ext2fs_lookup.c revision 1.37
1/* $NetBSD: ext2fs_lookup.c,v 1.37 2005/08/30 22:01:12 xtraeme Exp $ */ 2 3/* 4 * Modified for NetBSD 1.2E 5 * May 1997, Manuel Bouyer 6 * Laboratoire d'informatique de Paris VI 7 */ 8/* 9 * modified for Lites 1.1 10 * 11 * Aug 1995, Godmar Back (gback@cs.utah.edu) 12 * University of Utah, Department of Computer Science 13 */ 14/* 15 * Copyright (c) 1989, 1993 16 * The Regents of the University of California. All rights reserved. 17 * (c) UNIX System Laboratories, Inc. 18 * All or some portions of this file are derived from material licensed 19 * to the University of California by American Telephone and Telegraph 20 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 21 * the permission of UNIX System Laboratories, Inc. 22 * 23 * Redistribution and use in source and binary forms, with or without 24 * modification, are permitted provided that the following conditions 25 * are met: 26 * 1. Redistributions of source code must retain the above copyright 27 * notice, this list of conditions and the following disclaimer. 28 * 2. Redistributions in binary form must reproduce the above copyright 29 * notice, this list of conditions and the following disclaimer in the 30 * documentation and/or other materials provided with the distribution. 31 * 3. Neither the name of the University nor the names of its contributors 32 * may be used to endorse or promote products derived from this software 33 * without specific prior written permission. 34 * 35 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 36 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 37 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 38 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 39 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 40 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 41 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 42 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 43 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 44 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 45 * SUCH DAMAGE. 46 * 47 * @(#)ufs_lookup.c 8.6 (Berkeley) 4/1/94 48 */ 49 50#include <sys/cdefs.h> 51__KERNEL_RCSID(0, "$NetBSD: ext2fs_lookup.c,v 1.37 2005/08/30 22:01:12 xtraeme Exp $"); 52 53#include <sys/param.h> 54#include <sys/systm.h> 55#include <sys/namei.h> 56#include <sys/buf.h> 57#include <sys/file.h> 58#include <sys/mount.h> 59#include <sys/vnode.h> 60#include <sys/malloc.h> 61#include <sys/dirent.h> 62 63#include <ufs/ufs/inode.h> 64#include <ufs/ufs/ufsmount.h> 65#include <ufs/ufs/ufs_extern.h> 66 67#include <ufs/ext2fs/ext2fs_extern.h> 68#include <ufs/ext2fs/ext2fs_dir.h> 69#include <ufs/ext2fs/ext2fs.h> 70 71extern int dirchk; 72 73static void ext2fs_dirconv2ffs(struct ext2fs_direct *e2dir, 74 struct dirent *ffsdir); 75static int ext2fs_dirbadentry(struct vnode *dp, 76 struct ext2fs_direct *de, 77 int entryoffsetinblock); 78 79/* 80 * the problem that is tackled below is the fact that FFS 81 * includes the terminating zero on disk while EXT2FS doesn't 82 * this implies that we need to introduce some padding. 83 * For instance, a filename "sbin" has normally a reclen 12 84 * in EXT2, but 16 in FFS. 85 * This reminds me of that Pepsi commercial: 'Kid saved a lousy nine cents...' 86 * If it wasn't for that, the complete ufs code for directories would 87 * have worked w/o changes (except for the difference in DIRBLKSIZ) 88 */ 89static void 90ext2fs_dirconv2ffs(struct ext2fs_direct *e2dir, struct dirent *ffsdir) 91{ 92 memset(ffsdir, 0, sizeof(struct dirent)); 93 ffsdir->d_fileno = fs2h32(e2dir->e2d_ino); 94 ffsdir->d_namlen = e2dir->e2d_namlen; 95 96 ffsdir->d_type = DT_UNKNOWN; /* don't know more here */ 97#ifdef DIAGNOSTIC 98#if MAXNAMLEN < E2FS_MAXNAMLEN 99 /* 100 * we should handle this more gracefully ! 101 */ 102 if (e2dir->e2d_namlen > MAXNAMLEN) 103 panic("ext2fs: e2dir->e2d_namlen"); 104#endif 105#endif 106 strncpy(ffsdir->d_name, e2dir->e2d_name, ffsdir->d_namlen); 107 108 /* Godmar thinks: since e2dir->e2d_reclen can be big and means 109 nothing anyway, we compute our own reclen according to what 110 we think is right 111 */ 112 ffsdir->d_reclen = _DIRENT_SIZE(ffsdir); 113} 114 115/* 116 * Vnode op for reading directories. 117 * 118 * Convert the on-disk entries to <sys/dirent.h> entries. 119 * the problem is that the conversion will blow up some entries by four bytes, 120 * so it can't be done in place. This is too bad. Right now the conversion is 121 * done entry by entry, the converted entry is sent via uiomove. 122 * 123 * XXX allocate a buffer, convert as many entries as possible, then send 124 * the whole buffer to uiomove 125 */ 126int 127ext2fs_readdir(void *v) 128{ 129 struct vop_readdir_args /* { 130 struct vnode *a_vp; 131 struct uio *a_uio; 132 struct ucred *a_cred; 133 int **a_eofflag; 134 off_t **a_cookies; 135 int ncookies; 136 } */ *ap = v; 137 struct uio *uio = ap->a_uio; 138 int error; 139 size_t e2fs_count, readcnt; 140 struct vnode *vp = ap->a_vp; 141 struct m_ext2fs *fs = VTOI(vp)->i_e2fs; 142 143 struct ext2fs_direct *dp; 144 struct dirent dstd; 145 struct uio auio; 146 struct iovec aiov; 147 caddr_t dirbuf; 148 off_t off = uio->uio_offset; 149 off_t *cookies = NULL; 150 int nc = 0, ncookies = 0; 151 int e2d_reclen; 152 153 if (vp->v_type != VDIR) 154 return (ENOTDIR); 155 156 e2fs_count = uio->uio_resid; 157 /* Make sure we don't return partial entries. */ 158 e2fs_count -= (uio->uio_offset + e2fs_count) & (fs->e2fs_bsize -1); 159 if (e2fs_count <= 0) 160 return (EINVAL); 161 162 auio = *uio; 163 auio.uio_iov = &aiov; 164 auio.uio_iovcnt = 1; 165 auio.uio_segflg = UIO_SYSSPACE; 166 aiov.iov_len = e2fs_count; 167 auio.uio_resid = e2fs_count; 168 MALLOC(dirbuf, caddr_t, e2fs_count, M_TEMP, M_WAITOK); 169 if (ap->a_ncookies) { 170 nc = ncookies = e2fs_count / 16; 171 cookies = malloc(sizeof (off_t) * ncookies, M_TEMP, M_WAITOK); 172 *ap->a_cookies = cookies; 173 } 174 memset(dirbuf, 0, e2fs_count); 175 aiov.iov_base = dirbuf; 176 177 error = VOP_READ(ap->a_vp, &auio, 0, ap->a_cred); 178 if (error == 0) { 179 readcnt = e2fs_count - auio.uio_resid; 180 for (dp = (struct ext2fs_direct *)dirbuf; 181 (char *)dp < (char *)dirbuf + readcnt; ) { 182 e2d_reclen = fs2h16(dp->e2d_reclen); 183 if (e2d_reclen == 0) { 184 error = EIO; 185 break; 186 } 187 ext2fs_dirconv2ffs(dp, &dstd); 188 if(dstd.d_reclen > uio->uio_resid) { 189 break; 190 } 191 if ((error = uiomove((caddr_t)&dstd, dstd.d_reclen, uio)) != 0) { 192 break; 193 } 194 off = off + e2d_reclen; 195 if (cookies != NULL) { 196 *cookies++ = off; 197 if (--ncookies <= 0){ 198 break; /* out of cookies */ 199 } 200 } 201 /* advance dp */ 202 dp = (struct ext2fs_direct *) ((char *)dp + e2d_reclen); 203 } 204 /* we need to correct uio_offset */ 205 uio->uio_offset = off; 206 } 207 FREE(dirbuf, M_TEMP); 208 *ap->a_eofflag = ext2fs_size(VTOI(ap->a_vp)) <= uio->uio_offset; 209 if (ap->a_ncookies) { 210 if (error) { 211 free(*ap->a_cookies, M_TEMP); 212 *ap->a_ncookies = 0; 213 *ap->a_cookies = NULL; 214 } else 215 *ap->a_ncookies = nc - ncookies; 216 } 217 return (error); 218} 219 220/* 221 * Convert a component of a pathname into a pointer to a locked inode. 222 * This is a very central and rather complicated routine. 223 * If the file system is not maintained in a strict tree hierarchy, 224 * this can result in a deadlock situation (see comments in code below). 225 * 226 * The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending 227 * on whether the name is to be looked up, created, renamed, or deleted. 228 * When CREATE, RENAME, or DELETE is specified, information usable in 229 * creating, renaming, or deleting a directory entry may be calculated. 230 * If flag has LOCKPARENT or'ed into it and the target of the pathname 231 * exists, lookup returns both the target and its parent directory locked. 232 * When creating or renaming and LOCKPARENT is specified, the target may 233 * not be ".". When deleting and LOCKPARENT is specified, the target may 234 * be "."., but the caller must check to ensure it does an vrele and vput 235 * instead of two vputs. 236 * 237 * Overall outline of ext2fs_lookup: 238 * 239 * check accessibility of directory 240 * look for name in cache, if found, then if at end of path 241 * and deleting or creating, drop it, else return name 242 * search for name in directory, to found or notfound 243 * notfound: 244 * if creating, return locked directory, leaving info on available slots 245 * else return error 246 * found: 247 * if at end of path and deleting, return information to allow delete 248 * if at end of path and rewriting (RENAME and LOCKPARENT), lock target 249 * inode and return info to allow rewrite 250 * if not at end, add name to cache; if at end and neither creating 251 * nor deleting, add name to cache 252 */ 253int 254ext2fs_lookup(void *v) 255{ 256 struct vop_lookup_args /* { 257 struct vnode *a_dvp; 258 struct vnode **a_vpp; 259 struct componentname *a_cnp; 260 } */ *ap = v; 261 struct vnode *vdp = ap->a_dvp; /* vnode for directory being searched */ 262 struct inode *dp = VTOI(vdp); /* inode for directory being searched */ 263 struct buf *bp; /* a buffer of directory entries */ 264 struct ext2fs_direct *ep; /* the current directory entry */ 265 int entryoffsetinblock; /* offset of ep in bp's buffer */ 266 enum {NONE, COMPACT, FOUND} slotstatus; 267 doff_t slotoffset; /* offset of area with free space */ 268 int slotsize; /* size of area at slotoffset */ 269 int slotfreespace; /* amount of space free in slot */ 270 int slotneeded; /* size of the entry we're seeking */ 271 int numdirpasses; /* strategy for directory search */ 272 doff_t endsearch; /* offset to end directory search */ 273 doff_t prevoff; /* prev entry dp->i_offset */ 274 struct vnode *pdp; /* saved dp during symlink work */ 275 struct vnode *tdp; /* returned by VFS_VGET */ 276 doff_t enduseful; /* pointer past last used dir slot */ 277 u_long bmask; /* block offset mask */ 278 int lockparent; /* 1 => lockparent flag is set */ 279 int wantparent; /* 1 => wantparent or lockparent flag */ 280 int namlen, error; 281 struct vnode **vpp = ap->a_vpp; 282 struct componentname *cnp = ap->a_cnp; 283 struct ucred *cred = cnp->cn_cred; 284 int flags; 285 int nameiop = cnp->cn_nameiop; 286 struct ufsmount *ump = dp->i_ump; 287 int dirblksiz = ump->um_dirblksiz; 288 ino_t foundino; 289 290 cnp->cn_flags &= ~PDIRUNLOCK; 291 flags = cnp->cn_flags; 292 293 bp = NULL; 294 slotoffset = -1; 295 *vpp = NULL; 296 lockparent = flags & LOCKPARENT; 297 wantparent = flags & (LOCKPARENT|WANTPARENT); 298 /* 299 * Check accessiblity of directory. 300 */ 301 if ((error = VOP_ACCESS(vdp, VEXEC, cred, cnp->cn_proc)) != 0) 302 return (error); 303 304 if ((flags & ISLASTCN) && (vdp->v_mount->mnt_flag & MNT_RDONLY) && 305 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 306 return (EROFS); 307 308 /* 309 * We now have a segment name to search for, and a directory to search. 310 * 311 * Before tediously performing a linear scan of the directory, 312 * check the name cache to see if the directory/name pair 313 * we are looking for is known already. 314 */ 315 if ((error = cache_lookup(vdp, vpp, cnp)) >= 0) 316 return (error); 317 318 /* 319 * Suppress search for slots unless creating 320 * file and at end of pathname, in which case 321 * we watch for a place to put the new file in 322 * case it doesn't already exist. 323 */ 324 slotstatus = FOUND; 325 slotfreespace = slotsize = slotneeded = 0; 326 if ((nameiop == CREATE || nameiop == RENAME) && 327 (flags & ISLASTCN)) { 328 slotstatus = NONE; 329 slotneeded = EXT2FS_DIRSIZ(cnp->cn_namelen); 330 } 331 332 /* 333 * If there is cached information on a previous search of 334 * this directory, pick up where we last left off. 335 * We cache only lookups as these are the most common 336 * and have the greatest payoff. Caching CREATE has little 337 * benefit as it usually must search the entire directory 338 * to determine that the entry does not exist. Caching the 339 * location of the last DELETE or RENAME has not reduced 340 * profiling time and hence has been removed in the interest 341 * of simplicity. 342 */ 343 bmask = vdp->v_mount->mnt_stat.f_iosize - 1; 344 if (nameiop != LOOKUP || dp->i_diroff == 0 || 345 dp->i_diroff >= ext2fs_size(dp)) { 346 entryoffsetinblock = 0; 347 dp->i_offset = 0; 348 numdirpasses = 1; 349 } else { 350 dp->i_offset = dp->i_diroff; 351 if ((entryoffsetinblock = dp->i_offset & bmask) && 352 (error = VOP_BLKATOFF(vdp, (off_t)dp->i_offset, NULL, &bp))) 353 return (error); 354 numdirpasses = 2; 355 nchstats.ncs_2passes++; 356 } 357 prevoff = dp->i_offset; 358 endsearch = roundup(ext2fs_size(dp), dirblksiz); 359 enduseful = 0; 360 361searchloop: 362 while (dp->i_offset < endsearch) { 363 if (curcpu()->ci_schedstate.spc_flags & SPCF_SHOULDYIELD) 364 preempt(1); 365 /* 366 * If necessary, get the next directory block. 367 */ 368 if ((dp->i_offset & bmask) == 0) { 369 if (bp != NULL) 370 brelse(bp); 371 error = VOP_BLKATOFF(vdp, (off_t)dp->i_offset, NULL, 372 &bp); 373 if (error != 0) 374 return (error); 375 entryoffsetinblock = 0; 376 } 377 /* 378 * If still looking for a slot, and at a dirblksize 379 * boundary, have to start looking for free space again. 380 */ 381 if (slotstatus == NONE && 382 (entryoffsetinblock & (dirblksiz - 1)) == 0) { 383 slotoffset = -1; 384 slotfreespace = 0; 385 } 386 /* 387 * Get pointer to next entry. 388 * Full validation checks are slow, so we only check 389 * enough to insure forward progress through the 390 * directory. Complete checks can be run by patching 391 * "dirchk" to be true. 392 */ 393 ep = (struct ext2fs_direct *) 394 ((char *)bp->b_data + entryoffsetinblock); 395 if (ep->e2d_reclen == 0 || 396 (dirchk && 397 ext2fs_dirbadentry(vdp, ep, entryoffsetinblock))) { 398 int i; 399 400 ufs_dirbad(dp, dp->i_offset, "mangled entry"); 401 i = dirblksiz - (entryoffsetinblock & (dirblksiz - 1)); 402 dp->i_offset += i; 403 entryoffsetinblock += i; 404 continue; 405 } 406 407 /* 408 * If an appropriate sized slot has not yet been found, 409 * check to see if one is available. Also accumulate space 410 * in the current block so that we can determine if 411 * compaction is viable. 412 */ 413 if (slotstatus != FOUND) { 414 int size = fs2h16(ep->e2d_reclen); 415 416 if (ep->e2d_ino != 0) 417 size -= EXT2FS_DIRSIZ(ep->e2d_namlen); 418 if (size > 0) { 419 if (size >= slotneeded) { 420 slotstatus = FOUND; 421 slotoffset = dp->i_offset; 422 slotsize = fs2h16(ep->e2d_reclen); 423 } else if (slotstatus == NONE) { 424 slotfreespace += size; 425 if (slotoffset == -1) 426 slotoffset = dp->i_offset; 427 if (slotfreespace >= slotneeded) { 428 slotstatus = COMPACT; 429 slotsize = dp->i_offset + 430 fs2h16(ep->e2d_reclen) - 431 slotoffset; 432 } 433 } 434 } 435 } 436 437 /* 438 * Check for a name match. 439 */ 440 if (ep->e2d_ino) { 441 namlen = ep->e2d_namlen; 442 if (namlen == cnp->cn_namelen && 443 !memcmp(cnp->cn_nameptr, ep->e2d_name, 444 (unsigned)namlen)) { 445 /* 446 * Save directory entry's inode number and 447 * reclen in ndp->ni_ufs area, and release 448 * directory buffer. 449 */ 450 foundino = fs2h32(ep->e2d_ino); 451 dp->i_reclen = fs2h16(ep->e2d_reclen); 452 goto found; 453 } 454 } 455 prevoff = dp->i_offset; 456 dp->i_offset += fs2h16(ep->e2d_reclen); 457 entryoffsetinblock += fs2h16(ep->e2d_reclen); 458 if (ep->e2d_ino) 459 enduseful = dp->i_offset; 460 } 461/* notfound: */ 462 /* 463 * If we started in the middle of the directory and failed 464 * to find our target, we must check the beginning as well. 465 */ 466 if (numdirpasses == 2) { 467 numdirpasses--; 468 dp->i_offset = 0; 469 endsearch = dp->i_diroff; 470 goto searchloop; 471 } 472 if (bp != NULL) 473 brelse(bp); 474 /* 475 * If creating, and at end of pathname and current 476 * directory has not been removed, then can consider 477 * allowing file to be created. 478 */ 479 if ((nameiop == CREATE || nameiop == RENAME) && 480 (flags & ISLASTCN) && dp->i_e2fs_nlink != 0) { 481 /* 482 * Access for write is interpreted as allowing 483 * creation of files in the directory. 484 */ 485 error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_proc); 486 if (error) 487 return (error); 488 /* 489 * Return an indication of where the new directory 490 * entry should be put. If we didn't find a slot, 491 * then set dp->i_count to 0 indicating 492 * that the new slot belongs at the end of the 493 * directory. If we found a slot, then the new entry 494 * can be put in the range from dp->i_offset to 495 * dp->i_offset + dp->i_count. 496 */ 497 if (slotstatus == NONE) { 498 dp->i_offset = roundup(ext2fs_size(dp), dirblksiz); 499 dp->i_count = 0; 500 enduseful = dp->i_offset; 501 } else if (nameiop == DELETE) { 502 dp->i_offset = slotoffset; 503 if ((dp->i_offset & (dirblksiz - 1)) == 0) 504 dp->i_count = 0; 505 else 506 dp->i_count = dp->i_offset - prevoff; 507 } else { 508 dp->i_offset = slotoffset; 509 dp->i_count = slotsize; 510 if (enduseful < slotoffset + slotsize) 511 enduseful = slotoffset + slotsize; 512 } 513 dp->i_endoff = roundup(enduseful, dirblksiz); 514#if 0 515 dp->i_flag |= IN_CHANGE | IN_UPDATE; 516#endif 517 /* 518 * We return with the directory locked, so that 519 * the parameters we set up above will still be 520 * valid if we actually decide to do a direnter(). 521 * We return ni_vp == NULL to indicate that the entry 522 * does not currently exist; we leave a pointer to 523 * the (locked) directory inode in ndp->ni_dvp. 524 * The pathname buffer is saved so that the name 525 * can be obtained later. 526 * 527 * NB - if the directory is unlocked, then this 528 * information cannot be used. 529 */ 530 cnp->cn_flags |= SAVENAME; 531 if (!lockparent) { 532 VOP_UNLOCK(vdp, 0); 533 cnp->cn_flags |= PDIRUNLOCK; 534 } 535 return (EJUSTRETURN); 536 } 537 /* 538 * Insert name into cache (as non-existent) if appropriate. 539 */ 540 if ((cnp->cn_flags & MAKEENTRY) && nameiop != CREATE) 541 cache_enter(vdp, *vpp, cnp); 542 return (ENOENT); 543 544found: 545 if (numdirpasses == 2) 546 nchstats.ncs_pass2++; 547 /* 548 * Check that directory length properly reflects presence 549 * of this entry. 550 */ 551 if (dp->i_offset + EXT2FS_DIRSIZ(ep->e2d_namlen) > ext2fs_size(dp)) { 552 ufs_dirbad(dp, dp->i_offset, "i_size too small"); 553 error = ext2fs_setsize(dp, 554 dp->i_offset + EXT2FS_DIRSIZ(ep->e2d_namlen)); 555 if (error) { 556 brelse(bp); 557 return (error); 558 } 559 dp->i_flag |= IN_CHANGE | IN_UPDATE; 560 uvm_vnp_setsize(vdp, ext2fs_size(dp)); 561 } 562 brelse(bp); 563 564 /* 565 * Found component in pathname. 566 * If the final component of path name, save information 567 * in the cache as to where the entry was found. 568 */ 569 if ((flags & ISLASTCN) && nameiop == LOOKUP) 570 dp->i_diroff = dp->i_offset &~ (dirblksiz - 1); 571 572 /* 573 * If deleting, and at end of pathname, return 574 * parameters which can be used to remove file. 575 * If the wantparent flag isn't set, we return only 576 * the directory (in ndp->ni_dvp), otherwise we go 577 * on and lock the inode, being careful with ".". 578 */ 579 if (nameiop == DELETE && (flags & ISLASTCN)) { 580 /* 581 * Write access to directory required to delete files. 582 */ 583 if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_proc)) != 0) 584 return (error); 585 /* 586 * Return pointer to current entry in dp->i_offset, 587 * and distance past previous entry (if there 588 * is a previous entry in this block) in dp->i_count. 589 * Save directory inode pointer in ndp->ni_dvp for dirremove(). 590 */ 591 if ((dp->i_offset & (dirblksiz - 1)) == 0) 592 dp->i_count = 0; 593 else 594 dp->i_count = dp->i_offset - prevoff; 595 if (dp->i_number == foundino) { 596 VREF(vdp); 597 *vpp = vdp; 598 return (0); 599 } 600 if (flags & ISDOTDOT) 601 VOP_UNLOCK(vdp, 0); /* race to get the inode */ 602 error = VFS_VGET(vdp->v_mount, foundino, &tdp); 603 if (flags & ISDOTDOT) 604 vn_lock(vdp, LK_EXCLUSIVE | LK_RETRY); 605 if (error) 606 return (error); 607 /* 608 * If directory is "sticky", then user must own 609 * the directory, or the file in it, else she 610 * may not delete it (unless she's root). This 611 * implements append-only directories. 612 */ 613 if ((dp->i_e2fs_mode & ISVTX) && 614 cred->cr_uid != 0 && 615 cred->cr_uid != dp->i_e2fs_uid && 616 VTOI(tdp)->i_e2fs_uid != cred->cr_uid) { 617 vput(tdp); 618 return (EPERM); 619 } 620 *vpp = tdp; 621 if (!lockparent) { 622 VOP_UNLOCK(vdp, 0); 623 cnp->cn_flags |= PDIRUNLOCK; 624 } 625 return (0); 626 } 627 628 /* 629 * If rewriting (RENAME), return the inode and the 630 * information required to rewrite the present directory 631 * Must get inode of directory entry to verify it's a 632 * regular file, or empty directory. 633 */ 634 if (nameiop == RENAME && wantparent && (flags & ISLASTCN)) { 635 error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_proc); 636 if (error) 637 return (error); 638 /* 639 * Careful about locking second inode. 640 * This can only occur if the target is ".". 641 */ 642 if (dp->i_number == foundino) 643 return (EISDIR); 644 if (flags & ISDOTDOT) 645 VOP_UNLOCK(vdp, 0); /* race to get the inode */ 646 error = VFS_VGET(vdp->v_mount, foundino, &tdp); 647 if (flags & ISDOTDOT) 648 vn_lock(vdp, LK_EXCLUSIVE | LK_RETRY); 649 if (error) 650 return (error); 651 *vpp = tdp; 652 cnp->cn_flags |= SAVENAME; 653 if (!lockparent) { 654 VOP_UNLOCK(vdp, 0); 655 cnp->cn_flags |= PDIRUNLOCK; 656 } 657 return (0); 658 } 659 660 /* 661 * Step through the translation in the name. We do not `vput' the 662 * directory because we may need it again if a symbolic link 663 * is relative to the current directory. Instead we save it 664 * unlocked as "pdp". We must get the target inode before unlocking 665 * the directory to insure that the inode will not be removed 666 * before we get it. We prevent deadlock by always fetching 667 * inodes from the root, moving down the directory tree. Thus 668 * when following backward pointers ".." we must unlock the 669 * parent directory before getting the requested directory. 670 * There is a potential race condition here if both the current 671 * and parent directories are removed before the VFS_VGET for the 672 * inode associated with ".." returns. We hope that this occurs 673 * infrequently since we cannot avoid this race condition without 674 * implementing a sophisticated deadlock detection algorithm. 675 * Note also that this simple deadlock detection scheme will not 676 * work if the file system has any hard links other than ".." 677 * that point backwards in the directory structure. 678 */ 679 pdp = vdp; 680 if (flags & ISDOTDOT) { 681 VOP_UNLOCK(pdp, 0); /* race to get the inode */ 682 cnp->cn_flags |= PDIRUNLOCK; 683 error = VFS_VGET(vdp->v_mount, foundino, &tdp); 684 if (error) { 685 if (vn_lock(pdp, LK_EXCLUSIVE | LK_RETRY) == 0) 686 cnp->cn_flags &= ~PDIRUNLOCK; 687 return (error); 688 } 689 if (lockparent && (flags & ISLASTCN)) { 690 if ((error = vn_lock(pdp, LK_EXCLUSIVE))) { 691 vput(tdp); 692 return (error); 693 } 694 cnp->cn_flags &= ~PDIRUNLOCK; 695 } 696 *vpp = tdp; 697 } else if (dp->i_number == foundino) { 698 VREF(vdp); /* we want ourself, ie "." */ 699 *vpp = vdp; 700 } else { 701 error = VFS_VGET(vdp->v_mount, foundino, &tdp); 702 if (error) 703 return (error); 704 if (!lockparent || !(flags & ISLASTCN)) { 705 VOP_UNLOCK(pdp, 0); 706 cnp->cn_flags |= PDIRUNLOCK; 707 } 708 *vpp = tdp; 709 } 710 711 /* 712 * Insert name into cache if appropriate. 713 */ 714 if (cnp->cn_flags & MAKEENTRY) 715 cache_enter(vdp, *vpp, cnp); 716 return (0); 717} 718 719/* 720 * Do consistency checking on a directory entry: 721 * record length must be multiple of 4 722 * entry must fit in rest of its dirblksize block 723 * record must be large enough to contain entry 724 * name is not longer than EXT2FS_MAXNAMLEN 725 * name must be as long as advertised, and null terminated 726 */ 727/* 728 * changed so that it confirms to ext2fs_check_dir_entry 729 */ 730static int 731ext2fs_dirbadentry(struct vnode *dp, struct ext2fs_direct *de, 732 int entryoffsetinblock) 733{ 734 struct ufsmount *ump = VFSTOUFS(dp->v_mount); 735 int dirblksiz = ump->um_dirblksiz; 736 737 const char *error_msg = NULL; 738 int reclen = fs2h16(de->e2d_reclen); 739 int namlen = de->e2d_namlen; 740 741 if (reclen < EXT2FS_DIRSIZ(1)) /* e2d_namlen = 1 */ 742 error_msg = "rec_len is smaller than minimal"; 743 else if (reclen % 4 != 0) 744 error_msg = "rec_len % 4 != 0"; 745 else if (namlen > EXT2FS_MAXNAMLEN) 746 error_msg = "namlen > EXT2FS_MAXNAMLEN"; 747 else if (reclen < EXT2FS_DIRSIZ(namlen)) 748 error_msg = "reclen is too small for name_len"; 749 else if (entryoffsetinblock + reclen > dirblksiz) 750 error_msg = "directory entry across blocks"; 751 else if (fs2h32(de->e2d_ino) > 752 VTOI(dp)->i_e2fs->e2fs.e2fs_icount) 753 error_msg = "inode out of bounds"; 754 755 if (error_msg != NULL) { 756 printf( "bad directory entry: %s\n" 757 "offset=%d, inode=%lu, rec_len=%d, name_len=%d \n", 758 error_msg, entryoffsetinblock, 759 (unsigned long) fs2h32(de->e2d_ino), 760 reclen, namlen); 761 panic("ext2fs_dirbadentry"); 762 } 763 return error_msg == NULL ? 0 : 1; 764} 765 766/* 767 * Write a directory entry after a call to namei, using the parameters 768 * that it left in nameidata. The argument ip is the inode which the new 769 * directory entry will refer to. Dvp is a pointer to the directory to 770 * be written, which was left locked by namei. Remaining parameters 771 * (dp->i_offset, dp->i_count) indicate how the space for the new 772 * entry is to be obtained. 773 */ 774int 775ext2fs_direnter(struct inode *ip, struct vnode *dvp, struct componentname *cnp) 776{ 777 struct ext2fs_direct *ep, *nep; 778 struct inode *dp; 779 struct buf *bp; 780 struct ext2fs_direct newdir; 781 struct iovec aiov; 782 struct uio auio; 783 u_int dsize; 784 int error, loc, newentrysize, spacefree; 785 char *dirbuf; 786 struct ufsmount *ump = VFSTOUFS(dvp->v_mount); 787 int dirblksiz = ump->um_dirblksiz; 788 789#ifdef DIAGNOSTIC 790 if ((cnp->cn_flags & SAVENAME) == 0) 791 panic("direnter: missing name"); 792#endif 793 dp = VTOI(dvp); 794 newdir.e2d_ino = h2fs32(ip->i_number); 795 newdir.e2d_namlen = cnp->cn_namelen; 796 if (ip->i_e2fs->e2fs.e2fs_rev > E2FS_REV0 && 797 (ip->i_e2fs->e2fs.e2fs_features_incompat & EXT2F_INCOMPAT_FTYPE)) { 798 newdir.e2d_type = inot2ext2dt(IFTODT(ip->i_e2fs_mode)); 799 } else { 800 newdir.e2d_type = 0; 801 }; 802 memcpy(newdir.e2d_name, cnp->cn_nameptr, (unsigned)cnp->cn_namelen + 1); 803 newentrysize = EXT2FS_DIRSIZ(cnp->cn_namelen); 804 if (dp->i_count == 0) { 805 /* 806 * If dp->i_count is 0, then namei could find no 807 * space in the directory. Here, dp->i_offset will 808 * be on a directory block boundary and we will write the 809 * new entry into a fresh block. 810 */ 811 if (dp->i_offset & (dirblksiz - 1)) 812 panic("ext2fs_direnter: newblk"); 813 auio.uio_offset = dp->i_offset; 814 newdir.e2d_reclen = h2fs16(dirblksiz); 815 auio.uio_resid = newentrysize; 816 aiov.iov_len = newentrysize; 817 aiov.iov_base = (caddr_t)&newdir; 818 auio.uio_iov = &aiov; 819 auio.uio_iovcnt = 1; 820 auio.uio_rw = UIO_WRITE; 821 auio.uio_segflg = UIO_SYSSPACE; 822 auio.uio_procp = NULL; 823 error = VOP_WRITE(dvp, &auio, IO_SYNC, cnp->cn_cred); 824 if (dirblksiz > dvp->v_mount->mnt_stat.f_bsize) 825 /* XXX should grow with balloc() */ 826 panic("ext2fs_direnter: frag size"); 827 else if (!error) { 828 error = ext2fs_setsize(dp, 829 roundup(ext2fs_size(dp), dirblksiz)); 830 if (error) 831 return (error); 832 dp->i_flag |= IN_CHANGE; 833 uvm_vnp_setsize(dvp, ext2fs_size(dp)); 834 } 835 return (error); 836 } 837 838 /* 839 * If dp->i_count is non-zero, then namei found space 840 * for the new entry in the range dp->i_offset to 841 * dp->i_offset + dp->i_count in the directory. 842 * To use this space, we may have to compact the entries located 843 * there, by copying them together towards the beginning of the 844 * block, leaving the free space in one usable chunk at the end. 845 */ 846 847 /* 848 * Get the block containing the space for the new directory entry. 849 */ 850 if ((error = VOP_BLKATOFF(dvp, (off_t)dp->i_offset, &dirbuf, &bp)) != 0) 851 return (error); 852 /* 853 * Find space for the new entry. In the simple case, the entry at 854 * offset base will have the space. If it does not, then namei 855 * arranged that compacting the region dp->i_offset to 856 * dp->i_offset + dp->i_count would yield the 857 * space. 858 */ 859 ep = (struct ext2fs_direct *)dirbuf; 860 dsize = EXT2FS_DIRSIZ(ep->e2d_namlen); 861 spacefree = fs2h16(ep->e2d_reclen) - dsize; 862 for (loc = fs2h16(ep->e2d_reclen); loc < dp->i_count; ) { 863 nep = (struct ext2fs_direct *)(dirbuf + loc); 864 if (ep->e2d_ino) { 865 /* trim the existing slot */ 866 ep->e2d_reclen = h2fs16(dsize); 867 ep = (struct ext2fs_direct *)((char *)ep + dsize); 868 } else { 869 /* overwrite; nothing there; header is ours */ 870 spacefree += dsize; 871 } 872 dsize = EXT2FS_DIRSIZ(nep->e2d_namlen); 873 spacefree += fs2h16(nep->e2d_reclen) - dsize; 874 loc += fs2h16(nep->e2d_reclen); 875 memcpy((caddr_t)ep, (caddr_t)nep, dsize); 876 } 877 /* 878 * Update the pointer fields in the previous entry (if any), 879 * copy in the new entry, and write out the block. 880 */ 881 if (ep->e2d_ino == 0) { 882#ifdef DIAGNOSTIC 883 if (spacefree + dsize < newentrysize) 884 panic("ext2fs_direnter: compact1"); 885#endif 886 newdir.e2d_reclen = h2fs16(spacefree + dsize); 887 } else { 888#ifdef DIAGNOSTIC 889 if (spacefree < newentrysize) { 890 printf("ext2fs_direnter: compact2 %u %u", 891 (u_int)spacefree, (u_int)newentrysize); 892 panic("ext2fs_direnter: compact2"); 893 } 894#endif 895 newdir.e2d_reclen = h2fs16(spacefree); 896 ep->e2d_reclen = h2fs16(dsize); 897 ep = (struct ext2fs_direct *)((char *)ep + dsize); 898 } 899 memcpy((caddr_t)ep, (caddr_t)&newdir, (u_int)newentrysize); 900 error = VOP_BWRITE(bp); 901 dp->i_flag |= IN_CHANGE | IN_UPDATE; 902 if (!error && dp->i_endoff && dp->i_endoff < ext2fs_size(dp)) 903 error = VOP_TRUNCATE(dvp, (off_t)dp->i_endoff, IO_SYNC, 904 cnp->cn_cred, cnp->cn_proc); 905 return (error); 906} 907 908/* 909 * Remove a directory entry after a call to namei, using 910 * the parameters which it left in nameidata. The entry 911 * dp->i_offset contains the offset into the directory of the 912 * entry to be eliminated. The dp->i_count field contains the 913 * size of the previous record in the directory. If this 914 * is 0, the first entry is being deleted, so we need only 915 * zero the inode number to mark the entry as free. If the 916 * entry is not the first in the directory, we must reclaim 917 * the space of the now empty record by adding the record size 918 * to the size of the previous entry. 919 */ 920int 921ext2fs_dirremove(struct vnode *dvp, struct componentname *cnp) 922{ 923 struct inode *dp; 924 struct ext2fs_direct *ep; 925 struct buf *bp; 926 int error; 927 928 dp = VTOI(dvp); 929 if (dp->i_count == 0) { 930 /* 931 * First entry in block: set d_ino to zero. 932 */ 933 error = VOP_BLKATOFF(dvp, (off_t)dp->i_offset, 934 (void *)&ep, &bp); 935 if (error != 0) 936 return (error); 937 ep->e2d_ino = 0; 938 error = VOP_BWRITE(bp); 939 dp->i_flag |= IN_CHANGE | IN_UPDATE; 940 return (error); 941 } 942 /* 943 * Collapse new free space into previous entry. 944 */ 945 error = VOP_BLKATOFF(dvp, (off_t)(dp->i_offset - dp->i_count), 946 (void *)&ep, &bp); 947 if (error != 0) 948 return (error); 949 ep->e2d_reclen = h2fs16(fs2h16(ep->e2d_reclen) + dp->i_reclen); 950 error = VOP_BWRITE(bp); 951 dp->i_flag |= IN_CHANGE | IN_UPDATE; 952 return (error); 953} 954 955/* 956 * Rewrite an existing directory entry to point at the inode 957 * supplied. The parameters describing the directory entry are 958 * set up by a call to namei. 959 */ 960int 961ext2fs_dirrewrite(struct inode *dp, struct inode *ip, 962 struct componentname *cnp) 963{ 964 struct buf *bp; 965 struct ext2fs_direct *ep; 966 struct vnode *vdp = ITOV(dp); 967 int error; 968 969 error = VOP_BLKATOFF(vdp, (off_t)dp->i_offset, (void *)&ep, &bp); 970 if (error != 0) 971 return (error); 972 ep->e2d_ino = h2fs32(ip->i_number); 973 if (ip->i_e2fs->e2fs.e2fs_rev > E2FS_REV0 && 974 (ip->i_e2fs->e2fs.e2fs_features_incompat & EXT2F_INCOMPAT_FTYPE)) { 975 ep->e2d_type = inot2ext2dt(IFTODT(ip->i_e2fs_mode)); 976 } else { 977 ep->e2d_type = 0; 978 } 979 error = VOP_BWRITE(bp); 980 dp->i_flag |= IN_CHANGE | IN_UPDATE; 981 return (error); 982} 983 984/* 985 * Check if a directory is empty or not. 986 * Inode supplied must be locked. 987 * 988 * Using a struct dirtemplate here is not precisely 989 * what we want, but better than using a struct ext2fs_direct. 990 * 991 * NB: does not handle corrupted directories. 992 */ 993int 994ext2fs_dirempty(struct inode *ip, ino_t parentino, struct ucred *cred) 995{ 996 off_t off; 997 struct ext2fs_dirtemplate dbuf; 998 struct ext2fs_direct *dp = (struct ext2fs_direct *)&dbuf; 999 int error, namlen; 1000 size_t count; 1001 1002#define MINDIRSIZ (sizeof (struct ext2fs_dirtemplate) / 2) 1003 1004 for (off = 0; off < ext2fs_size(ip); off += fs2h16(dp->e2d_reclen)) { 1005 error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ, off, 1006 UIO_SYSSPACE, IO_NODELOCKED, cred, &count, NULL); 1007 /* 1008 * Since we read MINDIRSIZ, residual must 1009 * be 0 unless we're at end of file. 1010 */ 1011 if (error || count != 0) 1012 return (0); 1013 /* avoid infinite loops */ 1014 if (dp->e2d_reclen == 0) 1015 return (0); 1016 /* skip empty entries */ 1017 if (dp->e2d_ino == 0) 1018 continue; 1019 /* accept only "." and ".." */ 1020 namlen = dp->e2d_namlen; 1021 if (namlen > 2) 1022 return (0); 1023 if (dp->e2d_name[0] != '.') 1024 return (0); 1025 /* 1026 * At this point namlen must be 1 or 2. 1027 * 1 implies ".", 2 implies ".." if second 1028 * char is also "." 1029 */ 1030 if (namlen == 1) 1031 continue; 1032 if (dp->e2d_name[1] == '.' && fs2h32(dp->e2d_ino) == parentino) 1033 continue; 1034 return (0); 1035 } 1036 return (1); 1037} 1038 1039/* 1040 * Check if source directory is in the path of the target directory. 1041 * Target is supplied locked, source is unlocked. 1042 * The target is always vput before returning. 1043 */ 1044int 1045ext2fs_checkpath(struct inode *source, struct inode *target, 1046 struct ucred *cred) 1047{ 1048 struct vnode *vp; 1049 int error, rootino, namlen; 1050 struct ext2fs_dirtemplate dirbuf; 1051 u_int32_t ino; 1052 1053 vp = ITOV(target); 1054 if (target->i_number == source->i_number) { 1055 error = EEXIST; 1056 goto out; 1057 } 1058 rootino = ROOTINO; 1059 error = 0; 1060 if (target->i_number == rootino) 1061 goto out; 1062 1063 for (;;) { 1064 if (vp->v_type != VDIR) { 1065 error = ENOTDIR; 1066 break; 1067 } 1068 error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf, 1069 sizeof (struct ext2fs_dirtemplate), (off_t)0, 1070 UIO_SYSSPACE, IO_NODELOCKED, cred, (size_t *)0, 1071 NULL); 1072 if (error != 0) 1073 break; 1074 namlen = dirbuf.dotdot_namlen; 1075 if (namlen != 2 || 1076 dirbuf.dotdot_name[0] != '.' || 1077 dirbuf.dotdot_name[1] != '.') { 1078 error = ENOTDIR; 1079 break; 1080 } 1081 ino = fs2h32(dirbuf.dotdot_ino); 1082 if (ino == source->i_number) { 1083 error = EINVAL; 1084 break; 1085 } 1086 if (ino == rootino) 1087 break; 1088 vput(vp); 1089 error = VFS_VGET(vp->v_mount, ino, &vp); 1090 if (error != 0) { 1091 vp = NULL; 1092 break; 1093 } 1094 } 1095 1096out: 1097 if (error == ENOTDIR) { 1098 printf("checkpath: .. not a directory\n"); 1099 panic("checkpath"); 1100 } 1101 if (vp != NULL) 1102 vput(vp); 1103 return (error); 1104} 1105