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. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the University of 21 * California, Berkeley and its contributors. 22 * 4. Neither the name of the University nor the names of its contributors 23 * may be used to endorse or promote products derived from this software 24 * without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 36 * SUCH DAMAGE. 37 * 38 * @(#)ufs_vnops.c 8.27 (Berkeley) 5/27/95 39 * $FreeBSD: src/sys/ufs/ufs/ufs_vnops.c,v 1.131.2.8 2003/01/02 17:26:19 bde Exp $ 40 */ 41 42 #include "opt_quota.h" 43 #include "opt_suiddir.h" 44 #include "opt_ufs.h" 45 46 #include <sys/param.h> 47 #include <sys/systm.h> 48 #include <sys/kernel.h> 49 #include <sys/fcntl.h> 50 #include <sys/stat.h> 51 #include <sys/buf.h> 52 #include <sys/proc.h> 53 #include <sys/priv.h> 54 #include <sys/namei.h> 55 #include <sys/mount.h> 56 #include <sys/unistd.h> 57 #include <sys/vnode.h> 58 #include <sys/malloc.h> 59 #include <sys/dirent.h> 60 #include <sys/lockf.h> 61 #include <sys/event.h> 62 #include <sys/conf.h> 63 64 #include <sys/file.h> /* XXX */ 65 #include <sys/jail.h> 66 67 #include <vm/vm.h> 68 #include <vm/vm_extern.h> 69 70 #include <vfs/fifofs/fifo.h> 71 72 #include "quota.h" 73 #include "inode.h" 74 #include "dir.h" 75 #include "ufsmount.h" 76 #include "ufs_extern.h" 77 #include "ffs_extern.h" 78 #include "fs.h" 79 #ifdef UFS_DIRHASH 80 #include "dirhash.h" 81 #endif 82 83 static int ufs_access (struct vop_access_args *); 84 static int ufs_advlock (struct vop_advlock_args *); 85 static int ufs_chmod (struct vnode *, int, struct ucred *); 86 static int ufs_chown (struct vnode *, uid_t, gid_t, struct ucred *); 87 static int ufs_close (struct vop_close_args *); 88 static int ufs_create (struct vop_old_create_args *); 89 static int ufs_getattr (struct vop_getattr_args *); 90 static int ufs_link (struct vop_old_link_args *); 91 static int ufs_makeinode (int mode, struct vnode *, struct vnode **, struct componentname *); 92 static int ufs_markatime (struct vop_markatime_args *); 93 static int ufs_missingop (struct vop_generic_args *ap); 94 static int ufs_mkdir (struct vop_old_mkdir_args *); 95 static int ufs_mknod (struct vop_old_mknod_args *); 96 static int ufs_mmap (struct vop_mmap_args *); 97 static int ufs_print (struct vop_print_args *); 98 static int ufs_readdir (struct vop_readdir_args *); 99 static int ufs_readlink (struct vop_readlink_args *); 100 static int ufs_remove (struct vop_old_remove_args *); 101 static int ufs_rename (struct vop_old_rename_args *); 102 static int ufs_rmdir (struct vop_old_rmdir_args *); 103 static int ufs_setattr (struct vop_setattr_args *); 104 static int ufs_strategy (struct vop_strategy_args *); 105 static int ufs_symlink (struct vop_old_symlink_args *); 106 static int ufs_whiteout (struct vop_old_whiteout_args *); 107 static int ufsfifo_close (struct vop_close_args *); 108 static int ufsfifo_kqfilter (struct vop_kqfilter_args *); 109 static int ufsfifo_read (struct vop_read_args *); 110 static int ufsfifo_write (struct vop_write_args *); 111 static int filt_ufsread (struct knote *kn, long hint); 112 static int filt_ufswrite (struct knote *kn, long hint); 113 static int filt_ufsvnode (struct knote *kn, long hint); 114 static void filt_ufsdetach (struct knote *kn); 115 static int ufs_kqfilter (struct vop_kqfilter_args *ap); 116 117 union _qcvt { 118 int64_t qcvt; 119 int32_t val[2]; 120 }; 121 #define SETHIGH(q, h) { \ 122 union _qcvt tmp; \ 123 tmp.qcvt = (q); \ 124 tmp.val[_QUAD_HIGHWORD] = (h); \ 125 (q) = tmp.qcvt; \ 126 } 127 #define SETLOW(q, l) { \ 128 union _qcvt tmp; \ 129 tmp.qcvt = (q); \ 130 tmp.val[_QUAD_LOWWORD] = (l); \ 131 (q) = tmp.qcvt; \ 132 } 133 #define VN_KNOTE(vp, b) \ 134 KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, (b)) 135 136 #define OFSFMT(vp) ((vp)->v_mount->mnt_maxsymlinklen <= 0) 137 138 /* 139 * A virgin directory (no blushing please). 140 */ 141 static struct dirtemplate mastertemplate = { 142 0, 12, DT_DIR, 1, ".", 143 0, DIRBLKSIZ - 12, DT_DIR, 2, ".." 144 }; 145 static struct odirtemplate omastertemplate = { 146 0, 12, 1, ".", 147 0, DIRBLKSIZ - 12, 2, ".." 148 }; 149 150 void 151 ufs_itimes(struct vnode *vp) 152 { 153 struct inode *ip; 154 struct timespec ts; 155 156 ip = VTOI(vp); 157 if ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE)) == 0) 158 return; 159 if ((vp->v_type == VBLK || vp->v_type == VCHR) && !DOINGSOFTDEP(vp)) 160 ip->i_flag |= IN_LAZYMOD; 161 else 162 ip->i_flag |= IN_MODIFIED; 163 if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 164 vfs_timestamp(&ts); 165 if (ip->i_flag & IN_ACCESS) { 166 ip->i_atime = ts.tv_sec; 167 ip->i_atimensec = ts.tv_nsec; 168 } 169 if (ip->i_flag & IN_UPDATE) { 170 ip->i_mtime = ts.tv_sec; 171 ip->i_mtimensec = ts.tv_nsec; 172 ip->i_modrev++; 173 } 174 if (ip->i_flag & IN_CHANGE) { 175 ip->i_ctime = ts.tv_sec; 176 ip->i_ctimensec = ts.tv_nsec; 177 } 178 } 179 ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_UPDATE); 180 } 181 182 /* 183 * Create a regular file 184 * 185 * ufs_create(struct vnode *a_dvp, struct vnode **a_vpp, 186 * struct componentname *a_cnp, struct vattr *a_vap) 187 */ 188 static 189 int 190 ufs_create(struct vop_old_create_args *ap) 191 { 192 int error; 193 194 error = 195 ufs_makeinode(MAKEIMODE(ap->a_vap->va_type, ap->a_vap->va_mode), 196 ap->a_dvp, ap->a_vpp, ap->a_cnp); 197 if (error) 198 return (error); 199 VN_KNOTE(ap->a_dvp, NOTE_WRITE); 200 return (0); 201 } 202 203 /* 204 * Mknod vnode call 205 * 206 * ufs_mknod(struct vnode *a_dvp, struct vnode **a_vpp, 207 * struct componentname *a_cnp, struct vattr *a_vap) 208 */ 209 /* ARGSUSED */ 210 static 211 int 212 ufs_mknod(struct vop_old_mknod_args *ap) 213 { 214 struct vattr *vap = ap->a_vap; 215 struct vnode **vpp = ap->a_vpp; 216 struct inode *ip; 217 ino_t ino; 218 int error; 219 220 /* 221 * UFS cannot represent the entire major/minor range supported by 222 * the kernel. 223 */ 224 if (vap->va_rmajor != VNOVAL && 225 makeudev(vap->va_rmajor, vap->va_rminor) == NOUDEV) { 226 return(EINVAL); 227 } 228 229 /* no special directory support */ 230 if (vap->va_type == VDIR) 231 return(EINVAL); 232 233 error = ufs_makeinode(MAKEIMODE(vap->va_type, vap->va_mode), 234 ap->a_dvp, vpp, ap->a_cnp); 235 if (error) 236 return (error); 237 VN_KNOTE(ap->a_dvp, NOTE_WRITE); 238 ip = VTOI(*vpp); 239 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 240 if (vap->va_rmajor != VNOVAL) { 241 /* 242 * Want to be able to use this to make badblock 243 * inodes, so don't truncate the dev number. 244 */ 245 ip->i_rdev = makeudev(vap->va_rmajor, vap->va_rminor); 246 } 247 /* 248 * Remove inode, then reload it through VFS_VGET so it is 249 * checked to see if it is an alias of an existing entry in 250 * the inode cache. 251 */ 252 (*vpp)->v_type = VNON; 253 ino = ip->i_number; /* Save this before vgone() invalidates ip. */ 254 vgone_vxlocked(*vpp); 255 vput(*vpp); 256 error = VFS_VGET(ap->a_dvp->v_mount, NULL, ino, vpp); 257 if (error) { 258 *vpp = NULL; 259 return (error); 260 } 261 return (0); 262 } 263 264 /* 265 * Close called. 266 * 267 * Update the times on the inode. 268 * 269 * ufs_close(struct vnode *a_vp, int a_fflag) 270 */ 271 /* ARGSUSED */ 272 static 273 int 274 ufs_close(struct vop_close_args *ap) 275 { 276 struct vnode *vp = ap->a_vp; 277 278 if (vp->v_sysref.refcnt > 1) 279 ufs_itimes(vp); 280 return (vop_stdclose(ap)); 281 } 282 283 /* 284 * ufs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred) 285 */ 286 static 287 int 288 ufs_access(struct vop_access_args *ap) 289 { 290 struct vnode *vp = ap->a_vp; 291 struct inode *ip = VTOI(vp); 292 int error; 293 294 #ifdef QUOTA 295 if (ap->a_mode & VWRITE) { 296 switch (vp->v_type) { 297 case VDIR: 298 case VLNK: 299 case VREG: 300 if ((error = ufs_getinoquota(ip)) != 0) 301 return (error); 302 break; 303 default: 304 break; 305 } 306 } 307 #endif 308 309 error = vop_helper_access(ap, ip->i_uid, ip->i_gid, ip->i_mode, 0); 310 return (error); 311 } 312 313 /* 314 * ufs_getattr(struct vnode *a_vp, struct vattr *a_vap) 315 */ 316 /* ARGSUSED */ 317 static 318 int 319 ufs_getattr(struct vop_getattr_args *ap) 320 { 321 struct vnode *vp = ap->a_vp; 322 struct inode *ip = VTOI(vp); 323 struct vattr *vap = ap->a_vap; 324 325 ufs_itimes(vp); 326 /* 327 * Copy from inode table 328 */ 329 vap->va_fsid = dev2udev(ip->i_dev); 330 vap->va_fileid = ip->i_number; 331 vap->va_mode = ip->i_mode & ~IFMT; 332 vap->va_nlink = VFSTOUFS(vp->v_mount)->um_i_effnlink_valid ? 333 ip->i_effnlink : ip->i_nlink; 334 vap->va_uid = ip->i_uid; 335 vap->va_gid = ip->i_gid; 336 vap->va_rmajor = umajor(ip->i_rdev); 337 vap->va_rminor = uminor(ip->i_rdev); 338 vap->va_size = ip->i_din.di_size; 339 vap->va_atime.tv_sec = ip->i_atime; 340 vap->va_atime.tv_nsec = ip->i_atimensec; 341 vap->va_mtime.tv_sec = ip->i_mtime; 342 vap->va_mtime.tv_nsec = ip->i_mtimensec; 343 vap->va_ctime.tv_sec = ip->i_ctime; 344 vap->va_ctime.tv_nsec = ip->i_ctimensec; 345 vap->va_flags = ip->i_flags; 346 vap->va_gen = ip->i_gen; 347 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize; 348 vap->va_bytes = dbtob((u_quad_t)ip->i_blocks); 349 vap->va_type = IFTOVT(ip->i_mode); 350 vap->va_filerev = ip->i_modrev; 351 return (0); 352 } 353 354 static 355 int 356 ufs_markatime(struct vop_markatime_args *ap) 357 { 358 struct vnode *vp = ap->a_vp; 359 struct inode *ip = VTOI(vp); 360 361 if (vp->v_mount->mnt_flag & MNT_RDONLY) 362 return (EROFS); 363 if (vp->v_mount->mnt_flag & MNT_NOATIME) 364 return (0); 365 ip->i_flag |= IN_ACCESS; 366 VN_KNOTE(vp, NOTE_ATTRIB); 367 return (0); 368 } 369 370 /* 371 * Set attribute vnode op. called from several syscalls 372 * 373 * ufs_setattr(struct vnode *a_vp, struct vattr *a_vap, 374 * struct ucred *a_cred) 375 */ 376 static 377 int 378 ufs_setattr(struct vop_setattr_args *ap) 379 { 380 struct vattr *vap = ap->a_vap; 381 struct vnode *vp = ap->a_vp; 382 struct inode *ip = VTOI(vp); 383 struct ucred *cred = ap->a_cred; 384 int error; 385 386 /* 387 * Check for unsettable attributes. 388 */ 389 if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) || 390 (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) || 391 (vap->va_blocksize != VNOVAL) || (vap->va_rmajor != VNOVAL) || 392 ((int)vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) { 393 return (EINVAL); 394 } 395 if (vap->va_flags != VNOVAL) { 396 if (vp->v_mount->mnt_flag & MNT_RDONLY) 397 return (EROFS); 398 if (cred->cr_uid != ip->i_uid && 399 (error = priv_check_cred(cred, PRIV_VFS_SETATTR, 0))) 400 return (error); 401 /* 402 * Note that a root chflags becomes a user chflags when 403 * we are jailed, unless the jail.chflags_allowed sysctl 404 * is set. 405 */ 406 if (cred->cr_uid == 0 && 407 (!jailed(cred) || jail_chflags_allowed)) { 408 if ((ip->i_flags 409 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) && 410 securelevel > 0) 411 return (EPERM); 412 ip->i_flags = vap->va_flags; 413 } else { 414 if (ip->i_flags 415 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) || 416 (vap->va_flags & UF_SETTABLE) != vap->va_flags) 417 return (EPERM); 418 ip->i_flags &= SF_SETTABLE; 419 ip->i_flags |= (vap->va_flags & UF_SETTABLE); 420 } 421 ip->i_flag |= IN_CHANGE; 422 if (vap->va_flags & (IMMUTABLE | APPEND)) 423 return (0); 424 } 425 if (ip->i_flags & (IMMUTABLE | APPEND)) 426 return (EPERM); 427 /* 428 * Go through the fields and update iff not VNOVAL. 429 */ 430 if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) { 431 if (vp->v_mount->mnt_flag & MNT_RDONLY) 432 return (EROFS); 433 if ((error = ufs_chown(vp, vap->va_uid, vap->va_gid, cred)) != 0) 434 return (error); 435 } 436 if (vap->va_size != VNOVAL) { 437 /* 438 * Disallow write attempts on read-only filesystems; 439 * unless the file is a socket, fifo, or a block or 440 * character device resident on the filesystem. 441 */ 442 switch (vp->v_type) { 443 case VDIR: 444 return (EISDIR); 445 case VLNK: 446 case VREG: 447 if (vp->v_mount->mnt_flag & MNT_RDONLY) 448 return (EROFS); 449 break; 450 default: 451 break; 452 } 453 if ((error = ffs_truncate(vp, vap->va_size, 0, cred)) != 0) 454 return (error); 455 } 456 ip = VTOI(vp); 457 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) { 458 if (vp->v_mount->mnt_flag & MNT_RDONLY) 459 return (EROFS); 460 if (cred->cr_uid != ip->i_uid && 461 (error = priv_check_cred(cred, PRIV_VFS_SETATTR, 0)) && 462 ((vap->va_vaflags & VA_UTIMES_NULL) == 0 || 463 (error = VOP_EACCESS(vp, VWRITE, cred)))) 464 return (error); 465 if (vap->va_atime.tv_sec != VNOVAL) 466 ip->i_flag |= IN_ACCESS; 467 if (vap->va_mtime.tv_sec != VNOVAL) 468 ip->i_flag |= IN_CHANGE | IN_UPDATE; 469 ufs_itimes(vp); 470 if (vap->va_atime.tv_sec != VNOVAL) { 471 ip->i_atime = vap->va_atime.tv_sec; 472 ip->i_atimensec = vap->va_atime.tv_nsec; 473 } 474 if (vap->va_mtime.tv_sec != VNOVAL) { 475 ip->i_mtime = vap->va_mtime.tv_sec; 476 ip->i_mtimensec = vap->va_mtime.tv_nsec; 477 } 478 error = ffs_update(vp, 0); 479 if (error) 480 return (error); 481 } 482 error = 0; 483 if (vap->va_mode != (mode_t)VNOVAL) { 484 if (vp->v_mount->mnt_flag & MNT_RDONLY) 485 return (EROFS); 486 error = ufs_chmod(vp, (int)vap->va_mode, cred); 487 } 488 VN_KNOTE(vp, NOTE_ATTRIB); 489 return (error); 490 } 491 492 /* 493 * Change the mode on a file. 494 * Inode must be locked before calling. 495 */ 496 static int 497 ufs_chmod(struct vnode *vp, int mode, struct ucred *cred) 498 { 499 struct inode *ip = VTOI(vp); 500 int error; 501 mode_t cur_mode = ip->i_mode; 502 503 error = vop_helper_chmod(vp, mode, cred, ip->i_uid, ip->i_gid, 504 &cur_mode); 505 if (error) 506 return (error); 507 #if 0 508 if (cred->cr_uid != ip->i_uid) { 509 error = priv_check_cred(cred, PRIV_VFS_CHMOD, 0); 510 if (error) 511 return (error); 512 } 513 if (cred->cr_uid) { 514 if (vp->v_type != VDIR && (mode & S_ISTXT)) 515 return (EFTYPE); 516 if (!groupmember(ip->i_gid, cred) && (mode & ISGID)) 517 return (EPERM); 518 } 519 #endif 520 ip->i_mode = cur_mode; 521 ip->i_flag |= IN_CHANGE; 522 return (0); 523 } 524 525 /* 526 * Perform chown operation on inode ip; 527 * inode must be locked prior to call. 528 */ 529 static int 530 ufs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred) 531 { 532 struct inode *ip = VTOI(vp); 533 uid_t ouid; 534 gid_t ogid; 535 int error = 0; 536 #ifdef QUOTA 537 int i; 538 long change; 539 #endif 540 541 if (uid == (uid_t)VNOVAL) 542 uid = ip->i_uid; 543 if (gid == (gid_t)VNOVAL) 544 gid = ip->i_gid; 545 /* 546 * If we don't own the file, are trying to change the owner 547 * of the file, or are not a member of the target group, 548 * the caller must be superuser or the call fails. 549 */ 550 if ((cred->cr_uid != ip->i_uid || uid != ip->i_uid || 551 (gid != ip->i_gid && !(cred->cr_gid == gid || 552 groupmember(gid, cred)))) && 553 (error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0))) 554 return (error); 555 ogid = ip->i_gid; 556 ouid = ip->i_uid; 557 #ifdef QUOTA 558 if ((error = ufs_getinoquota(ip)) != 0) 559 return (error); 560 if (ouid == uid) { 561 ufs_dqrele(vp, ip->i_dquot[USRQUOTA]); 562 ip->i_dquot[USRQUOTA] = NODQUOT; 563 } 564 if (ogid == gid) { 565 ufs_dqrele(vp, ip->i_dquot[GRPQUOTA]); 566 ip->i_dquot[GRPQUOTA] = NODQUOT; 567 } 568 change = ip->i_blocks; 569 (void) ufs_chkdq(ip, -change, cred, CHOWN); 570 (void) ufs_chkiq(ip, -1, cred, CHOWN); 571 for (i = 0; i < MAXQUOTAS; i++) { 572 ufs_dqrele(vp, ip->i_dquot[i]); 573 ip->i_dquot[i] = NODQUOT; 574 } 575 #endif 576 ip->i_gid = gid; 577 ip->i_uid = uid; 578 #ifdef QUOTA 579 if ((error = ufs_getinoquota(ip)) == 0) { 580 if (ouid == uid) { 581 ufs_dqrele(vp, ip->i_dquot[USRQUOTA]); 582 ip->i_dquot[USRQUOTA] = NODQUOT; 583 } 584 if (ogid == gid) { 585 ufs_dqrele(vp, ip->i_dquot[GRPQUOTA]); 586 ip->i_dquot[GRPQUOTA] = NODQUOT; 587 } 588 if ((error = ufs_chkdq(ip, change, cred, CHOWN)) == 0) { 589 if ((error = ufs_chkiq(ip, 1, cred, CHOWN)) == 0) 590 goto good; 591 else 592 (void)ufs_chkdq(ip, -change, cred, CHOWN|FORCE); 593 } 594 for (i = 0; i < MAXQUOTAS; i++) { 595 ufs_dqrele(vp, ip->i_dquot[i]); 596 ip->i_dquot[i] = NODQUOT; 597 } 598 } 599 ip->i_gid = ogid; 600 ip->i_uid = ouid; 601 if (ufs_getinoquota(ip) == 0) { 602 if (ouid == uid) { 603 ufs_dqrele(vp, ip->i_dquot[USRQUOTA]); 604 ip->i_dquot[USRQUOTA] = NODQUOT; 605 } 606 if (ogid == gid) { 607 ufs_dqrele(vp, ip->i_dquot[GRPQUOTA]); 608 ip->i_dquot[GRPQUOTA] = NODQUOT; 609 } 610 (void) ufs_chkdq(ip, change, cred, FORCE|CHOWN); 611 (void) ufs_chkiq(ip, 1, cred, FORCE|CHOWN); 612 (void) ufs_getinoquota(ip); 613 } 614 return (error); 615 good: 616 if (ufs_getinoquota(ip)) 617 panic("ufs_chown: lost quota"); 618 #endif /* QUOTA */ 619 ip->i_flag |= IN_CHANGE; 620 if (cred->cr_uid != 0 && (ouid != uid || ogid != gid)) 621 ip->i_mode &= ~(ISUID | ISGID); 622 return (0); 623 } 624 625 /* 626 * Mmap a file 627 * 628 * NB Currently unsupported. 629 * 630 * ufs_mmap(struct vnode *a_vp, int a_fflags, struct ucred *a_cred) 631 */ 632 /* ARGSUSED */ 633 static 634 int 635 ufs_mmap(struct vop_mmap_args *ap) 636 { 637 return (EINVAL); 638 } 639 640 /* 641 * ufs_remove(struct vnode *a_dvp, struct vnode *a_vp, 642 * struct componentname *a_cnp) 643 */ 644 static 645 int 646 ufs_remove(struct vop_old_remove_args *ap) 647 { 648 struct inode *ip; 649 struct vnode *vp = ap->a_vp; 650 struct vnode *dvp = ap->a_dvp; 651 int error; 652 653 ip = VTOI(vp); 654 #if 0 /* handled by kernel now */ 655 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) || 656 (VTOI(dvp)->i_flags & APPEND)) { 657 error = EPERM; 658 goto out; 659 } 660 #endif 661 error = ufs_dirremove(dvp, ip, ap->a_cnp->cn_flags, 0); 662 VN_KNOTE(vp, NOTE_DELETE); 663 VN_KNOTE(dvp, NOTE_WRITE); 664 #if 0 665 out: 666 #endif 667 return (error); 668 } 669 670 /* 671 * link vnode call 672 * 673 * ufs_link(struct vnode *a_tdvp, struct vnode *a_vp, 674 * struct componentname *a_cnp) 675 */ 676 static 677 int 678 ufs_link(struct vop_old_link_args *ap) 679 { 680 struct vnode *vp = ap->a_vp; 681 struct vnode *tdvp = ap->a_tdvp; 682 struct componentname *cnp = ap->a_cnp; 683 struct inode *ip; 684 struct direct newdir; 685 int error; 686 687 if (tdvp->v_mount != vp->v_mount) { 688 error = EXDEV; 689 goto out2; 690 } 691 if (tdvp != vp && (error = vn_lock(vp, LK_EXCLUSIVE))) { 692 goto out2; 693 } 694 ip = VTOI(vp); 695 if ((nlink_t)ip->i_nlink >= LINK_MAX) { 696 error = EMLINK; 697 goto out1; 698 } 699 #if 0 /* handled by kernel now, also DragonFly allows this */ 700 if (ip->i_flags & (IMMUTABLE | APPEND)) { 701 error = EPERM; 702 goto out1; 703 } 704 #endif 705 ip->i_effnlink++; 706 ip->i_nlink++; 707 ip->i_flag |= IN_CHANGE; 708 if (DOINGSOFTDEP(vp)) 709 softdep_change_linkcnt(ip); 710 error = ffs_update(vp, !(DOINGSOFTDEP(vp) | DOINGASYNC(vp))); 711 if (!error) { 712 ufs_makedirentry(ip, cnp, &newdir); 713 error = ufs_direnter(tdvp, vp, &newdir, cnp, NULL); 714 } 715 716 if (error) { 717 ip->i_effnlink--; 718 ip->i_nlink--; 719 ip->i_flag |= IN_CHANGE; 720 if (DOINGSOFTDEP(vp)) 721 softdep_change_linkcnt(ip); 722 } 723 out1: 724 if (tdvp != vp) 725 vn_unlock(vp); 726 out2: 727 VN_KNOTE(vp, NOTE_LINK); 728 VN_KNOTE(tdvp, NOTE_WRITE); 729 return (error); 730 } 731 732 /* 733 * whiteout vnode call 734 * 735 * ufs_whiteout(struct vnode *a_dvp, struct componentname *a_cnp, int a_flags) 736 */ 737 static 738 int 739 ufs_whiteout(struct vop_old_whiteout_args *ap) 740 { 741 struct vnode *dvp = ap->a_dvp; 742 struct componentname *cnp = ap->a_cnp; 743 struct direct newdir; 744 int error = 0; 745 746 switch (ap->a_flags) { 747 case NAMEI_LOOKUP: 748 /* 4.4 format directories support whiteout operations */ 749 if (dvp->v_mount->mnt_maxsymlinklen > 0) 750 return (0); 751 return (EOPNOTSUPP); 752 753 case NAMEI_CREATE: 754 /* create a new directory whiteout */ 755 #ifdef DIAGNOSTIC 756 if (dvp->v_mount->mnt_maxsymlinklen <= 0) 757 panic("ufs_whiteout: old format filesystem"); 758 #endif 759 760 newdir.d_ino = WINO; 761 newdir.d_namlen = cnp->cn_namelen; 762 bcopy(cnp->cn_nameptr, newdir.d_name, (unsigned)cnp->cn_namelen + 1); 763 newdir.d_type = DT_WHT; 764 error = ufs_direnter(dvp, NULL, &newdir, cnp, NULL); 765 break; 766 767 case NAMEI_DELETE: 768 /* remove an existing directory whiteout */ 769 #ifdef DIAGNOSTIC 770 if (dvp->v_mount->mnt_maxsymlinklen <= 0) 771 panic("ufs_whiteout: old format filesystem"); 772 #endif 773 774 cnp->cn_flags &= ~CNP_DOWHITEOUT; 775 error = ufs_dirremove(dvp, NULL, cnp->cn_flags, 0); 776 break; 777 default: 778 panic("ufs_whiteout: unknown op"); 779 } 780 return (error); 781 } 782 783 /* 784 * Rename system call. 785 * rename("foo", "bar"); 786 * is essentially 787 * unlink("bar"); 788 * link("foo", "bar"); 789 * unlink("foo"); 790 * but ``atomically''. Can't do full commit without saving state in the 791 * inode on disk which isn't feasible at this time. Best we can do is 792 * always guarantee the target exists. 793 * 794 * Basic algorithm is: 795 * 796 * 1) Bump link count on source while we're linking it to the 797 * target. This also ensure the inode won't be deleted out 798 * from underneath us while we work (it may be truncated by 799 * a concurrent `trunc' or `open' for creation). 800 * 2) Link source to destination. If destination already exists, 801 * delete it first. 802 * 3) Unlink source reference to inode if still around. If a 803 * directory was moved and the parent of the destination 804 * is different from the source, patch the ".." entry in the 805 * directory. 806 * 807 * ufs_rename(struct vnode *a_fdvp, struct vnode *a_fvp, 808 * struct componentname *a_fcnp, struct vnode *a_tdvp, 809 * struct vnode *a_tvp, struct componentname *a_tcnp) 810 */ 811 static 812 int 813 ufs_rename(struct vop_old_rename_args *ap) 814 { 815 struct vnode *tvp = ap->a_tvp; 816 struct vnode *tdvp = ap->a_tdvp; 817 struct vnode *fvp = ap->a_fvp; 818 struct vnode *fdvp = ap->a_fdvp; 819 struct componentname *tcnp = ap->a_tcnp; 820 struct componentname *fcnp = ap->a_fcnp; 821 struct inode *ip, *xp, *dp; 822 struct direct newdir; 823 ino_t oldparent = 0, newparent = 0; 824 int doingdirectory = 0; 825 int error = 0, ioflag; 826 827 /* 828 * Check for cross-device rename. 829 */ 830 if ((fvp->v_mount != tdvp->v_mount) || 831 (tvp && (fvp->v_mount != tvp->v_mount))) { 832 error = EXDEV; 833 abortit: 834 if (tdvp == tvp) 835 vrele(tdvp); 836 else 837 vput(tdvp); 838 if (tvp) 839 vput(tvp); 840 vrele(fdvp); 841 vrele(fvp); 842 return (error); 843 } 844 845 #if 0 /* handled by kernel now */ 846 if (tvp && ((VTOI(tvp)->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) || 847 (VTOI(tdvp)->i_flags & APPEND))) { 848 error = EPERM; 849 goto abortit; 850 } 851 #endif 852 853 /* 854 * Renaming a file to itself has no effect. The upper layers should 855 * not call us in that case. Temporarily just warn if they do. 856 */ 857 if (fvp == tvp) { 858 kprintf("ufs_rename: fvp == tvp (can't happen)\n"); 859 error = 0; 860 goto abortit; 861 } 862 863 if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0) 864 goto abortit; 865 866 /* 867 * Note: now that fvp is locked we have to be sure to unlock it before 868 * using the 'abortit' target. 869 */ 870 dp = VTOI(fdvp); 871 ip = VTOI(fvp); 872 if (ip->i_nlink >= LINK_MAX) { 873 vn_unlock(fvp); 874 error = EMLINK; 875 goto abortit; 876 } 877 #if 0 /* handled by kernel now */ 878 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) 879 || (dp->i_flags & APPEND)) { 880 vn_unlock(fvp); 881 error = EPERM; 882 goto abortit; 883 } 884 #endif 885 if ((ip->i_mode & IFMT) == IFDIR) { 886 /* 887 * Avoid ".", "..", and aliases of "." for obvious reasons. 888 */ 889 if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') || 890 dp == ip || (fcnp->cn_flags | tcnp->cn_flags) & CNP_ISDOTDOT || 891 (ip->i_flag & IN_RENAME)) { 892 vn_unlock(fvp); 893 error = EINVAL; 894 goto abortit; 895 } 896 ip->i_flag |= IN_RENAME; 897 oldparent = dp->i_number; 898 doingdirectory = 1; 899 } 900 VN_KNOTE(fdvp, NOTE_WRITE); /* XXX right place? */ 901 902 /* 903 * fvp still locked. ip->i_flag has IN_RENAME set if doingdirectory. 904 * Cleanup fvp requirements so we can unlock it. 905 * 906 * tvp and tdvp are locked. tvp may be NULL. Now that dp and xp 907 * is setup we can use the 'bad' target if we unlock fvp. We cannot 908 * use the abortit target anymore because of IN_RENAME. 909 */ 910 dp = VTOI(tdvp); 911 if (tvp) 912 xp = VTOI(tvp); 913 else 914 xp = NULL; 915 916 /* 917 * 1) Bump link count while we're moving stuff 918 * around. If we crash somewhere before 919 * completing our work, the link count 920 * may be wrong, but correctable. 921 */ 922 ip->i_effnlink++; 923 ip->i_nlink++; 924 ip->i_flag |= IN_CHANGE; 925 if (DOINGSOFTDEP(fvp)) 926 softdep_change_linkcnt(ip); 927 if ((error = ffs_update(fvp, !(DOINGSOFTDEP(fvp) | 928 DOINGASYNC(fvp)))) != 0) { 929 vn_unlock(fvp); 930 goto bad; 931 } 932 933 /* 934 * If ".." must be changed (ie the directory gets a new 935 * parent) then the source directory must not be in the 936 * directory heirarchy above the target, as this would 937 * orphan everything below the source directory. Also 938 * the user must have write permission in the source so 939 * as to be able to change "..". We must repeat the call 940 * to namei, as the parent directory is unlocked by the 941 * call to checkpath(). 942 */ 943 error = VOP_EACCESS(fvp, VWRITE, tcnp->cn_cred); 944 vn_unlock(fvp); 945 946 /* 947 * We are now back to where we were in that fvp, fdvp are unlocked 948 * and tvp, tdvp are locked. tvp may be NULL. IN_RENAME may be 949 * set. Only the bad target or, if we clean up tvp and tdvp, the 950 * out target, may be used. 951 */ 952 if (oldparent != dp->i_number) 953 newparent = dp->i_number; 954 if (doingdirectory && newparent) { 955 if (error) /* write access check above */ 956 goto bad; 957 958 /* 959 * Once we start messing with tvp and tdvp we cannot use the 960 * 'bad' target, only finish cleaning tdvp and tvp up and 961 * use the 'out' target. 962 * 963 * This cleans up tvp. 964 */ 965 if (xp != NULL) { 966 vput(tvp); 967 xp = NULL; 968 } 969 970 /* 971 * This is a real mess. ufs_checkpath vput's the target 972 * directory so retain an extra ref and note that tdvp will 973 * lose its lock on return. This leaves us with one good 974 * ref after ufs_checkpath returns. 975 */ 976 vref(tdvp); 977 error = ufs_checkpath(ip, dp, tcnp->cn_cred); 978 tcnp->cn_flags |= CNP_PDIRUNLOCK; 979 if (error) { 980 vrele(tdvp); 981 goto out; 982 } 983 984 /* 985 * relookup no longer messes with tdvp's refs. tdvp must be 986 * unlocked on entry and will be locked on a successful 987 * return. 988 */ 989 error = relookup(tdvp, &tvp, tcnp); 990 if (error) { 991 if (tcnp->cn_flags & CNP_PDIRUNLOCK) 992 vrele(tdvp); 993 else 994 vput(tdvp); 995 goto out; 996 } 997 KKASSERT((tcnp->cn_flags & CNP_PDIRUNLOCK) == 0); 998 dp = VTOI(tdvp); 999 if (tvp) 1000 xp = VTOI(tvp); 1001 } 1002 1003 /* 1004 * We are back to fvp, fdvp unlocked, tvp, tdvp locked. tvp may 1005 * be NULL (xp will also be NULL in that case), and IN_RENAME will 1006 * be set if doingdirectory. This means we can use the 'bad' target 1007 * again. 1008 */ 1009 1010 /* 1011 * 2) If target doesn't exist, link the target 1012 * to the source and unlink the source. 1013 * Otherwise, rewrite the target directory 1014 * entry to reference the source inode and 1015 * expunge the original entry's existence. 1016 */ 1017 if (xp == NULL) { 1018 if (dp->i_dev != ip->i_dev) 1019 panic("ufs_rename: EXDEV"); 1020 /* 1021 * Account for ".." in new directory. 1022 * When source and destination have the same 1023 * parent we don't fool with the link count. 1024 */ 1025 if (doingdirectory && newparent) { 1026 if ((nlink_t)dp->i_nlink >= LINK_MAX) { 1027 error = EMLINK; 1028 goto bad; 1029 } 1030 dp->i_effnlink++; 1031 dp->i_nlink++; 1032 dp->i_flag |= IN_CHANGE; 1033 if (DOINGSOFTDEP(tdvp)) 1034 softdep_change_linkcnt(dp); 1035 error = ffs_update(tdvp, !(DOINGSOFTDEP(tdvp) | 1036 DOINGASYNC(tdvp))); 1037 if (error) 1038 goto bad; 1039 } 1040 ufs_makedirentry(ip, tcnp, &newdir); 1041 error = ufs_direnter(tdvp, NULL, &newdir, tcnp, NULL); 1042 if (error) { 1043 if (doingdirectory && newparent) { 1044 dp->i_effnlink--; 1045 dp->i_nlink--; 1046 dp->i_flag |= IN_CHANGE; 1047 if (DOINGSOFTDEP(tdvp)) 1048 softdep_change_linkcnt(dp); 1049 (void)ffs_update(tdvp, 1); 1050 } 1051 goto bad; 1052 } 1053 VN_KNOTE(tdvp, NOTE_WRITE); 1054 vput(tdvp); 1055 } else { 1056 if (xp->i_dev != dp->i_dev || xp->i_dev != ip->i_dev) 1057 panic("ufs_rename: EXDEV"); 1058 /* 1059 * Short circuit rename(foo, foo). 1060 */ 1061 if (xp->i_number == ip->i_number) 1062 panic("ufs_rename: same file"); 1063 /* 1064 * If the parent directory is "sticky", then the user must 1065 * own the parent directory, or the destination of the rename, 1066 * otherwise the destination may not be changed (except by 1067 * root). This implements append-only directories. 1068 */ 1069 if ((dp->i_mode & S_ISTXT) && tcnp->cn_cred->cr_uid != 0 && 1070 tcnp->cn_cred->cr_uid != dp->i_uid && 1071 xp->i_uid != tcnp->cn_cred->cr_uid) { 1072 error = EPERM; 1073 goto bad; 1074 } 1075 /* 1076 * Target must be empty if a directory and have no links 1077 * to it. Also, ensure source and target are compatible 1078 * (both directories, or both not directories). 1079 * 1080 * Purge the file or directory being replaced from the 1081 * nameccache. 1082 */ 1083 if ((xp->i_mode&IFMT) == IFDIR) { 1084 if ((xp->i_effnlink > 2) || 1085 !ufs_dirempty(xp, dp->i_number, tcnp->cn_cred)) { 1086 error = ENOTEMPTY; 1087 goto bad; 1088 } 1089 if (!doingdirectory) { 1090 error = ENOTDIR; 1091 goto bad; 1092 } 1093 /* cache_purge removed - handled by VFS compat layer */ 1094 } else if (doingdirectory == 0) { 1095 /* cache_purge removed - handled by VFS compat layer */ 1096 } else { 1097 error = EISDIR; 1098 goto bad; 1099 } 1100 /* 1101 * note: inode passed to ufs_dirrewrite() is 0 for a 1102 * non-directory file rename, 1 for a directory rename 1103 * in the same directory, and > 1 for an inode representing 1104 * the new directory. 1105 */ 1106 error = ufs_dirrewrite(dp, xp, ip->i_number, 1107 IFTODT(ip->i_mode), 1108 (doingdirectory && newparent) ? 1109 newparent : (ino_t)doingdirectory); 1110 if (error) 1111 goto bad; 1112 if (doingdirectory) { 1113 if (!newparent) { 1114 dp->i_effnlink--; 1115 if (DOINGSOFTDEP(tdvp)) 1116 softdep_change_linkcnt(dp); 1117 } 1118 xp->i_effnlink--; 1119 if (DOINGSOFTDEP(tvp)) 1120 softdep_change_linkcnt(xp); 1121 } 1122 if (doingdirectory && !DOINGSOFTDEP(tvp)) { 1123 /* 1124 * Truncate inode. The only stuff left in the directory 1125 * is "." and "..". The "." reference is inconsequential 1126 * since we are quashing it. We have removed the "." 1127 * reference and the reference in the parent directory, 1128 * but there may be other hard links. The soft 1129 * dependency code will arrange to do these operations 1130 * after the parent directory entry has been deleted on 1131 * disk, so when running with that code we avoid doing 1132 * them now. 1133 */ 1134 if (!newparent) { 1135 dp->i_nlink--; 1136 dp->i_flag |= IN_CHANGE; 1137 } 1138 xp->i_nlink--; 1139 xp->i_flag |= IN_CHANGE; 1140 ioflag = DOINGASYNC(tvp) ? 0 : IO_SYNC; 1141 error = ffs_truncate(tvp, (off_t)0, ioflag, 1142 tcnp->cn_cred); 1143 if (error) 1144 goto bad; 1145 } 1146 VN_KNOTE(tdvp, NOTE_WRITE); 1147 vput(tdvp); 1148 VN_KNOTE(tvp, NOTE_DELETE); 1149 vput(tvp); 1150 xp = NULL; 1151 } 1152 1153 /* 1154 * tvp and tdvp have been cleaned up. only fvp and fdvp (both 1155 * unlocked) remain. We are about to overwrite fvp but we have to 1156 * keep 'ip' intact so we cannot release the old fvp, which is still 1157 * refd and accessible via ap->a_fvp. 1158 * 1159 * This means we cannot use either 'bad' or 'out' to cleanup any 1160 * more. 1161 */ 1162 1163 /* 1164 * 3) Unlink the source. 1165 */ 1166 fcnp->cn_flags &= ~CNP_MODMASK; 1167 fcnp->cn_flags |= CNP_LOCKPARENT; 1168 error = relookup(fdvp, &fvp, fcnp); 1169 if (error || fvp == NULL) { 1170 /* 1171 * From name has disappeared. IN_RENAME will not be set if 1172 * we get past the panic so we don't have to clean it up. 1173 */ 1174 if (doingdirectory) 1175 panic("ufs_rename: lost dir entry"); 1176 vrele(ap->a_fvp); 1177 if (fcnp->cn_flags & CNP_PDIRUNLOCK) 1178 vrele(fdvp); 1179 else 1180 vput(fdvp); 1181 return(0); 1182 } 1183 KKASSERT((fcnp->cn_flags & CNP_PDIRUNLOCK) == 0); 1184 1185 /* 1186 * fdvp and fvp are locked. 1187 */ 1188 xp = VTOI(fvp); 1189 dp = VTOI(fdvp); 1190 1191 /* 1192 * Ensure that the directory entry still exists and has not 1193 * changed while the new name has been entered. If the source is 1194 * a file then the entry may have been unlinked or renamed. In 1195 * either case there is no further work to be done. If the source 1196 * is a directory then it cannot have been rmdir'ed; the IN_RENAME 1197 * flag ensures that it cannot be moved by another rename or removed 1198 * by a rmdir. Cleanup IN_RENAME. 1199 */ 1200 if (xp != ip) { 1201 if (doingdirectory) 1202 panic("ufs_rename: lost dir entry"); 1203 } else { 1204 /* 1205 * If the source is a directory with a 1206 * new parent, the link count of the old 1207 * parent directory must be decremented 1208 * and ".." set to point to the new parent. 1209 */ 1210 if (doingdirectory && newparent) { 1211 xp->i_offset = mastertemplate.dot_reclen; 1212 ufs_dirrewrite(xp, dp, newparent, DT_DIR, 0); 1213 /* cache_purge removed - handled by VFS compat layer */ 1214 } 1215 error = ufs_dirremove(fdvp, xp, fcnp->cn_flags, 0); 1216 xp->i_flag &= ~IN_RENAME; 1217 } 1218 1219 VN_KNOTE(fvp, NOTE_RENAME); 1220 vput(fdvp); 1221 vput(fvp); 1222 vrele(ap->a_fvp); 1223 return (error); 1224 1225 bad: 1226 if (xp) 1227 vput(ITOV(xp)); 1228 vput(ITOV(dp)); 1229 out: 1230 if (doingdirectory) 1231 ip->i_flag &= ~IN_RENAME; 1232 if (vn_lock(fvp, LK_EXCLUSIVE) == 0) { 1233 ip->i_effnlink--; 1234 ip->i_nlink--; 1235 ip->i_flag |= IN_CHANGE; 1236 ip->i_flag &= ~IN_RENAME; 1237 if (DOINGSOFTDEP(fvp)) 1238 softdep_change_linkcnt(ip); 1239 vput(fvp); 1240 } else { 1241 vrele(fvp); 1242 } 1243 return (error); 1244 } 1245 1246 /* 1247 * Mkdir system call 1248 * 1249 * ufs_mkdir(struct vnode *a_dvp, struct vnode **a_vpp, 1250 * struct componentname *a_cnp, struct vattr *a_vap) 1251 */ 1252 static 1253 int 1254 ufs_mkdir(struct vop_old_mkdir_args *ap) 1255 { 1256 struct vnode *dvp = ap->a_dvp; 1257 struct vattr *vap = ap->a_vap; 1258 struct componentname *cnp = ap->a_cnp; 1259 struct inode *ip, *dp; 1260 struct vnode *tvp; 1261 struct buf *bp; 1262 struct dirtemplate dirtemplate, *dtp; 1263 struct direct newdir; 1264 int error, dmode; 1265 long blkoff; 1266 1267 dp = VTOI(dvp); 1268 if ((nlink_t)dp->i_nlink >= LINK_MAX) { 1269 error = EMLINK; 1270 goto out; 1271 } 1272 dmode = vap->va_mode & 0777; 1273 dmode |= IFDIR; 1274 /* 1275 * Must simulate part of ufs_makeinode here to acquire the inode, 1276 * but not have it entered in the parent directory. The entry is 1277 * made later after writing "." and ".." entries. 1278 */ 1279 error = ffs_valloc(dvp, dmode, cnp->cn_cred, &tvp); 1280 if (error) 1281 goto out; 1282 ip = VTOI(tvp); 1283 ip->i_gid = dp->i_gid; 1284 #ifdef SUIDDIR 1285 { 1286 #ifdef QUOTA 1287 struct ucred ucred, *ucp; 1288 ucp = cnp->cn_cred; 1289 #endif 1290 /* 1291 * If we are hacking owners here, (only do this where told to) 1292 * and we are not giving it TO root, (would subvert quotas) 1293 * then go ahead and give it to the other user. 1294 * The new directory also inherits the SUID bit. 1295 * If user's UID and dir UID are the same, 1296 * 'give it away' so that the SUID is still forced on. 1297 */ 1298 if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) && 1299 (dp->i_mode & ISUID) && dp->i_uid) { 1300 dmode |= ISUID; 1301 ip->i_uid = dp->i_uid; 1302 #ifdef QUOTA 1303 if (dp->i_uid != cnp->cn_cred->cr_uid) { 1304 /* 1305 * Make sure the correct user gets charged 1306 * for the space. 1307 * Make a dummy credential for the victim. 1308 * XXX This seems to never be accessed out of 1309 * our context so a stack variable is ok. 1310 */ 1311 ucred.cr_ref = 1; 1312 ucred.cr_uid = ip->i_uid; 1313 ucred.cr_ngroups = 1; 1314 ucred.cr_groups[0] = dp->i_gid; 1315 ucp = &ucred; 1316 } 1317 #endif 1318 } else 1319 ip->i_uid = cnp->cn_cred->cr_uid; 1320 #ifdef QUOTA 1321 if ((error = ufs_getinoquota(ip)) || 1322 (error = ufs_chkiq(ip, 1, ucp, 0))) { 1323 ffs_vfree(tvp, ip->i_number, dmode); 1324 vput(tvp); 1325 return (error); 1326 } 1327 #endif 1328 } 1329 #else /* !SUIDDIR */ 1330 ip->i_uid = cnp->cn_cred->cr_uid; 1331 #ifdef QUOTA 1332 if ((error = ufs_getinoquota(ip)) || 1333 (error = ufs_chkiq(ip, 1, cnp->cn_cred, 0))) { 1334 ffs_vfree(tvp, ip->i_number, dmode); 1335 vput(tvp); 1336 return (error); 1337 } 1338 #endif 1339 #endif /* !SUIDDIR */ 1340 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 1341 ip->i_mode = dmode; 1342 tvp->v_type = VDIR; /* Rest init'd in getnewvnode(). */ 1343 ip->i_effnlink = 2; 1344 ip->i_nlink = 2; 1345 if (DOINGSOFTDEP(tvp)) 1346 softdep_change_linkcnt(ip); 1347 if (cnp->cn_flags & CNP_ISWHITEOUT) 1348 ip->i_flags |= UF_OPAQUE; 1349 1350 /* 1351 * Bump link count in parent directory to reflect work done below. 1352 * Should be done before reference is created so cleanup is 1353 * possible if we crash. 1354 */ 1355 dp->i_effnlink++; 1356 dp->i_nlink++; 1357 dp->i_flag |= IN_CHANGE; 1358 if (DOINGSOFTDEP(dvp)) 1359 softdep_change_linkcnt(dp); 1360 error = ffs_update(tvp, !(DOINGSOFTDEP(dvp) | DOINGASYNC(dvp))); 1361 if (error) 1362 goto bad; 1363 1364 /* 1365 * The vnode must have a VM object in order to issue buffer cache 1366 * ops on it. 1367 */ 1368 vinitvmio(tvp, DIRBLKSIZ, DIRBLKSIZ, -1); 1369 1370 /* 1371 * Initialize directory with "." and ".." from static template. 1372 */ 1373 if (dvp->v_mount->mnt_maxsymlinklen > 0) 1374 dtp = &mastertemplate; 1375 else 1376 dtp = (struct dirtemplate *)&omastertemplate; 1377 dirtemplate = *dtp; 1378 dirtemplate.dot_ino = ip->i_number; 1379 dirtemplate.dotdot_ino = dp->i_number; 1380 nvnode_pager_setsize(tvp, DIRBLKSIZ, DIRBLKSIZ, -1); 1381 error = VOP_BALLOC(tvp, 0LL, DIRBLKSIZ, cnp->cn_cred, B_CLRBUF, &bp); 1382 if (error) 1383 goto bad; 1384 ip->i_size = DIRBLKSIZ; 1385 ip->i_flag |= IN_CHANGE | IN_UPDATE; 1386 bcopy((caddr_t)&dirtemplate, (caddr_t)bp->b_data, sizeof dirtemplate); 1387 if (DOINGSOFTDEP(tvp)) { 1388 /* 1389 * Ensure that the entire newly allocated block is a 1390 * valid directory so that future growth within the 1391 * block does not have to ensure that the block is 1392 * written before the inode. 1393 */ 1394 blkoff = DIRBLKSIZ; 1395 while (blkoff < bp->b_bcount) { 1396 ((struct direct *) 1397 (bp->b_data + blkoff))->d_reclen = DIRBLKSIZ; 1398 blkoff += DIRBLKSIZ; 1399 } 1400 } 1401 if ((error = ffs_update(tvp, !(DOINGSOFTDEP(tvp) | 1402 DOINGASYNC(tvp)))) != 0) { 1403 bwrite(bp); 1404 goto bad; 1405 } 1406 /* 1407 * Directory set up, now install its entry in the parent directory. 1408 * 1409 * If we are not doing soft dependencies, then we must write out the 1410 * buffer containing the new directory body before entering the new 1411 * name in the parent. If we are doing soft dependencies, then the 1412 * buffer containing the new directory body will be passed to and 1413 * released in the soft dependency code after the code has attached 1414 * an appropriate ordering dependency to the buffer which ensures that 1415 * the buffer is written before the new name is written in the parent. 1416 */ 1417 if (DOINGASYNC(dvp)) 1418 bdwrite(bp); 1419 else if (!DOINGSOFTDEP(dvp) && (error = bwrite(bp)) != 0) 1420 goto bad; 1421 ufs_makedirentry(ip, cnp, &newdir); 1422 error = ufs_direnter(dvp, tvp, &newdir, cnp, bp); 1423 1424 bad: 1425 if (error == 0) { 1426 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK); 1427 *ap->a_vpp = tvp; 1428 } else { 1429 dp->i_effnlink--; 1430 dp->i_nlink--; 1431 dp->i_flag |= IN_CHANGE; 1432 if (DOINGSOFTDEP(dvp)) 1433 softdep_change_linkcnt(dp); 1434 /* 1435 * No need to do an explicit VOP_TRUNCATE here, vrele will 1436 * do this for us because we set the link count to 0. 1437 */ 1438 ip->i_effnlink = 0; 1439 ip->i_nlink = 0; 1440 ip->i_flag |= IN_CHANGE; 1441 if (DOINGSOFTDEP(tvp)) 1442 softdep_change_linkcnt(ip); 1443 vput(tvp); 1444 } 1445 out: 1446 return (error); 1447 } 1448 1449 /* 1450 * Rmdir system call. 1451 * 1452 * ufs_rmdir(struct vnode *a_dvp, struct vnode *a_vp, 1453 * struct componentname *a_cnp) 1454 */ 1455 static 1456 int 1457 ufs_rmdir(struct vop_old_rmdir_args *ap) 1458 { 1459 struct vnode *vp = ap->a_vp; 1460 struct vnode *dvp = ap->a_dvp; 1461 struct componentname *cnp = ap->a_cnp; 1462 struct inode *ip, *dp; 1463 int error, ioflag; 1464 1465 ip = VTOI(vp); 1466 dp = VTOI(dvp); 1467 1468 /* 1469 * Do not remove a directory that is in the process of being renamed. 1470 * Verify the directory is empty (and valid). Rmdir ".." will not be 1471 * valid since ".." will contain a reference to the current directory 1472 * and thus be non-empty. Do not allow the removal of mounted on 1473 * directories (this can happen when an NFS exported filesystem 1474 * tries to remove a locally mounted on directory). 1475 */ 1476 error = 0; 1477 if (ip->i_flag & IN_RENAME) { 1478 error = EINVAL; 1479 goto out; 1480 } 1481 if (ip->i_effnlink != 2 || 1482 !ufs_dirempty(ip, dp->i_number, cnp->cn_cred)) { 1483 error = ENOTEMPTY; 1484 goto out; 1485 } 1486 #if 0 /* handled by kernel now */ 1487 if ((dp->i_flags & APPEND) 1488 || (ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))) { 1489 error = EPERM; 1490 goto out; 1491 } 1492 #endif 1493 /* 1494 * Delete reference to directory before purging 1495 * inode. If we crash in between, the directory 1496 * will be reattached to lost+found, 1497 */ 1498 dp->i_effnlink--; 1499 ip->i_effnlink--; 1500 if (DOINGSOFTDEP(vp)) { 1501 softdep_change_linkcnt(dp); 1502 softdep_change_linkcnt(ip); 1503 } 1504 error = ufs_dirremove(dvp, ip, cnp->cn_flags, 1); 1505 if (error) { 1506 dp->i_effnlink++; 1507 ip->i_effnlink++; 1508 if (DOINGSOFTDEP(vp)) { 1509 softdep_change_linkcnt(dp); 1510 softdep_change_linkcnt(ip); 1511 } 1512 goto out; 1513 } 1514 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK); 1515 /* 1516 * Truncate inode. The only stuff left in the directory is "." and 1517 * "..". The "." reference is inconsequential since we are quashing 1518 * it. The soft dependency code will arrange to do these operations 1519 * after the parent directory entry has been deleted on disk, so 1520 * when running with that code we avoid doing them now. 1521 */ 1522 if (!DOINGSOFTDEP(vp)) { 1523 dp->i_nlink--; 1524 dp->i_flag |= IN_CHANGE; 1525 ip->i_nlink--; 1526 ip->i_flag |= IN_CHANGE; 1527 ioflag = DOINGASYNC(vp) ? 0 : IO_SYNC; 1528 error = ffs_truncate(vp, (off_t)0, ioflag, cnp->cn_cred); 1529 } 1530 /* cache_purge removed - handled by VFS compat layer */ 1531 #ifdef UFS_DIRHASH 1532 /* Kill any active hash; i_effnlink == 0, so it will not come back. */ 1533 if (ip->i_dirhash != NULL) 1534 ufsdirhash_free(ip); 1535 #endif 1536 out: 1537 VN_KNOTE(vp, NOTE_DELETE); 1538 return (error); 1539 } 1540 1541 /* 1542 * symlink -- make a symbolic link 1543 * 1544 * ufs_symlink(struct vnode *a_dvp, struct vnode **a_vpp, 1545 * struct componentname *a_cnp, struct vattr *a_vap, 1546 * char *a_target) 1547 */ 1548 static 1549 int 1550 ufs_symlink(struct vop_old_symlink_args *ap) 1551 { 1552 struct vnode *vp, **vpp = ap->a_vpp; 1553 struct inode *ip; 1554 int len, error; 1555 1556 error = ufs_makeinode(IFLNK | ap->a_vap->va_mode, ap->a_dvp, 1557 vpp, ap->a_cnp); 1558 if (error) 1559 return (error); 1560 VN_KNOTE(ap->a_dvp, NOTE_WRITE); 1561 vp = *vpp; 1562 len = strlen(ap->a_target); 1563 if (len < vp->v_mount->mnt_maxsymlinklen) { 1564 ip = VTOI(vp); 1565 bcopy(ap->a_target, (char *)ip->i_shortlink, len); 1566 ip->i_size = len; 1567 ip->i_flag |= IN_CHANGE | IN_UPDATE; 1568 } else { 1569 /* 1570 * Make sure we have a VM object in order to use 1571 * the buffer cache. 1572 */ 1573 if (vp->v_object == NULL) 1574 vinitvmio(vp, 0, PAGE_SIZE, -1); 1575 error = vn_rdwr(UIO_WRITE, vp, ap->a_target, len, (off_t)0, 1576 UIO_SYSSPACE, IO_NODELOCKED, 1577 ap->a_cnp->cn_cred, NULL); 1578 } 1579 if (error) 1580 vput(vp); 1581 return (error); 1582 } 1583 1584 /* 1585 * Vnode op for reading directories. 1586 * 1587 * ufs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred, 1588 * int *a_eofflag, int *ncookies, off_t **a_cookies) 1589 */ 1590 static 1591 int 1592 ufs_readdir(struct vop_readdir_args *ap) 1593 { 1594 struct uio *uio = ap->a_uio; 1595 struct vnode *vp = ap->a_vp; 1596 struct direct *dp; 1597 struct buf *bp; 1598 int retval; 1599 int error; 1600 int offset; /* offset into buffer cache buffer */ 1601 int eoffset; /* end of buffer clipped to file EOF */ 1602 int pickup; /* pickup point */ 1603 int ncookies; 1604 int cookie_index; 1605 off_t *cookies; 1606 1607 if (uio->uio_offset < 0) 1608 return (EINVAL); 1609 /* 1610 * Guess the number of cookies needed. Make sure we compute at 1611 * least 1, and no more then a reasonable limit. 1612 */ 1613 if (ap->a_ncookies) { 1614 ncookies = uio->uio_resid / 16 + 1; 1615 if (ncookies > 1024) 1616 ncookies = 1024; 1617 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK); 1618 } else { 1619 ncookies = -1; /* force conditionals below */ 1620 cookies = NULL; 1621 } 1622 cookie_index = 0; 1623 1624 if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0) 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 (vp->v_sysref.refcnt > 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