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/inode.h> 61 #include <fs/ext2fs/ext2_mount.h> 62 #include <fs/ext2fs/ext2fs.h> 63 #include <fs/ext2fs/ext2_dinode.h> 64 #include <fs/ext2fs/ext2_dir.h> 65 #include <fs/ext2fs/ext2_extern.h> 66 #include <fs/ext2fs/fs.h> 67 68 #ifdef INVARIANTS 69 static int dirchk = 1; 70 #else 71 static int dirchk = 0; 72 #endif 73 74 static SYSCTL_NODE(_vfs, OID_AUTO, e2fs, CTLFLAG_RD, 0, "EXT2FS filesystem"); 75 SYSCTL_INT(_vfs_e2fs, OID_AUTO, dircheck, CTLFLAG_RW, &dirchk, 0, ""); 76 77 /* 78 DIRBLKSIZE in ffs is DEV_BSIZE (in most cases 512) 79 while it is the native blocksize in ext2fs - thus, a #define 80 is no longer appropriate 81 */ 82 #undef DIRBLKSIZ 83 84 static u_char ext2_ft_to_dt[] = { 85 DT_UNKNOWN, /* EXT2_FT_UNKNOWN */ 86 DT_REG, /* EXT2_FT_REG_FILE */ 87 DT_DIR, /* EXT2_FT_DIR */ 88 DT_CHR, /* EXT2_FT_CHRDEV */ 89 DT_BLK, /* EXT2_FT_BLKDEV */ 90 DT_FIFO, /* EXT2_FT_FIFO */ 91 DT_SOCK, /* EXT2_FT_SOCK */ 92 DT_LNK, /* EXT2_FT_SYMLINK */ 93 }; 94 #define FTTODT(ft) \ 95 ((ft) < nitems(ext2_ft_to_dt) ? ext2_ft_to_dt[(ft)] : DT_UNKNOWN) 96 97 static u_char dt_to_ext2_ft[] = { 98 EXT2_FT_UNKNOWN, /* DT_UNKNOWN */ 99 EXT2_FT_FIFO, /* DT_FIFO */ 100 EXT2_FT_CHRDEV, /* DT_CHR */ 101 EXT2_FT_UNKNOWN, /* unused */ 102 EXT2_FT_DIR, /* DT_DIR */ 103 EXT2_FT_UNKNOWN, /* unused */ 104 EXT2_FT_BLKDEV, /* DT_BLK */ 105 EXT2_FT_UNKNOWN, /* unused */ 106 EXT2_FT_REG_FILE, /* DT_REG */ 107 EXT2_FT_UNKNOWN, /* unused */ 108 EXT2_FT_SYMLINK, /* DT_LNK */ 109 EXT2_FT_UNKNOWN, /* unused */ 110 EXT2_FT_SOCK, /* DT_SOCK */ 111 EXT2_FT_UNKNOWN, /* unused */ 112 EXT2_FT_UNKNOWN, /* DT_WHT */ 113 }; 114 #define DTTOFT(dt) \ 115 ((dt) < nitems(dt_to_ext2_ft) ? dt_to_ext2_ft[(dt)] : EXT2_FT_UNKNOWN) 116 117 static int ext2_dirbadentry(struct vnode *dp, struct ext2fs_direct_2 *de, 118 int entryoffsetinblock); 119 static int ext2_is_dot_entry(struct componentname *cnp); 120 static int ext2_lookup_ino(struct vnode *vdp, struct vnode **vpp, 121 struct componentname *cnp, ino_t *dd_ino); 122 123 static int 124 ext2_is_dot_entry(struct componentname *cnp) 125 { 126 if (cnp->cn_namelen <= 2 && cnp->cn_nameptr[0] == '.' && 127 (cnp->cn_nameptr[1] == '.' || cnp->cn_nameptr[1] == '\0')) 128 return (1); 129 return (0); 130 } 131 132 /* 133 * Vnode op for reading directories. 134 */ 135 int 136 ext2_readdir(struct vop_readdir_args *ap) 137 { 138 struct vnode *vp = ap->a_vp; 139 struct uio *uio = ap->a_uio; 140 struct buf *bp; 141 struct inode *ip; 142 struct ext2fs_direct_2 *dp, *edp; 143 u_long *cookies; 144 struct dirent dstdp; 145 off_t offset, startoffset; 146 size_t readcnt, skipcnt; 147 ssize_t startresid; 148 int DIRBLKSIZ = VTOI(ap->a_vp)->i_e2fs->e2fs_bsize; 149 int error; 150 u_int ncookies; 151 152 if (uio->uio_offset < 0) 153 return (EINVAL); 154 ip = VTOI(vp); 155 if (ap->a_ncookies != NULL) { 156 ncookies = uio->uio_resid; 157 if (uio->uio_offset >= ip->i_size) 158 ncookies = 0; 159 else if (ip->i_size - uio->uio_offset < ncookies) 160 ncookies = ip->i_size - uio->uio_offset; 161 ncookies = ncookies / (offsetof(struct ext2fs_direct_2, 162 e2d_namlen) + 4) + 1; 163 cookies = mallocarray(ncookies, sizeof(*cookies), M_TEMP, 164 M_WAITOK); 165 *ap->a_ncookies = ncookies; 166 *ap->a_cookies = cookies; 167 } else { 168 ncookies = 0; 169 cookies = NULL; 170 } 171 offset = startoffset = uio->uio_offset; 172 startresid = uio->uio_resid; 173 error = 0; 174 while (error == 0 && uio->uio_resid > 0 && 175 uio->uio_offset < ip->i_size) { 176 error = ext2_blkatoff(vp, uio->uio_offset, NULL, &bp); 177 if (error) 178 break; 179 if (bp->b_offset + bp->b_bcount > ip->i_size) 180 readcnt = ip->i_size - bp->b_offset; 181 else 182 readcnt = bp->b_bcount; 183 skipcnt = (size_t)(uio->uio_offset - bp->b_offset) & 184 ~(size_t)(DIRBLKSIZ - 1); 185 offset = bp->b_offset + skipcnt; 186 dp = (struct ext2fs_direct_2 *)&bp->b_data[skipcnt]; 187 edp = (struct ext2fs_direct_2 *)&bp->b_data[readcnt]; 188 while (error == 0 && uio->uio_resid > 0 && dp < edp) { 189 if (dp->e2d_reclen <= offsetof(struct ext2fs_direct_2, 190 e2d_namlen) || (caddr_t)dp + dp->e2d_reclen > 191 (caddr_t)edp) { 192 error = EIO; 193 break; 194 } 195 /*- 196 * "New" ext2fs directory entries differ in 3 ways 197 * from ufs on-disk ones: 198 * - the name is not necessarily NUL-terminated. 199 * - the file type field always exists and always 200 * follows the name length field. 201 * - the file type is encoded in a different way. 202 * 203 * "Old" ext2fs directory entries need no special 204 * conversions, since they are binary compatible 205 * with "new" entries having a file type of 0 (i.e., 206 * EXT2_FT_UNKNOWN). Splitting the old name length 207 * field didn't make a mess like it did in ufs, 208 * because ext2fs uses a machine-independent disk 209 * layout. 210 */ 211 dstdp.d_namlen = dp->e2d_namlen; 212 dstdp.d_type = FTTODT(dp->e2d_type); 213 if (offsetof(struct ext2fs_direct_2, e2d_namlen) + 214 dstdp.d_namlen > dp->e2d_reclen) { 215 error = EIO; 216 break; 217 } 218 if (offset < startoffset || dp->e2d_ino == 0) 219 goto nextentry; 220 dstdp.d_fileno = dp->e2d_ino; 221 dstdp.d_reclen = GENERIC_DIRSIZ(&dstdp); 222 bcopy(dp->e2d_name, dstdp.d_name, dstdp.d_namlen); 223 dstdp.d_name[dstdp.d_namlen] = '\0'; 224 if (dstdp.d_reclen > uio->uio_resid) { 225 if (uio->uio_resid == startresid) 226 error = EINVAL; 227 else 228 error = EJUSTRETURN; 229 break; 230 } 231 /* Advance dp. */ 232 error = uiomove((caddr_t)&dstdp, dstdp.d_reclen, uio); 233 if (error) 234 break; 235 if (cookies != NULL) { 236 KASSERT(ncookies > 0, 237 ("ext2_readdir: cookies buffer too small")); 238 *cookies = offset + dp->e2d_reclen; 239 cookies++; 240 ncookies--; 241 } 242 nextentry: 243 offset += dp->e2d_reclen; 244 dp = (struct ext2fs_direct_2 *)((caddr_t)dp + 245 dp->e2d_reclen); 246 } 247 bqrelse(bp); 248 uio->uio_offset = offset; 249 } 250 /* We need to correct uio_offset. */ 251 uio->uio_offset = offset; 252 if (error == EJUSTRETURN) 253 error = 0; 254 if (ap->a_ncookies != NULL) { 255 if (error == 0) { 256 ap->a_ncookies -= ncookies; 257 } else { 258 free(*ap->a_cookies, M_TEMP); 259 *ap->a_ncookies = 0; 260 *ap->a_cookies = NULL; 261 } 262 } 263 if (error == 0 && ap->a_eofflag) 264 *ap->a_eofflag = ip->i_size <= uio->uio_offset; 265 return (error); 266 } 267 268 /* 269 * Convert a component of a pathname into a pointer to a locked inode. 270 * This is a very central and rather complicated routine. 271 * If the file system is not maintained in a strict tree hierarchy, 272 * this can result in a deadlock situation (see comments in code below). 273 * 274 * The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending 275 * on whether the name is to be looked up, created, renamed, or deleted. 276 * When CREATE, RENAME, or DELETE is specified, information usable in 277 * creating, renaming, or deleting a directory entry may be calculated. 278 * If flag has LOCKPARENT or'ed into it and the target of the pathname 279 * exists, lookup returns both the target and its parent directory locked. 280 * When creating or renaming and LOCKPARENT is specified, the target may 281 * not be ".". When deleting and LOCKPARENT is specified, the target may 282 * be "."., but the caller must check to ensure it does an vrele and vput 283 * instead of two vputs. 284 * 285 * Overall outline of ext2_lookup: 286 * 287 * search for name in directory, to found or notfound 288 * notfound: 289 * if creating, return locked directory, leaving info on available slots 290 * else return error 291 * found: 292 * if at end of path and deleting, return information to allow delete 293 * if at end of path and rewriting (RENAME and LOCKPARENT), lock target 294 * inode and return info to allow rewrite 295 * if not at end, add name to cache; if at end and neither creating 296 * nor deleting, add name to cache 297 */ 298 int 299 ext2_lookup(struct vop_cachedlookup_args *ap) 300 { 301 302 return (ext2_lookup_ino(ap->a_dvp, ap->a_vpp, ap->a_cnp, NULL)); 303 } 304 305 static int 306 ext2_lookup_ino(struct vnode *vdp, struct vnode **vpp, struct componentname *cnp, 307 ino_t *dd_ino) 308 { 309 struct inode *dp; /* inode for directory being searched */ 310 struct buf *bp; /* a buffer of directory entries */ 311 struct ext2fs_direct_2 *ep; /* the current directory entry */ 312 int entryoffsetinblock; /* offset of ep in bp's buffer */ 313 struct ext2fs_searchslot ss; 314 doff_t i_diroff; /* cached i_diroff value */ 315 doff_t i_offset; /* cached i_offset value */ 316 int numdirpasses; /* strategy for directory search */ 317 doff_t endsearch; /* offset to end directory search */ 318 doff_t prevoff; /* prev entry dp->i_offset */ 319 struct vnode *pdp; /* saved dp during symlink work */ 320 struct vnode *tdp; /* returned by VFS_VGET */ 321 doff_t enduseful; /* pointer past last used dir slot */ 322 u_long bmask; /* block offset mask */ 323 int error; 324 struct ucred *cred = cnp->cn_cred; 325 int flags = cnp->cn_flags; 326 int nameiop = cnp->cn_nameiop; 327 ino_t ino, ino1; 328 int ltype; 329 int entry_found = 0; 330 331 int DIRBLKSIZ = VTOI(vdp)->i_e2fs->e2fs_bsize; 332 333 if (vpp != NULL) 334 *vpp = NULL; 335 336 dp = VTOI(vdp); 337 bmask = VFSTOEXT2(vdp->v_mount)->um_mountp->mnt_stat.f_iosize - 1; 338 restart: 339 bp = NULL; 340 ss.slotoffset = -1; 341 342 /* 343 * We now have a segment name to search for, and a directory to search. 344 * 345 * Suppress search for slots unless creating 346 * file and at end of pathname, in which case 347 * we watch for a place to put the new file in 348 * case it doesn't already exist. 349 */ 350 i_diroff = dp->i_diroff; 351 ss.slotstatus = FOUND; 352 ss.slotfreespace = ss.slotsize = ss.slotneeded = 0; 353 if ((nameiop == CREATE || nameiop == RENAME) && 354 (flags & ISLASTCN)) { 355 ss.slotstatus = NONE; 356 ss.slotneeded = EXT2_DIR_REC_LEN(cnp->cn_namelen); 357 /* 358 * was ss.slotneeded = (sizeof(struct direct) - MAXNAMLEN + 359 * cnp->cn_namelen + 3) &~ 3; 360 */ 361 } 362 /* 363 * Try to lookup dir entry using htree directory index. 364 * 365 * If we got an error or we want to find '.' or '..' entry, 366 * we will fall back to linear search. 367 */ 368 if (!ext2_is_dot_entry(cnp) && ext2_htree_has_idx(dp)) { 369 numdirpasses = 1; 370 entryoffsetinblock = 0; 371 switch (ext2_htree_lookup(dp, cnp->cn_nameptr, cnp->cn_namelen, 372 &bp, &entryoffsetinblock, &i_offset, &prevoff, 373 &enduseful, &ss)) { 374 case 0: 375 ep = (struct ext2fs_direct_2 *)((char *)bp->b_data + 376 (i_offset & bmask)); 377 goto foundentry; 378 case ENOENT: 379 i_offset = roundup2(dp->i_size, DIRBLKSIZ); 380 goto notfound; 381 default: 382 /* 383 * Something failed; just fallback to do a linear 384 * search. 385 */ 386 break; 387 } 388 } 389 390 /* 391 * If there is cached information on a previous search of 392 * this directory, pick up where we last left off. 393 * We cache only lookups as these are the most common 394 * and have the greatest payoff. Caching CREATE has little 395 * benefit as it usually must search the entire directory 396 * to determine that the entry does not exist. Caching the 397 * location of the last DELETE or RENAME has not reduced 398 * profiling time and hence has been removed in the interest 399 * of simplicity. 400 */ 401 if (nameiop != LOOKUP || i_diroff == 0 || 402 i_diroff > dp->i_size) { 403 entryoffsetinblock = 0; 404 i_offset = 0; 405 numdirpasses = 1; 406 } else { 407 i_offset = i_diroff; 408 if ((entryoffsetinblock = i_offset & bmask) && 409 (error = ext2_blkatoff(vdp, (off_t)i_offset, NULL, 410 &bp))) 411 return (error); 412 numdirpasses = 2; 413 nchstats.ncs_2passes++; 414 } 415 prevoff = i_offset; 416 endsearch = roundup2(dp->i_size, DIRBLKSIZ); 417 enduseful = 0; 418 419 searchloop: 420 while (i_offset < endsearch) { 421 /* 422 * If necessary, get the next directory block. 423 */ 424 if (bp != NULL) 425 brelse(bp); 426 error = ext2_blkatoff(vdp, (off_t)i_offset, NULL, &bp); 427 if (error != 0) 428 return (error); 429 entryoffsetinblock = 0; 430 /* 431 * If still looking for a slot, and at a DIRBLKSIZE 432 * boundary, have to start looking for free space again. 433 */ 434 if (ss.slotstatus == NONE && 435 (entryoffsetinblock & (DIRBLKSIZ - 1)) == 0) { 436 ss.slotoffset = -1; 437 ss.slotfreespace = 0; 438 } 439 error = ext2_search_dirblock(dp, bp->b_data, &entry_found, 440 cnp->cn_nameptr, cnp->cn_namelen, 441 &entryoffsetinblock, &i_offset, &prevoff, 442 &enduseful, &ss); 443 if (error != 0) { 444 brelse(bp); 445 return (error); 446 } 447 if (entry_found) { 448 ep = (struct ext2fs_direct_2 *)((char *)bp->b_data + 449 (entryoffsetinblock & bmask)); 450 foundentry: 451 ino = ep->e2d_ino; 452 goto found; 453 } 454 } 455 notfound: 456 /* 457 * If we started in the middle of the directory and failed 458 * to find our target, we must check the beginning as well. 459 */ 460 if (numdirpasses == 2) { 461 numdirpasses--; 462 i_offset = 0; 463 endsearch = i_diroff; 464 goto searchloop; 465 } 466 if (bp != NULL) 467 brelse(bp); 468 /* 469 * If creating, and at end of pathname and current 470 * directory has not been removed, then can consider 471 * allowing file to be created. 472 */ 473 if ((nameiop == CREATE || nameiop == RENAME) && 474 (flags & ISLASTCN) && dp->i_nlink != 0) { 475 /* 476 * Access for write is interpreted as allowing 477 * creation of files in the directory. 478 */ 479 if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread)) != 0) 480 return (error); 481 /* 482 * Return an indication of where the new directory 483 * entry should be put. If we didn't find a slot, 484 * then set dp->i_count to 0 indicating 485 * that the new slot belongs at the end of the 486 * directory. If we found a slot, then the new entry 487 * can be put in the range from dp->i_offset to 488 * dp->i_offset + dp->i_count. 489 */ 490 if (ss.slotstatus == NONE) { 491 dp->i_offset = roundup2(dp->i_size, DIRBLKSIZ); 492 dp->i_count = 0; 493 enduseful = dp->i_offset; 494 } else { 495 dp->i_offset = ss.slotoffset; 496 dp->i_count = ss.slotsize; 497 if (enduseful < ss.slotoffset + ss.slotsize) 498 enduseful = ss.slotoffset + ss.slotsize; 499 } 500 dp->i_endoff = roundup2(enduseful, DIRBLKSIZ); 501 /* 502 * We return with the directory locked, so that 503 * the parameters we set up above will still be 504 * valid if we actually decide to do a direnter(). 505 * We return ni_vp == NULL to indicate that the entry 506 * does not currently exist; we leave a pointer to 507 * the (locked) directory inode in ndp->ni_dvp. 508 * The pathname buffer is saved so that the name 509 * can be obtained later. 510 * 511 * NB - if the directory is unlocked, then this 512 * information cannot be used. 513 */ 514 cnp->cn_flags |= SAVENAME; 515 return (EJUSTRETURN); 516 } 517 /* 518 * Insert name into cache (as non-existent) if appropriate. 519 */ 520 if ((cnp->cn_flags & MAKEENTRY) != 0) 521 cache_enter(vdp, NULL, cnp); 522 return (ENOENT); 523 524 found: 525 if (dd_ino != NULL) 526 *dd_ino = ino; 527 if (numdirpasses == 2) 528 nchstats.ncs_pass2++; 529 /* 530 * Check that directory length properly reflects presence 531 * of this entry. 532 */ 533 if (entryoffsetinblock + EXT2_DIR_REC_LEN(ep->e2d_namlen) 534 > dp->i_size) { 535 ext2_dirbad(dp, i_offset, "i_size too small"); 536 dp->i_size = entryoffsetinblock + EXT2_DIR_REC_LEN(ep->e2d_namlen); 537 dp->i_flag |= IN_CHANGE | IN_UPDATE; 538 } 539 brelse(bp); 540 541 /* 542 * Found component in pathname. 543 * If the final component of path name, save information 544 * in the cache as to where the entry was found. 545 */ 546 if ((flags & ISLASTCN) && nameiop == LOOKUP) 547 dp->i_diroff = rounddown2(i_offset, DIRBLKSIZ); 548 /* 549 * If deleting, and at end of pathname, return 550 * parameters which can be used to remove file. 551 */ 552 if (nameiop == DELETE && (flags & ISLASTCN)) { 553 if (flags & LOCKPARENT) 554 ASSERT_VOP_ELOCKED(vdp, __FUNCTION__); 555 /* 556 * Write access to directory required to delete files. 557 */ 558 if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread)) != 0) 559 return (error); 560 /* 561 * Return pointer to current entry in dp->i_offset, 562 * and distance past previous entry (if there 563 * is a previous entry in this block) in dp->i_count. 564 * Save directory inode pointer in ndp->ni_dvp for dirremove(). 565 * 566 * Technically we shouldn't be setting these in the 567 * WANTPARENT case (first lookup in rename()), but any 568 * lookups that will result in directory changes will 569 * overwrite these. 570 */ 571 dp->i_offset = i_offset; 572 if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0) 573 dp->i_count = 0; 574 else 575 dp->i_count = dp->i_offset - prevoff; 576 if (dd_ino != NULL) 577 return (0); 578 if (dp->i_number == ino) { 579 VREF(vdp); 580 *vpp = vdp; 581 return (0); 582 } 583 if ((error = VFS_VGET(vdp->v_mount, ino, LK_EXCLUSIVE, 584 &tdp)) != 0) 585 return (error); 586 /* 587 * If directory is "sticky", then user must own 588 * the directory, or the file in it, else she 589 * may not delete it (unless she's root). This 590 * implements append-only directories. 591 */ 592 if ((dp->i_mode & ISVTX) && 593 cred->cr_uid != 0 && 594 cred->cr_uid != dp->i_uid && 595 VTOI(tdp)->i_uid != cred->cr_uid) { 596 vput(tdp); 597 return (EPERM); 598 } 599 *vpp = tdp; 600 return (0); 601 } 602 603 /* 604 * If rewriting (RENAME), return the inode and the 605 * information required to rewrite the present directory 606 * Must get inode of directory entry to verify it's a 607 * regular file, or empty directory. 608 */ 609 if (nameiop == RENAME && (flags & ISLASTCN)) { 610 if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread)) != 0) 611 return (error); 612 /* 613 * Careful about locking second inode. 614 * This can only occur if the target is ".". 615 */ 616 dp->i_offset = i_offset; 617 if (dp->i_number == ino) 618 return (EISDIR); 619 if (dd_ino != NULL) 620 return (0); 621 if ((error = VFS_VGET(vdp->v_mount, ino, LK_EXCLUSIVE, 622 &tdp)) != 0) 623 return (error); 624 *vpp = tdp; 625 cnp->cn_flags |= SAVENAME; 626 return (0); 627 } 628 if (dd_ino != NULL) 629 return (0); 630 631 /* 632 * Step through the translation in the name. We do not `vput' the 633 * directory because we may need it again if a symbolic link 634 * is relative to the current directory. Instead we save it 635 * unlocked as "pdp". We must get the target inode before unlocking 636 * the directory to insure that the inode will not be removed 637 * before we get it. We prevent deadlock by always fetching 638 * inodes from the root, moving down the directory tree. Thus 639 * when following backward pointers ".." we must unlock the 640 * parent directory before getting the requested directory. 641 * There is a potential race condition here if both the current 642 * and parent directories are removed before the VFS_VGET for the 643 * inode associated with ".." returns. We hope that this occurs 644 * infrequently since we cannot avoid this race condition without 645 * implementing a sophisticated deadlock detection algorithm. 646 * Note also that this simple deadlock detection scheme will not 647 * work if the file system has any hard links other than ".." 648 * that point backwards in the directory structure. 649 */ 650 pdp = vdp; 651 if (flags & ISDOTDOT) { 652 error = vn_vget_ino(pdp, ino, cnp->cn_lkflags, &tdp); 653 if (pdp->v_iflag & VI_DOOMED) { 654 if (error == 0) 655 vput(tdp); 656 error = ENOENT; 657 } 658 if (error) 659 return (error); 660 /* 661 * Recheck that ".." entry in the vdp directory points 662 * to the inode we looked up before vdp lock was 663 * dropped. 664 */ 665 error = ext2_lookup_ino(pdp, NULL, cnp, &ino1); 666 if (error) { 667 vput(tdp); 668 return (error); 669 } 670 if (ino1 != ino) { 671 vput(tdp); 672 goto restart; 673 } 674 *vpp = tdp; 675 } else if (dp->i_number == ino) { 676 VREF(vdp); /* we want ourself, ie "." */ 677 /* 678 * When we lookup "." we still can be asked to lock it 679 * differently. 680 */ 681 ltype = cnp->cn_lkflags & LK_TYPE_MASK; 682 if (ltype != VOP_ISLOCKED(vdp)) { 683 if (ltype == LK_EXCLUSIVE) 684 vn_lock(vdp, LK_UPGRADE | LK_RETRY); 685 else /* if (ltype == LK_SHARED) */ 686 vn_lock(vdp, LK_DOWNGRADE | LK_RETRY); 687 } 688 *vpp = vdp; 689 } else { 690 if ((error = VFS_VGET(vdp->v_mount, ino, cnp->cn_lkflags, 691 &tdp)) != 0) 692 return (error); 693 *vpp = tdp; 694 } 695 696 /* 697 * Insert name into cache if appropriate. 698 */ 699 if (cnp->cn_flags & MAKEENTRY) 700 cache_enter(vdp, *vpp, cnp); 701 return (0); 702 } 703 704 int 705 ext2_search_dirblock(struct inode *ip, void *data, int *foundp, 706 const char *name, int namelen, int *entryoffsetinblockp, 707 doff_t *offp, doff_t *prevoffp, doff_t *endusefulp, 708 struct ext2fs_searchslot *ssp) 709 { 710 struct vnode *vdp; 711 struct ext2fs_direct_2 *ep, *top; 712 uint32_t bsize = ip->i_e2fs->e2fs_bsize; 713 int offset = *entryoffsetinblockp; 714 int namlen; 715 716 vdp = ITOV(ip); 717 718 ep = (struct ext2fs_direct_2 *)((char *)data + offset); 719 top = (struct ext2fs_direct_2 *)((char *)data + 720 bsize - EXT2_DIR_REC_LEN(0)); 721 722 while (ep < top) { 723 /* 724 * Full validation checks are slow, so we only check 725 * enough to insure forward progress through the 726 * directory. Complete checks can be run by setting 727 * "vfs.e2fs.dirchk" to be true. 728 */ 729 if (ep->e2d_reclen == 0 || 730 (dirchk && ext2_dirbadentry(vdp, ep, offset))) { 731 int i; 732 733 ext2_dirbad(ip, *offp, "mangled entry"); 734 i = bsize - (offset & (bsize - 1)); 735 *offp += i; 736 offset += i; 737 continue; 738 } 739 740 /* 741 * If an appropriate sized slot has not yet been found, 742 * check to see if one is available. Also accumulate space 743 * in the current block so that we can determine if 744 * compaction is viable. 745 */ 746 if (ssp->slotstatus != FOUND) { 747 int size = ep->e2d_reclen; 748 749 if (ep->e2d_ino != 0) 750 size -= EXT2_DIR_REC_LEN(ep->e2d_namlen); 751 if (size > 0) { 752 if (size >= ssp->slotneeded) { 753 ssp->slotstatus = FOUND; 754 ssp->slotoffset = *offp; 755 ssp->slotsize = ep->e2d_reclen; 756 } else if (ssp->slotstatus == NONE) { 757 ssp->slotfreespace += size; 758 if (ssp->slotoffset == -1) 759 ssp->slotoffset = *offp; 760 if (ssp->slotfreespace >= ssp->slotneeded) { 761 ssp->slotstatus = COMPACT; 762 ssp->slotsize = *offp + 763 ep->e2d_reclen - 764 ssp->slotoffset; 765 } 766 } 767 } 768 } 769 /* 770 * Check for a name match. 771 */ 772 if (ep->e2d_ino) { 773 namlen = ep->e2d_namlen; 774 if (namlen == namelen && 775 !bcmp(name, ep->e2d_name, (unsigned)namlen)) { 776 /* 777 * Save directory entry's inode number and 778 * reclen in ndp->ni_ufs area, and release 779 * directory buffer. 780 */ 781 *foundp = 1; 782 return (0); 783 } 784 } 785 *prevoffp = *offp; 786 *offp += ep->e2d_reclen; 787 offset += ep->e2d_reclen; 788 *entryoffsetinblockp = offset; 789 if (ep->e2d_ino) 790 *endusefulp = *offp; 791 /* 792 * Get pointer to the next entry. 793 */ 794 ep = (struct ext2fs_direct_2 *)((char *)data + offset); 795 } 796 797 return (0); 798 } 799 800 void 801 ext2_dirbad(struct inode *ip, doff_t offset, char *how) 802 { 803 struct mount *mp; 804 805 mp = ITOV(ip)->v_mount; 806 if ((mp->mnt_flag & MNT_RDONLY) == 0) 807 panic("ext2_dirbad: %s: bad dir ino %ju at offset %ld: %s\n", 808 mp->mnt_stat.f_mntonname, (uintmax_t)ip->i_number, 809 (long)offset, how); 810 else 811 (void)printf("%s: bad dir ino %ju at offset %ld: %s\n", 812 mp->mnt_stat.f_mntonname, (uintmax_t)ip->i_number, 813 (long)offset, how); 814 815 } 816 817 /* 818 * Do consistency checking on a directory entry: 819 * record length must be multiple of 4 820 * entry must fit in rest of its DIRBLKSIZ block 821 * record must be large enough to contain entry 822 * name is not longer than MAXNAMLEN 823 * name must be as long as advertised, and null terminated 824 */ 825 /* 826 * changed so that it confirms to ext2_check_dir_entry 827 */ 828 static int 829 ext2_dirbadentry(struct vnode *dp, struct ext2fs_direct_2 *de, 830 int entryoffsetinblock) 831 { 832 int DIRBLKSIZ = VTOI(dp)->i_e2fs->e2fs_bsize; 833 834 char *error_msg = NULL; 835 836 if (de->e2d_reclen < EXT2_DIR_REC_LEN(1)) 837 error_msg = "rec_len is smaller than minimal"; 838 else if (de->e2d_reclen % 4 != 0) 839 error_msg = "rec_len % 4 != 0"; 840 else if (de->e2d_reclen < EXT2_DIR_REC_LEN(de->e2d_namlen)) 841 error_msg = "reclen is too small for name_len"; 842 else if (entryoffsetinblock + de->e2d_reclen > DIRBLKSIZ) 843 error_msg = "directory entry across blocks"; 844 /* else LATER 845 if (de->inode > dir->i_sb->u.ext2_sb.s_es->s_inodes_count) 846 error_msg = "inode out of bounds"; 847 */ 848 849 if (error_msg != NULL) { 850 printf("bad directory entry: %s\n", error_msg); 851 printf("offset=%d, inode=%lu, rec_len=%u, name_len=%u\n", 852 entryoffsetinblock, (unsigned long)de->e2d_ino, 853 de->e2d_reclen, de->e2d_namlen); 854 } 855 return error_msg == NULL ? 0 : 1; 856 } 857 858 /* 859 * Write a directory entry after a call to namei, using the parameters 860 * that it left in nameidata. The argument ip is the inode which the new 861 * directory entry will refer to. Dvp is a pointer to the directory to 862 * be written, which was left locked by namei. Remaining parameters 863 * (dp->i_offset, dp->i_count) indicate how the space for the new 864 * entry is to be obtained. 865 */ 866 int 867 ext2_direnter(struct inode *ip, struct vnode *dvp, struct componentname *cnp) 868 { 869 struct inode *dp; 870 struct ext2fs_direct_2 newdir; 871 struct buf *bp; 872 int error, newentrysize; 873 int DIRBLKSIZ = ip->i_e2fs->e2fs_bsize; 874 875 876 #ifdef INVARIANTS 877 if ((cnp->cn_flags & SAVENAME) == 0) 878 panic("ext2_direnter: missing name"); 879 #endif 880 dp = VTOI(dvp); 881 newdir.e2d_ino = ip->i_number; 882 newdir.e2d_namlen = cnp->cn_namelen; 883 if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs, 884 EXT2F_INCOMPAT_FTYPE)) 885 newdir.e2d_type = DTTOFT(IFTODT(ip->i_mode)); 886 else 887 newdir.e2d_type = EXT2_FT_UNKNOWN; 888 bcopy(cnp->cn_nameptr, newdir.e2d_name, (unsigned)cnp->cn_namelen + 1); 889 newentrysize = EXT2_DIR_REC_LEN(newdir.e2d_namlen); 890 891 if (ext2_htree_has_idx(dp)) { 892 error = ext2_htree_add_entry(dvp, &newdir, cnp); 893 if (error) { 894 dp->i_flag &= ~IN_E3INDEX; 895 dp->i_flag |= IN_CHANGE | IN_UPDATE; 896 } 897 return (error); 898 } 899 900 if (EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_DIRHASHINDEX) && 901 !ext2_htree_has_idx(dp)) { 902 if ((dp->i_size / DIRBLKSIZ) == 1 && 903 dp->i_offset == DIRBLKSIZ) { 904 /* 905 * Making indexed directory when one block is not 906 * enough to save all entries. 907 */ 908 return ext2_htree_create_index(dvp, cnp, &newdir); 909 } 910 } 911 912 if (dp->i_count == 0) { 913 /* 914 * If dp->i_count is 0, then namei could find no 915 * space in the directory. Here, dp->i_offset will 916 * be on a directory block boundary and we will write the 917 * new entry into a fresh block. 918 */ 919 if (dp->i_offset & (DIRBLKSIZ - 1)) 920 panic("ext2_direnter: newblk"); 921 922 newdir.e2d_reclen = DIRBLKSIZ; 923 924 bp = getblk(ip->i_devvp, lblkno(dp->i_e2fs, dp->i_offset), 925 DIRBLKSIZ, 0, 0, 0); 926 if (!bp) 927 return (EIO); 928 929 memcpy(bp->b_data, &newdir, sizeof(struct ext2fs_direct_2)); 930 931 ext2_dir_blk_csum_set(dp, bp); 932 error = bwrite(bp); 933 if (error) 934 return (error); 935 936 dp->i_size = roundup2(dp->i_size, DIRBLKSIZ); 937 dp->i_flag |= IN_CHANGE; 938 939 return (0); 940 } 941 942 error = ext2_add_entry(dvp, &newdir); 943 if (!error && dp->i_endoff && dp->i_endoff < dp->i_size) 944 error = ext2_truncate(dvp, (off_t)dp->i_endoff, IO_SYNC, 945 cnp->cn_cred, cnp->cn_thread); 946 return (error); 947 } 948 949 /* 950 * Insert an entry into the directory block. 951 * Compact the contents. 952 */ 953 int 954 ext2_add_entry(struct vnode *dvp, struct ext2fs_direct_2 *entry) 955 { 956 struct ext2fs_direct_2 *ep, *nep; 957 struct inode *dp; 958 struct buf *bp; 959 u_int dsize; 960 int error, loc, newentrysize, spacefree; 961 char *dirbuf; 962 963 dp = VTOI(dvp); 964 965 /* 966 * If dp->i_count is non-zero, then namei found space 967 * for the new entry in the range dp->i_offset to 968 * dp->i_offset + dp->i_count in the directory. 969 * To use this space, we may have to compact the entries located 970 * there, by copying them together towards the beginning of the 971 * block, leaving the free space in one usable chunk at the end. 972 */ 973 974 /* 975 * Increase size of directory if entry eats into new space. 976 * This should never push the size past a new multiple of 977 * DIRBLKSIZE. 978 * 979 * N.B. - THIS IS AN ARTIFACT OF 4.2 AND SHOULD NEVER HAPPEN. 980 */ 981 if (dp->i_offset + dp->i_count > dp->i_size) 982 dp->i_size = dp->i_offset + dp->i_count; 983 /* 984 * Get the block containing the space for the new directory entry. 985 */ 986 if ((error = ext2_blkatoff(dvp, (off_t)dp->i_offset, &dirbuf, 987 &bp)) != 0) 988 return (error); 989 /* 990 * Find space for the new entry. In the simple case, the entry at 991 * offset base will have the space. If it does not, then namei 992 * arranged that compacting the region dp->i_offset to 993 * dp->i_offset + dp->i_count would yield the 994 * space. 995 */ 996 newentrysize = EXT2_DIR_REC_LEN(entry->e2d_namlen); 997 ep = (struct ext2fs_direct_2 *)dirbuf; 998 dsize = EXT2_DIR_REC_LEN(ep->e2d_namlen); 999 spacefree = ep->e2d_reclen - dsize; 1000 for (loc = ep->e2d_reclen; loc < dp->i_count; ) { 1001 nep = (struct ext2fs_direct_2 *)(dirbuf + loc); 1002 if (ep->e2d_ino) { 1003 /* trim the existing slot */ 1004 ep->e2d_reclen = dsize; 1005 ep = (struct ext2fs_direct_2 *)((char *)ep + dsize); 1006 } else { 1007 /* overwrite; nothing there; header is ours */ 1008 spacefree += dsize; 1009 } 1010 dsize = EXT2_DIR_REC_LEN(nep->e2d_namlen); 1011 spacefree += nep->e2d_reclen - dsize; 1012 loc += nep->e2d_reclen; 1013 bcopy((caddr_t)nep, (caddr_t)ep, dsize); 1014 } 1015 /* 1016 * Update the pointer fields in the previous entry (if any), 1017 * copy in the new entry, and write out the block. 1018 */ 1019 if (ep->e2d_ino == 0) { 1020 if (spacefree + dsize < newentrysize) 1021 panic("ext2_direnter: compact1"); 1022 entry->e2d_reclen = spacefree + dsize; 1023 } else { 1024 if (spacefree < newentrysize) 1025 panic("ext2_direnter: compact2"); 1026 entry->e2d_reclen = spacefree; 1027 ep->e2d_reclen = dsize; 1028 ep = (struct ext2fs_direct_2 *)((char *)ep + dsize); 1029 } 1030 bcopy((caddr_t)entry, (caddr_t)ep, (u_int)newentrysize); 1031 ext2_dir_blk_csum_set(dp, bp); 1032 if (DOINGASYNC(dvp)) { 1033 bdwrite(bp); 1034 error = 0; 1035 } else { 1036 error = bwrite(bp); 1037 } 1038 dp->i_flag |= IN_CHANGE | IN_UPDATE; 1039 return (error); 1040 } 1041 1042 /* 1043 * Remove a directory entry after a call to namei, using 1044 * the parameters which it left in nameidata. The entry 1045 * dp->i_offset contains the offset into the directory of the 1046 * entry to be eliminated. The dp->i_count field contains the 1047 * size of the previous record in the directory. If this 1048 * is 0, the first entry is being deleted, so we need only 1049 * zero the inode number to mark the entry as free. If the 1050 * entry is not the first in the directory, we must reclaim 1051 * the space of the now empty record by adding the record size 1052 * to the size of the previous entry. 1053 */ 1054 int 1055 ext2_dirremove(struct vnode *dvp, struct componentname *cnp) 1056 { 1057 struct inode *dp; 1058 struct ext2fs_direct_2 *ep, *rep; 1059 struct buf *bp; 1060 int error; 1061 1062 dp = VTOI(dvp); 1063 if (dp->i_count == 0) { 1064 /* 1065 * First entry in block: set d_ino to zero. 1066 */ 1067 if ((error = 1068 ext2_blkatoff(dvp, (off_t)dp->i_offset, (char **)&ep, 1069 &bp)) != 0) 1070 return (error); 1071 ep->e2d_ino = 0; 1072 error = bwrite(bp); 1073 dp->i_flag |= IN_CHANGE | IN_UPDATE; 1074 return (error); 1075 } 1076 /* 1077 * Collapse new free space into previous entry. 1078 */ 1079 if ((error = ext2_blkatoff(dvp, (off_t)(dp->i_offset - dp->i_count), 1080 (char **)&ep, &bp)) != 0) 1081 return (error); 1082 1083 /* Set 'rep' to the entry being removed. */ 1084 if (dp->i_count == 0) 1085 rep = ep; 1086 else 1087 rep = (struct ext2fs_direct_2 *)((char *)ep + ep->e2d_reclen); 1088 ep->e2d_reclen += rep->e2d_reclen; 1089 ext2_dir_blk_csum_set(dp, bp); 1090 if (DOINGASYNC(dvp) && dp->i_count != 0) 1091 bdwrite(bp); 1092 else 1093 error = bwrite(bp); 1094 dp->i_flag |= IN_CHANGE | IN_UPDATE; 1095 return (error); 1096 } 1097 1098 /* 1099 * Rewrite an existing directory entry to point at the inode 1100 * supplied. The parameters describing the directory entry are 1101 * set up by a call to namei. 1102 */ 1103 int 1104 ext2_dirrewrite(struct inode *dp, struct inode *ip, struct componentname *cnp) 1105 { 1106 struct buf *bp; 1107 struct ext2fs_direct_2 *ep; 1108 struct vnode *vdp = ITOV(dp); 1109 int error; 1110 1111 if ((error = ext2_blkatoff(vdp, (off_t)dp->i_offset, (char **)&ep, 1112 &bp)) != 0) 1113 return (error); 1114 ep->e2d_ino = ip->i_number; 1115 if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs, 1116 EXT2F_INCOMPAT_FTYPE)) 1117 ep->e2d_type = DTTOFT(IFTODT(ip->i_mode)); 1118 else 1119 ep->e2d_type = EXT2_FT_UNKNOWN; 1120 ext2_dir_blk_csum_set(dp, bp); 1121 error = bwrite(bp); 1122 dp->i_flag |= IN_CHANGE | IN_UPDATE; 1123 return (error); 1124 } 1125 1126 /* 1127 * Check if a directory is empty or not. 1128 * Inode supplied must be locked. 1129 * 1130 * Using a struct dirtemplate here is not precisely 1131 * what we want, but better than using a struct direct. 1132 * 1133 * NB: does not handle corrupted directories. 1134 */ 1135 int 1136 ext2_dirempty(struct inode *ip, ino_t parentino, struct ucred *cred) 1137 { 1138 off_t off; 1139 struct dirtemplate dbuf; 1140 struct ext2fs_direct_2 *dp = (struct ext2fs_direct_2 *)&dbuf; 1141 int error, namlen; 1142 ssize_t count; 1143 #define MINDIRSIZ (sizeof(struct dirtemplate) / 2) 1144 1145 for (off = 0; off < ip->i_size; off += dp->e2d_reclen) { 1146 error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ, 1147 off, UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK, cred, 1148 NOCRED, &count, (struct thread *)0); 1149 /* 1150 * Since we read MINDIRSIZ, residual must 1151 * be 0 unless we're at end of file. 1152 */ 1153 if (error || count != 0) 1154 return (0); 1155 /* avoid infinite loops */ 1156 if (dp->e2d_reclen == 0) 1157 return (0); 1158 /* skip empty entries */ 1159 if (dp->e2d_ino == 0) 1160 continue; 1161 /* accept only "." and ".." */ 1162 namlen = dp->e2d_namlen; 1163 if (namlen > 2) 1164 return (0); 1165 if (dp->e2d_name[0] != '.') 1166 return (0); 1167 /* 1168 * At this point namlen must be 1 or 2. 1169 * 1 implies ".", 2 implies ".." if second 1170 * char is also "." 1171 */ 1172 if (namlen == 1) 1173 continue; 1174 if (dp->e2d_name[1] == '.' && dp->e2d_ino == parentino) 1175 continue; 1176 return (0); 1177 } 1178 return (1); 1179 } 1180 1181 /* 1182 * Check if source directory is in the path of the target directory. 1183 * Target is supplied locked, source is unlocked. 1184 * The target is always vput before returning. 1185 */ 1186 int 1187 ext2_checkpath(struct inode *source, struct inode *target, struct ucred *cred) 1188 { 1189 struct vnode *vp; 1190 int error, namlen; 1191 struct dirtemplate dirbuf; 1192 1193 vp = ITOV(target); 1194 if (target->i_number == source->i_number) { 1195 error = EEXIST; 1196 goto out; 1197 } 1198 if (target->i_number == EXT2_ROOTINO) { 1199 error = 0; 1200 goto out; 1201 } 1202 1203 for (;;) { 1204 if (vp->v_type != VDIR) { 1205 error = ENOTDIR; 1206 break; 1207 } 1208 error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf, 1209 sizeof(struct dirtemplate), (off_t)0, UIO_SYSSPACE, 1210 IO_NODELOCKED | IO_NOMACCHECK, cred, NOCRED, NULL, 1211 NULL); 1212 if (error != 0) 1213 break; 1214 namlen = dirbuf.dotdot_type; /* like ufs little-endian */ 1215 if (namlen != 2 || 1216 dirbuf.dotdot_name[0] != '.' || 1217 dirbuf.dotdot_name[1] != '.') { 1218 error = ENOTDIR; 1219 break; 1220 } 1221 if (dirbuf.dotdot_ino == source->i_number) { 1222 error = EINVAL; 1223 break; 1224 } 1225 if (dirbuf.dotdot_ino == EXT2_ROOTINO) 1226 break; 1227 vput(vp); 1228 if ((error = VFS_VGET(vp->v_mount, dirbuf.dotdot_ino, 1229 LK_EXCLUSIVE, &vp)) != 0) { 1230 vp = NULL; 1231 break; 1232 } 1233 } 1234 1235 out: 1236 if (error == ENOTDIR) 1237 printf("checkpath: .. not a directory\n"); 1238 if (vp != NULL) 1239 vput(vp); 1240 return (error); 1241 } 1242