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/stat.h> 55 #include <sys/bio.h> 56 #include <sys/buf.h> 57 #include <sys/endian.h> 58 #include <sys/priv.h> 59 #include <sys/rwlock.h> 60 #include <sys/mount.h> 61 #include <sys/unistd.h> 62 #include <sys/time.h> 63 #include <sys/vnode.h> 64 #include <sys/namei.h> 65 #include <sys/lockf.h> 66 #include <sys/event.h> 67 #include <sys/conf.h> 68 #include <sys/file.h> 69 #include <sys/extattr.h> 70 #include <sys/vmmeter.h> 71 72 #include <vm/vm.h> 73 #include <vm/vm_param.h> 74 #include <vm/vm_extern.h> 75 #include <vm/vm_object.h> 76 #include <vm/vm_page.h> 77 #include <vm/vm_pager.h> 78 #include <vm/vnode_pager.h> 79 80 #include "opt_directio.h" 81 82 #include <ufs/ufs/dir.h> 83 84 #include <fs/ext2fs/fs.h> 85 #include <fs/ext2fs/inode.h> 86 #include <fs/ext2fs/ext2_acl.h> 87 #include <fs/ext2fs/ext2fs.h> 88 #include <fs/ext2fs/ext2_extern.h> 89 #include <fs/ext2fs/ext2_dinode.h> 90 #include <fs/ext2fs/ext2_dir.h> 91 #include <fs/ext2fs/ext2_mount.h> 92 #include <fs/ext2fs/ext2_extattr.h> 93 #include <fs/ext2fs/ext2_extents.h> 94 95 static int ext2_makeinode(int mode, struct vnode *, struct vnode **, struct componentname *); 96 static void ext2_itimes_locked(struct vnode *); 97 98 static vop_access_t ext2_access; 99 static int ext2_chmod(struct vnode *, int, struct ucred *, struct thread *); 100 static int ext2_chown(struct vnode *, uid_t, gid_t, struct ucred *, 101 struct thread *); 102 static vop_close_t ext2_close; 103 static vop_create_t ext2_create; 104 static vop_fsync_t ext2_fsync; 105 static vop_getattr_t ext2_getattr; 106 static vop_ioctl_t ext2_ioctl; 107 static vop_link_t ext2_link; 108 static vop_mkdir_t ext2_mkdir; 109 static vop_mknod_t ext2_mknod; 110 static vop_open_t ext2_open; 111 static vop_pathconf_t ext2_pathconf; 112 static vop_print_t ext2_print; 113 static vop_read_t ext2_read; 114 static vop_readlink_t ext2_readlink; 115 static vop_remove_t ext2_remove; 116 static vop_rename_t ext2_rename; 117 static vop_rmdir_t ext2_rmdir; 118 static vop_setattr_t ext2_setattr; 119 static vop_strategy_t ext2_strategy; 120 static vop_symlink_t ext2_symlink; 121 static vop_write_t ext2_write; 122 static vop_deleteextattr_t ext2_deleteextattr; 123 static vop_getextattr_t ext2_getextattr; 124 static vop_listextattr_t ext2_listextattr; 125 static vop_setextattr_t ext2_setextattr; 126 static vop_vptofh_t ext2_vptofh; 127 static vop_close_t ext2fifo_close; 128 static vop_kqfilter_t ext2fifo_kqfilter; 129 130 /* Global vfs data structures for ext2. */ 131 struct vop_vector ext2_vnodeops = { 132 .vop_default = &default_vnodeops, 133 .vop_access = ext2_access, 134 .vop_bmap = ext2_bmap, 135 .vop_cachedlookup = ext2_lookup, 136 .vop_close = ext2_close, 137 .vop_create = ext2_create, 138 .vop_fsync = ext2_fsync, 139 .vop_getpages = vnode_pager_local_getpages, 140 .vop_getpages_async = vnode_pager_local_getpages_async, 141 .vop_getattr = ext2_getattr, 142 .vop_inactive = ext2_inactive, 143 .vop_ioctl = ext2_ioctl, 144 .vop_link = ext2_link, 145 .vop_lookup = vfs_cache_lookup, 146 .vop_mkdir = ext2_mkdir, 147 .vop_mknod = ext2_mknod, 148 .vop_open = ext2_open, 149 .vop_pathconf = ext2_pathconf, 150 .vop_poll = vop_stdpoll, 151 .vop_print = ext2_print, 152 .vop_read = ext2_read, 153 .vop_readdir = ext2_readdir, 154 .vop_readlink = ext2_readlink, 155 .vop_reallocblks = ext2_reallocblks, 156 .vop_reclaim = ext2_reclaim, 157 .vop_remove = ext2_remove, 158 .vop_rename = ext2_rename, 159 .vop_rmdir = ext2_rmdir, 160 .vop_setattr = ext2_setattr, 161 .vop_strategy = ext2_strategy, 162 .vop_symlink = ext2_symlink, 163 .vop_write = ext2_write, 164 .vop_deleteextattr = ext2_deleteextattr, 165 .vop_getextattr = ext2_getextattr, 166 .vop_listextattr = ext2_listextattr, 167 .vop_setextattr = ext2_setextattr, 168 #ifdef UFS_ACL 169 .vop_getacl = ext2_getacl, 170 .vop_setacl = ext2_setacl, 171 .vop_aclcheck = ext2_aclcheck, 172 #endif /* UFS_ACL */ 173 .vop_vptofh = ext2_vptofh, 174 }; 175 176 struct vop_vector ext2_fifoops = { 177 .vop_default = &fifo_specops, 178 .vop_access = ext2_access, 179 .vop_close = ext2fifo_close, 180 .vop_fsync = ext2_fsync, 181 .vop_getattr = ext2_getattr, 182 .vop_inactive = ext2_inactive, 183 .vop_kqfilter = ext2fifo_kqfilter, 184 .vop_pathconf = ext2_pathconf, 185 .vop_print = ext2_print, 186 .vop_read = VOP_PANIC, 187 .vop_reclaim = ext2_reclaim, 188 .vop_setattr = ext2_setattr, 189 .vop_write = VOP_PANIC, 190 .vop_vptofh = ext2_vptofh, 191 }; 192 193 /* 194 * A virgin directory (no blushing please). 195 * Note that the type and namlen fields are reversed relative to ext2. 196 * Also, we don't use `struct odirtemplate', since it would just cause 197 * endianness problems. 198 */ 199 static struct dirtemplate mastertemplate = { 200 0, 12, 1, EXT2_FT_DIR, ".", 201 0, DIRBLKSIZ - 12, 2, EXT2_FT_DIR, ".." 202 }; 203 static struct dirtemplate omastertemplate = { 204 0, 12, 1, EXT2_FT_UNKNOWN, ".", 205 0, DIRBLKSIZ - 12, 2, EXT2_FT_UNKNOWN, ".." 206 }; 207 208 static void 209 ext2_itimes_locked(struct vnode *vp) 210 { 211 struct inode *ip; 212 struct timespec ts; 213 214 ASSERT_VI_LOCKED(vp, __func__); 215 216 ip = VTOI(vp); 217 if ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE)) == 0) 218 return; 219 if ((vp->v_type == VBLK || vp->v_type == VCHR)) 220 ip->i_flag |= IN_LAZYMOD; 221 else 222 ip->i_flag |= IN_MODIFIED; 223 if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 224 vfs_timestamp(&ts); 225 if (ip->i_flag & IN_ACCESS) { 226 ip->i_atime = ts.tv_sec; 227 ip->i_atimensec = ts.tv_nsec; 228 } 229 if (ip->i_flag & IN_UPDATE) { 230 ip->i_mtime = ts.tv_sec; 231 ip->i_mtimensec = ts.tv_nsec; 232 ip->i_modrev++; 233 } 234 if (ip->i_flag & IN_CHANGE) { 235 ip->i_ctime = ts.tv_sec; 236 ip->i_ctimensec = ts.tv_nsec; 237 } 238 } 239 ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_UPDATE); 240 } 241 242 void 243 ext2_itimes(struct vnode *vp) 244 { 245 246 VI_LOCK(vp); 247 ext2_itimes_locked(vp); 248 VI_UNLOCK(vp); 249 } 250 251 /* 252 * Create a regular file 253 */ 254 static int 255 ext2_create(struct vop_create_args *ap) 256 { 257 int error; 258 259 error = 260 ext2_makeinode(MAKEIMODE(ap->a_vap->va_type, ap->a_vap->va_mode), 261 ap->a_dvp, ap->a_vpp, ap->a_cnp); 262 if (error != 0) 263 return (error); 264 if ((ap->a_cnp->cn_flags & MAKEENTRY) != 0) 265 cache_enter(ap->a_dvp, *ap->a_vpp, ap->a_cnp); 266 return (0); 267 } 268 269 static int 270 ext2_open(struct vop_open_args *ap) 271 { 272 273 if (ap->a_vp->v_type == VBLK || ap->a_vp->v_type == VCHR) 274 return (EOPNOTSUPP); 275 276 /* 277 * Files marked append-only must be opened for appending. 278 */ 279 if ((VTOI(ap->a_vp)->i_flags & APPEND) && 280 (ap->a_mode & (FWRITE | O_APPEND)) == FWRITE) 281 return (EPERM); 282 283 vnode_create_vobject(ap->a_vp, VTOI(ap->a_vp)->i_size, ap->a_td); 284 285 return (0); 286 } 287 288 /* 289 * Close called. 290 * 291 * Update the times on the inode. 292 */ 293 static int 294 ext2_close(struct vop_close_args *ap) 295 { 296 struct vnode *vp = ap->a_vp; 297 298 VI_LOCK(vp); 299 if (vp->v_usecount > 1) 300 ext2_itimes_locked(vp); 301 VI_UNLOCK(vp); 302 return (0); 303 } 304 305 static int 306 ext2_access(struct vop_access_args *ap) 307 { 308 struct vnode *vp = ap->a_vp; 309 struct inode *ip = VTOI(vp); 310 accmode_t accmode = ap->a_accmode; 311 int error; 312 313 if (vp->v_type == VBLK || vp->v_type == VCHR) 314 return (EOPNOTSUPP); 315 316 /* 317 * Disallow write attempts on read-only file systems; 318 * unless the file is a socket, fifo, or a block or 319 * character device resident on the file system. 320 */ 321 if (accmode & VWRITE) { 322 switch (vp->v_type) { 323 case VDIR: 324 case VLNK: 325 case VREG: 326 if (vp->v_mount->mnt_flag & MNT_RDONLY) 327 return (EROFS); 328 break; 329 default: 330 break; 331 } 332 } 333 334 /* If immutable bit set, nobody gets to write it. */ 335 if ((accmode & VWRITE) && (ip->i_flags & (SF_IMMUTABLE | SF_SNAPSHOT))) 336 return (EPERM); 337 338 error = vaccess(vp->v_type, ip->i_mode, ip->i_uid, ip->i_gid, 339 ap->a_accmode, ap->a_cred, NULL); 340 return (error); 341 } 342 343 static int 344 ext2_getattr(struct vop_getattr_args *ap) 345 { 346 struct vnode *vp = ap->a_vp; 347 struct inode *ip = VTOI(vp); 348 struct vattr *vap = ap->a_vap; 349 350 ext2_itimes(vp); 351 /* 352 * Copy from inode table 353 */ 354 vap->va_fsid = dev2udev(ip->i_devvp->v_rdev); 355 vap->va_fileid = ip->i_number; 356 vap->va_mode = ip->i_mode & ~IFMT; 357 vap->va_nlink = ip->i_nlink; 358 vap->va_uid = ip->i_uid; 359 vap->va_gid = ip->i_gid; 360 vap->va_rdev = ip->i_rdev; 361 vap->va_size = ip->i_size; 362 vap->va_atime.tv_sec = ip->i_atime; 363 vap->va_atime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_atimensec : 0; 364 vap->va_mtime.tv_sec = ip->i_mtime; 365 vap->va_mtime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_mtimensec : 0; 366 vap->va_ctime.tv_sec = ip->i_ctime; 367 vap->va_ctime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_ctimensec : 0; 368 if E2DI_HAS_XTIME(ip) { 369 vap->va_birthtime.tv_sec = ip->i_birthtime; 370 vap->va_birthtime.tv_nsec = ip->i_birthnsec; 371 } 372 vap->va_flags = ip->i_flags; 373 vap->va_gen = ip->i_gen; 374 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize; 375 vap->va_bytes = dbtob((u_quad_t)ip->i_blocks); 376 vap->va_type = IFTOVT(ip->i_mode); 377 vap->va_filerev = ip->i_modrev; 378 return (0); 379 } 380 381 /* 382 * Set attribute vnode op. called from several syscalls 383 */ 384 static int 385 ext2_setattr(struct vop_setattr_args *ap) 386 { 387 struct vattr *vap = ap->a_vap; 388 struct vnode *vp = ap->a_vp; 389 struct inode *ip = VTOI(vp); 390 struct ucred *cred = ap->a_cred; 391 struct thread *td = curthread; 392 int error; 393 394 /* 395 * Check for unsettable attributes. 396 */ 397 if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) || 398 (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) || 399 (vap->va_blocksize != VNOVAL) || (vap->va_rdev != VNOVAL) || 400 ((int)vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) { 401 return (EINVAL); 402 } 403 if (vap->va_flags != VNOVAL) { 404 /* Disallow flags not supported by ext2fs. */ 405 if (vap->va_flags & ~(SF_APPEND | SF_IMMUTABLE | UF_NODUMP)) 406 return (EOPNOTSUPP); 407 408 if (vp->v_mount->mnt_flag & MNT_RDONLY) 409 return (EROFS); 410 /* 411 * Callers may only modify the file flags on objects they 412 * have VADMIN rights for. 413 */ 414 if ((error = VOP_ACCESS(vp, VADMIN, cred, td))) 415 return (error); 416 /* 417 * Unprivileged processes and privileged processes in 418 * jail() are not permitted to unset system flags, or 419 * modify flags if any system flags are set. 420 * Privileged non-jail processes may not modify system flags 421 * if securelevel > 0 and any existing system flags are set. 422 */ 423 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) { 424 if (ip->i_flags & (SF_IMMUTABLE | SF_APPEND)) { 425 error = securelevel_gt(cred, 0); 426 if (error) 427 return (error); 428 } 429 } else { 430 if (ip->i_flags & (SF_IMMUTABLE | SF_APPEND) || 431 ((vap->va_flags ^ ip->i_flags) & SF_SETTABLE)) 432 return (EPERM); 433 } 434 ip->i_flags = vap->va_flags; 435 ip->i_flag |= IN_CHANGE; 436 if (ip->i_flags & (IMMUTABLE | APPEND)) 437 return (0); 438 } 439 if (ip->i_flags & (IMMUTABLE | APPEND)) 440 return (EPERM); 441 /* 442 * Go through the fields and update iff not VNOVAL. 443 */ 444 if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) { 445 if (vp->v_mount->mnt_flag & MNT_RDONLY) 446 return (EROFS); 447 if ((error = ext2_chown(vp, vap->va_uid, vap->va_gid, cred, 448 td)) != 0) 449 return (error); 450 } 451 if (vap->va_size != VNOVAL) { 452 /* 453 * Disallow write attempts on read-only file systems; 454 * unless the file is a socket, fifo, or a block or 455 * character device resident on the file system. 456 */ 457 switch (vp->v_type) { 458 case VDIR: 459 return (EISDIR); 460 case VLNK: 461 case VREG: 462 if (vp->v_mount->mnt_flag & MNT_RDONLY) 463 return (EROFS); 464 break; 465 default: 466 break; 467 } 468 if ((error = ext2_truncate(vp, vap->va_size, 0, cred, td)) != 0) 469 return (error); 470 } 471 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) { 472 if (vp->v_mount->mnt_flag & MNT_RDONLY) 473 return (EROFS); 474 /* 475 * From utimes(2): 476 * If times is NULL, ... The caller must be the owner of 477 * the file, have permission to write the file, or be the 478 * super-user. 479 * If times is non-NULL, ... The caller must be the owner of 480 * the file or be the super-user. 481 */ 482 if ((error = VOP_ACCESS(vp, VADMIN, cred, td)) && 483 ((vap->va_vaflags & VA_UTIMES_NULL) == 0 || 484 (error = VOP_ACCESS(vp, VWRITE, cred, td)))) 485 return (error); 486 ip->i_flag |= IN_CHANGE | IN_MODIFIED; 487 if (vap->va_atime.tv_sec != VNOVAL) { 488 ip->i_flag &= ~IN_ACCESS; 489 ip->i_atime = vap->va_atime.tv_sec; 490 ip->i_atimensec = vap->va_atime.tv_nsec; 491 } 492 if (vap->va_mtime.tv_sec != VNOVAL) { 493 ip->i_flag &= ~IN_UPDATE; 494 ip->i_mtime = vap->va_mtime.tv_sec; 495 ip->i_mtimensec = vap->va_mtime.tv_nsec; 496 } 497 ip->i_birthtime = vap->va_birthtime.tv_sec; 498 ip->i_birthnsec = vap->va_birthtime.tv_nsec; 499 error = ext2_update(vp, 0); 500 if (error) 501 return (error); 502 } 503 error = 0; 504 if (vap->va_mode != (mode_t)VNOVAL) { 505 if (vp->v_mount->mnt_flag & MNT_RDONLY) 506 return (EROFS); 507 error = ext2_chmod(vp, (int)vap->va_mode, cred, td); 508 } 509 return (error); 510 } 511 512 /* 513 * Change the mode on a file. 514 * Inode must be locked before calling. 515 */ 516 static int 517 ext2_chmod(struct vnode *vp, int mode, struct ucred *cred, struct thread *td) 518 { 519 struct inode *ip = VTOI(vp); 520 int error; 521 522 /* 523 * To modify the permissions on a file, must possess VADMIN 524 * for that file. 525 */ 526 if ((error = VOP_ACCESS(vp, VADMIN, cred, td))) 527 return (error); 528 /* 529 * Privileged processes may set the sticky bit on non-directories, 530 * as well as set the setgid bit on a file with a group that the 531 * process is not a member of. 532 */ 533 if (vp->v_type != VDIR && (mode & S_ISTXT)) { 534 error = priv_check_cred(cred, PRIV_VFS_STICKYFILE, 0); 535 if (error) 536 return (EFTYPE); 537 } 538 if (!groupmember(ip->i_gid, cred) && (mode & ISGID)) { 539 error = priv_check_cred(cred, PRIV_VFS_SETGID, 0); 540 if (error) 541 return (error); 542 } 543 ip->i_mode &= ~ALLPERMS; 544 ip->i_mode |= (mode & ALLPERMS); 545 ip->i_flag |= IN_CHANGE; 546 return (0); 547 } 548 549 /* 550 * Perform chown operation on inode ip; 551 * inode must be locked prior to call. 552 */ 553 static int 554 ext2_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred, 555 struct thread *td) 556 { 557 struct inode *ip = VTOI(vp); 558 uid_t ouid; 559 gid_t ogid; 560 int error = 0; 561 562 if (uid == (uid_t)VNOVAL) 563 uid = ip->i_uid; 564 if (gid == (gid_t)VNOVAL) 565 gid = ip->i_gid; 566 /* 567 * To modify the ownership of a file, must possess VADMIN 568 * for that file. 569 */ 570 if ((error = VOP_ACCESS(vp, VADMIN, cred, td))) 571 return (error); 572 /* 573 * To change the owner of a file, or change the group of a file 574 * to a group of which we are not a member, the caller must 575 * have privilege. 576 */ 577 if (uid != ip->i_uid || (gid != ip->i_gid && 578 !groupmember(gid, cred))) { 579 error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0); 580 if (error) 581 return (error); 582 } 583 ogid = ip->i_gid; 584 ouid = ip->i_uid; 585 ip->i_gid = gid; 586 ip->i_uid = uid; 587 ip->i_flag |= IN_CHANGE; 588 if ((ip->i_mode & (ISUID | ISGID)) && (ouid != uid || ogid != gid)) { 589 if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID, 0) != 0) 590 ip->i_mode &= ~(ISUID | ISGID); 591 } 592 return (0); 593 } 594 595 /* 596 * Synch an open file. 597 */ 598 /* ARGSUSED */ 599 static int 600 ext2_fsync(struct vop_fsync_args *ap) 601 { 602 /* 603 * Flush all dirty buffers associated with a vnode. 604 */ 605 606 vop_stdfsync(ap); 607 608 return (ext2_update(ap->a_vp, ap->a_waitfor == MNT_WAIT)); 609 } 610 611 /* 612 * Mknod vnode call 613 */ 614 /* ARGSUSED */ 615 static int 616 ext2_mknod(struct vop_mknod_args *ap) 617 { 618 struct vattr *vap = ap->a_vap; 619 struct vnode **vpp = ap->a_vpp; 620 struct inode *ip; 621 ino_t ino; 622 int error; 623 624 error = ext2_makeinode(MAKEIMODE(vap->va_type, vap->va_mode), 625 ap->a_dvp, vpp, ap->a_cnp); 626 if (error) 627 return (error); 628 ip = VTOI(*vpp); 629 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 630 if (vap->va_rdev != VNOVAL) { 631 /* 632 * Want to be able to use this to make badblock 633 * inodes, so don't truncate the dev number. 634 */ 635 if (!(ip->i_flag & IN_E4EXTENTS)) 636 ip->i_rdev = vap->va_rdev; 637 } 638 /* 639 * Remove inode, then reload it through VFS_VGET so it is 640 * checked to see if it is an alias of an existing entry in 641 * the inode cache. XXX I don't believe this is necessary now. 642 */ 643 (*vpp)->v_type = VNON; 644 ino = ip->i_number; /* Save this before vgone() invalidates ip. */ 645 vgone(*vpp); 646 vput(*vpp); 647 error = VFS_VGET(ap->a_dvp->v_mount, ino, LK_EXCLUSIVE, vpp); 648 if (error) { 649 *vpp = NULL; 650 return (error); 651 } 652 return (0); 653 } 654 655 static int 656 ext2_remove(struct vop_remove_args *ap) 657 { 658 struct inode *ip; 659 struct vnode *vp = ap->a_vp; 660 struct vnode *dvp = ap->a_dvp; 661 int error; 662 663 ip = VTOI(vp); 664 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) || 665 (VTOI(dvp)->i_flags & APPEND)) { 666 error = EPERM; 667 goto out; 668 } 669 error = ext2_dirremove(dvp, ap->a_cnp); 670 if (error == 0) { 671 ip->i_nlink--; 672 ip->i_flag |= IN_CHANGE; 673 } 674 out: 675 return (error); 676 } 677 678 static unsigned short 679 ext2_max_nlink(struct inode *ip) 680 { 681 struct m_ext2fs *fs; 682 683 fs = ip->i_e2fs; 684 685 if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_DIR_NLINK)) 686 return (EXT4_LINK_MAX); 687 else 688 return (EXT2_LINK_MAX); 689 } 690 691 /* 692 * link vnode call 693 */ 694 static int 695 ext2_link(struct vop_link_args *ap) 696 { 697 struct vnode *vp = ap->a_vp; 698 struct vnode *tdvp = ap->a_tdvp; 699 struct componentname *cnp = ap->a_cnp; 700 struct inode *ip; 701 int error; 702 703 #ifdef INVARIANTS 704 if ((cnp->cn_flags & HASBUF) == 0) 705 panic("ext2_link: no name"); 706 #endif 707 ip = VTOI(vp); 708 if ((nlink_t)ip->i_nlink >= ext2_max_nlink(ip)) { 709 error = EMLINK; 710 goto out; 711 } 712 if (ip->i_flags & (IMMUTABLE | APPEND)) { 713 error = EPERM; 714 goto out; 715 } 716 ip->i_nlink++; 717 ip->i_flag |= IN_CHANGE; 718 error = ext2_update(vp, !DOINGASYNC(vp)); 719 if (!error) 720 error = ext2_direnter(ip, tdvp, cnp); 721 if (error) { 722 ip->i_nlink--; 723 ip->i_flag |= IN_CHANGE; 724 } 725 out: 726 return (error); 727 } 728 729 static int 730 ext2_inc_nlink(struct inode *ip) 731 { 732 733 ip->i_nlink++; 734 735 if (ext2_htree_has_idx(ip) && ip->i_nlink > 1) { 736 if (ip->i_nlink >= ext2_max_nlink(ip) || ip->i_nlink == 2) 737 ip->i_nlink = 1; 738 } else if (ip->i_nlink > ext2_max_nlink(ip)) { 739 ip->i_nlink--; 740 return (EMLINK); 741 } 742 743 return (0); 744 } 745 746 static void 747 ext2_dec_nlink(struct inode *ip) 748 { 749 750 if (!S_ISDIR(ip->i_mode) || ip->i_nlink > 2) 751 ip->i_nlink--; 752 } 753 754 /* 755 * Rename system call. 756 * rename("foo", "bar"); 757 * is essentially 758 * unlink("bar"); 759 * link("foo", "bar"); 760 * unlink("foo"); 761 * but ``atomically''. Can't do full commit without saving state in the 762 * inode on disk which isn't feasible at this time. Best we can do is 763 * always guarantee the target exists. 764 * 765 * Basic algorithm is: 766 * 767 * 1) Bump link count on source while we're linking it to the 768 * target. This also ensure the inode won't be deleted out 769 * from underneath us while we work (it may be truncated by 770 * a concurrent `trunc' or `open' for creation). 771 * 2) Link source to destination. If destination already exists, 772 * delete it first. 773 * 3) Unlink source reference to inode if still around. If a 774 * directory was moved and the parent of the destination 775 * is different from the source, patch the ".." entry in the 776 * directory. 777 */ 778 static int 779 ext2_rename(struct vop_rename_args *ap) 780 { 781 struct vnode *tvp = ap->a_tvp; 782 struct vnode *tdvp = ap->a_tdvp; 783 struct vnode *fvp = ap->a_fvp; 784 struct vnode *fdvp = ap->a_fdvp; 785 struct componentname *tcnp = ap->a_tcnp; 786 struct componentname *fcnp = ap->a_fcnp; 787 struct inode *ip, *xp, *dp; 788 struct dirtemplate *dirbuf; 789 int doingdirectory = 0, oldparent = 0, newparent = 0; 790 int error = 0; 791 u_char namlen; 792 793 #ifdef INVARIANTS 794 if ((tcnp->cn_flags & HASBUF) == 0 || 795 (fcnp->cn_flags & HASBUF) == 0) 796 panic("ext2_rename: no name"); 797 #endif 798 /* 799 * Check for cross-device rename. 800 */ 801 if ((fvp->v_mount != tdvp->v_mount) || 802 (tvp && (fvp->v_mount != tvp->v_mount))) { 803 error = EXDEV; 804 abortit: 805 if (tdvp == tvp) 806 vrele(tdvp); 807 else 808 vput(tdvp); 809 if (tvp) 810 vput(tvp); 811 vrele(fdvp); 812 vrele(fvp); 813 return (error); 814 } 815 816 if (tvp && ((VTOI(tvp)->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) || 817 (VTOI(tdvp)->i_flags & APPEND))) { 818 error = EPERM; 819 goto abortit; 820 } 821 822 /* 823 * Renaming a file to itself has no effect. The upper layers should 824 * not call us in that case. Temporarily just warn if they do. 825 */ 826 if (fvp == tvp) { 827 printf("ext2_rename: fvp == tvp (can't happen)\n"); 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 >= ext2_max_nlink(ip) && !ext2_htree_has_idx(ip)) { 837 VOP_UNLOCK(fvp, 0); 838 error = EMLINK; 839 goto abortit; 840 } 841 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) 842 || (dp->i_flags & APPEND)) { 843 VOP_UNLOCK(fvp, 0); 844 error = EPERM; 845 goto abortit; 846 } 847 if ((ip->i_mode & IFMT) == IFDIR) { 848 /* 849 * Avoid ".", "..", and aliases of "." for obvious reasons. 850 */ 851 if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') || 852 dp == ip || (fcnp->cn_flags | tcnp->cn_flags) & ISDOTDOT || 853 (ip->i_flag & IN_RENAME)) { 854 VOP_UNLOCK(fvp, 0); 855 error = EINVAL; 856 goto abortit; 857 } 858 ip->i_flag |= IN_RENAME; 859 oldparent = dp->i_number; 860 doingdirectory++; 861 } 862 vrele(fdvp); 863 864 /* 865 * When the target exists, both the directory 866 * and target vnodes are returned locked. 867 */ 868 dp = VTOI(tdvp); 869 xp = NULL; 870 if (tvp) 871 xp = VTOI(tvp); 872 873 /* 874 * 1) Bump link count while we're moving stuff 875 * around. If we crash somewhere before 876 * completing our work, the link count 877 * may be wrong, but correctable. 878 */ 879 ext2_inc_nlink(ip); 880 ip->i_flag |= IN_CHANGE; 881 if ((error = ext2_update(fvp, !DOINGASYNC(fvp))) != 0) { 882 VOP_UNLOCK(fvp, 0); 883 goto bad; 884 } 885 886 /* 887 * If ".." must be changed (ie the directory gets a new 888 * parent) then the source directory must not be in the 889 * directory hierarchy above the target, as this would 890 * orphan everything below the source directory. Also 891 * the user must have write permission in the source so 892 * as to be able to change "..". We must repeat the call 893 * to namei, as the parent directory is unlocked by the 894 * call to checkpath(). 895 */ 896 error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, tcnp->cn_thread); 897 VOP_UNLOCK(fvp, 0); 898 if (oldparent != dp->i_number) 899 newparent = dp->i_number; 900 if (doingdirectory && newparent) { 901 if (error) /* write access check above */ 902 goto bad; 903 if (xp != NULL) 904 vput(tvp); 905 error = ext2_checkpath(ip, dp, tcnp->cn_cred); 906 if (error) 907 goto out; 908 VREF(tdvp); 909 error = relookup(tdvp, &tvp, tcnp); 910 if (error) 911 goto out; 912 vrele(tdvp); 913 dp = VTOI(tdvp); 914 xp = NULL; 915 if (tvp) 916 xp = VTOI(tvp); 917 } 918 /* 919 * 2) If target doesn't exist, link the target 920 * to the source and unlink the source. 921 * Otherwise, rewrite the target directory 922 * entry to reference the source inode and 923 * expunge the original entry's existence. 924 */ 925 if (xp == NULL) { 926 if (dp->i_devvp != ip->i_devvp) 927 panic("ext2_rename: EXDEV"); 928 /* 929 * Account for ".." in new directory. 930 * When source and destination have the same 931 * parent we don't fool with the link count. 932 */ 933 if (doingdirectory && newparent) { 934 error = ext2_inc_nlink(dp); 935 if (error) 936 goto bad; 937 938 dp->i_flag |= IN_CHANGE; 939 error = ext2_update(tdvp, !DOINGASYNC(tdvp)); 940 if (error) 941 goto bad; 942 } 943 error = ext2_direnter(ip, tdvp, tcnp); 944 if (error) { 945 if (doingdirectory && newparent) { 946 ext2_dec_nlink(dp); 947 dp->i_flag |= IN_CHANGE; 948 (void)ext2_update(tdvp, 1); 949 } 950 goto bad; 951 } 952 vput(tdvp); 953 } else { 954 if (xp->i_devvp != dp->i_devvp || xp->i_devvp != ip->i_devvp) 955 panic("ext2_rename: EXDEV"); 956 /* 957 * Short circuit rename(foo, foo). 958 */ 959 if (xp->i_number == ip->i_number) 960 panic("ext2_rename: same file"); 961 /* 962 * If the parent directory is "sticky", then the user must 963 * own the parent directory, or the destination of the rename, 964 * otherwise the destination may not be changed (except by 965 * root). This implements append-only directories. 966 */ 967 if ((dp->i_mode & S_ISTXT) && tcnp->cn_cred->cr_uid != 0 && 968 tcnp->cn_cred->cr_uid != dp->i_uid && 969 xp->i_uid != tcnp->cn_cred->cr_uid) { 970 error = EPERM; 971 goto bad; 972 } 973 /* 974 * Target must be empty if a directory and have no links 975 * to it. Also, ensure source and target are compatible 976 * (both directories, or both not directories). 977 */ 978 if ((xp->i_mode & IFMT) == IFDIR) { 979 if (!ext2_dirempty(xp, dp->i_number, tcnp->cn_cred)) { 980 error = ENOTEMPTY; 981 goto bad; 982 } 983 if (!doingdirectory) { 984 error = ENOTDIR; 985 goto bad; 986 } 987 cache_purge(tdvp); 988 } else if (doingdirectory) { 989 error = EISDIR; 990 goto bad; 991 } 992 error = ext2_dirrewrite(dp, ip, tcnp); 993 if (error) 994 goto bad; 995 /* 996 * If the target directory is in the same 997 * directory as the source directory, 998 * decrement the link count on the parent 999 * of the target directory. 1000 */ 1001 if (doingdirectory && !newparent) { 1002 ext2_dec_nlink(dp); 1003 dp->i_flag |= IN_CHANGE; 1004 } 1005 vput(tdvp); 1006 /* 1007 * Adjust the link count of the target to 1008 * reflect the dirrewrite above. If this is 1009 * a directory it is empty and there are 1010 * no links to it, so we can squash the inode and 1011 * any space associated with it. We disallowed 1012 * renaming over top of a directory with links to 1013 * it above, as the remaining link would point to 1014 * a directory without "." or ".." entries. 1015 */ 1016 ext2_dec_nlink(xp); 1017 if (doingdirectory) { 1018 if (--xp->i_nlink != 0) 1019 panic("ext2_rename: linked directory"); 1020 error = ext2_truncate(tvp, (off_t)0, IO_SYNC, 1021 tcnp->cn_cred, tcnp->cn_thread); 1022 } 1023 xp->i_flag |= IN_CHANGE; 1024 vput(tvp); 1025 xp = NULL; 1026 } 1027 1028 /* 1029 * 3) Unlink the source. 1030 */ 1031 fcnp->cn_flags &= ~MODMASK; 1032 fcnp->cn_flags |= LOCKPARENT | LOCKLEAF; 1033 VREF(fdvp); 1034 error = relookup(fdvp, &fvp, fcnp); 1035 if (error == 0) 1036 vrele(fdvp); 1037 if (fvp != NULL) { 1038 xp = VTOI(fvp); 1039 dp = VTOI(fdvp); 1040 } else { 1041 /* 1042 * From name has disappeared. IN_RENAME is not sufficient 1043 * to protect against directory races due to timing windows, 1044 * so we can't panic here. 1045 */ 1046 vrele(ap->a_fvp); 1047 return (0); 1048 } 1049 /* 1050 * Ensure that the directory entry still exists and has not 1051 * changed while the new name has been entered. If the source is 1052 * a file then the entry may have been unlinked or renamed. In 1053 * either case there is no further work to be done. If the source 1054 * is a directory then it cannot have been rmdir'ed; its link 1055 * count of three would cause a rmdir to fail with ENOTEMPTY. 1056 * The IN_RENAME flag ensures that it cannot be moved by another 1057 * rename. 1058 */ 1059 if (xp != ip) { 1060 /* 1061 * From name resolves to a different inode. IN_RENAME is 1062 * not sufficient protection against timing window races 1063 * so we can't panic here. 1064 */ 1065 } else { 1066 /* 1067 * If the source is a directory with a 1068 * new parent, the link count of the old 1069 * parent directory must be decremented 1070 * and ".." set to point to the new parent. 1071 */ 1072 if (doingdirectory && newparent) { 1073 ext2_dec_nlink(dp); 1074 dp->i_flag |= IN_CHANGE; 1075 dirbuf = malloc(dp->i_e2fs->e2fs_bsize, M_TEMP, M_WAITOK | M_ZERO); 1076 if (!dirbuf) { 1077 error = ENOMEM; 1078 goto bad; 1079 } 1080 error = vn_rdwr(UIO_READ, fvp, (caddr_t)dirbuf, 1081 ip->i_e2fs->e2fs_bsize, (off_t)0, 1082 UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK, 1083 tcnp->cn_cred, NOCRED, NULL, NULL); 1084 if (error == 0) { 1085 /* Like ufs little-endian: */ 1086 namlen = dirbuf->dotdot_type; 1087 if (namlen != 2 || 1088 dirbuf->dotdot_name[0] != '.' || 1089 dirbuf->dotdot_name[1] != '.') { 1090 ext2_dirbad(xp, (doff_t)12, 1091 "rename: mangled dir"); 1092 } else { 1093 dirbuf->dotdot_ino = newparent; 1094 ext2_dirent_csum_set(ip, 1095 (struct ext2fs_direct_2 *)dirbuf); 1096 (void)vn_rdwr(UIO_WRITE, fvp, 1097 (caddr_t)dirbuf, 1098 ip->i_e2fs->e2fs_bsize, 1099 (off_t)0, UIO_SYSSPACE, 1100 IO_NODELOCKED | IO_SYNC | 1101 IO_NOMACCHECK, tcnp->cn_cred, 1102 NOCRED, NULL, NULL); 1103 cache_purge(fdvp); 1104 } 1105 } 1106 free(dirbuf, M_TEMP); 1107 } 1108 error = ext2_dirremove(fdvp, fcnp); 1109 if (!error) { 1110 ext2_dec_nlink(xp); 1111 xp->i_flag |= IN_CHANGE; 1112 } 1113 xp->i_flag &= ~IN_RENAME; 1114 } 1115 if (dp) 1116 vput(fdvp); 1117 if (xp) 1118 vput(fvp); 1119 vrele(ap->a_fvp); 1120 return (error); 1121 1122 bad: 1123 if (xp) 1124 vput(ITOV(xp)); 1125 vput(ITOV(dp)); 1126 out: 1127 if (doingdirectory) 1128 ip->i_flag &= ~IN_RENAME; 1129 if (vn_lock(fvp, LK_EXCLUSIVE) == 0) { 1130 ext2_dec_nlink(ip); 1131 ip->i_flag |= IN_CHANGE; 1132 ip->i_flag &= ~IN_RENAME; 1133 vput(fvp); 1134 } else 1135 vrele(fvp); 1136 return (error); 1137 } 1138 1139 #ifdef UFS_ACL 1140 static int 1141 ext2_do_posix1e_acl_inheritance_dir(struct vnode *dvp, struct vnode *tvp, 1142 mode_t dmode, struct ucred *cred, struct thread *td) 1143 { 1144 int error; 1145 struct inode *ip = VTOI(tvp); 1146 struct acl *dacl, *acl; 1147 1148 acl = acl_alloc(M_WAITOK); 1149 dacl = acl_alloc(M_WAITOK); 1150 1151 /* 1152 * Retrieve default ACL from parent, if any. 1153 */ 1154 error = VOP_GETACL(dvp, ACL_TYPE_DEFAULT, acl, cred, td); 1155 switch (error) { 1156 case 0: 1157 /* 1158 * Retrieved a default ACL, so merge mode and ACL if 1159 * necessary. If the ACL is empty, fall through to 1160 * the "not defined or available" case. 1161 */ 1162 if (acl->acl_cnt != 0) { 1163 dmode = acl_posix1e_newfilemode(dmode, acl); 1164 ip->i_mode = dmode; 1165 *dacl = *acl; 1166 ext2_sync_acl_from_inode(ip, acl); 1167 break; 1168 } 1169 /* FALLTHROUGH */ 1170 1171 case EOPNOTSUPP: 1172 /* 1173 * Just use the mode as-is. 1174 */ 1175 ip->i_mode = dmode; 1176 error = 0; 1177 goto out; 1178 1179 default: 1180 goto out; 1181 } 1182 1183 error = VOP_SETACL(tvp, ACL_TYPE_ACCESS, acl, cred, td); 1184 if (error == 0) 1185 error = VOP_SETACL(tvp, ACL_TYPE_DEFAULT, dacl, cred, td); 1186 switch (error) { 1187 case 0: 1188 break; 1189 1190 case EOPNOTSUPP: 1191 /* 1192 * XXX: This should not happen, as EOPNOTSUPP above 1193 * was supposed to free acl. 1194 */ 1195 #ifdef DEBUG 1196 printf("ext2_mkdir: VOP_GETACL() but no VOP_SETACL()\n"); 1197 #endif /* DEBUG */ 1198 break; 1199 1200 default: 1201 goto out; 1202 } 1203 1204 out: 1205 acl_free(acl); 1206 acl_free(dacl); 1207 1208 return (error); 1209 } 1210 1211 static int 1212 ext2_do_posix1e_acl_inheritance_file(struct vnode *dvp, struct vnode *tvp, 1213 mode_t mode, struct ucred *cred, struct thread *td) 1214 { 1215 int error; 1216 struct inode *ip = VTOI(tvp); 1217 struct acl *acl; 1218 1219 acl = acl_alloc(M_WAITOK); 1220 1221 /* 1222 * Retrieve default ACL for parent, if any. 1223 */ 1224 error = VOP_GETACL(dvp, ACL_TYPE_DEFAULT, acl, cred, td); 1225 switch (error) { 1226 case 0: 1227 /* 1228 * Retrieved a default ACL, so merge mode and ACL if 1229 * necessary. 1230 */ 1231 if (acl->acl_cnt != 0) { 1232 /* 1233 * Two possible ways for default ACL to not 1234 * be present. First, the EA can be 1235 * undefined, or second, the default ACL can 1236 * be blank. If it's blank, fall through to 1237 * the it's not defined case. 1238 */ 1239 mode = acl_posix1e_newfilemode(mode, acl); 1240 ip->i_mode = mode; 1241 ext2_sync_acl_from_inode(ip, acl); 1242 break; 1243 } 1244 /* FALLTHROUGH */ 1245 1246 case EOPNOTSUPP: 1247 /* 1248 * Just use the mode as-is. 1249 */ 1250 ip->i_mode = mode; 1251 error = 0; 1252 goto out; 1253 1254 default: 1255 goto out; 1256 } 1257 1258 error = VOP_SETACL(tvp, ACL_TYPE_ACCESS, acl, cred, td); 1259 switch (error) { 1260 case 0: 1261 break; 1262 1263 case EOPNOTSUPP: 1264 /* 1265 * XXX: This should not happen, as EOPNOTSUPP above was 1266 * supposed to free acl. 1267 */ 1268 printf("ufs_do_posix1e_acl_inheritance_file: VOP_GETACL() " 1269 "but no VOP_SETACL()\n"); 1270 /* panic("ufs_do_posix1e_acl_inheritance_file: VOP_GETACL() " 1271 "but no VOP_SETACL()"); */ 1272 break; 1273 1274 default: 1275 goto out; 1276 } 1277 1278 out: 1279 acl_free(acl); 1280 1281 return (error); 1282 } 1283 1284 #endif /* UFS_ACL */ 1285 1286 /* 1287 * Mkdir system call 1288 */ 1289 static int 1290 ext2_mkdir(struct vop_mkdir_args *ap) 1291 { 1292 struct m_ext2fs *fs; 1293 struct vnode *dvp = ap->a_dvp; 1294 struct vattr *vap = ap->a_vap; 1295 struct componentname *cnp = ap->a_cnp; 1296 struct inode *ip, *dp; 1297 struct vnode *tvp; 1298 struct dirtemplate dirtemplate, *dtp; 1299 char *buf = NULL; 1300 int error, dmode; 1301 1302 #ifdef INVARIANTS 1303 if ((cnp->cn_flags & HASBUF) == 0) 1304 panic("ext2_mkdir: no name"); 1305 #endif 1306 dp = VTOI(dvp); 1307 if ((nlink_t)dp->i_nlink >= ext2_max_nlink(dp) && 1308 !ext2_htree_has_idx(dp)) { 1309 error = EMLINK; 1310 goto out; 1311 } 1312 dmode = vap->va_mode & 0777; 1313 dmode |= IFDIR; 1314 /* 1315 * Must simulate part of ext2_makeinode here to acquire the inode, 1316 * but not have it entered in the parent directory. The entry is 1317 * made later after writing "." and ".." entries. 1318 */ 1319 error = ext2_valloc(dvp, dmode, cnp->cn_cred, &tvp); 1320 if (error) 1321 goto out; 1322 ip = VTOI(tvp); 1323 fs = ip->i_e2fs; 1324 ip->i_gid = dp->i_gid; 1325 #ifdef SUIDDIR 1326 { 1327 /* 1328 * if we are hacking owners here, (only do this where told to) 1329 * and we are not giving it TOO root, (would subvert quotas) 1330 * then go ahead and give it to the other user. 1331 * The new directory also inherits the SUID bit. 1332 * If user's UID and dir UID are the same, 1333 * 'give it away' so that the SUID is still forced on. 1334 */ 1335 if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) && 1336 (dp->i_mode & ISUID) && dp->i_uid) { 1337 dmode |= ISUID; 1338 ip->i_uid = dp->i_uid; 1339 } else { 1340 ip->i_uid = cnp->cn_cred->cr_uid; 1341 } 1342 } 1343 #else 1344 ip->i_uid = cnp->cn_cred->cr_uid; 1345 #endif 1346 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 1347 ip->i_mode = dmode; 1348 tvp->v_type = VDIR; /* Rest init'd in getnewvnode(). */ 1349 ip->i_nlink = 2; 1350 if (cnp->cn_flags & ISWHITEOUT) 1351 ip->i_flags |= UF_OPAQUE; 1352 error = ext2_update(tvp, 1); 1353 1354 /* 1355 * Bump link count in parent directory 1356 * to reflect work done below. Should 1357 * be done before reference is created 1358 * so reparation is possible if we crash. 1359 */ 1360 ext2_inc_nlink(dp); 1361 dp->i_flag |= IN_CHANGE; 1362 error = ext2_update(dvp, !DOINGASYNC(dvp)); 1363 if (error) 1364 goto bad; 1365 1366 /* Initialize directory with "." and ".." from static template. */ 1367 if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs, 1368 EXT2F_INCOMPAT_FTYPE)) 1369 dtp = &mastertemplate; 1370 else 1371 dtp = &omastertemplate; 1372 dirtemplate = *dtp; 1373 dirtemplate.dot_ino = ip->i_number; 1374 dirtemplate.dotdot_ino = dp->i_number; 1375 /* 1376 * note that in ext2 DIRBLKSIZ == blocksize, not DEV_BSIZE so let's 1377 * just redefine it - for this function only 1378 */ 1379 #undef DIRBLKSIZ 1380 #define DIRBLKSIZ VTOI(dvp)->i_e2fs->e2fs_bsize 1381 dirtemplate.dotdot_reclen = DIRBLKSIZ - 12; 1382 buf = malloc(DIRBLKSIZ, M_TEMP, M_WAITOK | M_ZERO); 1383 if (!buf) { 1384 error = ENOMEM; 1385 ext2_dec_nlink(dp); 1386 dp->i_flag |= IN_CHANGE; 1387 goto bad; 1388 } 1389 if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM)) { 1390 dirtemplate.dotdot_reclen -= sizeof(struct ext2fs_direct_tail); 1391 ext2_init_dirent_tail(EXT2_DIRENT_TAIL(buf, DIRBLKSIZ)); 1392 } 1393 memcpy(buf, &dirtemplate, sizeof(dirtemplate)); 1394 ext2_dirent_csum_set(ip, (struct ext2fs_direct_2 *)buf); 1395 error = vn_rdwr(UIO_WRITE, tvp, (caddr_t)buf, 1396 DIRBLKSIZ, (off_t)0, UIO_SYSSPACE, 1397 IO_NODELOCKED | IO_SYNC | IO_NOMACCHECK, cnp->cn_cred, NOCRED, 1398 NULL, NULL); 1399 if (error) { 1400 ext2_dec_nlink(dp); 1401 dp->i_flag |= IN_CHANGE; 1402 goto bad; 1403 } 1404 if (DIRBLKSIZ > VFSTOEXT2(dvp->v_mount)->um_mountp->mnt_stat.f_bsize) 1405 /* XXX should grow with balloc() */ 1406 panic("ext2_mkdir: blksize"); 1407 else { 1408 ip->i_size = DIRBLKSIZ; 1409 ip->i_flag |= IN_CHANGE; 1410 } 1411 1412 #ifdef UFS_ACL 1413 if (dvp->v_mount->mnt_flag & MNT_ACLS) { 1414 error = ext2_do_posix1e_acl_inheritance_dir(dvp, tvp, dmode, 1415 cnp->cn_cred, cnp->cn_thread); 1416 if (error) 1417 goto bad; 1418 } 1419 1420 #endif /* UFS_ACL */ 1421 1422 /* Directory set up, now install its entry in the parent directory. */ 1423 error = ext2_direnter(ip, dvp, cnp); 1424 if (error) { 1425 ext2_dec_nlink(dp); 1426 dp->i_flag |= IN_CHANGE; 1427 } 1428 bad: 1429 /* 1430 * No need to do an explicit VOP_TRUNCATE here, vrele will do this 1431 * for us because we set the link count to 0. 1432 */ 1433 if (error) { 1434 ip->i_nlink = 0; 1435 ip->i_flag |= IN_CHANGE; 1436 vput(tvp); 1437 } else 1438 *ap->a_vpp = tvp; 1439 out: 1440 free(buf, M_TEMP); 1441 return (error); 1442 #undef DIRBLKSIZ 1443 #define DIRBLKSIZ DEV_BSIZE 1444 } 1445 1446 /* 1447 * Rmdir system call. 1448 */ 1449 static int 1450 ext2_rmdir(struct vop_rmdir_args *ap) 1451 { 1452 struct vnode *vp = ap->a_vp; 1453 struct vnode *dvp = ap->a_dvp; 1454 struct componentname *cnp = ap->a_cnp; 1455 struct inode *ip, *dp; 1456 int error; 1457 1458 ip = VTOI(vp); 1459 dp = VTOI(dvp); 1460 1461 /* 1462 * Verify the directory is empty (and valid). 1463 * (Rmdir ".." won't be valid since 1464 * ".." will contain a reference to 1465 * the current directory and thus be 1466 * non-empty.) 1467 */ 1468 if (!ext2_dirempty(ip, dp->i_number, cnp->cn_cred)) { 1469 error = ENOTEMPTY; 1470 goto out; 1471 } 1472 if ((dp->i_flags & APPEND) 1473 || (ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))) { 1474 error = EPERM; 1475 goto out; 1476 } 1477 /* 1478 * Delete reference to directory before purging 1479 * inode. If we crash in between, the directory 1480 * will be reattached to lost+found, 1481 */ 1482 error = ext2_dirremove(dvp, cnp); 1483 if (error) 1484 goto out; 1485 ext2_dec_nlink(dp); 1486 dp->i_flag |= IN_CHANGE; 1487 cache_purge(dvp); 1488 VOP_UNLOCK(dvp, 0); 1489 /* 1490 * Truncate inode. The only stuff left 1491 * in the directory is "." and "..". 1492 */ 1493 ip->i_nlink = 0; 1494 error = ext2_truncate(vp, (off_t)0, IO_SYNC, cnp->cn_cred, 1495 cnp->cn_thread); 1496 cache_purge(ITOV(ip)); 1497 if (vn_lock(dvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) { 1498 VOP_UNLOCK(vp, 0); 1499 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 1500 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1501 } 1502 out: 1503 return (error); 1504 } 1505 1506 /* 1507 * symlink -- make a symbolic link 1508 */ 1509 static int 1510 ext2_symlink(struct vop_symlink_args *ap) 1511 { 1512 struct vnode *vp, **vpp = ap->a_vpp; 1513 struct inode *ip; 1514 int len, error; 1515 1516 error = ext2_makeinode(IFLNK | ap->a_vap->va_mode, ap->a_dvp, 1517 vpp, ap->a_cnp); 1518 if (error) 1519 return (error); 1520 vp = *vpp; 1521 len = strlen(ap->a_target); 1522 if (len < vp->v_mount->mnt_maxsymlinklen) { 1523 ip = VTOI(vp); 1524 bcopy(ap->a_target, (char *)ip->i_shortlink, len); 1525 ip->i_size = len; 1526 ip->i_flag |= IN_CHANGE | IN_UPDATE; 1527 } else 1528 error = vn_rdwr(UIO_WRITE, vp, ap->a_target, len, (off_t)0, 1529 UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK, 1530 ap->a_cnp->cn_cred, NOCRED, NULL, NULL); 1531 if (error) 1532 vput(vp); 1533 return (error); 1534 } 1535 1536 /* 1537 * Return target name of a symbolic link 1538 */ 1539 static int 1540 ext2_readlink(struct vop_readlink_args *ap) 1541 { 1542 struct vnode *vp = ap->a_vp; 1543 struct inode *ip = VTOI(vp); 1544 int isize; 1545 1546 isize = ip->i_size; 1547 if (isize < vp->v_mount->mnt_maxsymlinklen) { 1548 uiomove((char *)ip->i_shortlink, isize, ap->a_uio); 1549 return (0); 1550 } 1551 return (VOP_READ(vp, ap->a_uio, 0, ap->a_cred)); 1552 } 1553 1554 /* 1555 * Calculate the logical to physical mapping if not done already, 1556 * then call the device strategy routine. 1557 * 1558 * In order to be able to swap to a file, the ext2_bmaparray() operation may not 1559 * deadlock on memory. See ext2_bmap() for details. 1560 */ 1561 static int 1562 ext2_strategy(struct vop_strategy_args *ap) 1563 { 1564 struct buf *bp = ap->a_bp; 1565 struct vnode *vp = ap->a_vp; 1566 struct bufobj *bo; 1567 daddr_t blkno; 1568 int error; 1569 1570 if (vp->v_type == VBLK || vp->v_type == VCHR) 1571 panic("ext2_strategy: spec"); 1572 if (bp->b_blkno == bp->b_lblkno) { 1573 1574 if (VTOI(ap->a_vp)->i_flag & IN_E4EXTENTS) 1575 error = ext4_bmapext(vp, bp->b_lblkno, &blkno, NULL, NULL); 1576 else 1577 error = ext2_bmaparray(vp, bp->b_lblkno, &blkno, NULL, NULL); 1578 1579 bp->b_blkno = blkno; 1580 if (error) { 1581 bp->b_error = error; 1582 bp->b_ioflags |= BIO_ERROR; 1583 bufdone(bp); 1584 return (0); 1585 } 1586 if ((long)bp->b_blkno == -1) 1587 vfs_bio_clrbuf(bp); 1588 } 1589 if ((long)bp->b_blkno == -1) { 1590 bufdone(bp); 1591 return (0); 1592 } 1593 bp->b_iooffset = dbtob(bp->b_blkno); 1594 bo = VFSTOEXT2(vp->v_mount)->um_bo; 1595 BO_STRATEGY(bo, bp); 1596 return (0); 1597 } 1598 1599 /* 1600 * Print out the contents of an inode. 1601 */ 1602 static int 1603 ext2_print(struct vop_print_args *ap) 1604 { 1605 struct vnode *vp = ap->a_vp; 1606 struct inode *ip = VTOI(vp); 1607 1608 vn_printf(ip->i_devvp, "\tino %ju", (uintmax_t)ip->i_number); 1609 if (vp->v_type == VFIFO) 1610 fifo_printinfo(vp); 1611 printf("\n"); 1612 return (0); 1613 } 1614 1615 /* 1616 * Close wrapper for fifos. 1617 * 1618 * Update the times on the inode then do device close. 1619 */ 1620 static int 1621 ext2fifo_close(struct vop_close_args *ap) 1622 { 1623 struct vnode *vp = ap->a_vp; 1624 1625 VI_LOCK(vp); 1626 if (vp->v_usecount > 1) 1627 ext2_itimes_locked(vp); 1628 VI_UNLOCK(vp); 1629 return (fifo_specops.vop_close(ap)); 1630 } 1631 1632 /* 1633 * Kqfilter wrapper for fifos. 1634 * 1635 * Fall through to ext2 kqfilter routines if needed 1636 */ 1637 static int 1638 ext2fifo_kqfilter(struct vop_kqfilter_args *ap) 1639 { 1640 int error; 1641 1642 error = fifo_specops.vop_kqfilter(ap); 1643 if (error) 1644 error = vfs_kqfilter(ap); 1645 return (error); 1646 } 1647 1648 /* 1649 * Return POSIX pathconf information applicable to ext2 filesystems. 1650 */ 1651 static int 1652 ext2_pathconf(struct vop_pathconf_args *ap) 1653 { 1654 int error = 0; 1655 1656 switch (ap->a_name) { 1657 case _PC_LINK_MAX: 1658 if (ext2_htree_has_idx(VTOI(ap->a_vp))) 1659 *ap->a_retval = INT_MAX; 1660 else 1661 *ap->a_retval = ext2_max_nlink(VTOI(ap->a_vp)); 1662 break; 1663 case _PC_NAME_MAX: 1664 *ap->a_retval = NAME_MAX; 1665 break; 1666 case _PC_PIPE_BUF: 1667 if (ap->a_vp->v_type == VDIR || ap->a_vp->v_type == VFIFO) 1668 *ap->a_retval = PIPE_BUF; 1669 else 1670 error = EINVAL; 1671 break; 1672 case _PC_CHOWN_RESTRICTED: 1673 *ap->a_retval = 1; 1674 break; 1675 case _PC_NO_TRUNC: 1676 *ap->a_retval = 1; 1677 break; 1678 1679 #ifdef UFS_ACL 1680 case _PC_ACL_EXTENDED: 1681 if (ap->a_vp->v_mount->mnt_flag & MNT_ACLS) 1682 *ap->a_retval = 1; 1683 else 1684 *ap->a_retval = 0; 1685 break; 1686 case _PC_ACL_PATH_MAX: 1687 if (ap->a_vp->v_mount->mnt_flag & MNT_ACLS) 1688 *ap->a_retval = ACL_MAX_ENTRIES; 1689 else 1690 *ap->a_retval = 3; 1691 break; 1692 #endif /* UFS_ACL */ 1693 1694 case _PC_MIN_HOLE_SIZE: 1695 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize; 1696 break; 1697 case _PC_PRIO_IO: 1698 *ap->a_retval = 0; 1699 break; 1700 case _PC_SYNC_IO: 1701 *ap->a_retval = 0; 1702 break; 1703 case _PC_ALLOC_SIZE_MIN: 1704 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_bsize; 1705 break; 1706 case _PC_FILESIZEBITS: 1707 *ap->a_retval = 64; 1708 break; 1709 case _PC_REC_INCR_XFER_SIZE: 1710 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize; 1711 break; 1712 case _PC_REC_MAX_XFER_SIZE: 1713 *ap->a_retval = -1; /* means ``unlimited'' */ 1714 break; 1715 case _PC_REC_MIN_XFER_SIZE: 1716 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize; 1717 break; 1718 case _PC_REC_XFER_ALIGN: 1719 *ap->a_retval = PAGE_SIZE; 1720 break; 1721 case _PC_SYMLINK_MAX: 1722 *ap->a_retval = MAXPATHLEN; 1723 break; 1724 1725 default: 1726 error = vop_stdpathconf(ap); 1727 break; 1728 } 1729 return (error); 1730 } 1731 1732 /* 1733 * Vnode operation to remove a named attribute. 1734 */ 1735 static int 1736 ext2_deleteextattr(struct vop_deleteextattr_args *ap) 1737 { 1738 struct inode *ip; 1739 struct m_ext2fs *fs; 1740 int error; 1741 1742 ip = VTOI(ap->a_vp); 1743 fs = ip->i_e2fs; 1744 1745 if (!EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_EXT_ATTR)) 1746 return (EOPNOTSUPP); 1747 1748 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK) 1749 return (EOPNOTSUPP); 1750 1751 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 1752 ap->a_cred, ap->a_td, VWRITE); 1753 if (error) 1754 return (error); 1755 1756 error = ENOATTR; 1757 1758 if (EXT2_INODE_SIZE(fs) != E2FS_REV0_INODE_SIZE) { 1759 error = ext2_extattr_inode_delete(ip, ap->a_attrnamespace, ap->a_name); 1760 if (error != ENOATTR) 1761 return (error); 1762 } 1763 1764 if (ip->i_facl) 1765 error = ext2_extattr_block_delete(ip, ap->a_attrnamespace, ap->a_name); 1766 1767 return (error); 1768 } 1769 1770 /* 1771 * Vnode operation to retrieve a named extended attribute. 1772 */ 1773 static int 1774 ext2_getextattr(struct vop_getextattr_args *ap) 1775 { 1776 struct inode *ip; 1777 struct m_ext2fs *fs; 1778 int error; 1779 1780 ip = VTOI(ap->a_vp); 1781 fs = ip->i_e2fs; 1782 1783 if (!EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_EXT_ATTR)) 1784 return (EOPNOTSUPP); 1785 1786 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK) 1787 return (EOPNOTSUPP); 1788 1789 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 1790 ap->a_cred, ap->a_td, VREAD); 1791 if (error) 1792 return (error); 1793 1794 if (ap->a_size != NULL) 1795 *ap->a_size = 0; 1796 1797 error = ENOATTR; 1798 1799 if (EXT2_INODE_SIZE(fs) != E2FS_REV0_INODE_SIZE) { 1800 error = ext2_extattr_inode_get(ip, ap->a_attrnamespace, 1801 ap->a_name, ap->a_uio, ap->a_size); 1802 if (error != ENOATTR) 1803 return (error); 1804 } 1805 1806 if (ip->i_facl) 1807 error = ext2_extattr_block_get(ip, ap->a_attrnamespace, 1808 ap->a_name, ap->a_uio, ap->a_size); 1809 1810 return (error); 1811 } 1812 1813 /* 1814 * Vnode operation to retrieve extended attributes on a vnode. 1815 */ 1816 static int 1817 ext2_listextattr(struct vop_listextattr_args *ap) 1818 { 1819 struct inode *ip; 1820 struct m_ext2fs *fs; 1821 int error; 1822 1823 ip = VTOI(ap->a_vp); 1824 fs = ip->i_e2fs; 1825 1826 if (!EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_EXT_ATTR)) 1827 return (EOPNOTSUPP); 1828 1829 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK) 1830 return (EOPNOTSUPP); 1831 1832 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 1833 ap->a_cred, ap->a_td, VREAD); 1834 if (error) 1835 return (error); 1836 1837 if (ap->a_size != NULL) 1838 *ap->a_size = 0; 1839 1840 if (EXT2_INODE_SIZE(fs) != E2FS_REV0_INODE_SIZE) { 1841 error = ext2_extattr_inode_list(ip, ap->a_attrnamespace, 1842 ap->a_uio, ap->a_size); 1843 if (error) 1844 return (error); 1845 } 1846 1847 if (ip->i_facl) 1848 error = ext2_extattr_block_list(ip, ap->a_attrnamespace, 1849 ap->a_uio, ap->a_size); 1850 1851 return (error); 1852 } 1853 1854 /* 1855 * Vnode operation to set a named attribute. 1856 */ 1857 static int 1858 ext2_setextattr(struct vop_setextattr_args *ap) 1859 { 1860 struct inode *ip; 1861 struct m_ext2fs *fs; 1862 int error; 1863 1864 ip = VTOI(ap->a_vp); 1865 fs = ip->i_e2fs; 1866 1867 if (!EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_EXT_ATTR)) 1868 return (EOPNOTSUPP); 1869 1870 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK) 1871 return (EOPNOTSUPP); 1872 1873 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 1874 ap->a_cred, ap->a_td, VWRITE); 1875 if (error) 1876 return (error); 1877 1878 error = ext2_extattr_valid_attrname(ap->a_attrnamespace, ap->a_name); 1879 if (error) 1880 return (error); 1881 1882 if (EXT2_INODE_SIZE(fs) != E2FS_REV0_INODE_SIZE) { 1883 error = ext2_extattr_inode_set(ip, ap->a_attrnamespace, 1884 ap->a_name, ap->a_uio); 1885 if (error != ENOSPC) 1886 return (error); 1887 } 1888 1889 error = ext2_extattr_block_set(ip, ap->a_attrnamespace, 1890 ap->a_name, ap->a_uio); 1891 1892 return (error); 1893 } 1894 1895 /* 1896 * Vnode pointer to File handle 1897 */ 1898 /* ARGSUSED */ 1899 static int 1900 ext2_vptofh(struct vop_vptofh_args *ap) 1901 { 1902 struct inode *ip; 1903 struct ufid *ufhp; 1904 1905 ip = VTOI(ap->a_vp); 1906 ufhp = (struct ufid *)ap->a_fhp; 1907 ufhp->ufid_len = sizeof(struct ufid); 1908 ufhp->ufid_ino = ip->i_number; 1909 ufhp->ufid_gen = ip->i_gen; 1910 return (0); 1911 } 1912 1913 /* 1914 * Initialize the vnode associated with a new inode, handle aliased 1915 * vnodes. 1916 */ 1917 int 1918 ext2_vinit(struct mount *mntp, struct vop_vector *fifoops, struct vnode **vpp) 1919 { 1920 struct inode *ip; 1921 struct vnode *vp; 1922 1923 vp = *vpp; 1924 ip = VTOI(vp); 1925 vp->v_type = IFTOVT(ip->i_mode); 1926 if (vp->v_type == VFIFO) 1927 vp->v_op = fifoops; 1928 1929 if (ip->i_number == EXT2_ROOTINO) 1930 vp->v_vflag |= VV_ROOT; 1931 ip->i_modrev = init_va_filerev(); 1932 *vpp = vp; 1933 return (0); 1934 } 1935 1936 /* 1937 * Allocate a new inode. 1938 */ 1939 static int 1940 ext2_makeinode(int mode, struct vnode *dvp, struct vnode **vpp, 1941 struct componentname *cnp) 1942 { 1943 struct inode *ip, *pdir; 1944 struct vnode *tvp; 1945 int error; 1946 1947 pdir = VTOI(dvp); 1948 #ifdef INVARIANTS 1949 if ((cnp->cn_flags & HASBUF) == 0) 1950 panic("ext2_makeinode: no name"); 1951 #endif 1952 *vpp = NULL; 1953 if ((mode & IFMT) == 0) 1954 mode |= IFREG; 1955 1956 error = ext2_valloc(dvp, mode, cnp->cn_cred, &tvp); 1957 if (error) { 1958 return (error); 1959 } 1960 ip = VTOI(tvp); 1961 ip->i_gid = pdir->i_gid; 1962 #ifdef SUIDDIR 1963 { 1964 /* 1965 * if we are 1966 * not the owner of the directory, 1967 * and we are hacking owners here, (only do this where told to) 1968 * and we are not giving it TOO root, (would subvert quotas) 1969 * then go ahead and give it to the other user. 1970 * Note that this drops off the execute bits for security. 1971 */ 1972 if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) && 1973 (pdir->i_mode & ISUID) && 1974 (pdir->i_uid != cnp->cn_cred->cr_uid) && pdir->i_uid) { 1975 ip->i_uid = pdir->i_uid; 1976 mode &= ~07111; 1977 } else { 1978 ip->i_uid = cnp->cn_cred->cr_uid; 1979 } 1980 } 1981 #else 1982 ip->i_uid = cnp->cn_cred->cr_uid; 1983 #endif 1984 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 1985 ip->i_mode = mode; 1986 tvp->v_type = IFTOVT(mode); /* Rest init'd in getnewvnode(). */ 1987 ip->i_nlink = 1; 1988 if ((ip->i_mode & ISGID) && !groupmember(ip->i_gid, cnp->cn_cred)) { 1989 if (priv_check_cred(cnp->cn_cred, PRIV_VFS_RETAINSUGID, 0)) 1990 ip->i_mode &= ~ISGID; 1991 } 1992 1993 if (cnp->cn_flags & ISWHITEOUT) 1994 ip->i_flags |= UF_OPAQUE; 1995 1996 /* 1997 * Make sure inode goes to disk before directory entry. 1998 */ 1999 error = ext2_update(tvp, !DOINGASYNC(tvp)); 2000 if (error) 2001 goto bad; 2002 2003 #ifdef UFS_ACL 2004 if (dvp->v_mount->mnt_flag & MNT_ACLS) { 2005 error = ext2_do_posix1e_acl_inheritance_file(dvp, tvp, mode, 2006 cnp->cn_cred, cnp->cn_thread); 2007 if (error) 2008 goto bad; 2009 } 2010 #endif /* UFS_ACL */ 2011 2012 error = ext2_direnter(ip, dvp, cnp); 2013 if (error) 2014 goto bad; 2015 2016 *vpp = tvp; 2017 return (0); 2018 2019 bad: 2020 /* 2021 * Write error occurred trying to update the inode 2022 * or the directory so must deallocate the inode. 2023 */ 2024 ip->i_nlink = 0; 2025 ip->i_flag |= IN_CHANGE; 2026 vput(tvp); 2027 return (error); 2028 } 2029 2030 /* 2031 * Vnode op for reading. 2032 */ 2033 static int 2034 ext2_read(struct vop_read_args *ap) 2035 { 2036 struct vnode *vp; 2037 struct inode *ip; 2038 struct uio *uio; 2039 struct m_ext2fs *fs; 2040 struct buf *bp; 2041 daddr_t lbn, nextlbn; 2042 off_t bytesinfile; 2043 long size, xfersize, blkoffset; 2044 int error, orig_resid, seqcount; 2045 int ioflag; 2046 2047 vp = ap->a_vp; 2048 uio = ap->a_uio; 2049 ioflag = ap->a_ioflag; 2050 2051 seqcount = ap->a_ioflag >> IO_SEQSHIFT; 2052 ip = VTOI(vp); 2053 2054 #ifdef INVARIANTS 2055 if (uio->uio_rw != UIO_READ) 2056 panic("%s: mode", "ext2_read"); 2057 2058 if (vp->v_type == VLNK) { 2059 if ((int)ip->i_size < vp->v_mount->mnt_maxsymlinklen) 2060 panic("%s: short symlink", "ext2_read"); 2061 } else if (vp->v_type != VREG && vp->v_type != VDIR) 2062 panic("%s: type %d", "ext2_read", vp->v_type); 2063 #endif 2064 orig_resid = uio->uio_resid; 2065 KASSERT(orig_resid >= 0, ("ext2_read: uio->uio_resid < 0")); 2066 if (orig_resid == 0) 2067 return (0); 2068 KASSERT(uio->uio_offset >= 0, ("ext2_read: uio->uio_offset < 0")); 2069 fs = ip->i_e2fs; 2070 if (uio->uio_offset < ip->i_size && 2071 uio->uio_offset >= fs->e2fs_maxfilesize) 2072 return (EOVERFLOW); 2073 2074 for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) { 2075 if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0) 2076 break; 2077 lbn = lblkno(fs, uio->uio_offset); 2078 nextlbn = lbn + 1; 2079 size = blksize(fs, ip, lbn); 2080 blkoffset = blkoff(fs, uio->uio_offset); 2081 2082 xfersize = fs->e2fs_fsize - blkoffset; 2083 if (uio->uio_resid < xfersize) 2084 xfersize = uio->uio_resid; 2085 if (bytesinfile < xfersize) 2086 xfersize = bytesinfile; 2087 2088 if (lblktosize(fs, nextlbn) >= ip->i_size) 2089 error = bread(vp, lbn, size, NOCRED, &bp); 2090 else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) { 2091 error = cluster_read(vp, ip->i_size, lbn, size, 2092 NOCRED, blkoffset + uio->uio_resid, seqcount, 2093 0, &bp); 2094 } else if (seqcount > 1) { 2095 u_int nextsize = blksize(fs, ip, nextlbn); 2096 2097 error = breadn(vp, lbn, 2098 size, &nextlbn, &nextsize, 1, NOCRED, &bp); 2099 } else 2100 error = bread(vp, lbn, size, NOCRED, &bp); 2101 if (error) { 2102 brelse(bp); 2103 bp = NULL; 2104 break; 2105 } 2106 2107 /* 2108 * We should only get non-zero b_resid when an I/O error 2109 * has occurred, which should cause us to break above. 2110 * However, if the short read did not cause an error, 2111 * then we want to ensure that we do not uiomove bad 2112 * or uninitialized data. 2113 */ 2114 size -= bp->b_resid; 2115 if (size < xfersize) { 2116 if (size == 0) 2117 break; 2118 xfersize = size; 2119 } 2120 error = uiomove((char *)bp->b_data + blkoffset, 2121 (int)xfersize, uio); 2122 if (error) 2123 break; 2124 vfs_bio_brelse(bp, ioflag); 2125 } 2126 2127 /* 2128 * This can only happen in the case of an error because the loop 2129 * above resets bp to NULL on each iteration and on normal 2130 * completion has not set a new value into it. so it must have come 2131 * from a 'break' statement 2132 */ 2133 if (bp != NULL) 2134 vfs_bio_brelse(bp, ioflag); 2135 2136 if ((error == 0 || uio->uio_resid != orig_resid) && 2137 (vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) 2138 ip->i_flag |= IN_ACCESS; 2139 return (error); 2140 } 2141 2142 static int 2143 ext2_ioctl(struct vop_ioctl_args *ap) 2144 { 2145 2146 switch (ap->a_command) { 2147 case FIOSEEKDATA: 2148 case FIOSEEKHOLE: 2149 return (vn_bmap_seekhole(ap->a_vp, ap->a_command, 2150 (off_t *)ap->a_data, ap->a_cred)); 2151 default: 2152 return (ENOTTY); 2153 } 2154 } 2155 2156 /* 2157 * Vnode op for writing. 2158 */ 2159 static int 2160 ext2_write(struct vop_write_args *ap) 2161 { 2162 struct vnode *vp; 2163 struct uio *uio; 2164 struct inode *ip; 2165 struct m_ext2fs *fs; 2166 struct buf *bp; 2167 daddr_t lbn; 2168 off_t osize; 2169 int blkoffset, error, flags, ioflag, resid, size, seqcount, xfersize; 2170 2171 ioflag = ap->a_ioflag; 2172 uio = ap->a_uio; 2173 vp = ap->a_vp; 2174 2175 seqcount = ioflag >> IO_SEQSHIFT; 2176 ip = VTOI(vp); 2177 2178 #ifdef INVARIANTS 2179 if (uio->uio_rw != UIO_WRITE) 2180 panic("%s: mode", "ext2_write"); 2181 #endif 2182 2183 switch (vp->v_type) { 2184 case VREG: 2185 if (ioflag & IO_APPEND) 2186 uio->uio_offset = ip->i_size; 2187 if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size) 2188 return (EPERM); 2189 /* FALLTHROUGH */ 2190 case VLNK: 2191 break; 2192 case VDIR: 2193 /* XXX differs from ffs -- this is called from ext2_mkdir(). */ 2194 if ((ioflag & IO_SYNC) == 0) 2195 panic("ext2_write: nonsync dir write"); 2196 break; 2197 default: 2198 panic("ext2_write: type %p %d (%jd,%jd)", (void *)vp, 2199 vp->v_type, (intmax_t)uio->uio_offset, 2200 (intmax_t)uio->uio_resid); 2201 } 2202 2203 KASSERT(uio->uio_resid >= 0, ("ext2_write: uio->uio_resid < 0")); 2204 KASSERT(uio->uio_offset >= 0, ("ext2_write: uio->uio_offset < 0")); 2205 fs = ip->i_e2fs; 2206 if ((uoff_t)uio->uio_offset + uio->uio_resid > fs->e2fs_maxfilesize) 2207 return (EFBIG); 2208 /* 2209 * Maybe this should be above the vnode op call, but so long as 2210 * file servers have no limits, I don't think it matters. 2211 */ 2212 if (vn_rlimit_fsize(vp, uio, uio->uio_td)) 2213 return (EFBIG); 2214 2215 resid = uio->uio_resid; 2216 osize = ip->i_size; 2217 if (seqcount > BA_SEQMAX) 2218 flags = BA_SEQMAX << BA_SEQSHIFT; 2219 else 2220 flags = seqcount << BA_SEQSHIFT; 2221 if ((ioflag & IO_SYNC) && !DOINGASYNC(vp)) 2222 flags |= IO_SYNC; 2223 2224 for (error = 0; uio->uio_resid > 0;) { 2225 lbn = lblkno(fs, uio->uio_offset); 2226 blkoffset = blkoff(fs, uio->uio_offset); 2227 xfersize = fs->e2fs_fsize - blkoffset; 2228 if (uio->uio_resid < xfersize) 2229 xfersize = uio->uio_resid; 2230 if (uio->uio_offset + xfersize > ip->i_size) 2231 vnode_pager_setsize(vp, uio->uio_offset + xfersize); 2232 2233 /* 2234 * We must perform a read-before-write if the transfer size 2235 * does not cover the entire buffer. 2236 */ 2237 if (fs->e2fs_bsize > xfersize) 2238 flags |= BA_CLRBUF; 2239 else 2240 flags &= ~BA_CLRBUF; 2241 error = ext2_balloc(ip, lbn, blkoffset + xfersize, 2242 ap->a_cred, &bp, flags); 2243 if (error != 0) 2244 break; 2245 2246 if ((ioflag & (IO_SYNC | IO_INVAL)) == (IO_SYNC | IO_INVAL)) 2247 bp->b_flags |= B_NOCACHE; 2248 if (uio->uio_offset + xfersize > ip->i_size) 2249 ip->i_size = uio->uio_offset + xfersize; 2250 size = blksize(fs, ip, lbn) - bp->b_resid; 2251 if (size < xfersize) 2252 xfersize = size; 2253 2254 error = 2255 uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio); 2256 /* 2257 * If the buffer is not already filled and we encounter an 2258 * error while trying to fill it, we have to clear out any 2259 * garbage data from the pages instantiated for the buffer. 2260 * If we do not, a failed uiomove() during a write can leave 2261 * the prior contents of the pages exposed to a userland mmap. 2262 * 2263 * Note that we need only clear buffers with a transfer size 2264 * equal to the block size because buffers with a shorter 2265 * transfer size were cleared above by the call to ext2_balloc() 2266 * with the BA_CLRBUF flag set. 2267 * 2268 * If the source region for uiomove identically mmaps the 2269 * buffer, uiomove() performed the NOP copy, and the buffer 2270 * content remains valid because the page fault handler 2271 * validated the pages. 2272 */ 2273 if (error != 0 && (bp->b_flags & B_CACHE) == 0 && 2274 fs->e2fs_bsize == xfersize) 2275 vfs_bio_clrbuf(bp); 2276 2277 vfs_bio_set_flags(bp, ioflag); 2278 2279 /* 2280 * If IO_SYNC each buffer is written synchronously. Otherwise 2281 * if we have a severe page deficiency write the buffer 2282 * asynchronously. Otherwise try to cluster, and if that 2283 * doesn't do it then either do an async write (if O_DIRECT), 2284 * or a delayed write (if not). 2285 */ 2286 if (ioflag & IO_SYNC) { 2287 (void)bwrite(bp); 2288 } else if (vm_page_count_severe() || 2289 buf_dirty_count_severe() || 2290 (ioflag & IO_ASYNC)) { 2291 bp->b_flags |= B_CLUSTEROK; 2292 bawrite(bp); 2293 } else if (xfersize + blkoffset == fs->e2fs_fsize) { 2294 if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) { 2295 bp->b_flags |= B_CLUSTEROK; 2296 cluster_write(vp, bp, ip->i_size, seqcount, 0); 2297 } else { 2298 bawrite(bp); 2299 } 2300 } else if (ioflag & IO_DIRECT) { 2301 bp->b_flags |= B_CLUSTEROK; 2302 bawrite(bp); 2303 } else { 2304 bp->b_flags |= B_CLUSTEROK; 2305 bdwrite(bp); 2306 } 2307 if (error || xfersize == 0) 2308 break; 2309 } 2310 /* 2311 * If we successfully wrote any data, and we are not the superuser 2312 * we clear the setuid and setgid bits as a precaution against 2313 * tampering. 2314 */ 2315 if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid && 2316 ap->a_cred) { 2317 if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0)) 2318 ip->i_mode &= ~(ISUID | ISGID); 2319 } 2320 if (error) { 2321 if (ioflag & IO_UNIT) { 2322 (void)ext2_truncate(vp, osize, 2323 ioflag & IO_SYNC, ap->a_cred, uio->uio_td); 2324 uio->uio_offset -= resid - uio->uio_resid; 2325 uio->uio_resid = resid; 2326 } 2327 } 2328 if (uio->uio_resid != resid) { 2329 ip->i_flag |= IN_CHANGE | IN_UPDATE; 2330 if (ioflag & IO_SYNC) 2331 error = ext2_update(vp, 1); 2332 } 2333 return (error); 2334 } 2335