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