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 .vop_fplookup_vexec = VOP_EAGAIN, 157 .vop_fplookup_symlink = VOP_EAGAIN, 158 }; 159 VFS_VOP_VECTOR_REGISTER(ffs_vnodeops1); 160 161 struct vop_vector ffs_fifoops1 = { 162 .vop_default = &ufs_fifoops, 163 .vop_fsync = ffs_fsync, 164 .vop_fdatasync = ffs_fdatasync, 165 .vop_lock1 = ffs_lock, 166 #ifdef INVARIANTS 167 .vop_unlock = ffs_unlock_debug, 168 #endif 169 .vop_vptofh = ffs_vptofh, 170 }; 171 VFS_VOP_VECTOR_REGISTER(ffs_fifoops1); 172 173 /* Global vfs data structures for ufs. */ 174 struct vop_vector ffs_vnodeops2 = { 175 .vop_default = &ufs_vnodeops, 176 .vop_fsync = ffs_fsync, 177 .vop_fdatasync = ffs_fdatasync, 178 .vop_getpages = ffs_getpages, 179 .vop_getpages_async = ffs_getpages_async, 180 .vop_lock1 = ffs_lock, 181 #ifdef INVARIANTS 182 .vop_unlock = ffs_unlock_debug, 183 #endif 184 .vop_read = ffs_read, 185 .vop_reallocblks = ffs_reallocblks, 186 .vop_write = ffs_write, 187 .vop_closeextattr = ffs_closeextattr, 188 .vop_deleteextattr = ffs_deleteextattr, 189 .vop_getextattr = ffs_getextattr, 190 .vop_listextattr = ffs_listextattr, 191 .vop_openextattr = ffs_openextattr, 192 .vop_setextattr = ffs_setextattr, 193 .vop_vptofh = ffs_vptofh, 194 .vop_vput_pair = ffs_vput_pair, 195 .vop_fplookup_vexec = VOP_EAGAIN, 196 .vop_fplookup_symlink = VOP_EAGAIN, 197 }; 198 VFS_VOP_VECTOR_REGISTER(ffs_vnodeops2); 199 200 struct vop_vector ffs_fifoops2 = { 201 .vop_default = &ufs_fifoops, 202 .vop_fsync = ffs_fsync, 203 .vop_fdatasync = ffs_fdatasync, 204 .vop_lock1 = ffs_lock, 205 #ifdef INVARIANTS 206 .vop_unlock = ffs_unlock_debug, 207 #endif 208 .vop_reallocblks = ffs_reallocblks, 209 .vop_strategy = ffsext_strategy, 210 .vop_closeextattr = ffs_closeextattr, 211 .vop_deleteextattr = ffs_deleteextattr, 212 .vop_getextattr = ffs_getextattr, 213 .vop_listextattr = ffs_listextattr, 214 .vop_openextattr = ffs_openextattr, 215 .vop_setextattr = ffs_setextattr, 216 .vop_vptofh = ffs_vptofh, 217 }; 218 VFS_VOP_VECTOR_REGISTER(ffs_fifoops2); 219 220 /* 221 * Synch an open file. 222 */ 223 /* ARGSUSED */ 224 static int 225 ffs_fsync(struct vop_fsync_args *ap) 226 { 227 struct vnode *vp; 228 struct bufobj *bo; 229 int error; 230 231 vp = ap->a_vp; 232 bo = &vp->v_bufobj; 233 retry: 234 error = ffs_syncvnode(vp, ap->a_waitfor, 0); 235 if (error) 236 return (error); 237 if (ap->a_waitfor == MNT_WAIT && DOINGSOFTDEP(vp)) { 238 error = softdep_fsync(vp); 239 if (error) 240 return (error); 241 242 /* 243 * The softdep_fsync() function may drop vp lock, 244 * allowing for dirty buffers to reappear on the 245 * bo_dirty list. Recheck and resync as needed. 246 */ 247 BO_LOCK(bo); 248 if ((vp->v_type == VREG || vp->v_type == VDIR) && 249 (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)) { 250 BO_UNLOCK(bo); 251 goto retry; 252 } 253 BO_UNLOCK(bo); 254 } 255 if (ffs_fsfail_cleanup(VFSTOUFS(vp->v_mount), 0)) 256 return (ENXIO); 257 return (0); 258 } 259 260 int 261 ffs_syncvnode(struct vnode *vp, int waitfor, int flags) 262 { 263 struct inode *ip; 264 struct bufobj *bo; 265 struct ufsmount *ump; 266 struct buf *bp, *nbp; 267 ufs_lbn_t lbn; 268 int error, passes, wflag; 269 bool still_dirty, unlocked, wait; 270 271 ip = VTOI(vp); 272 bo = &vp->v_bufobj; 273 ump = VFSTOUFS(vp->v_mount); 274 #ifdef WITNESS 275 wflag = IS_SNAPSHOT(ip) ? LK_NOWITNESS : 0; 276 #else 277 wflag = 0; 278 #endif 279 280 /* 281 * When doing MNT_WAIT we must first flush all dependencies 282 * on the inode. 283 */ 284 if (DOINGSOFTDEP(vp) && waitfor == MNT_WAIT && 285 (error = softdep_sync_metadata(vp)) != 0) { 286 if (ffs_fsfail_cleanup(ump, error)) 287 error = 0; 288 return (error); 289 } 290 291 /* 292 * Flush all dirty buffers associated with a vnode. 293 */ 294 error = 0; 295 passes = 0; 296 wait = false; /* Always do an async pass first. */ 297 unlocked = false; 298 lbn = lblkno(ITOFS(ip), (ip->i_size + ITOFS(ip)->fs_bsize - 1)); 299 BO_LOCK(bo); 300 loop: 301 TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs) 302 bp->b_vflags &= ~BV_SCANNED; 303 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 304 /* 305 * Reasons to skip this buffer: it has already been considered 306 * on this pass, the buffer has dependencies that will cause 307 * it to be redirtied and it has not already been deferred, 308 * or it is already being written. 309 */ 310 if ((bp->b_vflags & BV_SCANNED) != 0) 311 continue; 312 bp->b_vflags |= BV_SCANNED; 313 /* 314 * Flush indirects in order, if requested. 315 * 316 * Note that if only datasync is requested, we can 317 * skip indirect blocks when softupdates are not 318 * active. Otherwise we must flush them with data, 319 * since dependencies prevent data block writes. 320 */ 321 if (waitfor == MNT_WAIT && bp->b_lblkno <= -UFS_NDADDR && 322 (lbn_level(bp->b_lblkno) >= passes || 323 ((flags & DATA_ONLY) != 0 && !DOINGSOFTDEP(vp)))) 324 continue; 325 if (bp->b_lblkno > lbn) 326 panic("ffs_syncvnode: syncing truncated data."); 327 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL) == 0) { 328 BO_UNLOCK(bo); 329 } else if (wait) { 330 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL | 331 LK_INTERLOCK | wflag, BO_LOCKPTR(bo)) != 0) { 332 BO_LOCK(bo); 333 bp->b_vflags &= ~BV_SCANNED; 334 goto next_locked; 335 } 336 } else 337 continue; 338 if ((bp->b_flags & B_DELWRI) == 0) 339 panic("ffs_fsync: not dirty"); 340 /* 341 * Check for dependencies and potentially complete them. 342 */ 343 if (!LIST_EMPTY(&bp->b_dep) && 344 (error = softdep_sync_buf(vp, bp, 345 wait ? MNT_WAIT : MNT_NOWAIT)) != 0) { 346 /* 347 * Lock order conflict, buffer was already unlocked, 348 * and vnode possibly unlocked. 349 */ 350 if (error == ERELOOKUP) { 351 if (vp->v_data == NULL) 352 return (EBADF); 353 unlocked = true; 354 if (DOINGSOFTDEP(vp) && waitfor == MNT_WAIT && 355 (error = softdep_sync_metadata(vp)) != 0) { 356 if (ffs_fsfail_cleanup(ump, error)) 357 error = 0; 358 return (unlocked && error == 0 ? 359 ERELOOKUP : error); 360 } 361 /* Re-evaluate inode size */ 362 lbn = lblkno(ITOFS(ip), (ip->i_size + 363 ITOFS(ip)->fs_bsize - 1)); 364 goto next; 365 } 366 /* I/O error. */ 367 if (error != EBUSY) { 368 BUF_UNLOCK(bp); 369 return (error); 370 } 371 /* If we deferred once, don't defer again. */ 372 if ((bp->b_flags & B_DEFERRED) == 0) { 373 bp->b_flags |= B_DEFERRED; 374 BUF_UNLOCK(bp); 375 goto next; 376 } 377 } 378 if (wait) { 379 bremfree(bp); 380 error = bwrite(bp); 381 if (ffs_fsfail_cleanup(ump, error)) 382 error = 0; 383 if (error != 0) 384 return (error); 385 } else if ((bp->b_flags & B_CLUSTEROK)) { 386 (void) vfs_bio_awrite(bp); 387 } else { 388 bremfree(bp); 389 (void) bawrite(bp); 390 } 391 next: 392 /* 393 * Since we may have slept during the I/O, we need 394 * to start from a known point. 395 */ 396 BO_LOCK(bo); 397 next_locked: 398 nbp = TAILQ_FIRST(&bo->bo_dirty.bv_hd); 399 } 400 if (waitfor != MNT_WAIT) { 401 BO_UNLOCK(bo); 402 if ((flags & NO_INO_UPDT) != 0) 403 return (unlocked ? ERELOOKUP : 0); 404 error = ffs_update(vp, 0); 405 if (error == 0 && unlocked) 406 error = ERELOOKUP; 407 return (error); 408 } 409 /* Drain IO to see if we're done. */ 410 bufobj_wwait(bo, 0, 0); 411 /* 412 * Block devices associated with filesystems may have new I/O 413 * requests posted for them even if the vnode is locked, so no 414 * amount of trying will get them clean. We make several passes 415 * as a best effort. 416 * 417 * Regular files may need multiple passes to flush all dependency 418 * work as it is possible that we must write once per indirect 419 * level, once for the leaf, and once for the inode and each of 420 * these will be done with one sync and one async pass. 421 */ 422 if (bo->bo_dirty.bv_cnt > 0) { 423 if ((flags & DATA_ONLY) == 0) { 424 still_dirty = true; 425 } else { 426 /* 427 * For data-only sync, dirty indirect buffers 428 * are ignored. 429 */ 430 still_dirty = false; 431 TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs) { 432 if (bp->b_lblkno > -UFS_NDADDR) { 433 still_dirty = true; 434 break; 435 } 436 } 437 } 438 439 if (still_dirty) { 440 /* Write the inode after sync passes to flush deps. */ 441 if (wait && DOINGSOFTDEP(vp) && 442 (flags & NO_INO_UPDT) == 0) { 443 BO_UNLOCK(bo); 444 ffs_update(vp, 1); 445 BO_LOCK(bo); 446 } 447 /* switch between sync/async. */ 448 wait = !wait; 449 if (wait || ++passes < UFS_NIADDR + 2) 450 goto loop; 451 } 452 } 453 BO_UNLOCK(bo); 454 error = 0; 455 if ((flags & DATA_ONLY) == 0) { 456 if ((flags & NO_INO_UPDT) == 0) 457 error = ffs_update(vp, 1); 458 if (DOINGSUJ(vp)) 459 softdep_journal_fsync(VTOI(vp)); 460 } else if ((ip->i_flags & (IN_SIZEMOD | IN_IBLKDATA)) != 0) { 461 error = ffs_update(vp, 1); 462 } 463 if (error == 0 && unlocked) 464 error = ERELOOKUP; 465 if (error == 0) 466 ip->i_flag &= ~IN_NEEDSYNC; 467 return (error); 468 } 469 470 static int 471 ffs_fdatasync(struct vop_fdatasync_args *ap) 472 { 473 474 return (ffs_syncvnode(ap->a_vp, MNT_WAIT, DATA_ONLY)); 475 } 476 477 static int 478 ffs_lock( 479 struct vop_lock1_args /* { 480 struct vnode *a_vp; 481 int a_flags; 482 char *file; 483 int line; 484 } */ *ap) 485 { 486 #if !defined(NO_FFS_SNAPSHOT) || defined(DIAGNOSTIC) 487 struct vnode *vp = ap->a_vp; 488 #endif /* !NO_FFS_SNAPSHOT || DIAGNOSTIC */ 489 #ifdef DIAGNOSTIC 490 struct inode *ip; 491 #endif /* DIAGNOSTIC */ 492 int result; 493 #ifndef NO_FFS_SNAPSHOT 494 int flags; 495 struct lock *lkp; 496 497 /* 498 * Adaptive spinning mixed with SU leads to trouble. use a giant hammer 499 * and only use it when LK_NODDLKTREAT is set. Currently this means it 500 * is only used during path lookup. 501 */ 502 if ((ap->a_flags & LK_NODDLKTREAT) != 0) 503 ap->a_flags |= LK_ADAPTIVE; 504 switch (ap->a_flags & LK_TYPE_MASK) { 505 case LK_SHARED: 506 case LK_UPGRADE: 507 case LK_EXCLUSIVE: 508 flags = ap->a_flags; 509 for (;;) { 510 #ifdef DEBUG_VFS_LOCKS 511 VNPASS(vp->v_holdcnt != 0, vp); 512 #endif /* DEBUG_VFS_LOCKS */ 513 lkp = vp->v_vnlock; 514 result = lockmgr_lock_flags(lkp, flags, 515 &VI_MTX(vp)->lock_object, ap->a_file, ap->a_line); 516 if (lkp == vp->v_vnlock || result != 0) 517 break; 518 /* 519 * Apparent success, except that the vnode 520 * mutated between snapshot file vnode and 521 * regular file vnode while this process 522 * slept. The lock currently held is not the 523 * right lock. Release it, and try to get the 524 * new lock. 525 */ 526 lockmgr_unlock(lkp); 527 if ((flags & (LK_INTERLOCK | LK_NOWAIT)) == 528 (LK_INTERLOCK | LK_NOWAIT)) 529 return (EBUSY); 530 if ((flags & LK_TYPE_MASK) == LK_UPGRADE) 531 flags = (flags & ~LK_TYPE_MASK) | LK_EXCLUSIVE; 532 flags &= ~LK_INTERLOCK; 533 } 534 #ifdef DIAGNOSTIC 535 switch (ap->a_flags & LK_TYPE_MASK) { 536 case LK_UPGRADE: 537 case LK_EXCLUSIVE: 538 if (result == 0 && vp->v_vnlock->lk_recurse == 0) { 539 ip = VTOI(vp); 540 if (ip != NULL) 541 ip->i_lock_gen++; 542 } 543 } 544 #endif /* DIAGNOSTIC */ 545 break; 546 default: 547 #ifdef DIAGNOSTIC 548 if ((ap->a_flags & LK_TYPE_MASK) == LK_DOWNGRADE) { 549 ip = VTOI(vp); 550 if (ip != NULL) 551 ufs_unlock_tracker(ip); 552 } 553 #endif /* DIAGNOSTIC */ 554 result = VOP_LOCK1_APV(&ufs_vnodeops, ap); 555 break; 556 } 557 #else /* NO_FFS_SNAPSHOT */ 558 /* 559 * See above for an explanation. 560 */ 561 if ((ap->a_flags & LK_NODDLKTREAT) != 0) 562 ap->a_flags |= LK_ADAPTIVE; 563 #ifdef DIAGNOSTIC 564 if ((ap->a_flags & LK_TYPE_MASK) == LK_DOWNGRADE) { 565 ip = VTOI(vp); 566 if (ip != NULL) 567 ufs_unlock_tracker(ip); 568 } 569 #endif /* DIAGNOSTIC */ 570 result = VOP_LOCK1_APV(&ufs_vnodeops, ap); 571 #endif /* NO_FFS_SNAPSHOT */ 572 #ifdef DIAGNOSTIC 573 switch (ap->a_flags & LK_TYPE_MASK) { 574 case LK_UPGRADE: 575 case LK_EXCLUSIVE: 576 if (result == 0 && vp->v_vnlock->lk_recurse == 0) { 577 ip = VTOI(vp); 578 if (ip != NULL) 579 ip->i_lock_gen++; 580 } 581 } 582 #endif /* DIAGNOSTIC */ 583 return (result); 584 } 585 586 #ifdef INVARIANTS 587 static int 588 ffs_unlock_debug(struct vop_unlock_args *ap) 589 { 590 struct vnode *vp; 591 struct inode *ip; 592 593 vp = ap->a_vp; 594 ip = VTOI(vp); 595 if (ip->i_flag & UFS_INODE_FLAG_LAZY_MASK_ASSERTABLE) { 596 if ((vp->v_mflag & VMP_LAZYLIST) == 0) { 597 VI_LOCK(vp); 598 VNASSERT((vp->v_mflag & VMP_LAZYLIST), vp, 599 ("%s: modified vnode (%x) not on lazy list", 600 __func__, ip->i_flag)); 601 VI_UNLOCK(vp); 602 } 603 } 604 KASSERT(vp->v_type != VDIR || vp->v_vnlock->lk_recurse != 0 || 605 (ip->i_flag & IN_ENDOFF) == 0, 606 ("ufs dir vp %p ip %p flags %#x", vp, ip, ip->i_flag)); 607 #ifdef DIAGNOSTIC 608 if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE && ip != NULL && 609 vp->v_vnlock->lk_recurse == 0) 610 ufs_unlock_tracker(ip); 611 #endif 612 return (VOP_UNLOCK_APV(&ufs_vnodeops, ap)); 613 } 614 #endif 615 616 static int 617 ffs_read_hole(struct uio *uio, long xfersize, long *size) 618 { 619 ssize_t saved_resid, tlen; 620 int error; 621 622 while (xfersize > 0) { 623 tlen = min(xfersize, ZERO_REGION_SIZE); 624 saved_resid = uio->uio_resid; 625 error = vn_io_fault_uiomove(__DECONST(void *, zero_region), 626 tlen, uio); 627 if (error != 0) 628 return (error); 629 tlen = saved_resid - uio->uio_resid; 630 xfersize -= tlen; 631 *size -= tlen; 632 } 633 return (0); 634 } 635 636 /* 637 * Vnode op for reading. 638 */ 639 static int 640 ffs_read( 641 struct vop_read_args /* { 642 struct vnode *a_vp; 643 struct uio *a_uio; 644 int a_ioflag; 645 struct ucred *a_cred; 646 } */ *ap) 647 { 648 struct vnode *vp; 649 struct inode *ip; 650 struct uio *uio; 651 struct fs *fs; 652 struct buf *bp; 653 ufs_lbn_t lbn, nextlbn; 654 off_t bytesinfile; 655 long size, xfersize, blkoffset; 656 ssize_t orig_resid; 657 int bflag, error, ioflag, seqcount; 658 659 vp = ap->a_vp; 660 uio = ap->a_uio; 661 ioflag = ap->a_ioflag; 662 if (ap->a_ioflag & IO_EXT) 663 #ifdef notyet 664 return (ffs_extread(vp, uio, ioflag)); 665 #else 666 panic("ffs_read+IO_EXT"); 667 #endif 668 #ifdef DIRECTIO 669 if ((ioflag & IO_DIRECT) != 0) { 670 int workdone; 671 672 error = ffs_rawread(vp, uio, &workdone); 673 if (error != 0 || workdone != 0) 674 return error; 675 } 676 #endif 677 678 seqcount = ap->a_ioflag >> IO_SEQSHIFT; 679 ip = VTOI(vp); 680 681 #ifdef INVARIANTS 682 if (uio->uio_rw != UIO_READ) 683 panic("ffs_read: mode"); 684 685 if (vp->v_type == VLNK) { 686 if ((int)ip->i_size < VFSTOUFS(vp->v_mount)->um_maxsymlinklen) 687 panic("ffs_read: short symlink"); 688 } else if (vp->v_type != VREG && vp->v_type != VDIR) 689 panic("ffs_read: type %d", vp->v_type); 690 #endif 691 orig_resid = uio->uio_resid; 692 KASSERT(orig_resid >= 0, ("ffs_read: uio->uio_resid < 0")); 693 if (orig_resid == 0) 694 return (0); 695 KASSERT(uio->uio_offset >= 0, ("ffs_read: uio->uio_offset < 0")); 696 fs = ITOFS(ip); 697 if (uio->uio_offset < ip->i_size && 698 uio->uio_offset >= fs->fs_maxfilesize) 699 return (EOVERFLOW); 700 701 bflag = GB_UNMAPPED | (uio->uio_segflg == UIO_NOCOPY ? 0 : GB_NOSPARSE); 702 #ifdef WITNESS 703 bflag |= IS_SNAPSHOT(ip) ? GB_NOWITNESS : 0; 704 #endif 705 for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) { 706 if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0) 707 break; 708 lbn = lblkno(fs, uio->uio_offset); 709 nextlbn = lbn + 1; 710 711 /* 712 * size of buffer. The buffer representing the 713 * end of the file is rounded up to the size of 714 * the block type ( fragment or full block, 715 * depending ). 716 */ 717 size = blksize(fs, ip, lbn); 718 blkoffset = blkoff(fs, uio->uio_offset); 719 720 /* 721 * The amount we want to transfer in this iteration is 722 * one FS block less the amount of the data before 723 * our startpoint (duh!) 724 */ 725 xfersize = fs->fs_bsize - blkoffset; 726 727 /* 728 * But if we actually want less than the block, 729 * or the file doesn't have a whole block more of data, 730 * then use the lesser number. 731 */ 732 if (uio->uio_resid < xfersize) 733 xfersize = uio->uio_resid; 734 if (bytesinfile < xfersize) 735 xfersize = bytesinfile; 736 737 if (lblktosize(fs, nextlbn) >= ip->i_size) { 738 /* 739 * Don't do readahead if this is the end of the file. 740 */ 741 error = bread_gb(vp, lbn, size, NOCRED, bflag, &bp); 742 } else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) { 743 /* 744 * Otherwise if we are allowed to cluster, 745 * grab as much as we can. 746 * 747 * XXX This may not be a win if we are not 748 * doing sequential access. 749 */ 750 error = cluster_read(vp, ip->i_size, lbn, 751 size, NOCRED, blkoffset + uio->uio_resid, 752 seqcount, bflag, &bp); 753 } else if (seqcount > 1) { 754 /* 755 * If we are NOT allowed to cluster, then 756 * if we appear to be acting sequentially, 757 * fire off a request for a readahead 758 * as well as a read. Note that the 4th and 5th 759 * arguments point to arrays of the size specified in 760 * the 6th argument. 761 */ 762 u_int nextsize = blksize(fs, ip, nextlbn); 763 error = breadn_flags(vp, lbn, lbn, size, &nextlbn, 764 &nextsize, 1, NOCRED, bflag, NULL, &bp); 765 } else { 766 /* 767 * Failing all of the above, just read what the 768 * user asked for. Interestingly, the same as 769 * the first option above. 770 */ 771 error = bread_gb(vp, lbn, size, NOCRED, bflag, &bp); 772 } 773 if (error == EJUSTRETURN) { 774 error = ffs_read_hole(uio, xfersize, &size); 775 if (error == 0) 776 continue; 777 } 778 if (error != 0) { 779 brelse(bp); 780 bp = NULL; 781 break; 782 } 783 784 /* 785 * We should only get non-zero b_resid when an I/O error 786 * has occurred, which should cause us to break above. 787 * However, if the short read did not cause an error, 788 * then we want to ensure that we do not uiomove bad 789 * or uninitialized data. 790 */ 791 size -= bp->b_resid; 792 if (size < xfersize) { 793 if (size == 0) 794 break; 795 xfersize = size; 796 } 797 798 if (buf_mapped(bp)) { 799 error = vn_io_fault_uiomove((char *)bp->b_data + 800 blkoffset, (int)xfersize, uio); 801 } else { 802 error = vn_io_fault_pgmove(bp->b_pages, 803 blkoffset + (bp->b_offset & PAGE_MASK), 804 (int)xfersize, uio); 805 } 806 if (error) 807 break; 808 809 vfs_bio_brelse(bp, ioflag); 810 } 811 812 /* 813 * This can only happen in the case of an error 814 * because the loop above resets bp to NULL on each iteration 815 * and on normal completion has not set a new value into it. 816 * so it must have come from a 'break' statement 817 */ 818 if (bp != NULL) 819 vfs_bio_brelse(bp, ioflag); 820 821 if ((error == 0 || uio->uio_resid != orig_resid) && 822 (vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) 823 UFS_INODE_SET_FLAG_SHARED(ip, IN_ACCESS); 824 return (error); 825 } 826 827 /* 828 * Vnode op for writing. 829 */ 830 static int 831 ffs_write( 832 struct vop_write_args /* { 833 struct vnode *a_vp; 834 struct uio *a_uio; 835 int a_ioflag; 836 struct ucred *a_cred; 837 } */ *ap) 838 { 839 struct vnode *vp; 840 struct uio *uio; 841 struct inode *ip; 842 struct fs *fs; 843 struct buf *bp; 844 ufs_lbn_t lbn; 845 off_t osize; 846 ssize_t resid, r; 847 int seqcount; 848 int blkoffset, error, flags, ioflag, size, xfersize; 849 850 vp = ap->a_vp; 851 if (DOINGSUJ(vp)) 852 softdep_prealloc(vp, MNT_WAIT); 853 if (vp->v_data == NULL) 854 return (EBADF); 855 856 uio = ap->a_uio; 857 ioflag = ap->a_ioflag; 858 if (ap->a_ioflag & IO_EXT) 859 #ifdef notyet 860 return (ffs_extwrite(vp, uio, ioflag, ap->a_cred)); 861 #else 862 panic("ffs_write+IO_EXT"); 863 #endif 864 865 seqcount = ap->a_ioflag >> IO_SEQSHIFT; 866 ip = VTOI(vp); 867 868 #ifdef INVARIANTS 869 if (uio->uio_rw != UIO_WRITE) 870 panic("ffs_write: mode"); 871 #endif 872 873 switch (vp->v_type) { 874 case VREG: 875 if (ioflag & IO_APPEND) 876 uio->uio_offset = ip->i_size; 877 if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size) 878 return (EPERM); 879 /* FALLTHROUGH */ 880 case VLNK: 881 break; 882 case VDIR: 883 panic("ffs_write: dir write"); 884 break; 885 default: 886 panic("ffs_write: type %p %d (%d,%d)", vp, (int)vp->v_type, 887 (int)uio->uio_offset, 888 (int)uio->uio_resid 889 ); 890 } 891 892 KASSERT(uio->uio_resid >= 0, ("ffs_write: uio->uio_resid < 0")); 893 KASSERT(uio->uio_offset >= 0, ("ffs_write: uio->uio_offset < 0")); 894 fs = ITOFS(ip); 895 896 /* 897 * Maybe this should be above the vnode op call, but so long as 898 * file servers have no limits, I don't think it matters. 899 */ 900 error = vn_rlimit_fsizex(vp, uio, fs->fs_maxfilesize, &r, 901 uio->uio_td); 902 if (error != 0) { 903 vn_rlimit_fsizex_res(uio, r); 904 return (error); 905 } 906 907 resid = uio->uio_resid; 908 osize = ip->i_size; 909 if (seqcount > BA_SEQMAX) 910 flags = BA_SEQMAX << BA_SEQSHIFT; 911 else 912 flags = seqcount << BA_SEQSHIFT; 913 if (ioflag & IO_SYNC) 914 flags |= IO_SYNC; 915 flags |= BA_UNMAPPED; 916 917 for (error = 0; uio->uio_resid > 0;) { 918 lbn = lblkno(fs, uio->uio_offset); 919 blkoffset = blkoff(fs, uio->uio_offset); 920 xfersize = fs->fs_bsize - blkoffset; 921 if (uio->uio_resid < xfersize) 922 xfersize = uio->uio_resid; 923 if (uio->uio_offset + xfersize > ip->i_size) 924 vnode_pager_setsize(vp, uio->uio_offset + xfersize); 925 926 /* 927 * We must perform a read-before-write if the transfer size 928 * does not cover the entire buffer. 929 */ 930 if (fs->fs_bsize > xfersize) 931 flags |= BA_CLRBUF; 932 else 933 flags &= ~BA_CLRBUF; 934 /* XXX is uio->uio_offset the right thing here? */ 935 error = UFS_BALLOC(vp, uio->uio_offset, xfersize, 936 ap->a_cred, flags, &bp); 937 if (error != 0) { 938 vnode_pager_setsize(vp, ip->i_size); 939 break; 940 } 941 if ((ioflag & (IO_SYNC|IO_INVAL)) == (IO_SYNC|IO_INVAL)) 942 bp->b_flags |= B_NOCACHE; 943 944 if (uio->uio_offset + xfersize > ip->i_size) { 945 ip->i_size = uio->uio_offset + xfersize; 946 DIP_SET(ip, i_size, ip->i_size); 947 UFS_INODE_SET_FLAG(ip, IN_SIZEMOD | IN_CHANGE); 948 } 949 950 size = blksize(fs, ip, lbn) - bp->b_resid; 951 if (size < xfersize) 952 xfersize = size; 953 954 if (buf_mapped(bp)) { 955 error = vn_io_fault_uiomove((char *)bp->b_data + 956 blkoffset, (int)xfersize, uio); 957 } else { 958 error = vn_io_fault_pgmove(bp->b_pages, 959 blkoffset + (bp->b_offset & PAGE_MASK), 960 (int)xfersize, uio); 961 } 962 /* 963 * If the buffer is not already filled and we encounter an 964 * error while trying to fill it, we have to clear out any 965 * garbage data from the pages instantiated for the buffer. 966 * If we do not, a failed uiomove() during a write can leave 967 * the prior contents of the pages exposed to a userland mmap. 968 * 969 * Note that we need only clear buffers with a transfer size 970 * equal to the block size because buffers with a shorter 971 * transfer size were cleared above by the call to UFS_BALLOC() 972 * with the BA_CLRBUF flag set. 973 * 974 * If the source region for uiomove identically mmaps the 975 * buffer, uiomove() performed the NOP copy, and the buffer 976 * content remains valid because the page fault handler 977 * validated the pages. 978 */ 979 if (error != 0 && (bp->b_flags & B_CACHE) == 0 && 980 fs->fs_bsize == xfersize) 981 vfs_bio_clrbuf(bp); 982 983 vfs_bio_set_flags(bp, ioflag); 984 985 /* 986 * If IO_SYNC each buffer is written synchronously. Otherwise 987 * if we have a severe page deficiency write the buffer 988 * asynchronously. Otherwise try to cluster, and if that 989 * doesn't do it then either do an async write (if O_DIRECT), 990 * or a delayed write (if not). 991 */ 992 if (ioflag & IO_SYNC) { 993 (void)bwrite(bp); 994 } else if (vm_page_count_severe() || 995 buf_dirty_count_severe() || 996 (ioflag & IO_ASYNC)) { 997 bp->b_flags |= B_CLUSTEROK; 998 bawrite(bp); 999 } else if (xfersize + blkoffset == fs->fs_bsize) { 1000 if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) { 1001 bp->b_flags |= B_CLUSTEROK; 1002 cluster_write(vp, &ip->i_clusterw, bp, 1003 ip->i_size, seqcount, GB_UNMAPPED); 1004 } else { 1005 bawrite(bp); 1006 } 1007 } else if (ioflag & IO_DIRECT) { 1008 bp->b_flags |= B_CLUSTEROK; 1009 bawrite(bp); 1010 } else { 1011 bp->b_flags |= B_CLUSTEROK; 1012 bdwrite(bp); 1013 } 1014 if (error || xfersize == 0) 1015 break; 1016 UFS_INODE_SET_FLAG(ip, IN_CHANGE | IN_UPDATE); 1017 } 1018 /* 1019 * If we successfully wrote any data, and we are not the superuser 1020 * we clear the setuid and setgid bits as a precaution against 1021 * tampering. 1022 */ 1023 if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid && 1024 ap->a_cred) { 1025 if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID)) { 1026 vn_seqc_write_begin(vp); 1027 UFS_INODE_SET_MODE(ip, ip->i_mode & ~(ISUID | ISGID)); 1028 DIP_SET(ip, i_mode, ip->i_mode); 1029 vn_seqc_write_end(vp); 1030 } 1031 } 1032 if (error) { 1033 if (ioflag & IO_UNIT) { 1034 (void)ffs_truncate(vp, osize, 1035 IO_NORMAL | (ioflag & IO_SYNC), ap->a_cred); 1036 uio->uio_offset -= resid - uio->uio_resid; 1037 uio->uio_resid = resid; 1038 } 1039 } else if (resid > uio->uio_resid && (ioflag & IO_SYNC)) { 1040 if (!(ioflag & IO_DATASYNC) || 1041 (ip->i_flags & (IN_SIZEMOD | IN_IBLKDATA))) 1042 error = ffs_update(vp, 1); 1043 if (ffs_fsfail_cleanup(VFSTOUFS(vp->v_mount), error)) 1044 error = ENXIO; 1045 } 1046 vn_rlimit_fsizex_res(uio, r); 1047 return (error); 1048 } 1049 1050 /* 1051 * Extended attribute area reading. 1052 */ 1053 static int 1054 ffs_extread(struct vnode *vp, struct uio *uio, int ioflag) 1055 { 1056 struct inode *ip; 1057 struct ufs2_dinode *dp; 1058 struct fs *fs; 1059 struct buf *bp; 1060 ufs_lbn_t lbn, nextlbn; 1061 off_t bytesinfile; 1062 long size, xfersize, blkoffset; 1063 ssize_t orig_resid; 1064 int error; 1065 1066 ip = VTOI(vp); 1067 fs = ITOFS(ip); 1068 dp = ip->i_din2; 1069 1070 #ifdef INVARIANTS 1071 if (uio->uio_rw != UIO_READ || fs->fs_magic != FS_UFS2_MAGIC) 1072 panic("ffs_extread: mode"); 1073 1074 #endif 1075 orig_resid = uio->uio_resid; 1076 KASSERT(orig_resid >= 0, ("ffs_extread: uio->uio_resid < 0")); 1077 if (orig_resid == 0) 1078 return (0); 1079 KASSERT(uio->uio_offset >= 0, ("ffs_extread: uio->uio_offset < 0")); 1080 1081 for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) { 1082 if ((bytesinfile = dp->di_extsize - uio->uio_offset) <= 0) 1083 break; 1084 lbn = lblkno(fs, uio->uio_offset); 1085 nextlbn = lbn + 1; 1086 1087 /* 1088 * size of buffer. The buffer representing the 1089 * end of the file is rounded up to the size of 1090 * the block type ( fragment or full block, 1091 * depending ). 1092 */ 1093 size = sblksize(fs, dp->di_extsize, lbn); 1094 blkoffset = blkoff(fs, uio->uio_offset); 1095 1096 /* 1097 * The amount we want to transfer in this iteration is 1098 * one FS block less the amount of the data before 1099 * our startpoint (duh!) 1100 */ 1101 xfersize = fs->fs_bsize - blkoffset; 1102 1103 /* 1104 * But if we actually want less than the block, 1105 * or the file doesn't have a whole block more of data, 1106 * then use the lesser number. 1107 */ 1108 if (uio->uio_resid < xfersize) 1109 xfersize = uio->uio_resid; 1110 if (bytesinfile < xfersize) 1111 xfersize = bytesinfile; 1112 1113 if (lblktosize(fs, nextlbn) >= dp->di_extsize) { 1114 /* 1115 * Don't do readahead if this is the end of the info. 1116 */ 1117 error = bread(vp, -1 - lbn, size, NOCRED, &bp); 1118 } else { 1119 /* 1120 * If we have a second block, then 1121 * fire off a request for a readahead 1122 * as well as a read. Note that the 4th and 5th 1123 * arguments point to arrays of the size specified in 1124 * the 6th argument. 1125 */ 1126 u_int nextsize = sblksize(fs, dp->di_extsize, nextlbn); 1127 1128 nextlbn = -1 - nextlbn; 1129 error = breadn(vp, -1 - lbn, 1130 size, &nextlbn, &nextsize, 1, NOCRED, &bp); 1131 } 1132 if (error) { 1133 brelse(bp); 1134 bp = NULL; 1135 break; 1136 } 1137 1138 /* 1139 * We should only get non-zero b_resid when an I/O error 1140 * has occurred, which should cause us to break above. 1141 * However, if the short read did not cause an error, 1142 * then we want to ensure that we do not uiomove bad 1143 * or uninitialized data. 1144 */ 1145 size -= bp->b_resid; 1146 if (size < xfersize) { 1147 if (size == 0) 1148 break; 1149 xfersize = size; 1150 } 1151 1152 error = uiomove((char *)bp->b_data + blkoffset, 1153 (int)xfersize, uio); 1154 if (error) 1155 break; 1156 vfs_bio_brelse(bp, ioflag); 1157 } 1158 1159 /* 1160 * This can only happen in the case of an error 1161 * because the loop above resets bp to NULL on each iteration 1162 * and on normal completion has not set a new value into it. 1163 * so it must have come from a 'break' statement 1164 */ 1165 if (bp != NULL) 1166 vfs_bio_brelse(bp, ioflag); 1167 return (error); 1168 } 1169 1170 /* 1171 * Extended attribute area writing. 1172 */ 1173 static int 1174 ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag, struct ucred *ucred) 1175 { 1176 struct inode *ip; 1177 struct ufs2_dinode *dp; 1178 struct fs *fs; 1179 struct buf *bp; 1180 ufs_lbn_t lbn; 1181 off_t osize; 1182 ssize_t resid; 1183 int blkoffset, error, flags, size, xfersize; 1184 1185 ip = VTOI(vp); 1186 fs = ITOFS(ip); 1187 dp = ip->i_din2; 1188 1189 #ifdef INVARIANTS 1190 if (uio->uio_rw != UIO_WRITE || fs->fs_magic != FS_UFS2_MAGIC) 1191 panic("ffs_extwrite: mode"); 1192 #endif 1193 1194 if (ioflag & IO_APPEND) 1195 uio->uio_offset = dp->di_extsize; 1196 KASSERT(uio->uio_offset >= 0, ("ffs_extwrite: uio->uio_offset < 0")); 1197 KASSERT(uio->uio_resid >= 0, ("ffs_extwrite: uio->uio_resid < 0")); 1198 if ((uoff_t)uio->uio_offset + uio->uio_resid > 1199 UFS_NXADDR * fs->fs_bsize) 1200 return (EFBIG); 1201 1202 resid = uio->uio_resid; 1203 osize = dp->di_extsize; 1204 flags = IO_EXT; 1205 if (ioflag & IO_SYNC) 1206 flags |= IO_SYNC; 1207 1208 for (error = 0; uio->uio_resid > 0;) { 1209 lbn = lblkno(fs, uio->uio_offset); 1210 blkoffset = blkoff(fs, uio->uio_offset); 1211 xfersize = fs->fs_bsize - blkoffset; 1212 if (uio->uio_resid < xfersize) 1213 xfersize = uio->uio_resid; 1214 1215 /* 1216 * We must perform a read-before-write if the transfer size 1217 * does not cover the entire buffer. 1218 */ 1219 if (fs->fs_bsize > xfersize) 1220 flags |= BA_CLRBUF; 1221 else 1222 flags &= ~BA_CLRBUF; 1223 error = UFS_BALLOC(vp, uio->uio_offset, xfersize, 1224 ucred, flags, &bp); 1225 if (error != 0) 1226 break; 1227 /* 1228 * If the buffer is not valid we have to clear out any 1229 * garbage data from the pages instantiated for the buffer. 1230 * If we do not, a failed uiomove() during a write can leave 1231 * the prior contents of the pages exposed to a userland 1232 * mmap(). XXX deal with uiomove() errors a better way. 1233 */ 1234 if ((bp->b_flags & B_CACHE) == 0 && fs->fs_bsize <= xfersize) 1235 vfs_bio_clrbuf(bp); 1236 1237 if (uio->uio_offset + xfersize > dp->di_extsize) { 1238 dp->di_extsize = uio->uio_offset + xfersize; 1239 UFS_INODE_SET_FLAG(ip, IN_SIZEMOD | IN_CHANGE); 1240 } 1241 1242 size = sblksize(fs, dp->di_extsize, lbn) - bp->b_resid; 1243 if (size < xfersize) 1244 xfersize = size; 1245 1246 error = 1247 uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio); 1248 1249 vfs_bio_set_flags(bp, ioflag); 1250 1251 /* 1252 * If IO_SYNC each buffer is written synchronously. Otherwise 1253 * if we have a severe page deficiency write the buffer 1254 * asynchronously. Otherwise try to cluster, and if that 1255 * doesn't do it then either do an async write (if O_DIRECT), 1256 * or a delayed write (if not). 1257 */ 1258 if (ioflag & IO_SYNC) { 1259 (void)bwrite(bp); 1260 } else if (vm_page_count_severe() || 1261 buf_dirty_count_severe() || 1262 xfersize + blkoffset == fs->fs_bsize || 1263 (ioflag & (IO_ASYNC | IO_DIRECT))) 1264 bawrite(bp); 1265 else 1266 bdwrite(bp); 1267 if (error || xfersize == 0) 1268 break; 1269 UFS_INODE_SET_FLAG(ip, IN_CHANGE); 1270 } 1271 /* 1272 * If we successfully wrote any data, and we are not the superuser 1273 * we clear the setuid and setgid bits as a precaution against 1274 * tampering. 1275 */ 1276 if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid && ucred) { 1277 if (priv_check_cred(ucred, PRIV_VFS_RETAINSUGID)) { 1278 vn_seqc_write_begin(vp); 1279 UFS_INODE_SET_MODE(ip, ip->i_mode & ~(ISUID | ISGID)); 1280 dp->di_mode = ip->i_mode; 1281 vn_seqc_write_end(vp); 1282 } 1283 } 1284 if (error) { 1285 if (ioflag & IO_UNIT) { 1286 (void)ffs_truncate(vp, osize, 1287 IO_EXT | (ioflag&IO_SYNC), ucred); 1288 uio->uio_offset -= resid - uio->uio_resid; 1289 uio->uio_resid = resid; 1290 } 1291 } else if (resid > uio->uio_resid && (ioflag & IO_SYNC)) 1292 error = ffs_update(vp, 1); 1293 return (error); 1294 } 1295 1296 /* 1297 * Vnode operating to retrieve a named extended attribute. 1298 * 1299 * Locate a particular EA (nspace:name) in the area (ptr:length), and return 1300 * the length of the EA, and possibly the pointer to the entry and to the data. 1301 */ 1302 static int 1303 ffs_findextattr(u_char *ptr, u_int length, int nspace, const char *name, 1304 struct extattr **eapp, u_char **eac) 1305 { 1306 struct extattr *eap, *eaend; 1307 size_t nlen; 1308 1309 nlen = strlen(name); 1310 KASSERT(ALIGNED_TO(ptr, struct extattr), ("unaligned")); 1311 eap = (struct extattr *)ptr; 1312 eaend = (struct extattr *)(ptr + length); 1313 for (; eap < eaend; eap = EXTATTR_NEXT(eap)) { 1314 KASSERT(EXTATTR_NEXT(eap) <= eaend, 1315 ("extattr next %p beyond %p", EXTATTR_NEXT(eap), eaend)); 1316 if (eap->ea_namespace != nspace || eap->ea_namelength != nlen 1317 || memcmp(eap->ea_name, name, nlen) != 0) 1318 continue; 1319 if (eapp != NULL) 1320 *eapp = eap; 1321 if (eac != NULL) 1322 *eac = EXTATTR_CONTENT(eap); 1323 return (EXTATTR_CONTENT_SIZE(eap)); 1324 } 1325 return (-1); 1326 } 1327 1328 static int 1329 ffs_rdextattr(u_char **p, struct vnode *vp, struct thread *td) 1330 { 1331 const struct extattr *eap, *eaend, *eapnext; 1332 struct inode *ip; 1333 struct ufs2_dinode *dp; 1334 struct fs *fs; 1335 struct uio luio; 1336 struct iovec liovec; 1337 u_int easize; 1338 int error; 1339 u_char *eae; 1340 1341 ip = VTOI(vp); 1342 fs = ITOFS(ip); 1343 dp = ip->i_din2; 1344 easize = dp->di_extsize; 1345 if ((uoff_t)easize > UFS_NXADDR * fs->fs_bsize) 1346 return (EFBIG); 1347 1348 eae = malloc(easize, M_TEMP, M_WAITOK); 1349 1350 liovec.iov_base = eae; 1351 liovec.iov_len = easize; 1352 luio.uio_iov = &liovec; 1353 luio.uio_iovcnt = 1; 1354 luio.uio_offset = 0; 1355 luio.uio_resid = easize; 1356 luio.uio_segflg = UIO_SYSSPACE; 1357 luio.uio_rw = UIO_READ; 1358 luio.uio_td = td; 1359 1360 error = ffs_extread(vp, &luio, IO_EXT | IO_SYNC); 1361 if (error) { 1362 free(eae, M_TEMP); 1363 return (error); 1364 } 1365 /* Validate disk xattrfile contents. */ 1366 for (eap = (void *)eae, eaend = (void *)(eae + easize); eap < eaend; 1367 eap = eapnext) { 1368 /* Detect zeroed out tail */ 1369 if (eap->ea_length < sizeof(*eap) || eap->ea_length == 0) { 1370 easize = (const u_char *)eap - eae; 1371 break; 1372 } 1373 1374 eapnext = EXTATTR_NEXT(eap); 1375 /* Bogusly long entry. */ 1376 if (eapnext > eaend) { 1377 free(eae, M_TEMP); 1378 return (EINTEGRITY); 1379 } 1380 } 1381 ip->i_ea_len = easize; 1382 *p = eae; 1383 return (0); 1384 } 1385 1386 static void 1387 ffs_lock_ea(struct vnode *vp) 1388 { 1389 struct inode *ip; 1390 1391 ip = VTOI(vp); 1392 VI_LOCK(vp); 1393 while (ip->i_flag & IN_EA_LOCKED) { 1394 UFS_INODE_SET_FLAG(ip, IN_EA_LOCKWAIT); 1395 msleep(&ip->i_ea_refs, &vp->v_interlock, PINOD + 2, "ufs_ea", 1396 0); 1397 } 1398 UFS_INODE_SET_FLAG(ip, IN_EA_LOCKED); 1399 VI_UNLOCK(vp); 1400 } 1401 1402 static void 1403 ffs_unlock_ea(struct vnode *vp) 1404 { 1405 struct inode *ip; 1406 1407 ip = VTOI(vp); 1408 VI_LOCK(vp); 1409 if (ip->i_flag & IN_EA_LOCKWAIT) 1410 wakeup(&ip->i_ea_refs); 1411 ip->i_flag &= ~(IN_EA_LOCKED | IN_EA_LOCKWAIT); 1412 VI_UNLOCK(vp); 1413 } 1414 1415 static int 1416 ffs_open_ea(struct vnode *vp, struct ucred *cred, struct thread *td) 1417 { 1418 struct inode *ip; 1419 int error; 1420 1421 ip = VTOI(vp); 1422 1423 ffs_lock_ea(vp); 1424 if (ip->i_ea_area != NULL) { 1425 ip->i_ea_refs++; 1426 ffs_unlock_ea(vp); 1427 return (0); 1428 } 1429 error = ffs_rdextattr(&ip->i_ea_area, vp, td); 1430 if (error) { 1431 ffs_unlock_ea(vp); 1432 return (error); 1433 } 1434 ip->i_ea_error = 0; 1435 ip->i_ea_refs++; 1436 ffs_unlock_ea(vp); 1437 return (0); 1438 } 1439 1440 /* 1441 * Vnode extattr transaction commit/abort 1442 */ 1443 static int 1444 ffs_close_ea(struct vnode *vp, int commit, struct ucred *cred, struct thread *td) 1445 { 1446 struct inode *ip; 1447 struct uio luio; 1448 struct iovec *liovec; 1449 struct ufs2_dinode *dp; 1450 size_t ea_len, tlen; 1451 int error, i, lcnt; 1452 bool truncate; 1453 1454 ip = VTOI(vp); 1455 1456 ffs_lock_ea(vp); 1457 if (ip->i_ea_area == NULL) { 1458 ffs_unlock_ea(vp); 1459 return (EINVAL); 1460 } 1461 dp = ip->i_din2; 1462 error = ip->i_ea_error; 1463 truncate = false; 1464 if (commit && error == 0) { 1465 ASSERT_VOP_ELOCKED(vp, "ffs_close_ea commit"); 1466 if (cred == NOCRED) 1467 cred = vp->v_mount->mnt_cred; 1468 1469 ea_len = MAX(ip->i_ea_len, dp->di_extsize); 1470 for (lcnt = 1, tlen = ea_len - ip->i_ea_len; tlen > 0;) { 1471 tlen -= MIN(ZERO_REGION_SIZE, tlen); 1472 lcnt++; 1473 } 1474 1475 liovec = __builtin_alloca(lcnt * sizeof(struct iovec)); 1476 luio.uio_iovcnt = lcnt; 1477 1478 liovec[0].iov_base = ip->i_ea_area; 1479 liovec[0].iov_len = ip->i_ea_len; 1480 for (i = 1, tlen = ea_len - ip->i_ea_len; i < lcnt; i++) { 1481 liovec[i].iov_base = __DECONST(void *, zero_region); 1482 liovec[i].iov_len = MIN(ZERO_REGION_SIZE, tlen); 1483 tlen -= liovec[i].iov_len; 1484 } 1485 MPASS(tlen == 0); 1486 1487 luio.uio_iov = liovec; 1488 luio.uio_offset = 0; 1489 luio.uio_resid = ea_len; 1490 luio.uio_segflg = UIO_SYSSPACE; 1491 luio.uio_rw = UIO_WRITE; 1492 luio.uio_td = td; 1493 error = ffs_extwrite(vp, &luio, IO_EXT | IO_SYNC, cred); 1494 if (error == 0 && ip->i_ea_len == 0) 1495 truncate = true; 1496 } 1497 if (--ip->i_ea_refs == 0) { 1498 free(ip->i_ea_area, M_TEMP); 1499 ip->i_ea_area = NULL; 1500 ip->i_ea_len = 0; 1501 ip->i_ea_error = 0; 1502 } 1503 ffs_unlock_ea(vp); 1504 1505 if (truncate) 1506 ffs_truncate(vp, 0, IO_EXT, cred); 1507 return (error); 1508 } 1509 1510 /* 1511 * Vnode extattr strategy routine for fifos. 1512 * 1513 * We need to check for a read or write of the external attributes. 1514 * Otherwise we just fall through and do the usual thing. 1515 */ 1516 static int 1517 ffsext_strategy( 1518 struct vop_strategy_args /* { 1519 struct vnodeop_desc *a_desc; 1520 struct vnode *a_vp; 1521 struct buf *a_bp; 1522 } */ *ap) 1523 { 1524 struct vnode *vp; 1525 daddr_t lbn; 1526 1527 vp = ap->a_vp; 1528 lbn = ap->a_bp->b_lblkno; 1529 if (I_IS_UFS2(VTOI(vp)) && lbn < 0 && lbn >= -UFS_NXADDR) 1530 return (VOP_STRATEGY_APV(&ufs_vnodeops, ap)); 1531 if (vp->v_type == VFIFO) 1532 return (VOP_STRATEGY_APV(&ufs_fifoops, ap)); 1533 panic("spec nodes went here"); 1534 } 1535 1536 /* 1537 * Vnode extattr transaction commit/abort 1538 */ 1539 static int 1540 ffs_openextattr( 1541 struct vop_openextattr_args /* { 1542 struct vnodeop_desc *a_desc; 1543 struct vnode *a_vp; 1544 IN struct ucred *a_cred; 1545 IN struct thread *a_td; 1546 } */ *ap) 1547 { 1548 1549 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK) 1550 return (EOPNOTSUPP); 1551 1552 return (ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td)); 1553 } 1554 1555 /* 1556 * Vnode extattr transaction commit/abort 1557 */ 1558 static int 1559 ffs_closeextattr( 1560 struct vop_closeextattr_args /* { 1561 struct vnodeop_desc *a_desc; 1562 struct vnode *a_vp; 1563 int a_commit; 1564 IN struct ucred *a_cred; 1565 IN struct thread *a_td; 1566 } */ *ap) 1567 { 1568 struct vnode *vp; 1569 1570 vp = ap->a_vp; 1571 if (vp->v_type == VCHR || vp->v_type == VBLK) 1572 return (EOPNOTSUPP); 1573 if (ap->a_commit && (vp->v_mount->mnt_flag & MNT_RDONLY) != 0) 1574 return (EROFS); 1575 1576 if (ap->a_commit && DOINGSUJ(vp)) { 1577 ASSERT_VOP_ELOCKED(vp, "ffs_closeextattr commit"); 1578 softdep_prealloc(vp, MNT_WAIT); 1579 if (vp->v_data == NULL) 1580 return (EBADF); 1581 } 1582 return (ffs_close_ea(vp, ap->a_commit, ap->a_cred, ap->a_td)); 1583 } 1584 1585 /* 1586 * Vnode operation to remove a named attribute. 1587 */ 1588 static int 1589 ffs_deleteextattr( 1590 struct vop_deleteextattr_args /* { 1591 IN struct vnode *a_vp; 1592 IN int a_attrnamespace; 1593 IN const char *a_name; 1594 IN struct ucred *a_cred; 1595 IN struct thread *a_td; 1596 } */ *ap) 1597 { 1598 struct vnode *vp; 1599 struct inode *ip; 1600 struct extattr *eap; 1601 uint32_t ul; 1602 int olen, error, i, easize; 1603 u_char *eae; 1604 void *tmp; 1605 1606 vp = ap->a_vp; 1607 ip = VTOI(vp); 1608 1609 if (vp->v_type == VCHR || vp->v_type == VBLK) 1610 return (EOPNOTSUPP); 1611 if (strlen(ap->a_name) == 0) 1612 return (EINVAL); 1613 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1614 return (EROFS); 1615 1616 error = extattr_check_cred(vp, ap->a_attrnamespace, 1617 ap->a_cred, ap->a_td, VWRITE); 1618 if (error) { 1619 /* 1620 * ffs_lock_ea is not needed there, because the vnode 1621 * must be exclusively locked. 1622 */ 1623 if (ip->i_ea_area != NULL && ip->i_ea_error == 0) 1624 ip->i_ea_error = error; 1625 return (error); 1626 } 1627 1628 if (DOINGSUJ(vp)) { 1629 ASSERT_VOP_ELOCKED(vp, "ffs_deleteextattr"); 1630 softdep_prealloc(vp, MNT_WAIT); 1631 if (vp->v_data == NULL) 1632 return (EBADF); 1633 } 1634 1635 error = ffs_open_ea(vp, ap->a_cred, ap->a_td); 1636 if (error) 1637 return (error); 1638 1639 /* CEM: delete could be done in-place instead */ 1640 eae = malloc(ip->i_ea_len, M_TEMP, M_WAITOK); 1641 bcopy(ip->i_ea_area, eae, ip->i_ea_len); 1642 easize = ip->i_ea_len; 1643 1644 olen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name, 1645 &eap, NULL); 1646 if (olen == -1) { 1647 /* delete but nonexistent */ 1648 free(eae, M_TEMP); 1649 ffs_close_ea(vp, 0, ap->a_cred, ap->a_td); 1650 return (ENOATTR); 1651 } 1652 ul = eap->ea_length; 1653 i = (u_char *)EXTATTR_NEXT(eap) - eae; 1654 bcopy(EXTATTR_NEXT(eap), eap, easize - i); 1655 easize -= ul; 1656 1657 tmp = ip->i_ea_area; 1658 ip->i_ea_area = eae; 1659 ip->i_ea_len = easize; 1660 free(tmp, M_TEMP); 1661 error = ffs_close_ea(vp, 1, ap->a_cred, ap->a_td); 1662 return (error); 1663 } 1664 1665 /* 1666 * Vnode operation to retrieve a named extended attribute. 1667 */ 1668 static int 1669 ffs_getextattr( 1670 struct vop_getextattr_args /* { 1671 IN struct vnode *a_vp; 1672 IN int a_attrnamespace; 1673 IN const char *a_name; 1674 INOUT struct uio *a_uio; 1675 OUT size_t *a_size; 1676 IN struct ucred *a_cred; 1677 IN struct thread *a_td; 1678 } */ *ap) 1679 { 1680 struct inode *ip; 1681 u_char *eae, *p; 1682 unsigned easize; 1683 int error, ealen; 1684 1685 ip = VTOI(ap->a_vp); 1686 1687 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK) 1688 return (EOPNOTSUPP); 1689 1690 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 1691 ap->a_cred, ap->a_td, VREAD); 1692 if (error) 1693 return (error); 1694 1695 error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td); 1696 if (error) 1697 return (error); 1698 1699 eae = ip->i_ea_area; 1700 easize = ip->i_ea_len; 1701 1702 ealen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name, 1703 NULL, &p); 1704 if (ealen >= 0) { 1705 error = 0; 1706 if (ap->a_size != NULL) 1707 *ap->a_size = ealen; 1708 else if (ap->a_uio != NULL) 1709 error = uiomove(p, ealen, ap->a_uio); 1710 } else 1711 error = ENOATTR; 1712 1713 ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td); 1714 return (error); 1715 } 1716 1717 /* 1718 * Vnode operation to retrieve extended attributes on a vnode. 1719 */ 1720 static int 1721 ffs_listextattr( 1722 struct vop_listextattr_args /* { 1723 IN struct vnode *a_vp; 1724 IN int a_attrnamespace; 1725 INOUT struct uio *a_uio; 1726 OUT size_t *a_size; 1727 IN struct ucred *a_cred; 1728 IN struct thread *a_td; 1729 } */ *ap) 1730 { 1731 struct inode *ip; 1732 struct extattr *eap, *eaend; 1733 int error, ealen; 1734 1735 ip = VTOI(ap->a_vp); 1736 1737 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK) 1738 return (EOPNOTSUPP); 1739 1740 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 1741 ap->a_cred, ap->a_td, VREAD); 1742 if (error) 1743 return (error); 1744 1745 error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td); 1746 if (error) 1747 return (error); 1748 1749 error = 0; 1750 if (ap->a_size != NULL) 1751 *ap->a_size = 0; 1752 1753 KASSERT(ALIGNED_TO(ip->i_ea_area, struct extattr), ("unaligned")); 1754 eap = (struct extattr *)ip->i_ea_area; 1755 eaend = (struct extattr *)(ip->i_ea_area + ip->i_ea_len); 1756 for (; error == 0 && eap < eaend; eap = EXTATTR_NEXT(eap)) { 1757 KASSERT(EXTATTR_NEXT(eap) <= eaend, 1758 ("extattr next %p beyond %p", EXTATTR_NEXT(eap), eaend)); 1759 if (eap->ea_namespace != ap->a_attrnamespace) 1760 continue; 1761 1762 ealen = eap->ea_namelength; 1763 if (ap->a_size != NULL) 1764 *ap->a_size += ealen + 1; 1765 else if (ap->a_uio != NULL) 1766 error = uiomove(&eap->ea_namelength, ealen + 1, 1767 ap->a_uio); 1768 } 1769 1770 ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td); 1771 return (error); 1772 } 1773 1774 /* 1775 * Vnode operation to set a named attribute. 1776 */ 1777 static int 1778 ffs_setextattr( 1779 struct vop_setextattr_args /* { 1780 IN struct vnode *a_vp; 1781 IN int a_attrnamespace; 1782 IN const char *a_name; 1783 INOUT struct uio *a_uio; 1784 IN struct ucred *a_cred; 1785 IN struct thread *a_td; 1786 } */ *ap) 1787 { 1788 struct vnode *vp; 1789 struct inode *ip; 1790 struct fs *fs; 1791 struct extattr *eap; 1792 uint32_t ealength, ul; 1793 ssize_t ealen; 1794 int olen, eapad1, eapad2, error, i, easize; 1795 u_char *eae; 1796 void *tmp; 1797 1798 vp = ap->a_vp; 1799 ip = VTOI(vp); 1800 fs = ITOFS(ip); 1801 1802 if (vp->v_type == VCHR || vp->v_type == VBLK) 1803 return (EOPNOTSUPP); 1804 if (strlen(ap->a_name) == 0) 1805 return (EINVAL); 1806 1807 /* XXX Now unsupported API to delete EAs using NULL uio. */ 1808 if (ap->a_uio == NULL) 1809 return (EOPNOTSUPP); 1810 1811 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1812 return (EROFS); 1813 1814 ealen = ap->a_uio->uio_resid; 1815 if (ealen < 0 || ealen > lblktosize(fs, UFS_NXADDR)) 1816 return (EINVAL); 1817 1818 error = extattr_check_cred(vp, ap->a_attrnamespace, 1819 ap->a_cred, ap->a_td, VWRITE); 1820 if (error) { 1821 /* 1822 * ffs_lock_ea is not needed there, because the vnode 1823 * must be exclusively locked. 1824 */ 1825 if (ip->i_ea_area != NULL && ip->i_ea_error == 0) 1826 ip->i_ea_error = error; 1827 return (error); 1828 } 1829 1830 if (DOINGSUJ(vp)) { 1831 ASSERT_VOP_ELOCKED(vp, "ffs_deleteextattr"); 1832 softdep_prealloc(vp, MNT_WAIT); 1833 if (vp->v_data == NULL) 1834 return (EBADF); 1835 } 1836 1837 error = ffs_open_ea(vp, ap->a_cred, ap->a_td); 1838 if (error) 1839 return (error); 1840 1841 ealength = sizeof(uint32_t) + 3 + strlen(ap->a_name); 1842 eapad1 = roundup2(ealength, 8) - ealength; 1843 eapad2 = roundup2(ealen, 8) - ealen; 1844 ealength += eapad1 + ealen + eapad2; 1845 1846 /* 1847 * CEM: rewrites of the same size or smaller could be done in-place 1848 * instead. (We don't acquire any fine-grained locks in here either, 1849 * so we could also do bigger writes in-place.) 1850 */ 1851 eae = malloc(ip->i_ea_len + ealength, M_TEMP, M_WAITOK); 1852 bcopy(ip->i_ea_area, eae, ip->i_ea_len); 1853 easize = ip->i_ea_len; 1854 1855 olen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name, 1856 &eap, NULL); 1857 if (olen == -1) { 1858 /* new, append at end */ 1859 KASSERT(ALIGNED_TO(eae + easize, struct extattr), 1860 ("unaligned")); 1861 eap = (struct extattr *)(eae + easize); 1862 easize += ealength; 1863 } else { 1864 ul = eap->ea_length; 1865 i = (u_char *)EXTATTR_NEXT(eap) - eae; 1866 if (ul != ealength) { 1867 bcopy(EXTATTR_NEXT(eap), (u_char *)eap + ealength, 1868 easize - i); 1869 easize += (ealength - ul); 1870 } 1871 } 1872 if (easize > lblktosize(fs, UFS_NXADDR)) { 1873 free(eae, M_TEMP); 1874 ffs_close_ea(vp, 0, ap->a_cred, ap->a_td); 1875 if (ip->i_ea_area != NULL && ip->i_ea_error == 0) 1876 ip->i_ea_error = ENOSPC; 1877 return (ENOSPC); 1878 } 1879 eap->ea_length = ealength; 1880 eap->ea_namespace = ap->a_attrnamespace; 1881 eap->ea_contentpadlen = eapad2; 1882 eap->ea_namelength = strlen(ap->a_name); 1883 memcpy(eap->ea_name, ap->a_name, strlen(ap->a_name)); 1884 bzero(&eap->ea_name[strlen(ap->a_name)], eapad1); 1885 error = uiomove(EXTATTR_CONTENT(eap), ealen, ap->a_uio); 1886 if (error) { 1887 free(eae, M_TEMP); 1888 ffs_close_ea(vp, 0, ap->a_cred, ap->a_td); 1889 if (ip->i_ea_area != NULL && ip->i_ea_error == 0) 1890 ip->i_ea_error = error; 1891 return (error); 1892 } 1893 bzero((u_char *)EXTATTR_CONTENT(eap) + ealen, eapad2); 1894 1895 tmp = ip->i_ea_area; 1896 ip->i_ea_area = eae; 1897 ip->i_ea_len = easize; 1898 free(tmp, M_TEMP); 1899 error = ffs_close_ea(vp, 1, ap->a_cred, ap->a_td); 1900 return (error); 1901 } 1902 1903 /* 1904 * Vnode pointer to File handle 1905 */ 1906 static int 1907 ffs_vptofh( 1908 struct vop_vptofh_args /* { 1909 IN struct vnode *a_vp; 1910 IN struct fid *a_fhp; 1911 } */ *ap) 1912 { 1913 struct inode *ip; 1914 struct ufid *ufhp; 1915 1916 ip = VTOI(ap->a_vp); 1917 ufhp = (struct ufid *)ap->a_fhp; 1918 ufhp->ufid_len = sizeof(struct ufid); 1919 ufhp->ufid_ino = ip->i_number; 1920 ufhp->ufid_gen = ip->i_gen; 1921 return (0); 1922 } 1923 1924 SYSCTL_DECL(_vfs_ffs); 1925 static int use_buf_pager = 1; 1926 SYSCTL_INT(_vfs_ffs, OID_AUTO, use_buf_pager, CTLFLAG_RWTUN, &use_buf_pager, 0, 1927 "Always use buffer pager instead of bmap"); 1928 1929 static daddr_t 1930 ffs_gbp_getblkno(struct vnode *vp, vm_ooffset_t off) 1931 { 1932 1933 return (lblkno(VFSTOUFS(vp->v_mount)->um_fs, off)); 1934 } 1935 1936 static int 1937 ffs_gbp_getblksz(struct vnode *vp, daddr_t lbn, long *sz) 1938 { 1939 1940 *sz = blksize(VFSTOUFS(vp->v_mount)->um_fs, VTOI(vp), lbn); 1941 return (0); 1942 } 1943 1944 static int 1945 ffs_getpages(struct vop_getpages_args *ap) 1946 { 1947 struct vnode *vp; 1948 struct ufsmount *um; 1949 1950 vp = ap->a_vp; 1951 um = VFSTOUFS(vp->v_mount); 1952 1953 if (!use_buf_pager && um->um_devvp->v_bufobj.bo_bsize <= PAGE_SIZE) 1954 return (vnode_pager_generic_getpages(vp, ap->a_m, ap->a_count, 1955 ap->a_rbehind, ap->a_rahead, NULL, NULL)); 1956 return (vfs_bio_getpages(vp, ap->a_m, ap->a_count, ap->a_rbehind, 1957 ap->a_rahead, ffs_gbp_getblkno, ffs_gbp_getblksz)); 1958 } 1959 1960 static int 1961 ffs_getpages_async(struct vop_getpages_async_args *ap) 1962 { 1963 struct vnode *vp; 1964 struct ufsmount *um; 1965 bool do_iodone; 1966 int error; 1967 1968 vp = ap->a_vp; 1969 um = VFSTOUFS(vp->v_mount); 1970 do_iodone = true; 1971 1972 if (um->um_devvp->v_bufobj.bo_bsize <= PAGE_SIZE) { 1973 error = vnode_pager_generic_getpages(vp, ap->a_m, ap->a_count, 1974 ap->a_rbehind, ap->a_rahead, ap->a_iodone, ap->a_arg); 1975 if (error == 0) 1976 do_iodone = false; 1977 } else { 1978 error = vfs_bio_getpages(vp, ap->a_m, ap->a_count, 1979 ap->a_rbehind, ap->a_rahead, ffs_gbp_getblkno, 1980 ffs_gbp_getblksz); 1981 } 1982 if (do_iodone && ap->a_iodone != NULL) 1983 ap->a_iodone(ap->a_arg, ap->a_m, ap->a_count, error); 1984 1985 return (error); 1986 } 1987 1988 static int 1989 ffs_vput_pair(struct vop_vput_pair_args *ap) 1990 { 1991 struct mount *mp; 1992 struct vnode *dvp, *vp, *vp1, **vpp; 1993 struct inode *dp, *ip; 1994 ino_t ip_ino; 1995 u_int64_t ip_gen; 1996 int error, vp_locked; 1997 1998 dvp = ap->a_dvp; 1999 dp = VTOI(dvp); 2000 vpp = ap->a_vpp; 2001 vp = vpp != NULL ? *vpp : NULL; 2002 2003 if ((dp->i_flag & (IN_NEEDSYNC | IN_ENDOFF)) == 0) { 2004 vput(dvp); 2005 if (vp != NULL && ap->a_unlock_vp) 2006 vput(vp); 2007 return (0); 2008 } 2009 2010 mp = dvp->v_mount; 2011 if (vp != NULL) { 2012 if (ap->a_unlock_vp) { 2013 vput(vp); 2014 } else { 2015 MPASS(vp->v_type != VNON); 2016 vp_locked = VOP_ISLOCKED(vp); 2017 ip = VTOI(vp); 2018 ip_ino = ip->i_number; 2019 ip_gen = ip->i_gen; 2020 VOP_UNLOCK(vp); 2021 } 2022 } 2023 2024 /* 2025 * If compaction or fsync was requested do it in ffs_vput_pair() 2026 * now that other locks are no longer held. 2027 */ 2028 if ((dp->i_flag & IN_ENDOFF) != 0) { 2029 VNASSERT(I_ENDOFF(dp) != 0 && I_ENDOFF(dp) < dp->i_size, dvp, 2030 ("IN_ENDOFF set but I_ENDOFF() is not")); 2031 dp->i_flag &= ~IN_ENDOFF; 2032 error = UFS_TRUNCATE(dvp, (off_t)I_ENDOFF(dp), IO_NORMAL | 2033 (DOINGASYNC(dvp) ? 0 : IO_SYNC), curthread->td_ucred); 2034 if (error != 0 && error != ERELOOKUP) { 2035 if (!ffs_fsfail_cleanup(VFSTOUFS(mp), error)) { 2036 vn_printf(dvp, 2037 "IN_ENDOFF: failed to truncate, " 2038 "error %d\n", error); 2039 } 2040 #ifdef UFS_DIRHASH 2041 ufsdirhash_free(dp); 2042 #endif 2043 } 2044 SET_I_ENDOFF(dp, 0); 2045 } 2046 if ((dp->i_flag & IN_NEEDSYNC) != 0) { 2047 do { 2048 error = ffs_syncvnode(dvp, MNT_WAIT, 0); 2049 } while (error == ERELOOKUP); 2050 } 2051 2052 vput(dvp); 2053 2054 if (vp == NULL || ap->a_unlock_vp) 2055 return (0); 2056 MPASS(mp != NULL); 2057 2058 /* 2059 * It is possible that vp is reclaimed at this point. Only 2060 * routines that call us with a_unlock_vp == false can find 2061 * that their vp has been reclaimed. There are three areas 2062 * that are affected: 2063 * 1) vn_open_cred() - later VOPs could fail, but 2064 * dead_open() returns 0 to simulate successful open. 2065 * 2) ffs_snapshot() - creation of snapshot fails with EBADF. 2066 * 3) NFS server (several places) - code is prepared to detect 2067 * and respond to dead vnodes by returning ESTALE. 2068 */ 2069 VOP_LOCK(vp, vp_locked | LK_RETRY); 2070 if (IS_UFS(vp)) 2071 return (0); 2072 2073 /* 2074 * Try harder to recover from reclaimed vp if reclaim was not 2075 * because underlying inode was cleared. We saved inode 2076 * number and inode generation, so we can try to reinstantiate 2077 * exactly same version of inode. If this fails, return 2078 * original doomed vnode and let caller to handle 2079 * consequences. 2080 * 2081 * Note that callers must keep write started around 2082 * VOP_VPUT_PAIR() calls, so it is safe to use mp without 2083 * busying it. 2084 */ 2085 VOP_UNLOCK(vp); 2086 error = ffs_inotovp(mp, ip_ino, ip_gen, LK_EXCLUSIVE, &vp1, 2087 FFSV_REPLACE_DOOMED); 2088 if (error != 0) { 2089 VOP_LOCK(vp, vp_locked | LK_RETRY); 2090 } else { 2091 vrele(vp); 2092 *vpp = vp1; 2093 } 2094 return (error); 2095 } 2096