xref: /freebsd/sys/ufs/ffs/ffs_vnops.c (revision 4798f1e8f28d1bd440987696b3474eb6ca764fde)
1 /*-
2  * SPDX-License-Identifier: (BSD-2-Clause 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  * from: $FreeBSD: .../ufs/ufs_readwrite.c,v 1.96 2002/08/12 09:22:11 phk ...
61  */
62 
63 #include <sys/cdefs.h>
64 #include "opt_directio.h"
65 #include "opt_ffs.h"
66 #include "opt_ufs.h"
67 
68 #include <sys/param.h>
69 #include <sys/bio.h>
70 #include <sys/systm.h>
71 #include <sys/buf.h>
72 #include <sys/conf.h>
73 #include <sys/extattr.h>
74 #include <sys/kernel.h>
75 #include <sys/limits.h>
76 #include <sys/malloc.h>
77 #include <sys/mount.h>
78 #include <sys/priv.h>
79 #include <sys/rwlock.h>
80 #include <sys/stat.h>
81 #include <sys/sysctl.h>
82 #include <sys/vmmeter.h>
83 #include <sys/vnode.h>
84 
85 #include <vm/vm.h>
86 #include <vm/vm_param.h>
87 #include <vm/vm_extern.h>
88 #include <vm/vm_object.h>
89 #include <vm/vm_page.h>
90 #include <vm/vm_pager.h>
91 #include <vm/vnode_pager.h>
92 
93 #include <ufs/ufs/extattr.h>
94 #include <ufs/ufs/quota.h>
95 #include <ufs/ufs/inode.h>
96 #include <ufs/ufs/ufs_extern.h>
97 #include <ufs/ufs/ufsmount.h>
98 #include <ufs/ufs/dir.h>
99 #ifdef UFS_DIRHASH
100 #include <ufs/ufs/dirhash.h>
101 #endif
102 
103 #include <ufs/ffs/fs.h>
104 #include <ufs/ffs/ffs_extern.h>
105 
106 #define	ALIGNED_TO(ptr, s)	\
107 	(((uintptr_t)(ptr) & (_Alignof(s) - 1)) == 0)
108 
109 #ifdef DIRECTIO
110 extern int	ffs_rawread(struct vnode *vp, struct uio *uio, int *workdone);
111 #endif
112 static vop_fdatasync_t	ffs_fdatasync;
113 static vop_fsync_t	ffs_fsync;
114 static vop_getpages_t	ffs_getpages;
115 static vop_getpages_async_t	ffs_getpages_async;
116 static vop_lock1_t	ffs_lock;
117 #ifdef INVARIANTS
118 static vop_unlock_t	ffs_unlock_debug;
119 #endif
120 static vop_read_t	ffs_read;
121 static vop_write_t	ffs_write;
122 static int	ffs_extread(struct vnode *vp, struct uio *uio, int ioflag);
123 static int	ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag,
124 		    struct ucred *cred);
125 static vop_strategy_t	ffsext_strategy;
126 static vop_closeextattr_t	ffs_closeextattr;
127 static vop_deleteextattr_t	ffs_deleteextattr;
128 static vop_getextattr_t	ffs_getextattr;
129 static vop_listextattr_t	ffs_listextattr;
130 static vop_openextattr_t	ffs_openextattr;
131 static vop_setextattr_t	ffs_setextattr;
132 static vop_vptofh_t	ffs_vptofh;
133 static vop_vput_pair_t	ffs_vput_pair;
134 
135 vop_fplookup_vexec_t ufs_fplookup_vexec;
136 
137 /* Global vfs data structures for ufs. */
138 struct vop_vector ffs_vnodeops1 = {
139 	.vop_default =		&ufs_vnodeops,
140 	.vop_fsync =		ffs_fsync,
141 	.vop_fdatasync =	ffs_fdatasync,
142 	.vop_getpages =		ffs_getpages,
143 	.vop_getpages_async =	ffs_getpages_async,
144 	.vop_lock1 =		ffs_lock,
145 #ifdef INVARIANTS
146 	.vop_unlock =		ffs_unlock_debug,
147 #endif
148 	.vop_read =		ffs_read,
149 	.vop_reallocblks =	ffs_reallocblks,
150 	.vop_write =		ffs_write,
151 	.vop_vptofh =		ffs_vptofh,
152 	.vop_vput_pair =	ffs_vput_pair,
153 	.vop_fplookup_vexec =	ufs_fplookup_vexec,
154 	.vop_fplookup_symlink =	VOP_EAGAIN,
155 };
156 VFS_VOP_VECTOR_REGISTER(ffs_vnodeops1);
157 
158 struct vop_vector ffs_fifoops1 = {
159 	.vop_default =		&ufs_fifoops,
160 	.vop_fsync =		ffs_fsync,
161 	.vop_fdatasync =	ffs_fdatasync,
162 	.vop_lock1 =		ffs_lock,
163 #ifdef INVARIANTS
164 	.vop_unlock =		ffs_unlock_debug,
165 #endif
166 	.vop_vptofh =		ffs_vptofh,
167 	.vop_fplookup_vexec =   VOP_EAGAIN,
168 	.vop_fplookup_symlink = VOP_EAGAIN,
169 };
170 VFS_VOP_VECTOR_REGISTER(ffs_fifoops1);
171 
172 /* Global vfs data structures for ufs. */
173 struct vop_vector ffs_vnodeops2 = {
174 	.vop_default =		&ufs_vnodeops,
175 	.vop_fsync =		ffs_fsync,
176 	.vop_fdatasync =	ffs_fdatasync,
177 	.vop_getpages =		ffs_getpages,
178 	.vop_getpages_async =	ffs_getpages_async,
179 	.vop_lock1 =		ffs_lock,
180 #ifdef INVARIANTS
181 	.vop_unlock =		ffs_unlock_debug,
182 #endif
183 	.vop_read =		ffs_read,
184 	.vop_reallocblks =	ffs_reallocblks,
185 	.vop_write =		ffs_write,
186 	.vop_closeextattr =	ffs_closeextattr,
187 	.vop_deleteextattr =	ffs_deleteextattr,
188 	.vop_getextattr =	ffs_getextattr,
189 	.vop_listextattr =	ffs_listextattr,
190 	.vop_openextattr =	ffs_openextattr,
191 	.vop_setextattr =	ffs_setextattr,
192 	.vop_vptofh =		ffs_vptofh,
193 	.vop_vput_pair =	ffs_vput_pair,
194 	.vop_fplookup_vexec =	ufs_fplookup_vexec,
195 	.vop_fplookup_symlink =	VOP_EAGAIN,
196 };
197 VFS_VOP_VECTOR_REGISTER(ffs_vnodeops2);
198 
199 struct vop_vector ffs_fifoops2 = {
200 	.vop_default =		&ufs_fifoops,
201 	.vop_fsync =		ffs_fsync,
202 	.vop_fdatasync =	ffs_fdatasync,
203 	.vop_lock1 =		ffs_lock,
204 #ifdef INVARIANTS
205 	.vop_unlock =		ffs_unlock_debug,
206 #endif
207 	.vop_reallocblks =	ffs_reallocblks,
208 	.vop_strategy =		ffsext_strategy,
209 	.vop_closeextattr =	ffs_closeextattr,
210 	.vop_deleteextattr =	ffs_deleteextattr,
211 	.vop_getextattr =	ffs_getextattr,
212 	.vop_listextattr =	ffs_listextattr,
213 	.vop_openextattr =	ffs_openextattr,
214 	.vop_setextattr =	ffs_setextattr,
215 	.vop_vptofh =		ffs_vptofh,
216 	.vop_fplookup_vexec =   VOP_EAGAIN,
217 	.vop_fplookup_symlink = VOP_EAGAIN,
218 };
219 VFS_VOP_VECTOR_REGISTER(ffs_fifoops2);
220 
221 /*
222  * Synch an open file.
223  */
224 /* ARGSUSED */
225 static int
ffs_fsync(struct vop_fsync_args * ap)226 ffs_fsync(struct vop_fsync_args *ap)
227 {
228 	struct vnode *vp;
229 	struct bufobj *bo;
230 	int error;
231 
232 	vp = ap->a_vp;
233 	bo = &vp->v_bufobj;
234 retry:
235 	error = ffs_syncvnode(vp, ap->a_waitfor, 0);
236 	if (error)
237 		return (error);
238 	if (ap->a_waitfor == MNT_WAIT && DOINGSOFTDEP(vp)) {
239 		error = softdep_fsync(vp);
240 		if (error)
241 			return (error);
242 
243 		/*
244 		 * The softdep_fsync() function may drop vp lock,
245 		 * allowing for dirty buffers to reappear on the
246 		 * bo_dirty list. Recheck and resync as needed.
247 		 */
248 		BO_LOCK(bo);
249 		if ((vp->v_type == VREG || vp->v_type == VDIR) &&
250 		    (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)) {
251 			BO_UNLOCK(bo);
252 			goto retry;
253 		}
254 		BO_UNLOCK(bo);
255 	}
256 	if (ffs_fsfail_cleanup(VFSTOUFS(vp->v_mount), 0))
257 		return (ENXIO);
258 	return (0);
259 }
260 
261 int
ffs_syncvnode(struct vnode * vp,int waitfor,int flags)262 ffs_syncvnode(struct vnode *vp, int waitfor, int flags)
263 {
264 	struct inode *ip;
265 	struct bufobj *bo;
266 	struct ufsmount *ump;
267 	struct buf *bp, *nbp;
268 	ufs_lbn_t lbn;
269 	int error, passes, wflag;
270 	bool still_dirty, unlocked, wait;
271 
272 	ip = VTOI(vp);
273 	bo = &vp->v_bufobj;
274 	ump = VFSTOUFS(vp->v_mount);
275 #ifdef WITNESS
276 	wflag = IS_SNAPSHOT(ip) ? LK_NOWITNESS : 0;
277 #else
278 	wflag = 0;
279 #endif
280 
281 	/*
282 	 * When doing MNT_WAIT we must first flush all dependencies
283 	 * on the inode.
284 	 */
285 	if (DOINGSOFTDEP(vp) && waitfor == MNT_WAIT &&
286 	    (error = softdep_sync_metadata(vp)) != 0) {
287 		if (ffs_fsfail_cleanup(ump, error))
288 			error = 0;
289 		return (error);
290 	}
291 
292 	/*
293 	 * Flush all dirty buffers associated with a vnode.
294 	 */
295 	error = 0;
296 	passes = 0;
297 	wait = false;	/* Always do an async pass first. */
298 	unlocked = false;
299 	lbn = lblkno(ITOFS(ip), (ip->i_size + ITOFS(ip)->fs_bsize - 1));
300 	BO_LOCK(bo);
301 loop:
302 	TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs)
303 		bp->b_vflags &= ~BV_SCANNED;
304 	TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
305 		/*
306 		 * Reasons to skip this buffer: it has already been considered
307 		 * on this pass, the buffer has dependencies that will cause
308 		 * it to be redirtied and it has not already been deferred,
309 		 * or it is already being written.
310 		 */
311 		if ((bp->b_vflags & BV_SCANNED) != 0)
312 			continue;
313 		bp->b_vflags |= BV_SCANNED;
314 		/*
315 		 * Flush indirects in order, if requested.
316 		 *
317 		 * Note that if only datasync is requested, we can
318 		 * skip indirect blocks when softupdates are not
319 		 * active.  Otherwise we must flush them with data,
320 		 * since dependencies prevent data block writes.
321 		 */
322 		if (waitfor == MNT_WAIT && bp->b_lblkno <= -UFS_NDADDR &&
323 		    (lbn_level(bp->b_lblkno) >= passes ||
324 		    ((flags & DATA_ONLY) != 0 && !DOINGSOFTDEP(vp))))
325 			continue;
326 		if (bp->b_lblkno > lbn)
327 			panic("ffs_syncvnode: syncing truncated data.");
328 		if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL) == 0) {
329 			BO_UNLOCK(bo);
330 		} else if (wait) {
331 			if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL |
332 			    LK_INTERLOCK | wflag, BO_LOCKPTR(bo)) != 0) {
333 				BO_LOCK(bo);
334 				bp->b_vflags &= ~BV_SCANNED;
335 				goto next_locked;
336 			}
337 		} else
338 			continue;
339 		if ((bp->b_flags & B_DELWRI) == 0)
340 			panic("ffs_fsync: not dirty");
341 		/*
342 		 * Check for dependencies and potentially complete them.
343 		 */
344 		if (!LIST_EMPTY(&bp->b_dep) &&
345 		    (error = softdep_sync_buf(vp, bp,
346 		    wait ? MNT_WAIT : MNT_NOWAIT)) != 0) {
347 			/*
348 			 * Lock order conflict, buffer was already unlocked,
349 			 * and vnode possibly unlocked.
350 			 */
351 			if (error == ERELOOKUP) {
352 				if (vp->v_data == NULL)
353 					return (EBADF);
354 				unlocked = true;
355 				if (DOINGSOFTDEP(vp) && waitfor == MNT_WAIT &&
356 				    (error = softdep_sync_metadata(vp)) != 0) {
357 					if (ffs_fsfail_cleanup(ump, error))
358 						error = 0;
359 					return (unlocked && error == 0 ?
360 					    ERELOOKUP : error);
361 				}
362 				/* Re-evaluate inode size */
363 				lbn = lblkno(ITOFS(ip), (ip->i_size +
364 				    ITOFS(ip)->fs_bsize - 1));
365 				goto next;
366 			}
367 			/* I/O error. */
368 			if (error != EBUSY) {
369 				BUF_UNLOCK(bp);
370 				return (error);
371 			}
372 			/* If we deferred once, don't defer again. */
373 		    	if ((bp->b_flags & B_DEFERRED) == 0) {
374 				bp->b_flags |= B_DEFERRED;
375 				BUF_UNLOCK(bp);
376 				goto next;
377 			}
378 		}
379 		if (wait) {
380 			bremfree(bp);
381 			error = bwrite(bp);
382 			if (ffs_fsfail_cleanup(ump, error))
383 				error = 0;
384 			if (error != 0)
385 				return (error);
386 		} else if ((bp->b_flags & B_CLUSTEROK)) {
387 			(void) vfs_bio_awrite(bp);
388 		} else {
389 			bremfree(bp);
390 			(void) bawrite(bp);
391 		}
392 next:
393 		/*
394 		 * Since we may have slept during the I/O, we need
395 		 * to start from a known point.
396 		 */
397 		BO_LOCK(bo);
398 next_locked:
399 		nbp = TAILQ_FIRST(&bo->bo_dirty.bv_hd);
400 	}
401 	if (waitfor != MNT_WAIT) {
402 		BO_UNLOCK(bo);
403 		if ((flags & NO_INO_UPDT) != 0)
404 			return (unlocked ? ERELOOKUP : 0);
405 		error = ffs_update(vp, 0);
406 		if (error == 0 && unlocked)
407 			error = ERELOOKUP;
408 		return (error);
409 	}
410 	/* Drain IO to see if we're done. */
411 	bufobj_wwait(bo, 0, 0);
412 	/*
413 	 * Block devices associated with filesystems may have new I/O
414 	 * requests posted for them even if the vnode is locked, so no
415 	 * amount of trying will get them clean.  We make several passes
416 	 * as a best effort.
417 	 *
418 	 * Regular files may need multiple passes to flush all dependency
419 	 * work as it is possible that we must write once per indirect
420 	 * level, once for the leaf, and once for the inode and each of
421 	 * these will be done with one sync and one async pass.
422 	 */
423 	if (bo->bo_dirty.bv_cnt > 0) {
424 		if ((flags & DATA_ONLY) == 0) {
425 			still_dirty = true;
426 		} else {
427 			/*
428 			 * For data-only sync, dirty indirect buffers
429 			 * are ignored.
430 			 */
431 			still_dirty = false;
432 			TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs) {
433 				if (bp->b_lblkno > -UFS_NDADDR) {
434 					still_dirty = true;
435 					break;
436 				}
437 			}
438 		}
439 
440 		if (still_dirty) {
441 			/* Write the inode after sync passes to flush deps. */
442 			if (wait && DOINGSOFTDEP(vp) &&
443 			    (flags & NO_INO_UPDT) == 0) {
444 				BO_UNLOCK(bo);
445 				ffs_update(vp, 1);
446 				BO_LOCK(bo);
447 			}
448 			/* switch between sync/async. */
449 			wait = !wait;
450 			if (wait || ++passes < UFS_NIADDR + 2)
451 				goto loop;
452 		}
453 	}
454 	BO_UNLOCK(bo);
455 	error = 0;
456 	if ((flags & DATA_ONLY) == 0) {
457 		if ((flags & NO_INO_UPDT) == 0)
458 			error = ffs_update(vp, 1);
459 		if (DOINGSUJ(vp))
460 			softdep_journal_fsync(VTOI(vp));
461 	} else if ((ip->i_flags & (IN_SIZEMOD | IN_IBLKDATA)) != 0) {
462 		error = ffs_update(vp, 1);
463 	}
464 	if (error == 0 && unlocked)
465 		error = ERELOOKUP;
466 	if (error == 0)
467 		ip->i_flag &= ~IN_NEEDSYNC;
468 	return (error);
469 }
470 
471 static int
ffs_fdatasync(struct vop_fdatasync_args * ap)472 ffs_fdatasync(struct vop_fdatasync_args *ap)
473 {
474 
475 	return (ffs_syncvnode(ap->a_vp, MNT_WAIT, DATA_ONLY));
476 }
477 
478 static int
ffs_lock(struct vop_lock1_args * ap)479 ffs_lock(
480 	struct vop_lock1_args /* {
481 		struct vnode *a_vp;
482 		int a_flags;
483 		char *file;
484 		int line;
485 	} */ *ap)
486 {
487 #if !defined(NO_FFS_SNAPSHOT) || defined(DIAGNOSTIC)
488 	struct vnode *vp = ap->a_vp;
489 #endif	/* !NO_FFS_SNAPSHOT || DIAGNOSTIC */
490 #ifdef DIAGNOSTIC
491 	struct inode *ip;
492 #endif	/* DIAGNOSTIC */
493 	int result;
494 #ifndef NO_FFS_SNAPSHOT
495 	int flags;
496 	struct lock *lkp;
497 
498 	/*
499 	 * Adaptive spinning mixed with SU leads to trouble. use a giant hammer
500 	 * and only use it when LK_NODDLKTREAT is set. Currently this means it
501 	 * is only used during path lookup.
502 	 */
503 	if ((ap->a_flags & LK_NODDLKTREAT) != 0)
504 		ap->a_flags |= LK_ADAPTIVE;
505 	switch (ap->a_flags & LK_TYPE_MASK) {
506 	case LK_SHARED:
507 	case LK_UPGRADE:
508 	case LK_EXCLUSIVE:
509 		flags = ap->a_flags;
510 		for (;;) {
511 			VNPASS(vp->v_holdcnt != 0, vp);
512 			lkp = vp->v_vnlock;
513 			result = lockmgr_lock_flags(lkp, flags,
514 			    &VI_MTX(vp)->lock_object, ap->a_file, ap->a_line);
515 			if (lkp == vp->v_vnlock || result != 0)
516 				break;
517 			/*
518 			 * Apparent success, except that the vnode
519 			 * mutated between snapshot file vnode and
520 			 * regular file vnode while this process
521 			 * slept.  The lock currently held is not the
522 			 * right lock.  Release it, and try to get the
523 			 * new lock.
524 			 */
525 			lockmgr_unlock(lkp);
526 			if ((flags & (LK_INTERLOCK | LK_NOWAIT)) ==
527 			    (LK_INTERLOCK | LK_NOWAIT))
528 				return (EBUSY);
529 			if ((flags & LK_TYPE_MASK) == LK_UPGRADE)
530 				flags = (flags & ~LK_TYPE_MASK) | LK_EXCLUSIVE;
531 			flags &= ~LK_INTERLOCK;
532 		}
533 #ifdef DIAGNOSTIC
534 		switch (ap->a_flags & LK_TYPE_MASK) {
535 		case LK_UPGRADE:
536 		case LK_EXCLUSIVE:
537 			if (result == 0 && vp->v_vnlock->lk_recurse == 0) {
538 				ip = VTOI(vp);
539 				if (ip != NULL)
540 					ip->i_lock_gen++;
541 			}
542 		}
543 #endif	/* DIAGNOSTIC */
544 		break;
545 	default:
546 #ifdef DIAGNOSTIC
547 		if ((ap->a_flags & LK_TYPE_MASK) == LK_DOWNGRADE) {
548 			ip = VTOI(vp);
549 			if (ip != NULL)
550 				ufs_unlock_tracker(ip);
551 		}
552 #endif	/* DIAGNOSTIC */
553 		result = VOP_LOCK1_APV(&ufs_vnodeops, ap);
554 		break;
555 	}
556 #else	/* NO_FFS_SNAPSHOT */
557 	/*
558 	 * See above for an explanation.
559 	 */
560 	if ((ap->a_flags & LK_NODDLKTREAT) != 0)
561 		ap->a_flags |= LK_ADAPTIVE;
562 #ifdef DIAGNOSTIC
563 	if ((ap->a_flags & LK_TYPE_MASK) == LK_DOWNGRADE) {
564 		ip = VTOI(vp);
565 		if (ip != NULL)
566 			ufs_unlock_tracker(ip);
567 	}
568 #endif	/* DIAGNOSTIC */
569 	result =  VOP_LOCK1_APV(&ufs_vnodeops, ap);
570 #endif	/* NO_FFS_SNAPSHOT */
571 #ifdef DIAGNOSTIC
572 	switch (ap->a_flags & LK_TYPE_MASK) {
573 	case LK_UPGRADE:
574 	case LK_EXCLUSIVE:
575 		if (result == 0 && vp->v_vnlock->lk_recurse == 0) {
576 			ip = VTOI(vp);
577 			if (ip != NULL)
578 				ip->i_lock_gen++;
579 		}
580 	}
581 #endif	/* DIAGNOSTIC */
582 	return (result);
583 }
584 
585 #ifdef INVARIANTS
586 static int
ffs_unlock_debug(struct vop_unlock_args * ap)587 ffs_unlock_debug(struct vop_unlock_args *ap)
588 {
589 	struct vnode *vp;
590 	struct inode *ip;
591 
592 	vp = ap->a_vp;
593 	ip = VTOI(vp);
594 	if (ip->i_flag & UFS_INODE_FLAG_LAZY_MASK_ASSERTABLE) {
595 		if ((vp->v_mflag & VMP_LAZYLIST) == 0) {
596 			VI_LOCK(vp);
597 			VNASSERT((vp->v_mflag & VMP_LAZYLIST), vp,
598 			    ("%s: modified vnode (%x) not on lazy list",
599 			    __func__, ip->i_flag));
600 			VI_UNLOCK(vp);
601 		}
602 	}
603 	KASSERT(vp->v_type != VDIR || vp->v_vnlock->lk_recurse != 0 ||
604 	    (ip->i_flag & IN_ENDOFF) == 0,
605 	    ("ufs dir vp %p ip %p flags %#x", vp, ip, ip->i_flag));
606 #ifdef DIAGNOSTIC
607 	if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE && ip != NULL &&
608 	    vp->v_vnlock->lk_recurse == 0)
609 		ufs_unlock_tracker(ip);
610 #endif
611 	return (VOP_UNLOCK_APV(&ufs_vnodeops, ap));
612 }
613 #endif
614 
615 static int
ffs_read_hole(struct uio * uio,long xfersize,long * size)616 ffs_read_hole(struct uio *uio, long xfersize, long *size)
617 {
618 	ssize_t saved_resid, tlen;
619 	int error;
620 
621 	while (xfersize > 0) {
622 		tlen = min(xfersize, ZERO_REGION_SIZE);
623 		saved_resid = uio->uio_resid;
624 		error = vn_io_fault_uiomove(__DECONST(void *, zero_region),
625 		    tlen, uio);
626 		if (error != 0)
627 			return (error);
628 		tlen = saved_resid - uio->uio_resid;
629 		xfersize -= tlen;
630 		*size -= tlen;
631 	}
632 	return (0);
633 }
634 
635 /*
636  * Vnode op for reading.
637  */
638 static int
ffs_read(struct vop_read_args * ap)639 ffs_read(
640 	struct vop_read_args /* {
641 		struct vnode *a_vp;
642 		struct uio *a_uio;
643 		int a_ioflag;
644 		struct ucred *a_cred;
645 	} */ *ap)
646 {
647 	struct vnode *vp;
648 	struct inode *ip;
649 	struct uio *uio;
650 	struct fs *fs;
651 	struct buf *bp;
652 	ufs_lbn_t lbn, nextlbn;
653 	off_t bytesinfile;
654 	long size, xfersize, blkoffset;
655 	ssize_t orig_resid;
656 	int bflag, error, ioflag, seqcount;
657 
658 	vp = ap->a_vp;
659 	uio = ap->a_uio;
660 	ioflag = ap->a_ioflag;
661 	if (ap->a_ioflag & IO_EXT)
662 #ifdef notyet
663 		return (ffs_extread(vp, uio, ioflag));
664 #else
665 		panic("ffs_read+IO_EXT");
666 #endif
667 #ifdef DIRECTIO
668 	if ((ioflag & IO_DIRECT) != 0) {
669 		int workdone;
670 
671 		error = ffs_rawread(vp, uio, &workdone);
672 		if (error != 0 || workdone != 0)
673 			return error;
674 	}
675 #endif
676 
677 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
678 	ip = VTOI(vp);
679 
680 #ifdef INVARIANTS
681 	if (uio->uio_rw != UIO_READ)
682 		panic("ffs_read: mode");
683 
684 	if (vp->v_type == VLNK) {
685 		if ((int)ip->i_size < VFSTOUFS(vp->v_mount)->um_maxsymlinklen)
686 			panic("ffs_read: short symlink");
687 	} else if (vp->v_type != VREG && vp->v_type != VDIR)
688 		panic("ffs_read: type %d",  vp->v_type);
689 #endif
690 	orig_resid = uio->uio_resid;
691 	KASSERT(orig_resid >= 0, ("ffs_read: uio->uio_resid < 0"));
692 	if (orig_resid == 0)
693 		return (0);
694 	KASSERT(uio->uio_offset >= 0, ("ffs_read: uio->uio_offset < 0"));
695 	fs = ITOFS(ip);
696 	if (uio->uio_offset < ip->i_size &&
697 	    uio->uio_offset >= fs->fs_maxfilesize)
698 		return (EOVERFLOW);
699 
700 	bflag = GB_UNMAPPED | (uio->uio_segflg == UIO_NOCOPY ? 0 : GB_NOSPARSE);
701 #ifdef WITNESS
702 	bflag |= IS_SNAPSHOT(ip) ? GB_NOWITNESS : 0;
703 #endif
704 	for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
705 		if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
706 			break;
707 		lbn = lblkno(fs, uio->uio_offset);
708 		nextlbn = lbn + 1;
709 
710 		/*
711 		 * size of buffer.  The buffer representing the
712 		 * end of the file is rounded up to the size of
713 		 * the block type ( fragment or full block,
714 		 * depending ).
715 		 */
716 		size = blksize(fs, ip, lbn);
717 		blkoffset = blkoff(fs, uio->uio_offset);
718 
719 		/*
720 		 * The amount we want to transfer in this iteration is
721 		 * one FS block less the amount of the data before
722 		 * our startpoint (duh!)
723 		 */
724 		xfersize = fs->fs_bsize - blkoffset;
725 
726 		/*
727 		 * But if we actually want less than the block,
728 		 * or the file doesn't have a whole block more of data,
729 		 * then use the lesser number.
730 		 */
731 		if (uio->uio_resid < xfersize)
732 			xfersize = uio->uio_resid;
733 		if (bytesinfile < xfersize)
734 			xfersize = bytesinfile;
735 
736 		if (lblktosize(fs, nextlbn) >= ip->i_size) {
737 			/*
738 			 * Don't do readahead if this is the end of the file.
739 			 */
740 			error = bread_gb(vp, lbn, size, NOCRED, bflag, &bp);
741 		} else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
742 			/*
743 			 * Otherwise if we are allowed to cluster,
744 			 * grab as much as we can.
745 			 *
746 			 * XXX  This may not be a win if we are not
747 			 * doing sequential access.
748 			 */
749 			error = cluster_read(vp, ip->i_size, lbn,
750 			    size, NOCRED, blkoffset + uio->uio_resid,
751 			    seqcount, bflag, &bp);
752 		} else if (seqcount > 1) {
753 			/*
754 			 * If we are NOT allowed to cluster, then
755 			 * if we appear to be acting sequentially,
756 			 * fire off a request for a readahead
757 			 * as well as a read. Note that the 4th and 5th
758 			 * arguments point to arrays of the size specified in
759 			 * the 6th argument.
760 			 */
761 			int nextsize = blksize(fs, ip, nextlbn);
762 			error = breadn_flags(vp, lbn, lbn, size, &nextlbn,
763 			    &nextsize, 1, NOCRED, bflag, NULL, &bp);
764 		} else {
765 			/*
766 			 * Failing all of the above, just read what the
767 			 * user asked for. Interestingly, the same as
768 			 * the first option above.
769 			 */
770 			error = bread_gb(vp, lbn, size, NOCRED, bflag, &bp);
771 		}
772 		if (error == EJUSTRETURN) {
773 			error = ffs_read_hole(uio, xfersize, &size);
774 			if (error == 0)
775 				continue;
776 		}
777 		if (error != 0) {
778 			brelse(bp);
779 			bp = NULL;
780 			break;
781 		}
782 
783 		/*
784 		 * We should only get non-zero b_resid when an I/O error
785 		 * has occurred, which should cause us to break above.
786 		 * However, if the short read did not cause an error,
787 		 * then we want to ensure that we do not uiomove bad
788 		 * or uninitialized data.
789 		 */
790 		size -= bp->b_resid;
791 		if (size < xfersize) {
792 			if (size == 0)
793 				break;
794 			xfersize = size;
795 		}
796 
797 		if (buf_mapped(bp)) {
798 			error = vn_io_fault_uiomove((char *)bp->b_data +
799 			    blkoffset, (int)xfersize, uio);
800 		} else {
801 			error = vn_io_fault_pgmove(bp->b_pages,
802 			    blkoffset + (bp->b_offset & PAGE_MASK),
803 			    (int)xfersize, uio);
804 		}
805 		if (error)
806 			break;
807 
808 		vfs_bio_brelse(bp, ioflag);
809 	}
810 
811 	/*
812 	 * This can only happen in the case of an error
813 	 * because the loop above resets bp to NULL on each iteration
814 	 * and on normal completion has not set a new value into it.
815 	 * so it must have come from a 'break' statement
816 	 */
817 	if (bp != NULL)
818 		vfs_bio_brelse(bp, ioflag);
819 
820 	if ((error == 0 || uio->uio_resid != orig_resid) &&
821 	    (vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
822 		UFS_INODE_SET_FLAG_SHARED(ip, IN_ACCESS);
823 	return (error);
824 }
825 
826 /*
827  * Vnode op for writing.
828  */
829 static int
ffs_write(struct vop_write_args * ap)830 ffs_write(
831 	struct vop_write_args /* {
832 		struct vnode *a_vp;
833 		struct uio *a_uio;
834 		int a_ioflag;
835 		struct ucred *a_cred;
836 	} */ *ap)
837 {
838 	struct vnode *vp;
839 	struct uio *uio;
840 	struct inode *ip;
841 	struct fs *fs;
842 	struct buf *bp;
843 	ufs_lbn_t lbn;
844 	off_t osize;
845 	ssize_t resid, r;
846 	int seqcount;
847 	int blkoffset, error, flags, ioflag, size, xfersize;
848 
849 	vp = ap->a_vp;
850 	if (DOINGSUJ(vp))
851 		softdep_prealloc(vp, MNT_WAIT);
852 	if (vp->v_data == NULL)
853 		return (EBADF);
854 
855 	uio = ap->a_uio;
856 	ioflag = ap->a_ioflag;
857 	if (ap->a_ioflag & IO_EXT)
858 #ifdef notyet
859 		return (ffs_extwrite(vp, uio, ioflag, ap->a_cred));
860 #else
861 		panic("ffs_write+IO_EXT");
862 #endif
863 
864 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
865 	ip = VTOI(vp);
866 
867 #ifdef INVARIANTS
868 	if (uio->uio_rw != UIO_WRITE)
869 		panic("ffs_write: mode");
870 #endif
871 
872 	switch (vp->v_type) {
873 	case VREG:
874 		if (ioflag & IO_APPEND)
875 			uio->uio_offset = ip->i_size;
876 		if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size)
877 			return (EPERM);
878 		/* FALLTHROUGH */
879 	case VLNK:
880 		break;
881 	case VDIR:
882 		panic("ffs_write: dir write");
883 		break;
884 	default:
885 		panic("ffs_write: type %p %d (%d,%d)", vp, (int)vp->v_type,
886 			(int)uio->uio_offset,
887 			(int)uio->uio_resid
888 		);
889 	}
890 
891 	KASSERT(uio->uio_resid >= 0, ("ffs_write: uio->uio_resid < 0"));
892 	KASSERT(uio->uio_offset >= 0, ("ffs_write: uio->uio_offset < 0"));
893 	fs = ITOFS(ip);
894 
895 	/*
896 	 * Maybe this should be above the vnode op call, but so long as
897 	 * file servers have no limits, I don't think it matters.
898 	 */
899 	error = vn_rlimit_fsizex(vp, uio, fs->fs_maxfilesize, &r,
900 	    uio->uio_td);
901 	if (error != 0) {
902 		vn_rlimit_fsizex_res(uio, r);
903 		return (error);
904 	}
905 
906 	resid = uio->uio_resid;
907 	osize = ip->i_size;
908 	if (seqcount > BA_SEQMAX)
909 		flags = BA_SEQMAX << BA_SEQSHIFT;
910 	else
911 		flags = seqcount << BA_SEQSHIFT;
912 	if (ioflag & IO_SYNC)
913 		flags |= IO_SYNC;
914 	flags |= BA_UNMAPPED;
915 
916 	for (error = 0; uio->uio_resid > 0;) {
917 		lbn = lblkno(fs, uio->uio_offset);
918 		blkoffset = blkoff(fs, uio->uio_offset);
919 		xfersize = fs->fs_bsize - blkoffset;
920 		if (uio->uio_resid < xfersize)
921 			xfersize = uio->uio_resid;
922 		if (uio->uio_offset + xfersize > ip->i_size)
923 			vnode_pager_setsize(vp, uio->uio_offset + xfersize);
924 
925 		/*
926 		 * We must perform a read-before-write if the transfer size
927 		 * does not cover the entire buffer.
928 		 */
929 		if (fs->fs_bsize > xfersize)
930 			flags |= BA_CLRBUF;
931 		else
932 			flags &= ~BA_CLRBUF;
933 /* XXX is uio->uio_offset the right thing here? */
934 		error = UFS_BALLOC(vp, uio->uio_offset, xfersize,
935 		    ap->a_cred, flags, &bp);
936 		if (error != 0) {
937 			vnode_pager_setsize(vp, ip->i_size);
938 			break;
939 		}
940 		if ((ioflag & (IO_SYNC|IO_INVAL)) == (IO_SYNC|IO_INVAL))
941 			bp->b_flags |= B_NOCACHE;
942 
943 		if (uio->uio_offset + xfersize > ip->i_size) {
944 			ip->i_size = uio->uio_offset + xfersize;
945 			DIP_SET(ip, i_size, ip->i_size);
946 			UFS_INODE_SET_FLAG(ip, IN_SIZEMOD | IN_CHANGE);
947 		}
948 
949 		size = blksize(fs, ip, lbn) - bp->b_resid;
950 		if (size < xfersize)
951 			xfersize = size;
952 
953 		if (buf_mapped(bp)) {
954 			error = vn_io_fault_uiomove((char *)bp->b_data +
955 			    blkoffset, (int)xfersize, uio);
956 		} else {
957 			error = vn_io_fault_pgmove(bp->b_pages,
958 			    blkoffset + (bp->b_offset & PAGE_MASK),
959 			    (int)xfersize, uio);
960 		}
961 		/*
962 		 * If the buffer is not already filled and we encounter an
963 		 * error while trying to fill it, we have to clear out any
964 		 * garbage data from the pages instantiated for the buffer.
965 		 * If we do not, a failed uiomove() during a write can leave
966 		 * the prior contents of the pages exposed to a userland mmap.
967 		 *
968 		 * Note that we need only clear buffers with a transfer size
969 		 * equal to the block size because buffers with a shorter
970 		 * transfer size were cleared above by the call to UFS_BALLOC()
971 		 * with the BA_CLRBUF flag set.
972 		 *
973 		 * If the source region for uiomove identically mmaps the
974 		 * buffer, uiomove() performed the NOP copy, and the buffer
975 		 * content remains valid because the page fault handler
976 		 * validated the pages.
977 		 */
978 		if (error != 0 && (bp->b_flags & B_CACHE) == 0 &&
979 		    fs->fs_bsize == xfersize) {
980 			if (error == EFAULT && LIST_EMPTY(&bp->b_dep)) {
981 				bp->b_flags |= B_INVAL | B_RELBUF | B_NOCACHE;
982 				brelse(bp);
983 				break;
984 			} else {
985 				vfs_bio_clrbuf(bp);
986 			}
987 		}
988 
989 		vfs_bio_set_flags(bp, ioflag);
990 
991 		/*
992 		 * If IO_SYNC each buffer is written synchronously.  Otherwise
993 		 * if we have a severe page deficiency write the buffer
994 		 * asynchronously.  Otherwise try to cluster, and if that
995 		 * doesn't do it then either do an async write (if O_DIRECT),
996 		 * or a delayed write (if not).
997 		 */
998 		if (ioflag & IO_SYNC) {
999 			(void)bwrite(bp);
1000 		} else if (vm_page_count_severe() ||
1001 			    buf_dirty_count_severe() ||
1002 			    (ioflag & IO_ASYNC)) {
1003 			bp->b_flags |= B_CLUSTEROK;
1004 			bawrite(bp);
1005 		} else if (xfersize + blkoffset == fs->fs_bsize) {
1006 			if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) {
1007 				bp->b_flags |= B_CLUSTEROK;
1008 				cluster_write(vp, &ip->i_clusterw, bp,
1009 				    ip->i_size, seqcount, GB_UNMAPPED);
1010 			} else {
1011 				bawrite(bp);
1012 			}
1013 		} else if (ioflag & IO_DIRECT) {
1014 			bp->b_flags |= B_CLUSTEROK;
1015 			bawrite(bp);
1016 		} else {
1017 			bp->b_flags |= B_CLUSTEROK;
1018 			bdwrite(bp);
1019 		}
1020 		if (error || xfersize == 0)
1021 			break;
1022 		UFS_INODE_SET_FLAG(ip, IN_CHANGE | IN_UPDATE);
1023 	}
1024 	/*
1025 	 * If we successfully wrote any data, and we are not the superuser
1026 	 * we clear the setuid and setgid bits as a precaution against
1027 	 * tampering.
1028 	 */
1029 	if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid &&
1030 	    ap->a_cred) {
1031 		if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID)) {
1032 			vn_seqc_write_begin(vp);
1033 			UFS_INODE_SET_MODE(ip, ip->i_mode & ~(ISUID | ISGID));
1034 			DIP_SET(ip, i_mode, ip->i_mode);
1035 			vn_seqc_write_end(vp);
1036 		}
1037 	}
1038 	if (error) {
1039 		if (ioflag & IO_UNIT) {
1040 			(void)ffs_truncate(vp, osize,
1041 			    IO_NORMAL | (ioflag & IO_SYNC), ap->a_cred);
1042 			uio->uio_offset -= resid - uio->uio_resid;
1043 			uio->uio_resid = resid;
1044 		}
1045 	} else if (resid > uio->uio_resid && (ioflag & IO_SYNC)) {
1046 		if (!(ioflag & IO_DATASYNC) ||
1047 		    (ip->i_flags & (IN_SIZEMOD | IN_IBLKDATA)))
1048 			error = ffs_update(vp, 1);
1049 		if (ffs_fsfail_cleanup(VFSTOUFS(vp->v_mount), error))
1050 			error = ENXIO;
1051 	}
1052 	vn_rlimit_fsizex_res(uio, r);
1053 	return (error);
1054 }
1055 
1056 /*
1057  * Extended attribute area reading.
1058  */
1059 static int
ffs_extread(struct vnode * vp,struct uio * uio,int ioflag)1060 ffs_extread(struct vnode *vp, struct uio *uio, int ioflag)
1061 {
1062 	struct inode *ip;
1063 	struct ufs2_dinode *dp;
1064 	struct fs *fs;
1065 	struct buf *bp;
1066 	ufs_lbn_t lbn, nextlbn;
1067 	off_t bytesinfile;
1068 	long size, xfersize, blkoffset;
1069 	ssize_t orig_resid;
1070 	int error;
1071 
1072 	ip = VTOI(vp);
1073 	fs = ITOFS(ip);
1074 	dp = ip->i_din2;
1075 
1076 #ifdef INVARIANTS
1077 	if (uio->uio_rw != UIO_READ || fs->fs_magic != FS_UFS2_MAGIC)
1078 		panic("ffs_extread: mode");
1079 
1080 #endif
1081 	orig_resid = uio->uio_resid;
1082 	KASSERT(orig_resid >= 0, ("ffs_extread: uio->uio_resid < 0"));
1083 	if (orig_resid == 0)
1084 		return (0);
1085 	KASSERT(uio->uio_offset >= 0, ("ffs_extread: uio->uio_offset < 0"));
1086 
1087 	for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
1088 		if ((bytesinfile = dp->di_extsize - uio->uio_offset) <= 0)
1089 			break;
1090 		lbn = lblkno(fs, uio->uio_offset);
1091 		nextlbn = lbn + 1;
1092 
1093 		/*
1094 		 * size of buffer.  The buffer representing the
1095 		 * end of the file is rounded up to the size of
1096 		 * the block type ( fragment or full block,
1097 		 * depending ).
1098 		 */
1099 		size = sblksize(fs, dp->di_extsize, lbn);
1100 		blkoffset = blkoff(fs, uio->uio_offset);
1101 
1102 		/*
1103 		 * The amount we want to transfer in this iteration is
1104 		 * one FS block less the amount of the data before
1105 		 * our startpoint (duh!)
1106 		 */
1107 		xfersize = fs->fs_bsize - blkoffset;
1108 
1109 		/*
1110 		 * But if we actually want less than the block,
1111 		 * or the file doesn't have a whole block more of data,
1112 		 * then use the lesser number.
1113 		 */
1114 		if (uio->uio_resid < xfersize)
1115 			xfersize = uio->uio_resid;
1116 		if (bytesinfile < xfersize)
1117 			xfersize = bytesinfile;
1118 
1119 		if (lblktosize(fs, nextlbn) >= dp->di_extsize) {
1120 			/*
1121 			 * Don't do readahead if this is the end of the info.
1122 			 */
1123 			error = bread(vp, -1 - lbn, size, NOCRED, &bp);
1124 		} else {
1125 			/*
1126 			 * If we have a second block, then
1127 			 * fire off a request for a readahead
1128 			 * as well as a read. Note that the 4th and 5th
1129 			 * arguments point to arrays of the size specified in
1130 			 * the 6th argument.
1131 			 */
1132 			int nextsize = sblksize(fs, dp->di_extsize, nextlbn);
1133 			nextlbn = -1 - nextlbn;
1134 			error = breadn(vp, -1 - lbn,
1135 			    size, &nextlbn, &nextsize, 1, NOCRED, &bp);
1136 		}
1137 		if (error) {
1138 			brelse(bp);
1139 			bp = NULL;
1140 			break;
1141 		}
1142 
1143 		/*
1144 		 * We should only get non-zero b_resid when an I/O error
1145 		 * has occurred, which should cause us to break above.
1146 		 * However, if the short read did not cause an error,
1147 		 * then we want to ensure that we do not uiomove bad
1148 		 * or uninitialized data.
1149 		 */
1150 		size -= bp->b_resid;
1151 		if (size < xfersize) {
1152 			if (size == 0)
1153 				break;
1154 			xfersize = size;
1155 		}
1156 
1157 		error = uiomove((char *)bp->b_data + blkoffset,
1158 					(int)xfersize, uio);
1159 		if (error)
1160 			break;
1161 		vfs_bio_brelse(bp, ioflag);
1162 	}
1163 
1164 	/*
1165 	 * This can only happen in the case of an error
1166 	 * because the loop above resets bp to NULL on each iteration
1167 	 * and on normal completion has not set a new value into it.
1168 	 * so it must have come from a 'break' statement
1169 	 */
1170 	if (bp != NULL)
1171 		vfs_bio_brelse(bp, ioflag);
1172 	return (error);
1173 }
1174 
1175 /*
1176  * Extended attribute area writing.
1177  */
1178 static int
ffs_extwrite(struct vnode * vp,struct uio * uio,int ioflag,struct ucred * ucred)1179 ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag, struct ucred *ucred)
1180 {
1181 	struct inode *ip;
1182 	struct ufs2_dinode *dp;
1183 	struct fs *fs;
1184 	struct buf *bp;
1185 	ufs_lbn_t lbn;
1186 	off_t osize;
1187 	ssize_t resid;
1188 	int blkoffset, error, flags, size, xfersize;
1189 
1190 	ip = VTOI(vp);
1191 	fs = ITOFS(ip);
1192 	dp = ip->i_din2;
1193 
1194 #ifdef INVARIANTS
1195 	if (uio->uio_rw != UIO_WRITE || fs->fs_magic != FS_UFS2_MAGIC)
1196 		panic("ffs_extwrite: mode");
1197 #endif
1198 
1199 	if (ioflag & IO_APPEND)
1200 		uio->uio_offset = dp->di_extsize;
1201 	KASSERT(uio->uio_offset >= 0, ("ffs_extwrite: uio->uio_offset < 0"));
1202 	KASSERT(uio->uio_resid >= 0, ("ffs_extwrite: uio->uio_resid < 0"));
1203 	if ((uoff_t)uio->uio_offset + uio->uio_resid >
1204 	    UFS_NXADDR * fs->fs_bsize)
1205 		return (EFBIG);
1206 
1207 	resid = uio->uio_resid;
1208 	osize = dp->di_extsize;
1209 	flags = IO_EXT;
1210 	if (ioflag & IO_SYNC)
1211 		flags |= IO_SYNC;
1212 
1213 	for (error = 0; uio->uio_resid > 0;) {
1214 		lbn = lblkno(fs, uio->uio_offset);
1215 		blkoffset = blkoff(fs, uio->uio_offset);
1216 		xfersize = fs->fs_bsize - blkoffset;
1217 		if (uio->uio_resid < xfersize)
1218 			xfersize = uio->uio_resid;
1219 
1220 		/*
1221 		 * We must perform a read-before-write if the transfer size
1222 		 * does not cover the entire buffer.
1223 		 */
1224 		if (fs->fs_bsize > xfersize)
1225 			flags |= BA_CLRBUF;
1226 		else
1227 			flags &= ~BA_CLRBUF;
1228 		error = UFS_BALLOC(vp, uio->uio_offset, xfersize,
1229 		    ucred, flags, &bp);
1230 		if (error != 0)
1231 			break;
1232 		/*
1233 		 * If the buffer is not valid we have to clear out any
1234 		 * garbage data from the pages instantiated for the buffer.
1235 		 * If we do not, a failed uiomove() during a write can leave
1236 		 * the prior contents of the pages exposed to a userland
1237 		 * mmap().  XXX deal with uiomove() errors a better way.
1238 		 */
1239 		if ((bp->b_flags & B_CACHE) == 0 && fs->fs_bsize <= xfersize)
1240 			vfs_bio_clrbuf(bp);
1241 
1242 		if (uio->uio_offset + xfersize > dp->di_extsize) {
1243 			dp->di_extsize = uio->uio_offset + xfersize;
1244 			UFS_INODE_SET_FLAG(ip, IN_SIZEMOD | IN_CHANGE);
1245 		}
1246 
1247 		size = sblksize(fs, dp->di_extsize, lbn) - bp->b_resid;
1248 		if (size < xfersize)
1249 			xfersize = size;
1250 
1251 		error =
1252 		    uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
1253 
1254 		vfs_bio_set_flags(bp, ioflag);
1255 
1256 		/*
1257 		 * If IO_SYNC each buffer is written synchronously.  Otherwise
1258 		 * if we have a severe page deficiency write the buffer
1259 		 * asynchronously.  Otherwise try to cluster, and if that
1260 		 * doesn't do it then either do an async write (if O_DIRECT),
1261 		 * or a delayed write (if not).
1262 		 */
1263 		if (ioflag & IO_SYNC) {
1264 			(void)bwrite(bp);
1265 		} else if (vm_page_count_severe() ||
1266 			    buf_dirty_count_severe() ||
1267 			    xfersize + blkoffset == fs->fs_bsize ||
1268 			    (ioflag & (IO_ASYNC | IO_DIRECT)))
1269 			bawrite(bp);
1270 		else
1271 			bdwrite(bp);
1272 		if (error || xfersize == 0)
1273 			break;
1274 		UFS_INODE_SET_FLAG(ip, IN_CHANGE);
1275 	}
1276 	/*
1277 	 * If we successfully wrote any data, and we are not the superuser
1278 	 * we clear the setuid and setgid bits as a precaution against
1279 	 * tampering.
1280 	 */
1281 	if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid && ucred) {
1282 		if (priv_check_cred(ucred, PRIV_VFS_RETAINSUGID)) {
1283 			vn_seqc_write_begin(vp);
1284 			UFS_INODE_SET_MODE(ip, ip->i_mode & ~(ISUID | ISGID));
1285 			dp->di_mode = ip->i_mode;
1286 			vn_seqc_write_end(vp);
1287 		}
1288 	}
1289 	if (error) {
1290 		if (ioflag & IO_UNIT) {
1291 			(void)ffs_truncate(vp, osize,
1292 			    IO_EXT | (ioflag&IO_SYNC), ucred);
1293 			uio->uio_offset -= resid - uio->uio_resid;
1294 			uio->uio_resid = resid;
1295 		}
1296 	} else if (resid > uio->uio_resid && (ioflag & IO_SYNC))
1297 		error = ffs_update(vp, 1);
1298 	return (error);
1299 }
1300 
1301 /*
1302  * Vnode operating to retrieve a named extended attribute.
1303  *
1304  * Locate a particular EA (nspace:name) in the area (ptr:length), and return
1305  * the length of the EA, and possibly the pointer to the entry and to the data.
1306  */
1307 static int
ffs_findextattr(uint8_t * ptr,uint64_t length,int nspace,const char * name,struct extattr ** eapp,uint8_t ** eac)1308 ffs_findextattr(uint8_t *ptr, uint64_t length, int nspace, const char *name,
1309     struct extattr **eapp, uint8_t **eac)
1310 {
1311 	struct extattr *eap, *eaend;
1312 	size_t nlen;
1313 
1314 	nlen = strlen(name);
1315 	KASSERT(ALIGNED_TO(ptr, struct extattr), ("unaligned"));
1316 	eap = (struct extattr *)ptr;
1317 	eaend = (struct extattr *)(ptr + length);
1318 	for (; eap < eaend; eap = EXTATTR_NEXT(eap)) {
1319 		KASSERT(EXTATTR_NEXT(eap) <= eaend,
1320 		    ("extattr next %p beyond %p", EXTATTR_NEXT(eap), eaend));
1321 		if (eap->ea_namespace != nspace || eap->ea_namelength != nlen
1322 		    || memcmp(eap->ea_name, name, nlen) != 0)
1323 			continue;
1324 		if (eapp != NULL)
1325 			*eapp = eap;
1326 		if (eac != NULL)
1327 			*eac = EXTATTR_CONTENT(eap);
1328 		return (EXTATTR_CONTENT_SIZE(eap));
1329 	}
1330 	return (-1);
1331 }
1332 
1333 static int
ffs_rdextattr(uint8_t ** p,struct vnode * vp,struct thread * td)1334 ffs_rdextattr(uint8_t **p, struct vnode *vp, struct thread *td)
1335 {
1336 	const struct extattr *eap, *eaend, *eapnext;
1337 	struct inode *ip;
1338 	struct ufs2_dinode *dp;
1339 	struct fs *fs;
1340 	struct uio luio;
1341 	struct iovec liovec;
1342 	uint64_t easize;
1343 	int error;
1344 	uint8_t *eae;
1345 
1346 	ip = VTOI(vp);
1347 	fs = ITOFS(ip);
1348 	dp = ip->i_din2;
1349 	easize = dp->di_extsize;
1350 	if ((uoff_t)easize > UFS_NXADDR * fs->fs_bsize)
1351 		return (EFBIG);
1352 
1353 	eae = malloc(easize, M_TEMP, M_WAITOK);
1354 
1355 	liovec.iov_base = eae;
1356 	liovec.iov_len = easize;
1357 	luio.uio_iov = &liovec;
1358 	luio.uio_iovcnt = 1;
1359 	luio.uio_offset = 0;
1360 	luio.uio_resid = easize;
1361 	luio.uio_segflg = UIO_SYSSPACE;
1362 	luio.uio_rw = UIO_READ;
1363 	luio.uio_td = td;
1364 
1365 	error = ffs_extread(vp, &luio, IO_EXT | IO_SYNC);
1366 	if (error) {
1367 		free(eae, M_TEMP);
1368 		return (error);
1369 	}
1370 	/* Validate disk xattrfile contents. */
1371 	for (eap = (void *)eae, eaend = (void *)(eae + easize); eap < eaend;
1372 	    eap = eapnext) {
1373 		/* Detect zeroed out tail */
1374 		if (eap->ea_length < sizeof(*eap) || eap->ea_length == 0) {
1375 			easize = (const uint8_t *)eap - eae;
1376 			break;
1377 		}
1378 
1379 		eapnext = EXTATTR_NEXT(eap);
1380 		/* Bogusly long entry. */
1381 		if (eapnext > eaend) {
1382 			free(eae, M_TEMP);
1383 			return (EINTEGRITY);
1384 		}
1385 	}
1386 	ip->i_ea_len = easize;
1387 	*p = eae;
1388 	return (0);
1389 }
1390 
1391 static void
ffs_lock_ea(struct vnode * vp)1392 ffs_lock_ea(struct vnode *vp)
1393 {
1394 	struct inode *ip;
1395 
1396 	ip = VTOI(vp);
1397 	VI_LOCK(vp);
1398 	while (ip->i_flag & IN_EA_LOCKED) {
1399 		UFS_INODE_SET_FLAG(ip, IN_EA_LOCKWAIT);
1400 		msleep(&ip->i_ea_refs, &vp->v_interlock, PINOD, "ufs_ea", 0);
1401 	}
1402 	UFS_INODE_SET_FLAG(ip, IN_EA_LOCKED);
1403 	VI_UNLOCK(vp);
1404 }
1405 
1406 static void
ffs_unlock_ea(struct vnode * vp)1407 ffs_unlock_ea(struct vnode *vp)
1408 {
1409 	struct inode *ip;
1410 
1411 	ip = VTOI(vp);
1412 	VI_LOCK(vp);
1413 	if (ip->i_flag & IN_EA_LOCKWAIT)
1414 		wakeup(&ip->i_ea_refs);
1415 	ip->i_flag &= ~(IN_EA_LOCKED | IN_EA_LOCKWAIT);
1416 	VI_UNLOCK(vp);
1417 }
1418 
1419 static int
ffs_open_ea(struct vnode * vp,struct ucred * cred,struct thread * td)1420 ffs_open_ea(struct vnode *vp, struct ucred *cred, struct thread *td)
1421 {
1422 	struct inode *ip;
1423 	int error;
1424 
1425 	ip = VTOI(vp);
1426 
1427 	ffs_lock_ea(vp);
1428 	if (ip->i_ea_area != NULL) {
1429 		ip->i_ea_refs++;
1430 		ffs_unlock_ea(vp);
1431 		return (0);
1432 	}
1433 	error = ffs_rdextattr(&ip->i_ea_area, vp, td);
1434 	if (error) {
1435 		ffs_unlock_ea(vp);
1436 		return (error);
1437 	}
1438 	ip->i_ea_error = 0;
1439 	ip->i_ea_refs++;
1440 	ffs_unlock_ea(vp);
1441 	return (0);
1442 }
1443 
1444 /*
1445  * Vnode extattr transaction commit/abort
1446  */
1447 static int
ffs_close_ea(struct vnode * vp,int commit,struct ucred * cred,struct thread * td)1448 ffs_close_ea(struct vnode *vp, int commit, struct ucred *cred, struct thread *td)
1449 {
1450 	struct inode *ip;
1451 	struct uio luio;
1452 	struct iovec *liovec;
1453 	struct ufs2_dinode *dp;
1454 	size_t ea_len, tlen;
1455 	int error, i, lcnt;
1456 	bool truncate;
1457 
1458 	ip = VTOI(vp);
1459 
1460 	ffs_lock_ea(vp);
1461 	if (ip->i_ea_area == NULL) {
1462 		ffs_unlock_ea(vp);
1463 		return (EINVAL);
1464 	}
1465 	dp = ip->i_din2;
1466 	error = ip->i_ea_error;
1467 	truncate = false;
1468 	if (commit && error == 0) {
1469 		ASSERT_VOP_ELOCKED(vp, "ffs_close_ea commit");
1470 		if (cred == NOCRED)
1471 			cred =  vp->v_mount->mnt_cred;
1472 
1473 		ea_len = MAX(ip->i_ea_len, dp->di_extsize);
1474 		for (lcnt = 1, tlen = ea_len - ip->i_ea_len; tlen > 0;) {
1475 			tlen -= MIN(ZERO_REGION_SIZE, tlen);
1476 			lcnt++;
1477 		}
1478 
1479 		liovec = __builtin_alloca(lcnt * sizeof(struct iovec));
1480 		luio.uio_iovcnt = lcnt;
1481 
1482 		liovec[0].iov_base = ip->i_ea_area;
1483 		liovec[0].iov_len = ip->i_ea_len;
1484 		for (i = 1, tlen = ea_len - ip->i_ea_len; i < lcnt; i++) {
1485 			liovec[i].iov_base = __DECONST(void *, zero_region);
1486 			liovec[i].iov_len = MIN(ZERO_REGION_SIZE, tlen);
1487 			tlen -= liovec[i].iov_len;
1488 		}
1489 		MPASS(tlen == 0);
1490 
1491 		luio.uio_iov = liovec;
1492 		luio.uio_offset = 0;
1493 		luio.uio_resid = ea_len;
1494 		luio.uio_segflg = UIO_SYSSPACE;
1495 		luio.uio_rw = UIO_WRITE;
1496 		luio.uio_td = td;
1497 		error = ffs_extwrite(vp, &luio, IO_EXT | IO_SYNC, cred);
1498 		if (error == 0 && ip->i_ea_len == 0)
1499 			truncate = true;
1500 	}
1501 	if (--ip->i_ea_refs == 0) {
1502 		free(ip->i_ea_area, M_TEMP);
1503 		ip->i_ea_area = NULL;
1504 		ip->i_ea_len = 0;
1505 		ip->i_ea_error = 0;
1506 	}
1507 	ffs_unlock_ea(vp);
1508 
1509 	if (truncate)
1510 		ffs_truncate(vp, 0, IO_EXT, cred);
1511 	return (error);
1512 }
1513 
1514 /*
1515  * Vnode extattr strategy routine for fifos.
1516  *
1517  * We need to check for a read or write of the external attributes.
1518  * Otherwise we just fall through and do the usual thing.
1519  */
1520 static int
ffsext_strategy(struct vop_strategy_args * ap)1521 ffsext_strategy(
1522 	struct vop_strategy_args /* {
1523 		struct vnodeop_desc *a_desc;
1524 		struct vnode *a_vp;
1525 		struct buf *a_bp;
1526 	} */ *ap)
1527 {
1528 	struct vnode *vp;
1529 	daddr_t lbn;
1530 
1531 	vp = ap->a_vp;
1532 	lbn = ap->a_bp->b_lblkno;
1533 	if (I_IS_UFS2(VTOI(vp)) && lbn < 0 && lbn >= -UFS_NXADDR)
1534 		return (VOP_STRATEGY_APV(&ufs_vnodeops, ap));
1535 	if (vp->v_type == VFIFO)
1536 		return (VOP_STRATEGY_APV(&ufs_fifoops, ap));
1537 	panic("spec nodes went here");
1538 }
1539 
1540 /*
1541  * Vnode extattr transaction commit/abort
1542  */
1543 static int
ffs_openextattr(struct vop_openextattr_args * ap)1544 ffs_openextattr(
1545 	struct vop_openextattr_args /* {
1546 		struct vnodeop_desc *a_desc;
1547 		struct vnode *a_vp;
1548 		IN struct ucred *a_cred;
1549 		IN struct thread *a_td;
1550 	} */ *ap)
1551 {
1552 
1553 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1554 		return (EOPNOTSUPP);
1555 
1556 	return (ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td));
1557 }
1558 
1559 /*
1560  * Vnode extattr transaction commit/abort
1561  */
1562 static int
ffs_closeextattr(struct vop_closeextattr_args * ap)1563 ffs_closeextattr(
1564 	struct vop_closeextattr_args /* {
1565 		struct vnodeop_desc *a_desc;
1566 		struct vnode *a_vp;
1567 		int a_commit;
1568 		IN struct ucred *a_cred;
1569 		IN struct thread *a_td;
1570 	} */ *ap)
1571 {
1572 	struct vnode *vp;
1573 
1574 	vp = ap->a_vp;
1575 	if (vp->v_type == VCHR || vp->v_type == VBLK)
1576 		return (EOPNOTSUPP);
1577 	if (ap->a_commit && (vp->v_mount->mnt_flag & MNT_RDONLY) != 0)
1578 		return (EROFS);
1579 
1580 	if (ap->a_commit && DOINGSUJ(vp)) {
1581 		ASSERT_VOP_ELOCKED(vp, "ffs_closeextattr commit");
1582 		softdep_prealloc(vp, MNT_WAIT);
1583 		if (vp->v_data == NULL)
1584 			return (EBADF);
1585 	}
1586 	return (ffs_close_ea(vp, ap->a_commit, ap->a_cred, ap->a_td));
1587 }
1588 
1589 /*
1590  * Vnode operation to remove a named attribute.
1591  */
1592 static int
ffs_deleteextattr(struct vop_deleteextattr_args * ap)1593 ffs_deleteextattr(
1594 	struct vop_deleteextattr_args /* {
1595 		IN struct vnode *a_vp;
1596 		IN int a_attrnamespace;
1597 		IN const char *a_name;
1598 		IN struct ucred *a_cred;
1599 		IN struct thread *a_td;
1600 	} */ *ap)
1601 {
1602 	struct vnode *vp;
1603 	struct inode *ip;
1604 	struct extattr *eap;
1605 	uint32_t ul;
1606 	int olen, error, i, easize;
1607 	uint8_t *eae;
1608 	void *tmp;
1609 
1610 	vp = ap->a_vp;
1611 	ip = VTOI(vp);
1612 
1613 	if (vp->v_type == VCHR || vp->v_type == VBLK)
1614 		return (EOPNOTSUPP);
1615 	if (strlen(ap->a_name) == 0)
1616 		return (EINVAL);
1617 	if (vp->v_mount->mnt_flag & MNT_RDONLY)
1618 		return (EROFS);
1619 
1620 	error = extattr_check_cred(vp, ap->a_attrnamespace,
1621 	    ap->a_cred, ap->a_td, VWRITE);
1622 	if (error) {
1623 		/*
1624 		 * ffs_lock_ea is not needed there, because the vnode
1625 		 * must be exclusively locked.
1626 		 */
1627 		if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1628 			ip->i_ea_error = error;
1629 		return (error);
1630 	}
1631 
1632 	if (DOINGSUJ(vp)) {
1633 		ASSERT_VOP_ELOCKED(vp, "ffs_deleteextattr");
1634 		softdep_prealloc(vp, MNT_WAIT);
1635 		if (vp->v_data == NULL)
1636 			return (EBADF);
1637 	}
1638 
1639 	error = ffs_open_ea(vp, ap->a_cred, ap->a_td);
1640 	if (error)
1641 		return (error);
1642 
1643 	/* CEM: delete could be done in-place instead */
1644 	eae = malloc(ip->i_ea_len, M_TEMP, M_WAITOK);
1645 	bcopy(ip->i_ea_area, eae, ip->i_ea_len);
1646 	easize = ip->i_ea_len;
1647 
1648 	olen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name,
1649 	    &eap, NULL);
1650 	if (olen == -1) {
1651 		/* delete but nonexistent */
1652 		free(eae, M_TEMP);
1653 		ffs_close_ea(vp, 0, ap->a_cred, ap->a_td);
1654 		return (ENOATTR);
1655 	}
1656 	ul = eap->ea_length;
1657 	i = (uint8_t *)EXTATTR_NEXT(eap) - eae;
1658 	bcopy(EXTATTR_NEXT(eap), eap, easize - i);
1659 	easize -= ul;
1660 
1661 	tmp = ip->i_ea_area;
1662 	ip->i_ea_area = eae;
1663 	ip->i_ea_len = easize;
1664 	free(tmp, M_TEMP);
1665 	error = ffs_close_ea(vp, 1, ap->a_cred, ap->a_td);
1666 	return (error);
1667 }
1668 
1669 /*
1670  * Vnode operation to retrieve a named extended attribute.
1671  */
1672 static int
ffs_getextattr(struct vop_getextattr_args * ap)1673 ffs_getextattr(
1674 	struct vop_getextattr_args /* {
1675 		IN struct vnode *a_vp;
1676 		IN int a_attrnamespace;
1677 		IN const char *a_name;
1678 		INOUT struct uio *a_uio;
1679 		OUT size_t *a_size;
1680 		IN struct ucred *a_cred;
1681 		IN struct thread *a_td;
1682 	} */ *ap)
1683 {
1684 	struct inode *ip;
1685 	uint8_t *eae, *p;
1686 	unsigned easize;
1687 	int error, ealen;
1688 
1689 	ip = VTOI(ap->a_vp);
1690 
1691 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1692 		return (EOPNOTSUPP);
1693 
1694 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1695 	    ap->a_cred, ap->a_td, VREAD);
1696 	if (error)
1697 		return (error);
1698 
1699 	error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1700 	if (error)
1701 		return (error);
1702 
1703 	eae = ip->i_ea_area;
1704 	easize = ip->i_ea_len;
1705 
1706 	ealen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name,
1707 	    NULL, &p);
1708 	if (ealen >= 0) {
1709 		error = 0;
1710 		if (ap->a_size != NULL)
1711 			*ap->a_size = ealen;
1712 		else if (ap->a_uio != NULL)
1713 			error = uiomove(p, ealen, ap->a_uio);
1714 	} else
1715 		error = ENOATTR;
1716 
1717 	ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1718 	return (error);
1719 }
1720 
1721 /*
1722  * Vnode operation to retrieve extended attributes on a vnode.
1723  */
1724 static int
ffs_listextattr(struct vop_listextattr_args * ap)1725 ffs_listextattr(
1726 	struct vop_listextattr_args /* {
1727 		IN struct vnode *a_vp;
1728 		IN int a_attrnamespace;
1729 		INOUT struct uio *a_uio;
1730 		OUT size_t *a_size;
1731 		IN struct ucred *a_cred;
1732 		IN struct thread *a_td;
1733 	} */ *ap)
1734 {
1735 	struct inode *ip;
1736 	struct extattr *eap, *eaend;
1737 	int error, ealen;
1738 
1739 	ip = VTOI(ap->a_vp);
1740 
1741 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1742 		return (EOPNOTSUPP);
1743 
1744 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1745 	    ap->a_cred, ap->a_td, VREAD);
1746 	if (error)
1747 		return (error);
1748 
1749 	error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1750 	if (error)
1751 		return (error);
1752 
1753 	error = 0;
1754 	if (ap->a_size != NULL)
1755 		*ap->a_size = 0;
1756 
1757 	KASSERT(ALIGNED_TO(ip->i_ea_area, struct extattr), ("unaligned"));
1758 	eap = (struct extattr *)ip->i_ea_area;
1759 	eaend = (struct extattr *)(ip->i_ea_area + ip->i_ea_len);
1760 	for (; error == 0 && eap < eaend; eap = EXTATTR_NEXT(eap)) {
1761 		KASSERT(EXTATTR_NEXT(eap) <= eaend,
1762 		    ("extattr next %p beyond %p", EXTATTR_NEXT(eap), eaend));
1763 		if (eap->ea_namespace != ap->a_attrnamespace)
1764 			continue;
1765 
1766 		ealen = eap->ea_namelength;
1767 		if (ap->a_size != NULL)
1768 			*ap->a_size += ealen + 1;
1769 		else if (ap->a_uio != NULL)
1770 			error = uiomove(&eap->ea_namelength, ealen + 1,
1771 			    ap->a_uio);
1772 	}
1773 
1774 	ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1775 	return (error);
1776 }
1777 
1778 /*
1779  * Vnode operation to set a named attribute.
1780  */
1781 static int
ffs_setextattr(struct vop_setextattr_args * ap)1782 ffs_setextattr(
1783 	struct vop_setextattr_args /* {
1784 		IN struct vnode *a_vp;
1785 		IN int a_attrnamespace;
1786 		IN const char *a_name;
1787 		INOUT struct uio *a_uio;
1788 		IN struct ucred *a_cred;
1789 		IN struct thread *a_td;
1790 	} */ *ap)
1791 {
1792 	struct vnode *vp;
1793 	struct inode *ip;
1794 	struct fs *fs;
1795 	struct extattr *eap;
1796 	uint32_t ealength, ul;
1797 	ssize_t ealen;
1798 	int olen, eapad1, eapad2, error, i, easize;
1799 	uint8_t *eae;
1800 	void *tmp;
1801 
1802 	vp = ap->a_vp;
1803 	ip = VTOI(vp);
1804 	fs = ITOFS(ip);
1805 
1806 	if (vp->v_type == VCHR || vp->v_type == VBLK)
1807 		return (EOPNOTSUPP);
1808 	if (strlen(ap->a_name) == 0)
1809 		return (EINVAL);
1810 
1811 	/* XXX Now unsupported API to delete EAs using NULL uio. */
1812 	if (ap->a_uio == NULL)
1813 		return (EOPNOTSUPP);
1814 
1815 	if (vp->v_mount->mnt_flag & MNT_RDONLY)
1816 		return (EROFS);
1817 
1818 	ealen = ap->a_uio->uio_resid;
1819 	if (ealen < 0 || ealen > lblktosize(fs, UFS_NXADDR))
1820 		return (EINVAL);
1821 
1822 	error = extattr_check_cred(vp, ap->a_attrnamespace,
1823 	    ap->a_cred, ap->a_td, VWRITE);
1824 	if (error) {
1825 		/*
1826 		 * ffs_lock_ea is not needed there, because the vnode
1827 		 * must be exclusively locked.
1828 		 */
1829 		if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1830 			ip->i_ea_error = error;
1831 		return (error);
1832 	}
1833 
1834 	if (DOINGSUJ(vp)) {
1835 		ASSERT_VOP_ELOCKED(vp, "ffs_deleteextattr");
1836 		softdep_prealloc(vp, MNT_WAIT);
1837 		if (vp->v_data == NULL)
1838 			return (EBADF);
1839 	}
1840 
1841 	error = ffs_open_ea(vp, ap->a_cred, ap->a_td);
1842 	if (error)
1843 		return (error);
1844 
1845 	ealength = sizeof(uint32_t) + 3 + strlen(ap->a_name);
1846 	eapad1 = roundup2(ealength, 8) - ealength;
1847 	eapad2 = roundup2(ealen, 8) - ealen;
1848 	ealength += eapad1 + ealen + eapad2;
1849 
1850 	/*
1851 	 * CEM: rewrites of the same size or smaller could be done in-place
1852 	 * instead.  (We don't acquire any fine-grained locks in here either,
1853 	 * so we could also do bigger writes in-place.)
1854 	 */
1855 	eae = malloc(ip->i_ea_len + ealength, M_TEMP, M_WAITOK);
1856 	bcopy(ip->i_ea_area, eae, ip->i_ea_len);
1857 	easize = ip->i_ea_len;
1858 
1859 	olen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name,
1860 	    &eap, NULL);
1861         if (olen == -1) {
1862 		/* new, append at end */
1863 		KASSERT(ALIGNED_TO(eae + easize, struct extattr),
1864 		    ("unaligned"));
1865 		eap = (struct extattr *)(eae + easize);
1866 		easize += ealength;
1867 	} else {
1868 		ul = eap->ea_length;
1869 		i = (uint8_t *)EXTATTR_NEXT(eap) - eae;
1870 		if (ul != ealength) {
1871 			bcopy(EXTATTR_NEXT(eap), (uint8_t *)eap + ealength,
1872 			    easize - i);
1873 			easize += (ealength - ul);
1874 		}
1875 	}
1876 	if (easize > lblktosize(fs, UFS_NXADDR)) {
1877 		free(eae, M_TEMP);
1878 		ffs_close_ea(vp, 0, ap->a_cred, ap->a_td);
1879 		if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1880 			ip->i_ea_error = ENOSPC;
1881 		return (ENOSPC);
1882 	}
1883 	eap->ea_length = ealength;
1884 	eap->ea_namespace = ap->a_attrnamespace;
1885 	eap->ea_contentpadlen = eapad2;
1886 	eap->ea_namelength = strlen(ap->a_name);
1887 	memcpy(eap->ea_name, ap->a_name, strlen(ap->a_name));
1888 	bzero(&eap->ea_name[strlen(ap->a_name)], eapad1);
1889 	error = uiomove(EXTATTR_CONTENT(eap), ealen, ap->a_uio);
1890 	if (error) {
1891 		free(eae, M_TEMP);
1892 		ffs_close_ea(vp, 0, ap->a_cred, ap->a_td);
1893 		if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1894 			ip->i_ea_error = error;
1895 		return (error);
1896 	}
1897 	bzero((uint8_t *)EXTATTR_CONTENT(eap) + ealen, eapad2);
1898 
1899 	tmp = ip->i_ea_area;
1900 	ip->i_ea_area = eae;
1901 	ip->i_ea_len = easize;
1902 	free(tmp, M_TEMP);
1903 	error = ffs_close_ea(vp, 1, ap->a_cred, ap->a_td);
1904 	return (error);
1905 }
1906 
1907 /*
1908  * Vnode pointer to File handle
1909  */
1910 static int
ffs_vptofh(struct vop_vptofh_args * ap)1911 ffs_vptofh(
1912 	struct vop_vptofh_args /* {
1913 		IN struct vnode *a_vp;
1914 		IN struct fid *a_fhp;
1915 	} */ *ap)
1916 {
1917 	struct inode *ip;
1918 	struct ufid *ufhp;
1919 	_Static_assert(sizeof(struct ufid) <= sizeof(struct fid),
1920 	    "struct ufid cannot be larger than struct fid");
1921 
1922 	ip = VTOI(ap->a_vp);
1923 	ufhp = (struct ufid *)ap->a_fhp;
1924 	ufhp->ufid_len = sizeof(struct ufid);
1925 	ufhp->ufid_ino = ip->i_number;
1926 	ufhp->ufid_gen = ip->i_gen;
1927 	return (0);
1928 }
1929 
1930 SYSCTL_DECL(_vfs_ffs);
1931 static int use_buf_pager = 1;
1932 SYSCTL_INT(_vfs_ffs, OID_AUTO, use_buf_pager, CTLFLAG_RWTUN, &use_buf_pager, 0,
1933     "Always use buffer pager instead of bmap");
1934 
1935 static daddr_t
ffs_gbp_getblkno(struct vnode * vp,vm_ooffset_t off)1936 ffs_gbp_getblkno(struct vnode *vp, vm_ooffset_t off)
1937 {
1938 
1939 	return (lblkno(VFSTOUFS(vp->v_mount)->um_fs, off));
1940 }
1941 
1942 static int
ffs_gbp_getblksz(struct vnode * vp,daddr_t lbn,long * sz)1943 ffs_gbp_getblksz(struct vnode *vp, daddr_t lbn, long *sz)
1944 {
1945 
1946 	*sz = blksize(VFSTOUFS(vp->v_mount)->um_fs, VTOI(vp), lbn);
1947 	return (0);
1948 }
1949 
1950 static int
ffs_getpages(struct vop_getpages_args * ap)1951 ffs_getpages(struct vop_getpages_args *ap)
1952 {
1953 	struct vnode *vp;
1954 	struct ufsmount *um;
1955 
1956 	vp = ap->a_vp;
1957 	um = VFSTOUFS(vp->v_mount);
1958 
1959 	if (!use_buf_pager && um->um_devvp->v_bufobj.bo_bsize <= PAGE_SIZE)
1960 		return (vnode_pager_generic_getpages(vp, ap->a_m, ap->a_count,
1961 		    ap->a_rbehind, ap->a_rahead, NULL, NULL));
1962 	return (vfs_bio_getpages(vp, ap->a_m, ap->a_count, ap->a_rbehind,
1963 	    ap->a_rahead, ffs_gbp_getblkno, ffs_gbp_getblksz));
1964 }
1965 
1966 static int
ffs_getpages_async(struct vop_getpages_async_args * ap)1967 ffs_getpages_async(struct vop_getpages_async_args *ap)
1968 {
1969 	struct vnode *vp;
1970 	struct ufsmount *um;
1971 	bool do_iodone;
1972 	int error;
1973 
1974 	vp = ap->a_vp;
1975 	um = VFSTOUFS(vp->v_mount);
1976 	do_iodone = true;
1977 
1978 	if (um->um_devvp->v_bufobj.bo_bsize <= PAGE_SIZE) {
1979 		error = vnode_pager_generic_getpages(vp, ap->a_m, ap->a_count,
1980 		    ap->a_rbehind, ap->a_rahead, ap->a_iodone, ap->a_arg);
1981 		if (error == 0)
1982 			do_iodone = false;
1983 	} else {
1984 		error = vfs_bio_getpages(vp, ap->a_m, ap->a_count,
1985 		    ap->a_rbehind, ap->a_rahead, ffs_gbp_getblkno,
1986 		    ffs_gbp_getblksz);
1987 	}
1988 	if (do_iodone && ap->a_iodone != NULL)
1989 		ap->a_iodone(ap->a_arg, ap->a_m, ap->a_count, error);
1990 
1991 	return (error);
1992 }
1993 
1994 static int
ffs_vput_pair(struct vop_vput_pair_args * ap)1995 ffs_vput_pair(struct vop_vput_pair_args *ap)
1996 {
1997 	struct mount *mp;
1998 	struct vnode *dvp, *vp, *vp1, **vpp;
1999 	struct inode *dp, *ip;
2000 	ino_t ip_ino;
2001 	uint64_t ip_gen;
2002 	int error, vp_locked;
2003 
2004 	dvp = ap->a_dvp;
2005 	dp = VTOI(dvp);
2006 	vpp = ap->a_vpp;
2007 	vp = vpp != NULL ? *vpp : NULL;
2008 
2009 	if ((dp->i_flag & (IN_NEEDSYNC | IN_ENDOFF)) == 0) {
2010 		vput(dvp);
2011 		if (vp != NULL && ap->a_unlock_vp)
2012 			vput(vp);
2013 		return (0);
2014 	}
2015 
2016 	mp = dvp->v_mount;
2017 	if (vp != NULL) {
2018 		if (ap->a_unlock_vp) {
2019 			vput(vp);
2020 		} else {
2021 			MPASS(vp->v_type != VNON);
2022 			vp_locked = VOP_ISLOCKED(vp);
2023 			ip = VTOI(vp);
2024 			ip_ino = ip->i_number;
2025 			ip_gen = ip->i_gen;
2026 			VOP_UNLOCK(vp);
2027 		}
2028 	}
2029 
2030 	/*
2031 	 * If compaction or fsync was requested do it in ffs_vput_pair()
2032 	 * now that other locks are no longer held.
2033          */
2034 	if ((dp->i_flag & IN_ENDOFF) != 0) {
2035 		VNASSERT(I_ENDOFF(dp) != 0 && I_ENDOFF(dp) < dp->i_size, dvp,
2036 		    ("IN_ENDOFF set but I_ENDOFF() is not"));
2037 		dp->i_flag &= ~IN_ENDOFF;
2038 		error = UFS_TRUNCATE(dvp, (off_t)I_ENDOFF(dp), IO_NORMAL |
2039 		    (DOINGASYNC(dvp) ? 0 : IO_SYNC), curthread->td_ucred);
2040 		if (error != 0 && error != ERELOOKUP) {
2041 			if (!ffs_fsfail_cleanup(VFSTOUFS(mp), error)) {
2042 				vn_printf(dvp,
2043 				    "IN_ENDOFF: failed to truncate, "
2044 				    "error %d\n", error);
2045 			}
2046 #ifdef UFS_DIRHASH
2047 			ufsdirhash_free(dp);
2048 #endif
2049 		}
2050 		SET_I_ENDOFF(dp, 0);
2051 	}
2052 	if ((dp->i_flag & IN_NEEDSYNC) != 0) {
2053 		do {
2054 			error = ffs_syncvnode(dvp, MNT_WAIT, 0);
2055 		} while (error == ERELOOKUP);
2056 	}
2057 
2058 	vput(dvp);
2059 
2060 	if (vp == NULL || ap->a_unlock_vp)
2061 		return (0);
2062 	MPASS(mp != NULL);
2063 
2064 	/*
2065 	 * It is possible that vp is reclaimed at this point. Only
2066 	 * routines that call us with a_unlock_vp == false can find
2067 	 * that their vp has been reclaimed. There are three areas
2068 	 * that are affected:
2069 	 * 1) vn_open_cred() - later VOPs could fail, but
2070 	 *    dead_open() returns 0 to simulate successful open.
2071 	 * 2) ffs_snapshot() - creation of snapshot fails with EBADF.
2072 	 * 3) NFS server (several places) - code is prepared to detect
2073 	 *    and respond to dead vnodes by returning ESTALE.
2074 	 */
2075 	VOP_LOCK(vp, vp_locked | LK_RETRY);
2076 	if (IS_UFS(vp))
2077 		return (0);
2078 
2079 	/*
2080 	 * Try harder to recover from reclaimed vp if reclaim was not
2081 	 * because underlying inode was cleared.  We saved inode
2082 	 * number and inode generation, so we can try to reinstantiate
2083 	 * exactly same version of inode.  If this fails, return
2084 	 * original doomed vnode and let caller to handle
2085 	 * consequences.
2086 	 *
2087 	 * Note that callers must keep write started around
2088 	 * VOP_VPUT_PAIR() calls, so it is safe to use mp without
2089 	 * busying it.
2090 	 */
2091 	VOP_UNLOCK(vp);
2092 	error = ffs_inotovp(mp, ip_ino, ip_gen, LK_EXCLUSIVE, &vp1,
2093 	    FFSV_REPLACE_DOOMED);
2094 	if (error != 0) {
2095 		VOP_LOCK(vp, vp_locked | LK_RETRY);
2096 	} else {
2097 		vrele(vp);
2098 		*vpp = vp1;
2099 	}
2100 	return (error);
2101 }
2102