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