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 * 4. 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 ino = 0; 347 i_diroff = dp->i_diroff; 348 ss.slotstatus = FOUND; 349 ss.slotfreespace = ss.slotsize = ss.slotneeded = 0; 350 if ((nameiop == CREATE || nameiop == RENAME) && 351 (flags & ISLASTCN)) { 352 ss.slotstatus = NONE; 353 ss.slotneeded = EXT2_DIR_REC_LEN(cnp->cn_namelen); 354 /* was 355 ss.slotneeded = (sizeof(struct direct) - MAXNAMLEN + 356 cnp->cn_namelen + 3) &~ 3; */ 357 } 358 359 /* 360 * Try to lookup dir entry using htree directory index. 361 * 362 * If we got an error or we want to find '.' or '..' entry, 363 * we will fall back to linear search. 364 */ 365 if (!ext2_is_dot_entry(cnp) && ext2_htree_has_idx(dp)) { 366 numdirpasses = 1; 367 entryoffsetinblock = 0; 368 switch (ext2_htree_lookup(dp, cnp->cn_nameptr, cnp->cn_namelen, 369 &bp, &entryoffsetinblock, &i_offset, &prevoff, 370 &enduseful, &ss)) { 371 case 0: 372 ep = (struct ext2fs_direct_2 *)((char *)bp->b_data + 373 (i_offset & bmask)); 374 goto foundentry; 375 case ENOENT: 376 i_offset = roundup2(dp->i_size, DIRBLKSIZ); 377 goto notfound; 378 default: 379 /* 380 * Something failed; just fallback to do a linear 381 * search. 382 */ 383 break; 384 } 385 } 386 387 /* 388 * If there is cached information on a previous search of 389 * this directory, pick up where we last left off. 390 * We cache only lookups as these are the most common 391 * and have the greatest payoff. Caching CREATE has little 392 * benefit as it usually must search the entire directory 393 * to determine that the entry does not exist. Caching the 394 * location of the last DELETE or RENAME has not reduced 395 * profiling time and hence has been removed in the interest 396 * of simplicity. 397 */ 398 if (nameiop != LOOKUP || i_diroff == 0 || 399 i_diroff > dp->i_size) { 400 entryoffsetinblock = 0; 401 i_offset = 0; 402 numdirpasses = 1; 403 } else { 404 i_offset = i_diroff; 405 if ((entryoffsetinblock = i_offset & bmask) && 406 (error = ext2_blkatoff(vdp, (off_t)i_offset, NULL, 407 &bp))) 408 return (error); 409 numdirpasses = 2; 410 nchstats.ncs_2passes++; 411 } 412 prevoff = i_offset; 413 endsearch = roundup2(dp->i_size, DIRBLKSIZ); 414 enduseful = 0; 415 416 searchloop: 417 while (i_offset < endsearch) { 418 /* 419 * If necessary, get the next directory block. 420 */ 421 if (bp != NULL) 422 brelse(bp); 423 error = ext2_blkatoff(vdp, (off_t)i_offset, NULL, &bp); 424 if (error != 0) 425 return (error); 426 entryoffsetinblock = 0; 427 /* 428 * If still looking for a slot, and at a DIRBLKSIZE 429 * boundary, have to start looking for free space again. 430 */ 431 if (ss.slotstatus == NONE && 432 (entryoffsetinblock & (DIRBLKSIZ - 1)) == 0) { 433 ss.slotoffset = -1; 434 ss.slotfreespace = 0; 435 } 436 error = ext2_search_dirblock(dp, bp->b_data, &entry_found, 437 cnp->cn_nameptr, cnp->cn_namelen, 438 &entryoffsetinblock, &i_offset, &prevoff, 439 &enduseful, &ss); 440 if (error != 0) { 441 brelse(bp); 442 return (error); 443 } 444 if (entry_found) { 445 ep = (struct ext2fs_direct_2 *)((char *)bp->b_data + 446 (entryoffsetinblock & bmask)); 447 foundentry: 448 ino = ep->e2d_ino; 449 goto found; 450 } 451 } 452 notfound: 453 /* 454 * If we started in the middle of the directory and failed 455 * to find our target, we must check the beginning as well. 456 */ 457 if (numdirpasses == 2) { 458 numdirpasses--; 459 i_offset = 0; 460 endsearch = i_diroff; 461 goto searchloop; 462 } 463 if (bp != NULL) 464 brelse(bp); 465 /* 466 * If creating, and at end of pathname and current 467 * directory has not been removed, then can consider 468 * allowing file to be created. 469 */ 470 if ((nameiop == CREATE || nameiop == RENAME) && 471 (flags & ISLASTCN) && dp->i_nlink != 0) { 472 /* 473 * Access for write is interpreted as allowing 474 * creation of files in the directory. 475 */ 476 if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread)) != 0) 477 return (error); 478 /* 479 * Return an indication of where the new directory 480 * entry should be put. If we didn't find a slot, 481 * then set dp->i_count to 0 indicating 482 * that the new slot belongs at the end of the 483 * directory. If we found a slot, then the new entry 484 * can be put in the range from dp->i_offset to 485 * dp->i_offset + dp->i_count. 486 */ 487 if (ss.slotstatus == NONE) { 488 dp->i_offset = roundup2(dp->i_size, DIRBLKSIZ); 489 dp->i_count = 0; 490 enduseful = dp->i_offset; 491 } else { 492 dp->i_offset = ss.slotoffset; 493 dp->i_count = ss.slotsize; 494 if (enduseful < ss.slotoffset + ss.slotsize) 495 enduseful = ss.slotoffset + ss.slotsize; 496 } 497 dp->i_endoff = roundup2(enduseful, DIRBLKSIZ); 498 /* 499 * We return with the directory locked, so that 500 * the parameters we set up above will still be 501 * valid if we actually decide to do a direnter(). 502 * We return ni_vp == NULL to indicate that the entry 503 * does not currently exist; we leave a pointer to 504 * the (locked) directory inode in ndp->ni_dvp. 505 * The pathname buffer is saved so that the name 506 * can be obtained later. 507 * 508 * NB - if the directory is unlocked, then this 509 * information cannot be used. 510 */ 511 cnp->cn_flags |= SAVENAME; 512 return (EJUSTRETURN); 513 } 514 /* 515 * Insert name into cache (as non-existent) if appropriate. 516 */ 517 if ((cnp->cn_flags & MAKEENTRY) && nameiop != CREATE) 518 cache_enter(vdp, NULL, cnp); 519 return (ENOENT); 520 521 found: 522 if (dd_ino != NULL) 523 *dd_ino = ino; 524 if (numdirpasses == 2) 525 nchstats.ncs_pass2++; 526 /* 527 * Check that directory length properly reflects presence 528 * of this entry. 529 */ 530 if (entryoffsetinblock + EXT2_DIR_REC_LEN(ep->e2d_namlen) 531 > dp->i_size) { 532 ext2_dirbad(dp, i_offset, "i_size too small"); 533 dp->i_size = entryoffsetinblock+EXT2_DIR_REC_LEN(ep->e2d_namlen); 534 dp->i_flag |= IN_CHANGE | IN_UPDATE; 535 } 536 brelse(bp); 537 538 /* 539 * Found component in pathname. 540 * If the final component of path name, save information 541 * in the cache as to where the entry was found. 542 */ 543 if ((flags & ISLASTCN) && nameiop == LOOKUP) 544 dp->i_diroff = i_offset &~ (DIRBLKSIZ - 1); 545 /* 546 * If deleting, and at end of pathname, return 547 * parameters which can be used to remove file. 548 */ 549 if (nameiop == DELETE && (flags & ISLASTCN)) { 550 if (flags & LOCKPARENT) 551 ASSERT_VOP_ELOCKED(vdp, __FUNCTION__); 552 /* 553 * Write access to directory required to delete files. 554 */ 555 if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread)) != 0) 556 return (error); 557 /* 558 * Return pointer to current entry in dp->i_offset, 559 * and distance past previous entry (if there 560 * is a previous entry in this block) in dp->i_count. 561 * Save directory inode pointer in ndp->ni_dvp for dirremove(). 562 * 563 * Technically we shouldn't be setting these in the 564 * WANTPARENT case (first lookup in rename()), but any 565 * lookups that will result in directory changes will 566 * overwrite these. 567 */ 568 dp->i_offset = i_offset; 569 if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0) 570 dp->i_count = 0; 571 else 572 dp->i_count = dp->i_offset - prevoff; 573 if (dd_ino != NULL) 574 return (0); 575 if (dp->i_number == ino) { 576 VREF(vdp); 577 *vpp = vdp; 578 return (0); 579 } 580 if ((error = VFS_VGET(vdp->v_mount, ino, LK_EXCLUSIVE, 581 &tdp)) != 0) 582 return (error); 583 /* 584 * If directory is "sticky", then user must own 585 * the directory, or the file in it, else she 586 * may not delete it (unless she's root). This 587 * implements append-only directories. 588 */ 589 if ((dp->i_mode & ISVTX) && 590 cred->cr_uid != 0 && 591 cred->cr_uid != dp->i_uid && 592 VTOI(tdp)->i_uid != cred->cr_uid) { 593 vput(tdp); 594 return (EPERM); 595 } 596 *vpp = tdp; 597 return (0); 598 } 599 600 /* 601 * If rewriting (RENAME), return the inode and the 602 * information required to rewrite the present directory 603 * Must get inode of directory entry to verify it's a 604 * regular file, or empty directory. 605 */ 606 if (nameiop == RENAME && (flags & ISLASTCN)) { 607 if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread)) != 0) 608 return (error); 609 /* 610 * Careful about locking second inode. 611 * This can only occur if the target is ".". 612 */ 613 dp->i_offset = i_offset; 614 if (dp->i_number == ino) 615 return (EISDIR); 616 if (dd_ino != NULL) 617 return (0); 618 if ((error = VFS_VGET(vdp->v_mount, ino, LK_EXCLUSIVE, 619 &tdp)) != 0) 620 return (error); 621 *vpp = tdp; 622 cnp->cn_flags |= SAVENAME; 623 return (0); 624 } 625 if (dd_ino != NULL) 626 return (0); 627 628 /* 629 * Step through the translation in the name. We do not `vput' the 630 * directory because we may need it again if a symbolic link 631 * is relative to the current directory. Instead we save it 632 * unlocked as "pdp". We must get the target inode before unlocking 633 * the directory to insure that the inode will not be removed 634 * before we get it. We prevent deadlock by always fetching 635 * inodes from the root, moving down the directory tree. Thus 636 * when following backward pointers ".." we must unlock the 637 * parent directory before getting the requested directory. 638 * There is a potential race condition here if both the current 639 * and parent directories are removed before the VFS_VGET for the 640 * inode associated with ".." returns. We hope that this occurs 641 * infrequently since we cannot avoid this race condition without 642 * implementing a sophisticated deadlock detection algorithm. 643 * Note also that this simple deadlock detection scheme will not 644 * work if the file system has any hard links other than ".." 645 * that point backwards in the directory structure. 646 */ 647 pdp = vdp; 648 if (flags & ISDOTDOT) { 649 error = vn_vget_ino(pdp, ino, cnp->cn_lkflags, &tdp); 650 if (pdp->v_iflag & VI_DOOMED) { 651 if (error == 0) 652 vput(tdp); 653 error = ENOENT; 654 } 655 if (error) 656 return (error); 657 /* 658 * Recheck that ".." entry in the vdp directory points 659 * to the inode we looked up before vdp lock was 660 * dropped. 661 */ 662 error = ext2_lookup_ino(pdp, NULL, cnp, &ino1); 663 if (error) { 664 vput(tdp); 665 return (error); 666 } 667 if (ino1 != ino) { 668 vput(tdp); 669 goto restart; 670 } 671 *vpp = tdp; 672 } else if (dp->i_number == ino) { 673 VREF(vdp); /* we want ourself, ie "." */ 674 /* 675 * When we lookup "." we still can be asked to lock it 676 * differently. 677 */ 678 ltype = cnp->cn_lkflags & LK_TYPE_MASK; 679 if (ltype != VOP_ISLOCKED(vdp)) { 680 if (ltype == LK_EXCLUSIVE) 681 vn_lock(vdp, LK_UPGRADE | LK_RETRY); 682 else /* if (ltype == LK_SHARED) */ 683 vn_lock(vdp, LK_DOWNGRADE | LK_RETRY); 684 } 685 *vpp = vdp; 686 } else { 687 if ((error = VFS_VGET(vdp->v_mount, ino, cnp->cn_lkflags, 688 &tdp)) != 0) 689 return (error); 690 *vpp = tdp; 691 } 692 693 /* 694 * Insert name into cache if appropriate. 695 */ 696 if (cnp->cn_flags & MAKEENTRY) 697 cache_enter(vdp, *vpp, cnp); 698 return (0); 699 } 700 701 int 702 ext2_search_dirblock(struct inode *ip, void *data, int *foundp, 703 const char *name, int namelen, int *entryoffsetinblockp, 704 doff_t *offp, doff_t *prevoffp, doff_t *endusefulp, 705 struct ext2fs_searchslot *ssp) 706 { 707 struct vnode *vdp; 708 struct ext2fs_direct_2 *ep, *top; 709 uint32_t bsize = ip->i_e2fs->e2fs_bsize; 710 int offset = *entryoffsetinblockp; 711 int namlen; 712 713 vdp = ITOV(ip); 714 715 ep = (struct ext2fs_direct_2 *)((char *)data + offset); 716 top = (struct ext2fs_direct_2 *)((char *)data + 717 bsize - EXT2_DIR_REC_LEN(0)); 718 719 while (ep < top) { 720 /* 721 * Full validation checks are slow, so we only check 722 * enough to insure forward progress through the 723 * directory. Complete checks can be run by setting 724 * "vfs.e2fs.dirchk" to be true. 725 */ 726 if (ep->e2d_reclen == 0 || 727 (dirchk && ext2_dirbadentry(vdp, ep, offset))) { 728 int i; 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 /* 767 * Check for a name match. 768 */ 769 if (ep->e2d_ino) { 770 namlen = ep->e2d_namlen; 771 if (namlen == namelen && 772 !bcmp(name, ep->e2d_name, (unsigned)namlen)) { 773 /* 774 * Save directory entry's inode number and 775 * reclen in ndp->ni_ufs area, and release 776 * directory buffer. 777 */ 778 *foundp = 1; 779 return (0); 780 } 781 } 782 *prevoffp = *offp; 783 *offp += ep->e2d_reclen; 784 offset += ep->e2d_reclen; 785 *entryoffsetinblockp = offset; 786 if (ep->e2d_ino) 787 *endusefulp = *offp; 788 /* 789 * Get pointer to the next entry. 790 */ 791 ep = (struct ext2fs_direct_2 *)((char *)data + offset); 792 } 793 794 return (0); 795 } 796 797 void 798 ext2_dirbad(struct inode *ip, doff_t offset, char *how) 799 { 800 struct mount *mp; 801 802 mp = ITOV(ip)->v_mount; 803 if ((mp->mnt_flag & MNT_RDONLY) == 0) 804 panic("ext2_dirbad: %s: bad dir ino %lu at offset %ld: %s\n", 805 mp->mnt_stat.f_mntonname, (u_long)ip->i_number,(long)offset, how); 806 else 807 (void)printf("%s: bad dir ino %lu at offset %ld: %s\n", 808 mp->mnt_stat.f_mntonname, (u_long)ip->i_number, (long)offset, how); 809 810 } 811 812 /* 813 * Do consistency checking on a directory entry: 814 * record length must be multiple of 4 815 * entry must fit in rest of its DIRBLKSIZ block 816 * record must be large enough to contain entry 817 * name is not longer than MAXNAMLEN 818 * name must be as long as advertised, and null terminated 819 */ 820 /* 821 * changed so that it confirms to ext2_check_dir_entry 822 */ 823 static int 824 ext2_dirbadentry(struct vnode *dp, struct ext2fs_direct_2 *de, 825 int entryoffsetinblock) 826 { 827 int DIRBLKSIZ = VTOI(dp)->i_e2fs->e2fs_bsize; 828 829 char * error_msg = NULL; 830 831 if (de->e2d_reclen < EXT2_DIR_REC_LEN(1)) 832 error_msg = "rec_len is smaller than minimal"; 833 else if (de->e2d_reclen % 4 != 0) 834 error_msg = "rec_len % 4 != 0"; 835 else if (de->e2d_reclen < EXT2_DIR_REC_LEN(de->e2d_namlen)) 836 error_msg = "reclen is too small for name_len"; 837 else if (entryoffsetinblock + de->e2d_reclen > DIRBLKSIZ) 838 error_msg = "directory entry across blocks"; 839 /* else LATER 840 if (de->inode > dir->i_sb->u.ext2_sb.s_es->s_inodes_count) 841 error_msg = "inode out of bounds"; 842 */ 843 844 if (error_msg != NULL) { 845 printf("bad directory entry: %s\n", error_msg); 846 printf("offset=%d, inode=%lu, rec_len=%u, name_len=%u\n", 847 entryoffsetinblock, (unsigned long)de->e2d_ino, 848 de->e2d_reclen, de->e2d_namlen); 849 } 850 return error_msg == NULL ? 0 : 1; 851 } 852 853 /* 854 * Write a directory entry after a call to namei, using the parameters 855 * that it left in nameidata. The argument ip is the inode which the new 856 * directory entry will refer to. Dvp is a pointer to the directory to 857 * be written, which was left locked by namei. Remaining parameters 858 * (dp->i_offset, dp->i_count) indicate how the space for the new 859 * entry is to be obtained. 860 */ 861 int 862 ext2_direnter(struct inode *ip, struct vnode *dvp, struct componentname *cnp) 863 { 864 struct inode *dp; 865 struct ext2fs_direct_2 newdir; 866 struct iovec aiov; 867 struct uio auio; 868 int error, newentrysize; 869 int DIRBLKSIZ = ip->i_e2fs->e2fs_bsize; 870 871 872 #ifdef INVARIANTS 873 if ((cnp->cn_flags & SAVENAME) == 0) 874 panic("ext2_direnter: missing name"); 875 #endif 876 dp = VTOI(dvp); 877 newdir.e2d_ino = ip->i_number; 878 newdir.e2d_namlen = cnp->cn_namelen; 879 if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs, 880 EXT2F_INCOMPAT_FTYPE)) 881 newdir.e2d_type = DTTOFT(IFTODT(ip->i_mode)); 882 else 883 newdir.e2d_type = EXT2_FT_UNKNOWN; 884 bcopy(cnp->cn_nameptr, newdir.e2d_name, (unsigned)cnp->cn_namelen + 1); 885 newentrysize = EXT2_DIR_REC_LEN(newdir.e2d_namlen); 886 887 if (ext2_htree_has_idx(dp)) { 888 error = ext2_htree_add_entry(dvp, &newdir, cnp); 889 if (error) { 890 dp->i_flag &= ~IN_E4INDEX; 891 dp->i_flag |= IN_CHANGE | IN_UPDATE; 892 } 893 return (error); 894 } 895 896 if (EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_DIRHASHINDEX) && 897 !ext2_htree_has_idx(dp)) { 898 if ((dp->i_size / DIRBLKSIZ) == 1 && 899 dp->i_offset == DIRBLKSIZ) { 900 /* 901 * Making indexed directory when one block is not 902 * enough to save all entries. 903 */ 904 return ext2_htree_create_index(dvp, cnp, &newdir); 905 } 906 } 907 908 if (dp->i_count == 0) { 909 /* 910 * If dp->i_count is 0, then namei could find no 911 * space in the directory. Here, dp->i_offset will 912 * be on a directory block boundary and we will write the 913 * new entry into a fresh block. 914 */ 915 if (dp->i_offset & (DIRBLKSIZ - 1)) 916 panic("ext2_direnter: newblk"); 917 auio.uio_offset = dp->i_offset; 918 newdir.e2d_reclen = DIRBLKSIZ; 919 auio.uio_resid = newentrysize; 920 aiov.iov_len = newentrysize; 921 aiov.iov_base = (caddr_t)&newdir; 922 auio.uio_iov = &aiov; 923 auio.uio_iovcnt = 1; 924 auio.uio_rw = UIO_WRITE; 925 auio.uio_segflg = UIO_SYSSPACE; 926 auio.uio_td = (struct thread *)0; 927 error = VOP_WRITE(dvp, &auio, IO_SYNC, cnp->cn_cred); 928 if (DIRBLKSIZ > 929 VFSTOEXT2(dvp->v_mount)->um_mountp->mnt_stat.f_bsize) 930 /* XXX should grow with balloc() */ 931 panic("ext2_direnter: frag size"); 932 else if (!error) { 933 dp->i_size = roundup2(dp->i_size, DIRBLKSIZ); 934 dp->i_flag |= IN_CHANGE; 935 } 936 return (error); 937 } 938 939 error = ext2_add_entry(dvp, &newdir); 940 if (!error && dp->i_endoff && dp->i_endoff < dp->i_size) 941 error = ext2_truncate(dvp, (off_t)dp->i_endoff, IO_SYNC, 942 cnp->cn_cred, cnp->cn_thread); 943 return (error); 944 } 945 946 /* 947 * Insert an entry into the directory block. 948 * Compact the contents. 949 */ 950 int 951 ext2_add_entry(struct vnode *dvp, struct ext2fs_direct_2 *entry) 952 { 953 struct ext2fs_direct_2 *ep, *nep; 954 struct inode *dp; 955 struct buf *bp; 956 u_int dsize; 957 int error, loc, newentrysize, spacefree; 958 char *dirbuf; 959 960 dp = VTOI(dvp); 961 962 /* 963 * If dp->i_count is non-zero, then namei found space 964 * for the new entry in the range dp->i_offset to 965 * dp->i_offset + dp->i_count in the directory. 966 * To use this space, we may have to compact the entries located 967 * there, by copying them together towards the beginning of the 968 * block, leaving the free space in one usable chunk at the end. 969 */ 970 971 /* 972 * Increase size of directory if entry eats into new space. 973 * This should never push the size past a new multiple of 974 * DIRBLKSIZE. 975 * 976 * N.B. - THIS IS AN ARTIFACT OF 4.2 AND SHOULD NEVER HAPPEN. 977 */ 978 if (dp->i_offset + dp->i_count > dp->i_size) 979 dp->i_size = dp->i_offset + dp->i_count; 980 /* 981 * Get the block containing the space for the new directory entry. 982 */ 983 if ((error = ext2_blkatoff(dvp, (off_t)dp->i_offset, &dirbuf, 984 &bp)) != 0) 985 return (error); 986 /* 987 * Find space for the new entry. In the simple case, the entry at 988 * offset base will have the space. If it does not, then namei 989 * arranged that compacting the region dp->i_offset to 990 * dp->i_offset + dp->i_count would yield the 991 * space. 992 */ 993 newentrysize = EXT2_DIR_REC_LEN(entry->e2d_namlen); 994 ep = (struct ext2fs_direct_2 *)dirbuf; 995 dsize = EXT2_DIR_REC_LEN(ep->e2d_namlen); 996 spacefree = ep->e2d_reclen - dsize; 997 for (loc = ep->e2d_reclen; loc < dp->i_count; ) { 998 nep = (struct ext2fs_direct_2 *)(dirbuf + loc); 999 if (ep->e2d_ino) { 1000 /* trim the existing slot */ 1001 ep->e2d_reclen = dsize; 1002 ep = (struct ext2fs_direct_2 *)((char *)ep + dsize); 1003 } else { 1004 /* overwrite; nothing there; header is ours */ 1005 spacefree += dsize; 1006 } 1007 dsize = EXT2_DIR_REC_LEN(nep->e2d_namlen); 1008 spacefree += nep->e2d_reclen - dsize; 1009 loc += nep->e2d_reclen; 1010 bcopy((caddr_t)nep, (caddr_t)ep, dsize); 1011 } 1012 /* 1013 * Update the pointer fields in the previous entry (if any), 1014 * copy in the new entry, and write out the block. 1015 */ 1016 if (ep->e2d_ino == 0) { 1017 if (spacefree + dsize < newentrysize) 1018 panic("ext2_direnter: compact1"); 1019 entry->e2d_reclen = spacefree + dsize; 1020 } else { 1021 if (spacefree < newentrysize) 1022 panic("ext2_direnter: compact2"); 1023 entry->e2d_reclen = spacefree; 1024 ep->e2d_reclen = dsize; 1025 ep = (struct ext2fs_direct_2 *)((char *)ep + dsize); 1026 } 1027 bcopy((caddr_t)entry, (caddr_t)ep, (u_int)newentrysize); 1028 if (DOINGASYNC(dvp)) { 1029 bdwrite(bp); 1030 error = 0; 1031 } else { 1032 error = bwrite(bp); 1033 } 1034 dp->i_flag |= IN_CHANGE | IN_UPDATE; 1035 return (error); 1036 } 1037 1038 /* 1039 * Remove a directory entry after a call to namei, using 1040 * the parameters which it left in nameidata. The entry 1041 * dp->i_offset contains the offset into the directory of the 1042 * entry to be eliminated. The dp->i_count field contains the 1043 * size of the previous record in the directory. If this 1044 * is 0, the first entry is being deleted, so we need only 1045 * zero the inode number to mark the entry as free. If the 1046 * entry is not the first in the directory, we must reclaim 1047 * the space of the now empty record by adding the record size 1048 * to the size of the previous entry. 1049 */ 1050 int 1051 ext2_dirremove(struct vnode *dvp, struct componentname *cnp) 1052 { 1053 struct inode *dp; 1054 struct ext2fs_direct_2 *ep, *rep; 1055 struct buf *bp; 1056 int error; 1057 1058 dp = VTOI(dvp); 1059 if (dp->i_count == 0) { 1060 /* 1061 * First entry in block: set d_ino to zero. 1062 */ 1063 if ((error = 1064 ext2_blkatoff(dvp, (off_t)dp->i_offset, (char **)&ep, 1065 &bp)) != 0) 1066 return (error); 1067 ep->e2d_ino = 0; 1068 error = bwrite(bp); 1069 dp->i_flag |= IN_CHANGE | IN_UPDATE; 1070 return (error); 1071 } 1072 /* 1073 * Collapse new free space into previous entry. 1074 */ 1075 if ((error = ext2_blkatoff(dvp, (off_t)(dp->i_offset - dp->i_count), 1076 (char **)&ep, &bp)) != 0) 1077 return (error); 1078 1079 /* Set 'rep' to the entry being removed. */ 1080 if (dp->i_count == 0) 1081 rep = ep; 1082 else 1083 rep = (struct ext2fs_direct_2 *)((char *)ep + ep->e2d_reclen); 1084 ep->e2d_reclen += rep->e2d_reclen; 1085 if (DOINGASYNC(dvp) && dp->i_count != 0) 1086 bdwrite(bp); 1087 else 1088 error = bwrite(bp); 1089 dp->i_flag |= IN_CHANGE | IN_UPDATE; 1090 return (error); 1091 } 1092 1093 /* 1094 * Rewrite an existing directory entry to point at the inode 1095 * supplied. The parameters describing the directory entry are 1096 * set up by a call to namei. 1097 */ 1098 int 1099 ext2_dirrewrite(struct inode *dp, struct inode *ip, struct componentname *cnp) 1100 { 1101 struct buf *bp; 1102 struct ext2fs_direct_2 *ep; 1103 struct vnode *vdp = ITOV(dp); 1104 int error; 1105 1106 if ((error = ext2_blkatoff(vdp, (off_t)dp->i_offset, (char **)&ep, 1107 &bp)) != 0) 1108 return (error); 1109 ep->e2d_ino = ip->i_number; 1110 if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs, 1111 EXT2F_INCOMPAT_FTYPE)) 1112 ep->e2d_type = DTTOFT(IFTODT(ip->i_mode)); 1113 else 1114 ep->e2d_type = EXT2_FT_UNKNOWN; 1115 error = bwrite(bp); 1116 dp->i_flag |= IN_CHANGE | IN_UPDATE; 1117 return (error); 1118 } 1119 1120 /* 1121 * Check if a directory is empty or not. 1122 * Inode supplied must be locked. 1123 * 1124 * Using a struct dirtemplate here is not precisely 1125 * what we want, but better than using a struct direct. 1126 * 1127 * NB: does not handle corrupted directories. 1128 */ 1129 int 1130 ext2_dirempty(struct inode *ip, ino_t parentino, struct ucred *cred) 1131 { 1132 off_t off; 1133 struct dirtemplate dbuf; 1134 struct ext2fs_direct_2 *dp = (struct ext2fs_direct_2 *)&dbuf; 1135 int error, namlen; 1136 ssize_t count; 1137 #define MINDIRSIZ (sizeof(struct dirtemplate) / 2) 1138 1139 for (off = 0; off < ip->i_size; off += dp->e2d_reclen) { 1140 error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ, 1141 off, UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK, cred, 1142 NOCRED, &count, (struct thread *)0); 1143 /* 1144 * Since we read MINDIRSIZ, residual must 1145 * be 0 unless we're at end of file. 1146 */ 1147 if (error || count != 0) 1148 return (0); 1149 /* avoid infinite loops */ 1150 if (dp->e2d_reclen == 0) 1151 return (0); 1152 /* skip empty entries */ 1153 if (dp->e2d_ino == 0) 1154 continue; 1155 /* accept only "." and ".." */ 1156 namlen = dp->e2d_namlen; 1157 if (namlen > 2) 1158 return (0); 1159 if (dp->e2d_name[0] != '.') 1160 return (0); 1161 /* 1162 * At this point namlen must be 1 or 2. 1163 * 1 implies ".", 2 implies ".." if second 1164 * char is also "." 1165 */ 1166 if (namlen == 1) 1167 continue; 1168 if (dp->e2d_name[1] == '.' && dp->e2d_ino == parentino) 1169 continue; 1170 return (0); 1171 } 1172 return (1); 1173 } 1174 1175 /* 1176 * Check if source directory is in the path of the target directory. 1177 * Target is supplied locked, source is unlocked. 1178 * The target is always vput before returning. 1179 */ 1180 int 1181 ext2_checkpath(struct inode *source, struct inode *target, struct ucred *cred) 1182 { 1183 struct vnode *vp; 1184 int error, namlen; 1185 struct dirtemplate dirbuf; 1186 1187 vp = ITOV(target); 1188 if (target->i_number == source->i_number) { 1189 error = EEXIST; 1190 goto out; 1191 } 1192 if (target->i_number == EXT2_ROOTINO) { 1193 error = 0; 1194 goto out; 1195 } 1196 1197 for (;;) { 1198 if (vp->v_type != VDIR) { 1199 error = ENOTDIR; 1200 break; 1201 } 1202 error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf, 1203 sizeof(struct dirtemplate), (off_t)0, UIO_SYSSPACE, 1204 IO_NODELOCKED | IO_NOMACCHECK, cred, NOCRED, NULL, 1205 NULL); 1206 if (error != 0) 1207 break; 1208 namlen = dirbuf.dotdot_type; /* like ufs little-endian */ 1209 if (namlen != 2 || 1210 dirbuf.dotdot_name[0] != '.' || 1211 dirbuf.dotdot_name[1] != '.') { 1212 error = ENOTDIR; 1213 break; 1214 } 1215 if (dirbuf.dotdot_ino == source->i_number) { 1216 error = EINVAL; 1217 break; 1218 } 1219 if (dirbuf.dotdot_ino == EXT2_ROOTINO) 1220 break; 1221 vput(vp); 1222 if ((error = VFS_VGET(vp->v_mount, dirbuf.dotdot_ino, 1223 LK_EXCLUSIVE, &vp)) != 0) { 1224 vp = NULL; 1225 break; 1226 } 1227 } 1228 1229 out: 1230 if (error == ENOTDIR) 1231 printf("checkpath: .. not a directory\n"); 1232 if (vp != NULL) 1233 vput(vp); 1234 return (error); 1235 } 1236