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 1.53 2006/07/19 06:08:14 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/namei.h> 55 #include <sys/mount.h> 56 #include <sys/unistd.h> 57 #include <sys/vnode.h> 58 #include <sys/malloc.h> 59 #include <sys/dirent.h> 60 #include <sys/lockf.h> 61 #include <sys/event.h> 62 #include <sys/conf.h> 63 64 #include <sys/file.h> /* XXX */ 65 #include <sys/jail.h> 66 67 #include <vm/vm.h> 68 #include <vm/vm_extern.h> 69 70 #include <vfs/fifofs/fifo.h> 71 72 #include "quota.h" 73 #include "inode.h" 74 #include "dir.h" 75 #include "ufsmount.h" 76 #include "ufs_extern.h" 77 #include "ffs_extern.h" 78 #ifdef UFS_DIRHASH 79 #include "dirhash.h" 80 #endif 81 82 static int ufs_access (struct vop_access_args *); 83 static int ufs_advlock (struct vop_advlock_args *); 84 static int ufs_chmod (struct vnode *, int, struct ucred *); 85 static int ufs_chown (struct vnode *, uid_t, gid_t, struct ucred *); 86 static int ufs_close (struct vop_close_args *); 87 static int ufs_create (struct vop_old_create_args *); 88 static int ufs_getattr (struct vop_getattr_args *); 89 static int ufs_link (struct vop_old_link_args *); 90 static int ufs_makeinode (int mode, struct vnode *, struct vnode **, struct componentname *); 91 static int ufs_missingop (struct vop_generic_args *ap); 92 static int ufs_mkdir (struct vop_old_mkdir_args *); 93 static int ufs_mknod (struct vop_old_mknod_args *); 94 static int ufs_mmap (struct vop_mmap_args *); 95 static int ufs_open (struct vop_open_args *); 96 static int ufs_pathconf (struct vop_pathconf_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 error = ufs_makeinode(MAKEIMODE(vap->va_type, vap->va_mode), 224 ap->a_dvp, vpp, ap->a_cnp); 225 if (error) 226 return (error); 227 VN_KNOTE(ap->a_dvp, NOTE_WRITE); 228 ip = VTOI(*vpp); 229 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 230 if (vap->va_rdev != VNOVAL) { 231 /* 232 * Want to be able to use this to make badblock 233 * inodes, so don't truncate the dev number. 234 */ 235 ip->i_rdev = vap->va_rdev; 236 } 237 /* 238 * Remove inode, then reload it through VFS_VGET so it is 239 * checked to see if it is an alias of an existing entry in 240 * the inode cache. 241 */ 242 (*vpp)->v_type = VNON; 243 ino = ip->i_number; /* Save this before vgone() invalidates ip. */ 244 vgone(*vpp); 245 vput(*vpp); 246 error = VFS_VGET(ap->a_dvp->v_mount, ino, vpp); 247 if (error) { 248 *vpp = NULL; 249 return (error); 250 } 251 return (0); 252 } 253 254 /* 255 * Open called. 256 * 257 * Nothing to do. 258 * 259 * ufs_open(struct vnode *a_vp, int a_mode, struct ucred *a_cred, 260 * struct thread *a_td) 261 */ 262 /* ARGSUSED */ 263 static 264 int 265 ufs_open(struct vop_open_args *ap) 266 { 267 struct vnode *vp = ap->a_vp; 268 269 /* 270 * Files marked append-only must be opened for appending. 271 */ 272 if ((VTOI(vp)->i_flags & APPEND) && 273 (ap->a_mode & (FWRITE | O_APPEND)) == FWRITE) { 274 return (EPERM); 275 } 276 277 return (vop_stdopen(ap)); 278 } 279 280 /* 281 * Close called. 282 * 283 * Update the times on the inode. 284 * 285 * ufs_close(struct vnode *a_vp, int a_fflag, struct ucred *a_cred, 286 * struct thread *a_td) 287 */ 288 /* ARGSUSED */ 289 static 290 int 291 ufs_close(struct vop_close_args *ap) 292 { 293 struct vnode *vp = ap->a_vp; 294 295 if (vp->v_usecount > 1) 296 ufs_itimes(vp); 297 return (vop_stdclose(ap)); 298 } 299 300 /* 301 * ufs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred, 302 * struct thread *a_td) 303 */ 304 static 305 int 306 ufs_access(struct vop_access_args *ap) 307 { 308 struct vnode *vp = ap->a_vp; 309 struct inode *ip = VTOI(vp); 310 struct ucred *cred = ap->a_cred; 311 mode_t mask, mode = ap->a_mode; 312 gid_t *gp; 313 int i; 314 #ifdef QUOTA 315 int error; 316 #endif 317 318 /* 319 * Disallow write attempts on read-only filesystems; 320 * unless the file is a socket, fifo, or a block or 321 * character device resident on the filesystem. 322 */ 323 if (mode & VWRITE) { 324 switch (vp->v_type) { 325 case VDIR: 326 case VLNK: 327 case VREG: 328 if (vp->v_mount->mnt_flag & MNT_RDONLY) 329 return (EROFS); 330 #ifdef QUOTA 331 if ((error = ufs_getinoquota(ip)) != 0) 332 return (error); 333 #endif 334 break; 335 default: 336 break; 337 } 338 } 339 340 /* If immutable bit set, nobody gets to write it. */ 341 if ((mode & VWRITE) && (ip->i_flags & IMMUTABLE)) 342 return (EPERM); 343 344 /* Otherwise, user id 0 always gets access. */ 345 if (cred->cr_uid == 0) 346 return (0); 347 348 mask = 0; 349 350 /* Otherwise, check the owner. */ 351 if (cred->cr_uid == ip->i_uid) { 352 if (mode & VEXEC) 353 mask |= S_IXUSR; 354 if (mode & VREAD) 355 mask |= S_IRUSR; 356 if (mode & VWRITE) 357 mask |= S_IWUSR; 358 return ((ip->i_mode & mask) == mask ? 0 : EACCES); 359 } 360 361 /* Otherwise, check the groups. */ 362 for (i = 0, gp = cred->cr_groups; i < cred->cr_ngroups; i++, gp++) 363 if (ip->i_gid == *gp) { 364 if (mode & VEXEC) 365 mask |= S_IXGRP; 366 if (mode & VREAD) 367 mask |= S_IRGRP; 368 if (mode & VWRITE) 369 mask |= S_IWGRP; 370 return ((ip->i_mode & mask) == mask ? 0 : EACCES); 371 } 372 373 /* Otherwise, check everyone else. */ 374 if (mode & VEXEC) 375 mask |= S_IXOTH; 376 if (mode & VREAD) 377 mask |= S_IROTH; 378 if (mode & VWRITE) 379 mask |= S_IWOTH; 380 return ((ip->i_mode & mask) == mask ? 0 : EACCES); 381 } 382 383 /* 384 * ufs_getattr(struct vnode *a_vp, struct vattr *a_vap, 385 * struct thread *a_td) 386 */ 387 /* ARGSUSED */ 388 static 389 int 390 ufs_getattr(struct vop_getattr_args *ap) 391 { 392 struct vnode *vp = ap->a_vp; 393 struct inode *ip = VTOI(vp); 394 struct vattr *vap = ap->a_vap; 395 396 /* 397 * This may cause i_fsmid to be updated even if no change (0) 398 * is returned, but we should only write out the inode if non-zero 399 * is returned and if the mount is read-write. 400 */ 401 if (cache_check_fsmid_vp(vp, &ip->i_fsmid) && 402 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0 403 ) { 404 ip->i_flag |= IN_LAZYMOD; 405 } 406 407 ufs_itimes(vp); 408 /* 409 * Copy from inode table 410 */ 411 vap->va_fsid = dev2udev(ip->i_dev); 412 vap->va_fileid = ip->i_number; 413 vap->va_mode = ip->i_mode & ~IFMT; 414 vap->va_nlink = VFSTOUFS(vp->v_mount)->um_i_effnlink_valid ? 415 ip->i_effnlink : ip->i_nlink; 416 vap->va_uid = ip->i_uid; 417 vap->va_gid = ip->i_gid; 418 vap->va_rdev = ip->i_rdev; 419 vap->va_size = ip->i_din.di_size; 420 vap->va_atime.tv_sec = ip->i_atime; 421 vap->va_atime.tv_nsec = ip->i_atimensec; 422 vap->va_mtime.tv_sec = ip->i_mtime; 423 vap->va_mtime.tv_nsec = ip->i_mtimensec; 424 vap->va_ctime.tv_sec = ip->i_ctime; 425 vap->va_ctime.tv_nsec = ip->i_ctimensec; 426 vap->va_flags = ip->i_flags; 427 vap->va_gen = ip->i_gen; 428 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize; 429 vap->va_bytes = dbtob((u_quad_t)ip->i_blocks); 430 vap->va_type = IFTOVT(ip->i_mode); 431 vap->va_filerev = ip->i_modrev; 432 vap->va_fsmid = ip->i_fsmid; 433 return (0); 434 } 435 436 /* 437 * Set attribute vnode op. called from several syscalls 438 * 439 * ufs_setattr(struct vnode *a_vp, struct vattr *a_vap, 440 * struct ucred *a_cred, struct thread *a_td) 441 */ 442 static 443 int 444 ufs_setattr(struct vop_setattr_args *ap) 445 { 446 struct vattr *vap = ap->a_vap; 447 struct vnode *vp = ap->a_vp; 448 struct inode *ip = VTOI(vp); 449 struct ucred *cred = ap->a_cred; 450 int error; 451 452 /* 453 * Check for unsettable attributes. 454 */ 455 if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) || 456 (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) || 457 (vap->va_blocksize != VNOVAL) || (vap->va_rdev != VNOVAL) || 458 ((int)vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) { 459 return (EINVAL); 460 } 461 if (vap->va_flags != VNOVAL) { 462 if (vp->v_mount->mnt_flag & MNT_RDONLY) 463 return (EROFS); 464 if (cred->cr_uid != ip->i_uid && 465 (error = suser_cred(cred, PRISON_ROOT))) 466 return (error); 467 /* 468 * Note that a root chflags becomes a user chflags when 469 * we are jailed, unless the jail.chflags_allowed sysctl 470 * is set. 471 */ 472 if (cred->cr_uid == 0 && 473 (!jailed(cred) || jail_chflags_allowed)) { 474 if ((ip->i_flags 475 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) && 476 securelevel > 0) 477 return (EPERM); 478 ip->i_flags = vap->va_flags; 479 } else { 480 if (ip->i_flags 481 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) || 482 (vap->va_flags & UF_SETTABLE) != vap->va_flags) 483 return (EPERM); 484 ip->i_flags &= SF_SETTABLE; 485 ip->i_flags |= (vap->va_flags & UF_SETTABLE); 486 } 487 ip->i_flag |= IN_CHANGE; 488 if (vap->va_flags & (IMMUTABLE | APPEND)) 489 return (0); 490 } 491 if (ip->i_flags & (IMMUTABLE | APPEND)) 492 return (EPERM); 493 /* 494 * Go through the fields and update iff not VNOVAL. 495 */ 496 if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) { 497 if (vp->v_mount->mnt_flag & MNT_RDONLY) 498 return (EROFS); 499 if ((error = ufs_chown(vp, vap->va_uid, vap->va_gid, cred)) != 0) 500 return (error); 501 } 502 if (vap->va_size != VNOVAL) { 503 /* 504 * Disallow write attempts on read-only filesystems; 505 * unless the file is a socket, fifo, or a block or 506 * character device resident on the filesystem. 507 */ 508 switch (vp->v_type) { 509 case VDIR: 510 return (EISDIR); 511 case VLNK: 512 case VREG: 513 if (vp->v_mount->mnt_flag & MNT_RDONLY) 514 return (EROFS); 515 break; 516 default: 517 break; 518 } 519 if ((error = ffs_truncate(vp, vap->va_size, 0, cred)) != 0) 520 return (error); 521 } 522 ip = VTOI(vp); 523 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) { 524 if (vp->v_mount->mnt_flag & MNT_RDONLY) 525 return (EROFS); 526 if (cred->cr_uid != ip->i_uid && 527 (error = suser_cred(cred, PRISON_ROOT)) && 528 ((vap->va_vaflags & VA_UTIMES_NULL) == 0 || 529 (error = VOP_ACCESS(vp, VWRITE, cred)))) 530 return (error); 531 if (vap->va_atime.tv_sec != VNOVAL) 532 ip->i_flag |= IN_ACCESS; 533 if (vap->va_mtime.tv_sec != VNOVAL) 534 ip->i_flag |= IN_CHANGE | IN_UPDATE; 535 ufs_itimes(vp); 536 if (vap->va_atime.tv_sec != VNOVAL) { 537 ip->i_atime = vap->va_atime.tv_sec; 538 ip->i_atimensec = vap->va_atime.tv_nsec; 539 } 540 if (vap->va_mtime.tv_sec != VNOVAL) { 541 ip->i_mtime = vap->va_mtime.tv_sec; 542 ip->i_mtimensec = vap->va_mtime.tv_nsec; 543 } 544 error = ffs_update(vp, 0); 545 if (error) 546 return (error); 547 } 548 error = 0; 549 if (vap->va_mode != (mode_t)VNOVAL) { 550 if (vp->v_mount->mnt_flag & MNT_RDONLY) 551 return (EROFS); 552 error = ufs_chmod(vp, (int)vap->va_mode, cred); 553 } 554 VN_KNOTE(vp, NOTE_ATTRIB); 555 return (error); 556 } 557 558 /* 559 * Change the mode on a file. 560 * Inode must be locked before calling. 561 */ 562 static int 563 ufs_chmod(struct vnode *vp, int mode, struct ucred *cred) 564 { 565 struct inode *ip = VTOI(vp); 566 int error; 567 568 if (cred->cr_uid != ip->i_uid) { 569 error = suser_cred(cred, PRISON_ROOT); 570 if (error) 571 return (error); 572 } 573 if (cred->cr_uid) { 574 if (vp->v_type != VDIR && (mode & S_ISTXT)) 575 return (EFTYPE); 576 if (!groupmember(ip->i_gid, cred) && (mode & ISGID)) 577 return (EPERM); 578 } 579 ip->i_mode &= ~ALLPERMS; 580 ip->i_mode |= (mode & ALLPERMS); 581 ip->i_flag |= IN_CHANGE; 582 return (0); 583 } 584 585 /* 586 * Perform chown operation on inode ip; 587 * inode must be locked prior to call. 588 */ 589 static int 590 ufs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred) 591 { 592 struct inode *ip = VTOI(vp); 593 uid_t ouid; 594 gid_t ogid; 595 int error = 0; 596 #ifdef QUOTA 597 int i; 598 long change; 599 #endif 600 601 if (uid == (uid_t)VNOVAL) 602 uid = ip->i_uid; 603 if (gid == (gid_t)VNOVAL) 604 gid = ip->i_gid; 605 /* 606 * If we don't own the file, are trying to change the owner 607 * of the file, or are not a member of the target group, 608 * the caller must be superuser or the call fails. 609 */ 610 if ((cred->cr_uid != ip->i_uid || uid != ip->i_uid || 611 (gid != ip->i_gid && !(cred->cr_gid == gid || 612 groupmember((gid_t)gid, cred)))) && 613 (error = suser_cred(cred, PRISON_ROOT))) 614 return (error); 615 ogid = ip->i_gid; 616 ouid = ip->i_uid; 617 #ifdef QUOTA 618 if ((error = ufs_getinoquota(ip)) != 0) 619 return (error); 620 if (ouid == uid) { 621 ufs_dqrele(vp, ip->i_dquot[USRQUOTA]); 622 ip->i_dquot[USRQUOTA] = NODQUOT; 623 } 624 if (ogid == gid) { 625 ufs_dqrele(vp, ip->i_dquot[GRPQUOTA]); 626 ip->i_dquot[GRPQUOTA] = NODQUOT; 627 } 628 change = ip->i_blocks; 629 (void) ufs_chkdq(ip, -change, cred, CHOWN); 630 (void) ufs_chkiq(ip, -1, cred, CHOWN); 631 for (i = 0; i < MAXQUOTAS; i++) { 632 ufs_dqrele(vp, ip->i_dquot[i]); 633 ip->i_dquot[i] = NODQUOT; 634 } 635 #endif 636 ip->i_gid = gid; 637 ip->i_uid = uid; 638 #ifdef QUOTA 639 if ((error = ufs_getinoquota(ip)) == 0) { 640 if (ouid == uid) { 641 ufs_dqrele(vp, ip->i_dquot[USRQUOTA]); 642 ip->i_dquot[USRQUOTA] = NODQUOT; 643 } 644 if (ogid == gid) { 645 ufs_dqrele(vp, ip->i_dquot[GRPQUOTA]); 646 ip->i_dquot[GRPQUOTA] = NODQUOT; 647 } 648 if ((error = ufs_chkdq(ip, change, cred, CHOWN)) == 0) { 649 if ((error = ufs_chkiq(ip, 1, cred, CHOWN)) == 0) 650 goto good; 651 else 652 (void)ufs_chkdq(ip, -change, cred, CHOWN|FORCE); 653 } 654 for (i = 0; i < MAXQUOTAS; i++) { 655 ufs_dqrele(vp, ip->i_dquot[i]); 656 ip->i_dquot[i] = NODQUOT; 657 } 658 } 659 ip->i_gid = ogid; 660 ip->i_uid = ouid; 661 if (ufs_getinoquota(ip) == 0) { 662 if (ouid == uid) { 663 ufs_dqrele(vp, ip->i_dquot[USRQUOTA]); 664 ip->i_dquot[USRQUOTA] = NODQUOT; 665 } 666 if (ogid == gid) { 667 ufs_dqrele(vp, ip->i_dquot[GRPQUOTA]); 668 ip->i_dquot[GRPQUOTA] = NODQUOT; 669 } 670 (void) ufs_chkdq(ip, change, cred, FORCE|CHOWN); 671 (void) ufs_chkiq(ip, 1, cred, FORCE|CHOWN); 672 (void) ufs_getinoquota(ip); 673 } 674 return (error); 675 good: 676 if (ufs_getinoquota(ip)) 677 panic("ufs_chown: lost quota"); 678 #endif /* QUOTA */ 679 ip->i_flag |= IN_CHANGE; 680 if (cred->cr_uid != 0 && (ouid != uid || ogid != gid)) 681 ip->i_mode &= ~(ISUID | ISGID); 682 return (0); 683 } 684 685 /* 686 * Mmap a file 687 * 688 * NB Currently unsupported. 689 * 690 * ufs_mmap(struct vnode *a_vp, int a_fflags, struct ucred *a_cred, 691 * struct thread *a_td) 692 */ 693 /* ARGSUSED */ 694 static 695 int 696 ufs_mmap(struct vop_mmap_args *ap) 697 { 698 return (EINVAL); 699 } 700 701 /* 702 * ufs_remove(struct vnode *a_dvp, struct vnode *a_vp, 703 * struct componentname *a_cnp) 704 */ 705 static 706 int 707 ufs_remove(struct vop_old_remove_args *ap) 708 { 709 struct inode *ip; 710 struct vnode *vp = ap->a_vp; 711 struct vnode *dvp = ap->a_dvp; 712 int error; 713 714 ip = VTOI(vp); 715 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) || 716 (VTOI(dvp)->i_flags & APPEND)) { 717 error = EPERM; 718 goto out; 719 } 720 error = ufs_dirremove(dvp, ip, ap->a_cnp->cn_flags, 0); 721 VN_KNOTE(vp, NOTE_DELETE); 722 VN_KNOTE(dvp, NOTE_WRITE); 723 out: 724 return (error); 725 } 726 727 /* 728 * link vnode call 729 * 730 * ufs_link(struct vnode *a_tdvp, struct vnode *a_vp, 731 * struct componentname *a_cnp) 732 */ 733 static 734 int 735 ufs_link(struct vop_old_link_args *ap) 736 { 737 struct vnode *vp = ap->a_vp; 738 struct vnode *tdvp = ap->a_tdvp; 739 struct componentname *cnp = ap->a_cnp; 740 struct inode *ip; 741 struct direct newdir; 742 int error; 743 744 if (tdvp->v_mount != vp->v_mount) { 745 error = EXDEV; 746 goto out2; 747 } 748 if (tdvp != vp && (error = vn_lock(vp, LK_EXCLUSIVE))) { 749 goto out2; 750 } 751 ip = VTOI(vp); 752 if ((nlink_t)ip->i_nlink >= LINK_MAX) { 753 error = EMLINK; 754 goto out1; 755 } 756 if (ip->i_flags & (IMMUTABLE | APPEND)) { 757 error = EPERM; 758 goto out1; 759 } 760 ip->i_effnlink++; 761 ip->i_nlink++; 762 ip->i_flag |= IN_CHANGE; 763 if (DOINGSOFTDEP(vp)) 764 softdep_change_linkcnt(ip); 765 error = ffs_update(vp, !(DOINGSOFTDEP(vp) | DOINGASYNC(vp))); 766 if (!error) { 767 ufs_makedirentry(ip, cnp, &newdir); 768 error = ufs_direnter(tdvp, vp, &newdir, cnp, NULL); 769 } 770 771 if (error) { 772 ip->i_effnlink--; 773 ip->i_nlink--; 774 ip->i_flag |= IN_CHANGE; 775 if (DOINGSOFTDEP(vp)) 776 softdep_change_linkcnt(ip); 777 } 778 out1: 779 if (tdvp != vp) 780 VOP_UNLOCK(vp, 0); 781 out2: 782 VN_KNOTE(vp, NOTE_LINK); 783 VN_KNOTE(tdvp, NOTE_WRITE); 784 return (error); 785 } 786 787 /* 788 * whiteout vnode call 789 * 790 * ufs_whiteout(struct vnode *a_dvp, struct componentname *a_cnp, int a_flags) 791 */ 792 static 793 int 794 ufs_whiteout(struct vop_old_whiteout_args *ap) 795 { 796 struct vnode *dvp = ap->a_dvp; 797 struct componentname *cnp = ap->a_cnp; 798 struct direct newdir; 799 int error = 0; 800 801 switch (ap->a_flags) { 802 case NAMEI_LOOKUP: 803 /* 4.4 format directories support whiteout operations */ 804 if (dvp->v_mount->mnt_maxsymlinklen > 0) 805 return (0); 806 return (EOPNOTSUPP); 807 808 case NAMEI_CREATE: 809 /* create a new directory whiteout */ 810 #ifdef DIAGNOSTIC 811 if (dvp->v_mount->mnt_maxsymlinklen <= 0) 812 panic("ufs_whiteout: old format filesystem"); 813 #endif 814 815 newdir.d_ino = WINO; 816 newdir.d_namlen = cnp->cn_namelen; 817 bcopy(cnp->cn_nameptr, newdir.d_name, (unsigned)cnp->cn_namelen + 1); 818 newdir.d_type = DT_WHT; 819 error = ufs_direnter(dvp, NULL, &newdir, cnp, NULL); 820 break; 821 822 case NAMEI_DELETE: 823 /* remove an existing directory whiteout */ 824 #ifdef DIAGNOSTIC 825 if (dvp->v_mount->mnt_maxsymlinklen <= 0) 826 panic("ufs_whiteout: old format filesystem"); 827 #endif 828 829 cnp->cn_flags &= ~CNP_DOWHITEOUT; 830 error = ufs_dirremove(dvp, NULL, cnp->cn_flags, 0); 831 break; 832 default: 833 panic("ufs_whiteout: unknown op"); 834 } 835 return (error); 836 } 837 838 /* 839 * Rename system call. 840 * rename("foo", "bar"); 841 * is essentially 842 * unlink("bar"); 843 * link("foo", "bar"); 844 * unlink("foo"); 845 * but ``atomically''. Can't do full commit without saving state in the 846 * inode on disk which isn't feasible at this time. Best we can do is 847 * always guarantee the target exists. 848 * 849 * Basic algorithm is: 850 * 851 * 1) Bump link count on source while we're linking it to the 852 * target. This also ensure the inode won't be deleted out 853 * from underneath us while we work (it may be truncated by 854 * a concurrent `trunc' or `open' for creation). 855 * 2) Link source to destination. If destination already exists, 856 * delete it first. 857 * 3) Unlink source reference to inode if still around. If a 858 * directory was moved and the parent of the destination 859 * is different from the source, patch the ".." entry in the 860 * directory. 861 * 862 * ufs_rename(struct vnode *a_fdvp, struct vnode *a_fvp, 863 * struct componentname *a_fcnp, struct vnode *a_tdvp, 864 * struct vnode *a_tvp, struct componentname *a_tcnp) 865 */ 866 static 867 int 868 ufs_rename(struct vop_old_rename_args *ap) 869 { 870 struct vnode *tvp = ap->a_tvp; 871 struct vnode *tdvp = ap->a_tdvp; 872 struct vnode *fvp = ap->a_fvp; 873 struct vnode *fdvp = ap->a_fdvp; 874 struct componentname *tcnp = ap->a_tcnp; 875 struct componentname *fcnp = ap->a_fcnp; 876 struct inode *ip, *xp, *dp; 877 struct direct newdir; 878 ino_t oldparent = 0, newparent = 0; 879 int doingdirectory = 0; 880 int error = 0, ioflag; 881 882 /* 883 * Check for cross-device rename. 884 */ 885 if ((fvp->v_mount != tdvp->v_mount) || 886 (tvp && (fvp->v_mount != tvp->v_mount))) { 887 error = EXDEV; 888 abortit: 889 if (tdvp == tvp) 890 vrele(tdvp); 891 else 892 vput(tdvp); 893 if (tvp) 894 vput(tvp); 895 vrele(fdvp); 896 vrele(fvp); 897 return (error); 898 } 899 900 if (tvp && ((VTOI(tvp)->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) || 901 (VTOI(tdvp)->i_flags & APPEND))) { 902 error = EPERM; 903 goto abortit; 904 } 905 906 /* 907 * Renaming a file to itself has no effect. The upper layers should 908 * not call us in that case. Temporarily just warn if they do. 909 */ 910 if (fvp == tvp) { 911 printf("ufs_rename: fvp == tvp (can't happen)\n"); 912 error = 0; 913 goto abortit; 914 } 915 916 if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0) 917 goto abortit; 918 919 /* 920 * Note: now that fvp is locked we have to be sure to unlock it before 921 * using the 'abortit' target. 922 */ 923 dp = VTOI(fdvp); 924 ip = VTOI(fvp); 925 if (ip->i_nlink >= LINK_MAX) { 926 VOP_UNLOCK(fvp, 0); 927 error = EMLINK; 928 goto abortit; 929 } 930 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) 931 || (dp->i_flags & APPEND)) { 932 VOP_UNLOCK(fvp, 0); 933 error = EPERM; 934 goto abortit; 935 } 936 if ((ip->i_mode & IFMT) == IFDIR) { 937 /* 938 * Avoid ".", "..", and aliases of "." for obvious reasons. 939 */ 940 if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') || 941 dp == ip || (fcnp->cn_flags | tcnp->cn_flags) & CNP_ISDOTDOT || 942 (ip->i_flag & IN_RENAME)) { 943 VOP_UNLOCK(fvp, 0); 944 error = EINVAL; 945 goto abortit; 946 } 947 ip->i_flag |= IN_RENAME; 948 oldparent = dp->i_number; 949 doingdirectory = 1; 950 } 951 VN_KNOTE(fdvp, NOTE_WRITE); /* XXX right place? */ 952 953 /* 954 * fvp still locked. ip->i_flag has IN_RENAME set if doingdirectory. 955 * Cleanup fvp requirements so we can unlock it. 956 * 957 * tvp and tdvp are locked. tvp may be NULL. Now that dp and xp 958 * is setup we can use the 'bad' target if we unlock fvp. We cannot 959 * use the abortit target anymore because of IN_RENAME. 960 */ 961 dp = VTOI(tdvp); 962 if (tvp) 963 xp = VTOI(tvp); 964 else 965 xp = NULL; 966 967 /* 968 * 1) Bump link count while we're moving stuff 969 * around. If we crash somewhere before 970 * completing our work, the link count 971 * may be wrong, but correctable. 972 */ 973 ip->i_effnlink++; 974 ip->i_nlink++; 975 ip->i_flag |= IN_CHANGE; 976 if (DOINGSOFTDEP(fvp)) 977 softdep_change_linkcnt(ip); 978 if ((error = ffs_update(fvp, !(DOINGSOFTDEP(fvp) | 979 DOINGASYNC(fvp)))) != 0) { 980 VOP_UNLOCK(fvp, 0); 981 goto bad; 982 } 983 984 /* 985 * If ".." must be changed (ie the directory gets a new 986 * parent) then the source directory must not be in the 987 * directory heirarchy above the target, as this would 988 * orphan everything below the source directory. Also 989 * the user must have write permission in the source so 990 * as to be able to change "..". We must repeat the call 991 * to namei, as the parent directory is unlocked by the 992 * call to checkpath(). 993 */ 994 error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred); 995 VOP_UNLOCK(fvp, 0); 996 997 /* 998 * We are now back to where we were in that fvp, fdvp are unlocked 999 * and tvp, tdvp are locked. tvp may be NULL. IN_RENAME may be 1000 * set. Only the bad target or, if we clean up tvp and tdvp, the 1001 * out target, may be used. 1002 */ 1003 if (oldparent != dp->i_number) 1004 newparent = dp->i_number; 1005 if (doingdirectory && newparent) { 1006 if (error) /* write access check above */ 1007 goto bad; 1008 1009 /* 1010 * Once we start messing with tvp and tdvp we cannot use the 1011 * 'bad' target, only finish cleaning tdvp and tvp up and 1012 * use the 'out' target. 1013 * 1014 * This cleans up tvp. 1015 */ 1016 if (xp != NULL) { 1017 vput(tvp); 1018 xp = NULL; 1019 } 1020 1021 /* 1022 * This is a real mess. ufs_checkpath vput's the target 1023 * directory so retain an extra ref and note that tdvp will 1024 * lose its lock on return. This leaves us with one good 1025 * ref after ufs_checkpath returns. 1026 */ 1027 vref(tdvp); 1028 error = ufs_checkpath(ip, dp, tcnp->cn_cred); 1029 tcnp->cn_flags |= CNP_PDIRUNLOCK; 1030 if (error) { 1031 vrele(tdvp); 1032 goto out; 1033 } 1034 1035 /* 1036 * relookup no longer messes with tdvp's refs. tdvp must be 1037 * unlocked on entry and will be locked on a successful 1038 * return. 1039 */ 1040 error = relookup(tdvp, &tvp, tcnp); 1041 if (error) { 1042 if (tcnp->cn_flags & CNP_PDIRUNLOCK) 1043 vrele(tdvp); 1044 else 1045 vput(tdvp); 1046 goto out; 1047 } 1048 KKASSERT((tcnp->cn_flags & CNP_PDIRUNLOCK) == 0); 1049 dp = VTOI(tdvp); 1050 if (tvp) 1051 xp = VTOI(tvp); 1052 } 1053 1054 /* 1055 * We are back to fvp, fdvp unlocked, tvp, tdvp locked. tvp may 1056 * be NULL (xp will also be NULL in that case), and IN_RENAME will 1057 * be set if doingdirectory. This means we can use the 'bad' target 1058 * again. 1059 */ 1060 1061 /* 1062 * 2) If target doesn't exist, link the target 1063 * to the source and unlink the source. 1064 * Otherwise, rewrite the target directory 1065 * entry to reference the source inode and 1066 * expunge the original entry's existence. 1067 */ 1068 if (xp == NULL) { 1069 if (dp->i_dev != ip->i_dev) 1070 panic("ufs_rename: EXDEV"); 1071 /* 1072 * Account for ".." in new directory. 1073 * When source and destination have the same 1074 * parent we don't fool with the link count. 1075 */ 1076 if (doingdirectory && newparent) { 1077 if ((nlink_t)dp->i_nlink >= LINK_MAX) { 1078 error = EMLINK; 1079 goto bad; 1080 } 1081 dp->i_effnlink++; 1082 dp->i_nlink++; 1083 dp->i_flag |= IN_CHANGE; 1084 if (DOINGSOFTDEP(tdvp)) 1085 softdep_change_linkcnt(dp); 1086 error = ffs_update(tdvp, !(DOINGSOFTDEP(tdvp) | 1087 DOINGASYNC(tdvp))); 1088 if (error) 1089 goto bad; 1090 } 1091 ufs_makedirentry(ip, tcnp, &newdir); 1092 error = ufs_direnter(tdvp, NULL, &newdir, tcnp, NULL); 1093 if (error) { 1094 if (doingdirectory && newparent) { 1095 dp->i_effnlink--; 1096 dp->i_nlink--; 1097 dp->i_flag |= IN_CHANGE; 1098 if (DOINGSOFTDEP(tdvp)) 1099 softdep_change_linkcnt(dp); 1100 (void)ffs_update(tdvp, 1); 1101 } 1102 goto bad; 1103 } 1104 VN_KNOTE(tdvp, NOTE_WRITE); 1105 vput(tdvp); 1106 } else { 1107 if (xp->i_dev != dp->i_dev || xp->i_dev != ip->i_dev) 1108 panic("ufs_rename: EXDEV"); 1109 /* 1110 * Short circuit rename(foo, foo). 1111 */ 1112 if (xp->i_number == ip->i_number) 1113 panic("ufs_rename: same file"); 1114 /* 1115 * If the parent directory is "sticky", then the user must 1116 * own the parent directory, or the destination of the rename, 1117 * otherwise the destination may not be changed (except by 1118 * root). This implements append-only directories. 1119 */ 1120 if ((dp->i_mode & S_ISTXT) && tcnp->cn_cred->cr_uid != 0 && 1121 tcnp->cn_cred->cr_uid != dp->i_uid && 1122 xp->i_uid != tcnp->cn_cred->cr_uid) { 1123 error = EPERM; 1124 goto bad; 1125 } 1126 /* 1127 * Target must be empty if a directory and have no links 1128 * to it. Also, ensure source and target are compatible 1129 * (both directories, or both not directories). 1130 * 1131 * Purge the file or directory being replaced from the 1132 * nameccache. 1133 */ 1134 if ((xp->i_mode&IFMT) == IFDIR) { 1135 if ((xp->i_effnlink > 2) || 1136 !ufs_dirempty(xp, dp->i_number, tcnp->cn_cred)) { 1137 error = ENOTEMPTY; 1138 goto bad; 1139 } 1140 if (!doingdirectory) { 1141 error = ENOTDIR; 1142 goto bad; 1143 } 1144 /* cache_purge removed - handled by VFS compat layer */ 1145 } else if (doingdirectory == 0) { 1146 /* cache_purge removed - handled by VFS compat layer */ 1147 } else { 1148 error = EISDIR; 1149 goto bad; 1150 } 1151 /* 1152 * note: inode passed to ufs_dirrewrite() is 0 for a 1153 * non-directory file rename, 1 for a directory rename 1154 * in the same directory, and > 1 for an inode representing 1155 * the new directory. 1156 */ 1157 error = ufs_dirrewrite(dp, xp, ip->i_number, 1158 IFTODT(ip->i_mode), 1159 (doingdirectory && newparent) ? 1160 newparent : (ino_t)doingdirectory); 1161 if (error) 1162 goto bad; 1163 if (doingdirectory) { 1164 if (!newparent) { 1165 dp->i_effnlink--; 1166 if (DOINGSOFTDEP(tdvp)) 1167 softdep_change_linkcnt(dp); 1168 } 1169 xp->i_effnlink--; 1170 if (DOINGSOFTDEP(tvp)) 1171 softdep_change_linkcnt(xp); 1172 } 1173 if (doingdirectory && !DOINGSOFTDEP(tvp)) { 1174 /* 1175 * Truncate inode. The only stuff left in the directory 1176 * is "." and "..". The "." reference is inconsequential 1177 * since we are quashing it. We have removed the "." 1178 * reference and the reference in the parent directory, 1179 * but there may be other hard links. The soft 1180 * dependency code will arrange to do these operations 1181 * after the parent directory entry has been deleted on 1182 * disk, so when running with that code we avoid doing 1183 * them now. 1184 */ 1185 if (!newparent) { 1186 dp->i_nlink--; 1187 dp->i_flag |= IN_CHANGE; 1188 } 1189 xp->i_nlink--; 1190 xp->i_flag |= IN_CHANGE; 1191 ioflag = DOINGASYNC(tvp) ? 0 : IO_SYNC; 1192 error = ffs_truncate(tvp, (off_t)0, ioflag, 1193 tcnp->cn_cred); 1194 if (error) 1195 goto bad; 1196 } 1197 VN_KNOTE(tdvp, NOTE_WRITE); 1198 vput(tdvp); 1199 VN_KNOTE(tvp, NOTE_DELETE); 1200 vput(tvp); 1201 xp = NULL; 1202 } 1203 1204 /* 1205 * tvp and tdvp have been cleaned up. only fvp and fdvp (both 1206 * unlocked) remain. We are about to overwrite fvp but we have to 1207 * keep 'ip' intact so we cannot release the old fvp, which is still 1208 * refd and accessible via ap->a_fvp. 1209 * 1210 * This means we cannot use either 'bad' or 'out' to cleanup any 1211 * more. 1212 */ 1213 1214 /* 1215 * 3) Unlink the source. 1216 */ 1217 fcnp->cn_flags &= ~CNP_MODMASK; 1218 fcnp->cn_flags |= CNP_LOCKPARENT; 1219 error = relookup(fdvp, &fvp, fcnp); 1220 if (error || fvp == NULL) { 1221 /* 1222 * From name has disappeared. IN_RENAME will not be set if 1223 * we get past the panic so we don't have to clean it up. 1224 */ 1225 if (doingdirectory) 1226 panic("ufs_rename: lost dir entry"); 1227 vrele(ap->a_fvp); 1228 if (fcnp->cn_flags & CNP_PDIRUNLOCK) 1229 vrele(fdvp); 1230 else 1231 vput(fdvp); 1232 return(0); 1233 } 1234 KKASSERT((fcnp->cn_flags & CNP_PDIRUNLOCK) == 0); 1235 1236 /* 1237 * fdvp and fvp are locked. 1238 */ 1239 xp = VTOI(fvp); 1240 dp = VTOI(fdvp); 1241 1242 /* 1243 * Ensure that the directory entry still exists and has not 1244 * changed while the new name has been entered. If the source is 1245 * a file then the entry may have been unlinked or renamed. In 1246 * either case there is no further work to be done. If the source 1247 * is a directory then it cannot have been rmdir'ed; the IN_RENAME 1248 * flag ensures that it cannot be moved by another rename or removed 1249 * by a rmdir. Cleanup IN_RENAME. 1250 */ 1251 if (xp != ip) { 1252 if (doingdirectory) 1253 panic("ufs_rename: lost dir entry"); 1254 } else { 1255 /* 1256 * If the source is a directory with a 1257 * new parent, the link count of the old 1258 * parent directory must be decremented 1259 * and ".." set to point to the new parent. 1260 */ 1261 if (doingdirectory && newparent) { 1262 xp->i_offset = mastertemplate.dot_reclen; 1263 ufs_dirrewrite(xp, dp, newparent, DT_DIR, 0); 1264 /* cache_purge removed - handled by VFS compat layer */ 1265 } 1266 error = ufs_dirremove(fdvp, xp, fcnp->cn_flags, 0); 1267 xp->i_flag &= ~IN_RENAME; 1268 } 1269 1270 VN_KNOTE(fvp, NOTE_RENAME); 1271 vput(fdvp); 1272 vput(fvp); 1273 vrele(ap->a_fvp); 1274 return (error); 1275 1276 bad: 1277 if (xp) 1278 vput(ITOV(xp)); 1279 vput(ITOV(dp)); 1280 out: 1281 if (doingdirectory) 1282 ip->i_flag &= ~IN_RENAME; 1283 if (vn_lock(fvp, LK_EXCLUSIVE) == 0) { 1284 ip->i_effnlink--; 1285 ip->i_nlink--; 1286 ip->i_flag |= IN_CHANGE; 1287 ip->i_flag &= ~IN_RENAME; 1288 if (DOINGSOFTDEP(fvp)) 1289 softdep_change_linkcnt(ip); 1290 vput(fvp); 1291 } else { 1292 vrele(fvp); 1293 } 1294 return (error); 1295 } 1296 1297 /* 1298 * Mkdir system call 1299 * 1300 * ufs_mkdir(struct vnode *a_dvp, struct vnode **a_vpp, 1301 * struct componentname *a_cnp, struct vattr *a_vap) 1302 */ 1303 static 1304 int 1305 ufs_mkdir(struct vop_old_mkdir_args *ap) 1306 { 1307 struct vnode *dvp = ap->a_dvp; 1308 struct vattr *vap = ap->a_vap; 1309 struct componentname *cnp = ap->a_cnp; 1310 struct inode *ip, *dp; 1311 struct vnode *tvp; 1312 struct buf *bp; 1313 struct dirtemplate dirtemplate, *dtp; 1314 struct direct newdir; 1315 int error, dmode; 1316 long blkoff; 1317 1318 dp = VTOI(dvp); 1319 if ((nlink_t)dp->i_nlink >= LINK_MAX) { 1320 error = EMLINK; 1321 goto out; 1322 } 1323 dmode = vap->va_mode & 0777; 1324 dmode |= IFDIR; 1325 /* 1326 * Must simulate part of ufs_makeinode here to acquire the inode, 1327 * but not have it entered in the parent directory. The entry is 1328 * made later after writing "." and ".." entries. 1329 */ 1330 error = ffs_valloc(dvp, dmode, cnp->cn_cred, &tvp); 1331 if (error) 1332 goto out; 1333 ip = VTOI(tvp); 1334 ip->i_gid = dp->i_gid; 1335 #ifdef SUIDDIR 1336 { 1337 #ifdef QUOTA 1338 struct ucred ucred, *ucp; 1339 ucp = cnp->cn_cred; 1340 #endif 1341 /* 1342 * If we are hacking owners here, (only do this where told to) 1343 * and we are not giving it TO root, (would subvert quotas) 1344 * then go ahead and give it to the other user. 1345 * The new directory also inherits the SUID bit. 1346 * If user's UID and dir UID are the same, 1347 * 'give it away' so that the SUID is still forced on. 1348 */ 1349 if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) && 1350 (dp->i_mode & ISUID) && dp->i_uid) { 1351 dmode |= ISUID; 1352 ip->i_uid = dp->i_uid; 1353 #ifdef QUOTA 1354 if (dp->i_uid != cnp->cn_cred->cr_uid) { 1355 /* 1356 * Make sure the correct user gets charged 1357 * for the space. 1358 * Make a dummy credential for the victim. 1359 * XXX This seems to never be accessed out of 1360 * our context so a stack variable is ok. 1361 */ 1362 ucred.cr_ref = 1; 1363 ucred.cr_uid = ip->i_uid; 1364 ucred.cr_ngroups = 1; 1365 ucred.cr_groups[0] = dp->i_gid; 1366 ucp = &ucred; 1367 } 1368 #endif 1369 } else 1370 ip->i_uid = cnp->cn_cred->cr_uid; 1371 #ifdef QUOTA 1372 if ((error = ufs_getinoquota(ip)) || 1373 (error = ufs_chkiq(ip, 1, ucp, 0))) { 1374 ffs_vfree(tvp, ip->i_number, dmode); 1375 vput(tvp); 1376 return (error); 1377 } 1378 #endif 1379 } 1380 #else /* !SUIDDIR */ 1381 ip->i_uid = cnp->cn_cred->cr_uid; 1382 #ifdef QUOTA 1383 if ((error = ufs_getinoquota(ip)) || 1384 (error = ufs_chkiq(ip, 1, cnp->cn_cred, 0))) { 1385 ffs_vfree(tvp, ip->i_number, dmode); 1386 vput(tvp); 1387 return (error); 1388 } 1389 #endif 1390 #endif /* !SUIDDIR */ 1391 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 1392 ip->i_mode = dmode; 1393 tvp->v_type = VDIR; /* Rest init'd in getnewvnode(). */ 1394 ip->i_effnlink = 2; 1395 ip->i_nlink = 2; 1396 if (DOINGSOFTDEP(tvp)) 1397 softdep_change_linkcnt(ip); 1398 if (cnp->cn_flags & CNP_ISWHITEOUT) 1399 ip->i_flags |= UF_OPAQUE; 1400 1401 /* 1402 * Bump link count in parent directory to reflect work done below. 1403 * Should be done before reference is created so cleanup is 1404 * possible if we crash. 1405 */ 1406 dp->i_effnlink++; 1407 dp->i_nlink++; 1408 dp->i_flag |= IN_CHANGE; 1409 if (DOINGSOFTDEP(dvp)) 1410 softdep_change_linkcnt(dp); 1411 error = ffs_update(tvp, !(DOINGSOFTDEP(dvp) | DOINGASYNC(dvp))); 1412 if (error) 1413 goto bad; 1414 1415 /* 1416 * The vnode must have a VM object in order to issue buffer cache 1417 * ops on it. 1418 */ 1419 vinitvmio(tvp, DIRBLKSIZ); 1420 1421 /* 1422 * Initialize directory with "." and ".." from static template. 1423 */ 1424 if (dvp->v_mount->mnt_maxsymlinklen > 0) 1425 dtp = &mastertemplate; 1426 else 1427 dtp = (struct dirtemplate *)&omastertemplate; 1428 dirtemplate = *dtp; 1429 dirtemplate.dot_ino = ip->i_number; 1430 dirtemplate.dotdot_ino = dp->i_number; 1431 vnode_pager_setsize(tvp, DIRBLKSIZ); 1432 error = VOP_BALLOC(tvp, 0LL, DIRBLKSIZ, cnp->cn_cred, B_CLRBUF, &bp); 1433 if (error) 1434 goto bad; 1435 ip->i_size = DIRBLKSIZ; 1436 ip->i_flag |= IN_CHANGE | IN_UPDATE; 1437 bcopy((caddr_t)&dirtemplate, (caddr_t)bp->b_data, sizeof dirtemplate); 1438 if (DOINGSOFTDEP(tvp)) { 1439 /* 1440 * Ensure that the entire newly allocated block is a 1441 * valid directory so that future growth within the 1442 * block does not have to ensure that the block is 1443 * written before the inode. 1444 */ 1445 blkoff = DIRBLKSIZ; 1446 while (blkoff < bp->b_bcount) { 1447 ((struct direct *) 1448 (bp->b_data + blkoff))->d_reclen = DIRBLKSIZ; 1449 blkoff += DIRBLKSIZ; 1450 } 1451 } 1452 if ((error = ffs_update(tvp, !(DOINGSOFTDEP(tvp) | 1453 DOINGASYNC(tvp)))) != 0) { 1454 bwrite(bp); 1455 goto bad; 1456 } 1457 /* 1458 * Directory set up, now install its entry in the parent directory. 1459 * 1460 * If we are not doing soft dependencies, then we must write out the 1461 * buffer containing the new directory body before entering the new 1462 * name in the parent. If we are doing soft dependencies, then the 1463 * buffer containing the new directory body will be passed to and 1464 * released in the soft dependency code after the code has attached 1465 * an appropriate ordering dependency to the buffer which ensures that 1466 * the buffer is written before the new name is written in the parent. 1467 */ 1468 if (DOINGASYNC(dvp)) 1469 bdwrite(bp); 1470 else if (!DOINGSOFTDEP(dvp) && (error = bwrite(bp)) != 0) 1471 goto bad; 1472 ufs_makedirentry(ip, cnp, &newdir); 1473 error = ufs_direnter(dvp, tvp, &newdir, cnp, bp); 1474 1475 bad: 1476 if (error == 0) { 1477 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK); 1478 *ap->a_vpp = tvp; 1479 } else { 1480 dp->i_effnlink--; 1481 dp->i_nlink--; 1482 dp->i_flag |= IN_CHANGE; 1483 if (DOINGSOFTDEP(dvp)) 1484 softdep_change_linkcnt(dp); 1485 /* 1486 * No need to do an explicit VOP_TRUNCATE here, vrele will 1487 * do this for us because we set the link count to 0. 1488 */ 1489 ip->i_effnlink = 0; 1490 ip->i_nlink = 0; 1491 ip->i_flag |= IN_CHANGE; 1492 if (DOINGSOFTDEP(tvp)) 1493 softdep_change_linkcnt(ip); 1494 vput(tvp); 1495 } 1496 out: 1497 return (error); 1498 } 1499 1500 /* 1501 * Rmdir system call. 1502 * 1503 * ufs_rmdir(struct vnode *a_dvp, struct vnode *a_vp, 1504 * struct componentname *a_cnp) 1505 */ 1506 static 1507 int 1508 ufs_rmdir(struct vop_old_rmdir_args *ap) 1509 { 1510 struct vnode *vp = ap->a_vp; 1511 struct vnode *dvp = ap->a_dvp; 1512 struct componentname *cnp = ap->a_cnp; 1513 struct inode *ip, *dp; 1514 int error, ioflag; 1515 1516 ip = VTOI(vp); 1517 dp = VTOI(dvp); 1518 1519 /* 1520 * Do not remove a directory that is in the process of being renamed. 1521 * Verify the directory is empty (and valid). Rmdir ".." will not be 1522 * valid since ".." will contain a reference to the current directory 1523 * and thus be non-empty. Do not allow the removal of mounted on 1524 * directories (this can happen when an NFS exported filesystem 1525 * tries to remove a locally mounted on directory). 1526 */ 1527 error = 0; 1528 if (ip->i_flag & IN_RENAME) { 1529 error = EINVAL; 1530 goto out; 1531 } 1532 if (ip->i_effnlink != 2 || 1533 !ufs_dirempty(ip, dp->i_number, cnp->cn_cred)) { 1534 error = ENOTEMPTY; 1535 goto out; 1536 } 1537 if ((dp->i_flags & APPEND) 1538 || (ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))) { 1539 error = EPERM; 1540 goto out; 1541 } 1542 if (vp->v_mountedhere != 0) { 1543 error = EINVAL; 1544 goto out; 1545 } 1546 /* 1547 * Delete reference to directory before purging 1548 * inode. If we crash in between, the directory 1549 * will be reattached to lost+found, 1550 */ 1551 dp->i_effnlink--; 1552 ip->i_effnlink--; 1553 if (DOINGSOFTDEP(vp)) { 1554 softdep_change_linkcnt(dp); 1555 softdep_change_linkcnt(ip); 1556 } 1557 error = ufs_dirremove(dvp, ip, cnp->cn_flags, 1); 1558 if (error) { 1559 dp->i_effnlink++; 1560 ip->i_effnlink++; 1561 if (DOINGSOFTDEP(vp)) { 1562 softdep_change_linkcnt(dp); 1563 softdep_change_linkcnt(ip); 1564 } 1565 goto out; 1566 } 1567 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK); 1568 /* 1569 * Truncate inode. The only stuff left in the directory is "." and 1570 * "..". The "." reference is inconsequential since we are quashing 1571 * it. The soft dependency code will arrange to do these operations 1572 * after the parent directory entry has been deleted on disk, so 1573 * when running with that code we avoid doing them now. 1574 */ 1575 if (!DOINGSOFTDEP(vp)) { 1576 dp->i_nlink--; 1577 dp->i_flag |= IN_CHANGE; 1578 ip->i_nlink--; 1579 ip->i_flag |= IN_CHANGE; 1580 ioflag = DOINGASYNC(vp) ? 0 : IO_SYNC; 1581 error = ffs_truncate(vp, (off_t)0, ioflag, cnp->cn_cred); 1582 } 1583 /* cache_purge removed - handled by VFS compat layer */ 1584 #ifdef UFS_DIRHASH 1585 /* Kill any active hash; i_effnlink == 0, so it will not come back. */ 1586 if (ip->i_dirhash != NULL) 1587 ufsdirhash_free(ip); 1588 #endif 1589 out: 1590 VN_KNOTE(vp, NOTE_DELETE); 1591 return (error); 1592 } 1593 1594 /* 1595 * symlink -- make a symbolic link 1596 * 1597 * ufs_symlink(struct vnode *a_dvp, struct vnode **a_vpp, 1598 * struct componentname *a_cnp, struct vattr *a_vap, 1599 * char *a_target) 1600 */ 1601 static 1602 int 1603 ufs_symlink(struct vop_old_symlink_args *ap) 1604 { 1605 struct vnode *vp, **vpp = ap->a_vpp; 1606 struct inode *ip; 1607 int len, error; 1608 1609 error = ufs_makeinode(IFLNK | ap->a_vap->va_mode, ap->a_dvp, 1610 vpp, ap->a_cnp); 1611 if (error) 1612 return (error); 1613 VN_KNOTE(ap->a_dvp, NOTE_WRITE); 1614 vp = *vpp; 1615 len = strlen(ap->a_target); 1616 if (len < vp->v_mount->mnt_maxsymlinklen) { 1617 ip = VTOI(vp); 1618 bcopy(ap->a_target, (char *)ip->i_shortlink, len); 1619 ip->i_size = len; 1620 ip->i_flag |= IN_CHANGE | IN_UPDATE; 1621 } else { 1622 /* 1623 * Make sure we have a VM object in order to use 1624 * the buffer cache. 1625 */ 1626 if (vp->v_object == NULL) 1627 vinitvmio(vp, 0); 1628 error = vn_rdwr(UIO_WRITE, vp, ap->a_target, len, (off_t)0, 1629 UIO_SYSSPACE, IO_NODELOCKED, 1630 ap->a_cnp->cn_cred, NULL); 1631 } 1632 if (error) 1633 vput(vp); 1634 return (error); 1635 } 1636 1637 /* 1638 * Vnode op for reading directories. 1639 * 1640 * ufs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred, 1641 * int *a_eofflag, int *ncookies, u_long **a_cookies) 1642 */ 1643 static 1644 int 1645 ufs_readdir(struct vop_readdir_args *ap) 1646 { 1647 struct uio *uio = ap->a_uio; 1648 struct vnode *vp = ap->a_vp; 1649 struct direct *dp; 1650 struct buf *bp; 1651 int retval; 1652 int error; 1653 int offset; /* offset into buffer cache buffer */ 1654 int eoffset; /* end of buffer clipped to file EOF */ 1655 int pickup; /* pickup point */ 1656 int ncookies; 1657 int cookie_index; 1658 u_long *cookies; 1659 1660 if (uio->uio_offset < 0) 1661 return (EINVAL); 1662 /* 1663 * Guess the number of cookies needed. Make sure we compute at 1664 * least 1, and no more then a reasonable limit. 1665 */ 1666 if (ap->a_ncookies) { 1667 ncookies = uio->uio_resid / 16 + 1; 1668 if (ncookies > 1024) 1669 ncookies = 1024; 1670 cookies = malloc(ncookies * sizeof(u_long), M_TEMP, M_WAITOK); 1671 } else { 1672 ncookies = -1; /* force conditionals below */ 1673 cookies = NULL; 1674 } 1675 cookie_index = 0; 1676 1677 /* 1678 * Past or at EOF 1679 */ 1680 if (uio->uio_offset >= VTOI(vp)->i_size) { 1681 if (ap->a_eofflag) 1682 *ap->a_eofflag = 1; 1683 if (ap->a_ncookies) { 1684 *ap->a_ncookies = cookie_index; 1685 *ap->a_cookies = cookies; 1686 } 1687 return(0); 1688 } 1689 1690 /* 1691 * Loop until we run out of cookies, we run out of user buffer, 1692 * or we hit the directory EOF. 1693 * 1694 * Always start scans at the beginning of the buffer, don't trust 1695 * the offset supplied by userland. 1696 */ 1697 while ((error = ffs_blkatoff_ra(vp, uio->uio_offset, NULL, &bp, 2)) == 0) { 1698 pickup = (int)(uio->uio_offset - bp->b_loffset); 1699 offset = 0; 1700 retval = 0; 1701 if (bp->b_loffset + bp->b_bcount > VTOI(vp)->i_size) 1702 eoffset = (int)(VTOI(vp)->i_size - bp->b_loffset); 1703 else 1704 eoffset = bp->b_bcount; 1705 1706 while (offset < eoffset) { 1707 dp = (struct direct *)(bp->b_data + offset); 1708 if (dp->d_reclen <= 0 || (dp->d_reclen & 3) || 1709 offset + dp->d_reclen > bp->b_bcount) { 1710 error = EIO; 1711 break; 1712 } 1713 if (offsetof(struct direct, d_name[dp->d_namlen]) > dp->d_reclen) { 1714 error = EIO; 1715 break; 1716 } 1717 if (offset < pickup) { 1718 offset += dp->d_reclen; 1719 continue; 1720 } 1721 #if BYTE_ORDER == LITTLE_ENDIAN 1722 if (OFSFMT(vp)) { 1723 retval = vop_write_dirent(&error, uio, 1724 dp->d_ino, dp->d_namlen, dp->d_type, 1725 dp->d_name); 1726 } else 1727 #endif 1728 { 1729 retval = vop_write_dirent(&error, uio, 1730 dp->d_ino, dp->d_type, dp->d_namlen, 1731 dp->d_name); 1732 } 1733 if (retval) 1734 break; 1735 if (cookies) { 1736 cookies[cookie_index] = 1737 (u_long)bp->b_loffset + offset; 1738 } 1739 ++cookie_index; 1740 offset += dp->d_reclen; 1741 if (cookie_index == ncookies) 1742 break; 1743 } 1744 1745 /* 1746 * This will align the next loop to the beginning of the 1747 * next block, and pickup will calculate to 0. 1748 */ 1749 uio->uio_offset = bp->b_loffset + offset; 1750 brelse(bp); 1751 1752 if (retval || error || cookie_index == ncookies || 1753 uio->uio_offset >= VTOI(vp)->i_size) { 1754 break; 1755 } 1756 } 1757 if (ap->a_eofflag) 1758 *ap->a_eofflag = VTOI(vp)->i_size <= uio->uio_offset; 1759 1760 /* 1761 * Report errors only if we didn't manage to read anything 1762 */ 1763 if (error && cookie_index == 0) { 1764 if (cookies) { 1765 free(cookies, M_TEMP); 1766 *ap->a_ncookies = 0; 1767 *ap->a_cookies = NULL; 1768 } 1769 } else { 1770 error = 0; 1771 if (cookies) { 1772 *ap->a_ncookies = cookie_index; 1773 *ap->a_cookies = cookies; 1774 } 1775 } 1776 return (error); 1777 } 1778 1779 /* 1780 * Return target name of a symbolic link 1781 * 1782 * ufs_readlink(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred) 1783 */ 1784 static 1785 int 1786 ufs_readlink(struct vop_readlink_args *ap) 1787 { 1788 struct vnode *vp = ap->a_vp; 1789 struct inode *ip = VTOI(vp); 1790 int isize; 1791 1792 isize = ip->i_size; 1793 if ((isize < vp->v_mount->mnt_maxsymlinklen) || 1794 (ip->i_din.di_blocks == 0)) { /* XXX - for old fastlink support */ 1795 uiomove((char *)ip->i_shortlink, isize, ap->a_uio); 1796 return (0); 1797 } 1798 1799 /* 1800 * Perform the equivalent of an OPEN on vp so we can issue a 1801 * VOP_READ. 1802 */ 1803 return (VOP_READ(vp, ap->a_uio, 0, ap->a_cred)); 1804 } 1805 1806 /* 1807 * Calculate the logical to physical mapping if not done already, 1808 * then call the device strategy routine. 1809 * 1810 * In order to be able to swap to a file, the VOP_BMAP operation may not 1811 * deadlock on memory. See ufs_bmap() for details. 1812 * 1813 * ufs_strategy(struct vnode *a_vp, struct bio *a_bio) 1814 */ 1815 static 1816 int 1817 ufs_strategy(struct vop_strategy_args *ap) 1818 { 1819 struct bio *bio = ap->a_bio; 1820 struct bio *nbio; 1821 struct buf *bp = bio->bio_buf; 1822 struct vnode *vp = ap->a_vp; 1823 struct inode *ip; 1824 int error; 1825 1826 ip = VTOI(vp); 1827 if (vp->v_type == VBLK || vp->v_type == VCHR) 1828 panic("ufs_strategy: spec"); 1829 nbio = push_bio(bio); 1830 if (nbio->bio_offset == NOOFFSET) { 1831 error = VOP_BMAP(vp, bio->bio_offset, NULL, &nbio->bio_offset, 1832 NULL, NULL); 1833 if (error) { 1834 bp->b_error = error; 1835 bp->b_flags |= B_ERROR; 1836 /* I/O was never started on nbio, must biodone(bio) */ 1837 biodone(bio); 1838 return (error); 1839 } 1840 if (nbio->bio_offset == NOOFFSET) 1841 vfs_bio_clrbuf(bp); 1842 } 1843 if (nbio->bio_offset == NOOFFSET) { 1844 /* 1845 * We hit a hole in the file. The buffer has been zero-filled 1846 * so just biodone() it. 1847 */ 1848 biodone(bio); 1849 } else { 1850 vn_strategy(ip->i_devvp, nbio); 1851 } 1852 return (0); 1853 } 1854 1855 /* 1856 * Print out the contents of an inode. 1857 * 1858 * ufs_print(struct vnode *a_vp) 1859 */ 1860 static 1861 int 1862 ufs_print(struct vop_print_args *ap) 1863 { 1864 struct vnode *vp = ap->a_vp; 1865 struct inode *ip = VTOI(vp); 1866 1867 printf("tag VT_UFS, ino %lu, on dev %s (%d, %d)", 1868 (u_long)ip->i_number, devtoname(ip->i_dev), major(ip->i_dev), 1869 minor(ip->i_dev)); 1870 if (vp->v_type == VFIFO) 1871 fifo_printinfo(vp); 1872 lockmgr_printinfo(&vp->v_lock); 1873 printf("\n"); 1874 return (0); 1875 } 1876 1877 /* 1878 * Read wrapper for special devices. 1879 * 1880 * ufsspec_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 1881 * struct ucred *a_cred) 1882 */ 1883 static 1884 int 1885 ufsspec_read(struct vop_read_args *ap) 1886 { 1887 int error, resid; 1888 struct inode *ip; 1889 struct uio *uio; 1890 1891 uio = ap->a_uio; 1892 resid = uio->uio_resid; 1893 error = VOCALL(&spec_vnode_vops, &ap->a_head); 1894 /* 1895 * The inode may have been revoked during the call, so it must not 1896 * be accessed blindly here or in the other wrapper functions. 1897 */ 1898 ip = VTOI(ap->a_vp); 1899 if (ip != NULL && (uio->uio_resid != resid || (error == 0 && resid != 0))) 1900 ip->i_flag |= IN_ACCESS; 1901 return (error); 1902 } 1903 1904 /* 1905 * Write wrapper for special devices. 1906 * 1907 * ufsspec_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 1908 * struct ucred *a_cred) 1909 */ 1910 static 1911 int 1912 ufsspec_write(struct vop_write_args *ap) 1913 { 1914 int error, resid; 1915 struct inode *ip; 1916 struct uio *uio; 1917 1918 uio = ap->a_uio; 1919 resid = uio->uio_resid; 1920 error = VOCALL(&spec_vnode_vops, &ap->a_head); 1921 ip = VTOI(ap->a_vp); 1922 if (ip != NULL && (uio->uio_resid != resid || (error == 0 && resid != 0))) 1923 VTOI(ap->a_vp)->i_flag |= IN_CHANGE | IN_UPDATE; 1924 return (error); 1925 } 1926 1927 /* 1928 * Close wrapper for special devices. 1929 * 1930 * Update the times on the inode then do device close. 1931 * 1932 * ufsspec_close(struct vnode *a_vp, int a_fflag, struct ucred *a_cred, 1933 * struct thread *a_td) 1934 */ 1935 static 1936 int 1937 ufsspec_close(struct vop_close_args *ap) 1938 { 1939 struct vnode *vp = ap->a_vp; 1940 1941 if (vp->v_usecount > 1) 1942 ufs_itimes(vp); 1943 return (VOCALL(&spec_vnode_vops, &ap->a_head)); 1944 } 1945 1946 /* 1947 * Read wrapper for fifos. 1948 * 1949 * ufsfifo_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 1950 * struct ucred *a_cred) 1951 */ 1952 static 1953 int 1954 ufsfifo_read(struct vop_read_args *ap) 1955 { 1956 int error, resid; 1957 struct inode *ip; 1958 struct uio *uio; 1959 1960 uio = ap->a_uio; 1961 resid = uio->uio_resid; 1962 error = VOCALL(&fifo_vnode_vops, &ap->a_head); 1963 ip = VTOI(ap->a_vp); 1964 if ((ap->a_vp->v_mount->mnt_flag & MNT_NOATIME) == 0 && ip != NULL && 1965 (uio->uio_resid != resid || (error == 0 && resid != 0))) 1966 VTOI(ap->a_vp)->i_flag |= IN_ACCESS; 1967 return (error); 1968 } 1969 1970 /* 1971 * Write wrapper for fifos. 1972 * 1973 * ufsfifo_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 1974 * struct ucred *a_cred) 1975 */ 1976 static 1977 int 1978 ufsfifo_write(struct vop_write_args *ap) 1979 { 1980 int error, resid; 1981 struct inode *ip; 1982 struct uio *uio; 1983 1984 uio = ap->a_uio; 1985 resid = uio->uio_resid; 1986 error = VOCALL(&fifo_vnode_vops, &ap->a_head); 1987 ip = VTOI(ap->a_vp); 1988 if (ip != NULL && (uio->uio_resid != resid || (error == 0 && resid != 0))) 1989 VTOI(ap->a_vp)->i_flag |= IN_CHANGE | IN_UPDATE; 1990 return (error); 1991 } 1992 1993 /* 1994 * Close wrapper for fifos. 1995 * 1996 * Update the times on the inode then do device close. 1997 * 1998 * ufsfifo_close(struct vnode *a_vp, int a_fflag, struct ucred *a_cred, 1999 * struct thread *a_td) 2000 */ 2001 static 2002 int 2003 ufsfifo_close(struct vop_close_args *ap) 2004 { 2005 struct vnode *vp = ap->a_vp; 2006 2007 if (vp->v_usecount > 1) 2008 ufs_itimes(vp); 2009 return (VOCALL(&fifo_vnode_vops, &ap->a_head)); 2010 } 2011 2012 /* 2013 * Kqfilter wrapper for fifos. 2014 * 2015 * Fall through to ufs kqfilter routines if needed 2016 */ 2017 static 2018 int 2019 ufsfifo_kqfilter(struct vop_kqfilter_args *ap) 2020 { 2021 int error; 2022 2023 error = VOCALL(&fifo_vnode_vops, &ap->a_head); 2024 if (error) 2025 error = ufs_kqfilter(ap); 2026 return (error); 2027 } 2028 2029 /* 2030 * Return POSIX pathconf information applicable to ufs filesystems. 2031 * 2032 * ufs_pathconf(struct vnode *a_vp, int a_name, int *a_retval) 2033 */ 2034 static 2035 int 2036 ufs_pathconf(struct vop_pathconf_args *ap) 2037 { 2038 switch (ap->a_name) { 2039 case _PC_LINK_MAX: 2040 *ap->a_retval = LINK_MAX; 2041 return (0); 2042 case _PC_NAME_MAX: 2043 *ap->a_retval = NAME_MAX; 2044 return (0); 2045 case _PC_PATH_MAX: 2046 *ap->a_retval = PATH_MAX; 2047 return (0); 2048 case _PC_PIPE_BUF: 2049 *ap->a_retval = PIPE_BUF; 2050 return (0); 2051 case _PC_CHOWN_RESTRICTED: 2052 *ap->a_retval = 1; 2053 return (0); 2054 case _PC_NO_TRUNC: 2055 *ap->a_retval = 1; 2056 return (0); 2057 default: 2058 return (EINVAL); 2059 } 2060 /* NOTREACHED */ 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, 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_gid = pdir->i_gid; 2155 #ifdef SUIDDIR 2156 { 2157 #ifdef QUOTA 2158 struct ucred ucred, *ucp; 2159 ucp = cnp->cn_cred; 2160 #endif 2161 /* 2162 * If we are not the owner of the directory, 2163 * and we are hacking owners here, (only do this where told to) 2164 * and we are not giving it TO root, (would subvert quotas) 2165 * then go ahead and give it to the other user. 2166 * Note that this drops off the execute bits for security. 2167 */ 2168 if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) && 2169 (pdir->i_mode & ISUID) && 2170 (pdir->i_uid != cnp->cn_cred->cr_uid) && pdir->i_uid) { 2171 ip->i_uid = pdir->i_uid; 2172 mode &= ~07111; 2173 #ifdef QUOTA 2174 /* 2175 * Make sure the correct user gets charged 2176 * for the space. 2177 * Quickly knock up a dummy credential for the victim. 2178 * XXX This seems to never be accessed out of our 2179 * context so a stack variable is ok. 2180 */ 2181 ucred.cr_ref = 1; 2182 ucred.cr_uid = ip->i_uid; 2183 ucred.cr_ngroups = 1; 2184 ucred.cr_groups[0] = pdir->i_gid; 2185 ucp = &ucred; 2186 #endif 2187 } else 2188 ip->i_uid = cnp->cn_cred->cr_uid; 2189 2190 #ifdef QUOTA 2191 if ((error = ufs_getinoquota(ip)) || 2192 (error = ufs_chkiq(ip, 1, ucp, 0))) { 2193 ffs_vfree(tvp, ip->i_number, mode); 2194 vput(tvp); 2195 return (error); 2196 } 2197 #endif 2198 } 2199 #else /* !SUIDDIR */ 2200 ip->i_uid = cnp->cn_cred->cr_uid; 2201 #ifdef QUOTA 2202 if ((error = ufs_getinoquota(ip)) || 2203 (error = ufs_chkiq(ip, 1, cnp->cn_cred, 0))) { 2204 ffs_vfree(tvp, ip->i_number, mode); 2205 vput(tvp); 2206 return (error); 2207 } 2208 #endif 2209 #endif /* !SUIDDIR */ 2210 ip->i_fsmid = cache_getnewfsmid(); 2211 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 2212 ip->i_mode = mode; 2213 tvp->v_type = IFTOVT(mode); /* Rest init'd in getnewvnode(). */ 2214 ip->i_effnlink = 1; 2215 ip->i_nlink = 1; 2216 if (DOINGSOFTDEP(tvp)) 2217 softdep_change_linkcnt(ip); 2218 if ((ip->i_mode & ISGID) && !groupmember(ip->i_gid, cnp->cn_cred) && 2219 suser_cred(cnp->cn_cred, 0)) { 2220 ip->i_mode &= ~ISGID; 2221 } 2222 2223 if (cnp->cn_flags & CNP_ISWHITEOUT) 2224 ip->i_flags |= UF_OPAQUE; 2225 2226 /* 2227 * Regular files and directories need VM objects. Softlinks do 2228 * not (not immediately anyway). 2229 */ 2230 if (tvp->v_type == VREG || tvp->v_type == VDIR) 2231 vinitvmio(tvp, 0); 2232 2233 /* 2234 * Make sure inode goes to disk before directory entry. 2235 */ 2236 error = ffs_update(tvp, !(DOINGSOFTDEP(tvp) | DOINGASYNC(tvp))); 2237 if (error) 2238 goto bad; 2239 ufs_makedirentry(ip, cnp, &newdir); 2240 error = ufs_direnter(dvp, tvp, &newdir, cnp, NULL); 2241 if (error) 2242 goto bad; 2243 *vpp = tvp; 2244 return (0); 2245 2246 bad: 2247 /* 2248 * Write error occurred trying to update the inode 2249 * or the directory so must deallocate the inode. 2250 */ 2251 ip->i_effnlink = 0; 2252 ip->i_nlink = 0; 2253 ip->i_flag |= IN_CHANGE; 2254 if (DOINGSOFTDEP(tvp)) 2255 softdep_change_linkcnt(ip); 2256 vput(tvp); 2257 return (error); 2258 } 2259 2260 static int 2261 ufs_missingop(struct vop_generic_args *ap) 2262 { 2263 panic("no vop function for %s in ufs child", ap->a_desc->sd_name); 2264 return (EOPNOTSUPP); 2265 } 2266 2267 static struct filterops ufsread_filtops = 2268 { 1, NULL, filt_ufsdetach, filt_ufsread }; 2269 static struct filterops ufswrite_filtops = 2270 { 1, NULL, filt_ufsdetach, filt_ufswrite }; 2271 static struct filterops ufsvnode_filtops = 2272 { 1, NULL, filt_ufsdetach, filt_ufsvnode }; 2273 2274 /* 2275 * ufs_kqfilter(struct vnode *a_vp, struct knote *a_kn) 2276 */ 2277 static int 2278 ufs_kqfilter(struct vop_kqfilter_args *ap) 2279 { 2280 struct vnode *vp = ap->a_vp; 2281 struct knote *kn = ap->a_kn; 2282 lwkt_tokref ilock; 2283 2284 switch (kn->kn_filter) { 2285 case EVFILT_READ: 2286 kn->kn_fop = &ufsread_filtops; 2287 break; 2288 case EVFILT_WRITE: 2289 kn->kn_fop = &ufswrite_filtops; 2290 break; 2291 case EVFILT_VNODE: 2292 kn->kn_fop = &ufsvnode_filtops; 2293 break; 2294 default: 2295 return (1); 2296 } 2297 2298 kn->kn_hook = (caddr_t)vp; 2299 2300 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token); 2301 SLIST_INSERT_HEAD(&vp->v_pollinfo.vpi_selinfo.si_note, kn, kn_selnext); 2302 lwkt_reltoken(&ilock); 2303 2304 return (0); 2305 } 2306 2307 static void 2308 filt_ufsdetach(struct knote *kn) 2309 { 2310 struct vnode *vp = (struct vnode *)kn->kn_hook; 2311 lwkt_tokref ilock; 2312 2313 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token); 2314 SLIST_REMOVE(&vp->v_pollinfo.vpi_selinfo.si_note, 2315 kn, knote, kn_selnext); 2316 lwkt_reltoken(&ilock); 2317 } 2318 2319 /*ARGSUSED*/ 2320 static int 2321 filt_ufsread(struct knote *kn, long hint) 2322 { 2323 struct vnode *vp = (struct vnode *)kn->kn_hook; 2324 struct inode *ip = VTOI(vp); 2325 2326 /* 2327 * filesystem is gone, so set the EOF flag and schedule 2328 * the knote for deletion. 2329 */ 2330 if (hint == NOTE_REVOKE) { 2331 kn->kn_flags |= (EV_EOF | EV_ONESHOT); 2332 return (1); 2333 } 2334 2335 kn->kn_data = ip->i_size - kn->kn_fp->f_offset; 2336 return (kn->kn_data != 0); 2337 } 2338 2339 /*ARGSUSED*/ 2340 static int 2341 filt_ufswrite(struct knote *kn, long hint) 2342 { 2343 /* 2344 * filesystem is gone, so set the EOF flag and schedule 2345 * the knote for deletion. 2346 */ 2347 if (hint == NOTE_REVOKE) 2348 kn->kn_flags |= (EV_EOF | EV_ONESHOT); 2349 2350 kn->kn_data = 0; 2351 return (1); 2352 } 2353 2354 static int 2355 filt_ufsvnode(struct knote *kn, long hint) 2356 { 2357 if (kn->kn_sfflags & hint) 2358 kn->kn_fflags |= hint; 2359 if (hint == NOTE_REVOKE) { 2360 kn->kn_flags |= EV_EOF; 2361 return (1); 2362 } 2363 return (kn->kn_fflags != 0); 2364 } 2365 2366 /* Global vfs data structures for ufs. */ 2367 static struct vop_ops ufs_vnode_vops = { 2368 .vop_default = vop_defaultop, 2369 .vop_fsync = (void *)ufs_missingop, 2370 .vop_read = (void *)ufs_missingop, 2371 .vop_reallocblks = (void *)ufs_missingop, 2372 .vop_write = (void *)ufs_missingop, 2373 .vop_access = ufs_access, 2374 .vop_advlock = ufs_advlock, 2375 .vop_bmap = ufs_bmap, 2376 .vop_old_lookup = ufs_lookup, 2377 .vop_close = ufs_close, 2378 .vop_old_create = ufs_create, 2379 .vop_getattr = ufs_getattr, 2380 .vop_inactive = ufs_inactive, 2381 .vop_islocked = vop_stdislocked, 2382 .vop_old_link = ufs_link, 2383 .vop_lock = vop_stdlock, 2384 .vop_old_mkdir = ufs_mkdir, 2385 .vop_old_mknod = ufs_mknod, 2386 .vop_mmap = ufs_mmap, 2387 .vop_open = ufs_open, 2388 .vop_pathconf = ufs_pathconf, 2389 .vop_poll = vop_stdpoll, 2390 .vop_kqfilter = ufs_kqfilter, 2391 .vop_print = ufs_print, 2392 .vop_readdir = ufs_readdir, 2393 .vop_readlink = ufs_readlink, 2394 .vop_reclaim = ufs_reclaim, 2395 .vop_old_remove = ufs_remove, 2396 .vop_old_rename = ufs_rename, 2397 .vop_old_rmdir = ufs_rmdir, 2398 .vop_setattr = ufs_setattr, 2399 .vop_strategy = ufs_strategy, 2400 .vop_old_symlink = ufs_symlink, 2401 .vop_unlock = vop_stdunlock, 2402 .vop_old_whiteout = ufs_whiteout 2403 }; 2404 2405 static struct vop_ops ufs_spec_vops = { 2406 .vop_default = spec_vnoperate, 2407 .vop_fsync = (void *)ufs_missingop, 2408 .vop_access = ufs_access, 2409 .vop_close = ufsspec_close, 2410 .vop_getattr = ufs_getattr, 2411 .vop_inactive = ufs_inactive, 2412 .vop_islocked = vop_stdislocked, 2413 .vop_lock = vop_stdlock, 2414 .vop_print = ufs_print, 2415 .vop_read = ufsspec_read, 2416 .vop_reclaim = ufs_reclaim, 2417 .vop_setattr = ufs_setattr, 2418 .vop_unlock = vop_stdunlock, 2419 .vop_write = ufsspec_write 2420 }; 2421 2422 static struct vop_ops ufs_fifo_vops = { 2423 .vop_default = fifo_vnoperate, 2424 .vop_fsync = (void *)ufs_missingop, 2425 .vop_access = ufs_access, 2426 .vop_close = ufsfifo_close, 2427 .vop_getattr = ufs_getattr, 2428 .vop_inactive = ufs_inactive, 2429 .vop_islocked = vop_stdislocked, 2430 .vop_kqfilter = ufsfifo_kqfilter, 2431 .vop_lock = vop_stdlock, 2432 .vop_print = ufs_print, 2433 .vop_read = ufsfifo_read, 2434 .vop_reclaim = ufs_reclaim, 2435 .vop_setattr = ufs_setattr, 2436 .vop_unlock = vop_stdunlock, 2437 .vop_write = ufsfifo_write 2438 }; 2439 2440 VNODEOP_SET(ufs_vnode_vops); 2441 VNODEOP_SET(ufs_spec_vops); 2442 VNODEOP_SET(ufs_fifo_vops); 2443 2444 /* 2445 * ufs_vnoperate() 2446 */ 2447 int 2448 ufs_vnoperate(struct vop_generic_args *ap) 2449 { 2450 return (VOCALL(&ufs_vnode_vops, ap)); 2451 } 2452 2453 /* 2454 * ufs_vnoperatefifo() 2455 */ 2456 int 2457 ufs_vnoperatefifo(struct vop_generic_args *ap) 2458 { 2459 return (VOCALL(&ufs_fifo_vops, ap)); 2460 } 2461 2462 /* 2463 * ufs_vnoperatespec() 2464 */ 2465 int 2466 ufs_vnoperatespec(struct vop_generic_args *ap) 2467 { 2468 return (VOCALL(&ufs_spec_vops, ap)); 2469 } 2470