zero otherwise. 600 */ 601 static int 602 mext_check_coverage(struct inode *inode, ext4_lblk_t from, ext4_lblk_t count, 603 int uninit, int *err) 604 { 605 struct ext4_ext_path *path = NULL; 606 struct ext4_extent *ext; 607 int ret = 0; 608 ext4_lblk_t last = from + count; 609 while (from < last) { 610 *err = get_ext_path(inode, from, &path); 611 if (*err) 612 goto out; 613 ext = path[ext_depth(inode)].p_ext; 614 if (uninit != ext4_ext_is_uninitialized(ext)) 615 goto out; 616 from += ext4_ext_get_actual_len(ext); 617 ext4_ext_drop_refs(path); 618 } 619 ret = 1; 620 out: 621 if (path) { 622 ext4_ext_drop_refs(path); 623 kfree(path); 624 } 625 return ret; 626 } 627 628 /** 629 * mext_replace_branches - Replace original extents with new extents 630 * 631 * @handle: journal handle 632 * @orig_inode: original inode 633 * @donor_inode: donor inode 634 * @from: block offset of orig_inode 635 * @count: block count to be replaced 636 * @err: pointer to save return value 637 * 638 * Replace original inode extents and donor inode extents page by page. 639 * We implement this replacement in the following three steps: 640 * 1. Save the block information of original and donor inodes into 641 * dummy extents. 642 * 2. Change the block information of original inode to point at the 643 * donor inode blocks. 644 * 3. Change the block information of donor inode to point at the saved 645 * original inode blocks in the dummy extents. 646 * 647 * Return replaced block count. 648 */ 649 static int 650 mext_replace_branches(handle_t *handle, struct inode *orig_inode, 651 struct inode *donor_inode, ext4_lblk_t from, 652 ext4_lblk_t count, int *err) 653 { 654 struct ext4_ext_path *orig_path = NULL; 655 struct ext4_ext_path *donor_path = NULL; 656 struct ext4_extent *oext, *dext; 657 struct ext4_extent tmp_dext, tmp_oext; 658 ext4_lblk_t orig_off = from, donor_off = from; 659 int depth; 660 int replaced_count = 0; 661 int dext_alen; 662 663 *err = ext4_es_remove_extent(orig_inode, from, count); 664 if (*err) 665 goto out; 666 667 *err = ext4_es_remove_extent(donor_inode, from, count); 668 if (*err) 669 goto out; 670 671 /* Get the original extent for the block "orig_off" */ 672 *err = get_ext_path(orig_inode, orig_off, &orig_path); 673 if (*err) 674 goto out; 675 676 /* Get the donor extent for the head */ 677 *err = get_ext_path(donor_inode, donor_off, &donor_path); 678 if (*err) 679 goto out; 680 depth = ext_depth(orig_inode); 681 oext = orig_path[depth].p_ext; 682 tmp_oext = *oext; 683 684 depth = ext_depth(donor_inode); 685 dext = donor_path[depth].p_ext; 686 if (unlikely(!dext)) 687 goto missing_donor_extent; 688 tmp_dext = *dext; 689 690 *err = mext_calc_swap_extents(&tmp_dext, &tmp_oext, orig_off, 691 donor_off, count); 692 if (*err) 693 goto out; 694 695 /* Loop for the donor extents */ 696 while (1) { 697 /* The extent for donor must be found. */ 698 if (unlikely(!dext)) { 699 missing_donor_extent: 700 EXT4_ERROR_INODE(donor_inode, 701 "The extent for donor must be found"); 702 *err = -EIO; 703 goto out; 704 } else if (donor_off != le32_to_cpu(tmp_dext.ee_block)) { 705 EXT4_ERROR_INODE(donor_inode, 706 "Donor offset(%u) and the first block of donor " 707 "extent(%u) should be equal", 708 donor_off, 709 le32_to_cpu(tmp_dext.ee_block)); 710 *err = -EIO; 711 goto out; 712 } 713 714 /* Set donor extent to orig extent */ 715 *err = mext_leaf_block(handle, orig_inode, 716 orig_path, &tmp_dext, &orig_off); 717 if (*err) 718 goto out; 719 720 /* Set orig extent to donor extent */ 721 *err = mext_leaf_block(handle, donor_inode, 722 donor_path, &tmp_oext, &donor_off); 723 if (*err) 724 goto out; 725 726 dext_alen = ext4_ext_get_actual_len(&tmp_dext); 727 replaced_count += dext_alen; 728 donor_off += dext_alen; 729 orig_off += dext_alen; 730 731 BUG_ON(replaced_count > count); 732 /* Already moved the expected blocks */ 733 if (replaced_count >= count) 734 break; 735 736 if (orig_path) 737 ext4_ext_drop_refs(orig_path); 738 *err = get_ext_path(orig_inode, orig_off, &orig_path); 739 if (*err) 740 goto out; 741 depth = ext_depth(orig_inode); 742 oext = orig_path[depth].p_ext; 743 tmp_oext = *oext; 744 745 if (donor_path) 746 ext4_ext_drop_refs(donor_path); 747 *err = get_ext_path(donor_inode, donor_off, &donor_path); 748 if (*err) 749 goto out; 750 depth = ext_depth(donor_inode); 751 dext = donor_path[depth].p_ext; 752 tmp_dext = *dext; 753 754 *err = mext_calc_swap_extents(&tmp_dext, &tmp_oext, orig_off, 755 donor_off, count - replaced_count); 756 if (*err) 757 goto out; 758 } 759 760 out: 761 if (orig_path) { 762 ext4_ext_drop_refs(orig_path); 763 kfree(orig_path); 764 } 765 if (donor_path) { 766 ext4_ext_drop_refs(donor_path); 767 kfree(donor_path); 768 } 769 770 return replaced_count; 771 } 772 773 /** 774 * mext_page_double_lock - Grab and lock pages on both @inode1 and @inode2 775 * 776 * @inode1: the inode structure 777 * @inode2: the inode structure 778 * @index: page index 779 * @page: result page vector 780 * 781 * Grab two locked pages for inode's by inode order 782 */ 783 static int 784 mext_page_double_lock(struct inode *inode1, struct inode *inode2, 785 pgoff_t index, struct page *page[2]) 786 { 787 struct address_space *mapping[2]; 788 unsigned fl = AOP_FLAG_NOFS; 789 790 BUG_ON(!inode1 || !inode2); 791 if (inode1 < inode2) { 792 mapping[0] = inode1->i_mapping; 793 mapping[1] = inode2->i_mapping; 794 } else { 795 mapping[0] = inode2->i_mapping; 796 mapping[1] = inode1->i_mapping; 797 } 798 799 page[0] = grab_cache_page_write_begin(mapping[0], index, fl); 800 if (!page[0]) 801 return -ENOMEM; 802 803 page[1] = grab_cache_page_write_begin(mapping[1], index, fl); 804 if (!page[1]) { 805 unlock_page(page[0]); 806 page_cache_release(page[0]); 807 return -ENOMEM; 808 } 809 /* 810 * grab_cache_page_write_begin() may not wait on page's writeback if 811 * BDI not demand that. But it is reasonable to be very conservative 812 * here and explicitly wait on page's writeback 813 */ 814 wait_on_page_writeback(page[0]); 815 wait_on_page_writeback(page[1]); 816 if (inode1 > inode2) { 817 struct page *tmp; 818 tmp = page[0]; 819 page[0] = page[1]; 820 page[1] = tmp; 821 } 822 return 0; 823 } 824 825 /* Force page buffers uptodate w/o dropping page's lock */ 826 static int 827 mext_page_mkuptodate(struct page *page, unsigned from, unsigned to) 828 { 829 struct inode *inode = page->mapping->host; 830 sector_t block; 831 struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE]; 832 unsigned int blocksize, block_start, block_end; 833 int i, err, nr = 0, partial = 0; 834 BUG_ON(!PageLocked(page)); 835 BUG_ON(PageWriteback(page)); 836 837 if (PageUptodate(page)) 838 return 0; 839 840 blocksize = 1 << inode->i_blkbits; 841 if (!page_has_buffers(page)) 842 create_empty_buffers(page, blocksize, 0); 843 844 head = page_buffers(page); 845 block = (sector_t)page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits); 846 for (bh = head, block_start = 0; bh != head || !block_start; 847 block++, block_start = block_end, bh = bh->b_this_page) { 848 block_end = block_start + blocksize; 849 if (block_end <= from || block_start >= to) { 850 if (!buffer_uptodate(bh)) 851 partial = 1; 852 continue; 853 } 854 if (buffer_uptodate(bh)) 855 continue; 856 if (!buffer_mapped(bh)) { 857 err = ext4_get_block(inode, block, bh, 0); 858 if (err) { 859 SetPageError(page); 860 return err; 861 } 862 if (!buffer_mapped(bh)) { 863 zero_user(page, block_start, blocksize); 864 if (!err) 865 set_buffer_uptodate(bh); 866 continue; 867 } 868 } 869 BUG_ON(nr >= MAX_BUF_PER_PAGE); 870 arr[nr++] = bh; 871 } 872 /* No io required */ 873 if (!nr) 874 goto out; 875 876 for (i = 0; i < nr; i++) { 877 bh = arr[i]; 878 if (!bh_uptodate_or_lock(bh)) { 879 err = bh_submit_read(bh); 880 if (err) 881 return err; 882 } 883 } 884 out: 885 if (!partial) 886 SetPageUptodate(page); 887 return 0; 888 } 889 890 /** 891 * move_extent_per_page - Move extent data per page 892 * 893 * @o_filp: file structure of original file 894 * @donor_inode: donor inode 895 * @orig_page_offset: page index on original file 896 * @data_offset_in_page: block index where data swapping starts 897 * @block_len_in_page: the number of blocks to be swapped 898 * @uninit: orig extent is uninitialized or not 899 * @err: pointer to save return value 900 * 901 * Save the data in original inode blocks and replace original inode extents 902 * with donor inode extents by calling mext_replace_branches(). 903 * Finally, write out the saved data in new original inode blocks. Return 904 * replaced block count. 905 */ 906 static int 907 move_extent_per_page(struct file *o_filp, struct inode *donor_inode, 908 pgoff_t orig_page_offset, int data_offset_in_page, 909 int block_len_in_page, int uninit, int *err) 910 { 911 struct inode *orig_inode = file_inode(o_filp); 912 struct page *pagep[2] = {NULL, NULL}; 913 handle_t *handle; 914 ext4_lblk_t orig_blk_offset; 915 long long offs = orig_page_offset << PAGE_CACHE_SHIFT; 916 unsigned long blocksize = orig_inode->i_sb->s_blocksize; 917 unsigned int w_flags = 0; 918 unsigned int tmp_data_size, data_size, replaced_size; 919 int err2, jblocks, retries = 0; 920 int replaced_count = 0; 921 int from = data_offset_in_page << orig_inode->i_blkbits; 922 int blocks_per_page = PAGE_CACHE_SIZE >> orig_inode->i_blkbits; 923 924 /* 925 * It needs twice the amount of ordinary journal buffers because 926 * inode and donor_inode may change each different metadata blocks. 927 */ 928 again: 929 *err = 0; 930 jblocks = ext4_writepage_trans_blocks(orig_inode) * 2; 931 handle = ext4_journal_start(orig_inode, EXT4_HT_MOVE_EXTENTS, jblocks); 932 if (IS_ERR(handle)) { 933 *err = PTR_ERR(handle); 934 return 0; 935 } 936 937 if (segment_eq(get_fs(), KERNEL_DS)) 938 w_flags |= AOP_FLAG_UNINTERRUPTIBLE; 939 940 orig_blk_offset = orig_page_offset * blocks_per_page + 941 data_offset_in_page; 942 943 offs = (long long)orig_blk_offset << orig_inode->i_blkbits; 944 945 /* Calculate data_size */ 946 if ((orig_blk_offset + block_len_in_page - 1) == 947 ((orig_inode->i_size - 1) >> orig_inode->i_blkbits)) { 948 /* Replace the last block */ 949 tmp_data_size = orig_inode->i_size & (blocksize - 1); 950 /* 951 * If data_size equal zero, it shows data_size is multiples of 952 * blocksize. So we set appropriate value. 953 */ 954 if (tmp_data_size == 0) 955 tmp_data_size = blocksize; 956 957 data_size = tmp_data_size + 958 ((block_len_in_page - 1) << orig_inode->i_blkbits); 959 } else 960 data_size = block_len_in_page << orig_inode->i_blkbits; 961 962 replaced_size = data_size; 963 964 *err = mext_page_double_lock(orig_inode, donor_inode, orig_page_offset, 965 pagep); 966 if (unlikely(*err < 0)) 967 goto stop_journal; 968 /* 969 * If orig extent was uninitialized it can become initialized 970 * at any time after i_data_sem was dropped, in order to 971 * serialize with delalloc we have recheck extent while we 972 * hold page's lock, if it is still the case data copy is not 973 * necessary, just swap data blocks between orig and donor. 974 */ 975 if (uninit) { 976 ext4_double_down_write_data_sem(orig_inode, donor_inode); 977 /* If any of extents in range became initialized we have to 978 * fallback to data copying */ 979 uninit = mext_check_coverage(orig_inode, orig_blk_offset, 980 block_len_in_page, 1, err); 981 if (*err) 982 goto drop_data_sem; 983 984 uninit &= mext_check_coverage(donor_inode, orig_blk_offset, 985 block_len_in_page, 1, err); 986 if (*err) 987 goto drop_data_sem; 988 989 if (!uninit) { 990 ext4_double_up_write_data_sem(orig_inode, donor_inode); 991 goto data_copy; 992 } 993 if ((page_has_private(pagep[0]) && 994 !try_to_release_page(pagep[0], 0)) || 995 (page_has_private(pagep[1]) && 996 !try_to_release_page(pagep[1], 0))) { 997 *err = -EBUSY; 998 goto drop_data_sem; 999 } 1000 replaced_count = mext_replace_branches(handle, orig_inode, 1001 donor_inode, orig_blk_offset, 1002 block_len_in_page, err); 1003 drop_data_sem: 1004 ext4_double_up_write_data_sem(orig_inode, donor_inode); 1005 goto unlock_pages; 1006 } 1007 data_copy: 1008 *err = mext_page_mkuptodate(pagep[0], from, from + replaced_size); 1009 if (*err) 1010 goto unlock_pages; 1011 1012 /* At this point all buffers in range are uptodate, old mapping layout 1013 * is no longer required, try to drop it now. */ 1014 if ((page_has_private(pagep[0]) && !try_to_release_page(pagep[0], 0)) || 1015 (page_has_private(pagep[1]) && !try_to_release_page(pagep[1], 0))) { 1016 *err = -EBUSY; 1017 goto unlock_pages; 1018 } 1019 1020 replaced_count = mext_replace_branches(handle, orig_inode, donor_inode, 1021 orig_blk_offset, 1022 block_len_in_page, err); 1023 if (*err) { 1024 if (replaced_count) { 1025 block_len_in_page = replaced_count; 1026 replaced_size = 1027 block_len_in_page << orig_inode->i_blkbits; 1028 } else 1029 goto unlock_pages; 1030 } 1031 /* Perform all necessary steps similar write_begin()/write_end() 1032 * but keeping in mind that i_size will not change */ 1033 *err = __block_write_begin(pagep[0], from, replaced_size, 1034 ext4_get_block); 1035 if (!*err) 1036 *err = block_commit_write(pagep[0], from, from + replaced_size); 1037 1038 if (unlikely(*err < 0)) 1039 goto repair_branches; 1040 1041 /* Even in case of data=writeback it is reasonable to pin 1042 * inode to transaction, to prevent unexpected data loss */ 1043 *err = ext4_jbd2_file_inode(handle, orig_inode); 1044 1045 unlock_pages: 1046 unlock_page(pagep[0]); 1047 page_cache_release(pagep[0]); 1048 unlock_page(pagep[1]); 1049 page_cache_release(pagep[1]); 1050 stop_journal: 1051 ext4_journal_stop(handle); 1052 /* Buffer was busy because probably is pinned to journal transaction, 1053 * force transaction commit may help to free it. */ 1054 if (*err == -EBUSY && ext4_should_retry_alloc(orig_inode->i_sb, 1055 &retries)) 1056 goto again; 1057 return replaced_count; 1058 1059 repair_branches: 1060 /* 1061 * This should never ever happen! 1062 * Extents are swapped already, but we are not able to copy data. 1063 * Try to swap extents to it's original places 1064 */ 1065 ext4_double_down_write_data_sem(orig_inode, donor_inode); 1066 replaced_count = mext_replace_branches(handle, donor_inode, orig_inode, 1067 orig_blk_offset, 1068 block_len_in_page, &err2); 1069 ext4_double_up_write_data_sem(orig_inode, donor_inode); 1070 if (replaced_count != block_len_in_page) { 1071 EXT4_ERROR_INODE_BLOCK(orig_inode, (sector_t)(orig_blk_offset), 1072 "Unable to copy data block," 1073 " data will be lost."); 1074 *err = -EIO; 1075 } 1076 replaced_count = 0; 1077 goto unlock_pages; 1078 } 1079 1080 /** 1081 * mext_check_arguments - Check whether move extent can be done 1082 * 1083 * @orig_inode: original inode 1084 * @donor_inode: donor inode 1085 * @orig_start: logical start offset in block for orig 1086 * @donor_start: logical start offset in block for donor 1087 * @len: the number of blocks to be moved 1088 * 1089 * Check the arguments of ext4_move_extents() whether the files can be 1090 * exchanged with each other. 1091 * Return 0 on success, or a negative error value on failure. 1092 */ 1093 static int 1094 mext_check_arguments(struct inode *orig_inode, 1095 struct inode *donor_inode, __u64 orig_start, 1096 __u64 donor_start, __u64 *len) 1097 { 1098 ext4_lblk_t orig_blocks, donor_blocks; 1099 unsigned int blkbits = orig_inode->i_blkbits; 1100 unsigned int blocksize = 1 << blkbits; 1101 1102 if (donor_inode->i_mode & (S_ISUID|S_ISGID)) { 1103 ext4_debug("ext4 move extent: suid or sgid is set" 1104 " to donor file [ino:orig %lu, donor %lu]\n", 1105 orig_inode->i_ino, donor_inode->i_ino); 1106 return -EINVAL; 1107 } 1108 1109 if (IS_IMMUTABLE(donor_inode) || IS_APPEND(donor_inode)) 1110 return -EPERM; 1111 1112 /* Ext4 move extent does not support swapfile */ 1113 if (IS_SWAPFILE(orig_inode) || IS_SWAPFILE(donor_inode)) { 1114 ext4_debug("ext4 move extent: The argument files should " 1115 "not be swapfile [ino:orig %lu, donor %lu]\n", 1116 orig_inode->i_ino, donor_inode->i_ino); 1117 return -EINVAL; 1118 } 1119 1120 /* Ext4 move extent supports only extent based file */ 1121 if (!(ext4_test_inode_flag(orig_inode, EXT4_INODE_EXTENTS))) { 1122 ext4_debug("ext4 move extent: orig file is not extents " 1123 "based file [ino:orig %lu]\n", orig_inode->i_ino); 1124 return -EOPNOTSUPP; 1125 } else if (!(ext4_test_inode_flag(donor_inode, EXT4_INODE_EXTENTS))) { 1126 ext4_debug("ext4 move extent: donor file is not extents " 1127 "based file [ino:donor %lu]\n", donor_inode->i_ino); 1128 return -EOPNOTSUPP; 1129 } 1130 1131 if ((!orig_inode->i_size) || (!donor_inode->i_size)) { 1132 ext4_debug("ext4 move extent: File size is 0 byte\n"); 1133 return -EINVAL; 1134 } 1135 1136 /* Start offset should be same */ 1137 if (orig_start != donor_start) { 1138 ext4_debug("ext4 move extent: orig and donor's start " 1139 "offset are not same [ino:orig %lu, donor %lu]\n", 1140 orig_inode->i_ino, donor_inode->i_ino); 1141 return -EINVAL; 1142 } 1143 1144 if ((orig_start >= EXT_MAX_BLOCKS) || 1145 (*len > EXT_MAX_BLOCKS) || 1146 (orig_start + *len >= EXT_MAX_BLOCKS)) { 1147 ext4_debug("ext4 move extent: Can't handle over [%u] blocks " 1148 "[ino:orig %lu, donor %lu]\n", EXT_MAX_BLOCKS, 1149 orig_inode->i_ino, donor_inode->i_ino); 1150 return -EINVAL; 1151 } 1152 1153 if (orig_inode->i_size > donor_inode->i_size) { 1154 donor_blocks = (donor_inode->i_size + blocksize - 1) >> blkbits; 1155 /* TODO: eliminate this artificial restriction */ 1156 if (orig_start >= donor_blocks) { 1157 ext4_debug("ext4 move extent: orig start offset " 1158 "[%llu] should be less than donor file blocks " 1159 "[%u] [ino:orig %lu, donor %lu]\n", 1160 orig_start, donor_blocks, 1161 orig_inode->i_ino, donor_inode->i_ino); 1162 return -EINVAL; 1163 } 1164 1165 /* TODO: eliminate this artificial restriction */ 1166 if (orig_start + *len > donor_blocks) { 1167 ext4_debug("ext4 move extent: End offset [%llu] should " 1168 "be less than donor file blocks [%u]." 1169 "So adjust length from %llu to %llu " 1170 "[ino:orig %lu, donor %lu]\n", 1171 orig_start + *len, donor_blocks, 1172 *len, donor_blocks - orig_start, 1173 orig_inode->i_ino, donor_inode->i_ino); 1174 *len = donor_blocks - orig_start; 1175 } 1176 } else { 1177 orig_blocks = (orig_inode->i_size + blocksize - 1) >> blkbits; 1178 if (orig_start >= orig_blocks) { 1179 ext4_debug("ext4 move extent: start offset [%llu] " 1180 "should be less than original file blocks " 1181 "[%u] [ino:orig %lu, donor %lu]\n", 1182 orig_start, orig_blocks, 1183 orig_inode->i_ino, donor_inode->i_ino); 1184 return -EINVAL; 1185 } 1186 1187 if (orig_start + *len > orig_blocks) { 1188 ext4_debug("ext4 move extent: Adjust length " 1189 "from %llu to %llu. Because it should be " 1190 "less than original file blocks " 1191 "[ino:orig %lu, donor %lu]\n", 1192 *len, orig_blocks - orig_start, 1193 orig_inode->i_ino, donor_inode->i_ino); 1194 *len = orig_blocks - orig_start; 1195 } 1196 } 1197 1198 if (!*len) { 1199 ext4_debug("ext4 move extent: len should not be 0 " 1200 "[ino:orig %lu, donor %lu]\n", orig_inode->i_ino, 1201 donor_inode->i_ino); 1202 return -EINVAL; 1203 } 1204 1205 return 0; 1206 } 1207 1208 /** 1209 * ext4_inode_double_lock - Lock i_mutex on both @inode1 and @inode2 1210 * 1211 * @inode1: the inode structure 1212 * @inode2: the inode structure 1213 * 1214 * Lock two inodes' i_mutex 1215 */ 1216 void 1217 ext4_inode_double_lock(struct inode *inode1, struct inode *inode2) 1218 { 1219 BUG_ON(inode1 == inode2); 1220 if (inode1 < inode2) { 1221 mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT); 1222 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD); 1223 } else { 1224 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_PARENT); 1225 mutex_lock_nested(&inode1->i_mutex, I_MUTEX_CHILD); 1226 } 1227 } 1228 1229 /** 1230 * ext4_inode_double_unlock - Release i_mutex on both @inode1 and @inode2 1231 * 1232 * @inode1: the inode that is released first 1233 * @inode2: the inode that is released second 1234 * 1235 */ 1236 1237 void 1238 ext4_inode_double_unlock(struct inode *inode1, struct inode *inode2) 1239 { 1240 mutex_unlock(&inode1->i_mutex); 1241 mutex_unlock(&inode2->i_mutex); 1242 } 1243 1244 /** 1245 * ext4_move_extents - Exchange the specified range of a file 1246 * 1247 * @o_filp: file structure of the original file 1248 * @d_filp: file structure of the donor file 1249 * @orig_start: start offset in block for orig 1250 * @donor_start: start offset in block for donor 1251 * @len: the number of blocks to be moved 1252 * @moved_len: moved block length 1253 * 1254 * This function returns 0 and moved block length is set in moved_len 1255 * if succeed, otherwise returns error value. 1256 * 1257 * Note: ext4_move_extents() proceeds the following order. 1258 * 1:ext4_move_extents() calculates the last block number of moving extent 1259 * function by the start block number (orig_start) and the number of blocks 1260 * to be moved (len) specified as arguments. 1261 * If the {orig, donor}_start points a hole, the extent's start offset 1262 * pointed by ext_cur (current extent), holecheck_path, orig_path are set 1263 * after hole behind. 1264 * 2:Continue step 3 to step 5, until the holecheck_path points to last_extent 1265 * or the ext_cur exceeds the block_end which is last logical block number. 1266 * 3:To get the length of continues area, call mext_next_extent() 1267 * specified with the ext_cur (initial value is holecheck_path) re-cursive, 1268 * until find un-continuous extent, the start logical block number exceeds 1269 * the block_end or the extent points to the last extent. 1270 * 4:Exchange the original inode data with donor inode data 1271 * from orig_page_offset to seq_end_page. 1272 * The start indexes of data are specified as arguments. 1273 * That of the original inode is orig_page_offset, 1274 * and the donor inode is also orig_page_offset 1275 * (To easily handle blocksize != pagesize case, the offset for the 1276 * donor inode is block unit). 1277 * 5:Update holecheck_path and orig_path to points a next proceeding extent, 1278 * then returns to step 2. 1279 * 6:Release holecheck_path, orig_path and set the len to moved_len 1280 * which shows the number of moved blocks. 1281 * The moved_len is useful for the command to calculate the file offset 1282 * for starting next move extent ioctl. 1283 * 7:Return 0 on success, or a negative error value on failure. 1284 */ 1285 int 1286 ext4_move_extents(struct file *o_filp, struct file *d_filp, 1287 __u64 orig_start, __u64 donor_start, __u64 len, 1288 __u64 *moved_len) 1289 { 1290 struct inode *orig_inode = file_inode(o_filp); 1291 struct inode *donor_inode = file_inode(d_filp); 1292 struct ext4_ext_path *orig_path = NULL, *holecheck_path = NULL; 1293 struct ext4_extent *ext_prev, *ext_cur, *ext_dummy; 1294 ext4_lblk_t block_start = orig_start; 1295 ext4_lblk_t block_end, seq_start, add_blocks, file_end, seq_blocks = 0; 1296 ext4_lblk_t rest_blocks; 1297 pgoff_t orig_page_offset = 0, seq_end_page; 1298 int ret, depth, last_extent = 0; 1299 int blocks_per_page = PAGE_CACHE_SIZE >> orig_inode->i_blkbits; 1300 int data_offset_in_page; 1301 int block_len_in_page; 1302 int uninit; 1303 1304 if (orig_inode->i_sb != donor_inode->i_sb) { 1305 ext4_debug("ext4 move extent: The argument files " 1306 "should be in same FS [ino:orig %lu, donor %lu]\n", 1307 orig_inode->i_ino, donor_inode->i_ino); 1308 return -EINVAL; 1309 } 1310 1311 /* orig and donor should be different inodes */ 1312 if (orig_inode == donor_inode) { 1313 ext4_debug("ext4 move extent: The argument files should not " 1314 "be same inode [ino:orig %lu, donor %lu]\n", 1315 orig_inode->i_ino, donor_inode->i_ino); 1316 return -EINVAL; 1317 } 1318 1319 /* Regular file check */ 1320 if (!S_ISREG(orig_inode->i_mode) || !S_ISREG(donor_inode->i_mode)) { 1321 ext4_debug("ext4 move extent: The argument files should be " 1322 "regular file [ino:orig %lu, donor %lu]\n", 1323 orig_inode->i_ino, donor_inode->i_ino); 1324 return -EINVAL; 1325 } 1326 /* TODO: This is non obvious task to swap blocks for inodes with full 1327 jornaling enabled */ 1328 if (ext4_should_journal_data(orig_inode) || 1329 ext4_should_journal_data(donor_inode)) { 1330 return -EINVAL; 1331 } 1332 /* Protect orig and donor inodes against a truncate */ 1333 ext4_inode_double_lock(orig_inode, donor_inode); 1334 1335 /* Wait for all existing dio workers */ 1336 ext4_inode_block_unlocked_dio(orig_inode); 1337 ext4_inode_block_unlocked_dio(donor_inode); 1338 inode_dio_wait(orig_inode); 1339 inode_dio_wait(donor_inode); 1340 1341 /* Protect extent tree against block allocations via delalloc */ 1342 ext4_double_down_write_data_sem(orig_inode, donor_inode); 1343 /* Check the filesystem environment whether move_extent can be done */ 1344 ret = mext_check_arguments(orig_inode, donor_inode, orig_start, 1345 donor_start, &len); 1346 if (ret) 1347 goto out; 1348 1349 file_end = (i_size_read(orig_inode) - 1) >> orig_inode->i_blkbits; 1350 block_end = block_start + len - 1; 1351 if (file_end < block_end) 1352 len -= block_end - file_end; 1353 1354 ret = get_ext_path(orig_inode, block_start, &orig_path); 1355 if (ret) 1356 goto out; 1357 1358 /* Get path structure to check the hole */ 1359 ret = get_ext_path(orig_inode, block_start, &holecheck_path); 1360 if (ret) 1361 goto out; 1362 1363 depth = ext_depth(orig_inode); 1364 ext_cur = holecheck_path[depth].p_ext; 1365 1366 /* 1367 * Get proper starting location of block replacement if block_start was 1368 * within the hole. 1369 */ 1370 if (le32_to_cpu(ext_cur->ee_block) + 1371 ext4_ext_get_actual_len(ext_cur) - 1 < block_start) { 1372 /* 1373 * The hole exists between extents or the tail of 1374 * original file. 1375 */ 1376 last_extent = mext_next_extent(orig_inode, 1377 holecheck_path, &ext_cur); 1378 if (last_extent < 0) { 1379 ret = last_extent; 1380 goto out; 1381 } 1382 last_extent = mext_next_extent(orig_inode, orig_path, 1383 &ext_dummy); 1384 if (last_extent < 0) { 1385 ret = last_extent; 1386 goto out; 1387 } 1388 seq_start = le32_to_cpu(ext_cur->ee_block); 1389 } else if (le32_to_cpu(ext_cur->ee_block) > block_start) 1390 /* The hole exists at the beginning of original file. */ 1391 seq_start = le32_to_cpu(ext_cur->ee_block); 1392 else 1393 seq_start = block_start; 1394 1395 /* No blocks within the specified range. */ 1396 if (le32_to_cpu(ext_cur->ee_block) > block_end) { 1397 ext4_debug("ext4 move extent: The specified range of file " 1398 "may be the hole\n"); 1399 ret = -EINVAL; 1400 goto out; 1401 } 1402 1403 /* Adjust start blocks */ 1404 add_blocks = min(le32_to_cpu(ext_cur->ee_block) + 1405 ext4_ext_get_actual_len(ext_cur), block_end + 1) - 1406 max(le32_to_cpu(ext_cur->ee_block), block_start); 1407 1408 while (!last_extent && le32_to_cpu(ext_cur->ee_block) <= block_end) { 1409 seq_blocks += add_blocks; 1410 1411 /* Adjust tail blocks */ 1412 if (seq_start + seq_blocks - 1 > block_end) 1413 seq_blocks = block_end - seq_start + 1; 1414 1415 ext_prev = ext_cur; 1416 last_extent = mext_next_extent(orig_inode, holecheck_path, 1417 &ext_cur); 1418 if (last_extent < 0) { 1419 ret = last_extent; 1420 break; 1421 } 1422 add_blocks = ext4_ext_get_actual_len(ext_cur); 1423 1424 /* 1425 * Extend the length of contiguous block (seq_blocks) 1426 * if extents are contiguous. 1427 */ 1428 if (ext4_can_extents_be_merged(orig_inode, 1429 ext_prev, ext_cur) && 1430 block_end >= le32_to_cpu(ext_cur->ee_block) && 1431 !last_extent) 1432 continue; 1433 1434 /* Is original extent is uninitialized */ 1435 uninit = ext4_ext_is_uninitialized(ext_prev); 1436 1437 data_offset_in_page = seq_start % blocks_per_page; 1438 1439 /* 1440 * Calculate data blocks count that should be swapped 1441 * at the first page. 1442 */ 1443 if (data_offset_in_page + seq_blocks > blocks_per_page) { 1444 /* Swapped blocks are across pages */ 1445 block_len_in_page = 1446 blocks_per_page - data_offset_in_page; 1447 } else { 1448 /* Swapped blocks are in a page */ 1449 block_len_in_page = seq_blocks; 1450 } 1451 1452 orig_page_offset = seq_start >> 1453 (PAGE_CACHE_SHIFT - orig_inode->i_blkbits); 1454 seq_end_page = (seq_start + seq_blocks - 1) >> 1455 (PAGE_CACHE_SHIFT - orig_inode->i_blkbits); 1456 seq_start = le32_to_cpu(ext_cur->ee_block); 1457 rest_blocks = seq_blocks; 1458 1459 /* 1460 * Up semaphore to avoid following problems: 1461 * a. transaction deadlock among ext4_journal_start, 1462 * ->write_begin via pagefault, and jbd2_journal_commit 1463 * b. racing with ->readpage, ->write_begin, and ext4_get_block 1464 * in move_extent_per_page 1465 */ 1466 ext4_double_up_write_data_sem(orig_inode, donor_inode); 1467 1468 while (orig_page_offset <= seq_end_page) { 1469 1470 /* Swap original branches with new branches */ 1471 block_len_in_page = move_extent_per_page( 1472 o_filp, donor_inode, 1473 orig_page_offset, 1474 data_offset_in_page, 1475 block_len_in_page, uninit, 1476 &ret); 1477 1478 /* Count how many blocks we have exchanged */ 1479 *moved_len += block_len_in_page; 1480 if (ret < 0) 1481 break; 1482 if (*moved_len > len) { 1483 EXT4_ERROR_INODE(orig_inode, 1484 "We replaced blocks too much! " 1485 "sum of replaced: %llu requested: %llu", 1486 *moved_len, len); 1487 ret = -EIO; 1488 break; 1489 } 1490 1491 orig_page_offset++; 1492 data_offset_in_page = 0; 1493 rest_blocks -= block_len_in_page; 1494 if (rest_blocks > blocks_per_page) 1495 block_len_in_page = blocks_per_page; 1496 else 1497 block_len_in_page = rest_blocks; 1498 } 1499 1500 ext4_double_down_write_data_sem(orig_inode, donor_inode); 1501 if (ret < 0) 1502 break; 1503 1504 /* Decrease buffer counter */ 1505 if (holecheck_path) 1506 ext4_ext_drop_refs(holecheck_path); 1507 ret = get_ext_path(orig_inode, seq_start, &holecheck_path); 1508 if (ret) 1509 break; 1510 depth = holecheck_path->p_depth; 1511 1512 /* Decrease buffer counter */ 1513 if (orig_path) 1514 ext4_ext_drop_refs(orig_path); 1515 ret = get_ext_path(orig_inode, seq_start, &orig_path); 1516 if (ret) 1517 break; 1518 1519 ext_cur = holecheck_path[depth].p_ext; 1520 add_blocks = ext4_ext_get_actual_len(ext_cur); 1521 seq_blocks = 0; 1522 1523 } 1524 out: 1525 if (*moved_len) { 1526 ext4_discard_preallocations(orig_inode); 1527 ext4_discard_preallocations(donor_inode); 1528 } 1529 1530 if (orig_path) { 1531 ext4_ext_drop_refs(orig_path); 1532 kfree(orig_path); 1533 } 1534 if (holecheck_path) { 1535 ext4_ext_drop_refs(holecheck_path); 1536 kfree(holecheck_path); 1537 } 1538 ext4_double_up_write_data_sem(orig_inode, donor_inode); 1539 ext4_inode_resume_unlocked_dio(orig_inode); 1540 ext4_inode_resume_unlocked_dio(donor_inode); 1541 ext4_inode_double_unlock(orig_inode, donor_inode); 1542 1543 return ret; 1544 } 1545