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