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_print (struct vop_print_args *); 93 static int ufs_readdir (struct vop_readdir_args *); 94 static int ufs_readlink (struct vop_readlink_args *); 95 static int ufs_remove (struct vop_old_remove_args *); 96 static int ufs_rename (struct vop_old_rename_args *); 97 static int ufs_rmdir (struct vop_old_rmdir_args *); 98 static int ufs_setattr (struct vop_setattr_args *); 99 static int ufs_strategy (struct vop_strategy_args *); 100 static int ufs_symlink (struct vop_old_symlink_args *); 101 static int ufs_whiteout (struct vop_old_whiteout_args *); 102 static int ufsfifo_close (struct vop_close_args *); 103 static int ufsfifo_kqfilter (struct vop_kqfilter_args *); 104 static int ufsfifo_read (struct vop_read_args *); 105 static int ufsfifo_write (struct vop_write_args *); 106 static int filt_ufsread (struct knote *kn, long hint); 107 static int filt_ufswrite (struct knote *kn, long hint); 108 static int filt_ufsvnode (struct knote *kn, long hint); 109 static void filt_ufsdetach (struct knote *kn); 110 static int ufs_kqfilter (struct vop_kqfilter_args *ap); 111 112 union _qcvt { 113 int64_t qcvt; 114 int32_t val[2]; 115 }; 116 #define SETHIGH(q, h) { \ 117 union _qcvt tmp; \ 118 tmp.qcvt = (q); \ 119 tmp.val[_QUAD_HIGHWORD] = (h); \ 120 (q) = tmp.qcvt; \ 121 } 122 #define SETLOW(q, l) { \ 123 union _qcvt tmp; \ 124 tmp.qcvt = (q); \ 125 tmp.val[_QUAD_LOWWORD] = (l); \ 126 (q) = tmp.qcvt; \ 127 } 128 #define VN_KNOTE(vp, b) \ 129 KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, (b)) 130 131 #define OFSFMT(vp) ((vp)->v_mount->mnt_maxsymlinklen <= 0) 132 133 /* 134 * A virgin directory (no blushing please). 135 */ 136 static struct dirtemplate mastertemplate = { 137 0, 12, DT_DIR, 1, ".", 138 0, DIRBLKSIZ - 12, DT_DIR, 2, ".." 139 }; 140 static struct odirtemplate omastertemplate = { 141 0, 12, 1, ".", 142 0, DIRBLKSIZ - 12, 2, ".." 143 }; 144 145 void 146 ufs_itimes(struct vnode *vp) 147 { 148 struct inode *ip; 149 struct timespec ts; 150 151 ip = VTOI(vp); 152 if ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE)) == 0) 153 return; 154 if ((vp->v_type == VBLK || vp->v_type == VCHR) && !DOINGSOFTDEP(vp)) 155 ip->i_flag |= IN_LAZYMOD; 156 else 157 ip->i_flag |= IN_MODIFIED; 158 159 if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 160 vfs_timestamp(&ts); 161 if (ip->i_flag & IN_ACCESS) { 162 ip->i_atime = (uint32_t)ts.tv_sec; 163 ip->i_atime_ext = ts.tv_sec >> 32; 164 ip->i_atimensec = ts.tv_nsec; 165 } 166 if (ip->i_flag & IN_CHANGE) { 167 ip->i_ctime = (uint32_t)ts.tv_sec; 168 ip->i_ctime_ext = ts.tv_sec >> 32; 169 ip->i_ctimensec = ts.tv_nsec; 170 } 171 if (ip->i_flag & IN_UPDATE) { 172 if (ip->i_flag & IN_NOCOPYWRITE) { 173 if (vp->v_flag & VLASTWRITETS) { 174 ip->i_mtime = (uint32_t) 175 vp->v_lastwrite_ts.tv_sec; 176 ip->i_mtime_ext = 177 vp->v_lastwrite_ts.tv_sec >> 32; 178 ip->i_mtimensec = 179 vp->v_lastwrite_ts.tv_nsec; 180 } 181 } else { 182 ip->i_mtime = (uint32_t)ts.tv_sec; 183 ip->i_mtime_ext = ts.tv_sec >> 32; 184 ip->i_mtimensec = ts.tv_nsec; 185 } 186 ip->i_modrev++; 187 } 188 } 189 ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_UPDATE); 190 } 191 192 /* 193 * Create a regular file 194 * 195 * ufs_create(struct vnode *a_dvp, struct vnode **a_vpp, 196 * struct componentname *a_cnp, struct vattr *a_vap) 197 */ 198 static 199 int 200 ufs_create(struct vop_old_create_args *ap) 201 { 202 int error; 203 204 error = 205 ufs_makeinode(MAKEIMODE(ap->a_vap->va_type, ap->a_vap->va_mode), 206 ap->a_dvp, ap->a_vpp, ap->a_cnp); 207 if (error) 208 return (error); 209 VN_KNOTE(ap->a_dvp, NOTE_WRITE); 210 return (0); 211 } 212 213 /* 214 * Mknod vnode call 215 * 216 * ufs_mknod(struct vnode *a_dvp, struct vnode **a_vpp, 217 * struct componentname *a_cnp, struct vattr *a_vap) 218 */ 219 /* ARGSUSED */ 220 static 221 int 222 ufs_mknod(struct vop_old_mknod_args *ap) 223 { 224 struct vattr *vap = ap->a_vap; 225 struct vnode **vpp = ap->a_vpp; 226 struct inode *ip; 227 ino_t ino; 228 int error; 229 230 /* 231 * UFS cannot represent the entire major/minor range supported by 232 * the kernel. 233 */ 234 if (vap->va_rmajor != VNOVAL && 235 makeudev(vap->va_rmajor, vap->va_rminor) == NOUDEV) { 236 return(EINVAL); 237 } 238 239 /* no special directory support */ 240 if (vap->va_type == VDIR) 241 return(EINVAL); 242 243 error = ufs_makeinode(MAKEIMODE(vap->va_type, vap->va_mode), 244 ap->a_dvp, vpp, ap->a_cnp); 245 if (error) 246 return (error); 247 VN_KNOTE(ap->a_dvp, NOTE_WRITE); 248 ip = VTOI(*vpp); 249 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 250 if (vap->va_rmajor != VNOVAL) { 251 /* 252 * Want to be able to use this to make badblock 253 * inodes, so don't truncate the dev number. 254 */ 255 ip->i_rdev = makeudev(vap->va_rmajor, vap->va_rminor); 256 } 257 /* 258 * Remove inode, then reload it through VFS_VGET so it is 259 * checked to see if it is an alias of an existing entry in 260 * the inode cache. 261 */ 262 (*vpp)->v_type = VNON; 263 ino = ip->i_number; /* Save this before vgone() invalidates ip. */ 264 vgone_vxlocked(*vpp); 265 vput(*vpp); 266 error = VFS_VGET(ap->a_dvp->v_mount, NULL, ino, vpp); 267 if (error) { 268 *vpp = NULL; 269 return (error); 270 } 271 return (0); 272 } 273 274 /* 275 * Close called. 276 * 277 * Update the times on the inode. 278 * 279 * ufs_close(struct vnode *a_vp, int a_fflag) 280 */ 281 /* ARGSUSED */ 282 static 283 int 284 ufs_close(struct vop_close_args *ap) 285 { 286 struct vnode *vp = ap->a_vp; 287 288 if (VREFCNT(vp) > 1) 289 ufs_itimes(vp); 290 return (vop_stdclose(ap)); 291 } 292 293 /* 294 * ufs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred) 295 */ 296 static 297 int 298 ufs_access(struct vop_access_args *ap) 299 { 300 struct vnode *vp = ap->a_vp; 301 struct inode *ip = VTOI(vp); 302 int error; 303 304 #ifdef QUOTA 305 if (ap->a_mode & VWRITE) { 306 switch (vp->v_type) { 307 case VDIR: 308 case VLNK: 309 case VREG: 310 if ((error = ufs_getinoquota(ip)) != 0) 311 return (error); 312 break; 313 default: 314 break; 315 } 316 } 317 #endif 318 319 error = vop_helper_access(ap, ip->i_uid, ip->i_gid, ip->i_mode, 0); 320 return (error); 321 } 322 323 /* 324 * ufs_getattr(struct vnode *a_vp, struct vattr *a_vap) 325 */ 326 /* ARGSUSED */ 327 static 328 int 329 ufs_getattr(struct vop_getattr_args *ap) 330 { 331 struct vnode *vp = ap->a_vp; 332 struct inode *ip = VTOI(vp); 333 struct vattr *vap = ap->a_vap; 334 335 ufs_itimes(vp); 336 /* 337 * Copy from inode table 338 */ 339 vap->va_fsid = dev2udev(ip->i_dev); 340 vap->va_fileid = ip->i_number; 341 vap->va_mode = ip->i_mode & ~IFMT; 342 vap->va_nlink = VFSTOUFS(vp->v_mount)->um_i_effnlink_valid ? 343 ip->i_effnlink : ip->i_nlink; 344 vap->va_uid = ip->i_uid; 345 vap->va_gid = ip->i_gid; 346 vap->va_rmajor = umajor(ip->i_rdev); 347 vap->va_rminor = uminor(ip->i_rdev); 348 vap->va_size = ip->i_din.di_size; 349 vap->va_atime.tv_sec = 350 (time_t)(ip->i_atime | ((uint64_t)ip->i_atime_ext << 32)); 351 vap->va_atime.tv_nsec = ip->i_atimensec; 352 vap->va_mtime.tv_sec = 353 (time_t)(ip->i_mtime | ((uint64_t)ip->i_mtime_ext << 32)); 354 vap->va_mtime.tv_nsec = ip->i_mtimensec; 355 vap->va_ctime.tv_sec = 356 (time_t)(ip->i_ctime | ((uint64_t)ip->i_ctime_ext << 32)); 357 vap->va_ctime.tv_nsec = ip->i_ctimensec; 358 vap->va_flags = ip->i_flags; 359 vap->va_gen = ip->i_gen; 360 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize; 361 vap->va_bytes = dbtob((u_quad_t)ip->i_blocks); 362 vap->va_type = IFTOVT(ip->i_mode); 363 vap->va_filerev = ip->i_modrev; 364 return (0); 365 } 366 367 static 368 int 369 ufs_markatime(struct vop_markatime_args *ap) 370 { 371 struct vnode *vp = ap->a_vp; 372 struct inode *ip = VTOI(vp); 373 374 if (vp->v_mount->mnt_flag & MNT_RDONLY) 375 return (EROFS); 376 if (vp->v_mount->mnt_flag & MNT_NOATIME) 377 return (0); 378 ip->i_flag |= IN_ACCESS; 379 VN_KNOTE(vp, NOTE_ATTRIB); 380 return (0); 381 } 382 383 /* 384 * Set attribute vnode op. called from several syscalls 385 * 386 * ufs_setattr(struct vnode *a_vp, struct vattr *a_vap, 387 * struct ucred *a_cred) 388 */ 389 static 390 int 391 ufs_setattr(struct vop_setattr_args *ap) 392 { 393 struct vattr *vap = ap->a_vap; 394 struct vnode *vp = ap->a_vp; 395 struct inode *ip = VTOI(vp); 396 struct ucred *cred = ap->a_cred; 397 int error; 398 399 /* 400 * Check for unsettable attributes. 401 */ 402 if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) || 403 (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) || 404 (vap->va_blocksize != VNOVAL) || (vap->va_rmajor != VNOVAL) || 405 ((int)vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) { 406 return (EINVAL); 407 } 408 if (vap->va_flags != VNOVAL) { 409 if (vp->v_mount->mnt_flag & MNT_RDONLY) 410 return (EROFS); 411 if (cred->cr_uid != ip->i_uid && 412 (error = priv_check_cred(cred, PRIV_VFS_SETATTR, 0))) 413 return (error); 414 /* 415 * Note that a root chflags becomes a user chflags when 416 * we are jailed, unless the jail.chflags_allowed sysctl 417 * is set. 418 */ 419 if (cred->cr_uid == 0 && 420 (!jailed(cred) || jail_chflags_allowed)) { 421 if ((ip->i_flags 422 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) && 423 securelevel > 0) 424 return (EPERM); 425 ip->i_flags = vap->va_flags; 426 } else { 427 if (ip->i_flags 428 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) || 429 (vap->va_flags & UF_SETTABLE) != vap->va_flags) 430 return (EPERM); 431 ip->i_flags &= SF_SETTABLE; 432 ip->i_flags |= (vap->va_flags & UF_SETTABLE); 433 } 434 ip->i_flag |= IN_CHANGE; 435 if (vap->va_flags & (IMMUTABLE | APPEND)) 436 return (0); 437 } 438 if (ip->i_flags & (IMMUTABLE | APPEND)) 439 return (EPERM); 440 /* 441 * Go through the fields and update iff not VNOVAL. 442 */ 443 if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) { 444 if (vp->v_mount->mnt_flag & MNT_RDONLY) 445 return (EROFS); 446 if ((error = ufs_chown(vp, vap->va_uid, vap->va_gid, cred)) != 0) 447 return (error); 448 } 449 if (vap->va_size != VNOVAL) { 450 /* 451 * Disallow write attempts on read-only filesystems; 452 * unless the file is a socket, fifo, or a block or 453 * character device resident on the filesystem. 454 */ 455 switch (vp->v_type) { 456 case VDIR: 457 return (EISDIR); 458 case VLNK: 459 case VREG: 460 if (vp->v_mount->mnt_flag & MNT_RDONLY) 461 return (EROFS); 462 break; 463 default: 464 break; 465 } 466 if ((error = ffs_truncate(vp, vap->va_size, 0, cred)) != 0) 467 return (error); 468 } 469 ip = VTOI(vp); 470 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) { 471 if (vp->v_mount->mnt_flag & MNT_RDONLY) 472 return (EROFS); 473 if (cred->cr_uid != ip->i_uid && 474 (error = priv_check_cred(cred, PRIV_VFS_SETATTR, 0)) && 475 ((vap->va_vaflags & VA_UTIMES_NULL) == 0 || 476 (error = VOP_EACCESS(vp, VWRITE, cred)))) 477 return (error); 478 if (vap->va_atime.tv_sec != VNOVAL) 479 ip->i_flag |= IN_ACCESS; 480 if (vap->va_mtime.tv_sec != VNOVAL) 481 ip->i_flag |= IN_CHANGE | IN_UPDATE; 482 ufs_itimes(vp); 483 if (vap->va_atime.tv_sec != VNOVAL) { 484 ip->i_atime = (uint32_t)vap->va_atime.tv_sec; 485 ip->i_atime_ext = vap->va_atime.tv_sec >> 32; 486 ip->i_atimensec = vap->va_atime.tv_nsec; 487 } 488 if (vap->va_mtime.tv_sec != VNOVAL) { 489 ip->i_mtime = (uint32_t)vap->va_mtime.tv_sec; 490 ip->i_mtime_ext = vap->va_mtime.tv_sec >> 32; 491 ip->i_mtimensec = vap->va_mtime.tv_nsec; 492 vclrflags(vp, VLASTWRITETS); 493 } 494 error = ffs_update(vp, 0); 495 if (error) 496 return (error); 497 } 498 error = 0; 499 if (vap->va_mode != (mode_t)VNOVAL) { 500 if (vp->v_mount->mnt_flag & MNT_RDONLY) 501 return (EROFS); 502 error = ufs_chmod(vp, (int)vap->va_mode, cred); 503 } 504 VN_KNOTE(vp, NOTE_ATTRIB); 505 return (error); 506 } 507 508 /* 509 * Change the mode on a file. 510 * Inode must be locked before calling. 511 */ 512 static int 513 ufs_chmod(struct vnode *vp, int mode, struct ucred *cred) 514 { 515 struct inode *ip = VTOI(vp); 516 int error; 517 mode_t cur_mode = ip->i_mode; 518 519 error = vop_helper_chmod(vp, mode, cred, ip->i_uid, ip->i_gid, 520 &cur_mode); 521 if (error) 522 return (error); 523 #if 0 524 if (cred->cr_uid != ip->i_uid) { 525 error = priv_check_cred(cred, PRIV_VFS_CHMOD, 0); 526 if (error) 527 return (error); 528 } 529 if (cred->cr_uid) { 530 if (vp->v_type != VDIR && (mode & S_ISTXT)) 531 return (EFTYPE); 532 if (!groupmember(ip->i_gid, cred) && (mode & ISGID)) 533 return (EPERM); 534 } 535 #endif 536 ip->i_mode = cur_mode; 537 ip->i_flag |= IN_CHANGE; 538 return (0); 539 } 540 541 /* 542 * Perform chown operation on inode ip; 543 * inode must be locked prior to call. 544 */ 545 static int 546 ufs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred) 547 { 548 struct inode *ip = VTOI(vp); 549 uid_t ouid; 550 gid_t ogid; 551 int error = 0; 552 #ifdef QUOTA 553 int i; 554 long change; 555 #endif 556 557 if (uid == (uid_t)VNOVAL) 558 uid = ip->i_uid; 559 if (gid == (gid_t)VNOVAL) 560 gid = ip->i_gid; 561 /* 562 * If we don't own the file, are trying to change the owner 563 * of the file, or are not a member of the target group, 564 * the caller must be superuser or the call fails. 565 */ 566 if ((cred->cr_uid != ip->i_uid || uid != ip->i_uid || 567 (gid != ip->i_gid && !(cred->cr_gid == gid || 568 groupmember(gid, cred)))) && 569 (error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0))) 570 return (error); 571 ogid = ip->i_gid; 572 ouid = ip->i_uid; 573 #ifdef QUOTA 574 if ((error = ufs_getinoquota(ip)) != 0) 575 return (error); 576 if (ouid == uid) { 577 ufs_dqrele(vp, ip->i_dquot[USRQUOTA]); 578 ip->i_dquot[USRQUOTA] = NODQUOT; 579 } 580 if (ogid == gid) { 581 ufs_dqrele(vp, ip->i_dquot[GRPQUOTA]); 582 ip->i_dquot[GRPQUOTA] = NODQUOT; 583 } 584 change = ip->i_blocks; 585 (void) ufs_chkdq(ip, -change, cred, CHOWN); 586 (void) ufs_chkiq(ip, -1, cred, CHOWN); 587 for (i = 0; i < MAXQUOTAS; i++) { 588 ufs_dqrele(vp, ip->i_dquot[i]); 589 ip->i_dquot[i] = NODQUOT; 590 } 591 #endif 592 ip->i_gid = gid; 593 ip->i_uid = uid; 594 #ifdef QUOTA 595 if ((error = ufs_getinoquota(ip)) == 0) { 596 if (ouid == uid) { 597 ufs_dqrele(vp, ip->i_dquot[USRQUOTA]); 598 ip->i_dquot[USRQUOTA] = NODQUOT; 599 } 600 if (ogid == gid) { 601 ufs_dqrele(vp, ip->i_dquot[GRPQUOTA]); 602 ip->i_dquot[GRPQUOTA] = NODQUOT; 603 } 604 if ((error = ufs_chkdq(ip, change, cred, CHOWN)) == 0) { 605 if ((error = ufs_chkiq(ip, 1, cred, CHOWN)) == 0) 606 goto good; 607 else 608 (void)ufs_chkdq(ip, -change, cred, CHOWN|FORCE); 609 } 610 for (i = 0; i < MAXQUOTAS; i++) { 611 ufs_dqrele(vp, ip->i_dquot[i]); 612 ip->i_dquot[i] = NODQUOT; 613 } 614 } 615 ip->i_gid = ogid; 616 ip->i_uid = ouid; 617 if (ufs_getinoquota(ip) == 0) { 618 if (ouid == uid) { 619 ufs_dqrele(vp, ip->i_dquot[USRQUOTA]); 620 ip->i_dquot[USRQUOTA] = NODQUOT; 621 } 622 if (ogid == gid) { 623 ufs_dqrele(vp, ip->i_dquot[GRPQUOTA]); 624 ip->i_dquot[GRPQUOTA] = NODQUOT; 625 } 626 (void) ufs_chkdq(ip, change, cred, FORCE|CHOWN); 627 (void) ufs_chkiq(ip, 1, cred, FORCE|CHOWN); 628 (void) ufs_getinoquota(ip); 629 } 630 return (error); 631 good: 632 if (ufs_getinoquota(ip)) 633 panic("ufs_chown: lost quota"); 634 #endif /* QUOTA */ 635 ip->i_flag |= IN_CHANGE; 636 if (cred->cr_uid != 0 && (ouid != uid || ogid != gid)) 637 ip->i_mode &= ~(ISUID | ISGID); 638 return (0); 639 } 640 641 /* 642 * ufs_remove(struct vnode *a_dvp, struct vnode *a_vp, 643 * struct componentname *a_cnp) 644 */ 645 static 646 int 647 ufs_remove(struct vop_old_remove_args *ap) 648 { 649 struct inode *ip; 650 struct vnode *vp = ap->a_vp; 651 struct vnode *dvp = ap->a_dvp; 652 int error; 653 654 ip = VTOI(vp); 655 #if 0 /* handled by kernel now */ 656 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) || 657 (VTOI(dvp)->i_flags & APPEND)) { 658 error = EPERM; 659 goto out; 660 } 661 #endif 662 error = ufs_dirremove(dvp, ip, ap->a_cnp->cn_flags, 0); 663 VN_KNOTE(vp, NOTE_DELETE); 664 VN_KNOTE(dvp, NOTE_WRITE); 665 #if 0 666 out: 667 #endif 668 return (error); 669 } 670 671 /* 672 * link vnode call 673 * 674 * ufs_link(struct vnode *a_tdvp, struct vnode *a_vp, 675 * struct componentname *a_cnp) 676 */ 677 static 678 int 679 ufs_link(struct vop_old_link_args *ap) 680 { 681 struct vnode *vp = ap->a_vp; 682 struct vnode *tdvp = ap->a_tdvp; 683 struct componentname *cnp = ap->a_cnp; 684 struct inode *ip; 685 struct direct newdir; 686 int error; 687 688 if (tdvp->v_mount != vp->v_mount) { 689 error = EXDEV; 690 goto out2; 691 } 692 if (tdvp != vp) { 693 error = vn_lock(vp, LK_EXCLUSIVE | LK_FAILRECLAIM); 694 if (error) 695 goto out2; 696 } 697 ip = VTOI(vp); 698 if ((nlink_t)ip->i_nlink >= LINK_MAX) { 699 error = EMLINK; 700 goto out1; 701 } 702 #if 0 /* handled by kernel now, also DragonFly allows this */ 703 if (ip->i_flags & (IMMUTABLE | APPEND)) { 704 error = EPERM; 705 goto out1; 706 } 707 #endif 708 ip->i_effnlink++; 709 ip->i_nlink++; 710 ip->i_flag |= IN_CHANGE; 711 if (DOINGSOFTDEP(vp)) 712 softdep_change_linkcnt(ip); 713 error = ffs_update(vp, !(DOINGSOFTDEP(vp) | DOINGASYNC(vp))); 714 if (!error) { 715 ufs_makedirentry(ip, cnp, &newdir); 716 error = ufs_direnter(tdvp, vp, &newdir, cnp, NULL); 717 } 718 719 if (error) { 720 ip->i_effnlink--; 721 ip->i_nlink--; 722 ip->i_flag |= IN_CHANGE; 723 if (DOINGSOFTDEP(vp)) 724 softdep_change_linkcnt(ip); 725 } 726 out1: 727 if (tdvp != vp) 728 vn_unlock(vp); 729 out2: 730 VN_KNOTE(vp, NOTE_LINK); 731 VN_KNOTE(tdvp, NOTE_WRITE); 732 return (error); 733 } 734 735 /* 736 * whiteout vnode call 737 * 738 * ufs_whiteout(struct vnode *a_dvp, struct componentname *a_cnp, int a_flags) 739 */ 740 static 741 int 742 ufs_whiteout(struct vop_old_whiteout_args *ap) 743 { 744 struct vnode *dvp = ap->a_dvp; 745 struct componentname *cnp = ap->a_cnp; 746 struct direct newdir; 747 int error = 0; 748 749 switch (ap->a_flags) { 750 case NAMEI_LOOKUP: 751 /* 4.4 format directories support whiteout operations */ 752 if (dvp->v_mount->mnt_maxsymlinklen > 0) 753 return (0); 754 return (EOPNOTSUPP); 755 756 case NAMEI_CREATE: 757 /* create a new directory whiteout */ 758 #ifdef DIAGNOSTIC 759 if (dvp->v_mount->mnt_maxsymlinklen <= 0) 760 panic("ufs_whiteout: old format filesystem"); 761 #endif 762 763 newdir.d_ino = UFS_WINO; 764 newdir.d_namlen = cnp->cn_namelen; 765 bcopy(cnp->cn_nameptr, newdir.d_name, (unsigned)cnp->cn_namelen + 1); 766 newdir.d_type = DT_WHT; 767 error = ufs_direnter(dvp, NULL, &newdir, cnp, NULL); 768 break; 769 770 case NAMEI_DELETE: 771 /* remove an existing directory whiteout */ 772 #ifdef DIAGNOSTIC 773 if (dvp->v_mount->mnt_maxsymlinklen <= 0) 774 panic("ufs_whiteout: old format filesystem"); 775 #endif 776 777 cnp->cn_flags &= ~CNP_DOWHITEOUT; 778 error = ufs_dirremove(dvp, NULL, cnp->cn_flags, 0); 779 break; 780 default: 781 panic("ufs_whiteout: unknown op"); 782 } 783 return (error); 784 } 785 786 /* 787 * Rename system call. 788 * rename("foo", "bar"); 789 * is essentially 790 * unlink("bar"); 791 * link("foo", "bar"); 792 * unlink("foo"); 793 * but ``atomically''. Can't do full commit without saving state in the 794 * inode on disk which isn't feasible at this time. Best we can do is 795 * always guarantee the target exists. 796 * 797 * Basic algorithm is: 798 * 799 * 1) Bump link count on source while we're linking it to the 800 * target. This also ensure the inode won't be deleted out 801 * from underneath us while we work (it may be truncated by 802 * a concurrent `trunc' or `open' for creation). 803 * 2) Link source to destination. If destination already exists, 804 * delete it first. 805 * 3) Unlink source reference to inode if still around. If a 806 * directory was moved and the parent of the destination 807 * is different from the source, patch the ".." entry in the 808 * directory. 809 * 810 * ufs_rename(struct vnode *a_fdvp, struct vnode *a_fvp, 811 * struct componentname *a_fcnp, struct vnode *a_tdvp, 812 * struct vnode *a_tvp, struct componentname *a_tcnp) 813 */ 814 static 815 int 816 ufs_rename(struct vop_old_rename_args *ap) 817 { 818 struct vnode *tvp = ap->a_tvp; 819 struct vnode *tdvp = ap->a_tdvp; 820 struct vnode *fvp = ap->a_fvp; 821 struct vnode *fdvp = ap->a_fdvp; 822 struct componentname *tcnp = ap->a_tcnp; 823 struct componentname *fcnp = ap->a_fcnp; 824 struct inode *ip, *xp, *dp; 825 struct direct newdir; 826 ino_t oldparent = 0, newparent = 0; 827 int doingdirectory = 0; 828 int error = 0, ioflag; 829 830 /* 831 * Check for cross-device rename. 832 */ 833 if ((fvp->v_mount != tdvp->v_mount) || 834 (tvp && (fvp->v_mount != tvp->v_mount))) { 835 error = EXDEV; 836 abortit: 837 if (tdvp == tvp) 838 vrele(tdvp); 839 else 840 vput(tdvp); 841 if (tvp) 842 vput(tvp); 843 vrele(fdvp); 844 vrele(fvp); 845 return (error); 846 } 847 848 #if 0 /* handled by kernel now */ 849 if (tvp && ((VTOI(tvp)->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) || 850 (VTOI(tdvp)->i_flags & APPEND))) { 851 error = EPERM; 852 goto abortit; 853 } 854 #endif 855 856 /* 857 * Renaming a file to itself has no effect. The upper layers should 858 * not call us in that case. Temporarily just warn if they do. 859 */ 860 if (fvp == tvp) { 861 kprintf("ufs_rename: fvp == tvp (can't happen)\n"); 862 error = 0; 863 goto abortit; 864 } 865 866 error = vn_lock(fvp, LK_EXCLUSIVE | LK_FAILRECLAIM); 867 if (error) 868 goto abortit; 869 870 /* 871 * Note: now that fvp is locked we have to be sure to unlock it before 872 * using the 'abortit' target. 873 */ 874 dp = VTOI(fdvp); 875 ip = VTOI(fvp); 876 if (ip->i_nlink >= LINK_MAX) { 877 vn_unlock(fvp); 878 error = EMLINK; 879 goto abortit; 880 } 881 #if 0 /* handled by kernel now */ 882 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) 883 || (dp->i_flags & APPEND)) { 884 vn_unlock(fvp); 885 error = EPERM; 886 goto abortit; 887 } 888 #endif 889 if ((ip->i_mode & IFMT) == IFDIR) { 890 /* 891 * Avoid ".", "..", and aliases of "." for obvious reasons. 892 */ 893 if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') || 894 dp == ip || (fcnp->cn_flags | tcnp->cn_flags) & CNP_ISDOTDOT || 895 (ip->i_flag & IN_RENAME)) { 896 vn_unlock(fvp); 897 error = EINVAL; 898 goto abortit; 899 } 900 ip->i_flag |= IN_RENAME; 901 oldparent = dp->i_number; 902 doingdirectory = 1; 903 } 904 VN_KNOTE(fdvp, NOTE_WRITE); /* XXX right place? */ 905 906 /* 907 * fvp still locked. ip->i_flag has IN_RENAME set if doingdirectory. 908 * Cleanup fvp requirements so we can unlock it. 909 * 910 * tvp and tdvp are locked. tvp may be NULL. Now that dp and xp 911 * is setup we can use the 'bad' target if we unlock fvp. We cannot 912 * use the abortit target anymore because of IN_RENAME. 913 */ 914 dp = VTOI(tdvp); 915 if (tvp) 916 xp = VTOI(tvp); 917 else 918 xp = NULL; 919 920 /* 921 * 1) Bump link count while we're moving stuff 922 * around. If we crash somewhere before 923 * completing our work, the link count 924 * may be wrong, but correctable. 925 */ 926 ip->i_effnlink++; 927 ip->i_nlink++; 928 ip->i_flag |= IN_CHANGE; 929 if (DOINGSOFTDEP(fvp)) 930 softdep_change_linkcnt(ip); 931 if ((error = ffs_update(fvp, !(DOINGSOFTDEP(fvp) | 932 DOINGASYNC(fvp)))) != 0) { 933 vn_unlock(fvp); 934 goto bad; 935 } 936 937 /* 938 * If ".." must be changed (ie the directory gets a new 939 * parent) then the source directory must not be in the 940 * directory heirarchy above the target, as this would 941 * orphan everything below the source directory. Also 942 * the user must have write permission in the source so 943 * as to be able to change "..". We must repeat the call 944 * to namei, as the parent directory is unlocked by the 945 * call to checkpath(). 946 */ 947 error = VOP_EACCESS(fvp, VWRITE, tcnp->cn_cred); 948 vn_unlock(fvp); 949 950 /* 951 * We are now back to where we were in that fvp, fdvp are unlocked 952 * and tvp, tdvp are locked. tvp may be NULL. IN_RENAME may be 953 * set. Only the bad target or, if we clean up tvp and tdvp, the 954 * out target, may be used. 955 */ 956 if (oldparent != dp->i_number) 957 newparent = dp->i_number; 958 if (doingdirectory && newparent) { 959 if (error) /* write access check above */ 960 goto bad; 961 962 /* 963 * Once we start messing with tvp and tdvp we cannot use the 964 * 'bad' target, only finish cleaning tdvp and tvp up and 965 * use the 'out' target. 966 * 967 * This cleans up tvp. 968 */ 969 if (xp != NULL) { 970 vput(tvp); 971 xp = NULL; 972 } 973 974 /* 975 * This is a real mess. ufs_checkpath vput's the target 976 * directory so retain an extra ref and note that tdvp will 977 * lose its lock on return. This leaves us with one good 978 * ref after ufs_checkpath returns. 979 */ 980 vref(tdvp); 981 error = ufs_checkpath(ip, dp, tcnp->cn_cred); 982 tcnp->cn_flags |= CNP_PDIRUNLOCK; 983 if (error) { 984 vrele(tdvp); 985 goto out; 986 } 987 988 /* 989 * relookup no longer messes with tdvp's refs. tdvp must be 990 * unlocked on entry and will be locked on a successful 991 * return. 992 */ 993 error = relookup(tdvp, &tvp, tcnp); 994 if (error) { 995 if (tcnp->cn_flags & CNP_PDIRUNLOCK) 996 vrele(tdvp); 997 else 998 vput(tdvp); 999 goto out; 1000 } 1001 KKASSERT((tcnp->cn_flags & CNP_PDIRUNLOCK) == 0); 1002 dp = VTOI(tdvp); 1003 if (tvp) 1004 xp = VTOI(tvp); 1005 } 1006 1007 /* 1008 * We are back to fvp, fdvp unlocked, tvp, tdvp locked. tvp may 1009 * be NULL (xp will also be NULL in that case), and IN_RENAME will 1010 * be set if doingdirectory. This means we can use the 'bad' target 1011 * again. 1012 */ 1013 1014 /* 1015 * 2) If target doesn't exist, link the target 1016 * to the source and unlink the source. 1017 * Otherwise, rewrite the target directory 1018 * entry to reference the source inode and 1019 * expunge the original entry's existence. 1020 */ 1021 if (xp == NULL) { 1022 if (dp->i_dev != ip->i_dev) 1023 panic("ufs_rename: EXDEV"); 1024 /* 1025 * Account for ".." in new directory. 1026 * When source and destination have the same 1027 * parent we don't fool with the link count. 1028 */ 1029 if (doingdirectory && newparent) { 1030 if ((nlink_t)dp->i_nlink >= LINK_MAX) { 1031 error = EMLINK; 1032 goto bad; 1033 } 1034 dp->i_effnlink++; 1035 dp->i_nlink++; 1036 dp->i_flag |= IN_CHANGE; 1037 if (DOINGSOFTDEP(tdvp)) 1038 softdep_change_linkcnt(dp); 1039 error = ffs_update(tdvp, !(DOINGSOFTDEP(tdvp) | 1040 DOINGASYNC(tdvp))); 1041 if (error) 1042 goto bad; 1043 } 1044 ufs_makedirentry(ip, tcnp, &newdir); 1045 error = ufs_direnter(tdvp, NULL, &newdir, tcnp, NULL); 1046 if (error) { 1047 if (doingdirectory && newparent) { 1048 dp->i_effnlink--; 1049 dp->i_nlink--; 1050 dp->i_flag |= IN_CHANGE; 1051 if (DOINGSOFTDEP(tdvp)) 1052 softdep_change_linkcnt(dp); 1053 (void)ffs_update(tdvp, 1); 1054 } 1055 goto bad; 1056 } 1057 VN_KNOTE(tdvp, NOTE_WRITE); 1058 vput(tdvp); 1059 } else { 1060 if (xp->i_dev != dp->i_dev || xp->i_dev != ip->i_dev) 1061 panic("ufs_rename: EXDEV"); 1062 /* 1063 * Short circuit rename(foo, foo). 1064 */ 1065 if (xp->i_number == ip->i_number) 1066 panic("ufs_rename: same file"); 1067 /* 1068 * If the parent directory is "sticky", then the user must 1069 * own the parent directory, or the destination of the rename, 1070 * otherwise the destination may not be changed (except by 1071 * root). This implements append-only directories. 1072 */ 1073 if ((dp->i_mode & S_ISTXT) && tcnp->cn_cred->cr_uid != 0 && 1074 tcnp->cn_cred->cr_uid != dp->i_uid && 1075 xp->i_uid != tcnp->cn_cred->cr_uid) { 1076 error = EPERM; 1077 goto bad; 1078 } 1079 /* 1080 * Target must be empty if a directory and have no links 1081 * to it. Also, ensure source and target are compatible 1082 * (both directories, or both not directories). 1083 * 1084 * Purge the file or directory being replaced from the 1085 * nameccache. 1086 */ 1087 if ((xp->i_mode&IFMT) == IFDIR) { 1088 if ((xp->i_effnlink > 2) || 1089 !ufs_dirempty(xp, dp->i_number, tcnp->cn_cred)) { 1090 error = ENOTEMPTY; 1091 goto bad; 1092 } 1093 if (!doingdirectory) { 1094 error = ENOTDIR; 1095 goto bad; 1096 } 1097 /* cache_purge removed - handled by VFS compat layer */ 1098 } else if (doingdirectory == 0) { 1099 /* cache_purge removed - handled by VFS compat layer */ 1100 } else { 1101 error = EISDIR; 1102 goto bad; 1103 } 1104 /* 1105 * note: inode passed to ufs_dirrewrite() is 0 for a 1106 * non-directory file rename, 1 for a directory rename 1107 * in the same directory, and > 1 for an inode representing 1108 * the new directory. 1109 */ 1110 error = ufs_dirrewrite(dp, xp, ip->i_number, 1111 IFTODT(ip->i_mode), 1112 (doingdirectory && newparent) ? 1113 newparent : (ino_t)doingdirectory); 1114 if (error) 1115 goto bad; 1116 if (doingdirectory) { 1117 if (!newparent) { 1118 dp->i_effnlink--; 1119 if (DOINGSOFTDEP(tdvp)) 1120 softdep_change_linkcnt(dp); 1121 } 1122 xp->i_effnlink--; 1123 if (DOINGSOFTDEP(tvp)) 1124 softdep_change_linkcnt(xp); 1125 } 1126 if (doingdirectory && !DOINGSOFTDEP(tvp)) { 1127 /* 1128 * Truncate inode. The only stuff left in the directory 1129 * is "." and "..". The "." reference is inconsequential 1130 * since we are quashing it. We have removed the "." 1131 * reference and the reference in the parent directory, 1132 * but there may be other hard links. The soft 1133 * dependency code will arrange to do these operations 1134 * after the parent directory entry has been deleted on 1135 * disk, so when running with that code we avoid doing 1136 * them now. 1137 */ 1138 if (!newparent) { 1139 dp->i_nlink--; 1140 dp->i_flag |= IN_CHANGE; 1141 } 1142 xp->i_nlink--; 1143 xp->i_flag |= IN_CHANGE; 1144 ioflag = DOINGASYNC(tvp) ? 0 : IO_SYNC; 1145 error = ffs_truncate(tvp, (off_t)0, ioflag, 1146 tcnp->cn_cred); 1147 if (error) 1148 goto bad; 1149 } 1150 VN_KNOTE(tdvp, NOTE_WRITE); 1151 vput(tdvp); 1152 VN_KNOTE(tvp, NOTE_DELETE); 1153 vput(tvp); 1154 xp = NULL; 1155 } 1156 1157 /* 1158 * tvp and tdvp have been cleaned up. only fvp and fdvp (both 1159 * unlocked) remain. We are about to overwrite fvp but we have to 1160 * keep 'ip' intact so we cannot release the old fvp, which is still 1161 * refd and accessible via ap->a_fvp. 1162 * 1163 * This means we cannot use either 'bad' or 'out' to cleanup any 1164 * more. 1165 */ 1166 1167 /* 1168 * 3) Unlink the source. 1169 */ 1170 fcnp->cn_flags &= ~CNP_MODMASK; 1171 fcnp->cn_flags |= CNP_LOCKPARENT; 1172 error = relookup(fdvp, &fvp, fcnp); 1173 if (error || fvp == NULL) { 1174 /* 1175 * From name has disappeared. IN_RENAME will not be set if 1176 * we get past the panic so we don't have to clean it up. 1177 */ 1178 if (doingdirectory) 1179 panic("ufs_rename: lost dir entry"); 1180 vrele(ap->a_fvp); 1181 if (fcnp->cn_flags & CNP_PDIRUNLOCK) 1182 vrele(fdvp); 1183 else 1184 vput(fdvp); 1185 return(0); 1186 } 1187 KKASSERT((fcnp->cn_flags & CNP_PDIRUNLOCK) == 0); 1188 1189 /* 1190 * fdvp and fvp are locked. 1191 */ 1192 xp = VTOI(fvp); 1193 dp = VTOI(fdvp); 1194 1195 /* 1196 * Ensure that the directory entry still exists and has not 1197 * changed while the new name has been entered. If the source is 1198 * a file then the entry may have been unlinked or renamed. In 1199 * either case there is no further work to be done. If the source 1200 * is a directory then it cannot have been rmdir'ed; the IN_RENAME 1201 * flag ensures that it cannot be moved by another rename or removed 1202 * by a rmdir. Cleanup IN_RENAME. 1203 */ 1204 if (xp != ip) { 1205 if (doingdirectory) 1206 panic("ufs_rename: lost dir entry"); 1207 } else { 1208 /* 1209 * If the source is a directory with a 1210 * new parent, the link count of the old 1211 * parent directory must be decremented 1212 * and ".." set to point to the new parent. 1213 */ 1214 if (doingdirectory && newparent) { 1215 xp->i_offset = mastertemplate.dot_reclen; 1216 ufs_dirrewrite(xp, dp, newparent, DT_DIR, 0); 1217 /* cache_purge removed - handled by VFS compat layer */ 1218 } 1219 error = ufs_dirremove(fdvp, xp, fcnp->cn_flags, 0); 1220 xp->i_flag &= ~IN_RENAME; 1221 } 1222 1223 VN_KNOTE(fvp, NOTE_RENAME); 1224 vput(fdvp); 1225 vput(fvp); 1226 vrele(ap->a_fvp); 1227 return (error); 1228 1229 bad: 1230 if (xp) 1231 vput(ITOV(xp)); 1232 vput(ITOV(dp)); 1233 out: 1234 if (doingdirectory) 1235 ip->i_flag &= ~IN_RENAME; 1236 if (vn_lock(fvp, LK_EXCLUSIVE | LK_FAILRECLAIM) == 0) { 1237 ip->i_effnlink--; 1238 ip->i_nlink--; 1239 ip->i_flag |= IN_CHANGE; 1240 ip->i_flag &= ~IN_RENAME; 1241 if (DOINGSOFTDEP(fvp)) 1242 softdep_change_linkcnt(ip); 1243 vput(fvp); 1244 } else { 1245 vrele(fvp); 1246 } 1247 return (error); 1248 } 1249 1250 /* 1251 * Mkdir system call 1252 * 1253 * ufs_mkdir(struct vnode *a_dvp, struct vnode **a_vpp, 1254 * struct componentname *a_cnp, struct vattr *a_vap) 1255 */ 1256 static 1257 int 1258 ufs_mkdir(struct vop_old_mkdir_args *ap) 1259 { 1260 struct vnode *dvp = ap->a_dvp; 1261 struct vattr *vap = ap->a_vap; 1262 struct componentname *cnp = ap->a_cnp; 1263 struct inode *ip, *dp; 1264 struct vnode *tvp; 1265 struct buf *bp; 1266 struct dirtemplate dirtemplate, *dtp; 1267 struct direct newdir; 1268 int error, dmode; 1269 long blkoff; 1270 1271 dp = VTOI(dvp); 1272 if ((nlink_t)dp->i_nlink >= LINK_MAX) { 1273 error = EMLINK; 1274 goto out; 1275 } 1276 dmode = vap->va_mode & 0777; 1277 dmode |= IFDIR; 1278 /* 1279 * Must simulate part of ufs_makeinode here to acquire the inode, 1280 * but not have it entered in the parent directory. The entry is 1281 * made later after writing "." and ".." entries. 1282 */ 1283 error = ffs_valloc(dvp, dmode, cnp->cn_cred, &tvp); 1284 if (error) 1285 goto out; 1286 ip = VTOI(tvp); 1287 ip->i_gid = dp->i_gid; 1288 #ifdef SUIDDIR 1289 { 1290 #ifdef QUOTA 1291 struct ucred ucred, *ucp; 1292 ucp = cnp->cn_cred; 1293 #endif 1294 /* 1295 * If we are hacking owners here, (only do this where told to) 1296 * and we are not giving it TO root, (would subvert quotas) 1297 * then go ahead and give it to the other user. 1298 * The new directory also inherits the SUID bit. 1299 * If user's UID and dir UID are the same, 1300 * 'give it away' so that the SUID is still forced on. 1301 */ 1302 if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) && 1303 (dp->i_mode & ISUID) && dp->i_uid) { 1304 dmode |= ISUID; 1305 ip->i_uid = dp->i_uid; 1306 #ifdef QUOTA 1307 if (dp->i_uid != cnp->cn_cred->cr_uid) { 1308 /* 1309 * Make sure the correct user gets charged 1310 * for the space. 1311 * Make a dummy credential for the victim. 1312 * XXX This seems to never be accessed out of 1313 * our context so a stack variable is ok. 1314 */ 1315 ucred.cr_ref = 1; 1316 ucred.cr_uid = ip->i_uid; 1317 ucred.cr_ngroups = 1; 1318 ucred.cr_groups[0] = dp->i_gid; 1319 ucp = &ucred; 1320 } 1321 #endif 1322 } else 1323 ip->i_uid = cnp->cn_cred->cr_uid; 1324 #ifdef QUOTA 1325 if ((error = ufs_getinoquota(ip)) || 1326 (error = ufs_chkiq(ip, 1, ucp, 0))) { 1327 ffs_vfree(tvp, ip->i_number, dmode); 1328 vput(tvp); 1329 return (error); 1330 } 1331 #endif 1332 } 1333 #else /* !SUIDDIR */ 1334 ip->i_uid = cnp->cn_cred->cr_uid; 1335 #ifdef QUOTA 1336 if ((error = ufs_getinoquota(ip)) || 1337 (error = ufs_chkiq(ip, 1, cnp->cn_cred, 0))) { 1338 ffs_vfree(tvp, ip->i_number, dmode); 1339 vput(tvp); 1340 return (error); 1341 } 1342 #endif 1343 #endif /* !SUIDDIR */ 1344 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 1345 ip->i_mode = dmode; 1346 tvp->v_type = VDIR; /* Rest init'd in getnewvnode(). */ 1347 ip->i_effnlink = 2; 1348 ip->i_nlink = 2; 1349 if (DOINGSOFTDEP(tvp)) 1350 softdep_change_linkcnt(ip); 1351 if (cnp->cn_flags & CNP_ISWHITEOUT) 1352 ip->i_flags |= UF_OPAQUE; 1353 1354 /* 1355 * Bump link count in parent directory to reflect work done below. 1356 * Should be done before reference is created so cleanup is 1357 * possible if we crash. 1358 */ 1359 dp->i_effnlink++; 1360 dp->i_nlink++; 1361 dp->i_flag |= IN_CHANGE; 1362 if (DOINGSOFTDEP(dvp)) 1363 softdep_change_linkcnt(dp); 1364 error = ffs_update(tvp, !(DOINGSOFTDEP(dvp) | DOINGASYNC(dvp))); 1365 if (error) 1366 goto bad; 1367 1368 /* 1369 * The vnode must have a VM object in order to issue buffer cache 1370 * ops on it. 1371 */ 1372 vinitvmio(tvp, DIRBLKSIZ, DIRBLKSIZ, -1); 1373 1374 /* 1375 * Initialize directory with "." and ".." from static template. 1376 */ 1377 if (dvp->v_mount->mnt_maxsymlinklen > 0) 1378 dtp = &mastertemplate; 1379 else 1380 dtp = (struct dirtemplate *)&omastertemplate; 1381 dirtemplate = *dtp; 1382 dirtemplate.dot_ino = ip->i_number; 1383 dirtemplate.dotdot_ino = dp->i_number; 1384 nvnode_pager_setsize(tvp, DIRBLKSIZ, DIRBLKSIZ, -1); 1385 error = VOP_BALLOC(tvp, 0LL, DIRBLKSIZ, cnp->cn_cred, B_CLRBUF, &bp); 1386 if (error) 1387 goto bad; 1388 ip->i_size = DIRBLKSIZ; 1389 ip->i_flag |= IN_CHANGE | IN_UPDATE; 1390 bcopy((caddr_t)&dirtemplate, (caddr_t)bp->b_data, sizeof dirtemplate); 1391 if (DOINGSOFTDEP(tvp)) { 1392 /* 1393 * Ensure that the entire newly allocated block is a 1394 * valid directory so that future growth within the 1395 * block does not have to ensure that the block is 1396 * written before the inode. 1397 */ 1398 blkoff = DIRBLKSIZ; 1399 while (blkoff < bp->b_bcount) { 1400 ((struct direct *) 1401 (bp->b_data + blkoff))->d_reclen = DIRBLKSIZ; 1402 blkoff += DIRBLKSIZ; 1403 } 1404 } 1405 if ((error = ffs_update(tvp, !(DOINGSOFTDEP(tvp) | 1406 DOINGASYNC(tvp)))) != 0) { 1407 bwrite(bp); 1408 goto bad; 1409 } 1410 /* 1411 * Directory set up, now install its entry in the parent directory. 1412 * 1413 * If we are not doing soft dependencies, then we must write out the 1414 * buffer containing the new directory body before entering the new 1415 * name in the parent. If we are doing soft dependencies, then the 1416 * buffer containing the new directory body will be passed to and 1417 * released in the soft dependency code after the code has attached 1418 * an appropriate ordering dependency to the buffer which ensures that 1419 * the buffer is written before the new name is written in the parent. 1420 */ 1421 if (DOINGASYNC(dvp)) 1422 bdwrite(bp); 1423 else if (!DOINGSOFTDEP(dvp) && (error = bwrite(bp)) != 0) 1424 goto bad; 1425 ufs_makedirentry(ip, cnp, &newdir); 1426 error = ufs_direnter(dvp, tvp, &newdir, cnp, bp); 1427 1428 bad: 1429 if (error == 0) { 1430 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK); 1431 *ap->a_vpp = tvp; 1432 } else { 1433 dp->i_effnlink--; 1434 dp->i_nlink--; 1435 dp->i_flag |= IN_CHANGE; 1436 if (DOINGSOFTDEP(dvp)) 1437 softdep_change_linkcnt(dp); 1438 /* 1439 * No need to do an explicit VOP_TRUNCATE here, vrele will 1440 * do this for us because we set the link count to 0. 1441 */ 1442 ip->i_effnlink = 0; 1443 ip->i_nlink = 0; 1444 ip->i_flag |= IN_CHANGE; 1445 if (DOINGSOFTDEP(tvp)) 1446 softdep_change_linkcnt(ip); 1447 vput(tvp); 1448 } 1449 out: 1450 return (error); 1451 } 1452 1453 /* 1454 * Rmdir system call. 1455 * 1456 * ufs_rmdir(struct vnode *a_dvp, struct vnode *a_vp, 1457 * struct componentname *a_cnp) 1458 */ 1459 static 1460 int 1461 ufs_rmdir(struct vop_old_rmdir_args *ap) 1462 { 1463 struct vnode *vp = ap->a_vp; 1464 struct vnode *dvp = ap->a_dvp; 1465 struct componentname *cnp = ap->a_cnp; 1466 struct inode *ip, *dp; 1467 int error, ioflag; 1468 1469 ip = VTOI(vp); 1470 dp = VTOI(dvp); 1471 1472 /* 1473 * Do not remove a directory that is in the process of being renamed. 1474 * Verify the directory is empty (and valid). Rmdir ".." will not be 1475 * valid since ".." will contain a reference to the current directory 1476 * and thus be non-empty. Do not allow the removal of mounted on 1477 * directories (this can happen when an NFS exported filesystem 1478 * tries to remove a locally mounted on directory). 1479 */ 1480 error = 0; 1481 if (ip->i_flag & IN_RENAME) { 1482 error = EINVAL; 1483 goto out; 1484 } 1485 if (ip->i_effnlink != 2 || 1486 !ufs_dirempty(ip, dp->i_number, cnp->cn_cred)) { 1487 error = ENOTEMPTY; 1488 goto out; 1489 } 1490 #if 0 /* handled by kernel now */ 1491 if ((dp->i_flags & APPEND) 1492 || (ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))) { 1493 error = EPERM; 1494 goto out; 1495 } 1496 #endif 1497 /* 1498 * Delete reference to directory before purging 1499 * inode. If we crash in between, the directory 1500 * will be reattached to lost+found, 1501 */ 1502 dp->i_effnlink--; 1503 ip->i_effnlink--; 1504 if (DOINGSOFTDEP(vp)) { 1505 softdep_change_linkcnt(dp); 1506 softdep_change_linkcnt(ip); 1507 } 1508 error = ufs_dirremove(dvp, ip, cnp->cn_flags, 1); 1509 if (error) { 1510 dp->i_effnlink++; 1511 ip->i_effnlink++; 1512 if (DOINGSOFTDEP(vp)) { 1513 softdep_change_linkcnt(dp); 1514 softdep_change_linkcnt(ip); 1515 } 1516 goto out; 1517 } 1518 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK); 1519 /* 1520 * Truncate inode. The only stuff left in the directory is "." and 1521 * "..". The "." reference is inconsequential since we are quashing 1522 * it. The soft dependency code will arrange to do these operations 1523 * after the parent directory entry has been deleted on disk, so 1524 * when running with that code we avoid doing them now. 1525 */ 1526 if (!DOINGSOFTDEP(vp)) { 1527 dp->i_nlink--; 1528 dp->i_flag |= IN_CHANGE; 1529 ip->i_nlink--; 1530 ip->i_flag |= IN_CHANGE; 1531 ioflag = DOINGASYNC(vp) ? 0 : IO_SYNC; 1532 error = ffs_truncate(vp, (off_t)0, ioflag, cnp->cn_cred); 1533 } 1534 /* cache_purge removed - handled by VFS compat layer */ 1535 #ifdef UFS_DIRHASH 1536 /* Kill any active hash; i_effnlink == 0, so it will not come back. */ 1537 if (ip->i_dirhash != NULL) 1538 ufsdirhash_free(ip); 1539 #endif 1540 out: 1541 VN_KNOTE(vp, NOTE_DELETE); 1542 return (error); 1543 } 1544 1545 /* 1546 * symlink -- make a symbolic link 1547 * 1548 * ufs_symlink(struct vnode *a_dvp, struct vnode **a_vpp, 1549 * struct componentname *a_cnp, struct vattr *a_vap, 1550 * char *a_target) 1551 */ 1552 static 1553 int 1554 ufs_symlink(struct vop_old_symlink_args *ap) 1555 { 1556 struct vnode *vp, **vpp = ap->a_vpp; 1557 struct inode *ip; 1558 int len, error; 1559 1560 error = ufs_makeinode(IFLNK | ap->a_vap->va_mode, ap->a_dvp, 1561 vpp, ap->a_cnp); 1562 if (error) 1563 return (error); 1564 VN_KNOTE(ap->a_dvp, NOTE_WRITE); 1565 vp = *vpp; 1566 len = strlen(ap->a_target); 1567 if (len < vp->v_mount->mnt_maxsymlinklen) { 1568 ip = VTOI(vp); 1569 bcopy(ap->a_target, (char *)ip->i_shortlink, len); 1570 ip->i_size = len; 1571 ip->i_flag |= IN_CHANGE | IN_UPDATE; 1572 } else { 1573 /* 1574 * Make sure we have a VM object in order to use 1575 * the buffer cache. 1576 */ 1577 if (vp->v_object == NULL) 1578 vinitvmio(vp, 0, PAGE_SIZE, -1); 1579 error = vn_rdwr(UIO_WRITE, vp, ap->a_target, len, (off_t)0, 1580 UIO_SYSSPACE, IO_NODELOCKED, 1581 ap->a_cnp->cn_cred, NULL); 1582 } 1583 if (error) 1584 vput(vp); 1585 return (error); 1586 } 1587 1588 /* 1589 * Vnode op for reading directories. 1590 * 1591 * ufs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred, 1592 * int *a_eofflag, int *ncookies, off_t **a_cookies) 1593 */ 1594 static 1595 int 1596 ufs_readdir(struct vop_readdir_args *ap) 1597 { 1598 struct uio *uio = ap->a_uio; 1599 struct vnode *vp = ap->a_vp; 1600 struct direct *dp; 1601 struct buf *bp; 1602 int retval; 1603 int error; 1604 int offset; /* offset into buffer cache buffer */ 1605 int eoffset; /* end of buffer clipped to file EOF */ 1606 int pickup; /* pickup point */ 1607 int ncookies; 1608 int cookie_index; 1609 off_t *cookies; 1610 1611 if (uio->uio_offset < 0) 1612 return (EINVAL); 1613 /* 1614 * Guess the number of cookies needed. Make sure we compute at 1615 * least 1, and no more then a reasonable limit. 1616 */ 1617 if (ap->a_ncookies) { 1618 ncookies = uio->uio_resid / 16 + 1; 1619 if (ncookies > 1024) 1620 ncookies = 1024; 1621 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK); 1622 } else { 1623 ncookies = -1; /* force conditionals below */ 1624 cookies = NULL; 1625 } 1626 cookie_index = 0; 1627 1628 error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY | LK_FAILRECLAIM); 1629 if (error) 1630 return (error); 1631 1632 /* 1633 * Past or at EOF 1634 */ 1635 if (uio->uio_offset >= VTOI(vp)->i_size) { 1636 if (ap->a_eofflag) 1637 *ap->a_eofflag = 1; 1638 if (ap->a_ncookies) { 1639 *ap->a_ncookies = cookie_index; 1640 *ap->a_cookies = cookies; 1641 } 1642 goto done; 1643 } 1644 1645 /* 1646 * Loop until we run out of cookies, we run out of user buffer, 1647 * or we hit the directory EOF. 1648 * 1649 * Always start scans at the beginning of the buffer, don't trust 1650 * the offset supplied by userland. 1651 */ 1652 while ((error = ffs_blkatoff_ra(vp, uio->uio_offset, NULL, &bp, 2)) == 0) { 1653 pickup = (int)(uio->uio_offset - bp->b_loffset); 1654 offset = 0; 1655 retval = 0; 1656 if (bp->b_loffset + bp->b_bcount > VTOI(vp)->i_size) 1657 eoffset = (int)(VTOI(vp)->i_size - bp->b_loffset); 1658 else 1659 eoffset = bp->b_bcount; 1660 1661 while (offset < eoffset) { 1662 dp = (struct direct *)(bp->b_data + offset); 1663 if (dp->d_reclen <= 0 || (dp->d_reclen & 3) || 1664 offset + dp->d_reclen > bp->b_bcount) { 1665 error = EIO; 1666 break; 1667 } 1668 if (offsetof(struct direct, d_name[dp->d_namlen]) > dp->d_reclen) { 1669 error = EIO; 1670 break; 1671 } 1672 if (offset < pickup) { 1673 offset += dp->d_reclen; 1674 continue; 1675 } 1676 #if BYTE_ORDER == LITTLE_ENDIAN 1677 if (OFSFMT(vp)) { 1678 retval = vop_write_dirent(&error, uio, 1679 dp->d_ino, dp->d_namlen, dp->d_type, 1680 dp->d_name); 1681 } else 1682 #endif 1683 { 1684 retval = vop_write_dirent(&error, uio, 1685 dp->d_ino, dp->d_type, dp->d_namlen, 1686 dp->d_name); 1687 } 1688 if (retval) 1689 break; 1690 if (cookies) 1691 cookies[cookie_index] = bp->b_loffset + offset; 1692 ++cookie_index; 1693 offset += dp->d_reclen; 1694 if (cookie_index == ncookies) 1695 break; 1696 } 1697 1698 /* 1699 * This will align the next loop to the beginning of the 1700 * next block, and pickup will calculate to 0. 1701 */ 1702 uio->uio_offset = bp->b_loffset + offset; 1703 brelse(bp); 1704 1705 if (retval || error || cookie_index == ncookies || 1706 uio->uio_offset >= VTOI(vp)->i_size) { 1707 break; 1708 } 1709 } 1710 if (ap->a_eofflag) 1711 *ap->a_eofflag = VTOI(vp)->i_size <= uio->uio_offset; 1712 1713 /* 1714 * Report errors only if we didn't manage to read anything 1715 */ 1716 if (error && cookie_index == 0) { 1717 if (cookies) { 1718 kfree(cookies, M_TEMP); 1719 *ap->a_ncookies = 0; 1720 *ap->a_cookies = NULL; 1721 } 1722 } else { 1723 error = 0; 1724 if (cookies) { 1725 *ap->a_ncookies = cookie_index; 1726 *ap->a_cookies = cookies; 1727 } 1728 } 1729 done: 1730 vn_unlock(vp); 1731 return (error); 1732 } 1733 1734 /* 1735 * Return target name of a symbolic link 1736 * 1737 * ufs_readlink(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred) 1738 */ 1739 static 1740 int 1741 ufs_readlink(struct vop_readlink_args *ap) 1742 { 1743 struct vnode *vp = ap->a_vp; 1744 struct inode *ip = VTOI(vp); 1745 int isize; 1746 1747 isize = ip->i_size; 1748 if ((isize < vp->v_mount->mnt_maxsymlinklen) || 1749 (ip->i_din.di_blocks == 0)) { /* XXX - for old fastlink support */ 1750 uiomove((char *)ip->i_shortlink, isize, ap->a_uio); 1751 return (0); 1752 } 1753 1754 /* 1755 * Perform the equivalent of an OPEN on vp so we can issue a 1756 * VOP_READ. 1757 */ 1758 return (VOP_READ(vp, ap->a_uio, 0, ap->a_cred)); 1759 } 1760 1761 /* 1762 * Calculate the logical to physical mapping if not done already, 1763 * then call the device strategy routine. 1764 * 1765 * In order to be able to swap to a file, the VOP_BMAP operation may not 1766 * deadlock on memory. See ufs_bmap() for details. 1767 * 1768 * ufs_strategy(struct vnode *a_vp, struct bio *a_bio) 1769 */ 1770 static 1771 int 1772 ufs_strategy(struct vop_strategy_args *ap) 1773 { 1774 struct bio *bio = ap->a_bio; 1775 struct bio *nbio; 1776 struct buf *bp = bio->bio_buf; 1777 struct vnode *vp = ap->a_vp; 1778 struct inode *ip; 1779 int error; 1780 1781 ip = VTOI(vp); 1782 if (vp->v_type == VBLK || vp->v_type == VCHR) 1783 panic("ufs_strategy: spec"); 1784 nbio = push_bio(bio); 1785 if (nbio->bio_offset == NOOFFSET) { 1786 error = VOP_BMAP(vp, bio->bio_offset, &nbio->bio_offset, 1787 NULL, NULL, bp->b_cmd); 1788 if (error) { 1789 bp->b_error = error; 1790 bp->b_flags |= B_ERROR; 1791 /* I/O was never started on nbio, must biodone(bio) */ 1792 biodone(bio); 1793 return (error); 1794 } 1795 if (nbio->bio_offset == NOOFFSET) 1796 vfs_bio_clrbuf(bp); 1797 } 1798 if (nbio->bio_offset == NOOFFSET) { 1799 /* 1800 * We hit a hole in the file. The buffer has been zero-filled 1801 * so just biodone() it. 1802 */ 1803 biodone(bio); 1804 } else { 1805 vn_strategy(ip->i_devvp, nbio); 1806 } 1807 return (0); 1808 } 1809 1810 /* 1811 * Print out the contents of an inode. 1812 * 1813 * ufs_print(struct vnode *a_vp) 1814 */ 1815 static 1816 int 1817 ufs_print(struct vop_print_args *ap) 1818 { 1819 struct vnode *vp = ap->a_vp; 1820 struct inode *ip = VTOI(vp); 1821 1822 kprintf("tag VT_UFS, ino %lu, on dev %s (%d, %d)", 1823 (u_long)ip->i_number, devtoname(ip->i_dev), major(ip->i_dev), 1824 minor(ip->i_dev)); 1825 if (vp->v_type == VFIFO) 1826 fifo_printinfo(vp); 1827 lockmgr_printinfo(&vp->v_lock); 1828 kprintf("\n"); 1829 return (0); 1830 } 1831 1832 /* 1833 * Read wrapper for fifos. 1834 * 1835 * ufsfifo_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 1836 * struct ucred *a_cred) 1837 */ 1838 static 1839 int 1840 ufsfifo_read(struct vop_read_args *ap) 1841 { 1842 int error, resid; 1843 struct inode *ip; 1844 struct uio *uio; 1845 1846 uio = ap->a_uio; 1847 resid = uio->uio_resid; 1848 error = VOCALL(&fifo_vnode_vops, &ap->a_head); 1849 ip = VTOI(ap->a_vp); 1850 if ((ap->a_vp->v_mount->mnt_flag & MNT_NOATIME) == 0 && ip != NULL && 1851 (uio->uio_resid != resid || (error == 0 && resid != 0))) 1852 VTOI(ap->a_vp)->i_flag |= IN_ACCESS; 1853 return (error); 1854 } 1855 1856 /* 1857 * Write wrapper for fifos. 1858 * 1859 * ufsfifo_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 1860 * struct ucred *a_cred) 1861 */ 1862 static 1863 int 1864 ufsfifo_write(struct vop_write_args *ap) 1865 { 1866 int error, resid; 1867 struct inode *ip; 1868 struct uio *uio; 1869 1870 uio = ap->a_uio; 1871 resid = uio->uio_resid; 1872 error = VOCALL(&fifo_vnode_vops, &ap->a_head); 1873 ip = VTOI(ap->a_vp); 1874 if (ip != NULL && (uio->uio_resid != resid || (error == 0 && resid != 0))) 1875 VTOI(ap->a_vp)->i_flag |= IN_CHANGE | IN_UPDATE; 1876 return (error); 1877 } 1878 1879 /* 1880 * Close wrapper for fifos. 1881 * 1882 * Update the times on the inode then do device close. 1883 * 1884 * ufsfifo_close(struct vnode *a_vp, int a_fflag) 1885 */ 1886 static 1887 int 1888 ufsfifo_close(struct vop_close_args *ap) 1889 { 1890 struct vnode *vp = ap->a_vp; 1891 1892 if (VREFCNT(vp) > 1) 1893 ufs_itimes(vp); 1894 return (VOCALL(&fifo_vnode_vops, &ap->a_head)); 1895 } 1896 1897 /* 1898 * Kqfilter wrapper for fifos. 1899 * 1900 * Fall through to ufs kqfilter routines if needed 1901 */ 1902 static 1903 int 1904 ufsfifo_kqfilter(struct vop_kqfilter_args *ap) 1905 { 1906 int error; 1907 1908 error = VOCALL(&fifo_vnode_vops, &ap->a_head); 1909 if (error) 1910 error = ufs_kqfilter(ap); 1911 return (error); 1912 } 1913 1914 /* 1915 * Advisory record locking support 1916 * 1917 * ufs_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, struct flock *a_fl, 1918 * int a_flags) 1919 */ 1920 static 1921 int 1922 ufs_advlock(struct vop_advlock_args *ap) 1923 { 1924 struct inode *ip = VTOI(ap->a_vp); 1925 1926 return (lf_advlock(ap, &(ip->i_lockf), ip->i_size)); 1927 } 1928 1929 /* 1930 * Initialize the vnode associated with a new inode, handle aliased 1931 * vnodes. 1932 * 1933 * Make sure directories have their VM object now rather then later, 1934 * saving us from having to check on all the myrid directory VOPs 1935 * that might be executed without a VOP_OPEN being performed. 1936 */ 1937 int 1938 ufs_vinit(struct mount *mntp, struct vnode **vpp) 1939 { 1940 struct inode *ip; 1941 struct vnode *vp; 1942 struct timeval tv; 1943 1944 vp = *vpp; 1945 ip = VTOI(vp); 1946 1947 vp->v_type = IFTOVT(ip->i_mode); 1948 1949 switch(vp->v_type) { 1950 case VCHR: 1951 case VBLK: 1952 vp->v_ops = &mntp->mnt_vn_spec_ops; 1953 addaliasu(vp, umajor(ip->i_rdev), uminor(ip->i_rdev)); 1954 break; 1955 case VFIFO: 1956 vp->v_ops = &mntp->mnt_vn_fifo_ops; 1957 break; 1958 case VDIR: 1959 case VREG: 1960 vinitvmio(vp, ip->i_size, 1961 blkoffsize(ip->i_fs, ip, ip->i_size), 1962 blkoff(ip->i_fs, ip->i_size)); 1963 break; 1964 case VLNK: 1965 if (ip->i_size >= vp->v_mount->mnt_maxsymlinklen) { 1966 vinitvmio(vp, ip->i_size, 1967 blkoffsize(ip->i_fs, ip, ip->i_size), 1968 blkoff(ip->i_fs, ip->i_size)); 1969 } 1970 break; 1971 default: 1972 break; 1973 1974 } 1975 1976 if (ip->i_number == UFS_ROOTINO) 1977 vsetflags(vp, VROOT); 1978 /* 1979 * Initialize modrev times 1980 */ 1981 getmicrouptime(&tv); 1982 SETHIGH(ip->i_modrev, tv.tv_sec); 1983 SETLOW(ip->i_modrev, tv.tv_usec * 4294); 1984 *vpp = vp; 1985 return (0); 1986 } 1987 1988 /* 1989 * Allocate a new inode. 1990 */ 1991 static 1992 int 1993 ufs_makeinode(int mode, struct vnode *dvp, struct vnode **vpp, 1994 struct componentname *cnp) 1995 { 1996 struct inode *ip, *pdir; 1997 struct direct newdir; 1998 struct vnode *tvp; 1999 int error; 2000 2001 pdir = VTOI(dvp); 2002 *vpp = NULL; 2003 if ((mode & IFMT) == 0) 2004 mode |= IFREG; 2005 2006 error = ffs_valloc(dvp, mode, cnp->cn_cred, &tvp); 2007 if (error) 2008 return (error); 2009 ip = VTOI(tvp); 2010 ip->i_flags = pdir->i_flags & (SF_NOHISTORY|UF_NOHISTORY|UF_NODUMP); 2011 ip->i_gid = pdir->i_gid; 2012 #ifdef SUIDDIR 2013 { 2014 #ifdef QUOTA 2015 struct ucred ucred, *ucp; 2016 ucp = cnp->cn_cred; 2017 #endif 2018 /* 2019 * If we are not the owner of the directory, 2020 * and we are hacking owners here, (only do this where told to) 2021 * and we are not giving it TO root, (would subvert quotas) 2022 * then go ahead and give it to the other user. 2023 * Note that this drops off the execute bits for security. 2024 */ 2025 if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) && 2026 (pdir->i_mode & ISUID) && 2027 (pdir->i_uid != cnp->cn_cred->cr_uid) && pdir->i_uid) { 2028 ip->i_uid = pdir->i_uid; 2029 mode &= ~07111; 2030 #ifdef QUOTA 2031 /* 2032 * Make sure the correct user gets charged 2033 * for the space. 2034 * Quickly knock up a dummy credential for the victim. 2035 * XXX This seems to never be accessed out of our 2036 * context so a stack variable is ok. 2037 */ 2038 ucred.cr_ref = 1; 2039 ucred.cr_uid = ip->i_uid; 2040 ucred.cr_ngroups = 1; 2041 ucred.cr_groups[0] = pdir->i_gid; 2042 ucp = &ucred; 2043 #endif 2044 } else 2045 ip->i_uid = cnp->cn_cred->cr_uid; 2046 2047 #ifdef QUOTA 2048 if ((error = ufs_getinoquota(ip)) || 2049 (error = ufs_chkiq(ip, 1, ucp, 0))) { 2050 ffs_vfree(tvp, ip->i_number, mode); 2051 vput(tvp); 2052 return (error); 2053 } 2054 #endif 2055 } 2056 #else /* !SUIDDIR */ 2057 ip->i_uid = cnp->cn_cred->cr_uid; 2058 #ifdef QUOTA 2059 if ((error = ufs_getinoquota(ip)) || 2060 (error = ufs_chkiq(ip, 1, cnp->cn_cred, 0))) { 2061 ffs_vfree(tvp, ip->i_number, mode); 2062 vput(tvp); 2063 return (error); 2064 } 2065 #endif 2066 #endif /* !SUIDDIR */ 2067 ip->i_din.di_spare7E = 0; 2068 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 2069 ip->i_mode = mode; 2070 tvp->v_type = IFTOVT(mode); /* Rest init'd in getnewvnode(). */ 2071 ip->i_effnlink = 1; 2072 ip->i_nlink = 1; 2073 if (DOINGSOFTDEP(tvp)) 2074 softdep_change_linkcnt(ip); 2075 if ((ip->i_mode & ISGID) && !groupmember(ip->i_gid, cnp->cn_cred) && 2076 priv_check_cred(cnp->cn_cred, PRIV_VFS_SETGID, 0)) { 2077 ip->i_mode &= ~ISGID; 2078 } 2079 2080 if (cnp->cn_flags & CNP_ISWHITEOUT) 2081 ip->i_flags |= UF_OPAQUE; 2082 2083 /* 2084 * Regular files and directories need VM objects. Softlinks do 2085 * not (not immediately anyway). 2086 */ 2087 if (tvp->v_type == VREG || tvp->v_type == VDIR) 2088 vinitvmio(tvp, 0, PAGE_SIZE, -1); 2089 2090 /* 2091 * Make sure inode goes to disk before directory entry. 2092 */ 2093 error = ffs_update(tvp, !(DOINGSOFTDEP(tvp) | DOINGASYNC(tvp))); 2094 if (error) 2095 goto bad; 2096 ufs_makedirentry(ip, cnp, &newdir); 2097 error = ufs_direnter(dvp, tvp, &newdir, cnp, NULL); 2098 if (error) 2099 goto bad; 2100 *vpp = tvp; 2101 return (0); 2102 2103 bad: 2104 /* 2105 * Write error occurred trying to update the inode 2106 * or the directory so must deallocate the inode. 2107 */ 2108 ip->i_effnlink = 0; 2109 ip->i_nlink = 0; 2110 ip->i_flag |= IN_CHANGE; 2111 if (DOINGSOFTDEP(tvp)) 2112 softdep_change_linkcnt(ip); 2113 vput(tvp); 2114 return (error); 2115 } 2116 2117 static int 2118 ufs_missingop(struct vop_generic_args *ap) 2119 { 2120 panic("no vop function for %s in ufs child", ap->a_desc->sd_name); 2121 return (EOPNOTSUPP); 2122 } 2123 2124 static struct filterops ufsread_filtops = 2125 { FILTEROP_ISFD, NULL, filt_ufsdetach, filt_ufsread }; 2126 static struct filterops ufswrite_filtops = 2127 { FILTEROP_ISFD, NULL, filt_ufsdetach, filt_ufswrite }; 2128 static struct filterops ufsvnode_filtops = 2129 { FILTEROP_ISFD, NULL, filt_ufsdetach, filt_ufsvnode }; 2130 2131 /* 2132 * ufs_kqfilter(struct vnode *a_vp, struct knote *a_kn) 2133 */ 2134 static int 2135 ufs_kqfilter(struct vop_kqfilter_args *ap) 2136 { 2137 struct vnode *vp = ap->a_vp; 2138 struct knote *kn = ap->a_kn; 2139 2140 switch (kn->kn_filter) { 2141 case EVFILT_READ: 2142 kn->kn_fop = &ufsread_filtops; 2143 break; 2144 case EVFILT_WRITE: 2145 kn->kn_fop = &ufswrite_filtops; 2146 break; 2147 case EVFILT_VNODE: 2148 kn->kn_fop = &ufsvnode_filtops; 2149 break; 2150 default: 2151 return (EOPNOTSUPP); 2152 } 2153 2154 kn->kn_hook = (caddr_t)vp; 2155 2156 /* XXX: kq token actually protects the list */ 2157 lwkt_gettoken(&vp->v_token); 2158 knote_insert(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn); 2159 lwkt_reltoken(&vp->v_token); 2160 2161 return (0); 2162 } 2163 2164 static void 2165 filt_ufsdetach(struct knote *kn) 2166 { 2167 struct vnode *vp = (struct vnode *)kn->kn_hook; 2168 2169 lwkt_gettoken(&vp->v_token); 2170 knote_remove(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn); 2171 lwkt_reltoken(&vp->v_token); 2172 } 2173 2174 /*ARGSUSED*/ 2175 static int 2176 filt_ufsread(struct knote *kn, long hint) 2177 { 2178 struct vnode *vp = (struct vnode *)kn->kn_hook; 2179 struct inode *ip = VTOI(vp); 2180 off_t off; 2181 2182 /* 2183 * filesystem is gone, so set the EOF flag and schedule 2184 * the knote for deletion. 2185 */ 2186 if (hint == NOTE_REVOKE) { 2187 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT); 2188 return (1); 2189 } 2190 2191 off = ip->i_size - kn->kn_fp->f_offset; 2192 kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX; 2193 if (kn->kn_sfflags & NOTE_OLDAPI) 2194 return(1); 2195 return (kn->kn_data != 0); 2196 } 2197 2198 /*ARGSUSED*/ 2199 static int 2200 filt_ufswrite(struct knote *kn, long hint) 2201 { 2202 /* 2203 * filesystem is gone, so set the EOF flag and schedule 2204 * the knote for deletion. 2205 */ 2206 if (hint == NOTE_REVOKE) 2207 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT); 2208 2209 kn->kn_data = 0; 2210 return (1); 2211 } 2212 2213 static int 2214 filt_ufsvnode(struct knote *kn, long hint) 2215 { 2216 if (kn->kn_sfflags & hint) 2217 kn->kn_fflags |= hint; 2218 if (hint == NOTE_REVOKE) { 2219 kn->kn_flags |= (EV_EOF | EV_NODATA); 2220 return (1); 2221 } 2222 return (kn->kn_fflags != 0); 2223 } 2224 2225 /* Global vfs data structures for ufs. */ 2226 static struct vop_ops ufs_vnode_vops = { 2227 .vop_default = vop_defaultop, 2228 .vop_fsync = (void *)ufs_missingop, 2229 .vop_read = (void *)ufs_missingop, 2230 .vop_reallocblks = (void *)ufs_missingop, 2231 .vop_write = (void *)ufs_missingop, 2232 .vop_access = ufs_access, 2233 .vop_advlock = ufs_advlock, 2234 .vop_bmap = ufs_bmap, 2235 .vop_old_lookup = ufs_lookup, 2236 .vop_close = ufs_close, 2237 .vop_old_create = ufs_create, 2238 .vop_getattr = ufs_getattr, 2239 .vop_inactive = ufs_inactive, 2240 .vop_old_link = ufs_link, 2241 .vop_old_mkdir = ufs_mkdir, 2242 .vop_old_mknod = ufs_mknod, 2243 .vop_open = vop_stdopen, 2244 .vop_pathconf = vop_stdpathconf, 2245 .vop_kqfilter = ufs_kqfilter, 2246 .vop_print = ufs_print, 2247 .vop_readdir = ufs_readdir, 2248 .vop_readlink = ufs_readlink, 2249 .vop_reclaim = ufs_reclaim, 2250 .vop_old_remove = ufs_remove, 2251 .vop_old_rename = ufs_rename, 2252 .vop_old_rmdir = ufs_rmdir, 2253 .vop_setattr = ufs_setattr, 2254 .vop_markatime = ufs_markatime, 2255 .vop_strategy = ufs_strategy, 2256 .vop_old_symlink = ufs_symlink, 2257 .vop_old_whiteout = ufs_whiteout 2258 }; 2259 2260 static struct vop_ops ufs_spec_vops = { 2261 .vop_default = vop_defaultop, 2262 .vop_fsync = (void *)ufs_missingop, 2263 .vop_access = ufs_access, 2264 .vop_close = ufs_close, 2265 .vop_getattr = ufs_getattr, 2266 .vop_inactive = ufs_inactive, 2267 .vop_print = ufs_print, 2268 .vop_read = vop_stdnoread, 2269 .vop_reclaim = ufs_reclaim, 2270 .vop_setattr = ufs_setattr, 2271 .vop_markatime = ufs_markatime, 2272 .vop_write = vop_stdnowrite 2273 }; 2274 2275 static struct vop_ops ufs_fifo_vops = { 2276 .vop_default = fifo_vnoperate, 2277 .vop_fsync = (void *)ufs_missingop, 2278 .vop_access = ufs_access, 2279 .vop_close = ufsfifo_close, 2280 .vop_getattr = ufs_getattr, 2281 .vop_inactive = ufs_inactive, 2282 .vop_kqfilter = ufsfifo_kqfilter, 2283 .vop_print = ufs_print, 2284 .vop_read = ufsfifo_read, 2285 .vop_reclaim = ufs_reclaim, 2286 .vop_setattr = ufs_setattr, 2287 .vop_markatime = ufs_markatime, 2288 .vop_write = ufsfifo_write 2289 }; 2290 2291 VNODEOP_SET(ufs_vnode_vops); 2292 VNODEOP_SET(ufs_spec_vops); 2293 VNODEOP_SET(ufs_fifo_vops); 2294 2295 /* 2296 * ufs_vnoperate() 2297 */ 2298 int 2299 ufs_vnoperate(struct vop_generic_args *ap) 2300 { 2301 return (VOCALL(&ufs_vnode_vops, ap)); 2302 } 2303 2304 /* 2305 * ufs_vnoperatefifo() 2306 */ 2307 int 2308 ufs_vnoperatefifo(struct vop_generic_args *ap) 2309 { 2310 return (VOCALL(&ufs_fifo_vops, ap)); 2311 } 2312 2313 /* 2314 * ufs_vnoperatespec() 2315 */ 2316 int 2317 ufs_vnoperatespec(struct vop_generic_args *ap) 2318 { 2319 return (VOCALL(&ufs_spec_vops, ap)); 2320 } 2321