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; 159 160 if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 161 vfs_timestamp(&ts); 162 if (ip->i_flag & IN_ACCESS) { 163 ip->i_atime = ts.tv_sec; 164 ip->i_atimensec = ts.tv_nsec; 165 } 166 if (ip->i_flag & IN_CHANGE) { 167 ip->i_ctime = ts.tv_sec; 168 ip->i_ctimensec = ts.tv_nsec; 169 } 170 if (ip->i_flag & IN_UPDATE) { 171 if (ip->i_flag & IN_NOCOPYWRITE) { 172 if (vp->v_flag & VLASTWRITETS) { 173 ip->i_mtime = vp->v_lastwrite_ts.tv_sec; 174 ip->i_mtimensec = 175 vp->v_lastwrite_ts.tv_nsec; 176 } 177 } else { 178 ip->i_mtime = ts.tv_sec; 179 ip->i_mtimensec = ts.tv_nsec; 180 } 181 ip->i_modrev++; 182 } 183 } 184 ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_UPDATE); 185 } 186 187 /* 188 * Create a regular file 189 * 190 * ufs_create(struct vnode *a_dvp, struct vnode **a_vpp, 191 * struct componentname *a_cnp, struct vattr *a_vap) 192 */ 193 static 194 int 195 ufs_create(struct vop_old_create_args *ap) 196 { 197 int error; 198 199 error = 200 ufs_makeinode(MAKEIMODE(ap->a_vap->va_type, ap->a_vap->va_mode), 201 ap->a_dvp, ap->a_vpp, ap->a_cnp); 202 if (error) 203 return (error); 204 VN_KNOTE(ap->a_dvp, NOTE_WRITE); 205 return (0); 206 } 207 208 /* 209 * Mknod vnode call 210 * 211 * ufs_mknod(struct vnode *a_dvp, struct vnode **a_vpp, 212 * struct componentname *a_cnp, struct vattr *a_vap) 213 */ 214 /* ARGSUSED */ 215 static 216 int 217 ufs_mknod(struct vop_old_mknod_args *ap) 218 { 219 struct vattr *vap = ap->a_vap; 220 struct vnode **vpp = ap->a_vpp; 221 struct inode *ip; 222 ino_t ino; 223 int error; 224 225 /* 226 * UFS cannot represent the entire major/minor range supported by 227 * the kernel. 228 */ 229 if (vap->va_rmajor != VNOVAL && 230 makeudev(vap->va_rmajor, vap->va_rminor) == NOUDEV) { 231 return(EINVAL); 232 } 233 234 /* no special directory support */ 235 if (vap->va_type == VDIR) 236 return(EINVAL); 237 238 error = ufs_makeinode(MAKEIMODE(vap->va_type, vap->va_mode), 239 ap->a_dvp, vpp, ap->a_cnp); 240 if (error) 241 return (error); 242 VN_KNOTE(ap->a_dvp, NOTE_WRITE); 243 ip = VTOI(*vpp); 244 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 245 if (vap->va_rmajor != VNOVAL) { 246 /* 247 * Want to be able to use this to make badblock 248 * inodes, so don't truncate the dev number. 249 */ 250 ip->i_rdev = makeudev(vap->va_rmajor, vap->va_rminor); 251 } 252 /* 253 * Remove inode, then reload it through VFS_VGET so it is 254 * checked to see if it is an alias of an existing entry in 255 * the inode cache. 256 */ 257 (*vpp)->v_type = VNON; 258 ino = ip->i_number; /* Save this before vgone() invalidates ip. */ 259 vgone_vxlocked(*vpp); 260 vput(*vpp); 261 error = VFS_VGET(ap->a_dvp->v_mount, NULL, ino, vpp); 262 if (error) { 263 *vpp = NULL; 264 return (error); 265 } 266 return (0); 267 } 268 269 /* 270 * Close called. 271 * 272 * Update the times on the inode. 273 * 274 * ufs_close(struct vnode *a_vp, int a_fflag) 275 */ 276 /* ARGSUSED */ 277 static 278 int 279 ufs_close(struct vop_close_args *ap) 280 { 281 struct vnode *vp = ap->a_vp; 282 283 if (VREFCNT(vp) > 1) 284 ufs_itimes(vp); 285 return (vop_stdclose(ap)); 286 } 287 288 /* 289 * ufs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred) 290 */ 291 static 292 int 293 ufs_access(struct vop_access_args *ap) 294 { 295 struct vnode *vp = ap->a_vp; 296 struct inode *ip = VTOI(vp); 297 int error; 298 299 #ifdef QUOTA 300 if (ap->a_mode & VWRITE) { 301 switch (vp->v_type) { 302 case VDIR: 303 case VLNK: 304 case VREG: 305 if ((error = ufs_getinoquota(ip)) != 0) 306 return (error); 307 break; 308 default: 309 break; 310 } 311 } 312 #endif 313 314 error = vop_helper_access(ap, ip->i_uid, ip->i_gid, ip->i_mode, 0); 315 return (error); 316 } 317 318 /* 319 * ufs_getattr(struct vnode *a_vp, struct vattr *a_vap) 320 */ 321 /* ARGSUSED */ 322 static 323 int 324 ufs_getattr(struct vop_getattr_args *ap) 325 { 326 struct vnode *vp = ap->a_vp; 327 struct inode *ip = VTOI(vp); 328 struct vattr *vap = ap->a_vap; 329 330 ufs_itimes(vp); 331 /* 332 * Copy from inode table 333 */ 334 vap->va_fsid = dev2udev(ip->i_dev); 335 vap->va_fileid = ip->i_number; 336 vap->va_mode = ip->i_mode & ~IFMT; 337 vap->va_nlink = VFSTOUFS(vp->v_mount)->um_i_effnlink_valid ? 338 ip->i_effnlink : ip->i_nlink; 339 vap->va_uid = ip->i_uid; 340 vap->va_gid = ip->i_gid; 341 vap->va_rmajor = umajor(ip->i_rdev); 342 vap->va_rminor = uminor(ip->i_rdev); 343 vap->va_size = ip->i_din.di_size; 344 vap->va_atime.tv_sec = ip->i_atime; 345 vap->va_atime.tv_nsec = ip->i_atimensec; 346 vap->va_mtime.tv_sec = ip->i_mtime; 347 vap->va_mtime.tv_nsec = ip->i_mtimensec; 348 vap->va_ctime.tv_sec = ip->i_ctime; 349 vap->va_ctime.tv_nsec = ip->i_ctimensec; 350 vap->va_flags = ip->i_flags; 351 vap->va_gen = ip->i_gen; 352 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize; 353 vap->va_bytes = dbtob((u_quad_t)ip->i_blocks); 354 vap->va_type = IFTOVT(ip->i_mode); 355 vap->va_filerev = ip->i_modrev; 356 return (0); 357 } 358 359 static 360 int 361 ufs_markatime(struct vop_markatime_args *ap) 362 { 363 struct vnode *vp = ap->a_vp; 364 struct inode *ip = VTOI(vp); 365 366 if (vp->v_mount->mnt_flag & MNT_RDONLY) 367 return (EROFS); 368 if (vp->v_mount->mnt_flag & MNT_NOATIME) 369 return (0); 370 ip->i_flag |= IN_ACCESS; 371 VN_KNOTE(vp, NOTE_ATTRIB); 372 return (0); 373 } 374 375 /* 376 * Set attribute vnode op. called from several syscalls 377 * 378 * ufs_setattr(struct vnode *a_vp, struct vattr *a_vap, 379 * struct ucred *a_cred) 380 */ 381 static 382 int 383 ufs_setattr(struct vop_setattr_args *ap) 384 { 385 struct vattr *vap = ap->a_vap; 386 struct vnode *vp = ap->a_vp; 387 struct inode *ip = VTOI(vp); 388 struct ucred *cred = ap->a_cred; 389 int error; 390 391 /* 392 * Check for unsettable attributes. 393 */ 394 if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) || 395 (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) || 396 (vap->va_blocksize != VNOVAL) || (vap->va_rmajor != VNOVAL) || 397 ((int)vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) { 398 return (EINVAL); 399 } 400 if (vap->va_flags != VNOVAL) { 401 if (vp->v_mount->mnt_flag & MNT_RDONLY) 402 return (EROFS); 403 if (cred->cr_uid != ip->i_uid && 404 (error = priv_check_cred(cred, PRIV_VFS_SETATTR, 0))) 405 return (error); 406 /* 407 * Note that a root chflags becomes a user chflags when 408 * we are jailed, unless the jail.chflags_allowed sysctl 409 * is set. 410 */ 411 if (cred->cr_uid == 0 && 412 (!jailed(cred) || jail_chflags_allowed)) { 413 if ((ip->i_flags 414 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) && 415 securelevel > 0) 416 return (EPERM); 417 ip->i_flags = vap->va_flags; 418 } else { 419 if (ip->i_flags 420 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) || 421 (vap->va_flags & UF_SETTABLE) != vap->va_flags) 422 return (EPERM); 423 ip->i_flags &= SF_SETTABLE; 424 ip->i_flags |= (vap->va_flags & UF_SETTABLE); 425 } 426 ip->i_flag |= IN_CHANGE; 427 if (vap->va_flags & (IMMUTABLE | APPEND)) 428 return (0); 429 } 430 if (ip->i_flags & (IMMUTABLE | APPEND)) 431 return (EPERM); 432 /* 433 * Go through the fields and update iff not VNOVAL. 434 */ 435 if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) { 436 if (vp->v_mount->mnt_flag & MNT_RDONLY) 437 return (EROFS); 438 if ((error = ufs_chown(vp, vap->va_uid, vap->va_gid, cred)) != 0) 439 return (error); 440 } 441 if (vap->va_size != VNOVAL) { 442 /* 443 * Disallow write attempts on read-only filesystems; 444 * unless the file is a socket, fifo, or a block or 445 * character device resident on the filesystem. 446 */ 447 switch (vp->v_type) { 448 case VDIR: 449 return (EISDIR); 450 case VLNK: 451 case VREG: 452 if (vp->v_mount->mnt_flag & MNT_RDONLY) 453 return (EROFS); 454 break; 455 default: 456 break; 457 } 458 if ((error = ffs_truncate(vp, vap->va_size, 0, cred)) != 0) 459 return (error); 460 } 461 ip = VTOI(vp); 462 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) { 463 if (vp->v_mount->mnt_flag & MNT_RDONLY) 464 return (EROFS); 465 if (cred->cr_uid != ip->i_uid && 466 (error = priv_check_cred(cred, PRIV_VFS_SETATTR, 0)) && 467 ((vap->va_vaflags & VA_UTIMES_NULL) == 0 || 468 (error = VOP_EACCESS(vp, VWRITE, cred)))) 469 return (error); 470 if (vap->va_atime.tv_sec != VNOVAL) 471 ip->i_flag |= IN_ACCESS; 472 if (vap->va_mtime.tv_sec != VNOVAL) 473 ip->i_flag |= IN_CHANGE | IN_UPDATE; 474 ufs_itimes(vp); 475 if (vap->va_atime.tv_sec != VNOVAL) { 476 ip->i_atime = vap->va_atime.tv_sec; 477 ip->i_atimensec = vap->va_atime.tv_nsec; 478 } 479 if (vap->va_mtime.tv_sec != VNOVAL) { 480 ip->i_mtime = vap->va_mtime.tv_sec; 481 ip->i_mtimensec = vap->va_mtime.tv_nsec; 482 vclrflags(vp, VLASTWRITETS); 483 } 484 error = ffs_update(vp, 0); 485 if (error) 486 return (error); 487 } 488 error = 0; 489 if (vap->va_mode != (mode_t)VNOVAL) { 490 if (vp->v_mount->mnt_flag & MNT_RDONLY) 491 return (EROFS); 492 error = ufs_chmod(vp, (int)vap->va_mode, cred); 493 } 494 VN_KNOTE(vp, NOTE_ATTRIB); 495 return (error); 496 } 497 498 /* 499 * Change the mode on a file. 500 * Inode must be locked before calling. 501 */ 502 static int 503 ufs_chmod(struct vnode *vp, int mode, struct ucred *cred) 504 { 505 struct inode *ip = VTOI(vp); 506 int error; 507 mode_t cur_mode = ip->i_mode; 508 509 error = vop_helper_chmod(vp, mode, cred, ip->i_uid, ip->i_gid, 510 &cur_mode); 511 if (error) 512 return (error); 513 #if 0 514 if (cred->cr_uid != ip->i_uid) { 515 error = priv_check_cred(cred, PRIV_VFS_CHMOD, 0); 516 if (error) 517 return (error); 518 } 519 if (cred->cr_uid) { 520 if (vp->v_type != VDIR && (mode & S_ISTXT)) 521 return (EFTYPE); 522 if (!groupmember(ip->i_gid, cred) && (mode & ISGID)) 523 return (EPERM); 524 } 525 #endif 526 ip->i_mode = cur_mode; 527 ip->i_flag |= IN_CHANGE; 528 return (0); 529 } 530 531 /* 532 * Perform chown operation on inode ip; 533 * inode must be locked prior to call. 534 */ 535 static int 536 ufs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred) 537 { 538 struct inode *ip = VTOI(vp); 539 uid_t ouid; 540 gid_t ogid; 541 int error = 0; 542 #ifdef QUOTA 543 int i; 544 long change; 545 #endif 546 547 if (uid == (uid_t)VNOVAL) 548 uid = ip->i_uid; 549 if (gid == (gid_t)VNOVAL) 550 gid = ip->i_gid; 551 /* 552 * If we don't own the file, are trying to change the owner 553 * of the file, or are not a member of the target group, 554 * the caller must be superuser or the call fails. 555 */ 556 if ((cred->cr_uid != ip->i_uid || uid != ip->i_uid || 557 (gid != ip->i_gid && !(cred->cr_gid == gid || 558 groupmember(gid, cred)))) && 559 (error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0))) 560 return (error); 561 ogid = ip->i_gid; 562 ouid = ip->i_uid; 563 #ifdef QUOTA 564 if ((error = ufs_getinoquota(ip)) != 0) 565 return (error); 566 if (ouid == uid) { 567 ufs_dqrele(vp, ip->i_dquot[USRQUOTA]); 568 ip->i_dquot[USRQUOTA] = NODQUOT; 569 } 570 if (ogid == gid) { 571 ufs_dqrele(vp, ip->i_dquot[GRPQUOTA]); 572 ip->i_dquot[GRPQUOTA] = NODQUOT; 573 } 574 change = ip->i_blocks; 575 (void) ufs_chkdq(ip, -change, cred, CHOWN); 576 (void) ufs_chkiq(ip, -1, cred, CHOWN); 577 for (i = 0; i < MAXQUOTAS; i++) { 578 ufs_dqrele(vp, ip->i_dquot[i]); 579 ip->i_dquot[i] = NODQUOT; 580 } 581 #endif 582 ip->i_gid = gid; 583 ip->i_uid = uid; 584 #ifdef QUOTA 585 if ((error = ufs_getinoquota(ip)) == 0) { 586 if (ouid == uid) { 587 ufs_dqrele(vp, ip->i_dquot[USRQUOTA]); 588 ip->i_dquot[USRQUOTA] = NODQUOT; 589 } 590 if (ogid == gid) { 591 ufs_dqrele(vp, ip->i_dquot[GRPQUOTA]); 592 ip->i_dquot[GRPQUOTA] = NODQUOT; 593 } 594 if ((error = ufs_chkdq(ip, change, cred, CHOWN)) == 0) { 595 if ((error = ufs_chkiq(ip, 1, cred, CHOWN)) == 0) 596 goto good; 597 else 598 (void)ufs_chkdq(ip, -change, cred, CHOWN|FORCE); 599 } 600 for (i = 0; i < MAXQUOTAS; i++) { 601 ufs_dqrele(vp, ip->i_dquot[i]); 602 ip->i_dquot[i] = NODQUOT; 603 } 604 } 605 ip->i_gid = ogid; 606 ip->i_uid = ouid; 607 if (ufs_getinoquota(ip) == 0) { 608 if (ouid == uid) { 609 ufs_dqrele(vp, ip->i_dquot[USRQUOTA]); 610 ip->i_dquot[USRQUOTA] = NODQUOT; 611 } 612 if (ogid == gid) { 613 ufs_dqrele(vp, ip->i_dquot[GRPQUOTA]); 614 ip->i_dquot[GRPQUOTA] = NODQUOT; 615 } 616 (void) ufs_chkdq(ip, change, cred, FORCE|CHOWN); 617 (void) ufs_chkiq(ip, 1, cred, FORCE|CHOWN); 618 (void) ufs_getinoquota(ip); 619 } 620 return (error); 621 good: 622 if (ufs_getinoquota(ip)) 623 panic("ufs_chown: lost quota"); 624 #endif /* QUOTA */ 625 ip->i_flag |= IN_CHANGE; 626 if (cred->cr_uid != 0 && (ouid != uid || ogid != gid)) 627 ip->i_mode &= ~(ISUID | ISGID); 628 return (0); 629 } 630 631 /* 632 * Mmap a file 633 * 634 * NB Currently unsupported. 635 * 636 * ufs_mmap(struct vnode *a_vp, int a_fflags, struct ucred *a_cred) 637 */ 638 /* ARGSUSED */ 639 static 640 int 641 ufs_mmap(struct vop_mmap_args *ap) 642 { 643 return (EINVAL); 644 } 645 646 /* 647 * ufs_remove(struct vnode *a_dvp, struct vnode *a_vp, 648 * struct componentname *a_cnp) 649 */ 650 static 651 int 652 ufs_remove(struct vop_old_remove_args *ap) 653 { 654 struct inode *ip; 655 struct vnode *vp = ap->a_vp; 656 struct vnode *dvp = ap->a_dvp; 657 int error; 658 659 ip = VTOI(vp); 660 #if 0 /* handled by kernel now */ 661 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) || 662 (VTOI(dvp)->i_flags & APPEND)) { 663 error = EPERM; 664 goto out; 665 } 666 #endif 667 error = ufs_dirremove(dvp, ip, ap->a_cnp->cn_flags, 0); 668 VN_KNOTE(vp, NOTE_DELETE); 669 VN_KNOTE(dvp, NOTE_WRITE); 670 #if 0 671 out: 672 #endif 673 return (error); 674 } 675 676 /* 677 * link vnode call 678 * 679 * ufs_link(struct vnode *a_tdvp, struct vnode *a_vp, 680 * struct componentname *a_cnp) 681 */ 682 static 683 int 684 ufs_link(struct vop_old_link_args *ap) 685 { 686 struct vnode *vp = ap->a_vp; 687 struct vnode *tdvp = ap->a_tdvp; 688 struct componentname *cnp = ap->a_cnp; 689 struct inode *ip; 690 struct direct newdir; 691 int error; 692 693 if (tdvp->v_mount != vp->v_mount) { 694 error = EXDEV; 695 goto out2; 696 } 697 if (tdvp != vp) { 698 error = vn_lock(vp, LK_EXCLUSIVE | LK_FAILRECLAIM); 699 if (error) 700 goto out2; 701 } 702 ip = VTOI(vp); 703 if ((nlink_t)ip->i_nlink >= LINK_MAX) { 704 error = EMLINK; 705 goto out1; 706 } 707 #if 0 /* handled by kernel now, also DragonFly allows this */ 708 if (ip->i_flags & (IMMUTABLE | APPEND)) { 709 error = EPERM; 710 goto out1; 711 } 712 #endif 713 ip->i_effnlink++; 714 ip->i_nlink++; 715 ip->i_flag |= IN_CHANGE; 716 if (DOINGSOFTDEP(vp)) 717 softdep_change_linkcnt(ip); 718 error = ffs_update(vp, !(DOINGSOFTDEP(vp) | DOINGASYNC(vp))); 719 if (!error) { 720 ufs_makedirentry(ip, cnp, &newdir); 721 error = ufs_direnter(tdvp, vp, &newdir, cnp, NULL); 722 } 723 724 if (error) { 725 ip->i_effnlink--; 726 ip->i_nlink--; 727 ip->i_flag |= IN_CHANGE; 728 if (DOINGSOFTDEP(vp)) 729 softdep_change_linkcnt(ip); 730 } 731 out1: 732 if (tdvp != vp) 733 vn_unlock(vp); 734 out2: 735 VN_KNOTE(vp, NOTE_LINK); 736 VN_KNOTE(tdvp, NOTE_WRITE); 737 return (error); 738 } 739 740 /* 741 * whiteout vnode call 742 * 743 * ufs_whiteout(struct vnode *a_dvp, struct componentname *a_cnp, int a_flags) 744 */ 745 static 746 int 747 ufs_whiteout(struct vop_old_whiteout_args *ap) 748 { 749 struct vnode *dvp = ap->a_dvp; 750 struct componentname *cnp = ap->a_cnp; 751 struct direct newdir; 752 int error = 0; 753 754 switch (ap->a_flags) { 755 case NAMEI_LOOKUP: 756 /* 4.4 format directories support whiteout operations */ 757 if (dvp->v_mount->mnt_maxsymlinklen > 0) 758 return (0); 759 return (EOPNOTSUPP); 760 761 case NAMEI_CREATE: 762 /* create a new directory whiteout */ 763 #ifdef DIAGNOSTIC 764 if (dvp->v_mount->mnt_maxsymlinklen <= 0) 765 panic("ufs_whiteout: old format filesystem"); 766 #endif 767 768 newdir.d_ino = WINO; 769 newdir.d_namlen = cnp->cn_namelen; 770 bcopy(cnp->cn_nameptr, newdir.d_name, (unsigned)cnp->cn_namelen + 1); 771 newdir.d_type = DT_WHT; 772 error = ufs_direnter(dvp, NULL, &newdir, cnp, NULL); 773 break; 774 775 case NAMEI_DELETE: 776 /* remove an existing directory whiteout */ 777 #ifdef DIAGNOSTIC 778 if (dvp->v_mount->mnt_maxsymlinklen <= 0) 779 panic("ufs_whiteout: old format filesystem"); 780 #endif 781 782 cnp->cn_flags &= ~CNP_DOWHITEOUT; 783 error = ufs_dirremove(dvp, NULL, cnp->cn_flags, 0); 784 break; 785 default: 786 panic("ufs_whiteout: unknown op"); 787 } 788 return (error); 789 } 790 791 /* 792 * Rename system call. 793 * rename("foo", "bar"); 794 * is essentially 795 * unlink("bar"); 796 * link("foo", "bar"); 797 * unlink("foo"); 798 * but ``atomically''. Can't do full commit without saving state in the 799 * inode on disk which isn't feasible at this time. Best we can do is 800 * always guarantee the target exists. 801 * 802 * Basic algorithm is: 803 * 804 * 1) Bump link count on source while we're linking it to the 805 * target. This also ensure the inode won't be deleted out 806 * from underneath us while we work (it may be truncated by 807 * a concurrent `trunc' or `open' for creation). 808 * 2) Link source to destination. If destination already exists, 809 * delete it first. 810 * 3) Unlink source reference to inode if still around. If a 811 * directory was moved and the parent of the destination 812 * is different from the source, patch the ".." entry in the 813 * directory. 814 * 815 * ufs_rename(struct vnode *a_fdvp, struct vnode *a_fvp, 816 * struct componentname *a_fcnp, struct vnode *a_tdvp, 817 * struct vnode *a_tvp, struct componentname *a_tcnp) 818 */ 819 static 820 int 821 ufs_rename(struct vop_old_rename_args *ap) 822 { 823 struct vnode *tvp = ap->a_tvp; 824 struct vnode *tdvp = ap->a_tdvp; 825 struct vnode *fvp = ap->a_fvp; 826 struct vnode *fdvp = ap->a_fdvp; 827 struct componentname *tcnp = ap->a_tcnp; 828 struct componentname *fcnp = ap->a_fcnp; 829 struct inode *ip, *xp, *dp; 830 struct direct newdir; 831 ino_t oldparent = 0, newparent = 0; 832 int doingdirectory = 0; 833 int error = 0, ioflag; 834 835 /* 836 * Check for cross-device rename. 837 */ 838 if ((fvp->v_mount != tdvp->v_mount) || 839 (tvp && (fvp->v_mount != tvp->v_mount))) { 840 error = EXDEV; 841 abortit: 842 if (tdvp == tvp) 843 vrele(tdvp); 844 else 845 vput(tdvp); 846 if (tvp) 847 vput(tvp); 848 vrele(fdvp); 849 vrele(fvp); 850 return (error); 851 } 852 853 #if 0 /* handled by kernel now */ 854 if (tvp && ((VTOI(tvp)->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) || 855 (VTOI(tdvp)->i_flags & APPEND))) { 856 error = EPERM; 857 goto abortit; 858 } 859 #endif 860 861 /* 862 * Renaming a file to itself has no effect. The upper layers should 863 * not call us in that case. Temporarily just warn if they do. 864 */ 865 if (fvp == tvp) { 866 kprintf("ufs_rename: fvp == tvp (can't happen)\n"); 867 error = 0; 868 goto abortit; 869 } 870 871 error = vn_lock(fvp, LK_EXCLUSIVE | LK_FAILRECLAIM); 872 if (error) 873 goto abortit; 874 875 /* 876 * Note: now that fvp is locked we have to be sure to unlock it before 877 * using the 'abortit' target. 878 */ 879 dp = VTOI(fdvp); 880 ip = VTOI(fvp); 881 if (ip->i_nlink >= LINK_MAX) { 882 vn_unlock(fvp); 883 error = EMLINK; 884 goto abortit; 885 } 886 #if 0 /* handled by kernel now */ 887 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) 888 || (dp->i_flags & APPEND)) { 889 vn_unlock(fvp); 890 error = EPERM; 891 goto abortit; 892 } 893 #endif 894 if ((ip->i_mode & IFMT) == IFDIR) { 895 /* 896 * Avoid ".", "..", and aliases of "." for obvious reasons. 897 */ 898 if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') || 899 dp == ip || (fcnp->cn_flags | tcnp->cn_flags) & CNP_ISDOTDOT || 900 (ip->i_flag & IN_RENAME)) { 901 vn_unlock(fvp); 902 error = EINVAL; 903 goto abortit; 904 } 905 ip->i_flag |= IN_RENAME; 906 oldparent = dp->i_number; 907 doingdirectory = 1; 908 } 909 VN_KNOTE(fdvp, NOTE_WRITE); /* XXX right place? */ 910 911 /* 912 * fvp still locked. ip->i_flag has IN_RENAME set if doingdirectory. 913 * Cleanup fvp requirements so we can unlock it. 914 * 915 * tvp and tdvp are locked. tvp may be NULL. Now that dp and xp 916 * is setup we can use the 'bad' target if we unlock fvp. We cannot 917 * use the abortit target anymore because of IN_RENAME. 918 */ 919 dp = VTOI(tdvp); 920 if (tvp) 921 xp = VTOI(tvp); 922 else 923 xp = NULL; 924 925 /* 926 * 1) Bump link count while we're moving stuff 927 * around. If we crash somewhere before 928 * completing our work, the link count 929 * may be wrong, but correctable. 930 */ 931 ip->i_effnlink++; 932 ip->i_nlink++; 933 ip->i_flag |= IN_CHANGE; 934 if (DOINGSOFTDEP(fvp)) 935 softdep_change_linkcnt(ip); 936 if ((error = ffs_update(fvp, !(DOINGSOFTDEP(fvp) | 937 DOINGASYNC(fvp)))) != 0) { 938 vn_unlock(fvp); 939 goto bad; 940 } 941 942 /* 943 * If ".." must be changed (ie the directory gets a new 944 * parent) then the source directory must not be in the 945 * directory heirarchy above the target, as this would 946 * orphan everything below the source directory. Also 947 * the user must have write permission in the source so 948 * as to be able to change "..". We must repeat the call 949 * to namei, as the parent directory is unlocked by the 950 * call to checkpath(). 951 */ 952 error = VOP_EACCESS(fvp, VWRITE, tcnp->cn_cred); 953 vn_unlock(fvp); 954 955 /* 956 * We are now back to where we were in that fvp, fdvp are unlocked 957 * and tvp, tdvp are locked. tvp may be NULL. IN_RENAME may be 958 * set. Only the bad target or, if we clean up tvp and tdvp, the 959 * out target, may be used. 960 */ 961 if (oldparent != dp->i_number) 962 newparent = dp->i_number; 963 if (doingdirectory && newparent) { 964 if (error) /* write access check above */ 965 goto bad; 966 967 /* 968 * Once we start messing with tvp and tdvp we cannot use the 969 * 'bad' target, only finish cleaning tdvp and tvp up and 970 * use the 'out' target. 971 * 972 * This cleans up tvp. 973 */ 974 if (xp != NULL) { 975 vput(tvp); 976 xp = NULL; 977 } 978 979 /* 980 * This is a real mess. ufs_checkpath vput's the target 981 * directory so retain an extra ref and note that tdvp will 982 * lose its lock on return. This leaves us with one good 983 * ref after ufs_checkpath returns. 984 */ 985 vref(tdvp); 986 error = ufs_checkpath(ip, dp, tcnp->cn_cred); 987 tcnp->cn_flags |= CNP_PDIRUNLOCK; 988 if (error) { 989 vrele(tdvp); 990 goto out; 991 } 992 993 /* 994 * relookup no longer messes with tdvp's refs. tdvp must be 995 * unlocked on entry and will be locked on a successful 996 * return. 997 */ 998 error = relookup(tdvp, &tvp, tcnp); 999 if (error) { 1000 if (tcnp->cn_flags & CNP_PDIRUNLOCK) 1001 vrele(tdvp); 1002 else 1003 vput(tdvp); 1004 goto out; 1005 } 1006 KKASSERT((tcnp->cn_flags & CNP_PDIRUNLOCK) == 0); 1007 dp = VTOI(tdvp); 1008 if (tvp) 1009 xp = VTOI(tvp); 1010 } 1011 1012 /* 1013 * We are back to fvp, fdvp unlocked, tvp, tdvp locked. tvp may 1014 * be NULL (xp will also be NULL in that case), and IN_RENAME will 1015 * be set if doingdirectory. This means we can use the 'bad' target 1016 * again. 1017 */ 1018 1019 /* 1020 * 2) If target doesn't exist, link the target 1021 * to the source and unlink the source. 1022 * Otherwise, rewrite the target directory 1023 * entry to reference the source inode and 1024 * expunge the original entry's existence. 1025 */ 1026 if (xp == NULL) { 1027 if (dp->i_dev != ip->i_dev) 1028 panic("ufs_rename: EXDEV"); 1029 /* 1030 * Account for ".." in new directory. 1031 * When source and destination have the same 1032 * parent we don't fool with the link count. 1033 */ 1034 if (doingdirectory && newparent) { 1035 if ((nlink_t)dp->i_nlink >= LINK_MAX) { 1036 error = EMLINK; 1037 goto bad; 1038 } 1039 dp->i_effnlink++; 1040 dp->i_nlink++; 1041 dp->i_flag |= IN_CHANGE; 1042 if (DOINGSOFTDEP(tdvp)) 1043 softdep_change_linkcnt(dp); 1044 error = ffs_update(tdvp, !(DOINGSOFTDEP(tdvp) | 1045 DOINGASYNC(tdvp))); 1046 if (error) 1047 goto bad; 1048 } 1049 ufs_makedirentry(ip, tcnp, &newdir); 1050 error = ufs_direnter(tdvp, NULL, &newdir, tcnp, NULL); 1051 if (error) { 1052 if (doingdirectory && newparent) { 1053 dp->i_effnlink--; 1054 dp->i_nlink--; 1055 dp->i_flag |= IN_CHANGE; 1056 if (DOINGSOFTDEP(tdvp)) 1057 softdep_change_linkcnt(dp); 1058 (void)ffs_update(tdvp, 1); 1059 } 1060 goto bad; 1061 } 1062 VN_KNOTE(tdvp, NOTE_WRITE); 1063 vput(tdvp); 1064 } else { 1065 if (xp->i_dev != dp->i_dev || xp->i_dev != ip->i_dev) 1066 panic("ufs_rename: EXDEV"); 1067 /* 1068 * Short circuit rename(foo, foo). 1069 */ 1070 if (xp->i_number == ip->i_number) 1071 panic("ufs_rename: same file"); 1072 /* 1073 * If the parent directory is "sticky", then the user must 1074 * own the parent directory, or the destination of the rename, 1075 * otherwise the destination may not be changed (except by 1076 * root). This implements append-only directories. 1077 */ 1078 if ((dp->i_mode & S_ISTXT) && tcnp->cn_cred->cr_uid != 0 && 1079 tcnp->cn_cred->cr_uid != dp->i_uid && 1080 xp->i_uid != tcnp->cn_cred->cr_uid) { 1081 error = EPERM; 1082 goto bad; 1083 } 1084 /* 1085 * Target must be empty if a directory and have no links 1086 * to it. Also, ensure source and target are compatible 1087 * (both directories, or both not directories). 1088 * 1089 * Purge the file or directory being replaced from the 1090 * nameccache. 1091 */ 1092 if ((xp->i_mode&IFMT) == IFDIR) { 1093 if ((xp->i_effnlink > 2) || 1094 !ufs_dirempty(xp, dp->i_number, tcnp->cn_cred)) { 1095 error = ENOTEMPTY; 1096 goto bad; 1097 } 1098 if (!doingdirectory) { 1099 error = ENOTDIR; 1100 goto bad; 1101 } 1102 /* cache_purge removed - handled by VFS compat layer */ 1103 } else if (doingdirectory == 0) { 1104 /* cache_purge removed - handled by VFS compat layer */ 1105 } else { 1106 error = EISDIR; 1107 goto bad; 1108 } 1109 /* 1110 * note: inode passed to ufs_dirrewrite() is 0 for a 1111 * non-directory file rename, 1 for a directory rename 1112 * in the same directory, and > 1 for an inode representing 1113 * the new directory. 1114 */ 1115 error = ufs_dirrewrite(dp, xp, ip->i_number, 1116 IFTODT(ip->i_mode), 1117 (doingdirectory && newparent) ? 1118 newparent : (ino_t)doingdirectory); 1119 if (error) 1120 goto bad; 1121 if (doingdirectory) { 1122 if (!newparent) { 1123 dp->i_effnlink--; 1124 if (DOINGSOFTDEP(tdvp)) 1125 softdep_change_linkcnt(dp); 1126 } 1127 xp->i_effnlink--; 1128 if (DOINGSOFTDEP(tvp)) 1129 softdep_change_linkcnt(xp); 1130 } 1131 if (doingdirectory && !DOINGSOFTDEP(tvp)) { 1132 /* 1133 * Truncate inode. The only stuff left in the directory 1134 * is "." and "..". The "." reference is inconsequential 1135 * since we are quashing it. We have removed the "." 1136 * reference and the reference in the parent directory, 1137 * but there may be other hard links. The soft 1138 * dependency code will arrange to do these operations 1139 * after the parent directory entry has been deleted on 1140 * disk, so when running with that code we avoid doing 1141 * them now. 1142 */ 1143 if (!newparent) { 1144 dp->i_nlink--; 1145 dp->i_flag |= IN_CHANGE; 1146 } 1147 xp->i_nlink--; 1148 xp->i_flag |= IN_CHANGE; 1149 ioflag = DOINGASYNC(tvp) ? 0 : IO_SYNC; 1150 error = ffs_truncate(tvp, (off_t)0, ioflag, 1151 tcnp->cn_cred); 1152 if (error) 1153 goto bad; 1154 } 1155 VN_KNOTE(tdvp, NOTE_WRITE); 1156 vput(tdvp); 1157 VN_KNOTE(tvp, NOTE_DELETE); 1158 vput(tvp); 1159 xp = NULL; 1160 } 1161 1162 /* 1163 * tvp and tdvp have been cleaned up. only fvp and fdvp (both 1164 * unlocked) remain. We are about to overwrite fvp but we have to 1165 * keep 'ip' intact so we cannot release the old fvp, which is still 1166 * refd and accessible via ap->a_fvp. 1167 * 1168 * This means we cannot use either 'bad' or 'out' to cleanup any 1169 * more. 1170 */ 1171 1172 /* 1173 * 3) Unlink the source. 1174 */ 1175 fcnp->cn_flags &= ~CNP_MODMASK; 1176 fcnp->cn_flags |= CNP_LOCKPARENT; 1177 error = relookup(fdvp, &fvp, fcnp); 1178 if (error || fvp == NULL) { 1179 /* 1180 * From name has disappeared. IN_RENAME will not be set if 1181 * we get past the panic so we don't have to clean it up. 1182 */ 1183 if (doingdirectory) 1184 panic("ufs_rename: lost dir entry"); 1185 vrele(ap->a_fvp); 1186 if (fcnp->cn_flags & CNP_PDIRUNLOCK) 1187 vrele(fdvp); 1188 else 1189 vput(fdvp); 1190 return(0); 1191 } 1192 KKASSERT((fcnp->cn_flags & CNP_PDIRUNLOCK) == 0); 1193 1194 /* 1195 * fdvp and fvp are locked. 1196 */ 1197 xp = VTOI(fvp); 1198 dp = VTOI(fdvp); 1199 1200 /* 1201 * Ensure that the directory entry still exists and has not 1202 * changed while the new name has been entered. If the source is 1203 * a file then the entry may have been unlinked or renamed. In 1204 * either case there is no further work to be done. If the source 1205 * is a directory then it cannot have been rmdir'ed; the IN_RENAME 1206 * flag ensures that it cannot be moved by another rename or removed 1207 * by a rmdir. Cleanup IN_RENAME. 1208 */ 1209 if (xp != ip) { 1210 if (doingdirectory) 1211 panic("ufs_rename: lost dir entry"); 1212 } else { 1213 /* 1214 * If the source is a directory with a 1215 * new parent, the link count of the old 1216 * parent directory must be decremented 1217 * and ".." set to point to the new parent. 1218 */ 1219 if (doingdirectory && newparent) { 1220 xp->i_offset = mastertemplate.dot_reclen; 1221 ufs_dirrewrite(xp, dp, newparent, DT_DIR, 0); 1222 /* cache_purge removed - handled by VFS compat layer */ 1223 } 1224 error = ufs_dirremove(fdvp, xp, fcnp->cn_flags, 0); 1225 xp->i_flag &= ~IN_RENAME; 1226 } 1227 1228 VN_KNOTE(fvp, NOTE_RENAME); 1229 vput(fdvp); 1230 vput(fvp); 1231 vrele(ap->a_fvp); 1232 return (error); 1233 1234 bad: 1235 if (xp) 1236 vput(ITOV(xp)); 1237 vput(ITOV(dp)); 1238 out: 1239 if (doingdirectory) 1240 ip->i_flag &= ~IN_RENAME; 1241 if (vn_lock(fvp, LK_EXCLUSIVE | LK_FAILRECLAIM) == 0) { 1242 ip->i_effnlink--; 1243 ip->i_nlink--; 1244 ip->i_flag |= IN_CHANGE; 1245 ip->i_flag &= ~IN_RENAME; 1246 if (DOINGSOFTDEP(fvp)) 1247 softdep_change_linkcnt(ip); 1248 vput(fvp); 1249 } else { 1250 vrele(fvp); 1251 } 1252 return (error); 1253 } 1254 1255 /* 1256 * Mkdir system call 1257 * 1258 * ufs_mkdir(struct vnode *a_dvp, struct vnode **a_vpp, 1259 * struct componentname *a_cnp, struct vattr *a_vap) 1260 */ 1261 static 1262 int 1263 ufs_mkdir(struct vop_old_mkdir_args *ap) 1264 { 1265 struct vnode *dvp = ap->a_dvp; 1266 struct vattr *vap = ap->a_vap; 1267 struct componentname *cnp = ap->a_cnp; 1268 struct inode *ip, *dp; 1269 struct vnode *tvp; 1270 struct buf *bp; 1271 struct dirtemplate dirtemplate, *dtp; 1272 struct direct newdir; 1273 int error, dmode; 1274 long blkoff; 1275 1276 dp = VTOI(dvp); 1277 if ((nlink_t)dp->i_nlink >= LINK_MAX) { 1278 error = EMLINK; 1279 goto out; 1280 } 1281 dmode = vap->va_mode & 0777; 1282 dmode |= IFDIR; 1283 /* 1284 * Must simulate part of ufs_makeinode here to acquire the inode, 1285 * but not have it entered in the parent directory. The entry is 1286 * made later after writing "." and ".." entries. 1287 */ 1288 error = ffs_valloc(dvp, dmode, cnp->cn_cred, &tvp); 1289 if (error) 1290 goto out; 1291 ip = VTOI(tvp); 1292 ip->i_gid = dp->i_gid; 1293 #ifdef SUIDDIR 1294 { 1295 #ifdef QUOTA 1296 struct ucred ucred, *ucp; 1297 ucp = cnp->cn_cred; 1298 #endif 1299 /* 1300 * If we are hacking owners here, (only do this where told to) 1301 * and we are not giving it TO root, (would subvert quotas) 1302 * then go ahead and give it to the other user. 1303 * The new directory also inherits the SUID bit. 1304 * If user's UID and dir UID are the same, 1305 * 'give it away' so that the SUID is still forced on. 1306 */ 1307 if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) && 1308 (dp->i_mode & ISUID) && dp->i_uid) { 1309 dmode |= ISUID; 1310 ip->i_uid = dp->i_uid; 1311 #ifdef QUOTA 1312 if (dp->i_uid != cnp->cn_cred->cr_uid) { 1313 /* 1314 * Make sure the correct user gets charged 1315 * for the space. 1316 * Make a dummy credential for the victim. 1317 * XXX This seems to never be accessed out of 1318 * our context so a stack variable is ok. 1319 */ 1320 ucred.cr_ref = 1; 1321 ucred.cr_uid = ip->i_uid; 1322 ucred.cr_ngroups = 1; 1323 ucred.cr_groups[0] = dp->i_gid; 1324 ucp = &ucred; 1325 } 1326 #endif 1327 } else 1328 ip->i_uid = cnp->cn_cred->cr_uid; 1329 #ifdef QUOTA 1330 if ((error = ufs_getinoquota(ip)) || 1331 (error = ufs_chkiq(ip, 1, ucp, 0))) { 1332 ffs_vfree(tvp, ip->i_number, dmode); 1333 vput(tvp); 1334 return (error); 1335 } 1336 #endif 1337 } 1338 #else /* !SUIDDIR */ 1339 ip->i_uid = cnp->cn_cred->cr_uid; 1340 #ifdef QUOTA 1341 if ((error = ufs_getinoquota(ip)) || 1342 (error = ufs_chkiq(ip, 1, cnp->cn_cred, 0))) { 1343 ffs_vfree(tvp, ip->i_number, dmode); 1344 vput(tvp); 1345 return (error); 1346 } 1347 #endif 1348 #endif /* !SUIDDIR */ 1349 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 1350 ip->i_mode = dmode; 1351 tvp->v_type = VDIR; /* Rest init'd in getnewvnode(). */ 1352 ip->i_effnlink = 2; 1353 ip->i_nlink = 2; 1354 if (DOINGSOFTDEP(tvp)) 1355 softdep_change_linkcnt(ip); 1356 if (cnp->cn_flags & CNP_ISWHITEOUT) 1357 ip->i_flags |= UF_OPAQUE; 1358 1359 /* 1360 * Bump link count in parent directory to reflect work done below. 1361 * Should be done before reference is created so cleanup is 1362 * possible if we crash. 1363 */ 1364 dp->i_effnlink++; 1365 dp->i_nlink++; 1366 dp->i_flag |= IN_CHANGE; 1367 if (DOINGSOFTDEP(dvp)) 1368 softdep_change_linkcnt(dp); 1369 error = ffs_update(tvp, !(DOINGSOFTDEP(dvp) | DOINGASYNC(dvp))); 1370 if (error) 1371 goto bad; 1372 1373 /* 1374 * The vnode must have a VM object in order to issue buffer cache 1375 * ops on it. 1376 */ 1377 vinitvmio(tvp, DIRBLKSIZ, DIRBLKSIZ, -1); 1378 1379 /* 1380 * Initialize directory with "." and ".." from static template. 1381 */ 1382 if (dvp->v_mount->mnt_maxsymlinklen > 0) 1383 dtp = &mastertemplate; 1384 else 1385 dtp = (struct dirtemplate *)&omastertemplate; 1386 dirtemplate = *dtp; 1387 dirtemplate.dot_ino = ip->i_number; 1388 dirtemplate.dotdot_ino = dp->i_number; 1389 nvnode_pager_setsize(tvp, DIRBLKSIZ, DIRBLKSIZ, -1); 1390 error = VOP_BALLOC(tvp, 0LL, DIRBLKSIZ, cnp->cn_cred, B_CLRBUF, &bp); 1391 if (error) 1392 goto bad; 1393 ip->i_size = DIRBLKSIZ; 1394 ip->i_flag |= IN_CHANGE | IN_UPDATE; 1395 bcopy((caddr_t)&dirtemplate, (caddr_t)bp->b_data, sizeof dirtemplate); 1396 if (DOINGSOFTDEP(tvp)) { 1397 /* 1398 * Ensure that the entire newly allocated block is a 1399 * valid directory so that future growth within the 1400 * block does not have to ensure that the block is 1401 * written before the inode. 1402 */ 1403 blkoff = DIRBLKSIZ; 1404 while (blkoff < bp->b_bcount) { 1405 ((struct direct *) 1406 (bp->b_data + blkoff))->d_reclen = DIRBLKSIZ; 1407 blkoff += DIRBLKSIZ; 1408 } 1409 } 1410 if ((error = ffs_update(tvp, !(DOINGSOFTDEP(tvp) | 1411 DOINGASYNC(tvp)))) != 0) { 1412 bwrite(bp); 1413 goto bad; 1414 } 1415 /* 1416 * Directory set up, now install its entry in the parent directory. 1417 * 1418 * If we are not doing soft dependencies, then we must write out the 1419 * buffer containing the new directory body before entering the new 1420 * name in the parent. If we are doing soft dependencies, then the 1421 * buffer containing the new directory body will be passed to and 1422 * released in the soft dependency code after the code has attached 1423 * an appropriate ordering dependency to the buffer which ensures that 1424 * the buffer is written before the new name is written in the parent. 1425 */ 1426 if (DOINGASYNC(dvp)) 1427 bdwrite(bp); 1428 else if (!DOINGSOFTDEP(dvp) && (error = bwrite(bp)) != 0) 1429 goto bad; 1430 ufs_makedirentry(ip, cnp, &newdir); 1431 error = ufs_direnter(dvp, tvp, &newdir, cnp, bp); 1432 1433 bad: 1434 if (error == 0) { 1435 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK); 1436 *ap->a_vpp = tvp; 1437 } else { 1438 dp->i_effnlink--; 1439 dp->i_nlink--; 1440 dp->i_flag |= IN_CHANGE; 1441 if (DOINGSOFTDEP(dvp)) 1442 softdep_change_linkcnt(dp); 1443 /* 1444 * No need to do an explicit VOP_TRUNCATE here, vrele will 1445 * do this for us because we set the link count to 0. 1446 */ 1447 ip->i_effnlink = 0; 1448 ip->i_nlink = 0; 1449 ip->i_flag |= IN_CHANGE; 1450 if (DOINGSOFTDEP(tvp)) 1451 softdep_change_linkcnt(ip); 1452 vput(tvp); 1453 } 1454 out: 1455 return (error); 1456 } 1457 1458 /* 1459 * Rmdir system call. 1460 * 1461 * ufs_rmdir(struct vnode *a_dvp, struct vnode *a_vp, 1462 * struct componentname *a_cnp) 1463 */ 1464 static 1465 int 1466 ufs_rmdir(struct vop_old_rmdir_args *ap) 1467 { 1468 struct vnode *vp = ap->a_vp; 1469 struct vnode *dvp = ap->a_dvp; 1470 struct componentname *cnp = ap->a_cnp; 1471 struct inode *ip, *dp; 1472 int error, ioflag; 1473 1474 ip = VTOI(vp); 1475 dp = VTOI(dvp); 1476 1477 /* 1478 * Do not remove a directory that is in the process of being renamed. 1479 * Verify the directory is empty (and valid). Rmdir ".." will not be 1480 * valid since ".." will contain a reference to the current directory 1481 * and thus be non-empty. Do not allow the removal of mounted on 1482 * directories (this can happen when an NFS exported filesystem 1483 * tries to remove a locally mounted on directory). 1484 */ 1485 error = 0; 1486 if (ip->i_flag & IN_RENAME) { 1487 error = EINVAL; 1488 goto out; 1489 } 1490 if (ip->i_effnlink != 2 || 1491 !ufs_dirempty(ip, dp->i_number, cnp->cn_cred)) { 1492 error = ENOTEMPTY; 1493 goto out; 1494 } 1495 #if 0 /* handled by kernel now */ 1496 if ((dp->i_flags & APPEND) 1497 || (ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))) { 1498 error = EPERM; 1499 goto out; 1500 } 1501 #endif 1502 /* 1503 * Delete reference to directory before purging 1504 * inode. If we crash in between, the directory 1505 * will be reattached to lost+found, 1506 */ 1507 dp->i_effnlink--; 1508 ip->i_effnlink--; 1509 if (DOINGSOFTDEP(vp)) { 1510 softdep_change_linkcnt(dp); 1511 softdep_change_linkcnt(ip); 1512 } 1513 error = ufs_dirremove(dvp, ip, cnp->cn_flags, 1); 1514 if (error) { 1515 dp->i_effnlink++; 1516 ip->i_effnlink++; 1517 if (DOINGSOFTDEP(vp)) { 1518 softdep_change_linkcnt(dp); 1519 softdep_change_linkcnt(ip); 1520 } 1521 goto out; 1522 } 1523 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK); 1524 /* 1525 * Truncate inode. The only stuff left in the directory is "." and 1526 * "..". The "." reference is inconsequential since we are quashing 1527 * it. The soft dependency code will arrange to do these operations 1528 * after the parent directory entry has been deleted on disk, so 1529 * when running with that code we avoid doing them now. 1530 */ 1531 if (!DOINGSOFTDEP(vp)) { 1532 dp->i_nlink--; 1533 dp->i_flag |= IN_CHANGE; 1534 ip->i_nlink--; 1535 ip->i_flag |= IN_CHANGE; 1536 ioflag = DOINGASYNC(vp) ? 0 : IO_SYNC; 1537 error = ffs_truncate(vp, (off_t)0, ioflag, cnp->cn_cred); 1538 } 1539 /* cache_purge removed - handled by VFS compat layer */ 1540 #ifdef UFS_DIRHASH 1541 /* Kill any active hash; i_effnlink == 0, so it will not come back. */ 1542 if (ip->i_dirhash != NULL) 1543 ufsdirhash_free(ip); 1544 #endif 1545 out: 1546 VN_KNOTE(vp, NOTE_DELETE); 1547 return (error); 1548 } 1549 1550 /* 1551 * symlink -- make a symbolic link 1552 * 1553 * ufs_symlink(struct vnode *a_dvp, struct vnode **a_vpp, 1554 * struct componentname *a_cnp, struct vattr *a_vap, 1555 * char *a_target) 1556 */ 1557 static 1558 int 1559 ufs_symlink(struct vop_old_symlink_args *ap) 1560 { 1561 struct vnode *vp, **vpp = ap->a_vpp; 1562 struct inode *ip; 1563 int len, error; 1564 1565 error = ufs_makeinode(IFLNK | ap->a_vap->va_mode, ap->a_dvp, 1566 vpp, ap->a_cnp); 1567 if (error) 1568 return (error); 1569 VN_KNOTE(ap->a_dvp, NOTE_WRITE); 1570 vp = *vpp; 1571 len = strlen(ap->a_target); 1572 if (len < vp->v_mount->mnt_maxsymlinklen) { 1573 ip = VTOI(vp); 1574 bcopy(ap->a_target, (char *)ip->i_shortlink, len); 1575 ip->i_size = len; 1576 ip->i_flag |= IN_CHANGE | IN_UPDATE; 1577 } else { 1578 /* 1579 * Make sure we have a VM object in order to use 1580 * the buffer cache. 1581 */ 1582 if (vp->v_object == NULL) 1583 vinitvmio(vp, 0, PAGE_SIZE, -1); 1584 error = vn_rdwr(UIO_WRITE, vp, ap->a_target, len, (off_t)0, 1585 UIO_SYSSPACE, IO_NODELOCKED, 1586 ap->a_cnp->cn_cred, NULL); 1587 } 1588 if (error) 1589 vput(vp); 1590 return (error); 1591 } 1592 1593 /* 1594 * Vnode op for reading directories. 1595 * 1596 * ufs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred, 1597 * int *a_eofflag, int *ncookies, off_t **a_cookies) 1598 */ 1599 static 1600 int 1601 ufs_readdir(struct vop_readdir_args *ap) 1602 { 1603 struct uio *uio = ap->a_uio; 1604 struct vnode *vp = ap->a_vp; 1605 struct direct *dp; 1606 struct buf *bp; 1607 int retval; 1608 int error; 1609 int offset; /* offset into buffer cache buffer */ 1610 int eoffset; /* end of buffer clipped to file EOF */ 1611 int pickup; /* pickup point */ 1612 int ncookies; 1613 int cookie_index; 1614 off_t *cookies; 1615 1616 if (uio->uio_offset < 0) 1617 return (EINVAL); 1618 /* 1619 * Guess the number of cookies needed. Make sure we compute at 1620 * least 1, and no more then a reasonable limit. 1621 */ 1622 if (ap->a_ncookies) { 1623 ncookies = uio->uio_resid / 16 + 1; 1624 if (ncookies > 1024) 1625 ncookies = 1024; 1626 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK); 1627 } else { 1628 ncookies = -1; /* force conditionals below */ 1629 cookies = NULL; 1630 } 1631 cookie_index = 0; 1632 1633 error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY | LK_FAILRECLAIM); 1634 if (error) 1635 return (error); 1636 1637 /* 1638 * Past or at EOF 1639 */ 1640 if (uio->uio_offset >= VTOI(vp)->i_size) { 1641 if (ap->a_eofflag) 1642 *ap->a_eofflag = 1; 1643 if (ap->a_ncookies) { 1644 *ap->a_ncookies = cookie_index; 1645 *ap->a_cookies = cookies; 1646 } 1647 goto done; 1648 } 1649 1650 /* 1651 * Loop until we run out of cookies, we run out of user buffer, 1652 * or we hit the directory EOF. 1653 * 1654 * Always start scans at the beginning of the buffer, don't trust 1655 * the offset supplied by userland. 1656 */ 1657 while ((error = ffs_blkatoff_ra(vp, uio->uio_offset, NULL, &bp, 2)) == 0) { 1658 pickup = (int)(uio->uio_offset - bp->b_loffset); 1659 offset = 0; 1660 retval = 0; 1661 if (bp->b_loffset + bp->b_bcount > VTOI(vp)->i_size) 1662 eoffset = (int)(VTOI(vp)->i_size - bp->b_loffset); 1663 else 1664 eoffset = bp->b_bcount; 1665 1666 while (offset < eoffset) { 1667 dp = (struct direct *)(bp->b_data + offset); 1668 if (dp->d_reclen <= 0 || (dp->d_reclen & 3) || 1669 offset + dp->d_reclen > bp->b_bcount) { 1670 error = EIO; 1671 break; 1672 } 1673 if (offsetof(struct direct, d_name[dp->d_namlen]) > dp->d_reclen) { 1674 error = EIO; 1675 break; 1676 } 1677 if (offset < pickup) { 1678 offset += dp->d_reclen; 1679 continue; 1680 } 1681 #if BYTE_ORDER == LITTLE_ENDIAN 1682 if (OFSFMT(vp)) { 1683 retval = vop_write_dirent(&error, uio, 1684 dp->d_ino, dp->d_namlen, dp->d_type, 1685 dp->d_name); 1686 } else 1687 #endif 1688 { 1689 retval = vop_write_dirent(&error, uio, 1690 dp->d_ino, dp->d_type, dp->d_namlen, 1691 dp->d_name); 1692 } 1693 if (retval) 1694 break; 1695 if (cookies) 1696 cookies[cookie_index] = bp->b_loffset + offset; 1697 ++cookie_index; 1698 offset += dp->d_reclen; 1699 if (cookie_index == ncookies) 1700 break; 1701 } 1702 1703 /* 1704 * This will align the next loop to the beginning of the 1705 * next block, and pickup will calculate to 0. 1706 */ 1707 uio->uio_offset = bp->b_loffset + offset; 1708 brelse(bp); 1709 1710 if (retval || error || cookie_index == ncookies || 1711 uio->uio_offset >= VTOI(vp)->i_size) { 1712 break; 1713 } 1714 } 1715 if (ap->a_eofflag) 1716 *ap->a_eofflag = VTOI(vp)->i_size <= uio->uio_offset; 1717 1718 /* 1719 * Report errors only if we didn't manage to read anything 1720 */ 1721 if (error && cookie_index == 0) { 1722 if (cookies) { 1723 kfree(cookies, M_TEMP); 1724 *ap->a_ncookies = 0; 1725 *ap->a_cookies = NULL; 1726 } 1727 } else { 1728 error = 0; 1729 if (cookies) { 1730 *ap->a_ncookies = cookie_index; 1731 *ap->a_cookies = cookies; 1732 } 1733 } 1734 done: 1735 vn_unlock(vp); 1736 return (error); 1737 } 1738 1739 /* 1740 * Return target name of a symbolic link 1741 * 1742 * ufs_readlink(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred) 1743 */ 1744 static 1745 int 1746 ufs_readlink(struct vop_readlink_args *ap) 1747 { 1748 struct vnode *vp = ap->a_vp; 1749 struct inode *ip = VTOI(vp); 1750 int isize; 1751 1752 isize = ip->i_size; 1753 if ((isize < vp->v_mount->mnt_maxsymlinklen) || 1754 (ip->i_din.di_blocks == 0)) { /* XXX - for old fastlink support */ 1755 uiomove((char *)ip->i_shortlink, isize, ap->a_uio); 1756 return (0); 1757 } 1758 1759 /* 1760 * Perform the equivalent of an OPEN on vp so we can issue a 1761 * VOP_READ. 1762 */ 1763 return (VOP_READ(vp, ap->a_uio, 0, ap->a_cred)); 1764 } 1765 1766 /* 1767 * Calculate the logical to physical mapping if not done already, 1768 * then call the device strategy routine. 1769 * 1770 * In order to be able to swap to a file, the VOP_BMAP operation may not 1771 * deadlock on memory. See ufs_bmap() for details. 1772 * 1773 * ufs_strategy(struct vnode *a_vp, struct bio *a_bio) 1774 */ 1775 static 1776 int 1777 ufs_strategy(struct vop_strategy_args *ap) 1778 { 1779 struct bio *bio = ap->a_bio; 1780 struct bio *nbio; 1781 struct buf *bp = bio->bio_buf; 1782 struct vnode *vp = ap->a_vp; 1783 struct inode *ip; 1784 int error; 1785 1786 ip = VTOI(vp); 1787 if (vp->v_type == VBLK || vp->v_type == VCHR) 1788 panic("ufs_strategy: spec"); 1789 nbio = push_bio(bio); 1790 if (nbio->bio_offset == NOOFFSET) { 1791 error = VOP_BMAP(vp, bio->bio_offset, &nbio->bio_offset, 1792 NULL, NULL, bp->b_cmd); 1793 if (error) { 1794 bp->b_error = error; 1795 bp->b_flags |= B_ERROR; 1796 /* I/O was never started on nbio, must biodone(bio) */ 1797 biodone(bio); 1798 return (error); 1799 } 1800 if (nbio->bio_offset == NOOFFSET) 1801 vfs_bio_clrbuf(bp); 1802 } 1803 if (nbio->bio_offset == NOOFFSET) { 1804 /* 1805 * We hit a hole in the file. The buffer has been zero-filled 1806 * so just biodone() it. 1807 */ 1808 biodone(bio); 1809 } else { 1810 vn_strategy(ip->i_devvp, nbio); 1811 } 1812 return (0); 1813 } 1814 1815 /* 1816 * Print out the contents of an inode. 1817 * 1818 * ufs_print(struct vnode *a_vp) 1819 */ 1820 static 1821 int 1822 ufs_print(struct vop_print_args *ap) 1823 { 1824 struct vnode *vp = ap->a_vp; 1825 struct inode *ip = VTOI(vp); 1826 1827 kprintf("tag VT_UFS, ino %lu, on dev %s (%d, %d)", 1828 (u_long)ip->i_number, devtoname(ip->i_dev), major(ip->i_dev), 1829 minor(ip->i_dev)); 1830 if (vp->v_type == VFIFO) 1831 fifo_printinfo(vp); 1832 lockmgr_printinfo(&vp->v_lock); 1833 kprintf("\n"); 1834 return (0); 1835 } 1836 1837 /* 1838 * Read wrapper for fifos. 1839 * 1840 * ufsfifo_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 1841 * struct ucred *a_cred) 1842 */ 1843 static 1844 int 1845 ufsfifo_read(struct vop_read_args *ap) 1846 { 1847 int error, resid; 1848 struct inode *ip; 1849 struct uio *uio; 1850 1851 uio = ap->a_uio; 1852 resid = uio->uio_resid; 1853 error = VOCALL(&fifo_vnode_vops, &ap->a_head); 1854 ip = VTOI(ap->a_vp); 1855 if ((ap->a_vp->v_mount->mnt_flag & MNT_NOATIME) == 0 && ip != NULL && 1856 (uio->uio_resid != resid || (error == 0 && resid != 0))) 1857 VTOI(ap->a_vp)->i_flag |= IN_ACCESS; 1858 return (error); 1859 } 1860 1861 /* 1862 * Write wrapper for fifos. 1863 * 1864 * ufsfifo_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 1865 * struct ucred *a_cred) 1866 */ 1867 static 1868 int 1869 ufsfifo_write(struct vop_write_args *ap) 1870 { 1871 int error, resid; 1872 struct inode *ip; 1873 struct uio *uio; 1874 1875 uio = ap->a_uio; 1876 resid = uio->uio_resid; 1877 error = VOCALL(&fifo_vnode_vops, &ap->a_head); 1878 ip = VTOI(ap->a_vp); 1879 if (ip != NULL && (uio->uio_resid != resid || (error == 0 && resid != 0))) 1880 VTOI(ap->a_vp)->i_flag |= IN_CHANGE | IN_UPDATE; 1881 return (error); 1882 } 1883 1884 /* 1885 * Close wrapper for fifos. 1886 * 1887 * Update the times on the inode then do device close. 1888 * 1889 * ufsfifo_close(struct vnode *a_vp, int a_fflag) 1890 */ 1891 static 1892 int 1893 ufsfifo_close(struct vop_close_args *ap) 1894 { 1895 struct vnode *vp = ap->a_vp; 1896 1897 if (VREFCNT(vp) > 1) 1898 ufs_itimes(vp); 1899 return (VOCALL(&fifo_vnode_vops, &ap->a_head)); 1900 } 1901 1902 /* 1903 * Kqfilter wrapper for fifos. 1904 * 1905 * Fall through to ufs kqfilter routines if needed 1906 */ 1907 static 1908 int 1909 ufsfifo_kqfilter(struct vop_kqfilter_args *ap) 1910 { 1911 int error; 1912 1913 error = VOCALL(&fifo_vnode_vops, &ap->a_head); 1914 if (error) 1915 error = ufs_kqfilter(ap); 1916 return (error); 1917 } 1918 1919 /* 1920 * Advisory record locking support 1921 * 1922 * ufs_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, struct flock *a_fl, 1923 * int a_flags) 1924 */ 1925 static 1926 int 1927 ufs_advlock(struct vop_advlock_args *ap) 1928 { 1929 struct inode *ip = VTOI(ap->a_vp); 1930 1931 return (lf_advlock(ap, &(ip->i_lockf), ip->i_size)); 1932 } 1933 1934 /* 1935 * Initialize the vnode associated with a new inode, handle aliased 1936 * vnodes. 1937 * 1938 * Make sure directories have their VM object now rather then later, 1939 * saving us from having to check on all the myrid directory VOPs 1940 * that might be executed without a VOP_OPEN being performed. 1941 */ 1942 int 1943 ufs_vinit(struct mount *mntp, struct vnode **vpp) 1944 { 1945 struct inode *ip; 1946 struct vnode *vp; 1947 struct timeval tv; 1948 1949 vp = *vpp; 1950 ip = VTOI(vp); 1951 1952 vp->v_type = IFTOVT(ip->i_mode); 1953 1954 switch(vp->v_type) { 1955 case VCHR: 1956 case VBLK: 1957 vp->v_ops = &mntp->mnt_vn_spec_ops; 1958 addaliasu(vp, umajor(ip->i_rdev), uminor(ip->i_rdev)); 1959 break; 1960 case VFIFO: 1961 vp->v_ops = &mntp->mnt_vn_fifo_ops; 1962 break; 1963 case VDIR: 1964 case VREG: 1965 vinitvmio(vp, ip->i_size, 1966 blkoffsize(ip->i_fs, ip, ip->i_size), 1967 blkoff(ip->i_fs, ip->i_size)); 1968 break; 1969 case VLNK: 1970 if (ip->i_size >= vp->v_mount->mnt_maxsymlinklen) { 1971 vinitvmio(vp, ip->i_size, 1972 blkoffsize(ip->i_fs, ip, ip->i_size), 1973 blkoff(ip->i_fs, ip->i_size)); 1974 } 1975 break; 1976 default: 1977 break; 1978 1979 } 1980 1981 if (ip->i_number == ROOTINO) 1982 vsetflags(vp, VROOT); 1983 /* 1984 * Initialize modrev times 1985 */ 1986 getmicrouptime(&tv); 1987 SETHIGH(ip->i_modrev, tv.tv_sec); 1988 SETLOW(ip->i_modrev, tv.tv_usec * 4294); 1989 *vpp = vp; 1990 return (0); 1991 } 1992 1993 /* 1994 * Allocate a new inode. 1995 */ 1996 static 1997 int 1998 ufs_makeinode(int mode, struct vnode *dvp, struct vnode **vpp, 1999 struct componentname *cnp) 2000 { 2001 struct inode *ip, *pdir; 2002 struct direct newdir; 2003 struct vnode *tvp; 2004 int error; 2005 2006 pdir = VTOI(dvp); 2007 *vpp = NULL; 2008 if ((mode & IFMT) == 0) 2009 mode |= IFREG; 2010 2011 error = ffs_valloc(dvp, mode, cnp->cn_cred, &tvp); 2012 if (error) 2013 return (error); 2014 ip = VTOI(tvp); 2015 ip->i_flags = pdir->i_flags & (SF_NOHISTORY|UF_NOHISTORY|UF_NODUMP); 2016 ip->i_gid = pdir->i_gid; 2017 #ifdef SUIDDIR 2018 { 2019 #ifdef QUOTA 2020 struct ucred ucred, *ucp; 2021 ucp = cnp->cn_cred; 2022 #endif 2023 /* 2024 * If we are not the owner of the directory, 2025 * and we are hacking owners here, (only do this where told to) 2026 * and we are not giving it TO root, (would subvert quotas) 2027 * then go ahead and give it to the other user. 2028 * Note that this drops off the execute bits for security. 2029 */ 2030 if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) && 2031 (pdir->i_mode & ISUID) && 2032 (pdir->i_uid != cnp->cn_cred->cr_uid) && pdir->i_uid) { 2033 ip->i_uid = pdir->i_uid; 2034 mode &= ~07111; 2035 #ifdef QUOTA 2036 /* 2037 * Make sure the correct user gets charged 2038 * for the space. 2039 * Quickly knock up a dummy credential for the victim. 2040 * XXX This seems to never be accessed out of our 2041 * context so a stack variable is ok. 2042 */ 2043 ucred.cr_ref = 1; 2044 ucred.cr_uid = ip->i_uid; 2045 ucred.cr_ngroups = 1; 2046 ucred.cr_groups[0] = pdir->i_gid; 2047 ucp = &ucred; 2048 #endif 2049 } else 2050 ip->i_uid = cnp->cn_cred->cr_uid; 2051 2052 #ifdef QUOTA 2053 if ((error = ufs_getinoquota(ip)) || 2054 (error = ufs_chkiq(ip, 1, ucp, 0))) { 2055 ffs_vfree(tvp, ip->i_number, mode); 2056 vput(tvp); 2057 return (error); 2058 } 2059 #endif 2060 } 2061 #else /* !SUIDDIR */ 2062 ip->i_uid = cnp->cn_cred->cr_uid; 2063 #ifdef QUOTA 2064 if ((error = ufs_getinoquota(ip)) || 2065 (error = ufs_chkiq(ip, 1, cnp->cn_cred, 0))) { 2066 ffs_vfree(tvp, ip->i_number, mode); 2067 vput(tvp); 2068 return (error); 2069 } 2070 #endif 2071 #endif /* !SUIDDIR */ 2072 ip->i_din.di_spare[0] = 0; 2073 ip->i_din.di_spare[1] = 0; 2074 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 2075 ip->i_mode = mode; 2076 tvp->v_type = IFTOVT(mode); /* Rest init'd in getnewvnode(). */ 2077 ip->i_effnlink = 1; 2078 ip->i_nlink = 1; 2079 if (DOINGSOFTDEP(tvp)) 2080 softdep_change_linkcnt(ip); 2081 if ((ip->i_mode & ISGID) && !groupmember(ip->i_gid, cnp->cn_cred) && 2082 priv_check_cred(cnp->cn_cred, PRIV_VFS_SETGID, 0)) { 2083 ip->i_mode &= ~ISGID; 2084 } 2085 2086 if (cnp->cn_flags & CNP_ISWHITEOUT) 2087 ip->i_flags |= UF_OPAQUE; 2088 2089 /* 2090 * Regular files and directories need VM objects. Softlinks do 2091 * not (not immediately anyway). 2092 */ 2093 if (tvp->v_type == VREG || tvp->v_type == VDIR) 2094 vinitvmio(tvp, 0, PAGE_SIZE, -1); 2095 2096 /* 2097 * Make sure inode goes to disk before directory entry. 2098 */ 2099 error = ffs_update(tvp, !(DOINGSOFTDEP(tvp) | DOINGASYNC(tvp))); 2100 if (error) 2101 goto bad; 2102 ufs_makedirentry(ip, cnp, &newdir); 2103 error = ufs_direnter(dvp, tvp, &newdir, cnp, NULL); 2104 if (error) 2105 goto bad; 2106 *vpp = tvp; 2107 return (0); 2108 2109 bad: 2110 /* 2111 * Write error occurred trying to update the inode 2112 * or the directory so must deallocate the inode. 2113 */ 2114 ip->i_effnlink = 0; 2115 ip->i_nlink = 0; 2116 ip->i_flag |= IN_CHANGE; 2117 if (DOINGSOFTDEP(tvp)) 2118 softdep_change_linkcnt(ip); 2119 vput(tvp); 2120 return (error); 2121 } 2122 2123 static int 2124 ufs_missingop(struct vop_generic_args *ap) 2125 { 2126 panic("no vop function for %s in ufs child", ap->a_desc->sd_name); 2127 return (EOPNOTSUPP); 2128 } 2129 2130 static struct filterops ufsread_filtops = 2131 { FILTEROP_ISFD, NULL, filt_ufsdetach, filt_ufsread }; 2132 static struct filterops ufswrite_filtops = 2133 { FILTEROP_ISFD, NULL, filt_ufsdetach, filt_ufswrite }; 2134 static struct filterops ufsvnode_filtops = 2135 { FILTEROP_ISFD, NULL, filt_ufsdetach, filt_ufsvnode }; 2136 2137 /* 2138 * ufs_kqfilter(struct vnode *a_vp, struct knote *a_kn) 2139 */ 2140 static int 2141 ufs_kqfilter(struct vop_kqfilter_args *ap) 2142 { 2143 struct vnode *vp = ap->a_vp; 2144 struct knote *kn = ap->a_kn; 2145 2146 switch (kn->kn_filter) { 2147 case EVFILT_READ: 2148 kn->kn_fop = &ufsread_filtops; 2149 break; 2150 case EVFILT_WRITE: 2151 kn->kn_fop = &ufswrite_filtops; 2152 break; 2153 case EVFILT_VNODE: 2154 kn->kn_fop = &ufsvnode_filtops; 2155 break; 2156 default: 2157 return (EOPNOTSUPP); 2158 } 2159 2160 kn->kn_hook = (caddr_t)vp; 2161 2162 /* XXX: kq token actually protects the list */ 2163 lwkt_gettoken(&vp->v_token); 2164 knote_insert(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn); 2165 lwkt_reltoken(&vp->v_token); 2166 2167 return (0); 2168 } 2169 2170 static void 2171 filt_ufsdetach(struct knote *kn) 2172 { 2173 struct vnode *vp = (struct vnode *)kn->kn_hook; 2174 2175 lwkt_gettoken(&vp->v_token); 2176 knote_remove(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn); 2177 lwkt_reltoken(&vp->v_token); 2178 } 2179 2180 /*ARGSUSED*/ 2181 static int 2182 filt_ufsread(struct knote *kn, long hint) 2183 { 2184 struct vnode *vp = (struct vnode *)kn->kn_hook; 2185 struct inode *ip = VTOI(vp); 2186 off_t off; 2187 2188 /* 2189 * filesystem is gone, so set the EOF flag and schedule 2190 * the knote for deletion. 2191 */ 2192 if (hint == NOTE_REVOKE) { 2193 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT); 2194 return (1); 2195 } 2196 2197 off = ip->i_size - kn->kn_fp->f_offset; 2198 kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX; 2199 if (kn->kn_sfflags & NOTE_OLDAPI) 2200 return(1); 2201 return (kn->kn_data != 0); 2202 } 2203 2204 /*ARGSUSED*/ 2205 static int 2206 filt_ufswrite(struct knote *kn, long hint) 2207 { 2208 /* 2209 * filesystem is gone, so set the EOF flag and schedule 2210 * the knote for deletion. 2211 */ 2212 if (hint == NOTE_REVOKE) 2213 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT); 2214 2215 kn->kn_data = 0; 2216 return (1); 2217 } 2218 2219 static int 2220 filt_ufsvnode(struct knote *kn, long hint) 2221 { 2222 if (kn->kn_sfflags & hint) 2223 kn->kn_fflags |= hint; 2224 if (hint == NOTE_REVOKE) { 2225 kn->kn_flags |= (EV_EOF | EV_NODATA); 2226 return (1); 2227 } 2228 return (kn->kn_fflags != 0); 2229 } 2230 2231 /* Global vfs data structures for ufs. */ 2232 static struct vop_ops ufs_vnode_vops = { 2233 .vop_default = vop_defaultop, 2234 .vop_fsync = (void *)ufs_missingop, 2235 .vop_read = (void *)ufs_missingop, 2236 .vop_reallocblks = (void *)ufs_missingop, 2237 .vop_write = (void *)ufs_missingop, 2238 .vop_access = ufs_access, 2239 .vop_advlock = ufs_advlock, 2240 .vop_bmap = ufs_bmap, 2241 .vop_old_lookup = ufs_lookup, 2242 .vop_close = ufs_close, 2243 .vop_old_create = ufs_create, 2244 .vop_getattr = ufs_getattr, 2245 .vop_inactive = ufs_inactive, 2246 .vop_old_link = ufs_link, 2247 .vop_old_mkdir = ufs_mkdir, 2248 .vop_old_mknod = ufs_mknod, 2249 .vop_mmap = ufs_mmap, 2250 .vop_open = vop_stdopen, 2251 .vop_pathconf = vop_stdpathconf, 2252 .vop_kqfilter = ufs_kqfilter, 2253 .vop_print = ufs_print, 2254 .vop_readdir = ufs_readdir, 2255 .vop_readlink = ufs_readlink, 2256 .vop_reclaim = ufs_reclaim, 2257 .vop_old_remove = ufs_remove, 2258 .vop_old_rename = ufs_rename, 2259 .vop_old_rmdir = ufs_rmdir, 2260 .vop_setattr = ufs_setattr, 2261 .vop_markatime = ufs_markatime, 2262 .vop_strategy = ufs_strategy, 2263 .vop_old_symlink = ufs_symlink, 2264 .vop_old_whiteout = ufs_whiteout 2265 }; 2266 2267 static struct vop_ops ufs_spec_vops = { 2268 .vop_default = vop_defaultop, 2269 .vop_fsync = (void *)ufs_missingop, 2270 .vop_access = ufs_access, 2271 .vop_close = ufs_close, 2272 .vop_getattr = ufs_getattr, 2273 .vop_inactive = ufs_inactive, 2274 .vop_print = ufs_print, 2275 .vop_read = vop_stdnoread, 2276 .vop_reclaim = ufs_reclaim, 2277 .vop_setattr = ufs_setattr, 2278 .vop_markatime = ufs_markatime, 2279 .vop_write = vop_stdnowrite 2280 }; 2281 2282 static struct vop_ops ufs_fifo_vops = { 2283 .vop_default = fifo_vnoperate, 2284 .vop_fsync = (void *)ufs_missingop, 2285 .vop_access = ufs_access, 2286 .vop_close = ufsfifo_close, 2287 .vop_getattr = ufs_getattr, 2288 .vop_inactive = ufs_inactive, 2289 .vop_kqfilter = ufsfifo_kqfilter, 2290 .vop_print = ufs_print, 2291 .vop_read = ufsfifo_read, 2292 .vop_reclaim = ufs_reclaim, 2293 .vop_setattr = ufs_setattr, 2294 .vop_markatime = ufs_markatime, 2295 .vop_write = ufsfifo_write 2296 }; 2297 2298 VNODEOP_SET(ufs_vnode_vops); 2299 VNODEOP_SET(ufs_spec_vops); 2300 VNODEOP_SET(ufs_fifo_vops); 2301 2302 /* 2303 * ufs_vnoperate() 2304 */ 2305 int 2306 ufs_vnoperate(struct vop_generic_args *ap) 2307 { 2308 return (VOCALL(&ufs_vnode_vops, ap)); 2309 } 2310 2311 /* 2312 * ufs_vnoperatefifo() 2313 */ 2314 int 2315 ufs_vnoperatefifo(struct vop_generic_args *ap) 2316 { 2317 return (VOCALL(&ufs_fifo_vops, ap)); 2318 } 2319 2320 /* 2321 * ufs_vnoperatespec() 2322 */ 2323 int 2324 ufs_vnoperatespec(struct vop_generic_args *ap) 2325 { 2326 return (VOCALL(&ufs_spec_vops, ap)); 2327 } 2328