1 /*- 2 * modified for EXT2FS support in Lites 1.1 3 * 4 * Aug 1995, Godmar Back (gback@cs.utah.edu) 5 * University of Utah, Department of Computer Science 6 */ 7 /*- 8 * SPDX-License-Identifier: BSD-3-Clause 9 * 10 * Copyright (c) 1982, 1986, 1989, 1993 11 * The Regents of the University of California. All rights reserved. 12 * (c) UNIX System Laboratories, Inc. 13 * All or some portions of this file are derived from material licensed 14 * to the University of California by American Telephone and Telegraph 15 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 16 * the permission of UNIX System Laboratories, Inc. 17 * 18 * Redistribution and use in source and binary forms, with or without 19 * modification, are permitted provided that the following conditions 20 * are met: 21 * 1. Redistributions of source code must retain the above copyright 22 * notice, this list of conditions and the following disclaimer. 23 * 2. Redistributions in binary form must reproduce the above copyright 24 * notice, this list of conditions and the following disclaimer in the 25 * documentation and/or other materials provided with the distribution. 26 * 3. Neither the name of the University nor the names of its contributors 27 * may be used to endorse or promote products derived from this software 28 * without specific prior written permission. 29 * 30 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 31 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 32 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 33 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 34 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 35 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 36 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 37 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 38 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 39 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 40 * SUCH DAMAGE. 41 * 42 * @(#)ufs_vnops.c 8.7 (Berkeley) 2/3/94 43 * @(#)ufs_vnops.c 8.27 (Berkeley) 5/27/95 44 * $FreeBSD$ 45 */ 46 47 #include "opt_suiddir.h" 48 49 #include <sys/param.h> 50 #include <sys/systm.h> 51 #include <sys/kernel.h> 52 #include <sys/fcntl.h> 53 #include <sys/filio.h> 54 #include <sys/limits.h> 55 #include <sys/sdt.h> 56 #include <sys/stat.h> 57 #include <sys/bio.h> 58 #include <sys/buf.h> 59 #include <sys/endian.h> 60 #include <sys/priv.h> 61 #include <sys/rwlock.h> 62 #include <sys/mount.h> 63 #include <sys/unistd.h> 64 #include <sys/time.h> 65 #include <sys/vnode.h> 66 #include <sys/namei.h> 67 #include <sys/lockf.h> 68 #include <sys/event.h> 69 #include <sys/conf.h> 70 #include <sys/file.h> 71 #include <sys/extattr.h> 72 #include <sys/vmmeter.h> 73 74 #include <vm/vm.h> 75 #include <vm/vm_param.h> 76 #include <vm/vm_extern.h> 77 #include <vm/vm_object.h> 78 #include <vm/vm_page.h> 79 #include <vm/vm_pager.h> 80 #include <vm/vnode_pager.h> 81 82 #include "opt_directio.h" 83 84 #include <ufs/ufs/dir.h> 85 86 #include <fs/ext2fs/fs.h> 87 #include <fs/ext2fs/inode.h> 88 #include <fs/ext2fs/ext2_acl.h> 89 #include <fs/ext2fs/ext2fs.h> 90 #include <fs/ext2fs/ext2_extern.h> 91 #include <fs/ext2fs/ext2_dinode.h> 92 #include <fs/ext2fs/ext2_dir.h> 93 #include <fs/ext2fs/ext2_mount.h> 94 #include <fs/ext2fs/ext2_extattr.h> 95 #include <fs/ext2fs/ext2_extents.h> 96 97 SDT_PROVIDER_DECLARE(ext2fs); 98 /* 99 * ext2fs trace probe: 100 * arg0: verbosity. Higher numbers give more verbose messages 101 * arg1: Textual message 102 */ 103 SDT_PROBE_DEFINE2(ext2fs, , vnops, trace, "int", "char*"); 104 105 static int ext2_makeinode(int mode, struct vnode *, struct vnode **, struct componentname *); 106 static void ext2_itimes_locked(struct vnode *); 107 108 static vop_access_t ext2_access; 109 static int ext2_chmod(struct vnode *, int, struct ucred *, struct thread *); 110 static int ext2_chown(struct vnode *, uid_t, gid_t, struct ucred *, 111 struct thread *); 112 static vop_close_t ext2_close; 113 static vop_create_t ext2_create; 114 static vop_fsync_t ext2_fsync; 115 static vop_getattr_t ext2_getattr; 116 static vop_ioctl_t ext2_ioctl; 117 static vop_link_t ext2_link; 118 static vop_mkdir_t ext2_mkdir; 119 static vop_mknod_t ext2_mknod; 120 static vop_open_t ext2_open; 121 static vop_pathconf_t ext2_pathconf; 122 static vop_print_t ext2_print; 123 static vop_read_t ext2_read; 124 static vop_readlink_t ext2_readlink; 125 static vop_remove_t ext2_remove; 126 static vop_rename_t ext2_rename; 127 static vop_rmdir_t ext2_rmdir; 128 static vop_setattr_t ext2_setattr; 129 static vop_strategy_t ext2_strategy; 130 static vop_symlink_t ext2_symlink; 131 static vop_write_t ext2_write; 132 static vop_deleteextattr_t ext2_deleteextattr; 133 static vop_getextattr_t ext2_getextattr; 134 static vop_listextattr_t ext2_listextattr; 135 static vop_setextattr_t ext2_setextattr; 136 static vop_vptofh_t ext2_vptofh; 137 static vop_close_t ext2fifo_close; 138 static vop_kqfilter_t ext2fifo_kqfilter; 139 140 /* Global vfs data structures for ext2. */ 141 struct vop_vector ext2_vnodeops = { 142 .vop_default = &default_vnodeops, 143 .vop_access = ext2_access, 144 .vop_bmap = ext2_bmap, 145 .vop_cachedlookup = ext2_lookup, 146 .vop_close = ext2_close, 147 .vop_create = ext2_create, 148 .vop_fsync = ext2_fsync, 149 .vop_getpages = vnode_pager_local_getpages, 150 .vop_getpages_async = vnode_pager_local_getpages_async, 151 .vop_getattr = ext2_getattr, 152 .vop_inactive = ext2_inactive, 153 .vop_ioctl = ext2_ioctl, 154 .vop_link = ext2_link, 155 .vop_lookup = vfs_cache_lookup, 156 .vop_mkdir = ext2_mkdir, 157 .vop_mknod = ext2_mknod, 158 .vop_open = ext2_open, 159 .vop_pathconf = ext2_pathconf, 160 .vop_poll = vop_stdpoll, 161 .vop_print = ext2_print, 162 .vop_read = ext2_read, 163 .vop_readdir = ext2_readdir, 164 .vop_readlink = ext2_readlink, 165 .vop_reallocblks = ext2_reallocblks, 166 .vop_reclaim = ext2_reclaim, 167 .vop_remove = ext2_remove, 168 .vop_rename = ext2_rename, 169 .vop_rmdir = ext2_rmdir, 170 .vop_setattr = ext2_setattr, 171 .vop_strategy = ext2_strategy, 172 .vop_symlink = ext2_symlink, 173 .vop_write = ext2_write, 174 .vop_deleteextattr = ext2_deleteextattr, 175 .vop_getextattr = ext2_getextattr, 176 .vop_listextattr = ext2_listextattr, 177 .vop_setextattr = ext2_setextattr, 178 #ifdef UFS_ACL 179 .vop_getacl = ext2_getacl, 180 .vop_setacl = ext2_setacl, 181 .vop_aclcheck = ext2_aclcheck, 182 #endif /* UFS_ACL */ 183 .vop_vptofh = ext2_vptofh, 184 }; 185 186 struct vop_vector ext2_fifoops = { 187 .vop_default = &fifo_specops, 188 .vop_access = ext2_access, 189 .vop_close = ext2fifo_close, 190 .vop_fsync = ext2_fsync, 191 .vop_getattr = ext2_getattr, 192 .vop_inactive = ext2_inactive, 193 .vop_kqfilter = ext2fifo_kqfilter, 194 .vop_pathconf = ext2_pathconf, 195 .vop_print = ext2_print, 196 .vop_read = VOP_PANIC, 197 .vop_reclaim = ext2_reclaim, 198 .vop_setattr = ext2_setattr, 199 .vop_write = VOP_PANIC, 200 .vop_vptofh = ext2_vptofh, 201 }; 202 203 /* 204 * A virgin directory (no blushing please). 205 * Note that the type and namlen fields are reversed relative to ext2. 206 * Also, we don't use `struct odirtemplate', since it would just cause 207 * endianness problems. 208 */ 209 static struct dirtemplate mastertemplate = { 210 0, 12, 1, EXT2_FT_DIR, ".", 211 0, DIRBLKSIZ - 12, 2, EXT2_FT_DIR, ".." 212 }; 213 static struct dirtemplate omastertemplate = { 214 0, 12, 1, EXT2_FT_UNKNOWN, ".", 215 0, DIRBLKSIZ - 12, 2, EXT2_FT_UNKNOWN, ".." 216 }; 217 218 static void 219 ext2_itimes_locked(struct vnode *vp) 220 { 221 struct inode *ip; 222 struct timespec ts; 223 224 ASSERT_VI_LOCKED(vp, __func__); 225 226 ip = VTOI(vp); 227 if ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE)) == 0) 228 return; 229 if ((vp->v_type == VBLK || vp->v_type == VCHR)) 230 ip->i_flag |= IN_LAZYMOD; 231 else 232 ip->i_flag |= IN_MODIFIED; 233 if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 234 vfs_timestamp(&ts); 235 if (ip->i_flag & IN_ACCESS) { 236 ip->i_atime = ts.tv_sec; 237 ip->i_atimensec = ts.tv_nsec; 238 } 239 if (ip->i_flag & IN_UPDATE) { 240 ip->i_mtime = ts.tv_sec; 241 ip->i_mtimensec = ts.tv_nsec; 242 ip->i_modrev++; 243 } 244 if (ip->i_flag & IN_CHANGE) { 245 ip->i_ctime = ts.tv_sec; 246 ip->i_ctimensec = ts.tv_nsec; 247 } 248 } 249 ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_UPDATE); 250 } 251 252 void 253 ext2_itimes(struct vnode *vp) 254 { 255 256 VI_LOCK(vp); 257 ext2_itimes_locked(vp); 258 VI_UNLOCK(vp); 259 } 260 261 /* 262 * Create a regular file 263 */ 264 static int 265 ext2_create(struct vop_create_args *ap) 266 { 267 int error; 268 269 error = 270 ext2_makeinode(MAKEIMODE(ap->a_vap->va_type, ap->a_vap->va_mode), 271 ap->a_dvp, ap->a_vpp, ap->a_cnp); 272 if (error != 0) 273 return (error); 274 if ((ap->a_cnp->cn_flags & MAKEENTRY) != 0) 275 cache_enter(ap->a_dvp, *ap->a_vpp, ap->a_cnp); 276 return (0); 277 } 278 279 static int 280 ext2_open(struct vop_open_args *ap) 281 { 282 283 if (ap->a_vp->v_type == VBLK || ap->a_vp->v_type == VCHR) 284 return (EOPNOTSUPP); 285 286 /* 287 * Files marked append-only must be opened for appending. 288 */ 289 if ((VTOI(ap->a_vp)->i_flags & APPEND) && 290 (ap->a_mode & (FWRITE | O_APPEND)) == FWRITE) 291 return (EPERM); 292 293 vnode_create_vobject(ap->a_vp, VTOI(ap->a_vp)->i_size, ap->a_td); 294 295 return (0); 296 } 297 298 /* 299 * Close called. 300 * 301 * Update the times on the inode. 302 */ 303 static int 304 ext2_close(struct vop_close_args *ap) 305 { 306 struct vnode *vp = ap->a_vp; 307 308 VI_LOCK(vp); 309 if (vp->v_usecount > 1) 310 ext2_itimes_locked(vp); 311 VI_UNLOCK(vp); 312 return (0); 313 } 314 315 static int 316 ext2_access(struct vop_access_args *ap) 317 { 318 struct vnode *vp = ap->a_vp; 319 struct inode *ip = VTOI(vp); 320 accmode_t accmode = ap->a_accmode; 321 int error; 322 323 if (vp->v_type == VBLK || vp->v_type == VCHR) 324 return (EOPNOTSUPP); 325 326 /* 327 * Disallow write attempts on read-only file systems; 328 * unless the file is a socket, fifo, or a block or 329 * character device resident on the file system. 330 */ 331 if (accmode & VWRITE) { 332 switch (vp->v_type) { 333 case VDIR: 334 case VLNK: 335 case VREG: 336 if (vp->v_mount->mnt_flag & MNT_RDONLY) 337 return (EROFS); 338 break; 339 default: 340 break; 341 } 342 } 343 344 /* If immutable bit set, nobody gets to write it. */ 345 if ((accmode & VWRITE) && (ip->i_flags & (SF_IMMUTABLE | SF_SNAPSHOT))) 346 return (EPERM); 347 348 error = vaccess(vp->v_type, ip->i_mode, ip->i_uid, ip->i_gid, 349 ap->a_accmode, ap->a_cred, NULL); 350 return (error); 351 } 352 353 static int 354 ext2_getattr(struct vop_getattr_args *ap) 355 { 356 struct vnode *vp = ap->a_vp; 357 struct inode *ip = VTOI(vp); 358 struct vattr *vap = ap->a_vap; 359 360 ext2_itimes(vp); 361 /* 362 * Copy from inode table 363 */ 364 vap->va_fsid = dev2udev(ip->i_devvp->v_rdev); 365 vap->va_fileid = ip->i_number; 366 vap->va_mode = ip->i_mode & ~IFMT; 367 vap->va_nlink = ip->i_nlink; 368 vap->va_uid = ip->i_uid; 369 vap->va_gid = ip->i_gid; 370 vap->va_rdev = ip->i_rdev; 371 vap->va_size = ip->i_size; 372 vap->va_atime.tv_sec = ip->i_atime; 373 vap->va_atime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_atimensec : 0; 374 vap->va_mtime.tv_sec = ip->i_mtime; 375 vap->va_mtime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_mtimensec : 0; 376 vap->va_ctime.tv_sec = ip->i_ctime; 377 vap->va_ctime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_ctimensec : 0; 378 if E2DI_HAS_XTIME(ip) { 379 vap->va_birthtime.tv_sec = ip->i_birthtime; 380 vap->va_birthtime.tv_nsec = ip->i_birthnsec; 381 } 382 vap->va_flags = ip->i_flags; 383 vap->va_gen = ip->i_gen; 384 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize; 385 vap->va_bytes = dbtob((u_quad_t)ip->i_blocks); 386 vap->va_type = IFTOVT(ip->i_mode); 387 vap->va_filerev = ip->i_modrev; 388 return (0); 389 } 390 391 /* 392 * Set attribute vnode op. called from several syscalls 393 */ 394 static int 395 ext2_setattr(struct vop_setattr_args *ap) 396 { 397 struct vattr *vap = ap->a_vap; 398 struct vnode *vp = ap->a_vp; 399 struct inode *ip = VTOI(vp); 400 struct ucred *cred = ap->a_cred; 401 struct thread *td = curthread; 402 int error; 403 404 /* 405 * Check for unsettable attributes. 406 */ 407 if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) || 408 (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) || 409 (vap->va_blocksize != VNOVAL) || (vap->va_rdev != VNOVAL) || 410 ((int)vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) { 411 return (EINVAL); 412 } 413 if (vap->va_flags != VNOVAL) { 414 /* Disallow flags not supported by ext2fs. */ 415 if (vap->va_flags & ~(SF_APPEND | SF_IMMUTABLE | UF_NODUMP)) 416 return (EOPNOTSUPP); 417 418 if (vp->v_mount->mnt_flag & MNT_RDONLY) 419 return (EROFS); 420 /* 421 * Callers may only modify the file flags on objects they 422 * have VADMIN rights for. 423 */ 424 if ((error = VOP_ACCESS(vp, VADMIN, cred, td))) 425 return (error); 426 /* 427 * Unprivileged processes and privileged processes in 428 * jail() are not permitted to unset system flags, or 429 * modify flags if any system flags are set. 430 * Privileged non-jail processes may not modify system flags 431 * if securelevel > 0 and any existing system flags are set. 432 */ 433 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS)) { 434 if (ip->i_flags & (SF_IMMUTABLE | SF_APPEND)) { 435 error = securelevel_gt(cred, 0); 436 if (error) 437 return (error); 438 } 439 } else { 440 if (ip->i_flags & (SF_IMMUTABLE | SF_APPEND) || 441 ((vap->va_flags ^ ip->i_flags) & SF_SETTABLE)) 442 return (EPERM); 443 } 444 ip->i_flags = vap->va_flags; 445 ip->i_flag |= IN_CHANGE; 446 if (ip->i_flags & (IMMUTABLE | APPEND)) 447 return (0); 448 } 449 if (ip->i_flags & (IMMUTABLE | APPEND)) 450 return (EPERM); 451 /* 452 * Go through the fields and update iff not VNOVAL. 453 */ 454 if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) { 455 if (vp->v_mount->mnt_flag & MNT_RDONLY) 456 return (EROFS); 457 if ((error = ext2_chown(vp, vap->va_uid, vap->va_gid, cred, 458 td)) != 0) 459 return (error); 460 } 461 if (vap->va_size != VNOVAL) { 462 /* 463 * Disallow write attempts on read-only file systems; 464 * unless the file is a socket, fifo, or a block or 465 * character device resident on the file system. 466 */ 467 switch (vp->v_type) { 468 case VDIR: 469 return (EISDIR); 470 case VLNK: 471 case VREG: 472 if (vp->v_mount->mnt_flag & MNT_RDONLY) 473 return (EROFS); 474 break; 475 default: 476 break; 477 } 478 if ((error = ext2_truncate(vp, vap->va_size, 0, cred, td)) != 0) 479 return (error); 480 } 481 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) { 482 if (vp->v_mount->mnt_flag & MNT_RDONLY) 483 return (EROFS); 484 /* 485 * From utimes(2): 486 * If times is NULL, ... The caller must be the owner of 487 * the file, have permission to write the file, or be the 488 * super-user. 489 * If times is non-NULL, ... The caller must be the owner of 490 * the file or be the super-user. 491 */ 492 if ((error = VOP_ACCESS(vp, VADMIN, cred, td)) && 493 ((vap->va_vaflags & VA_UTIMES_NULL) == 0 || 494 (error = VOP_ACCESS(vp, VWRITE, cred, td)))) 495 return (error); 496 ip->i_flag |= IN_CHANGE | IN_MODIFIED; 497 if (vap->va_atime.tv_sec != VNOVAL) { 498 ip->i_flag &= ~IN_ACCESS; 499 ip->i_atime = vap->va_atime.tv_sec; 500 ip->i_atimensec = vap->va_atime.tv_nsec; 501 } 502 if (vap->va_mtime.tv_sec != VNOVAL) { 503 ip->i_flag &= ~IN_UPDATE; 504 ip->i_mtime = vap->va_mtime.tv_sec; 505 ip->i_mtimensec = vap->va_mtime.tv_nsec; 506 } 507 ip->i_birthtime = vap->va_birthtime.tv_sec; 508 ip->i_birthnsec = vap->va_birthtime.tv_nsec; 509 error = ext2_update(vp, 0); 510 if (error) 511 return (error); 512 } 513 error = 0; 514 if (vap->va_mode != (mode_t)VNOVAL) { 515 if (vp->v_mount->mnt_flag & MNT_RDONLY) 516 return (EROFS); 517 error = ext2_chmod(vp, (int)vap->va_mode, cred, td); 518 } 519 return (error); 520 } 521 522 /* 523 * Change the mode on a file. 524 * Inode must be locked before calling. 525 */ 526 static int 527 ext2_chmod(struct vnode *vp, int mode, struct ucred *cred, struct thread *td) 528 { 529 struct inode *ip = VTOI(vp); 530 int error; 531 532 /* 533 * To modify the permissions on a file, must possess VADMIN 534 * for that file. 535 */ 536 if ((error = VOP_ACCESS(vp, VADMIN, cred, td))) 537 return (error); 538 /* 539 * Privileged processes may set the sticky bit on non-directories, 540 * as well as set the setgid bit on a file with a group that the 541 * process is not a member of. 542 */ 543 if (vp->v_type != VDIR && (mode & S_ISTXT)) { 544 error = priv_check_cred(cred, PRIV_VFS_STICKYFILE); 545 if (error) 546 return (EFTYPE); 547 } 548 if (!groupmember(ip->i_gid, cred) && (mode & ISGID)) { 549 error = priv_check_cred(cred, PRIV_VFS_SETGID); 550 if (error) 551 return (error); 552 } 553 ip->i_mode &= ~ALLPERMS; 554 ip->i_mode |= (mode & ALLPERMS); 555 ip->i_flag |= IN_CHANGE; 556 return (0); 557 } 558 559 /* 560 * Perform chown operation on inode ip; 561 * inode must be locked prior to call. 562 */ 563 static int 564 ext2_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred, 565 struct thread *td) 566 { 567 struct inode *ip = VTOI(vp); 568 uid_t ouid; 569 gid_t ogid; 570 int error = 0; 571 572 if (uid == (uid_t)VNOVAL) 573 uid = ip->i_uid; 574 if (gid == (gid_t)VNOVAL) 575 gid = ip->i_gid; 576 /* 577 * To modify the ownership of a file, must possess VADMIN 578 * for that file. 579 */ 580 if ((error = VOP_ACCESS(vp, VADMIN, cred, td))) 581 return (error); 582 /* 583 * To change the owner of a file, or change the group of a file 584 * to a group of which we are not a member, the caller must 585 * have privilege. 586 */ 587 if (uid != ip->i_uid || (gid != ip->i_gid && 588 !groupmember(gid, cred))) { 589 error = priv_check_cred(cred, PRIV_VFS_CHOWN); 590 if (error) 591 return (error); 592 } 593 ogid = ip->i_gid; 594 ouid = ip->i_uid; 595 ip->i_gid = gid; 596 ip->i_uid = uid; 597 ip->i_flag |= IN_CHANGE; 598 if ((ip->i_mode & (ISUID | ISGID)) && (ouid != uid || ogid != gid)) { 599 if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID) != 0) 600 ip->i_mode &= ~(ISUID | ISGID); 601 } 602 return (0); 603 } 604 605 /* 606 * Synch an open file. 607 */ 608 /* ARGSUSED */ 609 static int 610 ext2_fsync(struct vop_fsync_args *ap) 611 { 612 /* 613 * Flush all dirty buffers associated with a vnode. 614 */ 615 616 vop_stdfsync(ap); 617 618 return (ext2_update(ap->a_vp, ap->a_waitfor == MNT_WAIT)); 619 } 620 621 /* 622 * Mknod vnode call 623 */ 624 /* ARGSUSED */ 625 static int 626 ext2_mknod(struct vop_mknod_args *ap) 627 { 628 struct vattr *vap = ap->a_vap; 629 struct vnode **vpp = ap->a_vpp; 630 struct inode *ip; 631 ino_t ino; 632 int error; 633 634 error = ext2_makeinode(MAKEIMODE(vap->va_type, vap->va_mode), 635 ap->a_dvp, vpp, ap->a_cnp); 636 if (error) 637 return (error); 638 ip = VTOI(*vpp); 639 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 640 if (vap->va_rdev != VNOVAL) { 641 /* 642 * Want to be able to use this to make badblock 643 * inodes, so don't truncate the dev number. 644 */ 645 if (!(ip->i_flag & IN_E4EXTENTS)) 646 ip->i_rdev = vap->va_rdev; 647 } 648 /* 649 * Remove inode, then reload it through VFS_VGET so it is 650 * checked to see if it is an alias of an existing entry in 651 * the inode cache. XXX I don't believe this is necessary now. 652 */ 653 (*vpp)->v_type = VNON; 654 ino = ip->i_number; /* Save this before vgone() invalidates ip. */ 655 vgone(*vpp); 656 vput(*vpp); 657 error = VFS_VGET(ap->a_dvp->v_mount, ino, LK_EXCLUSIVE, vpp); 658 if (error) { 659 *vpp = NULL; 660 return (error); 661 } 662 return (0); 663 } 664 665 static int 666 ext2_remove(struct vop_remove_args *ap) 667 { 668 struct inode *ip; 669 struct vnode *vp = ap->a_vp; 670 struct vnode *dvp = ap->a_dvp; 671 int error; 672 673 ip = VTOI(vp); 674 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) || 675 (VTOI(dvp)->i_flags & APPEND)) { 676 error = EPERM; 677 goto out; 678 } 679 error = ext2_dirremove(dvp, ap->a_cnp); 680 if (error == 0) { 681 ip->i_nlink--; 682 ip->i_flag |= IN_CHANGE; 683 } 684 out: 685 return (error); 686 } 687 688 /* 689 * link vnode call 690 */ 691 static int 692 ext2_link(struct vop_link_args *ap) 693 { 694 struct vnode *vp = ap->a_vp; 695 struct vnode *tdvp = ap->a_tdvp; 696 struct componentname *cnp = ap->a_cnp; 697 struct inode *ip; 698 int error; 699 700 #ifdef INVARIANTS 701 if ((cnp->cn_flags & HASBUF) == 0) 702 panic("ext2_link: no name"); 703 #endif 704 ip = VTOI(vp); 705 if ((nlink_t)ip->i_nlink >= EXT4_LINK_MAX) { 706 error = EMLINK; 707 goto out; 708 } 709 if (ip->i_flags & (IMMUTABLE | APPEND)) { 710 error = EPERM; 711 goto out; 712 } 713 ip->i_nlink++; 714 ip->i_flag |= IN_CHANGE; 715 error = ext2_update(vp, !DOINGASYNC(vp)); 716 if (!error) 717 error = ext2_direnter(ip, tdvp, cnp); 718 if (error) { 719 ip->i_nlink--; 720 ip->i_flag |= IN_CHANGE; 721 } 722 out: 723 return (error); 724 } 725 726 static int 727 ext2_inc_nlink(struct inode *ip) 728 { 729 730 ip->i_nlink++; 731 732 if (S_ISDIR(ip->i_mode) && 733 EXT2_HAS_RO_COMPAT_FEATURE(ip->i_e2fs, EXT2F_ROCOMPAT_DIR_NLINK) && 734 ip->i_nlink > 1) { 735 if (ip->i_nlink >= EXT4_LINK_MAX || ip->i_nlink == 2) 736 ip->i_nlink = 1; 737 } else if (ip->i_nlink > EXT4_LINK_MAX) { 738 ip->i_nlink--; 739 return (EMLINK); 740 } 741 742 return (0); 743 } 744 745 static void 746 ext2_dec_nlink(struct inode *ip) 747 { 748 749 if (!S_ISDIR(ip->i_mode) || ip->i_nlink > 2) 750 ip->i_nlink--; 751 } 752 753 /* 754 * Rename system call. 755 * rename("foo", "bar"); 756 * is essentially 757 * unlink("bar"); 758 * link("foo", "bar"); 759 * unlink("foo"); 760 * but ``atomically''. Can't do full commit without saving state in the 761 * inode on disk which isn't feasible at this time. Best we can do is 762 * always guarantee the target exists. 763 * 764 * Basic algorithm is: 765 * 766 * 1) Bump link count on source while we're linking it to the 767 * target. This also ensure the inode won't be deleted out 768 * from underneath us while we work (it may be truncated by 769 * a concurrent `trunc' or `open' for creation). 770 * 2) Link source to destination. If destination already exists, 771 * delete it first. 772 * 3) Unlink source reference to inode if still around. If a 773 * directory was moved and the parent of the destination 774 * is different from the source, patch the ".." entry in the 775 * directory. 776 */ 777 static int 778 ext2_rename(struct vop_rename_args *ap) 779 { 780 struct vnode *tvp = ap->a_tvp; 781 struct vnode *tdvp = ap->a_tdvp; 782 struct vnode *fvp = ap->a_fvp; 783 struct vnode *fdvp = ap->a_fdvp; 784 struct componentname *tcnp = ap->a_tcnp; 785 struct componentname *fcnp = ap->a_fcnp; 786 struct inode *ip, *xp, *dp; 787 struct dirtemplate *dirbuf; 788 int doingdirectory = 0, oldparent = 0, newparent = 0; 789 int error = 0; 790 u_char namlen; 791 792 #ifdef INVARIANTS 793 if ((tcnp->cn_flags & HASBUF) == 0 || 794 (fcnp->cn_flags & HASBUF) == 0) 795 panic("ext2_rename: no name"); 796 #endif 797 /* 798 * Check for cross-device rename. 799 */ 800 if ((fvp->v_mount != tdvp->v_mount) || 801 (tvp && (fvp->v_mount != tvp->v_mount))) { 802 error = EXDEV; 803 abortit: 804 if (tdvp == tvp) 805 vrele(tdvp); 806 else 807 vput(tdvp); 808 if (tvp) 809 vput(tvp); 810 vrele(fdvp); 811 vrele(fvp); 812 return (error); 813 } 814 815 if (tvp && ((VTOI(tvp)->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) || 816 (VTOI(tdvp)->i_flags & APPEND))) { 817 error = EPERM; 818 goto abortit; 819 } 820 821 /* 822 * Renaming a file to itself has no effect. The upper layers should 823 * not call us in that case. Temporarily just warn if they do. 824 */ 825 if (fvp == tvp) { 826 SDT_PROBE2(ext2fs, , vnops, trace, 1, 827 "rename: fvp == tvp (can't happen)"); 828 error = 0; 829 goto abortit; 830 } 831 832 if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0) 833 goto abortit; 834 dp = VTOI(fdvp); 835 ip = VTOI(fvp); 836 if (ip->i_nlink >= EXT4_LINK_MAX && 837 !EXT2_HAS_RO_COMPAT_FEATURE(ip->i_e2fs, EXT2F_ROCOMPAT_DIR_NLINK)) { 838 VOP_UNLOCK(fvp, 0); 839 error = EMLINK; 840 goto abortit; 841 } 842 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) 843 || (dp->i_flags & APPEND)) { 844 VOP_UNLOCK(fvp, 0); 845 error = EPERM; 846 goto abortit; 847 } 848 if ((ip->i_mode & IFMT) == IFDIR) { 849 /* 850 * Avoid ".", "..", and aliases of "." for obvious reasons. 851 */ 852 if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') || 853 dp == ip || (fcnp->cn_flags | tcnp->cn_flags) & ISDOTDOT || 854 (ip->i_flag & IN_RENAME)) { 855 VOP_UNLOCK(fvp, 0); 856 error = EINVAL; 857 goto abortit; 858 } 859 ip->i_flag |= IN_RENAME; 860 oldparent = dp->i_number; 861 doingdirectory++; 862 } 863 vrele(fdvp); 864 865 /* 866 * When the target exists, both the directory 867 * and target vnodes are returned locked. 868 */ 869 dp = VTOI(tdvp); 870 xp = NULL; 871 if (tvp) 872 xp = VTOI(tvp); 873 874 /* 875 * 1) Bump link count while we're moving stuff 876 * around. If we crash somewhere before 877 * completing our work, the link count 878 * may be wrong, but correctable. 879 */ 880 ext2_inc_nlink(ip); 881 ip->i_flag |= IN_CHANGE; 882 if ((error = ext2_update(fvp, !DOINGASYNC(fvp))) != 0) { 883 VOP_UNLOCK(fvp, 0); 884 goto bad; 885 } 886 887 /* 888 * If ".." must be changed (ie the directory gets a new 889 * parent) then the source directory must not be in the 890 * directory hierarchy above the target, as this would 891 * orphan everything below the source directory. Also 892 * the user must have write permission in the source so 893 * as to be able to change "..". We must repeat the call 894 * to namei, as the parent directory is unlocked by the 895 * call to checkpath(). 896 */ 897 error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, tcnp->cn_thread); 898 VOP_UNLOCK(fvp, 0); 899 if (oldparent != dp->i_number) 900 newparent = dp->i_number; 901 if (doingdirectory && newparent) { 902 if (error) /* write access check above */ 903 goto bad; 904 if (xp != NULL) 905 vput(tvp); 906 error = ext2_checkpath(ip, dp, tcnp->cn_cred); 907 if (error) 908 goto out; 909 VREF(tdvp); 910 error = relookup(tdvp, &tvp, tcnp); 911 if (error) 912 goto out; 913 vrele(tdvp); 914 dp = VTOI(tdvp); 915 xp = NULL; 916 if (tvp) 917 xp = VTOI(tvp); 918 } 919 /* 920 * 2) If target doesn't exist, link the target 921 * to the source and unlink the source. 922 * Otherwise, rewrite the target directory 923 * entry to reference the source inode and 924 * expunge the original entry's existence. 925 */ 926 if (xp == NULL) { 927 if (dp->i_devvp != ip->i_devvp) 928 panic("ext2_rename: EXDEV"); 929 /* 930 * Account for ".." in new directory. 931 * When source and destination have the same 932 * parent we don't fool with the link count. 933 */ 934 if (doingdirectory && newparent) { 935 error = ext2_inc_nlink(dp); 936 if (error) 937 goto bad; 938 939 dp->i_flag |= IN_CHANGE; 940 error = ext2_update(tdvp, !DOINGASYNC(tdvp)); 941 if (error) 942 goto bad; 943 } 944 error = ext2_direnter(ip, tdvp, tcnp); 945 if (error) { 946 if (doingdirectory && newparent) { 947 ext2_dec_nlink(dp); 948 dp->i_flag |= IN_CHANGE; 949 (void)ext2_update(tdvp, 1); 950 } 951 goto bad; 952 } 953 vput(tdvp); 954 } else { 955 if (xp->i_devvp != dp->i_devvp || xp->i_devvp != ip->i_devvp) 956 panic("ext2_rename: EXDEV"); 957 /* 958 * Short circuit rename(foo, foo). 959 */ 960 if (xp->i_number == ip->i_number) 961 panic("ext2_rename: same file"); 962 /* 963 * If the parent directory is "sticky", then the user must 964 * own the parent directory, or the destination of the rename, 965 * otherwise the destination may not be changed (except by 966 * root). This implements append-only directories. 967 */ 968 if ((dp->i_mode & S_ISTXT) && tcnp->cn_cred->cr_uid != 0 && 969 tcnp->cn_cred->cr_uid != dp->i_uid && 970 xp->i_uid != tcnp->cn_cred->cr_uid) { 971 error = EPERM; 972 goto bad; 973 } 974 /* 975 * Target must be empty if a directory and have no links 976 * to it. Also, ensure source and target are compatible 977 * (both directories, or both not directories). 978 */ 979 if ((xp->i_mode & IFMT) == IFDIR) { 980 if (!ext2_dirempty(xp, dp->i_number, tcnp->cn_cred)) { 981 error = ENOTEMPTY; 982 goto bad; 983 } 984 if (!doingdirectory) { 985 error = ENOTDIR; 986 goto bad; 987 } 988 cache_purge(tdvp); 989 } else if (doingdirectory) { 990 error = EISDIR; 991 goto bad; 992 } 993 error = ext2_dirrewrite(dp, ip, tcnp); 994 if (error) 995 goto bad; 996 /* 997 * If the target directory is in the same 998 * directory as the source directory, 999 * decrement the link count on the parent 1000 * of the target directory. 1001 */ 1002 if (doingdirectory && !newparent) { 1003 ext2_dec_nlink(dp); 1004 dp->i_flag |= IN_CHANGE; 1005 } 1006 vput(tdvp); 1007 /* 1008 * Adjust the link count of the target to 1009 * reflect the dirrewrite above. If this is 1010 * a directory it is empty and there are 1011 * no links to it, so we can squash the inode and 1012 * any space associated with it. We disallowed 1013 * renaming over top of a directory with links to 1014 * it above, as the remaining link would point to 1015 * a directory without "." or ".." entries. 1016 */ 1017 ext2_dec_nlink(xp); 1018 if (doingdirectory) { 1019 if (--xp->i_nlink != 0) 1020 panic("ext2_rename: linked directory"); 1021 error = ext2_truncate(tvp, (off_t)0, IO_SYNC, 1022 tcnp->cn_cred, tcnp->cn_thread); 1023 } 1024 xp->i_flag |= IN_CHANGE; 1025 vput(tvp); 1026 xp = NULL; 1027 } 1028 1029 /* 1030 * 3) Unlink the source. 1031 */ 1032 fcnp->cn_flags &= ~MODMASK; 1033 fcnp->cn_flags |= LOCKPARENT | LOCKLEAF; 1034 VREF(fdvp); 1035 error = relookup(fdvp, &fvp, fcnp); 1036 if (error == 0) 1037 vrele(fdvp); 1038 if (fvp != NULL) { 1039 xp = VTOI(fvp); 1040 dp = VTOI(fdvp); 1041 } else { 1042 /* 1043 * From name has disappeared. IN_RENAME is not sufficient 1044 * to protect against directory races due to timing windows, 1045 * so we can't panic here. 1046 */ 1047 vrele(ap->a_fvp); 1048 return (0); 1049 } 1050 /* 1051 * Ensure that the directory entry still exists and has not 1052 * changed while the new name has been entered. If the source is 1053 * a file then the entry may have been unlinked or renamed. In 1054 * either case there is no further work to be done. If the source 1055 * is a directory then it cannot have been rmdir'ed; its link 1056 * count of three would cause a rmdir to fail with ENOTEMPTY. 1057 * The IN_RENAME flag ensures that it cannot be moved by another 1058 * rename. 1059 */ 1060 if (xp != ip) { 1061 /* 1062 * From name resolves to a different inode. IN_RENAME is 1063 * not sufficient protection against timing window races 1064 * so we can't panic here. 1065 */ 1066 } else { 1067 /* 1068 * If the source is a directory with a 1069 * new parent, the link count of the old 1070 * parent directory must be decremented 1071 * and ".." set to point to the new parent. 1072 */ 1073 if (doingdirectory && newparent) { 1074 ext2_dec_nlink(dp); 1075 dp->i_flag |= IN_CHANGE; 1076 dirbuf = malloc(dp->i_e2fs->e2fs_bsize, M_TEMP, M_WAITOK | M_ZERO); 1077 if (!dirbuf) { 1078 error = ENOMEM; 1079 goto bad; 1080 } 1081 error = vn_rdwr(UIO_READ, fvp, (caddr_t)dirbuf, 1082 ip->i_e2fs->e2fs_bsize, (off_t)0, 1083 UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK, 1084 tcnp->cn_cred, NOCRED, NULL, NULL); 1085 if (error == 0) { 1086 /* Like ufs little-endian: */ 1087 namlen = dirbuf->dotdot_type; 1088 if (namlen != 2 || 1089 dirbuf->dotdot_name[0] != '.' || 1090 dirbuf->dotdot_name[1] != '.') { 1091 ext2_dirbad(xp, (doff_t)12, 1092 "rename: mangled dir"); 1093 } else { 1094 dirbuf->dotdot_ino = newparent; 1095 /* 1096 * dirblock 0 could be htree root, 1097 * try both csum update functions. 1098 */ 1099 ext2_dirent_csum_set(ip, 1100 (struct ext2fs_direct_2 *)dirbuf); 1101 ext2_dx_csum_set(ip, 1102 (struct ext2fs_direct_2 *)dirbuf); 1103 (void)vn_rdwr(UIO_WRITE, fvp, 1104 (caddr_t)dirbuf, 1105 ip->i_e2fs->e2fs_bsize, 1106 (off_t)0, UIO_SYSSPACE, 1107 IO_NODELOCKED | IO_SYNC | 1108 IO_NOMACCHECK, tcnp->cn_cred, 1109 NOCRED, NULL, NULL); 1110 cache_purge(fdvp); 1111 } 1112 } 1113 free(dirbuf, M_TEMP); 1114 } 1115 error = ext2_dirremove(fdvp, fcnp); 1116 if (!error) { 1117 ext2_dec_nlink(xp); 1118 xp->i_flag |= IN_CHANGE; 1119 } 1120 xp->i_flag &= ~IN_RENAME; 1121 } 1122 if (dp) 1123 vput(fdvp); 1124 if (xp) 1125 vput(fvp); 1126 vrele(ap->a_fvp); 1127 return (error); 1128 1129 bad: 1130 if (xp) 1131 vput(ITOV(xp)); 1132 vput(ITOV(dp)); 1133 out: 1134 if (doingdirectory) 1135 ip->i_flag &= ~IN_RENAME; 1136 if (vn_lock(fvp, LK_EXCLUSIVE) == 0) { 1137 ext2_dec_nlink(ip); 1138 ip->i_flag |= IN_CHANGE; 1139 ip->i_flag &= ~IN_RENAME; 1140 vput(fvp); 1141 } else 1142 vrele(fvp); 1143 return (error); 1144 } 1145 1146 #ifdef UFS_ACL 1147 static int 1148 ext2_do_posix1e_acl_inheritance_dir(struct vnode *dvp, struct vnode *tvp, 1149 mode_t dmode, struct ucred *cred, struct thread *td) 1150 { 1151 int error; 1152 struct inode *ip = VTOI(tvp); 1153 struct acl *dacl, *acl; 1154 1155 acl = acl_alloc(M_WAITOK); 1156 dacl = acl_alloc(M_WAITOK); 1157 1158 /* 1159 * Retrieve default ACL from parent, if any. 1160 */ 1161 error = VOP_GETACL(dvp, ACL_TYPE_DEFAULT, acl, cred, td); 1162 switch (error) { 1163 case 0: 1164 /* 1165 * Retrieved a default ACL, so merge mode and ACL if 1166 * necessary. If the ACL is empty, fall through to 1167 * the "not defined or available" case. 1168 */ 1169 if (acl->acl_cnt != 0) { 1170 dmode = acl_posix1e_newfilemode(dmode, acl); 1171 ip->i_mode = dmode; 1172 *dacl = *acl; 1173 ext2_sync_acl_from_inode(ip, acl); 1174 break; 1175 } 1176 /* FALLTHROUGH */ 1177 1178 case EOPNOTSUPP: 1179 /* 1180 * Just use the mode as-is. 1181 */ 1182 ip->i_mode = dmode; 1183 error = 0; 1184 goto out; 1185 1186 default: 1187 goto out; 1188 } 1189 1190 error = VOP_SETACL(tvp, ACL_TYPE_ACCESS, acl, cred, td); 1191 if (error == 0) 1192 error = VOP_SETACL(tvp, ACL_TYPE_DEFAULT, dacl, cred, td); 1193 switch (error) { 1194 case 0: 1195 break; 1196 1197 case EOPNOTSUPP: 1198 /* 1199 * XXX: This should not happen, as EOPNOTSUPP above 1200 * was supposed to free acl. 1201 */ 1202 #ifdef DEBUG 1203 printf("ext2_mkdir: VOP_GETACL() but no VOP_SETACL()\n"); 1204 #endif /* DEBUG */ 1205 break; 1206 1207 default: 1208 goto out; 1209 } 1210 1211 out: 1212 acl_free(acl); 1213 acl_free(dacl); 1214 1215 return (error); 1216 } 1217 1218 static int 1219 ext2_do_posix1e_acl_inheritance_file(struct vnode *dvp, struct vnode *tvp, 1220 mode_t mode, struct ucred *cred, struct thread *td) 1221 { 1222 int error; 1223 struct inode *ip = VTOI(tvp); 1224 struct acl *acl; 1225 1226 acl = acl_alloc(M_WAITOK); 1227 1228 /* 1229 * Retrieve default ACL for parent, if any. 1230 */ 1231 error = VOP_GETACL(dvp, ACL_TYPE_DEFAULT, acl, cred, td); 1232 switch (error) { 1233 case 0: 1234 /* 1235 * Retrieved a default ACL, so merge mode and ACL if 1236 * necessary. 1237 */ 1238 if (acl->acl_cnt != 0) { 1239 /* 1240 * Two possible ways for default ACL to not 1241 * be present. First, the EA can be 1242 * undefined, or second, the default ACL can 1243 * be blank. If it's blank, fall through to 1244 * the it's not defined case. 1245 */ 1246 mode = acl_posix1e_newfilemode(mode, acl); 1247 ip->i_mode = mode; 1248 ext2_sync_acl_from_inode(ip, acl); 1249 break; 1250 } 1251 /* FALLTHROUGH */ 1252 1253 case EOPNOTSUPP: 1254 /* 1255 * Just use the mode as-is. 1256 */ 1257 ip->i_mode = mode; 1258 error = 0; 1259 goto out; 1260 1261 default: 1262 goto out; 1263 } 1264 1265 error = VOP_SETACL(tvp, ACL_TYPE_ACCESS, acl, cred, td); 1266 switch (error) { 1267 case 0: 1268 break; 1269 1270 case EOPNOTSUPP: 1271 /* 1272 * XXX: This should not happen, as EOPNOTSUPP above was 1273 * supposed to free acl. 1274 */ 1275 printf("ufs_do_posix1e_acl_inheritance_file: VOP_GETACL() " 1276 "but no VOP_SETACL()\n"); 1277 /* panic("ufs_do_posix1e_acl_inheritance_file: VOP_GETACL() " 1278 "but no VOP_SETACL()"); */ 1279 break; 1280 1281 default: 1282 goto out; 1283 } 1284 1285 out: 1286 acl_free(acl); 1287 1288 return (error); 1289 } 1290 1291 #endif /* UFS_ACL */ 1292 1293 /* 1294 * Mkdir system call 1295 */ 1296 static int 1297 ext2_mkdir(struct vop_mkdir_args *ap) 1298 { 1299 struct m_ext2fs *fs; 1300 struct vnode *dvp = ap->a_dvp; 1301 struct vattr *vap = ap->a_vap; 1302 struct componentname *cnp = ap->a_cnp; 1303 struct inode *ip, *dp; 1304 struct vnode *tvp; 1305 struct dirtemplate dirtemplate, *dtp; 1306 char *buf = NULL; 1307 int error, dmode; 1308 1309 #ifdef INVARIANTS 1310 if ((cnp->cn_flags & HASBUF) == 0) 1311 panic("ext2_mkdir: no name"); 1312 #endif 1313 dp = VTOI(dvp); 1314 if ((nlink_t)dp->i_nlink >= EXT4_LINK_MAX && 1315 !EXT2_HAS_RO_COMPAT_FEATURE(dp->i_e2fs, EXT2F_ROCOMPAT_DIR_NLINK)) { 1316 error = EMLINK; 1317 goto out; 1318 } 1319 dmode = vap->va_mode & 0777; 1320 dmode |= IFDIR; 1321 /* 1322 * Must simulate part of ext2_makeinode here to acquire the inode, 1323 * but not have it entered in the parent directory. The entry is 1324 * made later after writing "." and ".." entries. 1325 */ 1326 error = ext2_valloc(dvp, dmode, cnp->cn_cred, &tvp); 1327 if (error) 1328 goto out; 1329 ip = VTOI(tvp); 1330 fs = ip->i_e2fs; 1331 ip->i_gid = dp->i_gid; 1332 #ifdef SUIDDIR 1333 { 1334 /* 1335 * if we are hacking owners here, (only do this where told to) 1336 * and we are not giving it TOO root, (would subvert quotas) 1337 * then go ahead and give it to the other user. 1338 * The new directory also inherits the SUID bit. 1339 * If user's UID and dir UID are the same, 1340 * 'give it away' so that the SUID is still forced on. 1341 */ 1342 if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) && 1343 (dp->i_mode & ISUID) && dp->i_uid) { 1344 dmode |= ISUID; 1345 ip->i_uid = dp->i_uid; 1346 } else { 1347 ip->i_uid = cnp->cn_cred->cr_uid; 1348 } 1349 } 1350 #else 1351 ip->i_uid = cnp->cn_cred->cr_uid; 1352 #endif 1353 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 1354 ip->i_mode = dmode; 1355 tvp->v_type = VDIR; /* Rest init'd in getnewvnode(). */ 1356 ip->i_nlink = 2; 1357 if (cnp->cn_flags & ISWHITEOUT) 1358 ip->i_flags |= UF_OPAQUE; 1359 error = ext2_update(tvp, 1); 1360 1361 /* 1362 * Bump link count in parent directory 1363 * to reflect work done below. Should 1364 * be done before reference is created 1365 * so reparation is possible if we crash. 1366 */ 1367 ext2_inc_nlink(dp); 1368 dp->i_flag |= IN_CHANGE; 1369 error = ext2_update(dvp, !DOINGASYNC(dvp)); 1370 if (error) 1371 goto bad; 1372 1373 /* Initialize directory with "." and ".." from static template. */ 1374 if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs, 1375 EXT2F_INCOMPAT_FTYPE)) 1376 dtp = &mastertemplate; 1377 else 1378 dtp = &omastertemplate; 1379 dirtemplate = *dtp; 1380 dirtemplate.dot_ino = ip->i_number; 1381 dirtemplate.dotdot_ino = dp->i_number; 1382 /* 1383 * note that in ext2 DIRBLKSIZ == blocksize, not DEV_BSIZE so let's 1384 * just redefine it - for this function only 1385 */ 1386 #undef DIRBLKSIZ 1387 #define DIRBLKSIZ VTOI(dvp)->i_e2fs->e2fs_bsize 1388 dirtemplate.dotdot_reclen = DIRBLKSIZ - 12; 1389 buf = malloc(DIRBLKSIZ, M_TEMP, M_WAITOK | M_ZERO); 1390 if (!buf) { 1391 error = ENOMEM; 1392 ext2_dec_nlink(dp); 1393 dp->i_flag |= IN_CHANGE; 1394 goto bad; 1395 } 1396 if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM)) { 1397 dirtemplate.dotdot_reclen -= sizeof(struct ext2fs_direct_tail); 1398 ext2_init_dirent_tail(EXT2_DIRENT_TAIL(buf, DIRBLKSIZ)); 1399 } 1400 memcpy(buf, &dirtemplate, sizeof(dirtemplate)); 1401 ext2_dirent_csum_set(ip, (struct ext2fs_direct_2 *)buf); 1402 error = vn_rdwr(UIO_WRITE, tvp, (caddr_t)buf, 1403 DIRBLKSIZ, (off_t)0, UIO_SYSSPACE, 1404 IO_NODELOCKED | IO_SYNC | IO_NOMACCHECK, cnp->cn_cred, NOCRED, 1405 NULL, NULL); 1406 if (error) { 1407 ext2_dec_nlink(dp); 1408 dp->i_flag |= IN_CHANGE; 1409 goto bad; 1410 } 1411 if (DIRBLKSIZ > VFSTOEXT2(dvp->v_mount)->um_mountp->mnt_stat.f_bsize) 1412 /* XXX should grow with balloc() */ 1413 panic("ext2_mkdir: blksize"); 1414 else { 1415 ip->i_size = DIRBLKSIZ; 1416 ip->i_flag |= IN_CHANGE; 1417 } 1418 1419 #ifdef UFS_ACL 1420 if (dvp->v_mount->mnt_flag & MNT_ACLS) { 1421 error = ext2_do_posix1e_acl_inheritance_dir(dvp, tvp, dmode, 1422 cnp->cn_cred, cnp->cn_thread); 1423 if (error) 1424 goto bad; 1425 } 1426 1427 #endif /* UFS_ACL */ 1428 1429 /* Directory set up, now install its entry in the parent directory. */ 1430 error = ext2_direnter(ip, dvp, cnp); 1431 if (error) { 1432 ext2_dec_nlink(dp); 1433 dp->i_flag |= IN_CHANGE; 1434 } 1435 bad: 1436 /* 1437 * No need to do an explicit VOP_TRUNCATE here, vrele will do this 1438 * for us because we set the link count to 0. 1439 */ 1440 if (error) { 1441 ip->i_nlink = 0; 1442 ip->i_flag |= IN_CHANGE; 1443 vput(tvp); 1444 } else 1445 *ap->a_vpp = tvp; 1446 out: 1447 free(buf, M_TEMP); 1448 return (error); 1449 #undef DIRBLKSIZ 1450 #define DIRBLKSIZ DEV_BSIZE 1451 } 1452 1453 /* 1454 * Rmdir system call. 1455 */ 1456 static int 1457 ext2_rmdir(struct vop_rmdir_args *ap) 1458 { 1459 struct vnode *vp = ap->a_vp; 1460 struct vnode *dvp = ap->a_dvp; 1461 struct componentname *cnp = ap->a_cnp; 1462 struct inode *ip, *dp; 1463 int error; 1464 1465 ip = VTOI(vp); 1466 dp = VTOI(dvp); 1467 1468 /* 1469 * Verify the directory is empty (and valid). 1470 * (Rmdir ".." won't be valid since 1471 * ".." will contain a reference to 1472 * the current directory and thus be 1473 * non-empty.) 1474 */ 1475 if (!ext2_dirempty(ip, dp->i_number, cnp->cn_cred)) { 1476 error = ENOTEMPTY; 1477 goto out; 1478 } 1479 if ((dp->i_flags & APPEND) 1480 || (ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))) { 1481 error = EPERM; 1482 goto out; 1483 } 1484 /* 1485 * Delete reference to directory before purging 1486 * inode. If we crash in between, the directory 1487 * will be reattached to lost+found, 1488 */ 1489 error = ext2_dirremove(dvp, cnp); 1490 if (error) 1491 goto out; 1492 ext2_dec_nlink(dp); 1493 dp->i_flag |= IN_CHANGE; 1494 cache_purge(dvp); 1495 VOP_UNLOCK(dvp, 0); 1496 /* 1497 * Truncate inode. The only stuff left 1498 * in the directory is "." and "..". 1499 */ 1500 ip->i_nlink = 0; 1501 error = ext2_truncate(vp, (off_t)0, IO_SYNC, cnp->cn_cred, 1502 cnp->cn_thread); 1503 cache_purge(ITOV(ip)); 1504 if (vn_lock(dvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) { 1505 VOP_UNLOCK(vp, 0); 1506 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 1507 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1508 } 1509 out: 1510 return (error); 1511 } 1512 1513 /* 1514 * symlink -- make a symbolic link 1515 */ 1516 static int 1517 ext2_symlink(struct vop_symlink_args *ap) 1518 { 1519 struct vnode *vp, **vpp = ap->a_vpp; 1520 struct inode *ip; 1521 int len, error; 1522 1523 error = ext2_makeinode(IFLNK | ap->a_vap->va_mode, ap->a_dvp, 1524 vpp, ap->a_cnp); 1525 if (error) 1526 return (error); 1527 vp = *vpp; 1528 len = strlen(ap->a_target); 1529 if (len < vp->v_mount->mnt_maxsymlinklen) { 1530 ip = VTOI(vp); 1531 bcopy(ap->a_target, (char *)ip->i_shortlink, len); 1532 ip->i_size = len; 1533 ip->i_flag |= IN_CHANGE | IN_UPDATE; 1534 } else 1535 error = vn_rdwr(UIO_WRITE, vp, __DECONST(void *, ap->a_target), 1536 len, (off_t)0, UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK, 1537 ap->a_cnp->cn_cred, NOCRED, NULL, NULL); 1538 if (error) 1539 vput(vp); 1540 return (error); 1541 } 1542 1543 /* 1544 * Return target name of a symbolic link 1545 */ 1546 static int 1547 ext2_readlink(struct vop_readlink_args *ap) 1548 { 1549 struct vnode *vp = ap->a_vp; 1550 struct inode *ip = VTOI(vp); 1551 int isize; 1552 1553 isize = ip->i_size; 1554 if (isize < vp->v_mount->mnt_maxsymlinklen) { 1555 uiomove((char *)ip->i_shortlink, isize, ap->a_uio); 1556 return (0); 1557 } 1558 return (VOP_READ(vp, ap->a_uio, 0, ap->a_cred)); 1559 } 1560 1561 /* 1562 * Calculate the logical to physical mapping if not done already, 1563 * then call the device strategy routine. 1564 * 1565 * In order to be able to swap to a file, the ext2_bmaparray() operation may not 1566 * deadlock on memory. See ext2_bmap() for details. 1567 */ 1568 static int 1569 ext2_strategy(struct vop_strategy_args *ap) 1570 { 1571 struct buf *bp = ap->a_bp; 1572 struct vnode *vp = ap->a_vp; 1573 struct bufobj *bo; 1574 daddr_t blkno; 1575 int error; 1576 1577 if (vp->v_type == VBLK || vp->v_type == VCHR) 1578 panic("ext2_strategy: spec"); 1579 if (bp->b_blkno == bp->b_lblkno) { 1580 1581 if (VTOI(ap->a_vp)->i_flag & IN_E4EXTENTS) 1582 error = ext4_bmapext(vp, bp->b_lblkno, &blkno, NULL, NULL); 1583 else 1584 error = ext2_bmaparray(vp, bp->b_lblkno, &blkno, NULL, NULL); 1585 1586 bp->b_blkno = blkno; 1587 if (error) { 1588 bp->b_error = error; 1589 bp->b_ioflags |= BIO_ERROR; 1590 bufdone(bp); 1591 return (0); 1592 } 1593 if ((long)bp->b_blkno == -1) 1594 vfs_bio_clrbuf(bp); 1595 } 1596 if ((long)bp->b_blkno == -1) { 1597 bufdone(bp); 1598 return (0); 1599 } 1600 bp->b_iooffset = dbtob(bp->b_blkno); 1601 bo = VFSTOEXT2(vp->v_mount)->um_bo; 1602 BO_STRATEGY(bo, bp); 1603 return (0); 1604 } 1605 1606 /* 1607 * Print out the contents of an inode. 1608 */ 1609 static int 1610 ext2_print(struct vop_print_args *ap) 1611 { 1612 struct vnode *vp = ap->a_vp; 1613 struct inode *ip = VTOI(vp); 1614 1615 vn_printf(ip->i_devvp, "\tino %ju", (uintmax_t)ip->i_number); 1616 if (vp->v_type == VFIFO) 1617 fifo_printinfo(vp); 1618 printf("\n"); 1619 return (0); 1620 } 1621 1622 /* 1623 * Close wrapper for fifos. 1624 * 1625 * Update the times on the inode then do device close. 1626 */ 1627 static int 1628 ext2fifo_close(struct vop_close_args *ap) 1629 { 1630 struct vnode *vp = ap->a_vp; 1631 1632 VI_LOCK(vp); 1633 if (vp->v_usecount > 1) 1634 ext2_itimes_locked(vp); 1635 VI_UNLOCK(vp); 1636 return (fifo_specops.vop_close(ap)); 1637 } 1638 1639 /* 1640 * Kqfilter wrapper for fifos. 1641 * 1642 * Fall through to ext2 kqfilter routines if needed 1643 */ 1644 static int 1645 ext2fifo_kqfilter(struct vop_kqfilter_args *ap) 1646 { 1647 int error; 1648 1649 error = fifo_specops.vop_kqfilter(ap); 1650 if (error) 1651 error = vfs_kqfilter(ap); 1652 return (error); 1653 } 1654 1655 /* 1656 * Return POSIX pathconf information applicable to ext2 filesystems. 1657 */ 1658 static int 1659 ext2_pathconf(struct vop_pathconf_args *ap) 1660 { 1661 int error = 0; 1662 1663 switch (ap->a_name) { 1664 case _PC_LINK_MAX: 1665 if (EXT2_HAS_RO_COMPAT_FEATURE(VTOI(ap->a_vp)->i_e2fs, 1666 EXT2F_ROCOMPAT_DIR_NLINK)) 1667 *ap->a_retval = INT_MAX; 1668 else 1669 *ap->a_retval = EXT4_LINK_MAX; 1670 break; 1671 case _PC_NAME_MAX: 1672 *ap->a_retval = NAME_MAX; 1673 break; 1674 case _PC_PIPE_BUF: 1675 if (ap->a_vp->v_type == VDIR || ap->a_vp->v_type == VFIFO) 1676 *ap->a_retval = PIPE_BUF; 1677 else 1678 error = EINVAL; 1679 break; 1680 case _PC_CHOWN_RESTRICTED: 1681 *ap->a_retval = 1; 1682 break; 1683 case _PC_NO_TRUNC: 1684 *ap->a_retval = 1; 1685 break; 1686 1687 #ifdef UFS_ACL 1688 case _PC_ACL_EXTENDED: 1689 if (ap->a_vp->v_mount->mnt_flag & MNT_ACLS) 1690 *ap->a_retval = 1; 1691 else 1692 *ap->a_retval = 0; 1693 break; 1694 case _PC_ACL_PATH_MAX: 1695 if (ap->a_vp->v_mount->mnt_flag & MNT_ACLS) 1696 *ap->a_retval = ACL_MAX_ENTRIES; 1697 else 1698 *ap->a_retval = 3; 1699 break; 1700 #endif /* UFS_ACL */ 1701 1702 case _PC_MIN_HOLE_SIZE: 1703 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize; 1704 break; 1705 case _PC_PRIO_IO: 1706 *ap->a_retval = 0; 1707 break; 1708 case _PC_SYNC_IO: 1709 *ap->a_retval = 0; 1710 break; 1711 case _PC_ALLOC_SIZE_MIN: 1712 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_bsize; 1713 break; 1714 case _PC_FILESIZEBITS: 1715 *ap->a_retval = 64; 1716 break; 1717 case _PC_REC_INCR_XFER_SIZE: 1718 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize; 1719 break; 1720 case _PC_REC_MAX_XFER_SIZE: 1721 *ap->a_retval = -1; /* means ``unlimited'' */ 1722 break; 1723 case _PC_REC_MIN_XFER_SIZE: 1724 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize; 1725 break; 1726 case _PC_REC_XFER_ALIGN: 1727 *ap->a_retval = PAGE_SIZE; 1728 break; 1729 case _PC_SYMLINK_MAX: 1730 *ap->a_retval = MAXPATHLEN; 1731 break; 1732 1733 default: 1734 error = vop_stdpathconf(ap); 1735 break; 1736 } 1737 return (error); 1738 } 1739 1740 /* 1741 * Vnode operation to remove a named attribute. 1742 */ 1743 static int 1744 ext2_deleteextattr(struct vop_deleteextattr_args *ap) 1745 { 1746 struct inode *ip; 1747 struct m_ext2fs *fs; 1748 int error; 1749 1750 ip = VTOI(ap->a_vp); 1751 fs = ip->i_e2fs; 1752 1753 if (!EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_EXT_ATTR)) 1754 return (EOPNOTSUPP); 1755 1756 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK) 1757 return (EOPNOTSUPP); 1758 1759 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 1760 ap->a_cred, ap->a_td, VWRITE); 1761 if (error) 1762 return (error); 1763 1764 error = ENOATTR; 1765 1766 if (EXT2_INODE_SIZE(fs) != E2FS_REV0_INODE_SIZE) { 1767 error = ext2_extattr_inode_delete(ip, ap->a_attrnamespace, ap->a_name); 1768 if (error != ENOATTR) 1769 return (error); 1770 } 1771 1772 if (ip->i_facl) 1773 error = ext2_extattr_block_delete(ip, ap->a_attrnamespace, ap->a_name); 1774 1775 return (error); 1776 } 1777 1778 /* 1779 * Vnode operation to retrieve a named extended attribute. 1780 */ 1781 static int 1782 ext2_getextattr(struct vop_getextattr_args *ap) 1783 { 1784 struct inode *ip; 1785 struct m_ext2fs *fs; 1786 int error; 1787 1788 ip = VTOI(ap->a_vp); 1789 fs = ip->i_e2fs; 1790 1791 if (!EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_EXT_ATTR)) 1792 return (EOPNOTSUPP); 1793 1794 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK) 1795 return (EOPNOTSUPP); 1796 1797 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 1798 ap->a_cred, ap->a_td, VREAD); 1799 if (error) 1800 return (error); 1801 1802 if (ap->a_size != NULL) 1803 *ap->a_size = 0; 1804 1805 error = ENOATTR; 1806 1807 if (EXT2_INODE_SIZE(fs) != E2FS_REV0_INODE_SIZE) { 1808 error = ext2_extattr_inode_get(ip, ap->a_attrnamespace, 1809 ap->a_name, ap->a_uio, ap->a_size); 1810 if (error != ENOATTR) 1811 return (error); 1812 } 1813 1814 if (ip->i_facl) 1815 error = ext2_extattr_block_get(ip, ap->a_attrnamespace, 1816 ap->a_name, ap->a_uio, ap->a_size); 1817 1818 return (error); 1819 } 1820 1821 /* 1822 * Vnode operation to retrieve extended attributes on a vnode. 1823 */ 1824 static int 1825 ext2_listextattr(struct vop_listextattr_args *ap) 1826 { 1827 struct inode *ip; 1828 struct m_ext2fs *fs; 1829 int error; 1830 1831 ip = VTOI(ap->a_vp); 1832 fs = ip->i_e2fs; 1833 1834 if (!EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_EXT_ATTR)) 1835 return (EOPNOTSUPP); 1836 1837 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK) 1838 return (EOPNOTSUPP); 1839 1840 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 1841 ap->a_cred, ap->a_td, VREAD); 1842 if (error) 1843 return (error); 1844 1845 if (ap->a_size != NULL) 1846 *ap->a_size = 0; 1847 1848 if (EXT2_INODE_SIZE(fs) != E2FS_REV0_INODE_SIZE) { 1849 error = ext2_extattr_inode_list(ip, ap->a_attrnamespace, 1850 ap->a_uio, ap->a_size); 1851 if (error) 1852 return (error); 1853 } 1854 1855 if (ip->i_facl) 1856 error = ext2_extattr_block_list(ip, ap->a_attrnamespace, 1857 ap->a_uio, ap->a_size); 1858 1859 return (error); 1860 } 1861 1862 /* 1863 * Vnode operation to set a named attribute. 1864 */ 1865 static int 1866 ext2_setextattr(struct vop_setextattr_args *ap) 1867 { 1868 struct inode *ip; 1869 struct m_ext2fs *fs; 1870 int error; 1871 1872 ip = VTOI(ap->a_vp); 1873 fs = ip->i_e2fs; 1874 1875 if (!EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_EXT_ATTR)) 1876 return (EOPNOTSUPP); 1877 1878 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK) 1879 return (EOPNOTSUPP); 1880 1881 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 1882 ap->a_cred, ap->a_td, VWRITE); 1883 if (error) 1884 return (error); 1885 1886 error = ext2_extattr_valid_attrname(ap->a_attrnamespace, ap->a_name); 1887 if (error) 1888 return (error); 1889 1890 if (EXT2_INODE_SIZE(fs) != E2FS_REV0_INODE_SIZE) { 1891 error = ext2_extattr_inode_set(ip, ap->a_attrnamespace, 1892 ap->a_name, ap->a_uio); 1893 if (error != ENOSPC) 1894 return (error); 1895 } 1896 1897 error = ext2_extattr_block_set(ip, ap->a_attrnamespace, 1898 ap->a_name, ap->a_uio); 1899 1900 return (error); 1901 } 1902 1903 /* 1904 * Vnode pointer to File handle 1905 */ 1906 /* ARGSUSED */ 1907 static int 1908 ext2_vptofh(struct vop_vptofh_args *ap) 1909 { 1910 struct inode *ip; 1911 struct ufid *ufhp; 1912 1913 ip = VTOI(ap->a_vp); 1914 ufhp = (struct ufid *)ap->a_fhp; 1915 ufhp->ufid_len = sizeof(struct ufid); 1916 ufhp->ufid_ino = ip->i_number; 1917 ufhp->ufid_gen = ip->i_gen; 1918 return (0); 1919 } 1920 1921 /* 1922 * Initialize the vnode associated with a new inode, handle aliased 1923 * vnodes. 1924 */ 1925 int 1926 ext2_vinit(struct mount *mntp, struct vop_vector *fifoops, struct vnode **vpp) 1927 { 1928 struct inode *ip; 1929 struct vnode *vp; 1930 1931 vp = *vpp; 1932 ip = VTOI(vp); 1933 vp->v_type = IFTOVT(ip->i_mode); 1934 /* 1935 * Only unallocated inodes should be of type VNON. 1936 */ 1937 if (ip->i_mode != 0 && vp->v_type == VNON) 1938 return (EINVAL); 1939 if (vp->v_type == VFIFO) 1940 vp->v_op = fifoops; 1941 1942 if (ip->i_number == EXT2_ROOTINO) 1943 vp->v_vflag |= VV_ROOT; 1944 ip->i_modrev = init_va_filerev(); 1945 *vpp = vp; 1946 return (0); 1947 } 1948 1949 /* 1950 * Allocate a new inode. 1951 */ 1952 static int 1953 ext2_makeinode(int mode, struct vnode *dvp, struct vnode **vpp, 1954 struct componentname *cnp) 1955 { 1956 struct inode *ip, *pdir; 1957 struct vnode *tvp; 1958 int error; 1959 1960 pdir = VTOI(dvp); 1961 #ifdef INVARIANTS 1962 if ((cnp->cn_flags & HASBUF) == 0) 1963 panic("ext2_makeinode: no name"); 1964 #endif 1965 *vpp = NULL; 1966 if ((mode & IFMT) == 0) 1967 mode |= IFREG; 1968 1969 error = ext2_valloc(dvp, mode, cnp->cn_cred, &tvp); 1970 if (error) { 1971 return (error); 1972 } 1973 ip = VTOI(tvp); 1974 ip->i_gid = pdir->i_gid; 1975 #ifdef SUIDDIR 1976 { 1977 /* 1978 * if we are 1979 * not the owner of the directory, 1980 * and we are hacking owners here, (only do this where told to) 1981 * and we are not giving it TOO root, (would subvert quotas) 1982 * then go ahead and give it to the other user. 1983 * Note that this drops off the execute bits for security. 1984 */ 1985 if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) && 1986 (pdir->i_mode & ISUID) && 1987 (pdir->i_uid != cnp->cn_cred->cr_uid) && pdir->i_uid) { 1988 ip->i_uid = pdir->i_uid; 1989 mode &= ~07111; 1990 } else { 1991 ip->i_uid = cnp->cn_cred->cr_uid; 1992 } 1993 } 1994 #else 1995 ip->i_uid = cnp->cn_cred->cr_uid; 1996 #endif 1997 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 1998 ip->i_mode = mode; 1999 tvp->v_type = IFTOVT(mode); /* Rest init'd in getnewvnode(). */ 2000 ip->i_nlink = 1; 2001 if ((ip->i_mode & ISGID) && !groupmember(ip->i_gid, cnp->cn_cred)) { 2002 if (priv_check_cred(cnp->cn_cred, PRIV_VFS_RETAINSUGID)) 2003 ip->i_mode &= ~ISGID; 2004 } 2005 2006 if (cnp->cn_flags & ISWHITEOUT) 2007 ip->i_flags |= UF_OPAQUE; 2008 2009 /* 2010 * Make sure inode goes to disk before directory entry. 2011 */ 2012 error = ext2_update(tvp, !DOINGASYNC(tvp)); 2013 if (error) 2014 goto bad; 2015 2016 #ifdef UFS_ACL 2017 if (dvp->v_mount->mnt_flag & MNT_ACLS) { 2018 error = ext2_do_posix1e_acl_inheritance_file(dvp, tvp, mode, 2019 cnp->cn_cred, cnp->cn_thread); 2020 if (error) 2021 goto bad; 2022 } 2023 #endif /* UFS_ACL */ 2024 2025 error = ext2_direnter(ip, dvp, cnp); 2026 if (error) 2027 goto bad; 2028 2029 *vpp = tvp; 2030 return (0); 2031 2032 bad: 2033 /* 2034 * Write error occurred trying to update the inode 2035 * or the directory so must deallocate the inode. 2036 */ 2037 ip->i_nlink = 0; 2038 ip->i_flag |= IN_CHANGE; 2039 vput(tvp); 2040 return (error); 2041 } 2042 2043 /* 2044 * Vnode op for reading. 2045 */ 2046 static int 2047 ext2_read(struct vop_read_args *ap) 2048 { 2049 struct vnode *vp; 2050 struct inode *ip; 2051 struct uio *uio; 2052 struct m_ext2fs *fs; 2053 struct buf *bp; 2054 daddr_t lbn, nextlbn; 2055 off_t bytesinfile; 2056 long size, xfersize, blkoffset; 2057 int error, orig_resid, seqcount; 2058 int ioflag; 2059 2060 vp = ap->a_vp; 2061 uio = ap->a_uio; 2062 ioflag = ap->a_ioflag; 2063 2064 seqcount = ap->a_ioflag >> IO_SEQSHIFT; 2065 ip = VTOI(vp); 2066 2067 #ifdef INVARIANTS 2068 if (uio->uio_rw != UIO_READ) 2069 panic("%s: mode", "ext2_read"); 2070 2071 if (vp->v_type == VLNK) { 2072 if ((int)ip->i_size < vp->v_mount->mnt_maxsymlinklen) 2073 panic("%s: short symlink", "ext2_read"); 2074 } else if (vp->v_type != VREG && vp->v_type != VDIR) 2075 panic("%s: type %d", "ext2_read", vp->v_type); 2076 #endif 2077 orig_resid = uio->uio_resid; 2078 KASSERT(orig_resid >= 0, ("ext2_read: uio->uio_resid < 0")); 2079 if (orig_resid == 0) 2080 return (0); 2081 KASSERT(uio->uio_offset >= 0, ("ext2_read: uio->uio_offset < 0")); 2082 fs = ip->i_e2fs; 2083 if (uio->uio_offset < ip->i_size && 2084 uio->uio_offset >= fs->e2fs_maxfilesize) 2085 return (EOVERFLOW); 2086 2087 for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) { 2088 if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0) 2089 break; 2090 lbn = lblkno(fs, uio->uio_offset); 2091 nextlbn = lbn + 1; 2092 size = blksize(fs, ip, lbn); 2093 blkoffset = blkoff(fs, uio->uio_offset); 2094 2095 xfersize = fs->e2fs_fsize - blkoffset; 2096 if (uio->uio_resid < xfersize) 2097 xfersize = uio->uio_resid; 2098 if (bytesinfile < xfersize) 2099 xfersize = bytesinfile; 2100 2101 if (lblktosize(fs, nextlbn) >= ip->i_size) 2102 error = bread(vp, lbn, size, NOCRED, &bp); 2103 else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) { 2104 error = cluster_read(vp, ip->i_size, lbn, size, 2105 NOCRED, blkoffset + uio->uio_resid, seqcount, 2106 0, &bp); 2107 } else if (seqcount > 1) { 2108 u_int nextsize = blksize(fs, ip, nextlbn); 2109 2110 error = breadn(vp, lbn, 2111 size, &nextlbn, &nextsize, 1, NOCRED, &bp); 2112 } else 2113 error = bread(vp, lbn, size, NOCRED, &bp); 2114 if (error) { 2115 brelse(bp); 2116 bp = NULL; 2117 break; 2118 } 2119 2120 /* 2121 * We should only get non-zero b_resid when an I/O error 2122 * has occurred, which should cause us to break above. 2123 * However, if the short read did not cause an error, 2124 * then we want to ensure that we do not uiomove bad 2125 * or uninitialized data. 2126 */ 2127 size -= bp->b_resid; 2128 if (size < xfersize) { 2129 if (size == 0) 2130 break; 2131 xfersize = size; 2132 } 2133 error = uiomove((char *)bp->b_data + blkoffset, 2134 (int)xfersize, uio); 2135 if (error) 2136 break; 2137 vfs_bio_brelse(bp, ioflag); 2138 } 2139 2140 /* 2141 * This can only happen in the case of an error because the loop 2142 * above resets bp to NULL on each iteration and on normal 2143 * completion has not set a new value into it. so it must have come 2144 * from a 'break' statement 2145 */ 2146 if (bp != NULL) 2147 vfs_bio_brelse(bp, ioflag); 2148 2149 if ((error == 0 || uio->uio_resid != orig_resid) && 2150 (vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) 2151 ip->i_flag |= IN_ACCESS; 2152 return (error); 2153 } 2154 2155 static int 2156 ext2_ioctl(struct vop_ioctl_args *ap) 2157 { 2158 2159 switch (ap->a_command) { 2160 case FIOSEEKDATA: 2161 case FIOSEEKHOLE: 2162 return (vn_bmap_seekhole(ap->a_vp, ap->a_command, 2163 (off_t *)ap->a_data, ap->a_cred)); 2164 default: 2165 return (ENOTTY); 2166 } 2167 } 2168 2169 /* 2170 * Vnode op for writing. 2171 */ 2172 static int 2173 ext2_write(struct vop_write_args *ap) 2174 { 2175 struct vnode *vp; 2176 struct uio *uio; 2177 struct inode *ip; 2178 struct m_ext2fs *fs; 2179 struct buf *bp; 2180 daddr_t lbn; 2181 off_t osize; 2182 int blkoffset, error, flags, ioflag, resid, size, seqcount, xfersize; 2183 2184 ioflag = ap->a_ioflag; 2185 uio = ap->a_uio; 2186 vp = ap->a_vp; 2187 2188 seqcount = ioflag >> IO_SEQSHIFT; 2189 ip = VTOI(vp); 2190 2191 #ifdef INVARIANTS 2192 if (uio->uio_rw != UIO_WRITE) 2193 panic("%s: mode", "ext2_write"); 2194 #endif 2195 2196 switch (vp->v_type) { 2197 case VREG: 2198 if (ioflag & IO_APPEND) 2199 uio->uio_offset = ip->i_size; 2200 if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size) 2201 return (EPERM); 2202 /* FALLTHROUGH */ 2203 case VLNK: 2204 break; 2205 case VDIR: 2206 /* XXX differs from ffs -- this is called from ext2_mkdir(). */ 2207 if ((ioflag & IO_SYNC) == 0) 2208 panic("ext2_write: nonsync dir write"); 2209 break; 2210 default: 2211 panic("ext2_write: type %p %d (%jd,%jd)", (void *)vp, 2212 vp->v_type, (intmax_t)uio->uio_offset, 2213 (intmax_t)uio->uio_resid); 2214 } 2215 2216 KASSERT(uio->uio_resid >= 0, ("ext2_write: uio->uio_resid < 0")); 2217 KASSERT(uio->uio_offset >= 0, ("ext2_write: uio->uio_offset < 0")); 2218 fs = ip->i_e2fs; 2219 if ((uoff_t)uio->uio_offset + uio->uio_resid > fs->e2fs_maxfilesize) 2220 return (EFBIG); 2221 /* 2222 * Maybe this should be above the vnode op call, but so long as 2223 * file servers have no limits, I don't think it matters. 2224 */ 2225 if (vn_rlimit_fsize(vp, uio, uio->uio_td)) 2226 return (EFBIG); 2227 2228 resid = uio->uio_resid; 2229 osize = ip->i_size; 2230 if (seqcount > BA_SEQMAX) 2231 flags = BA_SEQMAX << BA_SEQSHIFT; 2232 else 2233 flags = seqcount << BA_SEQSHIFT; 2234 if ((ioflag & IO_SYNC) && !DOINGASYNC(vp)) 2235 flags |= IO_SYNC; 2236 2237 for (error = 0; uio->uio_resid > 0;) { 2238 lbn = lblkno(fs, uio->uio_offset); 2239 blkoffset = blkoff(fs, uio->uio_offset); 2240 xfersize = fs->e2fs_fsize - blkoffset; 2241 if (uio->uio_resid < xfersize) 2242 xfersize = uio->uio_resid; 2243 if (uio->uio_offset + xfersize > ip->i_size) 2244 vnode_pager_setsize(vp, uio->uio_offset + xfersize); 2245 2246 /* 2247 * We must perform a read-before-write if the transfer size 2248 * does not cover the entire buffer. 2249 */ 2250 if (fs->e2fs_bsize > xfersize) 2251 flags |= BA_CLRBUF; 2252 else 2253 flags &= ~BA_CLRBUF; 2254 error = ext2_balloc(ip, lbn, blkoffset + xfersize, 2255 ap->a_cred, &bp, flags); 2256 if (error != 0) 2257 break; 2258 2259 if ((ioflag & (IO_SYNC | IO_INVAL)) == (IO_SYNC | IO_INVAL)) 2260 bp->b_flags |= B_NOCACHE; 2261 if (uio->uio_offset + xfersize > ip->i_size) 2262 ip->i_size = uio->uio_offset + xfersize; 2263 size = blksize(fs, ip, lbn) - bp->b_resid; 2264 if (size < xfersize) 2265 xfersize = size; 2266 2267 error = 2268 uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio); 2269 /* 2270 * If the buffer is not already filled and we encounter an 2271 * error while trying to fill it, we have to clear out any 2272 * garbage data from the pages instantiated for the buffer. 2273 * If we do not, a failed uiomove() during a write can leave 2274 * the prior contents of the pages exposed to a userland mmap. 2275 * 2276 * Note that we need only clear buffers with a transfer size 2277 * equal to the block size because buffers with a shorter 2278 * transfer size were cleared above by the call to ext2_balloc() 2279 * with the BA_CLRBUF flag set. 2280 * 2281 * If the source region for uiomove identically mmaps the 2282 * buffer, uiomove() performed the NOP copy, and the buffer 2283 * content remains valid because the page fault handler 2284 * validated the pages. 2285 */ 2286 if (error != 0 && (bp->b_flags & B_CACHE) == 0 && 2287 fs->e2fs_bsize == xfersize) 2288 vfs_bio_clrbuf(bp); 2289 2290 vfs_bio_set_flags(bp, ioflag); 2291 2292 /* 2293 * If IO_SYNC each buffer is written synchronously. Otherwise 2294 * if we have a severe page deficiency write the buffer 2295 * asynchronously. Otherwise try to cluster, and if that 2296 * doesn't do it then either do an async write (if O_DIRECT), 2297 * or a delayed write (if not). 2298 */ 2299 if (ioflag & IO_SYNC) { 2300 (void)bwrite(bp); 2301 } else if (vm_page_count_severe() || 2302 buf_dirty_count_severe() || 2303 (ioflag & IO_ASYNC)) { 2304 bp->b_flags |= B_CLUSTEROK; 2305 bawrite(bp); 2306 } else if (xfersize + blkoffset == fs->e2fs_fsize) { 2307 if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) { 2308 bp->b_flags |= B_CLUSTEROK; 2309 cluster_write(vp, bp, ip->i_size, seqcount, 0); 2310 } else { 2311 bawrite(bp); 2312 } 2313 } else if (ioflag & IO_DIRECT) { 2314 bp->b_flags |= B_CLUSTEROK; 2315 bawrite(bp); 2316 } else { 2317 bp->b_flags |= B_CLUSTEROK; 2318 bdwrite(bp); 2319 } 2320 if (error || xfersize == 0) 2321 break; 2322 } 2323 /* 2324 * If we successfully wrote any data, and we are not the superuser 2325 * we clear the setuid and setgid bits as a precaution against 2326 * tampering. 2327 */ 2328 if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid && 2329 ap->a_cred) { 2330 if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID)) 2331 ip->i_mode &= ~(ISUID | ISGID); 2332 } 2333 if (error) { 2334 if (ioflag & IO_UNIT) { 2335 (void)ext2_truncate(vp, osize, 2336 ioflag & IO_SYNC, ap->a_cred, uio->uio_td); 2337 uio->uio_offset -= resid - uio->uio_resid; 2338 uio->uio_resid = resid; 2339 } 2340 } 2341 if (uio->uio_resid != resid) { 2342 ip->i_flag |= IN_CHANGE | IN_UPDATE; 2343 if (ioflag & IO_SYNC) 2344 error = ext2_update(vp, 1); 2345 } 2346 return (error); 2347 } 2348