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 u_int ncookies; 149 int DIRBLKSIZ = VTOI(ap->a_vp)->i_e2fs->e2fs_bsize; 150 int error; 151 152 if (uio->uio_offset < 0) 153 return (EINVAL); 154 ip = VTOI(vp); 155 if (ap->a_ncookies != NULL) { 156 if (uio->uio_resid < 0) 157 ncookies = 0; 158 else 159 ncookies = uio->uio_resid; 160 if (uio->uio_offset >= ip->i_size) 161 ncookies = 0; 162 else if (ip->i_size - uio->uio_offset < ncookies) 163 ncookies = ip->i_size - uio->uio_offset; 164 ncookies = ncookies / (offsetof(struct ext2fs_direct_2, 165 e2d_namlen) + 4) + 1; 166 cookies = malloc(ncookies * sizeof(*cookies), M_TEMP, M_WAITOK); 167 *ap->a_ncookies = ncookies; 168 *ap->a_cookies = cookies; 169 } else { 170 ncookies = 0; 171 cookies = NULL; 172 } 173 offset = startoffset = uio->uio_offset; 174 startresid = uio->uio_resid; 175 error = 0; 176 while (error == 0 && uio->uio_resid > 0 && 177 uio->uio_offset < ip->i_size) { 178 error = ext2_blkatoff(vp, uio->uio_offset, NULL, &bp); 179 if (error) 180 break; 181 if (bp->b_offset + bp->b_bcount > ip->i_size) 182 readcnt = ip->i_size - bp->b_offset; 183 else 184 readcnt = bp->b_bcount; 185 skipcnt = (size_t)(uio->uio_offset - bp->b_offset) & 186 ~(size_t)(DIRBLKSIZ - 1); 187 offset = bp->b_offset + skipcnt; 188 dp = (struct ext2fs_direct_2 *)&bp->b_data[skipcnt]; 189 edp = (struct ext2fs_direct_2 *)&bp->b_data[readcnt]; 190 while (error == 0 && uio->uio_resid > 0 && dp < edp) { 191 if (dp->e2d_reclen <= offsetof(struct ext2fs_direct_2, 192 e2d_namlen) || (caddr_t)dp + dp->e2d_reclen > 193 (caddr_t)edp) { 194 error = EIO; 195 break; 196 } 197 /*- 198 * "New" ext2fs directory entries differ in 3 ways 199 * from ufs on-disk ones: 200 * - the name is not necessarily NUL-terminated. 201 * - the file type field always exists and always 202 * follows the name length field. 203 * - the file type is encoded in a different way. 204 * 205 * "Old" ext2fs directory entries need no special 206 * conversions, since they are binary compatible 207 * with "new" entries having a file type of 0 (i.e., 208 * EXT2_FT_UNKNOWN). Splitting the old name length 209 * field didn't make a mess like it did in ufs, 210 * because ext2fs uses a machine-independent disk 211 * layout. 212 */ 213 dstdp.d_namlen = dp->e2d_namlen; 214 dstdp.d_type = FTTODT(dp->e2d_type); 215 if (offsetof(struct ext2fs_direct_2, e2d_namlen) + 216 dstdp.d_namlen > dp->e2d_reclen) { 217 error = EIO; 218 break; 219 } 220 if (offset < startoffset || dp->e2d_ino == 0) 221 goto nextentry; 222 dstdp.d_fileno = dp->e2d_ino; 223 dstdp.d_reclen = GENERIC_DIRSIZ(&dstdp); 224 bcopy(dp->e2d_name, dstdp.d_name, dstdp.d_namlen); 225 dstdp.d_name[dstdp.d_namlen] = '\0'; 226 if (dstdp.d_reclen > uio->uio_resid) { 227 if (uio->uio_resid == startresid) 228 error = EINVAL; 229 else 230 error = EJUSTRETURN; 231 break; 232 } 233 /* Advance dp. */ 234 error = uiomove((caddr_t)&dstdp, dstdp.d_reclen, uio); 235 if (error) 236 break; 237 if (cookies != NULL) { 238 KASSERT(ncookies > 0, 239 ("ext2_readdir: cookies buffer too small")); 240 *cookies = offset + dp->e2d_reclen; 241 cookies++; 242 ncookies--; 243 } 244 nextentry: 245 offset += dp->e2d_reclen; 246 dp = (struct ext2fs_direct_2 *)((caddr_t)dp + 247 dp->e2d_reclen); 248 } 249 bqrelse(bp); 250 uio->uio_offset = offset; 251 } 252 /* We need to correct uio_offset. */ 253 uio->uio_offset = offset; 254 if (error == EJUSTRETURN) 255 error = 0; 256 if (ap->a_ncookies != NULL) { 257 if (error == 0) { 258 ap->a_ncookies -= ncookies; 259 } else { 260 free(*ap->a_cookies, M_TEMP); 261 *ap->a_ncookies = 0; 262 *ap->a_cookies = NULL; 263 } 264 } 265 if (error == 0 && ap->a_eofflag) 266 *ap->a_eofflag = ip->i_size <= uio->uio_offset; 267 return (error); 268 } 269 270 /* 271 * Convert a component of a pathname into a pointer to a locked inode. 272 * This is a very central and rather complicated routine. 273 * If the file system is not maintained in a strict tree hierarchy, 274 * this can result in a deadlock situation (see comments in code below). 275 * 276 * The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending 277 * on whether the name is to be looked up, created, renamed, or deleted. 278 * When CREATE, RENAME, or DELETE is specified, information usable in 279 * creating, renaming, or deleting a directory entry may be calculated. 280 * If flag has LOCKPARENT or'ed into it and the target of the pathname 281 * exists, lookup returns both the target and its parent directory locked. 282 * When creating or renaming and LOCKPARENT is specified, the target may 283 * not be ".". When deleting and LOCKPARENT is specified, the target may 284 * be "."., but the caller must check to ensure it does an vrele and vput 285 * instead of two vputs. 286 * 287 * Overall outline of ext2_lookup: 288 * 289 * search for name in directory, to found or notfound 290 * notfound: 291 * if creating, return locked directory, leaving info on available slots 292 * else return error 293 * found: 294 * if at end of path and deleting, return information to allow delete 295 * if at end of path and rewriting (RENAME and LOCKPARENT), lock target 296 * inode and return info to allow rewrite 297 * if not at end, add name to cache; if at end and neither creating 298 * nor deleting, add name to cache 299 */ 300 int 301 ext2_lookup(struct vop_cachedlookup_args *ap) 302 { 303 304 return (ext2_lookup_ino(ap->a_dvp, ap->a_vpp, ap->a_cnp, NULL)); 305 } 306 307 static int 308 ext2_lookup_ino(struct vnode *vdp, struct vnode **vpp, struct componentname *cnp, 309 ino_t *dd_ino) 310 { 311 struct inode *dp; /* inode for directory being searched */ 312 struct buf *bp; /* a buffer of directory entries */ 313 struct ext2fs_direct_2 *ep; /* the current directory entry */ 314 int entryoffsetinblock; /* offset of ep in bp's buffer */ 315 struct ext2fs_searchslot ss; 316 doff_t i_diroff; /* cached i_diroff value */ 317 doff_t i_offset; /* cached i_offset value */ 318 int numdirpasses; /* strategy for directory search */ 319 doff_t endsearch; /* offset to end directory search */ 320 doff_t prevoff; /* prev entry dp->i_offset */ 321 struct vnode *pdp; /* saved dp during symlink work */ 322 struct vnode *tdp; /* returned by VFS_VGET */ 323 doff_t enduseful; /* pointer past last used dir slot */ 324 u_long bmask; /* block offset mask */ 325 int error; 326 struct ucred *cred = cnp->cn_cred; 327 int flags = cnp->cn_flags; 328 int nameiop = cnp->cn_nameiop; 329 ino_t ino, ino1; 330 int ltype; 331 int entry_found = 0; 332 333 int DIRBLKSIZ = VTOI(vdp)->i_e2fs->e2fs_bsize; 334 335 if (vpp != NULL) 336 *vpp = NULL; 337 338 dp = VTOI(vdp); 339 bmask = VFSTOEXT2(vdp->v_mount)->um_mountp->mnt_stat.f_iosize - 1; 340 restart: 341 bp = NULL; 342 ss.slotoffset = -1; 343 344 /* 345 * We now have a segment name to search for, and a directory to search. 346 * 347 * Suppress search for slots unless creating 348 * file and at end of pathname, in which case 349 * we watch for a place to put the new file in 350 * case it doesn't already exist. 351 */ 352 i_diroff = dp->i_diroff; 353 ss.slotstatus = FOUND; 354 ss.slotfreespace = ss.slotsize = ss.slotneeded = 0; 355 if ((nameiop == CREATE || nameiop == RENAME) && 356 (flags & ISLASTCN)) { 357 ss.slotstatus = NONE; 358 ss.slotneeded = EXT2_DIR_REC_LEN(cnp->cn_namelen); 359 /* 360 * was ss.slotneeded = (sizeof(struct direct) - MAXNAMLEN + 361 * cnp->cn_namelen + 3) &~ 3; 362 */ 363 } 364 /* 365 * Try to lookup dir entry using htree directory index. 366 * 367 * If we got an error or we want to find '.' or '..' entry, 368 * we will fall back to linear search. 369 */ 370 if (!ext2_is_dot_entry(cnp) && ext2_htree_has_idx(dp)) { 371 numdirpasses = 1; 372 entryoffsetinblock = 0; 373 switch (ext2_htree_lookup(dp, cnp->cn_nameptr, cnp->cn_namelen, 374 &bp, &entryoffsetinblock, &i_offset, &prevoff, 375 &enduseful, &ss)) { 376 case 0: 377 ep = (struct ext2fs_direct_2 *)((char *)bp->b_data + 378 (i_offset & bmask)); 379 goto foundentry; 380 case ENOENT: 381 i_offset = roundup2(dp->i_size, DIRBLKSIZ); 382 goto notfound; 383 default: 384 /* 385 * Something failed; just fallback to do a linear 386 * search. 387 */ 388 break; 389 } 390 } 391 392 /* 393 * If there is cached information on a previous search of 394 * this directory, pick up where we last left off. 395 * We cache only lookups as these are the most common 396 * and have the greatest payoff. Caching CREATE has little 397 * benefit as it usually must search the entire directory 398 * to determine that the entry does not exist. Caching the 399 * location of the last DELETE or RENAME has not reduced 400 * profiling time and hence has been removed in the interest 401 * of simplicity. 402 */ 403 if (nameiop != LOOKUP || i_diroff == 0 || 404 i_diroff > dp->i_size) { 405 entryoffsetinblock = 0; 406 i_offset = 0; 407 numdirpasses = 1; 408 } else { 409 i_offset = i_diroff; 410 if ((entryoffsetinblock = i_offset & bmask) && 411 (error = ext2_blkatoff(vdp, (off_t)i_offset, NULL, 412 &bp))) 413 return (error); 414 numdirpasses = 2; 415 nchstats.ncs_2passes++; 416 } 417 prevoff = i_offset; 418 endsearch = roundup2(dp->i_size, DIRBLKSIZ); 419 enduseful = 0; 420 421 searchloop: 422 while (i_offset < endsearch) { 423 /* 424 * If necessary, get the next directory block. 425 */ 426 if (bp != NULL) 427 brelse(bp); 428 error = ext2_blkatoff(vdp, (off_t)i_offset, NULL, &bp); 429 if (error != 0) 430 return (error); 431 entryoffsetinblock = 0; 432 /* 433 * If still looking for a slot, and at a DIRBLKSIZE 434 * boundary, have to start looking for free space again. 435 */ 436 if (ss.slotstatus == NONE && 437 (entryoffsetinblock & (DIRBLKSIZ - 1)) == 0) { 438 ss.slotoffset = -1; 439 ss.slotfreespace = 0; 440 } 441 error = ext2_search_dirblock(dp, bp->b_data, &entry_found, 442 cnp->cn_nameptr, cnp->cn_namelen, 443 &entryoffsetinblock, &i_offset, &prevoff, 444 &enduseful, &ss); 445 if (error != 0) { 446 brelse(bp); 447 return (error); 448 } 449 if (entry_found) { 450 ep = (struct ext2fs_direct_2 *)((char *)bp->b_data + 451 (entryoffsetinblock & bmask)); 452 foundentry: 453 ino = ep->e2d_ino; 454 goto found; 455 } 456 } 457 notfound: 458 /* 459 * If we started in the middle of the directory and failed 460 * to find our target, we must check the beginning as well. 461 */ 462 if (numdirpasses == 2) { 463 numdirpasses--; 464 i_offset = 0; 465 endsearch = i_diroff; 466 goto searchloop; 467 } 468 if (bp != NULL) 469 brelse(bp); 470 /* 471 * If creating, and at end of pathname and current 472 * directory has not been removed, then can consider 473 * allowing file to be created. 474 */ 475 if ((nameiop == CREATE || nameiop == RENAME) && 476 (flags & ISLASTCN) && dp->i_nlink != 0) { 477 /* 478 * Access for write is interpreted as allowing 479 * creation of files in the directory. 480 */ 481 if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread)) != 0) 482 return (error); 483 /* 484 * Return an indication of where the new directory 485 * entry should be put. If we didn't find a slot, 486 * then set dp->i_count to 0 indicating 487 * that the new slot belongs at the end of the 488 * directory. If we found a slot, then the new entry 489 * can be put in the range from dp->i_offset to 490 * dp->i_offset + dp->i_count. 491 */ 492 if (ss.slotstatus == NONE) { 493 dp->i_offset = roundup2(dp->i_size, DIRBLKSIZ); 494 dp->i_count = 0; 495 enduseful = dp->i_offset; 496 } else { 497 dp->i_offset = ss.slotoffset; 498 dp->i_count = ss.slotsize; 499 if (enduseful < ss.slotoffset + ss.slotsize) 500 enduseful = ss.slotoffset + ss.slotsize; 501 } 502 dp->i_endoff = roundup2(enduseful, DIRBLKSIZ); 503 /* 504 * We return with the directory locked, so that 505 * the parameters we set up above will still be 506 * valid if we actually decide to do a direnter(). 507 * We return ni_vp == NULL to indicate that the entry 508 * does not currently exist; we leave a pointer to 509 * the (locked) directory inode in ndp->ni_dvp. 510 * The pathname buffer is saved so that the name 511 * can be obtained later. 512 * 513 * NB - if the directory is unlocked, then this 514 * information cannot be used. 515 */ 516 cnp->cn_flags |= SAVENAME; 517 return (EJUSTRETURN); 518 } 519 /* 520 * Insert name into cache (as non-existent) if appropriate. 521 */ 522 if ((cnp->cn_flags & MAKEENTRY) != 0) 523 cache_enter(vdp, NULL, cnp); 524 return (ENOENT); 525 526 found: 527 if (dd_ino != NULL) 528 *dd_ino = ino; 529 if (numdirpasses == 2) 530 nchstats.ncs_pass2++; 531 /* 532 * Check that directory length properly reflects presence 533 * of this entry. 534 */ 535 if (entryoffsetinblock + EXT2_DIR_REC_LEN(ep->e2d_namlen) 536 > dp->i_size) { 537 ext2_dirbad(dp, i_offset, "i_size too small"); 538 dp->i_size = entryoffsetinblock + EXT2_DIR_REC_LEN(ep->e2d_namlen); 539 dp->i_flag |= IN_CHANGE | IN_UPDATE; 540 } 541 brelse(bp); 542 543 /* 544 * Found component in pathname. 545 * If the final component of path name, save information 546 * in the cache as to where the entry was found. 547 */ 548 if ((flags & ISLASTCN) && nameiop == LOOKUP) 549 dp->i_diroff = rounddown2(i_offset, DIRBLKSIZ); 550 /* 551 * If deleting, and at end of pathname, return 552 * parameters which can be used to remove file. 553 */ 554 if (nameiop == DELETE && (flags & ISLASTCN)) { 555 if (flags & LOCKPARENT) 556 ASSERT_VOP_ELOCKED(vdp, __FUNCTION__); 557 /* 558 * Write access to directory required to delete files. 559 */ 560 if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread)) != 0) 561 return (error); 562 /* 563 * Return pointer to current entry in dp->i_offset, 564 * and distance past previous entry (if there 565 * is a previous entry in this block) in dp->i_count. 566 * Save directory inode pointer in ndp->ni_dvp for dirremove(). 567 * 568 * Technically we shouldn't be setting these in the 569 * WANTPARENT case (first lookup in rename()), but any 570 * lookups that will result in directory changes will 571 * overwrite these. 572 */ 573 dp->i_offset = i_offset; 574 if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0) 575 dp->i_count = 0; 576 else 577 dp->i_count = dp->i_offset - prevoff; 578 if (dd_ino != NULL) 579 return (0); 580 if (dp->i_number == ino) { 581 VREF(vdp); 582 *vpp = vdp; 583 return (0); 584 } 585 if ((error = VFS_VGET(vdp->v_mount, ino, LK_EXCLUSIVE, 586 &tdp)) != 0) 587 return (error); 588 /* 589 * If directory is "sticky", then user must own 590 * the directory, or the file in it, else she 591 * may not delete it (unless she's root). This 592 * implements append-only directories. 593 */ 594 if ((dp->i_mode & ISVTX) && 595 cred->cr_uid != 0 && 596 cred->cr_uid != dp->i_uid && 597 VTOI(tdp)->i_uid != cred->cr_uid) { 598 vput(tdp); 599 return (EPERM); 600 } 601 *vpp = tdp; 602 return (0); 603 } 604 605 /* 606 * If rewriting (RENAME), return the inode and the 607 * information required to rewrite the present directory 608 * Must get inode of directory entry to verify it's a 609 * regular file, or empty directory. 610 */ 611 if (nameiop == RENAME && (flags & ISLASTCN)) { 612 if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread)) != 0) 613 return (error); 614 /* 615 * Careful about locking second inode. 616 * This can only occur if the target is ".". 617 */ 618 dp->i_offset = i_offset; 619 if (dp->i_number == ino) 620 return (EISDIR); 621 if (dd_ino != NULL) 622 return (0); 623 if ((error = VFS_VGET(vdp->v_mount, ino, LK_EXCLUSIVE, 624 &tdp)) != 0) 625 return (error); 626 *vpp = tdp; 627 cnp->cn_flags |= SAVENAME; 628 return (0); 629 } 630 if (dd_ino != NULL) 631 return (0); 632 633 /* 634 * Step through the translation in the name. We do not `vput' the 635 * directory because we may need it again if a symbolic link 636 * is relative to the current directory. Instead we save it 637 * unlocked as "pdp". We must get the target inode before unlocking 638 * the directory to insure that the inode will not be removed 639 * before we get it. We prevent deadlock by always fetching 640 * inodes from the root, moving down the directory tree. Thus 641 * when following backward pointers ".." we must unlock the 642 * parent directory before getting the requested directory. 643 * There is a potential race condition here if both the current 644 * and parent directories are removed before the VFS_VGET for the 645 * inode associated with ".." returns. We hope that this occurs 646 * infrequently since we cannot avoid this race condition without 647 * implementing a sophisticated deadlock detection algorithm. 648 * Note also that this simple deadlock detection scheme will not 649 * work if the file system has any hard links other than ".." 650 * that point backwards in the directory structure. 651 */ 652 pdp = vdp; 653 if (flags & ISDOTDOT) { 654 error = vn_vget_ino(pdp, ino, cnp->cn_lkflags, &tdp); 655 if (pdp->v_iflag & VI_DOOMED) { 656 if (error == 0) 657 vput(tdp); 658 error = ENOENT; 659 } 660 if (error) 661 return (error); 662 /* 663 * Recheck that ".." entry in the vdp directory points 664 * to the inode we looked up before vdp lock was 665 * dropped. 666 */ 667 error = ext2_lookup_ino(pdp, NULL, cnp, &ino1); 668 if (error) { 669 vput(tdp); 670 return (error); 671 } 672 if (ino1 != ino) { 673 vput(tdp); 674 goto restart; 675 } 676 *vpp = tdp; 677 } else if (dp->i_number == ino) { 678 VREF(vdp); /* we want ourself, ie "." */ 679 /* 680 * When we lookup "." we still can be asked to lock it 681 * differently. 682 */ 683 ltype = cnp->cn_lkflags & LK_TYPE_MASK; 684 if (ltype != VOP_ISLOCKED(vdp)) { 685 if (ltype == LK_EXCLUSIVE) 686 vn_lock(vdp, LK_UPGRADE | LK_RETRY); 687 else /* if (ltype == LK_SHARED) */ 688 vn_lock(vdp, LK_DOWNGRADE | LK_RETRY); 689 } 690 *vpp = vdp; 691 } else { 692 if ((error = VFS_VGET(vdp->v_mount, ino, cnp->cn_lkflags, 693 &tdp)) != 0) 694 return (error); 695 *vpp = tdp; 696 } 697 698 /* 699 * Insert name into cache if appropriate. 700 */ 701 if (cnp->cn_flags & MAKEENTRY) 702 cache_enter(vdp, *vpp, cnp); 703 return (0); 704 } 705 706 int 707 ext2_search_dirblock(struct inode *ip, void *data, int *foundp, 708 const char *name, int namelen, int *entryoffsetinblockp, 709 doff_t *offp, doff_t *prevoffp, doff_t *endusefulp, 710 struct ext2fs_searchslot *ssp) 711 { 712 struct vnode *vdp; 713 struct ext2fs_direct_2 *ep, *top; 714 uint32_t bsize = ip->i_e2fs->e2fs_bsize; 715 int offset = *entryoffsetinblockp; 716 int namlen; 717 718 vdp = ITOV(ip); 719 720 ep = (struct ext2fs_direct_2 *)((char *)data + offset); 721 top = (struct ext2fs_direct_2 *)((char *)data + 722 bsize - EXT2_DIR_REC_LEN(0)); 723 724 while (ep < top) { 725 /* 726 * Full validation checks are slow, so we only check 727 * enough to insure forward progress through the 728 * directory. Complete checks can be run by setting 729 * "vfs.e2fs.dirchk" to be true. 730 */ 731 if (ep->e2d_reclen == 0 || 732 (dirchk && ext2_dirbadentry(vdp, ep, offset))) { 733 int i; 734 735 ext2_dirbad(ip, *offp, "mangled entry"); 736 i = bsize - (offset & (bsize - 1)); 737 *offp += i; 738 offset += i; 739 continue; 740 } 741 742 /* 743 * If an appropriate sized slot has not yet been found, 744 * check to see if one is available. Also accumulate space 745 * in the current block so that we can determine if 746 * compaction is viable. 747 */ 748 if (ssp->slotstatus != FOUND) { 749 int size = ep->e2d_reclen; 750 751 if (ep->e2d_ino != 0) 752 size -= EXT2_DIR_REC_LEN(ep->e2d_namlen); 753 if (size > 0) { 754 if (size >= ssp->slotneeded) { 755 ssp->slotstatus = FOUND; 756 ssp->slotoffset = *offp; 757 ssp->slotsize = ep->e2d_reclen; 758 } else if (ssp->slotstatus == NONE) { 759 ssp->slotfreespace += size; 760 if (ssp->slotoffset == -1) 761 ssp->slotoffset = *offp; 762 if (ssp->slotfreespace >= ssp->slotneeded) { 763 ssp->slotstatus = COMPACT; 764 ssp->slotsize = *offp + 765 ep->e2d_reclen - 766 ssp->slotoffset; 767 } 768 } 769 } 770 } 771 /* 772 * Check for a name match. 773 */ 774 if (ep->e2d_ino) { 775 namlen = ep->e2d_namlen; 776 if (namlen == namelen && 777 !bcmp(name, ep->e2d_name, (unsigned)namlen)) { 778 /* 779 * Save directory entry's inode number and 780 * reclen in ndp->ni_ufs area, and release 781 * directory buffer. 782 */ 783 *foundp = 1; 784 return (0); 785 } 786 } 787 *prevoffp = *offp; 788 *offp += ep->e2d_reclen; 789 offset += ep->e2d_reclen; 790 *entryoffsetinblockp = offset; 791 if (ep->e2d_ino) 792 *endusefulp = *offp; 793 /* 794 * Get pointer to the next entry. 795 */ 796 ep = (struct ext2fs_direct_2 *)((char *)data + offset); 797 } 798 799 return (0); 800 } 801 802 void 803 ext2_dirbad(struct inode *ip, doff_t offset, char *how) 804 { 805 struct mount *mp; 806 807 mp = ITOV(ip)->v_mount; 808 if ((mp->mnt_flag & MNT_RDONLY) == 0) 809 panic("ext2_dirbad: %s: bad dir ino %ju at offset %ld: %s\n", 810 mp->mnt_stat.f_mntonname, (uintmax_t)ip->i_number, 811 (long)offset, how); 812 else 813 (void)printf("%s: bad dir ino %ju at offset %ld: %s\n", 814 mp->mnt_stat.f_mntonname, (uintmax_t)ip->i_number, 815 (long)offset, how); 816 817 } 818 819 /* 820 * Do consistency checking on a directory entry: 821 * record length must be multiple of 4 822 * entry must fit in rest of its DIRBLKSIZ block 823 * record must be large enough to contain entry 824 * name is not longer than MAXNAMLEN 825 * name must be as long as advertised, and null terminated 826 */ 827 /* 828 * changed so that it confirms to ext2_check_dir_entry 829 */ 830 static int 831 ext2_dirbadentry(struct vnode *dp, struct ext2fs_direct_2 *de, 832 int entryoffsetinblock) 833 { 834 int DIRBLKSIZ = VTOI(dp)->i_e2fs->e2fs_bsize; 835 836 char *error_msg = NULL; 837 838 if (de->e2d_reclen < EXT2_DIR_REC_LEN(1)) 839 error_msg = "rec_len is smaller than minimal"; 840 else if (de->e2d_reclen % 4 != 0) 841 error_msg = "rec_len % 4 != 0"; 842 else if (de->e2d_reclen < EXT2_DIR_REC_LEN(de->e2d_namlen)) 843 error_msg = "reclen is too small for name_len"; 844 else if (entryoffsetinblock + de->e2d_reclen > DIRBLKSIZ) 845 error_msg = "directory entry across blocks"; 846 /* else LATER 847 if (de->inode > dir->i_sb->u.ext2_sb.s_es->s_inodes_count) 848 error_msg = "inode out of bounds"; 849 */ 850 851 if (error_msg != NULL) { 852 printf("bad directory entry: %s\n", error_msg); 853 printf("offset=%d, inode=%lu, rec_len=%u, name_len=%u\n", 854 entryoffsetinblock, (unsigned long)de->e2d_ino, 855 de->e2d_reclen, de->e2d_namlen); 856 } 857 return error_msg == NULL ? 0 : 1; 858 } 859 860 /* 861 * Write a directory entry after a call to namei, using the parameters 862 * that it left in nameidata. The argument ip is the inode which the new 863 * directory entry will refer to. Dvp is a pointer to the directory to 864 * be written, which was left locked by namei. Remaining parameters 865 * (dp->i_offset, dp->i_count) indicate how the space for the new 866 * entry is to be obtained. 867 */ 868 int 869 ext2_direnter(struct inode *ip, struct vnode *dvp, struct componentname *cnp) 870 { 871 struct inode *dp; 872 struct ext2fs_direct_2 newdir; 873 struct buf *bp; 874 int DIRBLKSIZ = ip->i_e2fs->e2fs_bsize; 875 int error; 876 877 878 #ifdef INVARIANTS 879 if ((cnp->cn_flags & SAVENAME) == 0) 880 panic("ext2_direnter: missing name"); 881 #endif 882 dp = VTOI(dvp); 883 newdir.e2d_ino = ip->i_number; 884 newdir.e2d_namlen = cnp->cn_namelen; 885 if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs, 886 EXT2F_INCOMPAT_FTYPE)) 887 newdir.e2d_type = DTTOFT(IFTODT(ip->i_mode)); 888 else 889 newdir.e2d_type = EXT2_FT_UNKNOWN; 890 bcopy(cnp->cn_nameptr, newdir.e2d_name, (unsigned)cnp->cn_namelen + 1); 891 892 if (ext2_htree_has_idx(dp)) { 893 error = ext2_htree_add_entry(dvp, &newdir, cnp); 894 if (error) { 895 dp->i_flag &= ~IN_E3INDEX; 896 dp->i_flag |= IN_CHANGE | IN_UPDATE; 897 } 898 return (error); 899 } 900 901 if (EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_DIRHASHINDEX) && 902 !ext2_htree_has_idx(dp)) { 903 if ((dp->i_size / DIRBLKSIZ) == 1 && 904 dp->i_offset == DIRBLKSIZ) { 905 /* 906 * Making indexed directory when one block is not 907 * enough to save all entries. 908 */ 909 return ext2_htree_create_index(dvp, cnp, &newdir); 910 } 911 } 912 913 if (dp->i_count == 0) { 914 /* 915 * If dp->i_count is 0, then namei could find no 916 * space in the directory. Here, dp->i_offset will 917 * be on a directory block boundary and we will write the 918 * new entry into a fresh block. 919 */ 920 if (dp->i_offset & (DIRBLKSIZ - 1)) 921 panic("ext2_direnter: newblk"); 922 923 newdir.e2d_reclen = DIRBLKSIZ; 924 925 bp = getblk(ip->i_devvp, lblkno(dp->i_e2fs, dp->i_offset), 926 DIRBLKSIZ, 0, 0, 0); 927 if (!bp) 928 return (EIO); 929 930 memcpy(bp->b_data, &newdir, sizeof(struct ext2fs_direct_2)); 931 932 ext2_dir_blk_csum_set(dp, bp); 933 error = bwrite(bp); 934 if (error) 935 return (error); 936 937 dp->i_size = roundup2(dp->i_size, DIRBLKSIZ); 938 dp->i_flag |= IN_CHANGE; 939 940 return (0); 941 } 942 943 error = ext2_add_entry(dvp, &newdir); 944 if (!error && dp->i_endoff && dp->i_endoff < dp->i_size) 945 error = ext2_truncate(dvp, (off_t)dp->i_endoff, IO_SYNC, 946 cnp->cn_cred, cnp->cn_thread); 947 return (error); 948 } 949 950 /* 951 * Insert an entry into the directory block. 952 * Compact the contents. 953 */ 954 int 955 ext2_add_entry(struct vnode *dvp, struct ext2fs_direct_2 *entry) 956 { 957 struct ext2fs_direct_2 *ep, *nep; 958 struct inode *dp; 959 struct buf *bp; 960 u_int dsize; 961 int error, loc, newentrysize, spacefree; 962 char *dirbuf; 963 964 dp = VTOI(dvp); 965 966 /* 967 * If dp->i_count is non-zero, then namei found space 968 * for the new entry in the range dp->i_offset to 969 * dp->i_offset + dp->i_count in the directory. 970 * To use this space, we may have to compact the entries located 971 * there, by copying them together towards the beginning of the 972 * block, leaving the free space in one usable chunk at the end. 973 */ 974 975 /* 976 * Increase size of directory if entry eats into new space. 977 * This should never push the size past a new multiple of 978 * DIRBLKSIZE. 979 * 980 * N.B. - THIS IS AN ARTIFACT OF 4.2 AND SHOULD NEVER HAPPEN. 981 */ 982 if (dp->i_offset + dp->i_count > dp->i_size) 983 dp->i_size = dp->i_offset + dp->i_count; 984 /* 985 * Get the block containing the space for the new directory entry. 986 */ 987 if ((error = ext2_blkatoff(dvp, (off_t)dp->i_offset, &dirbuf, 988 &bp)) != 0) 989 return (error); 990 /* 991 * Find space for the new entry. In the simple case, the entry at 992 * offset base will have the space. If it does not, then namei 993 * arranged that compacting the region dp->i_offset to 994 * dp->i_offset + dp->i_count would yield the 995 * space. 996 */ 997 newentrysize = EXT2_DIR_REC_LEN(entry->e2d_namlen); 998 ep = (struct ext2fs_direct_2 *)dirbuf; 999 dsize = EXT2_DIR_REC_LEN(ep->e2d_namlen); 1000 spacefree = ep->e2d_reclen - dsize; 1001 for (loc = ep->e2d_reclen; loc < dp->i_count; ) { 1002 nep = (struct ext2fs_direct_2 *)(dirbuf + loc); 1003 if (ep->e2d_ino) { 1004 /* trim the existing slot */ 1005 ep->e2d_reclen = dsize; 1006 ep = (struct ext2fs_direct_2 *)((char *)ep + dsize); 1007 } else { 1008 /* overwrite; nothing there; header is ours */ 1009 spacefree += dsize; 1010 } 1011 dsize = EXT2_DIR_REC_LEN(nep->e2d_namlen); 1012 spacefree += nep->e2d_reclen - dsize; 1013 loc += nep->e2d_reclen; 1014 bcopy((caddr_t)nep, (caddr_t)ep, dsize); 1015 } 1016 /* 1017 * Update the pointer fields in the previous entry (if any), 1018 * copy in the new entry, and write out the block. 1019 */ 1020 if (ep->e2d_ino == 0) { 1021 if (spacefree + dsize < newentrysize) 1022 panic("ext2_direnter: compact1"); 1023 entry->e2d_reclen = spacefree + dsize; 1024 } else { 1025 if (spacefree < newentrysize) 1026 panic("ext2_direnter: compact2"); 1027 entry->e2d_reclen = spacefree; 1028 ep->e2d_reclen = dsize; 1029 ep = (struct ext2fs_direct_2 *)((char *)ep + dsize); 1030 } 1031 bcopy((caddr_t)entry, (caddr_t)ep, (u_int)newentrysize); 1032 ext2_dir_blk_csum_set(dp, bp); 1033 if (DOINGASYNC(dvp)) { 1034 bdwrite(bp); 1035 error = 0; 1036 } else { 1037 error = bwrite(bp); 1038 } 1039 dp->i_flag |= IN_CHANGE | IN_UPDATE; 1040 return (error); 1041 } 1042 1043 /* 1044 * Remove a directory entry after a call to namei, using 1045 * the parameters which it left in nameidata. The entry 1046 * dp->i_offset contains the offset into the directory of the 1047 * entry to be eliminated. The dp->i_count field contains the 1048 * size of the previous record in the directory. If this 1049 * is 0, the first entry is being deleted, so we need only 1050 * zero the inode number to mark the entry as free. If the 1051 * entry is not the first in the directory, we must reclaim 1052 * the space of the now empty record by adding the record size 1053 * to the size of the previous entry. 1054 */ 1055 int 1056 ext2_dirremove(struct vnode *dvp, struct componentname *cnp) 1057 { 1058 struct inode *dp; 1059 struct ext2fs_direct_2 *ep, *rep; 1060 struct buf *bp; 1061 int error; 1062 1063 dp = VTOI(dvp); 1064 if (dp->i_count == 0) { 1065 /* 1066 * First entry in block: set d_ino to zero. 1067 */ 1068 if ((error = 1069 ext2_blkatoff(dvp, (off_t)dp->i_offset, (char **)&ep, 1070 &bp)) != 0) 1071 return (error); 1072 ep->e2d_ino = 0; 1073 error = bwrite(bp); 1074 dp->i_flag |= IN_CHANGE | IN_UPDATE; 1075 return (error); 1076 } 1077 /* 1078 * Collapse new free space into previous entry. 1079 */ 1080 if ((error = ext2_blkatoff(dvp, (off_t)(dp->i_offset - dp->i_count), 1081 (char **)&ep, &bp)) != 0) 1082 return (error); 1083 1084 /* Set 'rep' to the entry being removed. */ 1085 if (dp->i_count == 0) 1086 rep = ep; 1087 else 1088 rep = (struct ext2fs_direct_2 *)((char *)ep + ep->e2d_reclen); 1089 ep->e2d_reclen += rep->e2d_reclen; 1090 ext2_dir_blk_csum_set(dp, bp); 1091 if (DOINGASYNC(dvp) && dp->i_count != 0) 1092 bdwrite(bp); 1093 else 1094 error = bwrite(bp); 1095 dp->i_flag |= IN_CHANGE | IN_UPDATE; 1096 return (error); 1097 } 1098 1099 /* 1100 * Rewrite an existing directory entry to point at the inode 1101 * supplied. The parameters describing the directory entry are 1102 * set up by a call to namei. 1103 */ 1104 int 1105 ext2_dirrewrite(struct inode *dp, struct inode *ip, struct componentname *cnp) 1106 { 1107 struct buf *bp; 1108 struct ext2fs_direct_2 *ep; 1109 struct vnode *vdp = ITOV(dp); 1110 int error; 1111 1112 if ((error = ext2_blkatoff(vdp, (off_t)dp->i_offset, (char **)&ep, 1113 &bp)) != 0) 1114 return (error); 1115 ep->e2d_ino = ip->i_number; 1116 if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs, 1117 EXT2F_INCOMPAT_FTYPE)) 1118 ep->e2d_type = DTTOFT(IFTODT(ip->i_mode)); 1119 else 1120 ep->e2d_type = EXT2_FT_UNKNOWN; 1121 ext2_dir_blk_csum_set(dp, bp); 1122 error = bwrite(bp); 1123 dp->i_flag |= IN_CHANGE | IN_UPDATE; 1124 return (error); 1125 } 1126 1127 /* 1128 * Check if a directory is empty or not. 1129 * Inode supplied must be locked. 1130 * 1131 * Using a struct dirtemplate here is not precisely 1132 * what we want, but better than using a struct direct. 1133 * 1134 * NB: does not handle corrupted directories. 1135 */ 1136 int 1137 ext2_dirempty(struct inode *ip, ino_t parentino, struct ucred *cred) 1138 { 1139 off_t off; 1140 struct dirtemplate dbuf; 1141 struct ext2fs_direct_2 *dp = (struct ext2fs_direct_2 *)&dbuf; 1142 int error, namlen; 1143 ssize_t count; 1144 #define MINDIRSIZ (sizeof(struct dirtemplate) / 2) 1145 1146 for (off = 0; off < ip->i_size; off += dp->e2d_reclen) { 1147 error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ, 1148 off, UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK, cred, 1149 NOCRED, &count, (struct thread *)0); 1150 /* 1151 * Since we read MINDIRSIZ, residual must 1152 * be 0 unless we're at end of file. 1153 */ 1154 if (error || count != 0) 1155 return (0); 1156 /* avoid infinite loops */ 1157 if (dp->e2d_reclen == 0) 1158 return (0); 1159 /* skip empty entries */ 1160 if (dp->e2d_ino == 0) 1161 continue; 1162 /* accept only "." and ".." */ 1163 namlen = dp->e2d_namlen; 1164 if (namlen > 2) 1165 return (0); 1166 if (dp->e2d_name[0] != '.') 1167 return (0); 1168 /* 1169 * At this point namlen must be 1 or 2. 1170 * 1 implies ".", 2 implies ".." if second 1171 * char is also "." 1172 */ 1173 if (namlen == 1) 1174 continue; 1175 if (dp->e2d_name[1] == '.' && dp->e2d_ino == parentino) 1176 continue; 1177 return (0); 1178 } 1179 return (1); 1180 } 1181 1182 /* 1183 * Check if source directory is in the path of the target directory. 1184 * Target is supplied locked, source is unlocked. 1185 * The target is always vput before returning. 1186 */ 1187 int 1188 ext2_checkpath(struct inode *source, struct inode *target, struct ucred *cred) 1189 { 1190 struct vnode *vp; 1191 int error, namlen; 1192 struct dirtemplate dirbuf; 1193 1194 vp = ITOV(target); 1195 if (target->i_number == source->i_number) { 1196 error = EEXIST; 1197 goto out; 1198 } 1199 if (target->i_number == EXT2_ROOTINO) { 1200 error = 0; 1201 goto out; 1202 } 1203 1204 for (;;) { 1205 if (vp->v_type != VDIR) { 1206 error = ENOTDIR; 1207 break; 1208 } 1209 error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf, 1210 sizeof(struct dirtemplate), (off_t)0, UIO_SYSSPACE, 1211 IO_NODELOCKED | IO_NOMACCHECK, cred, NOCRED, NULL, 1212 NULL); 1213 if (error != 0) 1214 break; 1215 namlen = dirbuf.dotdot_type; /* like ufs little-endian */ 1216 if (namlen != 2 || 1217 dirbuf.dotdot_name[0] != '.' || 1218 dirbuf.dotdot_name[1] != '.') { 1219 error = ENOTDIR; 1220 break; 1221 } 1222 if (dirbuf.dotdot_ino == source->i_number) { 1223 error = EINVAL; 1224 break; 1225 } 1226 if (dirbuf.dotdot_ino == EXT2_ROOTINO) 1227 break; 1228 vput(vp); 1229 if ((error = VFS_VGET(vp->v_mount, dirbuf.dotdot_ino, 1230 LK_EXCLUSIVE, &vp)) != 0) { 1231 vp = NULL; 1232 break; 1233 } 1234 } 1235 1236 out: 1237 if (error == ENOTDIR) 1238 printf("checkpath: .. not a directory\n"); 1239 if (vp != NULL) 1240 vput(vp); 1241 return (error); 1242 } 1243