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/uio.h> 45 #include <sys/kernel.h> 46 #include <sys/fcntl.h> 47 #include <sys/stat.h> 48 #include <sys/buf.h> 49 #include <sys/proc.h> 50 #include <sys/priv.h> 51 #include <sys/namei.h> 52 #include <sys/mount.h> 53 #include <sys/unistd.h> 54 #include <sys/vnode.h> 55 #include <sys/malloc.h> 56 #include <sys/dirent.h> 57 #include <sys/lockf.h> 58 #include <sys/event.h> 59 #include <sys/conf.h> 60 61 #include <sys/file.h> /* XXX */ 62 #include <sys/jail.h> 63 64 #include <vm/vm.h> 65 #include <vm/vm_extern.h> 66 67 #include <vfs/fifofs/fifo.h> 68 69 #include "quota.h" 70 #include "inode.h" 71 #include "dir.h" 72 #include "ufsmount.h" 73 #include "ufs_extern.h" 74 #include "ffs_extern.h" 75 #include "fs.h" 76 #ifdef UFS_DIRHASH 77 #include "dirhash.h" 78 #endif 79 80 static int ufs_access (struct vop_access_args *); 81 static int ufs_advlock (struct vop_advlock_args *); 82 static int ufs_chmod (struct vnode *, int, struct ucred *); 83 static int ufs_chown (struct vnode *, uid_t, gid_t, struct ucred *); 84 static int ufs_close (struct vop_close_args *); 85 static int ufs_create (struct vop_old_create_args *); 86 static int ufs_getattr (struct vop_getattr_args *); 87 static int ufs_link (struct vop_old_link_args *); 88 static int ufs_makeinode (int mode, struct vnode *, struct vnode **, struct componentname *); 89 static int ufs_markatime (struct vop_markatime_args *); 90 static int ufs_missingop (struct vop_generic_args *ap); 91 static int ufs_mkdir (struct vop_old_mkdir_args *); 92 static int ufs_mknod (struct vop_old_mknod_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 #define VN_KNOTE(vp, b) \ 114 KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, (b)) 115 116 #define OFSFMT(vp) ((vp)->v_mount->mnt_maxsymlinklen <= 0) 117 118 /* 119 * A virgin directory (no blushing please). 120 */ 121 static struct dirtemplate mastertemplate = { 122 0, 12, DT_DIR, 1, ".", 123 0, DIRBLKSIZ - 12, DT_DIR, 2, ".." 124 }; 125 static struct odirtemplate omastertemplate = { 126 0, 12, 1, ".", 127 0, DIRBLKSIZ - 12, 2, ".." 128 }; 129 130 void 131 ufs_itimes(struct vnode *vp) 132 { 133 struct inode *ip; 134 struct timespec ts; 135 136 ip = VTOI(vp); 137 if ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE)) == 0) 138 return; 139 if ((vp->v_type == VBLK || vp->v_type == VCHR) && !DOINGSOFTDEP(vp)) 140 ip->i_flag |= IN_LAZYMOD; 141 else 142 ip->i_flag |= IN_MODIFIED; 143 144 if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 145 vfs_timestamp(&ts); 146 if (ip->i_flag & IN_ACCESS) { 147 ip->i_atime = (uint32_t)ts.tv_sec; 148 ip->i_atime_ext = ts.tv_sec >> 32; 149 ip->i_atimensec = ts.tv_nsec; 150 } 151 if (ip->i_flag & IN_CHANGE) { 152 ip->i_ctime = (uint32_t)ts.tv_sec; 153 ip->i_ctime_ext = ts.tv_sec >> 32; 154 ip->i_ctimensec = ts.tv_nsec; 155 } 156 if (ip->i_flag & IN_UPDATE) { 157 if (ip->i_flag & IN_NOCOPYWRITE) { 158 if (vp->v_flag & VLASTWRITETS) { 159 ip->i_mtime = (uint32_t) 160 vp->v_lastwrite_ts.tv_sec; 161 ip->i_mtime_ext = 162 vp->v_lastwrite_ts.tv_sec >> 32; 163 ip->i_mtimensec = 164 vp->v_lastwrite_ts.tv_nsec; 165 } 166 } else { 167 ip->i_mtime = (uint32_t)ts.tv_sec; 168 ip->i_mtime_ext = ts.tv_sec >> 32; 169 ip->i_mtimensec = ts.tv_nsec; 170 } 171 ip->i_modrev++; 172 } 173 } 174 ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_UPDATE); 175 } 176 177 /* 178 * Create a regular file 179 * 180 * ufs_create(struct vnode *a_dvp, struct vnode **a_vpp, 181 * struct componentname *a_cnp, struct vattr *a_vap) 182 */ 183 static 184 int 185 ufs_create(struct vop_old_create_args *ap) 186 { 187 int error; 188 189 error = 190 ufs_makeinode(MAKEIMODE(ap->a_vap->va_type, ap->a_vap->va_mode), 191 ap->a_dvp, ap->a_vpp, ap->a_cnp); 192 if (error) 193 return (error); 194 VN_KNOTE(ap->a_dvp, NOTE_WRITE); 195 return (0); 196 } 197 198 /* 199 * Mknod vnode call 200 * 201 * ufs_mknod(struct vnode *a_dvp, struct vnode **a_vpp, 202 * struct componentname *a_cnp, struct vattr *a_vap) 203 */ 204 /* ARGSUSED */ 205 static 206 int 207 ufs_mknod(struct vop_old_mknod_args *ap) 208 { 209 struct vattr *vap = ap->a_vap; 210 struct vnode **vpp = ap->a_vpp; 211 struct inode *ip; 212 ino_t ino; 213 int error; 214 215 /* 216 * UFS cannot represent the entire major/minor range supported by 217 * the kernel. 218 */ 219 if (vap->va_rmajor != VNOVAL && 220 makeudev(vap->va_rmajor, vap->va_rminor) == NOUDEV) { 221 return(EINVAL); 222 } 223 224 /* no special directory support */ 225 if (vap->va_type == VDIR) 226 return(EINVAL); 227 228 error = ufs_makeinode(MAKEIMODE(vap->va_type, vap->va_mode), 229 ap->a_dvp, vpp, ap->a_cnp); 230 if (error) 231 return (error); 232 VN_KNOTE(ap->a_dvp, NOTE_WRITE); 233 ip = VTOI(*vpp); 234 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 235 if (vap->va_rmajor != VNOVAL) { 236 /* 237 * Want to be able to use this to make badblock 238 * inodes, so don't truncate the dev number. 239 */ 240 ip->i_rdev = makeudev(vap->va_rmajor, vap->va_rminor); 241 } 242 /* 243 * Remove inode, then reload it through VFS_VGET so it is 244 * checked to see if it is an alias of an existing entry in 245 * the inode cache. 246 */ 247 (*vpp)->v_type = VNON; 248 ino = ip->i_number; /* Save this before vgone() invalidates ip. */ 249 vgone_vxlocked(*vpp); 250 vput(*vpp); 251 error = VFS_VGET(ap->a_dvp->v_mount, NULL, ino, vpp); 252 if (error) { 253 *vpp = NULL; 254 return (error); 255 } 256 return (0); 257 } 258 259 /* 260 * Close called. 261 * 262 * Update the times on the inode. 263 * 264 * ufs_close(struct vnode *a_vp, int a_fflag) 265 */ 266 /* ARGSUSED */ 267 static 268 int 269 ufs_close(struct vop_close_args *ap) 270 { 271 struct vnode *vp = ap->a_vp; 272 273 if (VREFCNT(vp) > 1) 274 ufs_itimes(vp); 275 return (vop_stdclose(ap)); 276 } 277 278 /* 279 * ufs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred) 280 */ 281 static 282 int 283 ufs_access(struct vop_access_args *ap) 284 { 285 struct vnode *vp = ap->a_vp; 286 struct inode *ip = VTOI(vp); 287 int error; 288 289 #ifdef QUOTA 290 if (ap->a_mode & VWRITE) { 291 switch (vp->v_type) { 292 case VDIR: 293 case VLNK: 294 case VREG: 295 if ((error = ufs_getinoquota(ip)) != 0) 296 return (error); 297 break; 298 default: 299 break; 300 } 301 } 302 #endif 303 304 error = vop_helper_access(ap, ip->i_uid, ip->i_gid, ip->i_mode, 0); 305 return (error); 306 } 307 308 /* 309 * ufs_getattr(struct vnode *a_vp, struct vattr *a_vap) 310 */ 311 /* ARGSUSED */ 312 static 313 int 314 ufs_getattr(struct vop_getattr_args *ap) 315 { 316 struct vnode *vp = ap->a_vp; 317 struct inode *ip = VTOI(vp); 318 struct vattr *vap = ap->a_vap; 319 320 ufs_itimes(vp); 321 /* 322 * Copy from inode table 323 */ 324 vap->va_fsid = devid_from_dev(ip->i_dev); 325 vap->va_fileid = ip->i_number; 326 vap->va_mode = ip->i_mode & ~IFMT; 327 vap->va_nlink = VFSTOUFS(vp->v_mount)->um_i_effnlink_valid ? 328 ip->i_effnlink : ip->i_nlink; 329 vap->va_uid = ip->i_uid; 330 vap->va_gid = ip->i_gid; 331 vap->va_rmajor = umajor(ip->i_rdev); 332 vap->va_rminor = uminor(ip->i_rdev); 333 vap->va_size = ip->i_din.di_size; 334 vap->va_atime.tv_sec = 335 (time_t)(ip->i_atime | ((uint64_t)ip->i_atime_ext << 32)); 336 vap->va_atime.tv_nsec = ip->i_atimensec; 337 vap->va_mtime.tv_sec = 338 (time_t)(ip->i_mtime | ((uint64_t)ip->i_mtime_ext << 32)); 339 vap->va_mtime.tv_nsec = ip->i_mtimensec; 340 vap->va_ctime.tv_sec = 341 (time_t)(ip->i_ctime | ((uint64_t)ip->i_ctime_ext << 32)); 342 vap->va_ctime.tv_nsec = ip->i_ctimensec; 343 vap->va_flags = ip->i_flags; 344 vap->va_gen = ip->i_gen; 345 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize; 346 vap->va_bytes = dbtob((u_quad_t)ip->i_blocks); 347 vap->va_type = IFTOVT(ip->i_mode); 348 vap->va_filerev = ip->i_modrev; 349 return (0); 350 } 351 352 static 353 int 354 ufs_markatime(struct vop_markatime_args *ap) 355 { 356 struct vnode *vp = ap->a_vp; 357 struct inode *ip = VTOI(vp); 358 359 if (vp->v_mount->mnt_flag & MNT_RDONLY) 360 return (EROFS); 361 if (vp->v_mount->mnt_flag & MNT_NOATIME) 362 return (0); 363 ip->i_flag |= IN_ACCESS; 364 VN_KNOTE(vp, NOTE_ATTRIB); 365 return (0); 366 } 367 368 /* 369 * Set attribute vnode op. called from several syscalls 370 * 371 * ufs_setattr(struct vnode *a_vp, struct vattr *a_vap, 372 * struct ucred *a_cred) 373 */ 374 static 375 int 376 ufs_setattr(struct vop_setattr_args *ap) 377 { 378 struct vattr *vap = ap->a_vap; 379 struct vnode *vp = ap->a_vp; 380 struct inode *ip = VTOI(vp); 381 struct ucred *cred = ap->a_cred; 382 int error; 383 384 /* 385 * Check for unsettable attributes. 386 */ 387 if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) || 388 (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) || 389 (vap->va_blocksize != VNOVAL) || (vap->va_rmajor != VNOVAL) || 390 ((int)vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) { 391 return (EINVAL); 392 } 393 if (vap->va_flags != VNOVAL) { 394 if (vp->v_mount->mnt_flag & MNT_RDONLY) 395 return (EROFS); 396 if (cred->cr_uid != ip->i_uid && 397 (error = priv_check_cred(cred, PRIV_VFS_SETATTR, 0))) 398 return (error); 399 /* 400 * Note that a root chflags becomes a user chflags when 401 * we are jailed, unless the jail vfs_chflags sysctl 402 * is set. 403 */ 404 if (cred->cr_uid == 0 && 405 (!jailed(cred) || 406 PRISON_CAP_ISSET(cred->cr_prison->pr_caps, 407 PRISON_CAP_VFS_CHFLAGS))) { 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 = (uint32_t)vap->va_atime.tv_sec; 472 ip->i_atime_ext = vap->va_atime.tv_sec >> 32; 473 ip->i_atimensec = vap->va_atime.tv_nsec; 474 } 475 if (vap->va_mtime.tv_sec != VNOVAL) { 476 ip->i_mtime = (uint32_t)vap->va_mtime.tv_sec; 477 ip->i_mtime_ext = vap->va_mtime.tv_sec >> 32; 478 ip->i_mtimensec = vap->va_mtime.tv_nsec; 479 vclrflags(vp, VLASTWRITETS); 480 } 481 error = ffs_update(vp, 0); 482 if (error) 483 return (error); 484 } 485 error = 0; 486 if (vap->va_mode != (mode_t)VNOVAL) { 487 if (vp->v_mount->mnt_flag & MNT_RDONLY) 488 return (EROFS); 489 error = ufs_chmod(vp, (int)vap->va_mode, cred); 490 } 491 VN_KNOTE(vp, NOTE_ATTRIB); 492 return (error); 493 } 494 495 /* 496 * Change the mode on a file. 497 * Inode must be locked before calling. 498 */ 499 static int 500 ufs_chmod(struct vnode *vp, int mode, struct ucred *cred) 501 { 502 struct inode *ip = VTOI(vp); 503 int error; 504 mode_t cur_mode = ip->i_mode; 505 506 error = vop_helper_chmod(vp, mode, cred, ip->i_uid, ip->i_gid, 507 &cur_mode); 508 if (error) 509 return (error); 510 #if 0 511 if (cred->cr_uid != ip->i_uid) { 512 error = priv_check_cred(cred, PRIV_VFS_CHMOD, 0); 513 if (error) 514 return (error); 515 } 516 if (cred->cr_uid) { 517 if (vp->v_type != VDIR && (mode & S_ISTXT)) 518 return (EFTYPE); 519 if (!groupmember(ip->i_gid, cred) && (mode & ISGID)) 520 return (EPERM); 521 } 522 #endif 523 ip->i_mode = cur_mode; 524 ip->i_flag |= IN_CHANGE; 525 return (0); 526 } 527 528 /* 529 * Perform chown operation on inode ip; 530 * inode must be locked prior to call. 531 */ 532 static int 533 ufs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred) 534 { 535 struct inode *ip = VTOI(vp); 536 uid_t ouid; 537 gid_t ogid; 538 int error = 0; 539 #ifdef QUOTA 540 int i; 541 long change; 542 #endif 543 544 if (uid == (uid_t)VNOVAL) 545 uid = ip->i_uid; 546 if (gid == (gid_t)VNOVAL) 547 gid = ip->i_gid; 548 /* 549 * If we don't own the file, are trying to change the owner 550 * of the file, or are not a member of the target group, 551 * the caller must be superuser or the call fails. 552 */ 553 if ((cred->cr_uid != ip->i_uid || uid != ip->i_uid || 554 (gid != ip->i_gid && !(cred->cr_gid == gid || 555 groupmember(gid, cred)))) && 556 (error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0))) 557 return (error); 558 ogid = ip->i_gid; 559 ouid = ip->i_uid; 560 #ifdef QUOTA 561 if ((error = ufs_getinoquota(ip)) != 0) 562 return (error); 563 if (ouid == uid) { 564 ufs_dqrele(vp, ip->i_dquot[USRQUOTA]); 565 ip->i_dquot[USRQUOTA] = NODQUOT; 566 } 567 if (ogid == gid) { 568 ufs_dqrele(vp, ip->i_dquot[GRPQUOTA]); 569 ip->i_dquot[GRPQUOTA] = NODQUOT; 570 } 571 change = ip->i_blocks; 572 (void) ufs_chkdq(ip, -change, cred, CHOWN); 573 (void) ufs_chkiq(ip, -1, cred, CHOWN); 574 for (i = 0; i < MAXQUOTAS; i++) { 575 ufs_dqrele(vp, ip->i_dquot[i]); 576 ip->i_dquot[i] = NODQUOT; 577 } 578 #endif 579 ip->i_gid = gid; 580 ip->i_uid = uid; 581 #ifdef QUOTA 582 if ((error = ufs_getinoquota(ip)) == 0) { 583 if (ouid == uid) { 584 ufs_dqrele(vp, ip->i_dquot[USRQUOTA]); 585 ip->i_dquot[USRQUOTA] = NODQUOT; 586 } 587 if (ogid == gid) { 588 ufs_dqrele(vp, ip->i_dquot[GRPQUOTA]); 589 ip->i_dquot[GRPQUOTA] = NODQUOT; 590 } 591 if ((error = ufs_chkdq(ip, change, cred, CHOWN)) == 0) { 592 if ((error = ufs_chkiq(ip, 1, cred, CHOWN)) == 0) 593 goto good; 594 else 595 (void)ufs_chkdq(ip, -change, cred, CHOWN|FORCE); 596 } 597 for (i = 0; i < MAXQUOTAS; i++) { 598 ufs_dqrele(vp, ip->i_dquot[i]); 599 ip->i_dquot[i] = NODQUOT; 600 } 601 } 602 ip->i_gid = ogid; 603 ip->i_uid = ouid; 604 if (ufs_getinoquota(ip) == 0) { 605 if (ouid == uid) { 606 ufs_dqrele(vp, ip->i_dquot[USRQUOTA]); 607 ip->i_dquot[USRQUOTA] = NODQUOT; 608 } 609 if (ogid == gid) { 610 ufs_dqrele(vp, ip->i_dquot[GRPQUOTA]); 611 ip->i_dquot[GRPQUOTA] = NODQUOT; 612 } 613 (void) ufs_chkdq(ip, change, cred, FORCE|CHOWN); 614 (void) ufs_chkiq(ip, 1, cred, FORCE|CHOWN); 615 (void) ufs_getinoquota(ip); 616 } 617 return (error); 618 good: 619 if (ufs_getinoquota(ip)) 620 panic("ufs_chown: lost quota"); 621 #endif /* QUOTA */ 622 ip->i_flag |= IN_CHANGE; 623 if (cred->cr_uid != 0 && (ouid != uid || ogid != gid)) 624 ip->i_mode &= ~(ISUID | ISGID); 625 return (0); 626 } 627 628 /* 629 * ufs_remove(struct vnode *a_dvp, struct vnode *a_vp, 630 * struct componentname *a_cnp) 631 */ 632 static 633 int 634 ufs_remove(struct vop_old_remove_args *ap) 635 { 636 struct inode *ip; 637 struct vnode *vp = ap->a_vp; 638 struct vnode *dvp = ap->a_dvp; 639 int error; 640 641 ip = VTOI(vp); 642 #if 0 /* handled by kernel now */ 643 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) || 644 (VTOI(dvp)->i_flags & APPEND)) { 645 error = EPERM; 646 goto out; 647 } 648 #endif 649 error = ufs_dirremove(dvp, ip, ap->a_cnp->cn_flags, 0); 650 VN_KNOTE(vp, NOTE_DELETE); 651 VN_KNOTE(dvp, NOTE_WRITE); 652 #if 0 653 out: 654 #endif 655 return (error); 656 } 657 658 /* 659 * link vnode call 660 * 661 * ufs_link(struct vnode *a_tdvp, struct vnode *a_vp, 662 * struct componentname *a_cnp) 663 */ 664 static 665 int 666 ufs_link(struct vop_old_link_args *ap) 667 { 668 struct vnode *vp = ap->a_vp; 669 struct vnode *tdvp = ap->a_tdvp; 670 struct componentname *cnp = ap->a_cnp; 671 struct inode *ip; 672 struct direct newdir; 673 int error; 674 675 if (tdvp->v_mount != vp->v_mount) { 676 error = EXDEV; 677 goto out2; 678 } 679 if (tdvp != vp) { 680 error = vn_lock(vp, LK_EXCLUSIVE | LK_FAILRECLAIM); 681 if (error) 682 goto out2; 683 } 684 ip = VTOI(vp); 685 if ((nlink_t)ip->i_nlink >= LINK_MAX) { 686 error = EMLINK; 687 goto out1; 688 } 689 #if 0 /* handled by kernel now, also DragonFly allows this */ 690 if (ip->i_flags & (IMMUTABLE | APPEND)) { 691 error = EPERM; 692 goto out1; 693 } 694 #endif 695 ip->i_effnlink++; 696 ip->i_nlink++; 697 ip->i_flag |= IN_CHANGE; 698 if (DOINGSOFTDEP(vp)) 699 softdep_change_linkcnt(ip); 700 error = ffs_update(vp, !(DOINGSOFTDEP(vp) | DOINGASYNC(vp))); 701 if (!error) { 702 ufs_makedirentry(ip, cnp, &newdir); 703 error = ufs_direnter(tdvp, vp, &newdir, cnp, NULL); 704 } 705 706 if (error) { 707 ip->i_effnlink--; 708 ip->i_nlink--; 709 ip->i_flag |= IN_CHANGE; 710 if (DOINGSOFTDEP(vp)) 711 softdep_change_linkcnt(ip); 712 } 713 out1: 714 if (tdvp != vp) 715 vn_unlock(vp); 716 out2: 717 VN_KNOTE(vp, NOTE_LINK); 718 VN_KNOTE(tdvp, NOTE_WRITE); 719 return (error); 720 } 721 722 /* 723 * whiteout vnode call 724 * 725 * ufs_whiteout(struct vnode *a_dvp, struct componentname *a_cnp, int a_flags) 726 */ 727 static 728 int 729 ufs_whiteout(struct vop_old_whiteout_args *ap) 730 { 731 struct vnode *dvp = ap->a_dvp; 732 struct componentname *cnp = ap->a_cnp; 733 struct direct newdir; 734 int error = 0; 735 736 switch (ap->a_flags) { 737 case NAMEI_LOOKUP: 738 /* 4.4 format directories support whiteout operations */ 739 if (dvp->v_mount->mnt_maxsymlinklen > 0) 740 return (0); 741 return (EOPNOTSUPP); 742 743 case NAMEI_CREATE: 744 /* create a new directory whiteout */ 745 #ifdef DIAGNOSTIC 746 if (dvp->v_mount->mnt_maxsymlinklen <= 0) 747 panic("ufs_whiteout: old format filesystem"); 748 #endif 749 750 newdir.d_ino = UFS_WINO; 751 newdir.d_namlen = cnp->cn_namelen; 752 bcopy(cnp->cn_nameptr, newdir.d_name, (unsigned)cnp->cn_namelen + 1); 753 newdir.d_type = DT_WHT; 754 error = ufs_direnter(dvp, NULL, &newdir, cnp, NULL); 755 break; 756 757 case NAMEI_DELETE: 758 /* remove an existing directory whiteout */ 759 #ifdef DIAGNOSTIC 760 if (dvp->v_mount->mnt_maxsymlinklen <= 0) 761 panic("ufs_whiteout: old format filesystem"); 762 #endif 763 764 cnp->cn_flags &= ~CNP_DOWHITEOUT; 765 error = ufs_dirremove(dvp, NULL, cnp->cn_flags, 0); 766 break; 767 default: 768 panic("ufs_whiteout: unknown op"); 769 } 770 return (error); 771 } 772 773 /* 774 * Rename system call. 775 * rename("foo", "bar"); 776 * is essentially 777 * unlink("bar"); 778 * link("foo", "bar"); 779 * unlink("foo"); 780 * but ``atomically''. Can't do full commit without saving state in the 781 * inode on disk which isn't feasible at this time. Best we can do is 782 * always guarantee the target exists. 783 * 784 * Basic algorithm is: 785 * 786 * 1) Bump link count on source while we're linking it to the 787 * target. This also ensure the inode won't be deleted out 788 * from underneath us while we work (it may be truncated by 789 * a concurrent `trunc' or `open' for creation). 790 * 2) Link source to destination. If destination already exists, 791 * delete it first. 792 * 3) Unlink source reference to inode if still around. If a 793 * directory was moved and the parent of the destination 794 * is different from the source, patch the ".." entry in the 795 * directory. 796 * 797 * ufs_rename(struct vnode *a_fdvp, struct vnode *a_fvp, 798 * struct componentname *a_fcnp, struct vnode *a_tdvp, 799 * struct vnode *a_tvp, struct componentname *a_tcnp) 800 */ 801 static 802 int 803 ufs_rename(struct vop_old_rename_args *ap) 804 { 805 struct vnode *tvp = ap->a_tvp; 806 struct vnode *tdvp = ap->a_tdvp; 807 struct vnode *fvp = ap->a_fvp; 808 struct vnode *fdvp = ap->a_fdvp; 809 struct componentname *tcnp = ap->a_tcnp; 810 struct componentname *fcnp = ap->a_fcnp; 811 struct inode *ip, *xp, *dp; 812 struct direct newdir; 813 ino_t oldparent = 0, newparent = 0; 814 int doingdirectory = 0; 815 int error = 0, ioflag; 816 817 /* 818 * Check for cross-device rename. 819 */ 820 if ((fvp->v_mount != tdvp->v_mount) || 821 (tvp && (fvp->v_mount != tvp->v_mount))) { 822 error = EXDEV; 823 abortit: 824 if (tdvp == tvp) 825 vrele(tdvp); 826 else 827 vput(tdvp); 828 if (tvp) 829 vput(tvp); 830 vrele(fdvp); 831 vrele(fvp); 832 return (error); 833 } 834 835 #if 0 /* handled by kernel now */ 836 if (tvp && ((VTOI(tvp)->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) || 837 (VTOI(tdvp)->i_flags & APPEND))) { 838 error = EPERM; 839 goto abortit; 840 } 841 #endif 842 843 /* 844 * Renaming a file to itself has no effect. The upper layers should 845 * not call us in that case. Temporarily just warn if they do. 846 */ 847 if (fvp == tvp) { 848 kprintf("ufs_rename: fvp == tvp (can't happen)\n"); 849 error = 0; 850 goto abortit; 851 } 852 853 error = vn_lock(fvp, LK_EXCLUSIVE | LK_FAILRECLAIM); 854 if (error) 855 goto abortit; 856 857 /* 858 * Note: now that fvp is locked we have to be sure to unlock it before 859 * using the 'abortit' target. 860 */ 861 dp = VTOI(fdvp); 862 ip = VTOI(fvp); 863 if (ip->i_nlink >= LINK_MAX) { 864 vn_unlock(fvp); 865 error = EMLINK; 866 goto abortit; 867 } 868 #if 0 /* handled by kernel now */ 869 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) 870 || (dp->i_flags & APPEND)) { 871 vn_unlock(fvp); 872 error = EPERM; 873 goto abortit; 874 } 875 #endif 876 if ((ip->i_mode & IFMT) == IFDIR) { 877 /* 878 * Avoid ".", "..", and aliases of "." for obvious reasons. 879 */ 880 if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') || 881 dp == ip || (fcnp->cn_flags | tcnp->cn_flags) & CNP_ISDOTDOT || 882 (ip->i_flag & IN_RENAME)) { 883 vn_unlock(fvp); 884 error = EINVAL; 885 goto abortit; 886 } 887 ip->i_flag |= IN_RENAME; 888 oldparent = dp->i_number; 889 doingdirectory = 1; 890 } 891 VN_KNOTE(fdvp, NOTE_WRITE); /* XXX right place? */ 892 893 /* 894 * fvp still locked. ip->i_flag has IN_RENAME set if doingdirectory. 895 * Cleanup fvp requirements so we can unlock it. 896 * 897 * tvp and tdvp are locked. tvp may be NULL. Now that dp and xp 898 * is setup we can use the 'bad' target if we unlock fvp. We cannot 899 * use the abortit target anymore because of IN_RENAME. 900 */ 901 dp = VTOI(tdvp); 902 if (tvp) 903 xp = VTOI(tvp); 904 else 905 xp = NULL; 906 907 /* 908 * 1) Bump link count while we're moving stuff 909 * around. If we crash somewhere before 910 * completing our work, the link count 911 * may be wrong, but correctable. 912 */ 913 ip->i_effnlink++; 914 ip->i_nlink++; 915 ip->i_flag |= IN_CHANGE; 916 if (DOINGSOFTDEP(fvp)) 917 softdep_change_linkcnt(ip); 918 if ((error = ffs_update(fvp, !(DOINGSOFTDEP(fvp) | 919 DOINGASYNC(fvp)))) != 0) { 920 vn_unlock(fvp); 921 goto bad; 922 } 923 924 /* 925 * If ".." must be changed (ie the directory gets a new 926 * parent) then the source directory must not be in the 927 * directory heirarchy above the target, as this would 928 * orphan everything below the source directory. Also 929 * the user must have write permission in the source so 930 * as to be able to change "..". We must repeat the call 931 * to namei, as the parent directory is unlocked by the 932 * call to checkpath(). 933 */ 934 error = VOP_EACCESS(fvp, VWRITE, tcnp->cn_cred); 935 vn_unlock(fvp); 936 937 /* 938 * We are now back to where we were in that fvp, fdvp are unlocked 939 * and tvp, tdvp are locked. tvp may be NULL. IN_RENAME may be 940 * set. Only the bad target or, if we clean up tvp and tdvp, the 941 * out target, may be used. 942 */ 943 if (oldparent != dp->i_number) 944 newparent = dp->i_number; 945 if (doingdirectory && newparent) { 946 if (error) /* write access check above */ 947 goto bad; 948 949 /* 950 * Once we start messing with tvp and tdvp we cannot use the 951 * 'bad' target, only finish cleaning tdvp and tvp up and 952 * use the 'out' target. 953 * 954 * This cleans up tvp. 955 */ 956 if (xp != NULL) { 957 vput(tvp); 958 xp = NULL; 959 } 960 961 /* 962 * This is a real mess. ufs_checkpath vput's the target 963 * directory so retain an extra ref and note that tdvp will 964 * lose its lock on return. This leaves us with one good 965 * ref after ufs_checkpath returns. 966 */ 967 vref(tdvp); 968 error = ufs_checkpath(ip, dp, tcnp->cn_cred); 969 tcnp->cn_flags |= CNP_PDIRUNLOCK; 970 if (error) { 971 vrele(tdvp); 972 goto out; 973 } 974 975 /* 976 * relookup no longer messes with tdvp's refs. tdvp must be 977 * unlocked on entry and will be locked on a successful 978 * return. 979 */ 980 error = relookup(tdvp, &tvp, tcnp); 981 if (error) { 982 if (tcnp->cn_flags & CNP_PDIRUNLOCK) 983 vrele(tdvp); 984 else 985 vput(tdvp); 986 goto out; 987 } 988 KKASSERT((tcnp->cn_flags & CNP_PDIRUNLOCK) == 0); 989 dp = VTOI(tdvp); 990 if (tvp) 991 xp = VTOI(tvp); 992 } 993 994 /* 995 * We are back to fvp, fdvp unlocked, tvp, tdvp locked. tvp may 996 * be NULL (xp will also be NULL in that case), and IN_RENAME will 997 * be set if doingdirectory. This means we can use the 'bad' target 998 * again. 999 */ 1000 1001 /* 1002 * 2) If target doesn't exist, link the target 1003 * to the source and unlink the source. 1004 * Otherwise, rewrite the target directory 1005 * entry to reference the source inode and 1006 * expunge the original entry's existence. 1007 */ 1008 if (xp == NULL) { 1009 if (dp->i_dev != ip->i_dev) 1010 panic("ufs_rename: EXDEV"); 1011 /* 1012 * Account for ".." in new directory. 1013 * When source and destination have the same 1014 * parent we don't fool with the link count. 1015 */ 1016 if (doingdirectory && newparent) { 1017 if ((nlink_t)dp->i_nlink >= LINK_MAX) { 1018 error = EMLINK; 1019 goto bad; 1020 } 1021 dp->i_effnlink++; 1022 dp->i_nlink++; 1023 dp->i_flag |= IN_CHANGE; 1024 if (DOINGSOFTDEP(tdvp)) 1025 softdep_change_linkcnt(dp); 1026 error = ffs_update(tdvp, !(DOINGSOFTDEP(tdvp) | 1027 DOINGASYNC(tdvp))); 1028 if (error) 1029 goto bad; 1030 } 1031 ufs_makedirentry(ip, tcnp, &newdir); 1032 error = ufs_direnter(tdvp, NULL, &newdir, tcnp, NULL); 1033 if (error) { 1034 if (doingdirectory && newparent) { 1035 dp->i_effnlink--; 1036 dp->i_nlink--; 1037 dp->i_flag |= IN_CHANGE; 1038 if (DOINGSOFTDEP(tdvp)) 1039 softdep_change_linkcnt(dp); 1040 (void)ffs_update(tdvp, 1); 1041 } 1042 goto bad; 1043 } 1044 VN_KNOTE(tdvp, NOTE_WRITE); 1045 vput(tdvp); 1046 } else { 1047 if (xp->i_dev != dp->i_dev || xp->i_dev != ip->i_dev) 1048 panic("ufs_rename: EXDEV"); 1049 /* 1050 * Short circuit rename(foo, foo). 1051 */ 1052 if (xp->i_number == ip->i_number) 1053 panic("ufs_rename: same file"); 1054 /* 1055 * If the parent directory is "sticky", then the user must 1056 * own the parent directory, or the destination of the rename, 1057 * otherwise the destination may not be changed (except by 1058 * root). This implements append-only directories. 1059 */ 1060 if ((dp->i_mode & S_ISTXT) && tcnp->cn_cred->cr_uid != 0 && 1061 tcnp->cn_cred->cr_uid != dp->i_uid && 1062 xp->i_uid != tcnp->cn_cred->cr_uid) { 1063 error = EPERM; 1064 goto bad; 1065 } 1066 /* 1067 * Target must be empty if a directory and have no links 1068 * to it. Also, ensure source and target are compatible 1069 * (both directories, or both not directories). 1070 * 1071 * Purge the file or directory being replaced from the 1072 * nameccache. 1073 */ 1074 if ((xp->i_mode&IFMT) == IFDIR) { 1075 if ((xp->i_effnlink > 2) || 1076 !ufs_dirempty(xp, dp->i_number, tcnp->cn_cred)) { 1077 error = ENOTEMPTY; 1078 goto bad; 1079 } 1080 if (!doingdirectory) { 1081 error = ENOTDIR; 1082 goto bad; 1083 } 1084 /* cache_purge removed - handled by VFS compat layer */ 1085 } else if (doingdirectory == 0) { 1086 /* cache_purge removed - handled by VFS compat layer */ 1087 } else { 1088 error = EISDIR; 1089 goto bad; 1090 } 1091 /* 1092 * note: inode passed to ufs_dirrewrite() is 0 for a 1093 * non-directory file rename, 1 for a directory rename 1094 * in the same directory, and > 1 for an inode representing 1095 * the new directory. 1096 */ 1097 error = ufs_dirrewrite(dp, xp, ip->i_number, 1098 IFTODT(ip->i_mode), 1099 (doingdirectory && newparent) ? 1100 newparent : (ino_t)doingdirectory); 1101 if (error) 1102 goto bad; 1103 if (doingdirectory) { 1104 if (!newparent) { 1105 dp->i_effnlink--; 1106 if (DOINGSOFTDEP(tdvp)) 1107 softdep_change_linkcnt(dp); 1108 } 1109 xp->i_effnlink--; 1110 if (DOINGSOFTDEP(tvp)) 1111 softdep_change_linkcnt(xp); 1112 } 1113 if (doingdirectory && !DOINGSOFTDEP(tvp)) { 1114 /* 1115 * Truncate inode. The only stuff left in the directory 1116 * is "." and "..". The "." reference is inconsequential 1117 * since we are quashing it. We have removed the "." 1118 * reference and the reference in the parent directory, 1119 * but there may be other hard links. The soft 1120 * dependency code will arrange to do these operations 1121 * after the parent directory entry has been deleted on 1122 * disk, so when running with that code we avoid doing 1123 * them now. 1124 */ 1125 if (!newparent) { 1126 dp->i_nlink--; 1127 dp->i_flag |= IN_CHANGE; 1128 } 1129 xp->i_nlink--; 1130 xp->i_flag |= IN_CHANGE; 1131 ioflag = DOINGASYNC(tvp) ? 0 : IO_SYNC; 1132 error = ffs_truncate(tvp, (off_t)0, ioflag, 1133 tcnp->cn_cred); 1134 if (error) 1135 goto bad; 1136 } 1137 VN_KNOTE(tdvp, NOTE_WRITE); 1138 vput(tdvp); 1139 VN_KNOTE(tvp, NOTE_DELETE); 1140 vput(tvp); 1141 xp = NULL; 1142 } 1143 1144 /* 1145 * tvp and tdvp have been cleaned up. only fvp and fdvp (both 1146 * unlocked) remain. We are about to overwrite fvp but we have to 1147 * keep 'ip' intact so we cannot release the old fvp, which is still 1148 * refd and accessible via ap->a_fvp. 1149 * 1150 * This means we cannot use either 'bad' or 'out' to cleanup any 1151 * more. 1152 */ 1153 1154 /* 1155 * 3) Unlink the source. 1156 */ 1157 fcnp->cn_flags &= ~CNP_MODMASK; 1158 fcnp->cn_flags |= CNP_LOCKPARENT; 1159 error = relookup(fdvp, &fvp, fcnp); 1160 if (error || fvp == NULL) { 1161 /* 1162 * From name has disappeared. IN_RENAME will not be set if 1163 * we get past the panic so we don't have to clean it up. 1164 */ 1165 if (doingdirectory) 1166 panic("ufs_rename: lost dir entry"); 1167 vrele(ap->a_fvp); 1168 if (fcnp->cn_flags & CNP_PDIRUNLOCK) 1169 vrele(fdvp); 1170 else 1171 vput(fdvp); 1172 return(0); 1173 } 1174 KKASSERT((fcnp->cn_flags & CNP_PDIRUNLOCK) == 0); 1175 1176 /* 1177 * fdvp and fvp are locked. 1178 */ 1179 xp = VTOI(fvp); 1180 dp = VTOI(fdvp); 1181 1182 /* 1183 * Ensure that the directory entry still exists and has not 1184 * changed while the new name has been entered. If the source is 1185 * a file then the entry may have been unlinked or renamed. In 1186 * either case there is no further work to be done. If the source 1187 * is a directory then it cannot have been rmdir'ed; the IN_RENAME 1188 * flag ensures that it cannot be moved by another rename or removed 1189 * by a rmdir. Cleanup IN_RENAME. 1190 */ 1191 if (xp != ip) { 1192 if (doingdirectory) 1193 panic("ufs_rename: lost dir entry"); 1194 } else { 1195 /* 1196 * If the source is a directory with a 1197 * new parent, the link count of the old 1198 * parent directory must be decremented 1199 * and ".." set to point to the new parent. 1200 */ 1201 if (doingdirectory && newparent) { 1202 xp->i_offset = mastertemplate.dot_reclen; 1203 ufs_dirrewrite(xp, dp, newparent, DT_DIR, 0); 1204 /* cache_purge removed - handled by VFS compat layer */ 1205 } 1206 error = ufs_dirremove(fdvp, xp, fcnp->cn_flags, 0); 1207 xp->i_flag &= ~IN_RENAME; 1208 } 1209 1210 VN_KNOTE(fvp, NOTE_RENAME); 1211 vput(fdvp); 1212 vput(fvp); 1213 vrele(ap->a_fvp); 1214 return (error); 1215 1216 bad: 1217 if (xp) 1218 vput(ITOV(xp)); 1219 vput(ITOV(dp)); 1220 out: 1221 if (doingdirectory) 1222 ip->i_flag &= ~IN_RENAME; 1223 if (vn_lock(fvp, LK_EXCLUSIVE | LK_FAILRECLAIM) == 0) { 1224 ip->i_effnlink--; 1225 ip->i_nlink--; 1226 ip->i_flag |= IN_CHANGE; 1227 ip->i_flag &= ~IN_RENAME; 1228 if (DOINGSOFTDEP(fvp)) 1229 softdep_change_linkcnt(ip); 1230 vput(fvp); 1231 } else { 1232 vrele(fvp); 1233 } 1234 return (error); 1235 } 1236 1237 /* 1238 * Mkdir system call 1239 * 1240 * ufs_mkdir(struct vnode *a_dvp, struct vnode **a_vpp, 1241 * struct componentname *a_cnp, struct vattr *a_vap) 1242 */ 1243 static 1244 int 1245 ufs_mkdir(struct vop_old_mkdir_args *ap) 1246 { 1247 struct vnode *dvp = ap->a_dvp; 1248 struct vattr *vap = ap->a_vap; 1249 struct componentname *cnp = ap->a_cnp; 1250 struct inode *ip, *dp; 1251 struct vnode *tvp; 1252 struct buf *bp; 1253 struct dirtemplate dirtemplate, *dtp; 1254 struct direct newdir; 1255 int error, dmode; 1256 long blkoff; 1257 1258 dp = VTOI(dvp); 1259 if ((nlink_t)dp->i_nlink >= LINK_MAX) { 1260 error = EMLINK; 1261 goto out; 1262 } 1263 dmode = vap->va_mode & 0777; 1264 dmode |= IFDIR; 1265 /* 1266 * Must simulate part of ufs_makeinode here to acquire the inode, 1267 * but not have it entered in the parent directory. The entry is 1268 * made later after writing "." and ".." entries. 1269 */ 1270 error = ffs_valloc(dvp, dmode, cnp->cn_cred, &tvp); 1271 if (error) 1272 goto out; 1273 ip = VTOI(tvp); 1274 ip->i_gid = dp->i_gid; 1275 #ifdef SUIDDIR 1276 { 1277 #ifdef QUOTA 1278 struct ucred ucred, *ucp; 1279 ucp = cnp->cn_cred; 1280 #endif 1281 /* 1282 * If we are hacking owners here, (only do this where told to) 1283 * and we are not giving it TO root, (would subvert quotas) 1284 * then go ahead and give it to the other user. 1285 * The new directory also inherits the SUID bit. 1286 * If user's UID and dir UID are the same, 1287 * 'give it away' so that the SUID is still forced on. 1288 */ 1289 if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) && 1290 (dp->i_mode & ISUID) && dp->i_uid) { 1291 dmode |= ISUID; 1292 ip->i_uid = dp->i_uid; 1293 #ifdef QUOTA 1294 if (dp->i_uid != cnp->cn_cred->cr_uid) { 1295 /* 1296 * Make sure the correct user gets charged 1297 * for the space. 1298 * Make a dummy credential for the victim. 1299 * XXX This seems to never be accessed out of 1300 * our context so a stack variable is ok. 1301 */ 1302 ucred.cr_ref = 1; 1303 ucred.cr_uid = ip->i_uid; 1304 ucred.cr_ngroups = 1; 1305 ucred.cr_groups[0] = dp->i_gid; 1306 ucp = &ucred; 1307 } 1308 #endif 1309 } else 1310 ip->i_uid = cnp->cn_cred->cr_uid; 1311 #ifdef QUOTA 1312 if ((error = ufs_getinoquota(ip)) || 1313 (error = ufs_chkiq(ip, 1, ucp, 0))) { 1314 ffs_vfree(tvp, ip->i_number, dmode); 1315 vput(tvp); 1316 return (error); 1317 } 1318 #endif 1319 } 1320 #else /* !SUIDDIR */ 1321 ip->i_uid = cnp->cn_cred->cr_uid; 1322 #ifdef QUOTA 1323 if ((error = ufs_getinoquota(ip)) || 1324 (error = ufs_chkiq(ip, 1, cnp->cn_cred, 0))) { 1325 ffs_vfree(tvp, ip->i_number, dmode); 1326 vput(tvp); 1327 return (error); 1328 } 1329 #endif 1330 #endif /* !SUIDDIR */ 1331 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 1332 ip->i_mode = dmode; 1333 tvp->v_type = VDIR; /* Rest init'd in getnewvnode(). */ 1334 ip->i_effnlink = 2; 1335 ip->i_nlink = 2; 1336 if (DOINGSOFTDEP(tvp)) 1337 softdep_change_linkcnt(ip); 1338 if (cnp->cn_flags & CNP_ISWHITEOUT) 1339 ip->i_flags |= UF_OPAQUE; 1340 1341 /* 1342 * Bump link count in parent directory to reflect work done below. 1343 * Should be done before reference is created so cleanup is 1344 * possible if we crash. 1345 */ 1346 dp->i_effnlink++; 1347 dp->i_nlink++; 1348 dp->i_flag |= IN_CHANGE; 1349 if (DOINGSOFTDEP(dvp)) 1350 softdep_change_linkcnt(dp); 1351 error = ffs_update(tvp, !(DOINGSOFTDEP(dvp) | DOINGASYNC(dvp))); 1352 if (error) 1353 goto bad; 1354 1355 /* 1356 * The vnode must have a VM object in order to issue buffer cache 1357 * ops on it. 1358 */ 1359 vinitvmio(tvp, DIRBLKSIZ, DIRBLKSIZ, -1); 1360 1361 /* 1362 * Initialize directory with "." and ".." from static template. 1363 */ 1364 if (dvp->v_mount->mnt_maxsymlinklen > 0) 1365 dtp = &mastertemplate; 1366 else 1367 dtp = (struct dirtemplate *)&omastertemplate; 1368 dirtemplate = *dtp; 1369 dirtemplate.dot_ino = ip->i_number; 1370 dirtemplate.dotdot_ino = dp->i_number; 1371 nvnode_pager_setsize(tvp, DIRBLKSIZ, DIRBLKSIZ, -1); 1372 error = VOP_BALLOC(tvp, 0LL, DIRBLKSIZ, cnp->cn_cred, B_CLRBUF, &bp); 1373 if (error) 1374 goto bad; 1375 ip->i_size = DIRBLKSIZ; 1376 ip->i_flag |= IN_CHANGE | IN_UPDATE; 1377 bcopy((caddr_t)&dirtemplate, (caddr_t)bp->b_data, sizeof dirtemplate); 1378 if (DOINGSOFTDEP(tvp)) { 1379 /* 1380 * Ensure that the entire newly allocated block is a 1381 * valid directory so that future growth within the 1382 * block does not have to ensure that the block is 1383 * written before the inode. 1384 */ 1385 blkoff = DIRBLKSIZ; 1386 while (blkoff < bp->b_bcount) { 1387 ((struct direct *) 1388 (bp->b_data + blkoff))->d_reclen = DIRBLKSIZ; 1389 blkoff += DIRBLKSIZ; 1390 } 1391 } 1392 if ((error = ffs_update(tvp, !(DOINGSOFTDEP(tvp) | 1393 DOINGASYNC(tvp)))) != 0) { 1394 bwrite(bp); 1395 goto bad; 1396 } 1397 /* 1398 * Directory set up, now install its entry in the parent directory. 1399 * 1400 * If we are not doing soft dependencies, then we must write out the 1401 * buffer containing the new directory body before entering the new 1402 * name in the parent. If we are doing soft dependencies, then the 1403 * buffer containing the new directory body will be passed to and 1404 * released in the soft dependency code after the code has attached 1405 * an appropriate ordering dependency to the buffer which ensures that 1406 * the buffer is written before the new name is written in the parent. 1407 */ 1408 if (DOINGASYNC(dvp)) 1409 bdwrite(bp); 1410 else if (!DOINGSOFTDEP(dvp) && (error = bwrite(bp)) != 0) 1411 goto bad; 1412 ufs_makedirentry(ip, cnp, &newdir); 1413 error = ufs_direnter(dvp, tvp, &newdir, cnp, bp); 1414 1415 bad: 1416 if (error == 0) { 1417 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK); 1418 *ap->a_vpp = tvp; 1419 } else { 1420 dp->i_effnlink--; 1421 dp->i_nlink--; 1422 dp->i_flag |= IN_CHANGE; 1423 if (DOINGSOFTDEP(dvp)) 1424 softdep_change_linkcnt(dp); 1425 /* 1426 * No need to do an explicit VOP_TRUNCATE here, vrele will 1427 * do this for us because we set the link count to 0. 1428 */ 1429 ip->i_effnlink = 0; 1430 ip->i_nlink = 0; 1431 ip->i_flag |= IN_CHANGE; 1432 if (DOINGSOFTDEP(tvp)) 1433 softdep_change_linkcnt(ip); 1434 vput(tvp); 1435 } 1436 out: 1437 return (error); 1438 } 1439 1440 /* 1441 * Rmdir system call. 1442 * 1443 * ufs_rmdir(struct vnode *a_dvp, struct vnode *a_vp, 1444 * struct componentname *a_cnp) 1445 */ 1446 static 1447 int 1448 ufs_rmdir(struct vop_old_rmdir_args *ap) 1449 { 1450 struct vnode *vp = ap->a_vp; 1451 struct vnode *dvp = ap->a_dvp; 1452 struct componentname *cnp = ap->a_cnp; 1453 struct inode *ip, *dp; 1454 int error, ioflag; 1455 1456 ip = VTOI(vp); 1457 dp = VTOI(dvp); 1458 1459 /* 1460 * Do not remove a directory that is in the process of being renamed. 1461 * Verify the directory is empty (and valid). Rmdir ".." will not be 1462 * valid since ".." will contain a reference to the current directory 1463 * and thus be non-empty. Do not allow the removal of mounted on 1464 * directories (this can happen when an NFS exported filesystem 1465 * tries to remove a locally mounted on directory). 1466 */ 1467 error = 0; 1468 if (ip->i_flag & IN_RENAME) { 1469 error = EINVAL; 1470 goto out; 1471 } 1472 if (ip->i_effnlink != 2 || 1473 !ufs_dirempty(ip, dp->i_number, cnp->cn_cred)) { 1474 error = ENOTEMPTY; 1475 goto out; 1476 } 1477 #if 0 /* handled by kernel now */ 1478 if ((dp->i_flags & APPEND) 1479 || (ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))) { 1480 error = EPERM; 1481 goto out; 1482 } 1483 #endif 1484 /* 1485 * Delete reference to directory before purging 1486 * inode. If we crash in between, the directory 1487 * will be reattached to lost+found, 1488 */ 1489 dp->i_effnlink--; 1490 ip->i_effnlink--; 1491 if (DOINGSOFTDEP(vp)) { 1492 softdep_change_linkcnt(dp); 1493 softdep_change_linkcnt(ip); 1494 } 1495 error = ufs_dirremove(dvp, ip, cnp->cn_flags, 1); 1496 if (error) { 1497 dp->i_effnlink++; 1498 ip->i_effnlink++; 1499 if (DOINGSOFTDEP(vp)) { 1500 softdep_change_linkcnt(dp); 1501 softdep_change_linkcnt(ip); 1502 } 1503 goto out; 1504 } 1505 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK); 1506 /* 1507 * Truncate inode. The only stuff left in the directory is "." and 1508 * "..". The "." reference is inconsequential since we are quashing 1509 * it. The soft dependency code will arrange to do these operations 1510 * after the parent directory entry has been deleted on disk, so 1511 * when running with that code we avoid doing them now. 1512 */ 1513 if (!DOINGSOFTDEP(vp)) { 1514 dp->i_nlink--; 1515 dp->i_flag |= IN_CHANGE; 1516 ip->i_nlink--; 1517 ip->i_flag |= IN_CHANGE; 1518 ioflag = DOINGASYNC(vp) ? 0 : IO_SYNC; 1519 error = ffs_truncate(vp, (off_t)0, ioflag, cnp->cn_cred); 1520 } 1521 /* cache_purge removed - handled by VFS compat layer */ 1522 #ifdef UFS_DIRHASH 1523 /* Kill any active hash; i_effnlink == 0, so it will not come back. */ 1524 if (ip->i_dirhash != NULL) 1525 ufsdirhash_free(ip); 1526 #endif 1527 out: 1528 VN_KNOTE(vp, NOTE_DELETE); 1529 return (error); 1530 } 1531 1532 /* 1533 * symlink -- make a symbolic link 1534 * 1535 * ufs_symlink(struct vnode *a_dvp, struct vnode **a_vpp, 1536 * struct componentname *a_cnp, struct vattr *a_vap, 1537 * char *a_target) 1538 */ 1539 static 1540 int 1541 ufs_symlink(struct vop_old_symlink_args *ap) 1542 { 1543 struct vnode *vp, **vpp = ap->a_vpp; 1544 struct inode *ip; 1545 int len, error; 1546 1547 error = ufs_makeinode(IFLNK | ap->a_vap->va_mode, ap->a_dvp, 1548 vpp, ap->a_cnp); 1549 if (error) 1550 return (error); 1551 VN_KNOTE(ap->a_dvp, NOTE_WRITE); 1552 vp = *vpp; 1553 len = strlen(ap->a_target); 1554 if (len < vp->v_mount->mnt_maxsymlinklen) { 1555 ip = VTOI(vp); 1556 bcopy(ap->a_target, (char *)ip->i_shortlink, len); 1557 ip->i_size = len; 1558 ip->i_flag |= IN_CHANGE | IN_UPDATE; 1559 } else { 1560 /* 1561 * Make sure we have a VM object in order to use 1562 * the buffer cache. 1563 */ 1564 if (vp->v_object == NULL) 1565 vinitvmio(vp, 0, PAGE_SIZE, -1); 1566 error = vn_rdwr(UIO_WRITE, vp, ap->a_target, len, (off_t)0, 1567 UIO_SYSSPACE, IO_NODELOCKED, 1568 ap->a_cnp->cn_cred, NULL); 1569 } 1570 if (error) 1571 vput(vp); 1572 return (error); 1573 } 1574 1575 /* 1576 * Vnode op for reading directories. 1577 * 1578 * ufs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred, 1579 * int *a_eofflag, int *ncookies, off_t **a_cookies) 1580 */ 1581 static 1582 int 1583 ufs_readdir(struct vop_readdir_args *ap) 1584 { 1585 struct uio *uio = ap->a_uio; 1586 struct vnode *vp = ap->a_vp; 1587 struct direct *dp; 1588 struct buf *bp; 1589 int retval; 1590 int error; 1591 int offset; /* offset into buffer cache buffer */ 1592 int eoffset; /* end of buffer clipped to file EOF */ 1593 int pickup; /* pickup point */ 1594 int ncookies; 1595 int cookie_index; 1596 off_t *cookies; 1597 1598 if (uio->uio_offset < 0) 1599 return (EINVAL); 1600 /* 1601 * Guess the number of cookies needed. Make sure we compute at 1602 * least 1, and no more then a reasonable limit. 1603 */ 1604 if (ap->a_ncookies) { 1605 ncookies = uio->uio_resid / 16 + 1; 1606 if (ncookies > 1024) 1607 ncookies = 1024; 1608 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK); 1609 } else { 1610 ncookies = -1; /* force conditionals below */ 1611 cookies = NULL; 1612 } 1613 cookie_index = 0; 1614 1615 error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY | LK_FAILRECLAIM); 1616 if (error) 1617 return (error); 1618 1619 /* 1620 * Past or at EOF 1621 */ 1622 if (uio->uio_offset >= VTOI(vp)->i_size) { 1623 if (ap->a_eofflag) 1624 *ap->a_eofflag = 1; 1625 if (ap->a_ncookies) { 1626 *ap->a_ncookies = cookie_index; 1627 *ap->a_cookies = cookies; 1628 } 1629 goto done; 1630 } 1631 1632 /* 1633 * Loop until we run out of cookies, we run out of user buffer, 1634 * or we hit the directory EOF. 1635 * 1636 * Always start scans at the beginning of the buffer, don't trust 1637 * the offset supplied by userland. 1638 */ 1639 while ((error = ffs_blkatoff_ra(vp, uio->uio_offset, NULL, &bp, 2)) == 0) { 1640 pickup = (int)(uio->uio_offset - bp->b_loffset); 1641 offset = 0; 1642 retval = 0; 1643 if (bp->b_loffset + bp->b_bcount > VTOI(vp)->i_size) 1644 eoffset = (int)(VTOI(vp)->i_size - bp->b_loffset); 1645 else 1646 eoffset = bp->b_bcount; 1647 1648 while (offset < eoffset) { 1649 dp = (struct direct *)(bp->b_data + offset); 1650 if (dp->d_reclen <= 0 || (dp->d_reclen & 3) || 1651 offset + dp->d_reclen > bp->b_bcount) { 1652 error = EIO; 1653 break; 1654 } 1655 if (offsetof(struct direct, d_name[dp->d_namlen]) > dp->d_reclen) { 1656 error = EIO; 1657 break; 1658 } 1659 if (offset < pickup) { 1660 offset += dp->d_reclen; 1661 continue; 1662 } 1663 #if BYTE_ORDER == LITTLE_ENDIAN 1664 if (OFSFMT(vp)) { 1665 retval = vop_write_dirent(&error, uio, 1666 dp->d_ino, dp->d_namlen, dp->d_type, 1667 dp->d_name); 1668 } else 1669 #endif 1670 { 1671 retval = vop_write_dirent(&error, uio, 1672 dp->d_ino, dp->d_type, dp->d_namlen, 1673 dp->d_name); 1674 } 1675 if (retval) 1676 break; 1677 if (cookies) 1678 cookies[cookie_index] = bp->b_loffset + offset; 1679 ++cookie_index; 1680 offset += dp->d_reclen; 1681 if (cookie_index == ncookies) 1682 break; 1683 } 1684 1685 /* 1686 * This will align the next loop to the beginning of the 1687 * next block, and pickup will calculate to 0. 1688 */ 1689 uio->uio_offset = bp->b_loffset + offset; 1690 brelse(bp); 1691 1692 if (retval || error || cookie_index == ncookies || 1693 uio->uio_offset >= VTOI(vp)->i_size) { 1694 break; 1695 } 1696 } 1697 if (ap->a_eofflag) 1698 *ap->a_eofflag = VTOI(vp)->i_size <= uio->uio_offset; 1699 1700 /* 1701 * Report errors only if we didn't manage to read anything 1702 */ 1703 if (error && cookie_index == 0) { 1704 if (cookies) { 1705 kfree(cookies, M_TEMP); 1706 *ap->a_ncookies = 0; 1707 *ap->a_cookies = NULL; 1708 } 1709 } else { 1710 error = 0; 1711 if (cookies) { 1712 *ap->a_ncookies = cookie_index; 1713 *ap->a_cookies = cookies; 1714 } 1715 } 1716 done: 1717 vn_unlock(vp); 1718 return (error); 1719 } 1720 1721 /* 1722 * Return target name of a symbolic link 1723 * 1724 * ufs_readlink(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred) 1725 */ 1726 static 1727 int 1728 ufs_readlink(struct vop_readlink_args *ap) 1729 { 1730 struct vnode *vp = ap->a_vp; 1731 struct inode *ip = VTOI(vp); 1732 int isize; 1733 1734 isize = ip->i_size; 1735 if ((isize < vp->v_mount->mnt_maxsymlinklen) || 1736 (ip->i_din.di_blocks == 0)) { /* XXX - for old fastlink support */ 1737 uiomove((char *)ip->i_shortlink, isize, ap->a_uio); 1738 return (0); 1739 } 1740 1741 /* 1742 * Perform the equivalent of an OPEN on vp so we can issue a 1743 * VOP_READ. 1744 */ 1745 return (VOP_READ(vp, ap->a_uio, 0, ap->a_cred)); 1746 } 1747 1748 /* 1749 * Calculate the logical to physical mapping if not done already, 1750 * then call the device strategy routine. 1751 * 1752 * In order to be able to swap to a file, the VOP_BMAP operation may not 1753 * deadlock on memory. See ufs_bmap() for details. 1754 * 1755 * ufs_strategy(struct vnode *a_vp, struct bio *a_bio) 1756 */ 1757 static 1758 int 1759 ufs_strategy(struct vop_strategy_args *ap) 1760 { 1761 struct bio *bio = ap->a_bio; 1762 struct bio *nbio; 1763 struct buf *bp = bio->bio_buf; 1764 struct vnode *vp = ap->a_vp; 1765 struct inode *ip; 1766 int error; 1767 1768 ip = VTOI(vp); 1769 if (vp->v_type == VBLK || vp->v_type == VCHR) 1770 panic("ufs_strategy: spec"); 1771 nbio = push_bio(bio); 1772 if (nbio->bio_offset == NOOFFSET) { 1773 error = VOP_BMAP(vp, bio->bio_offset, &nbio->bio_offset, 1774 NULL, NULL, bp->b_cmd); 1775 if (error) { 1776 bp->b_error = error; 1777 bp->b_flags |= B_ERROR; 1778 /* I/O was never started on nbio, must biodone(bio) */ 1779 biodone(bio); 1780 return (error); 1781 } 1782 if (nbio->bio_offset == NOOFFSET) 1783 vfs_bio_clrbuf(bp); 1784 } 1785 if (nbio->bio_offset == NOOFFSET) { 1786 /* 1787 * We hit a hole in the file. The buffer has been zero-filled 1788 * so just biodone() it. 1789 */ 1790 biodone(bio); 1791 } else { 1792 vn_strategy(ip->i_devvp, nbio); 1793 } 1794 return (0); 1795 } 1796 1797 /* 1798 * Print out the contents of an inode. 1799 * 1800 * ufs_print(struct vnode *a_vp) 1801 */ 1802 static 1803 int 1804 ufs_print(struct vop_print_args *ap) 1805 { 1806 struct vnode *vp = ap->a_vp; 1807 struct inode *ip = VTOI(vp); 1808 1809 kprintf("tag VT_UFS, ino %lu, on dev %s (%d, %d)", 1810 (u_long)ip->i_number, devtoname(ip->i_dev), major(ip->i_dev), 1811 minor(ip->i_dev)); 1812 if (vp->v_type == VFIFO) 1813 fifo_printinfo(vp); 1814 lockmgr_printinfo(&vp->v_lock); 1815 kprintf("\n"); 1816 return (0); 1817 } 1818 1819 /* 1820 * Read wrapper for fifos. 1821 * 1822 * ufsfifo_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 1823 * struct ucred *a_cred) 1824 */ 1825 static 1826 int 1827 ufsfifo_read(struct vop_read_args *ap) 1828 { 1829 int error, resid; 1830 struct inode *ip; 1831 struct uio *uio; 1832 1833 uio = ap->a_uio; 1834 resid = uio->uio_resid; 1835 error = VOCALL(&fifo_vnode_vops, &ap->a_head); 1836 ip = VTOI(ap->a_vp); 1837 if ((ap->a_vp->v_mount->mnt_flag & MNT_NOATIME) == 0 && ip != NULL && 1838 (uio->uio_resid != resid || (error == 0 && resid != 0))) 1839 VTOI(ap->a_vp)->i_flag |= IN_ACCESS; 1840 return (error); 1841 } 1842 1843 /* 1844 * Write wrapper for fifos. 1845 * 1846 * ufsfifo_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 1847 * struct ucred *a_cred) 1848 */ 1849 static 1850 int 1851 ufsfifo_write(struct vop_write_args *ap) 1852 { 1853 int error, resid; 1854 struct inode *ip; 1855 struct uio *uio; 1856 1857 uio = ap->a_uio; 1858 resid = uio->uio_resid; 1859 error = VOCALL(&fifo_vnode_vops, &ap->a_head); 1860 ip = VTOI(ap->a_vp); 1861 if (ip != NULL && (uio->uio_resid != resid || (error == 0 && resid != 0))) 1862 VTOI(ap->a_vp)->i_flag |= IN_CHANGE | IN_UPDATE; 1863 return (error); 1864 } 1865 1866 /* 1867 * Close wrapper for fifos. 1868 * 1869 * Update the times on the inode then do device close. 1870 * 1871 * ufsfifo_close(struct vnode *a_vp, int a_fflag) 1872 */ 1873 static 1874 int 1875 ufsfifo_close(struct vop_close_args *ap) 1876 { 1877 struct vnode *vp = ap->a_vp; 1878 1879 if (VREFCNT(vp) > 1) 1880 ufs_itimes(vp); 1881 return (VOCALL(&fifo_vnode_vops, &ap->a_head)); 1882 } 1883 1884 /* 1885 * Kqfilter wrapper for fifos. 1886 * 1887 * Fall through to ufs kqfilter routines if needed 1888 */ 1889 static 1890 int 1891 ufsfifo_kqfilter(struct vop_kqfilter_args *ap) 1892 { 1893 int error; 1894 1895 error = VOCALL(&fifo_vnode_vops, &ap->a_head); 1896 if (error) 1897 error = ufs_kqfilter(ap); 1898 return (error); 1899 } 1900 1901 /* 1902 * Advisory record locking support 1903 * 1904 * ufs_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, struct flock *a_fl, 1905 * int a_flags) 1906 */ 1907 static 1908 int 1909 ufs_advlock(struct vop_advlock_args *ap) 1910 { 1911 struct inode *ip = VTOI(ap->a_vp); 1912 1913 return (lf_advlock(ap, &(ip->i_lockf), ip->i_size)); 1914 } 1915 1916 /* 1917 * Initialize the vnode associated with a new inode, handle aliased 1918 * vnodes. 1919 * 1920 * Make sure directories have their VM object now rather then later, 1921 * saving us from having to check on all the myrid directory VOPs 1922 * that might be executed without a VOP_OPEN being performed. 1923 */ 1924 int 1925 ufs_vinit(struct mount *mntp, struct vnode **vpp) 1926 { 1927 struct inode *ip; 1928 struct vnode *vp; 1929 1930 vp = *vpp; 1931 ip = VTOI(vp); 1932 1933 vp->v_type = IFTOVT(ip->i_mode); 1934 1935 switch(vp->v_type) { 1936 case VCHR: 1937 case VBLK: 1938 vp->v_ops = &mntp->mnt_vn_spec_ops; 1939 addaliasu(vp, umajor(ip->i_rdev), uminor(ip->i_rdev)); 1940 break; 1941 case VFIFO: 1942 vp->v_ops = &mntp->mnt_vn_fifo_ops; 1943 break; 1944 case VDIR: 1945 case VREG: 1946 vinitvmio(vp, ip->i_size, 1947 blkoffsize(ip->i_fs, ip, ip->i_size), 1948 blkoff(ip->i_fs, ip->i_size)); 1949 break; 1950 case VLNK: 1951 if (ip->i_size >= vp->v_mount->mnt_maxsymlinklen) { 1952 vinitvmio(vp, ip->i_size, 1953 blkoffsize(ip->i_fs, ip, ip->i_size), 1954 blkoff(ip->i_fs, ip->i_size)); 1955 } 1956 break; 1957 default: 1958 break; 1959 1960 } 1961 1962 if (ip->i_number == UFS_ROOTINO) 1963 vsetflags(vp, VROOT); 1964 /* 1965 * Initialize modrev times 1966 */ 1967 ip->i_modrev = init_va_filerev(); 1968 *vpp = vp; 1969 return (0); 1970 } 1971 1972 /* 1973 * Allocate a new inode. 1974 */ 1975 static 1976 int 1977 ufs_makeinode(int mode, struct vnode *dvp, struct vnode **vpp, 1978 struct componentname *cnp) 1979 { 1980 struct inode *ip, *pdir; 1981 struct direct newdir; 1982 struct vnode *tvp; 1983 int error; 1984 1985 pdir = VTOI(dvp); 1986 *vpp = NULL; 1987 if ((mode & IFMT) == 0) 1988 mode |= IFREG; 1989 1990 error = ffs_valloc(dvp, mode, cnp->cn_cred, &tvp); 1991 if (error) 1992 return (error); 1993 ip = VTOI(tvp); 1994 ip->i_flags = pdir->i_flags & (SF_NOHISTORY|UF_NOHISTORY|UF_NODUMP); 1995 ip->i_gid = pdir->i_gid; 1996 #ifdef SUIDDIR 1997 { 1998 #ifdef QUOTA 1999 struct ucred ucred, *ucp; 2000 ucp = cnp->cn_cred; 2001 #endif 2002 /* 2003 * If we are not the owner of the directory, 2004 * and we are hacking owners here, (only do this where told to) 2005 * and we are not giving it TO root, (would subvert quotas) 2006 * then go ahead and give it to the other user. 2007 * Note that this drops off the execute bits for security. 2008 */ 2009 if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) && 2010 (pdir->i_mode & ISUID) && 2011 (pdir->i_uid != cnp->cn_cred->cr_uid) && pdir->i_uid) { 2012 ip->i_uid = pdir->i_uid; 2013 mode &= ~07111; 2014 #ifdef QUOTA 2015 /* 2016 * Make sure the correct user gets charged 2017 * for the space. 2018 * Quickly knock up a dummy credential for the victim. 2019 * XXX This seems to never be accessed out of our 2020 * context so a stack variable is ok. 2021 */ 2022 ucred.cr_ref = 1; 2023 ucred.cr_uid = ip->i_uid; 2024 ucred.cr_ngroups = 1; 2025 ucred.cr_groups[0] = pdir->i_gid; 2026 ucp = &ucred; 2027 #endif 2028 } else 2029 ip->i_uid = cnp->cn_cred->cr_uid; 2030 2031 #ifdef QUOTA 2032 if ((error = ufs_getinoquota(ip)) || 2033 (error = ufs_chkiq(ip, 1, ucp, 0))) { 2034 ffs_vfree(tvp, ip->i_number, mode); 2035 vput(tvp); 2036 return (error); 2037 } 2038 #endif 2039 } 2040 #else /* !SUIDDIR */ 2041 ip->i_uid = cnp->cn_cred->cr_uid; 2042 #ifdef QUOTA 2043 if ((error = ufs_getinoquota(ip)) || 2044 (error = ufs_chkiq(ip, 1, cnp->cn_cred, 0))) { 2045 ffs_vfree(tvp, ip->i_number, mode); 2046 vput(tvp); 2047 return (error); 2048 } 2049 #endif 2050 #endif /* !SUIDDIR */ 2051 ip->i_din.di_spare7E = 0; 2052 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 2053 ip->i_mode = mode; 2054 tvp->v_type = IFTOVT(mode); /* Rest init'd in getnewvnode(). */ 2055 ip->i_effnlink = 1; 2056 ip->i_nlink = 1; 2057 if (DOINGSOFTDEP(tvp)) 2058 softdep_change_linkcnt(ip); 2059 if ((ip->i_mode & ISGID) && !groupmember(ip->i_gid, cnp->cn_cred) && 2060 priv_check_cred(cnp->cn_cred, PRIV_VFS_SETGID, 0)) { 2061 ip->i_mode &= ~ISGID; 2062 } 2063 2064 if (cnp->cn_flags & CNP_ISWHITEOUT) 2065 ip->i_flags |= UF_OPAQUE; 2066 2067 /* 2068 * Regular files and directories need VM objects. Softlinks do 2069 * not (not immediately anyway). 2070 */ 2071 if (tvp->v_type == VREG || tvp->v_type == VDIR) 2072 vinitvmio(tvp, 0, PAGE_SIZE, -1); 2073 2074 /* 2075 * Make sure inode goes to disk before directory entry. 2076 */ 2077 error = ffs_update(tvp, !(DOINGSOFTDEP(tvp) | DOINGASYNC(tvp))); 2078 if (error) 2079 goto bad; 2080 ufs_makedirentry(ip, cnp, &newdir); 2081 error = ufs_direnter(dvp, tvp, &newdir, cnp, NULL); 2082 if (error) 2083 goto bad; 2084 *vpp = tvp; 2085 return (0); 2086 2087 bad: 2088 /* 2089 * Write error occurred trying to update the inode 2090 * or the directory so must deallocate the inode. 2091 */ 2092 ip->i_effnlink = 0; 2093 ip->i_nlink = 0; 2094 ip->i_flag |= IN_CHANGE; 2095 if (DOINGSOFTDEP(tvp)) 2096 softdep_change_linkcnt(ip); 2097 vput(tvp); 2098 return (error); 2099 } 2100 2101 static int 2102 ufs_missingop(struct vop_generic_args *ap) 2103 { 2104 panic("no vop function for %s in ufs child", ap->a_desc->sd_name); 2105 return (EOPNOTSUPP); 2106 } 2107 2108 static struct filterops ufsread_filtops = 2109 { FILTEROP_ISFD, NULL, filt_ufsdetach, filt_ufsread }; 2110 static struct filterops ufswrite_filtops = 2111 { FILTEROP_ISFD, NULL, filt_ufsdetach, filt_ufswrite }; 2112 static struct filterops ufsvnode_filtops = 2113 { FILTEROP_ISFD, NULL, filt_ufsdetach, filt_ufsvnode }; 2114 2115 /* 2116 * ufs_kqfilter(struct vnode *a_vp, struct knote *a_kn) 2117 */ 2118 static int 2119 ufs_kqfilter(struct vop_kqfilter_args *ap) 2120 { 2121 struct vnode *vp = ap->a_vp; 2122 struct knote *kn = ap->a_kn; 2123 2124 switch (kn->kn_filter) { 2125 case EVFILT_READ: 2126 kn->kn_fop = &ufsread_filtops; 2127 break; 2128 case EVFILT_WRITE: 2129 kn->kn_fop = &ufswrite_filtops; 2130 break; 2131 case EVFILT_VNODE: 2132 kn->kn_fop = &ufsvnode_filtops; 2133 break; 2134 default: 2135 return (EOPNOTSUPP); 2136 } 2137 2138 kn->kn_hook = (caddr_t)vp; 2139 2140 /* XXX: kq token actually protects the list */ 2141 lwkt_gettoken(&vp->v_token); 2142 knote_insert(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn); 2143 lwkt_reltoken(&vp->v_token); 2144 2145 return (0); 2146 } 2147 2148 static void 2149 filt_ufsdetach(struct knote *kn) 2150 { 2151 struct vnode *vp = (struct vnode *)kn->kn_hook; 2152 2153 lwkt_gettoken(&vp->v_token); 2154 knote_remove(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn); 2155 lwkt_reltoken(&vp->v_token); 2156 } 2157 2158 /*ARGSUSED*/ 2159 static int 2160 filt_ufsread(struct knote *kn, long hint) 2161 { 2162 struct vnode *vp = (struct vnode *)kn->kn_hook; 2163 struct inode *ip = VTOI(vp); 2164 off_t off; 2165 2166 /* 2167 * filesystem is gone, so set the EOF flag and schedule 2168 * the knote for deletion. 2169 */ 2170 if (hint == NOTE_REVOKE) { 2171 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT); 2172 return (1); 2173 } 2174 2175 off = ip->i_size - kn->kn_fp->f_offset; 2176 kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX; 2177 if (kn->kn_sfflags & NOTE_OLDAPI) 2178 return(1); 2179 return (kn->kn_data != 0); 2180 } 2181 2182 /*ARGSUSED*/ 2183 static int 2184 filt_ufswrite(struct knote *kn, long hint) 2185 { 2186 /* 2187 * filesystem is gone, so set the EOF flag and schedule 2188 * the knote for deletion. 2189 */ 2190 if (hint == NOTE_REVOKE) 2191 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT); 2192 2193 kn->kn_data = 0; 2194 return (1); 2195 } 2196 2197 static int 2198 filt_ufsvnode(struct knote *kn, long hint) 2199 { 2200 if (kn->kn_sfflags & hint) 2201 kn->kn_fflags |= hint; 2202 if (hint == NOTE_REVOKE) { 2203 kn->kn_flags |= (EV_EOF | EV_NODATA); 2204 return (1); 2205 } 2206 return (kn->kn_fflags != 0); 2207 } 2208 2209 /* Global vfs data structures for ufs. */ 2210 static struct vop_ops ufs_vnode_vops = { 2211 .vop_default = vop_defaultop, 2212 .vop_fsync = (void *)ufs_missingop, 2213 .vop_read = (void *)ufs_missingop, 2214 .vop_reallocblks = (void *)ufs_missingop, 2215 .vop_write = (void *)ufs_missingop, 2216 .vop_access = ufs_access, 2217 .vop_advlock = ufs_advlock, 2218 .vop_bmap = ufs_bmap, 2219 .vop_old_lookup = ufs_lookup, 2220 .vop_close = ufs_close, 2221 .vop_old_create = ufs_create, 2222 .vop_getattr = ufs_getattr, 2223 .vop_inactive = ufs_inactive, 2224 .vop_old_link = ufs_link, 2225 .vop_old_mkdir = ufs_mkdir, 2226 .vop_old_mknod = ufs_mknod, 2227 .vop_open = vop_stdopen, 2228 .vop_pathconf = vop_stdpathconf, 2229 .vop_kqfilter = ufs_kqfilter, 2230 .vop_print = ufs_print, 2231 .vop_readdir = ufs_readdir, 2232 .vop_readlink = ufs_readlink, 2233 .vop_reclaim = ufs_reclaim, 2234 .vop_old_remove = ufs_remove, 2235 .vop_old_rename = ufs_rename, 2236 .vop_old_rmdir = ufs_rmdir, 2237 .vop_setattr = ufs_setattr, 2238 .vop_markatime = ufs_markatime, 2239 .vop_strategy = ufs_strategy, 2240 .vop_old_symlink = ufs_symlink, 2241 .vop_old_whiteout = ufs_whiteout 2242 }; 2243 2244 static struct vop_ops ufs_spec_vops = { 2245 .vop_default = vop_defaultop, 2246 .vop_fsync = (void *)ufs_missingop, 2247 .vop_access = ufs_access, 2248 .vop_close = ufs_close, 2249 .vop_getattr = ufs_getattr, 2250 .vop_inactive = ufs_inactive, 2251 .vop_print = ufs_print, 2252 .vop_read = vop_stdnoread, 2253 .vop_reclaim = ufs_reclaim, 2254 .vop_setattr = ufs_setattr, 2255 .vop_markatime = ufs_markatime, 2256 .vop_write = vop_stdnowrite 2257 }; 2258 2259 static struct vop_ops ufs_fifo_vops = { 2260 .vop_default = fifo_vnoperate, 2261 .vop_fsync = (void *)ufs_missingop, 2262 .vop_access = ufs_access, 2263 .vop_close = ufsfifo_close, 2264 .vop_getattr = ufs_getattr, 2265 .vop_inactive = ufs_inactive, 2266 .vop_kqfilter = ufsfifo_kqfilter, 2267 .vop_print = ufs_print, 2268 .vop_read = ufsfifo_read, 2269 .vop_reclaim = ufs_reclaim, 2270 .vop_setattr = ufs_setattr, 2271 .vop_markatime = ufs_markatime, 2272 .vop_write = ufsfifo_write 2273 }; 2274 2275 VNODEOP_SET(ufs_vnode_vops); 2276 VNODEOP_SET(ufs_spec_vops); 2277 VNODEOP_SET(ufs_fifo_vops); 2278 2279 /* 2280 * ufs_vnoperate() 2281 */ 2282 int 2283 ufs_vnoperate(struct vop_generic_args *ap) 2284 { 2285 return (VOCALL(&ufs_vnode_vops, ap)); 2286 } 2287 2288 /* 2289 * ufs_vnoperatefifo() 2290 */ 2291 int 2292 ufs_vnoperatefifo(struct vop_generic_args *ap) 2293 { 2294 return (VOCALL(&ufs_fifo_vops, ap)); 2295 } 2296 2297 /* 2298 * ufs_vnoperatespec() 2299 */ 2300 int 2301 ufs_vnoperatespec(struct vop_generic_args *ap) 2302 { 2303 return (VOCALL(&ufs_spec_vops, ap)); 2304 } 2305