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