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 * Copyright (c) 1982, 1986, 1989, 1993 9 * The Regents of the University of California. All rights reserved. 10 * (c) UNIX System Laboratories, Inc. 11 * All or some portions of this file are derived from material licensed 12 * to the University of California by American Telephone and Telegraph 13 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 14 * the permission of UNIX System Laboratories, Inc. 15 * 16 * Redistribution and use in source and binary forms, with or without 17 * modification, are permitted provided that the following conditions 18 * are met: 19 * 1. Redistributions of source code must retain the above copyright 20 * notice, this list of conditions and the following disclaimer. 21 * 2. Redistributions in binary form must reproduce the above copyright 22 * notice, this list of conditions and the following disclaimer in the 23 * documentation and/or other materials provided with the distribution. 24 * 4. Neither the name of the University nor the names of its contributors 25 * may be used to endorse or promote products derived from this software 26 * without specific prior written permission. 27 * 28 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 29 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 30 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 31 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 32 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 33 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 34 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 35 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 36 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 37 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 38 * SUCH DAMAGE. 39 * 40 * @(#)ufs_vnops.c 8.7 (Berkeley) 2/3/94 41 * @(#)ufs_vnops.c 8.27 (Berkeley) 5/27/95 42 * $FreeBSD$ 43 */ 44 45 #include "opt_suiddir.h" 46 47 #include <sys/param.h> 48 #include <sys/systm.h> 49 #include <sys/kernel.h> 50 #include <sys/fcntl.h> 51 #include <sys/filio.h> 52 #include <sys/stat.h> 53 #include <sys/bio.h> 54 #include <sys/buf.h> 55 #include <sys/endian.h> 56 #include <sys/priv.h> 57 #include <sys/rwlock.h> 58 #include <sys/mount.h> 59 #include <sys/unistd.h> 60 #include <sys/time.h> 61 #include <sys/vnode.h> 62 #include <sys/namei.h> 63 #include <sys/lockf.h> 64 #include <sys/event.h> 65 #include <sys/conf.h> 66 #include <sys/file.h> 67 68 #include <vm/vm.h> 69 #include <vm/vm_param.h> 70 #include <vm/vm_extern.h> 71 #include <vm/vm_object.h> 72 #include <vm/vm_page.h> 73 #include <vm/vm_pager.h> 74 #include <vm/vnode_pager.h> 75 76 #include "opt_directio.h" 77 78 #include <ufs/ufs/dir.h> 79 80 #include <fs/ext2fs/fs.h> 81 #include <fs/ext2fs/inode.h> 82 #include <fs/ext2fs/ext2_extern.h> 83 #include <fs/ext2fs/ext2fs.h> 84 #include <fs/ext2fs/ext2_dinode.h> 85 #include <fs/ext2fs/ext2_dir.h> 86 #include <fs/ext2fs/ext2_mount.h> 87 88 static int ext2_makeinode(int mode, struct vnode *, struct vnode **, struct componentname *); 89 static void ext2_itimes_locked(struct vnode *); 90 static int ext4_ext_read(struct vop_read_args *); 91 static int ext2_ind_read(struct vop_read_args *); 92 93 static vop_access_t ext2_access; 94 static int ext2_chmod(struct vnode *, int, struct ucred *, struct thread *); 95 static int ext2_chown(struct vnode *, uid_t, gid_t, struct ucred *, 96 struct thread *); 97 static vop_close_t ext2_close; 98 static vop_create_t ext2_create; 99 static vop_fsync_t ext2_fsync; 100 static vop_getattr_t ext2_getattr; 101 static vop_ioctl_t ext2_ioctl; 102 static vop_link_t ext2_link; 103 static vop_mkdir_t ext2_mkdir; 104 static vop_mknod_t ext2_mknod; 105 static vop_open_t ext2_open; 106 static vop_pathconf_t ext2_pathconf; 107 static vop_print_t ext2_print; 108 static vop_read_t ext2_read; 109 static vop_readlink_t ext2_readlink; 110 static vop_remove_t ext2_remove; 111 static vop_rename_t ext2_rename; 112 static vop_rmdir_t ext2_rmdir; 113 static vop_setattr_t ext2_setattr; 114 static vop_strategy_t ext2_strategy; 115 static vop_symlink_t ext2_symlink; 116 static vop_write_t ext2_write; 117 static vop_vptofh_t ext2_vptofh; 118 static vop_close_t ext2fifo_close; 119 static vop_kqfilter_t ext2fifo_kqfilter; 120 121 /* Global vfs data structures for ext2. */ 122 struct vop_vector ext2_vnodeops = { 123 .vop_default = &default_vnodeops, 124 .vop_access = ext2_access, 125 .vop_bmap = ext2_bmap, 126 .vop_cachedlookup = ext2_lookup, 127 .vop_close = ext2_close, 128 .vop_create = ext2_create, 129 .vop_fsync = ext2_fsync, 130 .vop_getpages = vnode_pager_local_getpages, 131 .vop_getattr = ext2_getattr, 132 .vop_inactive = ext2_inactive, 133 .vop_ioctl = ext2_ioctl, 134 .vop_link = ext2_link, 135 .vop_lookup = vfs_cache_lookup, 136 .vop_mkdir = ext2_mkdir, 137 .vop_mknod = ext2_mknod, 138 .vop_open = ext2_open, 139 .vop_pathconf = ext2_pathconf, 140 .vop_poll = vop_stdpoll, 141 .vop_print = ext2_print, 142 .vop_read = ext2_read, 143 .vop_readdir = ext2_readdir, 144 .vop_readlink = ext2_readlink, 145 .vop_reallocblks = ext2_reallocblks, 146 .vop_reclaim = ext2_reclaim, 147 .vop_remove = ext2_remove, 148 .vop_rename = ext2_rename, 149 .vop_rmdir = ext2_rmdir, 150 .vop_setattr = ext2_setattr, 151 .vop_strategy = ext2_strategy, 152 .vop_symlink = ext2_symlink, 153 .vop_write = ext2_write, 154 .vop_vptofh = ext2_vptofh, 155 }; 156 157 struct vop_vector ext2_fifoops = { 158 .vop_default = &fifo_specops, 159 .vop_access = ext2_access, 160 .vop_close = ext2fifo_close, 161 .vop_fsync = ext2_fsync, 162 .vop_getattr = ext2_getattr, 163 .vop_inactive = ext2_inactive, 164 .vop_kqfilter = ext2fifo_kqfilter, 165 .vop_print = ext2_print, 166 .vop_read = VOP_PANIC, 167 .vop_reclaim = ext2_reclaim, 168 .vop_setattr = ext2_setattr, 169 .vop_write = VOP_PANIC, 170 .vop_vptofh = ext2_vptofh, 171 }; 172 173 /* 174 * A virgin directory (no blushing please). 175 * Note that the type and namlen fields are reversed relative to ext2. 176 * Also, we don't use `struct odirtemplate', since it would just cause 177 * endianness problems. 178 */ 179 static struct dirtemplate mastertemplate = { 180 0, 12, 1, EXT2_FT_DIR, ".", 181 0, DIRBLKSIZ - 12, 2, EXT2_FT_DIR, ".." 182 }; 183 static struct dirtemplate omastertemplate = { 184 0, 12, 1, EXT2_FT_UNKNOWN, ".", 185 0, DIRBLKSIZ - 12, 2, EXT2_FT_UNKNOWN, ".." 186 }; 187 188 static void 189 ext2_itimes_locked(struct vnode *vp) 190 { 191 struct inode *ip; 192 struct timespec ts; 193 194 ASSERT_VI_LOCKED(vp, __func__); 195 196 ip = VTOI(vp); 197 if ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE)) == 0) 198 return; 199 if ((vp->v_type == VBLK || vp->v_type == VCHR)) 200 ip->i_flag |= IN_LAZYMOD; 201 else 202 ip->i_flag |= IN_MODIFIED; 203 if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 204 vfs_timestamp(&ts); 205 if (ip->i_flag & IN_ACCESS) { 206 ip->i_atime = ts.tv_sec; 207 ip->i_atimensec = ts.tv_nsec; 208 } 209 if (ip->i_flag & IN_UPDATE) { 210 ip->i_mtime = ts.tv_sec; 211 ip->i_mtimensec = ts.tv_nsec; 212 ip->i_modrev++; 213 } 214 if (ip->i_flag & IN_CHANGE) { 215 ip->i_ctime = ts.tv_sec; 216 ip->i_ctimensec = ts.tv_nsec; 217 } 218 } 219 ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_UPDATE); 220 } 221 222 void 223 ext2_itimes(struct vnode *vp) 224 { 225 226 VI_LOCK(vp); 227 ext2_itimes_locked(vp); 228 VI_UNLOCK(vp); 229 } 230 231 /* 232 * Create a regular file 233 */ 234 static int 235 ext2_create(struct vop_create_args *ap) 236 { 237 int error; 238 239 error = 240 ext2_makeinode(MAKEIMODE(ap->a_vap->va_type, ap->a_vap->va_mode), 241 ap->a_dvp, ap->a_vpp, ap->a_cnp); 242 if (error != 0) 243 return (error); 244 if ((ap->a_cnp->cn_flags & MAKEENTRY) != 0) 245 cache_enter(ap->a_dvp, *ap->a_vpp, ap->a_cnp); 246 return (0); 247 } 248 249 static int 250 ext2_open(struct vop_open_args *ap) 251 { 252 253 if (ap->a_vp->v_type == VBLK || ap->a_vp->v_type == VCHR) 254 return (EOPNOTSUPP); 255 256 /* 257 * Files marked append-only must be opened for appending. 258 */ 259 if ((VTOI(ap->a_vp)->i_flags & APPEND) && 260 (ap->a_mode & (FWRITE | O_APPEND)) == FWRITE) 261 return (EPERM); 262 263 vnode_create_vobject(ap->a_vp, VTOI(ap->a_vp)->i_size, ap->a_td); 264 265 return (0); 266 } 267 268 /* 269 * Close called. 270 * 271 * Update the times on the inode. 272 */ 273 static int 274 ext2_close(struct vop_close_args *ap) 275 { 276 struct vnode *vp = ap->a_vp; 277 278 VI_LOCK(vp); 279 if (vp->v_usecount > 1) 280 ext2_itimes_locked(vp); 281 VI_UNLOCK(vp); 282 return (0); 283 } 284 285 static int 286 ext2_access(struct vop_access_args *ap) 287 { 288 struct vnode *vp = ap->a_vp; 289 struct inode *ip = VTOI(vp); 290 accmode_t accmode = ap->a_accmode; 291 int error; 292 293 if (vp->v_type == VBLK || vp->v_type == VCHR) 294 return (EOPNOTSUPP); 295 296 /* 297 * Disallow write attempts on read-only file systems; 298 * unless the file is a socket, fifo, or a block or 299 * character device resident on the file system. 300 */ 301 if (accmode & VWRITE) { 302 switch (vp->v_type) { 303 case VDIR: 304 case VLNK: 305 case VREG: 306 if (vp->v_mount->mnt_flag & MNT_RDONLY) 307 return (EROFS); 308 break; 309 default: 310 break; 311 } 312 } 313 314 /* If immutable bit set, nobody gets to write it. */ 315 if ((accmode & VWRITE) && (ip->i_flags & (SF_IMMUTABLE | SF_SNAPSHOT))) 316 return (EPERM); 317 318 error = vaccess(vp->v_type, ip->i_mode, ip->i_uid, ip->i_gid, 319 ap->a_accmode, ap->a_cred, NULL); 320 return (error); 321 } 322 323 static int 324 ext2_getattr(struct vop_getattr_args *ap) 325 { 326 struct vnode *vp = ap->a_vp; 327 struct inode *ip = VTOI(vp); 328 struct vattr *vap = ap->a_vap; 329 330 ext2_itimes(vp); 331 /* 332 * Copy from inode table 333 */ 334 vap->va_fsid = dev2udev(ip->i_devvp->v_rdev); 335 vap->va_fileid = ip->i_number; 336 vap->va_mode = ip->i_mode & ~IFMT; 337 vap->va_nlink = ip->i_nlink; 338 vap->va_uid = ip->i_uid; 339 vap->va_gid = ip->i_gid; 340 vap->va_rdev = ip->i_rdev; 341 vap->va_size = ip->i_size; 342 vap->va_atime.tv_sec = ip->i_atime; 343 vap->va_atime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_atimensec : 0; 344 vap->va_mtime.tv_sec = ip->i_mtime; 345 vap->va_mtime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_mtimensec : 0; 346 vap->va_ctime.tv_sec = ip->i_ctime; 347 vap->va_ctime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_ctimensec : 0; 348 if E2DI_HAS_XTIME(ip) { 349 vap->va_birthtime.tv_sec = ip->i_birthtime; 350 vap->va_birthtime.tv_nsec = ip->i_birthnsec; 351 } 352 vap->va_flags = ip->i_flags; 353 vap->va_gen = ip->i_gen; 354 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize; 355 vap->va_bytes = dbtob((u_quad_t)ip->i_blocks); 356 vap->va_type = IFTOVT(ip->i_mode); 357 vap->va_filerev = ip->i_modrev; 358 return (0); 359 } 360 361 /* 362 * Set attribute vnode op. called from several syscalls 363 */ 364 static int 365 ext2_setattr(struct vop_setattr_args *ap) 366 { 367 struct vattr *vap = ap->a_vap; 368 struct vnode *vp = ap->a_vp; 369 struct inode *ip = VTOI(vp); 370 struct ucred *cred = ap->a_cred; 371 struct thread *td = curthread; 372 int error; 373 374 /* 375 * Check for unsettable attributes. 376 */ 377 if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) || 378 (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) || 379 (vap->va_blocksize != VNOVAL) || (vap->va_rdev != VNOVAL) || 380 ((int)vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) { 381 return (EINVAL); 382 } 383 if (vap->va_flags != VNOVAL) { 384 /* Disallow flags not supported by ext2fs. */ 385 if(vap->va_flags & ~(SF_APPEND | SF_IMMUTABLE | UF_NODUMP)) 386 return (EOPNOTSUPP); 387 388 if (vp->v_mount->mnt_flag & MNT_RDONLY) 389 return (EROFS); 390 /* 391 * Callers may only modify the file flags on objects they 392 * have VADMIN rights for. 393 */ 394 if ((error = VOP_ACCESS(vp, VADMIN, cred, td))) 395 return (error); 396 /* 397 * Unprivileged processes and privileged processes in 398 * jail() are not permitted to unset system flags, or 399 * modify flags if any system flags are set. 400 * Privileged non-jail processes may not modify system flags 401 * if securelevel > 0 and any existing system flags are set. 402 */ 403 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) { 404 if (ip->i_flags & (SF_IMMUTABLE | SF_APPEND)) { 405 error = securelevel_gt(cred, 0); 406 if (error) 407 return (error); 408 } 409 } else { 410 if (ip->i_flags & (SF_IMMUTABLE | SF_APPEND) || 411 ((vap->va_flags ^ ip->i_flags) & SF_SETTABLE)) 412 return (EPERM); 413 } 414 ip->i_flags = vap->va_flags; 415 ip->i_flag |= IN_CHANGE; 416 if (ip->i_flags & (IMMUTABLE | APPEND)) 417 return (0); 418 } 419 if (ip->i_flags & (IMMUTABLE | APPEND)) 420 return (EPERM); 421 /* 422 * Go through the fields and update iff not VNOVAL. 423 */ 424 if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) { 425 if (vp->v_mount->mnt_flag & MNT_RDONLY) 426 return (EROFS); 427 if ((error = ext2_chown(vp, vap->va_uid, vap->va_gid, cred, 428 td)) != 0) 429 return (error); 430 } 431 if (vap->va_size != VNOVAL) { 432 /* 433 * Disallow write attempts on read-only file systems; 434 * unless the file is a socket, fifo, or a block or 435 * character device resident on the file system. 436 */ 437 switch (vp->v_type) { 438 case VDIR: 439 return (EISDIR); 440 case VLNK: 441 case VREG: 442 if (vp->v_mount->mnt_flag & MNT_RDONLY) 443 return (EROFS); 444 break; 445 default: 446 break; 447 } 448 if ((error = ext2_truncate(vp, vap->va_size, 0, cred, td)) != 0) 449 return (error); 450 } 451 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) { 452 if (vp->v_mount->mnt_flag & MNT_RDONLY) 453 return (EROFS); 454 /* 455 * From utimes(2): 456 * If times is NULL, ... The caller must be the owner of 457 * the file, have permission to write the file, or be the 458 * super-user. 459 * If times is non-NULL, ... The caller must be the owner of 460 * the file or be the super-user. 461 */ 462 if ((error = VOP_ACCESS(vp, VADMIN, cred, td)) && 463 ((vap->va_vaflags & VA_UTIMES_NULL) == 0 || 464 (error = VOP_ACCESS(vp, VWRITE, cred, td)))) 465 return (error); 466 if (vap->va_atime.tv_sec != VNOVAL) 467 ip->i_flag |= IN_ACCESS; 468 if (vap->va_mtime.tv_sec != VNOVAL) 469 ip->i_flag |= IN_CHANGE | IN_UPDATE; 470 ext2_itimes(vp); 471 if (vap->va_atime.tv_sec != VNOVAL) { 472 ip->i_atime = vap->va_atime.tv_sec; 473 ip->i_atimensec = vap->va_atime.tv_nsec; 474 } 475 if (vap->va_mtime.tv_sec != VNOVAL) { 476 ip->i_mtime = vap->va_mtime.tv_sec; 477 ip->i_mtimensec = vap->va_mtime.tv_nsec; 478 } 479 ip->i_birthtime = vap->va_birthtime.tv_sec; 480 ip->i_birthnsec = vap->va_birthtime.tv_nsec; 481 error = ext2_update(vp, 0); 482 if (error) 483 return (error); 484 } 485 error = 0; 486 if (vap->va_mode != (mode_t)VNOVAL) { 487 if (vp->v_mount->mnt_flag & MNT_RDONLY) 488 return (EROFS); 489 error = ext2_chmod(vp, (int)vap->va_mode, cred, td); 490 } 491 return (error); 492 } 493 494 /* 495 * Change the mode on a file. 496 * Inode must be locked before calling. 497 */ 498 static int 499 ext2_chmod(struct vnode *vp, int mode, struct ucred *cred, struct thread *td) 500 { 501 struct inode *ip = VTOI(vp); 502 int error; 503 504 /* 505 * To modify the permissions on a file, must possess VADMIN 506 * for that file. 507 */ 508 if ((error = VOP_ACCESS(vp, VADMIN, cred, td))) 509 return (error); 510 /* 511 * Privileged processes may set the sticky bit on non-directories, 512 * as well as set the setgid bit on a file with a group that the 513 * process is not a member of. 514 */ 515 if (vp->v_type != VDIR && (mode & S_ISTXT)) { 516 error = priv_check_cred(cred, PRIV_VFS_STICKYFILE, 0); 517 if (error) 518 return (EFTYPE); 519 } 520 if (!groupmember(ip->i_gid, cred) && (mode & ISGID)) { 521 error = priv_check_cred(cred, PRIV_VFS_SETGID, 0); 522 if (error) 523 return (error); 524 } 525 ip->i_mode &= ~ALLPERMS; 526 ip->i_mode |= (mode & ALLPERMS); 527 ip->i_flag |= IN_CHANGE; 528 return (0); 529 } 530 531 /* 532 * Perform chown operation on inode ip; 533 * inode must be locked prior to call. 534 */ 535 static int 536 ext2_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred, 537 struct thread *td) 538 { 539 struct inode *ip = VTOI(vp); 540 uid_t ouid; 541 gid_t ogid; 542 int error = 0; 543 544 if (uid == (uid_t)VNOVAL) 545 uid = ip->i_uid; 546 if (gid == (gid_t)VNOVAL) 547 gid = ip->i_gid; 548 /* 549 * To modify the ownership of a file, must possess VADMIN 550 * for that file. 551 */ 552 if ((error = VOP_ACCESS(vp, VADMIN, cred, td))) 553 return (error); 554 /* 555 * To change the owner of a file, or change the group of a file 556 * to a group of which we are not a member, the caller must 557 * have privilege. 558 */ 559 if (uid != ip->i_uid || (gid != ip->i_gid && 560 !groupmember(gid, cred))) { 561 error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0); 562 if (error) 563 return (error); 564 } 565 ogid = ip->i_gid; 566 ouid = ip->i_uid; 567 ip->i_gid = gid; 568 ip->i_uid = uid; 569 ip->i_flag |= IN_CHANGE; 570 if ((ip->i_mode & (ISUID | ISGID)) && (ouid != uid || ogid != gid)) { 571 if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID, 0) != 0) 572 ip->i_mode &= ~(ISUID | ISGID); 573 } 574 return (0); 575 } 576 577 /* 578 * Synch an open file. 579 */ 580 /* ARGSUSED */ 581 static int 582 ext2_fsync(struct vop_fsync_args *ap) 583 { 584 /* 585 * Flush all dirty buffers associated with a vnode. 586 */ 587 588 vop_stdfsync(ap); 589 590 return (ext2_update(ap->a_vp, ap->a_waitfor == MNT_WAIT)); 591 } 592 593 /* 594 * Mknod vnode call 595 */ 596 /* ARGSUSED */ 597 static int 598 ext2_mknod(struct vop_mknod_args *ap) 599 { 600 struct vattr *vap = ap->a_vap; 601 struct vnode **vpp = ap->a_vpp; 602 struct inode *ip; 603 ino_t ino; 604 int error; 605 606 error = ext2_makeinode(MAKEIMODE(vap->va_type, vap->va_mode), 607 ap->a_dvp, vpp, ap->a_cnp); 608 if (error) 609 return (error); 610 ip = VTOI(*vpp); 611 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 612 if (vap->va_rdev != VNOVAL) { 613 /* 614 * Want to be able to use this to make badblock 615 * inodes, so don't truncate the dev number. 616 */ 617 ip->i_rdev = vap->va_rdev; 618 } 619 /* 620 * Remove inode, then reload it through VFS_VGET so it is 621 * checked to see if it is an alias of an existing entry in 622 * the inode cache. XXX I don't believe this is necessary now. 623 */ 624 (*vpp)->v_type = VNON; 625 ino = ip->i_number; /* Save this before vgone() invalidates ip. */ 626 vgone(*vpp); 627 vput(*vpp); 628 error = VFS_VGET(ap->a_dvp->v_mount, ino, LK_EXCLUSIVE, vpp); 629 if (error) { 630 *vpp = NULL; 631 return (error); 632 } 633 return (0); 634 } 635 636 static int 637 ext2_remove(struct vop_remove_args *ap) 638 { 639 struct inode *ip; 640 struct vnode *vp = ap->a_vp; 641 struct vnode *dvp = ap->a_dvp; 642 int error; 643 644 ip = VTOI(vp); 645 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) || 646 (VTOI(dvp)->i_flags & APPEND)) { 647 error = EPERM; 648 goto out; 649 } 650 error = ext2_dirremove(dvp, ap->a_cnp); 651 if (error == 0) { 652 ip->i_nlink--; 653 ip->i_flag |= IN_CHANGE; 654 } 655 out: 656 return (error); 657 } 658 659 /* 660 * link vnode call 661 */ 662 static int 663 ext2_link(struct vop_link_args *ap) 664 { 665 struct vnode *vp = ap->a_vp; 666 struct vnode *tdvp = ap->a_tdvp; 667 struct componentname *cnp = ap->a_cnp; 668 struct inode *ip; 669 int error; 670 671 #ifdef INVARIANTS 672 if ((cnp->cn_flags & HASBUF) == 0) 673 panic("ext2_link: no name"); 674 #endif 675 ip = VTOI(vp); 676 if ((nlink_t)ip->i_nlink >= EXT2_LINK_MAX) { 677 error = EMLINK; 678 goto out; 679 } 680 if (ip->i_flags & (IMMUTABLE | APPEND)) { 681 error = EPERM; 682 goto out; 683 } 684 ip->i_nlink++; 685 ip->i_flag |= IN_CHANGE; 686 error = ext2_update(vp, !DOINGASYNC(vp)); 687 if (!error) 688 error = ext2_direnter(ip, tdvp, cnp); 689 if (error) { 690 ip->i_nlink--; 691 ip->i_flag |= IN_CHANGE; 692 } 693 out: 694 return (error); 695 } 696 697 /* 698 * Rename system call. 699 * rename("foo", "bar"); 700 * is essentially 701 * unlink("bar"); 702 * link("foo", "bar"); 703 * unlink("foo"); 704 * but ``atomically''. Can't do full commit without saving state in the 705 * inode on disk which isn't feasible at this time. Best we can do is 706 * always guarantee the target exists. 707 * 708 * Basic algorithm is: 709 * 710 * 1) Bump link count on source while we're linking it to the 711 * target. This also ensure the inode won't be deleted out 712 * from underneath us while we work (it may be truncated by 713 * a concurrent `trunc' or `open' for creation). 714 * 2) Link source to destination. If destination already exists, 715 * delete it first. 716 * 3) Unlink source reference to inode if still around. If a 717 * directory was moved and the parent of the destination 718 * is different from the source, patch the ".." entry in the 719 * directory. 720 */ 721 static int 722 ext2_rename(struct vop_rename_args *ap) 723 { 724 struct vnode *tvp = ap->a_tvp; 725 struct vnode *tdvp = ap->a_tdvp; 726 struct vnode *fvp = ap->a_fvp; 727 struct vnode *fdvp = ap->a_fdvp; 728 struct componentname *tcnp = ap->a_tcnp; 729 struct componentname *fcnp = ap->a_fcnp; 730 struct inode *ip, *xp, *dp; 731 struct dirtemplate dirbuf; 732 int doingdirectory = 0, oldparent = 0, newparent = 0; 733 int error = 0; 734 u_char namlen; 735 736 #ifdef INVARIANTS 737 if ((tcnp->cn_flags & HASBUF) == 0 || 738 (fcnp->cn_flags & HASBUF) == 0) 739 panic("ext2_rename: no name"); 740 #endif 741 /* 742 * Check for cross-device rename. 743 */ 744 if ((fvp->v_mount != tdvp->v_mount) || 745 (tvp && (fvp->v_mount != tvp->v_mount))) { 746 error = EXDEV; 747 abortit: 748 if (tdvp == tvp) 749 vrele(tdvp); 750 else 751 vput(tdvp); 752 if (tvp) 753 vput(tvp); 754 vrele(fdvp); 755 vrele(fvp); 756 return (error); 757 } 758 759 if (tvp && ((VTOI(tvp)->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) || 760 (VTOI(tdvp)->i_flags & APPEND))) { 761 error = EPERM; 762 goto abortit; 763 } 764 765 /* 766 * Renaming a file to itself has no effect. The upper layers should 767 * not call us in that case. Temporarily just warn if they do. 768 */ 769 if (fvp == tvp) { 770 printf("ext2_rename: fvp == tvp (can't happen)\n"); 771 error = 0; 772 goto abortit; 773 } 774 775 if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0) 776 goto abortit; 777 dp = VTOI(fdvp); 778 ip = VTOI(fvp); 779 if (ip->i_nlink >= EXT2_LINK_MAX) { 780 VOP_UNLOCK(fvp, 0); 781 error = EMLINK; 782 goto abortit; 783 } 784 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) 785 || (dp->i_flags & APPEND)) { 786 VOP_UNLOCK(fvp, 0); 787 error = EPERM; 788 goto abortit; 789 } 790 if ((ip->i_mode & IFMT) == IFDIR) { 791 /* 792 * Avoid ".", "..", and aliases of "." for obvious reasons. 793 */ 794 if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') || 795 dp == ip || (fcnp->cn_flags | tcnp->cn_flags) & ISDOTDOT || 796 (ip->i_flag & IN_RENAME)) { 797 VOP_UNLOCK(fvp, 0); 798 error = EINVAL; 799 goto abortit; 800 } 801 ip->i_flag |= IN_RENAME; 802 oldparent = dp->i_number; 803 doingdirectory++; 804 } 805 vrele(fdvp); 806 807 /* 808 * When the target exists, both the directory 809 * and target vnodes are returned locked. 810 */ 811 dp = VTOI(tdvp); 812 xp = NULL; 813 if (tvp) 814 xp = VTOI(tvp); 815 816 /* 817 * 1) Bump link count while we're moving stuff 818 * around. If we crash somewhere before 819 * completing our work, the link count 820 * may be wrong, but correctable. 821 */ 822 ip->i_nlink++; 823 ip->i_flag |= IN_CHANGE; 824 if ((error = ext2_update(fvp, !DOINGASYNC(fvp))) != 0) { 825 VOP_UNLOCK(fvp, 0); 826 goto bad; 827 } 828 829 /* 830 * If ".." must be changed (ie the directory gets a new 831 * parent) then the source directory must not be in the 832 * directory hierarchy above the target, as this would 833 * orphan everything below the source directory. Also 834 * the user must have write permission in the source so 835 * as to be able to change "..". We must repeat the call 836 * to namei, as the parent directory is unlocked by the 837 * call to checkpath(). 838 */ 839 error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, tcnp->cn_thread); 840 VOP_UNLOCK(fvp, 0); 841 if (oldparent != dp->i_number) 842 newparent = dp->i_number; 843 if (doingdirectory && newparent) { 844 if (error) /* write access check above */ 845 goto bad; 846 if (xp != NULL) 847 vput(tvp); 848 error = ext2_checkpath(ip, dp, tcnp->cn_cred); 849 if (error) 850 goto out; 851 VREF(tdvp); 852 error = relookup(tdvp, &tvp, tcnp); 853 if (error) 854 goto out; 855 vrele(tdvp); 856 dp = VTOI(tdvp); 857 xp = NULL; 858 if (tvp) 859 xp = VTOI(tvp); 860 } 861 /* 862 * 2) If target doesn't exist, link the target 863 * to the source and unlink the source. 864 * Otherwise, rewrite the target directory 865 * entry to reference the source inode and 866 * expunge the original entry's existence. 867 */ 868 if (xp == NULL) { 869 if (dp->i_devvp != ip->i_devvp) 870 panic("ext2_rename: EXDEV"); 871 /* 872 * Account for ".." in new directory. 873 * When source and destination have the same 874 * parent we don't fool with the link count. 875 */ 876 if (doingdirectory && newparent) { 877 if ((nlink_t)dp->i_nlink >= EXT2_LINK_MAX) { 878 error = EMLINK; 879 goto bad; 880 } 881 dp->i_nlink++; 882 dp->i_flag |= IN_CHANGE; 883 error = ext2_update(tdvp, !DOINGASYNC(tdvp)); 884 if (error) 885 goto bad; 886 } 887 error = ext2_direnter(ip, tdvp, tcnp); 888 if (error) { 889 if (doingdirectory && newparent) { 890 dp->i_nlink--; 891 dp->i_flag |= IN_CHANGE; 892 (void)ext2_update(tdvp, 1); 893 } 894 goto bad; 895 } 896 vput(tdvp); 897 } else { 898 if (xp->i_devvp != dp->i_devvp || xp->i_devvp != ip->i_devvp) 899 panic("ext2_rename: EXDEV"); 900 /* 901 * Short circuit rename(foo, foo). 902 */ 903 if (xp->i_number == ip->i_number) 904 panic("ext2_rename: same file"); 905 /* 906 * If the parent directory is "sticky", then the user must 907 * own the parent directory, or the destination of the rename, 908 * otherwise the destination may not be changed (except by 909 * root). This implements append-only directories. 910 */ 911 if ((dp->i_mode & S_ISTXT) && tcnp->cn_cred->cr_uid != 0 && 912 tcnp->cn_cred->cr_uid != dp->i_uid && 913 xp->i_uid != tcnp->cn_cred->cr_uid) { 914 error = EPERM; 915 goto bad; 916 } 917 /* 918 * Target must be empty if a directory and have no links 919 * to it. Also, ensure source and target are compatible 920 * (both directories, or both not directories). 921 */ 922 if ((xp->i_mode&IFMT) == IFDIR) { 923 if (! ext2_dirempty(xp, dp->i_number, tcnp->cn_cred) || 924 xp->i_nlink > 2) { 925 error = ENOTEMPTY; 926 goto bad; 927 } 928 if (!doingdirectory) { 929 error = ENOTDIR; 930 goto bad; 931 } 932 cache_purge(tdvp); 933 } else if (doingdirectory) { 934 error = EISDIR; 935 goto bad; 936 } 937 error = ext2_dirrewrite(dp, ip, tcnp); 938 if (error) 939 goto bad; 940 /* 941 * If the target directory is in the same 942 * directory as the source directory, 943 * decrement the link count on the parent 944 * of the target directory. 945 */ 946 if (doingdirectory && !newparent) { 947 dp->i_nlink--; 948 dp->i_flag |= IN_CHANGE; 949 } 950 vput(tdvp); 951 /* 952 * Adjust the link count of the target to 953 * reflect the dirrewrite above. If this is 954 * a directory it is empty and there are 955 * no links to it, so we can squash the inode and 956 * any space associated with it. We disallowed 957 * renaming over top of a directory with links to 958 * it above, as the remaining link would point to 959 * a directory without "." or ".." entries. 960 */ 961 xp->i_nlink--; 962 if (doingdirectory) { 963 if (--xp->i_nlink != 0) 964 panic("ext2_rename: linked directory"); 965 error = ext2_truncate(tvp, (off_t)0, IO_SYNC, 966 tcnp->cn_cred, tcnp->cn_thread); 967 } 968 xp->i_flag |= IN_CHANGE; 969 vput(tvp); 970 xp = NULL; 971 } 972 973 /* 974 * 3) Unlink the source. 975 */ 976 fcnp->cn_flags &= ~MODMASK; 977 fcnp->cn_flags |= LOCKPARENT | LOCKLEAF; 978 VREF(fdvp); 979 error = relookup(fdvp, &fvp, fcnp); 980 if (error == 0) 981 vrele(fdvp); 982 if (fvp != NULL) { 983 xp = VTOI(fvp); 984 dp = VTOI(fdvp); 985 } else { 986 /* 987 * From name has disappeared. 988 */ 989 if (doingdirectory) 990 panic("ext2_rename: lost dir entry"); 991 vrele(ap->a_fvp); 992 return (0); 993 } 994 /* 995 * Ensure that the directory entry still exists and has not 996 * changed while the new name has been entered. If the source is 997 * a file then the entry may have been unlinked or renamed. In 998 * either case there is no further work to be done. If the source 999 * is a directory then it cannot have been rmdir'ed; its link 1000 * count of three would cause a rmdir to fail with ENOTEMPTY. 1001 * The IN_RENAME flag ensures that it cannot be moved by another 1002 * rename. 1003 */ 1004 if (xp != ip) { 1005 if (doingdirectory) 1006 panic("ext2_rename: lost dir entry"); 1007 } else { 1008 /* 1009 * If the source is a directory with a 1010 * new parent, the link count of the old 1011 * parent directory must be decremented 1012 * and ".." set to point to the new parent. 1013 */ 1014 if (doingdirectory && newparent) { 1015 dp->i_nlink--; 1016 dp->i_flag |= IN_CHANGE; 1017 error = vn_rdwr(UIO_READ, fvp, (caddr_t)&dirbuf, 1018 sizeof(struct dirtemplate), (off_t)0, 1019 UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK, 1020 tcnp->cn_cred, NOCRED, NULL, NULL); 1021 if (error == 0) { 1022 /* Like ufs little-endian: */ 1023 namlen = dirbuf.dotdot_type; 1024 if (namlen != 2 || 1025 dirbuf.dotdot_name[0] != '.' || 1026 dirbuf.dotdot_name[1] != '.') { 1027 ext2_dirbad(xp, (doff_t)12, 1028 "rename: mangled dir"); 1029 } else { 1030 dirbuf.dotdot_ino = newparent; 1031 (void) vn_rdwr(UIO_WRITE, fvp, 1032 (caddr_t)&dirbuf, 1033 sizeof(struct dirtemplate), 1034 (off_t)0, UIO_SYSSPACE, 1035 IO_NODELOCKED | IO_SYNC | 1036 IO_NOMACCHECK, tcnp->cn_cred, 1037 NOCRED, NULL, NULL); 1038 cache_purge(fdvp); 1039 } 1040 } 1041 } 1042 error = ext2_dirremove(fdvp, fcnp); 1043 if (!error) { 1044 xp->i_nlink--; 1045 xp->i_flag |= IN_CHANGE; 1046 } 1047 xp->i_flag &= ~IN_RENAME; 1048 } 1049 if (dp) 1050 vput(fdvp); 1051 if (xp) 1052 vput(fvp); 1053 vrele(ap->a_fvp); 1054 return (error); 1055 1056 bad: 1057 if (xp) 1058 vput(ITOV(xp)); 1059 vput(ITOV(dp)); 1060 out: 1061 if (doingdirectory) 1062 ip->i_flag &= ~IN_RENAME; 1063 if (vn_lock(fvp, LK_EXCLUSIVE) == 0) { 1064 ip->i_nlink--; 1065 ip->i_flag |= IN_CHANGE; 1066 ip->i_flag &= ~IN_RENAME; 1067 vput(fvp); 1068 } else 1069 vrele(fvp); 1070 return (error); 1071 } 1072 1073 /* 1074 * Mkdir system call 1075 */ 1076 static int 1077 ext2_mkdir(struct vop_mkdir_args *ap) 1078 { 1079 struct vnode *dvp = ap->a_dvp; 1080 struct vattr *vap = ap->a_vap; 1081 struct componentname *cnp = ap->a_cnp; 1082 struct inode *ip, *dp; 1083 struct vnode *tvp; 1084 struct dirtemplate dirtemplate, *dtp; 1085 int error, dmode; 1086 1087 #ifdef INVARIANTS 1088 if ((cnp->cn_flags & HASBUF) == 0) 1089 panic("ext2_mkdir: no name"); 1090 #endif 1091 dp = VTOI(dvp); 1092 if ((nlink_t)dp->i_nlink >= EXT2_LINK_MAX) { 1093 error = EMLINK; 1094 goto out; 1095 } 1096 dmode = vap->va_mode & 0777; 1097 dmode |= IFDIR; 1098 /* 1099 * Must simulate part of ext2_makeinode here to acquire the inode, 1100 * but not have it entered in the parent directory. The entry is 1101 * made later after writing "." and ".." entries. 1102 */ 1103 error = ext2_valloc(dvp, dmode, cnp->cn_cred, &tvp); 1104 if (error) 1105 goto out; 1106 ip = VTOI(tvp); 1107 ip->i_gid = dp->i_gid; 1108 #ifdef SUIDDIR 1109 { 1110 /* 1111 * if we are hacking owners here, (only do this where told to) 1112 * and we are not giving it TOO root, (would subvert quotas) 1113 * then go ahead and give it to the other user. 1114 * The new directory also inherits the SUID bit. 1115 * If user's UID and dir UID are the same, 1116 * 'give it away' so that the SUID is still forced on. 1117 */ 1118 if ( (dvp->v_mount->mnt_flag & MNT_SUIDDIR) && 1119 (dp->i_mode & ISUID) && dp->i_uid) { 1120 dmode |= ISUID; 1121 ip->i_uid = dp->i_uid; 1122 } else { 1123 ip->i_uid = cnp->cn_cred->cr_uid; 1124 } 1125 } 1126 #else 1127 ip->i_uid = cnp->cn_cred->cr_uid; 1128 #endif 1129 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 1130 ip->i_mode = dmode; 1131 tvp->v_type = VDIR; /* Rest init'd in getnewvnode(). */ 1132 ip->i_nlink = 2; 1133 if (cnp->cn_flags & ISWHITEOUT) 1134 ip->i_flags |= UF_OPAQUE; 1135 error = ext2_update(tvp, 1); 1136 1137 /* 1138 * Bump link count in parent directory 1139 * to reflect work done below. Should 1140 * be done before reference is created 1141 * so reparation is possible if we crash. 1142 */ 1143 dp->i_nlink++; 1144 dp->i_flag |= IN_CHANGE; 1145 error = ext2_update(dvp, !DOINGASYNC(dvp)); 1146 if (error) 1147 goto bad; 1148 1149 /* Initialize directory with "." and ".." from static template. */ 1150 if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs, 1151 EXT2F_INCOMPAT_FTYPE)) 1152 dtp = &mastertemplate; 1153 else 1154 dtp = &omastertemplate; 1155 dirtemplate = *dtp; 1156 dirtemplate.dot_ino = ip->i_number; 1157 dirtemplate.dotdot_ino = dp->i_number; 1158 /* note that in ext2 DIRBLKSIZ == blocksize, not DEV_BSIZE 1159 * so let's just redefine it - for this function only 1160 */ 1161 #undef DIRBLKSIZ 1162 #define DIRBLKSIZ VTOI(dvp)->i_e2fs->e2fs_bsize 1163 dirtemplate.dotdot_reclen = DIRBLKSIZ - 12; 1164 error = vn_rdwr(UIO_WRITE, tvp, (caddr_t)&dirtemplate, 1165 sizeof(dirtemplate), (off_t)0, UIO_SYSSPACE, 1166 IO_NODELOCKED | IO_SYNC | IO_NOMACCHECK, cnp->cn_cred, NOCRED, 1167 NULL, NULL); 1168 if (error) { 1169 dp->i_nlink--; 1170 dp->i_flag |= IN_CHANGE; 1171 goto bad; 1172 } 1173 if (DIRBLKSIZ > VFSTOEXT2(dvp->v_mount)->um_mountp->mnt_stat.f_bsize) 1174 /* XXX should grow with balloc() */ 1175 panic("ext2_mkdir: blksize"); 1176 else { 1177 ip->i_size = DIRBLKSIZ; 1178 ip->i_flag |= IN_CHANGE; 1179 } 1180 1181 /* Directory set up, now install its entry in the parent directory. */ 1182 error = ext2_direnter(ip, dvp, cnp); 1183 if (error) { 1184 dp->i_nlink--; 1185 dp->i_flag |= IN_CHANGE; 1186 } 1187 bad: 1188 /* 1189 * No need to do an explicit VOP_TRUNCATE here, vrele will do this 1190 * for us because we set the link count to 0. 1191 */ 1192 if (error) { 1193 ip->i_nlink = 0; 1194 ip->i_flag |= IN_CHANGE; 1195 vput(tvp); 1196 } else 1197 *ap->a_vpp = tvp; 1198 out: 1199 return (error); 1200 #undef DIRBLKSIZ 1201 #define DIRBLKSIZ DEV_BSIZE 1202 } 1203 1204 /* 1205 * Rmdir system call. 1206 */ 1207 static int 1208 ext2_rmdir(struct vop_rmdir_args *ap) 1209 { 1210 struct vnode *vp = ap->a_vp; 1211 struct vnode *dvp = ap->a_dvp; 1212 struct componentname *cnp = ap->a_cnp; 1213 struct inode *ip, *dp; 1214 int error; 1215 1216 ip = VTOI(vp); 1217 dp = VTOI(dvp); 1218 1219 /* 1220 * Verify the directory is empty (and valid). 1221 * (Rmdir ".." won't be valid since 1222 * ".." will contain a reference to 1223 * the current directory and thus be 1224 * non-empty.) 1225 */ 1226 if (ip->i_nlink != 2 || !ext2_dirempty(ip, dp->i_number, cnp->cn_cred)) { 1227 error = ENOTEMPTY; 1228 goto out; 1229 } 1230 if ((dp->i_flags & APPEND) 1231 || (ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))) { 1232 error = EPERM; 1233 goto out; 1234 } 1235 /* 1236 * Delete reference to directory before purging 1237 * inode. If we crash in between, the directory 1238 * will be reattached to lost+found, 1239 */ 1240 error = ext2_dirremove(dvp, cnp); 1241 if (error) 1242 goto out; 1243 dp->i_nlink--; 1244 dp->i_flag |= IN_CHANGE; 1245 cache_purge(dvp); 1246 VOP_UNLOCK(dvp, 0); 1247 /* 1248 * Truncate inode. The only stuff left 1249 * in the directory is "." and "..". The 1250 * "." reference is inconsequential since 1251 * we're quashing it. The ".." reference 1252 * has already been adjusted above. We've 1253 * removed the "." reference and the reference 1254 * in the parent directory, but there may be 1255 * other hard links so decrement by 2 and 1256 * worry about them later. 1257 */ 1258 ip->i_nlink -= 2; 1259 error = ext2_truncate(vp, (off_t)0, IO_SYNC, cnp->cn_cred, 1260 cnp->cn_thread); 1261 cache_purge(ITOV(ip)); 1262 if (vn_lock(dvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) { 1263 VOP_UNLOCK(vp, 0); 1264 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 1265 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1266 } 1267 out: 1268 return (error); 1269 } 1270 1271 /* 1272 * symlink -- make a symbolic link 1273 */ 1274 static int 1275 ext2_symlink(struct vop_symlink_args *ap) 1276 { 1277 struct vnode *vp, **vpp = ap->a_vpp; 1278 struct inode *ip; 1279 int len, error; 1280 1281 error = ext2_makeinode(IFLNK | ap->a_vap->va_mode, ap->a_dvp, 1282 vpp, ap->a_cnp); 1283 if (error) 1284 return (error); 1285 vp = *vpp; 1286 len = strlen(ap->a_target); 1287 if (len < vp->v_mount->mnt_maxsymlinklen) { 1288 ip = VTOI(vp); 1289 bcopy(ap->a_target, (char *)ip->i_shortlink, len); 1290 ip->i_size = len; 1291 ip->i_flag |= IN_CHANGE | IN_UPDATE; 1292 } else 1293 error = vn_rdwr(UIO_WRITE, vp, ap->a_target, len, (off_t)0, 1294 UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK, 1295 ap->a_cnp->cn_cred, NOCRED, NULL, NULL); 1296 if (error) 1297 vput(vp); 1298 return (error); 1299 } 1300 1301 /* 1302 * Return target name of a symbolic link 1303 */ 1304 static int 1305 ext2_readlink(struct vop_readlink_args *ap) 1306 { 1307 struct vnode *vp = ap->a_vp; 1308 struct inode *ip = VTOI(vp); 1309 int isize; 1310 1311 isize = ip->i_size; 1312 if (isize < vp->v_mount->mnt_maxsymlinklen) { 1313 uiomove((char *)ip->i_shortlink, isize, ap->a_uio); 1314 return (0); 1315 } 1316 return (VOP_READ(vp, ap->a_uio, 0, ap->a_cred)); 1317 } 1318 1319 /* 1320 * Calculate the logical to physical mapping if not done already, 1321 * then call the device strategy routine. 1322 * 1323 * In order to be able to swap to a file, the ext2_bmaparray() operation may not 1324 * deadlock on memory. See ext2_bmap() for details. 1325 */ 1326 static int 1327 ext2_strategy(struct vop_strategy_args *ap) 1328 { 1329 struct buf *bp = ap->a_bp; 1330 struct vnode *vp = ap->a_vp; 1331 struct bufobj *bo; 1332 daddr_t blkno; 1333 int error; 1334 1335 if (vp->v_type == VBLK || vp->v_type == VCHR) 1336 panic("ext2_strategy: spec"); 1337 if (bp->b_blkno == bp->b_lblkno) { 1338 error = ext2_bmaparray(vp, bp->b_lblkno, &blkno, NULL, NULL); 1339 bp->b_blkno = blkno; 1340 if (error) { 1341 bp->b_error = error; 1342 bp->b_ioflags |= BIO_ERROR; 1343 bufdone(bp); 1344 return (0); 1345 } 1346 if ((long)bp->b_blkno == -1) 1347 vfs_bio_clrbuf(bp); 1348 } 1349 if ((long)bp->b_blkno == -1) { 1350 bufdone(bp); 1351 return (0); 1352 } 1353 bp->b_iooffset = dbtob(bp->b_blkno); 1354 bo = VFSTOEXT2(vp->v_mount)->um_bo; 1355 BO_STRATEGY(bo, bp); 1356 return (0); 1357 } 1358 1359 /* 1360 * Print out the contents of an inode. 1361 */ 1362 static int 1363 ext2_print(struct vop_print_args *ap) 1364 { 1365 struct vnode *vp = ap->a_vp; 1366 struct inode *ip = VTOI(vp); 1367 1368 vn_printf(ip->i_devvp, "\tino %ju", (uintmax_t)ip->i_number); 1369 if (vp->v_type == VFIFO) 1370 fifo_printinfo(vp); 1371 printf("\n"); 1372 return (0); 1373 } 1374 1375 /* 1376 * Close wrapper for fifos. 1377 * 1378 * Update the times on the inode then do device close. 1379 */ 1380 static int 1381 ext2fifo_close(struct vop_close_args *ap) 1382 { 1383 struct vnode *vp = ap->a_vp; 1384 1385 VI_LOCK(vp); 1386 if (vp->v_usecount > 1) 1387 ext2_itimes_locked(vp); 1388 VI_UNLOCK(vp); 1389 return (fifo_specops.vop_close(ap)); 1390 } 1391 1392 /* 1393 * Kqfilter wrapper for fifos. 1394 * 1395 * Fall through to ext2 kqfilter routines if needed 1396 */ 1397 static int 1398 ext2fifo_kqfilter(struct vop_kqfilter_args *ap) 1399 { 1400 int error; 1401 1402 error = fifo_specops.vop_kqfilter(ap); 1403 if (error) 1404 error = vfs_kqfilter(ap); 1405 return (error); 1406 } 1407 1408 /* 1409 * Return POSIX pathconf information applicable to ext2 filesystems. 1410 */ 1411 static int 1412 ext2_pathconf(struct vop_pathconf_args *ap) 1413 { 1414 int error = 0; 1415 1416 switch (ap->a_name) { 1417 case _PC_LINK_MAX: 1418 *ap->a_retval = EXT2_LINK_MAX; 1419 break; 1420 case _PC_NAME_MAX: 1421 *ap->a_retval = NAME_MAX; 1422 break; 1423 case _PC_PATH_MAX: 1424 *ap->a_retval = PATH_MAX; 1425 break; 1426 case _PC_PIPE_BUF: 1427 *ap->a_retval = PIPE_BUF; 1428 break; 1429 case _PC_CHOWN_RESTRICTED: 1430 *ap->a_retval = 1; 1431 break; 1432 case _PC_NO_TRUNC: 1433 *ap->a_retval = 1; 1434 break; 1435 case _PC_MIN_HOLE_SIZE: 1436 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize; 1437 break; 1438 case _PC_ASYNC_IO: 1439 /* _PC_ASYNC_IO should have been handled by upper layers. */ 1440 KASSERT(0, ("_PC_ASYNC_IO should not get here")); 1441 error = EINVAL; 1442 break; 1443 case _PC_PRIO_IO: 1444 *ap->a_retval = 0; 1445 break; 1446 case _PC_SYNC_IO: 1447 *ap->a_retval = 0; 1448 break; 1449 case _PC_ALLOC_SIZE_MIN: 1450 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_bsize; 1451 break; 1452 case _PC_FILESIZEBITS: 1453 *ap->a_retval = 64; 1454 break; 1455 case _PC_REC_INCR_XFER_SIZE: 1456 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize; 1457 break; 1458 case _PC_REC_MAX_XFER_SIZE: 1459 *ap->a_retval = -1; /* means ``unlimited'' */ 1460 break; 1461 case _PC_REC_MIN_XFER_SIZE: 1462 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize; 1463 break; 1464 case _PC_REC_XFER_ALIGN: 1465 *ap->a_retval = PAGE_SIZE; 1466 break; 1467 case _PC_SYMLINK_MAX: 1468 *ap->a_retval = MAXPATHLEN; 1469 break; 1470 1471 default: 1472 error = EINVAL; 1473 break; 1474 } 1475 return (error); 1476 } 1477 1478 /* 1479 * Vnode pointer to File handle 1480 */ 1481 /* ARGSUSED */ 1482 static int 1483 ext2_vptofh(struct vop_vptofh_args *ap) 1484 { 1485 struct inode *ip; 1486 struct ufid *ufhp; 1487 1488 ip = VTOI(ap->a_vp); 1489 ufhp = (struct ufid *)ap->a_fhp; 1490 ufhp->ufid_len = sizeof(struct ufid); 1491 ufhp->ufid_ino = ip->i_number; 1492 ufhp->ufid_gen = ip->i_gen; 1493 return (0); 1494 } 1495 1496 /* 1497 * Initialize the vnode associated with a new inode, handle aliased 1498 * vnodes. 1499 */ 1500 int 1501 ext2_vinit(struct mount *mntp, struct vop_vector *fifoops, struct vnode **vpp) 1502 { 1503 struct inode *ip; 1504 struct vnode *vp; 1505 1506 vp = *vpp; 1507 ip = VTOI(vp); 1508 vp->v_type = IFTOVT(ip->i_mode); 1509 if (vp->v_type == VFIFO) 1510 vp->v_op = fifoops; 1511 1512 if (ip->i_number == EXT2_ROOTINO) 1513 vp->v_vflag |= VV_ROOT; 1514 ip->i_modrev = init_va_filerev(); 1515 *vpp = vp; 1516 return (0); 1517 } 1518 1519 /* 1520 * Allocate a new inode. 1521 */ 1522 static int 1523 ext2_makeinode(int mode, struct vnode *dvp, struct vnode **vpp, 1524 struct componentname *cnp) 1525 { 1526 struct inode *ip, *pdir; 1527 struct vnode *tvp; 1528 int error; 1529 1530 pdir = VTOI(dvp); 1531 #ifdef INVARIANTS 1532 if ((cnp->cn_flags & HASBUF) == 0) 1533 panic("ext2_makeinode: no name"); 1534 #endif 1535 *vpp = NULL; 1536 if ((mode & IFMT) == 0) 1537 mode |= IFREG; 1538 1539 error = ext2_valloc(dvp, mode, cnp->cn_cred, &tvp); 1540 if (error) { 1541 return (error); 1542 } 1543 ip = VTOI(tvp); 1544 ip->i_gid = pdir->i_gid; 1545 #ifdef SUIDDIR 1546 { 1547 /* 1548 * if we are 1549 * not the owner of the directory, 1550 * and we are hacking owners here, (only do this where told to) 1551 * and we are not giving it TOO root, (would subvert quotas) 1552 * then go ahead and give it to the other user. 1553 * Note that this drops off the execute bits for security. 1554 */ 1555 if ( (dvp->v_mount->mnt_flag & MNT_SUIDDIR) && 1556 (pdir->i_mode & ISUID) && 1557 (pdir->i_uid != cnp->cn_cred->cr_uid) && pdir->i_uid) { 1558 ip->i_uid = pdir->i_uid; 1559 mode &= ~07111; 1560 } else { 1561 ip->i_uid = cnp->cn_cred->cr_uid; 1562 } 1563 } 1564 #else 1565 ip->i_uid = cnp->cn_cred->cr_uid; 1566 #endif 1567 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 1568 ip->i_mode = mode; 1569 tvp->v_type = IFTOVT(mode); /* Rest init'd in getnewvnode(). */ 1570 ip->i_nlink = 1; 1571 if ((ip->i_mode & ISGID) && !groupmember(ip->i_gid, cnp->cn_cred)) { 1572 if (priv_check_cred(cnp->cn_cred, PRIV_VFS_RETAINSUGID, 0)) 1573 ip->i_mode &= ~ISGID; 1574 } 1575 1576 if (cnp->cn_flags & ISWHITEOUT) 1577 ip->i_flags |= UF_OPAQUE; 1578 1579 /* 1580 * Make sure inode goes to disk before directory entry. 1581 */ 1582 error = ext2_update(tvp, !DOINGASYNC(tvp)); 1583 if (error) 1584 goto bad; 1585 error = ext2_direnter(ip, dvp, cnp); 1586 if (error) 1587 goto bad; 1588 1589 *vpp = tvp; 1590 return (0); 1591 1592 bad: 1593 /* 1594 * Write error occurred trying to update the inode 1595 * or the directory so must deallocate the inode. 1596 */ 1597 ip->i_nlink = 0; 1598 ip->i_flag |= IN_CHANGE; 1599 vput(tvp); 1600 return (error); 1601 } 1602 1603 /* 1604 * Vnode op for reading. 1605 */ 1606 static int 1607 ext2_read(struct vop_read_args *ap) 1608 { 1609 struct vnode *vp; 1610 struct inode *ip; 1611 int error; 1612 1613 vp = ap->a_vp; 1614 ip = VTOI(vp); 1615 1616 /*EXT4_EXT_LOCK(ip);*/ 1617 if (ip->i_flag & IN_E4EXTENTS) 1618 error = ext4_ext_read(ap); 1619 else 1620 error = ext2_ind_read(ap); 1621 /*EXT4_EXT_UNLOCK(ip);*/ 1622 return (error); 1623 } 1624 1625 /* 1626 * Vnode op for reading. 1627 */ 1628 static int 1629 ext2_ind_read(struct vop_read_args *ap) 1630 { 1631 struct vnode *vp; 1632 struct inode *ip; 1633 struct uio *uio; 1634 struct m_ext2fs *fs; 1635 struct buf *bp; 1636 daddr_t lbn, nextlbn; 1637 off_t bytesinfile; 1638 long size, xfersize, blkoffset; 1639 int error, orig_resid, seqcount; 1640 int ioflag; 1641 1642 vp = ap->a_vp; 1643 uio = ap->a_uio; 1644 ioflag = ap->a_ioflag; 1645 1646 seqcount = ap->a_ioflag >> IO_SEQSHIFT; 1647 ip = VTOI(vp); 1648 1649 #ifdef INVARIANTS 1650 if (uio->uio_rw != UIO_READ) 1651 panic("%s: mode", "ext2_read"); 1652 1653 if (vp->v_type == VLNK) { 1654 if ((int)ip->i_size < vp->v_mount->mnt_maxsymlinklen) 1655 panic("%s: short symlink", "ext2_read"); 1656 } else if (vp->v_type != VREG && vp->v_type != VDIR) 1657 panic("%s: type %d", "ext2_read", vp->v_type); 1658 #endif 1659 orig_resid = uio->uio_resid; 1660 KASSERT(orig_resid >= 0, ("ext2_read: uio->uio_resid < 0")); 1661 if (orig_resid == 0) 1662 return (0); 1663 KASSERT(uio->uio_offset >= 0, ("ext2_read: uio->uio_offset < 0")); 1664 fs = ip->i_e2fs; 1665 if (uio->uio_offset < ip->i_size && 1666 uio->uio_offset >= fs->e2fs_maxfilesize) 1667 return (EOVERFLOW); 1668 1669 for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) { 1670 if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0) 1671 break; 1672 lbn = lblkno(fs, uio->uio_offset); 1673 nextlbn = lbn + 1; 1674 size = blksize(fs, ip, lbn); 1675 blkoffset = blkoff(fs, uio->uio_offset); 1676 1677 xfersize = fs->e2fs_fsize - blkoffset; 1678 if (uio->uio_resid < xfersize) 1679 xfersize = uio->uio_resid; 1680 if (bytesinfile < xfersize) 1681 xfersize = bytesinfile; 1682 1683 if (lblktosize(fs, nextlbn) >= ip->i_size) 1684 error = bread(vp, lbn, size, NOCRED, &bp); 1685 else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) { 1686 error = cluster_read(vp, ip->i_size, lbn, size, 1687 NOCRED, blkoffset + uio->uio_resid, seqcount, 1688 0, &bp); 1689 } else if (seqcount > 1) { 1690 u_int nextsize = blksize(fs, ip, nextlbn); 1691 error = breadn(vp, lbn, 1692 size, &nextlbn, &nextsize, 1, NOCRED, &bp); 1693 } else 1694 error = bread(vp, lbn, size, NOCRED, &bp); 1695 if (error) { 1696 brelse(bp); 1697 bp = NULL; 1698 break; 1699 } 1700 1701 /* 1702 * If IO_DIRECT then set B_DIRECT for the buffer. This 1703 * will cause us to attempt to release the buffer later on 1704 * and will cause the buffer cache to attempt to free the 1705 * underlying pages. 1706 */ 1707 if (ioflag & IO_DIRECT) 1708 bp->b_flags |= B_DIRECT; 1709 1710 /* 1711 * We should only get non-zero b_resid when an I/O error 1712 * has occurred, which should cause us to break above. 1713 * However, if the short read did not cause an error, 1714 * then we want to ensure that we do not uiomove bad 1715 * or uninitialized data. 1716 */ 1717 size -= bp->b_resid; 1718 if (size < xfersize) { 1719 if (size == 0) 1720 break; 1721 xfersize = size; 1722 } 1723 error = uiomove((char *)bp->b_data + blkoffset, 1724 (int)xfersize, uio); 1725 if (error) 1726 break; 1727 1728 if (ioflag & (IO_VMIO|IO_DIRECT)) { 1729 /* 1730 * If it's VMIO or direct I/O, then we don't 1731 * need the buf, mark it available for 1732 * freeing. If it's non-direct VMIO, the VM has 1733 * the data. 1734 */ 1735 bp->b_flags |= B_RELBUF; 1736 brelse(bp); 1737 } else { 1738 /* 1739 * Otherwise let whoever 1740 * made the request take care of 1741 * freeing it. We just queue 1742 * it onto another list. 1743 */ 1744 bqrelse(bp); 1745 } 1746 } 1747 1748 /* 1749 * This can only happen in the case of an error 1750 * because the loop above resets bp to NULL on each iteration 1751 * and on normal completion has not set a new value into it. 1752 * so it must have come from a 'break' statement 1753 */ 1754 if (bp != NULL) { 1755 if (ioflag & (IO_VMIO|IO_DIRECT)) { 1756 bp->b_flags |= B_RELBUF; 1757 brelse(bp); 1758 } else { 1759 bqrelse(bp); 1760 } 1761 } 1762 1763 if ((error == 0 || uio->uio_resid != orig_resid) && 1764 (vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) 1765 ip->i_flag |= IN_ACCESS; 1766 return (error); 1767 } 1768 1769 static int 1770 ext2_ioctl(struct vop_ioctl_args *ap) 1771 { 1772 1773 switch (ap->a_command) { 1774 case FIOSEEKDATA: 1775 case FIOSEEKHOLE: 1776 return (vn_bmap_seekhole(ap->a_vp, ap->a_command, 1777 (off_t *)ap->a_data, ap->a_cred)); 1778 default: 1779 return (ENOTTY); 1780 } 1781 } 1782 1783 /* 1784 * this function handles ext4 extents block mapping 1785 */ 1786 static int 1787 ext4_ext_read(struct vop_read_args *ap) 1788 { 1789 struct vnode *vp; 1790 struct inode *ip; 1791 struct uio *uio; 1792 struct m_ext2fs *fs; 1793 struct buf *bp; 1794 struct ext4_extent nex, *ep; 1795 struct ext4_extent_path path; 1796 daddr_t lbn, newblk; 1797 off_t bytesinfile; 1798 int cache_type; 1799 ssize_t orig_resid; 1800 int error; 1801 long size, xfersize, blkoffset; 1802 1803 vp = ap->a_vp; 1804 ip = VTOI(vp); 1805 uio = ap->a_uio; 1806 memset(&path, 0, sizeof(path)); 1807 1808 orig_resid = uio->uio_resid; 1809 KASSERT(orig_resid >= 0, ("%s: uio->uio_resid < 0", __func__)); 1810 if (orig_resid == 0) 1811 return (0); 1812 KASSERT(uio->uio_offset >= 0, ("%s: uio->uio_offset < 0", __func__)); 1813 fs = ip->i_e2fs; 1814 if (uio->uio_offset < ip->i_size && uio->uio_offset >= fs->e2fs_maxfilesize) 1815 return (EOVERFLOW); 1816 1817 while (uio->uio_resid > 0) { 1818 if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0) 1819 break; 1820 lbn = lblkno(fs, uio->uio_offset); 1821 size = blksize(fs, ip, lbn); 1822 blkoffset = blkoff(fs, uio->uio_offset); 1823 1824 xfersize = fs->e2fs_fsize - blkoffset; 1825 xfersize = MIN(xfersize, uio->uio_resid); 1826 xfersize = MIN(xfersize, bytesinfile); 1827 1828 /* get block from ext4 extent cache */ 1829 cache_type = ext4_ext_in_cache(ip, lbn, &nex); 1830 switch (cache_type) { 1831 case EXT4_EXT_CACHE_NO: 1832 ext4_ext_find_extent(fs, ip, lbn, &path); 1833 ep = path.ep_ext; 1834 if (ep == NULL) 1835 return (EIO); 1836 1837 ext4_ext_put_cache(ip, ep, EXT4_EXT_CACHE_IN); 1838 1839 newblk = lbn - ep->e_blk + (ep->e_start_lo | 1840 (daddr_t)ep->e_start_hi << 32); 1841 1842 if (path.ep_bp != NULL) { 1843 brelse(path.ep_bp); 1844 path.ep_bp = NULL; 1845 } 1846 break; 1847 1848 case EXT4_EXT_CACHE_GAP: 1849 /* block has not been allocated yet */ 1850 return (0); 1851 1852 case EXT4_EXT_CACHE_IN: 1853 newblk = lbn - nex.e_blk + (nex.e_start_lo | 1854 (daddr_t)nex.e_start_hi << 32); 1855 break; 1856 1857 default: 1858 panic("%s: invalid cache type", __func__); 1859 } 1860 1861 error = bread(ip->i_devvp, fsbtodb(fs, newblk), size, NOCRED, &bp); 1862 if (error) { 1863 brelse(bp); 1864 return (error); 1865 } 1866 1867 size -= bp->b_resid; 1868 if (size < xfersize) { 1869 if (size == 0) { 1870 bqrelse(bp); 1871 break; 1872 } 1873 xfersize = size; 1874 } 1875 error = uiomove(bp->b_data + blkoffset, (int)xfersize, uio); 1876 bqrelse(bp); 1877 if (error) 1878 return (error); 1879 } 1880 1881 return (0); 1882 } 1883 1884 /* 1885 * Vnode op for writing. 1886 */ 1887 static int 1888 ext2_write(struct vop_write_args *ap) 1889 { 1890 struct vnode *vp; 1891 struct uio *uio; 1892 struct inode *ip; 1893 struct m_ext2fs *fs; 1894 struct buf *bp; 1895 daddr_t lbn; 1896 off_t osize; 1897 int blkoffset, error, flags, ioflag, resid, size, seqcount, xfersize; 1898 1899 ioflag = ap->a_ioflag; 1900 uio = ap->a_uio; 1901 vp = ap->a_vp; 1902 1903 seqcount = ioflag >> IO_SEQSHIFT; 1904 ip = VTOI(vp); 1905 1906 #ifdef INVARIANTS 1907 if (uio->uio_rw != UIO_WRITE) 1908 panic("%s: mode", "ext2_write"); 1909 #endif 1910 1911 switch (vp->v_type) { 1912 case VREG: 1913 if (ioflag & IO_APPEND) 1914 uio->uio_offset = ip->i_size; 1915 if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size) 1916 return (EPERM); 1917 /* FALLTHROUGH */ 1918 case VLNK: 1919 break; 1920 case VDIR: 1921 /* XXX differs from ffs -- this is called from ext2_mkdir(). */ 1922 if ((ioflag & IO_SYNC) == 0) 1923 panic("ext2_write: nonsync dir write"); 1924 break; 1925 default: 1926 panic("ext2_write: type %p %d (%jd,%jd)", (void *)vp, 1927 vp->v_type, (intmax_t)uio->uio_offset, 1928 (intmax_t)uio->uio_resid); 1929 } 1930 1931 KASSERT(uio->uio_resid >= 0, ("ext2_write: uio->uio_resid < 0")); 1932 KASSERT(uio->uio_offset >= 0, ("ext2_write: uio->uio_offset < 0")); 1933 fs = ip->i_e2fs; 1934 if ((uoff_t)uio->uio_offset + uio->uio_resid > fs->e2fs_maxfilesize) 1935 return (EFBIG); 1936 /* 1937 * Maybe this should be above the vnode op call, but so long as 1938 * file servers have no limits, I don't think it matters. 1939 */ 1940 if (vn_rlimit_fsize(vp, uio, uio->uio_td)) 1941 return (EFBIG); 1942 1943 resid = uio->uio_resid; 1944 osize = ip->i_size; 1945 if (seqcount > BA_SEQMAX) 1946 flags = BA_SEQMAX << BA_SEQSHIFT; 1947 else 1948 flags = seqcount << BA_SEQSHIFT; 1949 if ((ioflag & IO_SYNC) && !DOINGASYNC(vp)) 1950 flags |= IO_SYNC; 1951 1952 for (error = 0; uio->uio_resid > 0;) { 1953 lbn = lblkno(fs, uio->uio_offset); 1954 blkoffset = blkoff(fs, uio->uio_offset); 1955 xfersize = fs->e2fs_fsize - blkoffset; 1956 if (uio->uio_resid < xfersize) 1957 xfersize = uio->uio_resid; 1958 if (uio->uio_offset + xfersize > ip->i_size) 1959 vnode_pager_setsize(vp, uio->uio_offset + xfersize); 1960 1961 /* 1962 * We must perform a read-before-write if the transfer size 1963 * does not cover the entire buffer. 1964 */ 1965 if (fs->e2fs_bsize > xfersize) 1966 flags |= BA_CLRBUF; 1967 else 1968 flags &= ~BA_CLRBUF; 1969 error = ext2_balloc(ip, lbn, blkoffset + xfersize, 1970 ap->a_cred, &bp, flags); 1971 if (error != 0) 1972 break; 1973 1974 if ((ioflag & (IO_SYNC|IO_INVAL)) == (IO_SYNC|IO_INVAL)) 1975 bp->b_flags |= B_NOCACHE; 1976 if (uio->uio_offset + xfersize > ip->i_size) 1977 ip->i_size = uio->uio_offset + xfersize; 1978 size = blksize(fs, ip, lbn) - bp->b_resid; 1979 if (size < xfersize) 1980 xfersize = size; 1981 1982 error = 1983 uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio); 1984 /* 1985 * If the buffer is not already filled and we encounter an 1986 * error while trying to fill it, we have to clear out any 1987 * garbage data from the pages instantiated for the buffer. 1988 * If we do not, a failed uiomove() during a write can leave 1989 * the prior contents of the pages exposed to a userland mmap. 1990 * 1991 * Note that we need only clear buffers with a transfer size 1992 * equal to the block size because buffers with a shorter 1993 * transfer size were cleared above by the call to ext2_balloc() 1994 * with the BA_CLRBUF flag set. 1995 * 1996 * If the source region for uiomove identically mmaps the 1997 * buffer, uiomove() performed the NOP copy, and the buffer 1998 * content remains valid because the page fault handler 1999 * validated the pages. 2000 */ 2001 if (error != 0 && (bp->b_flags & B_CACHE) == 0 && 2002 fs->e2fs_bsize == xfersize) 2003 vfs_bio_clrbuf(bp); 2004 if (ioflag & (IO_VMIO|IO_DIRECT)) { 2005 bp->b_flags |= B_RELBUF; 2006 } 2007 2008 /* 2009 * If IO_SYNC each buffer is written synchronously. Otherwise 2010 * if we have a severe page deficiency write the buffer 2011 * asynchronously. Otherwise try to cluster, and if that 2012 * doesn't do it then either do an async write (if O_DIRECT), 2013 * or a delayed write (if not). 2014 */ 2015 if (ioflag & IO_SYNC) { 2016 (void)bwrite(bp); 2017 } else if (vm_page_count_severe() || 2018 buf_dirty_count_severe() || 2019 (ioflag & IO_ASYNC)) { 2020 bp->b_flags |= B_CLUSTEROK; 2021 bawrite(bp); 2022 } else if (xfersize + blkoffset == fs->e2fs_fsize) { 2023 if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) { 2024 bp->b_flags |= B_CLUSTEROK; 2025 cluster_write(vp, bp, ip->i_size, seqcount, 0); 2026 } else { 2027 bawrite(bp); 2028 } 2029 } else if (ioflag & IO_DIRECT) { 2030 bp->b_flags |= B_CLUSTEROK; 2031 bawrite(bp); 2032 } else { 2033 bp->b_flags |= B_CLUSTEROK; 2034 bdwrite(bp); 2035 } 2036 if (error || xfersize == 0) 2037 break; 2038 } 2039 /* 2040 * If we successfully wrote any data, and we are not the superuser 2041 * we clear the setuid and setgid bits as a precaution against 2042 * tampering. 2043 */ 2044 if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid && 2045 ap->a_cred) { 2046 if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0)) 2047 ip->i_mode &= ~(ISUID | ISGID); 2048 } 2049 if (error) { 2050 if (ioflag & IO_UNIT) { 2051 (void)ext2_truncate(vp, osize, 2052 ioflag & IO_SYNC, ap->a_cred, uio->uio_td); 2053 uio->uio_offset -= resid - uio->uio_resid; 2054 uio->uio_resid = resid; 2055 } 2056 } 2057 if (uio->uio_resid != resid) { 2058 ip->i_flag |= IN_CHANGE | IN_UPDATE; 2059 if (ioflag & IO_SYNC) 2060 error = ext2_update(vp, 1); 2061 } 2062 return (error); 2063 } 2064