ufs_vnops.c revision 306182
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 * 4. 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 */ 36 37#include <sys/cdefs.h> 38__FBSDID("$FreeBSD: stable/10/sys/ufs/ufs/ufs_vnops.c 306182 2016-09-22 10:54:30Z kib $"); 39 40#include "opt_quota.h" 41#include "opt_suiddir.h" 42#include "opt_ufs.h" 43#include "opt_ffs.h" 44 45#include <sys/param.h> 46#include <sys/systm.h> 47#include <sys/malloc.h> 48#include <sys/namei.h> 49#include <sys/kernel.h> 50#include <sys/fcntl.h> 51#include <sys/filio.h> 52#include <sys/stat.h> 53#include <sys/bio.h> 54#include <sys/buf.h> 55#include <sys/mount.h> 56#include <sys/priv.h> 57#include <sys/refcount.h> 58#include <sys/unistd.h> 59#include <sys/vnode.h> 60#include <sys/dirent.h> 61#include <sys/lockf.h> 62#include <sys/conf.h> 63#include <sys/acl.h> 64 65#include <security/mac/mac_framework.h> 66 67#include <sys/file.h> /* XXX */ 68 69#include <vm/vm.h> 70#include <vm/vm_extern.h> 71 72#include <ufs/ufs/acl.h> 73#include <ufs/ufs/extattr.h> 74#include <ufs/ufs/quota.h> 75#include <ufs/ufs/inode.h> 76#include <ufs/ufs/dir.h> 77#include <ufs/ufs/ufsmount.h> 78#include <ufs/ufs/ufs_extern.h> 79#ifdef UFS_DIRHASH 80#include <ufs/ufs/dirhash.h> 81#endif 82#ifdef UFS_GJOURNAL 83#include <ufs/ufs/gjournal.h> 84FEATURE(ufs_gjournal, "Journaling support through GEOM for UFS"); 85#endif 86 87#ifdef QUOTA 88FEATURE(ufs_quota, "UFS disk quotas support"); 89FEATURE(ufs_quota64, "64bit UFS disk quotas support"); 90#endif 91 92#ifdef SUIDDIR 93FEATURE(suiddir, 94 "Give all new files in directory the same ownership as the directory"); 95#endif 96 97 98#include <ufs/ffs/ffs_extern.h> 99 100static vop_accessx_t ufs_accessx; 101static int ufs_chmod(struct vnode *, int, struct ucred *, struct thread *); 102static int ufs_chown(struct vnode *, uid_t, gid_t, struct ucred *, struct thread *); 103static vop_close_t ufs_close; 104static vop_create_t ufs_create; 105static vop_getattr_t ufs_getattr; 106static vop_ioctl_t ufs_ioctl; 107static vop_link_t ufs_link; 108static int ufs_makeinode(int mode, struct vnode *, struct vnode **, struct componentname *); 109static vop_markatime_t ufs_markatime; 110static vop_mkdir_t ufs_mkdir; 111static vop_mknod_t ufs_mknod; 112static vop_open_t ufs_open; 113static vop_pathconf_t ufs_pathconf; 114static vop_print_t ufs_print; 115static vop_readlink_t ufs_readlink; 116static vop_remove_t ufs_remove; 117static vop_rename_t ufs_rename; 118static vop_rmdir_t ufs_rmdir; 119static vop_setattr_t ufs_setattr; 120static vop_strategy_t ufs_strategy; 121static vop_symlink_t ufs_symlink; 122static vop_whiteout_t ufs_whiteout; 123static vop_close_t ufsfifo_close; 124static vop_kqfilter_t ufsfifo_kqfilter; 125static vop_pathconf_t ufsfifo_pathconf; 126 127SYSCTL_NODE(_vfs, OID_AUTO, ufs, CTLFLAG_RD, 0, "UFS filesystem"); 128 129/* 130 * A virgin directory (no blushing please). 131 */ 132static struct dirtemplate mastertemplate = { 133 0, 12, DT_DIR, 1, ".", 134 0, DIRBLKSIZ - 12, DT_DIR, 2, ".." 135}; 136static struct odirtemplate omastertemplate = { 137 0, 12, 1, ".", 138 0, DIRBLKSIZ - 12, 2, ".." 139}; 140 141static void 142ufs_itimes_locked(struct vnode *vp) 143{ 144 struct inode *ip; 145 struct timespec ts; 146 147 ASSERT_VI_LOCKED(vp, __func__); 148 149 ip = VTOI(vp); 150 if (UFS_RDONLY(ip)) 151 goto out; 152 if ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE)) == 0) 153 return; 154 155 if ((vp->v_type == VBLK || vp->v_type == VCHR) && !DOINGSOFTDEP(vp)) 156 ip->i_flag |= IN_LAZYMOD; 157 else if (((vp->v_mount->mnt_kern_flag & 158 (MNTK_SUSPENDED | MNTK_SUSPEND)) == 0) || 159 (ip->i_flag & (IN_CHANGE | IN_UPDATE))) 160 ip->i_flag |= IN_MODIFIED; 161 else if (ip->i_flag & IN_ACCESS) 162 ip->i_flag |= IN_LAZYACCESS; 163 vfs_timestamp(&ts); 164 if (ip->i_flag & IN_ACCESS) { 165 DIP_SET(ip, i_atime, ts.tv_sec); 166 DIP_SET(ip, i_atimensec, ts.tv_nsec); 167 } 168 if (ip->i_flag & IN_UPDATE) { 169 DIP_SET(ip, i_mtime, ts.tv_sec); 170 DIP_SET(ip, i_mtimensec, ts.tv_nsec); 171 } 172 if (ip->i_flag & IN_CHANGE) { 173 DIP_SET(ip, i_ctime, ts.tv_sec); 174 DIP_SET(ip, i_ctimensec, ts.tv_nsec); 175 DIP_SET(ip, i_modrev, DIP(ip, i_modrev) + 1); 176 } 177 178 out: 179 ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_UPDATE); 180} 181 182void 183ufs_itimes(struct vnode *vp) 184{ 185 186 VI_LOCK(vp); 187 ufs_itimes_locked(vp); 188 VI_UNLOCK(vp); 189} 190 191/* 192 * Create a regular file 193 */ 194static int 195ufs_create(ap) 196 struct vop_create_args /* { 197 struct vnode *a_dvp; 198 struct vnode **a_vpp; 199 struct componentname *a_cnp; 200 struct vattr *a_vap; 201 } */ *ap; 202{ 203 int error; 204 205 error = 206 ufs_makeinode(MAKEIMODE(ap->a_vap->va_type, ap->a_vap->va_mode), 207 ap->a_dvp, ap->a_vpp, ap->a_cnp); 208 if (error != 0) 209 return (error); 210 if ((ap->a_cnp->cn_flags & MAKEENTRY) != 0) 211 cache_enter(ap->a_dvp, *ap->a_vpp, ap->a_cnp); 212 return (0); 213} 214 215/* 216 * Mknod vnode call 217 */ 218/* ARGSUSED */ 219static int 220ufs_mknod(ap) 221 struct vop_mknod_args /* { 222 struct vnode *a_dvp; 223 struct vnode **a_vpp; 224 struct componentname *a_cnp; 225 struct vattr *a_vap; 226 } */ *ap; 227{ 228 struct vattr *vap = ap->a_vap; 229 struct vnode **vpp = ap->a_vpp; 230 struct inode *ip; 231 ino_t ino; 232 int error; 233 234 error = ufs_makeinode(MAKEIMODE(vap->va_type, vap->va_mode), 235 ap->a_dvp, vpp, ap->a_cnp); 236 if (error) 237 return (error); 238 ip = VTOI(*vpp); 239 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 240 if (vap->va_rdev != VNOVAL) { 241 /* 242 * Want to be able to use this to make badblock 243 * inodes, so don't truncate the dev number. 244 */ 245 DIP_SET(ip, i_rdev, vap->va_rdev); 246 } 247 /* 248 * Remove inode, then reload it through VFS_VGET so it is 249 * checked to see if it is an alias of an existing entry in 250 * the inode cache. XXX I don't believe this is necessary now. 251 */ 252 (*vpp)->v_type = VNON; 253 ino = ip->i_number; /* Save this before vgone() invalidates ip. */ 254 vgone(*vpp); 255 vput(*vpp); 256 error = VFS_VGET(ap->a_dvp->v_mount, ino, LK_EXCLUSIVE, vpp); 257 if (error) { 258 *vpp = NULL; 259 return (error); 260 } 261 return (0); 262} 263 264/* 265 * Open called. 266 */ 267/* ARGSUSED */ 268static int 269ufs_open(struct vop_open_args *ap) 270{ 271 struct vnode *vp = ap->a_vp; 272 struct inode *ip; 273 274 if (vp->v_type == VCHR || vp->v_type == VBLK) 275 return (EOPNOTSUPP); 276 277 ip = VTOI(vp); 278 /* 279 * Files marked append-only must be opened for appending. 280 */ 281 if ((ip->i_flags & APPEND) && 282 (ap->a_mode & (FWRITE | O_APPEND)) == FWRITE) 283 return (EPERM); 284 vnode_create_vobject(vp, DIP(ip, i_size), ap->a_td); 285 return (0); 286} 287 288/* 289 * Close called. 290 * 291 * Update the times on the inode. 292 */ 293/* ARGSUSED */ 294static int 295ufs_close(ap) 296 struct vop_close_args /* { 297 struct vnode *a_vp; 298 int a_fflag; 299 struct ucred *a_cred; 300 struct thread *a_td; 301 } */ *ap; 302{ 303 struct vnode *vp = ap->a_vp; 304 int usecount; 305 306 VI_LOCK(vp); 307 usecount = vp->v_usecount; 308 if (usecount > 1) 309 ufs_itimes_locked(vp); 310 VI_UNLOCK(vp); 311 return (0); 312} 313 314static int 315ufs_accessx(ap) 316 struct vop_accessx_args /* { 317 struct vnode *a_vp; 318 accmode_t a_accmode; 319 struct ucred *a_cred; 320 struct thread *a_td; 321 } */ *ap; 322{ 323 struct vnode *vp = ap->a_vp; 324 struct inode *ip = VTOI(vp); 325 accmode_t accmode = ap->a_accmode; 326 int error; 327#ifdef QUOTA 328 int relocked; 329#endif 330#ifdef UFS_ACL 331 struct acl *acl; 332 acl_type_t type; 333#endif 334 335 /* 336 * Disallow write attempts on read-only filesystems; 337 * unless the file is a socket, fifo, or a block or 338 * character device resident on the filesystem. 339 */ 340 if (accmode & VMODIFY_PERMS) { 341 switch (vp->v_type) { 342 case VDIR: 343 case VLNK: 344 case VREG: 345 if (vp->v_mount->mnt_flag & MNT_RDONLY) 346 return (EROFS); 347#ifdef QUOTA 348 /* 349 * Inode is accounted in the quotas only if struct 350 * dquot is attached to it. VOP_ACCESS() is called 351 * from vn_open_cred() and provides a convenient 352 * point to call getinoquota(). 353 */ 354 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { 355 356 /* 357 * Upgrade vnode lock, since getinoquota() 358 * requires exclusive lock to modify inode. 359 */ 360 relocked = 1; 361 vhold(vp); 362 vn_lock(vp, LK_UPGRADE | LK_RETRY); 363 VI_LOCK(vp); 364 if (vp->v_iflag & VI_DOOMED) { 365 vdropl(vp); 366 error = ENOENT; 367 goto relock; 368 } 369 vdropl(vp); 370 } else 371 relocked = 0; 372 error = getinoquota(ip); 373relock: 374 if (relocked) 375 vn_lock(vp, LK_DOWNGRADE | LK_RETRY); 376 if (error != 0) 377 return (error); 378#endif 379 break; 380 default: 381 break; 382 } 383 } 384 385 /* 386 * If immutable bit set, nobody gets to write it. "& ~VADMIN_PERMS" 387 * is here, because without it, * it would be impossible for the owner 388 * to remove the IMMUTABLE flag. 389 */ 390 if ((accmode & (VMODIFY_PERMS & ~VADMIN_PERMS)) && 391 (ip->i_flags & (IMMUTABLE | SF_SNAPSHOT))) 392 return (EPERM); 393 394#ifdef UFS_ACL 395 if ((vp->v_mount->mnt_flag & (MNT_ACLS | MNT_NFS4ACLS)) != 0) { 396 if (vp->v_mount->mnt_flag & MNT_NFS4ACLS) 397 type = ACL_TYPE_NFS4; 398 else 399 type = ACL_TYPE_ACCESS; 400 401 acl = acl_alloc(M_WAITOK); 402 if (type == ACL_TYPE_NFS4) 403 error = ufs_getacl_nfs4_internal(vp, acl, ap->a_td); 404 else 405 error = VOP_GETACL(vp, type, acl, ap->a_cred, ap->a_td); 406 switch (error) { 407 case 0: 408 if (type == ACL_TYPE_NFS4) { 409 error = vaccess_acl_nfs4(vp->v_type, ip->i_uid, 410 ip->i_gid, acl, accmode, ap->a_cred, NULL); 411 } else { 412 error = vfs_unixify_accmode(&accmode); 413 if (error == 0) 414 error = vaccess_acl_posix1e(vp->v_type, ip->i_uid, 415 ip->i_gid, acl, accmode, ap->a_cred, NULL); 416 } 417 break; 418 default: 419 if (error != EOPNOTSUPP) 420 printf( 421"ufs_accessx(): Error retrieving ACL on object (%d).\n", 422 error); 423 /* 424 * XXX: Fall back until debugged. Should 425 * eventually possibly log an error, and return 426 * EPERM for safety. 427 */ 428 error = vfs_unixify_accmode(&accmode); 429 if (error == 0) 430 error = vaccess(vp->v_type, ip->i_mode, ip->i_uid, 431 ip->i_gid, accmode, ap->a_cred, NULL); 432 } 433 acl_free(acl); 434 435 return (error); 436 } 437#endif /* !UFS_ACL */ 438 error = vfs_unixify_accmode(&accmode); 439 if (error == 0) 440 error = vaccess(vp->v_type, ip->i_mode, ip->i_uid, ip->i_gid, 441 accmode, ap->a_cred, NULL); 442 return (error); 443} 444 445/* ARGSUSED */ 446static int 447ufs_getattr(ap) 448 struct vop_getattr_args /* { 449 struct vnode *a_vp; 450 struct vattr *a_vap; 451 struct ucred *a_cred; 452 } */ *ap; 453{ 454 struct vnode *vp = ap->a_vp; 455 struct inode *ip = VTOI(vp); 456 struct vattr *vap = ap->a_vap; 457 458 VI_LOCK(vp); 459 ufs_itimes_locked(vp); 460 if (ip->i_ump->um_fstype == UFS1) { 461 vap->va_atime.tv_sec = ip->i_din1->di_atime; 462 vap->va_atime.tv_nsec = ip->i_din1->di_atimensec; 463 } else { 464 vap->va_atime.tv_sec = ip->i_din2->di_atime; 465 vap->va_atime.tv_nsec = ip->i_din2->di_atimensec; 466 } 467 VI_UNLOCK(vp); 468 /* 469 * Copy from inode table 470 */ 471 vap->va_fsid = dev2udev(ip->i_dev); 472 vap->va_fileid = ip->i_number; 473 vap->va_mode = ip->i_mode & ~IFMT; 474 vap->va_nlink = ip->i_effnlink; 475 vap->va_uid = ip->i_uid; 476 vap->va_gid = ip->i_gid; 477 if (ip->i_ump->um_fstype == UFS1) { 478 vap->va_rdev = ip->i_din1->di_rdev; 479 vap->va_size = ip->i_din1->di_size; 480 vap->va_mtime.tv_sec = ip->i_din1->di_mtime; 481 vap->va_mtime.tv_nsec = ip->i_din1->di_mtimensec; 482 vap->va_ctime.tv_sec = ip->i_din1->di_ctime; 483 vap->va_ctime.tv_nsec = ip->i_din1->di_ctimensec; 484 vap->va_bytes = dbtob((u_quad_t)ip->i_din1->di_blocks); 485 vap->va_filerev = ip->i_din1->di_modrev; 486 } else { 487 vap->va_rdev = ip->i_din2->di_rdev; 488 vap->va_size = ip->i_din2->di_size; 489 vap->va_mtime.tv_sec = ip->i_din2->di_mtime; 490 vap->va_mtime.tv_nsec = ip->i_din2->di_mtimensec; 491 vap->va_ctime.tv_sec = ip->i_din2->di_ctime; 492 vap->va_ctime.tv_nsec = ip->i_din2->di_ctimensec; 493 vap->va_birthtime.tv_sec = ip->i_din2->di_birthtime; 494 vap->va_birthtime.tv_nsec = ip->i_din2->di_birthnsec; 495 vap->va_bytes = dbtob((u_quad_t)ip->i_din2->di_blocks); 496 vap->va_filerev = ip->i_din2->di_modrev; 497 } 498 vap->va_flags = ip->i_flags; 499 vap->va_gen = ip->i_gen; 500 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize; 501 vap->va_type = IFTOVT(ip->i_mode); 502 return (0); 503} 504 505/* 506 * Set attribute vnode op. called from several syscalls 507 */ 508static int 509ufs_setattr(ap) 510 struct vop_setattr_args /* { 511 struct vnode *a_vp; 512 struct vattr *a_vap; 513 struct ucred *a_cred; 514 } */ *ap; 515{ 516 struct vattr *vap = ap->a_vap; 517 struct vnode *vp = ap->a_vp; 518 struct inode *ip = VTOI(vp); 519 struct ucred *cred = ap->a_cred; 520 struct thread *td = curthread; 521 int error; 522 523 /* 524 * Check for unsettable attributes. 525 */ 526 if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) || 527 (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) || 528 (vap->va_blocksize != VNOVAL) || (vap->va_rdev != VNOVAL) || 529 ((int)vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) { 530 return (EINVAL); 531 } 532 if (vap->va_flags != VNOVAL) { 533 if ((vap->va_flags & ~(SF_APPEND | SF_ARCHIVED | SF_IMMUTABLE | 534 SF_NOUNLINK | SF_SNAPSHOT | UF_APPEND | UF_ARCHIVE | 535 UF_HIDDEN | UF_IMMUTABLE | UF_NODUMP | UF_NOUNLINK | 536 UF_OFFLINE | UF_OPAQUE | UF_READONLY | UF_REPARSE | 537 UF_SPARSE | UF_SYSTEM)) != 0) 538 return (EOPNOTSUPP); 539 if (vp->v_mount->mnt_flag & MNT_RDONLY) 540 return (EROFS); 541 /* 542 * Callers may only modify the file flags on objects they 543 * have VADMIN rights for. 544 */ 545 if ((error = VOP_ACCESS(vp, VADMIN, cred, td))) 546 return (error); 547 /* 548 * Unprivileged processes are not permitted to unset system 549 * flags, or modify flags if any system flags are set. 550 * Privileged non-jail processes may not modify system flags 551 * if securelevel > 0 and any existing system flags are set. 552 * Privileged jail processes behave like privileged non-jail 553 * processes if the security.jail.chflags_allowed sysctl is 554 * is non-zero; otherwise, they behave like unprivileged 555 * processes. 556 */ 557 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) { 558 if (ip->i_flags & 559 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) { 560 error = securelevel_gt(cred, 0); 561 if (error) 562 return (error); 563 } 564 /* The snapshot flag cannot be toggled. */ 565 if ((vap->va_flags ^ ip->i_flags) & SF_SNAPSHOT) 566 return (EPERM); 567 } else { 568 if (ip->i_flags & 569 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) || 570 ((vap->va_flags ^ ip->i_flags) & SF_SETTABLE)) 571 return (EPERM); 572 } 573 ip->i_flags = vap->va_flags; 574 DIP_SET(ip, i_flags, vap->va_flags); 575 ip->i_flag |= IN_CHANGE; 576 error = UFS_UPDATE(vp, 0); 577 if (ip->i_flags & (IMMUTABLE | APPEND)) 578 return (error); 579 } 580 /* 581 * If immutable or append, no one can change any of its attributes 582 * except the ones already handled (in some cases, file flags 583 * including the immutability flags themselves for the superuser). 584 */ 585 if (ip->i_flags & (IMMUTABLE | APPEND)) 586 return (EPERM); 587 /* 588 * Go through the fields and update iff not VNOVAL. 589 */ 590 if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) { 591 if (vp->v_mount->mnt_flag & MNT_RDONLY) 592 return (EROFS); 593 if ((error = ufs_chown(vp, vap->va_uid, vap->va_gid, cred, 594 td)) != 0) 595 return (error); 596 } 597 if (vap->va_size != VNOVAL) { 598 /* 599 * XXX most of the following special cases should be in 600 * callers instead of in N filesystems. The VDIR check 601 * mostly already is. 602 */ 603 switch (vp->v_type) { 604 case VDIR: 605 return (EISDIR); 606 case VLNK: 607 case VREG: 608 /* 609 * Truncation should have an effect in these cases. 610 * Disallow it if the filesystem is read-only or 611 * the file is being snapshotted. 612 */ 613 if (vp->v_mount->mnt_flag & MNT_RDONLY) 614 return (EROFS); 615 if ((ip->i_flags & SF_SNAPSHOT) != 0) 616 return (EPERM); 617 break; 618 default: 619 /* 620 * According to POSIX, the result is unspecified 621 * for file types other than regular files, 622 * directories and shared memory objects. We 623 * don't support shared memory objects in the file 624 * system, and have dubious support for truncating 625 * symlinks. Just ignore the request in other cases. 626 */ 627 return (0); 628 } 629 if ((error = UFS_TRUNCATE(vp, vap->va_size, IO_NORMAL | 630 ((vap->va_vaflags & VA_SYNC) != 0 ? IO_SYNC : 0), 631 cred)) != 0) 632 return (error); 633 } 634 if (vap->va_atime.tv_sec != VNOVAL || 635 vap->va_mtime.tv_sec != VNOVAL || 636 vap->va_birthtime.tv_sec != VNOVAL) { 637 if (vp->v_mount->mnt_flag & MNT_RDONLY) 638 return (EROFS); 639 if ((ip->i_flags & SF_SNAPSHOT) != 0) 640 return (EPERM); 641 error = vn_utimes_perm(vp, vap, cred, td); 642 if (error != 0) 643 return (error); 644 ip->i_flag |= IN_CHANGE | IN_MODIFIED; 645 if (vap->va_atime.tv_sec != VNOVAL) { 646 ip->i_flag &= ~IN_ACCESS; 647 DIP_SET(ip, i_atime, vap->va_atime.tv_sec); 648 DIP_SET(ip, i_atimensec, vap->va_atime.tv_nsec); 649 } 650 if (vap->va_mtime.tv_sec != VNOVAL) { 651 ip->i_flag &= ~IN_UPDATE; 652 DIP_SET(ip, i_mtime, vap->va_mtime.tv_sec); 653 DIP_SET(ip, i_mtimensec, vap->va_mtime.tv_nsec); 654 } 655 if (vap->va_birthtime.tv_sec != VNOVAL && 656 ip->i_ump->um_fstype == UFS2) { 657 ip->i_din2->di_birthtime = vap->va_birthtime.tv_sec; 658 ip->i_din2->di_birthnsec = vap->va_birthtime.tv_nsec; 659 } 660 error = UFS_UPDATE(vp, 0); 661 if (error) 662 return (error); 663 } 664 error = 0; 665 if (vap->va_mode != (mode_t)VNOVAL) { 666 if (vp->v_mount->mnt_flag & MNT_RDONLY) 667 return (EROFS); 668 if ((ip->i_flags & SF_SNAPSHOT) != 0 && (vap->va_mode & 669 (S_IXUSR | S_IWUSR | S_IXGRP | S_IWGRP | S_IXOTH | S_IWOTH))) 670 return (EPERM); 671 error = ufs_chmod(vp, (int)vap->va_mode, cred, td); 672 } 673 return (error); 674} 675 676#ifdef UFS_ACL 677static int 678ufs_update_nfs4_acl_after_mode_change(struct vnode *vp, int mode, 679 int file_owner_id, struct ucred *cred, struct thread *td) 680{ 681 int error; 682 struct acl *aclp; 683 684 aclp = acl_alloc(M_WAITOK); 685 error = ufs_getacl_nfs4_internal(vp, aclp, td); 686 /* 687 * We don't have to handle EOPNOTSUPP here, as the filesystem claims 688 * it supports ACLs. 689 */ 690 if (error) 691 goto out; 692 693 acl_nfs4_sync_acl_from_mode(aclp, mode, file_owner_id); 694 error = ufs_setacl_nfs4_internal(vp, aclp, td); 695 696out: 697 acl_free(aclp); 698 return (error); 699} 700#endif /* UFS_ACL */ 701 702/* 703 * Mark this file's access time for update for vfs_mark_atime(). This 704 * is called from execve() and mmap(). 705 */ 706static int 707ufs_markatime(ap) 708 struct vop_markatime_args /* { 709 struct vnode *a_vp; 710 } */ *ap; 711{ 712 struct vnode *vp = ap->a_vp; 713 struct inode *ip = VTOI(vp); 714 715 VI_LOCK(vp); 716 ip->i_flag |= IN_ACCESS; 717 VI_UNLOCK(vp); 718 /* 719 * XXXKIB No UFS_UPDATE(ap->a_vp, 0) there. 720 */ 721 return (0); 722} 723 724/* 725 * Change the mode on a file. 726 * Inode must be locked before calling. 727 */ 728static int 729ufs_chmod(vp, mode, cred, td) 730 struct vnode *vp; 731 int mode; 732 struct ucred *cred; 733 struct thread *td; 734{ 735 struct inode *ip = VTOI(vp); 736 int error; 737 738 /* 739 * To modify the permissions on a file, must possess VADMIN 740 * for that file. 741 */ 742 if ((error = VOP_ACCESSX(vp, VWRITE_ACL, cred, td))) 743 return (error); 744 /* 745 * Privileged processes may set the sticky bit on non-directories, 746 * as well as set the setgid bit on a file with a group that the 747 * process is not a member of. Both of these are allowed in 748 * jail(8). 749 */ 750 if (vp->v_type != VDIR && (mode & S_ISTXT)) { 751 if (priv_check_cred(cred, PRIV_VFS_STICKYFILE, 0)) 752 return (EFTYPE); 753 } 754 if (!groupmember(ip->i_gid, cred) && (mode & ISGID)) { 755 error = priv_check_cred(cred, PRIV_VFS_SETGID, 0); 756 if (error) 757 return (error); 758 } 759 760 /* 761 * Deny setting setuid if we are not the file owner. 762 */ 763 if ((mode & ISUID) && ip->i_uid != cred->cr_uid) { 764 error = priv_check_cred(cred, PRIV_VFS_ADMIN, 0); 765 if (error) 766 return (error); 767 } 768 769 ip->i_mode &= ~ALLPERMS; 770 ip->i_mode |= (mode & ALLPERMS); 771 DIP_SET(ip, i_mode, ip->i_mode); 772 ip->i_flag |= IN_CHANGE; 773#ifdef UFS_ACL 774 if ((vp->v_mount->mnt_flag & MNT_NFS4ACLS) != 0) 775 error = ufs_update_nfs4_acl_after_mode_change(vp, mode, ip->i_uid, cred, td); 776#endif 777 if (error == 0 && (ip->i_flag & IN_CHANGE) != 0) 778 error = UFS_UPDATE(vp, 0); 779 780 return (error); 781} 782 783/* 784 * Perform chown operation on inode ip; 785 * inode must be locked prior to call. 786 */ 787static int 788ufs_chown(vp, uid, gid, cred, td) 789 struct vnode *vp; 790 uid_t uid; 791 gid_t gid; 792 struct ucred *cred; 793 struct thread *td; 794{ 795 struct inode *ip = VTOI(vp); 796 uid_t ouid; 797 gid_t ogid; 798 int error = 0; 799#ifdef QUOTA 800 int i; 801 ufs2_daddr_t change; 802#endif 803 804 if (uid == (uid_t)VNOVAL) 805 uid = ip->i_uid; 806 if (gid == (gid_t)VNOVAL) 807 gid = ip->i_gid; 808 /* 809 * To modify the ownership of a file, must possess VADMIN for that 810 * file. 811 */ 812 if ((error = VOP_ACCESSX(vp, VWRITE_OWNER, cred, td))) 813 return (error); 814 /* 815 * To change the owner of a file, or change the group of a file to a 816 * group of which we are not a member, the caller must have 817 * privilege. 818 */ 819 if (((uid != ip->i_uid && uid != cred->cr_uid) || 820 (gid != ip->i_gid && !groupmember(gid, cred))) && 821 (error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0))) 822 return (error); 823 ogid = ip->i_gid; 824 ouid = ip->i_uid; 825#ifdef QUOTA 826 if ((error = getinoquota(ip)) != 0) 827 return (error); 828 if (ouid == uid) { 829 dqrele(vp, ip->i_dquot[USRQUOTA]); 830 ip->i_dquot[USRQUOTA] = NODQUOT; 831 } 832 if (ogid == gid) { 833 dqrele(vp, ip->i_dquot[GRPQUOTA]); 834 ip->i_dquot[GRPQUOTA] = NODQUOT; 835 } 836 change = DIP(ip, i_blocks); 837 (void) chkdq(ip, -change, cred, CHOWN); 838 (void) chkiq(ip, -1, cred, CHOWN); 839 for (i = 0; i < MAXQUOTAS; i++) { 840 dqrele(vp, ip->i_dquot[i]); 841 ip->i_dquot[i] = NODQUOT; 842 } 843#endif 844 ip->i_gid = gid; 845 DIP_SET(ip, i_gid, gid); 846 ip->i_uid = uid; 847 DIP_SET(ip, i_uid, uid); 848#ifdef QUOTA 849 if ((error = getinoquota(ip)) == 0) { 850 if (ouid == uid) { 851 dqrele(vp, ip->i_dquot[USRQUOTA]); 852 ip->i_dquot[USRQUOTA] = NODQUOT; 853 } 854 if (ogid == gid) { 855 dqrele(vp, ip->i_dquot[GRPQUOTA]); 856 ip->i_dquot[GRPQUOTA] = NODQUOT; 857 } 858 if ((error = chkdq(ip, change, cred, CHOWN)) == 0) { 859 if ((error = chkiq(ip, 1, cred, CHOWN)) == 0) 860 goto good; 861 else 862 (void) chkdq(ip, -change, cred, CHOWN|FORCE); 863 } 864 for (i = 0; i < MAXQUOTAS; i++) { 865 dqrele(vp, ip->i_dquot[i]); 866 ip->i_dquot[i] = NODQUOT; 867 } 868 } 869 ip->i_gid = ogid; 870 DIP_SET(ip, i_gid, ogid); 871 ip->i_uid = ouid; 872 DIP_SET(ip, i_uid, ouid); 873 if (getinoquota(ip) == 0) { 874 if (ouid == uid) { 875 dqrele(vp, ip->i_dquot[USRQUOTA]); 876 ip->i_dquot[USRQUOTA] = NODQUOT; 877 } 878 if (ogid == gid) { 879 dqrele(vp, ip->i_dquot[GRPQUOTA]); 880 ip->i_dquot[GRPQUOTA] = NODQUOT; 881 } 882 (void) chkdq(ip, change, cred, FORCE|CHOWN); 883 (void) chkiq(ip, 1, cred, FORCE|CHOWN); 884 (void) getinoquota(ip); 885 } 886 return (error); 887good: 888 if (getinoquota(ip)) 889 panic("ufs_chown: lost quota"); 890#endif /* QUOTA */ 891 ip->i_flag |= IN_CHANGE; 892 if ((ip->i_mode & (ISUID | ISGID)) && (ouid != uid || ogid != gid)) { 893 if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID, 0)) { 894 ip->i_mode &= ~(ISUID | ISGID); 895 DIP_SET(ip, i_mode, ip->i_mode); 896 } 897 } 898 error = UFS_UPDATE(vp, 0); 899 return (error); 900} 901 902static int 903ufs_remove(ap) 904 struct vop_remove_args /* { 905 struct vnode *a_dvp; 906 struct vnode *a_vp; 907 struct componentname *a_cnp; 908 } */ *ap; 909{ 910 struct inode *ip; 911 struct vnode *vp = ap->a_vp; 912 struct vnode *dvp = ap->a_dvp; 913 int error; 914 struct thread *td; 915 916 td = curthread; 917 ip = VTOI(vp); 918 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) || 919 (VTOI(dvp)->i_flags & APPEND)) { 920 error = EPERM; 921 goto out; 922 } 923#ifdef UFS_GJOURNAL 924 ufs_gjournal_orphan(vp); 925#endif 926 error = ufs_dirremove(dvp, ip, ap->a_cnp->cn_flags, 0); 927 if (ip->i_nlink <= 0) 928 vp->v_vflag |= VV_NOSYNC; 929 if ((ip->i_flags & SF_SNAPSHOT) != 0) { 930 /* 931 * Avoid deadlock where another thread is trying to 932 * update the inodeblock for dvp and is waiting on 933 * snaplk. Temporary unlock the vnode lock for the 934 * unlinked file and sync the directory. This should 935 * allow vput() of the directory to not block later on 936 * while holding the snapshot vnode locked, assuming 937 * that the directory hasn't been unlinked too. 938 */ 939 VOP_UNLOCK(vp, 0); 940 (void) VOP_FSYNC(dvp, MNT_WAIT, td); 941 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 942 } 943out: 944 return (error); 945} 946 947/* 948 * link vnode call 949 */ 950static int 951ufs_link(ap) 952 struct vop_link_args /* { 953 struct vnode *a_tdvp; 954 struct vnode *a_vp; 955 struct componentname *a_cnp; 956 } */ *ap; 957{ 958 struct vnode *vp = ap->a_vp; 959 struct vnode *tdvp = ap->a_tdvp; 960 struct componentname *cnp = ap->a_cnp; 961 struct inode *ip; 962 struct direct newdir; 963 int error; 964 965#ifdef INVARIANTS 966 if ((cnp->cn_flags & HASBUF) == 0) 967 panic("ufs_link: no name"); 968#endif 969 if (VTOI(tdvp)->i_effnlink < 2) 970 panic("ufs_link: Bad link count %d on parent", 971 VTOI(tdvp)->i_effnlink); 972 ip = VTOI(vp); 973 if ((nlink_t)ip->i_nlink >= LINK_MAX) { 974 error = EMLINK; 975 goto out; 976 } 977 /* 978 * The file may have been removed after namei droped the original 979 * lock. 980 */ 981 if (ip->i_effnlink == 0) { 982 error = ENOENT; 983 goto out; 984 } 985 if (ip->i_flags & (IMMUTABLE | APPEND)) { 986 error = EPERM; 987 goto out; 988 } 989 ip->i_effnlink++; 990 ip->i_nlink++; 991 DIP_SET(ip, i_nlink, ip->i_nlink); 992 ip->i_flag |= IN_CHANGE; 993 if (DOINGSOFTDEP(vp)) 994 softdep_setup_link(VTOI(tdvp), ip); 995 error = UFS_UPDATE(vp, !(DOINGSOFTDEP(vp) | DOINGASYNC(vp))); 996 if (!error) { 997 ufs_makedirentry(ip, cnp, &newdir); 998 error = ufs_direnter(tdvp, vp, &newdir, cnp, NULL, 0); 999 } 1000 1001 if (error) { 1002 ip->i_effnlink--; 1003 ip->i_nlink--; 1004 DIP_SET(ip, i_nlink, ip->i_nlink); 1005 ip->i_flag |= IN_CHANGE; 1006 if (DOINGSOFTDEP(vp)) 1007 softdep_revert_link(VTOI(tdvp), ip); 1008 } 1009out: 1010 return (error); 1011} 1012 1013/* 1014 * whiteout vnode call 1015 */ 1016static int 1017ufs_whiteout(ap) 1018 struct vop_whiteout_args /* { 1019 struct vnode *a_dvp; 1020 struct componentname *a_cnp; 1021 int a_flags; 1022 } */ *ap; 1023{ 1024 struct vnode *dvp = ap->a_dvp; 1025 struct componentname *cnp = ap->a_cnp; 1026 struct direct newdir; 1027 int error = 0; 1028 1029 switch (ap->a_flags) { 1030 case LOOKUP: 1031 /* 4.4 format directories support whiteout operations */ 1032 if (dvp->v_mount->mnt_maxsymlinklen > 0) 1033 return (0); 1034 return (EOPNOTSUPP); 1035 1036 case CREATE: 1037 /* create a new directory whiteout */ 1038#ifdef INVARIANTS 1039 if ((cnp->cn_flags & SAVENAME) == 0) 1040 panic("ufs_whiteout: missing name"); 1041 if (dvp->v_mount->mnt_maxsymlinklen <= 0) 1042 panic("ufs_whiteout: old format filesystem"); 1043#endif 1044 1045 newdir.d_ino = WINO; 1046 newdir.d_namlen = cnp->cn_namelen; 1047 bcopy(cnp->cn_nameptr, newdir.d_name, (unsigned)cnp->cn_namelen + 1); 1048 newdir.d_type = DT_WHT; 1049 error = ufs_direnter(dvp, NULL, &newdir, cnp, NULL, 0); 1050 break; 1051 1052 case DELETE: 1053 /* remove an existing directory whiteout */ 1054#ifdef INVARIANTS 1055 if (dvp->v_mount->mnt_maxsymlinklen <= 0) 1056 panic("ufs_whiteout: old format filesystem"); 1057#endif 1058 1059 cnp->cn_flags &= ~DOWHITEOUT; 1060 error = ufs_dirremove(dvp, NULL, cnp->cn_flags, 0); 1061 break; 1062 default: 1063 panic("ufs_whiteout: unknown op"); 1064 } 1065 return (error); 1066} 1067 1068static volatile int rename_restarts; 1069SYSCTL_INT(_vfs_ufs, OID_AUTO, rename_restarts, CTLFLAG_RD, 1070 __DEVOLATILE(int *, &rename_restarts), 0, 1071 "Times rename had to restart due to lock contention"); 1072 1073/* 1074 * Rename system call. 1075 * rename("foo", "bar"); 1076 * is essentially 1077 * unlink("bar"); 1078 * link("foo", "bar"); 1079 * unlink("foo"); 1080 * but ``atomically''. Can't do full commit without saving state in the 1081 * inode on disk which isn't feasible at this time. Best we can do is 1082 * always guarantee the target exists. 1083 * 1084 * Basic algorithm is: 1085 * 1086 * 1) Bump link count on source while we're linking it to the 1087 * target. This also ensure the inode won't be deleted out 1088 * from underneath us while we work (it may be truncated by 1089 * a concurrent `trunc' or `open' for creation). 1090 * 2) Link source to destination. If destination already exists, 1091 * delete it first. 1092 * 3) Unlink source reference to inode if still around. If a 1093 * directory was moved and the parent of the destination 1094 * is different from the source, patch the ".." entry in the 1095 * directory. 1096 */ 1097static int 1098ufs_rename(ap) 1099 struct vop_rename_args /* { 1100 struct vnode *a_fdvp; 1101 struct vnode *a_fvp; 1102 struct componentname *a_fcnp; 1103 struct vnode *a_tdvp; 1104 struct vnode *a_tvp; 1105 struct componentname *a_tcnp; 1106 } */ *ap; 1107{ 1108 struct vnode *tvp = ap->a_tvp; 1109 struct vnode *tdvp = ap->a_tdvp; 1110 struct vnode *fvp = ap->a_fvp; 1111 struct vnode *fdvp = ap->a_fdvp; 1112 struct vnode *nvp; 1113 struct componentname *tcnp = ap->a_tcnp; 1114 struct componentname *fcnp = ap->a_fcnp; 1115 struct thread *td = fcnp->cn_thread; 1116 struct inode *fip, *tip, *tdp, *fdp; 1117 struct direct newdir; 1118 off_t endoff; 1119 int doingdirectory, newparent; 1120 int error = 0; 1121 struct mount *mp; 1122 ino_t ino; 1123 1124#ifdef INVARIANTS 1125 if ((tcnp->cn_flags & HASBUF) == 0 || 1126 (fcnp->cn_flags & HASBUF) == 0) 1127 panic("ufs_rename: no name"); 1128#endif 1129 endoff = 0; 1130 mp = tdvp->v_mount; 1131 VOP_UNLOCK(tdvp, 0); 1132 if (tvp && tvp != tdvp) 1133 VOP_UNLOCK(tvp, 0); 1134 /* 1135 * Check for cross-device rename. 1136 */ 1137 if ((fvp->v_mount != tdvp->v_mount) || 1138 (tvp && (fvp->v_mount != tvp->v_mount))) { 1139 error = EXDEV; 1140 mp = NULL; 1141 goto releout; 1142 } 1143relock: 1144 /* 1145 * We need to acquire 2 to 4 locks depending on whether tvp is NULL 1146 * and fdvp and tdvp are the same directory. Subsequently we need 1147 * to double-check all paths and in the directory rename case we 1148 * need to verify that we are not creating a directory loop. To 1149 * handle this we acquire all but fdvp using non-blocking 1150 * acquisitions. If we fail to acquire any lock in the path we will 1151 * drop all held locks, acquire the new lock in a blocking fashion, 1152 * and then release it and restart the rename. This acquire/release 1153 * step ensures that we do not spin on a lock waiting for release. 1154 */ 1155 error = vn_lock(fdvp, LK_EXCLUSIVE); 1156 if (error) 1157 goto releout; 1158 if (vn_lock(tdvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) { 1159 VOP_UNLOCK(fdvp, 0); 1160 error = vn_lock(tdvp, LK_EXCLUSIVE); 1161 if (error) 1162 goto releout; 1163 VOP_UNLOCK(tdvp, 0); 1164 atomic_add_int(&rename_restarts, 1); 1165 goto relock; 1166 } 1167 /* 1168 * Re-resolve fvp to be certain it still exists and fetch the 1169 * correct vnode. 1170 */ 1171 error = ufs_lookup_ino(fdvp, NULL, fcnp, &ino); 1172 if (error) { 1173 VOP_UNLOCK(fdvp, 0); 1174 VOP_UNLOCK(tdvp, 0); 1175 goto releout; 1176 } 1177 error = VFS_VGET(mp, ino, LK_EXCLUSIVE | LK_NOWAIT, &nvp); 1178 if (error) { 1179 VOP_UNLOCK(fdvp, 0); 1180 VOP_UNLOCK(tdvp, 0); 1181 if (error != EBUSY) 1182 goto releout; 1183 error = VFS_VGET(mp, ino, LK_EXCLUSIVE, &nvp); 1184 if (error != 0) 1185 goto releout; 1186 VOP_UNLOCK(nvp, 0); 1187 vrele(fvp); 1188 fvp = nvp; 1189 atomic_add_int(&rename_restarts, 1); 1190 goto relock; 1191 } 1192 vrele(fvp); 1193 fvp = nvp; 1194 /* 1195 * Re-resolve tvp and acquire the vnode lock if present. 1196 */ 1197 error = ufs_lookup_ino(tdvp, NULL, tcnp, &ino); 1198 if (error != 0 && error != EJUSTRETURN) { 1199 VOP_UNLOCK(fdvp, 0); 1200 VOP_UNLOCK(tdvp, 0); 1201 VOP_UNLOCK(fvp, 0); 1202 goto releout; 1203 } 1204 /* 1205 * If tvp disappeared we just carry on. 1206 */ 1207 if (error == EJUSTRETURN && tvp != NULL) { 1208 vrele(tvp); 1209 tvp = NULL; 1210 } 1211 /* 1212 * Get the tvp ino if the lookup succeeded. We may have to restart 1213 * if the non-blocking acquire fails. 1214 */ 1215 if (error == 0) { 1216 nvp = NULL; 1217 error = VFS_VGET(mp, ino, LK_EXCLUSIVE | LK_NOWAIT, &nvp); 1218 if (tvp) 1219 vrele(tvp); 1220 tvp = nvp; 1221 if (error) { 1222 VOP_UNLOCK(fdvp, 0); 1223 VOP_UNLOCK(tdvp, 0); 1224 VOP_UNLOCK(fvp, 0); 1225 if (error != EBUSY) 1226 goto releout; 1227 error = VFS_VGET(mp, ino, LK_EXCLUSIVE, &nvp); 1228 if (error != 0) 1229 goto releout; 1230 vput(nvp); 1231 atomic_add_int(&rename_restarts, 1); 1232 goto relock; 1233 } 1234 } 1235 fdp = VTOI(fdvp); 1236 fip = VTOI(fvp); 1237 tdp = VTOI(tdvp); 1238 tip = NULL; 1239 if (tvp) 1240 tip = VTOI(tvp); 1241 if (tvp && ((VTOI(tvp)->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) || 1242 (VTOI(tdvp)->i_flags & APPEND))) { 1243 error = EPERM; 1244 goto unlockout; 1245 } 1246 /* 1247 * Renaming a file to itself has no effect. The upper layers should 1248 * not call us in that case. However, things could change after 1249 * we drop the locks above. 1250 */ 1251 if (fvp == tvp) { 1252 error = 0; 1253 goto unlockout; 1254 } 1255 doingdirectory = 0; 1256 newparent = 0; 1257 ino = fip->i_number; 1258 if (fip->i_nlink >= LINK_MAX) { 1259 error = EMLINK; 1260 goto unlockout; 1261 } 1262 if ((fip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) 1263 || (fdp->i_flags & APPEND)) { 1264 error = EPERM; 1265 goto unlockout; 1266 } 1267 if ((fip->i_mode & IFMT) == IFDIR) { 1268 /* 1269 * Avoid ".", "..", and aliases of "." for obvious reasons. 1270 */ 1271 if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') || 1272 fdp == fip || 1273 (fcnp->cn_flags | tcnp->cn_flags) & ISDOTDOT) { 1274 error = EINVAL; 1275 goto unlockout; 1276 } 1277 if (fdp->i_number != tdp->i_number) 1278 newparent = tdp->i_number; 1279 doingdirectory = 1; 1280 } 1281 if ((fvp->v_type == VDIR && fvp->v_mountedhere != NULL) || 1282 (tvp != NULL && tvp->v_type == VDIR && 1283 tvp->v_mountedhere != NULL)) { 1284 error = EXDEV; 1285 goto unlockout; 1286 } 1287 1288 /* 1289 * If ".." must be changed (ie the directory gets a new 1290 * parent) then the source directory must not be in the 1291 * directory hierarchy above the target, as this would 1292 * orphan everything below the source directory. Also 1293 * the user must have write permission in the source so 1294 * as to be able to change "..". 1295 */ 1296 if (doingdirectory && newparent) { 1297 error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, tcnp->cn_thread); 1298 if (error) 1299 goto unlockout; 1300 error = ufs_checkpath(ino, fdp->i_number, tdp, tcnp->cn_cred, 1301 &ino); 1302 /* 1303 * We encountered a lock that we have to wait for. Unlock 1304 * everything else and VGET before restarting. 1305 */ 1306 if (ino) { 1307 VOP_UNLOCK(fdvp, 0); 1308 VOP_UNLOCK(fvp, 0); 1309 VOP_UNLOCK(tdvp, 0); 1310 if (tvp) 1311 VOP_UNLOCK(tvp, 0); 1312 error = VFS_VGET(mp, ino, LK_SHARED, &nvp); 1313 if (error == 0) 1314 vput(nvp); 1315 atomic_add_int(&rename_restarts, 1); 1316 goto relock; 1317 } 1318 if (error) 1319 goto unlockout; 1320 if ((tcnp->cn_flags & SAVESTART) == 0) 1321 panic("ufs_rename: lost to startdir"); 1322 } 1323 if (fip->i_effnlink == 0 || fdp->i_effnlink == 0 || 1324 tdp->i_effnlink == 0) 1325 panic("Bad effnlink fip %p, fdp %p, tdp %p", fip, fdp, tdp); 1326 1327 /* 1328 * 1) Bump link count while we're moving stuff 1329 * around. If we crash somewhere before 1330 * completing our work, the link count 1331 * may be wrong, but correctable. 1332 */ 1333 fip->i_effnlink++; 1334 fip->i_nlink++; 1335 DIP_SET(fip, i_nlink, fip->i_nlink); 1336 fip->i_flag |= IN_CHANGE; 1337 if (DOINGSOFTDEP(fvp)) 1338 softdep_setup_link(tdp, fip); 1339 error = UFS_UPDATE(fvp, !(DOINGSOFTDEP(fvp) | DOINGASYNC(fvp))); 1340 if (error) 1341 goto bad; 1342 1343 /* 1344 * 2) If target doesn't exist, link the target 1345 * to the source and unlink the source. 1346 * Otherwise, rewrite the target directory 1347 * entry to reference the source inode and 1348 * expunge the original entry's existence. 1349 */ 1350 if (tip == NULL) { 1351 if (tdp->i_dev != fip->i_dev) 1352 panic("ufs_rename: EXDEV"); 1353 if (doingdirectory && newparent) { 1354 /* 1355 * Account for ".." in new directory. 1356 * When source and destination have the same 1357 * parent we don't adjust the link count. The 1358 * actual link modification is completed when 1359 * .. is rewritten below. 1360 */ 1361 if ((nlink_t)tdp->i_nlink >= LINK_MAX) { 1362 error = EMLINK; 1363 goto bad; 1364 } 1365 } 1366 ufs_makedirentry(fip, tcnp, &newdir); 1367 error = ufs_direnter(tdvp, NULL, &newdir, tcnp, NULL, 1); 1368 if (error) 1369 goto bad; 1370 /* Setup tdvp for directory compaction if needed. */ 1371 if (tdp->i_count && tdp->i_endoff && 1372 tdp->i_endoff < tdp->i_size) 1373 endoff = tdp->i_endoff; 1374 } else { 1375 if (tip->i_dev != tdp->i_dev || tip->i_dev != fip->i_dev) 1376 panic("ufs_rename: EXDEV"); 1377 /* 1378 * Short circuit rename(foo, foo). 1379 */ 1380 if (tip->i_number == fip->i_number) 1381 panic("ufs_rename: same file"); 1382 /* 1383 * If the parent directory is "sticky", then the caller 1384 * must possess VADMIN for the parent directory, or the 1385 * destination of the rename. This implements append-only 1386 * directories. 1387 */ 1388 if ((tdp->i_mode & S_ISTXT) && 1389 VOP_ACCESS(tdvp, VADMIN, tcnp->cn_cred, td) && 1390 VOP_ACCESS(tvp, VADMIN, tcnp->cn_cred, td)) { 1391 error = EPERM; 1392 goto bad; 1393 } 1394 /* 1395 * Target must be empty if a directory and have no links 1396 * to it. Also, ensure source and target are compatible 1397 * (both directories, or both not directories). 1398 */ 1399 if ((tip->i_mode & IFMT) == IFDIR) { 1400 if ((tip->i_effnlink > 2) || 1401 !ufs_dirempty(tip, tdp->i_number, tcnp->cn_cred)) { 1402 error = ENOTEMPTY; 1403 goto bad; 1404 } 1405 if (!doingdirectory) { 1406 error = ENOTDIR; 1407 goto bad; 1408 } 1409 cache_purge(tdvp); 1410 } else if (doingdirectory) { 1411 error = EISDIR; 1412 goto bad; 1413 } 1414 if (doingdirectory) { 1415 if (!newparent) { 1416 tdp->i_effnlink--; 1417 if (DOINGSOFTDEP(tdvp)) 1418 softdep_change_linkcnt(tdp); 1419 } 1420 tip->i_effnlink--; 1421 if (DOINGSOFTDEP(tvp)) 1422 softdep_change_linkcnt(tip); 1423 } 1424 error = ufs_dirrewrite(tdp, tip, fip->i_number, 1425 IFTODT(fip->i_mode), 1426 (doingdirectory && newparent) ? newparent : doingdirectory); 1427 if (error) { 1428 if (doingdirectory) { 1429 if (!newparent) { 1430 tdp->i_effnlink++; 1431 if (DOINGSOFTDEP(tdvp)) 1432 softdep_change_linkcnt(tdp); 1433 } 1434 tip->i_effnlink++; 1435 if (DOINGSOFTDEP(tvp)) 1436 softdep_change_linkcnt(tip); 1437 } 1438 } 1439 if (doingdirectory && !DOINGSOFTDEP(tvp)) { 1440 /* 1441 * The only stuff left in the directory is "." 1442 * and "..". The "." reference is inconsequential 1443 * since we are quashing it. We have removed the "." 1444 * reference and the reference in the parent directory, 1445 * but there may be other hard links. The soft 1446 * dependency code will arrange to do these operations 1447 * after the parent directory entry has been deleted on 1448 * disk, so when running with that code we avoid doing 1449 * them now. 1450 */ 1451 if (!newparent) { 1452 tdp->i_nlink--; 1453 DIP_SET(tdp, i_nlink, tdp->i_nlink); 1454 tdp->i_flag |= IN_CHANGE; 1455 } 1456 tip->i_nlink--; 1457 DIP_SET(tip, i_nlink, tip->i_nlink); 1458 tip->i_flag |= IN_CHANGE; 1459 } 1460 } 1461 1462 /* 1463 * 3) Unlink the source. We have to resolve the path again to 1464 * fixup the directory offset and count for ufs_dirremove. 1465 */ 1466 if (fdvp == tdvp) { 1467 error = ufs_lookup_ino(fdvp, NULL, fcnp, &ino); 1468 if (error) 1469 panic("ufs_rename: from entry went away!"); 1470 if (ino != fip->i_number) 1471 panic("ufs_rename: ino mismatch %ju != %ju\n", 1472 (uintmax_t)ino, (uintmax_t)fip->i_number); 1473 } 1474 /* 1475 * If the source is a directory with a 1476 * new parent, the link count of the old 1477 * parent directory must be decremented 1478 * and ".." set to point to the new parent. 1479 */ 1480 if (doingdirectory && newparent) { 1481 /* 1482 * If tip exists we simply use its link, otherwise we must 1483 * add a new one. 1484 */ 1485 if (tip == NULL) { 1486 tdp->i_effnlink++; 1487 tdp->i_nlink++; 1488 DIP_SET(tdp, i_nlink, tdp->i_nlink); 1489 tdp->i_flag |= IN_CHANGE; 1490 if (DOINGSOFTDEP(tdvp)) 1491 softdep_setup_dotdot_link(tdp, fip); 1492 error = UFS_UPDATE(tdvp, !(DOINGSOFTDEP(tdvp) | 1493 DOINGASYNC(tdvp))); 1494 /* Don't go to bad here as the new link exists. */ 1495 if (error) 1496 goto unlockout; 1497 } else if (DOINGSUJ(tdvp)) 1498 /* Journal must account for each new link. */ 1499 softdep_setup_dotdot_link(tdp, fip); 1500 fip->i_offset = mastertemplate.dot_reclen; 1501 ufs_dirrewrite(fip, fdp, newparent, DT_DIR, 0); 1502 cache_purge(fdvp); 1503 } 1504 error = ufs_dirremove(fdvp, fip, fcnp->cn_flags, 0); 1505 /* 1506 * The kern_renameat() looks up the fvp using the DELETE flag, which 1507 * causes the removal of the name cache entry for fvp. 1508 * As the relookup of the fvp is done in two steps: 1509 * ufs_lookup_ino() and then VFS_VGET(), another thread might do a 1510 * normal lookup of the from name just before the VFS_VGET() call, 1511 * causing the cache entry to be re-instantiated. 1512 * 1513 * The same issue also applies to tvp if it exists as 1514 * otherwise we may have a stale name cache entry for the new 1515 * name that references the old i-node if it has other links 1516 * or open file descriptors. 1517 */ 1518 cache_purge(fvp); 1519 if (tvp) 1520 cache_purge(tvp); 1521 cache_purge_negative(tdvp); 1522 1523unlockout: 1524 vput(fdvp); 1525 vput(fvp); 1526 if (tvp) 1527 vput(tvp); 1528 /* 1529 * If compaction or fsync was requested do it now that other locks 1530 * are no longer needed. 1531 */ 1532 if (error == 0 && endoff != 0) { 1533 error = UFS_TRUNCATE(tdvp, endoff, IO_NORMAL | IO_SYNC, 1534 tcnp->cn_cred); 1535 if (error != 0) 1536 vn_printf(tdvp, "ufs_rename: failed to truncate " 1537 "err %d", error); 1538#ifdef UFS_DIRHASH 1539 else if (tdp->i_dirhash != NULL) 1540 ufsdirhash_dirtrunc(tdp, endoff); 1541#endif 1542 /* 1543 * Even if the directory compaction failed, rename was 1544 * succesful. Do not propagate a UFS_TRUNCATE() error 1545 * to the caller. 1546 */ 1547 error = 0; 1548 } 1549 if (error == 0 && tdp->i_flag & IN_NEEDSYNC) 1550 error = VOP_FSYNC(tdvp, MNT_WAIT, td); 1551 vput(tdvp); 1552 return (error); 1553 1554bad: 1555 fip->i_effnlink--; 1556 fip->i_nlink--; 1557 DIP_SET(fip, i_nlink, fip->i_nlink); 1558 fip->i_flag |= IN_CHANGE; 1559 if (DOINGSOFTDEP(fvp)) 1560 softdep_revert_link(tdp, fip); 1561 goto unlockout; 1562 1563releout: 1564 vrele(fdvp); 1565 vrele(fvp); 1566 vrele(tdvp); 1567 if (tvp) 1568 vrele(tvp); 1569 1570 return (error); 1571} 1572 1573#ifdef UFS_ACL 1574static int 1575ufs_do_posix1e_acl_inheritance_dir(struct vnode *dvp, struct vnode *tvp, 1576 mode_t dmode, struct ucred *cred, struct thread *td) 1577{ 1578 int error; 1579 struct inode *ip = VTOI(tvp); 1580 struct acl *dacl, *acl; 1581 1582 acl = acl_alloc(M_WAITOK); 1583 dacl = acl_alloc(M_WAITOK); 1584 1585 /* 1586 * Retrieve default ACL from parent, if any. 1587 */ 1588 error = VOP_GETACL(dvp, ACL_TYPE_DEFAULT, acl, cred, td); 1589 switch (error) { 1590 case 0: 1591 /* 1592 * Retrieved a default ACL, so merge mode and ACL if 1593 * necessary. If the ACL is empty, fall through to 1594 * the "not defined or available" case. 1595 */ 1596 if (acl->acl_cnt != 0) { 1597 dmode = acl_posix1e_newfilemode(dmode, acl); 1598 ip->i_mode = dmode; 1599 DIP_SET(ip, i_mode, dmode); 1600 *dacl = *acl; 1601 ufs_sync_acl_from_inode(ip, acl); 1602 break; 1603 } 1604 /* FALLTHROUGH */ 1605 1606 case EOPNOTSUPP: 1607 /* 1608 * Just use the mode as-is. 1609 */ 1610 ip->i_mode = dmode; 1611 DIP_SET(ip, i_mode, dmode); 1612 error = 0; 1613 goto out; 1614 1615 default: 1616 goto out; 1617 } 1618 1619 /* 1620 * XXX: If we abort now, will Soft Updates notify the extattr 1621 * code that the EAs for the file need to be released? 1622 */ 1623 error = VOP_SETACL(tvp, ACL_TYPE_ACCESS, acl, cred, td); 1624 if (error == 0) 1625 error = VOP_SETACL(tvp, ACL_TYPE_DEFAULT, dacl, cred, td); 1626 switch (error) { 1627 case 0: 1628 break; 1629 1630 case EOPNOTSUPP: 1631 /* 1632 * XXX: This should not happen, as EOPNOTSUPP above 1633 * was supposed to free acl. 1634 */ 1635 printf("ufs_mkdir: VOP_GETACL() but no VOP_SETACL()\n"); 1636 /* 1637 panic("ufs_mkdir: VOP_GETACL() but no VOP_SETACL()"); 1638 */ 1639 break; 1640 1641 default: 1642 goto out; 1643 } 1644 1645out: 1646 acl_free(acl); 1647 acl_free(dacl); 1648 1649 return (error); 1650} 1651 1652static int 1653ufs_do_posix1e_acl_inheritance_file(struct vnode *dvp, struct vnode *tvp, 1654 mode_t mode, struct ucred *cred, struct thread *td) 1655{ 1656 int error; 1657 struct inode *ip = VTOI(tvp); 1658 struct acl *acl; 1659 1660 acl = acl_alloc(M_WAITOK); 1661 1662 /* 1663 * Retrieve default ACL for parent, if any. 1664 */ 1665 error = VOP_GETACL(dvp, ACL_TYPE_DEFAULT, acl, cred, td); 1666 switch (error) { 1667 case 0: 1668 /* 1669 * Retrieved a default ACL, so merge mode and ACL if 1670 * necessary. 1671 */ 1672 if (acl->acl_cnt != 0) { 1673 /* 1674 * Two possible ways for default ACL to not 1675 * be present. First, the EA can be 1676 * undefined, or second, the default ACL can 1677 * be blank. If it's blank, fall through to 1678 * the it's not defined case. 1679 */ 1680 mode = acl_posix1e_newfilemode(mode, acl); 1681 ip->i_mode = mode; 1682 DIP_SET(ip, i_mode, mode); 1683 ufs_sync_acl_from_inode(ip, acl); 1684 break; 1685 } 1686 /* FALLTHROUGH */ 1687 1688 case EOPNOTSUPP: 1689 /* 1690 * Just use the mode as-is. 1691 */ 1692 ip->i_mode = mode; 1693 DIP_SET(ip, i_mode, mode); 1694 error = 0; 1695 goto out; 1696 1697 default: 1698 goto out; 1699 } 1700 1701 /* 1702 * XXX: If we abort now, will Soft Updates notify the extattr 1703 * code that the EAs for the file need to be released? 1704 */ 1705 error = VOP_SETACL(tvp, ACL_TYPE_ACCESS, acl, cred, td); 1706 switch (error) { 1707 case 0: 1708 break; 1709 1710 case EOPNOTSUPP: 1711 /* 1712 * XXX: This should not happen, as EOPNOTSUPP above was 1713 * supposed to free acl. 1714 */ 1715 printf("ufs_makeinode: VOP_GETACL() but no " 1716 "VOP_SETACL()\n"); 1717 /* panic("ufs_makeinode: VOP_GETACL() but no " 1718 "VOP_SETACL()"); */ 1719 break; 1720 1721 default: 1722 goto out; 1723 } 1724 1725out: 1726 acl_free(acl); 1727 1728 return (error); 1729} 1730 1731static int 1732ufs_do_nfs4_acl_inheritance(struct vnode *dvp, struct vnode *tvp, 1733 mode_t child_mode, struct ucred *cred, struct thread *td) 1734{ 1735 int error; 1736 struct acl *parent_aclp, *child_aclp; 1737 1738 parent_aclp = acl_alloc(M_WAITOK); 1739 child_aclp = acl_alloc(M_WAITOK | M_ZERO); 1740 1741 error = ufs_getacl_nfs4_internal(dvp, parent_aclp, td); 1742 if (error) 1743 goto out; 1744 acl_nfs4_compute_inherited_acl(parent_aclp, child_aclp, 1745 child_mode, VTOI(tvp)->i_uid, tvp->v_type == VDIR); 1746 error = ufs_setacl_nfs4_internal(tvp, child_aclp, td); 1747 if (error) 1748 goto out; 1749out: 1750 acl_free(parent_aclp); 1751 acl_free(child_aclp); 1752 1753 return (error); 1754} 1755#endif 1756 1757/* 1758 * Mkdir system call 1759 */ 1760static int 1761ufs_mkdir(ap) 1762 struct vop_mkdir_args /* { 1763 struct vnode *a_dvp; 1764 struct vnode **a_vpp; 1765 struct componentname *a_cnp; 1766 struct vattr *a_vap; 1767 } */ *ap; 1768{ 1769 struct vnode *dvp = ap->a_dvp; 1770 struct vattr *vap = ap->a_vap; 1771 struct componentname *cnp = ap->a_cnp; 1772 struct inode *ip, *dp; 1773 struct vnode *tvp; 1774 struct buf *bp; 1775 struct dirtemplate dirtemplate, *dtp; 1776 struct direct newdir; 1777 int error, dmode; 1778 long blkoff; 1779 1780#ifdef INVARIANTS 1781 if ((cnp->cn_flags & HASBUF) == 0) 1782 panic("ufs_mkdir: no name"); 1783#endif 1784 dp = VTOI(dvp); 1785 if ((nlink_t)dp->i_nlink >= LINK_MAX) { 1786 error = EMLINK; 1787 goto out; 1788 } 1789 dmode = vap->va_mode & 0777; 1790 dmode |= IFDIR; 1791 /* 1792 * Must simulate part of ufs_makeinode here to acquire the inode, 1793 * but not have it entered in the parent directory. The entry is 1794 * made later after writing "." and ".." entries. 1795 */ 1796 error = UFS_VALLOC(dvp, dmode, cnp->cn_cred, &tvp); 1797 if (error) 1798 goto out; 1799 ip = VTOI(tvp); 1800 ip->i_gid = dp->i_gid; 1801 DIP_SET(ip, i_gid, dp->i_gid); 1802#ifdef SUIDDIR 1803 { 1804#ifdef QUOTA 1805 struct ucred ucred, *ucp; 1806 gid_t ucred_group; 1807 ucp = cnp->cn_cred; 1808#endif 1809 /* 1810 * If we are hacking owners here, (only do this where told to) 1811 * and we are not giving it TO root, (would subvert quotas) 1812 * then go ahead and give it to the other user. 1813 * The new directory also inherits the SUID bit. 1814 * If user's UID and dir UID are the same, 1815 * 'give it away' so that the SUID is still forced on. 1816 */ 1817 if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) && 1818 (dp->i_mode & ISUID) && dp->i_uid) { 1819 dmode |= ISUID; 1820 ip->i_uid = dp->i_uid; 1821 DIP_SET(ip, i_uid, dp->i_uid); 1822#ifdef QUOTA 1823 if (dp->i_uid != cnp->cn_cred->cr_uid) { 1824 /* 1825 * Make sure the correct user gets charged 1826 * for the space. 1827 * Make a dummy credential for the victim. 1828 * XXX This seems to never be accessed out of 1829 * our context so a stack variable is ok. 1830 */ 1831 refcount_init(&ucred.cr_ref, 1); 1832 ucred.cr_uid = ip->i_uid; 1833 ucred.cr_ngroups = 1; 1834 ucred.cr_groups = &ucred_group; 1835 ucred.cr_groups[0] = dp->i_gid; 1836 ucp = &ucred; 1837 } 1838#endif 1839 } else { 1840 ip->i_uid = cnp->cn_cred->cr_uid; 1841 DIP_SET(ip, i_uid, ip->i_uid); 1842 } 1843#ifdef QUOTA 1844 if ((error = getinoquota(ip)) || 1845 (error = chkiq(ip, 1, ucp, 0))) { 1846 if (DOINGSOFTDEP(tvp)) 1847 softdep_revert_link(dp, ip); 1848 UFS_VFREE(tvp, ip->i_number, dmode); 1849 vput(tvp); 1850 return (error); 1851 } 1852#endif 1853 } 1854#else /* !SUIDDIR */ 1855 ip->i_uid = cnp->cn_cred->cr_uid; 1856 DIP_SET(ip, i_uid, ip->i_uid); 1857#ifdef QUOTA 1858 if ((error = getinoquota(ip)) || 1859 (error = chkiq(ip, 1, cnp->cn_cred, 0))) { 1860 if (DOINGSOFTDEP(tvp)) 1861 softdep_revert_link(dp, ip); 1862 UFS_VFREE(tvp, ip->i_number, dmode); 1863 vput(tvp); 1864 return (error); 1865 } 1866#endif 1867#endif /* !SUIDDIR */ 1868 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 1869 ip->i_mode = dmode; 1870 DIP_SET(ip, i_mode, dmode); 1871 tvp->v_type = VDIR; /* Rest init'd in getnewvnode(). */ 1872 ip->i_effnlink = 2; 1873 ip->i_nlink = 2; 1874 DIP_SET(ip, i_nlink, 2); 1875 1876 if (cnp->cn_flags & ISWHITEOUT) { 1877 ip->i_flags |= UF_OPAQUE; 1878 DIP_SET(ip, i_flags, ip->i_flags); 1879 } 1880 1881 /* 1882 * Bump link count in parent directory to reflect work done below. 1883 * Should be done before reference is created so cleanup is 1884 * possible if we crash. 1885 */ 1886 dp->i_effnlink++; 1887 dp->i_nlink++; 1888 DIP_SET(dp, i_nlink, dp->i_nlink); 1889 dp->i_flag |= IN_CHANGE; 1890 if (DOINGSOFTDEP(dvp)) 1891 softdep_setup_mkdir(dp, ip); 1892 error = UFS_UPDATE(dvp, !(DOINGSOFTDEP(dvp) | DOINGASYNC(dvp))); 1893 if (error) 1894 goto bad; 1895#ifdef MAC 1896 if (dvp->v_mount->mnt_flag & MNT_MULTILABEL) { 1897 error = mac_vnode_create_extattr(cnp->cn_cred, dvp->v_mount, 1898 dvp, tvp, cnp); 1899 if (error) 1900 goto bad; 1901 } 1902#endif 1903#ifdef UFS_ACL 1904 if (dvp->v_mount->mnt_flag & MNT_ACLS) { 1905 error = ufs_do_posix1e_acl_inheritance_dir(dvp, tvp, dmode, 1906 cnp->cn_cred, cnp->cn_thread); 1907 if (error) 1908 goto bad; 1909 } else if (dvp->v_mount->mnt_flag & MNT_NFS4ACLS) { 1910 error = ufs_do_nfs4_acl_inheritance(dvp, tvp, dmode, 1911 cnp->cn_cred, cnp->cn_thread); 1912 if (error) 1913 goto bad; 1914 } 1915#endif /* !UFS_ACL */ 1916 1917 /* 1918 * Initialize directory with "." and ".." from static template. 1919 */ 1920 if (dvp->v_mount->mnt_maxsymlinklen > 0) 1921 dtp = &mastertemplate; 1922 else 1923 dtp = (struct dirtemplate *)&omastertemplate; 1924 dirtemplate = *dtp; 1925 dirtemplate.dot_ino = ip->i_number; 1926 dirtemplate.dotdot_ino = dp->i_number; 1927 vnode_pager_setsize(tvp, DIRBLKSIZ); 1928 if ((error = UFS_BALLOC(tvp, (off_t)0, DIRBLKSIZ, cnp->cn_cred, 1929 BA_CLRBUF, &bp)) != 0) 1930 goto bad; 1931 ip->i_size = DIRBLKSIZ; 1932 DIP_SET(ip, i_size, DIRBLKSIZ); 1933 ip->i_flag |= IN_CHANGE | IN_UPDATE; 1934 bcopy((caddr_t)&dirtemplate, (caddr_t)bp->b_data, sizeof dirtemplate); 1935 if (DOINGSOFTDEP(tvp)) { 1936 /* 1937 * Ensure that the entire newly allocated block is a 1938 * valid directory so that future growth within the 1939 * block does not have to ensure that the block is 1940 * written before the inode. 1941 */ 1942 blkoff = DIRBLKSIZ; 1943 while (blkoff < bp->b_bcount) { 1944 ((struct direct *) 1945 (bp->b_data + blkoff))->d_reclen = DIRBLKSIZ; 1946 blkoff += DIRBLKSIZ; 1947 } 1948 } 1949 if ((error = UFS_UPDATE(tvp, !(DOINGSOFTDEP(tvp) | 1950 DOINGASYNC(tvp)))) != 0) { 1951 (void)bwrite(bp); 1952 goto bad; 1953 } 1954 /* 1955 * Directory set up, now install its entry in the parent directory. 1956 * 1957 * If we are not doing soft dependencies, then we must write out the 1958 * buffer containing the new directory body before entering the new 1959 * name in the parent. If we are doing soft dependencies, then the 1960 * buffer containing the new directory body will be passed to and 1961 * released in the soft dependency code after the code has attached 1962 * an appropriate ordering dependency to the buffer which ensures that 1963 * the buffer is written before the new name is written in the parent. 1964 */ 1965 if (DOINGASYNC(dvp)) 1966 bdwrite(bp); 1967 else if (!DOINGSOFTDEP(dvp) && ((error = bwrite(bp)))) 1968 goto bad; 1969 ufs_makedirentry(ip, cnp, &newdir); 1970 error = ufs_direnter(dvp, tvp, &newdir, cnp, bp, 0); 1971 1972bad: 1973 if (error == 0) { 1974 *ap->a_vpp = tvp; 1975 } else { 1976 dp->i_effnlink--; 1977 dp->i_nlink--; 1978 DIP_SET(dp, i_nlink, dp->i_nlink); 1979 dp->i_flag |= IN_CHANGE; 1980 /* 1981 * No need to do an explicit VOP_TRUNCATE here, vrele will 1982 * do this for us because we set the link count to 0. 1983 */ 1984 ip->i_effnlink = 0; 1985 ip->i_nlink = 0; 1986 DIP_SET(ip, i_nlink, 0); 1987 ip->i_flag |= IN_CHANGE; 1988 if (DOINGSOFTDEP(tvp)) 1989 softdep_revert_mkdir(dp, ip); 1990 1991 vput(tvp); 1992 } 1993out: 1994 return (error); 1995} 1996 1997/* 1998 * Rmdir system call. 1999 */ 2000static int 2001ufs_rmdir(ap) 2002 struct vop_rmdir_args /* { 2003 struct vnode *a_dvp; 2004 struct vnode *a_vp; 2005 struct componentname *a_cnp; 2006 } */ *ap; 2007{ 2008 struct vnode *vp = ap->a_vp; 2009 struct vnode *dvp = ap->a_dvp; 2010 struct componentname *cnp = ap->a_cnp; 2011 struct inode *ip, *dp; 2012 int error; 2013 2014 ip = VTOI(vp); 2015 dp = VTOI(dvp); 2016 2017 /* 2018 * Do not remove a directory that is in the process of being renamed. 2019 * Verify the directory is empty (and valid). Rmdir ".." will not be 2020 * valid since ".." will contain a reference to the current directory 2021 * and thus be non-empty. Do not allow the removal of mounted on 2022 * directories (this can happen when an NFS exported filesystem 2023 * tries to remove a locally mounted on directory). 2024 */ 2025 error = 0; 2026 if (ip->i_effnlink < 2) { 2027 error = EINVAL; 2028 goto out; 2029 } 2030 if (dp->i_effnlink < 3) 2031 panic("ufs_dirrem: Bad link count %d on parent", 2032 dp->i_effnlink); 2033 if (!ufs_dirempty(ip, dp->i_number, cnp->cn_cred)) { 2034 error = ENOTEMPTY; 2035 goto out; 2036 } 2037 if ((dp->i_flags & APPEND) 2038 || (ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))) { 2039 error = EPERM; 2040 goto out; 2041 } 2042 if (vp->v_mountedhere != 0) { 2043 error = EINVAL; 2044 goto out; 2045 } 2046#ifdef UFS_GJOURNAL 2047 ufs_gjournal_orphan(vp); 2048#endif 2049 /* 2050 * Delete reference to directory before purging 2051 * inode. If we crash in between, the directory 2052 * will be reattached to lost+found, 2053 */ 2054 dp->i_effnlink--; 2055 ip->i_effnlink--; 2056 if (DOINGSOFTDEP(vp)) 2057 softdep_setup_rmdir(dp, ip); 2058 error = ufs_dirremove(dvp, ip, cnp->cn_flags, 1); 2059 if (error) { 2060 dp->i_effnlink++; 2061 ip->i_effnlink++; 2062 if (DOINGSOFTDEP(vp)) 2063 softdep_revert_rmdir(dp, ip); 2064 goto out; 2065 } 2066 cache_purge(dvp); 2067 /* 2068 * The only stuff left in the directory is "." and "..". The "." 2069 * reference is inconsequential since we are quashing it. The soft 2070 * dependency code will arrange to do these operations after 2071 * the parent directory entry has been deleted on disk, so 2072 * when running with that code we avoid doing them now. 2073 */ 2074 if (!DOINGSOFTDEP(vp)) { 2075 dp->i_nlink--; 2076 DIP_SET(dp, i_nlink, dp->i_nlink); 2077 dp->i_flag |= IN_CHANGE; 2078 error = UFS_UPDATE(dvp, 0); 2079 ip->i_nlink--; 2080 DIP_SET(ip, i_nlink, ip->i_nlink); 2081 ip->i_flag |= IN_CHANGE; 2082 } 2083 cache_purge(vp); 2084#ifdef UFS_DIRHASH 2085 /* Kill any active hash; i_effnlink == 0, so it will not come back. */ 2086 if (ip->i_dirhash != NULL) 2087 ufsdirhash_free(ip); 2088#endif 2089out: 2090 return (error); 2091} 2092 2093/* 2094 * symlink -- make a symbolic link 2095 */ 2096static int 2097ufs_symlink(ap) 2098 struct vop_symlink_args /* { 2099 struct vnode *a_dvp; 2100 struct vnode **a_vpp; 2101 struct componentname *a_cnp; 2102 struct vattr *a_vap; 2103 char *a_target; 2104 } */ *ap; 2105{ 2106 struct vnode *vp, **vpp = ap->a_vpp; 2107 struct inode *ip; 2108 int len, error; 2109 2110 error = ufs_makeinode(IFLNK | ap->a_vap->va_mode, ap->a_dvp, 2111 vpp, ap->a_cnp); 2112 if (error) 2113 return (error); 2114 vp = *vpp; 2115 len = strlen(ap->a_target); 2116 if (len < vp->v_mount->mnt_maxsymlinklen) { 2117 ip = VTOI(vp); 2118 bcopy(ap->a_target, SHORTLINK(ip), len); 2119 ip->i_size = len; 2120 DIP_SET(ip, i_size, len); 2121 ip->i_flag |= IN_CHANGE | IN_UPDATE; 2122 error = UFS_UPDATE(vp, 0); 2123 } else 2124 error = vn_rdwr(UIO_WRITE, vp, ap->a_target, len, (off_t)0, 2125 UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK, 2126 ap->a_cnp->cn_cred, NOCRED, NULL, NULL); 2127 if (error) 2128 vput(vp); 2129 return (error); 2130} 2131 2132/* 2133 * Vnode op for reading directories. 2134 */ 2135int 2136ufs_readdir(ap) 2137 struct vop_readdir_args /* { 2138 struct vnode *a_vp; 2139 struct uio *a_uio; 2140 struct ucred *a_cred; 2141 int *a_eofflag; 2142 int *a_ncookies; 2143 u_long **a_cookies; 2144 } */ *ap; 2145{ 2146 struct vnode *vp = ap->a_vp; 2147 struct uio *uio = ap->a_uio; 2148 struct buf *bp; 2149 struct inode *ip; 2150 struct direct *dp, *edp; 2151 u_long *cookies; 2152 struct dirent dstdp; 2153 off_t offset, startoffset; 2154 size_t readcnt, skipcnt; 2155 ssize_t startresid; 2156 int ncookies; 2157 int error; 2158 2159 if (uio->uio_offset < 0) 2160 return (EINVAL); 2161 ip = VTOI(vp); 2162 if (ip->i_effnlink == 0) 2163 return (0); 2164 if (ap->a_ncookies != NULL) { 2165 ncookies = uio->uio_resid; 2166 if (uio->uio_offset >= ip->i_size) 2167 ncookies = 0; 2168 else if (ip->i_size - uio->uio_offset < ncookies) 2169 ncookies = ip->i_size - uio->uio_offset; 2170 ncookies = ncookies / (offsetof(struct direct, d_name) + 4) + 1; 2171 cookies = malloc(ncookies * sizeof(*cookies), M_TEMP, M_WAITOK); 2172 *ap->a_ncookies = ncookies; 2173 *ap->a_cookies = cookies; 2174 } else { 2175 ncookies = 0; 2176 cookies = NULL; 2177 } 2178 offset = startoffset = uio->uio_offset; 2179 startresid = uio->uio_resid; 2180 error = 0; 2181 while (error == 0 && uio->uio_resid > 0 && 2182 uio->uio_offset < ip->i_size) { 2183 error = ffs_blkatoff(vp, uio->uio_offset, NULL, &bp); 2184 if (error) 2185 break; 2186 if (bp->b_offset + bp->b_bcount > ip->i_size) 2187 readcnt = ip->i_size - bp->b_offset; 2188 else 2189 readcnt = bp->b_bcount; 2190 skipcnt = (size_t)(uio->uio_offset - bp->b_offset) & 2191 ~(size_t)(DIRBLKSIZ - 1); 2192 offset = bp->b_offset + skipcnt; 2193 dp = (struct direct *)&bp->b_data[skipcnt]; 2194 edp = (struct direct *)&bp->b_data[readcnt]; 2195 while (error == 0 && uio->uio_resid > 0 && dp < edp) { 2196 if (dp->d_reclen <= offsetof(struct direct, d_name) || 2197 (caddr_t)dp + dp->d_reclen > (caddr_t)edp) { 2198 error = EIO; 2199 break; 2200 } 2201#if BYTE_ORDER == LITTLE_ENDIAN 2202 /* Old filesystem format. */ 2203 if (vp->v_mount->mnt_maxsymlinklen <= 0) { 2204 dstdp.d_namlen = dp->d_type; 2205 dstdp.d_type = dp->d_namlen; 2206 } else 2207#endif 2208 { 2209 dstdp.d_namlen = dp->d_namlen; 2210 dstdp.d_type = dp->d_type; 2211 } 2212 if (offsetof(struct direct, d_name) + dstdp.d_namlen > 2213 dp->d_reclen) { 2214 error = EIO; 2215 break; 2216 } 2217 if (offset < startoffset || dp->d_ino == 0) 2218 goto nextentry; 2219 dstdp.d_fileno = dp->d_ino; 2220 dstdp.d_reclen = GENERIC_DIRSIZ(&dstdp); 2221 bcopy(dp->d_name, dstdp.d_name, dstdp.d_namlen); 2222 dstdp.d_name[dstdp.d_namlen] = '\0'; 2223 if (dstdp.d_reclen > uio->uio_resid) { 2224 if (uio->uio_resid == startresid) 2225 error = EINVAL; 2226 else 2227 error = EJUSTRETURN; 2228 break; 2229 } 2230 /* Advance dp. */ 2231 error = uiomove((caddr_t)&dstdp, dstdp.d_reclen, uio); 2232 if (error) 2233 break; 2234 if (cookies != NULL) { 2235 KASSERT(ncookies > 0, 2236 ("ufs_readdir: cookies buffer too small")); 2237 *cookies = offset + dp->d_reclen; 2238 cookies++; 2239 ncookies--; 2240 } 2241nextentry: 2242 offset += dp->d_reclen; 2243 dp = (struct direct *)((caddr_t)dp + dp->d_reclen); 2244 } 2245 bqrelse(bp); 2246 uio->uio_offset = offset; 2247 } 2248 /* We need to correct uio_offset. */ 2249 uio->uio_offset = offset; 2250 if (error == EJUSTRETURN) 2251 error = 0; 2252 if (ap->a_ncookies != NULL) { 2253 if (error == 0) { 2254 ap->a_ncookies -= ncookies; 2255 } else { 2256 free(*ap->a_cookies, M_TEMP); 2257 *ap->a_ncookies = 0; 2258 *ap->a_cookies = NULL; 2259 } 2260 } 2261 if (error == 0 && ap->a_eofflag) 2262 *ap->a_eofflag = ip->i_size <= uio->uio_offset; 2263 return (error); 2264} 2265 2266/* 2267 * Return target name of a symbolic link 2268 */ 2269static int 2270ufs_readlink(ap) 2271 struct vop_readlink_args /* { 2272 struct vnode *a_vp; 2273 struct uio *a_uio; 2274 struct ucred *a_cred; 2275 } */ *ap; 2276{ 2277 struct vnode *vp = ap->a_vp; 2278 struct inode *ip = VTOI(vp); 2279 doff_t isize; 2280 2281 isize = ip->i_size; 2282 if ((isize < vp->v_mount->mnt_maxsymlinklen) || 2283 DIP(ip, i_blocks) == 0) { /* XXX - for old fastlink support */ 2284 return (uiomove(SHORTLINK(ip), isize, ap->a_uio)); 2285 } 2286 return (VOP_READ(vp, ap->a_uio, 0, ap->a_cred)); 2287} 2288 2289/* 2290 * Calculate the logical to physical mapping if not done already, 2291 * then call the device strategy routine. 2292 * 2293 * In order to be able to swap to a file, the ufs_bmaparray() operation may not 2294 * deadlock on memory. See ufs_bmap() for details. 2295 */ 2296static int 2297ufs_strategy(ap) 2298 struct vop_strategy_args /* { 2299 struct vnode *a_vp; 2300 struct buf *a_bp; 2301 } */ *ap; 2302{ 2303 struct buf *bp = ap->a_bp; 2304 struct vnode *vp = ap->a_vp; 2305 struct bufobj *bo; 2306 struct inode *ip; 2307 ufs2_daddr_t blkno; 2308 int error; 2309 2310 ip = VTOI(vp); 2311 if (bp->b_blkno == bp->b_lblkno) { 2312 error = ufs_bmaparray(vp, bp->b_lblkno, &blkno, bp, NULL, NULL); 2313 bp->b_blkno = blkno; 2314 if (error) { 2315 bp->b_error = error; 2316 bp->b_ioflags |= BIO_ERROR; 2317 bufdone(bp); 2318 return (0); 2319 } 2320 if ((long)bp->b_blkno == -1) 2321 vfs_bio_clrbuf(bp); 2322 } 2323 if ((long)bp->b_blkno == -1) { 2324 bufdone(bp); 2325 return (0); 2326 } 2327 bp->b_iooffset = dbtob(bp->b_blkno); 2328 bo = ip->i_umbufobj; 2329 BO_STRATEGY(bo, bp); 2330 return (0); 2331} 2332 2333/* 2334 * Print out the contents of an inode. 2335 */ 2336static int 2337ufs_print(ap) 2338 struct vop_print_args /* { 2339 struct vnode *a_vp; 2340 } */ *ap; 2341{ 2342 struct vnode *vp = ap->a_vp; 2343 struct inode *ip = VTOI(vp); 2344 2345 printf("\tino %lu, on dev %s", (u_long)ip->i_number, 2346 devtoname(ip->i_dev)); 2347 if (vp->v_type == VFIFO) 2348 fifo_printinfo(vp); 2349 printf("\n"); 2350 return (0); 2351} 2352 2353/* 2354 * Close wrapper for fifos. 2355 * 2356 * Update the times on the inode then do device close. 2357 */ 2358static int 2359ufsfifo_close(ap) 2360 struct vop_close_args /* { 2361 struct vnode *a_vp; 2362 int a_fflag; 2363 struct ucred *a_cred; 2364 struct thread *a_td; 2365 } */ *ap; 2366{ 2367 struct vnode *vp = ap->a_vp; 2368 int usecount; 2369 2370 VI_LOCK(vp); 2371 usecount = vp->v_usecount; 2372 if (usecount > 1) 2373 ufs_itimes_locked(vp); 2374 VI_UNLOCK(vp); 2375 return (fifo_specops.vop_close(ap)); 2376} 2377 2378/* 2379 * Kqfilter wrapper for fifos. 2380 * 2381 * Fall through to ufs kqfilter routines if needed 2382 */ 2383static int 2384ufsfifo_kqfilter(ap) 2385 struct vop_kqfilter_args *ap; 2386{ 2387 int error; 2388 2389 error = fifo_specops.vop_kqfilter(ap); 2390 if (error) 2391 error = vfs_kqfilter(ap); 2392 return (error); 2393} 2394 2395/* 2396 * Return POSIX pathconf information applicable to fifos. 2397 */ 2398static int 2399ufsfifo_pathconf(ap) 2400 struct vop_pathconf_args /* { 2401 struct vnode *a_vp; 2402 int a_name; 2403 int *a_retval; 2404 } */ *ap; 2405{ 2406 2407 switch (ap->a_name) { 2408 case _PC_ACL_EXTENDED: 2409 case _PC_ACL_NFS4: 2410 case _PC_ACL_PATH_MAX: 2411 case _PC_MAC_PRESENT: 2412 return (ufs_pathconf(ap)); 2413 default: 2414 return (fifo_specops.vop_pathconf(ap)); 2415 } 2416 /* NOTREACHED */ 2417} 2418 2419/* 2420 * Return POSIX pathconf information applicable to ufs filesystems. 2421 */ 2422static int 2423ufs_pathconf(ap) 2424 struct vop_pathconf_args /* { 2425 struct vnode *a_vp; 2426 int a_name; 2427 int *a_retval; 2428 } */ *ap; 2429{ 2430 int error; 2431 2432 error = 0; 2433 switch (ap->a_name) { 2434 case _PC_LINK_MAX: 2435 *ap->a_retval = LINK_MAX; 2436 break; 2437 case _PC_NAME_MAX: 2438 *ap->a_retval = NAME_MAX; 2439 break; 2440 case _PC_PATH_MAX: 2441 *ap->a_retval = PATH_MAX; 2442 break; 2443 case _PC_PIPE_BUF: 2444 *ap->a_retval = PIPE_BUF; 2445 break; 2446 case _PC_CHOWN_RESTRICTED: 2447 *ap->a_retval = 1; 2448 break; 2449 case _PC_NO_TRUNC: 2450 *ap->a_retval = 1; 2451 break; 2452 case _PC_ACL_EXTENDED: 2453#ifdef UFS_ACL 2454 if (ap->a_vp->v_mount->mnt_flag & MNT_ACLS) 2455 *ap->a_retval = 1; 2456 else 2457 *ap->a_retval = 0; 2458#else 2459 *ap->a_retval = 0; 2460#endif 2461 break; 2462 2463 case _PC_ACL_NFS4: 2464#ifdef UFS_ACL 2465 if (ap->a_vp->v_mount->mnt_flag & MNT_NFS4ACLS) 2466 *ap->a_retval = 1; 2467 else 2468 *ap->a_retval = 0; 2469#else 2470 *ap->a_retval = 0; 2471#endif 2472 break; 2473 2474 case _PC_ACL_PATH_MAX: 2475#ifdef UFS_ACL 2476 if (ap->a_vp->v_mount->mnt_flag & (MNT_ACLS | MNT_NFS4ACLS)) 2477 *ap->a_retval = ACL_MAX_ENTRIES; 2478 else 2479 *ap->a_retval = 3; 2480#else 2481 *ap->a_retval = 3; 2482#endif 2483 break; 2484 case _PC_MAC_PRESENT: 2485#ifdef MAC 2486 if (ap->a_vp->v_mount->mnt_flag & MNT_MULTILABEL) 2487 *ap->a_retval = 1; 2488 else 2489 *ap->a_retval = 0; 2490#else 2491 *ap->a_retval = 0; 2492#endif 2493 break; 2494 case _PC_MIN_HOLE_SIZE: 2495 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize; 2496 break; 2497 case _PC_ASYNC_IO: 2498 /* _PC_ASYNC_IO should have been handled by upper layers. */ 2499 KASSERT(0, ("_PC_ASYNC_IO should not get here")); 2500 error = EINVAL; 2501 break; 2502 case _PC_PRIO_IO: 2503 *ap->a_retval = 0; 2504 break; 2505 case _PC_SYNC_IO: 2506 *ap->a_retval = 0; 2507 break; 2508 case _PC_ALLOC_SIZE_MIN: 2509 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_bsize; 2510 break; 2511 case _PC_FILESIZEBITS: 2512 *ap->a_retval = 64; 2513 break; 2514 case _PC_REC_INCR_XFER_SIZE: 2515 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize; 2516 break; 2517 case _PC_REC_MAX_XFER_SIZE: 2518 *ap->a_retval = -1; /* means ``unlimited'' */ 2519 break; 2520 case _PC_REC_MIN_XFER_SIZE: 2521 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize; 2522 break; 2523 case _PC_REC_XFER_ALIGN: 2524 *ap->a_retval = PAGE_SIZE; 2525 break; 2526 case _PC_SYMLINK_MAX: 2527 *ap->a_retval = MAXPATHLEN; 2528 break; 2529 2530 default: 2531 error = EINVAL; 2532 break; 2533 } 2534 return (error); 2535} 2536 2537/* 2538 * Initialize the vnode associated with a new inode, handle aliased 2539 * vnodes. 2540 */ 2541int 2542ufs_vinit(mntp, fifoops, vpp) 2543 struct mount *mntp; 2544 struct vop_vector *fifoops; 2545 struct vnode **vpp; 2546{ 2547 struct inode *ip; 2548 struct vnode *vp; 2549 2550 vp = *vpp; 2551 ip = VTOI(vp); 2552 vp->v_type = IFTOVT(ip->i_mode); 2553 if (vp->v_type == VFIFO) 2554 vp->v_op = fifoops; 2555 ASSERT_VOP_LOCKED(vp, "ufs_vinit"); 2556 if (ip->i_number == ROOTINO) 2557 vp->v_vflag |= VV_ROOT; 2558 *vpp = vp; 2559 return (0); 2560} 2561 2562/* 2563 * Allocate a new inode. 2564 * Vnode dvp must be locked. 2565 */ 2566static int 2567ufs_makeinode(mode, dvp, vpp, cnp) 2568 int mode; 2569 struct vnode *dvp; 2570 struct vnode **vpp; 2571 struct componentname *cnp; 2572{ 2573 struct inode *ip, *pdir; 2574 struct direct newdir; 2575 struct vnode *tvp; 2576 int error; 2577 2578 pdir = VTOI(dvp); 2579#ifdef INVARIANTS 2580 if ((cnp->cn_flags & HASBUF) == 0) 2581 panic("ufs_makeinode: no name"); 2582#endif 2583 *vpp = NULL; 2584 if ((mode & IFMT) == 0) 2585 mode |= IFREG; 2586 2587 if (VTOI(dvp)->i_effnlink < 2) 2588 panic("ufs_makeinode: Bad link count %d on parent", 2589 VTOI(dvp)->i_effnlink); 2590 error = UFS_VALLOC(dvp, mode, cnp->cn_cred, &tvp); 2591 if (error) 2592 return (error); 2593 ip = VTOI(tvp); 2594 ip->i_gid = pdir->i_gid; 2595 DIP_SET(ip, i_gid, pdir->i_gid); 2596#ifdef SUIDDIR 2597 { 2598#ifdef QUOTA 2599 struct ucred ucred, *ucp; 2600 gid_t ucred_group; 2601 ucp = cnp->cn_cred; 2602#endif 2603 /* 2604 * If we are not the owner of the directory, 2605 * and we are hacking owners here, (only do this where told to) 2606 * and we are not giving it TO root, (would subvert quotas) 2607 * then go ahead and give it to the other user. 2608 * Note that this drops off the execute bits for security. 2609 */ 2610 if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) && 2611 (pdir->i_mode & ISUID) && 2612 (pdir->i_uid != cnp->cn_cred->cr_uid) && pdir->i_uid) { 2613 ip->i_uid = pdir->i_uid; 2614 DIP_SET(ip, i_uid, ip->i_uid); 2615 mode &= ~07111; 2616#ifdef QUOTA 2617 /* 2618 * Make sure the correct user gets charged 2619 * for the space. 2620 * Quickly knock up a dummy credential for the victim. 2621 * XXX This seems to never be accessed out of our 2622 * context so a stack variable is ok. 2623 */ 2624 refcount_init(&ucred.cr_ref, 1); 2625 ucred.cr_uid = ip->i_uid; 2626 ucred.cr_ngroups = 1; 2627 ucred.cr_groups = &ucred_group; 2628 ucred.cr_groups[0] = pdir->i_gid; 2629 ucp = &ucred; 2630#endif 2631 } else { 2632 ip->i_uid = cnp->cn_cred->cr_uid; 2633 DIP_SET(ip, i_uid, ip->i_uid); 2634 } 2635 2636#ifdef QUOTA 2637 if ((error = getinoquota(ip)) || 2638 (error = chkiq(ip, 1, ucp, 0))) { 2639 if (DOINGSOFTDEP(tvp)) 2640 softdep_revert_link(pdir, ip); 2641 UFS_VFREE(tvp, ip->i_number, mode); 2642 vput(tvp); 2643 return (error); 2644 } 2645#endif 2646 } 2647#else /* !SUIDDIR */ 2648 ip->i_uid = cnp->cn_cred->cr_uid; 2649 DIP_SET(ip, i_uid, ip->i_uid); 2650#ifdef QUOTA 2651 if ((error = getinoquota(ip)) || 2652 (error = chkiq(ip, 1, cnp->cn_cred, 0))) { 2653 if (DOINGSOFTDEP(tvp)) 2654 softdep_revert_link(pdir, ip); 2655 UFS_VFREE(tvp, ip->i_number, mode); 2656 vput(tvp); 2657 return (error); 2658 } 2659#endif 2660#endif /* !SUIDDIR */ 2661 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 2662 ip->i_mode = mode; 2663 DIP_SET(ip, i_mode, mode); 2664 tvp->v_type = IFTOVT(mode); /* Rest init'd in getnewvnode(). */ 2665 ip->i_effnlink = 1; 2666 ip->i_nlink = 1; 2667 DIP_SET(ip, i_nlink, 1); 2668 if (DOINGSOFTDEP(tvp)) 2669 softdep_setup_create(VTOI(dvp), ip); 2670 if ((ip->i_mode & ISGID) && !groupmember(ip->i_gid, cnp->cn_cred) && 2671 priv_check_cred(cnp->cn_cred, PRIV_VFS_SETGID, 0)) { 2672 ip->i_mode &= ~ISGID; 2673 DIP_SET(ip, i_mode, ip->i_mode); 2674 } 2675 2676 if (cnp->cn_flags & ISWHITEOUT) { 2677 ip->i_flags |= UF_OPAQUE; 2678 DIP_SET(ip, i_flags, ip->i_flags); 2679 } 2680 2681 /* 2682 * Make sure inode goes to disk before directory entry. 2683 */ 2684 error = UFS_UPDATE(tvp, !(DOINGSOFTDEP(tvp) | DOINGASYNC(tvp))); 2685 if (error) 2686 goto bad; 2687#ifdef MAC 2688 if (dvp->v_mount->mnt_flag & MNT_MULTILABEL) { 2689 error = mac_vnode_create_extattr(cnp->cn_cred, dvp->v_mount, 2690 dvp, tvp, cnp); 2691 if (error) 2692 goto bad; 2693 } 2694#endif 2695#ifdef UFS_ACL 2696 if (dvp->v_mount->mnt_flag & MNT_ACLS) { 2697 error = ufs_do_posix1e_acl_inheritance_file(dvp, tvp, mode, 2698 cnp->cn_cred, cnp->cn_thread); 2699 if (error) 2700 goto bad; 2701 } else if (dvp->v_mount->mnt_flag & MNT_NFS4ACLS) { 2702 error = ufs_do_nfs4_acl_inheritance(dvp, tvp, mode, 2703 cnp->cn_cred, cnp->cn_thread); 2704 if (error) 2705 goto bad; 2706 } 2707#endif /* !UFS_ACL */ 2708 ufs_makedirentry(ip, cnp, &newdir); 2709 error = ufs_direnter(dvp, tvp, &newdir, cnp, NULL, 0); 2710 if (error) 2711 goto bad; 2712 *vpp = tvp; 2713 return (0); 2714 2715bad: 2716 /* 2717 * Write error occurred trying to update the inode 2718 * or the directory so must deallocate the inode. 2719 */ 2720 ip->i_effnlink = 0; 2721 ip->i_nlink = 0; 2722 DIP_SET(ip, i_nlink, 0); 2723 ip->i_flag |= IN_CHANGE; 2724 if (DOINGSOFTDEP(tvp)) 2725 softdep_revert_create(VTOI(dvp), ip); 2726 vput(tvp); 2727 return (error); 2728} 2729 2730static int 2731ufs_ioctl(struct vop_ioctl_args *ap) 2732{ 2733 2734 switch (ap->a_command) { 2735 case FIOSEEKDATA: 2736 case FIOSEEKHOLE: 2737 return (vn_bmap_seekhole(ap->a_vp, ap->a_command, 2738 (off_t *)ap->a_data, ap->a_cred)); 2739 default: 2740 return (ENOTTY); 2741 } 2742} 2743 2744/* Global vfs data structures for ufs. */ 2745struct vop_vector ufs_vnodeops = { 2746 .vop_default = &default_vnodeops, 2747 .vop_fsync = VOP_PANIC, 2748 .vop_read = VOP_PANIC, 2749 .vop_reallocblks = VOP_PANIC, 2750 .vop_write = VOP_PANIC, 2751 .vop_accessx = ufs_accessx, 2752 .vop_bmap = ufs_bmap, 2753 .vop_cachedlookup = ufs_lookup, 2754 .vop_close = ufs_close, 2755 .vop_create = ufs_create, 2756 .vop_getattr = ufs_getattr, 2757 .vop_inactive = ufs_inactive, 2758 .vop_ioctl = ufs_ioctl, 2759 .vop_link = ufs_link, 2760 .vop_lookup = vfs_cache_lookup, 2761 .vop_markatime = ufs_markatime, 2762 .vop_mkdir = ufs_mkdir, 2763 .vop_mknod = ufs_mknod, 2764 .vop_open = ufs_open, 2765 .vop_pathconf = ufs_pathconf, 2766 .vop_poll = vop_stdpoll, 2767 .vop_print = ufs_print, 2768 .vop_readdir = ufs_readdir, 2769 .vop_readlink = ufs_readlink, 2770 .vop_reclaim = ufs_reclaim, 2771 .vop_remove = ufs_remove, 2772 .vop_rename = ufs_rename, 2773 .vop_rmdir = ufs_rmdir, 2774 .vop_setattr = ufs_setattr, 2775#ifdef MAC 2776 .vop_setlabel = vop_stdsetlabel_ea, 2777#endif 2778 .vop_strategy = ufs_strategy, 2779 .vop_symlink = ufs_symlink, 2780 .vop_whiteout = ufs_whiteout, 2781#ifdef UFS_EXTATTR 2782 .vop_getextattr = ufs_getextattr, 2783 .vop_deleteextattr = ufs_deleteextattr, 2784 .vop_setextattr = ufs_setextattr, 2785#endif 2786#ifdef UFS_ACL 2787 .vop_getacl = ufs_getacl, 2788 .vop_setacl = ufs_setacl, 2789 .vop_aclcheck = ufs_aclcheck, 2790#endif 2791}; 2792 2793struct vop_vector ufs_fifoops = { 2794 .vop_default = &fifo_specops, 2795 .vop_fsync = VOP_PANIC, 2796 .vop_accessx = ufs_accessx, 2797 .vop_close = ufsfifo_close, 2798 .vop_getattr = ufs_getattr, 2799 .vop_inactive = ufs_inactive, 2800 .vop_kqfilter = ufsfifo_kqfilter, 2801 .vop_markatime = ufs_markatime, 2802 .vop_pathconf = ufsfifo_pathconf, 2803 .vop_print = ufs_print, 2804 .vop_read = VOP_PANIC, 2805 .vop_reclaim = ufs_reclaim, 2806 .vop_setattr = ufs_setattr, 2807#ifdef MAC 2808 .vop_setlabel = vop_stdsetlabel_ea, 2809#endif 2810 .vop_write = VOP_PANIC, 2811#ifdef UFS_EXTATTR 2812 .vop_getextattr = ufs_getextattr, 2813 .vop_deleteextattr = ufs_deleteextattr, 2814 .vop_setextattr = ufs_setextattr, 2815#endif 2816#ifdef UFS_ACL 2817 .vop_getacl = ufs_getacl, 2818 .vop_setacl = ufs_setacl, 2819 .vop_aclcheck = ufs_aclcheck, 2820#endif 2821}; 2822