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