1 /* 2 * Copyright (c) 1982, 1986, 1989, 1993, 1995 3 * The Regents of the University of California. All rights reserved. 4 * (c) UNIX System Laboratories, Inc. 5 * All or some portions of this file are derived from material licensed 6 * to the University of California by American Telephone and Telegraph 7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 8 * the permission of UNIX System Laboratories, Inc. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the University of 21 * California, Berkeley and its contributors. 22 * 4. Neither the name of the University nor the names of its contributors 23 * may be used to endorse or promote products derived from this software 24 * without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 36 * SUCH DAMAGE. 37 * 38 * @(#)ufs_vnops.c 8.27 (Berkeley) 5/27/95 39 * $FreeBSD: src/sys/ufs/ufs/ufs_vnops.c,v 1.131.2.8 2003/01/02 17:26:19 bde Exp $ 40 * $DragonFly: src/sys/vfs/ufs/ufs_vnops.c,v 1.67 2008/09/28 05:04:22 dillon Exp $ 41 */ 42 43 #include "opt_quota.h" 44 #include "opt_suiddir.h" 45 #include "opt_ufs.h" 46 47 #include <sys/param.h> 48 #include <sys/systm.h> 49 #include <sys/kernel.h> 50 #include <sys/fcntl.h> 51 #include <sys/stat.h> 52 #include <sys/buf.h> 53 #include <sys/proc.h> 54 #include <sys/priv.h> 55 #include <sys/namei.h> 56 #include <sys/mount.h> 57 #include <sys/unistd.h> 58 #include <sys/vnode.h> 59 #include <sys/malloc.h> 60 #include <sys/dirent.h> 61 #include <sys/lockf.h> 62 #include <sys/event.h> 63 #include <sys/conf.h> 64 65 #include <sys/file.h> /* XXX */ 66 #include <sys/jail.h> 67 68 #include <vm/vm.h> 69 #include <vm/vm_extern.h> 70 71 #include <vfs/fifofs/fifo.h> 72 73 #include "quota.h" 74 #include "inode.h" 75 #include "dir.h" 76 #include "ufsmount.h" 77 #include "ufs_extern.h" 78 #include "ffs_extern.h" 79 #ifdef UFS_DIRHASH 80 #include "dirhash.h" 81 #endif 82 83 static int ufs_access (struct vop_access_args *); 84 static int ufs_advlock (struct vop_advlock_args *); 85 static int ufs_chmod (struct vnode *, int, struct ucred *); 86 static int ufs_chown (struct vnode *, uid_t, gid_t, struct ucred *); 87 static int ufs_close (struct vop_close_args *); 88 static int ufs_create (struct vop_old_create_args *); 89 static int ufs_getattr (struct vop_getattr_args *); 90 static int ufs_link (struct vop_old_link_args *); 91 static int ufs_makeinode (int mode, struct vnode *, struct vnode **, struct componentname *); 92 static int ufs_markatime (struct vop_markatime_args *); 93 static int ufs_missingop (struct vop_generic_args *ap); 94 static int ufs_mkdir (struct vop_old_mkdir_args *); 95 static int ufs_mknod (struct vop_old_mknod_args *); 96 static int ufs_mmap (struct vop_mmap_args *); 97 static int ufs_print (struct vop_print_args *); 98 static int ufs_readdir (struct vop_readdir_args *); 99 static int ufs_readlink (struct vop_readlink_args *); 100 static int ufs_remove (struct vop_old_remove_args *); 101 static int ufs_rename (struct vop_old_rename_args *); 102 static int ufs_rmdir (struct vop_old_rmdir_args *); 103 static int ufs_setattr (struct vop_setattr_args *); 104 static int ufs_strategy (struct vop_strategy_args *); 105 static int ufs_symlink (struct vop_old_symlink_args *); 106 static int ufs_whiteout (struct vop_old_whiteout_args *); 107 static int ufsfifo_close (struct vop_close_args *); 108 static int ufsfifo_kqfilter (struct vop_kqfilter_args *); 109 static int ufsfifo_read (struct vop_read_args *); 110 static int ufsfifo_write (struct vop_write_args *); 111 static int ufsspec_close (struct vop_close_args *); 112 static int ufsspec_read (struct vop_read_args *); 113 static int ufsspec_write (struct vop_write_args *); 114 static int ufsspec_getattr (struct vop_getattr_args *); 115 static int filt_ufsread (struct knote *kn, long hint); 116 static int filt_ufswrite (struct knote *kn, long hint); 117 static int filt_ufsvnode (struct knote *kn, long hint); 118 static void filt_ufsdetach (struct knote *kn); 119 static int ufs_kqfilter (struct vop_kqfilter_args *ap); 120 121 union _qcvt { 122 int64_t qcvt; 123 int32_t val[2]; 124 }; 125 #define SETHIGH(q, h) { \ 126 union _qcvt tmp; \ 127 tmp.qcvt = (q); \ 128 tmp.val[_QUAD_HIGHWORD] = (h); \ 129 (q) = tmp.qcvt; \ 130 } 131 #define SETLOW(q, l) { \ 132 union _qcvt tmp; \ 133 tmp.qcvt = (q); \ 134 tmp.val[_QUAD_LOWWORD] = (l); \ 135 (q) = tmp.qcvt; \ 136 } 137 #define VN_KNOTE(vp, b) \ 138 KNOTE(&vp->v_pollinfo.vpi_selinfo.si_note, (b)) 139 140 #define OFSFMT(vp) ((vp)->v_mount->mnt_maxsymlinklen <= 0) 141 142 /* 143 * A virgin directory (no blushing please). 144 */ 145 static struct dirtemplate mastertemplate = { 146 0, 12, DT_DIR, 1, ".", 147 0, DIRBLKSIZ - 12, DT_DIR, 2, ".." 148 }; 149 static struct odirtemplate omastertemplate = { 150 0, 12, 1, ".", 151 0, DIRBLKSIZ - 12, 2, ".." 152 }; 153 154 void 155 ufs_itimes(struct vnode *vp) 156 { 157 struct inode *ip; 158 struct timespec ts; 159 160 ip = VTOI(vp); 161 if ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE)) == 0) 162 return; 163 if ((vp->v_type == VBLK || vp->v_type == VCHR) && !DOINGSOFTDEP(vp)) 164 ip->i_flag |= IN_LAZYMOD; 165 else 166 ip->i_flag |= IN_MODIFIED; 167 if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 168 vfs_timestamp(&ts); 169 if (ip->i_flag & IN_ACCESS) { 170 ip->i_atime = ts.tv_sec; 171 ip->i_atimensec = ts.tv_nsec; 172 } 173 if (ip->i_flag & IN_UPDATE) { 174 ip->i_mtime = ts.tv_sec; 175 ip->i_mtimensec = ts.tv_nsec; 176 ip->i_modrev++; 177 } 178 if (ip->i_flag & IN_CHANGE) { 179 ip->i_ctime = ts.tv_sec; 180 ip->i_ctimensec = ts.tv_nsec; 181 } 182 } 183 ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_UPDATE); 184 } 185 186 /* 187 * Create a regular file 188 * 189 * ufs_create(struct vnode *a_dvp, struct vnode **a_vpp, 190 * struct componentname *a_cnp, struct vattr *a_vap) 191 */ 192 static 193 int 194 ufs_create(struct vop_old_create_args *ap) 195 { 196 int error; 197 198 error = 199 ufs_makeinode(MAKEIMODE(ap->a_vap->va_type, ap->a_vap->va_mode), 200 ap->a_dvp, ap->a_vpp, ap->a_cnp); 201 if (error) 202 return (error); 203 VN_KNOTE(ap->a_dvp, NOTE_WRITE); 204 return (0); 205 } 206 207 /* 208 * Mknod vnode call 209 * 210 * ufs_mknod(struct vnode *a_dvp, struct vnode **a_vpp, 211 * struct componentname *a_cnp, struct vattr *a_vap) 212 */ 213 /* ARGSUSED */ 214 static 215 int 216 ufs_mknod(struct vop_old_mknod_args *ap) 217 { 218 struct vattr *vap = ap->a_vap; 219 struct vnode **vpp = ap->a_vpp; 220 struct inode *ip; 221 ino_t ino; 222 int error; 223 224 /* 225 * UFS cannot represent the entire major/minor range supported by 226 * the kernel. 227 */ 228 if (vap->va_rmajor != VNOVAL && 229 makeudev(vap->va_rmajor, vap->va_rminor) == NOUDEV) { 230 return(EINVAL); 231 } 232 233 /* no special directory support */ 234 if (vap->va_type == VDIR) 235 return(EINVAL); 236 237 error = ufs_makeinode(MAKEIMODE(vap->va_type, vap->va_mode), 238 ap->a_dvp, vpp, ap->a_cnp); 239 if (error) 240 return (error); 241 VN_KNOTE(ap->a_dvp, NOTE_WRITE); 242 ip = VTOI(*vpp); 243 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 244 if (vap->va_rmajor != VNOVAL) { 245 /* 246 * Want to be able to use this to make badblock 247 * inodes, so don't truncate the dev number. 248 */ 249 ip->i_rdev = makeudev(vap->va_rmajor, vap->va_rminor); 250 } 251 /* 252 * Remove inode, then reload it through VFS_VGET so it is 253 * checked to see if it is an alias of an existing entry in 254 * the inode cache. 255 */ 256 (*vpp)->v_type = VNON; 257 ino = ip->i_number; /* Save this before vgone() invalidates ip. */ 258 vgone_vxlocked(*vpp); 259 vput(*vpp); 260 error = VFS_VGET(ap->a_dvp->v_mount, ino, vpp); 261 if (error) { 262 *vpp = NULL; 263 return (error); 264 } 265 return (0); 266 } 267 268 /* 269 * Close called. 270 * 271 * Update the times on the inode. 272 * 273 * ufs_close(struct vnode *a_vp, int a_fflag) 274 */ 275 /* ARGSUSED */ 276 static 277 int 278 ufs_close(struct vop_close_args *ap) 279 { 280 struct vnode *vp = ap->a_vp; 281 282 if (vp->v_sysref.refcnt > 1) 283 ufs_itimes(vp); 284 return (vop_stdclose(ap)); 285 } 286 287 /* 288 * ufs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred) 289 */ 290 static 291 int 292 ufs_access(struct vop_access_args *ap) 293 { 294 struct vnode *vp = ap->a_vp; 295 struct inode *ip = VTOI(vp); 296 struct ucred *cred = ap->a_cred; 297 mode_t mask, mode = ap->a_mode; 298 gid_t *gp; 299 int i; 300 #ifdef QUOTA 301 int error; 302 #endif 303 304 /* 305 * Disallow write attempts on read-only filesystems; 306 * unless the file is a socket, fifo, or a block or 307 * character device resident on the filesystem. 308 */ 309 if (mode & VWRITE) { 310 switch (vp->v_type) { 311 case VDIR: 312 case VLNK: 313 case VREG: 314 if (vp->v_mount->mnt_flag & MNT_RDONLY) 315 return (EROFS); 316 #ifdef QUOTA 317 if ((error = ufs_getinoquota(ip)) != 0) 318 return (error); 319 #endif 320 break; 321 default: 322 break; 323 } 324 } 325 326 #if 0 /* handled by kernel now */ 327 /* If immutable bit set, nobody gets to write it. */ 328 if ((mode & VWRITE) && (ip->i_flags & IMMUTABLE)) 329 return (EPERM); 330 #endif 331 332 /* Otherwise, user id 0 always gets access. */ 333 if (cred->cr_uid == 0) 334 return (0); 335 336 mask = 0; 337 338 /* Otherwise, check the owner. */ 339 if (cred->cr_uid == ip->i_uid) { 340 if (mode & VEXEC) 341 mask |= S_IXUSR; 342 if (mode & VREAD) 343 mask |= S_IRUSR; 344 if (mode & VWRITE) 345 mask |= S_IWUSR; 346 return ((ip->i_mode & mask) == mask ? 0 : EACCES); 347 } 348 349 /* Otherwise, check the groups. */ 350 for (i = 0, gp = cred->cr_groups; i < cred->cr_ngroups; i++, gp++) 351 if (ip->i_gid == *gp) { 352 if (mode & VEXEC) 353 mask |= S_IXGRP; 354 if (mode & VREAD) 355 mask |= S_IRGRP; 356 if (mode & VWRITE) 357 mask |= S_IWGRP; 358 return ((ip->i_mode & mask) == mask ? 0 : EACCES); 359 } 360 361 /* Otherwise, check everyone else. */ 362 if (mode & VEXEC) 363 mask |= S_IXOTH; 364 if (mode & VREAD) 365 mask |= S_IROTH; 366 if (mode & VWRITE) 367 mask |= S_IWOTH; 368 return ((ip->i_mode & mask) == mask ? 0 : EACCES); 369 } 370 371 /* 372 * ufs_getattr(struct vnode *a_vp, struct vattr *a_vap) 373 */ 374 /* ARGSUSED */ 375 static 376 int 377 ufs_getattr(struct vop_getattr_args *ap) 378 { 379 struct vnode *vp = ap->a_vp; 380 struct inode *ip = VTOI(vp); 381 struct vattr *vap = ap->a_vap; 382 383 /* 384 * This may cause i_fsmid to be updated even if no change (0) 385 * is returned, but we should only write out the inode if non-zero 386 * is returned and if the mount is read-write. 387 */ 388 if (cache_check_fsmid_vp(vp, &ip->i_fsmid) && 389 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0 390 ) { 391 ip->i_flag |= IN_LAZYMOD; 392 } 393 394 ufs_itimes(vp); 395 /* 396 * Copy from inode table 397 */ 398 vap->va_fsid = dev2udev(ip->i_dev); 399 vap->va_fileid = ip->i_number; 400 vap->va_mode = ip->i_mode & ~IFMT; 401 vap->va_nlink = VFSTOUFS(vp->v_mount)->um_i_effnlink_valid ? 402 ip->i_effnlink : ip->i_nlink; 403 vap->va_uid = ip->i_uid; 404 vap->va_gid = ip->i_gid; 405 vap->va_rmajor = umajor(ip->i_rdev); 406 vap->va_rminor = uminor(ip->i_rdev); 407 vap->va_size = ip->i_din.di_size; 408 vap->va_atime.tv_sec = ip->i_atime; 409 vap->va_atime.tv_nsec = ip->i_atimensec; 410 vap->va_mtime.tv_sec = ip->i_mtime; 411 vap->va_mtime.tv_nsec = ip->i_mtimensec; 412 vap->va_ctime.tv_sec = ip->i_ctime; 413 vap->va_ctime.tv_nsec = ip->i_ctimensec; 414 vap->va_flags = ip->i_flags; 415 vap->va_gen = ip->i_gen; 416 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize; 417 vap->va_bytes = dbtob((u_quad_t)ip->i_blocks); 418 vap->va_type = IFTOVT(ip->i_mode); 419 vap->va_filerev = ip->i_modrev; 420 vap->va_fsmid = ip->i_fsmid; 421 return (0); 422 } 423 424 static 425 int 426 ufs_markatime(struct vop_markatime_args *ap) 427 { 428 struct vnode *vp = ap->a_vp; 429 struct inode *ip = VTOI(vp); 430 431 if (vp->v_mount->mnt_flag & MNT_RDONLY) 432 return (EROFS); 433 if (vp->v_mount->mnt_flag & MNT_NOATIME) 434 return (0); 435 ip->i_flag |= IN_ACCESS; 436 VN_KNOTE(vp, NOTE_ATTRIB); 437 return (0); 438 } 439 440 /* 441 * Set attribute vnode op. called from several syscalls 442 * 443 * ufs_setattr(struct vnode *a_vp, struct vattr *a_vap, 444 * struct ucred *a_cred) 445 */ 446 static 447 int 448 ufs_setattr(struct vop_setattr_args *ap) 449 { 450 struct vattr *vap = ap->a_vap; 451 struct vnode *vp = ap->a_vp; 452 struct inode *ip = VTOI(vp); 453 struct ucred *cred = ap->a_cred; 454 int error; 455 456 /* 457 * Check for unsettable attributes. 458 */ 459 if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) || 460 (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) || 461 (vap->va_blocksize != VNOVAL) || (vap->va_rmajor != VNOVAL) || 462 ((int)vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) { 463 return (EINVAL); 464 } 465 if (vap->va_flags != VNOVAL) { 466 if (vp->v_mount->mnt_flag & MNT_RDONLY) 467 return (EROFS); 468 if (cred->cr_uid != ip->i_uid && 469 (error = priv_check_cred(cred, PRIV_ROOT, PRISON_ROOT))) 470 return (error); 471 /* 472 * Note that a root chflags becomes a user chflags when 473 * we are jailed, unless the jail.chflags_allowed sysctl 474 * is set. 475 */ 476 if (cred->cr_uid == 0 && 477 (!jailed(cred) || jail_chflags_allowed)) { 478 if ((ip->i_flags 479 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) && 480 securelevel > 0) 481 return (EPERM); 482 ip->i_flags = vap->va_flags; 483 } else { 484 if (ip->i_flags 485 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) || 486 (vap->va_flags & UF_SETTABLE) != vap->va_flags) 487 return (EPERM); 488 ip->i_flags &= SF_SETTABLE; 489 ip->i_flags |= (vap->va_flags & UF_SETTABLE); 490 } 491 ip->i_flag |= IN_CHANGE; 492 if (vap->va_flags & (IMMUTABLE | APPEND)) 493 return (0); 494 } 495 if (ip->i_flags & (IMMUTABLE | APPEND)) 496 return (EPERM); 497 /* 498 * Go through the fields and update iff not VNOVAL. 499 */ 500 if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) { 501 if (vp->v_mount->mnt_flag & MNT_RDONLY) 502 return (EROFS); 503 if ((error = ufs_chown(vp, vap->va_uid, vap->va_gid, cred)) != 0) 504 return (error); 505 } 506 if (vap->va_size != VNOVAL) { 507 /* 508 * Disallow write attempts on read-only filesystems; 509 * unless the file is a socket, fifo, or a block or 510 * character device resident on the filesystem. 511 */ 512 switch (vp->v_type) { 513 case VDIR: 514 return (EISDIR); 515 case VLNK: 516 case VREG: 517 if (vp->v_mount->mnt_flag & MNT_RDONLY) 518 return (EROFS); 519 break; 520 default: 521 break; 522 } 523 if ((error = ffs_truncate(vp, vap->va_size, 0, cred)) != 0) 524 return (error); 525 } 526 ip = VTOI(vp); 527 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) { 528 if (vp->v_mount->mnt_flag & MNT_RDONLY) 529 return (EROFS); 530 if (cred->cr_uid != ip->i_uid && 531 (error = priv_check_cred(cred, PRIV_ROOT, PRISON_ROOT)) && 532 ((vap->va_vaflags & VA_UTIMES_NULL) == 0 || 533 (error = VOP_ACCESS(vp, VWRITE, cred)))) 534 return (error); 535 if (vap->va_atime.tv_sec != VNOVAL) 536 ip->i_flag |= IN_ACCESS; 537 if (vap->va_mtime.tv_sec != VNOVAL) 538 ip->i_flag |= IN_CHANGE | IN_UPDATE; 539 ufs_itimes(vp); 540 if (vap->va_atime.tv_sec != VNOVAL) { 541 ip->i_atime = vap->va_atime.tv_sec; 542 ip->i_atimensec = vap->va_atime.tv_nsec; 543 } 544 if (vap->va_mtime.tv_sec != VNOVAL) { 545 ip->i_mtime = vap->va_mtime.tv_sec; 546 ip->i_mtimensec = vap->va_mtime.tv_nsec; 547 } 548 error = ffs_update(vp, 0); 549 if (error) 550 return (error); 551 } 552 error = 0; 553 if (vap->va_mode != (mode_t)VNOVAL) { 554 if (vp->v_mount->mnt_flag & MNT_RDONLY) 555 return (EROFS); 556 error = ufs_chmod(vp, (int)vap->va_mode, cred); 557 } 558 VN_KNOTE(vp, NOTE_ATTRIB); 559 return (error); 560 } 561 562 /* 563 * Change the mode on a file. 564 * Inode must be locked before calling. 565 */ 566 static int 567 ufs_chmod(struct vnode *vp, int mode, struct ucred *cred) 568 { 569 struct inode *ip = VTOI(vp); 570 int error; 571 mode_t cur_mode = ip->i_mode; 572 573 error = vop_helper_chmod(vp, mode, cred, ip->i_uid, ip->i_gid, 574 &cur_mode); 575 if (error) 576 return (error); 577 #if 0 578 if (cred->cr_uid != ip->i_uid) { 579 error = priv_check_cred(cred, PRIV_ROOT, PRISON_ROOT); 580 if (error) 581 return (error); 582 } 583 if (cred->cr_uid) { 584 if (vp->v_type != VDIR && (mode & S_ISTXT)) 585 return (EFTYPE); 586 if (!groupmember(ip->i_gid, cred) && (mode & ISGID)) 587 return (EPERM); 588 } 589 #endif 590 ip->i_mode = cur_mode; 591 ip->i_flag |= IN_CHANGE; 592 return (0); 593 } 594 595 /* 596 * Perform chown operation on inode ip; 597 * inode must be locked prior to call. 598 */ 599 static int 600 ufs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred) 601 { 602 struct inode *ip = VTOI(vp); 603 uid_t ouid; 604 gid_t ogid; 605 int error = 0; 606 #ifdef QUOTA 607 int i; 608 long change; 609 #endif 610 611 if (uid == (uid_t)VNOVAL) 612 uid = ip->i_uid; 613 if (gid == (gid_t)VNOVAL) 614 gid = ip->i_gid; 615 /* 616 * If we don't own the file, are trying to change the owner 617 * of the file, or are not a member of the target group, 618 * the caller must be superuser or the call fails. 619 */ 620 if ((cred->cr_uid != ip->i_uid || uid != ip->i_uid || 621 (gid != ip->i_gid && !(cred->cr_gid == gid || 622 groupmember((gid_t)gid, cred)))) && 623 (error = priv_check_cred(cred, PRIV_ROOT, PRISON_ROOT))) 624 return (error); 625 ogid = ip->i_gid; 626 ouid = ip->i_uid; 627 #ifdef QUOTA 628 if ((error = ufs_getinoquota(ip)) != 0) 629 return (error); 630 if (ouid == uid) { 631 ufs_dqrele(vp, ip->i_dquot[USRQUOTA]); 632 ip->i_dquot[USRQUOTA] = NODQUOT; 633 } 634 if (ogid == gid) { 635 ufs_dqrele(vp, ip->i_dquot[GRPQUOTA]); 636 ip->i_dquot[GRPQUOTA] = NODQUOT; 637 } 638 change = ip->i_blocks; 639 (void) ufs_chkdq(ip, -change, cred, CHOWN); 640 (void) ufs_chkiq(ip, -1, cred, CHOWN); 641 for (i = 0; i < MAXQUOTAS; i++) { 642 ufs_dqrele(vp, ip->i_dquot[i]); 643 ip->i_dquot[i] = NODQUOT; 644 } 645 #endif 646 ip->i_gid = gid; 647 ip->i_uid = uid; 648 #ifdef QUOTA 649 if ((error = ufs_getinoquota(ip)) == 0) { 650 if (ouid == uid) { 651 ufs_dqrele(vp, ip->i_dquot[USRQUOTA]); 652 ip->i_dquot[USRQUOTA] = NODQUOT; 653 } 654 if (ogid == gid) { 655 ufs_dqrele(vp, ip->i_dquot[GRPQUOTA]); 656 ip->i_dquot[GRPQUOTA] = NODQUOT; 657 } 658 if ((error = ufs_chkdq(ip, change, cred, CHOWN)) == 0) { 659 if ((error = ufs_chkiq(ip, 1, cred, CHOWN)) == 0) 660 goto good; 661 else 662 (void)ufs_chkdq(ip, -change, cred, CHOWN|FORCE); 663 } 664 for (i = 0; i < MAXQUOTAS; i++) { 665 ufs_dqrele(vp, ip->i_dquot[i]); 666 ip->i_dquot[i] = NODQUOT; 667 } 668 } 669 ip->i_gid = ogid; 670 ip->i_uid = ouid; 671 if (ufs_getinoquota(ip) == 0) { 672 if (ouid == uid) { 673 ufs_dqrele(vp, ip->i_dquot[USRQUOTA]); 674 ip->i_dquot[USRQUOTA] = NODQUOT; 675 } 676 if (ogid == gid) { 677 ufs_dqrele(vp, ip->i_dquot[GRPQUOTA]); 678 ip->i_dquot[GRPQUOTA] = NODQUOT; 679 } 680 (void) ufs_chkdq(ip, change, cred, FORCE|CHOWN); 681 (void) ufs_chkiq(ip, 1, cred, FORCE|CHOWN); 682 (void) ufs_getinoquota(ip); 683 } 684 return (error); 685 good: 686 if (ufs_getinoquota(ip)) 687 panic("ufs_chown: lost quota"); 688 #endif /* QUOTA */ 689 ip->i_flag |= IN_CHANGE; 690 if (cred->cr_uid != 0 && (ouid != uid || ogid != gid)) 691 ip->i_mode &= ~(ISUID | ISGID); 692 return (0); 693 } 694 695 /* 696 * Mmap a file 697 * 698 * NB Currently unsupported. 699 * 700 * ufs_mmap(struct vnode *a_vp, int a_fflags, struct ucred *a_cred) 701 */ 702 /* ARGSUSED */ 703 static 704 int 705 ufs_mmap(struct vop_mmap_args *ap) 706 { 707 return (EINVAL); 708 } 709 710 /* 711 * ufs_remove(struct vnode *a_dvp, struct vnode *a_vp, 712 * struct componentname *a_cnp) 713 */ 714 static 715 int 716 ufs_remove(struct vop_old_remove_args *ap) 717 { 718 struct inode *ip; 719 struct vnode *vp = ap->a_vp; 720 struct vnode *dvp = ap->a_dvp; 721 int error; 722 723 ip = VTOI(vp); 724 #if 0 /* handled by kernel now */ 725 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) || 726 (VTOI(dvp)->i_flags & APPEND)) { 727 error = EPERM; 728 goto out; 729 } 730 #endif 731 error = ufs_dirremove(dvp, ip, ap->a_cnp->cn_flags, 0); 732 VN_KNOTE(vp, NOTE_DELETE); 733 VN_KNOTE(dvp, NOTE_WRITE); 734 #if 0 735 out: 736 #endif 737 return (error); 738 } 739 740 /* 741 * link vnode call 742 * 743 * ufs_link(struct vnode *a_tdvp, struct vnode *a_vp, 744 * struct componentname *a_cnp) 745 */ 746 static 747 int 748 ufs_link(struct vop_old_link_args *ap) 749 { 750 struct vnode *vp = ap->a_vp; 751 struct vnode *tdvp = ap->a_tdvp; 752 struct componentname *cnp = ap->a_cnp; 753 struct inode *ip; 754 struct direct newdir; 755 int error; 756 757 if (tdvp->v_mount != vp->v_mount) { 758 error = EXDEV; 759 goto out2; 760 } 761 if (tdvp != vp && (error = vn_lock(vp, LK_EXCLUSIVE))) { 762 goto out2; 763 } 764 ip = VTOI(vp); 765 if ((nlink_t)ip->i_nlink >= LINK_MAX) { 766 error = EMLINK; 767 goto out1; 768 } 769 #if 0 /* handled by kernel now, also DragonFly allows this */ 770 if (ip->i_flags & (IMMUTABLE | APPEND)) { 771 error = EPERM; 772 goto out1; 773 } 774 #endif 775 ip->i_effnlink++; 776 ip->i_nlink++; 777 ip->i_flag |= IN_CHANGE; 778 if (DOINGSOFTDEP(vp)) 779 softdep_change_linkcnt(ip); 780 error = ffs_update(vp, !(DOINGSOFTDEP(vp) | DOINGASYNC(vp))); 781 if (!error) { 782 ufs_makedirentry(ip, cnp, &newdir); 783 error = ufs_direnter(tdvp, vp, &newdir, cnp, NULL); 784 } 785 786 if (error) { 787 ip->i_effnlink--; 788 ip->i_nlink--; 789 ip->i_flag |= IN_CHANGE; 790 if (DOINGSOFTDEP(vp)) 791 softdep_change_linkcnt(ip); 792 } 793 out1: 794 if (tdvp != vp) 795 vn_unlock(vp); 796 out2: 797 VN_KNOTE(vp, NOTE_LINK); 798 VN_KNOTE(tdvp, NOTE_WRITE); 799 return (error); 800 } 801 802 /* 803 * whiteout vnode call 804 * 805 * ufs_whiteout(struct vnode *a_dvp, struct componentname *a_cnp, int a_flags) 806 */ 807 static 808 int 809 ufs_whiteout(struct vop_old_whiteout_args *ap) 810 { 811 struct vnode *dvp = ap->a_dvp; 812 struct componentname *cnp = ap->a_cnp; 813 struct direct newdir; 814 int error = 0; 815 816 switch (ap->a_flags) { 817 case NAMEI_LOOKUP: 818 /* 4.4 format directories support whiteout operations */ 819 if (dvp->v_mount->mnt_maxsymlinklen > 0) 820 return (0); 821 return (EOPNOTSUPP); 822 823 case NAMEI_CREATE: 824 /* create a new directory whiteout */ 825 #ifdef DIAGNOSTIC 826 if (dvp->v_mount->mnt_maxsymlinklen <= 0) 827 panic("ufs_whiteout: old format filesystem"); 828 #endif 829 830 newdir.d_ino = WINO; 831 newdir.d_namlen = cnp->cn_namelen; 832 bcopy(cnp->cn_nameptr, newdir.d_name, (unsigned)cnp->cn_namelen + 1); 833 newdir.d_type = DT_WHT; 834 error = ufs_direnter(dvp, NULL, &newdir, cnp, NULL); 835 break; 836 837 case NAMEI_DELETE: 838 /* remove an existing directory whiteout */ 839 #ifdef DIAGNOSTIC 840 if (dvp->v_mount->mnt_maxsymlinklen <= 0) 841 panic("ufs_whiteout: old format filesystem"); 842 #endif 843 844 cnp->cn_flags &= ~CNP_DOWHITEOUT; 845 error = ufs_dirremove(dvp, NULL, cnp->cn_flags, 0); 846 break; 847 default: 848 panic("ufs_whiteout: unknown op"); 849 } 850 return (error); 851 } 852 853 /* 854 * Rename system call. 855 * rename("foo", "bar"); 856 * is essentially 857 * unlink("bar"); 858 * link("foo", "bar"); 859 * unlink("foo"); 860 * but ``atomically''. Can't do full commit without saving state in the 861 * inode on disk which isn't feasible at this time. Best we can do is 862 * always guarantee the target exists. 863 * 864 * Basic algorithm is: 865 * 866 * 1) Bump link count on source while we're linking it to the 867 * target. This also ensure the inode won't be deleted out 868 * from underneath us while we work (it may be truncated by 869 * a concurrent `trunc' or `open' for creation). 870 * 2) Link source to destination. If destination already exists, 871 * delete it first. 872 * 3) Unlink source reference to inode if still around. If a 873 * directory was moved and the parent of the destination 874 * is different from the source, patch the ".." entry in the 875 * directory. 876 * 877 * ufs_rename(struct vnode *a_fdvp, struct vnode *a_fvp, 878 * struct componentname *a_fcnp, struct vnode *a_tdvp, 879 * struct vnode *a_tvp, struct componentname *a_tcnp) 880 */ 881 static 882 int 883 ufs_rename(struct vop_old_rename_args *ap) 884 { 885 struct vnode *tvp = ap->a_tvp; 886 struct vnode *tdvp = ap->a_tdvp; 887 struct vnode *fvp = ap->a_fvp; 888 struct vnode *fdvp = ap->a_fdvp; 889 struct componentname *tcnp = ap->a_tcnp; 890 struct componentname *fcnp = ap->a_fcnp; 891 struct inode *ip, *xp, *dp; 892 struct direct newdir; 893 ino_t oldparent = 0, newparent = 0; 894 int doingdirectory = 0; 895 int error = 0, ioflag; 896 897 /* 898 * Check for cross-device rename. 899 */ 900 if ((fvp->v_mount != tdvp->v_mount) || 901 (tvp && (fvp->v_mount != tvp->v_mount))) { 902 error = EXDEV; 903 abortit: 904 if (tdvp == tvp) 905 vrele(tdvp); 906 else 907 vput(tdvp); 908 if (tvp) 909 vput(tvp); 910 vrele(fdvp); 911 vrele(fvp); 912 return (error); 913 } 914 915 #if 0 /* handled by kernel now */ 916 if (tvp && ((VTOI(tvp)->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) || 917 (VTOI(tdvp)->i_flags & APPEND))) { 918 error = EPERM; 919 goto abortit; 920 } 921 #endif 922 923 /* 924 * Renaming a file to itself has no effect. The upper layers should 925 * not call us in that case. Temporarily just warn if they do. 926 */ 927 if (fvp == tvp) { 928 kprintf("ufs_rename: fvp == tvp (can't happen)\n"); 929 error = 0; 930 goto abortit; 931 } 932 933 if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0) 934 goto abortit; 935 936 /* 937 * Note: now that fvp is locked we have to be sure to unlock it before 938 * using the 'abortit' target. 939 */ 940 dp = VTOI(fdvp); 941 ip = VTOI(fvp); 942 if (ip->i_nlink >= LINK_MAX) { 943 vn_unlock(fvp); 944 error = EMLINK; 945 goto abortit; 946 } 947 #if 0 /* handled by kernel now */ 948 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) 949 || (dp->i_flags & APPEND)) { 950 vn_unlock(fvp); 951 error = EPERM; 952 goto abortit; 953 } 954 #endif 955 if ((ip->i_mode & IFMT) == IFDIR) { 956 /* 957 * Avoid ".", "..", and aliases of "." for obvious reasons. 958 */ 959 if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') || 960 dp == ip || (fcnp->cn_flags | tcnp->cn_flags) & CNP_ISDOTDOT || 961 (ip->i_flag & IN_RENAME)) { 962 vn_unlock(fvp); 963 error = EINVAL; 964 goto abortit; 965 } 966 ip->i_flag |= IN_RENAME; 967 oldparent = dp->i_number; 968 doingdirectory = 1; 969 } 970 VN_KNOTE(fdvp, NOTE_WRITE); /* XXX right place? */ 971 972 /* 973 * fvp still locked. ip->i_flag has IN_RENAME set if doingdirectory. 974 * Cleanup fvp requirements so we can unlock it. 975 * 976 * tvp and tdvp are locked. tvp may be NULL. Now that dp and xp 977 * is setup we can use the 'bad' target if we unlock fvp. We cannot 978 * use the abortit target anymore because of IN_RENAME. 979 */ 980 dp = VTOI(tdvp); 981 if (tvp) 982 xp = VTOI(tvp); 983 else 984 xp = NULL; 985 986 /* 987 * 1) Bump link count while we're moving stuff 988 * around. If we crash somewhere before 989 * completing our work, the link count 990 * may be wrong, but correctable. 991 */ 992 ip->i_effnlink++; 993 ip->i_nlink++; 994 ip->i_flag |= IN_CHANGE; 995 if (DOINGSOFTDEP(fvp)) 996 softdep_change_linkcnt(ip); 997 if ((error = ffs_update(fvp, !(DOINGSOFTDEP(fvp) | 998 DOINGASYNC(fvp)))) != 0) { 999 vn_unlock(fvp); 1000 goto bad; 1001 } 1002 1003 /* 1004 * If ".." must be changed (ie the directory gets a new 1005 * parent) then the source directory must not be in the 1006 * directory heirarchy above the target, as this would 1007 * orphan everything below the source directory. Also 1008 * the user must have write permission in the source so 1009 * as to be able to change "..". We must repeat the call 1010 * to namei, as the parent directory is unlocked by the 1011 * call to checkpath(). 1012 */ 1013 error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred); 1014 vn_unlock(fvp); 1015 1016 /* 1017 * We are now back to where we were in that fvp, fdvp are unlocked 1018 * and tvp, tdvp are locked. tvp may be NULL. IN_RENAME may be 1019 * set. Only the bad target or, if we clean up tvp and tdvp, the 1020 * out target, may be used. 1021 */ 1022 if (oldparent != dp->i_number) 1023 newparent = dp->i_number; 1024 if (doingdirectory && newparent) { 1025 if (error) /* write access check above */ 1026 goto bad; 1027 1028 /* 1029 * Once we start messing with tvp and tdvp we cannot use the 1030 * 'bad' target, only finish cleaning tdvp and tvp up and 1031 * use the 'out' target. 1032 * 1033 * This cleans up tvp. 1034 */ 1035 if (xp != NULL) { 1036 vput(tvp); 1037 xp = NULL; 1038 } 1039 1040 /* 1041 * This is a real mess. ufs_checkpath vput's the target 1042 * directory so retain an extra ref and note that tdvp will 1043 * lose its lock on return. This leaves us with one good 1044 * ref after ufs_checkpath returns. 1045 */ 1046 vref(tdvp); 1047 error = ufs_checkpath(ip, dp, tcnp->cn_cred); 1048 tcnp->cn_flags |= CNP_PDIRUNLOCK; 1049 if (error) { 1050 vrele(tdvp); 1051 goto out; 1052 } 1053 1054 /* 1055 * relookup no longer messes with tdvp's refs. tdvp must be 1056 * unlocked on entry and will be locked on a successful 1057 * return. 1058 */ 1059 error = relookup(tdvp, &tvp, tcnp); 1060 if (error) { 1061 if (tcnp->cn_flags & CNP_PDIRUNLOCK) 1062 vrele(tdvp); 1063 else 1064 vput(tdvp); 1065 goto out; 1066 } 1067 KKASSERT((tcnp->cn_flags & CNP_PDIRUNLOCK) == 0); 1068 dp = VTOI(tdvp); 1069 if (tvp) 1070 xp = VTOI(tvp); 1071 } 1072 1073 /* 1074 * We are back to fvp, fdvp unlocked, tvp, tdvp locked. tvp may 1075 * be NULL (xp will also be NULL in that case), and IN_RENAME will 1076 * be set if doingdirectory. This means we can use the 'bad' target 1077 * again. 1078 */ 1079 1080 /* 1081 * 2) If target doesn't exist, link the target 1082 * to the source and unlink the source. 1083 * Otherwise, rewrite the target directory 1084 * entry to reference the source inode and 1085 * expunge the original entry's existence. 1086 */ 1087 if (xp == NULL) { 1088 if (dp->i_dev != ip->i_dev) 1089 panic("ufs_rename: EXDEV"); 1090 /* 1091 * Account for ".." in new directory. 1092 * When source and destination have the same 1093 * parent we don't fool with the link count. 1094 */ 1095 if (doingdirectory && newparent) { 1096 if ((nlink_t)dp->i_nlink >= LINK_MAX) { 1097 error = EMLINK; 1098 goto bad; 1099 } 1100 dp->i_effnlink++; 1101 dp->i_nlink++; 1102 dp->i_flag |= IN_CHANGE; 1103 if (DOINGSOFTDEP(tdvp)) 1104 softdep_change_linkcnt(dp); 1105 error = ffs_update(tdvp, !(DOINGSOFTDEP(tdvp) | 1106 DOINGASYNC(tdvp))); 1107 if (error) 1108 goto bad; 1109 } 1110 ufs_makedirentry(ip, tcnp, &newdir); 1111 error = ufs_direnter(tdvp, NULL, &newdir, tcnp, NULL); 1112 if (error) { 1113 if (doingdirectory && newparent) { 1114 dp->i_effnlink--; 1115 dp->i_nlink--; 1116 dp->i_flag |= IN_CHANGE; 1117 if (DOINGSOFTDEP(tdvp)) 1118 softdep_change_linkcnt(dp); 1119 (void)ffs_update(tdvp, 1); 1120 } 1121 goto bad; 1122 } 1123 VN_KNOTE(tdvp, NOTE_WRITE); 1124 vput(tdvp); 1125 } else { 1126 if (xp->i_dev != dp->i_dev || xp->i_dev != ip->i_dev) 1127 panic("ufs_rename: EXDEV"); 1128 /* 1129 * Short circuit rename(foo, foo). 1130 */ 1131 if (xp->i_number == ip->i_number) 1132 panic("ufs_rename: same file"); 1133 /* 1134 * If the parent directory is "sticky", then the user must 1135 * own the parent directory, or the destination of the rename, 1136 * otherwise the destination may not be changed (except by 1137 * root). This implements append-only directories. 1138 */ 1139 if ((dp->i_mode & S_ISTXT) && tcnp->cn_cred->cr_uid != 0 && 1140 tcnp->cn_cred->cr_uid != dp->i_uid && 1141 xp->i_uid != tcnp->cn_cred->cr_uid) { 1142 error = EPERM; 1143 goto bad; 1144 } 1145 /* 1146 * Target must be empty if a directory and have no links 1147 * to it. Also, ensure source and target are compatible 1148 * (both directories, or both not directories). 1149 * 1150 * Purge the file or directory being replaced from the 1151 * nameccache. 1152 */ 1153 if ((xp->i_mode&IFMT) == IFDIR) { 1154 if ((xp->i_effnlink > 2) || 1155 !ufs_dirempty(xp, dp->i_number, tcnp->cn_cred)) { 1156 error = ENOTEMPTY; 1157 goto bad; 1158 } 1159 if (!doingdirectory) { 1160 error = ENOTDIR; 1161 goto bad; 1162 } 1163 /* cache_purge removed - handled by VFS compat layer */ 1164 } else if (doingdirectory == 0) { 1165 /* cache_purge removed - handled by VFS compat layer */ 1166 } else { 1167 error = EISDIR; 1168 goto bad; 1169 } 1170 /* 1171 * note: inode passed to ufs_dirrewrite() is 0 for a 1172 * non-directory file rename, 1 for a directory rename 1173 * in the same directory, and > 1 for an inode representing 1174 * the new directory. 1175 */ 1176 error = ufs_dirrewrite(dp, xp, ip->i_number, 1177 IFTODT(ip->i_mode), 1178 (doingdirectory && newparent) ? 1179 newparent : (ino_t)doingdirectory); 1180 if (error) 1181 goto bad; 1182 if (doingdirectory) { 1183 if (!newparent) { 1184 dp->i_effnlink--; 1185 if (DOINGSOFTDEP(tdvp)) 1186 softdep_change_linkcnt(dp); 1187 } 1188 xp->i_effnlink--; 1189 if (DOINGSOFTDEP(tvp)) 1190 softdep_change_linkcnt(xp); 1191 } 1192 if (doingdirectory && !DOINGSOFTDEP(tvp)) { 1193 /* 1194 * Truncate inode. The only stuff left in the directory 1195 * is "." and "..". The "." reference is inconsequential 1196 * since we are quashing it. We have removed the "." 1197 * reference and the reference in the parent directory, 1198 * but there may be other hard links. The soft 1199 * dependency code will arrange to do these operations 1200 * after the parent directory entry has been deleted on 1201 * disk, so when running with that code we avoid doing 1202 * them now. 1203 */ 1204 if (!newparent) { 1205 dp->i_nlink--; 1206 dp->i_flag |= IN_CHANGE; 1207 } 1208 xp->i_nlink--; 1209 xp->i_flag |= IN_CHANGE; 1210 ioflag = DOINGASYNC(tvp) ? 0 : IO_SYNC; 1211 error = ffs_truncate(tvp, (off_t)0, ioflag, 1212 tcnp->cn_cred); 1213 if (error) 1214 goto bad; 1215 } 1216 VN_KNOTE(tdvp, NOTE_WRITE); 1217 vput(tdvp); 1218 VN_KNOTE(tvp, NOTE_DELETE); 1219 vput(tvp); 1220 xp = NULL; 1221 } 1222 1223 /* 1224 * tvp and tdvp have been cleaned up. only fvp and fdvp (both 1225 * unlocked) remain. We are about to overwrite fvp but we have to 1226 * keep 'ip' intact so we cannot release the old fvp, which is still 1227 * refd and accessible via ap->a_fvp. 1228 * 1229 * This means we cannot use either 'bad' or 'out' to cleanup any 1230 * more. 1231 */ 1232 1233 /* 1234 * 3) Unlink the source. 1235 */ 1236 fcnp->cn_flags &= ~CNP_MODMASK; 1237 fcnp->cn_flags |= CNP_LOCKPARENT; 1238 error = relookup(fdvp, &fvp, fcnp); 1239 if (error || fvp == NULL) { 1240 /* 1241 * From name has disappeared. IN_RENAME will not be set if 1242 * we get past the panic so we don't have to clean it up. 1243 */ 1244 if (doingdirectory) 1245 panic("ufs_rename: lost dir entry"); 1246 vrele(ap->a_fvp); 1247 if (fcnp->cn_flags & CNP_PDIRUNLOCK) 1248 vrele(fdvp); 1249 else 1250 vput(fdvp); 1251 return(0); 1252 } 1253 KKASSERT((fcnp->cn_flags & CNP_PDIRUNLOCK) == 0); 1254 1255 /* 1256 * fdvp and fvp are locked. 1257 */ 1258 xp = VTOI(fvp); 1259 dp = VTOI(fdvp); 1260 1261 /* 1262 * Ensure that the directory entry still exists and has not 1263 * changed while the new name has been entered. If the source is 1264 * a file then the entry may have been unlinked or renamed. In 1265 * either case there is no further work to be done. If the source 1266 * is a directory then it cannot have been rmdir'ed; the IN_RENAME 1267 * flag ensures that it cannot be moved by another rename or removed 1268 * by a rmdir. Cleanup IN_RENAME. 1269 */ 1270 if (xp != ip) { 1271 if (doingdirectory) 1272 panic("ufs_rename: lost dir entry"); 1273 } else { 1274 /* 1275 * If the source is a directory with a 1276 * new parent, the link count of the old 1277 * parent directory must be decremented 1278 * and ".." set to point to the new parent. 1279 */ 1280 if (doingdirectory && newparent) { 1281 xp->i_offset = mastertemplate.dot_reclen; 1282 ufs_dirrewrite(xp, dp, newparent, DT_DIR, 0); 1283 /* cache_purge removed - handled by VFS compat layer */ 1284 } 1285 error = ufs_dirremove(fdvp, xp, fcnp->cn_flags, 0); 1286 xp->i_flag &= ~IN_RENAME; 1287 } 1288 1289 VN_KNOTE(fvp, NOTE_RENAME); 1290 vput(fdvp); 1291 vput(fvp); 1292 vrele(ap->a_fvp); 1293 return (error); 1294 1295 bad: 1296 if (xp) 1297 vput(ITOV(xp)); 1298 vput(ITOV(dp)); 1299 out: 1300 if (doingdirectory) 1301 ip->i_flag &= ~IN_RENAME; 1302 if (vn_lock(fvp, LK_EXCLUSIVE) == 0) { 1303 ip->i_effnlink--; 1304 ip->i_nlink--; 1305 ip->i_flag |= IN_CHANGE; 1306 ip->i_flag &= ~IN_RENAME; 1307 if (DOINGSOFTDEP(fvp)) 1308 softdep_change_linkcnt(ip); 1309 vput(fvp); 1310 } else { 1311 vrele(fvp); 1312 } 1313 return (error); 1314 } 1315 1316 /* 1317 * Mkdir system call 1318 * 1319 * ufs_mkdir(struct vnode *a_dvp, struct vnode **a_vpp, 1320 * struct componentname *a_cnp, struct vattr *a_vap) 1321 */ 1322 static 1323 int 1324 ufs_mkdir(struct vop_old_mkdir_args *ap) 1325 { 1326 struct vnode *dvp = ap->a_dvp; 1327 struct vattr *vap = ap->a_vap; 1328 struct componentname *cnp = ap->a_cnp; 1329 struct inode *ip, *dp; 1330 struct vnode *tvp; 1331 struct buf *bp; 1332 struct dirtemplate dirtemplate, *dtp; 1333 struct direct newdir; 1334 int error, dmode; 1335 long blkoff; 1336 1337 dp = VTOI(dvp); 1338 if ((nlink_t)dp->i_nlink >= LINK_MAX) { 1339 error = EMLINK; 1340 goto out; 1341 } 1342 dmode = vap->va_mode & 0777; 1343 dmode |= IFDIR; 1344 /* 1345 * Must simulate part of ufs_makeinode here to acquire the inode, 1346 * but not have it entered in the parent directory. The entry is 1347 * made later after writing "." and ".." entries. 1348 */ 1349 error = ffs_valloc(dvp, dmode, cnp->cn_cred, &tvp); 1350 if (error) 1351 goto out; 1352 ip = VTOI(tvp); 1353 ip->i_gid = dp->i_gid; 1354 #ifdef SUIDDIR 1355 { 1356 #ifdef QUOTA 1357 struct ucred ucred, *ucp; 1358 ucp = cnp->cn_cred; 1359 #endif 1360 /* 1361 * If we are hacking owners here, (only do this where told to) 1362 * and we are not giving it TO root, (would subvert quotas) 1363 * then go ahead and give it to the other user. 1364 * The new directory also inherits the SUID bit. 1365 * If user's UID and dir UID are the same, 1366 * 'give it away' so that the SUID is still forced on. 1367 */ 1368 if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) && 1369 (dp->i_mode & ISUID) && dp->i_uid) { 1370 dmode |= ISUID; 1371 ip->i_uid = dp->i_uid; 1372 #ifdef QUOTA 1373 if (dp->i_uid != cnp->cn_cred->cr_uid) { 1374 /* 1375 * Make sure the correct user gets charged 1376 * for the space. 1377 * Make a dummy credential for the victim. 1378 * XXX This seems to never be accessed out of 1379 * our context so a stack variable is ok. 1380 */ 1381 ucred.cr_ref = 1; 1382 ucred.cr_uid = ip->i_uid; 1383 ucred.cr_ngroups = 1; 1384 ucred.cr_groups[0] = dp->i_gid; 1385 ucp = &ucred; 1386 } 1387 #endif 1388 } else 1389 ip->i_uid = cnp->cn_cred->cr_uid; 1390 #ifdef QUOTA 1391 if ((error = ufs_getinoquota(ip)) || 1392 (error = ufs_chkiq(ip, 1, ucp, 0))) { 1393 ffs_vfree(tvp, ip->i_number, dmode); 1394 vput(tvp); 1395 return (error); 1396 } 1397 #endif 1398 } 1399 #else /* !SUIDDIR */ 1400 ip->i_uid = cnp->cn_cred->cr_uid; 1401 #ifdef QUOTA 1402 if ((error = ufs_getinoquota(ip)) || 1403 (error = ufs_chkiq(ip, 1, cnp->cn_cred, 0))) { 1404 ffs_vfree(tvp, ip->i_number, dmode); 1405 vput(tvp); 1406 return (error); 1407 } 1408 #endif 1409 #endif /* !SUIDDIR */ 1410 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 1411 ip->i_mode = dmode; 1412 tvp->v_type = VDIR; /* Rest init'd in getnewvnode(). */ 1413 ip->i_effnlink = 2; 1414 ip->i_nlink = 2; 1415 if (DOINGSOFTDEP(tvp)) 1416 softdep_change_linkcnt(ip); 1417 if (cnp->cn_flags & CNP_ISWHITEOUT) 1418 ip->i_flags |= UF_OPAQUE; 1419 1420 /* 1421 * Bump link count in parent directory to reflect work done below. 1422 * Should be done before reference is created so cleanup is 1423 * possible if we crash. 1424 */ 1425 dp->i_effnlink++; 1426 dp->i_nlink++; 1427 dp->i_flag |= IN_CHANGE; 1428 if (DOINGSOFTDEP(dvp)) 1429 softdep_change_linkcnt(dp); 1430 error = ffs_update(tvp, !(DOINGSOFTDEP(dvp) | DOINGASYNC(dvp))); 1431 if (error) 1432 goto bad; 1433 1434 /* 1435 * The vnode must have a VM object in order to issue buffer cache 1436 * ops on it. 1437 */ 1438 vinitvmio(tvp, DIRBLKSIZ); 1439 1440 /* 1441 * Initialize directory with "." and ".." from static template. 1442 */ 1443 if (dvp->v_mount->mnt_maxsymlinklen > 0) 1444 dtp = &mastertemplate; 1445 else 1446 dtp = (struct dirtemplate *)&omastertemplate; 1447 dirtemplate = *dtp; 1448 dirtemplate.dot_ino = ip->i_number; 1449 dirtemplate.dotdot_ino = dp->i_number; 1450 vnode_pager_setsize(tvp, DIRBLKSIZ); 1451 error = VOP_BALLOC(tvp, 0LL, DIRBLKSIZ, cnp->cn_cred, B_CLRBUF, &bp); 1452 if (error) 1453 goto bad; 1454 ip->i_size = DIRBLKSIZ; 1455 ip->i_flag |= IN_CHANGE | IN_UPDATE; 1456 bcopy((caddr_t)&dirtemplate, (caddr_t)bp->b_data, sizeof dirtemplate); 1457 if (DOINGSOFTDEP(tvp)) { 1458 /* 1459 * Ensure that the entire newly allocated block is a 1460 * valid directory so that future growth within the 1461 * block does not have to ensure that the block is 1462 * written before the inode. 1463 */ 1464 blkoff = DIRBLKSIZ; 1465 while (blkoff < bp->b_bcount) { 1466 ((struct direct *) 1467 (bp->b_data + blkoff))->d_reclen = DIRBLKSIZ; 1468 blkoff += DIRBLKSIZ; 1469 } 1470 } 1471 if ((error = ffs_update(tvp, !(DOINGSOFTDEP(tvp) | 1472 DOINGASYNC(tvp)))) != 0) { 1473 bwrite(bp); 1474 goto bad; 1475 } 1476 /* 1477 * Directory set up, now install its entry in the parent directory. 1478 * 1479 * If we are not doing soft dependencies, then we must write out the 1480 * buffer containing the new directory body before entering the new 1481 * name in the parent. If we are doing soft dependencies, then the 1482 * buffer containing the new directory body will be passed to and 1483 * released in the soft dependency code after the code has attached 1484 * an appropriate ordering dependency to the buffer which ensures that 1485 * the buffer is written before the new name is written in the parent. 1486 */ 1487 if (DOINGASYNC(dvp)) 1488 bdwrite(bp); 1489 else if (!DOINGSOFTDEP(dvp) && (error = bwrite(bp)) != 0) 1490 goto bad; 1491 ufs_makedirentry(ip, cnp, &newdir); 1492 error = ufs_direnter(dvp, tvp, &newdir, cnp, bp); 1493 1494 bad: 1495 if (error == 0) { 1496 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK); 1497 *ap->a_vpp = tvp; 1498 } else { 1499 dp->i_effnlink--; 1500 dp->i_nlink--; 1501 dp->i_flag |= IN_CHANGE; 1502 if (DOINGSOFTDEP(dvp)) 1503 softdep_change_linkcnt(dp); 1504 /* 1505 * No need to do an explicit VOP_TRUNCATE here, vrele will 1506 * do this for us because we set the link count to 0. 1507 */ 1508 ip->i_effnlink = 0; 1509 ip->i_nlink = 0; 1510 ip->i_flag |= IN_CHANGE; 1511 if (DOINGSOFTDEP(tvp)) 1512 softdep_change_linkcnt(ip); 1513 vput(tvp); 1514 } 1515 out: 1516 return (error); 1517 } 1518 1519 /* 1520 * Rmdir system call. 1521 * 1522 * ufs_rmdir(struct vnode *a_dvp, struct vnode *a_vp, 1523 * struct componentname *a_cnp) 1524 */ 1525 static 1526 int 1527 ufs_rmdir(struct vop_old_rmdir_args *ap) 1528 { 1529 struct vnode *vp = ap->a_vp; 1530 struct vnode *dvp = ap->a_dvp; 1531 struct componentname *cnp = ap->a_cnp; 1532 struct inode *ip, *dp; 1533 int error, ioflag; 1534 1535 ip = VTOI(vp); 1536 dp = VTOI(dvp); 1537 1538 /* 1539 * Do not remove a directory that is in the process of being renamed. 1540 * Verify the directory is empty (and valid). Rmdir ".." will not be 1541 * valid since ".." will contain a reference to the current directory 1542 * and thus be non-empty. Do not allow the removal of mounted on 1543 * directories (this can happen when an NFS exported filesystem 1544 * tries to remove a locally mounted on directory). 1545 */ 1546 error = 0; 1547 if (ip->i_flag & IN_RENAME) { 1548 error = EINVAL; 1549 goto out; 1550 } 1551 if (ip->i_effnlink != 2 || 1552 !ufs_dirempty(ip, dp->i_number, cnp->cn_cred)) { 1553 error = ENOTEMPTY; 1554 goto out; 1555 } 1556 #if 0 /* handled by kernel now */ 1557 if ((dp->i_flags & APPEND) 1558 || (ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))) { 1559 error = EPERM; 1560 goto out; 1561 } 1562 #endif 1563 /* 1564 * Delete reference to directory before purging 1565 * inode. If we crash in between, the directory 1566 * will be reattached to lost+found, 1567 */ 1568 dp->i_effnlink--; 1569 ip->i_effnlink--; 1570 if (DOINGSOFTDEP(vp)) { 1571 softdep_change_linkcnt(dp); 1572 softdep_change_linkcnt(ip); 1573 } 1574 error = ufs_dirremove(dvp, ip, cnp->cn_flags, 1); 1575 if (error) { 1576 dp->i_effnlink++; 1577 ip->i_effnlink++; 1578 if (DOINGSOFTDEP(vp)) { 1579 softdep_change_linkcnt(dp); 1580 softdep_change_linkcnt(ip); 1581 } 1582 goto out; 1583 } 1584 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK); 1585 /* 1586 * Truncate inode. The only stuff left in the directory is "." and 1587 * "..". The "." reference is inconsequential since we are quashing 1588 * it. The soft dependency code will arrange to do these operations 1589 * after the parent directory entry has been deleted on disk, so 1590 * when running with that code we avoid doing them now. 1591 */ 1592 if (!DOINGSOFTDEP(vp)) { 1593 dp->i_nlink--; 1594 dp->i_flag |= IN_CHANGE; 1595 ip->i_nlink--; 1596 ip->i_flag |= IN_CHANGE; 1597 ioflag = DOINGASYNC(vp) ? 0 : IO_SYNC; 1598 error = ffs_truncate(vp, (off_t)0, ioflag, cnp->cn_cred); 1599 } 1600 /* cache_purge removed - handled by VFS compat layer */ 1601 #ifdef UFS_DIRHASH 1602 /* Kill any active hash; i_effnlink == 0, so it will not come back. */ 1603 if (ip->i_dirhash != NULL) 1604 ufsdirhash_free(ip); 1605 #endif 1606 out: 1607 VN_KNOTE(vp, NOTE_DELETE); 1608 return (error); 1609 } 1610 1611 /* 1612 * symlink -- make a symbolic link 1613 * 1614 * ufs_symlink(struct vnode *a_dvp, struct vnode **a_vpp, 1615 * struct componentname *a_cnp, struct vattr *a_vap, 1616 * char *a_target) 1617 */ 1618 static 1619 int 1620 ufs_symlink(struct vop_old_symlink_args *ap) 1621 { 1622 struct vnode *vp, **vpp = ap->a_vpp; 1623 struct inode *ip; 1624 int len, error; 1625 1626 error = ufs_makeinode(IFLNK | ap->a_vap->va_mode, ap->a_dvp, 1627 vpp, ap->a_cnp); 1628 if (error) 1629 return (error); 1630 VN_KNOTE(ap->a_dvp, NOTE_WRITE); 1631 vp = *vpp; 1632 len = strlen(ap->a_target); 1633 if (len < vp->v_mount->mnt_maxsymlinklen) { 1634 ip = VTOI(vp); 1635 bcopy(ap->a_target, (char *)ip->i_shortlink, len); 1636 ip->i_size = len; 1637 ip->i_flag |= IN_CHANGE | IN_UPDATE; 1638 } else { 1639 /* 1640 * Make sure we have a VM object in order to use 1641 * the buffer cache. 1642 */ 1643 if (vp->v_object == NULL) 1644 vinitvmio(vp, 0); 1645 error = vn_rdwr(UIO_WRITE, vp, ap->a_target, len, (off_t)0, 1646 UIO_SYSSPACE, IO_NODELOCKED, 1647 ap->a_cnp->cn_cred, NULL); 1648 } 1649 if (error) 1650 vput(vp); 1651 return (error); 1652 } 1653 1654 /* 1655 * Vnode op for reading directories. 1656 * 1657 * ufs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred, 1658 * int *a_eofflag, int *ncookies, off_t **a_cookies) 1659 */ 1660 static 1661 int 1662 ufs_readdir(struct vop_readdir_args *ap) 1663 { 1664 struct uio *uio = ap->a_uio; 1665 struct vnode *vp = ap->a_vp; 1666 struct direct *dp; 1667 struct buf *bp; 1668 int retval; 1669 int error; 1670 int offset; /* offset into buffer cache buffer */ 1671 int eoffset; /* end of buffer clipped to file EOF */ 1672 int pickup; /* pickup point */ 1673 int ncookies; 1674 int cookie_index; 1675 off_t *cookies; 1676 1677 if (uio->uio_offset < 0) 1678 return (EINVAL); 1679 /* 1680 * Guess the number of cookies needed. Make sure we compute at 1681 * least 1, and no more then a reasonable limit. 1682 */ 1683 if (ap->a_ncookies) { 1684 ncookies = uio->uio_resid / 16 + 1; 1685 if (ncookies > 1024) 1686 ncookies = 1024; 1687 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK); 1688 } else { 1689 ncookies = -1; /* force conditionals below */ 1690 cookies = NULL; 1691 } 1692 cookie_index = 0; 1693 1694 if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0) 1695 return (error); 1696 1697 /* 1698 * Past or at EOF 1699 */ 1700 if (uio->uio_offset >= VTOI(vp)->i_size) { 1701 if (ap->a_eofflag) 1702 *ap->a_eofflag = 1; 1703 if (ap->a_ncookies) { 1704 *ap->a_ncookies = cookie_index; 1705 *ap->a_cookies = cookies; 1706 } 1707 goto done; 1708 } 1709 1710 /* 1711 * Loop until we run out of cookies, we run out of user buffer, 1712 * or we hit the directory EOF. 1713 * 1714 * Always start scans at the beginning of the buffer, don't trust 1715 * the offset supplied by userland. 1716 */ 1717 while ((error = ffs_blkatoff_ra(vp, uio->uio_offset, NULL, &bp, 2)) == 0) { 1718 pickup = (int)(uio->uio_offset - bp->b_loffset); 1719 offset = 0; 1720 retval = 0; 1721 if (bp->b_loffset + bp->b_bcount > VTOI(vp)->i_size) 1722 eoffset = (int)(VTOI(vp)->i_size - bp->b_loffset); 1723 else 1724 eoffset = bp->b_bcount; 1725 1726 while (offset < eoffset) { 1727 dp = (struct direct *)(bp->b_data + offset); 1728 if (dp->d_reclen <= 0 || (dp->d_reclen & 3) || 1729 offset + dp->d_reclen > bp->b_bcount) { 1730 error = EIO; 1731 break; 1732 } 1733 if (offsetof(struct direct, d_name[dp->d_namlen]) > dp->d_reclen) { 1734 error = EIO; 1735 break; 1736 } 1737 if (offset < pickup) { 1738 offset += dp->d_reclen; 1739 continue; 1740 } 1741 #if BYTE_ORDER == LITTLE_ENDIAN 1742 if (OFSFMT(vp)) { 1743 retval = vop_write_dirent(&error, uio, 1744 dp->d_ino, dp->d_namlen, dp->d_type, 1745 dp->d_name); 1746 } else 1747 #endif 1748 { 1749 retval = vop_write_dirent(&error, uio, 1750 dp->d_ino, dp->d_type, dp->d_namlen, 1751 dp->d_name); 1752 } 1753 if (retval) 1754 break; 1755 if (cookies) 1756 cookies[cookie_index] = bp->b_loffset + offset; 1757 ++cookie_index; 1758 offset += dp->d_reclen; 1759 if (cookie_index == ncookies) 1760 break; 1761 } 1762 1763 /* 1764 * This will align the next loop to the beginning of the 1765 * next block, and pickup will calculate to 0. 1766 */ 1767 uio->uio_offset = bp->b_loffset + offset; 1768 brelse(bp); 1769 1770 if (retval || error || cookie_index == ncookies || 1771 uio->uio_offset >= VTOI(vp)->i_size) { 1772 break; 1773 } 1774 } 1775 if (ap->a_eofflag) 1776 *ap->a_eofflag = VTOI(vp)->i_size <= uio->uio_offset; 1777 1778 /* 1779 * Report errors only if we didn't manage to read anything 1780 */ 1781 if (error && cookie_index == 0) { 1782 if (cookies) { 1783 kfree(cookies, M_TEMP); 1784 *ap->a_ncookies = 0; 1785 *ap->a_cookies = NULL; 1786 } 1787 } else { 1788 error = 0; 1789 if (cookies) { 1790 *ap->a_ncookies = cookie_index; 1791 *ap->a_cookies = cookies; 1792 } 1793 } 1794 done: 1795 vn_unlock(vp); 1796 return (error); 1797 } 1798 1799 /* 1800 * Return target name of a symbolic link 1801 * 1802 * ufs_readlink(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred) 1803 */ 1804 static 1805 int 1806 ufs_readlink(struct vop_readlink_args *ap) 1807 { 1808 struct vnode *vp = ap->a_vp; 1809 struct inode *ip = VTOI(vp); 1810 int isize; 1811 1812 isize = ip->i_size; 1813 if ((isize < vp->v_mount->mnt_maxsymlinklen) || 1814 (ip->i_din.di_blocks == 0)) { /* XXX - for old fastlink support */ 1815 uiomove((char *)ip->i_shortlink, isize, ap->a_uio); 1816 return (0); 1817 } 1818 1819 /* 1820 * Perform the equivalent of an OPEN on vp so we can issue a 1821 * VOP_READ. 1822 */ 1823 return (VOP_READ(vp, ap->a_uio, 0, ap->a_cred)); 1824 } 1825 1826 /* 1827 * Calculate the logical to physical mapping if not done already, 1828 * then call the device strategy routine. 1829 * 1830 * In order to be able to swap to a file, the VOP_BMAP operation may not 1831 * deadlock on memory. See ufs_bmap() for details. 1832 * 1833 * ufs_strategy(struct vnode *a_vp, struct bio *a_bio) 1834 */ 1835 static 1836 int 1837 ufs_strategy(struct vop_strategy_args *ap) 1838 { 1839 struct bio *bio = ap->a_bio; 1840 struct bio *nbio; 1841 struct buf *bp = bio->bio_buf; 1842 struct vnode *vp = ap->a_vp; 1843 struct inode *ip; 1844 int error; 1845 1846 ip = VTOI(vp); 1847 if (vp->v_type == VBLK || vp->v_type == VCHR) 1848 panic("ufs_strategy: spec"); 1849 nbio = push_bio(bio); 1850 if (nbio->bio_offset == NOOFFSET) { 1851 error = VOP_BMAP(vp, bio->bio_offset, &nbio->bio_offset, 1852 NULL, NULL, bp->b_cmd); 1853 if (error) { 1854 bp->b_error = error; 1855 bp->b_flags |= B_ERROR; 1856 /* I/O was never started on nbio, must biodone(bio) */ 1857 biodone(bio); 1858 return (error); 1859 } 1860 if (nbio->bio_offset == NOOFFSET) 1861 vfs_bio_clrbuf(bp); 1862 } 1863 if (nbio->bio_offset == NOOFFSET) { 1864 /* 1865 * We hit a hole in the file. The buffer has been zero-filled 1866 * so just biodone() it. 1867 */ 1868 biodone(bio); 1869 } else { 1870 vn_strategy(ip->i_devvp, nbio); 1871 } 1872 return (0); 1873 } 1874 1875 /* 1876 * Print out the contents of an inode. 1877 * 1878 * ufs_print(struct vnode *a_vp) 1879 */ 1880 static 1881 int 1882 ufs_print(struct vop_print_args *ap) 1883 { 1884 struct vnode *vp = ap->a_vp; 1885 struct inode *ip = VTOI(vp); 1886 1887 kprintf("tag VT_UFS, ino %lu, on dev %s (%d, %d)", 1888 (u_long)ip->i_number, devtoname(ip->i_dev), major(ip->i_dev), 1889 minor(ip->i_dev)); 1890 if (vp->v_type == VFIFO) 1891 fifo_printinfo(vp); 1892 lockmgr_printinfo(&vp->v_lock); 1893 kprintf("\n"); 1894 return (0); 1895 } 1896 1897 /* 1898 * Read wrapper for special devices. 1899 * 1900 * ufsspec_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 1901 * struct ucred *a_cred) 1902 */ 1903 static 1904 int 1905 ufsspec_read(struct vop_read_args *ap) 1906 { 1907 int error, resid; 1908 struct inode *ip; 1909 struct uio *uio; 1910 1911 uio = ap->a_uio; 1912 resid = uio->uio_resid; 1913 error = VOCALL(&spec_vnode_vops, &ap->a_head); 1914 /* 1915 * The inode may have been revoked during the call, so it must not 1916 * be accessed blindly here or in the other wrapper functions. 1917 */ 1918 ip = VTOI(ap->a_vp); 1919 if (ip != NULL && (uio->uio_resid != resid || (error == 0 && resid != 0))) 1920 ip->i_flag |= IN_ACCESS; 1921 return (error); 1922 } 1923 1924 /* 1925 * Write wrapper for special devices. 1926 * 1927 * ufsspec_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 1928 * struct ucred *a_cred) 1929 */ 1930 static 1931 int 1932 ufsspec_write(struct vop_write_args *ap) 1933 { 1934 int error, resid; 1935 struct inode *ip; 1936 struct uio *uio; 1937 1938 uio = ap->a_uio; 1939 resid = uio->uio_resid; 1940 error = VOCALL(&spec_vnode_vops, &ap->a_head); 1941 ip = VTOI(ap->a_vp); 1942 if (ip != NULL && (uio->uio_resid != resid || (error == 0 && resid != 0))) 1943 VTOI(ap->a_vp)->i_flag |= IN_CHANGE | IN_UPDATE; 1944 return (error); 1945 } 1946 1947 /* 1948 * SPECFS's getattr will override fields as necessary, but does not fill 1949 * stuff in from scratch. 1950 */ 1951 static 1952 int 1953 ufsspec_getattr (struct vop_getattr_args *ap) 1954 { 1955 int error; 1956 1957 error = ufs_getattr(ap); 1958 if (error == 0) 1959 VOCALL(&spec_vnode_vops, &ap->a_head); 1960 return (error); 1961 } 1962 1963 /* 1964 * Close wrapper for special devices. 1965 * 1966 * Update the times on the inode then do device close. 1967 * 1968 * ufsspec_close(struct vnode *a_vp, int a_fflag) 1969 */ 1970 static 1971 int 1972 ufsspec_close(struct vop_close_args *ap) 1973 { 1974 struct vnode *vp = ap->a_vp; 1975 1976 if (vp->v_sysref.refcnt > 1) 1977 ufs_itimes(vp); 1978 return (VOCALL(&spec_vnode_vops, &ap->a_head)); 1979 } 1980 1981 /* 1982 * Read wrapper for fifos. 1983 * 1984 * ufsfifo_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 1985 * struct ucred *a_cred) 1986 */ 1987 static 1988 int 1989 ufsfifo_read(struct vop_read_args *ap) 1990 { 1991 int error, resid; 1992 struct inode *ip; 1993 struct uio *uio; 1994 1995 uio = ap->a_uio; 1996 resid = uio->uio_resid; 1997 error = VOCALL(&fifo_vnode_vops, &ap->a_head); 1998 ip = VTOI(ap->a_vp); 1999 if ((ap->a_vp->v_mount->mnt_flag & MNT_NOATIME) == 0 && ip != NULL && 2000 (uio->uio_resid != resid || (error == 0 && resid != 0))) 2001 VTOI(ap->a_vp)->i_flag |= IN_ACCESS; 2002 return (error); 2003 } 2004 2005 /* 2006 * Write wrapper for fifos. 2007 * 2008 * ufsfifo_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 2009 * struct ucred *a_cred) 2010 */ 2011 static 2012 int 2013 ufsfifo_write(struct vop_write_args *ap) 2014 { 2015 int error, resid; 2016 struct inode *ip; 2017 struct uio *uio; 2018 2019 uio = ap->a_uio; 2020 resid = uio->uio_resid; 2021 error = VOCALL(&fifo_vnode_vops, &ap->a_head); 2022 ip = VTOI(ap->a_vp); 2023 if (ip != NULL && (uio->uio_resid != resid || (error == 0 && resid != 0))) 2024 VTOI(ap->a_vp)->i_flag |= IN_CHANGE | IN_UPDATE; 2025 return (error); 2026 } 2027 2028 /* 2029 * Close wrapper for fifos. 2030 * 2031 * Update the times on the inode then do device close. 2032 * 2033 * ufsfifo_close(struct vnode *a_vp, int a_fflag) 2034 */ 2035 static 2036 int 2037 ufsfifo_close(struct vop_close_args *ap) 2038 { 2039 struct vnode *vp = ap->a_vp; 2040 2041 if (vp->v_sysref.refcnt > 1) 2042 ufs_itimes(vp); 2043 return (VOCALL(&fifo_vnode_vops, &ap->a_head)); 2044 } 2045 2046 /* 2047 * Kqfilter wrapper for fifos. 2048 * 2049 * Fall through to ufs kqfilter routines if needed 2050 */ 2051 static 2052 int 2053 ufsfifo_kqfilter(struct vop_kqfilter_args *ap) 2054 { 2055 int error; 2056 2057 error = VOCALL(&fifo_vnode_vops, &ap->a_head); 2058 if (error) 2059 error = ufs_kqfilter(ap); 2060 return (error); 2061 } 2062 2063 /* 2064 * Advisory record locking support 2065 * 2066 * ufs_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, struct flock *a_fl, 2067 * int a_flags) 2068 */ 2069 static 2070 int 2071 ufs_advlock(struct vop_advlock_args *ap) 2072 { 2073 struct inode *ip = VTOI(ap->a_vp); 2074 2075 return (lf_advlock(ap, &(ip->i_lockf), ip->i_size)); 2076 } 2077 2078 /* 2079 * Initialize the vnode associated with a new inode, handle aliased 2080 * vnodes. 2081 * 2082 * Make sure directories have their VM object now rather then later, 2083 * saving us from having to check on all the myrid directory VOPs 2084 * that might be executed without a VOP_OPEN being performed. 2085 */ 2086 int 2087 ufs_vinit(struct mount *mntp, struct vnode **vpp) 2088 { 2089 struct inode *ip; 2090 struct vnode *vp; 2091 struct timeval tv; 2092 2093 vp = *vpp; 2094 ip = VTOI(vp); 2095 2096 vp->v_type = IFTOVT(ip->i_mode); 2097 2098 switch(vp->v_type) { 2099 case VCHR: 2100 case VBLK: 2101 vp->v_ops = &mntp->mnt_vn_spec_ops; 2102 addaliasu(vp, umajor(ip->i_rdev), uminor(ip->i_rdev)); 2103 break; 2104 case VFIFO: 2105 vp->v_ops = &mntp->mnt_vn_fifo_ops; 2106 break; 2107 case VDIR: 2108 case VREG: 2109 vinitvmio(vp, ip->i_size); 2110 break; 2111 case VLNK: 2112 if (ip->i_size >= vp->v_mount->mnt_maxsymlinklen) 2113 vinitvmio(vp, ip->i_size); 2114 break; 2115 default: 2116 break; 2117 2118 } 2119 2120 if (ip->i_number == ROOTINO) 2121 vp->v_flag |= VROOT; 2122 /* 2123 * Initialize modrev times 2124 */ 2125 getmicrouptime(&tv); 2126 SETHIGH(ip->i_modrev, tv.tv_sec); 2127 SETLOW(ip->i_modrev, tv.tv_usec * 4294); 2128 *vpp = vp; 2129 return (0); 2130 } 2131 2132 /* 2133 * Allocate a new inode. 2134 */ 2135 static 2136 int 2137 ufs_makeinode(int mode, struct vnode *dvp, struct vnode **vpp, 2138 struct componentname *cnp) 2139 { 2140 struct inode *ip, *pdir; 2141 struct direct newdir; 2142 struct vnode *tvp; 2143 int error; 2144 2145 pdir = VTOI(dvp); 2146 *vpp = NULL; 2147 if ((mode & IFMT) == 0) 2148 mode |= IFREG; 2149 2150 error = ffs_valloc(dvp, mode, cnp->cn_cred, &tvp); 2151 if (error) 2152 return (error); 2153 ip = VTOI(tvp); 2154 ip->i_flags = pdir->i_flags & (SF_NOHISTORY|UF_NOHISTORY|UF_NODUMP); 2155 ip->i_gid = pdir->i_gid; 2156 #ifdef SUIDDIR 2157 { 2158 #ifdef QUOTA 2159 struct ucred ucred, *ucp; 2160 ucp = cnp->cn_cred; 2161 #endif 2162 /* 2163 * If we are not the owner of the directory, 2164 * and we are hacking owners here, (only do this where told to) 2165 * and we are not giving it TO root, (would subvert quotas) 2166 * then go ahead and give it to the other user. 2167 * Note that this drops off the execute bits for security. 2168 */ 2169 if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) && 2170 (pdir->i_mode & ISUID) && 2171 (pdir->i_uid != cnp->cn_cred->cr_uid) && pdir->i_uid) { 2172 ip->i_uid = pdir->i_uid; 2173 mode &= ~07111; 2174 #ifdef QUOTA 2175 /* 2176 * Make sure the correct user gets charged 2177 * for the space. 2178 * Quickly knock up a dummy credential for the victim. 2179 * XXX This seems to never be accessed out of our 2180 * context so a stack variable is ok. 2181 */ 2182 ucred.cr_ref = 1; 2183 ucred.cr_uid = ip->i_uid; 2184 ucred.cr_ngroups = 1; 2185 ucred.cr_groups[0] = pdir->i_gid; 2186 ucp = &ucred; 2187 #endif 2188 } else 2189 ip->i_uid = cnp->cn_cred->cr_uid; 2190 2191 #ifdef QUOTA 2192 if ((error = ufs_getinoquota(ip)) || 2193 (error = ufs_chkiq(ip, 1, ucp, 0))) { 2194 ffs_vfree(tvp, ip->i_number, mode); 2195 vput(tvp); 2196 return (error); 2197 } 2198 #endif 2199 } 2200 #else /* !SUIDDIR */ 2201 ip->i_uid = cnp->cn_cred->cr_uid; 2202 #ifdef QUOTA 2203 if ((error = ufs_getinoquota(ip)) || 2204 (error = ufs_chkiq(ip, 1, cnp->cn_cred, 0))) { 2205 ffs_vfree(tvp, ip->i_number, mode); 2206 vput(tvp); 2207 return (error); 2208 } 2209 #endif 2210 #endif /* !SUIDDIR */ 2211 ip->i_fsmid = cache_getnewfsmid(); 2212 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 2213 ip->i_mode = mode; 2214 tvp->v_type = IFTOVT(mode); /* Rest init'd in getnewvnode(). */ 2215 ip->i_effnlink = 1; 2216 ip->i_nlink = 1; 2217 if (DOINGSOFTDEP(tvp)) 2218 softdep_change_linkcnt(ip); 2219 if ((ip->i_mode & ISGID) && !groupmember(ip->i_gid, cnp->cn_cred) && 2220 priv_check_cred(cnp->cn_cred, PRIV_ROOT, 0)) { 2221 ip->i_mode &= ~ISGID; 2222 } 2223 2224 if (cnp->cn_flags & CNP_ISWHITEOUT) 2225 ip->i_flags |= UF_OPAQUE; 2226 2227 /* 2228 * Regular files and directories need VM objects. Softlinks do 2229 * not (not immediately anyway). 2230 */ 2231 if (tvp->v_type == VREG || tvp->v_type == VDIR) 2232 vinitvmio(tvp, 0); 2233 2234 /* 2235 * Make sure inode goes to disk before directory entry. 2236 */ 2237 error = ffs_update(tvp, !(DOINGSOFTDEP(tvp) | DOINGASYNC(tvp))); 2238 if (error) 2239 goto bad; 2240 ufs_makedirentry(ip, cnp, &newdir); 2241 error = ufs_direnter(dvp, tvp, &newdir, cnp, NULL); 2242 if (error) 2243 goto bad; 2244 *vpp = tvp; 2245 return (0); 2246 2247 bad: 2248 /* 2249 * Write error occurred trying to update the inode 2250 * or the directory so must deallocate the inode. 2251 */ 2252 ip->i_effnlink = 0; 2253 ip->i_nlink = 0; 2254 ip->i_flag |= IN_CHANGE; 2255 if (DOINGSOFTDEP(tvp)) 2256 softdep_change_linkcnt(ip); 2257 vput(tvp); 2258 return (error); 2259 } 2260 2261 static int 2262 ufs_missingop(struct vop_generic_args *ap) 2263 { 2264 panic("no vop function for %s in ufs child", ap->a_desc->sd_name); 2265 return (EOPNOTSUPP); 2266 } 2267 2268 static struct filterops ufsread_filtops = 2269 { 1, NULL, filt_ufsdetach, filt_ufsread }; 2270 static struct filterops ufswrite_filtops = 2271 { 1, NULL, filt_ufsdetach, filt_ufswrite }; 2272 static struct filterops ufsvnode_filtops = 2273 { 1, NULL, filt_ufsdetach, filt_ufsvnode }; 2274 2275 /* 2276 * ufs_kqfilter(struct vnode *a_vp, struct knote *a_kn) 2277 */ 2278 static int 2279 ufs_kqfilter(struct vop_kqfilter_args *ap) 2280 { 2281 struct vnode *vp = ap->a_vp; 2282 struct knote *kn = ap->a_kn; 2283 lwkt_tokref ilock; 2284 2285 switch (kn->kn_filter) { 2286 case EVFILT_READ: 2287 kn->kn_fop = &ufsread_filtops; 2288 break; 2289 case EVFILT_WRITE: 2290 kn->kn_fop = &ufswrite_filtops; 2291 break; 2292 case EVFILT_VNODE: 2293 kn->kn_fop = &ufsvnode_filtops; 2294 break; 2295 default: 2296 return (1); 2297 } 2298 2299 kn->kn_hook = (caddr_t)vp; 2300 2301 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token); 2302 SLIST_INSERT_HEAD(&vp->v_pollinfo.vpi_selinfo.si_note, kn, kn_selnext); 2303 lwkt_reltoken(&ilock); 2304 2305 return (0); 2306 } 2307 2308 static void 2309 filt_ufsdetach(struct knote *kn) 2310 { 2311 struct vnode *vp = (struct vnode *)kn->kn_hook; 2312 lwkt_tokref ilock; 2313 2314 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token); 2315 SLIST_REMOVE(&vp->v_pollinfo.vpi_selinfo.si_note, 2316 kn, knote, kn_selnext); 2317 lwkt_reltoken(&ilock); 2318 } 2319 2320 /*ARGSUSED*/ 2321 static int 2322 filt_ufsread(struct knote *kn, long hint) 2323 { 2324 struct vnode *vp = (struct vnode *)kn->kn_hook; 2325 struct inode *ip = VTOI(vp); 2326 2327 /* 2328 * filesystem is gone, so set the EOF flag and schedule 2329 * the knote for deletion. 2330 */ 2331 if (hint == NOTE_REVOKE) { 2332 kn->kn_flags |= (EV_EOF | EV_ONESHOT); 2333 return (1); 2334 } 2335 2336 kn->kn_data = ip->i_size - kn->kn_fp->f_offset; 2337 return (kn->kn_data != 0); 2338 } 2339 2340 /*ARGSUSED*/ 2341 static int 2342 filt_ufswrite(struct knote *kn, long hint) 2343 { 2344 /* 2345 * filesystem is gone, so set the EOF flag and schedule 2346 * the knote for deletion. 2347 */ 2348 if (hint == NOTE_REVOKE) 2349 kn->kn_flags |= (EV_EOF | EV_ONESHOT); 2350 2351 kn->kn_data = 0; 2352 return (1); 2353 } 2354 2355 static int 2356 filt_ufsvnode(struct knote *kn, long hint) 2357 { 2358 if (kn->kn_sfflags & hint) 2359 kn->kn_fflags |= hint; 2360 if (hint == NOTE_REVOKE) { 2361 kn->kn_flags |= EV_EOF; 2362 return (1); 2363 } 2364 return (kn->kn_fflags != 0); 2365 } 2366 2367 /* Global vfs data structures for ufs. */ 2368 static struct vop_ops ufs_vnode_vops = { 2369 .vop_default = vop_defaultop, 2370 .vop_fsync = (void *)ufs_missingop, 2371 .vop_read = (void *)ufs_missingop, 2372 .vop_reallocblks = (void *)ufs_missingop, 2373 .vop_write = (void *)ufs_missingop, 2374 .vop_access = ufs_access, 2375 .vop_advlock = ufs_advlock, 2376 .vop_bmap = ufs_bmap, 2377 .vop_old_lookup = ufs_lookup, 2378 .vop_close = ufs_close, 2379 .vop_old_create = ufs_create, 2380 .vop_getattr = ufs_getattr, 2381 .vop_inactive = ufs_inactive, 2382 .vop_old_link = ufs_link, 2383 .vop_old_mkdir = ufs_mkdir, 2384 .vop_old_mknod = ufs_mknod, 2385 .vop_mmap = ufs_mmap, 2386 .vop_open = vop_stdopen, 2387 .vop_pathconf = vop_stdpathconf, 2388 .vop_poll = vop_stdpoll, 2389 .vop_kqfilter = ufs_kqfilter, 2390 .vop_print = ufs_print, 2391 .vop_readdir = ufs_readdir, 2392 .vop_readlink = ufs_readlink, 2393 .vop_reclaim = ufs_reclaim, 2394 .vop_old_remove = ufs_remove, 2395 .vop_old_rename = ufs_rename, 2396 .vop_old_rmdir = ufs_rmdir, 2397 .vop_setattr = ufs_setattr, 2398 .vop_markatime = ufs_markatime, 2399 .vop_strategy = ufs_strategy, 2400 .vop_old_symlink = ufs_symlink, 2401 .vop_old_whiteout = ufs_whiteout 2402 }; 2403 2404 static struct vop_ops ufs_spec_vops = { 2405 .vop_default = spec_vnoperate, 2406 .vop_fsync = (void *)ufs_missingop, 2407 .vop_access = ufs_access, 2408 .vop_close = ufsspec_close, 2409 .vop_getattr = ufsspec_getattr, 2410 .vop_inactive = ufs_inactive, 2411 .vop_print = ufs_print, 2412 .vop_read = ufsspec_read, 2413 .vop_reclaim = ufs_reclaim, 2414 .vop_setattr = ufs_setattr, 2415 .vop_markatime = ufs_markatime, 2416 .vop_write = ufsspec_write 2417 }; 2418 2419 static struct vop_ops ufs_fifo_vops = { 2420 .vop_default = fifo_vnoperate, 2421 .vop_fsync = (void *)ufs_missingop, 2422 .vop_access = ufs_access, 2423 .vop_close = ufsfifo_close, 2424 .vop_getattr = ufs_getattr, 2425 .vop_inactive = ufs_inactive, 2426 .vop_kqfilter = ufsfifo_kqfilter, 2427 .vop_print = ufs_print, 2428 .vop_read = ufsfifo_read, 2429 .vop_reclaim = ufs_reclaim, 2430 .vop_setattr = ufs_setattr, 2431 .vop_markatime = ufs_markatime, 2432 .vop_write = ufsfifo_write 2433 }; 2434 2435 VNODEOP_SET(ufs_vnode_vops); 2436 VNODEOP_SET(ufs_spec_vops); 2437 VNODEOP_SET(ufs_fifo_vops); 2438 2439 /* 2440 * ufs_vnoperate() 2441 */ 2442 int 2443 ufs_vnoperate(struct vop_generic_args *ap) 2444 { 2445 return (VOCALL(&ufs_vnode_vops, ap)); 2446 } 2447 2448 /* 2449 * ufs_vnoperatefifo() 2450 */ 2451 int 2452 ufs_vnoperatefifo(struct vop_generic_args *ap) 2453 { 2454 return (VOCALL(&ufs_fifo_vops, ap)); 2455 } 2456 2457 /* 2458 * ufs_vnoperatespec() 2459 */ 2460 int 2461 ufs_vnoperatespec(struct vop_generic_args *ap) 2462 { 2463 return (VOCALL(&ufs_spec_vops, ap)); 2464 } 2465