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 "opt_directio.h" 70 #include "opt_ffs.h" 71 #include "opt_ufs.h" 72 73 #include <sys/param.h> 74 #include <sys/bio.h> 75 #include <sys/systm.h> 76 #include <sys/buf.h> 77 #include <sys/conf.h> 78 #include <sys/extattr.h> 79 #include <sys/kernel.h> 80 #include <sys/limits.h> 81 #include <sys/malloc.h> 82 #include <sys/mount.h> 83 #include <sys/priv.h> 84 #include <sys/rwlock.h> 85 #include <sys/stat.h> 86 #include <sys/sysctl.h> 87 #include <sys/vmmeter.h> 88 #include <sys/vnode.h> 89 90 #include <vm/vm.h> 91 #include <vm/vm_param.h> 92 #include <vm/vm_extern.h> 93 #include <vm/vm_object.h> 94 #include <vm/vm_page.h> 95 #include <vm/vm_pager.h> 96 #include <vm/vnode_pager.h> 97 98 #include <ufs/ufs/extattr.h> 99 #include <ufs/ufs/quota.h> 100 #include <ufs/ufs/inode.h> 101 #include <ufs/ufs/ufs_extern.h> 102 #include <ufs/ufs/ufsmount.h> 103 #include <ufs/ufs/dir.h> 104 #ifdef UFS_DIRHASH 105 #include <ufs/ufs/dirhash.h> 106 #endif 107 108 #include <ufs/ffs/fs.h> 109 #include <ufs/ffs/ffs_extern.h> 110 111 #define ALIGNED_TO(ptr, s) \ 112 (((uintptr_t)(ptr) & (_Alignof(s) - 1)) == 0) 113 114 #ifdef DIRECTIO 115 extern int ffs_rawread(struct vnode *vp, struct uio *uio, int *workdone); 116 #endif 117 static vop_fdatasync_t ffs_fdatasync; 118 static vop_fsync_t ffs_fsync; 119 static vop_getpages_t ffs_getpages; 120 static vop_getpages_async_t ffs_getpages_async; 121 static vop_lock1_t ffs_lock; 122 #ifdef INVARIANTS 123 static vop_unlock_t ffs_unlock_debug; 124 #endif 125 static vop_read_t ffs_read; 126 static vop_write_t ffs_write; 127 static int ffs_extread(struct vnode *vp, struct uio *uio, int ioflag); 128 static int ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag, 129 struct ucred *cred); 130 static vop_strategy_t ffsext_strategy; 131 static vop_closeextattr_t ffs_closeextattr; 132 static vop_deleteextattr_t ffs_deleteextattr; 133 static vop_getextattr_t ffs_getextattr; 134 static vop_listextattr_t ffs_listextattr; 135 static vop_openextattr_t ffs_openextattr; 136 static vop_setextattr_t ffs_setextattr; 137 static vop_vptofh_t ffs_vptofh; 138 static vop_vput_pair_t ffs_vput_pair; 139 140 /* Global vfs data structures for ufs. */ 141 struct vop_vector ffs_vnodeops1 = { 142 .vop_default = &ufs_vnodeops, 143 .vop_fsync = ffs_fsync, 144 .vop_fdatasync = ffs_fdatasync, 145 .vop_getpages = ffs_getpages, 146 .vop_getpages_async = ffs_getpages_async, 147 .vop_lock1 = ffs_lock, 148 #ifdef INVARIANTS 149 .vop_unlock = ffs_unlock_debug, 150 #endif 151 .vop_read = ffs_read, 152 .vop_reallocblks = ffs_reallocblks, 153 .vop_write = ffs_write, 154 .vop_vptofh = ffs_vptofh, 155 .vop_vput_pair = ffs_vput_pair, 156 }; 157 VFS_VOP_VECTOR_REGISTER(ffs_vnodeops1); 158 159 struct vop_vector ffs_fifoops1 = { 160 .vop_default = &ufs_fifoops, 161 .vop_fsync = ffs_fsync, 162 .vop_fdatasync = ffs_fdatasync, 163 .vop_lock1 = ffs_lock, 164 #ifdef INVARIANTS 165 .vop_unlock = ffs_unlock_debug, 166 #endif 167 .vop_vptofh = ffs_vptofh, 168 }; 169 VFS_VOP_VECTOR_REGISTER(ffs_fifoops1); 170 171 /* Global vfs data structures for ufs. */ 172 struct vop_vector ffs_vnodeops2 = { 173 .vop_default = &ufs_vnodeops, 174 .vop_fsync = ffs_fsync, 175 .vop_fdatasync = ffs_fdatasync, 176 .vop_getpages = ffs_getpages, 177 .vop_getpages_async = ffs_getpages_async, 178 .vop_lock1 = ffs_lock, 179 #ifdef INVARIANTS 180 .vop_unlock = ffs_unlock_debug, 181 #endif 182 .vop_read = ffs_read, 183 .vop_reallocblks = ffs_reallocblks, 184 .vop_write = ffs_write, 185 .vop_closeextattr = ffs_closeextattr, 186 .vop_deleteextattr = ffs_deleteextattr, 187 .vop_getextattr = ffs_getextattr, 188 .vop_listextattr = ffs_listextattr, 189 .vop_openextattr = ffs_openextattr, 190 .vop_setextattr = ffs_setextattr, 191 .vop_vptofh = ffs_vptofh, 192 .vop_vput_pair = ffs_vput_pair, 193 }; 194 VFS_VOP_VECTOR_REGISTER(ffs_vnodeops2); 195 196 struct vop_vector ffs_fifoops2 = { 197 .vop_default = &ufs_fifoops, 198 .vop_fsync = ffs_fsync, 199 .vop_fdatasync = ffs_fdatasync, 200 .vop_lock1 = ffs_lock, 201 #ifdef INVARIANTS 202 .vop_unlock = ffs_unlock_debug, 203 #endif 204 .vop_reallocblks = ffs_reallocblks, 205 .vop_strategy = ffsext_strategy, 206 .vop_closeextattr = ffs_closeextattr, 207 .vop_deleteextattr = ffs_deleteextattr, 208 .vop_getextattr = ffs_getextattr, 209 .vop_listextattr = ffs_listextattr, 210 .vop_openextattr = ffs_openextattr, 211 .vop_setextattr = ffs_setextattr, 212 .vop_vptofh = ffs_vptofh, 213 }; 214 VFS_VOP_VECTOR_REGISTER(ffs_fifoops2); 215 216 /* 217 * Synch an open file. 218 */ 219 /* ARGSUSED */ 220 static int 221 ffs_fsync(struct vop_fsync_args *ap) 222 { 223 struct vnode *vp; 224 struct bufobj *bo; 225 int error; 226 227 vp = ap->a_vp; 228 bo = &vp->v_bufobj; 229 retry: 230 error = ffs_syncvnode(vp, ap->a_waitfor, 0); 231 if (error) 232 return (error); 233 if (ap->a_waitfor == MNT_WAIT && DOINGSOFTDEP(vp)) { 234 error = softdep_fsync(vp); 235 if (error) 236 return (error); 237 238 /* 239 * The softdep_fsync() function may drop vp lock, 240 * allowing for dirty buffers to reappear on the 241 * bo_dirty list. Recheck and resync as needed. 242 */ 243 BO_LOCK(bo); 244 if ((vp->v_type == VREG || vp->v_type == VDIR) && 245 (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)) { 246 BO_UNLOCK(bo); 247 goto retry; 248 } 249 BO_UNLOCK(bo); 250 } 251 if (ffs_fsfail_cleanup(VFSTOUFS(vp->v_mount), 0)) 252 return (ENXIO); 253 return (0); 254 } 255 256 int 257 ffs_syncvnode(struct vnode *vp, int waitfor, int flags) 258 { 259 struct inode *ip; 260 struct bufobj *bo; 261 struct ufsmount *ump; 262 struct buf *bp, *nbp; 263 ufs_lbn_t lbn; 264 int error, passes; 265 bool still_dirty, unlocked, wait; 266 267 ip = VTOI(vp); 268 bo = &vp->v_bufobj; 269 ump = VFSTOUFS(vp->v_mount); 270 271 /* 272 * When doing MNT_WAIT we must first flush all dependencies 273 * on the inode. 274 */ 275 if (DOINGSOFTDEP(vp) && waitfor == MNT_WAIT && 276 (error = softdep_sync_metadata(vp)) != 0) { 277 if (ffs_fsfail_cleanup(ump, error)) 278 error = 0; 279 return (error); 280 } 281 282 /* 283 * Flush all dirty buffers associated with a vnode. 284 */ 285 error = 0; 286 passes = 0; 287 wait = false; /* Always do an async pass first. */ 288 unlocked = false; 289 lbn = lblkno(ITOFS(ip), (ip->i_size + ITOFS(ip)->fs_bsize - 1)); 290 BO_LOCK(bo); 291 loop: 292 TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs) 293 bp->b_vflags &= ~BV_SCANNED; 294 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 295 /* 296 * Reasons to skip this buffer: it has already been considered 297 * on this pass, the buffer has dependencies that will cause 298 * it to be redirtied and it has not already been deferred, 299 * or it is already being written. 300 */ 301 if ((bp->b_vflags & BV_SCANNED) != 0) 302 continue; 303 bp->b_vflags |= BV_SCANNED; 304 /* 305 * Flush indirects in order, if requested. 306 * 307 * Note that if only datasync is requested, we can 308 * skip indirect blocks when softupdates are not 309 * active. Otherwise we must flush them with data, 310 * since dependencies prevent data block writes. 311 */ 312 if (waitfor == MNT_WAIT && bp->b_lblkno <= -UFS_NDADDR && 313 (lbn_level(bp->b_lblkno) >= passes || 314 ((flags & DATA_ONLY) != 0 && !DOINGSOFTDEP(vp)))) 315 continue; 316 if (bp->b_lblkno > lbn) 317 panic("ffs_syncvnode: syncing truncated data."); 318 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL) == 0) { 319 BO_UNLOCK(bo); 320 } else if (wait) { 321 if (BUF_LOCK(bp, 322 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, 323 BO_LOCKPTR(bo)) != 0) { 324 BO_LOCK(bo); 325 bp->b_vflags &= ~BV_SCANNED; 326 goto next_locked; 327 } 328 } else 329 continue; 330 if ((bp->b_flags & B_DELWRI) == 0) 331 panic("ffs_fsync: not dirty"); 332 /* 333 * Check for dependencies and potentially complete them. 334 */ 335 if (!LIST_EMPTY(&bp->b_dep) && 336 (error = softdep_sync_buf(vp, bp, 337 wait ? MNT_WAIT : MNT_NOWAIT)) != 0) { 338 /* 339 * Lock order conflict, buffer was already unlocked, 340 * and vnode possibly unlocked. 341 */ 342 if (error == ERELOOKUP) { 343 if (vp->v_data == NULL) 344 return (EBADF); 345 unlocked = true; 346 if (DOINGSOFTDEP(vp) && waitfor == MNT_WAIT && 347 (error = softdep_sync_metadata(vp)) != 0) { 348 if (ffs_fsfail_cleanup(ump, error)) 349 error = 0; 350 return (unlocked && error == 0 ? 351 ERELOOKUP : error); 352 } 353 /* Re-evaluate inode size */ 354 lbn = lblkno(ITOFS(ip), (ip->i_size + 355 ITOFS(ip)->fs_bsize - 1)); 356 goto next; 357 } 358 /* I/O error. */ 359 if (error != EBUSY) { 360 BUF_UNLOCK(bp); 361 return (error); 362 } 363 /* If we deferred once, don't defer again. */ 364 if ((bp->b_flags & B_DEFERRED) == 0) { 365 bp->b_flags |= B_DEFERRED; 366 BUF_UNLOCK(bp); 367 goto next; 368 } 369 } 370 if (wait) { 371 bremfree(bp); 372 error = bwrite(bp); 373 if (ffs_fsfail_cleanup(ump, error)) 374 error = 0; 375 if (error != 0) 376 return (error); 377 } else if ((bp->b_flags & B_CLUSTEROK)) { 378 (void) vfs_bio_awrite(bp); 379 } else { 380 bremfree(bp); 381 (void) bawrite(bp); 382 } 383 next: 384 /* 385 * Since we may have slept during the I/O, we need 386 * to start from a known point. 387 */ 388 BO_LOCK(bo); 389 next_locked: 390 nbp = TAILQ_FIRST(&bo->bo_dirty.bv_hd); 391 } 392 if (waitfor != MNT_WAIT) { 393 BO_UNLOCK(bo); 394 if ((flags & NO_INO_UPDT) != 0) 395 return (unlocked ? ERELOOKUP : 0); 396 error = ffs_update(vp, 0); 397 if (error == 0 && unlocked) 398 error = ERELOOKUP; 399 return (error); 400 } 401 /* Drain IO to see if we're done. */ 402 bufobj_wwait(bo, 0, 0); 403 /* 404 * Block devices associated with filesystems may have new I/O 405 * requests posted for them even if the vnode is locked, so no 406 * amount of trying will get them clean. We make several passes 407 * as a best effort. 408 * 409 * Regular files may need multiple passes to flush all dependency 410 * work as it is possible that we must write once per indirect 411 * level, once for the leaf, and once for the inode and each of 412 * these will be done with one sync and one async pass. 413 */ 414 if (bo->bo_dirty.bv_cnt > 0) { 415 if ((flags & DATA_ONLY) == 0) { 416 still_dirty = true; 417 } else { 418 /* 419 * For data-only sync, dirty indirect buffers 420 * are ignored. 421 */ 422 still_dirty = false; 423 TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs) { 424 if (bp->b_lblkno > -UFS_NDADDR) { 425 still_dirty = true; 426 break; 427 } 428 } 429 } 430 431 if (still_dirty) { 432 /* Write the inode after sync passes to flush deps. */ 433 if (wait && DOINGSOFTDEP(vp) && 434 (flags & NO_INO_UPDT) == 0) { 435 BO_UNLOCK(bo); 436 ffs_update(vp, 1); 437 BO_LOCK(bo); 438 } 439 /* switch between sync/async. */ 440 wait = !wait; 441 if (wait || ++passes < UFS_NIADDR + 2) 442 goto loop; 443 } 444 } 445 BO_UNLOCK(bo); 446 error = 0; 447 if ((flags & DATA_ONLY) == 0) { 448 if ((flags & NO_INO_UPDT) == 0) 449 error = ffs_update(vp, 1); 450 if (DOINGSUJ(vp)) 451 softdep_journal_fsync(VTOI(vp)); 452 } else if ((ip->i_flags & (IN_SIZEMOD | IN_IBLKDATA)) != 0) { 453 error = ffs_update(vp, 1); 454 } 455 if (error == 0 && unlocked) 456 error = ERELOOKUP; 457 if (error == 0) 458 ip->i_flag &= ~IN_NEEDSYNC; 459 return (error); 460 } 461 462 static int 463 ffs_fdatasync(struct vop_fdatasync_args *ap) 464 { 465 466 return (ffs_syncvnode(ap->a_vp, MNT_WAIT, DATA_ONLY)); 467 } 468 469 static int 470 ffs_lock(ap) 471 struct vop_lock1_args /* { 472 struct vnode *a_vp; 473 int a_flags; 474 char *file; 475 int line; 476 } */ *ap; 477 { 478 #if !defined(NO_FFS_SNAPSHOT) || defined(DIAGNOSTIC) 479 struct vnode *vp = ap->a_vp; 480 #endif /* !NO_FFS_SNAPSHOT || DIAGNOSTIC */ 481 #ifdef DIAGNOSTIC 482 struct inode *ip; 483 #endif /* DIAGNOSTIC */ 484 int result; 485 #ifndef NO_FFS_SNAPSHOT 486 int flags; 487 struct lock *lkp; 488 489 /* 490 * Adaptive spinning mixed with SU leads to trouble. use a giant hammer 491 * and only use it when LK_NODDLKTREAT is set. Currently this means it 492 * is only used during path lookup. 493 */ 494 if ((ap->a_flags & LK_NODDLKTREAT) != 0) 495 ap->a_flags |= LK_ADAPTIVE; 496 switch (ap->a_flags & LK_TYPE_MASK) { 497 case LK_SHARED: 498 case LK_UPGRADE: 499 case LK_EXCLUSIVE: 500 flags = ap->a_flags; 501 for (;;) { 502 #ifdef DEBUG_VFS_LOCKS 503 VNPASS(vp->v_holdcnt != 0, vp); 504 #endif /* DEBUG_VFS_LOCKS */ 505 lkp = vp->v_vnlock; 506 result = lockmgr_lock_flags(lkp, flags, 507 &VI_MTX(vp)->lock_object, ap->a_file, ap->a_line); 508 if (lkp == vp->v_vnlock || result != 0) 509 break; 510 /* 511 * Apparent success, except that the vnode 512 * mutated between snapshot file vnode and 513 * regular file vnode while this process 514 * slept. The lock currently held is not the 515 * right lock. Release it, and try to get the 516 * new lock. 517 */ 518 lockmgr_unlock(lkp); 519 if ((flags & (LK_INTERLOCK | LK_NOWAIT)) == 520 (LK_INTERLOCK | LK_NOWAIT)) 521 return (EBUSY); 522 if ((flags & LK_TYPE_MASK) == LK_UPGRADE) 523 flags = (flags & ~LK_TYPE_MASK) | LK_EXCLUSIVE; 524 flags &= ~LK_INTERLOCK; 525 } 526 #ifdef DIAGNOSTIC 527 switch (ap->a_flags & LK_TYPE_MASK) { 528 case LK_UPGRADE: 529 case LK_EXCLUSIVE: 530 if (result == 0 && vp->v_vnlock->lk_recurse == 0) { 531 ip = VTOI(vp); 532 if (ip != NULL) 533 ip->i_lock_gen++; 534 } 535 } 536 #endif /* DIAGNOSTIC */ 537 break; 538 default: 539 #ifdef DIAGNOSTIC 540 if ((ap->a_flags & LK_TYPE_MASK) == LK_DOWNGRADE) { 541 ip = VTOI(vp); 542 if (ip != NULL) 543 ufs_unlock_tracker(ip); 544 } 545 #endif /* DIAGNOSTIC */ 546 result = VOP_LOCK1_APV(&ufs_vnodeops, ap); 547 break; 548 } 549 #else /* NO_FFS_SNAPSHOT */ 550 /* 551 * See above for an explanation. 552 */ 553 if ((ap->a_flags & LK_NODDLKTREAT) != 0) 554 ap->a_flags |= LK_ADAPTIVE; 555 #ifdef DIAGNOSTIC 556 if ((ap->a_flags & LK_TYPE_MASK) == LK_DOWNGRADE) { 557 ip = VTOI(vp); 558 if (ip != NULL) 559 ufs_unlock_tracker(ip); 560 } 561 #endif /* DIAGNOSTIC */ 562 result = VOP_LOCK1_APV(&ufs_vnodeops, ap); 563 #endif /* NO_FFS_SNAPSHOT */ 564 #ifdef DIAGNOSTIC 565 switch (ap->a_flags & LK_TYPE_MASK) { 566 case LK_UPGRADE: 567 case LK_EXCLUSIVE: 568 if (result == 0 && vp->v_vnlock->lk_recurse == 0) { 569 ip = VTOI(vp); 570 if (ip != NULL) 571 ip->i_lock_gen++; 572 } 573 } 574 #endif /* DIAGNOSTIC */ 575 return (result); 576 } 577 578 #ifdef INVARIANTS 579 static int 580 ffs_unlock_debug(struct vop_unlock_args *ap) 581 { 582 struct vnode *vp; 583 struct inode *ip; 584 585 vp = ap->a_vp; 586 ip = VTOI(vp); 587 if (ip->i_flag & UFS_INODE_FLAG_LAZY_MASK_ASSERTABLE) { 588 if ((vp->v_mflag & VMP_LAZYLIST) == 0) { 589 VI_LOCK(vp); 590 VNASSERT((vp->v_mflag & VMP_LAZYLIST), vp, 591 ("%s: modified vnode (%x) not on lazy list", 592 __func__, ip->i_flag)); 593 VI_UNLOCK(vp); 594 } 595 } 596 KASSERT(vp->v_type != VDIR || vp->v_vnlock->lk_recurse != 0 || 597 (ip->i_flag & IN_ENDOFF) == 0, 598 ("ufs dir vp %p ip %p flags %#x", vp, ip, ip->i_flag)); 599 #ifdef DIAGNOSTIC 600 if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE && ip != NULL && 601 vp->v_vnlock->lk_recurse == 0) 602 ufs_unlock_tracker(ip); 603 #endif 604 return (VOP_UNLOCK_APV(&ufs_vnodeops, ap)); 605 } 606 #endif 607 608 static int 609 ffs_read_hole(struct uio *uio, long xfersize, long *size) 610 { 611 ssize_t saved_resid, tlen; 612 int error; 613 614 while (xfersize > 0) { 615 tlen = min(xfersize, ZERO_REGION_SIZE); 616 saved_resid = uio->uio_resid; 617 error = vn_io_fault_uiomove(__DECONST(void *, zero_region), 618 tlen, uio); 619 if (error != 0) 620 return (error); 621 tlen = saved_resid - uio->uio_resid; 622 xfersize -= tlen; 623 *size -= tlen; 624 } 625 return (0); 626 } 627 628 /* 629 * Vnode op for reading. 630 */ 631 static int 632 ffs_read(ap) 633 struct vop_read_args /* { 634 struct vnode *a_vp; 635 struct uio *a_uio; 636 int a_ioflag; 637 struct ucred *a_cred; 638 } */ *ap; 639 { 640 struct vnode *vp; 641 struct inode *ip; 642 struct uio *uio; 643 struct fs *fs; 644 struct buf *bp; 645 ufs_lbn_t lbn, nextlbn; 646 off_t bytesinfile; 647 long size, xfersize, blkoffset; 648 ssize_t orig_resid; 649 int bflag, error, ioflag, seqcount; 650 651 vp = ap->a_vp; 652 uio = ap->a_uio; 653 ioflag = ap->a_ioflag; 654 if (ap->a_ioflag & IO_EXT) 655 #ifdef notyet 656 return (ffs_extread(vp, uio, ioflag)); 657 #else 658 panic("ffs_read+IO_EXT"); 659 #endif 660 #ifdef DIRECTIO 661 if ((ioflag & IO_DIRECT) != 0) { 662 int workdone; 663 664 error = ffs_rawread(vp, uio, &workdone); 665 if (error != 0 || workdone != 0) 666 return error; 667 } 668 #endif 669 670 seqcount = ap->a_ioflag >> IO_SEQSHIFT; 671 ip = VTOI(vp); 672 673 #ifdef INVARIANTS 674 if (uio->uio_rw != UIO_READ) 675 panic("ffs_read: mode"); 676 677 if (vp->v_type == VLNK) { 678 if ((int)ip->i_size < VFSTOUFS(vp->v_mount)->um_maxsymlinklen) 679 panic("ffs_read: short symlink"); 680 } else if (vp->v_type != VREG && vp->v_type != VDIR) 681 panic("ffs_read: type %d", vp->v_type); 682 #endif 683 orig_resid = uio->uio_resid; 684 KASSERT(orig_resid >= 0, ("ffs_read: uio->uio_resid < 0")); 685 if (orig_resid == 0) 686 return (0); 687 KASSERT(uio->uio_offset >= 0, ("ffs_read: uio->uio_offset < 0")); 688 fs = ITOFS(ip); 689 if (uio->uio_offset < ip->i_size && 690 uio->uio_offset >= fs->fs_maxfilesize) 691 return (EOVERFLOW); 692 693 bflag = GB_UNMAPPED | (uio->uio_segflg == UIO_NOCOPY ? 0 : GB_NOSPARSE); 694 for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) { 695 if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0) 696 break; 697 lbn = lblkno(fs, uio->uio_offset); 698 nextlbn = lbn + 1; 699 700 /* 701 * size of buffer. The buffer representing the 702 * end of the file is rounded up to the size of 703 * the block type ( fragment or full block, 704 * depending ). 705 */ 706 size = blksize(fs, ip, lbn); 707 blkoffset = blkoff(fs, uio->uio_offset); 708 709 /* 710 * The amount we want to transfer in this iteration is 711 * one FS block less the amount of the data before 712 * our startpoint (duh!) 713 */ 714 xfersize = fs->fs_bsize - blkoffset; 715 716 /* 717 * But if we actually want less than the block, 718 * or the file doesn't have a whole block more of data, 719 * then use the lesser number. 720 */ 721 if (uio->uio_resid < xfersize) 722 xfersize = uio->uio_resid; 723 if (bytesinfile < xfersize) 724 xfersize = bytesinfile; 725 726 if (lblktosize(fs, nextlbn) >= ip->i_size) { 727 /* 728 * Don't do readahead if this is the end of the file. 729 */ 730 error = bread_gb(vp, lbn, size, NOCRED, bflag, &bp); 731 } else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) { 732 /* 733 * Otherwise if we are allowed to cluster, 734 * grab as much as we can. 735 * 736 * XXX This may not be a win if we are not 737 * doing sequential access. 738 */ 739 error = cluster_read(vp, ip->i_size, lbn, 740 size, NOCRED, blkoffset + uio->uio_resid, 741 seqcount, bflag, &bp); 742 } else if (seqcount > 1) { 743 /* 744 * If we are NOT allowed to cluster, then 745 * if we appear to be acting sequentially, 746 * fire off a request for a readahead 747 * as well as a read. Note that the 4th and 5th 748 * arguments point to arrays of the size specified in 749 * the 6th argument. 750 */ 751 u_int nextsize = blksize(fs, ip, nextlbn); 752 error = breadn_flags(vp, lbn, lbn, size, &nextlbn, 753 &nextsize, 1, NOCRED, bflag, NULL, &bp); 754 } else { 755 /* 756 * Failing all of the above, just read what the 757 * user asked for. Interestingly, the same as 758 * the first option above. 759 */ 760 error = bread_gb(vp, lbn, size, NOCRED, bflag, &bp); 761 } 762 if (error == EJUSTRETURN) { 763 error = ffs_read_hole(uio, xfersize, &size); 764 if (error == 0) 765 continue; 766 } 767 if (error != 0) { 768 brelse(bp); 769 bp = NULL; 770 break; 771 } 772 773 /* 774 * We should only get non-zero b_resid when an I/O error 775 * has occurred, which should cause us to break above. 776 * However, if the short read did not cause an error, 777 * then we want to ensure that we do not uiomove bad 778 * or uninitialized data. 779 */ 780 size -= bp->b_resid; 781 if (size < xfersize) { 782 if (size == 0) 783 break; 784 xfersize = size; 785 } 786 787 if (buf_mapped(bp)) { 788 error = vn_io_fault_uiomove((char *)bp->b_data + 789 blkoffset, (int)xfersize, uio); 790 } else { 791 error = vn_io_fault_pgmove(bp->b_pages, blkoffset, 792 (int)xfersize, uio); 793 } 794 if (error) 795 break; 796 797 vfs_bio_brelse(bp, ioflag); 798 } 799 800 /* 801 * This can only happen in the case of an error 802 * because the loop above resets bp to NULL on each iteration 803 * and on normal completion has not set a new value into it. 804 * so it must have come from a 'break' statement 805 */ 806 if (bp != NULL) 807 vfs_bio_brelse(bp, ioflag); 808 809 if ((error == 0 || uio->uio_resid != orig_resid) && 810 (vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) 811 UFS_INODE_SET_FLAG_SHARED(ip, IN_ACCESS); 812 return (error); 813 } 814 815 /* 816 * Vnode op for writing. 817 */ 818 static int 819 ffs_write(ap) 820 struct vop_write_args /* { 821 struct vnode *a_vp; 822 struct uio *a_uio; 823 int a_ioflag; 824 struct ucred *a_cred; 825 } */ *ap; 826 { 827 struct vnode *vp; 828 struct uio *uio; 829 struct inode *ip; 830 struct fs *fs; 831 struct buf *bp; 832 ufs_lbn_t lbn; 833 off_t osize; 834 ssize_t resid; 835 int seqcount; 836 int blkoffset, error, flags, ioflag, size, xfersize; 837 838 vp = ap->a_vp; 839 if (DOINGSUJ(vp)) 840 softdep_prealloc(vp, MNT_WAIT); 841 if (vp->v_data == NULL) 842 return (EBADF); 843 844 uio = ap->a_uio; 845 ioflag = ap->a_ioflag; 846 if (ap->a_ioflag & IO_EXT) 847 #ifdef notyet 848 return (ffs_extwrite(vp, uio, ioflag, ap->a_cred)); 849 #else 850 panic("ffs_write+IO_EXT"); 851 #endif 852 853 seqcount = ap->a_ioflag >> IO_SEQSHIFT; 854 ip = VTOI(vp); 855 856 #ifdef INVARIANTS 857 if (uio->uio_rw != UIO_WRITE) 858 panic("ffs_write: mode"); 859 #endif 860 861 switch (vp->v_type) { 862 case VREG: 863 if (ioflag & IO_APPEND) 864 uio->uio_offset = ip->i_size; 865 if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size) 866 return (EPERM); 867 /* FALLTHROUGH */ 868 case VLNK: 869 break; 870 case VDIR: 871 panic("ffs_write: dir write"); 872 break; 873 default: 874 panic("ffs_write: type %p %d (%d,%d)", vp, (int)vp->v_type, 875 (int)uio->uio_offset, 876 (int)uio->uio_resid 877 ); 878 } 879 880 KASSERT(uio->uio_resid >= 0, ("ffs_write: uio->uio_resid < 0")); 881 KASSERT(uio->uio_offset >= 0, ("ffs_write: uio->uio_offset < 0")); 882 fs = ITOFS(ip); 883 if ((uoff_t)uio->uio_offset + uio->uio_resid > fs->fs_maxfilesize) 884 return (EFBIG); 885 /* 886 * Maybe this should be above the vnode op call, but so long as 887 * file servers have no limits, I don't think it matters. 888 */ 889 if (vn_rlimit_fsize(vp, uio, uio->uio_td)) 890 return (EFBIG); 891 892 resid = uio->uio_resid; 893 osize = ip->i_size; 894 if (seqcount > BA_SEQMAX) 895 flags = BA_SEQMAX << BA_SEQSHIFT; 896 else 897 flags = seqcount << BA_SEQSHIFT; 898 if (ioflag & IO_SYNC) 899 flags |= IO_SYNC; 900 flags |= BA_UNMAPPED; 901 902 for (error = 0; uio->uio_resid > 0;) { 903 lbn = lblkno(fs, uio->uio_offset); 904 blkoffset = blkoff(fs, uio->uio_offset); 905 xfersize = fs->fs_bsize - blkoffset; 906 if (uio->uio_resid < xfersize) 907 xfersize = uio->uio_resid; 908 if (uio->uio_offset + xfersize > ip->i_size) 909 vnode_pager_setsize(vp, uio->uio_offset + xfersize); 910 911 /* 912 * We must perform a read-before-write if the transfer size 913 * does not cover the entire buffer. 914 */ 915 if (fs->fs_bsize > xfersize) 916 flags |= BA_CLRBUF; 917 else 918 flags &= ~BA_CLRBUF; 919 /* XXX is uio->uio_offset the right thing here? */ 920 error = UFS_BALLOC(vp, uio->uio_offset, xfersize, 921 ap->a_cred, flags, &bp); 922 if (error != 0) { 923 vnode_pager_setsize(vp, ip->i_size); 924 break; 925 } 926 if ((ioflag & (IO_SYNC|IO_INVAL)) == (IO_SYNC|IO_INVAL)) 927 bp->b_flags |= B_NOCACHE; 928 929 if (uio->uio_offset + xfersize > ip->i_size) { 930 ip->i_size = uio->uio_offset + xfersize; 931 DIP_SET(ip, i_size, ip->i_size); 932 UFS_INODE_SET_FLAG(ip, IN_SIZEMOD | IN_CHANGE); 933 } 934 935 size = blksize(fs, ip, lbn) - bp->b_resid; 936 if (size < xfersize) 937 xfersize = size; 938 939 if (buf_mapped(bp)) { 940 error = vn_io_fault_uiomove((char *)bp->b_data + 941 blkoffset, (int)xfersize, uio); 942 } else { 943 error = vn_io_fault_pgmove(bp->b_pages, blkoffset, 944 (int)xfersize, uio); 945 } 946 /* 947 * If the buffer is not already filled and we encounter an 948 * error while trying to fill it, we have to clear out any 949 * garbage data from the pages instantiated for the buffer. 950 * If we do not, a failed uiomove() during a write can leave 951 * the prior contents of the pages exposed to a userland mmap. 952 * 953 * Note that we need only clear buffers with a transfer size 954 * equal to the block size because buffers with a shorter 955 * transfer size were cleared above by the call to UFS_BALLOC() 956 * with the BA_CLRBUF flag set. 957 * 958 * If the source region for uiomove identically mmaps the 959 * buffer, uiomove() performed the NOP copy, and the buffer 960 * content remains valid because the page fault handler 961 * validated the pages. 962 */ 963 if (error != 0 && (bp->b_flags & B_CACHE) == 0 && 964 fs->fs_bsize == xfersize) 965 vfs_bio_clrbuf(bp); 966 967 vfs_bio_set_flags(bp, ioflag); 968 969 /* 970 * If IO_SYNC each buffer is written synchronously. Otherwise 971 * if we have a severe page deficiency write the buffer 972 * asynchronously. Otherwise try to cluster, and if that 973 * doesn't do it then either do an async write (if O_DIRECT), 974 * or a delayed write (if not). 975 */ 976 if (ioflag & IO_SYNC) { 977 (void)bwrite(bp); 978 } else if (vm_page_count_severe() || 979 buf_dirty_count_severe() || 980 (ioflag & IO_ASYNC)) { 981 bp->b_flags |= B_CLUSTEROK; 982 bawrite(bp); 983 } else if (xfersize + blkoffset == fs->fs_bsize) { 984 if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) { 985 bp->b_flags |= B_CLUSTEROK; 986 cluster_write(vp, &ip->i_clusterw, bp, 987 ip->i_size, seqcount, GB_UNMAPPED); 988 } else { 989 bawrite(bp); 990 } 991 } else if (ioflag & IO_DIRECT) { 992 bp->b_flags |= B_CLUSTEROK; 993 bawrite(bp); 994 } else { 995 bp->b_flags |= B_CLUSTEROK; 996 bdwrite(bp); 997 } 998 if (error || xfersize == 0) 999 break; 1000 UFS_INODE_SET_FLAG(ip, IN_CHANGE | IN_UPDATE); 1001 } 1002 /* 1003 * If we successfully wrote any data, and we are not the superuser 1004 * we clear the setuid and setgid bits as a precaution against 1005 * tampering. 1006 */ 1007 if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid && 1008 ap->a_cred) { 1009 if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID)) { 1010 vn_seqc_write_begin(vp); 1011 UFS_INODE_SET_MODE(ip, ip->i_mode & ~(ISUID | ISGID)); 1012 DIP_SET(ip, i_mode, ip->i_mode); 1013 vn_seqc_write_end(vp); 1014 } 1015 } 1016 if (error) { 1017 if (ioflag & IO_UNIT) { 1018 (void)ffs_truncate(vp, osize, 1019 IO_NORMAL | (ioflag & IO_SYNC), ap->a_cred); 1020 uio->uio_offset -= resid - uio->uio_resid; 1021 uio->uio_resid = resid; 1022 } 1023 } else if (resid > uio->uio_resid && (ioflag & IO_SYNC)) { 1024 if (!(ioflag & IO_DATASYNC) || 1025 (ip->i_flags & (IN_SIZEMOD | IN_IBLKDATA))) 1026 error = ffs_update(vp, 1); 1027 if (ffs_fsfail_cleanup(VFSTOUFS(vp->v_mount), error)) 1028 error = ENXIO; 1029 } 1030 return (error); 1031 } 1032 1033 /* 1034 * Extended attribute area reading. 1035 */ 1036 static int 1037 ffs_extread(struct vnode *vp, struct uio *uio, int ioflag) 1038 { 1039 struct inode *ip; 1040 struct ufs2_dinode *dp; 1041 struct fs *fs; 1042 struct buf *bp; 1043 ufs_lbn_t lbn, nextlbn; 1044 off_t bytesinfile; 1045 long size, xfersize, blkoffset; 1046 ssize_t orig_resid; 1047 int error; 1048 1049 ip = VTOI(vp); 1050 fs = ITOFS(ip); 1051 dp = ip->i_din2; 1052 1053 #ifdef INVARIANTS 1054 if (uio->uio_rw != UIO_READ || fs->fs_magic != FS_UFS2_MAGIC) 1055 panic("ffs_extread: mode"); 1056 1057 #endif 1058 orig_resid = uio->uio_resid; 1059 KASSERT(orig_resid >= 0, ("ffs_extread: uio->uio_resid < 0")); 1060 if (orig_resid == 0) 1061 return (0); 1062 KASSERT(uio->uio_offset >= 0, ("ffs_extread: uio->uio_offset < 0")); 1063 1064 for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) { 1065 if ((bytesinfile = dp->di_extsize - uio->uio_offset) <= 0) 1066 break; 1067 lbn = lblkno(fs, uio->uio_offset); 1068 nextlbn = lbn + 1; 1069 1070 /* 1071 * size of buffer. The buffer representing the 1072 * end of the file is rounded up to the size of 1073 * the block type ( fragment or full block, 1074 * depending ). 1075 */ 1076 size = sblksize(fs, dp->di_extsize, lbn); 1077 blkoffset = blkoff(fs, uio->uio_offset); 1078 1079 /* 1080 * The amount we want to transfer in this iteration is 1081 * one FS block less the amount of the data before 1082 * our startpoint (duh!) 1083 */ 1084 xfersize = fs->fs_bsize - blkoffset; 1085 1086 /* 1087 * But if we actually want less than the block, 1088 * or the file doesn't have a whole block more of data, 1089 * then use the lesser number. 1090 */ 1091 if (uio->uio_resid < xfersize) 1092 xfersize = uio->uio_resid; 1093 if (bytesinfile < xfersize) 1094 xfersize = bytesinfile; 1095 1096 if (lblktosize(fs, nextlbn) >= dp->di_extsize) { 1097 /* 1098 * Don't do readahead if this is the end of the info. 1099 */ 1100 error = bread(vp, -1 - lbn, size, NOCRED, &bp); 1101 } else { 1102 /* 1103 * If we have a second block, then 1104 * fire off a request for a readahead 1105 * as well as a read. Note that the 4th and 5th 1106 * arguments point to arrays of the size specified in 1107 * the 6th argument. 1108 */ 1109 u_int nextsize = sblksize(fs, dp->di_extsize, nextlbn); 1110 1111 nextlbn = -1 - nextlbn; 1112 error = breadn(vp, -1 - lbn, 1113 size, &nextlbn, &nextsize, 1, NOCRED, &bp); 1114 } 1115 if (error) { 1116 brelse(bp); 1117 bp = NULL; 1118 break; 1119 } 1120 1121 /* 1122 * We should only get non-zero b_resid when an I/O error 1123 * has occurred, which should cause us to break above. 1124 * However, if the short read did not cause an error, 1125 * then we want to ensure that we do not uiomove bad 1126 * or uninitialized data. 1127 */ 1128 size -= bp->b_resid; 1129 if (size < xfersize) { 1130 if (size == 0) 1131 break; 1132 xfersize = size; 1133 } 1134 1135 error = uiomove((char *)bp->b_data + blkoffset, 1136 (int)xfersize, uio); 1137 if (error) 1138 break; 1139 vfs_bio_brelse(bp, ioflag); 1140 } 1141 1142 /* 1143 * This can only happen in the case of an error 1144 * because the loop above resets bp to NULL on each iteration 1145 * and on normal completion has not set a new value into it. 1146 * so it must have come from a 'break' statement 1147 */ 1148 if (bp != NULL) 1149 vfs_bio_brelse(bp, ioflag); 1150 return (error); 1151 } 1152 1153 /* 1154 * Extended attribute area writing. 1155 */ 1156 static int 1157 ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag, struct ucred *ucred) 1158 { 1159 struct inode *ip; 1160 struct ufs2_dinode *dp; 1161 struct fs *fs; 1162 struct buf *bp; 1163 ufs_lbn_t lbn; 1164 off_t osize; 1165 ssize_t resid; 1166 int blkoffset, error, flags, size, xfersize; 1167 1168 ip = VTOI(vp); 1169 fs = ITOFS(ip); 1170 dp = ip->i_din2; 1171 1172 #ifdef INVARIANTS 1173 if (uio->uio_rw != UIO_WRITE || fs->fs_magic != FS_UFS2_MAGIC) 1174 panic("ffs_extwrite: mode"); 1175 #endif 1176 1177 if (ioflag & IO_APPEND) 1178 uio->uio_offset = dp->di_extsize; 1179 KASSERT(uio->uio_offset >= 0, ("ffs_extwrite: uio->uio_offset < 0")); 1180 KASSERT(uio->uio_resid >= 0, ("ffs_extwrite: uio->uio_resid < 0")); 1181 if ((uoff_t)uio->uio_offset + uio->uio_resid > 1182 UFS_NXADDR * fs->fs_bsize) 1183 return (EFBIG); 1184 1185 resid = uio->uio_resid; 1186 osize = dp->di_extsize; 1187 flags = IO_EXT; 1188 if (ioflag & IO_SYNC) 1189 flags |= IO_SYNC; 1190 1191 for (error = 0; uio->uio_resid > 0;) { 1192 lbn = lblkno(fs, uio->uio_offset); 1193 blkoffset = blkoff(fs, uio->uio_offset); 1194 xfersize = fs->fs_bsize - blkoffset; 1195 if (uio->uio_resid < xfersize) 1196 xfersize = uio->uio_resid; 1197 1198 /* 1199 * We must perform a read-before-write if the transfer size 1200 * does not cover the entire buffer. 1201 */ 1202 if (fs->fs_bsize > xfersize) 1203 flags |= BA_CLRBUF; 1204 else 1205 flags &= ~BA_CLRBUF; 1206 error = UFS_BALLOC(vp, uio->uio_offset, xfersize, 1207 ucred, flags, &bp); 1208 if (error != 0) 1209 break; 1210 /* 1211 * If the buffer is not valid we have to clear out any 1212 * garbage data from the pages instantiated for the buffer. 1213 * If we do not, a failed uiomove() during a write can leave 1214 * the prior contents of the pages exposed to a userland 1215 * mmap(). XXX deal with uiomove() errors a better way. 1216 */ 1217 if ((bp->b_flags & B_CACHE) == 0 && fs->fs_bsize <= xfersize) 1218 vfs_bio_clrbuf(bp); 1219 1220 if (uio->uio_offset + xfersize > dp->di_extsize) { 1221 dp->di_extsize = uio->uio_offset + xfersize; 1222 UFS_INODE_SET_FLAG(ip, IN_SIZEMOD | IN_CHANGE); 1223 } 1224 1225 size = sblksize(fs, dp->di_extsize, lbn) - bp->b_resid; 1226 if (size < xfersize) 1227 xfersize = size; 1228 1229 error = 1230 uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio); 1231 1232 vfs_bio_set_flags(bp, ioflag); 1233 1234 /* 1235 * If IO_SYNC each buffer is written synchronously. Otherwise 1236 * if we have a severe page deficiency write the buffer 1237 * asynchronously. Otherwise try to cluster, and if that 1238 * doesn't do it then either do an async write (if O_DIRECT), 1239 * or a delayed write (if not). 1240 */ 1241 if (ioflag & IO_SYNC) { 1242 (void)bwrite(bp); 1243 } else if (vm_page_count_severe() || 1244 buf_dirty_count_severe() || 1245 xfersize + blkoffset == fs->fs_bsize || 1246 (ioflag & (IO_ASYNC | IO_DIRECT))) 1247 bawrite(bp); 1248 else 1249 bdwrite(bp); 1250 if (error || xfersize == 0) 1251 break; 1252 UFS_INODE_SET_FLAG(ip, IN_CHANGE); 1253 } 1254 /* 1255 * If we successfully wrote any data, and we are not the superuser 1256 * we clear the setuid and setgid bits as a precaution against 1257 * tampering. 1258 */ 1259 if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid && ucred) { 1260 if (priv_check_cred(ucred, PRIV_VFS_RETAINSUGID)) { 1261 vn_seqc_write_begin(vp); 1262 UFS_INODE_SET_MODE(ip, ip->i_mode & ~(ISUID | ISGID)); 1263 dp->di_mode = ip->i_mode; 1264 vn_seqc_write_end(vp); 1265 } 1266 } 1267 if (error) { 1268 if (ioflag & IO_UNIT) { 1269 (void)ffs_truncate(vp, osize, 1270 IO_EXT | (ioflag&IO_SYNC), ucred); 1271 uio->uio_offset -= resid - uio->uio_resid; 1272 uio->uio_resid = resid; 1273 } 1274 } else if (resid > uio->uio_resid && (ioflag & IO_SYNC)) 1275 error = ffs_update(vp, 1); 1276 return (error); 1277 } 1278 1279 /* 1280 * Vnode operating to retrieve a named extended attribute. 1281 * 1282 * Locate a particular EA (nspace:name) in the area (ptr:length), and return 1283 * the length of the EA, and possibly the pointer to the entry and to the data. 1284 */ 1285 static int 1286 ffs_findextattr(u_char *ptr, u_int length, int nspace, const char *name, 1287 struct extattr **eapp, u_char **eac) 1288 { 1289 struct extattr *eap, *eaend; 1290 size_t nlen; 1291 1292 nlen = strlen(name); 1293 KASSERT(ALIGNED_TO(ptr, struct extattr), ("unaligned")); 1294 eap = (struct extattr *)ptr; 1295 eaend = (struct extattr *)(ptr + length); 1296 for (; eap < eaend; eap = EXTATTR_NEXT(eap)) { 1297 KASSERT(EXTATTR_NEXT(eap) <= eaend, 1298 ("extattr next %p beyond %p", EXTATTR_NEXT(eap), eaend)); 1299 if (eap->ea_namespace != nspace || eap->ea_namelength != nlen 1300 || memcmp(eap->ea_name, name, nlen) != 0) 1301 continue; 1302 if (eapp != NULL) 1303 *eapp = eap; 1304 if (eac != NULL) 1305 *eac = EXTATTR_CONTENT(eap); 1306 return (EXTATTR_CONTENT_SIZE(eap)); 1307 } 1308 return (-1); 1309 } 1310 1311 static int 1312 ffs_rdextattr(u_char **p, struct vnode *vp, struct thread *td) 1313 { 1314 const struct extattr *eap, *eaend, *eapnext; 1315 struct inode *ip; 1316 struct ufs2_dinode *dp; 1317 struct fs *fs; 1318 struct uio luio; 1319 struct iovec liovec; 1320 u_int easize; 1321 int error; 1322 u_char *eae; 1323 1324 ip = VTOI(vp); 1325 fs = ITOFS(ip); 1326 dp = ip->i_din2; 1327 easize = dp->di_extsize; 1328 if ((uoff_t)easize > UFS_NXADDR * fs->fs_bsize) 1329 return (EFBIG); 1330 1331 eae = malloc(easize, M_TEMP, M_WAITOK); 1332 1333 liovec.iov_base = eae; 1334 liovec.iov_len = easize; 1335 luio.uio_iov = &liovec; 1336 luio.uio_iovcnt = 1; 1337 luio.uio_offset = 0; 1338 luio.uio_resid = easize; 1339 luio.uio_segflg = UIO_SYSSPACE; 1340 luio.uio_rw = UIO_READ; 1341 luio.uio_td = td; 1342 1343 error = ffs_extread(vp, &luio, IO_EXT | IO_SYNC); 1344 if (error) { 1345 free(eae, M_TEMP); 1346 return (error); 1347 } 1348 /* Validate disk xattrfile contents. */ 1349 for (eap = (void *)eae, eaend = (void *)(eae + easize); eap < eaend; 1350 eap = eapnext) { 1351 /* Detect zeroed out tail */ 1352 if (eap->ea_length < sizeof(*eap) || eap->ea_length == 0) { 1353 easize = (const u_char *)eap - eae; 1354 break; 1355 } 1356 1357 eapnext = EXTATTR_NEXT(eap); 1358 /* Bogusly long entry. */ 1359 if (eapnext > eaend) { 1360 free(eae, M_TEMP); 1361 return (EINTEGRITY); 1362 } 1363 } 1364 ip->i_ea_len = easize; 1365 *p = eae; 1366 return (0); 1367 } 1368 1369 static void 1370 ffs_lock_ea(struct vnode *vp) 1371 { 1372 struct inode *ip; 1373 1374 ip = VTOI(vp); 1375 VI_LOCK(vp); 1376 while (ip->i_flag & IN_EA_LOCKED) { 1377 UFS_INODE_SET_FLAG(ip, IN_EA_LOCKWAIT); 1378 msleep(&ip->i_ea_refs, &vp->v_interlock, PINOD + 2, "ufs_ea", 1379 0); 1380 } 1381 UFS_INODE_SET_FLAG(ip, IN_EA_LOCKED); 1382 VI_UNLOCK(vp); 1383 } 1384 1385 static void 1386 ffs_unlock_ea(struct vnode *vp) 1387 { 1388 struct inode *ip; 1389 1390 ip = VTOI(vp); 1391 VI_LOCK(vp); 1392 if (ip->i_flag & IN_EA_LOCKWAIT) 1393 wakeup(&ip->i_ea_refs); 1394 ip->i_flag &= ~(IN_EA_LOCKED | IN_EA_LOCKWAIT); 1395 VI_UNLOCK(vp); 1396 } 1397 1398 static int 1399 ffs_open_ea(struct vnode *vp, struct ucred *cred, struct thread *td) 1400 { 1401 struct inode *ip; 1402 int error; 1403 1404 ip = VTOI(vp); 1405 1406 ffs_lock_ea(vp); 1407 if (ip->i_ea_area != NULL) { 1408 ip->i_ea_refs++; 1409 ffs_unlock_ea(vp); 1410 return (0); 1411 } 1412 error = ffs_rdextattr(&ip->i_ea_area, vp, td); 1413 if (error) { 1414 ffs_unlock_ea(vp); 1415 return (error); 1416 } 1417 ip->i_ea_error = 0; 1418 ip->i_ea_refs++; 1419 ffs_unlock_ea(vp); 1420 return (0); 1421 } 1422 1423 /* 1424 * Vnode extattr transaction commit/abort 1425 */ 1426 static int 1427 ffs_close_ea(struct vnode *vp, int commit, struct ucred *cred, struct thread *td) 1428 { 1429 struct inode *ip; 1430 struct uio luio; 1431 struct iovec *liovec; 1432 struct ufs2_dinode *dp; 1433 size_t ea_len, tlen; 1434 int error, i, lcnt; 1435 bool truncate; 1436 1437 ip = VTOI(vp); 1438 1439 ffs_lock_ea(vp); 1440 if (ip->i_ea_area == NULL) { 1441 ffs_unlock_ea(vp); 1442 return (EINVAL); 1443 } 1444 dp = ip->i_din2; 1445 error = ip->i_ea_error; 1446 truncate = false; 1447 if (commit && error == 0) { 1448 ASSERT_VOP_ELOCKED(vp, "ffs_close_ea commit"); 1449 if (cred == NOCRED) 1450 cred = vp->v_mount->mnt_cred; 1451 1452 ea_len = MAX(ip->i_ea_len, dp->di_extsize); 1453 for (lcnt = 1, tlen = ea_len - ip->i_ea_len; tlen > 0;) { 1454 tlen -= MIN(ZERO_REGION_SIZE, tlen); 1455 lcnt++; 1456 } 1457 1458 liovec = __builtin_alloca(lcnt * sizeof(struct iovec)); 1459 luio.uio_iovcnt = lcnt; 1460 1461 liovec[0].iov_base = ip->i_ea_area; 1462 liovec[0].iov_len = ip->i_ea_len; 1463 for (i = 1, tlen = ea_len - ip->i_ea_len; i < lcnt; i++) { 1464 liovec[i].iov_base = __DECONST(void *, zero_region); 1465 liovec[i].iov_len = MIN(ZERO_REGION_SIZE, tlen); 1466 tlen -= liovec[i].iov_len; 1467 } 1468 MPASS(tlen == 0); 1469 1470 luio.uio_iov = liovec; 1471 luio.uio_offset = 0; 1472 luio.uio_resid = ea_len; 1473 luio.uio_segflg = UIO_SYSSPACE; 1474 luio.uio_rw = UIO_WRITE; 1475 luio.uio_td = td; 1476 error = ffs_extwrite(vp, &luio, IO_EXT | IO_SYNC, cred); 1477 if (error == 0 && ip->i_ea_len == 0) 1478 truncate = true; 1479 } 1480 if (--ip->i_ea_refs == 0) { 1481 free(ip->i_ea_area, M_TEMP); 1482 ip->i_ea_area = NULL; 1483 ip->i_ea_len = 0; 1484 ip->i_ea_error = 0; 1485 } 1486 ffs_unlock_ea(vp); 1487 1488 if (truncate) 1489 ffs_truncate(vp, 0, IO_EXT, cred); 1490 return (error); 1491 } 1492 1493 /* 1494 * Vnode extattr strategy routine for fifos. 1495 * 1496 * We need to check for a read or write of the external attributes. 1497 * Otherwise we just fall through and do the usual thing. 1498 */ 1499 static int 1500 ffsext_strategy(struct vop_strategy_args *ap) 1501 /* 1502 struct vop_strategy_args { 1503 struct vnodeop_desc *a_desc; 1504 struct vnode *a_vp; 1505 struct buf *a_bp; 1506 }; 1507 */ 1508 { 1509 struct vnode *vp; 1510 daddr_t lbn; 1511 1512 vp = ap->a_vp; 1513 lbn = ap->a_bp->b_lblkno; 1514 if (I_IS_UFS2(VTOI(vp)) && lbn < 0 && lbn >= -UFS_NXADDR) 1515 return (VOP_STRATEGY_APV(&ufs_vnodeops, ap)); 1516 if (vp->v_type == VFIFO) 1517 return (VOP_STRATEGY_APV(&ufs_fifoops, ap)); 1518 panic("spec nodes went here"); 1519 } 1520 1521 /* 1522 * Vnode extattr transaction commit/abort 1523 */ 1524 static int 1525 ffs_openextattr(struct vop_openextattr_args *ap) 1526 /* 1527 struct vop_openextattr_args { 1528 struct vnodeop_desc *a_desc; 1529 struct vnode *a_vp; 1530 IN struct ucred *a_cred; 1531 IN struct thread *a_td; 1532 }; 1533 */ 1534 { 1535 1536 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK) 1537 return (EOPNOTSUPP); 1538 1539 return (ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td)); 1540 } 1541 1542 /* 1543 * Vnode extattr transaction commit/abort 1544 */ 1545 static int 1546 ffs_closeextattr(struct vop_closeextattr_args *ap) 1547 /* 1548 struct vop_closeextattr_args { 1549 struct vnodeop_desc *a_desc; 1550 struct vnode *a_vp; 1551 int a_commit; 1552 IN struct ucred *a_cred; 1553 IN struct thread *a_td; 1554 }; 1555 */ 1556 { 1557 struct vnode *vp; 1558 1559 vp = ap->a_vp; 1560 if (vp->v_type == VCHR || vp->v_type == VBLK) 1561 return (EOPNOTSUPP); 1562 if (ap->a_commit && (vp->v_mount->mnt_flag & MNT_RDONLY) != 0) 1563 return (EROFS); 1564 1565 if (ap->a_commit && DOINGSUJ(vp)) { 1566 ASSERT_VOP_ELOCKED(vp, "ffs_closeextattr commit"); 1567 softdep_prealloc(vp, MNT_WAIT); 1568 if (vp->v_data == NULL) 1569 return (EBADF); 1570 } 1571 return (ffs_close_ea(vp, ap->a_commit, ap->a_cred, ap->a_td)); 1572 } 1573 1574 /* 1575 * Vnode operation to remove a named attribute. 1576 */ 1577 static int 1578 ffs_deleteextattr(struct vop_deleteextattr_args *ap) 1579 /* 1580 vop_deleteextattr { 1581 IN struct vnode *a_vp; 1582 IN int a_attrnamespace; 1583 IN const char *a_name; 1584 IN struct ucred *a_cred; 1585 IN struct thread *a_td; 1586 }; 1587 */ 1588 { 1589 struct vnode *vp; 1590 struct inode *ip; 1591 struct extattr *eap; 1592 uint32_t ul; 1593 int olen, error, i, easize; 1594 u_char *eae; 1595 void *tmp; 1596 1597 vp = ap->a_vp; 1598 ip = VTOI(vp); 1599 1600 if (vp->v_type == VCHR || vp->v_type == VBLK) 1601 return (EOPNOTSUPP); 1602 if (strlen(ap->a_name) == 0) 1603 return (EINVAL); 1604 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1605 return (EROFS); 1606 1607 error = extattr_check_cred(vp, ap->a_attrnamespace, 1608 ap->a_cred, ap->a_td, VWRITE); 1609 if (error) { 1610 /* 1611 * ffs_lock_ea is not needed there, because the vnode 1612 * must be exclusively locked. 1613 */ 1614 if (ip->i_ea_area != NULL && ip->i_ea_error == 0) 1615 ip->i_ea_error = error; 1616 return (error); 1617 } 1618 1619 if (DOINGSUJ(vp)) { 1620 ASSERT_VOP_ELOCKED(vp, "ffs_deleteextattr"); 1621 softdep_prealloc(vp, MNT_WAIT); 1622 if (vp->v_data == NULL) 1623 return (EBADF); 1624 } 1625 1626 error = ffs_open_ea(vp, ap->a_cred, ap->a_td); 1627 if (error) 1628 return (error); 1629 1630 /* CEM: delete could be done in-place instead */ 1631 eae = malloc(ip->i_ea_len, M_TEMP, M_WAITOK); 1632 bcopy(ip->i_ea_area, eae, ip->i_ea_len); 1633 easize = ip->i_ea_len; 1634 1635 olen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name, 1636 &eap, NULL); 1637 if (olen == -1) { 1638 /* delete but nonexistent */ 1639 free(eae, M_TEMP); 1640 ffs_close_ea(vp, 0, ap->a_cred, ap->a_td); 1641 return (ENOATTR); 1642 } 1643 ul = eap->ea_length; 1644 i = (u_char *)EXTATTR_NEXT(eap) - eae; 1645 bcopy(EXTATTR_NEXT(eap), eap, easize - i); 1646 easize -= ul; 1647 1648 tmp = ip->i_ea_area; 1649 ip->i_ea_area = eae; 1650 ip->i_ea_len = easize; 1651 free(tmp, M_TEMP); 1652 error = ffs_close_ea(vp, 1, ap->a_cred, ap->a_td); 1653 return (error); 1654 } 1655 1656 /* 1657 * Vnode operation to retrieve a named extended attribute. 1658 */ 1659 static int 1660 ffs_getextattr(struct vop_getextattr_args *ap) 1661 /* 1662 vop_getextattr { 1663 IN struct vnode *a_vp; 1664 IN int a_attrnamespace; 1665 IN const char *a_name; 1666 INOUT struct uio *a_uio; 1667 OUT size_t *a_size; 1668 IN struct ucred *a_cred; 1669 IN struct thread *a_td; 1670 }; 1671 */ 1672 { 1673 struct inode *ip; 1674 u_char *eae, *p; 1675 unsigned easize; 1676 int error, ealen; 1677 1678 ip = VTOI(ap->a_vp); 1679 1680 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK) 1681 return (EOPNOTSUPP); 1682 1683 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 1684 ap->a_cred, ap->a_td, VREAD); 1685 if (error) 1686 return (error); 1687 1688 error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td); 1689 if (error) 1690 return (error); 1691 1692 eae = ip->i_ea_area; 1693 easize = ip->i_ea_len; 1694 1695 ealen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name, 1696 NULL, &p); 1697 if (ealen >= 0) { 1698 error = 0; 1699 if (ap->a_size != NULL) 1700 *ap->a_size = ealen; 1701 else if (ap->a_uio != NULL) 1702 error = uiomove(p, ealen, ap->a_uio); 1703 } else 1704 error = ENOATTR; 1705 1706 ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td); 1707 return (error); 1708 } 1709 1710 /* 1711 * Vnode operation to retrieve extended attributes on a vnode. 1712 */ 1713 static int 1714 ffs_listextattr(struct vop_listextattr_args *ap) 1715 /* 1716 vop_listextattr { 1717 IN struct vnode *a_vp; 1718 IN int a_attrnamespace; 1719 INOUT struct uio *a_uio; 1720 OUT size_t *a_size; 1721 IN struct ucred *a_cred; 1722 IN struct thread *a_td; 1723 }; 1724 */ 1725 { 1726 struct inode *ip; 1727 struct extattr *eap, *eaend; 1728 int error, ealen; 1729 1730 ip = VTOI(ap->a_vp); 1731 1732 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK) 1733 return (EOPNOTSUPP); 1734 1735 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 1736 ap->a_cred, ap->a_td, VREAD); 1737 if (error) 1738 return (error); 1739 1740 error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td); 1741 if (error) 1742 return (error); 1743 1744 error = 0; 1745 if (ap->a_size != NULL) 1746 *ap->a_size = 0; 1747 1748 KASSERT(ALIGNED_TO(ip->i_ea_area, struct extattr), ("unaligned")); 1749 eap = (struct extattr *)ip->i_ea_area; 1750 eaend = (struct extattr *)(ip->i_ea_area + ip->i_ea_len); 1751 for (; error == 0 && eap < eaend; eap = EXTATTR_NEXT(eap)) { 1752 KASSERT(EXTATTR_NEXT(eap) <= eaend, 1753 ("extattr next %p beyond %p", EXTATTR_NEXT(eap), eaend)); 1754 if (eap->ea_namespace != ap->a_attrnamespace) 1755 continue; 1756 1757 ealen = eap->ea_namelength; 1758 if (ap->a_size != NULL) 1759 *ap->a_size += ealen + 1; 1760 else if (ap->a_uio != NULL) 1761 error = uiomove(&eap->ea_namelength, ealen + 1, 1762 ap->a_uio); 1763 } 1764 1765 ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td); 1766 return (error); 1767 } 1768 1769 /* 1770 * Vnode operation to set a named attribute. 1771 */ 1772 static int 1773 ffs_setextattr(struct vop_setextattr_args *ap) 1774 /* 1775 vop_setextattr { 1776 IN struct vnode *a_vp; 1777 IN int a_attrnamespace; 1778 IN const char *a_name; 1779 INOUT struct uio *a_uio; 1780 IN struct ucred *a_cred; 1781 IN struct thread *a_td; 1782 }; 1783 */ 1784 { 1785 struct vnode *vp; 1786 struct inode *ip; 1787 struct fs *fs; 1788 struct extattr *eap; 1789 uint32_t ealength, ul; 1790 ssize_t ealen; 1791 int olen, eapad1, eapad2, error, i, easize; 1792 u_char *eae; 1793 void *tmp; 1794 1795 vp = ap->a_vp; 1796 ip = VTOI(vp); 1797 fs = ITOFS(ip); 1798 1799 if (vp->v_type == VCHR || vp->v_type == VBLK) 1800 return (EOPNOTSUPP); 1801 if (strlen(ap->a_name) == 0) 1802 return (EINVAL); 1803 1804 /* XXX Now unsupported API to delete EAs using NULL uio. */ 1805 if (ap->a_uio == NULL) 1806 return (EOPNOTSUPP); 1807 1808 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1809 return (EROFS); 1810 1811 ealen = ap->a_uio->uio_resid; 1812 if (ealen < 0 || ealen > lblktosize(fs, UFS_NXADDR)) 1813 return (EINVAL); 1814 1815 error = extattr_check_cred(vp, ap->a_attrnamespace, 1816 ap->a_cred, ap->a_td, VWRITE); 1817 if (error) { 1818 /* 1819 * ffs_lock_ea is not needed there, because the vnode 1820 * must be exclusively locked. 1821 */ 1822 if (ip->i_ea_area != NULL && ip->i_ea_error == 0) 1823 ip->i_ea_error = error; 1824 return (error); 1825 } 1826 1827 if (DOINGSUJ(vp)) { 1828 ASSERT_VOP_ELOCKED(vp, "ffs_deleteextattr"); 1829 softdep_prealloc(vp, MNT_WAIT); 1830 if (vp->v_data == NULL) 1831 return (EBADF); 1832 } 1833 1834 error = ffs_open_ea(vp, ap->a_cred, ap->a_td); 1835 if (error) 1836 return (error); 1837 1838 ealength = sizeof(uint32_t) + 3 + strlen(ap->a_name); 1839 eapad1 = roundup2(ealength, 8) - ealength; 1840 eapad2 = roundup2(ealen, 8) - ealen; 1841 ealength += eapad1 + ealen + eapad2; 1842 1843 /* 1844 * CEM: rewrites of the same size or smaller could be done in-place 1845 * instead. (We don't acquire any fine-grained locks in here either, 1846 * so we could also do bigger writes in-place.) 1847 */ 1848 eae = malloc(ip->i_ea_len + ealength, M_TEMP, M_WAITOK); 1849 bcopy(ip->i_ea_area, eae, ip->i_ea_len); 1850 easize = ip->i_ea_len; 1851 1852 olen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name, 1853 &eap, NULL); 1854 if (olen == -1) { 1855 /* new, append at end */ 1856 KASSERT(ALIGNED_TO(eae + easize, struct extattr), 1857 ("unaligned")); 1858 eap = (struct extattr *)(eae + easize); 1859 easize += ealength; 1860 } else { 1861 ul = eap->ea_length; 1862 i = (u_char *)EXTATTR_NEXT(eap) - eae; 1863 if (ul != ealength) { 1864 bcopy(EXTATTR_NEXT(eap), (u_char *)eap + ealength, 1865 easize - i); 1866 easize += (ealength - ul); 1867 } 1868 } 1869 if (easize > lblktosize(fs, UFS_NXADDR)) { 1870 free(eae, M_TEMP); 1871 ffs_close_ea(vp, 0, ap->a_cred, ap->a_td); 1872 if (ip->i_ea_area != NULL && ip->i_ea_error == 0) 1873 ip->i_ea_error = ENOSPC; 1874 return (ENOSPC); 1875 } 1876 eap->ea_length = ealength; 1877 eap->ea_namespace = ap->a_attrnamespace; 1878 eap->ea_contentpadlen = eapad2; 1879 eap->ea_namelength = strlen(ap->a_name); 1880 memcpy(eap->ea_name, ap->a_name, strlen(ap->a_name)); 1881 bzero(&eap->ea_name[strlen(ap->a_name)], eapad1); 1882 error = uiomove(EXTATTR_CONTENT(eap), ealen, ap->a_uio); 1883 if (error) { 1884 free(eae, M_TEMP); 1885 ffs_close_ea(vp, 0, ap->a_cred, ap->a_td); 1886 if (ip->i_ea_area != NULL && ip->i_ea_error == 0) 1887 ip->i_ea_error = error; 1888 return (error); 1889 } 1890 bzero((u_char *)EXTATTR_CONTENT(eap) + ealen, eapad2); 1891 1892 tmp = ip->i_ea_area; 1893 ip->i_ea_area = eae; 1894 ip->i_ea_len = easize; 1895 free(tmp, M_TEMP); 1896 error = ffs_close_ea(vp, 1, ap->a_cred, ap->a_td); 1897 return (error); 1898 } 1899 1900 /* 1901 * Vnode pointer to File handle 1902 */ 1903 static int 1904 ffs_vptofh(struct vop_vptofh_args *ap) 1905 /* 1906 vop_vptofh { 1907 IN struct vnode *a_vp; 1908 IN struct fid *a_fhp; 1909 }; 1910 */ 1911 { 1912 struct inode *ip; 1913 struct ufid *ufhp; 1914 1915 ip = VTOI(ap->a_vp); 1916 ufhp = (struct ufid *)ap->a_fhp; 1917 ufhp->ufid_len = sizeof(struct ufid); 1918 ufhp->ufid_ino = ip->i_number; 1919 ufhp->ufid_gen = ip->i_gen; 1920 return (0); 1921 } 1922 1923 SYSCTL_DECL(_vfs_ffs); 1924 static int use_buf_pager = 1; 1925 SYSCTL_INT(_vfs_ffs, OID_AUTO, use_buf_pager, CTLFLAG_RWTUN, &use_buf_pager, 0, 1926 "Always use buffer pager instead of bmap"); 1927 1928 static daddr_t 1929 ffs_gbp_getblkno(struct vnode *vp, vm_ooffset_t off) 1930 { 1931 1932 return (lblkno(VFSTOUFS(vp->v_mount)->um_fs, off)); 1933 } 1934 1935 static int 1936 ffs_gbp_getblksz(struct vnode *vp, daddr_t lbn, long *sz) 1937 { 1938 1939 *sz = blksize(VFSTOUFS(vp->v_mount)->um_fs, VTOI(vp), lbn); 1940 return (0); 1941 } 1942 1943 static int 1944 ffs_getpages(struct vop_getpages_args *ap) 1945 { 1946 struct vnode *vp; 1947 struct ufsmount *um; 1948 1949 vp = ap->a_vp; 1950 um = VFSTOUFS(vp->v_mount); 1951 1952 if (!use_buf_pager && um->um_devvp->v_bufobj.bo_bsize <= PAGE_SIZE) 1953 return (vnode_pager_generic_getpages(vp, ap->a_m, ap->a_count, 1954 ap->a_rbehind, ap->a_rahead, NULL, NULL)); 1955 return (vfs_bio_getpages(vp, ap->a_m, ap->a_count, ap->a_rbehind, 1956 ap->a_rahead, ffs_gbp_getblkno, ffs_gbp_getblksz)); 1957 } 1958 1959 static int 1960 ffs_getpages_async(struct vop_getpages_async_args *ap) 1961 { 1962 struct vnode *vp; 1963 struct ufsmount *um; 1964 bool do_iodone; 1965 int error; 1966 1967 vp = ap->a_vp; 1968 um = VFSTOUFS(vp->v_mount); 1969 do_iodone = true; 1970 1971 if (um->um_devvp->v_bufobj.bo_bsize <= PAGE_SIZE) { 1972 error = vnode_pager_generic_getpages(vp, ap->a_m, ap->a_count, 1973 ap->a_rbehind, ap->a_rahead, ap->a_iodone, ap->a_arg); 1974 if (error == 0) 1975 do_iodone = false; 1976 } else { 1977 error = vfs_bio_getpages(vp, ap->a_m, ap->a_count, 1978 ap->a_rbehind, ap->a_rahead, ffs_gbp_getblkno, 1979 ffs_gbp_getblksz); 1980 } 1981 if (do_iodone && ap->a_iodone != NULL) 1982 ap->a_iodone(ap->a_arg, ap->a_m, ap->a_count, error); 1983 1984 return (error); 1985 } 1986 1987 static int 1988 ffs_vput_pair(struct vop_vput_pair_args *ap) 1989 { 1990 struct mount *mp; 1991 struct vnode *dvp, *vp, *vp1, **vpp; 1992 struct inode *dp, *ip; 1993 ino_t ip_ino; 1994 u_int64_t ip_gen; 1995 int error, vp_locked; 1996 1997 dvp = ap->a_dvp; 1998 dp = VTOI(dvp); 1999 vpp = ap->a_vpp; 2000 vp = vpp != NULL ? *vpp : NULL; 2001 2002 if ((dp->i_flag & (IN_NEEDSYNC | IN_ENDOFF)) == 0) { 2003 vput(dvp); 2004 if (vp != NULL && ap->a_unlock_vp) 2005 vput(vp); 2006 return (0); 2007 } 2008 2009 mp = dvp->v_mount; 2010 if (vp != NULL) { 2011 if (ap->a_unlock_vp) { 2012 vput(vp); 2013 } else { 2014 MPASS(vp->v_type != VNON); 2015 vp_locked = VOP_ISLOCKED(vp); 2016 ip = VTOI(vp); 2017 ip_ino = ip->i_number; 2018 ip_gen = ip->i_gen; 2019 VOP_UNLOCK(vp); 2020 } 2021 } 2022 2023 /* 2024 * If compaction or fsync was requested do it in ffs_vput_pair() 2025 * now that other locks are no longer held. 2026 */ 2027 if ((dp->i_flag & IN_ENDOFF) != 0) { 2028 VNASSERT(I_ENDOFF(dp) != 0 && I_ENDOFF(dp) < dp->i_size, dvp, 2029 ("IN_ENDOFF set but I_ENDOFF() is not")); 2030 dp->i_flag &= ~IN_ENDOFF; 2031 error = UFS_TRUNCATE(dvp, (off_t)I_ENDOFF(dp), IO_NORMAL | 2032 (DOINGASYNC(dvp) ? 0 : IO_SYNC), curthread->td_ucred); 2033 if (error != 0 && error != ERELOOKUP) { 2034 if (!ffs_fsfail_cleanup(VFSTOUFS(mp), error)) { 2035 vn_printf(dvp, 2036 "IN_ENDOFF: failed to truncate, " 2037 "error %d\n", error); 2038 } 2039 #ifdef UFS_DIRHASH 2040 ufsdirhash_free(dp); 2041 #endif 2042 } 2043 SET_I_ENDOFF(dp, 0); 2044 } 2045 if ((dp->i_flag & IN_NEEDSYNC) != 0) { 2046 do { 2047 error = ffs_syncvnode(dvp, MNT_WAIT, 0); 2048 } while (error == ERELOOKUP); 2049 } 2050 2051 vput(dvp); 2052 2053 if (vp == NULL || ap->a_unlock_vp) 2054 return (0); 2055 MPASS(mp != NULL); 2056 2057 /* 2058 * It is possible that vp is reclaimed at this point. Only 2059 * routines that call us with a_unlock_vp == false can find 2060 * that their vp has been reclaimed. There are three areas 2061 * that are affected: 2062 * 1) vn_open_cred() - later VOPs could fail, but 2063 * dead_open() returns 0 to simulate successful open. 2064 * 2) ffs_snapshot() - creation of snapshot fails with EBADF. 2065 * 3) NFS server (several places) - code is prepared to detect 2066 * and respond to dead vnodes by returning ESTALE. 2067 */ 2068 VOP_LOCK(vp, vp_locked | LK_RETRY); 2069 if (!VN_IS_DOOMED(vp)) 2070 return (0); 2071 2072 /* 2073 * Try harder to recover from reclaimed vp if reclaim was not 2074 * because underlying inode was cleared. We saved inode 2075 * number and inode generation, so we can try to reinstantiate 2076 * exactly same version of inode. If this fails, return 2077 * original doomed vnode and let caller to handle 2078 * consequences. 2079 * 2080 * Note that callers must keep write started around 2081 * VOP_VPUT_PAIR() calls, so it is safe to use mp without 2082 * busying it. 2083 */ 2084 VOP_UNLOCK(vp); 2085 error = ffs_inotovp(mp, ip_ino, ip_gen, LK_EXCLUSIVE, &vp1, 2086 FFSV_REPLACE_DOOMED); 2087 if (error != 0) { 2088 VOP_LOCK(vp, vp_locked | LK_RETRY); 2089 } else { 2090 vrele(vp); 2091 *vpp = vp1; 2092 } 2093 return (error); 2094 } 2095