1 /*- 2 * modified for Lites 1.1 3 * 4 * Aug 1995, Godmar Back (gback@cs.utah.edu) 5 * University of Utah, Department of Computer Science 6 */ 7 /*- 8 * SPDX-License-Identifier: BSD-3-Clause 9 * 10 * Copyright (c) 1989, 1993 11 * The Regents of the University of California. All rights reserved. 12 * (c) UNIX System Laboratories, Inc. 13 * All or some portions of this file are derived from material licensed 14 * to the University of California by American Telephone and Telegraph 15 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 16 * the permission of UNIX System Laboratories, Inc. 17 * 18 * Redistribution and use in source and binary forms, with or without 19 * modification, are permitted provided that the following conditions 20 * are met: 21 * 1. Redistributions of source code must retain the above copyright 22 * notice, this list of conditions and the following disclaimer. 23 * 2. Redistributions in binary form must reproduce the above copyright 24 * notice, this list of conditions and the following disclaimer in the 25 * documentation and/or other materials provided with the distribution. 26 * 3. Neither the name of the University nor the names of its contributors 27 * may be used to endorse or promote products derived from this software 28 * without specific prior written permission. 29 * 30 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 31 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 32 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 33 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 34 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 35 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 36 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 37 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 38 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 39 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 40 * SUCH DAMAGE. 41 * 42 * @(#)ufs_lookup.c 8.6 (Berkeley) 4/1/94 43 * $FreeBSD$ 44 */ 45 46 #include <sys/param.h> 47 #include <sys/systm.h> 48 #include <sys/namei.h> 49 #include <sys/bio.h> 50 #include <sys/buf.h> 51 #include <sys/endian.h> 52 #include <sys/mount.h> 53 #include <sys/vnode.h> 54 #include <sys/malloc.h> 55 #include <sys/dirent.h> 56 #include <sys/sysctl.h> 57 58 #include <ufs/ufs/dir.h> 59 60 #include <fs/ext2fs/fs.h> 61 #include <fs/ext2fs/inode.h> 62 #include <fs/ext2fs/ext2_mount.h> 63 #include <fs/ext2fs/ext2fs.h> 64 #include <fs/ext2fs/ext2_dinode.h> 65 #include <fs/ext2fs/ext2_dir.h> 66 #include <fs/ext2fs/ext2_extern.h> 67 #include <fs/ext2fs/fs.h> 68 69 #ifdef INVARIANTS 70 static int dirchk = 1; 71 #else 72 static int dirchk = 0; 73 #endif 74 75 static SYSCTL_NODE(_vfs, OID_AUTO, e2fs, CTLFLAG_RD, 0, "EXT2FS filesystem"); 76 SYSCTL_INT(_vfs_e2fs, OID_AUTO, dircheck, CTLFLAG_RW, &dirchk, 0, ""); 77 78 /* 79 DIRBLKSIZE in ffs is DEV_BSIZE (in most cases 512) 80 while it is the native blocksize in ext2fs - thus, a #define 81 is no longer appropriate 82 */ 83 #undef DIRBLKSIZ 84 85 static u_char ext2_ft_to_dt[] = { 86 DT_UNKNOWN, /* EXT2_FT_UNKNOWN */ 87 DT_REG, /* EXT2_FT_REG_FILE */ 88 DT_DIR, /* EXT2_FT_DIR */ 89 DT_CHR, /* EXT2_FT_CHRDEV */ 90 DT_BLK, /* EXT2_FT_BLKDEV */ 91 DT_FIFO, /* EXT2_FT_FIFO */ 92 DT_SOCK, /* EXT2_FT_SOCK */ 93 DT_LNK, /* EXT2_FT_SYMLINK */ 94 }; 95 #define FTTODT(ft) \ 96 ((ft) < nitems(ext2_ft_to_dt) ? ext2_ft_to_dt[(ft)] : DT_UNKNOWN) 97 98 static u_char dt_to_ext2_ft[] = { 99 EXT2_FT_UNKNOWN, /* DT_UNKNOWN */ 100 EXT2_FT_FIFO, /* DT_FIFO */ 101 EXT2_FT_CHRDEV, /* DT_CHR */ 102 EXT2_FT_UNKNOWN, /* unused */ 103 EXT2_FT_DIR, /* DT_DIR */ 104 EXT2_FT_UNKNOWN, /* unused */ 105 EXT2_FT_BLKDEV, /* DT_BLK */ 106 EXT2_FT_UNKNOWN, /* unused */ 107 EXT2_FT_REG_FILE, /* DT_REG */ 108 EXT2_FT_UNKNOWN, /* unused */ 109 EXT2_FT_SYMLINK, /* DT_LNK */ 110 EXT2_FT_UNKNOWN, /* unused */ 111 EXT2_FT_SOCK, /* DT_SOCK */ 112 EXT2_FT_UNKNOWN, /* unused */ 113 EXT2_FT_UNKNOWN, /* DT_WHT */ 114 }; 115 #define DTTOFT(dt) \ 116 ((dt) < nitems(dt_to_ext2_ft) ? dt_to_ext2_ft[(dt)] : EXT2_FT_UNKNOWN) 117 118 static int ext2_dirbadentry(struct vnode *dp, struct ext2fs_direct_2 *de, 119 int entryoffsetinblock); 120 static int ext2_is_dot_entry(struct componentname *cnp); 121 static int ext2_lookup_ino(struct vnode *vdp, struct vnode **vpp, 122 struct componentname *cnp, ino_t *dd_ino); 123 124 static int 125 ext2_is_dot_entry(struct componentname *cnp) 126 { 127 if (cnp->cn_namelen <= 2 && cnp->cn_nameptr[0] == '.' && 128 (cnp->cn_nameptr[1] == '.' || cnp->cn_nameptr[1] == '\0')) 129 return (1); 130 return (0); 131 } 132 133 /* 134 * Vnode op for reading directories. 135 */ 136 int 137 ext2_readdir(struct vop_readdir_args *ap) 138 { 139 struct vnode *vp = ap->a_vp; 140 struct uio *uio = ap->a_uio; 141 struct buf *bp; 142 struct inode *ip; 143 struct ext2fs_direct_2 *dp, *edp; 144 u_long *cookies; 145 struct dirent dstdp; 146 off_t offset, startoffset; 147 size_t readcnt, skipcnt; 148 ssize_t startresid; 149 u_int ncookies; 150 int DIRBLKSIZ = VTOI(ap->a_vp)->i_e2fs->e2fs_bsize; 151 int error; 152 153 if (uio->uio_offset < 0) 154 return (EINVAL); 155 ip = VTOI(vp); 156 if (ap->a_ncookies != NULL) { 157 if (uio->uio_resid < 0) 158 ncookies = 0; 159 else 160 ncookies = uio->uio_resid; 161 if (uio->uio_offset >= ip->i_size) 162 ncookies = 0; 163 else if (ip->i_size - uio->uio_offset < ncookies) 164 ncookies = ip->i_size - uio->uio_offset; 165 ncookies = ncookies / (offsetof(struct ext2fs_direct_2, 166 e2d_namlen) + 4) + 1; 167 cookies = malloc(ncookies * sizeof(*cookies), M_TEMP, M_WAITOK); 168 *ap->a_ncookies = ncookies; 169 *ap->a_cookies = cookies; 170 } else { 171 ncookies = 0; 172 cookies = NULL; 173 } 174 offset = startoffset = uio->uio_offset; 175 startresid = uio->uio_resid; 176 error = 0; 177 while (error == 0 && uio->uio_resid > 0 && 178 uio->uio_offset < ip->i_size) { 179 error = ext2_blkatoff(vp, uio->uio_offset, NULL, &bp); 180 if (error) 181 break; 182 if (bp->b_offset + bp->b_bcount > ip->i_size) 183 readcnt = ip->i_size - bp->b_offset; 184 else 185 readcnt = bp->b_bcount; 186 skipcnt = (size_t)(uio->uio_offset - bp->b_offset) & 187 ~(size_t)(DIRBLKSIZ - 1); 188 offset = bp->b_offset + skipcnt; 189 dp = (struct ext2fs_direct_2 *)&bp->b_data[skipcnt]; 190 edp = (struct ext2fs_direct_2 *)&bp->b_data[readcnt]; 191 while (error == 0 && uio->uio_resid > 0 && dp < edp) { 192 if (dp->e2d_reclen <= offsetof(struct ext2fs_direct_2, 193 e2d_namlen) || (caddr_t)dp + dp->e2d_reclen > 194 (caddr_t)edp) { 195 error = EIO; 196 break; 197 } 198 /*- 199 * "New" ext2fs directory entries differ in 3 ways 200 * from ufs on-disk ones: 201 * - the name is not necessarily NUL-terminated. 202 * - the file type field always exists and always 203 * follows the name length field. 204 * - the file type is encoded in a different way. 205 * 206 * "Old" ext2fs directory entries need no special 207 * conversions, since they are binary compatible 208 * with "new" entries having a file type of 0 (i.e., 209 * EXT2_FT_UNKNOWN). Splitting the old name length 210 * field didn't make a mess like it did in ufs, 211 * because ext2fs uses a machine-independent disk 212 * layout. 213 */ 214 dstdp.d_namlen = dp->e2d_namlen; 215 dstdp.d_type = FTTODT(dp->e2d_type); 216 if (offsetof(struct ext2fs_direct_2, e2d_namlen) + 217 dstdp.d_namlen > dp->e2d_reclen) { 218 error = EIO; 219 break; 220 } 221 if (offset < startoffset || dp->e2d_ino == 0) 222 goto nextentry; 223 dstdp.d_fileno = dp->e2d_ino; 224 dstdp.d_reclen = GENERIC_DIRSIZ(&dstdp); 225 bcopy(dp->e2d_name, dstdp.d_name, dstdp.d_namlen); 226 dstdp.d_name[dstdp.d_namlen] = '\0'; 227 if (dstdp.d_reclen > uio->uio_resid) { 228 if (uio->uio_resid == startresid) 229 error = EINVAL; 230 else 231 error = EJUSTRETURN; 232 break; 233 } 234 /* Advance dp. */ 235 error = uiomove((caddr_t)&dstdp, dstdp.d_reclen, uio); 236 if (error) 237 break; 238 if (cookies != NULL) { 239 KASSERT(ncookies > 0, 240 ("ext2_readdir: cookies buffer too small")); 241 *cookies = offset + dp->e2d_reclen; 242 cookies++; 243 ncookies--; 244 } 245 nextentry: 246 offset += dp->e2d_reclen; 247 dp = (struct ext2fs_direct_2 *)((caddr_t)dp + 248 dp->e2d_reclen); 249 } 250 bqrelse(bp); 251 uio->uio_offset = offset; 252 } 253 /* We need to correct uio_offset. */ 254 uio->uio_offset = offset; 255 if (error == EJUSTRETURN) 256 error = 0; 257 if (ap->a_ncookies != NULL) { 258 if (error == 0) { 259 ap->a_ncookies -= ncookies; 260 } else { 261 free(*ap->a_cookies, M_TEMP); 262 *ap->a_ncookies = 0; 263 *ap->a_cookies = NULL; 264 } 265 } 266 if (error == 0 && ap->a_eofflag) 267 *ap->a_eofflag = ip->i_size <= uio->uio_offset; 268 return (error); 269 } 270 271 /* 272 * Convert a component of a pathname into a pointer to a locked inode. 273 * This is a very central and rather complicated routine. 274 * If the file system is not maintained in a strict tree hierarchy, 275 * this can result in a deadlock situation (see comments in code below). 276 * 277 * The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending 278 * on whether the name is to be looked up, created, renamed, or deleted. 279 * When CREATE, RENAME, or DELETE is specified, information usable in 280 * creating, renaming, or deleting a directory entry may be calculated. 281 * If flag has LOCKPARENT or'ed into it and the target of the pathname 282 * exists, lookup returns both the target and its parent directory locked. 283 * When creating or renaming and LOCKPARENT is specified, the target may 284 * not be ".". When deleting and LOCKPARENT is specified, the target may 285 * be "."., but the caller must check to ensure it does an vrele and vput 286 * instead of two vputs. 287 * 288 * Overall outline of ext2_lookup: 289 * 290 * search for name in directory, to found or notfound 291 * notfound: 292 * if creating, return locked directory, leaving info on available slots 293 * else return error 294 * found: 295 * if at end of path and deleting, return information to allow delete 296 * if at end of path and rewriting (RENAME and LOCKPARENT), lock target 297 * inode and return info to allow rewrite 298 * if not at end, add name to cache; if at end and neither creating 299 * nor deleting, add name to cache 300 */ 301 int 302 ext2_lookup(struct vop_cachedlookup_args *ap) 303 { 304 305 return (ext2_lookup_ino(ap->a_dvp, ap->a_vpp, ap->a_cnp, NULL)); 306 } 307 308 static int 309 ext2_lookup_ino(struct vnode *vdp, struct vnode **vpp, struct componentname *cnp, 310 ino_t *dd_ino) 311 { 312 struct inode *dp; /* inode for directory being searched */ 313 struct buf *bp; /* a buffer of directory entries */ 314 struct ext2fs_direct_2 *ep; /* the current directory entry */ 315 int entryoffsetinblock; /* offset of ep in bp's buffer */ 316 struct ext2fs_searchslot ss; 317 doff_t i_diroff; /* cached i_diroff value */ 318 doff_t i_offset; /* cached i_offset value */ 319 int numdirpasses; /* strategy for directory search */ 320 doff_t endsearch; /* offset to end directory search */ 321 doff_t prevoff; /* prev entry dp->i_offset */ 322 struct vnode *pdp; /* saved dp during symlink work */ 323 struct vnode *tdp; /* returned by VFS_VGET */ 324 doff_t enduseful; /* pointer past last used dir slot */ 325 u_long bmask; /* block offset mask */ 326 int error; 327 struct ucred *cred = cnp->cn_cred; 328 int flags = cnp->cn_flags; 329 int nameiop = cnp->cn_nameiop; 330 ino_t ino, ino1; 331 int ltype; 332 int entry_found = 0; 333 334 int DIRBLKSIZ = VTOI(vdp)->i_e2fs->e2fs_bsize; 335 336 if (vpp != NULL) 337 *vpp = NULL; 338 339 dp = VTOI(vdp); 340 bmask = VFSTOEXT2(vdp->v_mount)->um_mountp->mnt_stat.f_iosize - 1; 341 restart: 342 bp = NULL; 343 ss.slotoffset = -1; 344 345 /* 346 * We now have a segment name to search for, and a directory to search. 347 * 348 * Suppress search for slots unless creating 349 * file and at end of pathname, in which case 350 * we watch for a place to put the new file in 351 * case it doesn't already exist. 352 */ 353 i_diroff = dp->i_diroff; 354 ss.slotstatus = FOUND; 355 ss.slotfreespace = ss.slotsize = ss.slotneeded = 0; 356 if ((nameiop == CREATE || nameiop == RENAME) && 357 (flags & ISLASTCN)) { 358 ss.slotstatus = NONE; 359 ss.slotneeded = EXT2_DIR_REC_LEN(cnp->cn_namelen); 360 /* 361 * was ss.slotneeded = (sizeof(struct direct) - MAXNAMLEN + 362 * cnp->cn_namelen + 3) &~ 3; 363 */ 364 } 365 /* 366 * Try to lookup dir entry using htree directory index. 367 * 368 * If we got an error or we want to find '.' or '..' entry, 369 * we will fall back to linear search. 370 */ 371 if (!ext2_is_dot_entry(cnp) && ext2_htree_has_idx(dp)) { 372 numdirpasses = 1; 373 entryoffsetinblock = 0; 374 switch (ext2_htree_lookup(dp, cnp->cn_nameptr, cnp->cn_namelen, 375 &bp, &entryoffsetinblock, &i_offset, &prevoff, 376 &enduseful, &ss)) { 377 case 0: 378 ep = (struct ext2fs_direct_2 *)((char *)bp->b_data + 379 (i_offset & bmask)); 380 goto foundentry; 381 case ENOENT: 382 i_offset = roundup2(dp->i_size, DIRBLKSIZ); 383 goto notfound; 384 default: 385 /* 386 * Something failed; just fallback to do a linear 387 * search. 388 */ 389 break; 390 } 391 } 392 393 /* 394 * If there is cached information on a previous search of 395 * this directory, pick up where we last left off. 396 * We cache only lookups as these are the most common 397 * and have the greatest payoff. Caching CREATE has little 398 * benefit as it usually must search the entire directory 399 * to determine that the entry does not exist. Caching the 400 * location of the last DELETE or RENAME has not reduced 401 * profiling time and hence has been removed in the interest 402 * of simplicity. 403 */ 404 if (nameiop != LOOKUP || i_diroff == 0 || 405 i_diroff > dp->i_size) { 406 entryoffsetinblock = 0; 407 i_offset = 0; 408 numdirpasses = 1; 409 } else { 410 i_offset = i_diroff; 411 if ((entryoffsetinblock = i_offset & bmask) && 412 (error = ext2_blkatoff(vdp, (off_t)i_offset, NULL, 413 &bp))) 414 return (error); 415 numdirpasses = 2; 416 nchstats.ncs_2passes++; 417 } 418 prevoff = i_offset; 419 endsearch = roundup2(dp->i_size, DIRBLKSIZ); 420 enduseful = 0; 421 422 searchloop: 423 while (i_offset < endsearch) { 424 /* 425 * If necessary, get the next directory block. 426 */ 427 if (bp != NULL) 428 brelse(bp); 429 error = ext2_blkatoff(vdp, (off_t)i_offset, NULL, &bp); 430 if (error != 0) 431 return (error); 432 entryoffsetinblock = 0; 433 /* 434 * If still looking for a slot, and at a DIRBLKSIZE 435 * boundary, have to start looking for free space again. 436 */ 437 if (ss.slotstatus == NONE && 438 (entryoffsetinblock & (DIRBLKSIZ - 1)) == 0) { 439 ss.slotoffset = -1; 440 ss.slotfreespace = 0; 441 } 442 error = ext2_search_dirblock(dp, bp->b_data, &entry_found, 443 cnp->cn_nameptr, cnp->cn_namelen, 444 &entryoffsetinblock, &i_offset, &prevoff, 445 &enduseful, &ss); 446 if (error != 0) { 447 brelse(bp); 448 return (error); 449 } 450 if (entry_found) { 451 ep = (struct ext2fs_direct_2 *)((char *)bp->b_data + 452 (entryoffsetinblock & bmask)); 453 foundentry: 454 ino = ep->e2d_ino; 455 goto found; 456 } 457 } 458 notfound: 459 /* 460 * If we started in the middle of the directory and failed 461 * to find our target, we must check the beginning as well. 462 */ 463 if (numdirpasses == 2) { 464 numdirpasses--; 465 i_offset = 0; 466 endsearch = i_diroff; 467 goto searchloop; 468 } 469 if (bp != NULL) 470 brelse(bp); 471 /* 472 * If creating, and at end of pathname and current 473 * directory has not been removed, then can consider 474 * allowing file to be created. 475 */ 476 if ((nameiop == CREATE || nameiop == RENAME) && 477 (flags & ISLASTCN) && dp->i_nlink != 0) { 478 /* 479 * Access for write is interpreted as allowing 480 * creation of files in the directory. 481 */ 482 if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread)) != 0) 483 return (error); 484 /* 485 * Return an indication of where the new directory 486 * entry should be put. If we didn't find a slot, 487 * then set dp->i_count to 0 indicating 488 * that the new slot belongs at the end of the 489 * directory. If we found a slot, then the new entry 490 * can be put in the range from dp->i_offset to 491 * dp->i_offset + dp->i_count. 492 */ 493 if (ss.slotstatus == NONE) { 494 dp->i_offset = roundup2(dp->i_size, DIRBLKSIZ); 495 dp->i_count = 0; 496 enduseful = dp->i_offset; 497 } else { 498 dp->i_offset = ss.slotoffset; 499 dp->i_count = ss.slotsize; 500 if (enduseful < ss.slotoffset + ss.slotsize) 501 enduseful = ss.slotoffset + ss.slotsize; 502 } 503 dp->i_endoff = roundup2(enduseful, DIRBLKSIZ); 504 /* 505 * We return with the directory locked, so that 506 * the parameters we set up above will still be 507 * valid if we actually decide to do a direnter(). 508 * We return ni_vp == NULL to indicate that the entry 509 * does not currently exist; we leave a pointer to 510 * the (locked) directory inode in ndp->ni_dvp. 511 * The pathname buffer is saved so that the name 512 * can be obtained later. 513 * 514 * NB - if the directory is unlocked, then this 515 * information cannot be used. 516 */ 517 cnp->cn_flags |= SAVENAME; 518 return (EJUSTRETURN); 519 } 520 /* 521 * Insert name into cache (as non-existent) if appropriate. 522 */ 523 if ((cnp->cn_flags & MAKEENTRY) != 0) 524 cache_enter(vdp, NULL, cnp); 525 return (ENOENT); 526 527 found: 528 if (dd_ino != NULL) 529 *dd_ino = ino; 530 if (numdirpasses == 2) 531 nchstats.ncs_pass2++; 532 /* 533 * Check that directory length properly reflects presence 534 * of this entry. 535 */ 536 if (entryoffsetinblock + EXT2_DIR_REC_LEN(ep->e2d_namlen) 537 > dp->i_size) { 538 ext2_dirbad(dp, i_offset, "i_size too small"); 539 dp->i_size = entryoffsetinblock + EXT2_DIR_REC_LEN(ep->e2d_namlen); 540 dp->i_flag |= IN_CHANGE | IN_UPDATE; 541 } 542 brelse(bp); 543 544 /* 545 * Found component in pathname. 546 * If the final component of path name, save information 547 * in the cache as to where the entry was found. 548 */ 549 if ((flags & ISLASTCN) && nameiop == LOOKUP) 550 dp->i_diroff = rounddown2(i_offset, DIRBLKSIZ); 551 /* 552 * If deleting, and at end of pathname, return 553 * parameters which can be used to remove file. 554 */ 555 if (nameiop == DELETE && (flags & ISLASTCN)) { 556 if (flags & LOCKPARENT) 557 ASSERT_VOP_ELOCKED(vdp, __FUNCTION__); 558 /* 559 * Write access to directory required to delete files. 560 */ 561 if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread)) != 0) 562 return (error); 563 /* 564 * Return pointer to current entry in dp->i_offset, 565 * and distance past previous entry (if there 566 * is a previous entry in this block) in dp->i_count. 567 * Save directory inode pointer in ndp->ni_dvp for dirremove(). 568 * 569 * Technically we shouldn't be setting these in the 570 * WANTPARENT case (first lookup in rename()), but any 571 * lookups that will result in directory changes will 572 * overwrite these. 573 */ 574 dp->i_offset = i_offset; 575 if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0) 576 dp->i_count = 0; 577 else 578 dp->i_count = dp->i_offset - prevoff; 579 if (dd_ino != NULL) 580 return (0); 581 if (dp->i_number == ino) { 582 VREF(vdp); 583 *vpp = vdp; 584 return (0); 585 } 586 if ((error = VFS_VGET(vdp->v_mount, ino, LK_EXCLUSIVE, 587 &tdp)) != 0) 588 return (error); 589 /* 590 * If directory is "sticky", then user must own 591 * the directory, or the file in it, else she 592 * may not delete it (unless she's root). This 593 * implements append-only directories. 594 */ 595 if ((dp->i_mode & ISVTX) && 596 cred->cr_uid != 0 && 597 cred->cr_uid != dp->i_uid && 598 VTOI(tdp)->i_uid != cred->cr_uid) { 599 vput(tdp); 600 return (EPERM); 601 } 602 *vpp = tdp; 603 return (0); 604 } 605 606 /* 607 * If rewriting (RENAME), return the inode and the 608 * information required to rewrite the present directory 609 * Must get inode of directory entry to verify it's a 610 * regular file, or empty directory. 611 */ 612 if (nameiop == RENAME && (flags & ISLASTCN)) { 613 if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread)) != 0) 614 return (error); 615 /* 616 * Careful about locking second inode. 617 * This can only occur if the target is ".". 618 */ 619 dp->i_offset = i_offset; 620 if (dp->i_number == ino) 621 return (EISDIR); 622 if (dd_ino != NULL) 623 return (0); 624 if ((error = VFS_VGET(vdp->v_mount, ino, LK_EXCLUSIVE, 625 &tdp)) != 0) 626 return (error); 627 *vpp = tdp; 628 cnp->cn_flags |= SAVENAME; 629 return (0); 630 } 631 if (dd_ino != NULL) 632 return (0); 633 634 /* 635 * Step through the translation in the name. We do not `vput' the 636 * directory because we may need it again if a symbolic link 637 * is relative to the current directory. Instead we save it 638 * unlocked as "pdp". We must get the target inode before unlocking 639 * the directory to insure that the inode will not be removed 640 * before we get it. We prevent deadlock by always fetching 641 * inodes from the root, moving down the directory tree. Thus 642 * when following backward pointers ".." we must unlock the 643 * parent directory before getting the requested directory. 644 * There is a potential race condition here if both the current 645 * and parent directories are removed before the VFS_VGET for the 646 * inode associated with ".." returns. We hope that this occurs 647 * infrequently since we cannot avoid this race condition without 648 * implementing a sophisticated deadlock detection algorithm. 649 * Note also that this simple deadlock detection scheme will not 650 * work if the file system has any hard links other than ".." 651 * that point backwards in the directory structure. 652 */ 653 pdp = vdp; 654 if (flags & ISDOTDOT) { 655 error = vn_vget_ino(pdp, ino, cnp->cn_lkflags, &tdp); 656 if (pdp->v_iflag & VI_DOOMED) { 657 if (error == 0) 658 vput(tdp); 659 error = ENOENT; 660 } 661 if (error) 662 return (error); 663 /* 664 * Recheck that ".." entry in the vdp directory points 665 * to the inode we looked up before vdp lock was 666 * dropped. 667 */ 668 error = ext2_lookup_ino(pdp, NULL, cnp, &ino1); 669 if (error) { 670 vput(tdp); 671 return (error); 672 } 673 if (ino1 != ino) { 674 vput(tdp); 675 goto restart; 676 } 677 *vpp = tdp; 678 } else if (dp->i_number == ino) { 679 VREF(vdp); /* we want ourself, ie "." */ 680 /* 681 * When we lookup "." we still can be asked to lock it 682 * differently. 683 */ 684 ltype = cnp->cn_lkflags & LK_TYPE_MASK; 685 if (ltype != VOP_ISLOCKED(vdp)) { 686 if (ltype == LK_EXCLUSIVE) 687 vn_lock(vdp, LK_UPGRADE | LK_RETRY); 688 else /* if (ltype == LK_SHARED) */ 689 vn_lock(vdp, LK_DOWNGRADE | LK_RETRY); 690 } 691 *vpp = vdp; 692 } else { 693 if ((error = VFS_VGET(vdp->v_mount, ino, cnp->cn_lkflags, 694 &tdp)) != 0) 695 return (error); 696 *vpp = tdp; 697 } 698 699 /* 700 * Insert name into cache if appropriate. 701 */ 702 if (cnp->cn_flags & MAKEENTRY) 703 cache_enter(vdp, *vpp, cnp); 704 return (0); 705 } 706 707 int 708 ext2_search_dirblock(struct inode *ip, void *data, int *foundp, 709 const char *name, int namelen, int *entryoffsetinblockp, 710 doff_t *offp, doff_t *prevoffp, doff_t *endusefulp, 711 struct ext2fs_searchslot *ssp) 712 { 713 struct vnode *vdp; 714 struct ext2fs_direct_2 *ep, *top; 715 uint32_t bsize = ip->i_e2fs->e2fs_bsize; 716 int offset = *entryoffsetinblockp; 717 int namlen; 718 719 vdp = ITOV(ip); 720 721 ep = (struct ext2fs_direct_2 *)((char *)data + offset); 722 top = (struct ext2fs_direct_2 *)((char *)data + 723 bsize - EXT2_DIR_REC_LEN(0)); 724 725 while (ep < top) { 726 /* 727 * Full validation checks are slow, so we only check 728 * enough to insure forward progress through the 729 * directory. Complete checks can be run by setting 730 * "vfs.e2fs.dirchk" to be true. 731 */ 732 if (ep->e2d_reclen == 0 || 733 (dirchk && ext2_dirbadentry(vdp, ep, offset))) { 734 int i; 735 736 ext2_dirbad(ip, *offp, "mangled entry"); 737 i = bsize - (offset & (bsize - 1)); 738 *offp += i; 739 offset += i; 740 continue; 741 } 742 743 /* 744 * If an appropriate sized slot has not yet been found, 745 * check to see if one is available. Also accumulate space 746 * in the current block so that we can determine if 747 * compaction is viable. 748 */ 749 if (ssp->slotstatus != FOUND) { 750 int size = ep->e2d_reclen; 751 752 if (ep->e2d_ino != 0) 753 size -= EXT2_DIR_REC_LEN(ep->e2d_namlen); 754 if (size > 0) { 755 if (size >= ssp->slotneeded) { 756 ssp->slotstatus = FOUND; 757 ssp->slotoffset = *offp; 758 ssp->slotsize = ep->e2d_reclen; 759 } else if (ssp->slotstatus == NONE) { 760 ssp->slotfreespace += size; 761 if (ssp->slotoffset == -1) 762 ssp->slotoffset = *offp; 763 if (ssp->slotfreespace >= ssp->slotneeded) { 764 ssp->slotstatus = COMPACT; 765 ssp->slotsize = *offp + 766 ep->e2d_reclen - 767 ssp->slotoffset; 768 } 769 } 770 } 771 } 772 /* 773 * Check for a name match. 774 */ 775 if (ep->e2d_ino) { 776 namlen = ep->e2d_namlen; 777 if (namlen == namelen && 778 !bcmp(name, ep->e2d_name, (unsigned)namlen)) { 779 /* 780 * Save directory entry's inode number and 781 * reclen in ndp->ni_ufs area, and release 782 * directory buffer. 783 */ 784 *foundp = 1; 785 return (0); 786 } 787 } 788 *prevoffp = *offp; 789 *offp += ep->e2d_reclen; 790 offset += ep->e2d_reclen; 791 *entryoffsetinblockp = offset; 792 if (ep->e2d_ino) 793 *endusefulp = *offp; 794 /* 795 * Get pointer to the next entry. 796 */ 797 ep = (struct ext2fs_direct_2 *)((char *)data + offset); 798 } 799 800 return (0); 801 } 802 803 void 804 ext2_dirbad(struct inode *ip, doff_t offset, char *how) 805 { 806 struct mount *mp; 807 808 mp = ITOV(ip)->v_mount; 809 if ((mp->mnt_flag & MNT_RDONLY) == 0) 810 panic("ext2_dirbad: %s: bad dir ino %ju at offset %ld: %s\n", 811 mp->mnt_stat.f_mntonname, (uintmax_t)ip->i_number, 812 (long)offset, how); 813 else 814 (void)printf("%s: bad dir ino %ju at offset %ld: %s\n", 815 mp->mnt_stat.f_mntonname, (uintmax_t)ip->i_number, 816 (long)offset, how); 817 818 } 819 820 /* 821 * Do consistency checking on a directory entry: 822 * record length must be multiple of 4 823 * entry must fit in rest of its DIRBLKSIZ block 824 * record must be large enough to contain entry 825 * name is not longer than MAXNAMLEN 826 * name must be as long as advertised, and null terminated 827 */ 828 /* 829 * changed so that it confirms to ext2_check_dir_entry 830 */ 831 static int 832 ext2_dirbadentry(struct vnode *dp, struct ext2fs_direct_2 *de, 833 int entryoffsetinblock) 834 { 835 int DIRBLKSIZ = VTOI(dp)->i_e2fs->e2fs_bsize; 836 837 char *error_msg = NULL; 838 839 if (de->e2d_reclen < EXT2_DIR_REC_LEN(1)) 840 error_msg = "rec_len is smaller than minimal"; 841 else if (de->e2d_reclen % 4 != 0) 842 error_msg = "rec_len % 4 != 0"; 843 else if (de->e2d_reclen < EXT2_DIR_REC_LEN(de->e2d_namlen)) 844 error_msg = "reclen is too small for name_len"; 845 else if (entryoffsetinblock + de->e2d_reclen > DIRBLKSIZ) 846 error_msg = "directory entry across blocks"; 847 /* else LATER 848 if (de->inode > dir->i_sb->u.ext2_sb.s_es->s_inodes_count) 849 error_msg = "inode out of bounds"; 850 */ 851 852 if (error_msg != NULL) { 853 printf("bad directory entry: %s\n", error_msg); 854 printf("offset=%d, inode=%lu, rec_len=%u, name_len=%u\n", 855 entryoffsetinblock, (unsigned long)de->e2d_ino, 856 de->e2d_reclen, de->e2d_namlen); 857 } 858 return error_msg == NULL ? 0 : 1; 859 } 860 861 /* 862 * Write a directory entry after a call to namei, using the parameters 863 * that it left in nameidata. The argument ip is the inode which the new 864 * directory entry will refer to. Dvp is a pointer to the directory to 865 * be written, which was left locked by namei. Remaining parameters 866 * (dp->i_offset, dp->i_count) indicate how the space for the new 867 * entry is to be obtained. 868 */ 869 int 870 ext2_direnter(struct inode *ip, struct vnode *dvp, struct componentname *cnp) 871 { 872 struct inode *dp; 873 struct ext2fs_direct_2 newdir; 874 struct buf *bp; 875 int DIRBLKSIZ = ip->i_e2fs->e2fs_bsize; 876 int error; 877 878 879 #ifdef INVARIANTS 880 if ((cnp->cn_flags & SAVENAME) == 0) 881 panic("ext2_direnter: missing name"); 882 #endif 883 dp = VTOI(dvp); 884 newdir.e2d_ino = ip->i_number; 885 newdir.e2d_namlen = cnp->cn_namelen; 886 if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs, 887 EXT2F_INCOMPAT_FTYPE)) 888 newdir.e2d_type = DTTOFT(IFTODT(ip->i_mode)); 889 else 890 newdir.e2d_type = EXT2_FT_UNKNOWN; 891 bcopy(cnp->cn_nameptr, newdir.e2d_name, (unsigned)cnp->cn_namelen + 1); 892 893 if (ext2_htree_has_idx(dp)) { 894 error = ext2_htree_add_entry(dvp, &newdir, cnp); 895 if (error) { 896 dp->i_flag &= ~IN_E3INDEX; 897 dp->i_flag |= IN_CHANGE | IN_UPDATE; 898 } 899 return (error); 900 } 901 902 if (EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_DIRHASHINDEX) && 903 !ext2_htree_has_idx(dp)) { 904 if ((dp->i_size / DIRBLKSIZ) == 1 && 905 dp->i_offset == DIRBLKSIZ) { 906 /* 907 * Making indexed directory when one block is not 908 * enough to save all entries. 909 */ 910 return ext2_htree_create_index(dvp, cnp, &newdir); 911 } 912 } 913 914 if (dp->i_count == 0) { 915 /* 916 * If dp->i_count is 0, then namei could find no 917 * space in the directory. Here, dp->i_offset will 918 * be on a directory block boundary and we will write the 919 * new entry into a fresh block. 920 */ 921 if (dp->i_offset & (DIRBLKSIZ - 1)) 922 panic("ext2_direnter: newblk"); 923 924 newdir.e2d_reclen = DIRBLKSIZ; 925 926 bp = getblk(ip->i_devvp, lblkno(dp->i_e2fs, dp->i_offset), 927 DIRBLKSIZ, 0, 0, 0); 928 if (!bp) 929 return (EIO); 930 931 memcpy(bp->b_data, &newdir, sizeof(struct ext2fs_direct_2)); 932 933 ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)bp->b_data); 934 error = bwrite(bp); 935 if (error) 936 return (error); 937 938 dp->i_size = roundup2(dp->i_size, DIRBLKSIZ); 939 dp->i_flag |= IN_CHANGE; 940 941 return (0); 942 } 943 944 error = ext2_add_entry(dvp, &newdir); 945 if (!error && dp->i_endoff && dp->i_endoff < dp->i_size) 946 error = ext2_truncate(dvp, (off_t)dp->i_endoff, IO_SYNC, 947 cnp->cn_cred, cnp->cn_thread); 948 return (error); 949 } 950 951 /* 952 * Insert an entry into the directory block. 953 * Compact the contents. 954 */ 955 int 956 ext2_add_entry(struct vnode *dvp, struct ext2fs_direct_2 *entry) 957 { 958 struct ext2fs_direct_2 *ep, *nep; 959 struct inode *dp; 960 struct buf *bp; 961 u_int dsize; 962 int error, loc, newentrysize, spacefree; 963 char *dirbuf; 964 965 dp = VTOI(dvp); 966 967 /* 968 * If dp->i_count is non-zero, then namei found space 969 * for the new entry in the range dp->i_offset to 970 * dp->i_offset + dp->i_count in the directory. 971 * To use this space, we may have to compact the entries located 972 * there, by copying them together towards the beginning of the 973 * block, leaving the free space in one usable chunk at the end. 974 */ 975 976 /* 977 * Increase size of directory if entry eats into new space. 978 * This should never push the size past a new multiple of 979 * DIRBLKSIZE. 980 * 981 * N.B. - THIS IS AN ARTIFACT OF 4.2 AND SHOULD NEVER HAPPEN. 982 */ 983 if (dp->i_offset + dp->i_count > dp->i_size) 984 dp->i_size = dp->i_offset + dp->i_count; 985 /* 986 * Get the block containing the space for the new directory entry. 987 */ 988 if ((error = ext2_blkatoff(dvp, (off_t)dp->i_offset, &dirbuf, 989 &bp)) != 0) 990 return (error); 991 /* 992 * Find space for the new entry. In the simple case, the entry at 993 * offset base will have the space. If it does not, then namei 994 * arranged that compacting the region dp->i_offset to 995 * dp->i_offset + dp->i_count would yield the 996 * space. 997 */ 998 newentrysize = EXT2_DIR_REC_LEN(entry->e2d_namlen); 999 ep = (struct ext2fs_direct_2 *)dirbuf; 1000 dsize = EXT2_DIR_REC_LEN(ep->e2d_namlen); 1001 spacefree = ep->e2d_reclen - dsize; 1002 for (loc = ep->e2d_reclen; loc < dp->i_count; ) { 1003 nep = (struct ext2fs_direct_2 *)(dirbuf + loc); 1004 if (ep->e2d_ino) { 1005 /* trim the existing slot */ 1006 ep->e2d_reclen = dsize; 1007 ep = (struct ext2fs_direct_2 *)((char *)ep + dsize); 1008 } else { 1009 /* overwrite; nothing there; header is ours */ 1010 spacefree += dsize; 1011 } 1012 dsize = EXT2_DIR_REC_LEN(nep->e2d_namlen); 1013 spacefree += nep->e2d_reclen - dsize; 1014 loc += nep->e2d_reclen; 1015 bcopy((caddr_t)nep, (caddr_t)ep, dsize); 1016 } 1017 /* 1018 * Update the pointer fields in the previous entry (if any), 1019 * copy in the new entry, and write out the block. 1020 */ 1021 if (ep->e2d_ino == 0) { 1022 if (spacefree + dsize < newentrysize) 1023 panic("ext2_direnter: compact1"); 1024 entry->e2d_reclen = spacefree + dsize; 1025 } else { 1026 if (spacefree < newentrysize) 1027 panic("ext2_direnter: compact2"); 1028 entry->e2d_reclen = spacefree; 1029 ep->e2d_reclen = dsize; 1030 ep = (struct ext2fs_direct_2 *)((char *)ep + dsize); 1031 } 1032 bcopy((caddr_t)entry, (caddr_t)ep, (u_int)newentrysize); 1033 ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)bp->b_data); 1034 if (DOINGASYNC(dvp)) { 1035 bdwrite(bp); 1036 error = 0; 1037 } else { 1038 error = bwrite(bp); 1039 } 1040 dp->i_flag |= IN_CHANGE | IN_UPDATE; 1041 return (error); 1042 } 1043 1044 /* 1045 * Remove a directory entry after a call to namei, using 1046 * the parameters which it left in nameidata. The entry 1047 * dp->i_offset contains the offset into the directory of the 1048 * entry to be eliminated. The dp->i_count field contains the 1049 * size of the previous record in the directory. If this 1050 * is 0, the first entry is being deleted, so we need only 1051 * zero the inode number to mark the entry as free. If the 1052 * entry is not the first in the directory, we must reclaim 1053 * the space of the now empty record by adding the record size 1054 * to the size of the previous entry. 1055 */ 1056 int 1057 ext2_dirremove(struct vnode *dvp, struct componentname *cnp) 1058 { 1059 struct inode *dp; 1060 struct ext2fs_direct_2 *ep, *rep; 1061 struct buf *bp; 1062 int error; 1063 1064 dp = VTOI(dvp); 1065 if (dp->i_count == 0) { 1066 /* 1067 * First entry in block: set d_ino to zero. 1068 */ 1069 if ((error = 1070 ext2_blkatoff(dvp, (off_t)dp->i_offset, (char **)&ep, 1071 &bp)) != 0) 1072 return (error); 1073 ep->e2d_ino = 0; 1074 error = bwrite(bp); 1075 dp->i_flag |= IN_CHANGE | IN_UPDATE; 1076 return (error); 1077 } 1078 /* 1079 * Collapse new free space into previous entry. 1080 */ 1081 if ((error = ext2_blkatoff(dvp, (off_t)(dp->i_offset - dp->i_count), 1082 (char **)&ep, &bp)) != 0) 1083 return (error); 1084 1085 /* Set 'rep' to the entry being removed. */ 1086 if (dp->i_count == 0) 1087 rep = ep; 1088 else 1089 rep = (struct ext2fs_direct_2 *)((char *)ep + ep->e2d_reclen); 1090 ep->e2d_reclen += rep->e2d_reclen; 1091 ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)bp->b_data); 1092 if (DOINGASYNC(dvp) && dp->i_count != 0) 1093 bdwrite(bp); 1094 else 1095 error = bwrite(bp); 1096 dp->i_flag |= IN_CHANGE | IN_UPDATE; 1097 return (error); 1098 } 1099 1100 /* 1101 * Rewrite an existing directory entry to point at the inode 1102 * supplied. The parameters describing the directory entry are 1103 * set up by a call to namei. 1104 */ 1105 int 1106 ext2_dirrewrite(struct inode *dp, struct inode *ip, struct componentname *cnp) 1107 { 1108 struct buf *bp; 1109 struct ext2fs_direct_2 *ep; 1110 struct vnode *vdp = ITOV(dp); 1111 int error; 1112 1113 if ((error = ext2_blkatoff(vdp, (off_t)dp->i_offset, (char **)&ep, 1114 &bp)) != 0) 1115 return (error); 1116 ep->e2d_ino = ip->i_number; 1117 if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs, 1118 EXT2F_INCOMPAT_FTYPE)) 1119 ep->e2d_type = DTTOFT(IFTODT(ip->i_mode)); 1120 else 1121 ep->e2d_type = EXT2_FT_UNKNOWN; 1122 ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)bp->b_data); 1123 error = bwrite(bp); 1124 dp->i_flag |= IN_CHANGE | IN_UPDATE; 1125 return (error); 1126 } 1127 1128 /* 1129 * Check if a directory is empty or not. 1130 * Inode supplied must be locked. 1131 * 1132 * Using a struct dirtemplate here is not precisely 1133 * what we want, but better than using a struct direct. 1134 * 1135 * NB: does not handle corrupted directories. 1136 */ 1137 int 1138 ext2_dirempty(struct inode *ip, ino_t parentino, struct ucred *cred) 1139 { 1140 off_t off; 1141 struct dirtemplate dbuf; 1142 struct ext2fs_direct_2 *dp = (struct ext2fs_direct_2 *)&dbuf; 1143 int error, namlen; 1144 ssize_t count; 1145 #define MINDIRSIZ (sizeof(struct dirtemplate) / 2) 1146 1147 for (off = 0; off < ip->i_size; off += dp->e2d_reclen) { 1148 error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ, 1149 off, UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK, cred, 1150 NOCRED, &count, (struct thread *)0); 1151 /* 1152 * Since we read MINDIRSIZ, residual must 1153 * be 0 unless we're at end of file. 1154 */ 1155 if (error || count != 0) 1156 return (0); 1157 /* avoid infinite loops */ 1158 if (dp->e2d_reclen == 0) 1159 return (0); 1160 /* skip empty entries */ 1161 if (dp->e2d_ino == 0) 1162 continue; 1163 /* accept only "." and ".." */ 1164 namlen = dp->e2d_namlen; 1165 if (namlen > 2) 1166 return (0); 1167 if (dp->e2d_name[0] != '.') 1168 return (0); 1169 /* 1170 * At this point namlen must be 1 or 2. 1171 * 1 implies ".", 2 implies ".." if second 1172 * char is also "." 1173 */ 1174 if (namlen == 1) 1175 continue; 1176 if (dp->e2d_name[1] == '.' && dp->e2d_ino == parentino) 1177 continue; 1178 return (0); 1179 } 1180 return (1); 1181 } 1182 1183 /* 1184 * Check if source directory is in the path of the target directory. 1185 * Target is supplied locked, source is unlocked. 1186 * The target is always vput before returning. 1187 */ 1188 int 1189 ext2_checkpath(struct inode *source, struct inode *target, struct ucred *cred) 1190 { 1191 struct vnode *vp; 1192 int error, namlen; 1193 struct dirtemplate dirbuf; 1194 1195 vp = ITOV(target); 1196 if (target->i_number == source->i_number) { 1197 error = EEXIST; 1198 goto out; 1199 } 1200 if (target->i_number == EXT2_ROOTINO) { 1201 error = 0; 1202 goto out; 1203 } 1204 1205 for (;;) { 1206 if (vp->v_type != VDIR) { 1207 error = ENOTDIR; 1208 break; 1209 } 1210 error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf, 1211 sizeof(struct dirtemplate), (off_t)0, UIO_SYSSPACE, 1212 IO_NODELOCKED | IO_NOMACCHECK, cred, NOCRED, NULL, 1213 NULL); 1214 if (error != 0) 1215 break; 1216 namlen = dirbuf.dotdot_type; /* like ufs little-endian */ 1217 if (namlen != 2 || 1218 dirbuf.dotdot_name[0] != '.' || 1219 dirbuf.dotdot_name[1] != '.') { 1220 error = ENOTDIR; 1221 break; 1222 } 1223 if (dirbuf.dotdot_ino == source->i_number) { 1224 error = EINVAL; 1225 break; 1226 } 1227 if (dirbuf.dotdot_ino == EXT2_ROOTINO) 1228 break; 1229 vput(vp); 1230 if ((error = VFS_VGET(vp->v_mount, dirbuf.dotdot_ino, 1231 LK_EXCLUSIVE, &vp)) != 0) { 1232 vp = NULL; 1233 break; 1234 } 1235 } 1236 1237 out: 1238 if (error == ENOTDIR) 1239 printf("checkpath: .. not a directory\n"); 1240 if (vp != NULL) 1241 vput(vp); 1242 return (error); 1243 } 1244