xref: /NextBSD/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zfs_vnops.c (revision 4557fabb34e865d7f40be64b39c9e34fa41dbb60)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23  * Copyright (c) 2012, 2015 by Delphix. All rights reserved.
24  * Copyright 2014 Nexenta Systems, Inc.  All rights reserved.
25  */
26 
27 /* Portions Copyright 2007 Jeremy Teo */
28 /* Portions Copyright 2010 Robert Milkowski */
29 
30 #include <sys/types.h>
31 #include <sys/param.h>
32 #include <sys/time.h>
33 #include <sys/systm.h>
34 #include <sys/sysmacros.h>
35 #include <sys/resource.h>
36 #include <sys/vfs.h>
37 #include <sys/vm.h>
38 #include <sys/vnode.h>
39 #include <sys/file.h>
40 #include <sys/stat.h>
41 #include <sys/kmem.h>
42 #include <sys/taskq.h>
43 #include <sys/uio.h>
44 #include <sys/atomic.h>
45 #include <sys/namei.h>
46 #include <sys/mman.h>
47 #include <sys/cmn_err.h>
48 #include <sys/errno.h>
49 #include <sys/unistd.h>
50 #include <sys/zfs_dir.h>
51 #include <sys/zfs_ioctl.h>
52 #include <sys/fs/zfs.h>
53 #include <sys/dmu.h>
54 #include <sys/dmu_objset.h>
55 #include <sys/spa.h>
56 #include <sys/txg.h>
57 #include <sys/dbuf.h>
58 #include <sys/zap.h>
59 #include <sys/sa.h>
60 #include <sys/dirent.h>
61 #include <sys/policy.h>
62 #include <sys/sunddi.h>
63 #include <sys/filio.h>
64 #include <sys/sid.h>
65 #include <sys/zfs_ctldir.h>
66 #include <sys/zfs_fuid.h>
67 #include <sys/zfs_sa.h>
68 #include <sys/dnlc.h>
69 #include <sys/zfs_rlock.h>
70 #include <sys/extdirent.h>
71 #include <sys/kidmap.h>
72 #include <sys/bio.h>
73 #include <sys/buf.h>
74 #include <sys/sched.h>
75 #include <sys/acl.h>
76 #include <vm/vm_param.h>
77 #include <vm/vm_pageout.h>
78 
79 /*
80  * Programming rules.
81  *
82  * Each vnode op performs some logical unit of work.  To do this, the ZPL must
83  * properly lock its in-core state, create a DMU transaction, do the work,
84  * record this work in the intent log (ZIL), commit the DMU transaction,
85  * and wait for the intent log to commit if it is a synchronous operation.
86  * Moreover, the vnode ops must work in both normal and log replay context.
87  * The ordering of events is important to avoid deadlocks and references
88  * to freed memory.  The example below illustrates the following Big Rules:
89  *
90  *  (1)	A check must be made in each zfs thread for a mounted file system.
91  *	This is done avoiding races using ZFS_ENTER(zfsvfs).
92  *	A ZFS_EXIT(zfsvfs) is needed before all returns.  Any znodes
93  *	must be checked with ZFS_VERIFY_ZP(zp).  Both of these macros
94  *	can return EIO from the calling function.
95  *
96  *  (2)	VN_RELE() should always be the last thing except for zil_commit()
97  *	(if necessary) and ZFS_EXIT(). This is for 3 reasons:
98  *	First, if it's the last reference, the vnode/znode
99  *	can be freed, so the zp may point to freed memory.  Second, the last
100  *	reference will call zfs_zinactive(), which may induce a lot of work --
101  *	pushing cached pages (which acquires range locks) and syncing out
102  *	cached atime changes.  Third, zfs_zinactive() may require a new tx,
103  *	which could deadlock the system if you were already holding one.
104  *	If you must call VN_RELE() within a tx then use VN_RELE_ASYNC().
105  *
106  *  (3)	All range locks must be grabbed before calling dmu_tx_assign(),
107  *	as they can span dmu_tx_assign() calls.
108  *
109  *  (4) If ZPL locks are held, pass TXG_NOWAIT as the second argument to
110  *      dmu_tx_assign().  This is critical because we don't want to block
111  *      while holding locks.
112  *
113  *	If no ZPL locks are held (aside from ZFS_ENTER()), use TXG_WAIT.  This
114  *	reduces lock contention and CPU usage when we must wait (note that if
115  *	throughput is constrained by the storage, nearly every transaction
116  *	must wait).
117  *
118  *      Note, in particular, that if a lock is sometimes acquired before
119  *      the tx assigns, and sometimes after (e.g. z_lock), then failing
120  *      to use a non-blocking assign can deadlock the system.  The scenario:
121  *
122  *	Thread A has grabbed a lock before calling dmu_tx_assign().
123  *	Thread B is in an already-assigned tx, and blocks for this lock.
124  *	Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
125  *	forever, because the previous txg can't quiesce until B's tx commits.
126  *
127  *	If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
128  *	then drop all locks, call dmu_tx_wait(), and try again.  On subsequent
129  *	calls to dmu_tx_assign(), pass TXG_WAITED rather than TXG_NOWAIT,
130  *	to indicate that this operation has already called dmu_tx_wait().
131  *	This will ensure that we don't retry forever, waiting a short bit
132  *	each time.
133  *
134  *  (5)	If the operation succeeded, generate the intent log entry for it
135  *	before dropping locks.  This ensures that the ordering of events
136  *	in the intent log matches the order in which they actually occurred.
137  *	During ZIL replay the zfs_log_* functions will update the sequence
138  *	number to indicate the zil transaction has replayed.
139  *
140  *  (6)	At the end of each vnode op, the DMU tx must always commit,
141  *	regardless of whether there were any errors.
142  *
143  *  (7)	After dropping all locks, invoke zil_commit(zilog, foid)
144  *	to ensure that synchronous semantics are provided when necessary.
145  *
146  * In general, this is how things should be ordered in each vnode op:
147  *
148  *	ZFS_ENTER(zfsvfs);		// exit if unmounted
149  * top:
150  *	zfs_dirent_lock(&dl, ...)	// lock directory entry (may VN_HOLD())
151  *	rw_enter(...);			// grab any other locks you need
152  *	tx = dmu_tx_create(...);	// get DMU tx
153  *	dmu_tx_hold_*();		// hold each object you might modify
154  *	error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
155  *	if (error) {
156  *		rw_exit(...);		// drop locks
157  *		zfs_dirent_unlock(dl);	// unlock directory entry
158  *		VN_RELE(...);		// release held vnodes
159  *		if (error == ERESTART) {
160  *			waited = B_TRUE;
161  *			dmu_tx_wait(tx);
162  *			dmu_tx_abort(tx);
163  *			goto top;
164  *		}
165  *		dmu_tx_abort(tx);	// abort DMU tx
166  *		ZFS_EXIT(zfsvfs);	// finished in zfs
167  *		return (error);		// really out of space
168  *	}
169  *	error = do_real_work();		// do whatever this VOP does
170  *	if (error == 0)
171  *		zfs_log_*(...);		// on success, make ZIL entry
172  *	dmu_tx_commit(tx);		// commit DMU tx -- error or not
173  *	rw_exit(...);			// drop locks
174  *	zfs_dirent_unlock(dl);		// unlock directory entry
175  *	VN_RELE(...);			// release held vnodes
176  *	zil_commit(zilog, foid);	// synchronous when necessary
177  *	ZFS_EXIT(zfsvfs);		// finished in zfs
178  *	return (error);			// done, report error
179  */
180 
181 /* ARGSUSED */
182 static int
zfs_open(vnode_t ** vpp,int flag,cred_t * cr,caller_context_t * ct)183 zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct)
184 {
185 	znode_t	*zp = VTOZ(*vpp);
186 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
187 
188 	ZFS_ENTER(zfsvfs);
189 	ZFS_VERIFY_ZP(zp);
190 
191 	if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) &&
192 	    ((flag & FAPPEND) == 0)) {
193 		ZFS_EXIT(zfsvfs);
194 		return (SET_ERROR(EPERM));
195 	}
196 
197 	if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
198 	    ZTOV(zp)->v_type == VREG &&
199 	    !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) {
200 		if (fs_vscan(*vpp, cr, 0) != 0) {
201 			ZFS_EXIT(zfsvfs);
202 			return (SET_ERROR(EACCES));
203 		}
204 	}
205 
206 	/* Keep a count of the synchronous opens in the znode */
207 	if (flag & (FSYNC | FDSYNC))
208 		atomic_inc_32(&zp->z_sync_cnt);
209 
210 	ZFS_EXIT(zfsvfs);
211 	return (0);
212 }
213 
214 /* ARGSUSED */
215 static int
zfs_close(vnode_t * vp,int flag,int count,offset_t offset,cred_t * cr,caller_context_t * ct)216 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
217     caller_context_t *ct)
218 {
219 	znode_t	*zp = VTOZ(vp);
220 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
221 
222 	/*
223 	 * Clean up any locks held by this process on the vp.
224 	 */
225 	cleanlocks(vp, ddi_get_pid(), 0);
226 	cleanshares(vp, ddi_get_pid());
227 
228 	ZFS_ENTER(zfsvfs);
229 	ZFS_VERIFY_ZP(zp);
230 
231 	/* Decrement the synchronous opens in the znode */
232 	if ((flag & (FSYNC | FDSYNC)) && (count == 1))
233 		atomic_dec_32(&zp->z_sync_cnt);
234 
235 	if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
236 	    ZTOV(zp)->v_type == VREG &&
237 	    !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0)
238 		VERIFY(fs_vscan(vp, cr, 1) == 0);
239 
240 	ZFS_EXIT(zfsvfs);
241 	return (0);
242 }
243 
244 /*
245  * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and
246  * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter.
247  */
248 static int
zfs_holey(vnode_t * vp,u_long cmd,offset_t * off)249 zfs_holey(vnode_t *vp, u_long cmd, offset_t *off)
250 {
251 	znode_t	*zp = VTOZ(vp);
252 	uint64_t noff = (uint64_t)*off; /* new offset */
253 	uint64_t file_sz;
254 	int error;
255 	boolean_t hole;
256 
257 	file_sz = zp->z_size;
258 	if (noff >= file_sz)  {
259 		return (SET_ERROR(ENXIO));
260 	}
261 
262 	if (cmd == _FIO_SEEK_HOLE)
263 		hole = B_TRUE;
264 	else
265 		hole = B_FALSE;
266 
267 	error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff);
268 
269 	if (error == ESRCH)
270 		return (SET_ERROR(ENXIO));
271 
272 	/*
273 	 * We could find a hole that begins after the logical end-of-file,
274 	 * because dmu_offset_next() only works on whole blocks.  If the
275 	 * EOF falls mid-block, then indicate that the "virtual hole"
276 	 * at the end of the file begins at the logical EOF, rather than
277 	 * at the end of the last block.
278 	 */
279 	if (noff > file_sz) {
280 		ASSERT(hole);
281 		noff = file_sz;
282 	}
283 
284 	if (noff < *off)
285 		return (error);
286 	*off = noff;
287 	return (error);
288 }
289 
290 /* ARGSUSED */
291 static int
zfs_ioctl(vnode_t * vp,u_long com,intptr_t data,int flag,cred_t * cred,int * rvalp,caller_context_t * ct)292 zfs_ioctl(vnode_t *vp, u_long com, intptr_t data, int flag, cred_t *cred,
293     int *rvalp, caller_context_t *ct)
294 {
295 	offset_t off;
296 	offset_t ndata;
297 	dmu_object_info_t doi;
298 	int error;
299 	zfsvfs_t *zfsvfs;
300 	znode_t *zp;
301 
302 	switch (com) {
303 	case _FIOFFS:
304 	{
305 		return (0);
306 
307 		/*
308 		 * The following two ioctls are used by bfu.  Faking out,
309 		 * necessary to avoid bfu errors.
310 		 */
311 	}
312 	case _FIOGDIO:
313 	case _FIOSDIO:
314 	{
315 		return (0);
316 	}
317 
318 	case _FIO_SEEK_DATA:
319 	case _FIO_SEEK_HOLE:
320 	{
321 #ifdef illumos
322 		if (ddi_copyin((void *)data, &off, sizeof (off), flag))
323 			return (SET_ERROR(EFAULT));
324 #else
325 		off = *(offset_t *)data;
326 #endif
327 		zp = VTOZ(vp);
328 		zfsvfs = zp->z_zfsvfs;
329 		ZFS_ENTER(zfsvfs);
330 		ZFS_VERIFY_ZP(zp);
331 
332 		/* offset parameter is in/out */
333 		error = zfs_holey(vp, com, &off);
334 		ZFS_EXIT(zfsvfs);
335 		if (error)
336 			return (error);
337 #ifdef illumos
338 		if (ddi_copyout(&off, (void *)data, sizeof (off), flag))
339 			return (SET_ERROR(EFAULT));
340 #else
341 		*(offset_t *)data = off;
342 #endif
343 		return (0);
344 	}
345 #ifdef illumos
346 	case _FIO_COUNT_FILLED:
347 	{
348 		/*
349 		 * _FIO_COUNT_FILLED adds a new ioctl command which
350 		 * exposes the number of filled blocks in a
351 		 * ZFS object.
352 		 */
353 		zp = VTOZ(vp);
354 		zfsvfs = zp->z_zfsvfs;
355 		ZFS_ENTER(zfsvfs);
356 		ZFS_VERIFY_ZP(zp);
357 
358 		/*
359 		 * Wait for all dirty blocks for this object
360 		 * to get synced out to disk, and the DMU info
361 		 * updated.
362 		 */
363 		error = dmu_object_wait_synced(zfsvfs->z_os, zp->z_id);
364 		if (error) {
365 			ZFS_EXIT(zfsvfs);
366 			return (error);
367 		}
368 
369 		/*
370 		 * Retrieve fill count from DMU object.
371 		 */
372 		error = dmu_object_info(zfsvfs->z_os, zp->z_id, &doi);
373 		if (error) {
374 			ZFS_EXIT(zfsvfs);
375 			return (error);
376 		}
377 
378 		ndata = doi.doi_fill_count;
379 
380 		ZFS_EXIT(zfsvfs);
381 		if (ddi_copyout(&ndata, (void *)data, sizeof (ndata), flag))
382 			return (SET_ERROR(EFAULT));
383 		return (0);
384 	}
385 #endif
386 	}
387 	return (SET_ERROR(ENOTTY));
388 }
389 
390 static vm_page_t
page_busy(vnode_t * vp,int64_t start,int64_t off,int64_t nbytes)391 page_busy(vnode_t *vp, int64_t start, int64_t off, int64_t nbytes)
392 {
393 	vm_object_t obj;
394 	vm_page_t pp;
395 	int64_t end;
396 
397 	/*
398 	 * At present vm_page_clear_dirty extends the cleared range to DEV_BSIZE
399 	 * aligned boundaries, if the range is not aligned.  As a result a
400 	 * DEV_BSIZE subrange with partially dirty data may get marked as clean.
401 	 * It may happen that all DEV_BSIZE subranges are marked clean and thus
402 	 * the whole page would be considred clean despite have some dirty data.
403 	 * For this reason we should shrink the range to DEV_BSIZE aligned
404 	 * boundaries before calling vm_page_clear_dirty.
405 	 */
406 	end = rounddown2(off + nbytes, DEV_BSIZE);
407 	off = roundup2(off, DEV_BSIZE);
408 	nbytes = end - off;
409 
410 	obj = vp->v_object;
411 	zfs_vmobject_assert_wlocked(obj);
412 
413 	for (;;) {
414 		if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL &&
415 		    pp->valid) {
416 			if (vm_page_xbusied(pp)) {
417 				/*
418 				 * Reference the page before unlocking and
419 				 * sleeping so that the page daemon is less
420 				 * likely to reclaim it.
421 				 */
422 				vm_page_reference(pp);
423 				vm_page_lock(pp);
424 				zfs_vmobject_wunlock(obj);
425 				vm_page_busy_sleep(pp, "zfsmwb");
426 				zfs_vmobject_wlock(obj);
427 				continue;
428 			}
429 			vm_page_sbusy(pp);
430 		}
431 #ifdef VM_LEGACY
432 		else if (pp == NULL) {
433 			pp = vm_page_alloc(obj, OFF_TO_IDX(start),
434 			    VM_ALLOC_SYSTEM | VM_ALLOC_IFCACHED |
435 			    VM_ALLOC_SBUSY);
436 		} else {
437 			ASSERT(pp != NULL && !pp->valid);
438 			pp = NULL;
439 		}
440 #endif
441 		if (pp != NULL) {
442 			ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
443 			vm_object_pip_add(obj, 1);
444 			pmap_remove_write(pp);
445 			if (nbytes != 0)
446 				vm_page_clear_dirty(pp, off, nbytes);
447 		}
448 		break;
449 	}
450 	return (pp);
451 }
452 
453 static void
page_unbusy(vm_page_t pp)454 page_unbusy(vm_page_t pp)
455 {
456 
457 	vm_page_sunbusy(pp);
458 	vm_object_pip_subtract(pp->object, 1);
459 }
460 
461 static vm_page_t
page_hold(vnode_t * vp,int64_t start)462 page_hold(vnode_t *vp, int64_t start)
463 {
464 	vm_object_t obj;
465 	vm_page_t pp;
466 
467 	obj = vp->v_object;
468 	zfs_vmobject_assert_wlocked(obj);
469 
470 	for (;;) {
471 		if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL &&
472 		    pp->valid) {
473 			if (vm_page_xbusied(pp)) {
474 				/*
475 				 * Reference the page before unlocking and
476 				 * sleeping so that the page daemon is less
477 				 * likely to reclaim it.
478 				 */
479 				vm_page_reference(pp);
480 				vm_page_lock(pp);
481 				zfs_vmobject_wunlock(obj);
482 				vm_page_busy_sleep(pp, "zfsmwb");
483 				zfs_vmobject_wlock(obj);
484 				continue;
485 			}
486 
487 			ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
488 			vm_page_lock(pp);
489 			vm_page_hold(pp);
490 			vm_page_unlock(pp);
491 
492 		} else
493 			pp = NULL;
494 		break;
495 	}
496 	return (pp);
497 }
498 
499 static void
page_unhold(vm_page_t pp)500 page_unhold(vm_page_t pp)
501 {
502 
503 	vm_page_lock(pp);
504 	vm_page_unhold(pp);
505 	vm_page_unlock(pp);
506 }
507 
508 /*
509  * When a file is memory mapped, we must keep the IO data synchronized
510  * between the DMU cache and the memory mapped pages.  What this means:
511  *
512  * On Write:	If we find a memory mapped page, we write to *both*
513  *		the page and the dmu buffer.
514  */
515 static void
update_pages(vnode_t * vp,int64_t start,int len,objset_t * os,uint64_t oid,int segflg,dmu_tx_t * tx)516 update_pages(vnode_t *vp, int64_t start, int len, objset_t *os, uint64_t oid,
517     int segflg, dmu_tx_t *tx)
518 {
519 	vm_object_t obj;
520 	struct sf_buf *sf;
521 	caddr_t va;
522 	int off;
523 
524 	ASSERT(segflg != UIO_NOCOPY);
525 	ASSERT(vp->v_mount != NULL);
526 	obj = vp->v_object;
527 	ASSERT(obj != NULL);
528 
529 	off = start & PAGEOFFSET;
530 	zfs_vmobject_wlock(obj);
531 	for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
532 		vm_page_t pp;
533 		int nbytes = imin(PAGESIZE - off, len);
534 
535 		if ((pp = page_busy(vp, start, off, nbytes)) != NULL) {
536 			zfs_vmobject_wunlock(obj);
537 
538 			va = zfs_map_page(pp, &sf);
539 			(void) dmu_read(os, oid, start+off, nbytes,
540 			    va+off, DMU_READ_PREFETCH);;
541 			zfs_unmap_page(sf);
542 
543 			zfs_vmobject_wlock(obj);
544 			page_unbusy(pp);
545 		}
546 		len -= nbytes;
547 		off = 0;
548 	}
549 	vm_object_pip_wakeupn(obj, 0);
550 	zfs_vmobject_wunlock(obj);
551 }
552 
553 /*
554  * Read with UIO_NOCOPY flag means that sendfile(2) requests
555  * ZFS to populate a range of page cache pages with data.
556  *
557  * NOTE: this function could be optimized to pre-allocate
558  * all pages in advance, drain exclusive busy on all of them,
559  * map them into contiguous KVA region and populate them
560  * in one single dmu_read() call.
561  */
562 static int
mappedread_sf(vnode_t * vp,int nbytes,uio_t * uio)563 mappedread_sf(vnode_t *vp, int nbytes, uio_t *uio)
564 {
565 	znode_t *zp = VTOZ(vp);
566 	objset_t *os = zp->z_zfsvfs->z_os;
567 	struct sf_buf *sf;
568 	vm_object_t obj;
569 	vm_page_t pp;
570 	int64_t start;
571 	caddr_t va;
572 	int len = nbytes;
573 	int off;
574 	int error = 0;
575 
576 	ASSERT(uio->uio_segflg == UIO_NOCOPY);
577 	ASSERT(vp->v_mount != NULL);
578 	obj = vp->v_object;
579 	ASSERT(obj != NULL);
580 	ASSERT((uio->uio_loffset & PAGEOFFSET) == 0);
581 
582 	zfs_vmobject_wlock(obj);
583 	for (start = uio->uio_loffset; len > 0; start += PAGESIZE) {
584 		int bytes = MIN(PAGESIZE, len);
585 
586 		pp = vm_page_grab(obj, OFF_TO_IDX(start), VM_ALLOC_SBUSY |
587 		    VM_ALLOC_NORMAL | VM_ALLOC_IGN_SBUSY);
588 		if (pp->valid == 0) {
589 			zfs_vmobject_wunlock(obj);
590 			va = zfs_map_page(pp, &sf);
591 			error = dmu_read(os, zp->z_id, start, bytes, va,
592 			    DMU_READ_PREFETCH);
593 			if (bytes != PAGESIZE && error == 0)
594 				bzero(va + bytes, PAGESIZE - bytes);
595 			zfs_unmap_page(sf);
596 			zfs_vmobject_wlock(obj);
597 			vm_page_sunbusy(pp);
598 			vm_page_lock(pp);
599 			if (error) {
600 				if (pp->wire_count == 0 && pp->valid == 0 &&
601 				    !vm_page_busied(pp))
602 					vm_page_free(pp);
603 			} else {
604 				pp->valid = VM_PAGE_BITS_ALL;
605 				vm_page_activate(pp);
606 			}
607 			vm_page_unlock(pp);
608 		} else {
609 			ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
610 			vm_page_sunbusy(pp);
611 		}
612 		if (error)
613 			break;
614 		uio->uio_resid -= bytes;
615 		uio->uio_offset += bytes;
616 		len -= bytes;
617 	}
618 	zfs_vmobject_wunlock(obj);
619 	return (error);
620 }
621 
622 /*
623  * When a file is memory mapped, we must keep the IO data synchronized
624  * between the DMU cache and the memory mapped pages.  What this means:
625  *
626  * On Read:	We "read" preferentially from memory mapped pages,
627  *		else we default from the dmu buffer.
628  *
629  * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
630  *	 the file is memory mapped.
631  */
632 static int
mappedread(vnode_t * vp,int nbytes,uio_t * uio)633 mappedread(vnode_t *vp, int nbytes, uio_t *uio)
634 {
635 	znode_t *zp = VTOZ(vp);
636 	vm_object_t obj;
637 	int64_t start;
638 	caddr_t va;
639 	int len = nbytes;
640 	int off;
641 	int error = 0;
642 
643 	ASSERT(vp->v_mount != NULL);
644 	obj = vp->v_object;
645 	ASSERT(obj != NULL);
646 
647 	start = uio->uio_loffset;
648 	off = start & PAGEOFFSET;
649 	zfs_vmobject_wlock(obj);
650 	for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
651 		vm_page_t pp;
652 		uint64_t bytes = MIN(PAGESIZE - off, len);
653 
654 		if (pp = page_hold(vp, start)) {
655 			struct sf_buf *sf;
656 			caddr_t va;
657 
658 			zfs_vmobject_wunlock(obj);
659 			va = zfs_map_page(pp, &sf);
660 			error = uiomove(va + off, bytes, UIO_READ, uio);
661 			zfs_unmap_page(sf);
662 			zfs_vmobject_wlock(obj);
663 			page_unhold(pp);
664 		} else {
665 			zfs_vmobject_wunlock(obj);
666 			error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl),
667 			    uio, bytes);
668 			zfs_vmobject_wlock(obj);
669 		}
670 		len -= bytes;
671 		off = 0;
672 		if (error)
673 			break;
674 	}
675 	zfs_vmobject_wunlock(obj);
676 	return (error);
677 }
678 
679 offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */
680 
681 /*
682  * Read bytes from specified file into supplied buffer.
683  *
684  *	IN:	vp	- vnode of file to be read from.
685  *		uio	- structure supplying read location, range info,
686  *			  and return buffer.
687  *		ioflag	- SYNC flags; used to provide FRSYNC semantics.
688  *		cr	- credentials of caller.
689  *		ct	- caller context
690  *
691  *	OUT:	uio	- updated offset and range, buffer filled.
692  *
693  *	RETURN:	0 on success, error code on failure.
694  *
695  * Side Effects:
696  *	vp - atime updated if byte count > 0
697  */
698 /* ARGSUSED */
699 static int
zfs_read(vnode_t * vp,uio_t * uio,int ioflag,cred_t * cr,caller_context_t * ct)700 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
701 {
702 	znode_t		*zp = VTOZ(vp);
703 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
704 	ssize_t		n, nbytes;
705 	int		error = 0;
706 	rl_t		*rl;
707 	xuio_t		*xuio = NULL;
708 
709 	ZFS_ENTER(zfsvfs);
710 	ZFS_VERIFY_ZP(zp);
711 
712 	if (zp->z_pflags & ZFS_AV_QUARANTINED) {
713 		ZFS_EXIT(zfsvfs);
714 		return (SET_ERROR(EACCES));
715 	}
716 
717 	/*
718 	 * Validate file offset
719 	 */
720 	if (uio->uio_loffset < (offset_t)0) {
721 		ZFS_EXIT(zfsvfs);
722 		return (SET_ERROR(EINVAL));
723 	}
724 
725 	/*
726 	 * Fasttrack empty reads
727 	 */
728 	if (uio->uio_resid == 0) {
729 		ZFS_EXIT(zfsvfs);
730 		return (0);
731 	}
732 
733 	/*
734 	 * Check for mandatory locks
735 	 */
736 	if (MANDMODE(zp->z_mode)) {
737 		if (error = chklock(vp, FREAD,
738 		    uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) {
739 			ZFS_EXIT(zfsvfs);
740 			return (error);
741 		}
742 	}
743 
744 	/*
745 	 * If we're in FRSYNC mode, sync out this znode before reading it.
746 	 */
747 	if (zfsvfs->z_log &&
748 	    (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS))
749 		zil_commit(zfsvfs->z_log, zp->z_id);
750 
751 	/*
752 	 * Lock the range against changes.
753 	 */
754 	rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER);
755 
756 	/*
757 	 * If we are reading past end-of-file we can skip
758 	 * to the end; but we might still need to set atime.
759 	 */
760 	if (uio->uio_loffset >= zp->z_size) {
761 		error = 0;
762 		goto out;
763 	}
764 
765 	ASSERT(uio->uio_loffset < zp->z_size);
766 	n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset);
767 
768 #ifdef illumos
769 	if ((uio->uio_extflg == UIO_XUIO) &&
770 	    (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) {
771 		int nblk;
772 		int blksz = zp->z_blksz;
773 		uint64_t offset = uio->uio_loffset;
774 
775 		xuio = (xuio_t *)uio;
776 		if ((ISP2(blksz))) {
777 			nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset,
778 			    blksz)) / blksz;
779 		} else {
780 			ASSERT(offset + n <= blksz);
781 			nblk = 1;
782 		}
783 		(void) dmu_xuio_init(xuio, nblk);
784 
785 		if (vn_has_cached_data(vp)) {
786 			/*
787 			 * For simplicity, we always allocate a full buffer
788 			 * even if we only expect to read a portion of a block.
789 			 */
790 			while (--nblk >= 0) {
791 				(void) dmu_xuio_add(xuio,
792 				    dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
793 				    blksz), 0, blksz);
794 			}
795 		}
796 	}
797 #endif	/* illumos */
798 
799 	while (n > 0) {
800 		nbytes = MIN(n, zfs_read_chunk_size -
801 		    P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
802 
803 #ifdef __FreeBSD__
804 		if (uio->uio_segflg == UIO_NOCOPY)
805 			error = mappedread_sf(vp, nbytes, uio);
806 		else
807 #endif /* __FreeBSD__ */
808 		if (vn_has_cached_data(vp)) {
809 			error = mappedread(vp, nbytes, uio);
810 		} else {
811 			error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl),
812 			    uio, nbytes);
813 		}
814 		if (error) {
815 			/* convert checksum errors into IO errors */
816 			if (error == ECKSUM)
817 				error = SET_ERROR(EIO);
818 			break;
819 		}
820 
821 		n -= nbytes;
822 	}
823 out:
824 	zfs_range_unlock(rl);
825 
826 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
827 	ZFS_EXIT(zfsvfs);
828 	return (error);
829 }
830 
831 /*
832  * Write the bytes to a file.
833  *
834  *	IN:	vp	- vnode of file to be written to.
835  *		uio	- structure supplying write location, range info,
836  *			  and data buffer.
837  *		ioflag	- FAPPEND, FSYNC, and/or FDSYNC.  FAPPEND is
838  *			  set if in append mode.
839  *		cr	- credentials of caller.
840  *		ct	- caller context (NFS/CIFS fem monitor only)
841  *
842  *	OUT:	uio	- updated offset and range.
843  *
844  *	RETURN:	0 on success, error code on failure.
845  *
846  * Timestamps:
847  *	vp - ctime|mtime updated if byte count > 0
848  */
849 
850 /* ARGSUSED */
851 static int
zfs_write(vnode_t * vp,uio_t * uio,int ioflag,cred_t * cr,caller_context_t * ct)852 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
853 {
854 	znode_t		*zp = VTOZ(vp);
855 	rlim64_t	limit = MAXOFFSET_T;
856 	ssize_t		start_resid = uio->uio_resid;
857 	ssize_t		tx_bytes;
858 	uint64_t	end_size;
859 	dmu_tx_t	*tx;
860 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
861 	zilog_t		*zilog;
862 	offset_t	woff;
863 	ssize_t		n, nbytes;
864 	rl_t		*rl;
865 	int		max_blksz = zfsvfs->z_max_blksz;
866 	int		error = 0;
867 	arc_buf_t	*abuf;
868 	iovec_t		*aiov = NULL;
869 	xuio_t		*xuio = NULL;
870 	int		i_iov = 0;
871 	int		iovcnt = uio->uio_iovcnt;
872 	iovec_t		*iovp = uio->uio_iov;
873 	int		write_eof;
874 	int		count = 0;
875 	sa_bulk_attr_t	bulk[4];
876 	uint64_t	mtime[2], ctime[2];
877 
878 	/*
879 	 * Fasttrack empty write
880 	 */
881 	n = start_resid;
882 	if (n == 0)
883 		return (0);
884 
885 	if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
886 		limit = MAXOFFSET_T;
887 
888 	ZFS_ENTER(zfsvfs);
889 	ZFS_VERIFY_ZP(zp);
890 
891 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
892 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
893 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
894 	    &zp->z_size, 8);
895 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
896 	    &zp->z_pflags, 8);
897 
898 	/*
899 	 * In a case vp->v_vfsp != zp->z_zfsvfs->z_vfs (e.g. snapshots) our
900 	 * callers might not be able to detect properly that we are read-only,
901 	 * so check it explicitly here.
902 	 */
903 	if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
904 		ZFS_EXIT(zfsvfs);
905 		return (SET_ERROR(EROFS));
906 	}
907 
908 	/*
909 	 * If immutable or not appending then return EPERM
910 	 */
911 	if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
912 	    ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
913 	    (uio->uio_loffset < zp->z_size))) {
914 		ZFS_EXIT(zfsvfs);
915 		return (SET_ERROR(EPERM));
916 	}
917 
918 	zilog = zfsvfs->z_log;
919 
920 	/*
921 	 * Validate file offset
922 	 */
923 	woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset;
924 	if (woff < 0) {
925 		ZFS_EXIT(zfsvfs);
926 		return (SET_ERROR(EINVAL));
927 	}
928 
929 	/*
930 	 * Check for mandatory locks before calling zfs_range_lock()
931 	 * in order to prevent a deadlock with locks set via fcntl().
932 	 */
933 	if (MANDMODE((mode_t)zp->z_mode) &&
934 	    (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) {
935 		ZFS_EXIT(zfsvfs);
936 		return (error);
937 	}
938 
939 #ifdef illumos
940 	/*
941 	 * Pre-fault the pages to ensure slow (eg NFS) pages
942 	 * don't hold up txg.
943 	 * Skip this if uio contains loaned arc_buf.
944 	 */
945 	if ((uio->uio_extflg == UIO_XUIO) &&
946 	    (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY))
947 		xuio = (xuio_t *)uio;
948 	else
949 		uio_prefaultpages(MIN(n, max_blksz), uio);
950 #endif
951 
952 	/*
953 	 * If in append mode, set the io offset pointer to eof.
954 	 */
955 	if (ioflag & FAPPEND) {
956 		/*
957 		 * Obtain an appending range lock to guarantee file append
958 		 * semantics.  We reset the write offset once we have the lock.
959 		 */
960 		rl = zfs_range_lock(zp, 0, n, RL_APPEND);
961 		woff = rl->r_off;
962 		if (rl->r_len == UINT64_MAX) {
963 			/*
964 			 * We overlocked the file because this write will cause
965 			 * the file block size to increase.
966 			 * Note that zp_size cannot change with this lock held.
967 			 */
968 			woff = zp->z_size;
969 		}
970 		uio->uio_loffset = woff;
971 	} else {
972 		/*
973 		 * Note that if the file block size will change as a result of
974 		 * this write, then this range lock will lock the entire file
975 		 * so that we can re-write the block safely.
976 		 */
977 		rl = zfs_range_lock(zp, woff, n, RL_WRITER);
978 	}
979 
980 	if (vn_rlimit_fsize(vp, uio, uio->uio_td)) {
981 		zfs_range_unlock(rl);
982 		ZFS_EXIT(zfsvfs);
983 		return (EFBIG);
984 	}
985 
986 	if (woff >= limit) {
987 		zfs_range_unlock(rl);
988 		ZFS_EXIT(zfsvfs);
989 		return (SET_ERROR(EFBIG));
990 	}
991 
992 	if ((woff + n) > limit || woff > (limit - n))
993 		n = limit - woff;
994 
995 	/* Will this write extend the file length? */
996 	write_eof = (woff + n > zp->z_size);
997 
998 	end_size = MAX(zp->z_size, woff + n);
999 
1000 	/*
1001 	 * Write the file in reasonable size chunks.  Each chunk is written
1002 	 * in a separate transaction; this keeps the intent log records small
1003 	 * and allows us to do more fine-grained space accounting.
1004 	 */
1005 	while (n > 0) {
1006 		abuf = NULL;
1007 		woff = uio->uio_loffset;
1008 		if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
1009 		    zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
1010 			if (abuf != NULL)
1011 				dmu_return_arcbuf(abuf);
1012 			error = SET_ERROR(EDQUOT);
1013 			break;
1014 		}
1015 
1016 		if (xuio && abuf == NULL) {
1017 			ASSERT(i_iov < iovcnt);
1018 			aiov = &iovp[i_iov];
1019 			abuf = dmu_xuio_arcbuf(xuio, i_iov);
1020 			dmu_xuio_clear(xuio, i_iov);
1021 			DTRACE_PROBE3(zfs_cp_write, int, i_iov,
1022 			    iovec_t *, aiov, arc_buf_t *, abuf);
1023 			ASSERT((aiov->iov_base == abuf->b_data) ||
1024 			    ((char *)aiov->iov_base - (char *)abuf->b_data +
1025 			    aiov->iov_len == arc_buf_size(abuf)));
1026 			i_iov++;
1027 		} else if (abuf == NULL && n >= max_blksz &&
1028 		    woff >= zp->z_size &&
1029 		    P2PHASE(woff, max_blksz) == 0 &&
1030 		    zp->z_blksz == max_blksz) {
1031 			/*
1032 			 * This write covers a full block.  "Borrow" a buffer
1033 			 * from the dmu so that we can fill it before we enter
1034 			 * a transaction.  This avoids the possibility of
1035 			 * holding up the transaction if the data copy hangs
1036 			 * up on a pagefault (e.g., from an NFS server mapping).
1037 			 */
1038 			size_t cbytes;
1039 
1040 			abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
1041 			    max_blksz);
1042 			ASSERT(abuf != NULL);
1043 			ASSERT(arc_buf_size(abuf) == max_blksz);
1044 			if (error = uiocopy(abuf->b_data, max_blksz,
1045 			    UIO_WRITE, uio, &cbytes)) {
1046 				dmu_return_arcbuf(abuf);
1047 				break;
1048 			}
1049 			ASSERT(cbytes == max_blksz);
1050 		}
1051 
1052 		/*
1053 		 * Start a transaction.
1054 		 */
1055 		tx = dmu_tx_create(zfsvfs->z_os);
1056 		dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1057 		dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
1058 		zfs_sa_upgrade_txholds(tx, zp);
1059 		error = dmu_tx_assign(tx, TXG_WAIT);
1060 		if (error) {
1061 			dmu_tx_abort(tx);
1062 			if (abuf != NULL)
1063 				dmu_return_arcbuf(abuf);
1064 			break;
1065 		}
1066 
1067 		/*
1068 		 * If zfs_range_lock() over-locked we grow the blocksize
1069 		 * and then reduce the lock range.  This will only happen
1070 		 * on the first iteration since zfs_range_reduce() will
1071 		 * shrink down r_len to the appropriate size.
1072 		 */
1073 		if (rl->r_len == UINT64_MAX) {
1074 			uint64_t new_blksz;
1075 
1076 			if (zp->z_blksz > max_blksz) {
1077 				/*
1078 				 * File's blocksize is already larger than the
1079 				 * "recordsize" property.  Only let it grow to
1080 				 * the next power of 2.
1081 				 */
1082 				ASSERT(!ISP2(zp->z_blksz));
1083 				new_blksz = MIN(end_size,
1084 				    1 << highbit64(zp->z_blksz));
1085 			} else {
1086 				new_blksz = MIN(end_size, max_blksz);
1087 			}
1088 			zfs_grow_blocksize(zp, new_blksz, tx);
1089 			zfs_range_reduce(rl, woff, n);
1090 		}
1091 
1092 		/*
1093 		 * XXX - should we really limit each write to z_max_blksz?
1094 		 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
1095 		 */
1096 		nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
1097 
1098 		if (woff + nbytes > zp->z_size)
1099 			vnode_pager_setsize(vp, woff + nbytes);
1100 
1101 		if (abuf == NULL) {
1102 			tx_bytes = uio->uio_resid;
1103 			error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
1104 			    uio, nbytes, tx);
1105 			tx_bytes -= uio->uio_resid;
1106 		} else {
1107 			tx_bytes = nbytes;
1108 			ASSERT(xuio == NULL || tx_bytes == aiov->iov_len);
1109 			/*
1110 			 * If this is not a full block write, but we are
1111 			 * extending the file past EOF and this data starts
1112 			 * block-aligned, use assign_arcbuf().  Otherwise,
1113 			 * write via dmu_write().
1114 			 */
1115 			if (tx_bytes < max_blksz && (!write_eof ||
1116 			    aiov->iov_base != abuf->b_data)) {
1117 				ASSERT(xuio);
1118 				dmu_write(zfsvfs->z_os, zp->z_id, woff,
1119 				    aiov->iov_len, aiov->iov_base, tx);
1120 				dmu_return_arcbuf(abuf);
1121 				xuio_stat_wbuf_copied();
1122 			} else {
1123 				ASSERT(xuio || tx_bytes == max_blksz);
1124 				dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl),
1125 				    woff, abuf, tx);
1126 			}
1127 			ASSERT(tx_bytes <= uio->uio_resid);
1128 			uioskip(uio, tx_bytes);
1129 		}
1130 		if (tx_bytes && vn_has_cached_data(vp)) {
1131 			update_pages(vp, woff, tx_bytes, zfsvfs->z_os,
1132 			    zp->z_id, uio->uio_segflg, tx);
1133 		}
1134 
1135 		/*
1136 		 * If we made no progress, we're done.  If we made even
1137 		 * partial progress, update the znode and ZIL accordingly.
1138 		 */
1139 		if (tx_bytes == 0) {
1140 			(void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
1141 			    (void *)&zp->z_size, sizeof (uint64_t), tx);
1142 			dmu_tx_commit(tx);
1143 			ASSERT(error != 0);
1144 			break;
1145 		}
1146 
1147 		/*
1148 		 * Clear Set-UID/Set-GID bits on successful write if not
1149 		 * privileged and at least one of the excute bits is set.
1150 		 *
1151 		 * It would be nice to to this after all writes have
1152 		 * been done, but that would still expose the ISUID/ISGID
1153 		 * to another app after the partial write is committed.
1154 		 *
1155 		 * Note: we don't call zfs_fuid_map_id() here because
1156 		 * user 0 is not an ephemeral uid.
1157 		 */
1158 		mutex_enter(&zp->z_acl_lock);
1159 		if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
1160 		    (S_IXUSR >> 6))) != 0 &&
1161 		    (zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
1162 		    secpolicy_vnode_setid_retain(vp, cr,
1163 		    (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) {
1164 			uint64_t newmode;
1165 			zp->z_mode &= ~(S_ISUID | S_ISGID);
1166 			newmode = zp->z_mode;
1167 			(void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs),
1168 			    (void *)&newmode, sizeof (uint64_t), tx);
1169 		}
1170 		mutex_exit(&zp->z_acl_lock);
1171 
1172 		zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
1173 		    B_TRUE);
1174 
1175 		/*
1176 		 * Update the file size (zp_size) if it has changed;
1177 		 * account for possible concurrent updates.
1178 		 */
1179 		while ((end_size = zp->z_size) < uio->uio_loffset) {
1180 			(void) atomic_cas_64(&zp->z_size, end_size,
1181 			    uio->uio_loffset);
1182 			ASSERT(error == 0);
1183 		}
1184 		/*
1185 		 * If we are replaying and eof is non zero then force
1186 		 * the file size to the specified eof. Note, there's no
1187 		 * concurrency during replay.
1188 		 */
1189 		if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0)
1190 			zp->z_size = zfsvfs->z_replay_eof;
1191 
1192 		error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
1193 
1194 		zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag);
1195 		dmu_tx_commit(tx);
1196 
1197 		if (error != 0)
1198 			break;
1199 		ASSERT(tx_bytes == nbytes);
1200 		n -= nbytes;
1201 
1202 #ifdef illumos
1203 		if (!xuio && n > 0)
1204 			uio_prefaultpages(MIN(n, max_blksz), uio);
1205 #endif
1206 	}
1207 
1208 	zfs_range_unlock(rl);
1209 
1210 	/*
1211 	 * If we're in replay mode, or we made no progress, return error.
1212 	 * Otherwise, it's at least a partial write, so it's successful.
1213 	 */
1214 	if (zfsvfs->z_replay || uio->uio_resid == start_resid) {
1215 		ZFS_EXIT(zfsvfs);
1216 		return (error);
1217 	}
1218 
1219 	if (ioflag & (FSYNC | FDSYNC) ||
1220 	    zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1221 		zil_commit(zilog, zp->z_id);
1222 
1223 	ZFS_EXIT(zfsvfs);
1224 	return (0);
1225 }
1226 
1227 void
zfs_get_done(zgd_t * zgd,int error)1228 zfs_get_done(zgd_t *zgd, int error)
1229 {
1230 	znode_t *zp = zgd->zgd_private;
1231 	objset_t *os = zp->z_zfsvfs->z_os;
1232 
1233 	if (zgd->zgd_db)
1234 		dmu_buf_rele(zgd->zgd_db, zgd);
1235 
1236 	zfs_range_unlock(zgd->zgd_rl);
1237 
1238 	/*
1239 	 * Release the vnode asynchronously as we currently have the
1240 	 * txg stopped from syncing.
1241 	 */
1242 	VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1243 
1244 	if (error == 0 && zgd->zgd_bp)
1245 		zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1246 
1247 	kmem_free(zgd, sizeof (zgd_t));
1248 }
1249 
1250 #ifdef DEBUG
1251 static int zil_fault_io = 0;
1252 #endif
1253 
1254 /*
1255  * Get data to generate a TX_WRITE intent log record.
1256  */
1257 int
zfs_get_data(void * arg,lr_write_t * lr,char * buf,zio_t * zio)1258 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1259 {
1260 	zfsvfs_t *zfsvfs = arg;
1261 	objset_t *os = zfsvfs->z_os;
1262 	znode_t *zp;
1263 	uint64_t object = lr->lr_foid;
1264 	uint64_t offset = lr->lr_offset;
1265 	uint64_t size = lr->lr_length;
1266 	blkptr_t *bp = &lr->lr_blkptr;
1267 	dmu_buf_t *db;
1268 	zgd_t *zgd;
1269 	int error = 0;
1270 
1271 	ASSERT(zio != NULL);
1272 	ASSERT(size != 0);
1273 
1274 	/*
1275 	 * Nothing to do if the file has been removed
1276 	 */
1277 	if (zfs_zget(zfsvfs, object, &zp) != 0)
1278 		return (SET_ERROR(ENOENT));
1279 	if (zp->z_unlinked) {
1280 		/*
1281 		 * Release the vnode asynchronously as we currently have the
1282 		 * txg stopped from syncing.
1283 		 */
1284 		VN_RELE_ASYNC(ZTOV(zp),
1285 		    dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1286 		return (SET_ERROR(ENOENT));
1287 	}
1288 
1289 	zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1290 	zgd->zgd_zilog = zfsvfs->z_log;
1291 	zgd->zgd_private = zp;
1292 
1293 	/*
1294 	 * Write records come in two flavors: immediate and indirect.
1295 	 * For small writes it's cheaper to store the data with the
1296 	 * log record (immediate); for large writes it's cheaper to
1297 	 * sync the data and get a pointer to it (indirect) so that
1298 	 * we don't have to write the data twice.
1299 	 */
1300 	if (buf != NULL) { /* immediate write */
1301 		zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER);
1302 		/* test for truncation needs to be done while range locked */
1303 		if (offset >= zp->z_size) {
1304 			error = SET_ERROR(ENOENT);
1305 		} else {
1306 			error = dmu_read(os, object, offset, size, buf,
1307 			    DMU_READ_NO_PREFETCH);
1308 		}
1309 		ASSERT(error == 0 || error == ENOENT);
1310 	} else { /* indirect write */
1311 		/*
1312 		 * Have to lock the whole block to ensure when it's
1313 		 * written out and it's checksum is being calculated
1314 		 * that no one can change the data. We need to re-check
1315 		 * blocksize after we get the lock in case it's changed!
1316 		 */
1317 		for (;;) {
1318 			uint64_t blkoff;
1319 			size = zp->z_blksz;
1320 			blkoff = ISP2(size) ? P2PHASE(offset, size) : offset;
1321 			offset -= blkoff;
1322 			zgd->zgd_rl = zfs_range_lock(zp, offset, size,
1323 			    RL_READER);
1324 			if (zp->z_blksz == size)
1325 				break;
1326 			offset += blkoff;
1327 			zfs_range_unlock(zgd->zgd_rl);
1328 		}
1329 		/* test for truncation needs to be done while range locked */
1330 		if (lr->lr_offset >= zp->z_size)
1331 			error = SET_ERROR(ENOENT);
1332 #ifdef DEBUG
1333 		if (zil_fault_io) {
1334 			error = SET_ERROR(EIO);
1335 			zil_fault_io = 0;
1336 		}
1337 #endif
1338 		if (error == 0)
1339 			error = dmu_buf_hold(os, object, offset, zgd, &db,
1340 			    DMU_READ_NO_PREFETCH);
1341 
1342 		if (error == 0) {
1343 			blkptr_t *obp = dmu_buf_get_blkptr(db);
1344 			if (obp) {
1345 				ASSERT(BP_IS_HOLE(bp));
1346 				*bp = *obp;
1347 			}
1348 
1349 			zgd->zgd_db = db;
1350 			zgd->zgd_bp = bp;
1351 
1352 			ASSERT(db->db_offset == offset);
1353 			ASSERT(db->db_size == size);
1354 
1355 			error = dmu_sync(zio, lr->lr_common.lrc_txg,
1356 			    zfs_get_done, zgd);
1357 			ASSERT(error || lr->lr_length <= zp->z_blksz);
1358 
1359 			/*
1360 			 * On success, we need to wait for the write I/O
1361 			 * initiated by dmu_sync() to complete before we can
1362 			 * release this dbuf.  We will finish everything up
1363 			 * in the zfs_get_done() callback.
1364 			 */
1365 			if (error == 0)
1366 				return (0);
1367 
1368 			if (error == EALREADY) {
1369 				lr->lr_common.lrc_txtype = TX_WRITE2;
1370 				error = 0;
1371 			}
1372 		}
1373 	}
1374 
1375 	zfs_get_done(zgd, error);
1376 
1377 	return (error);
1378 }
1379 
1380 /*ARGSUSED*/
1381 static int
zfs_access(vnode_t * vp,int mode,int flag,cred_t * cr,caller_context_t * ct)1382 zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr,
1383     caller_context_t *ct)
1384 {
1385 	znode_t *zp = VTOZ(vp);
1386 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1387 	int error;
1388 
1389 	ZFS_ENTER(zfsvfs);
1390 	ZFS_VERIFY_ZP(zp);
1391 
1392 	if (flag & V_ACE_MASK)
1393 		error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
1394 	else
1395 		error = zfs_zaccess_rwx(zp, mode, flag, cr);
1396 
1397 	ZFS_EXIT(zfsvfs);
1398 	return (error);
1399 }
1400 
1401 /*
1402  * If vnode is for a device return a specfs vnode instead.
1403  */
1404 static int
specvp_check(vnode_t ** vpp,cred_t * cr)1405 specvp_check(vnode_t **vpp, cred_t *cr)
1406 {
1407 	int error = 0;
1408 
1409 	if (IS_DEVVP(*vpp)) {
1410 		struct vnode *svp;
1411 
1412 		svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1413 		VN_RELE(*vpp);
1414 		if (svp == NULL)
1415 			error = SET_ERROR(ENOSYS);
1416 		*vpp = svp;
1417 	}
1418 	return (error);
1419 }
1420 
1421 
1422 /*
1423  * Lookup an entry in a directory, or an extended attribute directory.
1424  * If it exists, return a held vnode reference for it.
1425  *
1426  *	IN:	dvp	- vnode of directory to search.
1427  *		nm	- name of entry to lookup.
1428  *		pnp	- full pathname to lookup [UNUSED].
1429  *		flags	- LOOKUP_XATTR set if looking for an attribute.
1430  *		rdir	- root directory vnode [UNUSED].
1431  *		cr	- credentials of caller.
1432  *		ct	- caller context
1433  *		direntflags - directory lookup flags
1434  *		realpnp - returned pathname.
1435  *
1436  *	OUT:	vpp	- vnode of located entry, NULL if not found.
1437  *
1438  *	RETURN:	0 on success, error code on failure.
1439  *
1440  * Timestamps:
1441  *	NA
1442  */
1443 /* ARGSUSED */
1444 static int
zfs_lookup(vnode_t * dvp,char * nm,vnode_t ** vpp,struct componentname * cnp,int nameiop,cred_t * cr,kthread_t * td,int flags)1445 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct componentname *cnp,
1446     int nameiop, cred_t *cr, kthread_t *td, int flags)
1447 {
1448 	znode_t *zdp = VTOZ(dvp);
1449 	zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
1450 	int	error = 0;
1451 	int *direntflags = NULL;
1452 	void *realpnp = NULL;
1453 
1454 	/* fast path */
1455 	if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
1456 
1457 		if (dvp->v_type != VDIR) {
1458 			return (SET_ERROR(ENOTDIR));
1459 		} else if (zdp->z_sa_hdl == NULL) {
1460 			return (SET_ERROR(EIO));
1461 		}
1462 
1463 		if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
1464 			error = zfs_fastaccesschk_execute(zdp, cr);
1465 			if (!error) {
1466 				*vpp = dvp;
1467 				VN_HOLD(*vpp);
1468 				return (0);
1469 			}
1470 			return (error);
1471 		} else {
1472 			vnode_t *tvp = dnlc_lookup(dvp, nm);
1473 
1474 			if (tvp) {
1475 				error = zfs_fastaccesschk_execute(zdp, cr);
1476 				if (error) {
1477 					VN_RELE(tvp);
1478 					return (error);
1479 				}
1480 				if (tvp == DNLC_NO_VNODE) {
1481 					VN_RELE(tvp);
1482 					return (SET_ERROR(ENOENT));
1483 				} else {
1484 					*vpp = tvp;
1485 					return (specvp_check(vpp, cr));
1486 				}
1487 			}
1488 		}
1489 	}
1490 
1491 	DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm);
1492 
1493 	ZFS_ENTER(zfsvfs);
1494 	ZFS_VERIFY_ZP(zdp);
1495 
1496 	*vpp = NULL;
1497 
1498 	if (flags & LOOKUP_XATTR) {
1499 #ifdef TODO
1500 		/*
1501 		 * If the xattr property is off, refuse the lookup request.
1502 		 */
1503 		if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) {
1504 			ZFS_EXIT(zfsvfs);
1505 			return (SET_ERROR(EINVAL));
1506 		}
1507 #endif
1508 
1509 		/*
1510 		 * We don't allow recursive attributes..
1511 		 * Maybe someday we will.
1512 		 */
1513 		if (zdp->z_pflags & ZFS_XATTR) {
1514 			ZFS_EXIT(zfsvfs);
1515 			return (SET_ERROR(EINVAL));
1516 		}
1517 
1518 		if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) {
1519 			ZFS_EXIT(zfsvfs);
1520 			return (error);
1521 		}
1522 
1523 		/*
1524 		 * Do we have permission to get into attribute directory?
1525 		 */
1526 
1527 		if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0,
1528 		    B_FALSE, cr)) {
1529 			VN_RELE(*vpp);
1530 			*vpp = NULL;
1531 		}
1532 
1533 		ZFS_EXIT(zfsvfs);
1534 		return (error);
1535 	}
1536 
1537 	if (dvp->v_type != VDIR) {
1538 		ZFS_EXIT(zfsvfs);
1539 		return (SET_ERROR(ENOTDIR));
1540 	}
1541 
1542 	/*
1543 	 * Check accessibility of directory.
1544 	 */
1545 
1546 	if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) {
1547 		ZFS_EXIT(zfsvfs);
1548 		return (error);
1549 	}
1550 
1551 	if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
1552 	    NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1553 		ZFS_EXIT(zfsvfs);
1554 		return (SET_ERROR(EILSEQ));
1555 	}
1556 
1557 	error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp);
1558 	if (error == 0)
1559 		error = specvp_check(vpp, cr);
1560 
1561 	/* Translate errors and add SAVENAME when needed. */
1562 	if (cnp->cn_flags & ISLASTCN) {
1563 		switch (nameiop) {
1564 		case CREATE:
1565 		case RENAME:
1566 			if (error == ENOENT) {
1567 				error = EJUSTRETURN;
1568 				cnp->cn_flags |= SAVENAME;
1569 				break;
1570 			}
1571 			/* FALLTHROUGH */
1572 		case DELETE:
1573 			if (error == 0)
1574 				cnp->cn_flags |= SAVENAME;
1575 			break;
1576 		}
1577 	}
1578 	if (error == 0 && (nm[0] != '.' || nm[1] != '\0')) {
1579 		int ltype = 0;
1580 
1581 		if (cnp->cn_flags & ISDOTDOT) {
1582 			ltype = VOP_ISLOCKED(dvp);
1583 			VOP_UNLOCK(dvp, 0);
1584 		}
1585 		ZFS_EXIT(zfsvfs);
1586 		error = vn_lock(*vpp, cnp->cn_lkflags);
1587 		if (cnp->cn_flags & ISDOTDOT)
1588 			vn_lock(dvp, ltype | LK_RETRY);
1589 		if (error != 0) {
1590 			VN_RELE(*vpp);
1591 			*vpp = NULL;
1592 			return (error);
1593 		}
1594 	} else {
1595 		ZFS_EXIT(zfsvfs);
1596 	}
1597 
1598 #ifdef FREEBSD_NAMECACHE
1599 	/*
1600 	 * Insert name into cache (as non-existent) if appropriate.
1601 	 */
1602 	if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) != 0)
1603 		cache_enter(dvp, *vpp, cnp);
1604 	/*
1605 	 * Insert name into cache if appropriate.
1606 	 */
1607 	if (error == 0 && (cnp->cn_flags & MAKEENTRY)) {
1608 		if (!(cnp->cn_flags & ISLASTCN) ||
1609 		    (nameiop != DELETE && nameiop != RENAME)) {
1610 			cache_enter(dvp, *vpp, cnp);
1611 		}
1612 	}
1613 #endif
1614 
1615 	return (error);
1616 }
1617 
1618 /*
1619  * Attempt to create a new entry in a directory.  If the entry
1620  * already exists, truncate the file if permissible, else return
1621  * an error.  Return the vp of the created or trunc'd file.
1622  *
1623  *	IN:	dvp	- vnode of directory to put new file entry in.
1624  *		name	- name of new file entry.
1625  *		vap	- attributes of new file.
1626  *		excl	- flag indicating exclusive or non-exclusive mode.
1627  *		mode	- mode to open file with.
1628  *		cr	- credentials of caller.
1629  *		flag	- large file flag [UNUSED].
1630  *		ct	- caller context
1631  *		vsecp	- ACL to be set
1632  *
1633  *	OUT:	vpp	- vnode of created or trunc'd entry.
1634  *
1635  *	RETURN:	0 on success, error code on failure.
1636  *
1637  * Timestamps:
1638  *	dvp - ctime|mtime updated if new entry created
1639  *	 vp - ctime|mtime always, atime if new
1640  */
1641 
1642 /* ARGSUSED */
1643 static int
zfs_create(vnode_t * dvp,char * name,vattr_t * vap,int excl,int mode,vnode_t ** vpp,cred_t * cr,kthread_t * td)1644 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, int excl, int mode,
1645     vnode_t **vpp, cred_t *cr, kthread_t *td)
1646 {
1647 	znode_t		*zp, *dzp = VTOZ(dvp);
1648 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1649 	zilog_t		*zilog;
1650 	objset_t	*os;
1651 	zfs_dirlock_t	*dl;
1652 	dmu_tx_t	*tx;
1653 	int		error;
1654 	ksid_t		*ksid;
1655 	uid_t		uid;
1656 	gid_t		gid = crgetgid(cr);
1657 	zfs_acl_ids_t   acl_ids;
1658 	boolean_t	fuid_dirtied;
1659 	boolean_t	have_acl = B_FALSE;
1660 	boolean_t	waited = B_FALSE;
1661 	void		*vsecp = NULL;
1662 	int		flag = 0;
1663 
1664 	/*
1665 	 * If we have an ephemeral id, ACL, or XVATTR then
1666 	 * make sure file system is at proper version
1667 	 */
1668 
1669 	ksid = crgetsid(cr, KSID_OWNER);
1670 	if (ksid)
1671 		uid = ksid_getid(ksid);
1672 	else
1673 		uid = crgetuid(cr);
1674 
1675 	if (zfsvfs->z_use_fuids == B_FALSE &&
1676 	    (vsecp || (vap->va_mask & AT_XVATTR) ||
1677 	    IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1678 		return (SET_ERROR(EINVAL));
1679 
1680 	ZFS_ENTER(zfsvfs);
1681 	ZFS_VERIFY_ZP(dzp);
1682 	os = zfsvfs->z_os;
1683 	zilog = zfsvfs->z_log;
1684 
1685 	if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1686 	    NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1687 		ZFS_EXIT(zfsvfs);
1688 		return (SET_ERROR(EILSEQ));
1689 	}
1690 
1691 	if (vap->va_mask & AT_XVATTR) {
1692 		if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap,
1693 		    crgetuid(cr), cr, vap->va_type)) != 0) {
1694 			ZFS_EXIT(zfsvfs);
1695 			return (error);
1696 		}
1697 	}
1698 
1699 	getnewvnode_reserve(1);
1700 
1701 top:
1702 	*vpp = NULL;
1703 
1704 	if ((vap->va_mode & S_ISVTX) && secpolicy_vnode_stky_modify(cr))
1705 		vap->va_mode &= ~S_ISVTX;
1706 
1707 	if (*name == '\0') {
1708 		/*
1709 		 * Null component name refers to the directory itself.
1710 		 */
1711 		VN_HOLD(dvp);
1712 		zp = dzp;
1713 		dl = NULL;
1714 		error = 0;
1715 	} else {
1716 		/* possible VN_HOLD(zp) */
1717 		int zflg = 0;
1718 
1719 		if (flag & FIGNORECASE)
1720 			zflg |= ZCILOOK;
1721 
1722 		error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1723 		    NULL, NULL);
1724 		if (error) {
1725 			if (have_acl)
1726 				zfs_acl_ids_free(&acl_ids);
1727 			if (strcmp(name, "..") == 0)
1728 				error = SET_ERROR(EISDIR);
1729 			getnewvnode_drop_reserve();
1730 			ZFS_EXIT(zfsvfs);
1731 			return (error);
1732 		}
1733 	}
1734 
1735 	if (zp == NULL) {
1736 		uint64_t txtype;
1737 
1738 		/*
1739 		 * Create a new file object and update the directory
1740 		 * to reference it.
1741 		 */
1742 		if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
1743 			if (have_acl)
1744 				zfs_acl_ids_free(&acl_ids);
1745 			goto out;
1746 		}
1747 
1748 		/*
1749 		 * We only support the creation of regular files in
1750 		 * extended attribute directories.
1751 		 */
1752 
1753 		if ((dzp->z_pflags & ZFS_XATTR) &&
1754 		    (vap->va_type != VREG)) {
1755 			if (have_acl)
1756 				zfs_acl_ids_free(&acl_ids);
1757 			error = SET_ERROR(EINVAL);
1758 			goto out;
1759 		}
1760 
1761 		if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1762 		    cr, vsecp, &acl_ids)) != 0)
1763 			goto out;
1764 		have_acl = B_TRUE;
1765 
1766 		if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1767 			zfs_acl_ids_free(&acl_ids);
1768 			error = SET_ERROR(EDQUOT);
1769 			goto out;
1770 		}
1771 
1772 		tx = dmu_tx_create(os);
1773 
1774 		dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1775 		    ZFS_SA_BASE_ATTR_SIZE);
1776 
1777 		fuid_dirtied = zfsvfs->z_fuid_dirty;
1778 		if (fuid_dirtied)
1779 			zfs_fuid_txhold(zfsvfs, tx);
1780 		dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1781 		dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1782 		if (!zfsvfs->z_use_sa &&
1783 		    acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1784 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1785 			    0, acl_ids.z_aclp->z_acl_bytes);
1786 		}
1787 		error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1788 		if (error) {
1789 			zfs_dirent_unlock(dl);
1790 			if (error == ERESTART) {
1791 				waited = B_TRUE;
1792 				dmu_tx_wait(tx);
1793 				dmu_tx_abort(tx);
1794 				goto top;
1795 			}
1796 			zfs_acl_ids_free(&acl_ids);
1797 			dmu_tx_abort(tx);
1798 			getnewvnode_drop_reserve();
1799 			ZFS_EXIT(zfsvfs);
1800 			return (error);
1801 		}
1802 		zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1803 
1804 		if (fuid_dirtied)
1805 			zfs_fuid_sync(zfsvfs, tx);
1806 
1807 		(void) zfs_link_create(dl, zp, tx, ZNEW);
1808 		txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1809 		if (flag & FIGNORECASE)
1810 			txtype |= TX_CI;
1811 		zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1812 		    vsecp, acl_ids.z_fuidp, vap);
1813 		zfs_acl_ids_free(&acl_ids);
1814 		dmu_tx_commit(tx);
1815 	} else {
1816 		int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1817 
1818 		if (have_acl)
1819 			zfs_acl_ids_free(&acl_ids);
1820 		have_acl = B_FALSE;
1821 
1822 		/*
1823 		 * A directory entry already exists for this name.
1824 		 */
1825 		/*
1826 		 * Can't truncate an existing file if in exclusive mode.
1827 		 */
1828 		if (excl == EXCL) {
1829 			error = SET_ERROR(EEXIST);
1830 			goto out;
1831 		}
1832 		/*
1833 		 * Can't open a directory for writing.
1834 		 */
1835 		if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) {
1836 			error = SET_ERROR(EISDIR);
1837 			goto out;
1838 		}
1839 		/*
1840 		 * Verify requested access to file.
1841 		 */
1842 		if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1843 			goto out;
1844 		}
1845 
1846 		mutex_enter(&dzp->z_lock);
1847 		dzp->z_seq++;
1848 		mutex_exit(&dzp->z_lock);
1849 
1850 		/*
1851 		 * Truncate regular files if requested.
1852 		 */
1853 		if ((ZTOV(zp)->v_type == VREG) &&
1854 		    (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) {
1855 			/* we can't hold any locks when calling zfs_freesp() */
1856 			zfs_dirent_unlock(dl);
1857 			dl = NULL;
1858 			error = zfs_freesp(zp, 0, 0, mode, TRUE);
1859 			if (error == 0) {
1860 				vnevent_create(ZTOV(zp), ct);
1861 			}
1862 		}
1863 	}
1864 out:
1865 	getnewvnode_drop_reserve();
1866 	if (dl)
1867 		zfs_dirent_unlock(dl);
1868 
1869 	if (error) {
1870 		if (zp)
1871 			VN_RELE(ZTOV(zp));
1872 	} else {
1873 		*vpp = ZTOV(zp);
1874 		error = specvp_check(vpp, cr);
1875 	}
1876 
1877 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1878 		zil_commit(zilog, 0);
1879 
1880 	ZFS_EXIT(zfsvfs);
1881 	return (error);
1882 }
1883 
1884 /*
1885  * Remove an entry from a directory.
1886  *
1887  *	IN:	dvp	- vnode of directory to remove entry from.
1888  *		name	- name of entry to remove.
1889  *		cr	- credentials of caller.
1890  *		ct	- caller context
1891  *		flags	- case flags
1892  *
1893  *	RETURN:	0 on success, error code on failure.
1894  *
1895  * Timestamps:
1896  *	dvp - ctime|mtime
1897  *	 vp - ctime (if nlink > 0)
1898  */
1899 
1900 uint64_t null_xattr = 0;
1901 
1902 /*ARGSUSED*/
1903 static int
zfs_remove(vnode_t * dvp,char * name,cred_t * cr,caller_context_t * ct,int flags)1904 zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct,
1905     int flags)
1906 {
1907 	znode_t		*zp, *dzp = VTOZ(dvp);
1908 	znode_t		*xzp;
1909 	vnode_t		*vp;
1910 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1911 	zilog_t		*zilog;
1912 	uint64_t	acl_obj, xattr_obj;
1913 	uint64_t	xattr_obj_unlinked = 0;
1914 	uint64_t	obj = 0;
1915 	zfs_dirlock_t	*dl;
1916 	dmu_tx_t	*tx;
1917 	boolean_t	may_delete_now, delete_now = FALSE;
1918 	boolean_t	unlinked, toobig = FALSE;
1919 	uint64_t	txtype;
1920 	pathname_t	*realnmp = NULL;
1921 	pathname_t	realnm;
1922 	int		error;
1923 	int		zflg = ZEXISTS;
1924 	boolean_t	waited = B_FALSE;
1925 
1926 	ZFS_ENTER(zfsvfs);
1927 	ZFS_VERIFY_ZP(dzp);
1928 	zilog = zfsvfs->z_log;
1929 
1930 	if (flags & FIGNORECASE) {
1931 		zflg |= ZCILOOK;
1932 		pn_alloc(&realnm);
1933 		realnmp = &realnm;
1934 	}
1935 
1936 top:
1937 	xattr_obj = 0;
1938 	xzp = NULL;
1939 	/*
1940 	 * Attempt to lock directory; fail if entry doesn't exist.
1941 	 */
1942 	if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1943 	    NULL, realnmp)) {
1944 		if (realnmp)
1945 			pn_free(realnmp);
1946 		ZFS_EXIT(zfsvfs);
1947 		return (error);
1948 	}
1949 
1950 	vp = ZTOV(zp);
1951 
1952 	if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1953 		goto out;
1954 	}
1955 
1956 	/*
1957 	 * Need to use rmdir for removing directories.
1958 	 */
1959 	if (vp->v_type == VDIR) {
1960 		error = SET_ERROR(EPERM);
1961 		goto out;
1962 	}
1963 
1964 	vnevent_remove(vp, dvp, name, ct);
1965 
1966 	if (realnmp)
1967 		dnlc_remove(dvp, realnmp->pn_buf);
1968 	else
1969 		dnlc_remove(dvp, name);
1970 
1971 	VI_LOCK(vp);
1972 	may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp);
1973 	VI_UNLOCK(vp);
1974 
1975 	/*
1976 	 * We may delete the znode now, or we may put it in the unlinked set;
1977 	 * it depends on whether we're the last link, and on whether there are
1978 	 * other holds on the vnode.  So we dmu_tx_hold() the right things to
1979 	 * allow for either case.
1980 	 */
1981 	obj = zp->z_id;
1982 	tx = dmu_tx_create(zfsvfs->z_os);
1983 	dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1984 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1985 	zfs_sa_upgrade_txholds(tx, zp);
1986 	zfs_sa_upgrade_txholds(tx, dzp);
1987 	if (may_delete_now) {
1988 		toobig =
1989 		    zp->z_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT;
1990 		/* if the file is too big, only hold_free a token amount */
1991 		dmu_tx_hold_free(tx, zp->z_id, 0,
1992 		    (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END));
1993 	}
1994 
1995 	/* are there any extended attributes? */
1996 	error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1997 	    &xattr_obj, sizeof (xattr_obj));
1998 	if (error == 0 && xattr_obj) {
1999 		error = zfs_zget(zfsvfs, xattr_obj, &xzp);
2000 		ASSERT0(error);
2001 		dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
2002 		dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
2003 	}
2004 
2005 	mutex_enter(&zp->z_lock);
2006 	if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now)
2007 		dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
2008 	mutex_exit(&zp->z_lock);
2009 
2010 	/* charge as an update -- would be nice not to charge at all */
2011 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
2012 
2013 	/*
2014 	 * Mark this transaction as typically resulting in a net free of
2015 	 * space, unless object removal will be delayed indefinitely
2016 	 * (due to active holds on the vnode due to the file being open).
2017 	 */
2018 	if (may_delete_now)
2019 		dmu_tx_mark_netfree(tx);
2020 
2021 	error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2022 	if (error) {
2023 		zfs_dirent_unlock(dl);
2024 		VN_RELE(vp);
2025 		if (xzp)
2026 			VN_RELE(ZTOV(xzp));
2027 		if (error == ERESTART) {
2028 			waited = B_TRUE;
2029 			dmu_tx_wait(tx);
2030 			dmu_tx_abort(tx);
2031 			goto top;
2032 		}
2033 		if (realnmp)
2034 			pn_free(realnmp);
2035 		dmu_tx_abort(tx);
2036 		ZFS_EXIT(zfsvfs);
2037 		return (error);
2038 	}
2039 
2040 	/*
2041 	 * Remove the directory entry.
2042 	 */
2043 	error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
2044 
2045 	if (error) {
2046 		dmu_tx_commit(tx);
2047 		goto out;
2048 	}
2049 
2050 	if (unlinked) {
2051 		/*
2052 		 * Hold z_lock so that we can make sure that the ACL obj
2053 		 * hasn't changed.  Could have been deleted due to
2054 		 * zfs_sa_upgrade().
2055 		 */
2056 		mutex_enter(&zp->z_lock);
2057 		VI_LOCK(vp);
2058 		(void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
2059 		    &xattr_obj_unlinked, sizeof (xattr_obj_unlinked));
2060 		delete_now = may_delete_now && !toobig &&
2061 		    vp->v_count == 1 && !vn_has_cached_data(vp) &&
2062 		    xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) ==
2063 		    acl_obj;
2064 		VI_UNLOCK(vp);
2065 	}
2066 
2067 	if (delete_now) {
2068 #ifdef __FreeBSD__
2069 		panic("zfs_remove: delete_now branch taken");
2070 #endif
2071 		if (xattr_obj_unlinked) {
2072 			ASSERT3U(xzp->z_links, ==, 2);
2073 			mutex_enter(&xzp->z_lock);
2074 			xzp->z_unlinked = 1;
2075 			xzp->z_links = 0;
2076 			error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
2077 			    &xzp->z_links, sizeof (xzp->z_links), tx);
2078 			ASSERT3U(error,  ==,  0);
2079 			mutex_exit(&xzp->z_lock);
2080 			zfs_unlinked_add(xzp, tx);
2081 
2082 			if (zp->z_is_sa)
2083 				error = sa_remove(zp->z_sa_hdl,
2084 				    SA_ZPL_XATTR(zfsvfs), tx);
2085 			else
2086 				error = sa_update(zp->z_sa_hdl,
2087 				    SA_ZPL_XATTR(zfsvfs), &null_xattr,
2088 				    sizeof (uint64_t), tx);
2089 			ASSERT0(error);
2090 		}
2091 		VI_LOCK(vp);
2092 		vp->v_count--;
2093 		ASSERT0(vp->v_count);
2094 		VI_UNLOCK(vp);
2095 		mutex_exit(&zp->z_lock);
2096 		zfs_znode_delete(zp, tx);
2097 	} else if (unlinked) {
2098 		mutex_exit(&zp->z_lock);
2099 		zfs_unlinked_add(zp, tx);
2100 #ifdef __FreeBSD__
2101 		vp->v_vflag |= VV_NOSYNC;
2102 #endif
2103 	}
2104 
2105 	txtype = TX_REMOVE;
2106 	if (flags & FIGNORECASE)
2107 		txtype |= TX_CI;
2108 	zfs_log_remove(zilog, tx, txtype, dzp, name, obj);
2109 
2110 	dmu_tx_commit(tx);
2111 out:
2112 	if (realnmp)
2113 		pn_free(realnmp);
2114 
2115 	zfs_dirent_unlock(dl);
2116 
2117 	if (!delete_now)
2118 		VN_RELE(vp);
2119 	if (xzp)
2120 		VN_RELE(ZTOV(xzp));
2121 
2122 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2123 		zil_commit(zilog, 0);
2124 
2125 	ZFS_EXIT(zfsvfs);
2126 	return (error);
2127 }
2128 
2129 /*
2130  * Create a new directory and insert it into dvp using the name
2131  * provided.  Return a pointer to the inserted directory.
2132  *
2133  *	IN:	dvp	- vnode of directory to add subdir to.
2134  *		dirname	- name of new directory.
2135  *		vap	- attributes of new directory.
2136  *		cr	- credentials of caller.
2137  *		ct	- caller context
2138  *		flags	- case flags
2139  *		vsecp	- ACL to be set
2140  *
2141  *	OUT:	vpp	- vnode of created directory.
2142  *
2143  *	RETURN:	0 on success, error code on failure.
2144  *
2145  * Timestamps:
2146  *	dvp - ctime|mtime updated
2147  *	 vp - ctime|mtime|atime updated
2148  */
2149 /*ARGSUSED*/
2150 static int
zfs_mkdir(vnode_t * dvp,char * dirname,vattr_t * vap,vnode_t ** vpp,cred_t * cr,caller_context_t * ct,int flags,vsecattr_t * vsecp)2151 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr,
2152     caller_context_t *ct, int flags, vsecattr_t *vsecp)
2153 {
2154 	znode_t		*zp, *dzp = VTOZ(dvp);
2155 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
2156 	zilog_t		*zilog;
2157 	zfs_dirlock_t	*dl;
2158 	uint64_t	txtype;
2159 	dmu_tx_t	*tx;
2160 	int		error;
2161 	int		zf = ZNEW;
2162 	ksid_t		*ksid;
2163 	uid_t		uid;
2164 	gid_t		gid = crgetgid(cr);
2165 	zfs_acl_ids_t   acl_ids;
2166 	boolean_t	fuid_dirtied;
2167 	boolean_t	waited = B_FALSE;
2168 
2169 	ASSERT(vap->va_type == VDIR);
2170 
2171 	/*
2172 	 * If we have an ephemeral id, ACL, or XVATTR then
2173 	 * make sure file system is at proper version
2174 	 */
2175 
2176 	ksid = crgetsid(cr, KSID_OWNER);
2177 	if (ksid)
2178 		uid = ksid_getid(ksid);
2179 	else
2180 		uid = crgetuid(cr);
2181 	if (zfsvfs->z_use_fuids == B_FALSE &&
2182 	    (vsecp || (vap->va_mask & AT_XVATTR) ||
2183 	    IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
2184 		return (SET_ERROR(EINVAL));
2185 
2186 	ZFS_ENTER(zfsvfs);
2187 	ZFS_VERIFY_ZP(dzp);
2188 	zilog = zfsvfs->z_log;
2189 
2190 	if (dzp->z_pflags & ZFS_XATTR) {
2191 		ZFS_EXIT(zfsvfs);
2192 		return (SET_ERROR(EINVAL));
2193 	}
2194 
2195 	if (zfsvfs->z_utf8 && u8_validate(dirname,
2196 	    strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
2197 		ZFS_EXIT(zfsvfs);
2198 		return (SET_ERROR(EILSEQ));
2199 	}
2200 	if (flags & FIGNORECASE)
2201 		zf |= ZCILOOK;
2202 
2203 	if (vap->va_mask & AT_XVATTR) {
2204 		if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap,
2205 		    crgetuid(cr), cr, vap->va_type)) != 0) {
2206 			ZFS_EXIT(zfsvfs);
2207 			return (error);
2208 		}
2209 	}
2210 
2211 	if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
2212 	    vsecp, &acl_ids)) != 0) {
2213 		ZFS_EXIT(zfsvfs);
2214 		return (error);
2215 	}
2216 
2217 	getnewvnode_reserve(1);
2218 
2219 	/*
2220 	 * First make sure the new directory doesn't exist.
2221 	 *
2222 	 * Existence is checked first to make sure we don't return
2223 	 * EACCES instead of EEXIST which can cause some applications
2224 	 * to fail.
2225 	 */
2226 top:
2227 	*vpp = NULL;
2228 
2229 	if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
2230 	    NULL, NULL)) {
2231 		zfs_acl_ids_free(&acl_ids);
2232 		getnewvnode_drop_reserve();
2233 		ZFS_EXIT(zfsvfs);
2234 		return (error);
2235 	}
2236 
2237 	if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) {
2238 		zfs_acl_ids_free(&acl_ids);
2239 		zfs_dirent_unlock(dl);
2240 		getnewvnode_drop_reserve();
2241 		ZFS_EXIT(zfsvfs);
2242 		return (error);
2243 	}
2244 
2245 	if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
2246 		zfs_acl_ids_free(&acl_ids);
2247 		zfs_dirent_unlock(dl);
2248 		getnewvnode_drop_reserve();
2249 		ZFS_EXIT(zfsvfs);
2250 		return (SET_ERROR(EDQUOT));
2251 	}
2252 
2253 	/*
2254 	 * Add a new entry to the directory.
2255 	 */
2256 	tx = dmu_tx_create(zfsvfs->z_os);
2257 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
2258 	dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2259 	fuid_dirtied = zfsvfs->z_fuid_dirty;
2260 	if (fuid_dirtied)
2261 		zfs_fuid_txhold(zfsvfs, tx);
2262 	if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2263 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
2264 		    acl_ids.z_aclp->z_acl_bytes);
2265 	}
2266 
2267 	dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
2268 	    ZFS_SA_BASE_ATTR_SIZE);
2269 
2270 	error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2271 	if (error) {
2272 		zfs_dirent_unlock(dl);
2273 		if (error == ERESTART) {
2274 			waited = B_TRUE;
2275 			dmu_tx_wait(tx);
2276 			dmu_tx_abort(tx);
2277 			goto top;
2278 		}
2279 		zfs_acl_ids_free(&acl_ids);
2280 		dmu_tx_abort(tx);
2281 		getnewvnode_drop_reserve();
2282 		ZFS_EXIT(zfsvfs);
2283 		return (error);
2284 	}
2285 
2286 	/*
2287 	 * Create new node.
2288 	 */
2289 	zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
2290 
2291 	if (fuid_dirtied)
2292 		zfs_fuid_sync(zfsvfs, tx);
2293 
2294 	/*
2295 	 * Now put new name in parent dir.
2296 	 */
2297 	(void) zfs_link_create(dl, zp, tx, ZNEW);
2298 
2299 	*vpp = ZTOV(zp);
2300 
2301 	txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
2302 	if (flags & FIGNORECASE)
2303 		txtype |= TX_CI;
2304 	zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
2305 	    acl_ids.z_fuidp, vap);
2306 
2307 	zfs_acl_ids_free(&acl_ids);
2308 
2309 	dmu_tx_commit(tx);
2310 
2311 	getnewvnode_drop_reserve();
2312 
2313 	zfs_dirent_unlock(dl);
2314 
2315 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2316 		zil_commit(zilog, 0);
2317 
2318 	ZFS_EXIT(zfsvfs);
2319 	return (0);
2320 }
2321 
2322 /*
2323  * Remove a directory subdir entry.  If the current working
2324  * directory is the same as the subdir to be removed, the
2325  * remove will fail.
2326  *
2327  *	IN:	dvp	- vnode of directory to remove from.
2328  *		name	- name of directory to be removed.
2329  *		cwd	- vnode of current working directory.
2330  *		cr	- credentials of caller.
2331  *		ct	- caller context
2332  *		flags	- case flags
2333  *
2334  *	RETURN:	0 on success, error code on failure.
2335  *
2336  * Timestamps:
2337  *	dvp - ctime|mtime updated
2338  */
2339 /*ARGSUSED*/
2340 static int
zfs_rmdir(vnode_t * dvp,char * name,vnode_t * cwd,cred_t * cr,caller_context_t * ct,int flags)2341 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
2342     caller_context_t *ct, int flags)
2343 {
2344 	znode_t		*dzp = VTOZ(dvp);
2345 	znode_t		*zp;
2346 	vnode_t		*vp;
2347 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
2348 	zilog_t		*zilog;
2349 	zfs_dirlock_t	*dl;
2350 	dmu_tx_t	*tx;
2351 	int		error;
2352 	int		zflg = ZEXISTS;
2353 	boolean_t	waited = B_FALSE;
2354 
2355 	ZFS_ENTER(zfsvfs);
2356 	ZFS_VERIFY_ZP(dzp);
2357 	zilog = zfsvfs->z_log;
2358 
2359 	if (flags & FIGNORECASE)
2360 		zflg |= ZCILOOK;
2361 top:
2362 	zp = NULL;
2363 
2364 	/*
2365 	 * Attempt to lock directory; fail if entry doesn't exist.
2366 	 */
2367 	if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
2368 	    NULL, NULL)) {
2369 		ZFS_EXIT(zfsvfs);
2370 		return (error);
2371 	}
2372 
2373 	vp = ZTOV(zp);
2374 
2375 	if (error = zfs_zaccess_delete(dzp, zp, cr)) {
2376 		goto out;
2377 	}
2378 
2379 	if (vp->v_type != VDIR) {
2380 		error = SET_ERROR(ENOTDIR);
2381 		goto out;
2382 	}
2383 
2384 	if (vp == cwd) {
2385 		error = SET_ERROR(EINVAL);
2386 		goto out;
2387 	}
2388 
2389 	vnevent_rmdir(vp, dvp, name, ct);
2390 
2391 	/*
2392 	 * Grab a lock on the directory to make sure that noone is
2393 	 * trying to add (or lookup) entries while we are removing it.
2394 	 */
2395 	rw_enter(&zp->z_name_lock, RW_WRITER);
2396 
2397 	/*
2398 	 * Grab a lock on the parent pointer to make sure we play well
2399 	 * with the treewalk and directory rename code.
2400 	 */
2401 	rw_enter(&zp->z_parent_lock, RW_WRITER);
2402 
2403 	tx = dmu_tx_create(zfsvfs->z_os);
2404 	dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
2405 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2406 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
2407 	zfs_sa_upgrade_txholds(tx, zp);
2408 	zfs_sa_upgrade_txholds(tx, dzp);
2409 	error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2410 	if (error) {
2411 		rw_exit(&zp->z_parent_lock);
2412 		rw_exit(&zp->z_name_lock);
2413 		zfs_dirent_unlock(dl);
2414 		VN_RELE(vp);
2415 		if (error == ERESTART) {
2416 			waited = B_TRUE;
2417 			dmu_tx_wait(tx);
2418 			dmu_tx_abort(tx);
2419 			goto top;
2420 		}
2421 		dmu_tx_abort(tx);
2422 		ZFS_EXIT(zfsvfs);
2423 		return (error);
2424 	}
2425 
2426 #ifdef FREEBSD_NAMECACHE
2427 	cache_purge(dvp);
2428 #endif
2429 
2430 	error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
2431 
2432 	if (error == 0) {
2433 		uint64_t txtype = TX_RMDIR;
2434 		if (flags & FIGNORECASE)
2435 			txtype |= TX_CI;
2436 		zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT);
2437 	}
2438 
2439 	dmu_tx_commit(tx);
2440 
2441 	rw_exit(&zp->z_parent_lock);
2442 	rw_exit(&zp->z_name_lock);
2443 #ifdef FREEBSD_NAMECACHE
2444 	cache_purge(vp);
2445 #endif
2446 out:
2447 	zfs_dirent_unlock(dl);
2448 
2449 	VN_RELE(vp);
2450 
2451 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2452 		zil_commit(zilog, 0);
2453 
2454 	ZFS_EXIT(zfsvfs);
2455 	return (error);
2456 }
2457 
2458 /*
2459  * Read as many directory entries as will fit into the provided
2460  * buffer from the given directory cursor position (specified in
2461  * the uio structure).
2462  *
2463  *	IN:	vp	- vnode of directory to read.
2464  *		uio	- structure supplying read location, range info,
2465  *			  and return buffer.
2466  *		cr	- credentials of caller.
2467  *		ct	- caller context
2468  *		flags	- case flags
2469  *
2470  *	OUT:	uio	- updated offset and range, buffer filled.
2471  *		eofp	- set to true if end-of-file detected.
2472  *
2473  *	RETURN:	0 on success, error code on failure.
2474  *
2475  * Timestamps:
2476  *	vp - atime updated
2477  *
2478  * Note that the low 4 bits of the cookie returned by zap is always zero.
2479  * This allows us to use the low range for "special" directory entries:
2480  * We use 0 for '.', and 1 for '..'.  If this is the root of the filesystem,
2481  * we use the offset 2 for the '.zfs' directory.
2482  */
2483 /* ARGSUSED */
2484 static int
zfs_readdir(vnode_t * vp,uio_t * uio,cred_t * cr,int * eofp,int * ncookies,u_long ** cookies)2485 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp, int *ncookies, u_long **cookies)
2486 {
2487 	znode_t		*zp = VTOZ(vp);
2488 	iovec_t		*iovp;
2489 	edirent_t	*eodp;
2490 	dirent64_t	*odp;
2491 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2492 	objset_t	*os;
2493 	caddr_t		outbuf;
2494 	size_t		bufsize;
2495 	zap_cursor_t	zc;
2496 	zap_attribute_t	zap;
2497 	uint_t		bytes_wanted;
2498 	uint64_t	offset; /* must be unsigned; checks for < 1 */
2499 	uint64_t	parent;
2500 	int		local_eof;
2501 	int		outcount;
2502 	int		error;
2503 	uint8_t		prefetch;
2504 	boolean_t	check_sysattrs;
2505 	uint8_t		type;
2506 	int		ncooks;
2507 	u_long		*cooks = NULL;
2508 	int		flags = 0;
2509 
2510 	ZFS_ENTER(zfsvfs);
2511 	ZFS_VERIFY_ZP(zp);
2512 
2513 	if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
2514 	    &parent, sizeof (parent))) != 0) {
2515 		ZFS_EXIT(zfsvfs);
2516 		return (error);
2517 	}
2518 
2519 	/*
2520 	 * If we are not given an eof variable,
2521 	 * use a local one.
2522 	 */
2523 	if (eofp == NULL)
2524 		eofp = &local_eof;
2525 
2526 	/*
2527 	 * Check for valid iov_len.
2528 	 */
2529 	if (uio->uio_iov->iov_len <= 0) {
2530 		ZFS_EXIT(zfsvfs);
2531 		return (SET_ERROR(EINVAL));
2532 	}
2533 
2534 	/*
2535 	 * Quit if directory has been removed (posix)
2536 	 */
2537 	if ((*eofp = zp->z_unlinked) != 0) {
2538 		ZFS_EXIT(zfsvfs);
2539 		return (0);
2540 	}
2541 
2542 	error = 0;
2543 	os = zfsvfs->z_os;
2544 	offset = uio->uio_loffset;
2545 	prefetch = zp->z_zn_prefetch;
2546 
2547 	/*
2548 	 * Initialize the iterator cursor.
2549 	 */
2550 	if (offset <= 3) {
2551 		/*
2552 		 * Start iteration from the beginning of the directory.
2553 		 */
2554 		zap_cursor_init(&zc, os, zp->z_id);
2555 	} else {
2556 		/*
2557 		 * The offset is a serialized cursor.
2558 		 */
2559 		zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
2560 	}
2561 
2562 	/*
2563 	 * Get space to change directory entries into fs independent format.
2564 	 */
2565 	iovp = uio->uio_iov;
2566 	bytes_wanted = iovp->iov_len;
2567 	if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) {
2568 		bufsize = bytes_wanted;
2569 		outbuf = kmem_alloc(bufsize, KM_SLEEP);
2570 		odp = (struct dirent64 *)outbuf;
2571 	} else {
2572 		bufsize = bytes_wanted;
2573 		outbuf = NULL;
2574 		odp = (struct dirent64 *)iovp->iov_base;
2575 	}
2576 	eodp = (struct edirent *)odp;
2577 
2578 	if (ncookies != NULL) {
2579 		/*
2580 		 * Minimum entry size is dirent size and 1 byte for a file name.
2581 		 */
2582 		ncooks = uio->uio_resid / (sizeof(struct dirent) - sizeof(((struct dirent *)NULL)->d_name) + 1);
2583 		cooks = malloc(ncooks * sizeof(u_long), M_TEMP, M_WAITOK);
2584 		*cookies = cooks;
2585 		*ncookies = ncooks;
2586 	}
2587 	/*
2588 	 * If this VFS supports the system attribute view interface; and
2589 	 * we're looking at an extended attribute directory; and we care
2590 	 * about normalization conflicts on this vfs; then we must check
2591 	 * for normalization conflicts with the sysattr name space.
2592 	 */
2593 #ifdef TODO
2594 	check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
2595 	    (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm &&
2596 	    (flags & V_RDDIR_ENTFLAGS);
2597 #else
2598 	check_sysattrs = 0;
2599 #endif
2600 
2601 	/*
2602 	 * Transform to file-system independent format
2603 	 */
2604 	outcount = 0;
2605 	while (outcount < bytes_wanted) {
2606 		ino64_t objnum;
2607 		ushort_t reclen;
2608 		off64_t *next = NULL;
2609 
2610 		/*
2611 		 * Special case `.', `..', and `.zfs'.
2612 		 */
2613 		if (offset == 0) {
2614 			(void) strcpy(zap.za_name, ".");
2615 			zap.za_normalization_conflict = 0;
2616 			objnum = zp->z_id;
2617 			type = DT_DIR;
2618 		} else if (offset == 1) {
2619 			(void) strcpy(zap.za_name, "..");
2620 			zap.za_normalization_conflict = 0;
2621 			objnum = parent;
2622 			type = DT_DIR;
2623 		} else if (offset == 2 && zfs_show_ctldir(zp)) {
2624 			(void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2625 			zap.za_normalization_conflict = 0;
2626 			objnum = ZFSCTL_INO_ROOT;
2627 			type = DT_DIR;
2628 		} else {
2629 			/*
2630 			 * Grab next entry.
2631 			 */
2632 			if (error = zap_cursor_retrieve(&zc, &zap)) {
2633 				if ((*eofp = (error == ENOENT)) != 0)
2634 					break;
2635 				else
2636 					goto update;
2637 			}
2638 
2639 			if (zap.za_integer_length != 8 ||
2640 			    zap.za_num_integers != 1) {
2641 				cmn_err(CE_WARN, "zap_readdir: bad directory "
2642 				    "entry, obj = %lld, offset = %lld\n",
2643 				    (u_longlong_t)zp->z_id,
2644 				    (u_longlong_t)offset);
2645 				error = SET_ERROR(ENXIO);
2646 				goto update;
2647 			}
2648 
2649 			objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2650 			/*
2651 			 * MacOS X can extract the object type here such as:
2652 			 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2653 			 */
2654 			type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2655 
2656 			if (check_sysattrs && !zap.za_normalization_conflict) {
2657 #ifdef TODO
2658 				zap.za_normalization_conflict =
2659 				    xattr_sysattr_casechk(zap.za_name);
2660 #else
2661 				panic("%s:%u: TODO", __func__, __LINE__);
2662 #endif
2663 			}
2664 		}
2665 
2666 		if (flags & V_RDDIR_ACCFILTER) {
2667 			/*
2668 			 * If we have no access at all, don't include
2669 			 * this entry in the returned information
2670 			 */
2671 			znode_t	*ezp;
2672 			if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0)
2673 				goto skip_entry;
2674 			if (!zfs_has_access(ezp, cr)) {
2675 				VN_RELE(ZTOV(ezp));
2676 				goto skip_entry;
2677 			}
2678 			VN_RELE(ZTOV(ezp));
2679 		}
2680 
2681 		if (flags & V_RDDIR_ENTFLAGS)
2682 			reclen = EDIRENT_RECLEN(strlen(zap.za_name));
2683 		else
2684 			reclen = DIRENT64_RECLEN(strlen(zap.za_name));
2685 
2686 		/*
2687 		 * Will this entry fit in the buffer?
2688 		 */
2689 		if (outcount + reclen > bufsize) {
2690 			/*
2691 			 * Did we manage to fit anything in the buffer?
2692 			 */
2693 			if (!outcount) {
2694 				error = SET_ERROR(EINVAL);
2695 				goto update;
2696 			}
2697 			break;
2698 		}
2699 		if (flags & V_RDDIR_ENTFLAGS) {
2700 			/*
2701 			 * Add extended flag entry:
2702 			 */
2703 			eodp->ed_ino = objnum;
2704 			eodp->ed_reclen = reclen;
2705 			/* NOTE: ed_off is the offset for the *next* entry */
2706 			next = &(eodp->ed_off);
2707 			eodp->ed_eflags = zap.za_normalization_conflict ?
2708 			    ED_CASE_CONFLICT : 0;
2709 			(void) strncpy(eodp->ed_name, zap.za_name,
2710 			    EDIRENT_NAMELEN(reclen));
2711 			eodp = (edirent_t *)((intptr_t)eodp + reclen);
2712 		} else {
2713 			/*
2714 			 * Add normal entry:
2715 			 */
2716 			odp->d_ino = objnum;
2717 			odp->d_reclen = reclen;
2718 			odp->d_namlen = strlen(zap.za_name);
2719 			(void) strlcpy(odp->d_name, zap.za_name, odp->d_namlen + 1);
2720 			odp->d_type = type;
2721 			odp = (dirent64_t *)((intptr_t)odp + reclen);
2722 		}
2723 		outcount += reclen;
2724 
2725 		ASSERT(outcount <= bufsize);
2726 
2727 		/* Prefetch znode */
2728 		if (prefetch)
2729 			dmu_prefetch(os, objnum, 0, 0, 0,
2730 			    ZIO_PRIORITY_SYNC_READ);
2731 
2732 	skip_entry:
2733 		/*
2734 		 * Move to the next entry, fill in the previous offset.
2735 		 */
2736 		if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2737 			zap_cursor_advance(&zc);
2738 			offset = zap_cursor_serialize(&zc);
2739 		} else {
2740 			offset += 1;
2741 		}
2742 
2743 		if (cooks != NULL) {
2744 			*cooks++ = offset;
2745 			ncooks--;
2746 			KASSERT(ncooks >= 0, ("ncookies=%d", ncooks));
2747 		}
2748 	}
2749 	zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2750 
2751 	/* Subtract unused cookies */
2752 	if (ncookies != NULL)
2753 		*ncookies -= ncooks;
2754 
2755 	if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) {
2756 		iovp->iov_base += outcount;
2757 		iovp->iov_len -= outcount;
2758 		uio->uio_resid -= outcount;
2759 	} else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) {
2760 		/*
2761 		 * Reset the pointer.
2762 		 */
2763 		offset = uio->uio_loffset;
2764 	}
2765 
2766 update:
2767 	zap_cursor_fini(&zc);
2768 	if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2769 		kmem_free(outbuf, bufsize);
2770 
2771 	if (error == ENOENT)
2772 		error = 0;
2773 
2774 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
2775 
2776 	uio->uio_loffset = offset;
2777 	ZFS_EXIT(zfsvfs);
2778 	if (error != 0 && cookies != NULL) {
2779 		free(*cookies, M_TEMP);
2780 		*cookies = NULL;
2781 		*ncookies = 0;
2782 	}
2783 	return (error);
2784 }
2785 
2786 ulong_t zfs_fsync_sync_cnt = 4;
2787 
2788 static int
zfs_fsync(vnode_t * vp,int syncflag,cred_t * cr,caller_context_t * ct)2789 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
2790 {
2791 	znode_t	*zp = VTOZ(vp);
2792 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2793 
2794 	(void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2795 
2796 	if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) {
2797 		ZFS_ENTER(zfsvfs);
2798 		ZFS_VERIFY_ZP(zp);
2799 		zil_commit(zfsvfs->z_log, zp->z_id);
2800 		ZFS_EXIT(zfsvfs);
2801 	}
2802 	return (0);
2803 }
2804 
2805 
2806 /*
2807  * Get the requested file attributes and place them in the provided
2808  * vattr structure.
2809  *
2810  *	IN:	vp	- vnode of file.
2811  *		vap	- va_mask identifies requested attributes.
2812  *			  If AT_XVATTR set, then optional attrs are requested
2813  *		flags	- ATTR_NOACLCHECK (CIFS server context)
2814  *		cr	- credentials of caller.
2815  *		ct	- caller context
2816  *
2817  *	OUT:	vap	- attribute values.
2818  *
2819  *	RETURN:	0 (always succeeds).
2820  */
2821 /* ARGSUSED */
2822 static int
zfs_getattr(vnode_t * vp,vattr_t * vap,int flags,cred_t * cr,caller_context_t * ct)2823 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2824     caller_context_t *ct)
2825 {
2826 	znode_t *zp = VTOZ(vp);
2827 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2828 	int	error = 0;
2829 	uint32_t blksize;
2830 	u_longlong_t nblocks;
2831 	uint64_t links;
2832 	uint64_t mtime[2], ctime[2], crtime[2], rdev;
2833 	xvattr_t *xvap = (xvattr_t *)vap;	/* vap may be an xvattr_t * */
2834 	xoptattr_t *xoap = NULL;
2835 	boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2836 	sa_bulk_attr_t bulk[4];
2837 	int count = 0;
2838 
2839 	ZFS_ENTER(zfsvfs);
2840 	ZFS_VERIFY_ZP(zp);
2841 
2842 	zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2843 
2844 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
2845 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
2846 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL, &crtime, 16);
2847 	if (vp->v_type == VBLK || vp->v_type == VCHR)
2848 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL,
2849 		    &rdev, 8);
2850 
2851 	if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2852 		ZFS_EXIT(zfsvfs);
2853 		return (error);
2854 	}
2855 
2856 	/*
2857 	 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2858 	 * Also, if we are the owner don't bother, since owner should
2859 	 * always be allowed to read basic attributes of file.
2860 	 */
2861 	if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2862 	    (vap->va_uid != crgetuid(cr))) {
2863 		if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2864 		    skipaclchk, cr)) {
2865 			ZFS_EXIT(zfsvfs);
2866 			return (error);
2867 		}
2868 	}
2869 
2870 	/*
2871 	 * Return all attributes.  It's cheaper to provide the answer
2872 	 * than to determine whether we were asked the question.
2873 	 */
2874 
2875 	mutex_enter(&zp->z_lock);
2876 	vap->va_type = IFTOVT(zp->z_mode);
2877 	vap->va_mode = zp->z_mode & ~S_IFMT;
2878 #ifdef illumos
2879 	vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev;
2880 #else
2881 	vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
2882 #endif
2883 	vap->va_nodeid = zp->z_id;
2884 	if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp))
2885 		links = zp->z_links + 1;
2886 	else
2887 		links = zp->z_links;
2888 	vap->va_nlink = MIN(links, LINK_MAX);	/* nlink_t limit! */
2889 	vap->va_size = zp->z_size;
2890 #ifdef illumos
2891 	vap->va_rdev = vp->v_rdev;
2892 #else
2893 	if (vp->v_type == VBLK || vp->v_type == VCHR)
2894 		vap->va_rdev = zfs_cmpldev(rdev);
2895 #endif
2896 	vap->va_seq = zp->z_seq;
2897 	vap->va_flags = 0;	/* FreeBSD: Reset chflags(2) flags. */
2898      	vap->va_filerev = zp->z_seq;
2899 
2900 	/*
2901 	 * Add in any requested optional attributes and the create time.
2902 	 * Also set the corresponding bits in the returned attribute bitmap.
2903 	 */
2904 	if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2905 		if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2906 			xoap->xoa_archive =
2907 			    ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2908 			XVA_SET_RTN(xvap, XAT_ARCHIVE);
2909 		}
2910 
2911 		if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2912 			xoap->xoa_readonly =
2913 			    ((zp->z_pflags & ZFS_READONLY) != 0);
2914 			XVA_SET_RTN(xvap, XAT_READONLY);
2915 		}
2916 
2917 		if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2918 			xoap->xoa_system =
2919 			    ((zp->z_pflags & ZFS_SYSTEM) != 0);
2920 			XVA_SET_RTN(xvap, XAT_SYSTEM);
2921 		}
2922 
2923 		if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2924 			xoap->xoa_hidden =
2925 			    ((zp->z_pflags & ZFS_HIDDEN) != 0);
2926 			XVA_SET_RTN(xvap, XAT_HIDDEN);
2927 		}
2928 
2929 		if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2930 			xoap->xoa_nounlink =
2931 			    ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2932 			XVA_SET_RTN(xvap, XAT_NOUNLINK);
2933 		}
2934 
2935 		if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2936 			xoap->xoa_immutable =
2937 			    ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2938 			XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2939 		}
2940 
2941 		if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2942 			xoap->xoa_appendonly =
2943 			    ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2944 			XVA_SET_RTN(xvap, XAT_APPENDONLY);
2945 		}
2946 
2947 		if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2948 			xoap->xoa_nodump =
2949 			    ((zp->z_pflags & ZFS_NODUMP) != 0);
2950 			XVA_SET_RTN(xvap, XAT_NODUMP);
2951 		}
2952 
2953 		if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2954 			xoap->xoa_opaque =
2955 			    ((zp->z_pflags & ZFS_OPAQUE) != 0);
2956 			XVA_SET_RTN(xvap, XAT_OPAQUE);
2957 		}
2958 
2959 		if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2960 			xoap->xoa_av_quarantined =
2961 			    ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2962 			XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2963 		}
2964 
2965 		if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2966 			xoap->xoa_av_modified =
2967 			    ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2968 			XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2969 		}
2970 
2971 		if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2972 		    vp->v_type == VREG) {
2973 			zfs_sa_get_scanstamp(zp, xvap);
2974 		}
2975 
2976 		if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2977 			uint64_t times[2];
2978 
2979 			(void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs),
2980 			    times, sizeof (times));
2981 			ZFS_TIME_DECODE(&xoap->xoa_createtime, times);
2982 			XVA_SET_RTN(xvap, XAT_CREATETIME);
2983 		}
2984 
2985 		if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2986 			xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2987 			XVA_SET_RTN(xvap, XAT_REPARSE);
2988 		}
2989 		if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2990 			xoap->xoa_generation = zp->z_gen;
2991 			XVA_SET_RTN(xvap, XAT_GEN);
2992 		}
2993 
2994 		if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2995 			xoap->xoa_offline =
2996 			    ((zp->z_pflags & ZFS_OFFLINE) != 0);
2997 			XVA_SET_RTN(xvap, XAT_OFFLINE);
2998 		}
2999 
3000 		if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
3001 			xoap->xoa_sparse =
3002 			    ((zp->z_pflags & ZFS_SPARSE) != 0);
3003 			XVA_SET_RTN(xvap, XAT_SPARSE);
3004 		}
3005 	}
3006 
3007 	ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime);
3008 	ZFS_TIME_DECODE(&vap->va_mtime, mtime);
3009 	ZFS_TIME_DECODE(&vap->va_ctime, ctime);
3010 	ZFS_TIME_DECODE(&vap->va_birthtime, crtime);
3011 
3012 	mutex_exit(&zp->z_lock);
3013 
3014 	sa_object_size(zp->z_sa_hdl, &blksize, &nblocks);
3015 	vap->va_blksize = blksize;
3016 	vap->va_bytes = nblocks << 9;	/* nblocks * 512 */
3017 
3018 	if (zp->z_blksz == 0) {
3019 		/*
3020 		 * Block size hasn't been set; suggest maximal I/O transfers.
3021 		 */
3022 		vap->va_blksize = zfsvfs->z_max_blksz;
3023 	}
3024 
3025 	ZFS_EXIT(zfsvfs);
3026 	return (0);
3027 }
3028 
3029 /*
3030  * Set the file attributes to the values contained in the
3031  * vattr structure.
3032  *
3033  *	IN:	vp	- vnode of file to be modified.
3034  *		vap	- new attribute values.
3035  *			  If AT_XVATTR set, then optional attrs are being set
3036  *		flags	- ATTR_UTIME set if non-default time values provided.
3037  *			- ATTR_NOACLCHECK (CIFS context only).
3038  *		cr	- credentials of caller.
3039  *		ct	- caller context
3040  *
3041  *	RETURN:	0 on success, error code on failure.
3042  *
3043  * Timestamps:
3044  *	vp - ctime updated, mtime updated if size changed.
3045  */
3046 /* ARGSUSED */
3047 static int
zfs_setattr(vnode_t * vp,vattr_t * vap,int flags,cred_t * cr,caller_context_t * ct)3048 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
3049     caller_context_t *ct)
3050 {
3051 	znode_t		*zp = VTOZ(vp);
3052 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
3053 	zilog_t		*zilog;
3054 	dmu_tx_t	*tx;
3055 	vattr_t		oldva;
3056 	xvattr_t	tmpxvattr;
3057 	uint_t		mask = vap->va_mask;
3058 	uint_t		saved_mask = 0;
3059 	uint64_t	saved_mode;
3060 	int		trim_mask = 0;
3061 	uint64_t	new_mode;
3062 	uint64_t	new_uid, new_gid;
3063 	uint64_t	xattr_obj;
3064 	uint64_t	mtime[2], ctime[2];
3065 	znode_t		*attrzp;
3066 	int		need_policy = FALSE;
3067 	int		err, err2;
3068 	zfs_fuid_info_t *fuidp = NULL;
3069 	xvattr_t *xvap = (xvattr_t *)vap;	/* vap may be an xvattr_t * */
3070 	xoptattr_t	*xoap;
3071 	zfs_acl_t	*aclp;
3072 	boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
3073 	boolean_t	fuid_dirtied = B_FALSE;
3074 	sa_bulk_attr_t	bulk[7], xattr_bulk[7];
3075 	int		count = 0, xattr_count = 0;
3076 
3077 	if (mask == 0)
3078 		return (0);
3079 
3080 	if (mask & AT_NOSET)
3081 		return (SET_ERROR(EINVAL));
3082 
3083 	ZFS_ENTER(zfsvfs);
3084 	ZFS_VERIFY_ZP(zp);
3085 
3086 	zilog = zfsvfs->z_log;
3087 
3088 	/*
3089 	 * Make sure that if we have ephemeral uid/gid or xvattr specified
3090 	 * that file system is at proper version level
3091 	 */
3092 
3093 	if (zfsvfs->z_use_fuids == B_FALSE &&
3094 	    (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
3095 	    ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) ||
3096 	    (mask & AT_XVATTR))) {
3097 		ZFS_EXIT(zfsvfs);
3098 		return (SET_ERROR(EINVAL));
3099 	}
3100 
3101 	if (mask & AT_SIZE && vp->v_type == VDIR) {
3102 		ZFS_EXIT(zfsvfs);
3103 		return (SET_ERROR(EISDIR));
3104 	}
3105 
3106 	if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) {
3107 		ZFS_EXIT(zfsvfs);
3108 		return (SET_ERROR(EINVAL));
3109 	}
3110 
3111 	/*
3112 	 * If this is an xvattr_t, then get a pointer to the structure of
3113 	 * optional attributes.  If this is NULL, then we have a vattr_t.
3114 	 */
3115 	xoap = xva_getxoptattr(xvap);
3116 
3117 	xva_init(&tmpxvattr);
3118 
3119 	/*
3120 	 * Immutable files can only alter immutable bit and atime
3121 	 */
3122 	if ((zp->z_pflags & ZFS_IMMUTABLE) &&
3123 	    ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) ||
3124 	    ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
3125 		ZFS_EXIT(zfsvfs);
3126 		return (SET_ERROR(EPERM));
3127 	}
3128 
3129 	if ((mask & AT_SIZE) && (zp->z_pflags & ZFS_READONLY)) {
3130 		ZFS_EXIT(zfsvfs);
3131 		return (SET_ERROR(EPERM));
3132 	}
3133 
3134 	/*
3135 	 * Verify timestamps doesn't overflow 32 bits.
3136 	 * ZFS can handle large timestamps, but 32bit syscalls can't
3137 	 * handle times greater than 2039.  This check should be removed
3138 	 * once large timestamps are fully supported.
3139 	 */
3140 	if (mask & (AT_ATIME | AT_MTIME)) {
3141 		if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
3142 		    ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
3143 			ZFS_EXIT(zfsvfs);
3144 			return (SET_ERROR(EOVERFLOW));
3145 		}
3146 	}
3147 
3148 top:
3149 	attrzp = NULL;
3150 	aclp = NULL;
3151 
3152 	/* Can this be moved to before the top label? */
3153 	if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
3154 		ZFS_EXIT(zfsvfs);
3155 		return (SET_ERROR(EROFS));
3156 	}
3157 
3158 	/*
3159 	 * First validate permissions
3160 	 */
3161 
3162 	if (mask & AT_SIZE) {
3163 		/*
3164 		 * XXX - Note, we are not providing any open
3165 		 * mode flags here (like FNDELAY), so we may
3166 		 * block if there are locks present... this
3167 		 * should be addressed in openat().
3168 		 */
3169 		/* XXX - would it be OK to generate a log record here? */
3170 		err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
3171 		if (err) {
3172 			ZFS_EXIT(zfsvfs);
3173 			return (err);
3174 		}
3175 	}
3176 
3177 	if (mask & (AT_ATIME|AT_MTIME) ||
3178 	    ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
3179 	    XVA_ISSET_REQ(xvap, XAT_READONLY) ||
3180 	    XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
3181 	    XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
3182 	    XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
3183 	    XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
3184 	    XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
3185 		need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
3186 		    skipaclchk, cr);
3187 	}
3188 
3189 	if (mask & (AT_UID|AT_GID)) {
3190 		int	idmask = (mask & (AT_UID|AT_GID));
3191 		int	take_owner;
3192 		int	take_group;
3193 
3194 		/*
3195 		 * NOTE: even if a new mode is being set,
3196 		 * we may clear S_ISUID/S_ISGID bits.
3197 		 */
3198 
3199 		if (!(mask & AT_MODE))
3200 			vap->va_mode = zp->z_mode;
3201 
3202 		/*
3203 		 * Take ownership or chgrp to group we are a member of
3204 		 */
3205 
3206 		take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
3207 		take_group = (mask & AT_GID) &&
3208 		    zfs_groupmember(zfsvfs, vap->va_gid, cr);
3209 
3210 		/*
3211 		 * If both AT_UID and AT_GID are set then take_owner and
3212 		 * take_group must both be set in order to allow taking
3213 		 * ownership.
3214 		 *
3215 		 * Otherwise, send the check through secpolicy_vnode_setattr()
3216 		 *
3217 		 */
3218 
3219 		if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
3220 		    ((idmask == AT_UID) && take_owner) ||
3221 		    ((idmask == AT_GID) && take_group)) {
3222 			if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
3223 			    skipaclchk, cr) == 0) {
3224 				/*
3225 				 * Remove setuid/setgid for non-privileged users
3226 				 */
3227 				secpolicy_setid_clear(vap, vp, cr);
3228 				trim_mask = (mask & (AT_UID|AT_GID));
3229 			} else {
3230 				need_policy =  TRUE;
3231 			}
3232 		} else {
3233 			need_policy =  TRUE;
3234 		}
3235 	}
3236 
3237 	mutex_enter(&zp->z_lock);
3238 	oldva.va_mode = zp->z_mode;
3239 	zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
3240 	if (mask & AT_XVATTR) {
3241 		/*
3242 		 * Update xvattr mask to include only those attributes
3243 		 * that are actually changing.
3244 		 *
3245 		 * the bits will be restored prior to actually setting
3246 		 * the attributes so the caller thinks they were set.
3247 		 */
3248 		if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
3249 			if (xoap->xoa_appendonly !=
3250 			    ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
3251 				need_policy = TRUE;
3252 			} else {
3253 				XVA_CLR_REQ(xvap, XAT_APPENDONLY);
3254 				XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY);
3255 			}
3256 		}
3257 
3258 		if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
3259 			if (xoap->xoa_nounlink !=
3260 			    ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
3261 				need_policy = TRUE;
3262 			} else {
3263 				XVA_CLR_REQ(xvap, XAT_NOUNLINK);
3264 				XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK);
3265 			}
3266 		}
3267 
3268 		if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
3269 			if (xoap->xoa_immutable !=
3270 			    ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
3271 				need_policy = TRUE;
3272 			} else {
3273 				XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
3274 				XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE);
3275 			}
3276 		}
3277 
3278 		if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
3279 			if (xoap->xoa_nodump !=
3280 			    ((zp->z_pflags & ZFS_NODUMP) != 0)) {
3281 				need_policy = TRUE;
3282 			} else {
3283 				XVA_CLR_REQ(xvap, XAT_NODUMP);
3284 				XVA_SET_REQ(&tmpxvattr, XAT_NODUMP);
3285 			}
3286 		}
3287 
3288 		if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
3289 			if (xoap->xoa_av_modified !=
3290 			    ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
3291 				need_policy = TRUE;
3292 			} else {
3293 				XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
3294 				XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED);
3295 			}
3296 		}
3297 
3298 		if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
3299 			if ((vp->v_type != VREG &&
3300 			    xoap->xoa_av_quarantined) ||
3301 			    xoap->xoa_av_quarantined !=
3302 			    ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
3303 				need_policy = TRUE;
3304 			} else {
3305 				XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
3306 				XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED);
3307 			}
3308 		}
3309 
3310 		if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
3311 			mutex_exit(&zp->z_lock);
3312 			ZFS_EXIT(zfsvfs);
3313 			return (SET_ERROR(EPERM));
3314 		}
3315 
3316 		if (need_policy == FALSE &&
3317 		    (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
3318 		    XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
3319 			need_policy = TRUE;
3320 		}
3321 	}
3322 
3323 	mutex_exit(&zp->z_lock);
3324 
3325 	if (mask & AT_MODE) {
3326 		if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
3327 			err = secpolicy_setid_setsticky_clear(vp, vap,
3328 			    &oldva, cr);
3329 			if (err) {
3330 				ZFS_EXIT(zfsvfs);
3331 				return (err);
3332 			}
3333 			trim_mask |= AT_MODE;
3334 		} else {
3335 			need_policy = TRUE;
3336 		}
3337 	}
3338 
3339 	if (need_policy) {
3340 		/*
3341 		 * If trim_mask is set then take ownership
3342 		 * has been granted or write_acl is present and user
3343 		 * has the ability to modify mode.  In that case remove
3344 		 * UID|GID and or MODE from mask so that
3345 		 * secpolicy_vnode_setattr() doesn't revoke it.
3346 		 */
3347 
3348 		if (trim_mask) {
3349 			saved_mask = vap->va_mask;
3350 			vap->va_mask &= ~trim_mask;
3351 			if (trim_mask & AT_MODE) {
3352 				/*
3353 				 * Save the mode, as secpolicy_vnode_setattr()
3354 				 * will overwrite it with ova.va_mode.
3355 				 */
3356 				saved_mode = vap->va_mode;
3357 			}
3358 		}
3359 		err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
3360 		    (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
3361 		if (err) {
3362 			ZFS_EXIT(zfsvfs);
3363 			return (err);
3364 		}
3365 
3366 		if (trim_mask) {
3367 			vap->va_mask |= saved_mask;
3368 			if (trim_mask & AT_MODE) {
3369 				/*
3370 				 * Recover the mode after
3371 				 * secpolicy_vnode_setattr().
3372 				 */
3373 				vap->va_mode = saved_mode;
3374 			}
3375 		}
3376 	}
3377 
3378 	/*
3379 	 * secpolicy_vnode_setattr, or take ownership may have
3380 	 * changed va_mask
3381 	 */
3382 	mask = vap->va_mask;
3383 
3384 	if ((mask & (AT_UID | AT_GID))) {
3385 		err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
3386 		    &xattr_obj, sizeof (xattr_obj));
3387 
3388 		if (err == 0 && xattr_obj) {
3389 			err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp);
3390 			if (err)
3391 				goto out2;
3392 		}
3393 		if (mask & AT_UID) {
3394 			new_uid = zfs_fuid_create(zfsvfs,
3395 			    (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
3396 			if (new_uid != zp->z_uid &&
3397 			    zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) {
3398 				if (attrzp)
3399 					VN_RELE(ZTOV(attrzp));
3400 				err = SET_ERROR(EDQUOT);
3401 				goto out2;
3402 			}
3403 		}
3404 
3405 		if (mask & AT_GID) {
3406 			new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid,
3407 			    cr, ZFS_GROUP, &fuidp);
3408 			if (new_gid != zp->z_gid &&
3409 			    zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) {
3410 				if (attrzp)
3411 					VN_RELE(ZTOV(attrzp));
3412 				err = SET_ERROR(EDQUOT);
3413 				goto out2;
3414 			}
3415 		}
3416 	}
3417 	tx = dmu_tx_create(zfsvfs->z_os);
3418 
3419 	if (mask & AT_MODE) {
3420 		uint64_t pmode = zp->z_mode;
3421 		uint64_t acl_obj;
3422 		new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
3423 
3424 		if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED &&
3425 		    !(zp->z_pflags & ZFS_ACL_TRIVIAL)) {
3426 			err = SET_ERROR(EPERM);
3427 			goto out;
3428 		}
3429 
3430 		if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode))
3431 			goto out;
3432 
3433 		mutex_enter(&zp->z_lock);
3434 		if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
3435 			/*
3436 			 * Are we upgrading ACL from old V0 format
3437 			 * to V1 format?
3438 			 */
3439 			if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
3440 			    zfs_znode_acl_version(zp) ==
3441 			    ZFS_ACL_VERSION_INITIAL) {
3442 				dmu_tx_hold_free(tx, acl_obj, 0,
3443 				    DMU_OBJECT_END);
3444 				dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3445 				    0, aclp->z_acl_bytes);
3446 			} else {
3447 				dmu_tx_hold_write(tx, acl_obj, 0,
3448 				    aclp->z_acl_bytes);
3449 			}
3450 		} else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3451 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3452 			    0, aclp->z_acl_bytes);
3453 		}
3454 		mutex_exit(&zp->z_lock);
3455 		dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3456 	} else {
3457 		if ((mask & AT_XVATTR) &&
3458 		    XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3459 			dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3460 		else
3461 			dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3462 	}
3463 
3464 	if (attrzp) {
3465 		dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
3466 	}
3467 
3468 	fuid_dirtied = zfsvfs->z_fuid_dirty;
3469 	if (fuid_dirtied)
3470 		zfs_fuid_txhold(zfsvfs, tx);
3471 
3472 	zfs_sa_upgrade_txholds(tx, zp);
3473 
3474 	err = dmu_tx_assign(tx, TXG_WAIT);
3475 	if (err)
3476 		goto out;
3477 
3478 	count = 0;
3479 	/*
3480 	 * Set each attribute requested.
3481 	 * We group settings according to the locks they need to acquire.
3482 	 *
3483 	 * Note: you cannot set ctime directly, although it will be
3484 	 * updated as a side-effect of calling this function.
3485 	 */
3486 
3487 
3488 	if (mask & (AT_UID|AT_GID|AT_MODE))
3489 		mutex_enter(&zp->z_acl_lock);
3490 	mutex_enter(&zp->z_lock);
3491 
3492 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
3493 	    &zp->z_pflags, sizeof (zp->z_pflags));
3494 
3495 	if (attrzp) {
3496 		if (mask & (AT_UID|AT_GID|AT_MODE))
3497 			mutex_enter(&attrzp->z_acl_lock);
3498 		mutex_enter(&attrzp->z_lock);
3499 		SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3500 		    SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags,
3501 		    sizeof (attrzp->z_pflags));
3502 	}
3503 
3504 	if (mask & (AT_UID|AT_GID)) {
3505 
3506 		if (mask & AT_UID) {
3507 			SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
3508 			    &new_uid, sizeof (new_uid));
3509 			zp->z_uid = new_uid;
3510 			if (attrzp) {
3511 				SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3512 				    SA_ZPL_UID(zfsvfs), NULL, &new_uid,
3513 				    sizeof (new_uid));
3514 				attrzp->z_uid = new_uid;
3515 			}
3516 		}
3517 
3518 		if (mask & AT_GID) {
3519 			SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs),
3520 			    NULL, &new_gid, sizeof (new_gid));
3521 			zp->z_gid = new_gid;
3522 			if (attrzp) {
3523 				SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3524 				    SA_ZPL_GID(zfsvfs), NULL, &new_gid,
3525 				    sizeof (new_gid));
3526 				attrzp->z_gid = new_gid;
3527 			}
3528 		}
3529 		if (!(mask & AT_MODE)) {
3530 			SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs),
3531 			    NULL, &new_mode, sizeof (new_mode));
3532 			new_mode = zp->z_mode;
3533 		}
3534 		err = zfs_acl_chown_setattr(zp);
3535 		ASSERT(err == 0);
3536 		if (attrzp) {
3537 			err = zfs_acl_chown_setattr(attrzp);
3538 			ASSERT(err == 0);
3539 		}
3540 	}
3541 
3542 	if (mask & AT_MODE) {
3543 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
3544 		    &new_mode, sizeof (new_mode));
3545 		zp->z_mode = new_mode;
3546 		ASSERT3U((uintptr_t)aclp, !=, 0);
3547 		err = zfs_aclset_common(zp, aclp, cr, tx);
3548 		ASSERT0(err);
3549 		if (zp->z_acl_cached)
3550 			zfs_acl_free(zp->z_acl_cached);
3551 		zp->z_acl_cached = aclp;
3552 		aclp = NULL;
3553 	}
3554 
3555 
3556 	if (mask & AT_ATIME) {
3557 		ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime);
3558 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
3559 		    &zp->z_atime, sizeof (zp->z_atime));
3560 	}
3561 
3562 	if (mask & AT_MTIME) {
3563 		ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
3564 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
3565 		    mtime, sizeof (mtime));
3566 	}
3567 
3568 	/* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
3569 	if (mask & AT_SIZE && !(mask & AT_MTIME)) {
3570 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs),
3571 		    NULL, mtime, sizeof (mtime));
3572 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3573 		    &ctime, sizeof (ctime));
3574 		zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
3575 		    B_TRUE);
3576 	} else if (mask != 0) {
3577 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3578 		    &ctime, sizeof (ctime));
3579 		zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime,
3580 		    B_TRUE);
3581 		if (attrzp) {
3582 			SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3583 			    SA_ZPL_CTIME(zfsvfs), NULL,
3584 			    &ctime, sizeof (ctime));
3585 			zfs_tstamp_update_setup(attrzp, STATE_CHANGED,
3586 			    mtime, ctime, B_TRUE);
3587 		}
3588 	}
3589 	/*
3590 	 * Do this after setting timestamps to prevent timestamp
3591 	 * update from toggling bit
3592 	 */
3593 
3594 	if (xoap && (mask & AT_XVATTR)) {
3595 
3596 		/*
3597 		 * restore trimmed off masks
3598 		 * so that return masks can be set for caller.
3599 		 */
3600 
3601 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) {
3602 			XVA_SET_REQ(xvap, XAT_APPENDONLY);
3603 		}
3604 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) {
3605 			XVA_SET_REQ(xvap, XAT_NOUNLINK);
3606 		}
3607 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) {
3608 			XVA_SET_REQ(xvap, XAT_IMMUTABLE);
3609 		}
3610 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) {
3611 			XVA_SET_REQ(xvap, XAT_NODUMP);
3612 		}
3613 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) {
3614 			XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
3615 		}
3616 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) {
3617 			XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
3618 		}
3619 
3620 		if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3621 			ASSERT(vp->v_type == VREG);
3622 
3623 		zfs_xvattr_set(zp, xvap, tx);
3624 	}
3625 
3626 	if (fuid_dirtied)
3627 		zfs_fuid_sync(zfsvfs, tx);
3628 
3629 	if (mask != 0)
3630 		zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
3631 
3632 	mutex_exit(&zp->z_lock);
3633 	if (mask & (AT_UID|AT_GID|AT_MODE))
3634 		mutex_exit(&zp->z_acl_lock);
3635 
3636 	if (attrzp) {
3637 		if (mask & (AT_UID|AT_GID|AT_MODE))
3638 			mutex_exit(&attrzp->z_acl_lock);
3639 		mutex_exit(&attrzp->z_lock);
3640 	}
3641 out:
3642 	if (err == 0 && attrzp) {
3643 		err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
3644 		    xattr_count, tx);
3645 		ASSERT(err2 == 0);
3646 	}
3647 
3648 	if (attrzp)
3649 		VN_RELE(ZTOV(attrzp));
3650 
3651 	if (aclp)
3652 		zfs_acl_free(aclp);
3653 
3654 	if (fuidp) {
3655 		zfs_fuid_info_free(fuidp);
3656 		fuidp = NULL;
3657 	}
3658 
3659 	if (err) {
3660 		dmu_tx_abort(tx);
3661 		if (err == ERESTART)
3662 			goto top;
3663 	} else {
3664 		err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
3665 		dmu_tx_commit(tx);
3666 	}
3667 
3668 out2:
3669 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3670 		zil_commit(zilog, 0);
3671 
3672 	ZFS_EXIT(zfsvfs);
3673 	return (err);
3674 }
3675 
3676 typedef struct zfs_zlock {
3677 	krwlock_t	*zl_rwlock;	/* lock we acquired */
3678 	znode_t		*zl_znode;	/* znode we held */
3679 	struct zfs_zlock *zl_next;	/* next in list */
3680 } zfs_zlock_t;
3681 
3682 /*
3683  * Drop locks and release vnodes that were held by zfs_rename_lock().
3684  */
3685 static void
zfs_rename_unlock(zfs_zlock_t ** zlpp)3686 zfs_rename_unlock(zfs_zlock_t **zlpp)
3687 {
3688 	zfs_zlock_t *zl;
3689 
3690 	while ((zl = *zlpp) != NULL) {
3691 		if (zl->zl_znode != NULL)
3692 			VN_RELE(ZTOV(zl->zl_znode));
3693 		rw_exit(zl->zl_rwlock);
3694 		*zlpp = zl->zl_next;
3695 		kmem_free(zl, sizeof (*zl));
3696 	}
3697 }
3698 
3699 /*
3700  * Search back through the directory tree, using the ".." entries.
3701  * Lock each directory in the chain to prevent concurrent renames.
3702  * Fail any attempt to move a directory into one of its own descendants.
3703  * XXX - z_parent_lock can overlap with map or grow locks
3704  */
3705 static int
zfs_rename_lock(znode_t * szp,znode_t * tdzp,znode_t * sdzp,zfs_zlock_t ** zlpp)3706 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
3707 {
3708 	zfs_zlock_t	*zl;
3709 	znode_t		*zp = tdzp;
3710 	uint64_t	rootid = zp->z_zfsvfs->z_root;
3711 	uint64_t	oidp = zp->z_id;
3712 	krwlock_t	*rwlp = &szp->z_parent_lock;
3713 	krw_t		rw = RW_WRITER;
3714 
3715 	/*
3716 	 * First pass write-locks szp and compares to zp->z_id.
3717 	 * Later passes read-lock zp and compare to zp->z_parent.
3718 	 */
3719 	do {
3720 		if (!rw_tryenter(rwlp, rw)) {
3721 			/*
3722 			 * Another thread is renaming in this path.
3723 			 * Note that if we are a WRITER, we don't have any
3724 			 * parent_locks held yet.
3725 			 */
3726 			if (rw == RW_READER && zp->z_id > szp->z_id) {
3727 				/*
3728 				 * Drop our locks and restart
3729 				 */
3730 				zfs_rename_unlock(&zl);
3731 				*zlpp = NULL;
3732 				zp = tdzp;
3733 				oidp = zp->z_id;
3734 				rwlp = &szp->z_parent_lock;
3735 				rw = RW_WRITER;
3736 				continue;
3737 			} else {
3738 				/*
3739 				 * Wait for other thread to drop its locks
3740 				 */
3741 				rw_enter(rwlp, rw);
3742 			}
3743 		}
3744 
3745 		zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
3746 		zl->zl_rwlock = rwlp;
3747 		zl->zl_znode = NULL;
3748 		zl->zl_next = *zlpp;
3749 		*zlpp = zl;
3750 
3751 		if (oidp == szp->z_id)		/* We're a descendant of szp */
3752 			return (SET_ERROR(EINVAL));
3753 
3754 		if (oidp == rootid)		/* We've hit the top */
3755 			return (0);
3756 
3757 		if (rw == RW_READER) {		/* i.e. not the first pass */
3758 			int error = zfs_zget(zp->z_zfsvfs, oidp, &zp);
3759 			if (error)
3760 				return (error);
3761 			zl->zl_znode = zp;
3762 		}
3763 		(void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zp->z_zfsvfs),
3764 		    &oidp, sizeof (oidp));
3765 		rwlp = &zp->z_parent_lock;
3766 		rw = RW_READER;
3767 
3768 	} while (zp->z_id != sdzp->z_id);
3769 
3770 	return (0);
3771 }
3772 
3773 /*
3774  * Move an entry from the provided source directory to the target
3775  * directory.  Change the entry name as indicated.
3776  *
3777  *	IN:	sdvp	- Source directory containing the "old entry".
3778  *		snm	- Old entry name.
3779  *		tdvp	- Target directory to contain the "new entry".
3780  *		tnm	- New entry name.
3781  *		cr	- credentials of caller.
3782  *		ct	- caller context
3783  *		flags	- case flags
3784  *
3785  *	RETURN:	0 on success, error code on failure.
3786  *
3787  * Timestamps:
3788  *	sdvp,tdvp - ctime|mtime updated
3789  */
3790 /*ARGSUSED*/
3791 static int
zfs_rename(vnode_t * sdvp,char * snm,vnode_t * tdvp,char * tnm,cred_t * cr,caller_context_t * ct,int flags)3792 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr,
3793     caller_context_t *ct, int flags)
3794 {
3795 	znode_t		*tdzp, *sdzp, *szp, *tzp;
3796 	zfsvfs_t 	*zfsvfs;
3797 	zilog_t		*zilog;
3798 	vnode_t		*realvp;
3799 	zfs_dirlock_t	*sdl, *tdl;
3800 	dmu_tx_t	*tx;
3801 	zfs_zlock_t	*zl;
3802 	int		cmp, serr, terr;
3803 	int		error = 0;
3804 	int		zflg = 0;
3805 	boolean_t	waited = B_FALSE;
3806 
3807 	tdzp = VTOZ(tdvp);
3808 	ZFS_VERIFY_ZP(tdzp);
3809 	zfsvfs = tdzp->z_zfsvfs;
3810 	ZFS_ENTER(zfsvfs);
3811 	zilog = zfsvfs->z_log;
3812 	sdzp = VTOZ(sdvp);
3813 
3814 	/*
3815 	 * In case sdzp is not valid, let's be sure to exit from the right
3816 	 * zfsvfs_t.
3817 	 */
3818 	if (sdzp->z_sa_hdl == NULL) {
3819 		ZFS_EXIT(zfsvfs);
3820 		return (SET_ERROR(EIO));
3821 	}
3822 
3823 	/*
3824 	 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the
3825 	 * ctldir appear to have the same v_vfsp.
3826 	 */
3827 	if (sdzp->z_zfsvfs != zfsvfs || zfsctl_is_node(tdvp)) {
3828 		ZFS_EXIT(zfsvfs);
3829 		return (SET_ERROR(EXDEV));
3830 	}
3831 
3832 	if (zfsvfs->z_utf8 && u8_validate(tnm,
3833 	    strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3834 		ZFS_EXIT(zfsvfs);
3835 		return (SET_ERROR(EILSEQ));
3836 	}
3837 
3838 	if (flags & FIGNORECASE)
3839 		zflg |= ZCILOOK;
3840 
3841 top:
3842 	szp = NULL;
3843 	tzp = NULL;
3844 	zl = NULL;
3845 
3846 	/*
3847 	 * This is to prevent the creation of links into attribute space
3848 	 * by renaming a linked file into/outof an attribute directory.
3849 	 * See the comment in zfs_link() for why this is considered bad.
3850 	 */
3851 	if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) {
3852 		ZFS_EXIT(zfsvfs);
3853 		return (SET_ERROR(EINVAL));
3854 	}
3855 
3856 	/*
3857 	 * Lock source and target directory entries.  To prevent deadlock,
3858 	 * a lock ordering must be defined.  We lock the directory with
3859 	 * the smallest object id first, or if it's a tie, the one with
3860 	 * the lexically first name.
3861 	 */
3862 	if (sdzp->z_id < tdzp->z_id) {
3863 		cmp = -1;
3864 	} else if (sdzp->z_id > tdzp->z_id) {
3865 		cmp = 1;
3866 	} else {
3867 		/*
3868 		 * First compare the two name arguments without
3869 		 * considering any case folding.
3870 		 */
3871 		int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER);
3872 
3873 		cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
3874 		ASSERT(error == 0 || !zfsvfs->z_utf8);
3875 		if (cmp == 0) {
3876 			/*
3877 			 * POSIX: "If the old argument and the new argument
3878 			 * both refer to links to the same existing file,
3879 			 * the rename() function shall return successfully
3880 			 * and perform no other action."
3881 			 */
3882 			ZFS_EXIT(zfsvfs);
3883 			return (0);
3884 		}
3885 		/*
3886 		 * If the file system is case-folding, then we may
3887 		 * have some more checking to do.  A case-folding file
3888 		 * system is either supporting mixed case sensitivity
3889 		 * access or is completely case-insensitive.  Note
3890 		 * that the file system is always case preserving.
3891 		 *
3892 		 * In mixed sensitivity mode case sensitive behavior
3893 		 * is the default.  FIGNORECASE must be used to
3894 		 * explicitly request case insensitive behavior.
3895 		 *
3896 		 * If the source and target names provided differ only
3897 		 * by case (e.g., a request to rename 'tim' to 'Tim'),
3898 		 * we will treat this as a special case in the
3899 		 * case-insensitive mode: as long as the source name
3900 		 * is an exact match, we will allow this to proceed as
3901 		 * a name-change request.
3902 		 */
3903 		if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
3904 		    (zfsvfs->z_case == ZFS_CASE_MIXED &&
3905 		    flags & FIGNORECASE)) &&
3906 		    u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST,
3907 		    &error) == 0) {
3908 			/*
3909 			 * case preserving rename request, require exact
3910 			 * name matches
3911 			 */
3912 			zflg |= ZCIEXACT;
3913 			zflg &= ~ZCILOOK;
3914 		}
3915 	}
3916 
3917 	/*
3918 	 * If the source and destination directories are the same, we should
3919 	 * grab the z_name_lock of that directory only once.
3920 	 */
3921 	if (sdzp == tdzp) {
3922 		zflg |= ZHAVELOCK;
3923 		rw_enter(&sdzp->z_name_lock, RW_READER);
3924 	}
3925 
3926 	if (cmp < 0) {
3927 		serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
3928 		    ZEXISTS | zflg, NULL, NULL);
3929 		terr = zfs_dirent_lock(&tdl,
3930 		    tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
3931 	} else {
3932 		terr = zfs_dirent_lock(&tdl,
3933 		    tdzp, tnm, &tzp, zflg, NULL, NULL);
3934 		serr = zfs_dirent_lock(&sdl,
3935 		    sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
3936 		    NULL, NULL);
3937 	}
3938 
3939 	if (serr) {
3940 		/*
3941 		 * Source entry invalid or not there.
3942 		 */
3943 		if (!terr) {
3944 			zfs_dirent_unlock(tdl);
3945 			if (tzp)
3946 				VN_RELE(ZTOV(tzp));
3947 		}
3948 
3949 		if (sdzp == tdzp)
3950 			rw_exit(&sdzp->z_name_lock);
3951 
3952 		/*
3953 		 * FreeBSD: In OpenSolaris they only check if rename source is
3954 		 * ".." here, because "." is handled in their lookup. This is
3955 		 * not the case for FreeBSD, so we check for "." explicitly.
3956 		 */
3957 		if (strcmp(snm, ".") == 0 || strcmp(snm, "..") == 0)
3958 			serr = SET_ERROR(EINVAL);
3959 		ZFS_EXIT(zfsvfs);
3960 		return (serr);
3961 	}
3962 	if (terr) {
3963 		zfs_dirent_unlock(sdl);
3964 		VN_RELE(ZTOV(szp));
3965 
3966 		if (sdzp == tdzp)
3967 			rw_exit(&sdzp->z_name_lock);
3968 
3969 		if (strcmp(tnm, "..") == 0)
3970 			terr = SET_ERROR(EINVAL);
3971 		ZFS_EXIT(zfsvfs);
3972 		return (terr);
3973 	}
3974 
3975 	/*
3976 	 * Must have write access at the source to remove the old entry
3977 	 * and write access at the target to create the new entry.
3978 	 * Note that if target and source are the same, this can be
3979 	 * done in a single check.
3980 	 */
3981 
3982 	if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr))
3983 		goto out;
3984 
3985 	if (ZTOV(szp)->v_type == VDIR) {
3986 		/*
3987 		 * Check to make sure rename is valid.
3988 		 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3989 		 */
3990 		if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl))
3991 			goto out;
3992 	}
3993 
3994 	/*
3995 	 * Does target exist?
3996 	 */
3997 	if (tzp) {
3998 		/*
3999 		 * Source and target must be the same type.
4000 		 */
4001 		if (ZTOV(szp)->v_type == VDIR) {
4002 			if (ZTOV(tzp)->v_type != VDIR) {
4003 				error = SET_ERROR(ENOTDIR);
4004 				goto out;
4005 			}
4006 		} else {
4007 			if (ZTOV(tzp)->v_type == VDIR) {
4008 				error = SET_ERROR(EISDIR);
4009 				goto out;
4010 			}
4011 		}
4012 		/*
4013 		 * POSIX dictates that when the source and target
4014 		 * entries refer to the same file object, rename
4015 		 * must do nothing and exit without error.
4016 		 */
4017 		if (szp->z_id == tzp->z_id) {
4018 			error = 0;
4019 			goto out;
4020 		}
4021 	}
4022 
4023 	vnevent_rename_src(ZTOV(szp), sdvp, snm, ct);
4024 	if (tzp)
4025 		vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct);
4026 
4027 	/*
4028 	 * notify the target directory if it is not the same
4029 	 * as source directory.
4030 	 */
4031 	if (tdvp != sdvp) {
4032 		vnevent_rename_dest_dir(tdvp, ct);
4033 	}
4034 
4035 	tx = dmu_tx_create(zfsvfs->z_os);
4036 	dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
4037 	dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
4038 	dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
4039 	dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
4040 	if (sdzp != tdzp) {
4041 		dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
4042 		zfs_sa_upgrade_txholds(tx, tdzp);
4043 	}
4044 	if (tzp) {
4045 		dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
4046 		zfs_sa_upgrade_txholds(tx, tzp);
4047 	}
4048 
4049 	zfs_sa_upgrade_txholds(tx, szp);
4050 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
4051 	error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
4052 	if (error) {
4053 		if (zl != NULL)
4054 			zfs_rename_unlock(&zl);
4055 		zfs_dirent_unlock(sdl);
4056 		zfs_dirent_unlock(tdl);
4057 
4058 		if (sdzp == tdzp)
4059 			rw_exit(&sdzp->z_name_lock);
4060 
4061 		VN_RELE(ZTOV(szp));
4062 		if (tzp)
4063 			VN_RELE(ZTOV(tzp));
4064 		if (error == ERESTART) {
4065 			waited = B_TRUE;
4066 			dmu_tx_wait(tx);
4067 			dmu_tx_abort(tx);
4068 			goto top;
4069 		}
4070 		dmu_tx_abort(tx);
4071 		ZFS_EXIT(zfsvfs);
4072 		return (error);
4073 	}
4074 
4075 	if (tzp)	/* Attempt to remove the existing target */
4076 		error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
4077 
4078 	if (error == 0) {
4079 		error = zfs_link_create(tdl, szp, tx, ZRENAMING);
4080 		if (error == 0) {
4081 			szp->z_pflags |= ZFS_AV_MODIFIED;
4082 
4083 			error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs),
4084 			    (void *)&szp->z_pflags, sizeof (uint64_t), tx);
4085 			ASSERT0(error);
4086 
4087 			error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
4088 			if (error == 0) {
4089 				zfs_log_rename(zilog, tx, TX_RENAME |
4090 				    (flags & FIGNORECASE ? TX_CI : 0), sdzp,
4091 				    sdl->dl_name, tdzp, tdl->dl_name, szp);
4092 
4093 				/*
4094 				 * Update path information for the target vnode
4095 				 */
4096 				vn_renamepath(tdvp, ZTOV(szp), tnm,
4097 				    strlen(tnm));
4098 			} else {
4099 				/*
4100 				 * At this point, we have successfully created
4101 				 * the target name, but have failed to remove
4102 				 * the source name.  Since the create was done
4103 				 * with the ZRENAMING flag, there are
4104 				 * complications; for one, the link count is
4105 				 * wrong.  The easiest way to deal with this
4106 				 * is to remove the newly created target, and
4107 				 * return the original error.  This must
4108 				 * succeed; fortunately, it is very unlikely to
4109 				 * fail, since we just created it.
4110 				 */
4111 				VERIFY3U(zfs_link_destroy(tdl, szp, tx,
4112 				    ZRENAMING, NULL), ==, 0);
4113 			}
4114 		}
4115 #ifdef FREEBSD_NAMECACHE
4116 		if (error == 0) {
4117 			cache_purge(sdvp);
4118 			cache_purge(tdvp);
4119 			cache_purge(ZTOV(szp));
4120 			if (tzp)
4121 				cache_purge(ZTOV(tzp));
4122 		}
4123 #endif
4124 	}
4125 
4126 	dmu_tx_commit(tx);
4127 out:
4128 	if (zl != NULL)
4129 		zfs_rename_unlock(&zl);
4130 
4131 	zfs_dirent_unlock(sdl);
4132 	zfs_dirent_unlock(tdl);
4133 
4134 	if (sdzp == tdzp)
4135 		rw_exit(&sdzp->z_name_lock);
4136 
4137 
4138 	VN_RELE(ZTOV(szp));
4139 	if (tzp)
4140 		VN_RELE(ZTOV(tzp));
4141 
4142 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4143 		zil_commit(zilog, 0);
4144 
4145 	ZFS_EXIT(zfsvfs);
4146 
4147 	return (error);
4148 }
4149 
4150 /*
4151  * Insert the indicated symbolic reference entry into the directory.
4152  *
4153  *	IN:	dvp	- Directory to contain new symbolic link.
4154  *		link	- Name for new symlink entry.
4155  *		vap	- Attributes of new entry.
4156  *		cr	- credentials of caller.
4157  *		ct	- caller context
4158  *		flags	- case flags
4159  *
4160  *	RETURN:	0 on success, error code on failure.
4161  *
4162  * Timestamps:
4163  *	dvp - ctime|mtime updated
4164  */
4165 /*ARGSUSED*/
4166 static int
zfs_symlink(vnode_t * dvp,vnode_t ** vpp,char * name,vattr_t * vap,char * link,cred_t * cr,kthread_t * td)4167 zfs_symlink(vnode_t *dvp, vnode_t **vpp, char *name, vattr_t *vap, char *link,
4168     cred_t *cr, kthread_t *td)
4169 {
4170 	znode_t		*zp, *dzp = VTOZ(dvp);
4171 	zfs_dirlock_t	*dl;
4172 	dmu_tx_t	*tx;
4173 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
4174 	zilog_t		*zilog;
4175 	uint64_t	len = strlen(link);
4176 	int		error;
4177 	int		zflg = ZNEW;
4178 	zfs_acl_ids_t	acl_ids;
4179 	boolean_t	fuid_dirtied;
4180 	uint64_t	txtype = TX_SYMLINK;
4181 	boolean_t	waited = B_FALSE;
4182 	int		flags = 0;
4183 
4184 	ASSERT(vap->va_type == VLNK);
4185 
4186 	ZFS_ENTER(zfsvfs);
4187 	ZFS_VERIFY_ZP(dzp);
4188 	zilog = zfsvfs->z_log;
4189 
4190 	if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
4191 	    NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4192 		ZFS_EXIT(zfsvfs);
4193 		return (SET_ERROR(EILSEQ));
4194 	}
4195 	if (flags & FIGNORECASE)
4196 		zflg |= ZCILOOK;
4197 
4198 	if (len > MAXPATHLEN) {
4199 		ZFS_EXIT(zfsvfs);
4200 		return (SET_ERROR(ENAMETOOLONG));
4201 	}
4202 
4203 	if ((error = zfs_acl_ids_create(dzp, 0,
4204 	    vap, cr, NULL, &acl_ids)) != 0) {
4205 		ZFS_EXIT(zfsvfs);
4206 		return (error);
4207 	}
4208 
4209 	getnewvnode_reserve(1);
4210 
4211 top:
4212 	/*
4213 	 * Attempt to lock directory; fail if entry already exists.
4214 	 */
4215 	error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
4216 	if (error) {
4217 		zfs_acl_ids_free(&acl_ids);
4218 		getnewvnode_drop_reserve();
4219 		ZFS_EXIT(zfsvfs);
4220 		return (error);
4221 	}
4222 
4223 	if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4224 		zfs_acl_ids_free(&acl_ids);
4225 		zfs_dirent_unlock(dl);
4226 		getnewvnode_drop_reserve();
4227 		ZFS_EXIT(zfsvfs);
4228 		return (error);
4229 	}
4230 
4231 	if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
4232 		zfs_acl_ids_free(&acl_ids);
4233 		zfs_dirent_unlock(dl);
4234 		getnewvnode_drop_reserve();
4235 		ZFS_EXIT(zfsvfs);
4236 		return (SET_ERROR(EDQUOT));
4237 	}
4238 	tx = dmu_tx_create(zfsvfs->z_os);
4239 	fuid_dirtied = zfsvfs->z_fuid_dirty;
4240 	dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
4241 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4242 	dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
4243 	    ZFS_SA_BASE_ATTR_SIZE + len);
4244 	dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
4245 	if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
4246 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
4247 		    acl_ids.z_aclp->z_acl_bytes);
4248 	}
4249 	if (fuid_dirtied)
4250 		zfs_fuid_txhold(zfsvfs, tx);
4251 	error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
4252 	if (error) {
4253 		zfs_dirent_unlock(dl);
4254 		if (error == ERESTART) {
4255 			waited = B_TRUE;
4256 			dmu_tx_wait(tx);
4257 			dmu_tx_abort(tx);
4258 			goto top;
4259 		}
4260 		zfs_acl_ids_free(&acl_ids);
4261 		dmu_tx_abort(tx);
4262 		getnewvnode_drop_reserve();
4263 		ZFS_EXIT(zfsvfs);
4264 		return (error);
4265 	}
4266 
4267 	/*
4268 	 * Create a new object for the symlink.
4269 	 * for version 4 ZPL datsets the symlink will be an SA attribute
4270 	 */
4271 	zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
4272 
4273 	if (fuid_dirtied)
4274 		zfs_fuid_sync(zfsvfs, tx);
4275 
4276 	mutex_enter(&zp->z_lock);
4277 	if (zp->z_is_sa)
4278 		error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs),
4279 		    link, len, tx);
4280 	else
4281 		zfs_sa_symlink(zp, link, len, tx);
4282 	mutex_exit(&zp->z_lock);
4283 
4284 	zp->z_size = len;
4285 	(void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
4286 	    &zp->z_size, sizeof (zp->z_size), tx);
4287 	/*
4288 	 * Insert the new object into the directory.
4289 	 */
4290 	(void) zfs_link_create(dl, zp, tx, ZNEW);
4291 
4292 	if (flags & FIGNORECASE)
4293 		txtype |= TX_CI;
4294 	zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
4295 	*vpp = ZTOV(zp);
4296 
4297 	zfs_acl_ids_free(&acl_ids);
4298 
4299 	dmu_tx_commit(tx);
4300 
4301 	getnewvnode_drop_reserve();
4302 
4303 	zfs_dirent_unlock(dl);
4304 
4305 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4306 		zil_commit(zilog, 0);
4307 
4308 	ZFS_EXIT(zfsvfs);
4309 	return (error);
4310 }
4311 
4312 /*
4313  * Return, in the buffer contained in the provided uio structure,
4314  * the symbolic path referred to by vp.
4315  *
4316  *	IN:	vp	- vnode of symbolic link.
4317  *		uio	- structure to contain the link path.
4318  *		cr	- credentials of caller.
4319  *		ct	- caller context
4320  *
4321  *	OUT:	uio	- structure containing the link path.
4322  *
4323  *	RETURN:	0 on success, error code on failure.
4324  *
4325  * Timestamps:
4326  *	vp - atime updated
4327  */
4328 /* ARGSUSED */
4329 static int
zfs_readlink(vnode_t * vp,uio_t * uio,cred_t * cr,caller_context_t * ct)4330 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct)
4331 {
4332 	znode_t		*zp = VTOZ(vp);
4333 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
4334 	int		error;
4335 
4336 	ZFS_ENTER(zfsvfs);
4337 	ZFS_VERIFY_ZP(zp);
4338 
4339 	mutex_enter(&zp->z_lock);
4340 	if (zp->z_is_sa)
4341 		error = sa_lookup_uio(zp->z_sa_hdl,
4342 		    SA_ZPL_SYMLINK(zfsvfs), uio);
4343 	else
4344 		error = zfs_sa_readlink(zp, uio);
4345 	mutex_exit(&zp->z_lock);
4346 
4347 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4348 
4349 	ZFS_EXIT(zfsvfs);
4350 	return (error);
4351 }
4352 
4353 /*
4354  * Insert a new entry into directory tdvp referencing svp.
4355  *
4356  *	IN:	tdvp	- Directory to contain new entry.
4357  *		svp	- vnode of new entry.
4358  *		name	- name of new entry.
4359  *		cr	- credentials of caller.
4360  *		ct	- caller context
4361  *
4362  *	RETURN:	0 on success, error code on failure.
4363  *
4364  * Timestamps:
4365  *	tdvp - ctime|mtime updated
4366  *	 svp - ctime updated
4367  */
4368 /* ARGSUSED */
4369 static int
zfs_link(vnode_t * tdvp,vnode_t * svp,char * name,cred_t * cr,caller_context_t * ct,int flags)4370 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr,
4371     caller_context_t *ct, int flags)
4372 {
4373 	znode_t		*dzp = VTOZ(tdvp);
4374 	znode_t		*tzp, *szp;
4375 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
4376 	zilog_t		*zilog;
4377 	zfs_dirlock_t	*dl;
4378 	dmu_tx_t	*tx;
4379 	vnode_t		*realvp;
4380 	int		error;
4381 	int		zf = ZNEW;
4382 	uint64_t	parent;
4383 	uid_t		owner;
4384 	boolean_t	waited = B_FALSE;
4385 
4386 	ASSERT(tdvp->v_type == VDIR);
4387 
4388 	ZFS_ENTER(zfsvfs);
4389 	ZFS_VERIFY_ZP(dzp);
4390 	zilog = zfsvfs->z_log;
4391 
4392 	if (VOP_REALVP(svp, &realvp, ct) == 0)
4393 		svp = realvp;
4394 
4395 	/*
4396 	 * POSIX dictates that we return EPERM here.
4397 	 * Better choices include ENOTSUP or EISDIR.
4398 	 */
4399 	if (svp->v_type == VDIR) {
4400 		ZFS_EXIT(zfsvfs);
4401 		return (SET_ERROR(EPERM));
4402 	}
4403 
4404 	szp = VTOZ(svp);
4405 	ZFS_VERIFY_ZP(szp);
4406 
4407 	if (szp->z_pflags & (ZFS_APPENDONLY | ZFS_IMMUTABLE | ZFS_READONLY)) {
4408 		ZFS_EXIT(zfsvfs);
4409 		return (SET_ERROR(EPERM));
4410 	}
4411 
4412 	/*
4413 	 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the
4414 	 * ctldir appear to have the same v_vfsp.
4415 	 */
4416 	if (szp->z_zfsvfs != zfsvfs || zfsctl_is_node(svp)) {
4417 		ZFS_EXIT(zfsvfs);
4418 		return (SET_ERROR(EXDEV));
4419 	}
4420 
4421 	/* Prevent links to .zfs/shares files */
4422 
4423 	if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
4424 	    &parent, sizeof (uint64_t))) != 0) {
4425 		ZFS_EXIT(zfsvfs);
4426 		return (error);
4427 	}
4428 	if (parent == zfsvfs->z_shares_dir) {
4429 		ZFS_EXIT(zfsvfs);
4430 		return (SET_ERROR(EPERM));
4431 	}
4432 
4433 	if (zfsvfs->z_utf8 && u8_validate(name,
4434 	    strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4435 		ZFS_EXIT(zfsvfs);
4436 		return (SET_ERROR(EILSEQ));
4437 	}
4438 	if (flags & FIGNORECASE)
4439 		zf |= ZCILOOK;
4440 
4441 	/*
4442 	 * We do not support links between attributes and non-attributes
4443 	 * because of the potential security risk of creating links
4444 	 * into "normal" file space in order to circumvent restrictions
4445 	 * imposed in attribute space.
4446 	 */
4447 	if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) {
4448 		ZFS_EXIT(zfsvfs);
4449 		return (SET_ERROR(EINVAL));
4450 	}
4451 
4452 
4453 	owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER);
4454 	if (owner != crgetuid(cr) && secpolicy_basic_link(svp, cr) != 0) {
4455 		ZFS_EXIT(zfsvfs);
4456 		return (SET_ERROR(EPERM));
4457 	}
4458 
4459 	if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4460 		ZFS_EXIT(zfsvfs);
4461 		return (error);
4462 	}
4463 
4464 top:
4465 	/*
4466 	 * Attempt to lock directory; fail if entry already exists.
4467 	 */
4468 	error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
4469 	if (error) {
4470 		ZFS_EXIT(zfsvfs);
4471 		return (error);
4472 	}
4473 
4474 	tx = dmu_tx_create(zfsvfs->z_os);
4475 	dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
4476 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4477 	zfs_sa_upgrade_txholds(tx, szp);
4478 	zfs_sa_upgrade_txholds(tx, dzp);
4479 	error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
4480 	if (error) {
4481 		zfs_dirent_unlock(dl);
4482 		if (error == ERESTART) {
4483 			waited = B_TRUE;
4484 			dmu_tx_wait(tx);
4485 			dmu_tx_abort(tx);
4486 			goto top;
4487 		}
4488 		dmu_tx_abort(tx);
4489 		ZFS_EXIT(zfsvfs);
4490 		return (error);
4491 	}
4492 
4493 	error = zfs_link_create(dl, szp, tx, 0);
4494 
4495 	if (error == 0) {
4496 		uint64_t txtype = TX_LINK;
4497 		if (flags & FIGNORECASE)
4498 			txtype |= TX_CI;
4499 		zfs_log_link(zilog, tx, txtype, dzp, szp, name);
4500 	}
4501 
4502 	dmu_tx_commit(tx);
4503 
4504 	zfs_dirent_unlock(dl);
4505 
4506 	if (error == 0) {
4507 		vnevent_link(svp, ct);
4508 	}
4509 
4510 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4511 		zil_commit(zilog, 0);
4512 
4513 	ZFS_EXIT(zfsvfs);
4514 	return (error);
4515 }
4516 
4517 #ifdef illumos
4518 /*
4519  * zfs_null_putapage() is used when the file system has been force
4520  * unmounted. It just drops the pages.
4521  */
4522 /* ARGSUSED */
4523 static int
zfs_null_putapage(vnode_t * vp,page_t * pp,u_offset_t * offp,size_t * lenp,int flags,cred_t * cr)4524 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4525     size_t *lenp, int flags, cred_t *cr)
4526 {
4527 	pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR);
4528 	return (0);
4529 }
4530 
4531 /*
4532  * Push a page out to disk, klustering if possible.
4533  *
4534  *	IN:	vp	- file to push page to.
4535  *		pp	- page to push.
4536  *		flags	- additional flags.
4537  *		cr	- credentials of caller.
4538  *
4539  *	OUT:	offp	- start of range pushed.
4540  *		lenp	- len of range pushed.
4541  *
4542  *	RETURN:	0 on success, error code on failure.
4543  *
4544  * NOTE: callers must have locked the page to be pushed.  On
4545  * exit, the page (and all other pages in the kluster) must be
4546  * unlocked.
4547  */
4548 /* ARGSUSED */
4549 static int
zfs_putapage(vnode_t * vp,page_t * pp,u_offset_t * offp,size_t * lenp,int flags,cred_t * cr)4550 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4551     size_t *lenp, int flags, cred_t *cr)
4552 {
4553 	znode_t		*zp = VTOZ(vp);
4554 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
4555 	dmu_tx_t	*tx;
4556 	u_offset_t	off, koff;
4557 	size_t		len, klen;
4558 	int		err;
4559 
4560 	off = pp->p_offset;
4561 	len = PAGESIZE;
4562 	/*
4563 	 * If our blocksize is bigger than the page size, try to kluster
4564 	 * multiple pages so that we write a full block (thus avoiding
4565 	 * a read-modify-write).
4566 	 */
4567 	if (off < zp->z_size && zp->z_blksz > PAGESIZE) {
4568 		klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
4569 		koff = ISP2(klen) ? P2ALIGN(off, (u_offset_t)klen) : 0;
4570 		ASSERT(koff <= zp->z_size);
4571 		if (koff + klen > zp->z_size)
4572 			klen = P2ROUNDUP(zp->z_size - koff, (uint64_t)PAGESIZE);
4573 		pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags);
4574 	}
4575 	ASSERT3U(btop(len), ==, btopr(len));
4576 
4577 	/*
4578 	 * Can't push pages past end-of-file.
4579 	 */
4580 	if (off >= zp->z_size) {
4581 		/* ignore all pages */
4582 		err = 0;
4583 		goto out;
4584 	} else if (off + len > zp->z_size) {
4585 		int npages = btopr(zp->z_size - off);
4586 		page_t *trunc;
4587 
4588 		page_list_break(&pp, &trunc, npages);
4589 		/* ignore pages past end of file */
4590 		if (trunc)
4591 			pvn_write_done(trunc, flags);
4592 		len = zp->z_size - off;
4593 	}
4594 
4595 	if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
4596 	    zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
4597 		err = SET_ERROR(EDQUOT);
4598 		goto out;
4599 	}
4600 	tx = dmu_tx_create(zfsvfs->z_os);
4601 	dmu_tx_hold_write(tx, zp->z_id, off, len);
4602 
4603 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4604 	zfs_sa_upgrade_txholds(tx, zp);
4605 	err = dmu_tx_assign(tx, TXG_WAIT);
4606 	if (err != 0) {
4607 		dmu_tx_abort(tx);
4608 		goto out;
4609 	}
4610 
4611 	if (zp->z_blksz <= PAGESIZE) {
4612 		caddr_t va = zfs_map_page(pp, S_READ);
4613 		ASSERT3U(len, <=, PAGESIZE);
4614 		dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx);
4615 		zfs_unmap_page(pp, va);
4616 	} else {
4617 		err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx);
4618 	}
4619 
4620 	if (err == 0) {
4621 		uint64_t mtime[2], ctime[2];
4622 		sa_bulk_attr_t bulk[3];
4623 		int count = 0;
4624 
4625 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
4626 		    &mtime, 16);
4627 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
4628 		    &ctime, 16);
4629 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
4630 		    &zp->z_pflags, 8);
4631 		zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
4632 		    B_TRUE);
4633 		zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0);
4634 	}
4635 	dmu_tx_commit(tx);
4636 
4637 out:
4638 	pvn_write_done(pp, (err ? B_ERROR : 0) | flags);
4639 	if (offp)
4640 		*offp = off;
4641 	if (lenp)
4642 		*lenp = len;
4643 
4644 	return (err);
4645 }
4646 
4647 /*
4648  * Copy the portion of the file indicated from pages into the file.
4649  * The pages are stored in a page list attached to the files vnode.
4650  *
4651  *	IN:	vp	- vnode of file to push page data to.
4652  *		off	- position in file to put data.
4653  *		len	- amount of data to write.
4654  *		flags	- flags to control the operation.
4655  *		cr	- credentials of caller.
4656  *		ct	- caller context.
4657  *
4658  *	RETURN:	0 on success, error code on failure.
4659  *
4660  * Timestamps:
4661  *	vp - ctime|mtime updated
4662  */
4663 /*ARGSUSED*/
4664 static int
zfs_putpage(vnode_t * vp,offset_t off,size_t len,int flags,cred_t * cr,caller_context_t * ct)4665 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr,
4666     caller_context_t *ct)
4667 {
4668 	znode_t		*zp = VTOZ(vp);
4669 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
4670 	page_t		*pp;
4671 	size_t		io_len;
4672 	u_offset_t	io_off;
4673 	uint_t		blksz;
4674 	rl_t		*rl;
4675 	int		error = 0;
4676 
4677 	ZFS_ENTER(zfsvfs);
4678 	ZFS_VERIFY_ZP(zp);
4679 
4680 	/*
4681 	 * Align this request to the file block size in case we kluster.
4682 	 * XXX - this can result in pretty aggresive locking, which can
4683 	 * impact simultanious read/write access.  One option might be
4684 	 * to break up long requests (len == 0) into block-by-block
4685 	 * operations to get narrower locking.
4686 	 */
4687 	blksz = zp->z_blksz;
4688 	if (ISP2(blksz))
4689 		io_off = P2ALIGN_TYPED(off, blksz, u_offset_t);
4690 	else
4691 		io_off = 0;
4692 	if (len > 0 && ISP2(blksz))
4693 		io_len = P2ROUNDUP_TYPED(len + (off - io_off), blksz, size_t);
4694 	else
4695 		io_len = 0;
4696 
4697 	if (io_len == 0) {
4698 		/*
4699 		 * Search the entire vp list for pages >= io_off.
4700 		 */
4701 		rl = zfs_range_lock(zp, io_off, UINT64_MAX, RL_WRITER);
4702 		error = pvn_vplist_dirty(vp, io_off, zfs_putapage, flags, cr);
4703 		goto out;
4704 	}
4705 	rl = zfs_range_lock(zp, io_off, io_len, RL_WRITER);
4706 
4707 	if (off > zp->z_size) {
4708 		/* past end of file */
4709 		zfs_range_unlock(rl);
4710 		ZFS_EXIT(zfsvfs);
4711 		return (0);
4712 	}
4713 
4714 	len = MIN(io_len, P2ROUNDUP(zp->z_size, PAGESIZE) - io_off);
4715 
4716 	for (off = io_off; io_off < off + len; io_off += io_len) {
4717 		if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
4718 			pp = page_lookup(vp, io_off,
4719 			    (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED);
4720 		} else {
4721 			pp = page_lookup_nowait(vp, io_off,
4722 			    (flags & B_FREE) ? SE_EXCL : SE_SHARED);
4723 		}
4724 
4725 		if (pp != NULL && pvn_getdirty(pp, flags)) {
4726 			int err;
4727 
4728 			/*
4729 			 * Found a dirty page to push
4730 			 */
4731 			err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr);
4732 			if (err)
4733 				error = err;
4734 		} else {
4735 			io_len = PAGESIZE;
4736 		}
4737 	}
4738 out:
4739 	zfs_range_unlock(rl);
4740 	if ((flags & B_ASYNC) == 0 || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4741 		zil_commit(zfsvfs->z_log, zp->z_id);
4742 	ZFS_EXIT(zfsvfs);
4743 	return (error);
4744 }
4745 #endif	/* illumos */
4746 
4747 /*ARGSUSED*/
4748 void
zfs_inactive(vnode_t * vp,cred_t * cr,caller_context_t * ct)4749 zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
4750 {
4751 	znode_t	*zp = VTOZ(vp);
4752 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4753 	int error;
4754 
4755 	rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
4756 	if (zp->z_sa_hdl == NULL) {
4757 		/*
4758 		 * The fs has been unmounted, or we did a
4759 		 * suspend/resume and this file no longer exists.
4760 		 */
4761 		rw_exit(&zfsvfs->z_teardown_inactive_lock);
4762 		vrecycle(vp);
4763 		return;
4764 	}
4765 
4766 	mutex_enter(&zp->z_lock);
4767 	if (zp->z_unlinked) {
4768 		/*
4769 		 * Fast path to recycle a vnode of a removed file.
4770 		 */
4771 		mutex_exit(&zp->z_lock);
4772 		rw_exit(&zfsvfs->z_teardown_inactive_lock);
4773 		vrecycle(vp);
4774 		return;
4775 	}
4776 	mutex_exit(&zp->z_lock);
4777 
4778 	if (zp->z_atime_dirty && zp->z_unlinked == 0) {
4779 		dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
4780 
4781 		dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4782 		zfs_sa_upgrade_txholds(tx, zp);
4783 		error = dmu_tx_assign(tx, TXG_WAIT);
4784 		if (error) {
4785 			dmu_tx_abort(tx);
4786 		} else {
4787 			mutex_enter(&zp->z_lock);
4788 			(void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs),
4789 			    (void *)&zp->z_atime, sizeof (zp->z_atime), tx);
4790 			zp->z_atime_dirty = 0;
4791 			mutex_exit(&zp->z_lock);
4792 			dmu_tx_commit(tx);
4793 		}
4794 	}
4795 	rw_exit(&zfsvfs->z_teardown_inactive_lock);
4796 }
4797 
4798 #ifdef illumos
4799 /*
4800  * Bounds-check the seek operation.
4801  *
4802  *	IN:	vp	- vnode seeking within
4803  *		ooff	- old file offset
4804  *		noffp	- pointer to new file offset
4805  *		ct	- caller context
4806  *
4807  *	RETURN:	0 on success, EINVAL if new offset invalid.
4808  */
4809 /* ARGSUSED */
4810 static int
zfs_seek(vnode_t * vp,offset_t ooff,offset_t * noffp,caller_context_t * ct)4811 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp,
4812     caller_context_t *ct)
4813 {
4814 	if (vp->v_type == VDIR)
4815 		return (0);
4816 	return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4817 }
4818 
4819 /*
4820  * Pre-filter the generic locking function to trap attempts to place
4821  * a mandatory lock on a memory mapped file.
4822  */
4823 static int
zfs_frlock(vnode_t * vp,int cmd,flock64_t * bfp,int flag,offset_t offset,flk_callback_t * flk_cbp,cred_t * cr,caller_context_t * ct)4824 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset,
4825     flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct)
4826 {
4827 	znode_t *zp = VTOZ(vp);
4828 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4829 
4830 	ZFS_ENTER(zfsvfs);
4831 	ZFS_VERIFY_ZP(zp);
4832 
4833 	/*
4834 	 * We are following the UFS semantics with respect to mapcnt
4835 	 * here: If we see that the file is mapped already, then we will
4836 	 * return an error, but we don't worry about races between this
4837 	 * function and zfs_map().
4838 	 */
4839 	if (zp->z_mapcnt > 0 && MANDMODE(zp->z_mode)) {
4840 		ZFS_EXIT(zfsvfs);
4841 		return (SET_ERROR(EAGAIN));
4842 	}
4843 	ZFS_EXIT(zfsvfs);
4844 	return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct));
4845 }
4846 
4847 /*
4848  * If we can't find a page in the cache, we will create a new page
4849  * and fill it with file data.  For efficiency, we may try to fill
4850  * multiple pages at once (klustering) to fill up the supplied page
4851  * list.  Note that the pages to be filled are held with an exclusive
4852  * lock to prevent access by other threads while they are being filled.
4853  */
4854 static int
zfs_fillpage(vnode_t * vp,u_offset_t off,struct seg * seg,caddr_t addr,page_t * pl[],size_t plsz,enum seg_rw rw)4855 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg,
4856     caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw)
4857 {
4858 	znode_t *zp = VTOZ(vp);
4859 	page_t *pp, *cur_pp;
4860 	objset_t *os = zp->z_zfsvfs->z_os;
4861 	u_offset_t io_off, total;
4862 	size_t io_len;
4863 	int err;
4864 
4865 	if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) {
4866 		/*
4867 		 * We only have a single page, don't bother klustering
4868 		 */
4869 		io_off = off;
4870 		io_len = PAGESIZE;
4871 		pp = page_create_va(vp, io_off, io_len,
4872 		    PG_EXCL | PG_WAIT, seg, addr);
4873 	} else {
4874 		/*
4875 		 * Try to find enough pages to fill the page list
4876 		 */
4877 		pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
4878 		    &io_len, off, plsz, 0);
4879 	}
4880 	if (pp == NULL) {
4881 		/*
4882 		 * The page already exists, nothing to do here.
4883 		 */
4884 		*pl = NULL;
4885 		return (0);
4886 	}
4887 
4888 	/*
4889 	 * Fill the pages in the kluster.
4890 	 */
4891 	cur_pp = pp;
4892 	for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
4893 		caddr_t va;
4894 
4895 		ASSERT3U(io_off, ==, cur_pp->p_offset);
4896 		va = zfs_map_page(cur_pp, S_WRITE);
4897 		err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va,
4898 		    DMU_READ_PREFETCH);
4899 		zfs_unmap_page(cur_pp, va);
4900 		if (err) {
4901 			/* On error, toss the entire kluster */
4902 			pvn_read_done(pp, B_ERROR);
4903 			/* convert checksum errors into IO errors */
4904 			if (err == ECKSUM)
4905 				err = SET_ERROR(EIO);
4906 			return (err);
4907 		}
4908 		cur_pp = cur_pp->p_next;
4909 	}
4910 
4911 	/*
4912 	 * Fill in the page list array from the kluster starting
4913 	 * from the desired offset `off'.
4914 	 * NOTE: the page list will always be null terminated.
4915 	 */
4916 	pvn_plist_init(pp, pl, plsz, off, io_len, rw);
4917 	ASSERT(pl == NULL || (*pl)->p_offset == off);
4918 
4919 	return (0);
4920 }
4921 
4922 /*
4923  * Return pointers to the pages for the file region [off, off + len]
4924  * in the pl array.  If plsz is greater than len, this function may
4925  * also return page pointers from after the specified region
4926  * (i.e. the region [off, off + plsz]).  These additional pages are
4927  * only returned if they are already in the cache, or were created as
4928  * part of a klustered read.
4929  *
4930  *	IN:	vp	- vnode of file to get data from.
4931  *		off	- position in file to get data from.
4932  *		len	- amount of data to retrieve.
4933  *		plsz	- length of provided page list.
4934  *		seg	- segment to obtain pages for.
4935  *		addr	- virtual address of fault.
4936  *		rw	- mode of created pages.
4937  *		cr	- credentials of caller.
4938  *		ct	- caller context.
4939  *
4940  *	OUT:	protp	- protection mode of created pages.
4941  *		pl	- list of pages created.
4942  *
4943  *	RETURN:	0 on success, error code on failure.
4944  *
4945  * Timestamps:
4946  *	vp - atime updated
4947  */
4948 /* ARGSUSED */
4949 static int
zfs_getpage(vnode_t * vp,offset_t off,size_t len,uint_t * protp,page_t * pl[],size_t plsz,struct seg * seg,caddr_t addr,enum seg_rw rw,cred_t * cr,caller_context_t * ct)4950 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
4951     page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
4952     enum seg_rw rw, cred_t *cr, caller_context_t *ct)
4953 {
4954 	znode_t		*zp = VTOZ(vp);
4955 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
4956 	page_t		**pl0 = pl;
4957 	int		err = 0;
4958 
4959 	/* we do our own caching, faultahead is unnecessary */
4960 	if (pl == NULL)
4961 		return (0);
4962 	else if (len > plsz)
4963 		len = plsz;
4964 	else
4965 		len = P2ROUNDUP(len, PAGESIZE);
4966 	ASSERT(plsz >= len);
4967 
4968 	ZFS_ENTER(zfsvfs);
4969 	ZFS_VERIFY_ZP(zp);
4970 
4971 	if (protp)
4972 		*protp = PROT_ALL;
4973 
4974 	/*
4975 	 * Loop through the requested range [off, off + len) looking
4976 	 * for pages.  If we don't find a page, we will need to create
4977 	 * a new page and fill it with data from the file.
4978 	 */
4979 	while (len > 0) {
4980 		if (*pl = page_lookup(vp, off, SE_SHARED))
4981 			*(pl+1) = NULL;
4982 		else if (err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw))
4983 			goto out;
4984 		while (*pl) {
4985 			ASSERT3U((*pl)->p_offset, ==, off);
4986 			off += PAGESIZE;
4987 			addr += PAGESIZE;
4988 			if (len > 0) {
4989 				ASSERT3U(len, >=, PAGESIZE);
4990 				len -= PAGESIZE;
4991 			}
4992 			ASSERT3U(plsz, >=, PAGESIZE);
4993 			plsz -= PAGESIZE;
4994 			pl++;
4995 		}
4996 	}
4997 
4998 	/*
4999 	 * Fill out the page array with any pages already in the cache.
5000 	 */
5001 	while (plsz > 0 &&
5002 	    (*pl++ = page_lookup_nowait(vp, off, SE_SHARED))) {
5003 			off += PAGESIZE;
5004 			plsz -= PAGESIZE;
5005 	}
5006 out:
5007 	if (err) {
5008 		/*
5009 		 * Release any pages we have previously locked.
5010 		 */
5011 		while (pl > pl0)
5012 			page_unlock(*--pl);
5013 	} else {
5014 		ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
5015 	}
5016 
5017 	*pl = NULL;
5018 
5019 	ZFS_EXIT(zfsvfs);
5020 	return (err);
5021 }
5022 
5023 /*
5024  * Request a memory map for a section of a file.  This code interacts
5025  * with common code and the VM system as follows:
5026  *
5027  * - common code calls mmap(), which ends up in smmap_common()
5028  * - this calls VOP_MAP(), which takes you into (say) zfs
5029  * - zfs_map() calls as_map(), passing segvn_create() as the callback
5030  * - segvn_create() creates the new segment and calls VOP_ADDMAP()
5031  * - zfs_addmap() updates z_mapcnt
5032  */
5033 /*ARGSUSED*/
5034 static int
zfs_map(vnode_t * vp,offset_t off,struct as * as,caddr_t * addrp,size_t len,uchar_t prot,uchar_t maxprot,uint_t flags,cred_t * cr,caller_context_t * ct)5035 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
5036     size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
5037     caller_context_t *ct)
5038 {
5039 	znode_t *zp = VTOZ(vp);
5040 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5041 	segvn_crargs_t	vn_a;
5042 	int		error;
5043 
5044 	ZFS_ENTER(zfsvfs);
5045 	ZFS_VERIFY_ZP(zp);
5046 
5047 	if ((prot & PROT_WRITE) && (zp->z_pflags &
5048 	    (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) {
5049 		ZFS_EXIT(zfsvfs);
5050 		return (SET_ERROR(EPERM));
5051 	}
5052 
5053 	if ((prot & (PROT_READ | PROT_EXEC)) &&
5054 	    (zp->z_pflags & ZFS_AV_QUARANTINED)) {
5055 		ZFS_EXIT(zfsvfs);
5056 		return (SET_ERROR(EACCES));
5057 	}
5058 
5059 	if (vp->v_flag & VNOMAP) {
5060 		ZFS_EXIT(zfsvfs);
5061 		return (SET_ERROR(ENOSYS));
5062 	}
5063 
5064 	if (off < 0 || len > MAXOFFSET_T - off) {
5065 		ZFS_EXIT(zfsvfs);
5066 		return (SET_ERROR(ENXIO));
5067 	}
5068 
5069 	if (vp->v_type != VREG) {
5070 		ZFS_EXIT(zfsvfs);
5071 		return (SET_ERROR(ENODEV));
5072 	}
5073 
5074 	/*
5075 	 * If file is locked, disallow mapping.
5076 	 */
5077 	if (MANDMODE(zp->z_mode) && vn_has_flocks(vp)) {
5078 		ZFS_EXIT(zfsvfs);
5079 		return (SET_ERROR(EAGAIN));
5080 	}
5081 
5082 	as_rangelock(as);
5083 	error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
5084 	if (error != 0) {
5085 		as_rangeunlock(as);
5086 		ZFS_EXIT(zfsvfs);
5087 		return (error);
5088 	}
5089 
5090 	vn_a.vp = vp;
5091 	vn_a.offset = (u_offset_t)off;
5092 	vn_a.type = flags & MAP_TYPE;
5093 	vn_a.prot = prot;
5094 	vn_a.maxprot = maxprot;
5095 	vn_a.cred = cr;
5096 	vn_a.amp = NULL;
5097 	vn_a.flags = flags & ~MAP_TYPE;
5098 	vn_a.szc = 0;
5099 	vn_a.lgrp_mem_policy_flags = 0;
5100 
5101 	error = as_map(as, *addrp, len, segvn_create, &vn_a);
5102 
5103 	as_rangeunlock(as);
5104 	ZFS_EXIT(zfsvfs);
5105 	return (error);
5106 }
5107 
5108 /* ARGSUSED */
5109 static int
zfs_addmap(vnode_t * vp,offset_t off,struct as * as,caddr_t addr,size_t len,uchar_t prot,uchar_t maxprot,uint_t flags,cred_t * cr,caller_context_t * ct)5110 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
5111     size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
5112     caller_context_t *ct)
5113 {
5114 	uint64_t pages = btopr(len);
5115 
5116 	atomic_add_64(&VTOZ(vp)->z_mapcnt, pages);
5117 	return (0);
5118 }
5119 
5120 /*
5121  * The reason we push dirty pages as part of zfs_delmap() is so that we get a
5122  * more accurate mtime for the associated file.  Since we don't have a way of
5123  * detecting when the data was actually modified, we have to resort to
5124  * heuristics.  If an explicit msync() is done, then we mark the mtime when the
5125  * last page is pushed.  The problem occurs when the msync() call is omitted,
5126  * which by far the most common case:
5127  *
5128  *	open()
5129  *	mmap()
5130  *	<modify memory>
5131  *	munmap()
5132  *	close()
5133  *	<time lapse>
5134  *	putpage() via fsflush
5135  *
5136  * If we wait until fsflush to come along, we can have a modification time that
5137  * is some arbitrary point in the future.  In order to prevent this in the
5138  * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is
5139  * torn down.
5140  */
5141 /* ARGSUSED */
5142 static int
zfs_delmap(vnode_t * vp,offset_t off,struct as * as,caddr_t addr,size_t len,uint_t prot,uint_t maxprot,uint_t flags,cred_t * cr,caller_context_t * ct)5143 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
5144     size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr,
5145     caller_context_t *ct)
5146 {
5147 	uint64_t pages = btopr(len);
5148 
5149 	ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages);
5150 	atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages);
5151 
5152 	if ((flags & MAP_SHARED) && (prot & PROT_WRITE) &&
5153 	    vn_has_cached_data(vp))
5154 		(void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct);
5155 
5156 	return (0);
5157 }
5158 
5159 /*
5160  * Free or allocate space in a file.  Currently, this function only
5161  * supports the `F_FREESP' command.  However, this command is somewhat
5162  * misnamed, as its functionality includes the ability to allocate as
5163  * well as free space.
5164  *
5165  *	IN:	vp	- vnode of file to free data in.
5166  *		cmd	- action to take (only F_FREESP supported).
5167  *		bfp	- section of file to free/alloc.
5168  *		flag	- current file open mode flags.
5169  *		offset	- current file offset.
5170  *		cr	- credentials of caller [UNUSED].
5171  *		ct	- caller context.
5172  *
5173  *	RETURN:	0 on success, error code on failure.
5174  *
5175  * Timestamps:
5176  *	vp - ctime|mtime updated
5177  */
5178 /* ARGSUSED */
5179 static int
zfs_space(vnode_t * vp,int cmd,flock64_t * bfp,int flag,offset_t offset,cred_t * cr,caller_context_t * ct)5180 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag,
5181     offset_t offset, cred_t *cr, caller_context_t *ct)
5182 {
5183 	znode_t		*zp = VTOZ(vp);
5184 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
5185 	uint64_t	off, len;
5186 	int		error;
5187 
5188 	ZFS_ENTER(zfsvfs);
5189 	ZFS_VERIFY_ZP(zp);
5190 
5191 	if (cmd != F_FREESP) {
5192 		ZFS_EXIT(zfsvfs);
5193 		return (SET_ERROR(EINVAL));
5194 	}
5195 
5196 	/*
5197 	 * In a case vp->v_vfsp != zp->z_zfsvfs->z_vfs (e.g. snapshots) our
5198 	 * callers might not be able to detect properly that we are read-only,
5199 	 * so check it explicitly here.
5200 	 */
5201 	if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
5202 		ZFS_EXIT(zfsvfs);
5203 		return (SET_ERROR(EROFS));
5204 	}
5205 
5206 	if (error = convoff(vp, bfp, 0, offset)) {
5207 		ZFS_EXIT(zfsvfs);
5208 		return (error);
5209 	}
5210 
5211 	if (bfp->l_len < 0) {
5212 		ZFS_EXIT(zfsvfs);
5213 		return (SET_ERROR(EINVAL));
5214 	}
5215 
5216 	off = bfp->l_start;
5217 	len = bfp->l_len; /* 0 means from off to end of file */
5218 
5219 	error = zfs_freesp(zp, off, len, flag, TRUE);
5220 
5221 	ZFS_EXIT(zfsvfs);
5222 	return (error);
5223 }
5224 #endif	/* illumos */
5225 
5226 CTASSERT(sizeof(struct zfid_short) <= sizeof(struct fid));
5227 CTASSERT(sizeof(struct zfid_long) <= sizeof(struct fid));
5228 
5229 /*ARGSUSED*/
5230 static int
zfs_fid(vnode_t * vp,fid_t * fidp,caller_context_t * ct)5231 zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
5232 {
5233 	znode_t		*zp = VTOZ(vp);
5234 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
5235 	uint32_t	gen;
5236 	uint64_t	gen64;
5237 	uint64_t	object = zp->z_id;
5238 	zfid_short_t	*zfid;
5239 	int		size, i, error;
5240 
5241 	ZFS_ENTER(zfsvfs);
5242 	ZFS_VERIFY_ZP(zp);
5243 
5244 	if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs),
5245 	    &gen64, sizeof (uint64_t))) != 0) {
5246 		ZFS_EXIT(zfsvfs);
5247 		return (error);
5248 	}
5249 
5250 	gen = (uint32_t)gen64;
5251 
5252 	size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN;
5253 
5254 #ifdef illumos
5255 	if (fidp->fid_len < size) {
5256 		fidp->fid_len = size;
5257 		ZFS_EXIT(zfsvfs);
5258 		return (SET_ERROR(ENOSPC));
5259 	}
5260 #else
5261 	fidp->fid_len = size;
5262 #endif
5263 
5264 	zfid = (zfid_short_t *)fidp;
5265 
5266 	zfid->zf_len = size;
5267 
5268 	for (i = 0; i < sizeof (zfid->zf_object); i++)
5269 		zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
5270 
5271 	/* Must have a non-zero generation number to distinguish from .zfs */
5272 	if (gen == 0)
5273 		gen = 1;
5274 	for (i = 0; i < sizeof (zfid->zf_gen); i++)
5275 		zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
5276 
5277 	if (size == LONG_FID_LEN) {
5278 		uint64_t	objsetid = dmu_objset_id(zfsvfs->z_os);
5279 		zfid_long_t	*zlfid;
5280 
5281 		zlfid = (zfid_long_t *)fidp;
5282 
5283 		for (i = 0; i < sizeof (zlfid->zf_setid); i++)
5284 			zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
5285 
5286 		/* XXX - this should be the generation number for the objset */
5287 		for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
5288 			zlfid->zf_setgen[i] = 0;
5289 	}
5290 
5291 	ZFS_EXIT(zfsvfs);
5292 	return (0);
5293 }
5294 
5295 static int
zfs_pathconf(vnode_t * vp,int cmd,ulong_t * valp,cred_t * cr,caller_context_t * ct)5296 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
5297     caller_context_t *ct)
5298 {
5299 	znode_t		*zp, *xzp;
5300 	zfsvfs_t	*zfsvfs;
5301 	zfs_dirlock_t	*dl;
5302 	int		error;
5303 
5304 	switch (cmd) {
5305 	case _PC_LINK_MAX:
5306 		*valp = INT_MAX;
5307 		return (0);
5308 
5309 	case _PC_FILESIZEBITS:
5310 		*valp = 64;
5311 		return (0);
5312 #ifdef illumos
5313 	case _PC_XATTR_EXISTS:
5314 		zp = VTOZ(vp);
5315 		zfsvfs = zp->z_zfsvfs;
5316 		ZFS_ENTER(zfsvfs);
5317 		ZFS_VERIFY_ZP(zp);
5318 		*valp = 0;
5319 		error = zfs_dirent_lock(&dl, zp, "", &xzp,
5320 		    ZXATTR | ZEXISTS | ZSHARED, NULL, NULL);
5321 		if (error == 0) {
5322 			zfs_dirent_unlock(dl);
5323 			if (!zfs_dirempty(xzp))
5324 				*valp = 1;
5325 			VN_RELE(ZTOV(xzp));
5326 		} else if (error == ENOENT) {
5327 			/*
5328 			 * If there aren't extended attributes, it's the
5329 			 * same as having zero of them.
5330 			 */
5331 			error = 0;
5332 		}
5333 		ZFS_EXIT(zfsvfs);
5334 		return (error);
5335 
5336 	case _PC_SATTR_ENABLED:
5337 	case _PC_SATTR_EXISTS:
5338 		*valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
5339 		    (vp->v_type == VREG || vp->v_type == VDIR);
5340 		return (0);
5341 
5342 	case _PC_ACCESS_FILTERING:
5343 		*valp = vfs_has_feature(vp->v_vfsp, VFSFT_ACCESS_FILTER) &&
5344 		    vp->v_type == VDIR;
5345 		return (0);
5346 
5347 	case _PC_ACL_ENABLED:
5348 		*valp = _ACL_ACE_ENABLED;
5349 		return (0);
5350 #endif	/* illumos */
5351 	case _PC_MIN_HOLE_SIZE:
5352 		*valp = (int)SPA_MINBLOCKSIZE;
5353 		return (0);
5354 #ifdef illumos
5355 	case _PC_TIMESTAMP_RESOLUTION:
5356 		/* nanosecond timestamp resolution */
5357 		*valp = 1L;
5358 		return (0);
5359 #endif
5360 	case _PC_ACL_EXTENDED:
5361 		*valp = 0;
5362 		return (0);
5363 
5364 	case _PC_ACL_NFS4:
5365 		*valp = 1;
5366 		return (0);
5367 
5368 	case _PC_ACL_PATH_MAX:
5369 		*valp = ACL_MAX_ENTRIES;
5370 		return (0);
5371 
5372 	default:
5373 		return (EOPNOTSUPP);
5374 	}
5375 }
5376 
5377 /*ARGSUSED*/
5378 static int
zfs_getsecattr(vnode_t * vp,vsecattr_t * vsecp,int flag,cred_t * cr,caller_context_t * ct)5379 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
5380     caller_context_t *ct)
5381 {
5382 	znode_t *zp = VTOZ(vp);
5383 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5384 	int error;
5385 	boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5386 
5387 	ZFS_ENTER(zfsvfs);
5388 	ZFS_VERIFY_ZP(zp);
5389 	error = zfs_getacl(zp, vsecp, skipaclchk, cr);
5390 	ZFS_EXIT(zfsvfs);
5391 
5392 	return (error);
5393 }
5394 
5395 /*ARGSUSED*/
5396 int
zfs_setsecattr(vnode_t * vp,vsecattr_t * vsecp,int flag,cred_t * cr,caller_context_t * ct)5397 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
5398     caller_context_t *ct)
5399 {
5400 	znode_t *zp = VTOZ(vp);
5401 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5402 	int error;
5403 	boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5404 	zilog_t	*zilog = zfsvfs->z_log;
5405 
5406 	ZFS_ENTER(zfsvfs);
5407 	ZFS_VERIFY_ZP(zp);
5408 
5409 	error = zfs_setacl(zp, vsecp, skipaclchk, cr);
5410 
5411 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
5412 		zil_commit(zilog, 0);
5413 
5414 	ZFS_EXIT(zfsvfs);
5415 	return (error);
5416 }
5417 
5418 #ifdef illumos
5419 /*
5420  * The smallest read we may consider to loan out an arcbuf.
5421  * This must be a power of 2.
5422  */
5423 int zcr_blksz_min = (1 << 10);	/* 1K */
5424 /*
5425  * If set to less than the file block size, allow loaning out of an
5426  * arcbuf for a partial block read.  This must be a power of 2.
5427  */
5428 int zcr_blksz_max = (1 << 17);	/* 128K */
5429 
5430 /*ARGSUSED*/
5431 static int
zfs_reqzcbuf(vnode_t * vp,enum uio_rw ioflag,xuio_t * xuio,cred_t * cr,caller_context_t * ct)5432 zfs_reqzcbuf(vnode_t *vp, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr,
5433     caller_context_t *ct)
5434 {
5435 	znode_t	*zp = VTOZ(vp);
5436 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5437 	int max_blksz = zfsvfs->z_max_blksz;
5438 	uio_t *uio = &xuio->xu_uio;
5439 	ssize_t size = uio->uio_resid;
5440 	offset_t offset = uio->uio_loffset;
5441 	int blksz;
5442 	int fullblk, i;
5443 	arc_buf_t *abuf;
5444 	ssize_t maxsize;
5445 	int preamble, postamble;
5446 
5447 	if (xuio->xu_type != UIOTYPE_ZEROCOPY)
5448 		return (SET_ERROR(EINVAL));
5449 
5450 	ZFS_ENTER(zfsvfs);
5451 	ZFS_VERIFY_ZP(zp);
5452 	switch (ioflag) {
5453 	case UIO_WRITE:
5454 		/*
5455 		 * Loan out an arc_buf for write if write size is bigger than
5456 		 * max_blksz, and the file's block size is also max_blksz.
5457 		 */
5458 		blksz = max_blksz;
5459 		if (size < blksz || zp->z_blksz != blksz) {
5460 			ZFS_EXIT(zfsvfs);
5461 			return (SET_ERROR(EINVAL));
5462 		}
5463 		/*
5464 		 * Caller requests buffers for write before knowing where the
5465 		 * write offset might be (e.g. NFS TCP write).
5466 		 */
5467 		if (offset == -1) {
5468 			preamble = 0;
5469 		} else {
5470 			preamble = P2PHASE(offset, blksz);
5471 			if (preamble) {
5472 				preamble = blksz - preamble;
5473 				size -= preamble;
5474 			}
5475 		}
5476 
5477 		postamble = P2PHASE(size, blksz);
5478 		size -= postamble;
5479 
5480 		fullblk = size / blksz;
5481 		(void) dmu_xuio_init(xuio,
5482 		    (preamble != 0) + fullblk + (postamble != 0));
5483 		DTRACE_PROBE3(zfs_reqzcbuf_align, int, preamble,
5484 		    int, postamble, int,
5485 		    (preamble != 0) + fullblk + (postamble != 0));
5486 
5487 		/*
5488 		 * Have to fix iov base/len for partial buffers.  They
5489 		 * currently represent full arc_buf's.
5490 		 */
5491 		if (preamble) {
5492 			/* data begins in the middle of the arc_buf */
5493 			abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5494 			    blksz);
5495 			ASSERT(abuf);
5496 			(void) dmu_xuio_add(xuio, abuf,
5497 			    blksz - preamble, preamble);
5498 		}
5499 
5500 		for (i = 0; i < fullblk; i++) {
5501 			abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5502 			    blksz);
5503 			ASSERT(abuf);
5504 			(void) dmu_xuio_add(xuio, abuf, 0, blksz);
5505 		}
5506 
5507 		if (postamble) {
5508 			/* data ends in the middle of the arc_buf */
5509 			abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5510 			    blksz);
5511 			ASSERT(abuf);
5512 			(void) dmu_xuio_add(xuio, abuf, 0, postamble);
5513 		}
5514 		break;
5515 	case UIO_READ:
5516 		/*
5517 		 * Loan out an arc_buf for read if the read size is larger than
5518 		 * the current file block size.  Block alignment is not
5519 		 * considered.  Partial arc_buf will be loaned out for read.
5520 		 */
5521 		blksz = zp->z_blksz;
5522 		if (blksz < zcr_blksz_min)
5523 			blksz = zcr_blksz_min;
5524 		if (blksz > zcr_blksz_max)
5525 			blksz = zcr_blksz_max;
5526 		/* avoid potential complexity of dealing with it */
5527 		if (blksz > max_blksz) {
5528 			ZFS_EXIT(zfsvfs);
5529 			return (SET_ERROR(EINVAL));
5530 		}
5531 
5532 		maxsize = zp->z_size - uio->uio_loffset;
5533 		if (size > maxsize)
5534 			size = maxsize;
5535 
5536 		if (size < blksz || vn_has_cached_data(vp)) {
5537 			ZFS_EXIT(zfsvfs);
5538 			return (SET_ERROR(EINVAL));
5539 		}
5540 		break;
5541 	default:
5542 		ZFS_EXIT(zfsvfs);
5543 		return (SET_ERROR(EINVAL));
5544 	}
5545 
5546 	uio->uio_extflg = UIO_XUIO;
5547 	XUIO_XUZC_RW(xuio) = ioflag;
5548 	ZFS_EXIT(zfsvfs);
5549 	return (0);
5550 }
5551 
5552 /*ARGSUSED*/
5553 static int
zfs_retzcbuf(vnode_t * vp,xuio_t * xuio,cred_t * cr,caller_context_t * ct)5554 zfs_retzcbuf(vnode_t *vp, xuio_t *xuio, cred_t *cr, caller_context_t *ct)
5555 {
5556 	int i;
5557 	arc_buf_t *abuf;
5558 	int ioflag = XUIO_XUZC_RW(xuio);
5559 
5560 	ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY);
5561 
5562 	i = dmu_xuio_cnt(xuio);
5563 	while (i-- > 0) {
5564 		abuf = dmu_xuio_arcbuf(xuio, i);
5565 		/*
5566 		 * if abuf == NULL, it must be a write buffer
5567 		 * that has been returned in zfs_write().
5568 		 */
5569 		if (abuf)
5570 			dmu_return_arcbuf(abuf);
5571 		ASSERT(abuf || ioflag == UIO_WRITE);
5572 	}
5573 
5574 	dmu_xuio_fini(xuio);
5575 	return (0);
5576 }
5577 
5578 /*
5579  * Predeclare these here so that the compiler assumes that
5580  * this is an "old style" function declaration that does
5581  * not include arguments => we won't get type mismatch errors
5582  * in the initializations that follow.
5583  */
5584 static int zfs_inval();
5585 static int zfs_isdir();
5586 
5587 static int
zfs_inval()5588 zfs_inval()
5589 {
5590 	return (SET_ERROR(EINVAL));
5591 }
5592 
5593 static int
zfs_isdir()5594 zfs_isdir()
5595 {
5596 	return (SET_ERROR(EISDIR));
5597 }
5598 /*
5599  * Directory vnode operations template
5600  */
5601 vnodeops_t *zfs_dvnodeops;
5602 const fs_operation_def_t zfs_dvnodeops_template[] = {
5603 	VOPNAME_OPEN,		{ .vop_open = zfs_open },
5604 	VOPNAME_CLOSE,		{ .vop_close = zfs_close },
5605 	VOPNAME_READ,		{ .error = zfs_isdir },
5606 	VOPNAME_WRITE,		{ .error = zfs_isdir },
5607 	VOPNAME_IOCTL,		{ .vop_ioctl = zfs_ioctl },
5608 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
5609 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
5610 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
5611 	VOPNAME_LOOKUP,		{ .vop_lookup = zfs_lookup },
5612 	VOPNAME_CREATE,		{ .vop_create = zfs_create },
5613 	VOPNAME_REMOVE,		{ .vop_remove = zfs_remove },
5614 	VOPNAME_LINK,		{ .vop_link = zfs_link },
5615 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
5616 	VOPNAME_MKDIR,		{ .vop_mkdir = zfs_mkdir },
5617 	VOPNAME_RMDIR,		{ .vop_rmdir = zfs_rmdir },
5618 	VOPNAME_READDIR,	{ .vop_readdir = zfs_readdir },
5619 	VOPNAME_SYMLINK,	{ .vop_symlink = zfs_symlink },
5620 	VOPNAME_FSYNC,		{ .vop_fsync = zfs_fsync },
5621 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
5622 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
5623 	VOPNAME_SEEK,		{ .vop_seek = zfs_seek },
5624 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
5625 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
5626 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
5627 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
5628 	NULL,			NULL
5629 };
5630 
5631 /*
5632  * Regular file vnode operations template
5633  */
5634 vnodeops_t *zfs_fvnodeops;
5635 const fs_operation_def_t zfs_fvnodeops_template[] = {
5636 	VOPNAME_OPEN,		{ .vop_open = zfs_open },
5637 	VOPNAME_CLOSE,		{ .vop_close = zfs_close },
5638 	VOPNAME_READ,		{ .vop_read = zfs_read },
5639 	VOPNAME_WRITE,		{ .vop_write = zfs_write },
5640 	VOPNAME_IOCTL,		{ .vop_ioctl = zfs_ioctl },
5641 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
5642 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
5643 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
5644 	VOPNAME_LOOKUP,		{ .vop_lookup = zfs_lookup },
5645 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
5646 	VOPNAME_FSYNC,		{ .vop_fsync = zfs_fsync },
5647 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
5648 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
5649 	VOPNAME_SEEK,		{ .vop_seek = zfs_seek },
5650 	VOPNAME_FRLOCK,		{ .vop_frlock = zfs_frlock },
5651 	VOPNAME_SPACE,		{ .vop_space = zfs_space },
5652 	VOPNAME_GETPAGE,	{ .vop_getpage = zfs_getpage },
5653 	VOPNAME_PUTPAGE,	{ .vop_putpage = zfs_putpage },
5654 	VOPNAME_MAP,		{ .vop_map = zfs_map },
5655 	VOPNAME_ADDMAP,		{ .vop_addmap = zfs_addmap },
5656 	VOPNAME_DELMAP,		{ .vop_delmap = zfs_delmap },
5657 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
5658 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
5659 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
5660 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
5661 	VOPNAME_REQZCBUF,	{ .vop_reqzcbuf = zfs_reqzcbuf },
5662 	VOPNAME_RETZCBUF,	{ .vop_retzcbuf = zfs_retzcbuf },
5663 	NULL,			NULL
5664 };
5665 
5666 /*
5667  * Symbolic link vnode operations template
5668  */
5669 vnodeops_t *zfs_symvnodeops;
5670 const fs_operation_def_t zfs_symvnodeops_template[] = {
5671 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
5672 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
5673 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
5674 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
5675 	VOPNAME_READLINK,	{ .vop_readlink = zfs_readlink },
5676 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
5677 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
5678 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
5679 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
5680 	NULL,			NULL
5681 };
5682 
5683 /*
5684  * special share hidden files vnode operations template
5685  */
5686 vnodeops_t *zfs_sharevnodeops;
5687 const fs_operation_def_t zfs_sharevnodeops_template[] = {
5688 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
5689 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
5690 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
5691 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
5692 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
5693 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
5694 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
5695 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
5696 	NULL,			NULL
5697 };
5698 
5699 /*
5700  * Extended attribute directory vnode operations template
5701  *
5702  * This template is identical to the directory vnodes
5703  * operation template except for restricted operations:
5704  *	VOP_MKDIR()
5705  *	VOP_SYMLINK()
5706  *
5707  * Note that there are other restrictions embedded in:
5708  *	zfs_create()	- restrict type to VREG
5709  *	zfs_link()	- no links into/out of attribute space
5710  *	zfs_rename()	- no moves into/out of attribute space
5711  */
5712 vnodeops_t *zfs_xdvnodeops;
5713 const fs_operation_def_t zfs_xdvnodeops_template[] = {
5714 	VOPNAME_OPEN,		{ .vop_open = zfs_open },
5715 	VOPNAME_CLOSE,		{ .vop_close = zfs_close },
5716 	VOPNAME_IOCTL,		{ .vop_ioctl = zfs_ioctl },
5717 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
5718 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
5719 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
5720 	VOPNAME_LOOKUP,		{ .vop_lookup = zfs_lookup },
5721 	VOPNAME_CREATE,		{ .vop_create = zfs_create },
5722 	VOPNAME_REMOVE,		{ .vop_remove = zfs_remove },
5723 	VOPNAME_LINK,		{ .vop_link = zfs_link },
5724 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
5725 	VOPNAME_MKDIR,		{ .error = zfs_inval },
5726 	VOPNAME_RMDIR,		{ .vop_rmdir = zfs_rmdir },
5727 	VOPNAME_READDIR,	{ .vop_readdir = zfs_readdir },
5728 	VOPNAME_SYMLINK,	{ .error = zfs_inval },
5729 	VOPNAME_FSYNC,		{ .vop_fsync = zfs_fsync },
5730 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
5731 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
5732 	VOPNAME_SEEK,		{ .vop_seek = zfs_seek },
5733 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
5734 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
5735 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
5736 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
5737 	NULL,			NULL
5738 };
5739 
5740 /*
5741  * Error vnode operations template
5742  */
5743 vnodeops_t *zfs_evnodeops;
5744 const fs_operation_def_t zfs_evnodeops_template[] = {
5745 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
5746 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
5747 	NULL,			NULL
5748 };
5749 #endif	/* illumos */
5750 
5751 static int
ioflags(int ioflags)5752 ioflags(int ioflags)
5753 {
5754 	int flags = 0;
5755 
5756 	if (ioflags & IO_APPEND)
5757 		flags |= FAPPEND;
5758 	if (ioflags & IO_NDELAY)
5759         	flags |= FNONBLOCK;
5760 	if (ioflags & IO_SYNC)
5761 		flags |= (FSYNC | FDSYNC | FRSYNC);
5762 
5763 	return (flags);
5764 }
5765 
5766 static int
zfs_getpages(struct vnode * vp,vm_page_t * m,int count,int * rbehind,int * rahead)5767 zfs_getpages(struct vnode *vp, vm_page_t *m, int count, int *rbehind,
5768     int *rahead)
5769 {
5770 	znode_t *zp = VTOZ(vp);
5771 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5772 	objset_t *os = zp->z_zfsvfs->z_os;
5773 	vm_page_t mlast;
5774 	vm_object_t object;
5775 	caddr_t va;
5776 	struct sf_buf *sf;
5777 	off_t startoff, endoff;
5778 	int i, error;
5779 	vm_pindex_t reqstart, reqend;
5780 	int lsize, reqsize, size;
5781 
5782 	object = m[0]->object;
5783 	error = 0;
5784 
5785 	ZFS_ENTER(zfsvfs);
5786 	ZFS_VERIFY_ZP(zp);
5787 
5788 	zfs_vmobject_wlock(object);
5789 	if (m[count - 1]->valid != 0 && --count == 0) {
5790 		zfs_vmobject_wunlock(object);
5791 		goto out;
5792 	}
5793 
5794 	mlast = m[count - 1];
5795 
5796 	if (IDX_TO_OFF(mlast->pindex) >=
5797 	    object->un_pager.vnp.vnp_size) {
5798 		zfs_vmobject_wunlock(object);
5799 		ZFS_EXIT(zfsvfs);
5800 		return (zfs_vm_pagerret_bad);
5801 	}
5802 
5803 	PCPU_INC(cnt.v_vnodein);
5804 	PCPU_ADD(cnt.v_vnodepgsin, reqsize);
5805 
5806 	lsize = PAGE_SIZE;
5807 	if (IDX_TO_OFF(mlast->pindex) + lsize > object->un_pager.vnp.vnp_size)
5808 		lsize = object->un_pager.vnp.vnp_size -
5809 		    IDX_TO_OFF(mlast->pindex);
5810 	zfs_vmobject_wunlock(object);
5811 
5812 	for (i = 0; i < count; i++) {
5813 		size = PAGE_SIZE;
5814 		if (i == count - 1)
5815 			size = lsize;
5816 		va = zfs_map_page(m[i], &sf);
5817 		error = dmu_read(os, zp->z_id, IDX_TO_OFF(m[i]->pindex),
5818 		    size, va, DMU_READ_PREFETCH);
5819 		if (size != PAGE_SIZE)
5820 			bzero(va + size, PAGE_SIZE - size);
5821 		zfs_unmap_page(sf);
5822 		if (error != 0)
5823 			goto out;
5824 	}
5825 
5826 	zfs_vmobject_wlock(object);
5827 	for (i = 0; i < count; i++)
5828 		m[i]->valid = VM_PAGE_BITS_ALL;
5829 	zfs_vmobject_wunlock(object);
5830 
5831 out:
5832 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
5833 	ZFS_EXIT(zfsvfs);
5834 	if (error == 0) {
5835 		if (rbehind)
5836 			*rbehind = 0;
5837 		if (rahead)
5838 			*rahead = 0;
5839 		return (zfs_vm_pagerret_ok);
5840 	} else
5841 		return (zfs_vm_pagerret_error);
5842 }
5843 
5844 static int
zfs_freebsd_getpages(ap)5845 zfs_freebsd_getpages(ap)
5846 	struct vop_getpages_args /* {
5847 		struct vnode *a_vp;
5848 		vm_page_t *a_m;
5849 		int a_count;
5850 		int *a_rbehind;
5851 		int *a_rahead;
5852 	} */ *ap;
5853 {
5854 
5855 	return (zfs_getpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_rbehind,
5856 	    ap->a_rahead));
5857 }
5858 
5859 static int
zfs_putpages(struct vnode * vp,vm_page_t * ma,size_t len,int flags,int * rtvals)5860 zfs_putpages(struct vnode *vp, vm_page_t *ma, size_t len, int flags,
5861     int *rtvals)
5862 {
5863 	znode_t		*zp = VTOZ(vp);
5864 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
5865 	rl_t		*rl;
5866 	dmu_tx_t	*tx;
5867 	struct sf_buf	*sf;
5868 	vm_object_t	object;
5869 	vm_page_t	m;
5870 	caddr_t		va;
5871 	size_t		tocopy;
5872 	size_t		lo_len;
5873 	vm_ooffset_t	lo_off;
5874 	vm_ooffset_t	off;
5875 	uint_t		blksz;
5876 	int		ncount;
5877 	int		pcount;
5878 	int		err;
5879 	int		i;
5880 
5881 	ZFS_ENTER(zfsvfs);
5882 	ZFS_VERIFY_ZP(zp);
5883 
5884 	object = vp->v_object;
5885 	pcount = btoc(len);
5886 	ncount = pcount;
5887 
5888 	KASSERT(ma[0]->object == object, ("mismatching object"));
5889 	KASSERT(len > 0 && (len & PAGE_MASK) == 0, ("unexpected length"));
5890 
5891 	for (i = 0; i < pcount; i++)
5892 		rtvals[i] = zfs_vm_pagerret_error;
5893 
5894 	off = IDX_TO_OFF(ma[0]->pindex);
5895 	blksz = zp->z_blksz;
5896 	lo_off = rounddown(off, blksz);
5897 	lo_len = roundup(len + (off - lo_off), blksz);
5898 	rl = zfs_range_lock(zp, lo_off, lo_len, RL_WRITER);
5899 
5900 	zfs_vmobject_wlock(object);
5901 	if (len + off > object->un_pager.vnp.vnp_size) {
5902 		if (object->un_pager.vnp.vnp_size > off) {
5903 			int pgoff;
5904 
5905 			len = object->un_pager.vnp.vnp_size - off;
5906 			ncount = btoc(len);
5907 			if ((pgoff = (int)len & PAGE_MASK) != 0) {
5908 				/*
5909 				 * If the object is locked and the following
5910 				 * conditions hold, then the page's dirty
5911 				 * field cannot be concurrently changed by a
5912 				 * pmap operation.
5913 				 */
5914 				m = ma[ncount - 1];
5915 				vm_page_assert_sbusied(m);
5916 				KASSERT(!pmap_page_is_write_mapped(m),
5917 				    ("zfs_putpages: page %p is not read-only", m));
5918 				vm_page_clear_dirty(m, pgoff, PAGE_SIZE -
5919 				    pgoff);
5920 			}
5921 		} else {
5922 			len = 0;
5923 			ncount = 0;
5924 		}
5925 		if (ncount < pcount) {
5926 			for (i = ncount; i < pcount; i++) {
5927 				rtvals[i] = zfs_vm_pagerret_bad;
5928 			}
5929 		}
5930 	}
5931 	zfs_vmobject_wunlock(object);
5932 
5933 	if (ncount == 0)
5934 		goto out;
5935 
5936 	if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
5937 	    zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
5938 		goto out;
5939 	}
5940 
5941 top:
5942 	tx = dmu_tx_create(zfsvfs->z_os);
5943 	dmu_tx_hold_write(tx, zp->z_id, off, len);
5944 
5945 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
5946 	zfs_sa_upgrade_txholds(tx, zp);
5947 	err = dmu_tx_assign(tx, TXG_NOWAIT);
5948 	if (err != 0) {
5949 		if (err == ERESTART) {
5950 			dmu_tx_wait(tx);
5951 			dmu_tx_abort(tx);
5952 			goto top;
5953 		}
5954 		dmu_tx_abort(tx);
5955 		goto out;
5956 	}
5957 
5958 	if (zp->z_blksz < PAGE_SIZE) {
5959 		i = 0;
5960 		for (i = 0; len > 0; off += tocopy, len -= tocopy, i++) {
5961 			tocopy = len > PAGE_SIZE ? PAGE_SIZE : len;
5962 			va = zfs_map_page(ma[i], &sf);
5963 			dmu_write(zfsvfs->z_os, zp->z_id, off, tocopy, va, tx);
5964 			zfs_unmap_page(sf);
5965 		}
5966 	} else {
5967 		err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, ma, tx);
5968 	}
5969 
5970 	if (err == 0) {
5971 		uint64_t mtime[2], ctime[2];
5972 		sa_bulk_attr_t bulk[3];
5973 		int count = 0;
5974 
5975 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
5976 		    &mtime, 16);
5977 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
5978 		    &ctime, 16);
5979 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
5980 		    &zp->z_pflags, 8);
5981 		zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
5982 		    B_TRUE);
5983 		(void)sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
5984 		zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0);
5985 
5986 		zfs_vmobject_wlock(object);
5987 		for (i = 0; i < ncount; i++) {
5988 			rtvals[i] = zfs_vm_pagerret_ok;
5989 			vm_page_undirty(ma[i]);
5990 		}
5991 		zfs_vmobject_wunlock(object);
5992 		PCPU_INC(cnt.v_vnodeout);
5993 		PCPU_ADD(cnt.v_vnodepgsout, ncount);
5994 	}
5995 	dmu_tx_commit(tx);
5996 
5997 out:
5998 	zfs_range_unlock(rl);
5999 	if ((flags & (zfs_vm_pagerput_sync | zfs_vm_pagerput_inval)) != 0 ||
6000 	    zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
6001 		zil_commit(zfsvfs->z_log, zp->z_id);
6002 	ZFS_EXIT(zfsvfs);
6003 	return (rtvals[0]);
6004 }
6005 
6006 int
zfs_freebsd_putpages(ap)6007 zfs_freebsd_putpages(ap)
6008 	struct vop_putpages_args /* {
6009 		struct vnode *a_vp;
6010 		vm_page_t *a_m;
6011 		int a_count;
6012 		int a_sync;
6013 		int *a_rtvals;
6014 	} */ *ap;
6015 {
6016 
6017 	return (zfs_putpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_sync,
6018 	    ap->a_rtvals));
6019 }
6020 
6021 static int
zfs_freebsd_bmap(ap)6022 zfs_freebsd_bmap(ap)
6023 	struct vop_bmap_args /* {
6024 		struct vnode *a_vp;
6025 		daddr_t  a_bn;
6026 		struct bufobj **a_bop;
6027 		daddr_t *a_bnp;
6028 		int *a_runp;
6029 		int *a_runb;
6030 	} */ *ap;
6031 {
6032 
6033 	if (ap->a_bop != NULL)
6034 		*ap->a_bop = &ap->a_vp->v_bufobj;
6035 	if (ap->a_bnp != NULL)
6036 		*ap->a_bnp = ap->a_bn;
6037 	if (ap->a_runp != NULL)
6038 		*ap->a_runp = 0;
6039 	if (ap->a_runb != NULL)
6040 		*ap->a_runb = 0;
6041 
6042 	return (0);
6043 }
6044 
6045 static int
zfs_freebsd_open(ap)6046 zfs_freebsd_open(ap)
6047 	struct vop_open_args /* {
6048 		struct vnode *a_vp;
6049 		int a_mode;
6050 		struct ucred *a_cred;
6051 		struct thread *a_td;
6052 	} */ *ap;
6053 {
6054 	vnode_t	*vp = ap->a_vp;
6055 	znode_t *zp = VTOZ(vp);
6056 	int error;
6057 
6058 	error = zfs_open(&vp, ap->a_mode, ap->a_cred, NULL);
6059 	if (error == 0)
6060 		vnode_create_vobject(vp, zp->z_size, ap->a_td);
6061 	return (error);
6062 }
6063 
6064 static int
zfs_freebsd_close(ap)6065 zfs_freebsd_close(ap)
6066 	struct vop_close_args /* {
6067 		struct vnode *a_vp;
6068 		int  a_fflag;
6069 		struct ucred *a_cred;
6070 		struct thread *a_td;
6071 	} */ *ap;
6072 {
6073 
6074 	return (zfs_close(ap->a_vp, ap->a_fflag, 1, 0, ap->a_cred, NULL));
6075 }
6076 
6077 static int
zfs_freebsd_ioctl(ap)6078 zfs_freebsd_ioctl(ap)
6079 	struct vop_ioctl_args /* {
6080 		struct vnode *a_vp;
6081 		u_long a_command;
6082 		caddr_t a_data;
6083 		int a_fflag;
6084 		struct ucred *cred;
6085 		struct thread *td;
6086 	} */ *ap;
6087 {
6088 
6089 	return (zfs_ioctl(ap->a_vp, ap->a_command, (intptr_t)ap->a_data,
6090 	    ap->a_fflag, ap->a_cred, NULL, NULL));
6091 }
6092 
6093 static int
zfs_freebsd_read(ap)6094 zfs_freebsd_read(ap)
6095 	struct vop_read_args /* {
6096 		struct vnode *a_vp;
6097 		struct uio *a_uio;
6098 		int a_ioflag;
6099 		struct ucred *a_cred;
6100 	} */ *ap;
6101 {
6102 
6103 	return (zfs_read(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag),
6104 	    ap->a_cred, NULL));
6105 }
6106 
6107 static int
zfs_freebsd_write(ap)6108 zfs_freebsd_write(ap)
6109 	struct vop_write_args /* {
6110 		struct vnode *a_vp;
6111 		struct uio *a_uio;
6112 		int a_ioflag;
6113 		struct ucred *a_cred;
6114 	} */ *ap;
6115 {
6116 
6117 	return (zfs_write(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag),
6118 	    ap->a_cred, NULL));
6119 }
6120 
6121 static int
zfs_freebsd_access(ap)6122 zfs_freebsd_access(ap)
6123 	struct vop_access_args /* {
6124 		struct vnode *a_vp;
6125 		accmode_t a_accmode;
6126 		struct ucred *a_cred;
6127 		struct thread *a_td;
6128 	} */ *ap;
6129 {
6130 	vnode_t *vp = ap->a_vp;
6131 	znode_t *zp = VTOZ(vp);
6132 	accmode_t accmode;
6133 	int error = 0;
6134 
6135 	/*
6136 	 * ZFS itself only knowns about VREAD, VWRITE, VEXEC and VAPPEND,
6137 	 */
6138 	accmode = ap->a_accmode & (VREAD|VWRITE|VEXEC|VAPPEND);
6139 	if (accmode != 0)
6140 		error = zfs_access(ap->a_vp, accmode, 0, ap->a_cred, NULL);
6141 
6142 	/*
6143 	 * VADMIN has to be handled by vaccess().
6144 	 */
6145 	if (error == 0) {
6146 		accmode = ap->a_accmode & ~(VREAD|VWRITE|VEXEC|VAPPEND);
6147 		if (accmode != 0) {
6148 			error = vaccess(vp->v_type, zp->z_mode, zp->z_uid,
6149 			    zp->z_gid, accmode, ap->a_cred, NULL);
6150 		}
6151 	}
6152 
6153 	/*
6154 	 * For VEXEC, ensure that at least one execute bit is set for
6155 	 * non-directories.
6156 	 */
6157 	if (error == 0 && (ap->a_accmode & VEXEC) != 0 && vp->v_type != VDIR &&
6158 	    (zp->z_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0) {
6159 		error = EACCES;
6160 	}
6161 
6162 	return (error);
6163 }
6164 
6165 static int
zfs_freebsd_lookup(ap)6166 zfs_freebsd_lookup(ap)
6167 	struct vop_lookup_args /* {
6168 		struct vnode *a_dvp;
6169 		struct vnode **a_vpp;
6170 		struct componentname *a_cnp;
6171 	} */ *ap;
6172 {
6173 	struct componentname *cnp = ap->a_cnp;
6174 	char nm[NAME_MAX + 1];
6175 
6176 	ASSERT(cnp->cn_namelen < sizeof(nm));
6177 	strlcpy(nm, cnp->cn_nameptr, MIN(cnp->cn_namelen + 1, sizeof(nm)));
6178 
6179 	return (zfs_lookup(ap->a_dvp, nm, ap->a_vpp, cnp, cnp->cn_nameiop,
6180 	    cnp->cn_cred, cnp->cn_thread, 0));
6181 }
6182 
6183 static int
zfs_freebsd_create(ap)6184 zfs_freebsd_create(ap)
6185 	struct vop_create_args /* {
6186 		struct vnode *a_dvp;
6187 		struct vnode **a_vpp;
6188 		struct componentname *a_cnp;
6189 		struct vattr *a_vap;
6190 	} */ *ap;
6191 {
6192 	struct componentname *cnp = ap->a_cnp;
6193 	vattr_t *vap = ap->a_vap;
6194 	int error, mode;
6195 
6196 	ASSERT(cnp->cn_flags & SAVENAME);
6197 
6198 	vattr_init_mask(vap);
6199 	mode = vap->va_mode & ALLPERMS;
6200 
6201 	error = zfs_create(ap->a_dvp, cnp->cn_nameptr, vap, !EXCL, mode,
6202 	    ap->a_vpp, cnp->cn_cred, cnp->cn_thread);
6203 #ifdef FREEBSD_NAMECACHE
6204 	if (error == 0 && (cnp->cn_flags & MAKEENTRY) != 0)
6205 		cache_enter(ap->a_dvp, *ap->a_vpp, cnp);
6206 #endif
6207 	return (error);
6208 }
6209 
6210 static int
zfs_freebsd_remove(ap)6211 zfs_freebsd_remove(ap)
6212 	struct vop_remove_args /* {
6213 		struct vnode *a_dvp;
6214 		struct vnode *a_vp;
6215 		struct componentname *a_cnp;
6216 	} */ *ap;
6217 {
6218 
6219 	ASSERT(ap->a_cnp->cn_flags & SAVENAME);
6220 
6221 	return (zfs_remove(ap->a_dvp, ap->a_cnp->cn_nameptr,
6222 	    ap->a_cnp->cn_cred, NULL, 0));
6223 }
6224 
6225 static int
zfs_freebsd_mkdir(ap)6226 zfs_freebsd_mkdir(ap)
6227 	struct vop_mkdir_args /* {
6228 		struct vnode *a_dvp;
6229 		struct vnode **a_vpp;
6230 		struct componentname *a_cnp;
6231 		struct vattr *a_vap;
6232 	} */ *ap;
6233 {
6234 	vattr_t *vap = ap->a_vap;
6235 
6236 	ASSERT(ap->a_cnp->cn_flags & SAVENAME);
6237 
6238 	vattr_init_mask(vap);
6239 
6240 	return (zfs_mkdir(ap->a_dvp, ap->a_cnp->cn_nameptr, vap, ap->a_vpp,
6241 	    ap->a_cnp->cn_cred, NULL, 0, NULL));
6242 }
6243 
6244 static int
zfs_freebsd_rmdir(ap)6245 zfs_freebsd_rmdir(ap)
6246 	struct vop_rmdir_args /* {
6247 		struct vnode *a_dvp;
6248 		struct vnode *a_vp;
6249 		struct componentname *a_cnp;
6250 	} */ *ap;
6251 {
6252 	struct componentname *cnp = ap->a_cnp;
6253 
6254 	ASSERT(cnp->cn_flags & SAVENAME);
6255 
6256 	return (zfs_rmdir(ap->a_dvp, cnp->cn_nameptr, NULL, cnp->cn_cred, NULL, 0));
6257 }
6258 
6259 static int
zfs_freebsd_readdir(ap)6260 zfs_freebsd_readdir(ap)
6261 	struct vop_readdir_args /* {
6262 		struct vnode *a_vp;
6263 		struct uio *a_uio;
6264 		struct ucred *a_cred;
6265 		int *a_eofflag;
6266 		int *a_ncookies;
6267 		u_long **a_cookies;
6268 	} */ *ap;
6269 {
6270 
6271 	return (zfs_readdir(ap->a_vp, ap->a_uio, ap->a_cred, ap->a_eofflag,
6272 	    ap->a_ncookies, ap->a_cookies));
6273 }
6274 
6275 static int
zfs_freebsd_fsync(ap)6276 zfs_freebsd_fsync(ap)
6277 	struct vop_fsync_args /* {
6278 		struct vnode *a_vp;
6279 		int a_waitfor;
6280 		struct thread *a_td;
6281 	} */ *ap;
6282 {
6283 
6284 	vop_stdfsync(ap);
6285 	return (zfs_fsync(ap->a_vp, 0, ap->a_td->td_ucred, NULL));
6286 }
6287 
6288 static int
zfs_freebsd_getattr(ap)6289 zfs_freebsd_getattr(ap)
6290 	struct vop_getattr_args /* {
6291 		struct vnode *a_vp;
6292 		struct vattr *a_vap;
6293 		struct ucred *a_cred;
6294 	} */ *ap;
6295 {
6296 	vattr_t *vap = ap->a_vap;
6297 	xvattr_t xvap;
6298 	u_long fflags = 0;
6299 	int error;
6300 
6301 	xva_init(&xvap);
6302 	xvap.xva_vattr = *vap;
6303 	xvap.xva_vattr.va_mask |= AT_XVATTR;
6304 
6305 	/* Convert chflags into ZFS-type flags. */
6306 	/* XXX: what about SF_SETTABLE?. */
6307 	XVA_SET_REQ(&xvap, XAT_IMMUTABLE);
6308 	XVA_SET_REQ(&xvap, XAT_APPENDONLY);
6309 	XVA_SET_REQ(&xvap, XAT_NOUNLINK);
6310 	XVA_SET_REQ(&xvap, XAT_NODUMP);
6311 	XVA_SET_REQ(&xvap, XAT_READONLY);
6312 	XVA_SET_REQ(&xvap, XAT_ARCHIVE);
6313 	XVA_SET_REQ(&xvap, XAT_SYSTEM);
6314 	XVA_SET_REQ(&xvap, XAT_HIDDEN);
6315 	XVA_SET_REQ(&xvap, XAT_REPARSE);
6316 	XVA_SET_REQ(&xvap, XAT_OFFLINE);
6317 	XVA_SET_REQ(&xvap, XAT_SPARSE);
6318 
6319 	error = zfs_getattr(ap->a_vp, (vattr_t *)&xvap, 0, ap->a_cred, NULL);
6320 	if (error != 0)
6321 		return (error);
6322 
6323 	/* Convert ZFS xattr into chflags. */
6324 #define	FLAG_CHECK(fflag, xflag, xfield)	do {			\
6325 	if (XVA_ISSET_RTN(&xvap, (xflag)) && (xfield) != 0)		\
6326 		fflags |= (fflag);					\
6327 } while (0)
6328 	FLAG_CHECK(SF_IMMUTABLE, XAT_IMMUTABLE,
6329 	    xvap.xva_xoptattrs.xoa_immutable);
6330 	FLAG_CHECK(SF_APPEND, XAT_APPENDONLY,
6331 	    xvap.xva_xoptattrs.xoa_appendonly);
6332 	FLAG_CHECK(SF_NOUNLINK, XAT_NOUNLINK,
6333 	    xvap.xva_xoptattrs.xoa_nounlink);
6334 	FLAG_CHECK(UF_ARCHIVE, XAT_ARCHIVE,
6335 	    xvap.xva_xoptattrs.xoa_archive);
6336 	FLAG_CHECK(UF_NODUMP, XAT_NODUMP,
6337 	    xvap.xva_xoptattrs.xoa_nodump);
6338 	FLAG_CHECK(UF_READONLY, XAT_READONLY,
6339 	    xvap.xva_xoptattrs.xoa_readonly);
6340 	FLAG_CHECK(UF_SYSTEM, XAT_SYSTEM,
6341 	    xvap.xva_xoptattrs.xoa_system);
6342 	FLAG_CHECK(UF_HIDDEN, XAT_HIDDEN,
6343 	    xvap.xva_xoptattrs.xoa_hidden);
6344 	FLAG_CHECK(UF_REPARSE, XAT_REPARSE,
6345 	    xvap.xva_xoptattrs.xoa_reparse);
6346 	FLAG_CHECK(UF_OFFLINE, XAT_OFFLINE,
6347 	    xvap.xva_xoptattrs.xoa_offline);
6348 	FLAG_CHECK(UF_SPARSE, XAT_SPARSE,
6349 	    xvap.xva_xoptattrs.xoa_sparse);
6350 
6351 #undef	FLAG_CHECK
6352 	*vap = xvap.xva_vattr;
6353 	vap->va_flags = fflags;
6354 	return (0);
6355 }
6356 
6357 static int
zfs_freebsd_setattr(ap)6358 zfs_freebsd_setattr(ap)
6359 	struct vop_setattr_args /* {
6360 		struct vnode *a_vp;
6361 		struct vattr *a_vap;
6362 		struct ucred *a_cred;
6363 	} */ *ap;
6364 {
6365 	vnode_t *vp = ap->a_vp;
6366 	vattr_t *vap = ap->a_vap;
6367 	cred_t *cred = ap->a_cred;
6368 	xvattr_t xvap;
6369 	u_long fflags;
6370 	uint64_t zflags;
6371 
6372 	vattr_init_mask(vap);
6373 	vap->va_mask &= ~AT_NOSET;
6374 
6375 	xva_init(&xvap);
6376 	xvap.xva_vattr = *vap;
6377 
6378 	zflags = VTOZ(vp)->z_pflags;
6379 
6380 	if (vap->va_flags != VNOVAL) {
6381 		zfsvfs_t *zfsvfs = VTOZ(vp)->z_zfsvfs;
6382 		int error;
6383 
6384 		if (zfsvfs->z_use_fuids == B_FALSE)
6385 			return (EOPNOTSUPP);
6386 
6387 		fflags = vap->va_flags;
6388 		/*
6389 		 * XXX KDM
6390 		 * We need to figure out whether it makes sense to allow
6391 		 * UF_REPARSE through, since we don't really have other
6392 		 * facilities to handle reparse points and zfs_setattr()
6393 		 * doesn't currently allow setting that attribute anyway.
6394 		 */
6395 		if ((fflags & ~(SF_IMMUTABLE|SF_APPEND|SF_NOUNLINK|UF_ARCHIVE|
6396 		     UF_NODUMP|UF_SYSTEM|UF_HIDDEN|UF_READONLY|UF_REPARSE|
6397 		     UF_OFFLINE|UF_SPARSE)) != 0)
6398 			return (EOPNOTSUPP);
6399 		/*
6400 		 * Unprivileged processes are not permitted to unset system
6401 		 * flags, or modify flags if any system flags are set.
6402 		 * Privileged non-jail processes may not modify system flags
6403 		 * if securelevel > 0 and any existing system flags are set.
6404 		 * Privileged jail processes behave like privileged non-jail
6405 		 * processes if the security.jail.chflags_allowed sysctl is
6406 		 * is non-zero; otherwise, they behave like unprivileged
6407 		 * processes.
6408 		 */
6409 		if (secpolicy_fs_owner(vp->v_mount, cred) == 0 ||
6410 		    priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0) == 0) {
6411 			if (zflags &
6412 			    (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) {
6413 				error = securelevel_gt(cred, 0);
6414 				if (error != 0)
6415 					return (error);
6416 			}
6417 		} else {
6418 			/*
6419 			 * Callers may only modify the file flags on objects they
6420 			 * have VADMIN rights for.
6421 			 */
6422 			if ((error = VOP_ACCESS(vp, VADMIN, cred, curthread)) != 0)
6423 				return (error);
6424 			if (zflags &
6425 			    (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) {
6426 				return (EPERM);
6427 			}
6428 			if (fflags &
6429 			    (SF_IMMUTABLE | SF_APPEND | SF_NOUNLINK)) {
6430 				return (EPERM);
6431 			}
6432 		}
6433 
6434 #define	FLAG_CHANGE(fflag, zflag, xflag, xfield)	do {		\
6435 	if (((fflags & (fflag)) && !(zflags & (zflag))) ||		\
6436 	    ((zflags & (zflag)) && !(fflags & (fflag)))) {		\
6437 		XVA_SET_REQ(&xvap, (xflag));				\
6438 		(xfield) = ((fflags & (fflag)) != 0);			\
6439 	}								\
6440 } while (0)
6441 		/* Convert chflags into ZFS-type flags. */
6442 		/* XXX: what about SF_SETTABLE?. */
6443 		FLAG_CHANGE(SF_IMMUTABLE, ZFS_IMMUTABLE, XAT_IMMUTABLE,
6444 		    xvap.xva_xoptattrs.xoa_immutable);
6445 		FLAG_CHANGE(SF_APPEND, ZFS_APPENDONLY, XAT_APPENDONLY,
6446 		    xvap.xva_xoptattrs.xoa_appendonly);
6447 		FLAG_CHANGE(SF_NOUNLINK, ZFS_NOUNLINK, XAT_NOUNLINK,
6448 		    xvap.xva_xoptattrs.xoa_nounlink);
6449 		FLAG_CHANGE(UF_ARCHIVE, ZFS_ARCHIVE, XAT_ARCHIVE,
6450 		    xvap.xva_xoptattrs.xoa_archive);
6451 		FLAG_CHANGE(UF_NODUMP, ZFS_NODUMP, XAT_NODUMP,
6452 		    xvap.xva_xoptattrs.xoa_nodump);
6453 		FLAG_CHANGE(UF_READONLY, ZFS_READONLY, XAT_READONLY,
6454 		    xvap.xva_xoptattrs.xoa_readonly);
6455 		FLAG_CHANGE(UF_SYSTEM, ZFS_SYSTEM, XAT_SYSTEM,
6456 		    xvap.xva_xoptattrs.xoa_system);
6457 		FLAG_CHANGE(UF_HIDDEN, ZFS_HIDDEN, XAT_HIDDEN,
6458 		    xvap.xva_xoptattrs.xoa_hidden);
6459 		FLAG_CHANGE(UF_REPARSE, ZFS_REPARSE, XAT_REPARSE,
6460 		    xvap.xva_xoptattrs.xoa_hidden);
6461 		FLAG_CHANGE(UF_OFFLINE, ZFS_OFFLINE, XAT_OFFLINE,
6462 		    xvap.xva_xoptattrs.xoa_offline);
6463 		FLAG_CHANGE(UF_SPARSE, ZFS_SPARSE, XAT_SPARSE,
6464 		    xvap.xva_xoptattrs.xoa_sparse);
6465 #undef	FLAG_CHANGE
6466 	}
6467 	return (zfs_setattr(vp, (vattr_t *)&xvap, 0, cred, NULL));
6468 }
6469 
6470 static int
zfs_freebsd_rename(ap)6471 zfs_freebsd_rename(ap)
6472 	struct vop_rename_args  /* {
6473 		struct vnode *a_fdvp;
6474 		struct vnode *a_fvp;
6475 		struct componentname *a_fcnp;
6476 		struct vnode *a_tdvp;
6477 		struct vnode *a_tvp;
6478 		struct componentname *a_tcnp;
6479 	} */ *ap;
6480 {
6481 	vnode_t *fdvp = ap->a_fdvp;
6482 	vnode_t *fvp = ap->a_fvp;
6483 	vnode_t *tdvp = ap->a_tdvp;
6484 	vnode_t *tvp = ap->a_tvp;
6485 	int error;
6486 
6487 	ASSERT(ap->a_fcnp->cn_flags & (SAVENAME|SAVESTART));
6488 	ASSERT(ap->a_tcnp->cn_flags & (SAVENAME|SAVESTART));
6489 
6490 	/*
6491 	 * Check for cross-device rename.
6492 	 */
6493 	if ((fdvp->v_mount != tdvp->v_mount) ||
6494 	    (tvp && (fdvp->v_mount != tvp->v_mount)))
6495 		error = EXDEV;
6496 	else
6497 		error = zfs_rename(fdvp, ap->a_fcnp->cn_nameptr, tdvp,
6498 		    ap->a_tcnp->cn_nameptr, ap->a_fcnp->cn_cred, NULL, 0);
6499 	if (tdvp == tvp)
6500 		VN_RELE(tdvp);
6501 	else
6502 		VN_URELE(tdvp);
6503 	if (tvp)
6504 		VN_URELE(tvp);
6505 	VN_RELE(fdvp);
6506 	VN_RELE(fvp);
6507 
6508 	return (error);
6509 }
6510 
6511 static int
zfs_freebsd_symlink(ap)6512 zfs_freebsd_symlink(ap)
6513 	struct vop_symlink_args /* {
6514 		struct vnode *a_dvp;
6515 		struct vnode **a_vpp;
6516 		struct componentname *a_cnp;
6517 		struct vattr *a_vap;
6518 		char *a_target;
6519 	} */ *ap;
6520 {
6521 	struct componentname *cnp = ap->a_cnp;
6522 	vattr_t *vap = ap->a_vap;
6523 
6524 	ASSERT(cnp->cn_flags & SAVENAME);
6525 
6526 	vap->va_type = VLNK;	/* FreeBSD: Syscall only sets va_mode. */
6527 	vattr_init_mask(vap);
6528 
6529 	return (zfs_symlink(ap->a_dvp, ap->a_vpp, cnp->cn_nameptr, vap,
6530 	    ap->a_target, cnp->cn_cred, cnp->cn_thread));
6531 }
6532 
6533 static int
zfs_freebsd_readlink(ap)6534 zfs_freebsd_readlink(ap)
6535 	struct vop_readlink_args /* {
6536 		struct vnode *a_vp;
6537 		struct uio *a_uio;
6538 		struct ucred *a_cred;
6539 	} */ *ap;
6540 {
6541 
6542 	return (zfs_readlink(ap->a_vp, ap->a_uio, ap->a_cred, NULL));
6543 }
6544 
6545 static int
zfs_freebsd_link(ap)6546 zfs_freebsd_link(ap)
6547 	struct vop_link_args /* {
6548 		struct vnode *a_tdvp;
6549 		struct vnode *a_vp;
6550 		struct componentname *a_cnp;
6551 	} */ *ap;
6552 {
6553 	struct componentname *cnp = ap->a_cnp;
6554 	vnode_t *vp = ap->a_vp;
6555 	vnode_t *tdvp = ap->a_tdvp;
6556 
6557 	if (tdvp->v_mount != vp->v_mount)
6558 		return (EXDEV);
6559 
6560 	ASSERT(cnp->cn_flags & SAVENAME);
6561 
6562 	return (zfs_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_cred, NULL, 0));
6563 }
6564 
6565 static int
zfs_freebsd_inactive(ap)6566 zfs_freebsd_inactive(ap)
6567 	struct vop_inactive_args /* {
6568 		struct vnode *a_vp;
6569 		struct thread *a_td;
6570 	} */ *ap;
6571 {
6572 	vnode_t *vp = ap->a_vp;
6573 
6574 	zfs_inactive(vp, ap->a_td->td_ucred, NULL);
6575 	return (0);
6576 }
6577 
6578 static int
zfs_freebsd_reclaim(ap)6579 zfs_freebsd_reclaim(ap)
6580 	struct vop_reclaim_args /* {
6581 		struct vnode *a_vp;
6582 		struct thread *a_td;
6583 	} */ *ap;
6584 {
6585 	vnode_t	*vp = ap->a_vp;
6586 	znode_t	*zp = VTOZ(vp);
6587 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
6588 
6589 	ASSERT(zp != NULL);
6590 
6591 	/* Destroy the vm object and flush associated pages. */
6592 	vnode_destroy_vobject(vp);
6593 
6594 	/*
6595 	 * z_teardown_inactive_lock protects from a race with
6596 	 * zfs_znode_dmu_fini in zfsvfs_teardown during
6597 	 * force unmount.
6598 	 */
6599 	rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
6600 	if (zp->z_sa_hdl == NULL)
6601 		zfs_znode_free(zp);
6602 	else
6603 		zfs_zinactive(zp);
6604 	rw_exit(&zfsvfs->z_teardown_inactive_lock);
6605 
6606 	vp->v_data = NULL;
6607 	return (0);
6608 }
6609 
6610 static int
zfs_freebsd_fid(ap)6611 zfs_freebsd_fid(ap)
6612 	struct vop_fid_args /* {
6613 		struct vnode *a_vp;
6614 		struct fid *a_fid;
6615 	} */ *ap;
6616 {
6617 
6618 	return (zfs_fid(ap->a_vp, (void *)ap->a_fid, NULL));
6619 }
6620 
6621 static int
zfs_freebsd_pathconf(ap)6622 zfs_freebsd_pathconf(ap)
6623 	struct vop_pathconf_args /* {
6624 		struct vnode *a_vp;
6625 		int a_name;
6626 		register_t *a_retval;
6627 	} */ *ap;
6628 {
6629 	ulong_t val;
6630 	int error;
6631 
6632 	error = zfs_pathconf(ap->a_vp, ap->a_name, &val, curthread->td_ucred, NULL);
6633 	if (error == 0)
6634 		*ap->a_retval = val;
6635 	else if (error == EOPNOTSUPP)
6636 		error = vop_stdpathconf(ap);
6637 	return (error);
6638 }
6639 
6640 static int
zfs_freebsd_fifo_pathconf(ap)6641 zfs_freebsd_fifo_pathconf(ap)
6642 	struct vop_pathconf_args /* {
6643 		struct vnode *a_vp;
6644 		int a_name;
6645 		register_t *a_retval;
6646 	} */ *ap;
6647 {
6648 
6649 	switch (ap->a_name) {
6650 	case _PC_ACL_EXTENDED:
6651 	case _PC_ACL_NFS4:
6652 	case _PC_ACL_PATH_MAX:
6653 	case _PC_MAC_PRESENT:
6654 		return (zfs_freebsd_pathconf(ap));
6655 	default:
6656 		return (fifo_specops.vop_pathconf(ap));
6657 	}
6658 }
6659 
6660 /*
6661  * FreeBSD's extended attributes namespace defines file name prefix for ZFS'
6662  * extended attribute name:
6663  *
6664  *	NAMESPACE	PREFIX
6665  *	system		freebsd:system:
6666  *	user		(none, can be used to access ZFS fsattr(5) attributes
6667  *			created on Solaris)
6668  */
6669 static int
zfs_create_attrname(int attrnamespace,const char * name,char * attrname,size_t size)6670 zfs_create_attrname(int attrnamespace, const char *name, char *attrname,
6671     size_t size)
6672 {
6673 	const char *namespace, *prefix, *suffix;
6674 
6675 	/* We don't allow '/' character in attribute name. */
6676 	if (strchr(name, '/') != NULL)
6677 		return (EINVAL);
6678 	/* We don't allow attribute names that start with "freebsd:" string. */
6679 	if (strncmp(name, "freebsd:", 8) == 0)
6680 		return (EINVAL);
6681 
6682 	bzero(attrname, size);
6683 
6684 	switch (attrnamespace) {
6685 	case EXTATTR_NAMESPACE_USER:
6686 #if 0
6687 		prefix = "freebsd:";
6688 		namespace = EXTATTR_NAMESPACE_USER_STRING;
6689 		suffix = ":";
6690 #else
6691 		/*
6692 		 * This is the default namespace by which we can access all
6693 		 * attributes created on Solaris.
6694 		 */
6695 		prefix = namespace = suffix = "";
6696 #endif
6697 		break;
6698 	case EXTATTR_NAMESPACE_SYSTEM:
6699 		prefix = "freebsd:";
6700 		namespace = EXTATTR_NAMESPACE_SYSTEM_STRING;
6701 		suffix = ":";
6702 		break;
6703 	case EXTATTR_NAMESPACE_EMPTY:
6704 	default:
6705 		return (EINVAL);
6706 	}
6707 	if (snprintf(attrname, size, "%s%s%s%s", prefix, namespace, suffix,
6708 	    name) >= size) {
6709 		return (ENAMETOOLONG);
6710 	}
6711 	return (0);
6712 }
6713 
6714 /*
6715  * Vnode operating to retrieve a named extended attribute.
6716  */
6717 static int
zfs_getextattr(struct vop_getextattr_args * ap)6718 zfs_getextattr(struct vop_getextattr_args *ap)
6719 /*
6720 vop_getextattr {
6721 	IN struct vnode *a_vp;
6722 	IN int a_attrnamespace;
6723 	IN const char *a_name;
6724 	INOUT struct uio *a_uio;
6725 	OUT size_t *a_size;
6726 	IN struct ucred *a_cred;
6727 	IN struct thread *a_td;
6728 };
6729 */
6730 {
6731 	zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6732 	struct thread *td = ap->a_td;
6733 	struct nameidata nd;
6734 	char attrname[255];
6735 	struct vattr va;
6736 	vnode_t *xvp = NULL, *vp;
6737 	int error, flags;
6738 
6739 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6740 	    ap->a_cred, ap->a_td, VREAD);
6741 	if (error != 0)
6742 		return (error);
6743 
6744 	error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6745 	    sizeof(attrname));
6746 	if (error != 0)
6747 		return (error);
6748 
6749 	ZFS_ENTER(zfsvfs);
6750 
6751 	error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6752 	    LOOKUP_XATTR);
6753 	if (error != 0) {
6754 		ZFS_EXIT(zfsvfs);
6755 		return (error);
6756 	}
6757 
6758 	flags = FREAD;
6759 	NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname,
6760 	    xvp, td);
6761 	error = vn_open_cred(&nd, &flags, 0, 0, ap->a_cred, NULL);
6762 	vp = nd.ni_vp;
6763 	NDFREE(&nd, NDF_ONLY_PNBUF);
6764 	if (error != 0) {
6765 		ZFS_EXIT(zfsvfs);
6766 		if (error == ENOENT)
6767 			error = ENOATTR;
6768 		return (error);
6769 	}
6770 
6771 	if (ap->a_size != NULL) {
6772 		error = VOP_GETATTR(vp, &va, ap->a_cred);
6773 		if (error == 0)
6774 			*ap->a_size = (size_t)va.va_size;
6775 	} else if (ap->a_uio != NULL)
6776 		error = VOP_READ(vp, ap->a_uio, IO_UNIT, ap->a_cred);
6777 
6778 	VOP_UNLOCK(vp, 0);
6779 	vn_close(vp, flags, ap->a_cred, td);
6780 	ZFS_EXIT(zfsvfs);
6781 
6782 	return (error);
6783 }
6784 
6785 /*
6786  * Vnode operation to remove a named attribute.
6787  */
6788 int
zfs_deleteextattr(struct vop_deleteextattr_args * ap)6789 zfs_deleteextattr(struct vop_deleteextattr_args *ap)
6790 /*
6791 vop_deleteextattr {
6792 	IN struct vnode *a_vp;
6793 	IN int a_attrnamespace;
6794 	IN const char *a_name;
6795 	IN struct ucred *a_cred;
6796 	IN struct thread *a_td;
6797 };
6798 */
6799 {
6800 	zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6801 	struct thread *td = ap->a_td;
6802 	struct nameidata nd;
6803 	char attrname[255];
6804 	struct vattr va;
6805 	vnode_t *xvp = NULL, *vp;
6806 	int error, flags;
6807 
6808 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6809 	    ap->a_cred, ap->a_td, VWRITE);
6810 	if (error != 0)
6811 		return (error);
6812 
6813 	error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6814 	    sizeof(attrname));
6815 	if (error != 0)
6816 		return (error);
6817 
6818 	ZFS_ENTER(zfsvfs);
6819 
6820 	error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6821 	    LOOKUP_XATTR);
6822 	if (error != 0) {
6823 		ZFS_EXIT(zfsvfs);
6824 		return (error);
6825 	}
6826 
6827 	NDINIT_ATVP(&nd, DELETE, NOFOLLOW | LOCKPARENT | LOCKLEAF,
6828 	    UIO_SYSSPACE, attrname, xvp, td);
6829 	error = namei(&nd);
6830 	vp = nd.ni_vp;
6831 	if (error != 0) {
6832 		ZFS_EXIT(zfsvfs);
6833 		NDFREE(&nd, NDF_ONLY_PNBUF);
6834 		if (error == ENOENT)
6835 			error = ENOATTR;
6836 		return (error);
6837 	}
6838 
6839 	error = VOP_REMOVE(nd.ni_dvp, vp, &nd.ni_cnd);
6840 	NDFREE(&nd, NDF_ONLY_PNBUF);
6841 
6842 	vput(nd.ni_dvp);
6843 	if (vp == nd.ni_dvp)
6844 		vrele(vp);
6845 	else
6846 		vput(vp);
6847 	ZFS_EXIT(zfsvfs);
6848 
6849 	return (error);
6850 }
6851 
6852 /*
6853  * Vnode operation to set a named attribute.
6854  */
6855 static int
zfs_setextattr(struct vop_setextattr_args * ap)6856 zfs_setextattr(struct vop_setextattr_args *ap)
6857 /*
6858 vop_setextattr {
6859 	IN struct vnode *a_vp;
6860 	IN int a_attrnamespace;
6861 	IN const char *a_name;
6862 	INOUT struct uio *a_uio;
6863 	IN struct ucred *a_cred;
6864 	IN struct thread *a_td;
6865 };
6866 */
6867 {
6868 	zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6869 	struct thread *td = ap->a_td;
6870 	struct nameidata nd;
6871 	char attrname[255];
6872 	struct vattr va;
6873 	vnode_t *xvp = NULL, *vp;
6874 	int error, flags;
6875 
6876 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6877 	    ap->a_cred, ap->a_td, VWRITE);
6878 	if (error != 0)
6879 		return (error);
6880 
6881 	error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6882 	    sizeof(attrname));
6883 	if (error != 0)
6884 		return (error);
6885 
6886 	ZFS_ENTER(zfsvfs);
6887 
6888 	error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6889 	    LOOKUP_XATTR | CREATE_XATTR_DIR);
6890 	if (error != 0) {
6891 		ZFS_EXIT(zfsvfs);
6892 		return (error);
6893 	}
6894 
6895 	flags = FFLAGS(O_WRONLY | O_CREAT);
6896 	NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname,
6897 	    xvp, td);
6898 	error = vn_open_cred(&nd, &flags, 0600, 0, ap->a_cred, NULL);
6899 	vp = nd.ni_vp;
6900 	NDFREE(&nd, NDF_ONLY_PNBUF);
6901 	if (error != 0) {
6902 		ZFS_EXIT(zfsvfs);
6903 		return (error);
6904 	}
6905 
6906 	VATTR_NULL(&va);
6907 	va.va_size = 0;
6908 	error = VOP_SETATTR(vp, &va, ap->a_cred);
6909 	if (error == 0)
6910 		VOP_WRITE(vp, ap->a_uio, IO_UNIT, ap->a_cred);
6911 
6912 	VOP_UNLOCK(vp, 0);
6913 	vn_close(vp, flags, ap->a_cred, td);
6914 	ZFS_EXIT(zfsvfs);
6915 
6916 	return (error);
6917 }
6918 
6919 /*
6920  * Vnode operation to retrieve extended attributes on a vnode.
6921  */
6922 static int
zfs_listextattr(struct vop_listextattr_args * ap)6923 zfs_listextattr(struct vop_listextattr_args *ap)
6924 /*
6925 vop_listextattr {
6926 	IN struct vnode *a_vp;
6927 	IN int a_attrnamespace;
6928 	INOUT struct uio *a_uio;
6929 	OUT size_t *a_size;
6930 	IN struct ucred *a_cred;
6931 	IN struct thread *a_td;
6932 };
6933 */
6934 {
6935 	zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6936 	struct thread *td = ap->a_td;
6937 	struct nameidata nd;
6938 	char attrprefix[16];
6939 	u_char dirbuf[sizeof(struct dirent)];
6940 	struct dirent *dp;
6941 	struct iovec aiov;
6942 	struct uio auio, *uio = ap->a_uio;
6943 	size_t *sizep = ap->a_size;
6944 	size_t plen;
6945 	vnode_t *xvp = NULL, *vp;
6946 	int done, error, eof, pos;
6947 
6948 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6949 	    ap->a_cred, ap->a_td, VREAD);
6950 	if (error != 0)
6951 		return (error);
6952 
6953 	error = zfs_create_attrname(ap->a_attrnamespace, "", attrprefix,
6954 	    sizeof(attrprefix));
6955 	if (error != 0)
6956 		return (error);
6957 	plen = strlen(attrprefix);
6958 
6959 	ZFS_ENTER(zfsvfs);
6960 
6961 	if (sizep != NULL)
6962 		*sizep = 0;
6963 
6964 	error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6965 	    LOOKUP_XATTR);
6966 	if (error != 0) {
6967 		ZFS_EXIT(zfsvfs);
6968 		/*
6969 		 * ENOATTR means that the EA directory does not yet exist,
6970 		 * i.e. there are no extended attributes there.
6971 		 */
6972 		if (error == ENOATTR)
6973 			error = 0;
6974 		return (error);
6975 	}
6976 
6977 	NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | LOCKLEAF | LOCKSHARED,
6978 	    UIO_SYSSPACE, ".", xvp, td);
6979 	error = namei(&nd);
6980 	vp = nd.ni_vp;
6981 	NDFREE(&nd, NDF_ONLY_PNBUF);
6982 	if (error != 0) {
6983 		ZFS_EXIT(zfsvfs);
6984 		return (error);
6985 	}
6986 
6987 	auio.uio_iov = &aiov;
6988 	auio.uio_iovcnt = 1;
6989 	auio.uio_segflg = UIO_SYSSPACE;
6990 	auio.uio_td = td;
6991 	auio.uio_rw = UIO_READ;
6992 	auio.uio_offset = 0;
6993 
6994 	do {
6995 		u_char nlen;
6996 
6997 		aiov.iov_base = (void *)dirbuf;
6998 		aiov.iov_len = sizeof(dirbuf);
6999 		auio.uio_resid = sizeof(dirbuf);
7000 		error = VOP_READDIR(vp, &auio, ap->a_cred, &eof, NULL, NULL);
7001 		done = sizeof(dirbuf) - auio.uio_resid;
7002 		if (error != 0)
7003 			break;
7004 		for (pos = 0; pos < done;) {
7005 			dp = (struct dirent *)(dirbuf + pos);
7006 			pos += dp->d_reclen;
7007 			/*
7008 			 * XXX: Temporarily we also accept DT_UNKNOWN, as this
7009 			 * is what we get when attribute was created on Solaris.
7010 			 */
7011 			if (dp->d_type != DT_REG && dp->d_type != DT_UNKNOWN)
7012 				continue;
7013 			if (plen == 0 && strncmp(dp->d_name, "freebsd:", 8) == 0)
7014 				continue;
7015 			else if (strncmp(dp->d_name, attrprefix, plen) != 0)
7016 				continue;
7017 			nlen = dp->d_namlen - plen;
7018 			if (sizep != NULL)
7019 				*sizep += 1 + nlen;
7020 			else if (uio != NULL) {
7021 				/*
7022 				 * Format of extattr name entry is one byte for
7023 				 * length and the rest for name.
7024 				 */
7025 				error = uiomove(&nlen, 1, uio->uio_rw, uio);
7026 				if (error == 0) {
7027 					error = uiomove(dp->d_name + plen, nlen,
7028 					    uio->uio_rw, uio);
7029 				}
7030 				if (error != 0)
7031 					break;
7032 			}
7033 		}
7034 	} while (!eof && error == 0);
7035 
7036 	vput(vp);
7037 	ZFS_EXIT(zfsvfs);
7038 
7039 	return (error);
7040 }
7041 
7042 int
zfs_freebsd_getacl(ap)7043 zfs_freebsd_getacl(ap)
7044 	struct vop_getacl_args /* {
7045 		struct vnode *vp;
7046 		acl_type_t type;
7047 		struct acl *aclp;
7048 		struct ucred *cred;
7049 		struct thread *td;
7050 	} */ *ap;
7051 {
7052 	int		error;
7053 	vsecattr_t      vsecattr;
7054 
7055 	if (ap->a_type != ACL_TYPE_NFS4)
7056 		return (EINVAL);
7057 
7058 	vsecattr.vsa_mask = VSA_ACE | VSA_ACECNT;
7059 	if (error = zfs_getsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL))
7060 		return (error);
7061 
7062 	error = acl_from_aces(ap->a_aclp, vsecattr.vsa_aclentp, vsecattr.vsa_aclcnt);
7063 	if (vsecattr.vsa_aclentp != NULL)
7064 		kmem_free(vsecattr.vsa_aclentp, vsecattr.vsa_aclentsz);
7065 
7066 	return (error);
7067 }
7068 
7069 int
zfs_freebsd_setacl(ap)7070 zfs_freebsd_setacl(ap)
7071 	struct vop_setacl_args /* {
7072 		struct vnode *vp;
7073 		acl_type_t type;
7074 		struct acl *aclp;
7075 		struct ucred *cred;
7076 		struct thread *td;
7077 	} */ *ap;
7078 {
7079 	int		error;
7080 	vsecattr_t      vsecattr;
7081 	int		aclbsize;	/* size of acl list in bytes */
7082 	aclent_t	*aaclp;
7083 
7084 	if (ap->a_type != ACL_TYPE_NFS4)
7085 		return (EINVAL);
7086 
7087 	if (ap->a_aclp->acl_cnt < 1 || ap->a_aclp->acl_cnt > MAX_ACL_ENTRIES)
7088 		return (EINVAL);
7089 
7090 	/*
7091 	 * With NFSv4 ACLs, chmod(2) may need to add additional entries,
7092 	 * splitting every entry into two and appending "canonical six"
7093 	 * entries at the end.  Don't allow for setting an ACL that would
7094 	 * cause chmod(2) to run out of ACL entries.
7095 	 */
7096 	if (ap->a_aclp->acl_cnt * 2 + 6 > ACL_MAX_ENTRIES)
7097 		return (ENOSPC);
7098 
7099 	error = acl_nfs4_check(ap->a_aclp, ap->a_vp->v_type == VDIR);
7100 	if (error != 0)
7101 		return (error);
7102 
7103 	vsecattr.vsa_mask = VSA_ACE;
7104 	aclbsize = ap->a_aclp->acl_cnt * sizeof(ace_t);
7105 	vsecattr.vsa_aclentp = kmem_alloc(aclbsize, KM_SLEEP);
7106 	aaclp = vsecattr.vsa_aclentp;
7107 	vsecattr.vsa_aclentsz = aclbsize;
7108 
7109 	aces_from_acl(vsecattr.vsa_aclentp, &vsecattr.vsa_aclcnt, ap->a_aclp);
7110 	error = zfs_setsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL);
7111 	kmem_free(aaclp, aclbsize);
7112 
7113 	return (error);
7114 }
7115 
7116 int
zfs_freebsd_aclcheck(ap)7117 zfs_freebsd_aclcheck(ap)
7118 	struct vop_aclcheck_args /* {
7119 		struct vnode *vp;
7120 		acl_type_t type;
7121 		struct acl *aclp;
7122 		struct ucred *cred;
7123 		struct thread *td;
7124 	} */ *ap;
7125 {
7126 
7127 	return (EOPNOTSUPP);
7128 }
7129 
7130 struct vop_vector zfs_vnodeops;
7131 struct vop_vector zfs_fifoops;
7132 struct vop_vector zfs_shareops;
7133 
7134 struct vop_vector zfs_vnodeops = {
7135 	.vop_default =		&default_vnodeops,
7136 	.vop_inactive =		zfs_freebsd_inactive,
7137 	.vop_reclaim =		zfs_freebsd_reclaim,
7138 	.vop_access =		zfs_freebsd_access,
7139 #ifdef FREEBSD_NAMECACHE
7140 	.vop_lookup =		vfs_cache_lookup,
7141 	.vop_cachedlookup =	zfs_freebsd_lookup,
7142 #else
7143 	.vop_lookup =		zfs_freebsd_lookup,
7144 #endif
7145 	.vop_getattr =		zfs_freebsd_getattr,
7146 	.vop_setattr =		zfs_freebsd_setattr,
7147 	.vop_create =		zfs_freebsd_create,
7148 	.vop_mknod =		zfs_freebsd_create,
7149 	.vop_mkdir =		zfs_freebsd_mkdir,
7150 	.vop_readdir =		zfs_freebsd_readdir,
7151 	.vop_fsync =		zfs_freebsd_fsync,
7152 	.vop_open =		zfs_freebsd_open,
7153 	.vop_close =		zfs_freebsd_close,
7154 	.vop_rmdir =		zfs_freebsd_rmdir,
7155 	.vop_ioctl =		zfs_freebsd_ioctl,
7156 	.vop_link =		zfs_freebsd_link,
7157 	.vop_symlink =		zfs_freebsd_symlink,
7158 	.vop_readlink =		zfs_freebsd_readlink,
7159 	.vop_read =		zfs_freebsd_read,
7160 	.vop_write =		zfs_freebsd_write,
7161 	.vop_remove =		zfs_freebsd_remove,
7162 	.vop_rename =		zfs_freebsd_rename,
7163 	.vop_pathconf =		zfs_freebsd_pathconf,
7164 	.vop_bmap =		zfs_freebsd_bmap,
7165 	.vop_fid =		zfs_freebsd_fid,
7166 	.vop_getextattr =	zfs_getextattr,
7167 	.vop_deleteextattr =	zfs_deleteextattr,
7168 	.vop_setextattr =	zfs_setextattr,
7169 	.vop_listextattr =	zfs_listextattr,
7170 	.vop_getacl =		zfs_freebsd_getacl,
7171 	.vop_setacl =		zfs_freebsd_setacl,
7172 	.vop_aclcheck =		zfs_freebsd_aclcheck,
7173 	.vop_getpages =		zfs_freebsd_getpages,
7174 	.vop_putpages =		zfs_freebsd_putpages,
7175 };
7176 
7177 struct vop_vector zfs_fifoops = {
7178 	.vop_default =		&fifo_specops,
7179 	.vop_fsync =		zfs_freebsd_fsync,
7180 	.vop_access =		zfs_freebsd_access,
7181 	.vop_getattr =		zfs_freebsd_getattr,
7182 	.vop_inactive =		zfs_freebsd_inactive,
7183 	.vop_read =		VOP_PANIC,
7184 	.vop_reclaim =		zfs_freebsd_reclaim,
7185 	.vop_setattr =		zfs_freebsd_setattr,
7186 	.vop_write =		VOP_PANIC,
7187 	.vop_pathconf = 	zfs_freebsd_fifo_pathconf,
7188 	.vop_fid =		zfs_freebsd_fid,
7189 	.vop_getacl =		zfs_freebsd_getacl,
7190 	.vop_setacl =		zfs_freebsd_setacl,
7191 	.vop_aclcheck =		zfs_freebsd_aclcheck,
7192 };
7193 
7194 /*
7195  * special share hidden files vnode operations template
7196  */
7197 struct vop_vector zfs_shareops = {
7198 	.vop_default =		&default_vnodeops,
7199 	.vop_access =		zfs_freebsd_access,
7200 	.vop_inactive =		zfs_freebsd_inactive,
7201 	.vop_reclaim =		zfs_freebsd_reclaim,
7202 	.vop_fid =		zfs_freebsd_fid,
7203 	.vop_pathconf =		zfs_freebsd_pathconf,
7204 };
7205