1 /* 2 * Copyright (c) 1982, 1986, 1989, 1993, 1995 3 * The Regents of the University of California. All rights reserved. 4 * (c) UNIX System Laboratories, Inc. 5 * All or some portions of this file are derived from material licensed 6 * to the University of California by American Telephone and Telegraph 7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 8 * the permission of UNIX System Laboratories, Inc. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)ufs_vnops.c 8.27 (Berkeley) 5/27/95 35 * $FreeBSD: src/sys/ufs/ufs/ufs_vnops.c,v 1.131.2.8 2003/01/02 17:26:19 bde Exp $ 36 */ 37 38 #include "opt_quota.h" 39 #include "opt_suiddir.h" 40 #include "opt_ufs.h" 41 42 #include <sys/param.h> 43 #include <sys/systm.h> 44 #include <sys/kernel.h> 45 #include <sys/fcntl.h> 46 #include <sys/stat.h> 47 #include <sys/buf.h> 48 #include <sys/proc.h> 49 #include <sys/priv.h> 50 #include <sys/namei.h> 51 #include <sys/mount.h> 52 #include <sys/unistd.h> 53 #include <sys/vnode.h> 54 #include <sys/malloc.h> 55 #include <sys/dirent.h> 56 #include <sys/lockf.h> 57 #include <sys/event.h> 58 #include <sys/conf.h> 59 60 #include <sys/file.h> /* XXX */ 61 #include <sys/jail.h> 62 63 #include <vm/vm.h> 64 #include <vm/vm_extern.h> 65 66 #include <vfs/fifofs/fifo.h> 67 68 #include "quota.h" 69 #include "inode.h" 70 #include "dir.h" 71 #include "ufsmount.h" 72 #include "ufs_extern.h" 73 #include "ffs_extern.h" 74 #include "fs.h" 75 #ifdef UFS_DIRHASH 76 #include "dirhash.h" 77 #endif 78 79 static int ufs_access (struct vop_access_args *); 80 static int ufs_advlock (struct vop_advlock_args *); 81 static int ufs_chmod (struct vnode *, int, struct ucred *); 82 static int ufs_chown (struct vnode *, uid_t, gid_t, struct ucred *); 83 static int ufs_close (struct vop_close_args *); 84 static int ufs_create (struct vop_old_create_args *); 85 static int ufs_getattr (struct vop_getattr_args *); 86 static int ufs_link (struct vop_old_link_args *); 87 static int ufs_makeinode (int mode, struct vnode *, struct vnode **, struct componentname *); 88 static int ufs_markatime (struct vop_markatime_args *); 89 static int ufs_missingop (struct vop_generic_args *ap); 90 static int ufs_mkdir (struct vop_old_mkdir_args *); 91 static int ufs_mknod (struct vop_old_mknod_args *); 92 static int ufs_mmap (struct vop_mmap_args *); 93 static int ufs_print (struct vop_print_args *); 94 static int ufs_readdir (struct vop_readdir_args *); 95 static int ufs_readlink (struct vop_readlink_args *); 96 static int ufs_remove (struct vop_old_remove_args *); 97 static int ufs_rename (struct vop_old_rename_args *); 98 static int ufs_rmdir (struct vop_old_rmdir_args *); 99 static int ufs_setattr (struct vop_setattr_args *); 100 static int ufs_strategy (struct vop_strategy_args *); 101 static int ufs_symlink (struct vop_old_symlink_args *); 102 static int ufs_whiteout (struct vop_old_whiteout_args *); 103 static int ufsfifo_close (struct vop_close_args *); 104 static int ufsfifo_kqfilter (struct vop_kqfilter_args *); 105 static int ufsfifo_read (struct vop_read_args *); 106 static int ufsfifo_write (struct vop_write_args *); 107 static int filt_ufsread (struct knote *kn, long hint); 108 static int filt_ufswrite (struct knote *kn, long hint); 109 static int filt_ufsvnode (struct knote *kn, long hint); 110 static void filt_ufsdetach (struct knote *kn); 111 static int ufs_kqfilter (struct vop_kqfilter_args *ap); 112 113 union _qcvt { 114 int64_t qcvt; 115 int32_t val[2]; 116 }; 117 #define SETHIGH(q, h) { \ 118 union _qcvt tmp; \ 119 tmp.qcvt = (q); \ 120 tmp.val[_QUAD_HIGHWORD] = (h); \ 121 (q) = tmp.qcvt; \ 122 } 123 #define SETLOW(q, l) { \ 124 union _qcvt tmp; \ 125 tmp.qcvt = (q); \ 126 tmp.val[_QUAD_LOWWORD] = (l); \ 127 (q) = tmp.qcvt; \ 128 } 129 #define VN_KNOTE(vp, b) \ 130 KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, (b)) 131 132 #define OFSFMT(vp) ((vp)->v_mount->mnt_maxsymlinklen <= 0) 133 134 /* 135 * A virgin directory (no blushing please). 136 */ 137 static struct dirtemplate mastertemplate = { 138 0, 12, DT_DIR, 1, ".", 139 0, DIRBLKSIZ - 12, DT_DIR, 2, ".." 140 }; 141 static struct odirtemplate omastertemplate = { 142 0, 12, 1, ".", 143 0, DIRBLKSIZ - 12, 2, ".." 144 }; 145 146 void 147 ufs_itimes(struct vnode *vp) 148 { 149 struct inode *ip; 150 struct timespec ts; 151 152 ip = VTOI(vp); 153 if ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE)) == 0) 154 return; 155 if ((vp->v_type == VBLK || vp->v_type == VCHR) && !DOINGSOFTDEP(vp)) 156 ip->i_flag |= IN_LAZYMOD; 157 else 158 ip->i_flag |= IN_MODIFIED; 159 if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 160 vfs_timestamp(&ts); 161 if (ip->i_flag & IN_ACCESS) { 162 ip->i_atime = ts.tv_sec; 163 ip->i_atimensec = ts.tv_nsec; 164 } 165 if (ip->i_flag & IN_UPDATE) { 166 ip->i_mtime = ts.tv_sec; 167 ip->i_mtimensec = ts.tv_nsec; 168 ip->i_modrev++; 169 } 170 if (ip->i_flag & IN_CHANGE) { 171 ip->i_ctime = ts.tv_sec; 172 ip->i_ctimensec = ts.tv_nsec; 173 } 174 } 175 ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_UPDATE); 176 } 177 178 /* 179 * Create a regular file 180 * 181 * ufs_create(struct vnode *a_dvp, struct vnode **a_vpp, 182 * struct componentname *a_cnp, struct vattr *a_vap) 183 */ 184 static 185 int 186 ufs_create(struct vop_old_create_args *ap) 187 { 188 int error; 189 190 error = 191 ufs_makeinode(MAKEIMODE(ap->a_vap->va_type, ap->a_vap->va_mode), 192 ap->a_dvp, ap->a_vpp, ap->a_cnp); 193 if (error) 194 return (error); 195 VN_KNOTE(ap->a_dvp, NOTE_WRITE); 196 return (0); 197 } 198 199 /* 200 * Mknod vnode call 201 * 202 * ufs_mknod(struct vnode *a_dvp, struct vnode **a_vpp, 203 * struct componentname *a_cnp, struct vattr *a_vap) 204 */ 205 /* ARGSUSED */ 206 static 207 int 208 ufs_mknod(struct vop_old_mknod_args *ap) 209 { 210 struct vattr *vap = ap->a_vap; 211 struct vnode **vpp = ap->a_vpp; 212 struct inode *ip; 213 ino_t ino; 214 int error; 215 216 /* 217 * UFS cannot represent the entire major/minor range supported by 218 * the kernel. 219 */ 220 if (vap->va_rmajor != VNOVAL && 221 makeudev(vap->va_rmajor, vap->va_rminor) == NOUDEV) { 222 return(EINVAL); 223 } 224 225 /* no special directory support */ 226 if (vap->va_type == VDIR) 227 return(EINVAL); 228 229 error = ufs_makeinode(MAKEIMODE(vap->va_type, vap->va_mode), 230 ap->a_dvp, vpp, ap->a_cnp); 231 if (error) 232 return (error); 233 VN_KNOTE(ap->a_dvp, NOTE_WRITE); 234 ip = VTOI(*vpp); 235 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 236 if (vap->va_rmajor != VNOVAL) { 237 /* 238 * Want to be able to use this to make badblock 239 * inodes, so don't truncate the dev number. 240 */ 241 ip->i_rdev = makeudev(vap->va_rmajor, vap->va_rminor); 242 } 243 /* 244 * Remove inode, then reload it through VFS_VGET so it is 245 * checked to see if it is an alias of an existing entry in 246 * the inode cache. 247 */ 248 (*vpp)->v_type = VNON; 249 ino = ip->i_number; /* Save this before vgone() invalidates ip. */ 250 vgone_vxlocked(*vpp); 251 vput(*vpp); 252 error = VFS_VGET(ap->a_dvp->v_mount, NULL, ino, vpp); 253 if (error) { 254 *vpp = NULL; 255 return (error); 256 } 257 return (0); 258 } 259 260 /* 261 * Close called. 262 * 263 * Update the times on the inode. 264 * 265 * ufs_close(struct vnode *a_vp, int a_fflag) 266 */ 267 /* ARGSUSED */ 268 static 269 int 270 ufs_close(struct vop_close_args *ap) 271 { 272 struct vnode *vp = ap->a_vp; 273 274 if (VREFCNT(vp) > 1) 275 ufs_itimes(vp); 276 return (vop_stdclose(ap)); 277 } 278 279 /* 280 * ufs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred) 281 */ 282 static 283 int 284 ufs_access(struct vop_access_args *ap) 285 { 286 struct vnode *vp = ap->a_vp; 287 struct inode *ip = VTOI(vp); 288 int error; 289 290 #ifdef QUOTA 291 if (ap->a_mode & VWRITE) { 292 switch (vp->v_type) { 293 case VDIR: 294 case VLNK: 295 case VREG: 296 if ((error = ufs_getinoquota(ip)) != 0) 297 return (error); 298 break; 299 default: 300 break; 301 } 302 } 303 #endif 304 305 error = vop_helper_access(ap, ip->i_uid, ip->i_gid, ip->i_mode, 0); 306 return (error); 307 } 308 309 /* 310 * ufs_getattr(struct vnode *a_vp, struct vattr *a_vap) 311 */ 312 /* ARGSUSED */ 313 static 314 int 315 ufs_getattr(struct vop_getattr_args *ap) 316 { 317 struct vnode *vp = ap->a_vp; 318 struct inode *ip = VTOI(vp); 319 struct vattr *vap = ap->a_vap; 320 321 ufs_itimes(vp); 322 /* 323 * Copy from inode table 324 */ 325 vap->va_fsid = dev2udev(ip->i_dev); 326 vap->va_fileid = ip->i_number; 327 vap->va_mode = ip->i_mode & ~IFMT; 328 vap->va_nlink = VFSTOUFS(vp->v_mount)->um_i_effnlink_valid ? 329 ip->i_effnlink : ip->i_nlink; 330 vap->va_uid = ip->i_uid; 331 vap->va_gid = ip->i_gid; 332 vap->va_rmajor = umajor(ip->i_rdev); 333 vap->va_rminor = uminor(ip->i_rdev); 334 vap->va_size = ip->i_din.di_size; 335 vap->va_atime.tv_sec = ip->i_atime; 336 vap->va_atime.tv_nsec = ip->i_atimensec; 337 vap->va_mtime.tv_sec = ip->i_mtime; 338 vap->va_mtime.tv_nsec = ip->i_mtimensec; 339 vap->va_ctime.tv_sec = ip->i_ctime; 340 vap->va_ctime.tv_nsec = ip->i_ctimensec; 341 vap->va_flags = ip->i_flags; 342 vap->va_gen = ip->i_gen; 343 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize; 344 vap->va_bytes = dbtob((u_quad_t)ip->i_blocks); 345 vap->va_type = IFTOVT(ip->i_mode); 346 vap->va_filerev = ip->i_modrev; 347 return (0); 348 } 349 350 static 351 int 352 ufs_markatime(struct vop_markatime_args *ap) 353 { 354 struct vnode *vp = ap->a_vp; 355 struct inode *ip = VTOI(vp); 356 357 if (vp->v_mount->mnt_flag & MNT_RDONLY) 358 return (EROFS); 359 if (vp->v_mount->mnt_flag & MNT_NOATIME) 360 return (0); 361 ip->i_flag |= IN_ACCESS; 362 VN_KNOTE(vp, NOTE_ATTRIB); 363 return (0); 364 } 365 366 /* 367 * Set attribute vnode op. called from several syscalls 368 * 369 * ufs_setattr(struct vnode *a_vp, struct vattr *a_vap, 370 * struct ucred *a_cred) 371 */ 372 static 373 int 374 ufs_setattr(struct vop_setattr_args *ap) 375 { 376 struct vattr *vap = ap->a_vap; 377 struct vnode *vp = ap->a_vp; 378 struct inode *ip = VTOI(vp); 379 struct ucred *cred = ap->a_cred; 380 int error; 381 382 /* 383 * Check for unsettable attributes. 384 */ 385 if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) || 386 (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) || 387 (vap->va_blocksize != VNOVAL) || (vap->va_rmajor != VNOVAL) || 388 ((int)vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) { 389 return (EINVAL); 390 } 391 if (vap->va_flags != VNOVAL) { 392 if (vp->v_mount->mnt_flag & MNT_RDONLY) 393 return (EROFS); 394 if (cred->cr_uid != ip->i_uid && 395 (error = priv_check_cred(cred, PRIV_VFS_SETATTR, 0))) 396 return (error); 397 /* 398 * Note that a root chflags becomes a user chflags when 399 * we are jailed, unless the jail.chflags_allowed sysctl 400 * is set. 401 */ 402 if (cred->cr_uid == 0 && 403 (!jailed(cred) || jail_chflags_allowed)) { 404 if ((ip->i_flags 405 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) && 406 securelevel > 0) 407 return (EPERM); 408 ip->i_flags = vap->va_flags; 409 } else { 410 if (ip->i_flags 411 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) || 412 (vap->va_flags & UF_SETTABLE) != vap->va_flags) 413 return (EPERM); 414 ip->i_flags &= SF_SETTABLE; 415 ip->i_flags |= (vap->va_flags & UF_SETTABLE); 416 } 417 ip->i_flag |= IN_CHANGE; 418 if (vap->va_flags & (IMMUTABLE | APPEND)) 419 return (0); 420 } 421 if (ip->i_flags & (IMMUTABLE | APPEND)) 422 return (EPERM); 423 /* 424 * Go through the fields and update iff not VNOVAL. 425 */ 426 if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) { 427 if (vp->v_mount->mnt_flag & MNT_RDONLY) 428 return (EROFS); 429 if ((error = ufs_chown(vp, vap->va_uid, vap->va_gid, cred)) != 0) 430 return (error); 431 } 432 if (vap->va_size != VNOVAL) { 433 /* 434 * Disallow write attempts on read-only filesystems; 435 * unless the file is a socket, fifo, or a block or 436 * character device resident on the filesystem. 437 */ 438 switch (vp->v_type) { 439 case VDIR: 440 return (EISDIR); 441 case VLNK: 442 case VREG: 443 if (vp->v_mount->mnt_flag & MNT_RDONLY) 444 return (EROFS); 445 break; 446 default: 447 break; 448 } 449 if ((error = ffs_truncate(vp, vap->va_size, 0, cred)) != 0) 450 return (error); 451 } 452 ip = VTOI(vp); 453 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) { 454 if (vp->v_mount->mnt_flag & MNT_RDONLY) 455 return (EROFS); 456 if (cred->cr_uid != ip->i_uid && 457 (error = priv_check_cred(cred, PRIV_VFS_SETATTR, 0)) && 458 ((vap->va_vaflags & VA_UTIMES_NULL) == 0 || 459 (error = VOP_EACCESS(vp, VWRITE, cred)))) 460 return (error); 461 if (vap->va_atime.tv_sec != VNOVAL) 462 ip->i_flag |= IN_ACCESS; 463 if (vap->va_mtime.tv_sec != VNOVAL) 464 ip->i_flag |= IN_CHANGE | IN_UPDATE; 465 ufs_itimes(vp); 466 if (vap->va_atime.tv_sec != VNOVAL) { 467 ip->i_atime = vap->va_atime.tv_sec; 468 ip->i_atimensec = vap->va_atime.tv_nsec; 469 } 470 if (vap->va_mtime.tv_sec != VNOVAL) { 471 ip->i_mtime = vap->va_mtime.tv_sec; 472 ip->i_mtimensec = vap->va_mtime.tv_nsec; 473 } 474 error = ffs_update(vp, 0); 475 if (error) 476 return (error); 477 } 478 error = 0; 479 if (vap->va_mode != (mode_t)VNOVAL) { 480 if (vp->v_mount->mnt_flag & MNT_RDONLY) 481 return (EROFS); 482 error = ufs_chmod(vp, (int)vap->va_mode, cred); 483 } 484 VN_KNOTE(vp, NOTE_ATTRIB); 485 return (error); 486 } 487 488 /* 489 * Change the mode on a file. 490 * Inode must be locked before calling. 491 */ 492 static int 493 ufs_chmod(struct vnode *vp, int mode, struct ucred *cred) 494 { 495 struct inode *ip = VTOI(vp); 496 int error; 497 mode_t cur_mode = ip->i_mode; 498 499 error = vop_helper_chmod(vp, mode, cred, ip->i_uid, ip->i_gid, 500 &cur_mode); 501 if (error) 502 return (error); 503 #if 0 504 if (cred->cr_uid != ip->i_uid) { 505 error = priv_check_cred(cred, PRIV_VFS_CHMOD, 0); 506 if (error) 507 return (error); 508 } 509 if (cred->cr_uid) { 510 if (vp->v_type != VDIR && (mode & S_ISTXT)) 511 return (EFTYPE); 512 if (!groupmember(ip->i_gid, cred) && (mode & ISGID)) 513 return (EPERM); 514 } 515 #endif 516 ip->i_mode = cur_mode; 517 ip->i_flag |= IN_CHANGE; 518 return (0); 519 } 520 521 /* 522 * Perform chown operation on inode ip; 523 * inode must be locked prior to call. 524 */ 525 static int 526 ufs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred) 527 { 528 struct inode *ip = VTOI(vp); 529 uid_t ouid; 530 gid_t ogid; 531 int error = 0; 532 #ifdef QUOTA 533 int i; 534 long change; 535 #endif 536 537 if (uid == (uid_t)VNOVAL) 538 uid = ip->i_uid; 539 if (gid == (gid_t)VNOVAL) 540 gid = ip->i_gid; 541 /* 542 * If we don't own the file, are trying to change the owner 543 * of the file, or are not a member of the target group, 544 * the caller must be superuser or the call fails. 545 */ 546 if ((cred->cr_uid != ip->i_uid || uid != ip->i_uid || 547 (gid != ip->i_gid && !(cred->cr_gid == gid || 548 groupmember(gid, cred)))) && 549 (error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0))) 550 return (error); 551 ogid = ip->i_gid; 552 ouid = ip->i_uid; 553 #ifdef QUOTA 554 if ((error = ufs_getinoquota(ip)) != 0) 555 return (error); 556 if (ouid == uid) { 557 ufs_dqrele(vp, ip->i_dquot[USRQUOTA]); 558 ip->i_dquot[USRQUOTA] = NODQUOT; 559 } 560 if (ogid == gid) { 561 ufs_dqrele(vp, ip->i_dquot[GRPQUOTA]); 562 ip->i_dquot[GRPQUOTA] = NODQUOT; 563 } 564 change = ip->i_blocks; 565 (void) ufs_chkdq(ip, -change, cred, CHOWN); 566 (void) ufs_chkiq(ip, -1, cred, CHOWN); 567 for (i = 0; i < MAXQUOTAS; i++) { 568 ufs_dqrele(vp, ip->i_dquot[i]); 569 ip->i_dquot[i] = NODQUOT; 570 } 571 #endif 572 ip->i_gid = gid; 573 ip->i_uid = uid; 574 #ifdef QUOTA 575 if ((error = ufs_getinoquota(ip)) == 0) { 576 if (ouid == uid) { 577 ufs_dqrele(vp, ip->i_dquot[USRQUOTA]); 578 ip->i_dquot[USRQUOTA] = NODQUOT; 579 } 580 if (ogid == gid) { 581 ufs_dqrele(vp, ip->i_dquot[GRPQUOTA]); 582 ip->i_dquot[GRPQUOTA] = NODQUOT; 583 } 584 if ((error = ufs_chkdq(ip, change, cred, CHOWN)) == 0) { 585 if ((error = ufs_chkiq(ip, 1, cred, CHOWN)) == 0) 586 goto good; 587 else 588 (void)ufs_chkdq(ip, -change, cred, CHOWN|FORCE); 589 } 590 for (i = 0; i < MAXQUOTAS; i++) { 591 ufs_dqrele(vp, ip->i_dquot[i]); 592 ip->i_dquot[i] = NODQUOT; 593 } 594 } 595 ip->i_gid = ogid; 596 ip->i_uid = ouid; 597 if (ufs_getinoquota(ip) == 0) { 598 if (ouid == uid) { 599 ufs_dqrele(vp, ip->i_dquot[USRQUOTA]); 600 ip->i_dquot[USRQUOTA] = NODQUOT; 601 } 602 if (ogid == gid) { 603 ufs_dqrele(vp, ip->i_dquot[GRPQUOTA]); 604 ip->i_dquot[GRPQUOTA] = NODQUOT; 605 } 606 (void) ufs_chkdq(ip, change, cred, FORCE|CHOWN); 607 (void) ufs_chkiq(ip, 1, cred, FORCE|CHOWN); 608 (void) ufs_getinoquota(ip); 609 } 610 return (error); 611 good: 612 if (ufs_getinoquota(ip)) 613 panic("ufs_chown: lost quota"); 614 #endif /* QUOTA */ 615 ip->i_flag |= IN_CHANGE; 616 if (cred->cr_uid != 0 && (ouid != uid || ogid != gid)) 617 ip->i_mode &= ~(ISUID | ISGID); 618 return (0); 619 } 620 621 /* 622 * Mmap a file 623 * 624 * NB Currently unsupported. 625 * 626 * ufs_mmap(struct vnode *a_vp, int a_fflags, struct ucred *a_cred) 627 */ 628 /* ARGSUSED */ 629 static 630 int 631 ufs_mmap(struct vop_mmap_args *ap) 632 { 633 return (EINVAL); 634 } 635 636 /* 637 * ufs_remove(struct vnode *a_dvp, struct vnode *a_vp, 638 * struct componentname *a_cnp) 639 */ 640 static 641 int 642 ufs_remove(struct vop_old_remove_args *ap) 643 { 644 struct inode *ip; 645 struct vnode *vp = ap->a_vp; 646 struct vnode *dvp = ap->a_dvp; 647 int error; 648 649 ip = VTOI(vp); 650 #if 0 /* handled by kernel now */ 651 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) || 652 (VTOI(dvp)->i_flags & APPEND)) { 653 error = EPERM; 654 goto out; 655 } 656 #endif 657 error = ufs_dirremove(dvp, ip, ap->a_cnp->cn_flags, 0); 658 VN_KNOTE(vp, NOTE_DELETE); 659 VN_KNOTE(dvp, NOTE_WRITE); 660 #if 0 661 out: 662 #endif 663 return (error); 664 } 665 666 /* 667 * link vnode call 668 * 669 * ufs_link(struct vnode *a_tdvp, struct vnode *a_vp, 670 * struct componentname *a_cnp) 671 */ 672 static 673 int 674 ufs_link(struct vop_old_link_args *ap) 675 { 676 struct vnode *vp = ap->a_vp; 677 struct vnode *tdvp = ap->a_tdvp; 678 struct componentname *cnp = ap->a_cnp; 679 struct inode *ip; 680 struct direct newdir; 681 int error; 682 683 if (tdvp->v_mount != vp->v_mount) { 684 error = EXDEV; 685 goto out2; 686 } 687 if (tdvp != vp) { 688 error = vn_lock(vp, LK_EXCLUSIVE | LK_FAILRECLAIM); 689 if (error) 690 goto out2; 691 } 692 ip = VTOI(vp); 693 if ((nlink_t)ip->i_nlink >= LINK_MAX) { 694 error = EMLINK; 695 goto out1; 696 } 697 #if 0 /* handled by kernel now, also DragonFly allows this */ 698 if (ip->i_flags & (IMMUTABLE | APPEND)) { 699 error = EPERM; 700 goto out1; 701 } 702 #endif 703 ip->i_effnlink++; 704 ip->i_nlink++; 705 ip->i_flag |= IN_CHANGE; 706 if (DOINGSOFTDEP(vp)) 707 softdep_change_linkcnt(ip); 708 error = ffs_update(vp, !(DOINGSOFTDEP(vp) | DOINGASYNC(vp))); 709 if (!error) { 710 ufs_makedirentry(ip, cnp, &newdir); 711 error = ufs_direnter(tdvp, vp, &newdir, cnp, NULL); 712 } 713 714 if (error) { 715 ip->i_effnlink--; 716 ip->i_nlink--; 717 ip->i_flag |= IN_CHANGE; 718 if (DOINGSOFTDEP(vp)) 719 softdep_change_linkcnt(ip); 720 } 721 out1: 722 if (tdvp != vp) 723 vn_unlock(vp); 724 out2: 725 VN_KNOTE(vp, NOTE_LINK); 726 VN_KNOTE(tdvp, NOTE_WRITE); 727 return (error); 728 } 729 730 /* 731 * whiteout vnode call 732 * 733 * ufs_whiteout(struct vnode *a_dvp, struct componentname *a_cnp, int a_flags) 734 */ 735 static 736 int 737 ufs_whiteout(struct vop_old_whiteout_args *ap) 738 { 739 struct vnode *dvp = ap->a_dvp; 740 struct componentname *cnp = ap->a_cnp; 741 struct direct newdir; 742 int error = 0; 743 744 switch (ap->a_flags) { 745 case NAMEI_LOOKUP: 746 /* 4.4 format directories support whiteout operations */ 747 if (dvp->v_mount->mnt_maxsymlinklen > 0) 748 return (0); 749 return (EOPNOTSUPP); 750 751 case NAMEI_CREATE: 752 /* create a new directory whiteout */ 753 #ifdef DIAGNOSTIC 754 if (dvp->v_mount->mnt_maxsymlinklen <= 0) 755 panic("ufs_whiteout: old format filesystem"); 756 #endif 757 758 newdir.d_ino = WINO; 759 newdir.d_namlen = cnp->cn_namelen; 760 bcopy(cnp->cn_nameptr, newdir.d_name, (unsigned)cnp->cn_namelen + 1); 761 newdir.d_type = DT_WHT; 762 error = ufs_direnter(dvp, NULL, &newdir, cnp, NULL); 763 break; 764 765 case NAMEI_DELETE: 766 /* remove an existing directory whiteout */ 767 #ifdef DIAGNOSTIC 768 if (dvp->v_mount->mnt_maxsymlinklen <= 0) 769 panic("ufs_whiteout: old format filesystem"); 770 #endif 771 772 cnp->cn_flags &= ~CNP_DOWHITEOUT; 773 error = ufs_dirremove(dvp, NULL, cnp->cn_flags, 0); 774 break; 775 default: 776 panic("ufs_whiteout: unknown op"); 777 } 778 return (error); 779 } 780 781 /* 782 * Rename system call. 783 * rename("foo", "bar"); 784 * is essentially 785 * unlink("bar"); 786 * link("foo", "bar"); 787 * unlink("foo"); 788 * but ``atomically''. Can't do full commit without saving state in the 789 * inode on disk which isn't feasible at this time. Best we can do is 790 * always guarantee the target exists. 791 * 792 * Basic algorithm is: 793 * 794 * 1) Bump link count on source while we're linking it to the 795 * target. This also ensure the inode won't be deleted out 796 * from underneath us while we work (it may be truncated by 797 * a concurrent `trunc' or `open' for creation). 798 * 2) Link source to destination. If destination already exists, 799 * delete it first. 800 * 3) Unlink source reference to inode if still around. If a 801 * directory was moved and the parent of the destination 802 * is different from the source, patch the ".." entry in the 803 * directory. 804 * 805 * ufs_rename(struct vnode *a_fdvp, struct vnode *a_fvp, 806 * struct componentname *a_fcnp, struct vnode *a_tdvp, 807 * struct vnode *a_tvp, struct componentname *a_tcnp) 808 */ 809 static 810 int 811 ufs_rename(struct vop_old_rename_args *ap) 812 { 813 struct vnode *tvp = ap->a_tvp; 814 struct vnode *tdvp = ap->a_tdvp; 815 struct vnode *fvp = ap->a_fvp; 816 struct vnode *fdvp = ap->a_fdvp; 817 struct componentname *tcnp = ap->a_tcnp; 818 struct componentname *fcnp = ap->a_fcnp; 819 struct inode *ip, *xp, *dp; 820 struct direct newdir; 821 ino_t oldparent = 0, newparent = 0; 822 int doingdirectory = 0; 823 int error = 0, ioflag; 824 825 /* 826 * Check for cross-device rename. 827 */ 828 if ((fvp->v_mount != tdvp->v_mount) || 829 (tvp && (fvp->v_mount != tvp->v_mount))) { 830 error = EXDEV; 831 abortit: 832 if (tdvp == tvp) 833 vrele(tdvp); 834 else 835 vput(tdvp); 836 if (tvp) 837 vput(tvp); 838 vrele(fdvp); 839 vrele(fvp); 840 return (error); 841 } 842 843 #if 0 /* handled by kernel now */ 844 if (tvp && ((VTOI(tvp)->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) || 845 (VTOI(tdvp)->i_flags & APPEND))) { 846 error = EPERM; 847 goto abortit; 848 } 849 #endif 850 851 /* 852 * Renaming a file to itself has no effect. The upper layers should 853 * not call us in that case. Temporarily just warn if they do. 854 */ 855 if (fvp == tvp) { 856 kprintf("ufs_rename: fvp == tvp (can't happen)\n"); 857 error = 0; 858 goto abortit; 859 } 860 861 error = vn_lock(fvp, LK_EXCLUSIVE | LK_FAILRECLAIM); 862 if (error) 863 goto abortit; 864 865 /* 866 * Note: now that fvp is locked we have to be sure to unlock it before 867 * using the 'abortit' target. 868 */ 869 dp = VTOI(fdvp); 870 ip = VTOI(fvp); 871 if (ip->i_nlink >= LINK_MAX) { 872 vn_unlock(fvp); 873 error = EMLINK; 874 goto abortit; 875 } 876 #if 0 /* handled by kernel now */ 877 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) 878 || (dp->i_flags & APPEND)) { 879 vn_unlock(fvp); 880 error = EPERM; 881 goto abortit; 882 } 883 #endif 884 if ((ip->i_mode & IFMT) == IFDIR) { 885 /* 886 * Avoid ".", "..", and aliases of "." for obvious reasons. 887 */ 888 if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') || 889 dp == ip || (fcnp->cn_flags | tcnp->cn_flags) & CNP_ISDOTDOT || 890 (ip->i_flag & IN_RENAME)) { 891 vn_unlock(fvp); 892 error = EINVAL; 893 goto abortit; 894 } 895 ip->i_flag |= IN_RENAME; 896 oldparent = dp->i_number; 897 doingdirectory = 1; 898 } 899 VN_KNOTE(fdvp, NOTE_WRITE); /* XXX right place? */ 900 901 /* 902 * fvp still locked. ip->i_flag has IN_RENAME set if doingdirectory. 903 * Cleanup fvp requirements so we can unlock it. 904 * 905 * tvp and tdvp are locked. tvp may be NULL. Now that dp and xp 906 * is setup we can use the 'bad' target if we unlock fvp. We cannot 907 * use the abortit target anymore because of IN_RENAME. 908 */ 909 dp = VTOI(tdvp); 910 if (tvp) 911 xp = VTOI(tvp); 912 else 913 xp = NULL; 914 915 /* 916 * 1) Bump link count while we're moving stuff 917 * around. If we crash somewhere before 918 * completing our work, the link count 919 * may be wrong, but correctable. 920 */ 921 ip->i_effnlink++; 922 ip->i_nlink++; 923 ip->i_flag |= IN_CHANGE; 924 if (DOINGSOFTDEP(fvp)) 925 softdep_change_linkcnt(ip); 926 if ((error = ffs_update(fvp, !(DOINGSOFTDEP(fvp) | 927 DOINGASYNC(fvp)))) != 0) { 928 vn_unlock(fvp); 929 goto bad; 930 } 931 932 /* 933 * If ".." must be changed (ie the directory gets a new 934 * parent) then the source directory must not be in the 935 * directory heirarchy above the target, as this would 936 * orphan everything below the source directory. Also 937 * the user must have write permission in the source so 938 * as to be able to change "..". We must repeat the call 939 * to namei, as the parent directory is unlocked by the 940 * call to checkpath(). 941 */ 942 error = VOP_EACCESS(fvp, VWRITE, tcnp->cn_cred); 943 vn_unlock(fvp); 944 945 /* 946 * We are now back to where we were in that fvp, fdvp are unlocked 947 * and tvp, tdvp are locked. tvp may be NULL. IN_RENAME may be 948 * set. Only the bad target or, if we clean up tvp and tdvp, the 949 * out target, may be used. 950 */ 951 if (oldparent != dp->i_number) 952 newparent = dp->i_number; 953 if (doingdirectory && newparent) { 954 if (error) /* write access check above */ 955 goto bad; 956 957 /* 958 * Once we start messing with tvp and tdvp we cannot use the 959 * 'bad' target, only finish cleaning tdvp and tvp up and 960 * use the 'out' target. 961 * 962 * This cleans up tvp. 963 */ 964 if (xp != NULL) { 965 vput(tvp); 966 xp = NULL; 967 } 968 969 /* 970 * This is a real mess. ufs_checkpath vput's the target 971 * directory so retain an extra ref and note that tdvp will 972 * lose its lock on return. This leaves us with one good 973 * ref after ufs_checkpath returns. 974 */ 975 vref(tdvp); 976 error = ufs_checkpath(ip, dp, tcnp->cn_cred); 977 tcnp->cn_flags |= CNP_PDIRUNLOCK; 978 if (error) { 979 vrele(tdvp); 980 goto out; 981 } 982 983 /* 984 * relookup no longer messes with tdvp's refs. tdvp must be 985 * unlocked on entry and will be locked on a successful 986 * return. 987 */ 988 error = relookup(tdvp, &tvp, tcnp); 989 if (error) { 990 if (tcnp->cn_flags & CNP_PDIRUNLOCK) 991 vrele(tdvp); 992 else 993 vput(tdvp); 994 goto out; 995 } 996 KKASSERT((tcnp->cn_flags & CNP_PDIRUNLOCK) == 0); 997 dp = VTOI(tdvp); 998 if (tvp) 999 xp = VTOI(tvp); 1000 } 1001 1002 /* 1003 * We are back to fvp, fdvp unlocked, tvp, tdvp locked. tvp may 1004 * be NULL (xp will also be NULL in that case), and IN_RENAME will 1005 * be set if doingdirectory. This means we can use the 'bad' target 1006 * again. 1007 */ 1008 1009 /* 1010 * 2) If target doesn't exist, link the target 1011 * to the source and unlink the source. 1012 * Otherwise, rewrite the target directory 1013 * entry to reference the source inode and 1014 * expunge the original entry's existence. 1015 */ 1016 if (xp == NULL) { 1017 if (dp->i_dev != ip->i_dev) 1018 panic("ufs_rename: EXDEV"); 1019 /* 1020 * Account for ".." in new directory. 1021 * When source and destination have the same 1022 * parent we don't fool with the link count. 1023 */ 1024 if (doingdirectory && newparent) { 1025 if ((nlink_t)dp->i_nlink >= LINK_MAX) { 1026 error = EMLINK; 1027 goto bad; 1028 } 1029 dp->i_effnlink++; 1030 dp->i_nlink++; 1031 dp->i_flag |= IN_CHANGE; 1032 if (DOINGSOFTDEP(tdvp)) 1033 softdep_change_linkcnt(dp); 1034 error = ffs_update(tdvp, !(DOINGSOFTDEP(tdvp) | 1035 DOINGASYNC(tdvp))); 1036 if (error) 1037 goto bad; 1038 } 1039 ufs_makedirentry(ip, tcnp, &newdir); 1040 error = ufs_direnter(tdvp, NULL, &newdir, tcnp, NULL); 1041 if (error) { 1042 if (doingdirectory && newparent) { 1043 dp->i_effnlink--; 1044 dp->i_nlink--; 1045 dp->i_flag |= IN_CHANGE; 1046 if (DOINGSOFTDEP(tdvp)) 1047 softdep_change_linkcnt(dp); 1048 (void)ffs_update(tdvp, 1); 1049 } 1050 goto bad; 1051 } 1052 VN_KNOTE(tdvp, NOTE_WRITE); 1053 vput(tdvp); 1054 } else { 1055 if (xp->i_dev != dp->i_dev || xp->i_dev != ip->i_dev) 1056 panic("ufs_rename: EXDEV"); 1057 /* 1058 * Short circuit rename(foo, foo). 1059 */ 1060 if (xp->i_number == ip->i_number) 1061 panic("ufs_rename: same file"); 1062 /* 1063 * If the parent directory is "sticky", then the user must 1064 * own the parent directory, or the destination of the rename, 1065 * otherwise the destination may not be changed (except by 1066 * root). This implements append-only directories. 1067 */ 1068 if ((dp->i_mode & S_ISTXT) && tcnp->cn_cred->cr_uid != 0 && 1069 tcnp->cn_cred->cr_uid != dp->i_uid && 1070 xp->i_uid != tcnp->cn_cred->cr_uid) { 1071 error = EPERM; 1072 goto bad; 1073 } 1074 /* 1075 * Target must be empty if a directory and have no links 1076 * to it. Also, ensure source and target are compatible 1077 * (both directories, or both not directories). 1078 * 1079 * Purge the file or directory being replaced from the 1080 * nameccache. 1081 */ 1082 if ((xp->i_mode&IFMT) == IFDIR) { 1083 if ((xp->i_effnlink > 2) || 1084 !ufs_dirempty(xp, dp->i_number, tcnp->cn_cred)) { 1085 error = ENOTEMPTY; 1086 goto bad; 1087 } 1088 if (!doingdirectory) { 1089 error = ENOTDIR; 1090 goto bad; 1091 } 1092 /* cache_purge removed - handled by VFS compat layer */ 1093 } else if (doingdirectory == 0) { 1094 /* cache_purge removed - handled by VFS compat layer */ 1095 } else { 1096 error = EISDIR; 1097 goto bad; 1098 } 1099 /* 1100 * note: inode passed to ufs_dirrewrite() is 0 for a 1101 * non-directory file rename, 1 for a directory rename 1102 * in the same directory, and > 1 for an inode representing 1103 * the new directory. 1104 */ 1105 error = ufs_dirrewrite(dp, xp, ip->i_number, 1106 IFTODT(ip->i_mode), 1107 (doingdirectory && newparent) ? 1108 newparent : (ino_t)doingdirectory); 1109 if (error) 1110 goto bad; 1111 if (doingdirectory) { 1112 if (!newparent) { 1113 dp->i_effnlink--; 1114 if (DOINGSOFTDEP(tdvp)) 1115 softdep_change_linkcnt(dp); 1116 } 1117 xp->i_effnlink--; 1118 if (DOINGSOFTDEP(tvp)) 1119 softdep_change_linkcnt(xp); 1120 } 1121 if (doingdirectory && !DOINGSOFTDEP(tvp)) { 1122 /* 1123 * Truncate inode. The only stuff left in the directory 1124 * is "." and "..". The "." reference is inconsequential 1125 * since we are quashing it. We have removed the "." 1126 * reference and the reference in the parent directory, 1127 * but there may be other hard links. The soft 1128 * dependency code will arrange to do these operations 1129 * after the parent directory entry has been deleted on 1130 * disk, so when running with that code we avoid doing 1131 * them now. 1132 */ 1133 if (!newparent) { 1134 dp->i_nlink--; 1135 dp->i_flag |= IN_CHANGE; 1136 } 1137 xp->i_nlink--; 1138 xp->i_flag |= IN_CHANGE; 1139 ioflag = DOINGASYNC(tvp) ? 0 : IO_SYNC; 1140 error = ffs_truncate(tvp, (off_t)0, ioflag, 1141 tcnp->cn_cred); 1142 if (error) 1143 goto bad; 1144 } 1145 VN_KNOTE(tdvp, NOTE_WRITE); 1146 vput(tdvp); 1147 VN_KNOTE(tvp, NOTE_DELETE); 1148 vput(tvp); 1149 xp = NULL; 1150 } 1151 1152 /* 1153 * tvp and tdvp have been cleaned up. only fvp and fdvp (both 1154 * unlocked) remain. We are about to overwrite fvp but we have to 1155 * keep 'ip' intact so we cannot release the old fvp, which is still 1156 * refd and accessible via ap->a_fvp. 1157 * 1158 * This means we cannot use either 'bad' or 'out' to cleanup any 1159 * more. 1160 */ 1161 1162 /* 1163 * 3) Unlink the source. 1164 */ 1165 fcnp->cn_flags &= ~CNP_MODMASK; 1166 fcnp->cn_flags |= CNP_LOCKPARENT; 1167 error = relookup(fdvp, &fvp, fcnp); 1168 if (error || fvp == NULL) { 1169 /* 1170 * From name has disappeared. IN_RENAME will not be set if 1171 * we get past the panic so we don't have to clean it up. 1172 */ 1173 if (doingdirectory) 1174 panic("ufs_rename: lost dir entry"); 1175 vrele(ap->a_fvp); 1176 if (fcnp->cn_flags & CNP_PDIRUNLOCK) 1177 vrele(fdvp); 1178 else 1179 vput(fdvp); 1180 return(0); 1181 } 1182 KKASSERT((fcnp->cn_flags & CNP_PDIRUNLOCK) == 0); 1183 1184 /* 1185 * fdvp and fvp are locked. 1186 */ 1187 xp = VTOI(fvp); 1188 dp = VTOI(fdvp); 1189 1190 /* 1191 * Ensure that the directory entry still exists and has not 1192 * changed while the new name has been entered. If the source is 1193 * a file then the entry may have been unlinked or renamed. In 1194 * either case there is no further work to be done. If the source 1195 * is a directory then it cannot have been rmdir'ed; the IN_RENAME 1196 * flag ensures that it cannot be moved by another rename or removed 1197 * by a rmdir. Cleanup IN_RENAME. 1198 */ 1199 if (xp != ip) { 1200 if (doingdirectory) 1201 panic("ufs_rename: lost dir entry"); 1202 } else { 1203 /* 1204 * If the source is a directory with a 1205 * new parent, the link count of the old 1206 * parent directory must be decremented 1207 * and ".." set to point to the new parent. 1208 */ 1209 if (doingdirectory && newparent) { 1210 xp->i_offset = mastertemplate.dot_reclen; 1211 ufs_dirrewrite(xp, dp, newparent, DT_DIR, 0); 1212 /* cache_purge removed - handled by VFS compat layer */ 1213 } 1214 error = ufs_dirremove(fdvp, xp, fcnp->cn_flags, 0); 1215 xp->i_flag &= ~IN_RENAME; 1216 } 1217 1218 VN_KNOTE(fvp, NOTE_RENAME); 1219 vput(fdvp); 1220 vput(fvp); 1221 vrele(ap->a_fvp); 1222 return (error); 1223 1224 bad: 1225 if (xp) 1226 vput(ITOV(xp)); 1227 vput(ITOV(dp)); 1228 out: 1229 if (doingdirectory) 1230 ip->i_flag &= ~IN_RENAME; 1231 if (vn_lock(fvp, LK_EXCLUSIVE | LK_FAILRECLAIM) == 0) { 1232 ip->i_effnlink--; 1233 ip->i_nlink--; 1234 ip->i_flag |= IN_CHANGE; 1235 ip->i_flag &= ~IN_RENAME; 1236 if (DOINGSOFTDEP(fvp)) 1237 softdep_change_linkcnt(ip); 1238 vput(fvp); 1239 } else { 1240 vrele(fvp); 1241 } 1242 return (error); 1243 } 1244 1245 /* 1246 * Mkdir system call 1247 * 1248 * ufs_mkdir(struct vnode *a_dvp, struct vnode **a_vpp, 1249 * struct componentname *a_cnp, struct vattr *a_vap) 1250 */ 1251 static 1252 int 1253 ufs_mkdir(struct vop_old_mkdir_args *ap) 1254 { 1255 struct vnode *dvp = ap->a_dvp; 1256 struct vattr *vap = ap->a_vap; 1257 struct componentname *cnp = ap->a_cnp; 1258 struct inode *ip, *dp; 1259 struct vnode *tvp; 1260 struct buf *bp; 1261 struct dirtemplate dirtemplate, *dtp; 1262 struct direct newdir; 1263 int error, dmode; 1264 long blkoff; 1265 1266 dp = VTOI(dvp); 1267 if ((nlink_t)dp->i_nlink >= LINK_MAX) { 1268 error = EMLINK; 1269 goto out; 1270 } 1271 dmode = vap->va_mode & 0777; 1272 dmode |= IFDIR; 1273 /* 1274 * Must simulate part of ufs_makeinode here to acquire the inode, 1275 * but not have it entered in the parent directory. The entry is 1276 * made later after writing "." and ".." entries. 1277 */ 1278 error = ffs_valloc(dvp, dmode, cnp->cn_cred, &tvp); 1279 if (error) 1280 goto out; 1281 ip = VTOI(tvp); 1282 ip->i_gid = dp->i_gid; 1283 #ifdef SUIDDIR 1284 { 1285 #ifdef QUOTA 1286 struct ucred ucred, *ucp; 1287 ucp = cnp->cn_cred; 1288 #endif 1289 /* 1290 * If we are hacking owners here, (only do this where told to) 1291 * and we are not giving it TO root, (would subvert quotas) 1292 * then go ahead and give it to the other user. 1293 * The new directory also inherits the SUID bit. 1294 * If user's UID and dir UID are the same, 1295 * 'give it away' so that the SUID is still forced on. 1296 */ 1297 if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) && 1298 (dp->i_mode & ISUID) && dp->i_uid) { 1299 dmode |= ISUID; 1300 ip->i_uid = dp->i_uid; 1301 #ifdef QUOTA 1302 if (dp->i_uid != cnp->cn_cred->cr_uid) { 1303 /* 1304 * Make sure the correct user gets charged 1305 * for the space. 1306 * Make a dummy credential for the victim. 1307 * XXX This seems to never be accessed out of 1308 * our context so a stack variable is ok. 1309 */ 1310 ucred.cr_ref = 1; 1311 ucred.cr_uid = ip->i_uid; 1312 ucred.cr_ngroups = 1; 1313 ucred.cr_groups[0] = dp->i_gid; 1314 ucp = &ucred; 1315 } 1316 #endif 1317 } else 1318 ip->i_uid = cnp->cn_cred->cr_uid; 1319 #ifdef QUOTA 1320 if ((error = ufs_getinoquota(ip)) || 1321 (error = ufs_chkiq(ip, 1, ucp, 0))) { 1322 ffs_vfree(tvp, ip->i_number, dmode); 1323 vput(tvp); 1324 return (error); 1325 } 1326 #endif 1327 } 1328 #else /* !SUIDDIR */ 1329 ip->i_uid = cnp->cn_cred->cr_uid; 1330 #ifdef QUOTA 1331 if ((error = ufs_getinoquota(ip)) || 1332 (error = ufs_chkiq(ip, 1, cnp->cn_cred, 0))) { 1333 ffs_vfree(tvp, ip->i_number, dmode); 1334 vput(tvp); 1335 return (error); 1336 } 1337 #endif 1338 #endif /* !SUIDDIR */ 1339 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 1340 ip->i_mode = dmode; 1341 tvp->v_type = VDIR; /* Rest init'd in getnewvnode(). */ 1342 ip->i_effnlink = 2; 1343 ip->i_nlink = 2; 1344 if (DOINGSOFTDEP(tvp)) 1345 softdep_change_linkcnt(ip); 1346 if (cnp->cn_flags & CNP_ISWHITEOUT) 1347 ip->i_flags |= UF_OPAQUE; 1348 1349 /* 1350 * Bump link count in parent directory to reflect work done below. 1351 * Should be done before reference is created so cleanup is 1352 * possible if we crash. 1353 */ 1354 dp->i_effnlink++; 1355 dp->i_nlink++; 1356 dp->i_flag |= IN_CHANGE; 1357 if (DOINGSOFTDEP(dvp)) 1358 softdep_change_linkcnt(dp); 1359 error = ffs_update(tvp, !(DOINGSOFTDEP(dvp) | DOINGASYNC(dvp))); 1360 if (error) 1361 goto bad; 1362 1363 /* 1364 * The vnode must have a VM object in order to issue buffer cache 1365 * ops on it. 1366 */ 1367 vinitvmio(tvp, DIRBLKSIZ, DIRBLKSIZ, -1); 1368 1369 /* 1370 * Initialize directory with "." and ".." from static template. 1371 */ 1372 if (dvp->v_mount->mnt_maxsymlinklen > 0) 1373 dtp = &mastertemplate; 1374 else 1375 dtp = (struct dirtemplate *)&omastertemplate; 1376 dirtemplate = *dtp; 1377 dirtemplate.dot_ino = ip->i_number; 1378 dirtemplate.dotdot_ino = dp->i_number; 1379 nvnode_pager_setsize(tvp, DIRBLKSIZ, DIRBLKSIZ, -1); 1380 error = VOP_BALLOC(tvp, 0LL, DIRBLKSIZ, cnp->cn_cred, B_CLRBUF, &bp); 1381 if (error) 1382 goto bad; 1383 ip->i_size = DIRBLKSIZ; 1384 ip->i_flag |= IN_CHANGE | IN_UPDATE; 1385 bcopy((caddr_t)&dirtemplate, (caddr_t)bp->b_data, sizeof dirtemplate); 1386 if (DOINGSOFTDEP(tvp)) { 1387 /* 1388 * Ensure that the entire newly allocated block is a 1389 * valid directory so that future growth within the 1390 * block does not have to ensure that the block is 1391 * written before the inode. 1392 */ 1393 blkoff = DIRBLKSIZ; 1394 while (blkoff < bp->b_bcount) { 1395 ((struct direct *) 1396 (bp->b_data + blkoff))->d_reclen = DIRBLKSIZ; 1397 blkoff += DIRBLKSIZ; 1398 } 1399 } 1400 if ((error = ffs_update(tvp, !(DOINGSOFTDEP(tvp) | 1401 DOINGASYNC(tvp)))) != 0) { 1402 bwrite(bp); 1403 goto bad; 1404 } 1405 /* 1406 * Directory set up, now install its entry in the parent directory. 1407 * 1408 * If we are not doing soft dependencies, then we must write out the 1409 * buffer containing the new directory body before entering the new 1410 * name in the parent. If we are doing soft dependencies, then the 1411 * buffer containing the new directory body will be passed to and 1412 * released in the soft dependency code after the code has attached 1413 * an appropriate ordering dependency to the buffer which ensures that 1414 * the buffer is written before the new name is written in the parent. 1415 */ 1416 if (DOINGASYNC(dvp)) 1417 bdwrite(bp); 1418 else if (!DOINGSOFTDEP(dvp) && (error = bwrite(bp)) != 0) 1419 goto bad; 1420 ufs_makedirentry(ip, cnp, &newdir); 1421 error = ufs_direnter(dvp, tvp, &newdir, cnp, bp); 1422 1423 bad: 1424 if (error == 0) { 1425 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK); 1426 *ap->a_vpp = tvp; 1427 } else { 1428 dp->i_effnlink--; 1429 dp->i_nlink--; 1430 dp->i_flag |= IN_CHANGE; 1431 if (DOINGSOFTDEP(dvp)) 1432 softdep_change_linkcnt(dp); 1433 /* 1434 * No need to do an explicit VOP_TRUNCATE here, vrele will 1435 * do this for us because we set the link count to 0. 1436 */ 1437 ip->i_effnlink = 0; 1438 ip->i_nlink = 0; 1439 ip->i_flag |= IN_CHANGE; 1440 if (DOINGSOFTDEP(tvp)) 1441 softdep_change_linkcnt(ip); 1442 vput(tvp); 1443 } 1444 out: 1445 return (error); 1446 } 1447 1448 /* 1449 * Rmdir system call. 1450 * 1451 * ufs_rmdir(struct vnode *a_dvp, struct vnode *a_vp, 1452 * struct componentname *a_cnp) 1453 */ 1454 static 1455 int 1456 ufs_rmdir(struct vop_old_rmdir_args *ap) 1457 { 1458 struct vnode *vp = ap->a_vp; 1459 struct vnode *dvp = ap->a_dvp; 1460 struct componentname *cnp = ap->a_cnp; 1461 struct inode *ip, *dp; 1462 int error, ioflag; 1463 1464 ip = VTOI(vp); 1465 dp = VTOI(dvp); 1466 1467 /* 1468 * Do not remove a directory that is in the process of being renamed. 1469 * Verify the directory is empty (and valid). Rmdir ".." will not be 1470 * valid since ".." will contain a reference to the current directory 1471 * and thus be non-empty. Do not allow the removal of mounted on 1472 * directories (this can happen when an NFS exported filesystem 1473 * tries to remove a locally mounted on directory). 1474 */ 1475 error = 0; 1476 if (ip->i_flag & IN_RENAME) { 1477 error = EINVAL; 1478 goto out; 1479 } 1480 if (ip->i_effnlink != 2 || 1481 !ufs_dirempty(ip, dp->i_number, cnp->cn_cred)) { 1482 error = ENOTEMPTY; 1483 goto out; 1484 } 1485 #if 0 /* handled by kernel now */ 1486 if ((dp->i_flags & APPEND) 1487 || (ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))) { 1488 error = EPERM; 1489 goto out; 1490 } 1491 #endif 1492 /* 1493 * Delete reference to directory before purging 1494 * inode. If we crash in between, the directory 1495 * will be reattached to lost+found, 1496 */ 1497 dp->i_effnlink--; 1498 ip->i_effnlink--; 1499 if (DOINGSOFTDEP(vp)) { 1500 softdep_change_linkcnt(dp); 1501 softdep_change_linkcnt(ip); 1502 } 1503 error = ufs_dirremove(dvp, ip, cnp->cn_flags, 1); 1504 if (error) { 1505 dp->i_effnlink++; 1506 ip->i_effnlink++; 1507 if (DOINGSOFTDEP(vp)) { 1508 softdep_change_linkcnt(dp); 1509 softdep_change_linkcnt(ip); 1510 } 1511 goto out; 1512 } 1513 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK); 1514 /* 1515 * Truncate inode. The only stuff left in the directory is "." and 1516 * "..". The "." reference is inconsequential since we are quashing 1517 * it. The soft dependency code will arrange to do these operations 1518 * after the parent directory entry has been deleted on disk, so 1519 * when running with that code we avoid doing them now. 1520 */ 1521 if (!DOINGSOFTDEP(vp)) { 1522 dp->i_nlink--; 1523 dp->i_flag |= IN_CHANGE; 1524 ip->i_nlink--; 1525 ip->i_flag |= IN_CHANGE; 1526 ioflag = DOINGASYNC(vp) ? 0 : IO_SYNC; 1527 error = ffs_truncate(vp, (off_t)0, ioflag, cnp->cn_cred); 1528 } 1529 /* cache_purge removed - handled by VFS compat layer */ 1530 #ifdef UFS_DIRHASH 1531 /* Kill any active hash; i_effnlink == 0, so it will not come back. */ 1532 if (ip->i_dirhash != NULL) 1533 ufsdirhash_free(ip); 1534 #endif 1535 out: 1536 VN_KNOTE(vp, NOTE_DELETE); 1537 return (error); 1538 } 1539 1540 /* 1541 * symlink -- make a symbolic link 1542 * 1543 * ufs_symlink(struct vnode *a_dvp, struct vnode **a_vpp, 1544 * struct componentname *a_cnp, struct vattr *a_vap, 1545 * char *a_target) 1546 */ 1547 static 1548 int 1549 ufs_symlink(struct vop_old_symlink_args *ap) 1550 { 1551 struct vnode *vp, **vpp = ap->a_vpp; 1552 struct inode *ip; 1553 int len, error; 1554 1555 error = ufs_makeinode(IFLNK | ap->a_vap->va_mode, ap->a_dvp, 1556 vpp, ap->a_cnp); 1557 if (error) 1558 return (error); 1559 VN_KNOTE(ap->a_dvp, NOTE_WRITE); 1560 vp = *vpp; 1561 len = strlen(ap->a_target); 1562 if (len < vp->v_mount->mnt_maxsymlinklen) { 1563 ip = VTOI(vp); 1564 bcopy(ap->a_target, (char *)ip->i_shortlink, len); 1565 ip->i_size = len; 1566 ip->i_flag |= IN_CHANGE | IN_UPDATE; 1567 } else { 1568 /* 1569 * Make sure we have a VM object in order to use 1570 * the buffer cache. 1571 */ 1572 if (vp->v_object == NULL) 1573 vinitvmio(vp, 0, PAGE_SIZE, -1); 1574 error = vn_rdwr(UIO_WRITE, vp, ap->a_target, len, (off_t)0, 1575 UIO_SYSSPACE, IO_NODELOCKED, 1576 ap->a_cnp->cn_cred, NULL); 1577 } 1578 if (error) 1579 vput(vp); 1580 return (error); 1581 } 1582 1583 /* 1584 * Vnode op for reading directories. 1585 * 1586 * ufs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred, 1587 * int *a_eofflag, int *ncookies, off_t **a_cookies) 1588 */ 1589 static 1590 int 1591 ufs_readdir(struct vop_readdir_args *ap) 1592 { 1593 struct uio *uio = ap->a_uio; 1594 struct vnode *vp = ap->a_vp; 1595 struct direct *dp; 1596 struct buf *bp; 1597 int retval; 1598 int error; 1599 int offset; /* offset into buffer cache buffer */ 1600 int eoffset; /* end of buffer clipped to file EOF */ 1601 int pickup; /* pickup point */ 1602 int ncookies; 1603 int cookie_index; 1604 off_t *cookies; 1605 1606 if (uio->uio_offset < 0) 1607 return (EINVAL); 1608 /* 1609 * Guess the number of cookies needed. Make sure we compute at 1610 * least 1, and no more then a reasonable limit. 1611 */ 1612 if (ap->a_ncookies) { 1613 ncookies = uio->uio_resid / 16 + 1; 1614 if (ncookies > 1024) 1615 ncookies = 1024; 1616 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK); 1617 } else { 1618 ncookies = -1; /* force conditionals below */ 1619 cookies = NULL; 1620 } 1621 cookie_index = 0; 1622 1623 error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY | LK_FAILRECLAIM); 1624 if (error) 1625 return (error); 1626 1627 /* 1628 * Past or at EOF 1629 */ 1630 if (uio->uio_offset >= VTOI(vp)->i_size) { 1631 if (ap->a_eofflag) 1632 *ap->a_eofflag = 1; 1633 if (ap->a_ncookies) { 1634 *ap->a_ncookies = cookie_index; 1635 *ap->a_cookies = cookies; 1636 } 1637 goto done; 1638 } 1639 1640 /* 1641 * Loop until we run out of cookies, we run out of user buffer, 1642 * or we hit the directory EOF. 1643 * 1644 * Always start scans at the beginning of the buffer, don't trust 1645 * the offset supplied by userland. 1646 */ 1647 while ((error = ffs_blkatoff_ra(vp, uio->uio_offset, NULL, &bp, 2)) == 0) { 1648 pickup = (int)(uio->uio_offset - bp->b_loffset); 1649 offset = 0; 1650 retval = 0; 1651 if (bp->b_loffset + bp->b_bcount > VTOI(vp)->i_size) 1652 eoffset = (int)(VTOI(vp)->i_size - bp->b_loffset); 1653 else 1654 eoffset = bp->b_bcount; 1655 1656 while (offset < eoffset) { 1657 dp = (struct direct *)(bp->b_data + offset); 1658 if (dp->d_reclen <= 0 || (dp->d_reclen & 3) || 1659 offset + dp->d_reclen > bp->b_bcount) { 1660 error = EIO; 1661 break; 1662 } 1663 if (offsetof(struct direct, d_name[dp->d_namlen]) > dp->d_reclen) { 1664 error = EIO; 1665 break; 1666 } 1667 if (offset < pickup) { 1668 offset += dp->d_reclen; 1669 continue; 1670 } 1671 #if BYTE_ORDER == LITTLE_ENDIAN 1672 if (OFSFMT(vp)) { 1673 retval = vop_write_dirent(&error, uio, 1674 dp->d_ino, dp->d_namlen, dp->d_type, 1675 dp->d_name); 1676 } else 1677 #endif 1678 { 1679 retval = vop_write_dirent(&error, uio, 1680 dp->d_ino, dp->d_type, dp->d_namlen, 1681 dp->d_name); 1682 } 1683 if (retval) 1684 break; 1685 if (cookies) 1686 cookies[cookie_index] = bp->b_loffset + offset; 1687 ++cookie_index; 1688 offset += dp->d_reclen; 1689 if (cookie_index == ncookies) 1690 break; 1691 } 1692 1693 /* 1694 * This will align the next loop to the beginning of the 1695 * next block, and pickup will calculate to 0. 1696 */ 1697 uio->uio_offset = bp->b_loffset + offset; 1698 brelse(bp); 1699 1700 if (retval || error || cookie_index == ncookies || 1701 uio->uio_offset >= VTOI(vp)->i_size) { 1702 break; 1703 } 1704 } 1705 if (ap->a_eofflag) 1706 *ap->a_eofflag = VTOI(vp)->i_size <= uio->uio_offset; 1707 1708 /* 1709 * Report errors only if we didn't manage to read anything 1710 */ 1711 if (error && cookie_index == 0) { 1712 if (cookies) { 1713 kfree(cookies, M_TEMP); 1714 *ap->a_ncookies = 0; 1715 *ap->a_cookies = NULL; 1716 } 1717 } else { 1718 error = 0; 1719 if (cookies) { 1720 *ap->a_ncookies = cookie_index; 1721 *ap->a_cookies = cookies; 1722 } 1723 } 1724 done: 1725 vn_unlock(vp); 1726 return (error); 1727 } 1728 1729 /* 1730 * Return target name of a symbolic link 1731 * 1732 * ufs_readlink(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred) 1733 */ 1734 static 1735 int 1736 ufs_readlink(struct vop_readlink_args *ap) 1737 { 1738 struct vnode *vp = ap->a_vp; 1739 struct inode *ip = VTOI(vp); 1740 int isize; 1741 1742 isize = ip->i_size; 1743 if ((isize < vp->v_mount->mnt_maxsymlinklen) || 1744 (ip->i_din.di_blocks == 0)) { /* XXX - for old fastlink support */ 1745 uiomove((char *)ip->i_shortlink, isize, ap->a_uio); 1746 return (0); 1747 } 1748 1749 /* 1750 * Perform the equivalent of an OPEN on vp so we can issue a 1751 * VOP_READ. 1752 */ 1753 return (VOP_READ(vp, ap->a_uio, 0, ap->a_cred)); 1754 } 1755 1756 /* 1757 * Calculate the logical to physical mapping if not done already, 1758 * then call the device strategy routine. 1759 * 1760 * In order to be able to swap to a file, the VOP_BMAP operation may not 1761 * deadlock on memory. See ufs_bmap() for details. 1762 * 1763 * ufs_strategy(struct vnode *a_vp, struct bio *a_bio) 1764 */ 1765 static 1766 int 1767 ufs_strategy(struct vop_strategy_args *ap) 1768 { 1769 struct bio *bio = ap->a_bio; 1770 struct bio *nbio; 1771 struct buf *bp = bio->bio_buf; 1772 struct vnode *vp = ap->a_vp; 1773 struct inode *ip; 1774 int error; 1775 1776 ip = VTOI(vp); 1777 if (vp->v_type == VBLK || vp->v_type == VCHR) 1778 panic("ufs_strategy: spec"); 1779 nbio = push_bio(bio); 1780 if (nbio->bio_offset == NOOFFSET) { 1781 error = VOP_BMAP(vp, bio->bio_offset, &nbio->bio_offset, 1782 NULL, NULL, bp->b_cmd); 1783 if (error) { 1784 bp->b_error = error; 1785 bp->b_flags |= B_ERROR; 1786 /* I/O was never started on nbio, must biodone(bio) */ 1787 biodone(bio); 1788 return (error); 1789 } 1790 if (nbio->bio_offset == NOOFFSET) 1791 vfs_bio_clrbuf(bp); 1792 } 1793 if (nbio->bio_offset == NOOFFSET) { 1794 /* 1795 * We hit a hole in the file. The buffer has been zero-filled 1796 * so just biodone() it. 1797 */ 1798 biodone(bio); 1799 } else { 1800 vn_strategy(ip->i_devvp, nbio); 1801 } 1802 return (0); 1803 } 1804 1805 /* 1806 * Print out the contents of an inode. 1807 * 1808 * ufs_print(struct vnode *a_vp) 1809 */ 1810 static 1811 int 1812 ufs_print(struct vop_print_args *ap) 1813 { 1814 struct vnode *vp = ap->a_vp; 1815 struct inode *ip = VTOI(vp); 1816 1817 kprintf("tag VT_UFS, ino %lu, on dev %s (%d, %d)", 1818 (u_long)ip->i_number, devtoname(ip->i_dev), major(ip->i_dev), 1819 minor(ip->i_dev)); 1820 if (vp->v_type == VFIFO) 1821 fifo_printinfo(vp); 1822 lockmgr_printinfo(&vp->v_lock); 1823 kprintf("\n"); 1824 return (0); 1825 } 1826 1827 /* 1828 * Read wrapper for fifos. 1829 * 1830 * ufsfifo_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 1831 * struct ucred *a_cred) 1832 */ 1833 static 1834 int 1835 ufsfifo_read(struct vop_read_args *ap) 1836 { 1837 int error, resid; 1838 struct inode *ip; 1839 struct uio *uio; 1840 1841 uio = ap->a_uio; 1842 resid = uio->uio_resid; 1843 error = VOCALL(&fifo_vnode_vops, &ap->a_head); 1844 ip = VTOI(ap->a_vp); 1845 if ((ap->a_vp->v_mount->mnt_flag & MNT_NOATIME) == 0 && ip != NULL && 1846 (uio->uio_resid != resid || (error == 0 && resid != 0))) 1847 VTOI(ap->a_vp)->i_flag |= IN_ACCESS; 1848 return (error); 1849 } 1850 1851 /* 1852 * Write wrapper for fifos. 1853 * 1854 * ufsfifo_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 1855 * struct ucred *a_cred) 1856 */ 1857 static 1858 int 1859 ufsfifo_write(struct vop_write_args *ap) 1860 { 1861 int error, resid; 1862 struct inode *ip; 1863 struct uio *uio; 1864 1865 uio = ap->a_uio; 1866 resid = uio->uio_resid; 1867 error = VOCALL(&fifo_vnode_vops, &ap->a_head); 1868 ip = VTOI(ap->a_vp); 1869 if (ip != NULL && (uio->uio_resid != resid || (error == 0 && resid != 0))) 1870 VTOI(ap->a_vp)->i_flag |= IN_CHANGE | IN_UPDATE; 1871 return (error); 1872 } 1873 1874 /* 1875 * Close wrapper for fifos. 1876 * 1877 * Update the times on the inode then do device close. 1878 * 1879 * ufsfifo_close(struct vnode *a_vp, int a_fflag) 1880 */ 1881 static 1882 int 1883 ufsfifo_close(struct vop_close_args *ap) 1884 { 1885 struct vnode *vp = ap->a_vp; 1886 1887 if (VREFCNT(vp) > 1) 1888 ufs_itimes(vp); 1889 return (VOCALL(&fifo_vnode_vops, &ap->a_head)); 1890 } 1891 1892 /* 1893 * Kqfilter wrapper for fifos. 1894 * 1895 * Fall through to ufs kqfilter routines if needed 1896 */ 1897 static 1898 int 1899 ufsfifo_kqfilter(struct vop_kqfilter_args *ap) 1900 { 1901 int error; 1902 1903 error = VOCALL(&fifo_vnode_vops, &ap->a_head); 1904 if (error) 1905 error = ufs_kqfilter(ap); 1906 return (error); 1907 } 1908 1909 /* 1910 * Advisory record locking support 1911 * 1912 * ufs_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, struct flock *a_fl, 1913 * int a_flags) 1914 */ 1915 static 1916 int 1917 ufs_advlock(struct vop_advlock_args *ap) 1918 { 1919 struct inode *ip = VTOI(ap->a_vp); 1920 1921 return (lf_advlock(ap, &(ip->i_lockf), ip->i_size)); 1922 } 1923 1924 /* 1925 * Initialize the vnode associated with a new inode, handle aliased 1926 * vnodes. 1927 * 1928 * Make sure directories have their VM object now rather then later, 1929 * saving us from having to check on all the myrid directory VOPs 1930 * that might be executed without a VOP_OPEN being performed. 1931 */ 1932 int 1933 ufs_vinit(struct mount *mntp, struct vnode **vpp) 1934 { 1935 struct inode *ip; 1936 struct vnode *vp; 1937 struct timeval tv; 1938 1939 vp = *vpp; 1940 ip = VTOI(vp); 1941 1942 vp->v_type = IFTOVT(ip->i_mode); 1943 1944 switch(vp->v_type) { 1945 case VCHR: 1946 case VBLK: 1947 vp->v_ops = &mntp->mnt_vn_spec_ops; 1948 addaliasu(vp, umajor(ip->i_rdev), uminor(ip->i_rdev)); 1949 break; 1950 case VFIFO: 1951 vp->v_ops = &mntp->mnt_vn_fifo_ops; 1952 break; 1953 case VDIR: 1954 case VREG: 1955 vinitvmio(vp, ip->i_size, 1956 blkoffsize(ip->i_fs, ip, ip->i_size), 1957 blkoff(ip->i_fs, ip->i_size)); 1958 break; 1959 case VLNK: 1960 if (ip->i_size >= vp->v_mount->mnt_maxsymlinklen) { 1961 vinitvmio(vp, ip->i_size, 1962 blkoffsize(ip->i_fs, ip, ip->i_size), 1963 blkoff(ip->i_fs, ip->i_size)); 1964 } 1965 break; 1966 default: 1967 break; 1968 1969 } 1970 1971 if (ip->i_number == ROOTINO) 1972 vsetflags(vp, VROOT); 1973 /* 1974 * Initialize modrev times 1975 */ 1976 getmicrouptime(&tv); 1977 SETHIGH(ip->i_modrev, tv.tv_sec); 1978 SETLOW(ip->i_modrev, tv.tv_usec * 4294); 1979 *vpp = vp; 1980 return (0); 1981 } 1982 1983 /* 1984 * Allocate a new inode. 1985 */ 1986 static 1987 int 1988 ufs_makeinode(int mode, struct vnode *dvp, struct vnode **vpp, 1989 struct componentname *cnp) 1990 { 1991 struct inode *ip, *pdir; 1992 struct direct newdir; 1993 struct vnode *tvp; 1994 int error; 1995 1996 pdir = VTOI(dvp); 1997 *vpp = NULL; 1998 if ((mode & IFMT) == 0) 1999 mode |= IFREG; 2000 2001 error = ffs_valloc(dvp, mode, cnp->cn_cred, &tvp); 2002 if (error) 2003 return (error); 2004 ip = VTOI(tvp); 2005 ip->i_flags = pdir->i_flags & (SF_NOHISTORY|UF_NOHISTORY|UF_NODUMP); 2006 ip->i_gid = pdir->i_gid; 2007 #ifdef SUIDDIR 2008 { 2009 #ifdef QUOTA 2010 struct ucred ucred, *ucp; 2011 ucp = cnp->cn_cred; 2012 #endif 2013 /* 2014 * If we are not the owner of the directory, 2015 * and we are hacking owners here, (only do this where told to) 2016 * and we are not giving it TO root, (would subvert quotas) 2017 * then go ahead and give it to the other user. 2018 * Note that this drops off the execute bits for security. 2019 */ 2020 if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) && 2021 (pdir->i_mode & ISUID) && 2022 (pdir->i_uid != cnp->cn_cred->cr_uid) && pdir->i_uid) { 2023 ip->i_uid = pdir->i_uid; 2024 mode &= ~07111; 2025 #ifdef QUOTA 2026 /* 2027 * Make sure the correct user gets charged 2028 * for the space. 2029 * Quickly knock up a dummy credential for the victim. 2030 * XXX This seems to never be accessed out of our 2031 * context so a stack variable is ok. 2032 */ 2033 ucred.cr_ref = 1; 2034 ucred.cr_uid = ip->i_uid; 2035 ucred.cr_ngroups = 1; 2036 ucred.cr_groups[0] = pdir->i_gid; 2037 ucp = &ucred; 2038 #endif 2039 } else 2040 ip->i_uid = cnp->cn_cred->cr_uid; 2041 2042 #ifdef QUOTA 2043 if ((error = ufs_getinoquota(ip)) || 2044 (error = ufs_chkiq(ip, 1, ucp, 0))) { 2045 ffs_vfree(tvp, ip->i_number, mode); 2046 vput(tvp); 2047 return (error); 2048 } 2049 #endif 2050 } 2051 #else /* !SUIDDIR */ 2052 ip->i_uid = cnp->cn_cred->cr_uid; 2053 #ifdef QUOTA 2054 if ((error = ufs_getinoquota(ip)) || 2055 (error = ufs_chkiq(ip, 1, cnp->cn_cred, 0))) { 2056 ffs_vfree(tvp, ip->i_number, mode); 2057 vput(tvp); 2058 return (error); 2059 } 2060 #endif 2061 #endif /* !SUIDDIR */ 2062 ip->i_din.di_spare[0] = 0; 2063 ip->i_din.di_spare[1] = 0; 2064 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 2065 ip->i_mode = mode; 2066 tvp->v_type = IFTOVT(mode); /* Rest init'd in getnewvnode(). */ 2067 ip->i_effnlink = 1; 2068 ip->i_nlink = 1; 2069 if (DOINGSOFTDEP(tvp)) 2070 softdep_change_linkcnt(ip); 2071 if ((ip->i_mode & ISGID) && !groupmember(ip->i_gid, cnp->cn_cred) && 2072 priv_check_cred(cnp->cn_cred, PRIV_VFS_SETGID, 0)) { 2073 ip->i_mode &= ~ISGID; 2074 } 2075 2076 if (cnp->cn_flags & CNP_ISWHITEOUT) 2077 ip->i_flags |= UF_OPAQUE; 2078 2079 /* 2080 * Regular files and directories need VM objects. Softlinks do 2081 * not (not immediately anyway). 2082 */ 2083 if (tvp->v_type == VREG || tvp->v_type == VDIR) 2084 vinitvmio(tvp, 0, PAGE_SIZE, -1); 2085 2086 /* 2087 * Make sure inode goes to disk before directory entry. 2088 */ 2089 error = ffs_update(tvp, !(DOINGSOFTDEP(tvp) | DOINGASYNC(tvp))); 2090 if (error) 2091 goto bad; 2092 ufs_makedirentry(ip, cnp, &newdir); 2093 error = ufs_direnter(dvp, tvp, &newdir, cnp, NULL); 2094 if (error) 2095 goto bad; 2096 *vpp = tvp; 2097 return (0); 2098 2099 bad: 2100 /* 2101 * Write error occurred trying to update the inode 2102 * or the directory so must deallocate the inode. 2103 */ 2104 ip->i_effnlink = 0; 2105 ip->i_nlink = 0; 2106 ip->i_flag |= IN_CHANGE; 2107 if (DOINGSOFTDEP(tvp)) 2108 softdep_change_linkcnt(ip); 2109 vput(tvp); 2110 return (error); 2111 } 2112 2113 static int 2114 ufs_missingop(struct vop_generic_args *ap) 2115 { 2116 panic("no vop function for %s in ufs child", ap->a_desc->sd_name); 2117 return (EOPNOTSUPP); 2118 } 2119 2120 static struct filterops ufsread_filtops = 2121 { FILTEROP_ISFD, NULL, filt_ufsdetach, filt_ufsread }; 2122 static struct filterops ufswrite_filtops = 2123 { FILTEROP_ISFD, NULL, filt_ufsdetach, filt_ufswrite }; 2124 static struct filterops ufsvnode_filtops = 2125 { FILTEROP_ISFD, NULL, filt_ufsdetach, filt_ufsvnode }; 2126 2127 /* 2128 * ufs_kqfilter(struct vnode *a_vp, struct knote *a_kn) 2129 */ 2130 static int 2131 ufs_kqfilter(struct vop_kqfilter_args *ap) 2132 { 2133 struct vnode *vp = ap->a_vp; 2134 struct knote *kn = ap->a_kn; 2135 2136 switch (kn->kn_filter) { 2137 case EVFILT_READ: 2138 kn->kn_fop = &ufsread_filtops; 2139 break; 2140 case EVFILT_WRITE: 2141 kn->kn_fop = &ufswrite_filtops; 2142 break; 2143 case EVFILT_VNODE: 2144 kn->kn_fop = &ufsvnode_filtops; 2145 break; 2146 default: 2147 return (EOPNOTSUPP); 2148 } 2149 2150 kn->kn_hook = (caddr_t)vp; 2151 2152 /* XXX: kq token actually protects the list */ 2153 lwkt_gettoken(&vp->v_token); 2154 knote_insert(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn); 2155 lwkt_reltoken(&vp->v_token); 2156 2157 return (0); 2158 } 2159 2160 static void 2161 filt_ufsdetach(struct knote *kn) 2162 { 2163 struct vnode *vp = (struct vnode *)kn->kn_hook; 2164 2165 lwkt_gettoken(&vp->v_token); 2166 knote_remove(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn); 2167 lwkt_reltoken(&vp->v_token); 2168 } 2169 2170 /*ARGSUSED*/ 2171 static int 2172 filt_ufsread(struct knote *kn, long hint) 2173 { 2174 struct vnode *vp = (struct vnode *)kn->kn_hook; 2175 struct inode *ip = VTOI(vp); 2176 off_t off; 2177 2178 /* 2179 * filesystem is gone, so set the EOF flag and schedule 2180 * the knote for deletion. 2181 */ 2182 if (hint == NOTE_REVOKE) { 2183 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT); 2184 return (1); 2185 } 2186 2187 off = ip->i_size - kn->kn_fp->f_offset; 2188 kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX; 2189 if (kn->kn_sfflags & NOTE_OLDAPI) 2190 return(1); 2191 return (kn->kn_data != 0); 2192 } 2193 2194 /*ARGSUSED*/ 2195 static int 2196 filt_ufswrite(struct knote *kn, long hint) 2197 { 2198 /* 2199 * filesystem is gone, so set the EOF flag and schedule 2200 * the knote for deletion. 2201 */ 2202 if (hint == NOTE_REVOKE) 2203 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT); 2204 2205 kn->kn_data = 0; 2206 return (1); 2207 } 2208 2209 static int 2210 filt_ufsvnode(struct knote *kn, long hint) 2211 { 2212 if (kn->kn_sfflags & hint) 2213 kn->kn_fflags |= hint; 2214 if (hint == NOTE_REVOKE) { 2215 kn->kn_flags |= (EV_EOF | EV_NODATA); 2216 return (1); 2217 } 2218 return (kn->kn_fflags != 0); 2219 } 2220 2221 /* Global vfs data structures for ufs. */ 2222 static struct vop_ops ufs_vnode_vops = { 2223 .vop_default = vop_defaultop, 2224 .vop_fsync = (void *)ufs_missingop, 2225 .vop_read = (void *)ufs_missingop, 2226 .vop_reallocblks = (void *)ufs_missingop, 2227 .vop_write = (void *)ufs_missingop, 2228 .vop_access = ufs_access, 2229 .vop_advlock = ufs_advlock, 2230 .vop_bmap = ufs_bmap, 2231 .vop_old_lookup = ufs_lookup, 2232 .vop_close = ufs_close, 2233 .vop_old_create = ufs_create, 2234 .vop_getattr = ufs_getattr, 2235 .vop_inactive = ufs_inactive, 2236 .vop_old_link = ufs_link, 2237 .vop_old_mkdir = ufs_mkdir, 2238 .vop_old_mknod = ufs_mknod, 2239 .vop_mmap = ufs_mmap, 2240 .vop_open = vop_stdopen, 2241 .vop_pathconf = vop_stdpathconf, 2242 .vop_kqfilter = ufs_kqfilter, 2243 .vop_print = ufs_print, 2244 .vop_readdir = ufs_readdir, 2245 .vop_readlink = ufs_readlink, 2246 .vop_reclaim = ufs_reclaim, 2247 .vop_old_remove = ufs_remove, 2248 .vop_old_rename = ufs_rename, 2249 .vop_old_rmdir = ufs_rmdir, 2250 .vop_setattr = ufs_setattr, 2251 .vop_markatime = ufs_markatime, 2252 .vop_strategy = ufs_strategy, 2253 .vop_old_symlink = ufs_symlink, 2254 .vop_old_whiteout = ufs_whiteout 2255 }; 2256 2257 static struct vop_ops ufs_spec_vops = { 2258 .vop_default = vop_defaultop, 2259 .vop_fsync = (void *)ufs_missingop, 2260 .vop_access = ufs_access, 2261 .vop_close = ufs_close, 2262 .vop_getattr = ufs_getattr, 2263 .vop_inactive = ufs_inactive, 2264 .vop_print = ufs_print, 2265 .vop_read = vop_stdnoread, 2266 .vop_reclaim = ufs_reclaim, 2267 .vop_setattr = ufs_setattr, 2268 .vop_markatime = ufs_markatime, 2269 .vop_write = vop_stdnowrite 2270 }; 2271 2272 static struct vop_ops ufs_fifo_vops = { 2273 .vop_default = fifo_vnoperate, 2274 .vop_fsync = (void *)ufs_missingop, 2275 .vop_access = ufs_access, 2276 .vop_close = ufsfifo_close, 2277 .vop_getattr = ufs_getattr, 2278 .vop_inactive = ufs_inactive, 2279 .vop_kqfilter = ufsfifo_kqfilter, 2280 .vop_print = ufs_print, 2281 .vop_read = ufsfifo_read, 2282 .vop_reclaim = ufs_reclaim, 2283 .vop_setattr = ufs_setattr, 2284 .vop_markatime = ufs_markatime, 2285 .vop_write = ufsfifo_write 2286 }; 2287 2288 VNODEOP_SET(ufs_vnode_vops); 2289 VNODEOP_SET(ufs_spec_vops); 2290 VNODEOP_SET(ufs_fifo_vops); 2291 2292 /* 2293 * ufs_vnoperate() 2294 */ 2295 int 2296 ufs_vnoperate(struct vop_generic_args *ap) 2297 { 2298 return (VOCALL(&ufs_vnode_vops, ap)); 2299 } 2300 2301 /* 2302 * ufs_vnoperatefifo() 2303 */ 2304 int 2305 ufs_vnoperatefifo(struct vop_generic_args *ap) 2306 { 2307 return (VOCALL(&ufs_fifo_vops, ap)); 2308 } 2309 2310 /* 2311 * ufs_vnoperatespec() 2312 */ 2313 int 2314 ufs_vnoperatespec(struct vop_generic_args *ap) 2315 { 2316 return (VOCALL(&ufs_spec_vops, ap)); 2317 } 2318