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