1 /*- 2 * SPDX-License-Identifier: (BSD-2-Clause-FreeBSD AND BSD-3-Clause) 3 * 4 * Copyright (c) 2002, 2003 Networks Associates Technology, Inc. 5 * All rights reserved. 6 * 7 * This software was developed for the FreeBSD Project by Marshall 8 * Kirk McKusick and Network Associates Laboratories, the Security 9 * Research Division of Network Associates, Inc. under DARPA/SPAWAR 10 * contract N66001-01-C-8035 ("CBOSS"), as part of the DARPA CHATS 11 * research program 12 * 13 * Redistribution and use in source and binary forms, with or without 14 * modification, are permitted provided that the following conditions 15 * are met: 16 * 1. Redistributions of source code must retain the above copyright 17 * notice, this list of conditions and the following disclaimer. 18 * 2. Redistributions in binary form must reproduce the above copyright 19 * notice, this list of conditions and the following disclaimer in the 20 * documentation and/or other materials provided with the distribution. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * Copyright (c) 1982, 1986, 1989, 1993 35 * The Regents of the University of California. All rights reserved. 36 * 37 * Redistribution and use in source and binary forms, with or without 38 * modification, are permitted provided that the following conditions 39 * are met: 40 * 1. Redistributions of source code must retain the above copyright 41 * notice, this list of conditions and the following disclaimer. 42 * 2. Redistributions in binary form must reproduce the above copyright 43 * notice, this list of conditions and the following disclaimer in the 44 * documentation and/or other materials provided with the distribution. 45 * 3. Neither the name of the University nor the names of its contributors 46 * may be used to endorse or promote products derived from this software 47 * without specific prior written permission. 48 * 49 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 50 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 51 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 52 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 53 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 54 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 55 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 56 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 57 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 58 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 59 * SUCH DAMAGE. 60 * 61 * from: @(#)ufs_readwrite.c 8.11 (Berkeley) 5/8/95 62 * from: $FreeBSD: .../ufs/ufs_readwrite.c,v 1.96 2002/08/12 09:22:11 phk ... 63 * @(#)ffs_vnops.c 8.15 (Berkeley) 5/14/95 64 */ 65 66 #include <sys/cdefs.h> 67 __FBSDID("$FreeBSD$"); 68 69 #include <sys/param.h> 70 #include <sys/bio.h> 71 #include <sys/systm.h> 72 #include <sys/buf.h> 73 #include <sys/conf.h> 74 #include <sys/extattr.h> 75 #include <sys/kernel.h> 76 #include <sys/limits.h> 77 #include <sys/malloc.h> 78 #include <sys/mount.h> 79 #include <sys/priv.h> 80 #include <sys/rwlock.h> 81 #include <sys/stat.h> 82 #include <sys/sysctl.h> 83 #include <sys/vmmeter.h> 84 #include <sys/vnode.h> 85 86 #include <vm/vm.h> 87 #include <vm/vm_param.h> 88 #include <vm/vm_extern.h> 89 #include <vm/vm_object.h> 90 #include <vm/vm_page.h> 91 #include <vm/vm_pager.h> 92 #include <vm/vnode_pager.h> 93 94 #include <ufs/ufs/extattr.h> 95 #include <ufs/ufs/quota.h> 96 #include <ufs/ufs/inode.h> 97 #include <ufs/ufs/ufs_extern.h> 98 #include <ufs/ufs/ufsmount.h> 99 100 #include <ufs/ffs/fs.h> 101 #include <ufs/ffs/ffs_extern.h> 102 #include "opt_directio.h" 103 #include "opt_ffs.h" 104 105 #define ALIGNED_TO(ptr, s) \ 106 (((uintptr_t)(ptr) & (_Alignof(s) - 1)) == 0) 107 108 #ifdef DIRECTIO 109 extern int ffs_rawread(struct vnode *vp, struct uio *uio, int *workdone); 110 #endif 111 static vop_fdatasync_t ffs_fdatasync; 112 static vop_fsync_t ffs_fsync; 113 static vop_getpages_t ffs_getpages; 114 static vop_getpages_async_t ffs_getpages_async; 115 static vop_lock1_t ffs_lock; 116 #ifdef INVARIANTS 117 static vop_unlock_t ffs_unlock_debug; 118 #endif 119 static vop_read_t ffs_read; 120 static vop_write_t ffs_write; 121 static int ffs_extread(struct vnode *vp, struct uio *uio, int ioflag); 122 static int ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag, 123 struct ucred *cred); 124 static vop_strategy_t ffsext_strategy; 125 static vop_closeextattr_t ffs_closeextattr; 126 static vop_deleteextattr_t ffs_deleteextattr; 127 static vop_getextattr_t ffs_getextattr; 128 static vop_listextattr_t ffs_listextattr; 129 static vop_openextattr_t ffs_openextattr; 130 static vop_setextattr_t ffs_setextattr; 131 static vop_vptofh_t ffs_vptofh; 132 133 /* Global vfs data structures for ufs. */ 134 struct vop_vector ffs_vnodeops1 = { 135 .vop_default = &ufs_vnodeops, 136 .vop_fsync = ffs_fsync, 137 .vop_fdatasync = ffs_fdatasync, 138 .vop_getpages = ffs_getpages, 139 .vop_getpages_async = ffs_getpages_async, 140 .vop_lock1 = ffs_lock, 141 #ifdef INVARIANTS 142 .vop_unlock = ffs_unlock_debug, 143 #endif 144 .vop_read = ffs_read, 145 .vop_reallocblks = ffs_reallocblks, 146 .vop_write = ffs_write, 147 .vop_vptofh = ffs_vptofh, 148 }; 149 VFS_VOP_VECTOR_REGISTER(ffs_vnodeops1); 150 151 struct vop_vector ffs_fifoops1 = { 152 .vop_default = &ufs_fifoops, 153 .vop_fsync = ffs_fsync, 154 .vop_fdatasync = ffs_fdatasync, 155 .vop_lock1 = ffs_lock, 156 #ifdef INVARIANTS 157 .vop_unlock = ffs_unlock_debug, 158 #endif 159 .vop_vptofh = ffs_vptofh, 160 }; 161 VFS_VOP_VECTOR_REGISTER(ffs_fifoops1); 162 163 /* Global vfs data structures for ufs. */ 164 struct vop_vector ffs_vnodeops2 = { 165 .vop_default = &ufs_vnodeops, 166 .vop_fsync = ffs_fsync, 167 .vop_fdatasync = ffs_fdatasync, 168 .vop_getpages = ffs_getpages, 169 .vop_getpages_async = ffs_getpages_async, 170 .vop_lock1 = ffs_lock, 171 #ifdef INVARIANTS 172 .vop_unlock = ffs_unlock_debug, 173 #endif 174 .vop_read = ffs_read, 175 .vop_reallocblks = ffs_reallocblks, 176 .vop_write = ffs_write, 177 .vop_closeextattr = ffs_closeextattr, 178 .vop_deleteextattr = ffs_deleteextattr, 179 .vop_getextattr = ffs_getextattr, 180 .vop_listextattr = ffs_listextattr, 181 .vop_openextattr = ffs_openextattr, 182 .vop_setextattr = ffs_setextattr, 183 .vop_vptofh = ffs_vptofh, 184 }; 185 VFS_VOP_VECTOR_REGISTER(ffs_vnodeops2); 186 187 struct vop_vector ffs_fifoops2 = { 188 .vop_default = &ufs_fifoops, 189 .vop_fsync = ffs_fsync, 190 .vop_fdatasync = ffs_fdatasync, 191 .vop_lock1 = ffs_lock, 192 #ifdef INVARIANTS 193 .vop_unlock = ffs_unlock_debug, 194 #endif 195 .vop_reallocblks = ffs_reallocblks, 196 .vop_strategy = ffsext_strategy, 197 .vop_closeextattr = ffs_closeextattr, 198 .vop_deleteextattr = ffs_deleteextattr, 199 .vop_getextattr = ffs_getextattr, 200 .vop_listextattr = ffs_listextattr, 201 .vop_openextattr = ffs_openextattr, 202 .vop_setextattr = ffs_setextattr, 203 .vop_vptofh = ffs_vptofh, 204 }; 205 VFS_VOP_VECTOR_REGISTER(ffs_fifoops2); 206 207 /* 208 * Synch an open file. 209 */ 210 /* ARGSUSED */ 211 static int 212 ffs_fsync(struct vop_fsync_args *ap) 213 { 214 struct vnode *vp; 215 struct bufobj *bo; 216 int error; 217 218 vp = ap->a_vp; 219 bo = &vp->v_bufobj; 220 retry: 221 error = ffs_syncvnode(vp, ap->a_waitfor, 0); 222 if (error) 223 return (error); 224 if (ap->a_waitfor == MNT_WAIT && DOINGSOFTDEP(vp)) { 225 error = softdep_fsync(vp); 226 if (error) 227 return (error); 228 229 /* 230 * The softdep_fsync() function may drop vp lock, 231 * allowing for dirty buffers to reappear on the 232 * bo_dirty list. Recheck and resync as needed. 233 */ 234 BO_LOCK(bo); 235 if ((vp->v_type == VREG || vp->v_type == VDIR) && 236 (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)) { 237 BO_UNLOCK(bo); 238 goto retry; 239 } 240 BO_UNLOCK(bo); 241 } 242 if (ffs_fsfail_cleanup(VFSTOUFS(vp->v_mount), 0)) 243 return (ENXIO); 244 return (0); 245 } 246 247 int 248 ffs_syncvnode(struct vnode *vp, int waitfor, int flags) 249 { 250 struct inode *ip; 251 struct bufobj *bo; 252 struct ufsmount *ump; 253 struct buf *bp, *nbp; 254 ufs_lbn_t lbn; 255 int error, passes; 256 bool still_dirty, wait; 257 258 ip = VTOI(vp); 259 ip->i_flag &= ~IN_NEEDSYNC; 260 bo = &vp->v_bufobj; 261 ump = VFSTOUFS(vp->v_mount); 262 263 /* 264 * When doing MNT_WAIT we must first flush all dependencies 265 * on the inode. 266 */ 267 if (DOINGSOFTDEP(vp) && waitfor == MNT_WAIT && 268 (error = softdep_sync_metadata(vp)) != 0) { 269 if (ffs_fsfail_cleanup(ump, error)) 270 error = 0; 271 return (error); 272 } 273 274 /* 275 * Flush all dirty buffers associated with a vnode. 276 */ 277 error = 0; 278 passes = 0; 279 wait = false; /* Always do an async pass first. */ 280 lbn = lblkno(ITOFS(ip), (ip->i_size + ITOFS(ip)->fs_bsize - 1)); 281 BO_LOCK(bo); 282 loop: 283 TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs) 284 bp->b_vflags &= ~BV_SCANNED; 285 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 286 /* 287 * Reasons to skip this buffer: it has already been considered 288 * on this pass, the buffer has dependencies that will cause 289 * it to be redirtied and it has not already been deferred, 290 * or it is already being written. 291 */ 292 if ((bp->b_vflags & BV_SCANNED) != 0) 293 continue; 294 bp->b_vflags |= BV_SCANNED; 295 /* 296 * Flush indirects in order, if requested. 297 * 298 * Note that if only datasync is requested, we can 299 * skip indirect blocks when softupdates are not 300 * active. Otherwise we must flush them with data, 301 * since dependencies prevent data block writes. 302 */ 303 if (waitfor == MNT_WAIT && bp->b_lblkno <= -UFS_NDADDR && 304 (lbn_level(bp->b_lblkno) >= passes || 305 ((flags & DATA_ONLY) != 0 && !DOINGSOFTDEP(vp)))) 306 continue; 307 if (bp->b_lblkno > lbn) 308 panic("ffs_syncvnode: syncing truncated data."); 309 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL) == 0) { 310 BO_UNLOCK(bo); 311 } else if (wait) { 312 if (BUF_LOCK(bp, 313 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, 314 BO_LOCKPTR(bo)) != 0) { 315 bp->b_vflags &= ~BV_SCANNED; 316 goto next; 317 } 318 } else 319 continue; 320 if ((bp->b_flags & B_DELWRI) == 0) 321 panic("ffs_fsync: not dirty"); 322 /* 323 * Check for dependencies and potentially complete them. 324 */ 325 if (!LIST_EMPTY(&bp->b_dep) && 326 (error = softdep_sync_buf(vp, bp, 327 wait ? MNT_WAIT : MNT_NOWAIT)) != 0) { 328 /* I/O error. */ 329 if (error != EBUSY) { 330 BUF_UNLOCK(bp); 331 return (error); 332 } 333 /* If we deferred once, don't defer again. */ 334 if ((bp->b_flags & B_DEFERRED) == 0) { 335 bp->b_flags |= B_DEFERRED; 336 BUF_UNLOCK(bp); 337 goto next; 338 } 339 } 340 if (wait) { 341 bremfree(bp); 342 error = bwrite(bp); 343 if (ffs_fsfail_cleanup(ump, error)) 344 error = 0; 345 if (error != 0) 346 return (error); 347 } else if ((bp->b_flags & B_CLUSTEROK)) { 348 (void) vfs_bio_awrite(bp); 349 } else { 350 bremfree(bp); 351 (void) bawrite(bp); 352 } 353 next: 354 /* 355 * Since we may have slept during the I/O, we need 356 * to start from a known point. 357 */ 358 BO_LOCK(bo); 359 nbp = TAILQ_FIRST(&bo->bo_dirty.bv_hd); 360 } 361 if (waitfor != MNT_WAIT) { 362 BO_UNLOCK(bo); 363 if ((flags & NO_INO_UPDT) != 0) 364 return (0); 365 else 366 return (ffs_update(vp, 0)); 367 } 368 /* Drain IO to see if we're done. */ 369 bufobj_wwait(bo, 0, 0); 370 /* 371 * Block devices associated with filesystems may have new I/O 372 * requests posted for them even if the vnode is locked, so no 373 * amount of trying will get them clean. We make several passes 374 * as a best effort. 375 * 376 * Regular files may need multiple passes to flush all dependency 377 * work as it is possible that we must write once per indirect 378 * level, once for the leaf, and once for the inode and each of 379 * these will be done with one sync and one async pass. 380 */ 381 if (bo->bo_dirty.bv_cnt > 0) { 382 if ((flags & DATA_ONLY) == 0) { 383 still_dirty = true; 384 } else { 385 /* 386 * For data-only sync, dirty indirect buffers 387 * are ignored. 388 */ 389 still_dirty = false; 390 TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs) { 391 if (bp->b_lblkno > -UFS_NDADDR) { 392 still_dirty = true; 393 break; 394 } 395 } 396 } 397 398 if (still_dirty) { 399 /* Write the inode after sync passes to flush deps. */ 400 if (wait && DOINGSOFTDEP(vp) && 401 (flags & NO_INO_UPDT) == 0) { 402 BO_UNLOCK(bo); 403 ffs_update(vp, 1); 404 BO_LOCK(bo); 405 } 406 /* switch between sync/async. */ 407 wait = !wait; 408 if (wait || ++passes < UFS_NIADDR + 2) 409 goto loop; 410 } 411 } 412 BO_UNLOCK(bo); 413 error = 0; 414 if ((flags & DATA_ONLY) == 0) { 415 if ((flags & NO_INO_UPDT) == 0) 416 error = ffs_update(vp, 1); 417 if (DOINGSUJ(vp)) 418 softdep_journal_fsync(VTOI(vp)); 419 } else if ((ip->i_flags & (IN_SIZEMOD | IN_IBLKDATA)) != 0) { 420 error = ffs_update(vp, 1); 421 } 422 return (error); 423 } 424 425 static int 426 ffs_fdatasync(struct vop_fdatasync_args *ap) 427 { 428 429 return (ffs_syncvnode(ap->a_vp, MNT_WAIT, DATA_ONLY)); 430 } 431 432 static int 433 ffs_lock(ap) 434 struct vop_lock1_args /* { 435 struct vnode *a_vp; 436 int a_flags; 437 struct thread *a_td; 438 char *file; 439 int line; 440 } */ *ap; 441 { 442 #ifndef NO_FFS_SNAPSHOT 443 struct vnode *vp; 444 int flags; 445 struct lock *lkp; 446 int result; 447 448 /* 449 * Adaptive spinning mixed with SU leads to trouble. use a giant hammer 450 * and only use it when LK_NODDLKTREAT is set. Currently this means it 451 * is only used during path lookup. 452 */ 453 if ((ap->a_flags & LK_NODDLKTREAT) != 0) 454 ap->a_flags |= LK_ADAPTIVE; 455 switch (ap->a_flags & LK_TYPE_MASK) { 456 case LK_SHARED: 457 case LK_UPGRADE: 458 case LK_EXCLUSIVE: 459 vp = ap->a_vp; 460 flags = ap->a_flags; 461 for (;;) { 462 #ifdef DEBUG_VFS_LOCKS 463 VNPASS(vp->v_holdcnt != 0, vp); 464 #endif 465 lkp = vp->v_vnlock; 466 result = lockmgr_lock_flags(lkp, flags, 467 &VI_MTX(vp)->lock_object, ap->a_file, ap->a_line); 468 if (lkp == vp->v_vnlock || result != 0) 469 break; 470 /* 471 * Apparent success, except that the vnode 472 * mutated between snapshot file vnode and 473 * regular file vnode while this process 474 * slept. The lock currently held is not the 475 * right lock. Release it, and try to get the 476 * new lock. 477 */ 478 lockmgr_unlock(lkp); 479 if ((flags & (LK_INTERLOCK | LK_NOWAIT)) == 480 (LK_INTERLOCK | LK_NOWAIT)) 481 return (EBUSY); 482 if ((flags & LK_TYPE_MASK) == LK_UPGRADE) 483 flags = (flags & ~LK_TYPE_MASK) | LK_EXCLUSIVE; 484 flags &= ~LK_INTERLOCK; 485 } 486 break; 487 default: 488 result = VOP_LOCK1_APV(&ufs_vnodeops, ap); 489 } 490 return (result); 491 #else 492 /* 493 * See above for an explanation. 494 */ 495 if ((ap->a_flags & LK_NODDLKTREAT) != 0) 496 ap->a_flags |= LK_ADAPTIVE; 497 return (VOP_LOCK1_APV(&ufs_vnodeops, ap)); 498 #endif 499 } 500 501 #ifdef INVARIANTS 502 static int 503 ffs_unlock_debug(struct vop_unlock_args *ap) 504 { 505 struct vnode *vp = ap->a_vp; 506 struct inode *ip = VTOI(vp); 507 508 if (ip->i_flag & UFS_INODE_FLAG_LAZY_MASK_ASSERTABLE) { 509 if ((vp->v_mflag & VMP_LAZYLIST) == 0) { 510 VI_LOCK(vp); 511 VNASSERT((vp->v_mflag & VMP_LAZYLIST), vp, 512 ("%s: modified vnode (%x) not on lazy list", 513 __func__, ip->i_flag)); 514 VI_UNLOCK(vp); 515 } 516 } 517 return (VOP_UNLOCK_APV(&ufs_vnodeops, ap)); 518 } 519 #endif 520 521 static int 522 ffs_read_hole(struct uio *uio, long xfersize, long *size) 523 { 524 ssize_t saved_resid, tlen; 525 int error; 526 527 while (xfersize > 0) { 528 tlen = min(xfersize, ZERO_REGION_SIZE); 529 saved_resid = uio->uio_resid; 530 error = vn_io_fault_uiomove(__DECONST(void *, zero_region), 531 tlen, uio); 532 if (error != 0) 533 return (error); 534 tlen = saved_resid - uio->uio_resid; 535 xfersize -= tlen; 536 *size -= tlen; 537 } 538 return (0); 539 } 540 541 /* 542 * Vnode op for reading. 543 */ 544 static int 545 ffs_read(ap) 546 struct vop_read_args /* { 547 struct vnode *a_vp; 548 struct uio *a_uio; 549 int a_ioflag; 550 struct ucred *a_cred; 551 } */ *ap; 552 { 553 struct vnode *vp; 554 struct inode *ip; 555 struct uio *uio; 556 struct fs *fs; 557 struct buf *bp; 558 ufs_lbn_t lbn, nextlbn; 559 off_t bytesinfile; 560 long size, xfersize, blkoffset; 561 ssize_t orig_resid; 562 int bflag, error, ioflag, seqcount; 563 564 vp = ap->a_vp; 565 uio = ap->a_uio; 566 ioflag = ap->a_ioflag; 567 if (ap->a_ioflag & IO_EXT) 568 #ifdef notyet 569 return (ffs_extread(vp, uio, ioflag)); 570 #else 571 panic("ffs_read+IO_EXT"); 572 #endif 573 #ifdef DIRECTIO 574 if ((ioflag & IO_DIRECT) != 0) { 575 int workdone; 576 577 error = ffs_rawread(vp, uio, &workdone); 578 if (error != 0 || workdone != 0) 579 return error; 580 } 581 #endif 582 583 seqcount = ap->a_ioflag >> IO_SEQSHIFT; 584 ip = VTOI(vp); 585 586 #ifdef INVARIANTS 587 if (uio->uio_rw != UIO_READ) 588 panic("ffs_read: mode"); 589 590 if (vp->v_type == VLNK) { 591 if ((int)ip->i_size < vp->v_mount->mnt_maxsymlinklen) 592 panic("ffs_read: short symlink"); 593 } else if (vp->v_type != VREG && vp->v_type != VDIR) 594 panic("ffs_read: type %d", vp->v_type); 595 #endif 596 orig_resid = uio->uio_resid; 597 KASSERT(orig_resid >= 0, ("ffs_read: uio->uio_resid < 0")); 598 if (orig_resid == 0) 599 return (0); 600 KASSERT(uio->uio_offset >= 0, ("ffs_read: uio->uio_offset < 0")); 601 fs = ITOFS(ip); 602 if (uio->uio_offset < ip->i_size && 603 uio->uio_offset >= fs->fs_maxfilesize) 604 return (EOVERFLOW); 605 606 bflag = GB_UNMAPPED | (uio->uio_segflg == UIO_NOCOPY ? 0 : GB_NOSPARSE); 607 for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) { 608 if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0) 609 break; 610 lbn = lblkno(fs, uio->uio_offset); 611 nextlbn = lbn + 1; 612 613 /* 614 * size of buffer. The buffer representing the 615 * end of the file is rounded up to the size of 616 * the block type ( fragment or full block, 617 * depending ). 618 */ 619 size = blksize(fs, ip, lbn); 620 blkoffset = blkoff(fs, uio->uio_offset); 621 622 /* 623 * The amount we want to transfer in this iteration is 624 * one FS block less the amount of the data before 625 * our startpoint (duh!) 626 */ 627 xfersize = fs->fs_bsize - blkoffset; 628 629 /* 630 * But if we actually want less than the block, 631 * or the file doesn't have a whole block more of data, 632 * then use the lesser number. 633 */ 634 if (uio->uio_resid < xfersize) 635 xfersize = uio->uio_resid; 636 if (bytesinfile < xfersize) 637 xfersize = bytesinfile; 638 639 if (lblktosize(fs, nextlbn) >= ip->i_size) { 640 /* 641 * Don't do readahead if this is the end of the file. 642 */ 643 error = bread_gb(vp, lbn, size, NOCRED, bflag, &bp); 644 } else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) { 645 /* 646 * Otherwise if we are allowed to cluster, 647 * grab as much as we can. 648 * 649 * XXX This may not be a win if we are not 650 * doing sequential access. 651 */ 652 error = cluster_read(vp, ip->i_size, lbn, 653 size, NOCRED, blkoffset + uio->uio_resid, 654 seqcount, bflag, &bp); 655 } else if (seqcount > 1) { 656 /* 657 * If we are NOT allowed to cluster, then 658 * if we appear to be acting sequentially, 659 * fire off a request for a readahead 660 * as well as a read. Note that the 4th and 5th 661 * arguments point to arrays of the size specified in 662 * the 6th argument. 663 */ 664 u_int nextsize = blksize(fs, ip, nextlbn); 665 error = breadn_flags(vp, lbn, lbn, size, &nextlbn, 666 &nextsize, 1, NOCRED, bflag, NULL, &bp); 667 } else { 668 /* 669 * Failing all of the above, just read what the 670 * user asked for. Interestingly, the same as 671 * the first option above. 672 */ 673 error = bread_gb(vp, lbn, size, NOCRED, bflag, &bp); 674 } 675 if (error == EJUSTRETURN) { 676 error = ffs_read_hole(uio, xfersize, &size); 677 if (error == 0) 678 continue; 679 } 680 if (error != 0) { 681 brelse(bp); 682 bp = NULL; 683 break; 684 } 685 686 /* 687 * We should only get non-zero b_resid when an I/O error 688 * has occurred, which should cause us to break above. 689 * However, if the short read did not cause an error, 690 * then we want to ensure that we do not uiomove bad 691 * or uninitialized data. 692 */ 693 size -= bp->b_resid; 694 if (size < xfersize) { 695 if (size == 0) 696 break; 697 xfersize = size; 698 } 699 700 if (buf_mapped(bp)) { 701 error = vn_io_fault_uiomove((char *)bp->b_data + 702 blkoffset, (int)xfersize, uio); 703 } else { 704 error = vn_io_fault_pgmove(bp->b_pages, blkoffset, 705 (int)xfersize, uio); 706 } 707 if (error) 708 break; 709 710 vfs_bio_brelse(bp, ioflag); 711 } 712 713 /* 714 * This can only happen in the case of an error 715 * because the loop above resets bp to NULL on each iteration 716 * and on normal completion has not set a new value into it. 717 * so it must have come from a 'break' statement 718 */ 719 if (bp != NULL) 720 vfs_bio_brelse(bp, ioflag); 721 722 if ((error == 0 || uio->uio_resid != orig_resid) && 723 (vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) 724 UFS_INODE_SET_FLAG_SHARED(ip, IN_ACCESS); 725 return (error); 726 } 727 728 /* 729 * Vnode op for writing. 730 */ 731 static int 732 ffs_write(ap) 733 struct vop_write_args /* { 734 struct vnode *a_vp; 735 struct uio *a_uio; 736 int a_ioflag; 737 struct ucred *a_cred; 738 } */ *ap; 739 { 740 struct vnode *vp; 741 struct uio *uio; 742 struct inode *ip; 743 struct fs *fs; 744 struct buf *bp; 745 ufs_lbn_t lbn; 746 off_t osize; 747 ssize_t resid; 748 int seqcount; 749 int blkoffset, error, flags, ioflag, size, xfersize; 750 751 vp = ap->a_vp; 752 uio = ap->a_uio; 753 ioflag = ap->a_ioflag; 754 if (ap->a_ioflag & IO_EXT) 755 #ifdef notyet 756 return (ffs_extwrite(vp, uio, ioflag, ap->a_cred)); 757 #else 758 panic("ffs_write+IO_EXT"); 759 #endif 760 761 seqcount = ap->a_ioflag >> IO_SEQSHIFT; 762 ip = VTOI(vp); 763 764 #ifdef INVARIANTS 765 if (uio->uio_rw != UIO_WRITE) 766 panic("ffs_write: mode"); 767 #endif 768 769 switch (vp->v_type) { 770 case VREG: 771 if (ioflag & IO_APPEND) 772 uio->uio_offset = ip->i_size; 773 if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size) 774 return (EPERM); 775 /* FALLTHROUGH */ 776 case VLNK: 777 break; 778 case VDIR: 779 panic("ffs_write: dir write"); 780 break; 781 default: 782 panic("ffs_write: type %p %d (%d,%d)", vp, (int)vp->v_type, 783 (int)uio->uio_offset, 784 (int)uio->uio_resid 785 ); 786 } 787 788 KASSERT(uio->uio_resid >= 0, ("ffs_write: uio->uio_resid < 0")); 789 KASSERT(uio->uio_offset >= 0, ("ffs_write: uio->uio_offset < 0")); 790 fs = ITOFS(ip); 791 if ((uoff_t)uio->uio_offset + uio->uio_resid > fs->fs_maxfilesize) 792 return (EFBIG); 793 /* 794 * Maybe this should be above the vnode op call, but so long as 795 * file servers have no limits, I don't think it matters. 796 */ 797 if (vn_rlimit_fsize(vp, uio, uio->uio_td)) 798 return (EFBIG); 799 800 resid = uio->uio_resid; 801 osize = ip->i_size; 802 if (seqcount > BA_SEQMAX) 803 flags = BA_SEQMAX << BA_SEQSHIFT; 804 else 805 flags = seqcount << BA_SEQSHIFT; 806 if (ioflag & IO_SYNC) 807 flags |= IO_SYNC; 808 flags |= BA_UNMAPPED; 809 810 for (error = 0; uio->uio_resid > 0;) { 811 lbn = lblkno(fs, uio->uio_offset); 812 blkoffset = blkoff(fs, uio->uio_offset); 813 xfersize = fs->fs_bsize - blkoffset; 814 if (uio->uio_resid < xfersize) 815 xfersize = uio->uio_resid; 816 if (uio->uio_offset + xfersize > ip->i_size) 817 vnode_pager_setsize(vp, uio->uio_offset + xfersize); 818 819 /* 820 * We must perform a read-before-write if the transfer size 821 * does not cover the entire buffer. 822 */ 823 if (fs->fs_bsize > xfersize) 824 flags |= BA_CLRBUF; 825 else 826 flags &= ~BA_CLRBUF; 827 /* XXX is uio->uio_offset the right thing here? */ 828 error = UFS_BALLOC(vp, uio->uio_offset, xfersize, 829 ap->a_cred, flags, &bp); 830 if (error != 0) { 831 vnode_pager_setsize(vp, ip->i_size); 832 break; 833 } 834 if ((ioflag & (IO_SYNC|IO_INVAL)) == (IO_SYNC|IO_INVAL)) 835 bp->b_flags |= B_NOCACHE; 836 837 if (uio->uio_offset + xfersize > ip->i_size) { 838 ip->i_size = uio->uio_offset + xfersize; 839 DIP_SET(ip, i_size, ip->i_size); 840 UFS_INODE_SET_FLAG(ip, IN_SIZEMOD | IN_CHANGE); 841 } 842 843 size = blksize(fs, ip, lbn) - bp->b_resid; 844 if (size < xfersize) 845 xfersize = size; 846 847 if (buf_mapped(bp)) { 848 error = vn_io_fault_uiomove((char *)bp->b_data + 849 blkoffset, (int)xfersize, uio); 850 } else { 851 error = vn_io_fault_pgmove(bp->b_pages, blkoffset, 852 (int)xfersize, uio); 853 } 854 /* 855 * If the buffer is not already filled and we encounter an 856 * error while trying to fill it, we have to clear out any 857 * garbage data from the pages instantiated for the buffer. 858 * If we do not, a failed uiomove() during a write can leave 859 * the prior contents of the pages exposed to a userland mmap. 860 * 861 * Note that we need only clear buffers with a transfer size 862 * equal to the block size because buffers with a shorter 863 * transfer size were cleared above by the call to UFS_BALLOC() 864 * with the BA_CLRBUF flag set. 865 * 866 * If the source region for uiomove identically mmaps the 867 * buffer, uiomove() performed the NOP copy, and the buffer 868 * content remains valid because the page fault handler 869 * validated the pages. 870 */ 871 if (error != 0 && (bp->b_flags & B_CACHE) == 0 && 872 fs->fs_bsize == xfersize) 873 vfs_bio_clrbuf(bp); 874 875 vfs_bio_set_flags(bp, ioflag); 876 877 /* 878 * If IO_SYNC each buffer is written synchronously. Otherwise 879 * if we have a severe page deficiency write the buffer 880 * asynchronously. Otherwise try to cluster, and if that 881 * doesn't do it then either do an async write (if O_DIRECT), 882 * or a delayed write (if not). 883 */ 884 if (ioflag & IO_SYNC) { 885 (void)bwrite(bp); 886 } else if (vm_page_count_severe() || 887 buf_dirty_count_severe() || 888 (ioflag & IO_ASYNC)) { 889 bp->b_flags |= B_CLUSTEROK; 890 bawrite(bp); 891 } else if (xfersize + blkoffset == fs->fs_bsize) { 892 if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) { 893 bp->b_flags |= B_CLUSTEROK; 894 cluster_write(vp, bp, ip->i_size, seqcount, 895 GB_UNMAPPED); 896 } else { 897 bawrite(bp); 898 } 899 } else if (ioflag & IO_DIRECT) { 900 bp->b_flags |= B_CLUSTEROK; 901 bawrite(bp); 902 } else { 903 bp->b_flags |= B_CLUSTEROK; 904 bdwrite(bp); 905 } 906 if (error || xfersize == 0) 907 break; 908 UFS_INODE_SET_FLAG(ip, IN_CHANGE | IN_UPDATE); 909 } 910 /* 911 * If we successfully wrote any data, and we are not the superuser 912 * we clear the setuid and setgid bits as a precaution against 913 * tampering. 914 */ 915 if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid && 916 ap->a_cred) { 917 if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID)) { 918 vn_seqc_write_begin(vp); 919 UFS_INODE_SET_MODE(ip, ip->i_mode & ~(ISUID | ISGID)); 920 DIP_SET(ip, i_mode, ip->i_mode); 921 vn_seqc_write_end(vp); 922 } 923 } 924 if (error) { 925 if (ioflag & IO_UNIT) { 926 (void)ffs_truncate(vp, osize, 927 IO_NORMAL | (ioflag & IO_SYNC), ap->a_cred); 928 uio->uio_offset -= resid - uio->uio_resid; 929 uio->uio_resid = resid; 930 } 931 } else if (resid > uio->uio_resid && (ioflag & IO_SYNC)) { 932 error = ffs_update(vp, 1); 933 if (ffs_fsfail_cleanup(VFSTOUFS(vp->v_mount), error)) 934 error = ENXIO; 935 } 936 return (error); 937 } 938 939 /* 940 * Extended attribute area reading. 941 */ 942 static int 943 ffs_extread(struct vnode *vp, struct uio *uio, int ioflag) 944 { 945 struct inode *ip; 946 struct ufs2_dinode *dp; 947 struct fs *fs; 948 struct buf *bp; 949 ufs_lbn_t lbn, nextlbn; 950 off_t bytesinfile; 951 long size, xfersize, blkoffset; 952 ssize_t orig_resid; 953 int error; 954 955 ip = VTOI(vp); 956 fs = ITOFS(ip); 957 dp = ip->i_din2; 958 959 #ifdef INVARIANTS 960 if (uio->uio_rw != UIO_READ || fs->fs_magic != FS_UFS2_MAGIC) 961 panic("ffs_extread: mode"); 962 963 #endif 964 orig_resid = uio->uio_resid; 965 KASSERT(orig_resid >= 0, ("ffs_extread: uio->uio_resid < 0")); 966 if (orig_resid == 0) 967 return (0); 968 KASSERT(uio->uio_offset >= 0, ("ffs_extread: uio->uio_offset < 0")); 969 970 for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) { 971 if ((bytesinfile = dp->di_extsize - uio->uio_offset) <= 0) 972 break; 973 lbn = lblkno(fs, uio->uio_offset); 974 nextlbn = lbn + 1; 975 976 /* 977 * size of buffer. The buffer representing the 978 * end of the file is rounded up to the size of 979 * the block type ( fragment or full block, 980 * depending ). 981 */ 982 size = sblksize(fs, dp->di_extsize, lbn); 983 blkoffset = blkoff(fs, uio->uio_offset); 984 985 /* 986 * The amount we want to transfer in this iteration is 987 * one FS block less the amount of the data before 988 * our startpoint (duh!) 989 */ 990 xfersize = fs->fs_bsize - blkoffset; 991 992 /* 993 * But if we actually want less than the block, 994 * or the file doesn't have a whole block more of data, 995 * then use the lesser number. 996 */ 997 if (uio->uio_resid < xfersize) 998 xfersize = uio->uio_resid; 999 if (bytesinfile < xfersize) 1000 xfersize = bytesinfile; 1001 1002 if (lblktosize(fs, nextlbn) >= dp->di_extsize) { 1003 /* 1004 * Don't do readahead if this is the end of the info. 1005 */ 1006 error = bread(vp, -1 - lbn, size, NOCRED, &bp); 1007 } else { 1008 /* 1009 * If we have a second block, then 1010 * fire off a request for a readahead 1011 * as well as a read. Note that the 4th and 5th 1012 * arguments point to arrays of the size specified in 1013 * the 6th argument. 1014 */ 1015 u_int nextsize = sblksize(fs, dp->di_extsize, nextlbn); 1016 1017 nextlbn = -1 - nextlbn; 1018 error = breadn(vp, -1 - lbn, 1019 size, &nextlbn, &nextsize, 1, NOCRED, &bp); 1020 } 1021 if (error) { 1022 brelse(bp); 1023 bp = NULL; 1024 break; 1025 } 1026 1027 /* 1028 * We should only get non-zero b_resid when an I/O error 1029 * has occurred, which should cause us to break above. 1030 * However, if the short read did not cause an error, 1031 * then we want to ensure that we do not uiomove bad 1032 * or uninitialized data. 1033 */ 1034 size -= bp->b_resid; 1035 if (size < xfersize) { 1036 if (size == 0) 1037 break; 1038 xfersize = size; 1039 } 1040 1041 error = uiomove((char *)bp->b_data + blkoffset, 1042 (int)xfersize, uio); 1043 if (error) 1044 break; 1045 vfs_bio_brelse(bp, ioflag); 1046 } 1047 1048 /* 1049 * This can only happen in the case of an error 1050 * because the loop above resets bp to NULL on each iteration 1051 * and on normal completion has not set a new value into it. 1052 * so it must have come from a 'break' statement 1053 */ 1054 if (bp != NULL) 1055 vfs_bio_brelse(bp, ioflag); 1056 return (error); 1057 } 1058 1059 /* 1060 * Extended attribute area writing. 1061 */ 1062 static int 1063 ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag, struct ucred *ucred) 1064 { 1065 struct inode *ip; 1066 struct ufs2_dinode *dp; 1067 struct fs *fs; 1068 struct buf *bp; 1069 ufs_lbn_t lbn; 1070 off_t osize; 1071 ssize_t resid; 1072 int blkoffset, error, flags, size, xfersize; 1073 1074 ip = VTOI(vp); 1075 fs = ITOFS(ip); 1076 dp = ip->i_din2; 1077 1078 #ifdef INVARIANTS 1079 if (uio->uio_rw != UIO_WRITE || fs->fs_magic != FS_UFS2_MAGIC) 1080 panic("ffs_extwrite: mode"); 1081 #endif 1082 1083 if (ioflag & IO_APPEND) 1084 uio->uio_offset = dp->di_extsize; 1085 KASSERT(uio->uio_offset >= 0, ("ffs_extwrite: uio->uio_offset < 0")); 1086 KASSERT(uio->uio_resid >= 0, ("ffs_extwrite: uio->uio_resid < 0")); 1087 if ((uoff_t)uio->uio_offset + uio->uio_resid > 1088 UFS_NXADDR * fs->fs_bsize) 1089 return (EFBIG); 1090 1091 resid = uio->uio_resid; 1092 osize = dp->di_extsize; 1093 flags = IO_EXT; 1094 if (ioflag & IO_SYNC) 1095 flags |= IO_SYNC; 1096 1097 for (error = 0; uio->uio_resid > 0;) { 1098 lbn = lblkno(fs, uio->uio_offset); 1099 blkoffset = blkoff(fs, uio->uio_offset); 1100 xfersize = fs->fs_bsize - blkoffset; 1101 if (uio->uio_resid < xfersize) 1102 xfersize = uio->uio_resid; 1103 1104 /* 1105 * We must perform a read-before-write if the transfer size 1106 * does not cover the entire buffer. 1107 */ 1108 if (fs->fs_bsize > xfersize) 1109 flags |= BA_CLRBUF; 1110 else 1111 flags &= ~BA_CLRBUF; 1112 error = UFS_BALLOC(vp, uio->uio_offset, xfersize, 1113 ucred, flags, &bp); 1114 if (error != 0) 1115 break; 1116 /* 1117 * If the buffer is not valid we have to clear out any 1118 * garbage data from the pages instantiated for the buffer. 1119 * If we do not, a failed uiomove() during a write can leave 1120 * the prior contents of the pages exposed to a userland 1121 * mmap(). XXX deal with uiomove() errors a better way. 1122 */ 1123 if ((bp->b_flags & B_CACHE) == 0 && fs->fs_bsize <= xfersize) 1124 vfs_bio_clrbuf(bp); 1125 1126 if (uio->uio_offset + xfersize > dp->di_extsize) { 1127 dp->di_extsize = uio->uio_offset + xfersize; 1128 UFS_INODE_SET_FLAG(ip, IN_SIZEMOD | IN_CHANGE); 1129 } 1130 1131 size = sblksize(fs, dp->di_extsize, lbn) - bp->b_resid; 1132 if (size < xfersize) 1133 xfersize = size; 1134 1135 error = 1136 uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio); 1137 1138 vfs_bio_set_flags(bp, ioflag); 1139 1140 /* 1141 * If IO_SYNC each buffer is written synchronously. Otherwise 1142 * if we have a severe page deficiency write the buffer 1143 * asynchronously. Otherwise try to cluster, and if that 1144 * doesn't do it then either do an async write (if O_DIRECT), 1145 * or a delayed write (if not). 1146 */ 1147 if (ioflag & IO_SYNC) { 1148 (void)bwrite(bp); 1149 } else if (vm_page_count_severe() || 1150 buf_dirty_count_severe() || 1151 xfersize + blkoffset == fs->fs_bsize || 1152 (ioflag & (IO_ASYNC | IO_DIRECT))) 1153 bawrite(bp); 1154 else 1155 bdwrite(bp); 1156 if (error || xfersize == 0) 1157 break; 1158 UFS_INODE_SET_FLAG(ip, IN_CHANGE); 1159 } 1160 /* 1161 * If we successfully wrote any data, and we are not the superuser 1162 * we clear the setuid and setgid bits as a precaution against 1163 * tampering. 1164 */ 1165 if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid && ucred) { 1166 if (priv_check_cred(ucred, PRIV_VFS_RETAINSUGID)) { 1167 vn_seqc_write_begin(vp); 1168 UFS_INODE_SET_MODE(ip, ip->i_mode & ~(ISUID | ISGID)); 1169 dp->di_mode = ip->i_mode; 1170 vn_seqc_write_end(vp); 1171 } 1172 } 1173 if (error) { 1174 if (ioflag & IO_UNIT) { 1175 (void)ffs_truncate(vp, osize, 1176 IO_EXT | (ioflag&IO_SYNC), ucred); 1177 uio->uio_offset -= resid - uio->uio_resid; 1178 uio->uio_resid = resid; 1179 } 1180 } else if (resid > uio->uio_resid && (ioflag & IO_SYNC)) 1181 error = ffs_update(vp, 1); 1182 return (error); 1183 } 1184 1185 /* 1186 * Vnode operating to retrieve a named extended attribute. 1187 * 1188 * Locate a particular EA (nspace:name) in the area (ptr:length), and return 1189 * the length of the EA, and possibly the pointer to the entry and to the data. 1190 */ 1191 static int 1192 ffs_findextattr(u_char *ptr, u_int length, int nspace, const char *name, 1193 struct extattr **eapp, u_char **eac) 1194 { 1195 struct extattr *eap, *eaend; 1196 size_t nlen; 1197 1198 nlen = strlen(name); 1199 KASSERT(ALIGNED_TO(ptr, struct extattr), ("unaligned")); 1200 eap = (struct extattr *)ptr; 1201 eaend = (struct extattr *)(ptr + length); 1202 for (; eap < eaend; eap = EXTATTR_NEXT(eap)) { 1203 KASSERT(EXTATTR_NEXT(eap) <= eaend, 1204 ("extattr next %p beyond %p", EXTATTR_NEXT(eap), eaend)); 1205 if (eap->ea_namespace != nspace || eap->ea_namelength != nlen 1206 || memcmp(eap->ea_name, name, nlen) != 0) 1207 continue; 1208 if (eapp != NULL) 1209 *eapp = eap; 1210 if (eac != NULL) 1211 *eac = EXTATTR_CONTENT(eap); 1212 return (EXTATTR_CONTENT_SIZE(eap)); 1213 } 1214 return (-1); 1215 } 1216 1217 static int 1218 ffs_rdextattr(u_char **p, struct vnode *vp, struct thread *td) 1219 { 1220 const struct extattr *eap, *eaend, *eapnext; 1221 struct inode *ip; 1222 struct ufs2_dinode *dp; 1223 struct fs *fs; 1224 struct uio luio; 1225 struct iovec liovec; 1226 u_int easize; 1227 int error; 1228 u_char *eae; 1229 1230 ip = VTOI(vp); 1231 fs = ITOFS(ip); 1232 dp = ip->i_din2; 1233 easize = dp->di_extsize; 1234 if ((uoff_t)easize > UFS_NXADDR * fs->fs_bsize) 1235 return (EFBIG); 1236 1237 eae = malloc(easize, M_TEMP, M_WAITOK); 1238 1239 liovec.iov_base = eae; 1240 liovec.iov_len = easize; 1241 luio.uio_iov = &liovec; 1242 luio.uio_iovcnt = 1; 1243 luio.uio_offset = 0; 1244 luio.uio_resid = easize; 1245 luio.uio_segflg = UIO_SYSSPACE; 1246 luio.uio_rw = UIO_READ; 1247 luio.uio_td = td; 1248 1249 error = ffs_extread(vp, &luio, IO_EXT | IO_SYNC); 1250 if (error) { 1251 free(eae, M_TEMP); 1252 return (error); 1253 } 1254 /* Validate disk xattrfile contents. */ 1255 for (eap = (void *)eae, eaend = (void *)(eae + easize); eap < eaend; 1256 eap = eapnext) { 1257 eapnext = EXTATTR_NEXT(eap); 1258 /* Bogusly short entry or bogusly long entry. */ 1259 if (eap->ea_length < sizeof(*eap) || eapnext > eaend) { 1260 free(eae, M_TEMP); 1261 return (EINTEGRITY); 1262 } 1263 } 1264 *p = eae; 1265 return (0); 1266 } 1267 1268 static void 1269 ffs_lock_ea(struct vnode *vp) 1270 { 1271 struct inode *ip; 1272 1273 ip = VTOI(vp); 1274 VI_LOCK(vp); 1275 while (ip->i_flag & IN_EA_LOCKED) { 1276 UFS_INODE_SET_FLAG(ip, IN_EA_LOCKWAIT); 1277 msleep(&ip->i_ea_refs, &vp->v_interlock, PINOD + 2, "ufs_ea", 1278 0); 1279 } 1280 UFS_INODE_SET_FLAG(ip, IN_EA_LOCKED); 1281 VI_UNLOCK(vp); 1282 } 1283 1284 static void 1285 ffs_unlock_ea(struct vnode *vp) 1286 { 1287 struct inode *ip; 1288 1289 ip = VTOI(vp); 1290 VI_LOCK(vp); 1291 if (ip->i_flag & IN_EA_LOCKWAIT) 1292 wakeup(&ip->i_ea_refs); 1293 ip->i_flag &= ~(IN_EA_LOCKED | IN_EA_LOCKWAIT); 1294 VI_UNLOCK(vp); 1295 } 1296 1297 static int 1298 ffs_open_ea(struct vnode *vp, struct ucred *cred, struct thread *td) 1299 { 1300 struct inode *ip; 1301 struct ufs2_dinode *dp; 1302 int error; 1303 1304 ip = VTOI(vp); 1305 1306 ffs_lock_ea(vp); 1307 if (ip->i_ea_area != NULL) { 1308 ip->i_ea_refs++; 1309 ffs_unlock_ea(vp); 1310 return (0); 1311 } 1312 dp = ip->i_din2; 1313 error = ffs_rdextattr(&ip->i_ea_area, vp, td); 1314 if (error) { 1315 ffs_unlock_ea(vp); 1316 return (error); 1317 } 1318 ip->i_ea_len = dp->di_extsize; 1319 ip->i_ea_error = 0; 1320 ip->i_ea_refs++; 1321 ffs_unlock_ea(vp); 1322 return (0); 1323 } 1324 1325 /* 1326 * Vnode extattr transaction commit/abort 1327 */ 1328 static int 1329 ffs_close_ea(struct vnode *vp, int commit, struct ucred *cred, struct thread *td) 1330 { 1331 struct inode *ip; 1332 struct uio luio; 1333 struct iovec liovec; 1334 int error; 1335 struct ufs2_dinode *dp; 1336 1337 ip = VTOI(vp); 1338 1339 ffs_lock_ea(vp); 1340 if (ip->i_ea_area == NULL) { 1341 ffs_unlock_ea(vp); 1342 return (EINVAL); 1343 } 1344 dp = ip->i_din2; 1345 error = ip->i_ea_error; 1346 if (commit && error == 0) { 1347 ASSERT_VOP_ELOCKED(vp, "ffs_close_ea commit"); 1348 if (cred == NOCRED) 1349 cred = vp->v_mount->mnt_cred; 1350 liovec.iov_base = ip->i_ea_area; 1351 liovec.iov_len = ip->i_ea_len; 1352 luio.uio_iov = &liovec; 1353 luio.uio_iovcnt = 1; 1354 luio.uio_offset = 0; 1355 luio.uio_resid = ip->i_ea_len; 1356 luio.uio_segflg = UIO_SYSSPACE; 1357 luio.uio_rw = UIO_WRITE; 1358 luio.uio_td = td; 1359 /* XXX: I'm not happy about truncating to zero size */ 1360 if (ip->i_ea_len < dp->di_extsize) 1361 error = ffs_truncate(vp, 0, IO_EXT, cred); 1362 error = ffs_extwrite(vp, &luio, IO_EXT | IO_SYNC, cred); 1363 } 1364 if (--ip->i_ea_refs == 0) { 1365 free(ip->i_ea_area, M_TEMP); 1366 ip->i_ea_area = NULL; 1367 ip->i_ea_len = 0; 1368 ip->i_ea_error = 0; 1369 } 1370 ffs_unlock_ea(vp); 1371 return (error); 1372 } 1373 1374 /* 1375 * Vnode extattr strategy routine for fifos. 1376 * 1377 * We need to check for a read or write of the external attributes. 1378 * Otherwise we just fall through and do the usual thing. 1379 */ 1380 static int 1381 ffsext_strategy(struct vop_strategy_args *ap) 1382 /* 1383 struct vop_strategy_args { 1384 struct vnodeop_desc *a_desc; 1385 struct vnode *a_vp; 1386 struct buf *a_bp; 1387 }; 1388 */ 1389 { 1390 struct vnode *vp; 1391 daddr_t lbn; 1392 1393 vp = ap->a_vp; 1394 lbn = ap->a_bp->b_lblkno; 1395 if (I_IS_UFS2(VTOI(vp)) && lbn < 0 && lbn >= -UFS_NXADDR) 1396 return (VOP_STRATEGY_APV(&ufs_vnodeops, ap)); 1397 if (vp->v_type == VFIFO) 1398 return (VOP_STRATEGY_APV(&ufs_fifoops, ap)); 1399 panic("spec nodes went here"); 1400 } 1401 1402 /* 1403 * Vnode extattr transaction commit/abort 1404 */ 1405 static int 1406 ffs_openextattr(struct vop_openextattr_args *ap) 1407 /* 1408 struct vop_openextattr_args { 1409 struct vnodeop_desc *a_desc; 1410 struct vnode *a_vp; 1411 IN struct ucred *a_cred; 1412 IN struct thread *a_td; 1413 }; 1414 */ 1415 { 1416 1417 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK) 1418 return (EOPNOTSUPP); 1419 1420 return (ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td)); 1421 } 1422 1423 /* 1424 * Vnode extattr transaction commit/abort 1425 */ 1426 static int 1427 ffs_closeextattr(struct vop_closeextattr_args *ap) 1428 /* 1429 struct vop_closeextattr_args { 1430 struct vnodeop_desc *a_desc; 1431 struct vnode *a_vp; 1432 int a_commit; 1433 IN struct ucred *a_cred; 1434 IN struct thread *a_td; 1435 }; 1436 */ 1437 { 1438 1439 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK) 1440 return (EOPNOTSUPP); 1441 1442 if (ap->a_commit && (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)) 1443 return (EROFS); 1444 1445 return (ffs_close_ea(ap->a_vp, ap->a_commit, ap->a_cred, ap->a_td)); 1446 } 1447 1448 /* 1449 * Vnode operation to remove a named attribute. 1450 */ 1451 static int 1452 ffs_deleteextattr(struct vop_deleteextattr_args *ap) 1453 /* 1454 vop_deleteextattr { 1455 IN struct vnode *a_vp; 1456 IN int a_attrnamespace; 1457 IN const char *a_name; 1458 IN struct ucred *a_cred; 1459 IN struct thread *a_td; 1460 }; 1461 */ 1462 { 1463 struct inode *ip; 1464 struct extattr *eap; 1465 uint32_t ul; 1466 int olen, error, i, easize; 1467 u_char *eae; 1468 void *tmp; 1469 1470 ip = VTOI(ap->a_vp); 1471 1472 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK) 1473 return (EOPNOTSUPP); 1474 1475 if (strlen(ap->a_name) == 0) 1476 return (EINVAL); 1477 1478 if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY) 1479 return (EROFS); 1480 1481 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 1482 ap->a_cred, ap->a_td, VWRITE); 1483 if (error) { 1484 /* 1485 * ffs_lock_ea is not needed there, because the vnode 1486 * must be exclusively locked. 1487 */ 1488 if (ip->i_ea_area != NULL && ip->i_ea_error == 0) 1489 ip->i_ea_error = error; 1490 return (error); 1491 } 1492 1493 error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td); 1494 if (error) 1495 return (error); 1496 1497 /* CEM: delete could be done in-place instead */ 1498 eae = malloc(ip->i_ea_len, M_TEMP, M_WAITOK); 1499 bcopy(ip->i_ea_area, eae, ip->i_ea_len); 1500 easize = ip->i_ea_len; 1501 1502 olen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name, 1503 &eap, NULL); 1504 if (olen == -1) { 1505 /* delete but nonexistent */ 1506 free(eae, M_TEMP); 1507 ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td); 1508 return (ENOATTR); 1509 } 1510 ul = eap->ea_length; 1511 i = (u_char *)EXTATTR_NEXT(eap) - eae; 1512 bcopy(EXTATTR_NEXT(eap), eap, easize - i); 1513 easize -= ul; 1514 1515 tmp = ip->i_ea_area; 1516 ip->i_ea_area = eae; 1517 ip->i_ea_len = easize; 1518 free(tmp, M_TEMP); 1519 error = ffs_close_ea(ap->a_vp, 1, ap->a_cred, ap->a_td); 1520 return (error); 1521 } 1522 1523 /* 1524 * Vnode operation to retrieve a named extended attribute. 1525 */ 1526 static int 1527 ffs_getextattr(struct vop_getextattr_args *ap) 1528 /* 1529 vop_getextattr { 1530 IN struct vnode *a_vp; 1531 IN int a_attrnamespace; 1532 IN const char *a_name; 1533 INOUT struct uio *a_uio; 1534 OUT size_t *a_size; 1535 IN struct ucred *a_cred; 1536 IN struct thread *a_td; 1537 }; 1538 */ 1539 { 1540 struct inode *ip; 1541 u_char *eae, *p; 1542 unsigned easize; 1543 int error, ealen; 1544 1545 ip = VTOI(ap->a_vp); 1546 1547 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK) 1548 return (EOPNOTSUPP); 1549 1550 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 1551 ap->a_cred, ap->a_td, VREAD); 1552 if (error) 1553 return (error); 1554 1555 error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td); 1556 if (error) 1557 return (error); 1558 1559 eae = ip->i_ea_area; 1560 easize = ip->i_ea_len; 1561 1562 ealen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name, 1563 NULL, &p); 1564 if (ealen >= 0) { 1565 error = 0; 1566 if (ap->a_size != NULL) 1567 *ap->a_size = ealen; 1568 else if (ap->a_uio != NULL) 1569 error = uiomove(p, ealen, ap->a_uio); 1570 } else 1571 error = ENOATTR; 1572 1573 ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td); 1574 return (error); 1575 } 1576 1577 /* 1578 * Vnode operation to retrieve extended attributes on a vnode. 1579 */ 1580 static int 1581 ffs_listextattr(struct vop_listextattr_args *ap) 1582 /* 1583 vop_listextattr { 1584 IN struct vnode *a_vp; 1585 IN int a_attrnamespace; 1586 INOUT struct uio *a_uio; 1587 OUT size_t *a_size; 1588 IN struct ucred *a_cred; 1589 IN struct thread *a_td; 1590 }; 1591 */ 1592 { 1593 struct inode *ip; 1594 struct extattr *eap, *eaend; 1595 int error, ealen; 1596 1597 ip = VTOI(ap->a_vp); 1598 1599 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK) 1600 return (EOPNOTSUPP); 1601 1602 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 1603 ap->a_cred, ap->a_td, VREAD); 1604 if (error) 1605 return (error); 1606 1607 error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td); 1608 if (error) 1609 return (error); 1610 1611 error = 0; 1612 if (ap->a_size != NULL) 1613 *ap->a_size = 0; 1614 1615 KASSERT(ALIGNED_TO(ip->i_ea_area, struct extattr), ("unaligned")); 1616 eap = (struct extattr *)ip->i_ea_area; 1617 eaend = (struct extattr *)(ip->i_ea_area + ip->i_ea_len); 1618 for (; error == 0 && eap < eaend; eap = EXTATTR_NEXT(eap)) { 1619 KASSERT(EXTATTR_NEXT(eap) <= eaend, 1620 ("extattr next %p beyond %p", EXTATTR_NEXT(eap), eaend)); 1621 if (eap->ea_namespace != ap->a_attrnamespace) 1622 continue; 1623 1624 ealen = eap->ea_namelength; 1625 if (ap->a_size != NULL) 1626 *ap->a_size += ealen + 1; 1627 else if (ap->a_uio != NULL) 1628 error = uiomove(&eap->ea_namelength, ealen + 1, 1629 ap->a_uio); 1630 } 1631 1632 ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td); 1633 return (error); 1634 } 1635 1636 /* 1637 * Vnode operation to set a named attribute. 1638 */ 1639 static int 1640 ffs_setextattr(struct vop_setextattr_args *ap) 1641 /* 1642 vop_setextattr { 1643 IN struct vnode *a_vp; 1644 IN int a_attrnamespace; 1645 IN const char *a_name; 1646 INOUT struct uio *a_uio; 1647 IN struct ucred *a_cred; 1648 IN struct thread *a_td; 1649 }; 1650 */ 1651 { 1652 struct inode *ip; 1653 struct fs *fs; 1654 struct extattr *eap; 1655 uint32_t ealength, ul; 1656 ssize_t ealen; 1657 int olen, eapad1, eapad2, error, i, easize; 1658 u_char *eae; 1659 void *tmp; 1660 1661 ip = VTOI(ap->a_vp); 1662 fs = ITOFS(ip); 1663 1664 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK) 1665 return (EOPNOTSUPP); 1666 1667 if (strlen(ap->a_name) == 0) 1668 return (EINVAL); 1669 1670 /* XXX Now unsupported API to delete EAs using NULL uio. */ 1671 if (ap->a_uio == NULL) 1672 return (EOPNOTSUPP); 1673 1674 if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY) 1675 return (EROFS); 1676 1677 ealen = ap->a_uio->uio_resid; 1678 if (ealen < 0 || ealen > lblktosize(fs, UFS_NXADDR)) 1679 return (EINVAL); 1680 1681 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 1682 ap->a_cred, ap->a_td, VWRITE); 1683 if (error) { 1684 /* 1685 * ffs_lock_ea is not needed there, because the vnode 1686 * must be exclusively locked. 1687 */ 1688 if (ip->i_ea_area != NULL && ip->i_ea_error == 0) 1689 ip->i_ea_error = error; 1690 return (error); 1691 } 1692 1693 error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td); 1694 if (error) 1695 return (error); 1696 1697 ealength = sizeof(uint32_t) + 3 + strlen(ap->a_name); 1698 eapad1 = roundup2(ealength, 8) - ealength; 1699 eapad2 = roundup2(ealen, 8) - ealen; 1700 ealength += eapad1 + ealen + eapad2; 1701 1702 /* 1703 * CEM: rewrites of the same size or smaller could be done in-place 1704 * instead. (We don't acquire any fine-grained locks in here either, 1705 * so we could also do bigger writes in-place.) 1706 */ 1707 eae = malloc(ip->i_ea_len + ealength, M_TEMP, M_WAITOK); 1708 bcopy(ip->i_ea_area, eae, ip->i_ea_len); 1709 easize = ip->i_ea_len; 1710 1711 olen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name, 1712 &eap, NULL); 1713 if (olen == -1) { 1714 /* new, append at end */ 1715 KASSERT(ALIGNED_TO(eae + easize, struct extattr), 1716 ("unaligned")); 1717 eap = (struct extattr *)(eae + easize); 1718 easize += ealength; 1719 } else { 1720 ul = eap->ea_length; 1721 i = (u_char *)EXTATTR_NEXT(eap) - eae; 1722 if (ul != ealength) { 1723 bcopy(EXTATTR_NEXT(eap), (u_char *)eap + ealength, 1724 easize - i); 1725 easize += (ealength - ul); 1726 } 1727 } 1728 if (easize > lblktosize(fs, UFS_NXADDR)) { 1729 free(eae, M_TEMP); 1730 ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td); 1731 if (ip->i_ea_area != NULL && ip->i_ea_error == 0) 1732 ip->i_ea_error = ENOSPC; 1733 return (ENOSPC); 1734 } 1735 eap->ea_length = ealength; 1736 eap->ea_namespace = ap->a_attrnamespace; 1737 eap->ea_contentpadlen = eapad2; 1738 eap->ea_namelength = strlen(ap->a_name); 1739 memcpy(eap->ea_name, ap->a_name, strlen(ap->a_name)); 1740 bzero(&eap->ea_name[strlen(ap->a_name)], eapad1); 1741 error = uiomove(EXTATTR_CONTENT(eap), ealen, ap->a_uio); 1742 if (error) { 1743 free(eae, M_TEMP); 1744 ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td); 1745 if (ip->i_ea_area != NULL && ip->i_ea_error == 0) 1746 ip->i_ea_error = error; 1747 return (error); 1748 } 1749 bzero((u_char *)EXTATTR_CONTENT(eap) + ealen, eapad2); 1750 1751 tmp = ip->i_ea_area; 1752 ip->i_ea_area = eae; 1753 ip->i_ea_len = easize; 1754 free(tmp, M_TEMP); 1755 error = ffs_close_ea(ap->a_vp, 1, ap->a_cred, ap->a_td); 1756 return (error); 1757 } 1758 1759 /* 1760 * Vnode pointer to File handle 1761 */ 1762 static int 1763 ffs_vptofh(struct vop_vptofh_args *ap) 1764 /* 1765 vop_vptofh { 1766 IN struct vnode *a_vp; 1767 IN struct fid *a_fhp; 1768 }; 1769 */ 1770 { 1771 struct inode *ip; 1772 struct ufid *ufhp; 1773 1774 ip = VTOI(ap->a_vp); 1775 ufhp = (struct ufid *)ap->a_fhp; 1776 ufhp->ufid_len = sizeof(struct ufid); 1777 ufhp->ufid_ino = ip->i_number; 1778 ufhp->ufid_gen = ip->i_gen; 1779 return (0); 1780 } 1781 1782 SYSCTL_DECL(_vfs_ffs); 1783 static int use_buf_pager = 1; 1784 SYSCTL_INT(_vfs_ffs, OID_AUTO, use_buf_pager, CTLFLAG_RWTUN, &use_buf_pager, 0, 1785 "Always use buffer pager instead of bmap"); 1786 1787 static daddr_t 1788 ffs_gbp_getblkno(struct vnode *vp, vm_ooffset_t off) 1789 { 1790 1791 return (lblkno(VFSTOUFS(vp->v_mount)->um_fs, off)); 1792 } 1793 1794 static int 1795 ffs_gbp_getblksz(struct vnode *vp, daddr_t lbn) 1796 { 1797 1798 return (blksize(VFSTOUFS(vp->v_mount)->um_fs, VTOI(vp), lbn)); 1799 } 1800 1801 static int 1802 ffs_getpages(struct vop_getpages_args *ap) 1803 { 1804 struct vnode *vp; 1805 struct ufsmount *um; 1806 1807 vp = ap->a_vp; 1808 um = VFSTOUFS(vp->v_mount); 1809 1810 if (!use_buf_pager && um->um_devvp->v_bufobj.bo_bsize <= PAGE_SIZE) 1811 return (vnode_pager_generic_getpages(vp, ap->a_m, ap->a_count, 1812 ap->a_rbehind, ap->a_rahead, NULL, NULL)); 1813 return (vfs_bio_getpages(vp, ap->a_m, ap->a_count, ap->a_rbehind, 1814 ap->a_rahead, ffs_gbp_getblkno, ffs_gbp_getblksz)); 1815 } 1816 1817 static int 1818 ffs_getpages_async(struct vop_getpages_async_args *ap) 1819 { 1820 struct vnode *vp; 1821 struct ufsmount *um; 1822 bool do_iodone; 1823 int error; 1824 1825 vp = ap->a_vp; 1826 um = VFSTOUFS(vp->v_mount); 1827 do_iodone = true; 1828 1829 if (um->um_devvp->v_bufobj.bo_bsize <= PAGE_SIZE) { 1830 error = vnode_pager_generic_getpages(vp, ap->a_m, ap->a_count, 1831 ap->a_rbehind, ap->a_rahead, ap->a_iodone, ap->a_arg); 1832 if (error == 0) 1833 do_iodone = false; 1834 } else { 1835 error = vfs_bio_getpages(vp, ap->a_m, ap->a_count, 1836 ap->a_rbehind, ap->a_rahead, ffs_gbp_getblkno, 1837 ffs_gbp_getblksz); 1838 } 1839 if (do_iodone && ap->a_iodone != NULL) 1840 ap->a_iodone(ap->a_arg, ap->a_m, ap->a_count, error); 1841 1842 return (error); 1843 } 1844