1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1989, 1993 5 * The Regents of the University of California. All rights reserved. 6 * (c) UNIX System Laboratories, Inc. 7 * All or some portions of this file are derived from material licensed 8 * to the University of California by American Telephone and Telegraph 9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 10 * the permission of UNIX System Laboratories, Inc. 11 * 12 * Redistribution and use in source and binary forms, with or without 13 * modification, are permitted provided that the following conditions 14 * are met: 15 * 1. Redistributions of source code must retain the above copyright 16 * notice, this list of conditions and the following disclaimer. 17 * 2. Redistributions in binary form must reproduce the above copyright 18 * notice, this list of conditions and the following disclaimer in the 19 * documentation and/or other materials provided with the distribution. 20 * 3. Neither the name of the University nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 * 36 * @(#)ufs_lookup.c 8.15 (Berkeley) 6/16/95 37 */ 38 39 #include <sys/cdefs.h> 40 __FBSDID("$FreeBSD$"); 41 42 #include "opt_ufs.h" 43 #include "opt_quota.h" 44 45 #include <sys/param.h> 46 #include <sys/systm.h> 47 #include <sys/kernel.h> 48 #include <sys/namei.h> 49 #include <sys/bio.h> 50 #include <sys/buf.h> 51 #include <sys/proc.h> 52 #include <sys/stat.h> 53 #include <sys/mount.h> 54 #include <sys/vnode.h> 55 #include <sys/sysctl.h> 56 57 #include <vm/vm.h> 58 #include <vm/vm_extern.h> 59 60 #include <ufs/ufs/extattr.h> 61 #include <ufs/ufs/quota.h> 62 #include <ufs/ufs/inode.h> 63 #include <ufs/ufs/dir.h> 64 #ifdef UFS_DIRHASH 65 #include <ufs/ufs/dirhash.h> 66 #endif 67 #include <ufs/ufs/ufsmount.h> 68 #include <ufs/ufs/ufs_extern.h> 69 #include <ufs/ffs/ffs_extern.h> 70 71 #ifdef DIAGNOSTIC 72 static int dirchk = 1; 73 #else 74 static int dirchk = 0; 75 #endif 76 77 SYSCTL_INT(_debug, OID_AUTO, dircheck, CTLFLAG_RW, &dirchk, 0, ""); 78 79 static int 80 ufs_delete_denied(struct vnode *vdp, struct vnode *tdp, struct ucred *cred, 81 struct thread *td) 82 { 83 int error; 84 85 #ifdef UFS_ACL 86 /* 87 * NFSv4 Minor Version 1, draft-ietf-nfsv4-minorversion1-03.txt 88 * 89 * 3.16.2.1. ACE4_DELETE vs. ACE4_DELETE_CHILD 90 */ 91 92 /* 93 * XXX: Is this check required? 94 */ 95 error = VOP_ACCESS(vdp, VEXEC, cred, td); 96 if (error) 97 return (error); 98 99 error = VOP_ACCESSX(tdp, VDELETE, cred, td); 100 if (error == 0) 101 return (0); 102 103 error = VOP_ACCESSX(vdp, VDELETE_CHILD, cred, td); 104 if (error == 0) 105 return (0); 106 107 error = VOP_ACCESSX(vdp, VEXPLICIT_DENY | VDELETE_CHILD, cred, td); 108 if (error) 109 return (error); 110 111 #endif /* !UFS_ACL */ 112 113 /* 114 * Standard Unix access control - delete access requires VWRITE. 115 */ 116 error = VOP_ACCESS(vdp, VWRITE, cred, td); 117 if (error) 118 return (error); 119 120 /* 121 * If directory is "sticky", then user must own 122 * the directory, or the file in it, else she 123 * may not delete it (unless she's root). This 124 * implements append-only directories. 125 */ 126 if ((VTOI(vdp)->i_mode & ISVTX) && 127 VOP_ACCESS(vdp, VADMIN, cred, td) && 128 VOP_ACCESS(tdp, VADMIN, cred, td)) 129 return (EPERM); 130 131 return (0); 132 } 133 134 /* 135 * Convert a component of a pathname into a pointer to a locked inode. 136 * This is a very central and rather complicated routine. 137 * If the filesystem is not maintained in a strict tree hierarchy, 138 * this can result in a deadlock situation (see comments in code below). 139 * 140 * The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending 141 * on whether the name is to be looked up, created, renamed, or deleted. 142 * When CREATE, RENAME, or DELETE is specified, information usable in 143 * creating, renaming, or deleting a directory entry may be calculated. 144 * If flag has LOCKPARENT or'ed into it and the target of the pathname 145 * exists, lookup returns both the target and its parent directory locked. 146 * When creating or renaming and LOCKPARENT is specified, the target may 147 * not be ".". When deleting and LOCKPARENT is specified, the target may 148 * be "."., but the caller must check to ensure it does an vrele and vput 149 * instead of two vputs. 150 * 151 * This routine is actually used as VOP_CACHEDLOOKUP method, and the 152 * filesystem employs the generic vfs_cache_lookup() as VOP_LOOKUP 153 * method. 154 * 155 * vfs_cache_lookup() performs the following for us: 156 * check that it is a directory 157 * check accessibility of directory 158 * check for modification attempts on read-only mounts 159 * if name found in cache 160 * if at end of path and deleting or creating 161 * drop it 162 * else 163 * return name. 164 * return VOP_CACHEDLOOKUP() 165 * 166 * Overall outline of ufs_lookup: 167 * 168 * search for name in directory, to found or notfound 169 * notfound: 170 * if creating, return locked directory, leaving info on available slots 171 * else return error 172 * found: 173 * if at end of path and deleting, return information to allow delete 174 * if at end of path and rewriting (RENAME and LOCKPARENT), lock target 175 * inode and return info to allow rewrite 176 * if not at end, add name to cache; if at end and neither creating 177 * nor deleting, add name to cache 178 */ 179 int 180 ufs_lookup(ap) 181 struct vop_cachedlookup_args /* { 182 struct vnode *a_dvp; 183 struct vnode **a_vpp; 184 struct componentname *a_cnp; 185 } */ *ap; 186 { 187 188 return (ufs_lookup_ino(ap->a_dvp, ap->a_vpp, ap->a_cnp, NULL)); 189 } 190 191 int 192 ufs_lookup_ino(struct vnode *vdp, struct vnode **vpp, struct componentname *cnp, 193 ino_t *dd_ino) 194 { 195 struct inode *dp; /* inode for directory being searched */ 196 struct buf *bp; /* a buffer of directory entries */ 197 struct direct *ep; /* the current directory entry */ 198 int entryoffsetinblock; /* offset of ep in bp's buffer */ 199 enum {NONE, COMPACT, FOUND} slotstatus; 200 doff_t slotoffset; /* offset of area with free space */ 201 doff_t i_diroff; /* cached i_diroff value. */ 202 doff_t i_offset; /* cached i_offset value. */ 203 int slotsize; /* size of area at slotoffset */ 204 int slotfreespace; /* amount of space free in slot */ 205 int slotneeded; /* size of the entry we're seeking */ 206 int numdirpasses; /* strategy for directory search */ 207 doff_t endsearch; /* offset to end directory search */ 208 doff_t prevoff; /* prev entry dp->i_offset */ 209 struct vnode *pdp; /* saved dp during symlink work */ 210 struct vnode *tdp; /* returned by VFS_VGET */ 211 doff_t enduseful; /* pointer past last used dir slot */ 212 u_long bmask; /* block offset mask */ 213 int namlen, error; 214 struct ucred *cred = cnp->cn_cred; 215 int flags = cnp->cn_flags; 216 int nameiop = cnp->cn_nameiop; 217 ino_t ino, ino1; 218 int ltype; 219 220 if (vpp != NULL) 221 *vpp = NULL; 222 223 dp = VTOI(vdp); 224 if (dp->i_effnlink == 0) 225 return (ENOENT); 226 227 /* 228 * Create a vm object if vmiodirenable is enabled. 229 * Alternatively we could call vnode_create_vobject 230 * in VFS_VGET but we could end up creating objects 231 * that are never used. 232 */ 233 vnode_create_vobject(vdp, DIP(dp, i_size), cnp->cn_thread); 234 235 bmask = VFSTOUFS(vdp->v_mount)->um_mountp->mnt_stat.f_iosize - 1; 236 237 #ifdef DEBUG_VFS_LOCKS 238 /* 239 * Assert that the directory vnode is locked, and locked 240 * exclusively for the last component lookup for modifying 241 * operations. 242 * 243 * The directory-modifying operations need to save 244 * intermediate state in the inode between namei() call and 245 * actual directory manipulations. See fields in the struct 246 * inode marked as 'used during directory lookup'. We must 247 * ensure that upgrade in namei() does not happen, since 248 * upgrade might need to unlock vdp. If quotas are enabled, 249 * getinoquota() also requires exclusive lock to modify inode. 250 */ 251 ASSERT_VOP_LOCKED(vdp, "ufs_lookup1"); 252 if ((nameiop == CREATE || nameiop == DELETE || nameiop == RENAME) && 253 (flags & (LOCKPARENT | ISLASTCN)) == (LOCKPARENT | ISLASTCN)) 254 ASSERT_VOP_ELOCKED(vdp, "ufs_lookup2"); 255 #endif 256 257 restart: 258 bp = NULL; 259 slotoffset = -1; 260 261 /* 262 * We now have a segment name to search for, and a directory to search. 263 * 264 * Suppress search for slots unless creating 265 * file and at end of pathname, in which case 266 * we watch for a place to put the new file in 267 * case it doesn't already exist. 268 */ 269 ino = 0; 270 i_diroff = dp->i_diroff; 271 slotstatus = FOUND; 272 slotfreespace = slotsize = slotneeded = 0; 273 if ((nameiop == CREATE || nameiop == RENAME) && 274 (flags & ISLASTCN)) { 275 slotstatus = NONE; 276 slotneeded = DIRECTSIZ(cnp->cn_namelen); 277 } 278 279 #ifdef UFS_DIRHASH 280 /* 281 * Use dirhash for fast operations on large directories. The logic 282 * to determine whether to hash the directory is contained within 283 * ufsdirhash_build(); a zero return means that it decided to hash 284 * this directory and it successfully built up the hash table. 285 */ 286 if (ufsdirhash_build(dp) == 0) { 287 /* Look for a free slot if needed. */ 288 enduseful = dp->i_size; 289 if (slotstatus != FOUND) { 290 slotoffset = ufsdirhash_findfree(dp, slotneeded, 291 &slotsize); 292 if (slotoffset >= 0) { 293 slotstatus = COMPACT; 294 enduseful = ufsdirhash_enduseful(dp); 295 if (enduseful < 0) 296 enduseful = dp->i_size; 297 } 298 } 299 /* Look up the component. */ 300 numdirpasses = 1; 301 entryoffsetinblock = 0; /* silence compiler warning */ 302 switch (ufsdirhash_lookup(dp, cnp->cn_nameptr, cnp->cn_namelen, 303 &i_offset, &bp, nameiop == DELETE ? &prevoff : NULL)) { 304 case 0: 305 ep = (struct direct *)((char *)bp->b_data + 306 (i_offset & bmask)); 307 goto foundentry; 308 case ENOENT: 309 i_offset = roundup2(dp->i_size, DIRBLKSIZ); 310 goto notfound; 311 default: 312 /* Something failed; just do a linear search. */ 313 break; 314 } 315 } 316 #endif /* UFS_DIRHASH */ 317 /* 318 * If there is cached information on a previous search of 319 * this directory, pick up where we last left off. 320 * We cache only lookups as these are the most common 321 * and have the greatest payoff. Caching CREATE has little 322 * benefit as it usually must search the entire directory 323 * to determine that the entry does not exist. Caching the 324 * location of the last DELETE or RENAME has not reduced 325 * profiling time and hence has been removed in the interest 326 * of simplicity. 327 */ 328 if (nameiop != LOOKUP || i_diroff == 0 || i_diroff >= dp->i_size) { 329 entryoffsetinblock = 0; 330 i_offset = 0; 331 numdirpasses = 1; 332 } else { 333 i_offset = i_diroff; 334 if ((entryoffsetinblock = i_offset & bmask) && 335 (error = UFS_BLKATOFF(vdp, (off_t)i_offset, NULL, &bp))) 336 return (error); 337 numdirpasses = 2; 338 nchstats.ncs_2passes++; 339 } 340 prevoff = i_offset; 341 endsearch = roundup2(dp->i_size, DIRBLKSIZ); 342 enduseful = 0; 343 344 searchloop: 345 while (i_offset < endsearch) { 346 /* 347 * If necessary, get the next directory block. 348 */ 349 if ((i_offset & bmask) == 0) { 350 if (bp != NULL) 351 brelse(bp); 352 error = 353 UFS_BLKATOFF(vdp, (off_t)i_offset, NULL, &bp); 354 if (error) 355 return (error); 356 entryoffsetinblock = 0; 357 } 358 /* 359 * If still looking for a slot, and at a DIRBLKSIZE 360 * boundary, have to start looking for free space again. 361 */ 362 if (slotstatus == NONE && 363 (entryoffsetinblock & (DIRBLKSIZ - 1)) == 0) { 364 slotoffset = -1; 365 slotfreespace = 0; 366 } 367 /* 368 * Get pointer to next entry. 369 * Full validation checks are slow, so we only check 370 * enough to insure forward progress through the 371 * directory. Complete checks can be run by patching 372 * "dirchk" to be true. 373 */ 374 ep = (struct direct *)((char *)bp->b_data + entryoffsetinblock); 375 if (ep->d_reclen == 0 || ep->d_reclen > 376 DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1)) || 377 (dirchk && ufs_dirbadentry(vdp, ep, entryoffsetinblock))) { 378 int i; 379 380 ufs_dirbad(dp, i_offset, "mangled entry"); 381 i = DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1)); 382 i_offset += i; 383 entryoffsetinblock += i; 384 continue; 385 } 386 387 /* 388 * If an appropriate sized slot has not yet been found, 389 * check to see if one is available. Also accumulate space 390 * in the current block so that we can determine if 391 * compaction is viable. 392 */ 393 if (slotstatus != FOUND) { 394 int size = ep->d_reclen; 395 396 if (ep->d_ino != 0) 397 size -= DIRSIZ(OFSFMT(vdp), ep); 398 if (size > 0) { 399 if (size >= slotneeded) { 400 slotstatus = FOUND; 401 slotoffset = i_offset; 402 slotsize = ep->d_reclen; 403 } else if (slotstatus == NONE) { 404 slotfreespace += size; 405 if (slotoffset == -1) 406 slotoffset = i_offset; 407 if (slotfreespace >= slotneeded) { 408 slotstatus = COMPACT; 409 slotsize = i_offset + 410 ep->d_reclen - slotoffset; 411 } 412 } 413 } 414 } 415 416 /* 417 * Check for a name match. 418 */ 419 if (ep->d_ino) { 420 # if (BYTE_ORDER == LITTLE_ENDIAN) 421 if (OFSFMT(vdp)) 422 namlen = ep->d_type; 423 else 424 namlen = ep->d_namlen; 425 # else 426 namlen = ep->d_namlen; 427 # endif 428 if (namlen == cnp->cn_namelen && 429 (cnp->cn_nameptr[0] == ep->d_name[0]) && 430 !bcmp(cnp->cn_nameptr, ep->d_name, 431 (unsigned)namlen)) { 432 #ifdef UFS_DIRHASH 433 foundentry: 434 #endif 435 /* 436 * Save directory entry's inode number and 437 * reclen in ndp->ni_ufs area, and release 438 * directory buffer. 439 */ 440 if (!OFSFMT(vdp) && ep->d_type == DT_WHT) { 441 slotstatus = FOUND; 442 slotoffset = i_offset; 443 slotsize = ep->d_reclen; 444 enduseful = dp->i_size; 445 cnp->cn_flags |= ISWHITEOUT; 446 numdirpasses--; 447 goto notfound; 448 } 449 ino = ep->d_ino; 450 goto found; 451 } 452 } 453 prevoff = i_offset; 454 i_offset += ep->d_reclen; 455 entryoffsetinblock += ep->d_reclen; 456 if (ep->d_ino) 457 enduseful = i_offset; 458 } 459 notfound: 460 /* 461 * If we started in the middle of the directory and failed 462 * to find our target, we must check the beginning as well. 463 */ 464 if (numdirpasses == 2) { 465 numdirpasses--; 466 i_offset = 0; 467 endsearch = i_diroff; 468 goto searchloop; 469 } 470 if (bp != NULL) 471 brelse(bp); 472 /* 473 * If creating, and at end of pathname and current 474 * directory has not been removed, then can consider 475 * allowing file to be created. 476 */ 477 if ((nameiop == CREATE || nameiop == RENAME || 478 (nameiop == DELETE && 479 (cnp->cn_flags & DOWHITEOUT) && 480 (cnp->cn_flags & ISWHITEOUT))) && 481 (flags & ISLASTCN) && dp->i_effnlink != 0) { 482 /* 483 * Access for write is interpreted as allowing 484 * creation of files in the directory. 485 * 486 * XXX: Fix the comment above. 487 */ 488 if (flags & WILLBEDIR) 489 error = VOP_ACCESSX(vdp, VWRITE | VAPPEND, cred, cnp->cn_thread); 490 else 491 error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread); 492 if (error) 493 return (error); 494 /* 495 * Return an indication of where the new directory 496 * entry should be put. If we didn't find a slot, 497 * then set dp->i_count to 0 indicating 498 * that the new slot belongs at the end of the 499 * directory. If we found a slot, then the new entry 500 * can be put in the range from dp->i_offset to 501 * dp->i_offset + dp->i_count. 502 */ 503 if (slotstatus == NONE) { 504 SET_I_OFFSET(dp, roundup2(dp->i_size, DIRBLKSIZ)); 505 SET_I_COUNT(dp, 0); 506 enduseful = I_OFFSET(dp); 507 } else if (nameiop == DELETE) { 508 SET_I_OFFSET(dp, slotoffset); 509 if ((I_OFFSET(dp) & (DIRBLKSIZ - 1)) == 0) 510 SET_I_COUNT(dp, 0); 511 else 512 SET_I_COUNT(dp, I_OFFSET(dp) - prevoff); 513 } else { 514 SET_I_OFFSET(dp, slotoffset); 515 SET_I_COUNT(dp, slotsize); 516 if (enduseful < slotoffset + slotsize) 517 enduseful = slotoffset + slotsize; 518 } 519 SET_I_ENDOFF(dp, roundup2(enduseful, DIRBLKSIZ)); 520 /* 521 * We return with the directory locked, so that 522 * the parameters we set up above will still be 523 * valid if we actually decide to do a direnter(). 524 * We return ni_vp == NULL to indicate that the entry 525 * does not currently exist; we leave a pointer to 526 * the (locked) directory inode in ndp->ni_dvp. 527 * The pathname buffer is saved so that the name 528 * can be obtained later. 529 * 530 * NB - if the directory is unlocked, then this 531 * information cannot be used. 532 */ 533 cnp->cn_flags |= SAVENAME; 534 return (EJUSTRETURN); 535 } 536 /* 537 * Insert name into cache (as non-existent) if appropriate. 538 */ 539 if ((cnp->cn_flags & MAKEENTRY) != 0) 540 cache_enter(vdp, NULL, cnp); 541 return (ENOENT); 542 543 found: 544 if (dd_ino != NULL) 545 *dd_ino = ino; 546 if (numdirpasses == 2) 547 nchstats.ncs_pass2++; 548 /* 549 * Check that directory length properly reflects presence 550 * of this entry. 551 */ 552 if (i_offset + DIRSIZ(OFSFMT(vdp), ep) > dp->i_size) { 553 ufs_dirbad(dp, i_offset, "i_size too small"); 554 dp->i_size = i_offset + DIRSIZ(OFSFMT(vdp), ep); 555 DIP_SET(dp, i_size, dp->i_size); 556 UFS_INODE_SET_FLAG(dp, IN_SIZEMOD | IN_CHANGE | IN_UPDATE); 557 } 558 brelse(bp); 559 560 /* 561 * Found component in pathname. 562 * If the final component of path name, save information 563 * in the cache as to where the entry was found. 564 */ 565 if ((flags & ISLASTCN) && nameiop == LOOKUP) 566 dp->i_diroff = rounddown2(i_offset, DIRBLKSIZ); 567 568 /* 569 * If deleting, and at end of pathname, return 570 * parameters which can be used to remove file. 571 */ 572 if (nameiop == DELETE && (flags & ISLASTCN)) { 573 if (flags & LOCKPARENT) 574 ASSERT_VOP_ELOCKED(vdp, __FUNCTION__); 575 576 if (VOP_ISLOCKED(vdp) == LK_EXCLUSIVE) { 577 /* 578 * Return pointer to current entry in 579 * dp->i_offset, and distance past previous 580 * entry (if there is a previous entry in this 581 * block) in dp->i_count. 582 * 583 * We shouldn't be setting these in the 584 * WANTPARENT case (first lookup in rename()), but any 585 * lookups that will result in directory changes will 586 * overwrite these. 587 */ 588 SET_I_OFFSET(dp, i_offset); 589 if ((I_OFFSET(dp) & (DIRBLKSIZ - 1)) == 0) 590 SET_I_COUNT(dp, 0); 591 else 592 SET_I_COUNT(dp, I_OFFSET(dp) - prevoff); 593 } 594 if (dd_ino != NULL) 595 return (0); 596 597 /* 598 * Save directory inode pointer in ndp->ni_dvp for 599 * dirremove(). 600 */ 601 if ((error = VFS_VGET(vdp->v_mount, ino, 602 LK_EXCLUSIVE, &tdp)) != 0) 603 return (error); 604 error = ufs_delete_denied(vdp, tdp, cred, cnp->cn_thread); 605 if (error) { 606 vput(tdp); 607 return (error); 608 } 609 if (dp->i_number == ino) { 610 VREF(vdp); 611 *vpp = vdp; 612 vput(tdp); 613 return (0); 614 } 615 616 *vpp = tdp; 617 return (0); 618 } 619 620 /* 621 * If rewriting (RENAME), return the inode and the 622 * information required to rewrite the present directory 623 * Must get inode of directory entry to verify it's a 624 * regular file, or empty directory. 625 */ 626 if (nameiop == RENAME && (flags & ISLASTCN)) { 627 if (flags & WILLBEDIR) 628 error = VOP_ACCESSX(vdp, VWRITE | VAPPEND, cred, cnp->cn_thread); 629 else 630 error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread); 631 if (error) 632 return (error); 633 /* 634 * Careful about locking second inode. 635 * This can only occur if the target is ".". 636 */ 637 SET_I_OFFSET(dp, i_offset); 638 if (dp->i_number == ino) 639 return (EISDIR); 640 if (dd_ino != NULL) 641 return (0); 642 if ((error = VFS_VGET(vdp->v_mount, ino, 643 LK_EXCLUSIVE, &tdp)) != 0) 644 return (error); 645 646 error = ufs_delete_denied(vdp, tdp, cred, cnp->cn_thread); 647 if (error) { 648 vput(tdp); 649 return (error); 650 } 651 652 #ifdef SunOS_doesnt_do_that 653 /* 654 * The only purpose of this check is to return the correct 655 * error. Assume that we want to rename directory "a" 656 * to a file "b", and that we have no ACL_WRITE_DATA on 657 * a containing directory, but we _do_ have ACL_APPEND_DATA. 658 * In that case, the VOP_ACCESS check above will return 0, 659 * and the operation will fail with ENOTDIR instead 660 * of EACCESS. 661 */ 662 if (tdp->v_type == VDIR) 663 error = VOP_ACCESSX(vdp, VWRITE | VAPPEND, cred, cnp->cn_thread); 664 else 665 error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread); 666 if (error) { 667 vput(tdp); 668 return (error); 669 } 670 #endif 671 672 *vpp = tdp; 673 cnp->cn_flags |= SAVENAME; 674 return (0); 675 } 676 if (dd_ino != NULL) 677 return (0); 678 679 /* 680 * Step through the translation in the name. We do not `vput' the 681 * directory because we may need it again if a symbolic link 682 * is relative to the current directory. Instead we save it 683 * unlocked as "pdp". We must get the target inode before unlocking 684 * the directory to insure that the inode will not be removed 685 * before we get it. We prevent deadlock by always fetching 686 * inodes from the root, moving down the directory tree. Thus 687 * when following backward pointers ".." we must unlock the 688 * parent directory before getting the requested directory. 689 * There is a potential race condition here if both the current 690 * and parent directories are removed before the VFS_VGET for the 691 * inode associated with ".." returns. We hope that this occurs 692 * infrequently since we cannot avoid this race condition without 693 * implementing a sophisticated deadlock detection algorithm. 694 * Note also that this simple deadlock detection scheme will not 695 * work if the filesystem has any hard links other than ".." 696 * that point backwards in the directory structure. 697 */ 698 pdp = vdp; 699 if (flags & ISDOTDOT) { 700 error = vn_vget_ino(pdp, ino, cnp->cn_lkflags, &tdp); 701 if (error) 702 return (error); 703 704 /* 705 * Recheck that ".." entry in the vdp directory points 706 * to the inode we looked up before vdp lock was 707 * dropped. 708 */ 709 error = ufs_lookup_ino(pdp, NULL, cnp, &ino1); 710 if (error) { 711 vput(tdp); 712 return (error); 713 } 714 if (ino1 != ino) { 715 vput(tdp); 716 goto restart; 717 } 718 719 *vpp = tdp; 720 } else if (dp->i_number == ino) { 721 VREF(vdp); /* we want ourself, ie "." */ 722 /* 723 * When we lookup "." we still can be asked to lock it 724 * differently. 725 */ 726 ltype = cnp->cn_lkflags & LK_TYPE_MASK; 727 if (ltype != VOP_ISLOCKED(vdp)) { 728 if (ltype == LK_EXCLUSIVE) 729 vn_lock(vdp, LK_UPGRADE | LK_RETRY); 730 else /* if (ltype == LK_SHARED) */ 731 vn_lock(vdp, LK_DOWNGRADE | LK_RETRY); 732 /* 733 * Relock for the "." case may left us with 734 * reclaimed vnode. 735 */ 736 if (VN_IS_DOOMED(vdp)) { 737 vrele(vdp); 738 return (ENOENT); 739 } 740 } 741 *vpp = vdp; 742 } else { 743 error = VFS_VGET(pdp->v_mount, ino, cnp->cn_lkflags, &tdp); 744 if (error == 0 && VTOI(tdp)->i_mode == 0) { 745 vgone(tdp); 746 vput(tdp); 747 error = ENOENT; 748 } 749 if (error) 750 return (error); 751 *vpp = tdp; 752 } 753 754 /* 755 * Insert name into cache if appropriate. 756 */ 757 if (cnp->cn_flags & MAKEENTRY) 758 cache_enter(vdp, *vpp, cnp); 759 return (0); 760 } 761 762 void 763 ufs_dirbad(ip, offset, how) 764 struct inode *ip; 765 doff_t offset; 766 char *how; 767 { 768 struct mount *mp; 769 770 mp = ITOV(ip)->v_mount; 771 if ((mp->mnt_flag & MNT_RDONLY) == 0) 772 panic("ufs_dirbad: %s: bad dir ino %ju at offset %ld: %s", 773 mp->mnt_stat.f_mntonname, (uintmax_t)ip->i_number, 774 (long)offset, how); 775 else 776 (void)printf("%s: bad dir ino %ju at offset %ld: %s\n", 777 mp->mnt_stat.f_mntonname, (uintmax_t)ip->i_number, 778 (long)offset, how); 779 } 780 781 /* 782 * Do consistency checking on a directory entry: 783 * record length must be multiple of 4 784 * entry must fit in rest of its DIRBLKSIZ block 785 * record must be large enough to contain entry 786 * name is not longer than UFS_MAXNAMLEN 787 * name must be as long as advertised, and null terminated 788 */ 789 int 790 ufs_dirbadentry(dp, ep, entryoffsetinblock) 791 struct vnode *dp; 792 struct direct *ep; 793 int entryoffsetinblock; 794 { 795 int i, namlen; 796 797 # if (BYTE_ORDER == LITTLE_ENDIAN) 798 if (OFSFMT(dp)) 799 namlen = ep->d_type; 800 else 801 namlen = ep->d_namlen; 802 # else 803 namlen = ep->d_namlen; 804 # endif 805 if ((ep->d_reclen & 0x3) != 0 || 806 ep->d_reclen > DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1)) || 807 ep->d_reclen < DIRSIZ(OFSFMT(dp), ep) || namlen > UFS_MAXNAMLEN) { 808 /*return (1); */ 809 printf("First bad\n"); 810 goto bad; 811 } 812 if (ep->d_ino == 0) 813 return (0); 814 for (i = 0; i < namlen; i++) 815 if (ep->d_name[i] == '\0') { 816 /*return (1); */ 817 printf("Second bad\n"); 818 goto bad; 819 } 820 if (ep->d_name[i]) 821 goto bad; 822 return (0); 823 bad: 824 return (1); 825 } 826 827 /* 828 * Construct a new directory entry after a call to namei, using the 829 * parameters that it left in the componentname argument cnp. The 830 * argument ip is the inode to which the new directory entry will refer. 831 */ 832 void 833 ufs_makedirentry(ip, cnp, newdirp) 834 struct inode *ip; 835 struct componentname *cnp; 836 struct direct *newdirp; 837 { 838 u_int namelen; 839 840 namelen = (unsigned)cnp->cn_namelen; 841 KASSERT((cnp->cn_flags & SAVENAME) != 0, 842 ("ufs_makedirentry: missing name")); 843 KASSERT(namelen <= UFS_MAXNAMLEN, 844 ("ufs_makedirentry: name too long")); 845 newdirp->d_ino = ip->i_number; 846 newdirp->d_namlen = namelen; 847 848 /* Zero out after-name padding */ 849 *(u_int32_t *)(&newdirp->d_name[namelen & ~(DIR_ROUNDUP - 1)]) = 0; 850 851 bcopy(cnp->cn_nameptr, newdirp->d_name, namelen); 852 853 if (!OFSFMT(ITOV(ip))) 854 newdirp->d_type = IFTODT(ip->i_mode); 855 else { 856 newdirp->d_type = 0; 857 # if (BYTE_ORDER == LITTLE_ENDIAN) 858 { u_char tmp = newdirp->d_namlen; 859 newdirp->d_namlen = newdirp->d_type; 860 newdirp->d_type = tmp; } 861 # endif 862 } 863 } 864 865 /* 866 * Write a directory entry after a call to namei, using the parameters 867 * that it left in nameidata. The argument dirp is the new directory 868 * entry contents. Dvp is a pointer to the directory to be written, 869 * which was left locked by namei. Remaining parameters (dp->i_offset, 870 * dp->i_count) indicate how the space for the new entry is to be obtained. 871 * Non-null bp indicates that a directory is being created (for the 872 * soft dependency code). 873 */ 874 int 875 ufs_direnter(dvp, tvp, dirp, cnp, newdirbp) 876 struct vnode *dvp; 877 struct vnode *tvp; 878 struct direct *dirp; 879 struct componentname *cnp; 880 struct buf *newdirbp; 881 { 882 struct ucred *cr; 883 struct thread *td; 884 int newentrysize; 885 struct inode *dp; 886 struct buf *bp; 887 u_int dsize; 888 struct direct *ep, *nep; 889 u_int64_t old_isize; 890 int error, ret, blkoff, loc, spacefree, flags, namlen; 891 char *dirbuf; 892 893 td = curthread; /* XXX */ 894 cr = td->td_ucred; 895 896 dp = VTOI(dvp); 897 newentrysize = DIRSIZ(OFSFMT(dvp), dirp); 898 899 if (I_COUNT(dp) == 0) { 900 /* 901 * If dp->i_count is 0, then namei could find no 902 * space in the directory. Here, dp->i_offset will 903 * be on a directory block boundary and we will write the 904 * new entry into a fresh block. 905 */ 906 if (I_OFFSET(dp) & (DIRBLKSIZ - 1)) 907 panic("ufs_direnter: newblk"); 908 flags = BA_CLRBUF; 909 if (!DOINGSOFTDEP(dvp) && !DOINGASYNC(dvp)) 910 flags |= IO_SYNC; 911 #ifdef QUOTA 912 if ((error = getinoquota(dp)) != 0) { 913 if (DOINGSOFTDEP(dvp) && newdirbp != NULL) 914 bdwrite(newdirbp); 915 return (error); 916 } 917 #endif 918 old_isize = dp->i_size; 919 vnode_pager_setsize(dvp, (u_long)I_OFFSET(dp) + DIRBLKSIZ); 920 if ((error = UFS_BALLOC(dvp, (off_t)I_OFFSET(dp), DIRBLKSIZ, 921 cr, flags, &bp)) != 0) { 922 if (DOINGSOFTDEP(dvp) && newdirbp != NULL) 923 bdwrite(newdirbp); 924 vnode_pager_setsize(dvp, (u_long)old_isize); 925 return (error); 926 } 927 dp->i_size = I_OFFSET(dp) + DIRBLKSIZ; 928 DIP_SET(dp, i_size, dp->i_size); 929 SET_I_ENDOFF(dp, dp->i_size); 930 UFS_INODE_SET_FLAG(dp, IN_SIZEMOD | IN_CHANGE | IN_UPDATE); 931 dirp->d_reclen = DIRBLKSIZ; 932 blkoff = I_OFFSET(dp) & 933 (VFSTOUFS(dvp->v_mount)->um_mountp->mnt_stat.f_iosize - 1); 934 bcopy((caddr_t)dirp, (caddr_t)bp->b_data + blkoff,newentrysize); 935 #ifdef UFS_DIRHASH 936 if (dp->i_dirhash != NULL) { 937 ufsdirhash_newblk(dp, I_OFFSET(dp)); 938 ufsdirhash_add(dp, dirp, I_OFFSET(dp)); 939 ufsdirhash_checkblock(dp, (char *)bp->b_data + blkoff, 940 I_OFFSET(dp)); 941 } 942 #endif 943 if (DOINGSOFTDEP(dvp)) { 944 /* 945 * Ensure that the entire newly allocated block is a 946 * valid directory so that future growth within the 947 * block does not have to ensure that the block is 948 * written before the inode. 949 */ 950 blkoff += DIRBLKSIZ; 951 while (blkoff < bp->b_bcount) { 952 ((struct direct *) 953 (bp->b_data + blkoff))->d_reclen = DIRBLKSIZ; 954 blkoff += DIRBLKSIZ; 955 } 956 if (softdep_setup_directory_add(bp, dp, I_OFFSET(dp), 957 dirp->d_ino, newdirbp, 1)) 958 UFS_INODE_SET_FLAG(dp, IN_NEEDSYNC); 959 if (newdirbp) 960 bdwrite(newdirbp); 961 bdwrite(bp); 962 return (UFS_UPDATE(dvp, 0)); 963 } 964 if (DOINGASYNC(dvp)) { 965 bdwrite(bp); 966 return (UFS_UPDATE(dvp, 0)); 967 } 968 error = bwrite(bp); 969 ret = UFS_UPDATE(dvp, 1); 970 if (error == 0) 971 return (ret); 972 return (error); 973 } 974 975 /* 976 * If dp->i_count is non-zero, then namei found space for the new 977 * entry in the range dp->i_offset to dp->i_offset + dp->i_count 978 * in the directory. To use this space, we may have to compact 979 * the entries located there, by copying them together towards the 980 * beginning of the block, leaving the free space in one usable 981 * chunk at the end. 982 */ 983 984 /* 985 * Increase size of directory if entry eats into new space. 986 * This should never push the size past a new multiple of 987 * DIRBLKSIZE. 988 * 989 * N.B. - THIS IS AN ARTIFACT OF 4.2 AND SHOULD NEVER HAPPEN. 990 */ 991 if (I_OFFSET(dp) + I_COUNT(dp) > dp->i_size) { 992 dp->i_size = I_OFFSET(dp) + I_COUNT(dp); 993 DIP_SET(dp, i_size, dp->i_size); 994 UFS_INODE_SET_FLAG(dp, IN_SIZEMOD | IN_MODIFIED); 995 } 996 /* 997 * Get the block containing the space for the new directory entry. 998 */ 999 error = UFS_BLKATOFF(dvp, (off_t)I_OFFSET(dp), &dirbuf, &bp); 1000 if (error) { 1001 if (DOINGSOFTDEP(dvp) && newdirbp != NULL) 1002 bdwrite(newdirbp); 1003 return (error); 1004 } 1005 /* 1006 * Find space for the new entry. In the simple case, the entry at 1007 * offset base will have the space. If it does not, then namei 1008 * arranged that compacting the region dp->i_offset to 1009 * dp->i_offset + dp->i_count would yield the space. 1010 */ 1011 ep = (struct direct *)dirbuf; 1012 dsize = ep->d_ino ? DIRSIZ(OFSFMT(dvp), ep) : 0; 1013 spacefree = ep->d_reclen - dsize; 1014 for (loc = ep->d_reclen; loc < I_COUNT(dp); ) { 1015 nep = (struct direct *)(dirbuf + loc); 1016 1017 /* Trim the existing slot (NB: dsize may be zero). */ 1018 ep->d_reclen = dsize; 1019 ep = (struct direct *)((char *)ep + dsize); 1020 1021 /* Read nep->d_reclen now as the bcopy() may clobber it. */ 1022 loc += nep->d_reclen; 1023 if (nep->d_ino == 0) { 1024 /* 1025 * A mid-block unused entry. Such entries are 1026 * never created by the kernel, but fsck_ffs 1027 * can create them (and it doesn't fix them). 1028 * 1029 * Add up the free space, and initialise the 1030 * relocated entry since we don't bcopy it. 1031 */ 1032 spacefree += nep->d_reclen; 1033 ep->d_ino = 0; 1034 dsize = 0; 1035 continue; 1036 } 1037 dsize = DIRSIZ(OFSFMT(dvp), nep); 1038 spacefree += nep->d_reclen - dsize; 1039 #ifdef UFS_DIRHASH 1040 if (dp->i_dirhash != NULL) 1041 ufsdirhash_move(dp, nep, 1042 I_OFFSET(dp) + ((char *)nep - dirbuf), 1043 I_OFFSET(dp) + ((char *)ep - dirbuf)); 1044 #endif 1045 if (DOINGSOFTDEP(dvp)) 1046 softdep_change_directoryentry_offset(bp, dp, dirbuf, 1047 (caddr_t)nep, (caddr_t)ep, dsize); 1048 else 1049 bcopy((caddr_t)nep, (caddr_t)ep, dsize); 1050 } 1051 /* 1052 * Here, `ep' points to a directory entry containing `dsize' in-use 1053 * bytes followed by `spacefree' unused bytes. If ep->d_ino == 0, 1054 * then the entry is completely unused (dsize == 0). The value 1055 * of ep->d_reclen is always indeterminate. 1056 * 1057 * Update the pointer fields in the previous entry (if any), 1058 * copy in the new entry, and write out the block. 1059 */ 1060 # if (BYTE_ORDER == LITTLE_ENDIAN) 1061 if (OFSFMT(dvp)) 1062 namlen = ep->d_type; 1063 else 1064 namlen = ep->d_namlen; 1065 # else 1066 namlen = ep->d_namlen; 1067 # endif 1068 if (ep->d_ino == 0 || 1069 (ep->d_ino == UFS_WINO && namlen == dirp->d_namlen && 1070 bcmp(ep->d_name, dirp->d_name, dirp->d_namlen) == 0)) { 1071 if (spacefree + dsize < newentrysize) 1072 panic("ufs_direnter: compact1"); 1073 dirp->d_reclen = spacefree + dsize; 1074 } else { 1075 if (spacefree < newentrysize) 1076 panic("ufs_direnter: compact2"); 1077 dirp->d_reclen = spacefree; 1078 ep->d_reclen = dsize; 1079 ep = (struct direct *)((char *)ep + dsize); 1080 } 1081 #ifdef UFS_DIRHASH 1082 if (dp->i_dirhash != NULL && (ep->d_ino == 0 || 1083 dirp->d_reclen == spacefree)) 1084 ufsdirhash_add(dp, dirp, I_OFFSET(dp) + ((char *)ep - dirbuf)); 1085 #endif 1086 bcopy((caddr_t)dirp, (caddr_t)ep, (u_int)newentrysize); 1087 #ifdef UFS_DIRHASH 1088 if (dp->i_dirhash != NULL) 1089 ufsdirhash_checkblock(dp, dirbuf - 1090 (I_OFFSET(dp) & (DIRBLKSIZ - 1)), 1091 rounddown2(I_OFFSET(dp), DIRBLKSIZ)); 1092 #endif 1093 1094 if (DOINGSOFTDEP(dvp)) { 1095 (void) softdep_setup_directory_add(bp, dp, 1096 I_OFFSET(dp) + (caddr_t)ep - dirbuf, 1097 dirp->d_ino, newdirbp, 0); 1098 if (newdirbp != NULL) 1099 bdwrite(newdirbp); 1100 bdwrite(bp); 1101 } else { 1102 if (DOINGASYNC(dvp)) { 1103 bdwrite(bp); 1104 error = 0; 1105 } else { 1106 error = bwrite(bp); 1107 } 1108 } 1109 1110 /* 1111 * If all went well, and the directory can be shortened, 1112 * mark directory inode with the truncation request. 1113 */ 1114 UFS_INODE_SET_FLAG(dp, IN_CHANGE | IN_UPDATE | (error == 0 && 1115 I_ENDOFF(dp) != 0 && I_ENDOFF(dp) < dp->i_size ? IN_ENDOFF : 0)); 1116 1117 return (error); 1118 } 1119 1120 /* 1121 * Remove a directory entry after a call to namei, using 1122 * the parameters which it left in nameidata. The entry 1123 * dp->i_offset contains the offset into the directory of the 1124 * entry to be eliminated. The dp->i_count field contains the 1125 * size of the previous record in the directory. If this 1126 * is 0, the first entry is being deleted, so we need only 1127 * zero the inode number to mark the entry as free. If the 1128 * entry is not the first in the directory, we must reclaim 1129 * the space of the now empty record by adding the record size 1130 * to the size of the previous entry. 1131 */ 1132 int 1133 ufs_dirremove(dvp, ip, flags, isrmdir) 1134 struct vnode *dvp; 1135 struct inode *ip; 1136 int flags; 1137 int isrmdir; 1138 { 1139 struct inode *dp; 1140 struct direct *ep, *rep; 1141 struct buf *bp; 1142 off_t offset; 1143 int error; 1144 1145 dp = VTOI(dvp); 1146 1147 /* 1148 * Adjust the link count early so softdep can block if necessary. 1149 */ 1150 if (ip) { 1151 ip->i_effnlink--; 1152 UFS_INODE_SET_FLAG(ip, IN_CHANGE); 1153 if (DOINGSOFTDEP(dvp)) { 1154 softdep_setup_unlink(dp, ip); 1155 } else { 1156 ip->i_nlink--; 1157 DIP_SET(ip, i_nlink, ip->i_nlink); 1158 UFS_INODE_SET_FLAG(ip, IN_CHANGE); 1159 } 1160 } 1161 if (flags & DOWHITEOUT) 1162 offset = I_OFFSET(dp); 1163 else 1164 offset = I_OFFSET(dp) - I_COUNT(dp); 1165 if ((error = UFS_BLKATOFF(dvp, offset, (char **)&ep, &bp)) != 0) { 1166 if (ip) { 1167 ip->i_effnlink++; 1168 UFS_INODE_SET_FLAG(ip, IN_CHANGE); 1169 if (DOINGSOFTDEP(dvp)) { 1170 softdep_change_linkcnt(ip); 1171 } else { 1172 ip->i_nlink++; 1173 DIP_SET(ip, i_nlink, ip->i_nlink); 1174 UFS_INODE_SET_FLAG(ip, IN_CHANGE); 1175 } 1176 } 1177 return (error); 1178 } 1179 if (flags & DOWHITEOUT) { 1180 /* 1181 * Whiteout entry: set d_ino to UFS_WINO. 1182 */ 1183 ep->d_ino = UFS_WINO; 1184 ep->d_type = DT_WHT; 1185 goto out; 1186 } 1187 /* Set 'rep' to the entry being removed. */ 1188 if (I_COUNT(dp) == 0) 1189 rep = ep; 1190 else 1191 rep = (struct direct *)((char *)ep + ep->d_reclen); 1192 #ifdef UFS_DIRHASH 1193 /* 1194 * Remove the dirhash entry. This is complicated by the fact 1195 * that `ep' is the previous entry when dp->i_count != 0. 1196 */ 1197 if (dp->i_dirhash != NULL) 1198 ufsdirhash_remove(dp, rep, I_OFFSET(dp)); 1199 #endif 1200 if (ip && rep->d_ino != ip->i_number) 1201 panic("ufs_dirremove: ip %ju does not match dirent ino %ju\n", 1202 (uintmax_t)ip->i_number, (uintmax_t)rep->d_ino); 1203 /* 1204 * Zero out the file directory entry metadata to reduce disk 1205 * scavenging disclosure. 1206 */ 1207 bzero(&rep->d_name[0], rep->d_namlen); 1208 rep->d_namlen = 0; 1209 rep->d_type = 0; 1210 rep->d_ino = 0; 1211 1212 if (I_COUNT(dp) != 0) { 1213 /* 1214 * Collapse new free space into previous entry. 1215 */ 1216 ep->d_reclen += rep->d_reclen; 1217 rep->d_reclen = 0; 1218 } 1219 #ifdef UFS_DIRHASH 1220 if (dp->i_dirhash != NULL) 1221 ufsdirhash_checkblock(dp, (char *)ep - 1222 ((I_OFFSET(dp) - I_COUNT(dp)) & (DIRBLKSIZ - 1)), 1223 rounddown2(I_OFFSET(dp), DIRBLKSIZ)); 1224 #endif 1225 out: 1226 error = 0; 1227 if (DOINGSOFTDEP(dvp)) { 1228 if (ip) 1229 softdep_setup_remove(bp, dp, ip, isrmdir); 1230 if (softdep_slowdown(dvp)) 1231 error = bwrite(bp); 1232 else 1233 bdwrite(bp); 1234 } else { 1235 if (flags & DOWHITEOUT) 1236 error = bwrite(bp); 1237 else if (DOINGASYNC(dvp)) 1238 bdwrite(bp); 1239 else 1240 error = bwrite(bp); 1241 } 1242 UFS_INODE_SET_FLAG(dp, IN_CHANGE | IN_UPDATE); 1243 /* 1244 * If the last named reference to a snapshot goes away, 1245 * drop its snapshot reference so that it will be reclaimed 1246 * when last open reference goes away. 1247 */ 1248 if (ip != NULL && (ip->i_flags & SF_SNAPSHOT) != 0 && 1249 ip->i_effnlink == 0) 1250 UFS_SNAPGONE(ip); 1251 return (error); 1252 } 1253 1254 /* 1255 * Rewrite an existing directory entry to point at the inode 1256 * supplied. The parameters describing the directory entry are 1257 * set up by a call to namei. 1258 */ 1259 int 1260 ufs_dirrewrite(dp, oip, newinum, newtype, isrmdir) 1261 struct inode *dp, *oip; 1262 ino_t newinum; 1263 int newtype; 1264 int isrmdir; 1265 { 1266 struct buf *bp; 1267 struct direct *ep; 1268 struct vnode *vdp = ITOV(dp); 1269 int error; 1270 1271 /* 1272 * Drop the link before we lock the buf so softdep can block if 1273 * necessary. 1274 */ 1275 oip->i_effnlink--; 1276 UFS_INODE_SET_FLAG(oip, IN_CHANGE); 1277 if (DOINGSOFTDEP(vdp)) { 1278 softdep_setup_unlink(dp, oip); 1279 } else { 1280 oip->i_nlink--; 1281 DIP_SET(oip, i_nlink, oip->i_nlink); 1282 UFS_INODE_SET_FLAG(oip, IN_CHANGE); 1283 } 1284 1285 error = UFS_BLKATOFF(vdp, (off_t)I_OFFSET(dp), (char **)&ep, &bp); 1286 if (error == 0 && ep->d_namlen == 2 && ep->d_name[1] == '.' && 1287 ep->d_name[0] == '.' && ep->d_ino != oip->i_number) { 1288 brelse(bp); 1289 error = EIDRM; 1290 } 1291 if (error) { 1292 oip->i_effnlink++; 1293 UFS_INODE_SET_FLAG(oip, IN_CHANGE); 1294 if (DOINGSOFTDEP(vdp)) { 1295 softdep_change_linkcnt(oip); 1296 } else { 1297 oip->i_nlink++; 1298 DIP_SET(oip, i_nlink, oip->i_nlink); 1299 UFS_INODE_SET_FLAG(oip, IN_CHANGE); 1300 } 1301 return (error); 1302 } 1303 ep->d_ino = newinum; 1304 if (!OFSFMT(vdp)) 1305 ep->d_type = newtype; 1306 if (DOINGSOFTDEP(vdp)) { 1307 softdep_setup_directory_change(bp, dp, oip, newinum, isrmdir); 1308 bdwrite(bp); 1309 } else { 1310 if (DOINGASYNC(vdp)) { 1311 bdwrite(bp); 1312 error = 0; 1313 } else { 1314 error = bwrite(bp); 1315 } 1316 } 1317 UFS_INODE_SET_FLAG(dp, IN_CHANGE | IN_UPDATE); 1318 /* 1319 * If the last named reference to a snapshot goes away, 1320 * drop its snapshot reference so that it will be reclaimed 1321 * when last open reference goes away. 1322 */ 1323 if ((oip->i_flags & SF_SNAPSHOT) != 0 && oip->i_effnlink == 0) 1324 UFS_SNAPGONE(oip); 1325 return (error); 1326 } 1327 1328 /* 1329 * Check if a directory is empty or not. 1330 * Inode supplied must be locked. 1331 * 1332 * Using a struct dirtemplate here is not precisely 1333 * what we want, but better than using a struct direct. 1334 * 1335 * NB: does not handle corrupted directories. 1336 */ 1337 int 1338 ufs_dirempty(ip, parentino, cred) 1339 struct inode *ip; 1340 ino_t parentino; 1341 struct ucred *cred; 1342 { 1343 doff_t off; 1344 struct dirtemplate dbuf; 1345 struct direct *dp = (struct direct *)&dbuf; 1346 int error, namlen; 1347 ssize_t count; 1348 #define MINDIRSIZ (sizeof (struct dirtemplate) / 2) 1349 1350 for (off = 0; off < ip->i_size; off += dp->d_reclen) { 1351 error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ, 1352 off, UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK, cred, 1353 NOCRED, &count, (struct thread *)0); 1354 /* 1355 * Since we read MINDIRSIZ, residual must 1356 * be 0 unless we're at end of file. 1357 */ 1358 if (error || count != 0) 1359 return (0); 1360 /* avoid infinite loops */ 1361 if (dp->d_reclen == 0) 1362 return (0); 1363 /* skip empty entries */ 1364 if (dp->d_ino == 0 || dp->d_ino == UFS_WINO) 1365 continue; 1366 /* accept only "." and ".." */ 1367 # if (BYTE_ORDER == LITTLE_ENDIAN) 1368 if (OFSFMT(ITOV(ip))) 1369 namlen = dp->d_type; 1370 else 1371 namlen = dp->d_namlen; 1372 # else 1373 namlen = dp->d_namlen; 1374 # endif 1375 if (namlen > 2) 1376 return (0); 1377 if (dp->d_name[0] != '.') 1378 return (0); 1379 /* 1380 * At this point namlen must be 1 or 2. 1381 * 1 implies ".", 2 implies ".." if second 1382 * char is also "." 1383 */ 1384 if (namlen == 1 && dp->d_ino == ip->i_number) 1385 continue; 1386 if (dp->d_name[1] == '.' && dp->d_ino == parentino) 1387 continue; 1388 return (0); 1389 } 1390 return (1); 1391 } 1392 1393 static int 1394 ufs_dir_dd_ino(struct vnode *vp, struct ucred *cred, ino_t *dd_ino, 1395 struct vnode **dd_vp) 1396 { 1397 struct dirtemplate dirbuf; 1398 struct vnode *ddvp; 1399 int error, namlen; 1400 1401 ASSERT_VOP_LOCKED(vp, "ufs_dir_dd_ino"); 1402 *dd_vp = NULL; 1403 if (vp->v_type != VDIR) 1404 return (ENOTDIR); 1405 /* 1406 * First check to see if we have it in the name cache. 1407 */ 1408 if ((ddvp = vn_dir_dd_ino(vp)) != NULL) { 1409 KASSERT(ddvp->v_mount == vp->v_mount, 1410 ("ufs_dir_dd_ino: Unexpected mount point crossing")); 1411 *dd_ino = VTOI(ddvp)->i_number; 1412 *dd_vp = ddvp; 1413 return (0); 1414 } 1415 /* 1416 * Have to read the directory. 1417 */ 1418 error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf, 1419 sizeof (struct dirtemplate), (off_t)0, UIO_SYSSPACE, 1420 IO_NODELOCKED | IO_NOMACCHECK, cred, NOCRED, NULL, NULL); 1421 if (error != 0) 1422 return (error); 1423 #if (BYTE_ORDER == LITTLE_ENDIAN) 1424 if (OFSFMT(vp)) 1425 namlen = dirbuf.dotdot_type; 1426 else 1427 namlen = dirbuf.dotdot_namlen; 1428 #else 1429 namlen = dirbuf.dotdot_namlen; 1430 #endif 1431 if (namlen != 2 || dirbuf.dotdot_name[0] != '.' || 1432 dirbuf.dotdot_name[1] != '.') 1433 return (ENOTDIR); 1434 *dd_ino = dirbuf.dotdot_ino; 1435 return (0); 1436 } 1437 1438 /* 1439 * Check if source directory is in the path of the target directory. 1440 */ 1441 int 1442 ufs_checkpath(ino_t source_ino, ino_t parent_ino, struct inode *target, struct ucred *cred, ino_t *wait_ino) 1443 { 1444 struct mount *mp; 1445 struct vnode *tvp, *vp, *vp1; 1446 int error; 1447 ino_t dd_ino; 1448 1449 vp = tvp = ITOV(target); 1450 mp = vp->v_mount; 1451 *wait_ino = 0; 1452 if (target->i_number == source_ino) 1453 return (EEXIST); 1454 if (target->i_number == parent_ino) 1455 return (0); 1456 if (target->i_number == UFS_ROOTINO) 1457 return (0); 1458 for (;;) { 1459 error = ufs_dir_dd_ino(vp, cred, &dd_ino, &vp1); 1460 if (error != 0) 1461 break; 1462 if (dd_ino == source_ino) { 1463 error = EINVAL; 1464 break; 1465 } 1466 if (dd_ino == UFS_ROOTINO) 1467 break; 1468 if (dd_ino == parent_ino) 1469 break; 1470 if (vp1 == NULL) { 1471 error = VFS_VGET(mp, dd_ino, LK_SHARED | LK_NOWAIT, 1472 &vp1); 1473 if (error != 0) { 1474 *wait_ino = dd_ino; 1475 break; 1476 } 1477 } 1478 KASSERT(dd_ino == VTOI(vp1)->i_number, 1479 ("directory %ju reparented\n", 1480 (uintmax_t)VTOI(vp1)->i_number)); 1481 if (vp != tvp) 1482 vput(vp); 1483 vp = vp1; 1484 } 1485 1486 if (error == ENOTDIR) 1487 panic("checkpath: .. not a directory\n"); 1488 if (vp1 != NULL) 1489 vput(vp1); 1490 if (vp != tvp) 1491 vput(vp); 1492 return (error); 1493 } 1494 1495 #ifdef DIAGNOSTIC 1496 static void 1497 ufs_assert_inode_offset_owner(struct inode *ip, struct iown_tracker *tr, 1498 const char *name, const char *file, int line) 1499 { 1500 char msg[128]; 1501 1502 snprintf(msg, sizeof(msg), "at %s@%d", file, line); 1503 ASSERT_VOP_ELOCKED(ITOV(ip), msg); 1504 MPASS((ip->i_mode & IFMT) == IFDIR); 1505 if (curthread == tr->tr_owner && ip->i_lock_gen == tr->tr_gen) 1506 return; 1507 printf("locked at\n"); 1508 stack_print(&tr->tr_st); 1509 printf("unlocked at\n"); 1510 stack_print(&tr->tr_unlock); 1511 panic("%s ip %p %jd offset owner %p %d gen %d " 1512 "curthread %p %d gen %d at %s@%d\n", 1513 name, ip, (uintmax_t)ip->i_number, tr->tr_owner, 1514 tr->tr_owner->td_tid, tr->tr_gen, 1515 curthread, curthread->td_tid, ip->i_lock_gen, 1516 file, line); 1517 } 1518 1519 static void 1520 ufs_set_inode_offset_owner(struct inode *ip, struct iown_tracker *tr, 1521 const char *file, int line) 1522 { 1523 char msg[128]; 1524 1525 snprintf(msg, sizeof(msg), "at %s@%d", file, line); 1526 ASSERT_VOP_ELOCKED(ITOV(ip), msg); 1527 MPASS((ip->i_mode & IFMT) == IFDIR); 1528 tr->tr_owner = curthread; 1529 tr->tr_gen = ip->i_lock_gen; 1530 stack_save(&tr->tr_st); 1531 } 1532 1533 static void 1534 ufs_init_one_tracker(struct iown_tracker *tr) 1535 { 1536 tr->tr_owner = NULL; 1537 stack_zero(&tr->tr_st); 1538 } 1539 1540 void 1541 ufs_init_trackers(struct inode *ip) 1542 { 1543 ufs_init_one_tracker(&ip->i_offset_tracker); 1544 ufs_init_one_tracker(&ip->i_count_tracker); 1545 ufs_init_one_tracker(&ip->i_endoff_tracker); 1546 } 1547 1548 void 1549 ufs_unlock_tracker(struct inode *ip) 1550 { 1551 if (ip->i_count_tracker.tr_gen == ip->i_lock_gen) 1552 stack_save(&ip->i_count_tracker.tr_unlock); 1553 if (ip->i_offset_tracker.tr_gen == ip->i_lock_gen) 1554 stack_save(&ip->i_offset_tracker.tr_unlock); 1555 if (ip->i_endoff_tracker.tr_gen == ip->i_lock_gen) 1556 stack_save(&ip->i_endoff_tracker.tr_unlock); 1557 ip->i_lock_gen++; 1558 } 1559 1560 doff_t 1561 ufs_get_i_offset(struct inode *ip, const char *file, int line) 1562 { 1563 ufs_assert_inode_offset_owner(ip, &ip->i_offset_tracker, "i_offset", 1564 file, line); 1565 return (ip->i_offset); 1566 } 1567 1568 void 1569 ufs_set_i_offset(struct inode *ip, doff_t off, const char *file, int line) 1570 { 1571 ufs_set_inode_offset_owner(ip, &ip->i_offset_tracker, file, line); 1572 ip->i_offset = off; 1573 } 1574 1575 int32_t 1576 ufs_get_i_count(struct inode *ip, const char *file, int line) 1577 { 1578 ufs_assert_inode_offset_owner(ip, &ip->i_count_tracker, "i_count", 1579 file, line); 1580 return (ip->i_count); 1581 } 1582 1583 void 1584 ufs_set_i_count(struct inode *ip, int32_t cnt, const char *file, int line) 1585 { 1586 ufs_set_inode_offset_owner(ip, &ip->i_count_tracker, file, line); 1587 ip->i_count = cnt; 1588 } 1589 1590 doff_t 1591 ufs_get_i_endoff(struct inode *ip, const char *file, int line) 1592 { 1593 ufs_assert_inode_offset_owner(ip, &ip->i_endoff_tracker, "i_endoff", 1594 file, line); 1595 return (ip->i_endoff); 1596 } 1597 1598 void 1599 ufs_set_i_endoff(struct inode *ip, doff_t off, const char *file, int line) 1600 { 1601 ufs_set_inode_offset_owner(ip, &ip->i_endoff_tracker, file, line); 1602 ip->i_endoff = off; 1603 } 1604 1605 #endif 1606