1// SPDX-License-Identifier: GPL-2.0 2/* 3 * Copyright (c) 2000-2006 Silicon Graphics, Inc. 4 * Copyright (c) 2012 Red Hat, Inc. 5 * All Rights Reserved. 6 */ 7#include "xfs.h" 8#include "xfs_fs.h" 9#include "xfs_shared.h" 10#include "xfs_format.h" 11#include "xfs_log_format.h" 12#include "xfs_trans_resv.h" 13#include "xfs_bit.h" 14#include "xfs_mount.h" 15#include "xfs_defer.h" 16#include "xfs_inode.h" 17#include "xfs_btree.h" 18#include "xfs_trans.h" 19#include "xfs_alloc.h" 20#include "xfs_bmap.h" 21#include "xfs_bmap_util.h" 22#include "xfs_bmap_btree.h" 23#include "xfs_rtalloc.h" 24#include "xfs_error.h" 25#include "xfs_quota.h" 26#include "xfs_trans_space.h" 27#include "xfs_trace.h" 28#include "xfs_icache.h" 29#include "xfs_iomap.h" 30#include "xfs_reflink.h" 31 32/* Kernel only BMAP related definitions and functions */ 33 34/* 35 * Convert the given file system block to a disk block. We have to treat it 36 * differently based on whether the file is a real time file or not, because the 37 * bmap code does. 38 */ 39xfs_daddr_t 40xfs_fsb_to_db(struct xfs_inode *ip, xfs_fsblock_t fsb) 41{ 42 if (XFS_IS_REALTIME_INODE(ip)) 43 return XFS_FSB_TO_BB(ip->i_mount, fsb); 44 return XFS_FSB_TO_DADDR(ip->i_mount, fsb); 45} 46 47/* 48 * Routine to zero an extent on disk allocated to the specific inode. 49 * 50 * The VFS functions take a linearised filesystem block offset, so we have to 51 * convert the sparse xfs fsb to the right format first. 52 * VFS types are real funky, too. 53 */ 54int 55xfs_zero_extent( 56 struct xfs_inode *ip, 57 xfs_fsblock_t start_fsb, 58 xfs_off_t count_fsb) 59{ 60 struct xfs_mount *mp = ip->i_mount; 61 struct xfs_buftarg *target = xfs_inode_buftarg(ip); 62 xfs_daddr_t sector = xfs_fsb_to_db(ip, start_fsb); 63 sector_t block = XFS_BB_TO_FSBT(mp, sector); 64 65 return blkdev_issue_zeroout(target->bt_bdev, 66 block << (mp->m_super->s_blocksize_bits - 9), 67 count_fsb << (mp->m_super->s_blocksize_bits - 9), 68 GFP_NOFS, 0); 69} 70 71#ifdef CONFIG_XFS_RT 72int 73xfs_bmap_rtalloc( 74 struct xfs_bmalloca *ap) 75{ 76 struct xfs_mount *mp = ap->ip->i_mount; 77 xfs_fileoff_t orig_offset = ap->offset; 78 xfs_rtblock_t rtb; 79 xfs_extlen_t prod = 0; /* product factor for allocators */ 80 xfs_extlen_t mod = 0; /* product factor for allocators */ 81 xfs_extlen_t ralen = 0; /* realtime allocation length */ 82 xfs_extlen_t align; /* minimum allocation alignment */ 83 xfs_extlen_t orig_length = ap->length; 84 xfs_extlen_t minlen = mp->m_sb.sb_rextsize; 85 xfs_extlen_t raminlen; 86 bool rtlocked = false; 87 bool ignore_locality = false; 88 int error; 89 90 align = xfs_get_extsz_hint(ap->ip); 91retry: 92 prod = align / mp->m_sb.sb_rextsize; 93 error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev, 94 align, 1, ap->eof, 0, 95 ap->conv, &ap->offset, &ap->length); 96 if (error) 97 return error; 98 ASSERT(ap->length); 99 ASSERT(ap->length % mp->m_sb.sb_rextsize == 0); 100 101 /* 102 * If we shifted the file offset downward to satisfy an extent size 103 * hint, increase minlen by that amount so that the allocator won't 104 * give us an allocation that's too short to cover at least one of the 105 * blocks that the caller asked for. 106 */ 107 if (ap->offset != orig_offset) 108 minlen += orig_offset - ap->offset; 109 110 /* 111 * If the offset & length are not perfectly aligned 112 * then kill prod, it will just get us in trouble. 113 */ 114 div_u64_rem(ap->offset, align, &mod); 115 if (mod || ap->length % align) 116 prod = 1; 117 /* 118 * Set ralen to be the actual requested length in rtextents. 119 */ 120 ralen = ap->length / mp->m_sb.sb_rextsize; 121 /* 122 * If the old value was close enough to XFS_BMBT_MAX_EXTLEN that 123 * we rounded up to it, cut it back so it's valid again. 124 * Note that if it's a really large request (bigger than 125 * XFS_BMBT_MAX_EXTLEN), we don't hear about that number, and can't 126 * adjust the starting point to match it. 127 */ 128 if (ralen * mp->m_sb.sb_rextsize >= XFS_MAX_BMBT_EXTLEN) 129 ralen = XFS_MAX_BMBT_EXTLEN / mp->m_sb.sb_rextsize; 130 131 /* 132 * Lock out modifications to both the RT bitmap and summary inodes 133 */ 134 if (!rtlocked) { 135 xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL|XFS_ILOCK_RTBITMAP); 136 xfs_trans_ijoin(ap->tp, mp->m_rbmip, XFS_ILOCK_EXCL); 137 xfs_ilock(mp->m_rsumip, XFS_ILOCK_EXCL|XFS_ILOCK_RTSUM); 138 xfs_trans_ijoin(ap->tp, mp->m_rsumip, XFS_ILOCK_EXCL); 139 rtlocked = true; 140 } 141 142 /* 143 * If it's an allocation to an empty file at offset 0, 144 * pick an extent that will space things out in the rt area. 145 */ 146 if (ap->eof && ap->offset == 0) { 147 xfs_rtblock_t rtx; /* realtime extent no */ 148 149 error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx); 150 if (error) 151 return error; 152 ap->blkno = rtx * mp->m_sb.sb_rextsize; 153 } else { 154 ap->blkno = 0; 155 } 156 157 xfs_bmap_adjacent(ap); 158 159 /* 160 * Realtime allocation, done through xfs_rtallocate_extent. 161 */ 162 if (ignore_locality) 163 ap->blkno = 0; 164 else 165 do_div(ap->blkno, mp->m_sb.sb_rextsize); 166 rtb = ap->blkno; 167 ap->length = ralen; 168 raminlen = max_t(xfs_extlen_t, 1, minlen / mp->m_sb.sb_rextsize); 169 error = xfs_rtallocate_extent(ap->tp, ap->blkno, raminlen, ap->length, 170 &ralen, ap->wasdel, prod, &rtb); 171 if (error) 172 return error; 173 174 if (rtb != NULLRTBLOCK) { 175 ap->blkno = rtb * mp->m_sb.sb_rextsize; 176 ap->length = ralen * mp->m_sb.sb_rextsize; 177 ap->ip->i_nblocks += ap->length; 178 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE); 179 if (ap->wasdel) 180 ap->ip->i_delayed_blks -= ap->length; 181 /* 182 * Adjust the disk quota also. This was reserved 183 * earlier. 184 */ 185 xfs_trans_mod_dquot_byino(ap->tp, ap->ip, 186 ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT : 187 XFS_TRANS_DQ_RTBCOUNT, ap->length); 188 return 0; 189 } 190 191 if (align > mp->m_sb.sb_rextsize) { 192 /* 193 * We previously enlarged the request length to try to satisfy 194 * an extent size hint. The allocator didn't return anything, 195 * so reset the parameters to the original values and try again 196 * without alignment criteria. 197 */ 198 ap->offset = orig_offset; 199 ap->length = orig_length; 200 minlen = align = mp->m_sb.sb_rextsize; 201 goto retry; 202 } 203 204 if (!ignore_locality && ap->blkno != 0) { 205 /* 206 * If we can't allocate near a specific rt extent, try again 207 * without locality criteria. 208 */ 209 ignore_locality = true; 210 goto retry; 211 } 212 213 ap->blkno = NULLFSBLOCK; 214 ap->length = 0; 215 return 0; 216} 217#endif /* CONFIG_XFS_RT */ 218 219/* 220 * Extent tree block counting routines. 221 */ 222 223/* 224 * Count leaf blocks given a range of extent records. Delayed allocation 225 * extents are not counted towards the totals. 226 */ 227xfs_extnum_t 228xfs_bmap_count_leaves( 229 struct xfs_ifork *ifp, 230 xfs_filblks_t *count) 231{ 232 struct xfs_iext_cursor icur; 233 struct xfs_bmbt_irec got; 234 xfs_extnum_t numrecs = 0; 235 236 for_each_xfs_iext(ifp, &icur, &got) { 237 if (!isnullstartblock(got.br_startblock)) { 238 *count += got.br_blockcount; 239 numrecs++; 240 } 241 } 242 243 return numrecs; 244} 245 246/* 247 * Count fsblocks of the given fork. Delayed allocation extents are 248 * not counted towards the totals. 249 */ 250int 251xfs_bmap_count_blocks( 252 struct xfs_trans *tp, 253 struct xfs_inode *ip, 254 int whichfork, 255 xfs_extnum_t *nextents, 256 xfs_filblks_t *count) 257{ 258 struct xfs_mount *mp = ip->i_mount; 259 struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork); 260 struct xfs_btree_cur *cur; 261 xfs_extlen_t btblocks = 0; 262 int error; 263 264 *nextents = 0; 265 *count = 0; 266 267 if (!ifp) 268 return 0; 269 270 switch (ifp->if_format) { 271 case XFS_DINODE_FMT_BTREE: 272 error = xfs_iread_extents(tp, ip, whichfork); 273 if (error) 274 return error; 275 276 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork); 277 error = xfs_btree_count_blocks(cur, &btblocks); 278 xfs_btree_del_cursor(cur, error); 279 if (error) 280 return error; 281 282 /* 283 * xfs_btree_count_blocks includes the root block contained in 284 * the inode fork in @btblocks, so subtract one because we're 285 * only interested in allocated disk blocks. 286 */ 287 *count += btblocks - 1; 288 289 fallthrough; 290 case XFS_DINODE_FMT_EXTENTS: 291 *nextents = xfs_bmap_count_leaves(ifp, count); 292 break; 293 } 294 295 return 0; 296} 297 298static int 299xfs_getbmap_report_one( 300 struct xfs_inode *ip, 301 struct getbmapx *bmv, 302 struct kgetbmap *out, 303 int64_t bmv_end, 304 struct xfs_bmbt_irec *got) 305{ 306 struct kgetbmap *p = out + bmv->bmv_entries; 307 bool shared = false; 308 int error; 309 310 error = xfs_reflink_trim_around_shared(ip, got, &shared); 311 if (error) 312 return error; 313 314 if (isnullstartblock(got->br_startblock) || 315 got->br_startblock == DELAYSTARTBLOCK) { 316 /* 317 * Take the flush completion as being a point-in-time snapshot 318 * where there are no delalloc extents, and if any new ones 319 * have been created racily, just skip them as being 'after' 320 * the flush and so don't get reported. 321 */ 322 if (!(bmv->bmv_iflags & BMV_IF_DELALLOC)) 323 return 0; 324 325 p->bmv_oflags |= BMV_OF_DELALLOC; 326 p->bmv_block = -2; 327 } else { 328 p->bmv_block = xfs_fsb_to_db(ip, got->br_startblock); 329 } 330 331 if (got->br_state == XFS_EXT_UNWRITTEN && 332 (bmv->bmv_iflags & BMV_IF_PREALLOC)) 333 p->bmv_oflags |= BMV_OF_PREALLOC; 334 335 if (shared) 336 p->bmv_oflags |= BMV_OF_SHARED; 337 338 p->bmv_offset = XFS_FSB_TO_BB(ip->i_mount, got->br_startoff); 339 p->bmv_length = XFS_FSB_TO_BB(ip->i_mount, got->br_blockcount); 340 341 bmv->bmv_offset = p->bmv_offset + p->bmv_length; 342 bmv->bmv_length = max(0LL, bmv_end - bmv->bmv_offset); 343 bmv->bmv_entries++; 344 return 0; 345} 346 347static void 348xfs_getbmap_report_hole( 349 struct xfs_inode *ip, 350 struct getbmapx *bmv, 351 struct kgetbmap *out, 352 int64_t bmv_end, 353 xfs_fileoff_t bno, 354 xfs_fileoff_t end) 355{ 356 struct kgetbmap *p = out + bmv->bmv_entries; 357 358 if (bmv->bmv_iflags & BMV_IF_NO_HOLES) 359 return; 360 361 p->bmv_block = -1; 362 p->bmv_offset = XFS_FSB_TO_BB(ip->i_mount, bno); 363 p->bmv_length = XFS_FSB_TO_BB(ip->i_mount, end - bno); 364 365 bmv->bmv_offset = p->bmv_offset + p->bmv_length; 366 bmv->bmv_length = max(0LL, bmv_end - bmv->bmv_offset); 367 bmv->bmv_entries++; 368} 369 370static inline bool 371xfs_getbmap_full( 372 struct getbmapx *bmv) 373{ 374 return bmv->bmv_length == 0 || bmv->bmv_entries >= bmv->bmv_count - 1; 375} 376 377static bool 378xfs_getbmap_next_rec( 379 struct xfs_bmbt_irec *rec, 380 xfs_fileoff_t total_end) 381{ 382 xfs_fileoff_t end = rec->br_startoff + rec->br_blockcount; 383 384 if (end == total_end) 385 return false; 386 387 rec->br_startoff += rec->br_blockcount; 388 if (!isnullstartblock(rec->br_startblock) && 389 rec->br_startblock != DELAYSTARTBLOCK) 390 rec->br_startblock += rec->br_blockcount; 391 rec->br_blockcount = total_end - end; 392 return true; 393} 394 395/* 396 * Get inode's extents as described in bmv, and format for output. 397 * Calls formatter to fill the user's buffer until all extents 398 * are mapped, until the passed-in bmv->bmv_count slots have 399 * been filled, or until the formatter short-circuits the loop, 400 * if it is tracking filled-in extents on its own. 401 */ 402int /* error code */ 403xfs_getbmap( 404 struct xfs_inode *ip, 405 struct getbmapx *bmv, /* user bmap structure */ 406 struct kgetbmap *out) 407{ 408 struct xfs_mount *mp = ip->i_mount; 409 int iflags = bmv->bmv_iflags; 410 int whichfork, lock, error = 0; 411 int64_t bmv_end, max_len; 412 xfs_fileoff_t bno, first_bno; 413 struct xfs_ifork *ifp; 414 struct xfs_bmbt_irec got, rec; 415 xfs_filblks_t len; 416 struct xfs_iext_cursor icur; 417 418 if (bmv->bmv_iflags & ~BMV_IF_VALID) 419 return -EINVAL; 420#ifndef DEBUG 421 /* Only allow CoW fork queries if we're debugging. */ 422 if (iflags & BMV_IF_COWFORK) 423 return -EINVAL; 424#endif 425 if ((iflags & BMV_IF_ATTRFORK) && (iflags & BMV_IF_COWFORK)) 426 return -EINVAL; 427 428 if (bmv->bmv_length < -1) 429 return -EINVAL; 430 bmv->bmv_entries = 0; 431 if (bmv->bmv_length == 0) 432 return 0; 433 434 if (iflags & BMV_IF_ATTRFORK) 435 whichfork = XFS_ATTR_FORK; 436 else if (iflags & BMV_IF_COWFORK) 437 whichfork = XFS_COW_FORK; 438 else 439 whichfork = XFS_DATA_FORK; 440 441 xfs_ilock(ip, XFS_IOLOCK_SHARED); 442 switch (whichfork) { 443 case XFS_ATTR_FORK: 444 lock = xfs_ilock_attr_map_shared(ip); 445 if (!xfs_inode_has_attr_fork(ip)) 446 goto out_unlock_ilock; 447 448 max_len = 1LL << 32; 449 break; 450 case XFS_COW_FORK: 451 lock = XFS_ILOCK_SHARED; 452 xfs_ilock(ip, lock); 453 454 /* No CoW fork? Just return */ 455 if (!xfs_ifork_ptr(ip, whichfork)) 456 goto out_unlock_ilock; 457 458 if (xfs_get_cowextsz_hint(ip)) 459 max_len = mp->m_super->s_maxbytes; 460 else 461 max_len = XFS_ISIZE(ip); 462 break; 463 case XFS_DATA_FORK: 464 if (!(iflags & BMV_IF_DELALLOC) && 465 (ip->i_delayed_blks || XFS_ISIZE(ip) > ip->i_disk_size)) { 466 error = filemap_write_and_wait(VFS_I(ip)->i_mapping); 467 if (error) 468 goto out_unlock_iolock; 469 470 /* 471 * Even after flushing the inode, there can still be 472 * delalloc blocks on the inode beyond EOF due to 473 * speculative preallocation. These are not removed 474 * until the release function is called or the inode 475 * is inactivated. Hence we cannot assert here that 476 * ip->i_delayed_blks == 0. 477 */ 478 } 479 480 if (xfs_get_extsz_hint(ip) || 481 (ip->i_diflags & 482 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND))) 483 max_len = mp->m_super->s_maxbytes; 484 else 485 max_len = XFS_ISIZE(ip); 486 487 lock = xfs_ilock_data_map_shared(ip); 488 break; 489 } 490 491 ifp = xfs_ifork_ptr(ip, whichfork); 492 493 switch (ifp->if_format) { 494 case XFS_DINODE_FMT_EXTENTS: 495 case XFS_DINODE_FMT_BTREE: 496 break; 497 case XFS_DINODE_FMT_LOCAL: 498 /* Local format inode forks report no extents. */ 499 goto out_unlock_ilock; 500 default: 501 error = -EINVAL; 502 goto out_unlock_ilock; 503 } 504 505 if (bmv->bmv_length == -1) { 506 max_len = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, max_len)); 507 bmv->bmv_length = max(0LL, max_len - bmv->bmv_offset); 508 } 509 510 bmv_end = bmv->bmv_offset + bmv->bmv_length; 511 512 first_bno = bno = XFS_BB_TO_FSBT(mp, bmv->bmv_offset); 513 len = XFS_BB_TO_FSB(mp, bmv->bmv_length); 514 515 error = xfs_iread_extents(NULL, ip, whichfork); 516 if (error) 517 goto out_unlock_ilock; 518 519 if (!xfs_iext_lookup_extent(ip, ifp, bno, &icur, &got)) { 520 /* 521 * Report a whole-file hole if the delalloc flag is set to 522 * stay compatible with the old implementation. 523 */ 524 if (iflags & BMV_IF_DELALLOC) 525 xfs_getbmap_report_hole(ip, bmv, out, bmv_end, bno, 526 XFS_B_TO_FSB(mp, XFS_ISIZE(ip))); 527 goto out_unlock_ilock; 528 } 529 530 while (!xfs_getbmap_full(bmv)) { 531 xfs_trim_extent(&got, first_bno, len); 532 533 /* 534 * Report an entry for a hole if this extent doesn't directly 535 * follow the previous one. 536 */ 537 if (got.br_startoff > bno) { 538 xfs_getbmap_report_hole(ip, bmv, out, bmv_end, bno, 539 got.br_startoff); 540 if (xfs_getbmap_full(bmv)) 541 break; 542 } 543 544 /* 545 * In order to report shared extents accurately, we report each 546 * distinct shared / unshared part of a single bmbt record with 547 * an individual getbmapx record. 548 */ 549 bno = got.br_startoff + got.br_blockcount; 550 rec = got; 551 do { 552 error = xfs_getbmap_report_one(ip, bmv, out, bmv_end, 553 &rec); 554 if (error || xfs_getbmap_full(bmv)) 555 goto out_unlock_ilock; 556 } while (xfs_getbmap_next_rec(&rec, bno)); 557 558 if (!xfs_iext_next_extent(ifp, &icur, &got)) { 559 xfs_fileoff_t end = XFS_B_TO_FSB(mp, XFS_ISIZE(ip)); 560 561 if (bmv->bmv_entries > 0) 562 out[bmv->bmv_entries - 1].bmv_oflags |= 563 BMV_OF_LAST; 564 565 if (whichfork != XFS_ATTR_FORK && bno < end && 566 !xfs_getbmap_full(bmv)) { 567 xfs_getbmap_report_hole(ip, bmv, out, bmv_end, 568 bno, end); 569 } 570 break; 571 } 572 573 if (bno >= first_bno + len) 574 break; 575 } 576 577out_unlock_ilock: 578 xfs_iunlock(ip, lock); 579out_unlock_iolock: 580 xfs_iunlock(ip, XFS_IOLOCK_SHARED); 581 return error; 582} 583 584/* 585 * Dead simple method of punching delalyed allocation blocks from a range in 586 * the inode. This will always punch out both the start and end blocks, even 587 * if the ranges only partially overlap them, so it is up to the caller to 588 * ensure that partial blocks are not passed in. 589 */ 590int 591xfs_bmap_punch_delalloc_range( 592 struct xfs_inode *ip, 593 xfs_off_t start_byte, 594 xfs_off_t end_byte) 595{ 596 struct xfs_mount *mp = ip->i_mount; 597 struct xfs_ifork *ifp = &ip->i_df; 598 xfs_fileoff_t start_fsb = XFS_B_TO_FSBT(mp, start_byte); 599 xfs_fileoff_t end_fsb = XFS_B_TO_FSB(mp, end_byte); 600 struct xfs_bmbt_irec got, del; 601 struct xfs_iext_cursor icur; 602 int error = 0; 603 604 ASSERT(!xfs_need_iread_extents(ifp)); 605 606 xfs_ilock(ip, XFS_ILOCK_EXCL); 607 if (!xfs_iext_lookup_extent_before(ip, ifp, &end_fsb, &icur, &got)) 608 goto out_unlock; 609 610 while (got.br_startoff + got.br_blockcount > start_fsb) { 611 del = got; 612 xfs_trim_extent(&del, start_fsb, end_fsb - start_fsb); 613 614 /* 615 * A delete can push the cursor forward. Step back to the 616 * previous extent on non-delalloc or extents outside the 617 * target range. 618 */ 619 if (!del.br_blockcount || 620 !isnullstartblock(del.br_startblock)) { 621 if (!xfs_iext_prev_extent(ifp, &icur, &got)) 622 break; 623 continue; 624 } 625 626 error = xfs_bmap_del_extent_delay(ip, XFS_DATA_FORK, &icur, 627 &got, &del); 628 if (error || !xfs_iext_get_extent(ifp, &icur, &got)) 629 break; 630 } 631 632out_unlock: 633 xfs_iunlock(ip, XFS_ILOCK_EXCL); 634 return error; 635} 636 637/* 638 * Test whether it is appropriate to check an inode for and free post EOF 639 * blocks. The 'force' parameter determines whether we should also consider 640 * regular files that are marked preallocated or append-only. 641 */ 642bool 643xfs_can_free_eofblocks( 644 struct xfs_inode *ip, 645 bool force) 646{ 647 struct xfs_bmbt_irec imap; 648 struct xfs_mount *mp = ip->i_mount; 649 xfs_fileoff_t end_fsb; 650 xfs_fileoff_t last_fsb; 651 int nimaps = 1; 652 int error; 653 654 /* 655 * Caller must either hold the exclusive io lock; or be inactivating 656 * the inode, which guarantees there are no other users of the inode. 657 */ 658 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL) || 659 (VFS_I(ip)->i_state & I_FREEING)); 660 661 /* prealloc/delalloc exists only on regular files */ 662 if (!S_ISREG(VFS_I(ip)->i_mode)) 663 return false; 664 665 /* 666 * Zero sized files with no cached pages and delalloc blocks will not 667 * have speculative prealloc/delalloc blocks to remove. 668 */ 669 if (VFS_I(ip)->i_size == 0 && 670 VFS_I(ip)->i_mapping->nrpages == 0 && 671 ip->i_delayed_blks == 0) 672 return false; 673 674 /* If we haven't read in the extent list, then don't do it now. */ 675 if (xfs_need_iread_extents(&ip->i_df)) 676 return false; 677 678 /* 679 * Do not free real preallocated or append-only files unless the file 680 * has delalloc blocks and we are forced to remove them. 681 */ 682 if (ip->i_diflags & (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)) 683 if (!force || ip->i_delayed_blks == 0) 684 return false; 685 686 /* 687 * Do not try to free post-EOF blocks if EOF is beyond the end of the 688 * range supported by the page cache, because the truncation will loop 689 * forever. 690 */ 691 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_ISIZE(ip)); 692 if (XFS_IS_REALTIME_INODE(ip) && mp->m_sb.sb_rextsize > 1) 693 end_fsb = roundup_64(end_fsb, mp->m_sb.sb_rextsize); 694 last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes); 695 if (last_fsb <= end_fsb) 696 return false; 697 698 /* 699 * Look up the mapping for the first block past EOF. If we can't find 700 * it, there's nothing to free. 701 */ 702 xfs_ilock(ip, XFS_ILOCK_SHARED); 703 error = xfs_bmapi_read(ip, end_fsb, last_fsb - end_fsb, &imap, &nimaps, 704 0); 705 xfs_iunlock(ip, XFS_ILOCK_SHARED); 706 if (error || nimaps == 0) 707 return false; 708 709 /* 710 * If there's a real mapping there or there are delayed allocation 711 * reservations, then we have post-EOF blocks to try to free. 712 */ 713 return imap.br_startblock != HOLESTARTBLOCK || ip->i_delayed_blks; 714} 715 716/* 717 * This is called to free any blocks beyond eof. The caller must hold 718 * IOLOCK_EXCL unless we are in the inode reclaim path and have the only 719 * reference to the inode. 720 */ 721int 722xfs_free_eofblocks( 723 struct xfs_inode *ip) 724{ 725 struct xfs_trans *tp; 726 struct xfs_mount *mp = ip->i_mount; 727 int error; 728 729 /* Attach the dquots to the inode up front. */ 730 error = xfs_qm_dqattach(ip); 731 if (error) 732 return error; 733 734 /* Wait on dio to ensure i_size has settled. */ 735 inode_dio_wait(VFS_I(ip)); 736 737 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp); 738 if (error) { 739 ASSERT(xfs_is_shutdown(mp)); 740 return error; 741 } 742 743 xfs_ilock(ip, XFS_ILOCK_EXCL); 744 xfs_trans_ijoin(tp, ip, 0); 745 746 /* 747 * Do not update the on-disk file size. If we update the on-disk file 748 * size and then the system crashes before the contents of the file are 749 * flushed to disk then the files may be full of holes (ie NULL files 750 * bug). 751 */ 752 error = xfs_itruncate_extents_flags(&tp, ip, XFS_DATA_FORK, 753 XFS_ISIZE(ip), XFS_BMAPI_NODISCARD); 754 if (error) 755 goto err_cancel; 756 757 error = xfs_trans_commit(tp); 758 if (error) 759 goto out_unlock; 760 761 xfs_inode_clear_eofblocks_tag(ip); 762 goto out_unlock; 763 764err_cancel: 765 /* 766 * If we get an error at this point we simply don't 767 * bother truncating the file. 768 */ 769 xfs_trans_cancel(tp); 770out_unlock: 771 xfs_iunlock(ip, XFS_ILOCK_EXCL); 772 return error; 773} 774 775int 776xfs_alloc_file_space( 777 struct xfs_inode *ip, 778 xfs_off_t offset, 779 xfs_off_t len) 780{ 781 xfs_mount_t *mp = ip->i_mount; 782 xfs_off_t count; 783 xfs_filblks_t allocatesize_fsb; 784 xfs_extlen_t extsz, temp; 785 xfs_fileoff_t startoffset_fsb; 786 xfs_fileoff_t endoffset_fsb; 787 int rt; 788 xfs_trans_t *tp; 789 xfs_bmbt_irec_t imaps[1], *imapp; 790 int error; 791 792 trace_xfs_alloc_file_space(ip); 793 794 if (xfs_is_shutdown(mp)) 795 return -EIO; 796 797 error = xfs_qm_dqattach(ip); 798 if (error) 799 return error; 800 801 if (len <= 0) 802 return -EINVAL; 803 804 rt = XFS_IS_REALTIME_INODE(ip); 805 extsz = xfs_get_extsz_hint(ip); 806 807 count = len; 808 imapp = &imaps[0]; 809 startoffset_fsb = XFS_B_TO_FSBT(mp, offset); 810 endoffset_fsb = XFS_B_TO_FSB(mp, offset + count); 811 allocatesize_fsb = endoffset_fsb - startoffset_fsb; 812 813 /* 814 * Allocate file space until done or until there is an error 815 */ 816 while (allocatesize_fsb && !error) { 817 xfs_fileoff_t s, e; 818 unsigned int dblocks, rblocks, resblks; 819 int nimaps = 1; 820 821 /* 822 * Determine space reservations for data/realtime. 823 */ 824 if (unlikely(extsz)) { 825 s = startoffset_fsb; 826 do_div(s, extsz); 827 s *= extsz; 828 e = startoffset_fsb + allocatesize_fsb; 829 div_u64_rem(startoffset_fsb, extsz, &temp); 830 if (temp) 831 e += temp; 832 div_u64_rem(e, extsz, &temp); 833 if (temp) 834 e += extsz - temp; 835 } else { 836 s = 0; 837 e = allocatesize_fsb; 838 } 839 840 /* 841 * The transaction reservation is limited to a 32-bit block 842 * count, hence we need to limit the number of blocks we are 843 * trying to reserve to avoid an overflow. We can't allocate 844 * more than @nimaps extents, and an extent is limited on disk 845 * to XFS_BMBT_MAX_EXTLEN (21 bits), so use that to enforce the 846 * limit. 847 */ 848 resblks = min_t(xfs_fileoff_t, (e - s), 849 (XFS_MAX_BMBT_EXTLEN * nimaps)); 850 if (unlikely(rt)) { 851 dblocks = XFS_DIOSTRAT_SPACE_RES(mp, 0); 852 rblocks = resblks; 853 } else { 854 dblocks = XFS_DIOSTRAT_SPACE_RES(mp, resblks); 855 rblocks = 0; 856 } 857 858 error = xfs_trans_alloc_inode(ip, &M_RES(mp)->tr_write, 859 dblocks, rblocks, false, &tp); 860 if (error) 861 break; 862 863 error = xfs_iext_count_may_overflow(ip, XFS_DATA_FORK, 864 XFS_IEXT_ADD_NOSPLIT_CNT); 865 if (error == -EFBIG) 866 error = xfs_iext_count_upgrade(tp, ip, 867 XFS_IEXT_ADD_NOSPLIT_CNT); 868 if (error) 869 goto error; 870 871 error = xfs_bmapi_write(tp, ip, startoffset_fsb, 872 allocatesize_fsb, XFS_BMAPI_PREALLOC, 0, imapp, 873 &nimaps); 874 if (error) 875 goto error; 876 877 ip->i_diflags |= XFS_DIFLAG_PREALLOC; 878 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 879 880 error = xfs_trans_commit(tp); 881 xfs_iunlock(ip, XFS_ILOCK_EXCL); 882 if (error) 883 break; 884 885 /* 886 * If the allocator cannot find a single free extent large 887 * enough to cover the start block of the requested range, 888 * xfs_bmapi_write will return 0 but leave *nimaps set to 0. 889 * 890 * In that case we simply need to keep looping with the same 891 * startoffset_fsb so that one of the following allocations 892 * will eventually reach the requested range. 893 */ 894 if (nimaps) { 895 startoffset_fsb += imapp->br_blockcount; 896 allocatesize_fsb -= imapp->br_blockcount; 897 } 898 } 899 900 return error; 901 902error: 903 xfs_trans_cancel(tp); 904 xfs_iunlock(ip, XFS_ILOCK_EXCL); 905 return error; 906} 907 908static int 909xfs_unmap_extent( 910 struct xfs_inode *ip, 911 xfs_fileoff_t startoffset_fsb, 912 xfs_filblks_t len_fsb, 913 int *done) 914{ 915 struct xfs_mount *mp = ip->i_mount; 916 struct xfs_trans *tp; 917 uint resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0); 918 int error; 919 920 error = xfs_trans_alloc_inode(ip, &M_RES(mp)->tr_write, resblks, 0, 921 false, &tp); 922 if (error) 923 return error; 924 925 error = xfs_iext_count_may_overflow(ip, XFS_DATA_FORK, 926 XFS_IEXT_PUNCH_HOLE_CNT); 927 if (error == -EFBIG) 928 error = xfs_iext_count_upgrade(tp, ip, XFS_IEXT_PUNCH_HOLE_CNT); 929 if (error) 930 goto out_trans_cancel; 931 932 error = xfs_bunmapi(tp, ip, startoffset_fsb, len_fsb, 0, 2, done); 933 if (error) 934 goto out_trans_cancel; 935 936 error = xfs_trans_commit(tp); 937out_unlock: 938 xfs_iunlock(ip, XFS_ILOCK_EXCL); 939 return error; 940 941out_trans_cancel: 942 xfs_trans_cancel(tp); 943 goto out_unlock; 944} 945 946/* Caller must first wait for the completion of any pending DIOs if required. */ 947int 948xfs_flush_unmap_range( 949 struct xfs_inode *ip, 950 xfs_off_t offset, 951 xfs_off_t len) 952{ 953 struct xfs_mount *mp = ip->i_mount; 954 struct inode *inode = VFS_I(ip); 955 xfs_off_t rounding, start, end; 956 int error; 957 958 rounding = max_t(xfs_off_t, mp->m_sb.sb_blocksize, PAGE_SIZE); 959 start = round_down(offset, rounding); 960 end = round_up(offset + len, rounding) - 1; 961 962 error = filemap_write_and_wait_range(inode->i_mapping, start, end); 963 if (error) 964 return error; 965 truncate_pagecache_range(inode, start, end); 966 return 0; 967} 968 969int 970xfs_free_file_space( 971 struct xfs_inode *ip, 972 xfs_off_t offset, 973 xfs_off_t len) 974{ 975 struct xfs_mount *mp = ip->i_mount; 976 xfs_fileoff_t startoffset_fsb; 977 xfs_fileoff_t endoffset_fsb; 978 int done = 0, error; 979 980 trace_xfs_free_file_space(ip); 981 982 error = xfs_qm_dqattach(ip); 983 if (error) 984 return error; 985 986 if (len <= 0) /* if nothing being freed */ 987 return 0; 988 989 startoffset_fsb = XFS_B_TO_FSB(mp, offset); 990 endoffset_fsb = XFS_B_TO_FSBT(mp, offset + len); 991 992 /* We can only free complete realtime extents. */ 993 if (XFS_IS_REALTIME_INODE(ip) && mp->m_sb.sb_rextsize > 1) { 994 startoffset_fsb = roundup_64(startoffset_fsb, 995 mp->m_sb.sb_rextsize); 996 endoffset_fsb = rounddown_64(endoffset_fsb, 997 mp->m_sb.sb_rextsize); 998 } 999 1000 /* 1001 * Need to zero the stuff we're not freeing, on disk. 1002 */ 1003 if (endoffset_fsb > startoffset_fsb) { 1004 while (!done) { 1005 error = xfs_unmap_extent(ip, startoffset_fsb, 1006 endoffset_fsb - startoffset_fsb, &done); 1007 if (error) 1008 return error; 1009 } 1010 } 1011 1012 /* 1013 * Now that we've unmap all full blocks we'll have to zero out any 1014 * partial block at the beginning and/or end. xfs_zero_range is smart 1015 * enough to skip any holes, including those we just created, but we 1016 * must take care not to zero beyond EOF and enlarge i_size. 1017 */ 1018 if (offset >= XFS_ISIZE(ip)) 1019 return 0; 1020 if (offset + len > XFS_ISIZE(ip)) 1021 len = XFS_ISIZE(ip) - offset; 1022 error = xfs_zero_range(ip, offset, len, NULL); 1023 if (error) 1024 return error; 1025 1026 /* 1027 * If we zeroed right up to EOF and EOF straddles a page boundary we 1028 * must make sure that the post-EOF area is also zeroed because the 1029 * page could be mmap'd and xfs_zero_range doesn't do that for us. 1030 * Writeback of the eof page will do this, albeit clumsily. 1031 */ 1032 if (offset + len >= XFS_ISIZE(ip) && offset_in_page(offset + len) > 0) { 1033 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping, 1034 round_down(offset + len, PAGE_SIZE), LLONG_MAX); 1035 } 1036 1037 return error; 1038} 1039 1040static int 1041xfs_prepare_shift( 1042 struct xfs_inode *ip, 1043 loff_t offset) 1044{ 1045 struct xfs_mount *mp = ip->i_mount; 1046 int error; 1047 1048 /* 1049 * Trim eofblocks to avoid shifting uninitialized post-eof preallocation 1050 * into the accessible region of the file. 1051 */ 1052 if (xfs_can_free_eofblocks(ip, true)) { 1053 error = xfs_free_eofblocks(ip); 1054 if (error) 1055 return error; 1056 } 1057 1058 /* 1059 * Shift operations must stabilize the start block offset boundary along 1060 * with the full range of the operation. If we don't, a COW writeback 1061 * completion could race with an insert, front merge with the start 1062 * extent (after split) during the shift and corrupt the file. Start 1063 * with the block just prior to the start to stabilize the boundary. 1064 */ 1065 offset = round_down(offset, mp->m_sb.sb_blocksize); 1066 if (offset) 1067 offset -= mp->m_sb.sb_blocksize; 1068 1069 /* 1070 * Writeback and invalidate cache for the remainder of the file as we're 1071 * about to shift down every extent from offset to EOF. 1072 */ 1073 error = xfs_flush_unmap_range(ip, offset, XFS_ISIZE(ip)); 1074 if (error) 1075 return error; 1076 1077 /* 1078 * Clean out anything hanging around in the cow fork now that 1079 * we've flushed all the dirty data out to disk to avoid having 1080 * CoW extents at the wrong offsets. 1081 */ 1082 if (xfs_inode_has_cow_data(ip)) { 1083 error = xfs_reflink_cancel_cow_range(ip, offset, NULLFILEOFF, 1084 true); 1085 if (error) 1086 return error; 1087 } 1088 1089 return 0; 1090} 1091 1092/* 1093 * xfs_collapse_file_space() 1094 * This routine frees disk space and shift extent for the given file. 1095 * The first thing we do is to free data blocks in the specified range 1096 * by calling xfs_free_file_space(). It would also sync dirty data 1097 * and invalidate page cache over the region on which collapse range 1098 * is working. And Shift extent records to the left to cover a hole. 1099 * RETURNS: 1100 * 0 on success 1101 * errno on error 1102 * 1103 */ 1104int 1105xfs_collapse_file_space( 1106 struct xfs_inode *ip, 1107 xfs_off_t offset, 1108 xfs_off_t len) 1109{ 1110 struct xfs_mount *mp = ip->i_mount; 1111 struct xfs_trans *tp; 1112 int error; 1113 xfs_fileoff_t next_fsb = XFS_B_TO_FSB(mp, offset + len); 1114 xfs_fileoff_t shift_fsb = XFS_B_TO_FSB(mp, len); 1115 bool done = false; 1116 1117 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); 1118 ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL)); 1119 1120 trace_xfs_collapse_file_space(ip); 1121 1122 error = xfs_free_file_space(ip, offset, len); 1123 if (error) 1124 return error; 1125 1126 error = xfs_prepare_shift(ip, offset); 1127 if (error) 1128 return error; 1129 1130 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0, 0, 0, &tp); 1131 if (error) 1132 return error; 1133 1134 xfs_ilock(ip, XFS_ILOCK_EXCL); 1135 xfs_trans_ijoin(tp, ip, 0); 1136 1137 while (!done) { 1138 error = xfs_bmap_collapse_extents(tp, ip, &next_fsb, shift_fsb, 1139 &done); 1140 if (error) 1141 goto out_trans_cancel; 1142 if (done) 1143 break; 1144 1145 /* finish any deferred frees and roll the transaction */ 1146 error = xfs_defer_finish(&tp); 1147 if (error) 1148 goto out_trans_cancel; 1149 } 1150 1151 error = xfs_trans_commit(tp); 1152 xfs_iunlock(ip, XFS_ILOCK_EXCL); 1153 return error; 1154 1155out_trans_cancel: 1156 xfs_trans_cancel(tp); 1157 xfs_iunlock(ip, XFS_ILOCK_EXCL); 1158 return error; 1159} 1160 1161/* 1162 * xfs_insert_file_space() 1163 * This routine create hole space by shifting extents for the given file. 1164 * The first thing we do is to sync dirty data and invalidate page cache 1165 * over the region on which insert range is working. And split an extent 1166 * to two extents at given offset by calling xfs_bmap_split_extent. 1167 * And shift all extent records which are laying between [offset, 1168 * last allocated extent] to the right to reserve hole range. 1169 * RETURNS: 1170 * 0 on success 1171 * errno on error 1172 */ 1173int 1174xfs_insert_file_space( 1175 struct xfs_inode *ip, 1176 loff_t offset, 1177 loff_t len) 1178{ 1179 struct xfs_mount *mp = ip->i_mount; 1180 struct xfs_trans *tp; 1181 int error; 1182 xfs_fileoff_t stop_fsb = XFS_B_TO_FSB(mp, offset); 1183 xfs_fileoff_t next_fsb = NULLFSBLOCK; 1184 xfs_fileoff_t shift_fsb = XFS_B_TO_FSB(mp, len); 1185 bool done = false; 1186 1187 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); 1188 ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL)); 1189 1190 trace_xfs_insert_file_space(ip); 1191 1192 error = xfs_bmap_can_insert_extents(ip, stop_fsb, shift_fsb); 1193 if (error) 1194 return error; 1195 1196 error = xfs_prepare_shift(ip, offset); 1197 if (error) 1198 return error; 1199 1200 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 1201 XFS_DIOSTRAT_SPACE_RES(mp, 0), 0, 0, &tp); 1202 if (error) 1203 return error; 1204 1205 xfs_ilock(ip, XFS_ILOCK_EXCL); 1206 xfs_trans_ijoin(tp, ip, 0); 1207 1208 error = xfs_iext_count_may_overflow(ip, XFS_DATA_FORK, 1209 XFS_IEXT_PUNCH_HOLE_CNT); 1210 if (error == -EFBIG) 1211 error = xfs_iext_count_upgrade(tp, ip, XFS_IEXT_PUNCH_HOLE_CNT); 1212 if (error) 1213 goto out_trans_cancel; 1214 1215 /* 1216 * The extent shifting code works on extent granularity. So, if stop_fsb 1217 * is not the starting block of extent, we need to split the extent at 1218 * stop_fsb. 1219 */ 1220 error = xfs_bmap_split_extent(tp, ip, stop_fsb); 1221 if (error) 1222 goto out_trans_cancel; 1223 1224 do { 1225 error = xfs_defer_finish(&tp); 1226 if (error) 1227 goto out_trans_cancel; 1228 1229 error = xfs_bmap_insert_extents(tp, ip, &next_fsb, shift_fsb, 1230 &done, stop_fsb); 1231 if (error) 1232 goto out_trans_cancel; 1233 } while (!done); 1234 1235 error = xfs_trans_commit(tp); 1236 xfs_iunlock(ip, XFS_ILOCK_EXCL); 1237 return error; 1238 1239out_trans_cancel: 1240 xfs_trans_cancel(tp); 1241 xfs_iunlock(ip, XFS_ILOCK_EXCL); 1242 return error; 1243} 1244 1245/* 1246 * We need to check that the format of the data fork in the temporary inode is 1247 * valid for the target inode before doing the swap. This is not a problem with 1248 * attr1 because of the fixed fork offset, but attr2 has a dynamically sized 1249 * data fork depending on the space the attribute fork is taking so we can get 1250 * invalid formats on the target inode. 1251 * 1252 * E.g. target has space for 7 extents in extent format, temp inode only has 1253 * space for 6. If we defragment down to 7 extents, then the tmp format is a 1254 * btree, but when swapped it needs to be in extent format. Hence we can't just 1255 * blindly swap data forks on attr2 filesystems. 1256 * 1257 * Note that we check the swap in both directions so that we don't end up with 1258 * a corrupt temporary inode, either. 1259 * 1260 * Note that fixing the way xfs_fsr sets up the attribute fork in the source 1261 * inode will prevent this situation from occurring, so all we do here is 1262 * reject and log the attempt. basically we are putting the responsibility on 1263 * userspace to get this right. 1264 */ 1265static int 1266xfs_swap_extents_check_format( 1267 struct xfs_inode *ip, /* target inode */ 1268 struct xfs_inode *tip) /* tmp inode */ 1269{ 1270 struct xfs_ifork *ifp = &ip->i_df; 1271 struct xfs_ifork *tifp = &tip->i_df; 1272 1273 /* User/group/project quota ids must match if quotas are enforced. */ 1274 if (XFS_IS_QUOTA_ON(ip->i_mount) && 1275 (!uid_eq(VFS_I(ip)->i_uid, VFS_I(tip)->i_uid) || 1276 !gid_eq(VFS_I(ip)->i_gid, VFS_I(tip)->i_gid) || 1277 ip->i_projid != tip->i_projid)) 1278 return -EINVAL; 1279 1280 /* Should never get a local format */ 1281 if (ifp->if_format == XFS_DINODE_FMT_LOCAL || 1282 tifp->if_format == XFS_DINODE_FMT_LOCAL) 1283 return -EINVAL; 1284 1285 /* 1286 * if the target inode has less extents that then temporary inode then 1287 * why did userspace call us? 1288 */ 1289 if (ifp->if_nextents < tifp->if_nextents) 1290 return -EINVAL; 1291 1292 /* 1293 * If we have to use the (expensive) rmap swap method, we can 1294 * handle any number of extents and any format. 1295 */ 1296 if (xfs_has_rmapbt(ip->i_mount)) 1297 return 0; 1298 1299 /* 1300 * if the target inode is in extent form and the temp inode is in btree 1301 * form then we will end up with the target inode in the wrong format 1302 * as we already know there are less extents in the temp inode. 1303 */ 1304 if (ifp->if_format == XFS_DINODE_FMT_EXTENTS && 1305 tifp->if_format == XFS_DINODE_FMT_BTREE) 1306 return -EINVAL; 1307 1308 /* Check temp in extent form to max in target */ 1309 if (tifp->if_format == XFS_DINODE_FMT_EXTENTS && 1310 tifp->if_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)) 1311 return -EINVAL; 1312 1313 /* Check target in extent form to max in temp */ 1314 if (ifp->if_format == XFS_DINODE_FMT_EXTENTS && 1315 ifp->if_nextents > XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK)) 1316 return -EINVAL; 1317 1318 /* 1319 * If we are in a btree format, check that the temp root block will fit 1320 * in the target and that it has enough extents to be in btree format 1321 * in the target. 1322 * 1323 * Note that we have to be careful to allow btree->extent conversions 1324 * (a common defrag case) which will occur when the temp inode is in 1325 * extent format... 1326 */ 1327 if (tifp->if_format == XFS_DINODE_FMT_BTREE) { 1328 if (xfs_inode_has_attr_fork(ip) && 1329 XFS_BMAP_BMDR_SPACE(tifp->if_broot) > xfs_inode_fork_boff(ip)) 1330 return -EINVAL; 1331 if (tifp->if_nextents <= XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)) 1332 return -EINVAL; 1333 } 1334 1335 /* Reciprocal target->temp btree format checks */ 1336 if (ifp->if_format == XFS_DINODE_FMT_BTREE) { 1337 if (xfs_inode_has_attr_fork(tip) && 1338 XFS_BMAP_BMDR_SPACE(ip->i_df.if_broot) > xfs_inode_fork_boff(tip)) 1339 return -EINVAL; 1340 if (ifp->if_nextents <= XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK)) 1341 return -EINVAL; 1342 } 1343 1344 return 0; 1345} 1346 1347static int 1348xfs_swap_extent_flush( 1349 struct xfs_inode *ip) 1350{ 1351 int error; 1352 1353 error = filemap_write_and_wait(VFS_I(ip)->i_mapping); 1354 if (error) 1355 return error; 1356 truncate_pagecache_range(VFS_I(ip), 0, -1); 1357 1358 /* Verify O_DIRECT for ftmp */ 1359 if (VFS_I(ip)->i_mapping->nrpages) 1360 return -EINVAL; 1361 return 0; 1362} 1363 1364/* 1365 * Move extents from one file to another, when rmap is enabled. 1366 */ 1367STATIC int 1368xfs_swap_extent_rmap( 1369 struct xfs_trans **tpp, 1370 struct xfs_inode *ip, 1371 struct xfs_inode *tip) 1372{ 1373 struct xfs_trans *tp = *tpp; 1374 struct xfs_bmbt_irec irec; 1375 struct xfs_bmbt_irec uirec; 1376 struct xfs_bmbt_irec tirec; 1377 xfs_fileoff_t offset_fsb; 1378 xfs_fileoff_t end_fsb; 1379 xfs_filblks_t count_fsb; 1380 int error; 1381 xfs_filblks_t ilen; 1382 xfs_filblks_t rlen; 1383 int nimaps; 1384 uint64_t tip_flags2; 1385 1386 /* 1387 * If the source file has shared blocks, we must flag the donor 1388 * file as having shared blocks so that we get the shared-block 1389 * rmap functions when we go to fix up the rmaps. The flags 1390 * will be switch for reals later. 1391 */ 1392 tip_flags2 = tip->i_diflags2; 1393 if (ip->i_diflags2 & XFS_DIFLAG2_REFLINK) 1394 tip->i_diflags2 |= XFS_DIFLAG2_REFLINK; 1395 1396 offset_fsb = 0; 1397 end_fsb = XFS_B_TO_FSB(ip->i_mount, i_size_read(VFS_I(ip))); 1398 count_fsb = (xfs_filblks_t)(end_fsb - offset_fsb); 1399 1400 while (count_fsb) { 1401 /* Read extent from the donor file */ 1402 nimaps = 1; 1403 error = xfs_bmapi_read(tip, offset_fsb, count_fsb, &tirec, 1404 &nimaps, 0); 1405 if (error) 1406 goto out; 1407 ASSERT(nimaps == 1); 1408 ASSERT(tirec.br_startblock != DELAYSTARTBLOCK); 1409 1410 trace_xfs_swap_extent_rmap_remap(tip, &tirec); 1411 ilen = tirec.br_blockcount; 1412 1413 /* Unmap the old blocks in the source file. */ 1414 while (tirec.br_blockcount) { 1415 ASSERT(tp->t_highest_agno == NULLAGNUMBER); 1416 trace_xfs_swap_extent_rmap_remap_piece(tip, &tirec); 1417 1418 /* Read extent from the source file */ 1419 nimaps = 1; 1420 error = xfs_bmapi_read(ip, tirec.br_startoff, 1421 tirec.br_blockcount, &irec, 1422 &nimaps, 0); 1423 if (error) 1424 goto out; 1425 ASSERT(nimaps == 1); 1426 ASSERT(tirec.br_startoff == irec.br_startoff); 1427 trace_xfs_swap_extent_rmap_remap_piece(ip, &irec); 1428 1429 /* Trim the extent. */ 1430 uirec = tirec; 1431 uirec.br_blockcount = rlen = min_t(xfs_filblks_t, 1432 tirec.br_blockcount, 1433 irec.br_blockcount); 1434 trace_xfs_swap_extent_rmap_remap_piece(tip, &uirec); 1435 1436 if (xfs_bmap_is_real_extent(&uirec)) { 1437 error = xfs_iext_count_may_overflow(ip, 1438 XFS_DATA_FORK, 1439 XFS_IEXT_SWAP_RMAP_CNT); 1440 if (error == -EFBIG) 1441 error = xfs_iext_count_upgrade(tp, ip, 1442 XFS_IEXT_SWAP_RMAP_CNT); 1443 if (error) 1444 goto out; 1445 } 1446 1447 if (xfs_bmap_is_real_extent(&irec)) { 1448 error = xfs_iext_count_may_overflow(tip, 1449 XFS_DATA_FORK, 1450 XFS_IEXT_SWAP_RMAP_CNT); 1451 if (error == -EFBIG) 1452 error = xfs_iext_count_upgrade(tp, ip, 1453 XFS_IEXT_SWAP_RMAP_CNT); 1454 if (error) 1455 goto out; 1456 } 1457 1458 /* Remove the mapping from the donor file. */ 1459 xfs_bmap_unmap_extent(tp, tip, &uirec); 1460 1461 /* Remove the mapping from the source file. */ 1462 xfs_bmap_unmap_extent(tp, ip, &irec); 1463 1464 /* Map the donor file's blocks into the source file. */ 1465 xfs_bmap_map_extent(tp, ip, &uirec); 1466 1467 /* Map the source file's blocks into the donor file. */ 1468 xfs_bmap_map_extent(tp, tip, &irec); 1469 1470 error = xfs_defer_finish(tpp); 1471 tp = *tpp; 1472 if (error) 1473 goto out; 1474 1475 tirec.br_startoff += rlen; 1476 if (tirec.br_startblock != HOLESTARTBLOCK && 1477 tirec.br_startblock != DELAYSTARTBLOCK) 1478 tirec.br_startblock += rlen; 1479 tirec.br_blockcount -= rlen; 1480 } 1481 1482 /* Roll on... */ 1483 count_fsb -= ilen; 1484 offset_fsb += ilen; 1485 } 1486 1487 tip->i_diflags2 = tip_flags2; 1488 return 0; 1489 1490out: 1491 trace_xfs_swap_extent_rmap_error(ip, error, _RET_IP_); 1492 tip->i_diflags2 = tip_flags2; 1493 return error; 1494} 1495 1496/* Swap the extents of two files by swapping data forks. */ 1497STATIC int 1498xfs_swap_extent_forks( 1499 struct xfs_trans *tp, 1500 struct xfs_inode *ip, 1501 struct xfs_inode *tip, 1502 int *src_log_flags, 1503 int *target_log_flags) 1504{ 1505 xfs_filblks_t aforkblks = 0; 1506 xfs_filblks_t taforkblks = 0; 1507 xfs_extnum_t junk; 1508 uint64_t tmp; 1509 int error; 1510 1511 /* 1512 * Count the number of extended attribute blocks 1513 */ 1514 if (xfs_inode_has_attr_fork(ip) && ip->i_af.if_nextents > 0 && 1515 ip->i_af.if_format != XFS_DINODE_FMT_LOCAL) { 1516 error = xfs_bmap_count_blocks(tp, ip, XFS_ATTR_FORK, &junk, 1517 &aforkblks); 1518 if (error) 1519 return error; 1520 } 1521 if (xfs_inode_has_attr_fork(tip) && tip->i_af.if_nextents > 0 && 1522 tip->i_af.if_format != XFS_DINODE_FMT_LOCAL) { 1523 error = xfs_bmap_count_blocks(tp, tip, XFS_ATTR_FORK, &junk, 1524 &taforkblks); 1525 if (error) 1526 return error; 1527 } 1528 1529 /* 1530 * Btree format (v3) inodes have the inode number stamped in the bmbt 1531 * block headers. We can't start changing the bmbt blocks until the 1532 * inode owner change is logged so recovery does the right thing in the 1533 * event of a crash. Set the owner change log flags now and leave the 1534 * bmbt scan as the last step. 1535 */ 1536 if (xfs_has_v3inodes(ip->i_mount)) { 1537 if (ip->i_df.if_format == XFS_DINODE_FMT_BTREE) 1538 (*target_log_flags) |= XFS_ILOG_DOWNER; 1539 if (tip->i_df.if_format == XFS_DINODE_FMT_BTREE) 1540 (*src_log_flags) |= XFS_ILOG_DOWNER; 1541 } 1542 1543 /* 1544 * Swap the data forks of the inodes 1545 */ 1546 swap(ip->i_df, tip->i_df); 1547 1548 /* 1549 * Fix the on-disk inode values 1550 */ 1551 tmp = (uint64_t)ip->i_nblocks; 1552 ip->i_nblocks = tip->i_nblocks - taforkblks + aforkblks; 1553 tip->i_nblocks = tmp + taforkblks - aforkblks; 1554 1555 /* 1556 * The extents in the source inode could still contain speculative 1557 * preallocation beyond EOF (e.g. the file is open but not modified 1558 * while defrag is in progress). In that case, we need to copy over the 1559 * number of delalloc blocks the data fork in the source inode is 1560 * tracking beyond EOF so that when the fork is truncated away when the 1561 * temporary inode is unlinked we don't underrun the i_delayed_blks 1562 * counter on that inode. 1563 */ 1564 ASSERT(tip->i_delayed_blks == 0); 1565 tip->i_delayed_blks = ip->i_delayed_blks; 1566 ip->i_delayed_blks = 0; 1567 1568 switch (ip->i_df.if_format) { 1569 case XFS_DINODE_FMT_EXTENTS: 1570 (*src_log_flags) |= XFS_ILOG_DEXT; 1571 break; 1572 case XFS_DINODE_FMT_BTREE: 1573 ASSERT(!xfs_has_v3inodes(ip->i_mount) || 1574 (*src_log_flags & XFS_ILOG_DOWNER)); 1575 (*src_log_flags) |= XFS_ILOG_DBROOT; 1576 break; 1577 } 1578 1579 switch (tip->i_df.if_format) { 1580 case XFS_DINODE_FMT_EXTENTS: 1581 (*target_log_flags) |= XFS_ILOG_DEXT; 1582 break; 1583 case XFS_DINODE_FMT_BTREE: 1584 (*target_log_flags) |= XFS_ILOG_DBROOT; 1585 ASSERT(!xfs_has_v3inodes(ip->i_mount) || 1586 (*target_log_flags & XFS_ILOG_DOWNER)); 1587 break; 1588 } 1589 1590 return 0; 1591} 1592 1593/* 1594 * Fix up the owners of the bmbt blocks to refer to the current inode. The 1595 * change owner scan attempts to order all modified buffers in the current 1596 * transaction. In the event of ordered buffer failure, the offending buffer is 1597 * physically logged as a fallback and the scan returns -EAGAIN. We must roll 1598 * the transaction in this case to replenish the fallback log reservation and 1599 * restart the scan. This process repeats until the scan completes. 1600 */ 1601static int 1602xfs_swap_change_owner( 1603 struct xfs_trans **tpp, 1604 struct xfs_inode *ip, 1605 struct xfs_inode *tmpip) 1606{ 1607 int error; 1608 struct xfs_trans *tp = *tpp; 1609 1610 do { 1611 error = xfs_bmbt_change_owner(tp, ip, XFS_DATA_FORK, ip->i_ino, 1612 NULL); 1613 /* success or fatal error */ 1614 if (error != -EAGAIN) 1615 break; 1616 1617 error = xfs_trans_roll(tpp); 1618 if (error) 1619 break; 1620 tp = *tpp; 1621 1622 /* 1623 * Redirty both inodes so they can relog and keep the log tail 1624 * moving forward. 1625 */ 1626 xfs_trans_ijoin(tp, ip, 0); 1627 xfs_trans_ijoin(tp, tmpip, 0); 1628 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 1629 xfs_trans_log_inode(tp, tmpip, XFS_ILOG_CORE); 1630 } while (true); 1631 1632 return error; 1633} 1634 1635int 1636xfs_swap_extents( 1637 struct xfs_inode *ip, /* target inode */ 1638 struct xfs_inode *tip, /* tmp inode */ 1639 struct xfs_swapext *sxp) 1640{ 1641 struct xfs_mount *mp = ip->i_mount; 1642 struct xfs_trans *tp; 1643 struct xfs_bstat *sbp = &sxp->sx_stat; 1644 int src_log_flags, target_log_flags; 1645 int error = 0; 1646 uint64_t f; 1647 int resblks = 0; 1648 unsigned int flags = 0; 1649 struct timespec64 ctime; 1650 1651 /* 1652 * Lock the inodes against other IO, page faults and truncate to 1653 * begin with. Then we can ensure the inodes are flushed and have no 1654 * page cache safely. Once we have done this we can take the ilocks and 1655 * do the rest of the checks. 1656 */ 1657 lock_two_nondirectories(VFS_I(ip), VFS_I(tip)); 1658 filemap_invalidate_lock_two(VFS_I(ip)->i_mapping, 1659 VFS_I(tip)->i_mapping); 1660 1661 /* Verify that both files have the same format */ 1662 if ((VFS_I(ip)->i_mode & S_IFMT) != (VFS_I(tip)->i_mode & S_IFMT)) { 1663 error = -EINVAL; 1664 goto out_unlock; 1665 } 1666 1667 /* Verify both files are either real-time or non-realtime */ 1668 if (XFS_IS_REALTIME_INODE(ip) != XFS_IS_REALTIME_INODE(tip)) { 1669 error = -EINVAL; 1670 goto out_unlock; 1671 } 1672 1673 error = xfs_qm_dqattach(ip); 1674 if (error) 1675 goto out_unlock; 1676 1677 error = xfs_qm_dqattach(tip); 1678 if (error) 1679 goto out_unlock; 1680 1681 error = xfs_swap_extent_flush(ip); 1682 if (error) 1683 goto out_unlock; 1684 error = xfs_swap_extent_flush(tip); 1685 if (error) 1686 goto out_unlock; 1687 1688 if (xfs_inode_has_cow_data(tip)) { 1689 error = xfs_reflink_cancel_cow_range(tip, 0, NULLFILEOFF, true); 1690 if (error) 1691 goto out_unlock; 1692 } 1693 1694 /* 1695 * Extent "swapping" with rmap requires a permanent reservation and 1696 * a block reservation because it's really just a remap operation 1697 * performed with log redo items! 1698 */ 1699 if (xfs_has_rmapbt(mp)) { 1700 int w = XFS_DATA_FORK; 1701 uint32_t ipnext = ip->i_df.if_nextents; 1702 uint32_t tipnext = tip->i_df.if_nextents; 1703 1704 /* 1705 * Conceptually this shouldn't affect the shape of either bmbt, 1706 * but since we atomically move extents one by one, we reserve 1707 * enough space to rebuild both trees. 1708 */ 1709 resblks = XFS_SWAP_RMAP_SPACE_RES(mp, ipnext, w); 1710 resblks += XFS_SWAP_RMAP_SPACE_RES(mp, tipnext, w); 1711 1712 /* 1713 * If either inode straddles a bmapbt block allocation boundary, 1714 * the rmapbt algorithm triggers repeated allocs and frees as 1715 * extents are remapped. This can exhaust the block reservation 1716 * prematurely and cause shutdown. Return freed blocks to the 1717 * transaction reservation to counter this behavior. 1718 */ 1719 flags |= XFS_TRANS_RES_FDBLKS; 1720 } 1721 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0, flags, 1722 &tp); 1723 if (error) 1724 goto out_unlock; 1725 1726 /* 1727 * Lock and join the inodes to the tansaction so that transaction commit 1728 * or cancel will unlock the inodes from this point onwards. 1729 */ 1730 xfs_lock_two_inodes(ip, XFS_ILOCK_EXCL, tip, XFS_ILOCK_EXCL); 1731 xfs_trans_ijoin(tp, ip, 0); 1732 xfs_trans_ijoin(tp, tip, 0); 1733 1734 1735 /* Verify all data are being swapped */ 1736 if (sxp->sx_offset != 0 || 1737 sxp->sx_length != ip->i_disk_size || 1738 sxp->sx_length != tip->i_disk_size) { 1739 error = -EFAULT; 1740 goto out_trans_cancel; 1741 } 1742 1743 trace_xfs_swap_extent_before(ip, 0); 1744 trace_xfs_swap_extent_before(tip, 1); 1745 1746 /* check inode formats now that data is flushed */ 1747 error = xfs_swap_extents_check_format(ip, tip); 1748 if (error) { 1749 xfs_notice(mp, 1750 "%s: inode 0x%llx format is incompatible for exchanging.", 1751 __func__, ip->i_ino); 1752 goto out_trans_cancel; 1753 } 1754 1755 /* 1756 * Compare the current change & modify times with that 1757 * passed in. If they differ, we abort this swap. 1758 * This is the mechanism used to ensure the calling 1759 * process that the file was not changed out from 1760 * under it. 1761 */ 1762 ctime = inode_get_ctime(VFS_I(ip)); 1763 if ((sbp->bs_ctime.tv_sec != ctime.tv_sec) || 1764 (sbp->bs_ctime.tv_nsec != ctime.tv_nsec) || 1765 (sbp->bs_mtime.tv_sec != VFS_I(ip)->i_mtime.tv_sec) || 1766 (sbp->bs_mtime.tv_nsec != VFS_I(ip)->i_mtime.tv_nsec)) { 1767 error = -EBUSY; 1768 goto out_trans_cancel; 1769 } 1770 1771 /* 1772 * Note the trickiness in setting the log flags - we set the owner log 1773 * flag on the opposite inode (i.e. the inode we are setting the new 1774 * owner to be) because once we swap the forks and log that, log 1775 * recovery is going to see the fork as owned by the swapped inode, 1776 * not the pre-swapped inodes. 1777 */ 1778 src_log_flags = XFS_ILOG_CORE; 1779 target_log_flags = XFS_ILOG_CORE; 1780 1781 if (xfs_has_rmapbt(mp)) 1782 error = xfs_swap_extent_rmap(&tp, ip, tip); 1783 else 1784 error = xfs_swap_extent_forks(tp, ip, tip, &src_log_flags, 1785 &target_log_flags); 1786 if (error) 1787 goto out_trans_cancel; 1788 1789 /* Do we have to swap reflink flags? */ 1790 if ((ip->i_diflags2 & XFS_DIFLAG2_REFLINK) ^ 1791 (tip->i_diflags2 & XFS_DIFLAG2_REFLINK)) { 1792 f = ip->i_diflags2 & XFS_DIFLAG2_REFLINK; 1793 ip->i_diflags2 &= ~XFS_DIFLAG2_REFLINK; 1794 ip->i_diflags2 |= tip->i_diflags2 & XFS_DIFLAG2_REFLINK; 1795 tip->i_diflags2 &= ~XFS_DIFLAG2_REFLINK; 1796 tip->i_diflags2 |= f & XFS_DIFLAG2_REFLINK; 1797 } 1798 1799 /* Swap the cow forks. */ 1800 if (xfs_has_reflink(mp)) { 1801 ASSERT(!ip->i_cowfp || 1802 ip->i_cowfp->if_format == XFS_DINODE_FMT_EXTENTS); 1803 ASSERT(!tip->i_cowfp || 1804 tip->i_cowfp->if_format == XFS_DINODE_FMT_EXTENTS); 1805 1806 swap(ip->i_cowfp, tip->i_cowfp); 1807 1808 if (ip->i_cowfp && ip->i_cowfp->if_bytes) 1809 xfs_inode_set_cowblocks_tag(ip); 1810 else 1811 xfs_inode_clear_cowblocks_tag(ip); 1812 if (tip->i_cowfp && tip->i_cowfp->if_bytes) 1813 xfs_inode_set_cowblocks_tag(tip); 1814 else 1815 xfs_inode_clear_cowblocks_tag(tip); 1816 } 1817 1818 xfs_trans_log_inode(tp, ip, src_log_flags); 1819 xfs_trans_log_inode(tp, tip, target_log_flags); 1820 1821 /* 1822 * The extent forks have been swapped, but crc=1,rmapbt=0 filesystems 1823 * have inode number owner values in the bmbt blocks that still refer to 1824 * the old inode. Scan each bmbt to fix up the owner values with the 1825 * inode number of the current inode. 1826 */ 1827 if (src_log_flags & XFS_ILOG_DOWNER) { 1828 error = xfs_swap_change_owner(&tp, ip, tip); 1829 if (error) 1830 goto out_trans_cancel; 1831 } 1832 if (target_log_flags & XFS_ILOG_DOWNER) { 1833 error = xfs_swap_change_owner(&tp, tip, ip); 1834 if (error) 1835 goto out_trans_cancel; 1836 } 1837 1838 /* 1839 * If this is a synchronous mount, make sure that the 1840 * transaction goes to disk before returning to the user. 1841 */ 1842 if (xfs_has_wsync(mp)) 1843 xfs_trans_set_sync(tp); 1844 1845 error = xfs_trans_commit(tp); 1846 1847 trace_xfs_swap_extent_after(ip, 0); 1848 trace_xfs_swap_extent_after(tip, 1); 1849 1850out_unlock_ilock: 1851 xfs_iunlock(ip, XFS_ILOCK_EXCL); 1852 xfs_iunlock(tip, XFS_ILOCK_EXCL); 1853out_unlock: 1854 filemap_invalidate_unlock_two(VFS_I(ip)->i_mapping, 1855 VFS_I(tip)->i_mapping); 1856 unlock_two_nondirectories(VFS_I(ip), VFS_I(tip)); 1857 return error; 1858 1859out_trans_cancel: 1860 xfs_trans_cancel(tp); 1861 goto out_unlock_ilock; 1862} 1863