1/*- 2 * Copyright (c) 2002, 2003 Networks Associates Technology, Inc. 3 * All rights reserved. 4 * 5 * This software was developed for the FreeBSD Project by Marshall 6 * Kirk McKusick and Network Associates Laboratories, the Security 7 * Research Division of Network Associates, Inc. under DARPA/SPAWAR 8 * contract N66001-01-C-8035 ("CBOSS"), as part of the DARPA CHATS 9 * research program 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 23 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 30 * SUCH DAMAGE. 31 * 32 * Copyright (c) 1982, 1986, 1989, 1993 33 * The Regents of the University of California. All rights reserved. 34 * 35 * Redistribution and use in source and binary forms, with or without 36 * modification, are permitted provided that the following conditions 37 * are met: 38 * 1. Redistributions of source code must retain the above copyright 39 * notice, this list of conditions and the following disclaimer. 40 * 2. Redistributions in binary form must reproduce the above copyright 41 * notice, this list of conditions and the following disclaimer in the 42 * documentation and/or other materials provided with the distribution. 43 * 4. Neither the name of the University nor the names of its contributors 44 * may be used to endorse or promote products derived from this software 45 * without specific prior written permission. 46 * 47 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 48 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 49 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 50 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 51 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 52 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 53 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 54 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 55 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 56 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 57 * SUCH DAMAGE. 58 * 59 * from: @(#)ufs_readwrite.c 8.11 (Berkeley) 5/8/95 60 * from: $FreeBSD: .../ufs/ufs_readwrite.c,v 1.96 2002/08/12 09:22:11 phk ... 61 * @(#)ffs_vnops.c 8.15 (Berkeley) 5/14/95 62 */ 63 64#include <sys/cdefs.h> 65__FBSDID("$FreeBSD$"); 66 67#include <sys/param.h> 68#include <sys/bio.h> 69#include <sys/systm.h> 70#include <sys/buf.h> 71#include <sys/conf.h> 72#include <sys/extattr.h> 73#include <sys/kernel.h> 74#include <sys/limits.h> 75#include <sys/malloc.h> 76#include <sys/mount.h> 77#include <sys/priv.h> 78#include <sys/stat.h> 79#include <sys/vmmeter.h> 80#include <sys/vnode.h> 81 82#include <vm/vm.h> 83#include <vm/vm_param.h> 84#include <vm/vm_extern.h> 85#include <vm/vm_object.h> 86#include <vm/vm_page.h> 87#include <vm/vm_pager.h> 88#include <vm/vnode_pager.h> 89 90#include <ufs/ufs/extattr.h> 91#include <ufs/ufs/quota.h> 92#include <ufs/ufs/inode.h> 93#include <ufs/ufs/ufs_extern.h> 94#include <ufs/ufs/ufsmount.h> 95 96#include <ufs/ffs/fs.h> 97#include <ufs/ffs/ffs_extern.h> 98#include "opt_directio.h" 99#include "opt_ffs.h" 100 101#ifdef DIRECTIO 102extern int ffs_rawread(struct vnode *vp, struct uio *uio, int *workdone); 103#endif 104static vop_fsync_t ffs_fsync; 105static vop_lock1_t ffs_lock; 106static vop_getpages_t ffs_getpages; 107static vop_read_t ffs_read; 108static vop_write_t ffs_write; 109static int ffs_extread(struct vnode *vp, struct uio *uio, int ioflag); 110static int ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag, 111 struct ucred *cred); 112static vop_strategy_t ffsext_strategy; 113static vop_closeextattr_t ffs_closeextattr; 114static vop_deleteextattr_t ffs_deleteextattr; 115static vop_getextattr_t ffs_getextattr; 116static vop_listextattr_t ffs_listextattr; 117static vop_openextattr_t ffs_openextattr; 118static vop_setextattr_t ffs_setextattr; 119static vop_vptofh_t ffs_vptofh; 120 121 122/* Global vfs data structures for ufs. */ 123struct vop_vector ffs_vnodeops1 = { 124 .vop_default = &ufs_vnodeops, 125 .vop_fsync = ffs_fsync, 126 .vop_getpages = ffs_getpages, 127 .vop_lock1 = ffs_lock, 128 .vop_read = ffs_read, 129 .vop_reallocblks = ffs_reallocblks, 130 .vop_write = ffs_write, 131 .vop_vptofh = ffs_vptofh, 132}; 133 134struct vop_vector ffs_fifoops1 = { 135 .vop_default = &ufs_fifoops, 136 .vop_fsync = ffs_fsync, 137 .vop_reallocblks = ffs_reallocblks, /* XXX: really ??? */ 138 .vop_vptofh = ffs_vptofh, 139}; 140 141/* Global vfs data structures for ufs. */ 142struct vop_vector ffs_vnodeops2 = { 143 .vop_default = &ufs_vnodeops, 144 .vop_fsync = ffs_fsync, 145 .vop_getpages = ffs_getpages, 146 .vop_lock1 = ffs_lock, 147 .vop_read = ffs_read, 148 .vop_reallocblks = ffs_reallocblks, 149 .vop_write = ffs_write, 150 .vop_closeextattr = ffs_closeextattr, 151 .vop_deleteextattr = ffs_deleteextattr, 152 .vop_getextattr = ffs_getextattr, 153 .vop_listextattr = ffs_listextattr, 154 .vop_openextattr = ffs_openextattr, 155 .vop_setextattr = ffs_setextattr, 156 .vop_vptofh = ffs_vptofh, 157}; 158 159struct vop_vector ffs_fifoops2 = { 160 .vop_default = &ufs_fifoops, 161 .vop_fsync = ffs_fsync, 162 .vop_lock1 = ffs_lock, 163 .vop_reallocblks = ffs_reallocblks, 164 .vop_strategy = ffsext_strategy, 165 .vop_closeextattr = ffs_closeextattr, 166 .vop_deleteextattr = ffs_deleteextattr, 167 .vop_getextattr = ffs_getextattr, 168 .vop_listextattr = ffs_listextattr, 169 .vop_openextattr = ffs_openextattr, 170 .vop_setextattr = ffs_setextattr, 171 .vop_vptofh = ffs_vptofh, 172}; 173 174/* 175 * Synch an open file. 176 */ 177/* ARGSUSED */ 178static int 179ffs_fsync(struct vop_fsync_args *ap) 180{ 181 struct vnode *vp; 182 struct bufobj *bo; 183 int error; 184 185 vp = ap->a_vp; 186 bo = &vp->v_bufobj; 187retry: 188 error = ffs_syncvnode(vp, ap->a_waitfor, 0); 189 if (error) 190 return (error); 191 if (ap->a_waitfor == MNT_WAIT && DOINGSOFTDEP(vp)) { 192 error = softdep_fsync(vp); 193 if (error) 194 return (error); 195 196 /* 197 * The softdep_fsync() function may drop vp lock, 198 * allowing for dirty buffers to reappear on the 199 * bo_dirty list. Recheck and resync as needed. 200 */ 201 BO_LOCK(bo); 202 if (vp->v_type == VREG && (bo->bo_numoutput > 0 || 203 bo->bo_dirty.bv_cnt > 0)) { 204 BO_UNLOCK(bo); 205 goto retry; 206 } 207 BO_UNLOCK(bo); 208 } 209 return (0); 210} 211 212int 213ffs_syncvnode(struct vnode *vp, int waitfor, int flags) 214{ 215 struct inode *ip; 216 struct bufobj *bo; 217 struct buf *bp; 218 struct buf *nbp; 219 ufs_lbn_t lbn; 220 int error, wait, passes; 221 222 ip = VTOI(vp); 223 ip->i_flag &= ~IN_NEEDSYNC; 224 bo = &vp->v_bufobj; 225 226 /* 227 * When doing MNT_WAIT we must first flush all dependencies 228 * on the inode. 229 */ 230 if (DOINGSOFTDEP(vp) && waitfor == MNT_WAIT && 231 (error = softdep_sync_metadata(vp)) != 0) 232 return (error); 233 234 /* 235 * Flush all dirty buffers associated with a vnode. 236 */ 237 error = 0; 238 passes = 0; 239 wait = 0; /* Always do an async pass first. */ 240 lbn = lblkno(ip->i_fs, (ip->i_size + ip->i_fs->fs_bsize - 1)); 241 BO_LOCK(bo); 242loop: 243 TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs) 244 bp->b_vflags &= ~BV_SCANNED; 245 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 246 /* 247 * Reasons to skip this buffer: it has already been considered 248 * on this pass, the buffer has dependencies that will cause 249 * it to be redirtied and it has not already been deferred, 250 * or it is already being written. 251 */ 252 if ((bp->b_vflags & BV_SCANNED) != 0) 253 continue; 254 bp->b_vflags |= BV_SCANNED; 255 /* Flush indirects in order. */ 256 if (waitfor == MNT_WAIT && bp->b_lblkno <= -NDADDR && 257 lbn_level(bp->b_lblkno) >= passes) 258 continue; 259 if (bp->b_lblkno > lbn) 260 panic("ffs_syncvnode: syncing truncated data."); 261 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) 262 continue; 263 BO_UNLOCK(bo); 264 if ((bp->b_flags & B_DELWRI) == 0) 265 panic("ffs_fsync: not dirty"); 266 /* 267 * Check for dependencies and potentially complete them. 268 */ 269 if (!LIST_EMPTY(&bp->b_dep) && 270 (error = softdep_sync_buf(vp, bp, 271 wait ? MNT_WAIT : MNT_NOWAIT)) != 0) { 272 /* I/O error. */ 273 if (error != EBUSY) { 274 BUF_UNLOCK(bp); 275 return (error); 276 } 277 /* If we deferred once, don't defer again. */ 278 if ((bp->b_flags & B_DEFERRED) == 0) { 279 bp->b_flags |= B_DEFERRED; 280 BUF_UNLOCK(bp); 281 goto next; 282 } 283 } 284 if (wait) { 285 bremfree(bp); 286 if ((error = bwrite(bp)) != 0) 287 return (error); 288 } else if ((bp->b_flags & B_CLUSTEROK)) { 289 (void) vfs_bio_awrite(bp); 290 } else { 291 bremfree(bp); 292 (void) bawrite(bp); 293 } 294next: 295 /* 296 * Since we may have slept during the I/O, we need 297 * to start from a known point. 298 */ 299 BO_LOCK(bo); 300 nbp = TAILQ_FIRST(&bo->bo_dirty.bv_hd); 301 } 302 if (waitfor != MNT_WAIT) { 303 BO_UNLOCK(bo); 304 if ((flags & NO_INO_UPDT) != 0) 305 return (0); 306 else 307 return (ffs_update(vp, 0)); 308 } 309 /* Drain IO to see if we're done. */ 310 bufobj_wwait(bo, 0, 0); 311 /* 312 * Block devices associated with filesystems may have new I/O 313 * requests posted for them even if the vnode is locked, so no 314 * amount of trying will get them clean. We make several passes 315 * as a best effort. 316 * 317 * Regular files may need multiple passes to flush all dependency 318 * work as it is possible that we must write once per indirect 319 * level, once for the leaf, and once for the inode and each of 320 * these will be done with one sync and one async pass. 321 */ 322 if (bo->bo_dirty.bv_cnt > 0) { 323 /* Write the inode after sync passes to flush deps. */ 324 if (wait && DOINGSOFTDEP(vp) && (flags & NO_INO_UPDT) == 0) { 325 BO_UNLOCK(bo); 326 ffs_update(vp, 1); 327 BO_LOCK(bo); 328 } 329 /* switch between sync/async. */ 330 wait = !wait; 331 if (wait == 1 || ++passes < NIADDR + 2) 332 goto loop; 333#ifdef INVARIANTS 334 if (!vn_isdisk(vp, NULL)) 335 vprint("ffs_fsync: dirty", vp); 336#endif 337 } 338 BO_UNLOCK(bo); 339 error = 0; 340 if ((flags & NO_INO_UPDT) == 0) 341 error = ffs_update(vp, 1); 342 if (DOINGSUJ(vp)) 343 softdep_journal_fsync(VTOI(vp)); 344 return (error); 345} 346 347static int 348ffs_lock(ap) 349 struct vop_lock1_args /* { 350 struct vnode *a_vp; 351 int a_flags; 352 struct thread *a_td; 353 char *file; 354 int line; 355 } */ *ap; 356{ 357#ifndef NO_FFS_SNAPSHOT 358 struct vnode *vp; 359 int flags; 360 struct lock *lkp; 361 int result; 362 363 switch (ap->a_flags & LK_TYPE_MASK) { 364 case LK_SHARED: 365 case LK_UPGRADE: 366 case LK_EXCLUSIVE: 367 vp = ap->a_vp; 368 flags = ap->a_flags; 369 for (;;) { 370#ifdef DEBUG_VFS_LOCKS 371 KASSERT(vp->v_holdcnt != 0, 372 ("ffs_lock %p: zero hold count", vp)); 373#endif 374 lkp = vp->v_vnlock; 375 result = _lockmgr_args(lkp, flags, VI_MTX(vp), 376 LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT, 377 ap->a_file, ap->a_line); 378 if (lkp == vp->v_vnlock || result != 0) 379 break; 380 /* 381 * Apparent success, except that the vnode 382 * mutated between snapshot file vnode and 383 * regular file vnode while this process 384 * slept. The lock currently held is not the 385 * right lock. Release it, and try to get the 386 * new lock. 387 */ 388 (void) _lockmgr_args(lkp, LK_RELEASE, NULL, 389 LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT, 390 ap->a_file, ap->a_line); 391 if ((flags & (LK_INTERLOCK | LK_NOWAIT)) == 392 (LK_INTERLOCK | LK_NOWAIT)) 393 return (EBUSY); 394 if ((flags & LK_TYPE_MASK) == LK_UPGRADE) 395 flags = (flags & ~LK_TYPE_MASK) | LK_EXCLUSIVE; 396 flags &= ~LK_INTERLOCK; 397 } 398 break; 399 default: 400 result = VOP_LOCK1_APV(&ufs_vnodeops, ap); 401 } 402 return (result); 403#else 404 return (VOP_LOCK1_APV(&ufs_vnodeops, ap)); 405#endif 406} 407 408/* 409 * Vnode op for reading. 410 */ 411static int 412ffs_read(ap) 413 struct vop_read_args /* { 414 struct vnode *a_vp; 415 struct uio *a_uio; 416 int a_ioflag; 417 struct ucred *a_cred; 418 } */ *ap; 419{ 420 struct vnode *vp; 421 struct inode *ip; 422 struct uio *uio; 423 struct fs *fs; 424 struct buf *bp; 425 ufs_lbn_t lbn, nextlbn; 426 off_t bytesinfile; 427 long size, xfersize, blkoffset; 428 ssize_t orig_resid; 429 int error; 430 int seqcount; 431 int ioflag; 432 433 vp = ap->a_vp; 434 uio = ap->a_uio; 435 ioflag = ap->a_ioflag; 436 if (ap->a_ioflag & IO_EXT) 437#ifdef notyet 438 return (ffs_extread(vp, uio, ioflag)); 439#else 440 panic("ffs_read+IO_EXT"); 441#endif 442#ifdef DIRECTIO 443 if ((ioflag & IO_DIRECT) != 0) { 444 int workdone; 445 446 error = ffs_rawread(vp, uio, &workdone); 447 if (error != 0 || workdone != 0) 448 return error; 449 } 450#endif 451 452 seqcount = ap->a_ioflag >> IO_SEQSHIFT; 453 ip = VTOI(vp); 454 455#ifdef INVARIANTS 456 if (uio->uio_rw != UIO_READ) 457 panic("ffs_read: mode"); 458 459 if (vp->v_type == VLNK) { 460 if ((int)ip->i_size < vp->v_mount->mnt_maxsymlinklen) 461 panic("ffs_read: short symlink"); 462 } else if (vp->v_type != VREG && vp->v_type != VDIR) 463 panic("ffs_read: type %d", vp->v_type); 464#endif 465 orig_resid = uio->uio_resid; 466 KASSERT(orig_resid >= 0, ("ffs_read: uio->uio_resid < 0")); 467 if (orig_resid == 0) 468 return (0); 469 KASSERT(uio->uio_offset >= 0, ("ffs_read: uio->uio_offset < 0")); 470 fs = ip->i_fs; 471 if (uio->uio_offset < ip->i_size && 472 uio->uio_offset >= fs->fs_maxfilesize) 473 return (EOVERFLOW); 474 475 for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) { 476 if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0) 477 break; 478 lbn = lblkno(fs, uio->uio_offset); 479 nextlbn = lbn + 1; 480 481 /* 482 * size of buffer. The buffer representing the 483 * end of the file is rounded up to the size of 484 * the block type ( fragment or full block, 485 * depending ). 486 */ 487 size = blksize(fs, ip, lbn); 488 blkoffset = blkoff(fs, uio->uio_offset); 489 490 /* 491 * The amount we want to transfer in this iteration is 492 * one FS block less the amount of the data before 493 * our startpoint (duh!) 494 */ 495 xfersize = fs->fs_bsize - blkoffset; 496 497 /* 498 * But if we actually want less than the block, 499 * or the file doesn't have a whole block more of data, 500 * then use the lesser number. 501 */ 502 if (uio->uio_resid < xfersize) 503 xfersize = uio->uio_resid; 504 if (bytesinfile < xfersize) 505 xfersize = bytesinfile; 506 507 if (lblktosize(fs, nextlbn) >= ip->i_size) { 508 /* 509 * Don't do readahead if this is the end of the file. 510 */ 511 error = bread_gb(vp, lbn, size, NOCRED, 512 GB_UNMAPPED, &bp); 513 } else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) { 514 /* 515 * Otherwise if we are allowed to cluster, 516 * grab as much as we can. 517 * 518 * XXX This may not be a win if we are not 519 * doing sequential access. 520 */ 521 error = cluster_read_gb(vp, ip->i_size, lbn, 522 size, NOCRED, blkoffset + uio->uio_resid, 523 seqcount, GB_UNMAPPED, &bp); 524 } else if (seqcount > 1) { 525 /* 526 * If we are NOT allowed to cluster, then 527 * if we appear to be acting sequentially, 528 * fire off a request for a readahead 529 * as well as a read. Note that the 4th and 5th 530 * arguments point to arrays of the size specified in 531 * the 6th argument. 532 */ 533 u_int nextsize = blksize(fs, ip, nextlbn); 534 error = breadn_flags(vp, lbn, size, &nextlbn, 535 &nextsize, 1, NOCRED, GB_UNMAPPED, &bp); 536 } else { 537 /* 538 * Failing all of the above, just read what the 539 * user asked for. Interestingly, the same as 540 * the first option above. 541 */ 542 error = bread_gb(vp, lbn, size, NOCRED, 543 GB_UNMAPPED, &bp); 544 } 545 if (error) { 546 brelse(bp); 547 bp = NULL; 548 break; 549 } 550 551 /* 552 * If IO_DIRECT then set B_DIRECT for the buffer. This 553 * will cause us to attempt to release the buffer later on 554 * and will cause the buffer cache to attempt to free the 555 * underlying pages. 556 */ 557 if (ioflag & IO_DIRECT) 558 bp->b_flags |= B_DIRECT; 559 560 /* 561 * We should only get non-zero b_resid when an I/O error 562 * has occurred, which should cause us to break above. 563 * However, if the short read did not cause an error, 564 * then we want to ensure that we do not uiomove bad 565 * or uninitialized data. 566 */ 567 size -= bp->b_resid; 568 if (size < xfersize) { 569 if (size == 0) 570 break; 571 xfersize = size; 572 } 573 574 if ((bp->b_flags & B_UNMAPPED) == 0) { 575 error = vn_io_fault_uiomove((char *)bp->b_data + 576 blkoffset, (int)xfersize, uio); 577 } else { 578 error = vn_io_fault_pgmove(bp->b_pages, blkoffset, 579 (int)xfersize, uio); 580 } 581 if (error) 582 break; 583 584 if ((ioflag & (IO_VMIO|IO_DIRECT)) && 585 (LIST_EMPTY(&bp->b_dep))) { 586 /* 587 * If there are no dependencies, and it's VMIO, 588 * then we don't need the buf, mark it available 589 * for freeing. For non-direct VMIO reads, the VM 590 * has the data. 591 */ 592 bp->b_flags |= B_RELBUF; 593 brelse(bp); 594 } else { 595 /* 596 * Otherwise let whoever 597 * made the request take care of 598 * freeing it. We just queue 599 * it onto another list. 600 */ 601 bqrelse(bp); 602 } 603 } 604 605 /* 606 * This can only happen in the case of an error 607 * because the loop above resets bp to NULL on each iteration 608 * and on normal completion has not set a new value into it. 609 * so it must have come from a 'break' statement 610 */ 611 if (bp != NULL) { 612 if ((ioflag & (IO_VMIO|IO_DIRECT)) && 613 (LIST_EMPTY(&bp->b_dep))) { 614 bp->b_flags |= B_RELBUF; 615 brelse(bp); 616 } else { 617 bqrelse(bp); 618 } 619 } 620 621 if ((error == 0 || uio->uio_resid != orig_resid) && 622 (vp->v_mount->mnt_flag & MNT_NOATIME) == 0 && 623 (ip->i_flag & IN_ACCESS) == 0) { 624 VI_LOCK(vp); 625 ip->i_flag |= IN_ACCESS; 626 VI_UNLOCK(vp); 627 } 628 return (error); 629} 630 631/* 632 * Vnode op for writing. 633 */ 634static int 635ffs_write(ap) 636 struct vop_write_args /* { 637 struct vnode *a_vp; 638 struct uio *a_uio; 639 int a_ioflag; 640 struct ucred *a_cred; 641 } */ *ap; 642{ 643 struct vnode *vp; 644 struct uio *uio; 645 struct inode *ip; 646 struct fs *fs; 647 struct buf *bp; 648 ufs_lbn_t lbn; 649 off_t osize; 650 ssize_t resid; 651 int seqcount; 652 int blkoffset, error, flags, ioflag, size, xfersize; 653 654 vp = ap->a_vp; 655 uio = ap->a_uio; 656 ioflag = ap->a_ioflag; 657 if (ap->a_ioflag & IO_EXT) 658#ifdef notyet 659 return (ffs_extwrite(vp, uio, ioflag, ap->a_cred)); 660#else 661 panic("ffs_write+IO_EXT"); 662#endif 663 664 seqcount = ap->a_ioflag >> IO_SEQSHIFT; 665 ip = VTOI(vp); 666 667#ifdef INVARIANTS 668 if (uio->uio_rw != UIO_WRITE) 669 panic("ffs_write: mode"); 670#endif 671 672 switch (vp->v_type) { 673 case VREG: 674 if (ioflag & IO_APPEND) 675 uio->uio_offset = ip->i_size; 676 if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size) 677 return (EPERM); 678 /* FALLTHROUGH */ 679 case VLNK: 680 break; 681 case VDIR: 682 panic("ffs_write: dir write"); 683 break; 684 default: 685 panic("ffs_write: type %p %d (%d,%d)", vp, (int)vp->v_type, 686 (int)uio->uio_offset, 687 (int)uio->uio_resid 688 ); 689 } 690 691 KASSERT(uio->uio_resid >= 0, ("ffs_write: uio->uio_resid < 0")); 692 KASSERT(uio->uio_offset >= 0, ("ffs_write: uio->uio_offset < 0")); 693 fs = ip->i_fs; 694 if ((uoff_t)uio->uio_offset + uio->uio_resid > fs->fs_maxfilesize) 695 return (EFBIG); 696 /* 697 * Maybe this should be above the vnode op call, but so long as 698 * file servers have no limits, I don't think it matters. 699 */ 700 if (vn_rlimit_fsize(vp, uio, uio->uio_td)) 701 return (EFBIG); 702 703 resid = uio->uio_resid; 704 osize = ip->i_size; 705 if (seqcount > BA_SEQMAX) 706 flags = BA_SEQMAX << BA_SEQSHIFT; 707 else 708 flags = seqcount << BA_SEQSHIFT; 709 if ((ioflag & IO_SYNC) && !DOINGASYNC(vp)) 710 flags |= IO_SYNC; 711 flags |= BA_UNMAPPED; 712 713 for (error = 0; uio->uio_resid > 0;) { 714 lbn = lblkno(fs, uio->uio_offset); 715 blkoffset = blkoff(fs, uio->uio_offset); 716 xfersize = fs->fs_bsize - blkoffset; 717 if (uio->uio_resid < xfersize) 718 xfersize = uio->uio_resid; 719 if (uio->uio_offset + xfersize > ip->i_size) 720 vnode_pager_setsize(vp, uio->uio_offset + xfersize); 721 722 /* 723 * We must perform a read-before-write if the transfer size 724 * does not cover the entire buffer. 725 */ 726 if (fs->fs_bsize > xfersize) 727 flags |= BA_CLRBUF; 728 else 729 flags &= ~BA_CLRBUF; 730/* XXX is uio->uio_offset the right thing here? */ 731 error = UFS_BALLOC(vp, uio->uio_offset, xfersize, 732 ap->a_cred, flags, &bp); 733 if (error != 0) { 734 vnode_pager_setsize(vp, ip->i_size); 735 break; 736 } 737 if (ioflag & IO_DIRECT) 738 bp->b_flags |= B_DIRECT; 739 if ((ioflag & (IO_SYNC|IO_INVAL)) == (IO_SYNC|IO_INVAL)) 740 bp->b_flags |= B_NOCACHE; 741 742 if (uio->uio_offset + xfersize > ip->i_size) { 743 ip->i_size = uio->uio_offset + xfersize; 744 DIP_SET(ip, i_size, ip->i_size); 745 } 746 747 size = blksize(fs, ip, lbn) - bp->b_resid; 748 if (size < xfersize) 749 xfersize = size; 750 751 if ((bp->b_flags & B_UNMAPPED) == 0) { 752 error = vn_io_fault_uiomove((char *)bp->b_data + 753 blkoffset, (int)xfersize, uio); 754 } else { 755 error = vn_io_fault_pgmove(bp->b_pages, blkoffset, 756 (int)xfersize, uio); 757 } 758 /* 759 * If the buffer is not already filled and we encounter an 760 * error while trying to fill it, we have to clear out any 761 * garbage data from the pages instantiated for the buffer. 762 * If we do not, a failed uiomove() during a write can leave 763 * the prior contents of the pages exposed to a userland mmap. 764 * 765 * Note that we need only clear buffers with a transfer size 766 * equal to the block size because buffers with a shorter 767 * transfer size were cleared above by the call to UFS_BALLOC() 768 * with the BA_CLRBUF flag set. 769 * 770 * If the source region for uiomove identically mmaps the 771 * buffer, uiomove() performed the NOP copy, and the buffer 772 * content remains valid because the page fault handler 773 * validated the pages. 774 */ 775 if (error != 0 && (bp->b_flags & B_CACHE) == 0 && 776 fs->fs_bsize == xfersize) 777 vfs_bio_clrbuf(bp); 778 if ((ioflag & (IO_VMIO|IO_DIRECT)) && 779 (LIST_EMPTY(&bp->b_dep))) { 780 bp->b_flags |= B_RELBUF; 781 } 782 783 /* 784 * If IO_SYNC each buffer is written synchronously. Otherwise 785 * if we have a severe page deficiency write the buffer 786 * asynchronously. Otherwise try to cluster, and if that 787 * doesn't do it then either do an async write (if O_DIRECT), 788 * or a delayed write (if not). 789 */ 790 if (ioflag & IO_SYNC) { 791 (void)bwrite(bp); 792 } else if (vm_page_count_severe() || 793 buf_dirty_count_severe() || 794 (ioflag & IO_ASYNC)) { 795 bp->b_flags |= B_CLUSTEROK; 796 bawrite(bp); 797 } else if (xfersize + blkoffset == fs->fs_bsize) { 798 if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) { 799 bp->b_flags |= B_CLUSTEROK; 800 cluster_write_gb(vp, bp, ip->i_size, seqcount, 801 GB_UNMAPPED); 802 } else { 803 bawrite(bp); 804 } 805 } else if (ioflag & IO_DIRECT) { 806 bp->b_flags |= B_CLUSTEROK; 807 bawrite(bp); 808 } else { 809 bp->b_flags |= B_CLUSTEROK; 810 bdwrite(bp); 811 } 812 if (error || xfersize == 0) 813 break; 814 ip->i_flag |= IN_CHANGE | IN_UPDATE; 815 } 816 /* 817 * If we successfully wrote any data, and we are not the superuser 818 * we clear the setuid and setgid bits as a precaution against 819 * tampering. 820 */ 821 if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid && 822 ap->a_cred) { 823 if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0)) { 824 ip->i_mode &= ~(ISUID | ISGID); 825 DIP_SET(ip, i_mode, ip->i_mode); 826 } 827 } 828 if (error) { 829 if (ioflag & IO_UNIT) { 830 (void)ffs_truncate(vp, osize, 831 IO_NORMAL | (ioflag & IO_SYNC), 832 ap->a_cred, uio->uio_td); 833 uio->uio_offset -= resid - uio->uio_resid; 834 uio->uio_resid = resid; 835 } 836 } else if (resid > uio->uio_resid && (ioflag & IO_SYNC)) 837 error = ffs_update(vp, 1); 838 return (error); 839} 840 841/* 842 * get page routine 843 */ 844static int 845ffs_getpages(ap) 846 struct vop_getpages_args *ap; 847{ 848 int i; 849 vm_page_t mreq; 850 int pcount; 851 852 pcount = round_page(ap->a_count) / PAGE_SIZE; 853 mreq = ap->a_m[ap->a_reqpage]; 854 855 /* 856 * if ANY DEV_BSIZE blocks are valid on a large filesystem block, 857 * then the entire page is valid. Since the page may be mapped, 858 * user programs might reference data beyond the actual end of file 859 * occuring within the page. We have to zero that data. 860 */ 861 VM_OBJECT_LOCK(mreq->object); 862 if (mreq->valid) { 863 if (mreq->valid != VM_PAGE_BITS_ALL) 864 vm_page_zero_invalid(mreq, TRUE); 865 for (i = 0; i < pcount; i++) { 866 if (i != ap->a_reqpage) { 867 vm_page_lock(ap->a_m[i]); 868 vm_page_free(ap->a_m[i]); 869 vm_page_unlock(ap->a_m[i]); 870 } 871 } 872 VM_OBJECT_UNLOCK(mreq->object); 873 return VM_PAGER_OK; 874 } 875 VM_OBJECT_UNLOCK(mreq->object); 876 877 return vnode_pager_generic_getpages(ap->a_vp, ap->a_m, 878 ap->a_count, 879 ap->a_reqpage); 880} 881 882 883/* 884 * Extended attribute area reading. 885 */ 886static int 887ffs_extread(struct vnode *vp, struct uio *uio, int ioflag) 888{ 889 struct inode *ip; 890 struct ufs2_dinode *dp; 891 struct fs *fs; 892 struct buf *bp; 893 ufs_lbn_t lbn, nextlbn; 894 off_t bytesinfile; 895 long size, xfersize, blkoffset; 896 ssize_t orig_resid; 897 int error; 898 899 ip = VTOI(vp); 900 fs = ip->i_fs; 901 dp = ip->i_din2; 902 903#ifdef INVARIANTS 904 if (uio->uio_rw != UIO_READ || fs->fs_magic != FS_UFS2_MAGIC) 905 panic("ffs_extread: mode"); 906 907#endif 908 orig_resid = uio->uio_resid; 909 KASSERT(orig_resid >= 0, ("ffs_extread: uio->uio_resid < 0")); 910 if (orig_resid == 0) 911 return (0); 912 KASSERT(uio->uio_offset >= 0, ("ffs_extread: uio->uio_offset < 0")); 913 914 for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) { 915 if ((bytesinfile = dp->di_extsize - uio->uio_offset) <= 0) 916 break; 917 lbn = lblkno(fs, uio->uio_offset); 918 nextlbn = lbn + 1; 919 920 /* 921 * size of buffer. The buffer representing the 922 * end of the file is rounded up to the size of 923 * the block type ( fragment or full block, 924 * depending ). 925 */ 926 size = sblksize(fs, dp->di_extsize, lbn); 927 blkoffset = blkoff(fs, uio->uio_offset); 928 929 /* 930 * The amount we want to transfer in this iteration is 931 * one FS block less the amount of the data before 932 * our startpoint (duh!) 933 */ 934 xfersize = fs->fs_bsize - blkoffset; 935 936 /* 937 * But if we actually want less than the block, 938 * or the file doesn't have a whole block more of data, 939 * then use the lesser number. 940 */ 941 if (uio->uio_resid < xfersize) 942 xfersize = uio->uio_resid; 943 if (bytesinfile < xfersize) 944 xfersize = bytesinfile; 945 946 if (lblktosize(fs, nextlbn) >= dp->di_extsize) { 947 /* 948 * Don't do readahead if this is the end of the info. 949 */ 950 error = bread(vp, -1 - lbn, size, NOCRED, &bp); 951 } else { 952 /* 953 * If we have a second block, then 954 * fire off a request for a readahead 955 * as well as a read. Note that the 4th and 5th 956 * arguments point to arrays of the size specified in 957 * the 6th argument. 958 */ 959 u_int nextsize = sblksize(fs, dp->di_extsize, nextlbn); 960 961 nextlbn = -1 - nextlbn; 962 error = breadn(vp, -1 - lbn, 963 size, &nextlbn, &nextsize, 1, NOCRED, &bp); 964 } 965 if (error) { 966 brelse(bp); 967 bp = NULL; 968 break; 969 } 970 971 /* 972 * If IO_DIRECT then set B_DIRECT for the buffer. This 973 * will cause us to attempt to release the buffer later on 974 * and will cause the buffer cache to attempt to free the 975 * underlying pages. 976 */ 977 if (ioflag & IO_DIRECT) 978 bp->b_flags |= B_DIRECT; 979 980 /* 981 * We should only get non-zero b_resid when an I/O error 982 * has occurred, which should cause us to break above. 983 * However, if the short read did not cause an error, 984 * then we want to ensure that we do not uiomove bad 985 * or uninitialized data. 986 */ 987 size -= bp->b_resid; 988 if (size < xfersize) { 989 if (size == 0) 990 break; 991 xfersize = size; 992 } 993 994 error = uiomove((char *)bp->b_data + blkoffset, 995 (int)xfersize, uio); 996 if (error) 997 break; 998 999 if ((ioflag & (IO_VMIO|IO_DIRECT)) && 1000 (LIST_EMPTY(&bp->b_dep))) { 1001 /* 1002 * If there are no dependencies, and it's VMIO, 1003 * then we don't need the buf, mark it available 1004 * for freeing. For non-direct VMIO reads, the VM 1005 * has the data. 1006 */ 1007 bp->b_flags |= B_RELBUF; 1008 brelse(bp); 1009 } else { 1010 /* 1011 * Otherwise let whoever 1012 * made the request take care of 1013 * freeing it. We just queue 1014 * it onto another list. 1015 */ 1016 bqrelse(bp); 1017 } 1018 } 1019 1020 /* 1021 * This can only happen in the case of an error 1022 * because the loop above resets bp to NULL on each iteration 1023 * and on normal completion has not set a new value into it. 1024 * so it must have come from a 'break' statement 1025 */ 1026 if (bp != NULL) { 1027 if ((ioflag & (IO_VMIO|IO_DIRECT)) && 1028 (LIST_EMPTY(&bp->b_dep))) { 1029 bp->b_flags |= B_RELBUF; 1030 brelse(bp); 1031 } else { 1032 bqrelse(bp); 1033 } 1034 } 1035 return (error); 1036} 1037 1038/* 1039 * Extended attribute area writing. 1040 */ 1041static int 1042ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag, struct ucred *ucred) 1043{ 1044 struct inode *ip; 1045 struct ufs2_dinode *dp; 1046 struct fs *fs; 1047 struct buf *bp; 1048 ufs_lbn_t lbn; 1049 off_t osize; 1050 ssize_t resid; 1051 int blkoffset, error, flags, size, xfersize; 1052 1053 ip = VTOI(vp); 1054 fs = ip->i_fs; 1055 dp = ip->i_din2; 1056 1057#ifdef INVARIANTS 1058 if (uio->uio_rw != UIO_WRITE || fs->fs_magic != FS_UFS2_MAGIC) 1059 panic("ffs_extwrite: mode"); 1060#endif 1061 1062 if (ioflag & IO_APPEND) 1063 uio->uio_offset = dp->di_extsize; 1064 KASSERT(uio->uio_offset >= 0, ("ffs_extwrite: uio->uio_offset < 0")); 1065 KASSERT(uio->uio_resid >= 0, ("ffs_extwrite: uio->uio_resid < 0")); 1066 if ((uoff_t)uio->uio_offset + uio->uio_resid > NXADDR * fs->fs_bsize) 1067 return (EFBIG); 1068 1069 resid = uio->uio_resid; 1070 osize = dp->di_extsize; 1071 flags = IO_EXT; 1072 if ((ioflag & IO_SYNC) && !DOINGASYNC(vp)) 1073 flags |= IO_SYNC; 1074 1075 for (error = 0; uio->uio_resid > 0;) { 1076 lbn = lblkno(fs, uio->uio_offset); 1077 blkoffset = blkoff(fs, uio->uio_offset); 1078 xfersize = fs->fs_bsize - blkoffset; 1079 if (uio->uio_resid < xfersize) 1080 xfersize = uio->uio_resid; 1081 1082 /* 1083 * We must perform a read-before-write if the transfer size 1084 * does not cover the entire buffer. 1085 */ 1086 if (fs->fs_bsize > xfersize) 1087 flags |= BA_CLRBUF; 1088 else 1089 flags &= ~BA_CLRBUF; 1090 error = UFS_BALLOC(vp, uio->uio_offset, xfersize, 1091 ucred, flags, &bp); 1092 if (error != 0) 1093 break; 1094 /* 1095 * If the buffer is not valid we have to clear out any 1096 * garbage data from the pages instantiated for the buffer. 1097 * If we do not, a failed uiomove() during a write can leave 1098 * the prior contents of the pages exposed to a userland 1099 * mmap(). XXX deal with uiomove() errors a better way. 1100 */ 1101 if ((bp->b_flags & B_CACHE) == 0 && fs->fs_bsize <= xfersize) 1102 vfs_bio_clrbuf(bp); 1103 if (ioflag & IO_DIRECT) 1104 bp->b_flags |= B_DIRECT; 1105 1106 if (uio->uio_offset + xfersize > dp->di_extsize) 1107 dp->di_extsize = uio->uio_offset + xfersize; 1108 1109 size = sblksize(fs, dp->di_extsize, lbn) - bp->b_resid; 1110 if (size < xfersize) 1111 xfersize = size; 1112 1113 error = 1114 uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio); 1115 if ((ioflag & (IO_VMIO|IO_DIRECT)) && 1116 (LIST_EMPTY(&bp->b_dep))) { 1117 bp->b_flags |= B_RELBUF; 1118 } 1119 1120 /* 1121 * If IO_SYNC each buffer is written synchronously. Otherwise 1122 * if we have a severe page deficiency write the buffer 1123 * asynchronously. Otherwise try to cluster, and if that 1124 * doesn't do it then either do an async write (if O_DIRECT), 1125 * or a delayed write (if not). 1126 */ 1127 if (ioflag & IO_SYNC) { 1128 (void)bwrite(bp); 1129 } else if (vm_page_count_severe() || 1130 buf_dirty_count_severe() || 1131 xfersize + blkoffset == fs->fs_bsize || 1132 (ioflag & (IO_ASYNC | IO_DIRECT))) 1133 bawrite(bp); 1134 else 1135 bdwrite(bp); 1136 if (error || xfersize == 0) 1137 break; 1138 ip->i_flag |= IN_CHANGE; 1139 } 1140 /* 1141 * If we successfully wrote any data, and we are not the superuser 1142 * we clear the setuid and setgid bits as a precaution against 1143 * tampering. 1144 */ 1145 if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid && ucred) { 1146 if (priv_check_cred(ucred, PRIV_VFS_RETAINSUGID, 0)) { 1147 ip->i_mode &= ~(ISUID | ISGID); 1148 dp->di_mode = ip->i_mode; 1149 } 1150 } 1151 if (error) { 1152 if (ioflag & IO_UNIT) { 1153 (void)ffs_truncate(vp, osize, 1154 IO_EXT | (ioflag&IO_SYNC), ucred, uio->uio_td); 1155 uio->uio_offset -= resid - uio->uio_resid; 1156 uio->uio_resid = resid; 1157 } 1158 } else if (resid > uio->uio_resid && (ioflag & IO_SYNC)) 1159 error = ffs_update(vp, 1); 1160 return (error); 1161} 1162 1163 1164/* 1165 * Vnode operating to retrieve a named extended attribute. 1166 * 1167 * Locate a particular EA (nspace:name) in the area (ptr:length), and return 1168 * the length of the EA, and possibly the pointer to the entry and to the data. 1169 */ 1170static int 1171ffs_findextattr(u_char *ptr, u_int length, int nspace, const char *name, u_char **eap, u_char **eac) 1172{ 1173 u_char *p, *pe, *pn, *p0; 1174 int eapad1, eapad2, ealength, ealen, nlen; 1175 uint32_t ul; 1176 1177 pe = ptr + length; 1178 nlen = strlen(name); 1179 1180 for (p = ptr; p < pe; p = pn) { 1181 p0 = p; 1182 bcopy(p, &ul, sizeof(ul)); 1183 pn = p + ul; 1184 /* make sure this entry is complete */ 1185 if (pn > pe) 1186 break; 1187 p += sizeof(uint32_t); 1188 if (*p != nspace) 1189 continue; 1190 p++; 1191 eapad2 = *p++; 1192 if (*p != nlen) 1193 continue; 1194 p++; 1195 if (bcmp(p, name, nlen)) 1196 continue; 1197 ealength = sizeof(uint32_t) + 3 + nlen; 1198 eapad1 = 8 - (ealength % 8); 1199 if (eapad1 == 8) 1200 eapad1 = 0; 1201 ealength += eapad1; 1202 ealen = ul - ealength - eapad2; 1203 p += nlen + eapad1; 1204 if (eap != NULL) 1205 *eap = p0; 1206 if (eac != NULL) 1207 *eac = p; 1208 return (ealen); 1209 } 1210 return(-1); 1211} 1212 1213static int 1214ffs_rdextattr(u_char **p, struct vnode *vp, struct thread *td, int extra) 1215{ 1216 struct inode *ip; 1217 struct ufs2_dinode *dp; 1218 struct fs *fs; 1219 struct uio luio; 1220 struct iovec liovec; 1221 u_int easize; 1222 int error; 1223 u_char *eae; 1224 1225 ip = VTOI(vp); 1226 fs = ip->i_fs; 1227 dp = ip->i_din2; 1228 easize = dp->di_extsize; 1229 if ((uoff_t)easize + extra > NXADDR * fs->fs_bsize) 1230 return (EFBIG); 1231 1232 eae = malloc(easize + extra, M_TEMP, M_WAITOK); 1233 1234 liovec.iov_base = eae; 1235 liovec.iov_len = easize; 1236 luio.uio_iov = &liovec; 1237 luio.uio_iovcnt = 1; 1238 luio.uio_offset = 0; 1239 luio.uio_resid = easize; 1240 luio.uio_segflg = UIO_SYSSPACE; 1241 luio.uio_rw = UIO_READ; 1242 luio.uio_td = td; 1243 1244 error = ffs_extread(vp, &luio, IO_EXT | IO_SYNC); 1245 if (error) { 1246 free(eae, M_TEMP); 1247 return(error); 1248 } 1249 *p = eae; 1250 return (0); 1251} 1252 1253static void 1254ffs_lock_ea(struct vnode *vp) 1255{ 1256 struct inode *ip; 1257 1258 ip = VTOI(vp); 1259 VI_LOCK(vp); 1260 while (ip->i_flag & IN_EA_LOCKED) { 1261 ip->i_flag |= IN_EA_LOCKWAIT; 1262 msleep(&ip->i_ea_refs, &vp->v_interlock, PINOD + 2, "ufs_ea", 1263 0); 1264 } 1265 ip->i_flag |= IN_EA_LOCKED; 1266 VI_UNLOCK(vp); 1267} 1268 1269static void 1270ffs_unlock_ea(struct vnode *vp) 1271{ 1272 struct inode *ip; 1273 1274 ip = VTOI(vp); 1275 VI_LOCK(vp); 1276 if (ip->i_flag & IN_EA_LOCKWAIT) 1277 wakeup(&ip->i_ea_refs); 1278 ip->i_flag &= ~(IN_EA_LOCKED | IN_EA_LOCKWAIT); 1279 VI_UNLOCK(vp); 1280} 1281 1282static int 1283ffs_open_ea(struct vnode *vp, struct ucred *cred, struct thread *td) 1284{ 1285 struct inode *ip; 1286 struct ufs2_dinode *dp; 1287 int error; 1288 1289 ip = VTOI(vp); 1290 1291 ffs_lock_ea(vp); 1292 if (ip->i_ea_area != NULL) { 1293 ip->i_ea_refs++; 1294 ffs_unlock_ea(vp); 1295 return (0); 1296 } 1297 dp = ip->i_din2; 1298 error = ffs_rdextattr(&ip->i_ea_area, vp, td, 0); 1299 if (error) { 1300 ffs_unlock_ea(vp); 1301 return (error); 1302 } 1303 ip->i_ea_len = dp->di_extsize; 1304 ip->i_ea_error = 0; 1305 ip->i_ea_refs++; 1306 ffs_unlock_ea(vp); 1307 return (0); 1308} 1309 1310/* 1311 * Vnode extattr transaction commit/abort 1312 */ 1313static int 1314ffs_close_ea(struct vnode *vp, int commit, struct ucred *cred, struct thread *td) 1315{ 1316 struct inode *ip; 1317 struct uio luio; 1318 struct iovec liovec; 1319 int error; 1320 struct ufs2_dinode *dp; 1321 1322 ip = VTOI(vp); 1323 1324 ffs_lock_ea(vp); 1325 if (ip->i_ea_area == NULL) { 1326 ffs_unlock_ea(vp); 1327 return (EINVAL); 1328 } 1329 dp = ip->i_din2; 1330 error = ip->i_ea_error; 1331 if (commit && error == 0) { 1332 ASSERT_VOP_ELOCKED(vp, "ffs_close_ea commit"); 1333 if (cred == NOCRED) 1334 cred = vp->v_mount->mnt_cred; 1335 liovec.iov_base = ip->i_ea_area; 1336 liovec.iov_len = ip->i_ea_len; 1337 luio.uio_iov = &liovec; 1338 luio.uio_iovcnt = 1; 1339 luio.uio_offset = 0; 1340 luio.uio_resid = ip->i_ea_len; 1341 luio.uio_segflg = UIO_SYSSPACE; 1342 luio.uio_rw = UIO_WRITE; 1343 luio.uio_td = td; 1344 /* XXX: I'm not happy about truncating to zero size */ 1345 if (ip->i_ea_len < dp->di_extsize) 1346 error = ffs_truncate(vp, 0, IO_EXT, cred, td); 1347 error = ffs_extwrite(vp, &luio, IO_EXT | IO_SYNC, cred); 1348 } 1349 if (--ip->i_ea_refs == 0) { 1350 free(ip->i_ea_area, M_TEMP); 1351 ip->i_ea_area = NULL; 1352 ip->i_ea_len = 0; 1353 ip->i_ea_error = 0; 1354 } 1355 ffs_unlock_ea(vp); 1356 return (error); 1357} 1358 1359/* 1360 * Vnode extattr strategy routine for fifos. 1361 * 1362 * We need to check for a read or write of the external attributes. 1363 * Otherwise we just fall through and do the usual thing. 1364 */ 1365static int 1366ffsext_strategy(struct vop_strategy_args *ap) 1367/* 1368struct vop_strategy_args { 1369 struct vnodeop_desc *a_desc; 1370 struct vnode *a_vp; 1371 struct buf *a_bp; 1372}; 1373*/ 1374{ 1375 struct vnode *vp; 1376 daddr_t lbn; 1377 1378 vp = ap->a_vp; 1379 lbn = ap->a_bp->b_lblkno; 1380 if (VTOI(vp)->i_fs->fs_magic == FS_UFS2_MAGIC && 1381 lbn < 0 && lbn >= -NXADDR) 1382 return (VOP_STRATEGY_APV(&ufs_vnodeops, ap)); 1383 if (vp->v_type == VFIFO) 1384 return (VOP_STRATEGY_APV(&ufs_fifoops, ap)); 1385 panic("spec nodes went here"); 1386} 1387 1388/* 1389 * Vnode extattr transaction commit/abort 1390 */ 1391static int 1392ffs_openextattr(struct vop_openextattr_args *ap) 1393/* 1394struct vop_openextattr_args { 1395 struct vnodeop_desc *a_desc; 1396 struct vnode *a_vp; 1397 IN struct ucred *a_cred; 1398 IN struct thread *a_td; 1399}; 1400*/ 1401{ 1402 struct inode *ip; 1403 struct fs *fs; 1404 1405 ip = VTOI(ap->a_vp); 1406 fs = ip->i_fs; 1407 1408 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK) 1409 return (EOPNOTSUPP); 1410 1411 return (ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td)); 1412} 1413 1414 1415/* 1416 * Vnode extattr transaction commit/abort 1417 */ 1418static int 1419ffs_closeextattr(struct vop_closeextattr_args *ap) 1420/* 1421struct vop_closeextattr_args { 1422 struct vnodeop_desc *a_desc; 1423 struct vnode *a_vp; 1424 int a_commit; 1425 IN struct ucred *a_cred; 1426 IN struct thread *a_td; 1427}; 1428*/ 1429{ 1430 struct inode *ip; 1431 struct fs *fs; 1432 1433 ip = VTOI(ap->a_vp); 1434 fs = ip->i_fs; 1435 1436 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK) 1437 return (EOPNOTSUPP); 1438 1439 if (ap->a_commit && (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)) 1440 return (EROFS); 1441 1442 return (ffs_close_ea(ap->a_vp, ap->a_commit, ap->a_cred, ap->a_td)); 1443} 1444 1445/* 1446 * Vnode operation to remove a named attribute. 1447 */ 1448static int 1449ffs_deleteextattr(struct vop_deleteextattr_args *ap) 1450/* 1451vop_deleteextattr { 1452 IN struct vnode *a_vp; 1453 IN int a_attrnamespace; 1454 IN const char *a_name; 1455 IN struct ucred *a_cred; 1456 IN struct thread *a_td; 1457}; 1458*/ 1459{ 1460 struct inode *ip; 1461 struct fs *fs; 1462 uint32_t ealength, ul; 1463 int ealen, olen, eapad1, eapad2, error, i, easize; 1464 u_char *eae, *p; 1465 1466 ip = VTOI(ap->a_vp); 1467 fs = ip->i_fs; 1468 1469 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK) 1470 return (EOPNOTSUPP); 1471 1472 if (strlen(ap->a_name) == 0) 1473 return (EINVAL); 1474 1475 if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY) 1476 return (EROFS); 1477 1478 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 1479 ap->a_cred, ap->a_td, VWRITE); 1480 if (error) { 1481 1482 /* 1483 * ffs_lock_ea is not needed there, because the vnode 1484 * must be exclusively locked. 1485 */ 1486 if (ip->i_ea_area != NULL && ip->i_ea_error == 0) 1487 ip->i_ea_error = error; 1488 return (error); 1489 } 1490 1491 error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td); 1492 if (error) 1493 return (error); 1494 1495 ealength = eapad1 = ealen = eapad2 = 0; 1496 1497 eae = malloc(ip->i_ea_len, M_TEMP, M_WAITOK); 1498 bcopy(ip->i_ea_area, eae, ip->i_ea_len); 1499 easize = ip->i_ea_len; 1500 1501 olen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name, 1502 &p, NULL); 1503 if (olen == -1) { 1504 /* delete but nonexistent */ 1505 free(eae, M_TEMP); 1506 ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td); 1507 return(ENOATTR); 1508 } 1509 bcopy(p, &ul, sizeof ul); 1510 i = p - eae + ul; 1511 if (ul != ealength) { 1512 bcopy(p + ul, p + ealength, easize - i); 1513 easize += (ealength - ul); 1514 } 1515 if (easize > NXADDR * fs->fs_bsize) { 1516 free(eae, M_TEMP); 1517 ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td); 1518 if (ip->i_ea_area != NULL && ip->i_ea_error == 0) 1519 ip->i_ea_error = ENOSPC; 1520 return(ENOSPC); 1521 } 1522 p = ip->i_ea_area; 1523 ip->i_ea_area = eae; 1524 ip->i_ea_len = easize; 1525 free(p, M_TEMP); 1526 error = ffs_close_ea(ap->a_vp, 1, ap->a_cred, ap->a_td); 1527 return(error); 1528} 1529 1530/* 1531 * Vnode operation to retrieve a named extended attribute. 1532 */ 1533static int 1534ffs_getextattr(struct vop_getextattr_args *ap) 1535/* 1536vop_getextattr { 1537 IN struct vnode *a_vp; 1538 IN int a_attrnamespace; 1539 IN const char *a_name; 1540 INOUT struct uio *a_uio; 1541 OUT size_t *a_size; 1542 IN struct ucred *a_cred; 1543 IN struct thread *a_td; 1544}; 1545*/ 1546{ 1547 struct inode *ip; 1548 struct fs *fs; 1549 u_char *eae, *p; 1550 unsigned easize; 1551 int error, ealen; 1552 1553 ip = VTOI(ap->a_vp); 1554 fs = ip->i_fs; 1555 1556 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK) 1557 return (EOPNOTSUPP); 1558 1559 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 1560 ap->a_cred, ap->a_td, VREAD); 1561 if (error) 1562 return (error); 1563 1564 error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td); 1565 if (error) 1566 return (error); 1567 1568 eae = ip->i_ea_area; 1569 easize = ip->i_ea_len; 1570 1571 ealen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name, 1572 NULL, &p); 1573 if (ealen >= 0) { 1574 error = 0; 1575 if (ap->a_size != NULL) 1576 *ap->a_size = ealen; 1577 else if (ap->a_uio != NULL) 1578 error = uiomove(p, ealen, ap->a_uio); 1579 } else 1580 error = ENOATTR; 1581 1582 ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td); 1583 return(error); 1584} 1585 1586/* 1587 * Vnode operation to retrieve extended attributes on a vnode. 1588 */ 1589static int 1590ffs_listextattr(struct vop_listextattr_args *ap) 1591/* 1592vop_listextattr { 1593 IN struct vnode *a_vp; 1594 IN int a_attrnamespace; 1595 INOUT struct uio *a_uio; 1596 OUT size_t *a_size; 1597 IN struct ucred *a_cred; 1598 IN struct thread *a_td; 1599}; 1600*/ 1601{ 1602 struct inode *ip; 1603 struct fs *fs; 1604 u_char *eae, *p, *pe, *pn; 1605 unsigned easize; 1606 uint32_t ul; 1607 int error, ealen; 1608 1609 ip = VTOI(ap->a_vp); 1610 fs = ip->i_fs; 1611 1612 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK) 1613 return (EOPNOTSUPP); 1614 1615 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 1616 ap->a_cred, ap->a_td, VREAD); 1617 if (error) 1618 return (error); 1619 1620 error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td); 1621 if (error) 1622 return (error); 1623 eae = ip->i_ea_area; 1624 easize = ip->i_ea_len; 1625 1626 error = 0; 1627 if (ap->a_size != NULL) 1628 *ap->a_size = 0; 1629 pe = eae + easize; 1630 for(p = eae; error == 0 && p < pe; p = pn) { 1631 bcopy(p, &ul, sizeof(ul)); 1632 pn = p + ul; 1633 if (pn > pe) 1634 break; 1635 p += sizeof(ul); 1636 if (*p++ != ap->a_attrnamespace) 1637 continue; 1638 p++; /* pad2 */ 1639 ealen = *p; 1640 if (ap->a_size != NULL) { 1641 *ap->a_size += ealen + 1; 1642 } else if (ap->a_uio != NULL) { 1643 error = uiomove(p, ealen + 1, ap->a_uio); 1644 } 1645 } 1646 ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td); 1647 return(error); 1648} 1649 1650/* 1651 * Vnode operation to set a named attribute. 1652 */ 1653static int 1654ffs_setextattr(struct vop_setextattr_args *ap) 1655/* 1656vop_setextattr { 1657 IN struct vnode *a_vp; 1658 IN int a_attrnamespace; 1659 IN const char *a_name; 1660 INOUT struct uio *a_uio; 1661 IN struct ucred *a_cred; 1662 IN struct thread *a_td; 1663}; 1664*/ 1665{ 1666 struct inode *ip; 1667 struct fs *fs; 1668 uint32_t ealength, ul; 1669 ssize_t ealen; 1670 int olen, eapad1, eapad2, error, i, easize; 1671 u_char *eae, *p; 1672 1673 ip = VTOI(ap->a_vp); 1674 fs = ip->i_fs; 1675 1676 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK) 1677 return (EOPNOTSUPP); 1678 1679 if (strlen(ap->a_name) == 0) 1680 return (EINVAL); 1681 1682 /* XXX Now unsupported API to delete EAs using NULL uio. */ 1683 if (ap->a_uio == NULL) 1684 return (EOPNOTSUPP); 1685 1686 if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY) 1687 return (EROFS); 1688 1689 ealen = ap->a_uio->uio_resid; 1690 if (ealen < 0 || ealen > lblktosize(fs, NXADDR)) 1691 return (EINVAL); 1692 1693 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 1694 ap->a_cred, ap->a_td, VWRITE); 1695 if (error) { 1696 1697 /* 1698 * ffs_lock_ea is not needed there, because the vnode 1699 * must be exclusively locked. 1700 */ 1701 if (ip->i_ea_area != NULL && ip->i_ea_error == 0) 1702 ip->i_ea_error = error; 1703 return (error); 1704 } 1705 1706 error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td); 1707 if (error) 1708 return (error); 1709 1710 ealength = sizeof(uint32_t) + 3 + strlen(ap->a_name); 1711 eapad1 = 8 - (ealength % 8); 1712 if (eapad1 == 8) 1713 eapad1 = 0; 1714 eapad2 = 8 - (ealen % 8); 1715 if (eapad2 == 8) 1716 eapad2 = 0; 1717 ealength += eapad1 + ealen + eapad2; 1718 1719 eae = malloc(ip->i_ea_len + ealength, M_TEMP, M_WAITOK); 1720 bcopy(ip->i_ea_area, eae, ip->i_ea_len); 1721 easize = ip->i_ea_len; 1722 1723 olen = ffs_findextattr(eae, easize, 1724 ap->a_attrnamespace, ap->a_name, &p, NULL); 1725 if (olen == -1) { 1726 /* new, append at end */ 1727 p = eae + easize; 1728 easize += ealength; 1729 } else { 1730 bcopy(p, &ul, sizeof ul); 1731 i = p - eae + ul; 1732 if (ul != ealength) { 1733 bcopy(p + ul, p + ealength, easize - i); 1734 easize += (ealength - ul); 1735 } 1736 } 1737 if (easize > lblktosize(fs, NXADDR)) { 1738 free(eae, M_TEMP); 1739 ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td); 1740 if (ip->i_ea_area != NULL && ip->i_ea_error == 0) 1741 ip->i_ea_error = ENOSPC; 1742 return(ENOSPC); 1743 } 1744 bcopy(&ealength, p, sizeof(ealength)); 1745 p += sizeof(ealength); 1746 *p++ = ap->a_attrnamespace; 1747 *p++ = eapad2; 1748 *p++ = strlen(ap->a_name); 1749 strcpy(p, ap->a_name); 1750 p += strlen(ap->a_name); 1751 bzero(p, eapad1); 1752 p += eapad1; 1753 error = uiomove(p, ealen, ap->a_uio); 1754 if (error) { 1755 free(eae, M_TEMP); 1756 ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td); 1757 if (ip->i_ea_area != NULL && ip->i_ea_error == 0) 1758 ip->i_ea_error = error; 1759 return(error); 1760 } 1761 p += ealen; 1762 bzero(p, eapad2); 1763 1764 p = ip->i_ea_area; 1765 ip->i_ea_area = eae; 1766 ip->i_ea_len = easize; 1767 free(p, M_TEMP); 1768 error = ffs_close_ea(ap->a_vp, 1, ap->a_cred, ap->a_td); 1769 return(error); 1770} 1771 1772/* 1773 * Vnode pointer to File handle 1774 */ 1775static int 1776ffs_vptofh(struct vop_vptofh_args *ap) 1777/* 1778vop_vptofh { 1779 IN struct vnode *a_vp; 1780 IN struct fid *a_fhp; 1781}; 1782*/ 1783{ 1784 struct inode *ip; 1785 struct ufid *ufhp; 1786 1787 ip = VTOI(ap->a_vp); 1788 ufhp = (struct ufid *)ap->a_fhp; 1789 ufhp->ufid_len = sizeof(struct ufid); 1790 ufhp->ufid_ino = ip->i_number; 1791 ufhp->ufid_gen = ip->i_gen; 1792 return (0); 1793} 1794