xref: /freebsd/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_vnops_os.c (revision 738a9a78fc92e035eb168b9d0a9ada146a9e9972)
1 // SPDX-License-Identifier: CDDL-1.0
2 /*
3  * CDDL HEADER START
4  *
5  * The contents of this file are subject to the terms of the
6  * Common Development and Distribution License (the "License").
7  * You may not use this file except in compliance with the License.
8  *
9  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10  * or https://opensource.org/licenses/CDDL-1.0.
11  * See the License for the specific language governing permissions
12  * and limitations under the License.
13  *
14  * When distributing Covered Code, include this CDDL HEADER in each
15  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16  * If applicable, add the following below this CDDL HEADER, with the
17  * fields enclosed by brackets "[]" replaced with your own identifying
18  * information: Portions Copyright [yyyy] [name of copyright owner]
19  *
20  * CDDL HEADER END
21  */
22 
23 /*
24  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
25  * Copyright (c) 2012, 2015 by Delphix. All rights reserved.
26  * Copyright (c) 2014 Integros [integros.com]
27  * Copyright 2017 Nexenta Systems, Inc.
28  */
29 
30 /* Portions Copyright 2007 Jeremy Teo */
31 /* Portions Copyright 2010 Robert Milkowski */
32 
33 #include <sys/param.h>
34 #include <sys/time.h>
35 #include <sys/systm.h>
36 #include <sys/sysmacros.h>
37 #include <sys/resource.h>
38 #include <security/mac/mac_framework.h>
39 #include <sys/vfs.h>
40 #include <sys/endian.h>
41 #include <sys/vm.h>
42 #include <sys/vnode.h>
43 #include <sys/smr.h>
44 #include <sys/dirent.h>
45 #include <sys/file.h>
46 #include <sys/stat.h>
47 #include <sys/kmem.h>
48 #include <sys/taskq.h>
49 #include <sys/uio.h>
50 #include <sys/atomic.h>
51 #include <sys/namei.h>
52 #include <sys/mman.h>
53 #include <sys/cmn_err.h>
54 #include <sys/kdb.h>
55 #include <sys/sysproto.h>
56 #include <sys/errno.h>
57 #include <sys/unistd.h>
58 #include <sys/zfs_dir.h>
59 #include <sys/zfs_ioctl.h>
60 #include <sys/fs/zfs.h>
61 #include <sys/dmu.h>
62 #include <sys/dmu_objset.h>
63 #include <sys/spa.h>
64 #include <sys/txg.h>
65 #include <sys/dbuf.h>
66 #include <sys/zap.h>
67 #include <sys/sa.h>
68 #include <sys/policy.h>
69 #include <sys/sunddi.h>
70 #include <sys/filio.h>
71 #include <sys/sid.h>
72 #include <sys/zfs_ctldir.h>
73 #include <sys/zfs_fuid.h>
74 #include <sys/zfs_quota.h>
75 #include <sys/zfs_sa.h>
76 #include <sys/zfs_rlock.h>
77 #include <sys/zfs_project.h>
78 #include <sys/bio.h>
79 #include <sys/buf.h>
80 #include <sys/sched.h>
81 #include <sys/acl.h>
82 #include <sys/vmmeter.h>
83 #include <vm/vm_param.h>
84 #include <sys/zil.h>
85 #include <sys/zfs_vnops.h>
86 #include <sys/module.h>
87 #include <sys/sysent.h>
88 #include <sys/dmu_impl.h>
89 #include <sys/brt.h>
90 #include <sys/zfeature.h>
91 
92 #include <vm/vm_object.h>
93 
94 #include <sys/extattr.h>
95 #include <sys/priv.h>
96 
97 #ifndef VN_OPEN_INVFS
98 #define	VN_OPEN_INVFS	0x0
99 #endif
100 
101 VFS_SMR_DECLARE;
102 
103 #ifdef DEBUG_VFS_LOCKS
104 #define	VNCHECKREF(vp)				  \
105 	VNASSERT((vp)->v_holdcnt > 0 && (vp)->v_usecount > 0, vp,	\
106 	    ("%s: wrong ref counts", __func__));
107 #else
108 #define	VNCHECKREF(vp)
109 #endif
110 
111 #if __FreeBSD_version >= 1400045
112 typedef uint64_t cookie_t;
113 #else
114 typedef ulong_t cookie_t;
115 #endif
116 
117 /*
118  * Programming rules.
119  *
120  * Each vnode op performs some logical unit of work.  To do this, the ZPL must
121  * properly lock its in-core state, create a DMU transaction, do the work,
122  * record this work in the intent log (ZIL), commit the DMU transaction,
123  * and wait for the intent log to commit if it is a synchronous operation.
124  * Moreover, the vnode ops must work in both normal and log replay context.
125  * The ordering of events is important to avoid deadlocks and references
126  * to freed memory.  The example below illustrates the following Big Rules:
127  *
128  *  (1)	A check must be made in each zfs thread for a mounted file system.
129  *	This is done avoiding races using zfs_enter(zfsvfs).
130  *	A zfs_exit(zfsvfs) is needed before all returns.  Any znodes
131  *	must be checked with zfs_verify_zp(zp).  Both of these macros
132  *	can return EIO from the calling function.
133  *
134  *  (2)	VN_RELE() should always be the last thing except for zil_commit()
135  *	(if necessary) and zfs_exit(). This is for 3 reasons:
136  *	First, if it's the last reference, the vnode/znode
137  *	can be freed, so the zp may point to freed memory.  Second, the last
138  *	reference will call zfs_zinactive(), which may induce a lot of work --
139  *	pushing cached pages (which acquires range locks) and syncing out
140  *	cached atime changes.  Third, zfs_zinactive() may require a new tx,
141  *	which could deadlock the system if you were already holding one.
142  *	If you must call VN_RELE() within a tx then use VN_RELE_ASYNC().
143  *
144  *  (3)	All range locks must be grabbed before calling dmu_tx_assign(),
145  *	as they can span dmu_tx_assign() calls.
146  *
147  *  (4) If ZPL locks are held, pass DMU_TX_NOWAIT as the second argument to
148  *      dmu_tx_assign().  This is critical because we don't want to block
149  *      while holding locks.
150  *
151  *	If no ZPL locks are held (aside from zfs_enter()), use DMU_TX_WAIT.
152  *	This reduces lock contention and CPU usage when we must wait (note
153  *	that if throughput is constrained by the storage, nearly every
154  *	transaction must wait).
155  *
156  *      Note, in particular, that if a lock is sometimes acquired before
157  *      the tx assigns, and sometimes after (e.g. z_lock), then failing
158  *      to use a non-blocking assign can deadlock the system.  The scenario:
159  *
160  *	Thread A has grabbed a lock before calling dmu_tx_assign().
161  *	Thread B is in an already-assigned tx, and blocks for this lock.
162  *	Thread A calls dmu_tx_assign(DMU_TX_WAIT) and blocks in
163  *	txg_wait_open() forever, because the previous txg can't quiesce
164  *	until B's tx commits.
165  *
166  *	If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is
167  *	DMU_TX_NOWAIT, then drop all locks, call dmu_tx_wait(), and try
168  *	again.  On subsequent calls to dmu_tx_assign(), pass
169  *	DMU_TX_NOTHROTTLE in addition to DMU_TX_NOWAIT, to indicate that
170  *	this operation has already called dmu_tx_wait().  This will ensure
171  *	that we don't retry forever, waiting a short bit each time.
172  *
173  *  (5)	If the operation succeeded, generate the intent log entry for it
174  *	before dropping locks.  This ensures that the ordering of events
175  *	in the intent log matches the order in which they actually occurred.
176  *	During ZIL replay the zfs_log_* functions will update the sequence
177  *	number to indicate the zil transaction has replayed.
178  *
179  *  (6)	At the end of each vnode op, the DMU tx must always commit,
180  *	regardless of whether there were any errors.
181  *
182  *  (7)	After dropping all locks, invoke zil_commit(zilog, foid)
183  *	to ensure that synchronous semantics are provided when necessary.
184  *
185  * In general, this is how things should be ordered in each vnode op:
186  *
187  *	zfs_enter(zfsvfs);		// exit if unmounted
188  * top:
189  *	zfs_dirent_lookup(&dl, ...)	// lock directory entry (may VN_HOLD())
190  *	rw_enter(...);			// grab any other locks you need
191  *	tx = dmu_tx_create(...);	// get DMU tx
192  *	dmu_tx_hold_*();		// hold each object you might modify
193  *	error = dmu_tx_assign(tx,
194  *	    (waited ? DMU_TX_NOTHROTTLE : 0) | DMU_TX_NOWAIT);
195  *	if (error) {
196  *		rw_exit(...);		// drop locks
197  *		zfs_dirent_unlock(dl);	// unlock directory entry
198  *		VN_RELE(...);		// release held vnodes
199  *		if (error == ERESTART) {
200  *			waited = B_TRUE;
201  *			dmu_tx_wait(tx);
202  *			dmu_tx_abort(tx);
203  *			goto top;
204  *		}
205  *		dmu_tx_abort(tx);	// abort DMU tx
206  *		zfs_exit(zfsvfs);	// finished in zfs
207  *		return (error);		// really out of space
208  *	}
209  *	error = do_real_work();		// do whatever this VOP does
210  *	if (error == 0)
211  *		zfs_log_*(...);		// on success, make ZIL entry
212  *	dmu_tx_commit(tx);		// commit DMU tx -- error or not
213  *	rw_exit(...);			// drop locks
214  *	zfs_dirent_unlock(dl);		// unlock directory entry
215  *	VN_RELE(...);			// release held vnodes
216  *	zil_commit(zilog, foid);	// synchronous when necessary
217  *	zfs_exit(zfsvfs);		// finished in zfs
218  *	return (error);			// done, report error
219  */
220 static int
zfs_open(vnode_t ** vpp,int flag,cred_t * cr)221 zfs_open(vnode_t **vpp, int flag, cred_t *cr)
222 {
223 	(void) cr;
224 	znode_t	*zp = VTOZ(*vpp);
225 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
226 	int error;
227 
228 	if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0)
229 		return (error);
230 
231 	if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) &&
232 	    ((flag & FAPPEND) == 0)) {
233 		zfs_exit(zfsvfs, FTAG);
234 		return (SET_ERROR(EPERM));
235 	}
236 
237 	/*
238 	 * Keep a count of the synchronous opens in the znode.  On first
239 	 * synchronous open we must convert all previous async transactions
240 	 * into sync to keep correct ordering.
241 	 */
242 	if (flag & O_SYNC) {
243 		if (atomic_inc_32_nv(&zp->z_sync_cnt) == 1)
244 			zil_async_to_sync(zfsvfs->z_log, zp->z_id);
245 	}
246 
247 	zfs_exit(zfsvfs, FTAG);
248 	return (0);
249 }
250 
251 static int
zfs_close(vnode_t * vp,int flag,int count,offset_t offset,cred_t * cr)252 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr)
253 {
254 	(void) offset, (void) cr;
255 	znode_t	*zp = VTOZ(vp);
256 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
257 	int error;
258 
259 	if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0)
260 		return (error);
261 
262 	/* Decrement the synchronous opens in the znode */
263 	if ((flag & O_SYNC) && (count == 1))
264 		atomic_dec_32(&zp->z_sync_cnt);
265 
266 	zfs_exit(zfsvfs, FTAG);
267 	return (0);
268 }
269 
270 static int
zfs_ioctl_getxattr(vnode_t * vp,zfsxattr_t * fsx)271 zfs_ioctl_getxattr(vnode_t *vp, zfsxattr_t *fsx)
272 {
273 	znode_t *zp = VTOZ(vp);
274 
275 	memset(fsx, 0, sizeof (*fsx));
276 	fsx->fsx_xflags = (zp->z_pflags & ZFS_PROJINHERIT) ?
277 	    ZFS_PROJINHERIT_FL : 0;
278 	fsx->fsx_projid = zp->z_projid;
279 
280 	return (0);
281 }
282 
283 static int
zfs_ioctl_setflags(vnode_t * vp,uint32_t ioctl_flags,xvattr_t * xva)284 zfs_ioctl_setflags(vnode_t *vp, uint32_t ioctl_flags, xvattr_t *xva)
285 {
286 	uint64_t zfs_flags = VTOZ(vp)->z_pflags;
287 	xoptattr_t *xoap;
288 
289 	if (ioctl_flags & ~(ZFS_PROJINHERIT_FL))
290 		return (SET_ERROR(EOPNOTSUPP));
291 
292 	xva_init(xva);
293 	xoap = xva_getxoptattr(xva);
294 
295 #define	FLAG_CHANGE(iflag, zflag, xflag, xfield)	do {		\
296 	if (((ioctl_flags & (iflag)) && !(zfs_flags & (zflag))) ||	\
297 	    ((zfs_flags & (zflag)) && !(ioctl_flags & (iflag)))) {	\
298 		XVA_SET_REQ(xva, (xflag));				\
299 		(xfield) = ((ioctl_flags & (iflag)) != 0);		\
300 	}								\
301 } while (0)
302 
303 	FLAG_CHANGE(ZFS_PROJINHERIT_FL, ZFS_PROJINHERIT, XAT_PROJINHERIT,
304 	    xoap->xoa_projinherit);
305 
306 #undef	FLAG_CHANGE
307 
308 	return (0);
309 }
310 
311 static int
zfs_ioctl_setxattr(vnode_t * vp,zfsxattr_t * fsx,cred_t * cr)312 zfs_ioctl_setxattr(vnode_t *vp, zfsxattr_t *fsx, cred_t *cr)
313 {
314 	znode_t *zp = VTOZ(vp);
315 	xvattr_t xva;
316 	xoptattr_t *xoap;
317 	int err;
318 
319 	if (!zpl_is_valid_projid(fsx->fsx_projid))
320 		return (SET_ERROR(EINVAL));
321 
322 	err = zfs_ioctl_setflags(vp, fsx->fsx_xflags, &xva);
323 	if (err)
324 		return (err);
325 
326 	xoap = xva_getxoptattr(&xva);
327 	XVA_SET_REQ(&xva, XAT_PROJID);
328 	xoap->xoa_projid = fsx->fsx_projid;
329 
330 	err = zfs_setattr(zp, (vattr_t *)&xva, 0, cr, NULL);
331 
332 	return (err);
333 }
334 
335 static int
zfs_ioctl(vnode_t * vp,ulong_t com,intptr_t data,int flag,cred_t * cred,int * rvalp)336 zfs_ioctl(vnode_t *vp, ulong_t com, intptr_t data, int flag, cred_t *cred,
337     int *rvalp)
338 {
339 	(void) flag, (void) cred, (void) rvalp;
340 	loff_t off;
341 	int error;
342 
343 	switch (com) {
344 	case _FIOFFS:
345 	{
346 		return (0);
347 
348 		/*
349 		 * The following two ioctls are used by bfu.  Faking out,
350 		 * necessary to avoid bfu errors.
351 		 */
352 	}
353 	case _FIOGDIO:
354 	case _FIOSDIO:
355 	{
356 		return (0);
357 	}
358 
359 	case F_SEEK_DATA:
360 	case F_SEEK_HOLE:
361 	{
362 		off = *(offset_t *)data;
363 		error = vn_lock(vp, LK_SHARED);
364 		if (error)
365 			return (error);
366 		/* offset parameter is in/out */
367 		error = zfs_holey(VTOZ(vp), com, &off);
368 		VOP_UNLOCK(vp);
369 		if (error)
370 			return (error);
371 		*(offset_t *)data = off;
372 		return (0);
373 	}
374 	case ZFS_IOC_FSGETXATTR: {
375 		zfsxattr_t *fsx = (zfsxattr_t *)data;
376 		error = vn_lock(vp, LK_SHARED);
377 		if (error)
378 			return (error);
379 		error = zfs_ioctl_getxattr(vp, fsx);
380 		VOP_UNLOCK(vp);
381 		return (error);
382 	}
383 	case ZFS_IOC_FSSETXATTR: {
384 		zfsxattr_t *fsx = (zfsxattr_t *)data;
385 		error = vn_lock(vp, LK_EXCLUSIVE);
386 		if (error)
387 			return (error);
388 		error = zfs_ioctl_setxattr(vp, fsx, cred);
389 		VOP_UNLOCK(vp);
390 		return (error);
391 	}
392 	case ZFS_IOC_REWRITE: {
393 		zfs_rewrite_args_t *args = (zfs_rewrite_args_t *)data;
394 		if ((flag & FWRITE) == 0)
395 			return (SET_ERROR(EBADF));
396 		error = vn_lock(vp, LK_SHARED);
397 		if (error)
398 			return (error);
399 		error = zfs_rewrite(VTOZ(vp), args->off, args->len,
400 		    args->flags, args->arg);
401 		VOP_UNLOCK(vp);
402 		return (error);
403 	}
404 	}
405 	return (SET_ERROR(ENOTTY));
406 }
407 
408 static vm_page_t
page_busy(vnode_t * vp,int64_t start,int64_t off,int64_t nbytes)409 page_busy(vnode_t *vp, int64_t start, int64_t off, int64_t nbytes)
410 {
411 	vm_object_t obj;
412 	vm_page_t pp;
413 	int64_t end;
414 
415 	/*
416 	 * At present vm_page_clear_dirty extends the cleared range to DEV_BSIZE
417 	 * aligned boundaries, if the range is not aligned.  As a result a
418 	 * DEV_BSIZE subrange with partially dirty data may get marked as clean.
419 	 * It may happen that all DEV_BSIZE subranges are marked clean and thus
420 	 * the whole page would be considered clean despite have some
421 	 * dirty data.
422 	 * For this reason we should shrink the range to DEV_BSIZE aligned
423 	 * boundaries before calling vm_page_clear_dirty.
424 	 */
425 	end = rounddown2(off + nbytes, DEV_BSIZE);
426 	off = roundup2(off, DEV_BSIZE);
427 	nbytes = end - off;
428 
429 	obj = vp->v_object;
430 	vm_page_grab_valid_unlocked(&pp, obj, OFF_TO_IDX(start),
431 	    VM_ALLOC_NOCREAT | VM_ALLOC_SBUSY | VM_ALLOC_NORMAL |
432 	    VM_ALLOC_IGN_SBUSY);
433 	if (pp != NULL) {
434 		ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
435 		vm_object_pip_add(obj, 1);
436 		pmap_remove_write(pp);
437 		if (nbytes != 0)
438 			vm_page_clear_dirty(pp, off, nbytes);
439 	}
440 	return (pp);
441 }
442 
443 static void
page_unbusy(vm_page_t pp)444 page_unbusy(vm_page_t pp)
445 {
446 
447 	vm_page_sunbusy(pp);
448 	vm_object_pip_wakeup(pp->object);
449 }
450 
451 static vm_page_t
page_hold(vnode_t * vp,int64_t start)452 page_hold(vnode_t *vp, int64_t start)
453 {
454 	vm_object_t obj;
455 	vm_page_t m;
456 
457 	obj = vp->v_object;
458 	vm_page_grab_valid_unlocked(&m, obj, OFF_TO_IDX(start),
459 	    VM_ALLOC_NOCREAT | VM_ALLOC_WIRED | VM_ALLOC_IGN_SBUSY |
460 	    VM_ALLOC_NOBUSY);
461 	return (m);
462 }
463 
464 static void
page_unhold(vm_page_t pp)465 page_unhold(vm_page_t pp)
466 {
467 	vm_page_unwire(pp, PQ_ACTIVE);
468 }
469 
470 /*
471  * When a file is memory mapped, we must keep the IO data synchronized
472  * between the DMU cache and the memory mapped pages.  What this means:
473  *
474  * On Write:	If we find a memory mapped page, we write to *both*
475  *		the page and the dmu buffer.
476  */
477 void
update_pages(znode_t * zp,int64_t start,int len,objset_t * os)478 update_pages(znode_t *zp, int64_t start, int len, objset_t *os)
479 {
480 	vm_object_t obj;
481 	struct sf_buf *sf;
482 	vnode_t *vp = ZTOV(zp);
483 	caddr_t va;
484 	int off;
485 
486 	ASSERT3P(vp->v_mount, !=, NULL);
487 	obj = vp->v_object;
488 	ASSERT3P(obj, !=, NULL);
489 
490 	off = start & PAGEOFFSET;
491 	vm_object_pip_add(obj, 1);
492 	for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
493 		vm_page_t pp;
494 		int nbytes = imin(PAGESIZE - off, len);
495 
496 		if ((pp = page_busy(vp, start, off, nbytes)) != NULL) {
497 			va = zfs_map_page(pp, &sf);
498 			(void) dmu_read(os, zp->z_id, start + off, nbytes,
499 			    va + off, DMU_READ_PREFETCH);
500 			zfs_unmap_page(sf);
501 			page_unbusy(pp);
502 		}
503 		len -= nbytes;
504 		off = 0;
505 	}
506 	vm_object_pip_wakeup(obj);
507 }
508 
509 /*
510  * Read with UIO_NOCOPY flag means that sendfile(2) requests
511  * ZFS to populate a range of page cache pages with data.
512  *
513  * NOTE: this function could be optimized to pre-allocate
514  * all pages in advance, drain exclusive busy on all of them,
515  * map them into contiguous KVA region and populate them
516  * in one single dmu_read() call.
517  */
518 int
mappedread_sf(znode_t * zp,int nbytes,zfs_uio_t * uio)519 mappedread_sf(znode_t *zp, int nbytes, zfs_uio_t *uio)
520 {
521 	vnode_t *vp = ZTOV(zp);
522 	objset_t *os = zp->z_zfsvfs->z_os;
523 	struct sf_buf *sf;
524 	vm_object_t obj;
525 	vm_page_t pp;
526 	int64_t start;
527 	caddr_t va;
528 	int len = nbytes;
529 	int error = 0;
530 
531 	ASSERT3U(zfs_uio_segflg(uio), ==, UIO_NOCOPY);
532 	ASSERT3P(vp->v_mount, !=, NULL);
533 	obj = vp->v_object;
534 	ASSERT3P(obj, !=, NULL);
535 	ASSERT0(zfs_uio_offset(uio) & PAGEOFFSET);
536 
537 	for (start = zfs_uio_offset(uio); len > 0; start += PAGESIZE) {
538 		int bytes = MIN(PAGESIZE, len);
539 
540 		pp = vm_page_grab_unlocked(obj, OFF_TO_IDX(start),
541 		    VM_ALLOC_SBUSY | VM_ALLOC_NORMAL | VM_ALLOC_IGN_SBUSY);
542 		if (vm_page_none_valid(pp)) {
543 			va = zfs_map_page(pp, &sf);
544 			error = dmu_read(os, zp->z_id, start, bytes, va,
545 			    DMU_READ_PREFETCH);
546 			if (bytes != PAGESIZE && error == 0)
547 				memset(va + bytes, 0, PAGESIZE - bytes);
548 			zfs_unmap_page(sf);
549 			if (error == 0) {
550 				vm_page_valid(pp);
551 				vm_page_activate(pp);
552 				vm_page_sunbusy(pp);
553 			} else {
554 				zfs_vmobject_wlock(obj);
555 				if (!vm_page_wired(pp) && pp->valid == 0 &&
556 				    vm_page_busy_tryupgrade(pp))
557 					vm_page_free(pp);
558 				else {
559 					vm_page_deactivate_noreuse(pp);
560 					vm_page_sunbusy(pp);
561 				}
562 				zfs_vmobject_wunlock(obj);
563 			}
564 		} else {
565 			ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
566 			vm_page_sunbusy(pp);
567 		}
568 		if (error)
569 			break;
570 		zfs_uio_advance(uio, bytes);
571 		len -= bytes;
572 	}
573 	return (error);
574 }
575 
576 /*
577  * When a file is memory mapped, we must keep the IO data synchronized
578  * between the DMU cache and the memory mapped pages.  What this means:
579  *
580  * On Read:	We "read" preferentially from memory mapped pages,
581  *		else we default from the dmu buffer.
582  *
583  * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
584  *	 the file is memory mapped.
585  */
586 int
mappedread(znode_t * zp,int nbytes,zfs_uio_t * uio)587 mappedread(znode_t *zp, int nbytes, zfs_uio_t *uio)
588 {
589 	vnode_t *vp = ZTOV(zp);
590 	vm_object_t obj;
591 	int64_t start;
592 	int len = nbytes;
593 	int off;
594 	int error = 0;
595 
596 	ASSERT3P(vp->v_mount, !=, NULL);
597 	obj = vp->v_object;
598 	ASSERT3P(obj, !=, NULL);
599 
600 	start = zfs_uio_offset(uio);
601 	off = start & PAGEOFFSET;
602 	for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
603 		vm_page_t pp;
604 		uint64_t bytes = MIN(PAGESIZE - off, len);
605 
606 		if ((pp = page_hold(vp, start))) {
607 			struct sf_buf *sf;
608 			caddr_t va;
609 
610 			va = zfs_map_page(pp, &sf);
611 			error = vn_io_fault_uiomove(va + off, bytes,
612 			    GET_UIO_STRUCT(uio));
613 			zfs_unmap_page(sf);
614 			page_unhold(pp);
615 		} else {
616 			error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl),
617 			    uio, bytes, DMU_READ_PREFETCH);
618 		}
619 		len -= bytes;
620 		off = 0;
621 		if (error)
622 			break;
623 	}
624 	return (error);
625 }
626 
627 int
zfs_write_simple(znode_t * zp,const void * data,size_t len,loff_t pos,size_t * presid)628 zfs_write_simple(znode_t *zp, const void *data, size_t len,
629     loff_t pos, size_t *presid)
630 {
631 	int error = 0;
632 	ssize_t resid;
633 
634 	error = vn_rdwr(UIO_WRITE, ZTOV(zp), __DECONST(void *, data), len, pos,
635 	    UIO_SYSSPACE, IO_SYNC, kcred, NOCRED, &resid, curthread);
636 
637 	if (error) {
638 		return (SET_ERROR(error));
639 	} else if (presid == NULL) {
640 		if (resid != 0) {
641 			error = SET_ERROR(EIO);
642 		}
643 	} else {
644 		*presid = resid;
645 	}
646 	return (error);
647 }
648 
649 void
zfs_zrele_async(znode_t * zp)650 zfs_zrele_async(znode_t *zp)
651 {
652 	vnode_t *vp = ZTOV(zp);
653 	objset_t *os = ITOZSB(vp)->z_os;
654 
655 	VN_RELE_ASYNC(vp, dsl_pool_zrele_taskq(dmu_objset_pool(os)));
656 }
657 
658 static int
zfs_dd_callback(struct mount * mp,void * arg,int lkflags,struct vnode ** vpp)659 zfs_dd_callback(struct mount *mp, void *arg, int lkflags, struct vnode **vpp)
660 {
661 	int error;
662 
663 	*vpp = arg;
664 	error = vn_lock(*vpp, lkflags);
665 	if (error != 0)
666 		vrele(*vpp);
667 	return (error);
668 }
669 
670 static int
zfs_lookup_lock(vnode_t * dvp,vnode_t * vp,const char * name,int lkflags)671 zfs_lookup_lock(vnode_t *dvp, vnode_t *vp, const char *name, int lkflags)
672 {
673 	znode_t *zdp = VTOZ(dvp);
674 	zfsvfs_t *zfsvfs __unused = zdp->z_zfsvfs;
675 	int error;
676 	int ltype;
677 
678 	if (zfsvfs->z_replay == B_FALSE)
679 		ASSERT_VOP_LOCKED(dvp, __func__);
680 
681 	if (name[0] == 0 || (name[0] == '.' && name[1] == 0)) {
682 		ASSERT3P(dvp, ==, vp);
683 		vref(dvp);
684 		ltype = lkflags & LK_TYPE_MASK;
685 		if (ltype != VOP_ISLOCKED(dvp)) {
686 			if (ltype == LK_EXCLUSIVE)
687 				vn_lock(dvp, LK_UPGRADE | LK_RETRY);
688 			else /* if (ltype == LK_SHARED) */
689 				vn_lock(dvp, LK_DOWNGRADE | LK_RETRY);
690 
691 			/*
692 			 * Relock for the "." case could leave us with
693 			 * reclaimed vnode.
694 			 */
695 			if (VN_IS_DOOMED(dvp)) {
696 				vrele(dvp);
697 				return (SET_ERROR(ENOENT));
698 			}
699 		}
700 		return (0);
701 	} else if (name[0] == '.' && name[1] == '.' && name[2] == 0) {
702 		/*
703 		 * Note that in this case, dvp is the child vnode, and we
704 		 * are looking up the parent vnode - exactly reverse from
705 		 * normal operation.  Unlocking dvp requires some rather
706 		 * tricky unlock/relock dance to prevent mp from being freed;
707 		 * use vn_vget_ino_gen() which takes care of all that.
708 		 *
709 		 * XXX Note that there is a time window when both vnodes are
710 		 * unlocked.  It is possible, although highly unlikely, that
711 		 * during that window the parent-child relationship between
712 		 * the vnodes may change, for example, get reversed.
713 		 * In that case we would have a wrong lock order for the vnodes.
714 		 * All other filesystems seem to ignore this problem, so we
715 		 * do the same here.
716 		 * A potential solution could be implemented as follows:
717 		 * - using LK_NOWAIT when locking the second vnode and retrying
718 		 *   if necessary
719 		 * - checking that the parent-child relationship still holds
720 		 *   after locking both vnodes and retrying if it doesn't
721 		 */
722 		error = vn_vget_ino_gen(dvp, zfs_dd_callback, vp, lkflags, &vp);
723 		return (error);
724 	} else {
725 		error = vn_lock(vp, lkflags);
726 		if (error != 0)
727 			vrele(vp);
728 		return (error);
729 	}
730 }
731 
732 /*
733  * Lookup an entry in a directory, or an extended attribute directory.
734  * If it exists, return a held vnode reference for it.
735  *
736  *	IN:	dvp	- vnode of directory to search.
737  *		nm	- name of entry to lookup.
738  *		pnp	- full pathname to lookup [UNUSED].
739  *		flags	- LOOKUP_XATTR set if looking for an attribute.
740  *		rdir	- root directory vnode [UNUSED].
741  *		cr	- credentials of caller.
742  *		ct	- caller context
743  *
744  *	OUT:	vpp	- vnode of located entry, NULL if not found.
745  *
746  *	RETURN:	0 on success, error code on failure.
747  *
748  * Timestamps:
749  *	NA
750  */
751 static int
zfs_lookup(vnode_t * dvp,const char * nm,vnode_t ** vpp,struct componentname * cnp,int nameiop,cred_t * cr,int flags,boolean_t cached)752 zfs_lookup(vnode_t *dvp, const char *nm, vnode_t **vpp,
753     struct componentname *cnp, int nameiop, cred_t *cr, int flags,
754     boolean_t cached)
755 {
756 	znode_t *zdp = VTOZ(dvp);
757 	znode_t *zp;
758 	zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
759 	seqc_t dvp_seqc;
760 	int	error = 0;
761 
762 	/*
763 	 * Fast path lookup, however we must skip DNLC lookup
764 	 * for case folding or normalizing lookups because the
765 	 * DNLC code only stores the passed in name.  This means
766 	 * creating 'a' and removing 'A' on a case insensitive
767 	 * file system would work, but DNLC still thinks 'a'
768 	 * exists and won't let you create it again on the next
769 	 * pass through fast path.
770 	 */
771 	if (!(flags & LOOKUP_XATTR)) {
772 		if (dvp->v_type != VDIR) {
773 			return (SET_ERROR(ENOTDIR));
774 		} else if (zdp->z_sa_hdl == NULL) {
775 			return (SET_ERROR(EIO));
776 		}
777 	}
778 
779 	DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp,
780 	    const char *, nm);
781 
782 	if ((error = zfs_enter_verify_zp(zfsvfs, zdp, FTAG)) != 0)
783 		return (error);
784 
785 	dvp_seqc = vn_seqc_read_notmodify(dvp);
786 
787 	*vpp = NULL;
788 
789 	if (flags & LOOKUP_XATTR) {
790 		/*
791 		 * If the xattr property is off, refuse the lookup request.
792 		 */
793 		if (!(zfsvfs->z_flags & ZSB_XATTR)) {
794 			zfs_exit(zfsvfs, FTAG);
795 			return (SET_ERROR(EOPNOTSUPP));
796 		}
797 
798 		/*
799 		 * We don't allow recursive attributes..
800 		 * Maybe someday we will.
801 		 */
802 		if (zdp->z_pflags & ZFS_XATTR) {
803 			zfs_exit(zfsvfs, FTAG);
804 			return (SET_ERROR(EINVAL));
805 		}
806 
807 		if ((error = zfs_get_xattrdir(VTOZ(dvp), &zp, cr, flags))) {
808 			zfs_exit(zfsvfs, FTAG);
809 			return (error);
810 		}
811 		*vpp = ZTOV(zp);
812 
813 		/*
814 		 * Do we have permission to get into attribute directory?
815 		 */
816 		error = zfs_zaccess(zp, ACE_EXECUTE, 0, B_FALSE, cr, NULL);
817 		if (error) {
818 			vrele(ZTOV(zp));
819 		}
820 
821 		zfs_exit(zfsvfs, FTAG);
822 		return (error);
823 	}
824 
825 	/*
826 	 * Check accessibility of directory if we're not coming in via
827 	 * VOP_CACHEDLOOKUP.
828 	 */
829 	if (!cached) {
830 #ifdef NOEXECCHECK
831 		if ((cnp->cn_flags & NOEXECCHECK) != 0) {
832 			cnp->cn_flags &= ~NOEXECCHECK;
833 		} else
834 #endif
835 		if ((error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr,
836 		    NULL))) {
837 			zfs_exit(zfsvfs, FTAG);
838 			return (error);
839 		}
840 	}
841 
842 	if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
843 	    NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
844 		zfs_exit(zfsvfs, FTAG);
845 		return (SET_ERROR(EILSEQ));
846 	}
847 
848 
849 	/*
850 	 * First handle the special cases.
851 	 */
852 	if ((cnp->cn_flags & ISDOTDOT) != 0) {
853 		/*
854 		 * If we are a snapshot mounted under .zfs, return
855 		 * the vp for the snapshot directory.
856 		 */
857 		if (zdp->z_id == zfsvfs->z_root && zfsvfs->z_parent != zfsvfs) {
858 			struct componentname cn;
859 			vnode_t *zfsctl_vp;
860 			int ltype;
861 
862 			zfs_exit(zfsvfs, FTAG);
863 			ltype = VOP_ISLOCKED(dvp);
864 			VOP_UNLOCK(dvp);
865 			error = zfsctl_root(zfsvfs->z_parent, LK_SHARED,
866 			    &zfsctl_vp);
867 			if (error == 0) {
868 				cn.cn_nameptr = "snapshot";
869 				cn.cn_namelen = strlen(cn.cn_nameptr);
870 				cn.cn_nameiop = cnp->cn_nameiop;
871 				cn.cn_flags = cnp->cn_flags & ~ISDOTDOT;
872 				cn.cn_lkflags = cnp->cn_lkflags;
873 				error = VOP_LOOKUP(zfsctl_vp, vpp, &cn);
874 				vput(zfsctl_vp);
875 			}
876 			vn_lock(dvp, ltype | LK_RETRY);
877 			return (error);
878 		}
879 	}
880 	if (zfs_has_ctldir(zdp) && strcmp(nm, ZFS_CTLDIR_NAME) == 0) {
881 		zfs_exit(zfsvfs, FTAG);
882 		if (zfsvfs->z_show_ctldir == ZFS_SNAPDIR_DISABLED)
883 			return (SET_ERROR(ENOENT));
884 		if ((cnp->cn_flags & ISLASTCN) != 0 && nameiop != LOOKUP)
885 			return (SET_ERROR(ENOTSUP));
886 		error = zfsctl_root(zfsvfs, cnp->cn_lkflags, vpp);
887 		return (error);
888 	}
889 
890 	/*
891 	 * The loop is retry the lookup if the parent-child relationship
892 	 * changes during the dot-dot locking complexities.
893 	 */
894 	for (;;) {
895 		uint64_t parent;
896 
897 		error = zfs_dirlook(zdp, nm, &zp);
898 		if (error == 0)
899 			*vpp = ZTOV(zp);
900 
901 		zfs_exit(zfsvfs, FTAG);
902 		if (error != 0)
903 			break;
904 
905 		error = zfs_lookup_lock(dvp, *vpp, nm, cnp->cn_lkflags);
906 		if (error != 0) {
907 			/*
908 			 * If we've got a locking error, then the vnode
909 			 * got reclaimed because of a force unmount.
910 			 * We never enter doomed vnodes into the name cache.
911 			 */
912 			*vpp = NULL;
913 			return (error);
914 		}
915 
916 		if ((cnp->cn_flags & ISDOTDOT) == 0)
917 			break;
918 
919 		if ((error = zfs_enter(zfsvfs, FTAG)) != 0) {
920 			vput(ZTOV(zp));
921 			*vpp = NULL;
922 			return (error);
923 		}
924 		if (zdp->z_sa_hdl == NULL) {
925 			error = SET_ERROR(EIO);
926 		} else {
927 			error = sa_lookup(zdp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
928 			    &parent, sizeof (parent));
929 		}
930 		if (error != 0) {
931 			zfs_exit(zfsvfs, FTAG);
932 			vput(ZTOV(zp));
933 			break;
934 		}
935 		if (zp->z_id == parent) {
936 			zfs_exit(zfsvfs, FTAG);
937 			break;
938 		}
939 		vput(ZTOV(zp));
940 	}
941 
942 	if (error != 0)
943 		*vpp = NULL;
944 
945 	/* Translate errors and add SAVENAME when needed. */
946 	if (cnp->cn_flags & ISLASTCN) {
947 		switch (nameiop) {
948 		case CREATE:
949 		case RENAME:
950 			if (error == ENOENT) {
951 				error = EJUSTRETURN;
952 #if __FreeBSD_version < 1400068
953 				cnp->cn_flags |= SAVENAME;
954 #endif
955 				break;
956 			}
957 			zfs_fallthrough;
958 		case DELETE:
959 #if __FreeBSD_version < 1400068
960 			if (error == 0)
961 				cnp->cn_flags |= SAVENAME;
962 #endif
963 			break;
964 		}
965 	}
966 
967 	if ((cnp->cn_flags & ISDOTDOT) != 0) {
968 		/*
969 		 * FIXME: zfs_lookup_lock relocks vnodes and does nothing to
970 		 * handle races. In particular different callers may end up
971 		 * with different vnodes and will try to add conflicting
972 		 * entries to the namecache.
973 		 *
974 		 * While finding different result may be acceptable in face
975 		 * of concurrent modification, adding conflicting entries
976 		 * trips over an assert in the namecache.
977 		 *
978 		 * Ultimately let an entry through once everything settles.
979 		 */
980 		if (!vn_seqc_consistent(dvp, dvp_seqc)) {
981 			cnp->cn_flags &= ~MAKEENTRY;
982 		}
983 	}
984 
985 	/* Insert name into cache (as non-existent) if appropriate. */
986 	if (zfsvfs->z_use_namecache && !zfsvfs->z_replay &&
987 	    error == ENOENT && (cnp->cn_flags & MAKEENTRY) != 0)
988 		cache_enter(dvp, NULL, cnp);
989 
990 	/* Insert name into cache if appropriate. */
991 	if (zfsvfs->z_use_namecache && !zfsvfs->z_replay &&
992 	    error == 0 && (cnp->cn_flags & MAKEENTRY)) {
993 		if (!(cnp->cn_flags & ISLASTCN) ||
994 		    (nameiop != DELETE && nameiop != RENAME)) {
995 			cache_enter(dvp, *vpp, cnp);
996 		}
997 	}
998 
999 	return (error);
1000 }
1001 
1002 static inline bool
is_nametoolong(zfsvfs_t * zfsvfs,const char * name)1003 is_nametoolong(zfsvfs_t *zfsvfs, const char *name)
1004 {
1005 	size_t dlen = strlen(name);
1006 	return ((!zfsvfs->z_longname && dlen >= ZAP_MAXNAMELEN) ||
1007 	    dlen >= ZAP_MAXNAMELEN_NEW);
1008 }
1009 
1010 /*
1011  * Attempt to create a new entry in a directory.  If the entry
1012  * already exists, truncate the file if permissible, else return
1013  * an error.  Return the vp of the created or trunc'd file.
1014  *
1015  *	IN:	dvp	- vnode of directory to put new file entry in.
1016  *		name	- name of new file entry.
1017  *		vap	- attributes of new file.
1018  *		excl	- flag indicating exclusive or non-exclusive mode.
1019  *		mode	- mode to open file with.
1020  *		cr	- credentials of caller.
1021  *		flag	- large file flag [UNUSED].
1022  *		ct	- caller context
1023  *		vsecp	- ACL to be set
1024  *		mnt_ns	- Unused on FreeBSD
1025  *
1026  *	OUT:	vpp	- vnode of created or trunc'd entry.
1027  *
1028  *	RETURN:	0 on success, error code on failure.
1029  *
1030  * Timestamps:
1031  *	dvp - ctime|mtime updated if new entry created
1032  *	 vp - ctime|mtime always, atime if new
1033  */
1034 int
zfs_create(znode_t * dzp,const char * name,vattr_t * vap,int excl,int mode,znode_t ** zpp,cred_t * cr,int flag,vsecattr_t * vsecp,zidmap_t * mnt_ns)1035 zfs_create(znode_t *dzp, const char *name, vattr_t *vap, int excl, int mode,
1036     znode_t **zpp, cred_t *cr, int flag, vsecattr_t *vsecp, zidmap_t *mnt_ns)
1037 {
1038 	(void) excl, (void) mode, (void) flag;
1039 	znode_t		*zp;
1040 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1041 	zilog_t		*zilog;
1042 	objset_t	*os;
1043 	dmu_tx_t	*tx;
1044 	int		error;
1045 	uid_t		uid = crgetuid(cr);
1046 	gid_t		gid = crgetgid(cr);
1047 	uint64_t	projid = ZFS_DEFAULT_PROJID;
1048 	zfs_acl_ids_t   acl_ids;
1049 	boolean_t	fuid_dirtied;
1050 	uint64_t	txtype;
1051 #ifdef DEBUG_VFS_LOCKS
1052 	vnode_t	*dvp = ZTOV(dzp);
1053 #endif
1054 
1055 	if (is_nametoolong(zfsvfs, name))
1056 		return (SET_ERROR(ENAMETOOLONG));
1057 
1058 	/*
1059 	 * If we have an ephemeral id, ACL, or XVATTR then
1060 	 * make sure file system is at proper version
1061 	 */
1062 	if (zfsvfs->z_use_fuids == B_FALSE &&
1063 	    (vsecp || (vap->va_mask & AT_XVATTR) ||
1064 	    IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1065 		return (SET_ERROR(EINVAL));
1066 
1067 	if ((error = zfs_enter_verify_zp(zfsvfs, dzp, FTAG)) != 0)
1068 		return (error);
1069 	os = zfsvfs->z_os;
1070 	zilog = zfsvfs->z_log;
1071 
1072 	if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1073 	    NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1074 		zfs_exit(zfsvfs, FTAG);
1075 		return (SET_ERROR(EILSEQ));
1076 	}
1077 
1078 	if (vap->va_mask & AT_XVATTR) {
1079 		if ((error = secpolicy_xvattr(ZTOV(dzp), (xvattr_t *)vap,
1080 		    crgetuid(cr), cr, vap->va_type)) != 0) {
1081 			zfs_exit(zfsvfs, FTAG);
1082 			return (error);
1083 		}
1084 	}
1085 
1086 	*zpp = NULL;
1087 
1088 	if ((vap->va_mode & S_ISVTX) && secpolicy_vnode_stky_modify(cr))
1089 		vap->va_mode &= ~S_ISVTX;
1090 
1091 	error = zfs_dirent_lookup(dzp, name, &zp, ZNEW);
1092 	if (error) {
1093 		zfs_exit(zfsvfs, FTAG);
1094 		return (error);
1095 	}
1096 	ASSERT3P(zp, ==, NULL);
1097 
1098 	/*
1099 	 * Create a new file object and update the directory
1100 	 * to reference it.
1101 	 */
1102 	if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr, mnt_ns))) {
1103 		goto out;
1104 	}
1105 
1106 	/*
1107 	 * We only support the creation of regular files in
1108 	 * extended attribute directories.
1109 	 */
1110 
1111 	if ((dzp->z_pflags & ZFS_XATTR) &&
1112 	    (vap->va_type != VREG)) {
1113 		error = SET_ERROR(EINVAL);
1114 		goto out;
1115 	}
1116 
1117 	if ((error = zfs_acl_ids_create(dzp, 0, vap,
1118 	    cr, vsecp, &acl_ids, NULL)) != 0)
1119 		goto out;
1120 
1121 	if (S_ISREG(vap->va_mode) || S_ISDIR(vap->va_mode))
1122 		projid = zfs_inherit_projid(dzp);
1123 	if (zfs_acl_ids_overquota(zfsvfs, &acl_ids, projid)) {
1124 		zfs_acl_ids_free(&acl_ids);
1125 		error = SET_ERROR(EDQUOT);
1126 		goto out;
1127 	}
1128 
1129 	getnewvnode_reserve();
1130 
1131 	tx = dmu_tx_create(os);
1132 
1133 	dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1134 	    ZFS_SA_BASE_ATTR_SIZE);
1135 
1136 	fuid_dirtied = zfsvfs->z_fuid_dirty;
1137 	if (fuid_dirtied)
1138 		zfs_fuid_txhold(zfsvfs, tx);
1139 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1140 	dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1141 	if (!zfsvfs->z_use_sa &&
1142 	    acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1143 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1144 		    0, acl_ids.z_aclp->z_acl_bytes);
1145 	}
1146 	error = dmu_tx_assign(tx, DMU_TX_WAIT);
1147 	if (error) {
1148 		zfs_acl_ids_free(&acl_ids);
1149 		dmu_tx_abort(tx);
1150 		getnewvnode_drop_reserve();
1151 		zfs_exit(zfsvfs, FTAG);
1152 		return (error);
1153 	}
1154 	zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1155 
1156 	error = zfs_link_create(dzp, name, zp, tx, ZNEW);
1157 	if (error != 0) {
1158 		/*
1159 		 * Since, we failed to add the directory entry for it,
1160 		 * delete the newly created dnode.
1161 		 */
1162 		zfs_znode_delete(zp, tx);
1163 		VOP_UNLOCK(ZTOV(zp));
1164 		zrele(zp);
1165 		zfs_acl_ids_free(&acl_ids);
1166 		dmu_tx_commit(tx);
1167 		getnewvnode_drop_reserve();
1168 		goto out;
1169 	}
1170 
1171 	if (fuid_dirtied)
1172 		zfs_fuid_sync(zfsvfs, tx);
1173 
1174 	txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1175 	zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1176 	    vsecp, acl_ids.z_fuidp, vap);
1177 	zfs_acl_ids_free(&acl_ids);
1178 	dmu_tx_commit(tx);
1179 
1180 	getnewvnode_drop_reserve();
1181 
1182 out:
1183 	VNCHECKREF(dvp);
1184 	if (error == 0) {
1185 		*zpp = zp;
1186 	}
1187 
1188 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1189 		zil_commit(zilog, 0);
1190 
1191 	zfs_exit(zfsvfs, FTAG);
1192 	return (error);
1193 }
1194 
1195 /*
1196  * Remove an entry from a directory.
1197  *
1198  *	IN:	dvp	- vnode of directory to remove entry from.
1199  *		name	- name of entry to remove.
1200  *		cr	- credentials of caller.
1201  *		ct	- caller context
1202  *		flags	- case flags
1203  *
1204  *	RETURN:	0 on success, error code on failure.
1205  *
1206  * Timestamps:
1207  *	dvp - ctime|mtime
1208  *	 vp - ctime (if nlink > 0)
1209  */
1210 static int
zfs_remove_(vnode_t * dvp,vnode_t * vp,const char * name,cred_t * cr)1211 zfs_remove_(vnode_t *dvp, vnode_t *vp, const char *name, cred_t *cr)
1212 {
1213 	znode_t		*dzp = VTOZ(dvp);
1214 	znode_t		*zp;
1215 	znode_t		*xzp;
1216 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1217 	zilog_t		*zilog;
1218 	uint64_t	xattr_obj;
1219 	uint64_t	obj = 0;
1220 	dmu_tx_t	*tx;
1221 	boolean_t	unlinked;
1222 	uint64_t	txtype;
1223 	int		error;
1224 
1225 
1226 	if ((error = zfs_enter_verify_zp(zfsvfs, dzp, FTAG)) != 0)
1227 		return (error);
1228 	zp = VTOZ(vp);
1229 	if ((error = zfs_verify_zp(zp)) != 0) {
1230 		zfs_exit(zfsvfs, FTAG);
1231 		return (error);
1232 	}
1233 	zilog = zfsvfs->z_log;
1234 
1235 	xattr_obj = 0;
1236 	xzp = NULL;
1237 
1238 	if ((error = zfs_zaccess_delete(dzp, zp, cr, NULL))) {
1239 		goto out;
1240 	}
1241 
1242 	/*
1243 	 * Need to use rmdir for removing directories.
1244 	 */
1245 	if (vp->v_type == VDIR) {
1246 		error = SET_ERROR(EPERM);
1247 		goto out;
1248 	}
1249 
1250 	vnevent_remove(vp, dvp, name, ct);
1251 
1252 	obj = zp->z_id;
1253 
1254 	/* are there any extended attributes? */
1255 	error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1256 	    &xattr_obj, sizeof (xattr_obj));
1257 	if (error == 0 && xattr_obj) {
1258 		error = zfs_zget(zfsvfs, xattr_obj, &xzp);
1259 		ASSERT0(error);
1260 	}
1261 
1262 	/*
1263 	 * We may delete the znode now, or we may put it in the unlinked set;
1264 	 * it depends on whether we're the last link, and on whether there are
1265 	 * other holds on the vnode.  So we dmu_tx_hold() the right things to
1266 	 * allow for either case.
1267 	 */
1268 	tx = dmu_tx_create(zfsvfs->z_os);
1269 	dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1270 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1271 	zfs_sa_upgrade_txholds(tx, zp);
1272 	zfs_sa_upgrade_txholds(tx, dzp);
1273 
1274 	if (xzp) {
1275 		dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1276 		dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1277 	}
1278 
1279 	/* charge as an update -- would be nice not to charge at all */
1280 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1281 
1282 	/*
1283 	 * Mark this transaction as typically resulting in a net free of space
1284 	 */
1285 	dmu_tx_mark_netfree(tx);
1286 
1287 	error = dmu_tx_assign(tx, DMU_TX_WAIT);
1288 	if (error) {
1289 		dmu_tx_abort(tx);
1290 		zfs_exit(zfsvfs, FTAG);
1291 		return (error);
1292 	}
1293 
1294 	/*
1295 	 * Remove the directory entry.
1296 	 */
1297 	error = zfs_link_destroy(dzp, name, zp, tx, ZEXISTS, &unlinked);
1298 
1299 	if (error) {
1300 		dmu_tx_commit(tx);
1301 		goto out;
1302 	}
1303 
1304 	if (unlinked) {
1305 		zfs_unlinked_add(zp, tx);
1306 		vp->v_vflag |= VV_NOSYNC;
1307 	}
1308 	/* XXX check changes to linux vnops */
1309 	txtype = TX_REMOVE;
1310 	zfs_log_remove(zilog, tx, txtype, dzp, name, obj, unlinked);
1311 
1312 	dmu_tx_commit(tx);
1313 out:
1314 
1315 	if (xzp)
1316 		vrele(ZTOV(xzp));
1317 
1318 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1319 		zil_commit(zilog, 0);
1320 
1321 
1322 	zfs_exit(zfsvfs, FTAG);
1323 	return (error);
1324 }
1325 
1326 
1327 static int
zfs_lookup_internal(znode_t * dzp,const char * name,vnode_t ** vpp,struct componentname * cnp,int nameiop)1328 zfs_lookup_internal(znode_t *dzp, const char *name, vnode_t **vpp,
1329     struct componentname *cnp, int nameiop)
1330 {
1331 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1332 	int error;
1333 
1334 	cnp->cn_nameptr = __DECONST(char *, name);
1335 	cnp->cn_namelen = strlen(name);
1336 	cnp->cn_nameiop = nameiop;
1337 	cnp->cn_flags = ISLASTCN;
1338 #if __FreeBSD_version < 1400068
1339 	cnp->cn_flags |= SAVENAME;
1340 #endif
1341 	cnp->cn_lkflags = LK_EXCLUSIVE | LK_RETRY;
1342 	cnp->cn_cred = kcred;
1343 #if __FreeBSD_version < 1400037
1344 	cnp->cn_thread = curthread;
1345 #endif
1346 
1347 	if (zfsvfs->z_use_namecache && !zfsvfs->z_replay) {
1348 		struct vop_lookup_args a;
1349 
1350 		a.a_gen.a_desc = &vop_lookup_desc;
1351 		a.a_dvp = ZTOV(dzp);
1352 		a.a_vpp = vpp;
1353 		a.a_cnp = cnp;
1354 		error = vfs_cache_lookup(&a);
1355 	} else {
1356 		error = zfs_lookup(ZTOV(dzp), name, vpp, cnp, nameiop, kcred, 0,
1357 		    B_FALSE);
1358 	}
1359 #ifdef ZFS_DEBUG
1360 	if (error) {
1361 		printf("got error %d on name %s on op %d\n", error, name,
1362 		    nameiop);
1363 		kdb_backtrace();
1364 	}
1365 #endif
1366 	return (error);
1367 }
1368 
1369 int
zfs_remove(znode_t * dzp,const char * name,cred_t * cr,int flags)1370 zfs_remove(znode_t *dzp, const char *name, cred_t *cr, int flags)
1371 {
1372 	vnode_t *vp;
1373 	int error;
1374 	struct componentname cn;
1375 
1376 	if ((error = zfs_lookup_internal(dzp, name, &vp, &cn, DELETE)))
1377 		return (error);
1378 
1379 	error = zfs_remove_(ZTOV(dzp), vp, name, cr);
1380 	vput(vp);
1381 	return (error);
1382 }
1383 /*
1384  * Create a new directory and insert it into dvp using the name
1385  * provided.  Return a pointer to the inserted directory.
1386  *
1387  *	IN:	dvp	- vnode of directory to add subdir to.
1388  *		dirname	- name of new directory.
1389  *		vap	- attributes of new directory.
1390  *		cr	- credentials of caller.
1391  *		ct	- caller context
1392  *		flags	- case flags
1393  *		vsecp	- ACL to be set
1394  *		mnt_ns	- Unused on FreeBSD
1395  *
1396  *	OUT:	vpp	- vnode of created directory.
1397  *
1398  *	RETURN:	0 on success, error code on failure.
1399  *
1400  * Timestamps:
1401  *	dvp - ctime|mtime updated
1402  *	 vp - ctime|mtime|atime updated
1403  */
1404 int
zfs_mkdir(znode_t * dzp,const char * dirname,vattr_t * vap,znode_t ** zpp,cred_t * cr,int flags,vsecattr_t * vsecp,zidmap_t * mnt_ns)1405 zfs_mkdir(znode_t *dzp, const char *dirname, vattr_t *vap, znode_t **zpp,
1406     cred_t *cr, int flags, vsecattr_t *vsecp, zidmap_t *mnt_ns)
1407 {
1408 	(void) flags, (void) vsecp;
1409 	znode_t		*zp;
1410 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1411 	zilog_t		*zilog;
1412 	uint64_t	txtype;
1413 	dmu_tx_t	*tx;
1414 	int		error;
1415 	uid_t		uid = crgetuid(cr);
1416 	gid_t		gid = crgetgid(cr);
1417 	zfs_acl_ids_t   acl_ids;
1418 	boolean_t	fuid_dirtied;
1419 
1420 	ASSERT3U(vap->va_type, ==, VDIR);
1421 
1422 	if (is_nametoolong(zfsvfs, dirname))
1423 		return (SET_ERROR(ENAMETOOLONG));
1424 
1425 	/*
1426 	 * If we have an ephemeral id, ACL, or XVATTR then
1427 	 * make sure file system is at proper version
1428 	 */
1429 	if (zfsvfs->z_use_fuids == B_FALSE &&
1430 	    ((vap->va_mask & AT_XVATTR) ||
1431 	    IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1432 		return (SET_ERROR(EINVAL));
1433 
1434 	if ((error = zfs_enter_verify_zp(zfsvfs, dzp, FTAG)) != 0)
1435 		return (error);
1436 	zilog = zfsvfs->z_log;
1437 
1438 	if (dzp->z_pflags & ZFS_XATTR) {
1439 		zfs_exit(zfsvfs, FTAG);
1440 		return (SET_ERROR(EINVAL));
1441 	}
1442 
1443 	if (zfsvfs->z_utf8 && u8_validate(dirname,
1444 	    strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1445 		zfs_exit(zfsvfs, FTAG);
1446 		return (SET_ERROR(EILSEQ));
1447 	}
1448 
1449 	if (vap->va_mask & AT_XVATTR) {
1450 		if ((error = secpolicy_xvattr(ZTOV(dzp), (xvattr_t *)vap,
1451 		    crgetuid(cr), cr, vap->va_type)) != 0) {
1452 			zfs_exit(zfsvfs, FTAG);
1453 			return (error);
1454 		}
1455 	}
1456 
1457 	if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
1458 	    NULL, &acl_ids, NULL)) != 0) {
1459 		zfs_exit(zfsvfs, FTAG);
1460 		return (error);
1461 	}
1462 
1463 	/*
1464 	 * First make sure the new directory doesn't exist.
1465 	 *
1466 	 * Existence is checked first to make sure we don't return
1467 	 * EACCES instead of EEXIST which can cause some applications
1468 	 * to fail.
1469 	 */
1470 	*zpp = NULL;
1471 
1472 	if ((error = zfs_dirent_lookup(dzp, dirname, &zp, ZNEW))) {
1473 		zfs_acl_ids_free(&acl_ids);
1474 		zfs_exit(zfsvfs, FTAG);
1475 		return (error);
1476 	}
1477 	ASSERT3P(zp, ==, NULL);
1478 
1479 	if ((error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr,
1480 	    mnt_ns))) {
1481 		zfs_acl_ids_free(&acl_ids);
1482 		zfs_exit(zfsvfs, FTAG);
1483 		return (error);
1484 	}
1485 
1486 	if (zfs_acl_ids_overquota(zfsvfs, &acl_ids, zfs_inherit_projid(dzp))) {
1487 		zfs_acl_ids_free(&acl_ids);
1488 		zfs_exit(zfsvfs, FTAG);
1489 		return (SET_ERROR(EDQUOT));
1490 	}
1491 
1492 	/*
1493 	 * Add a new entry to the directory.
1494 	 */
1495 	getnewvnode_reserve();
1496 	tx = dmu_tx_create(zfsvfs->z_os);
1497 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
1498 	dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
1499 	fuid_dirtied = zfsvfs->z_fuid_dirty;
1500 	if (fuid_dirtied)
1501 		zfs_fuid_txhold(zfsvfs, tx);
1502 	if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1503 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
1504 		    acl_ids.z_aclp->z_acl_bytes);
1505 	}
1506 
1507 	dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1508 	    ZFS_SA_BASE_ATTR_SIZE);
1509 
1510 	error = dmu_tx_assign(tx, DMU_TX_WAIT);
1511 	if (error) {
1512 		zfs_acl_ids_free(&acl_ids);
1513 		dmu_tx_abort(tx);
1514 		getnewvnode_drop_reserve();
1515 		zfs_exit(zfsvfs, FTAG);
1516 		return (error);
1517 	}
1518 
1519 	/*
1520 	 * Create new node.
1521 	 */
1522 	zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1523 
1524 	/*
1525 	 * Now put new name in parent dir.
1526 	 */
1527 	error = zfs_link_create(dzp, dirname, zp, tx, ZNEW);
1528 	if (error != 0) {
1529 		zfs_znode_delete(zp, tx);
1530 		VOP_UNLOCK(ZTOV(zp));
1531 		zrele(zp);
1532 		goto out;
1533 	}
1534 
1535 	if (fuid_dirtied)
1536 		zfs_fuid_sync(zfsvfs, tx);
1537 
1538 	*zpp = zp;
1539 
1540 	txtype = zfs_log_create_txtype(Z_DIR, NULL, vap);
1541 	zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, NULL,
1542 	    acl_ids.z_fuidp, vap);
1543 
1544 out:
1545 	zfs_acl_ids_free(&acl_ids);
1546 
1547 	dmu_tx_commit(tx);
1548 
1549 	getnewvnode_drop_reserve();
1550 
1551 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1552 		zil_commit(zilog, 0);
1553 
1554 	zfs_exit(zfsvfs, FTAG);
1555 	return (error);
1556 }
1557 
1558 /*
1559  * Remove a directory subdir entry.  If the current working
1560  * directory is the same as the subdir to be removed, the
1561  * remove will fail.
1562  *
1563  *	IN:	dvp	- vnode of directory to remove from.
1564  *		name	- name of directory to be removed.
1565  *		cwd	- vnode of current working directory.
1566  *		cr	- credentials of caller.
1567  *		ct	- caller context
1568  *		flags	- case flags
1569  *
1570  *	RETURN:	0 on success, error code on failure.
1571  *
1572  * Timestamps:
1573  *	dvp - ctime|mtime updated
1574  */
1575 static int
zfs_rmdir_(vnode_t * dvp,vnode_t * vp,const char * name,cred_t * cr)1576 zfs_rmdir_(vnode_t *dvp, vnode_t *vp, const char *name, cred_t *cr)
1577 {
1578 	znode_t		*dzp = VTOZ(dvp);
1579 	znode_t		*zp = VTOZ(vp);
1580 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1581 	zilog_t		*zilog;
1582 	dmu_tx_t	*tx;
1583 	int		error;
1584 
1585 	if ((error = zfs_enter_verify_zp(zfsvfs, dzp, FTAG)) != 0)
1586 		return (error);
1587 	if ((error = zfs_verify_zp(zp)) != 0) {
1588 		zfs_exit(zfsvfs, FTAG);
1589 		return (error);
1590 	}
1591 	zilog = zfsvfs->z_log;
1592 
1593 
1594 	if ((error = zfs_zaccess_delete(dzp, zp, cr, NULL))) {
1595 		goto out;
1596 	}
1597 
1598 	if (vp->v_type != VDIR) {
1599 		error = SET_ERROR(ENOTDIR);
1600 		goto out;
1601 	}
1602 
1603 	vnevent_rmdir(vp, dvp, name, ct);
1604 
1605 	tx = dmu_tx_create(zfsvfs->z_os);
1606 	dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1607 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1608 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1609 	zfs_sa_upgrade_txholds(tx, zp);
1610 	zfs_sa_upgrade_txholds(tx, dzp);
1611 	dmu_tx_mark_netfree(tx);
1612 	error = dmu_tx_assign(tx, DMU_TX_WAIT);
1613 	if (error) {
1614 		dmu_tx_abort(tx);
1615 		zfs_exit(zfsvfs, FTAG);
1616 		return (error);
1617 	}
1618 
1619 	error = zfs_link_destroy(dzp, name, zp, tx, ZEXISTS, NULL);
1620 
1621 	if (error == 0) {
1622 		uint64_t txtype = TX_RMDIR;
1623 		zfs_log_remove(zilog, tx, txtype, dzp, name,
1624 		    ZFS_NO_OBJECT, B_FALSE);
1625 	}
1626 
1627 	dmu_tx_commit(tx);
1628 
1629 	if (zfsvfs->z_use_namecache)
1630 		cache_vop_rmdir(dvp, vp);
1631 out:
1632 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1633 		zil_commit(zilog, 0);
1634 
1635 	zfs_exit(zfsvfs, FTAG);
1636 	return (error);
1637 }
1638 
1639 int
zfs_rmdir(znode_t * dzp,const char * name,znode_t * cwd,cred_t * cr,int flags)1640 zfs_rmdir(znode_t *dzp, const char *name, znode_t *cwd, cred_t *cr, int flags)
1641 {
1642 	struct componentname cn;
1643 	vnode_t *vp;
1644 	int error;
1645 
1646 	if ((error = zfs_lookup_internal(dzp, name, &vp, &cn, DELETE)))
1647 		return (error);
1648 
1649 	error = zfs_rmdir_(ZTOV(dzp), vp, name, cr);
1650 	vput(vp);
1651 	return (error);
1652 }
1653 
1654 /*
1655  * Read as many directory entries as will fit into the provided
1656  * buffer from the given directory cursor position (specified in
1657  * the uio structure).
1658  *
1659  *	IN:	vp	- vnode of directory to read.
1660  *		uio	- structure supplying read location, range info,
1661  *			  and return buffer.
1662  *		cr	- credentials of caller.
1663  *		ct	- caller context
1664  *
1665  *	OUT:	uio	- updated offset and range, buffer filled.
1666  *		eofp	- set to true if end-of-file detected.
1667  *		ncookies- number of entries in cookies
1668  *		cookies	- offsets to directory entries
1669  *
1670  *	RETURN:	0 on success, error code on failure.
1671  *
1672  * Timestamps:
1673  *	vp - atime updated
1674  *
1675  * Note that the low 4 bits of the cookie returned by zap is always zero.
1676  * This allows us to use the low range for "special" directory entries:
1677  * We use 0 for '.', and 1 for '..'.  If this is the root of the filesystem,
1678  * we use the offset 2 for the '.zfs' directory.
1679  */
1680 static int
zfs_readdir(vnode_t * vp,zfs_uio_t * uio,cred_t * cr,int * eofp,int * ncookies,cookie_t ** cookies)1681 zfs_readdir(vnode_t *vp, zfs_uio_t *uio, cred_t *cr, int *eofp,
1682     int *ncookies, cookie_t **cookies)
1683 {
1684 	znode_t		*zp = VTOZ(vp);
1685 	iovec_t		*iovp;
1686 	dirent64_t	*odp;
1687 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
1688 	objset_t	*os;
1689 	caddr_t		outbuf;
1690 	size_t		bufsize;
1691 	zap_cursor_t	zc;
1692 	zap_attribute_t	*zap;
1693 	uint_t		bytes_wanted;
1694 	uint64_t	offset; /* must be unsigned; checks for < 1 */
1695 	uint64_t	parent;
1696 	int		local_eof;
1697 	int		outcount;
1698 	int		error;
1699 	uint8_t		prefetch;
1700 	uint8_t		type;
1701 	int		ncooks;
1702 	cookie_t	*cooks = NULL;
1703 
1704 	if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0)
1705 		return (error);
1706 
1707 	if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
1708 	    &parent, sizeof (parent))) != 0) {
1709 		zfs_exit(zfsvfs, FTAG);
1710 		return (error);
1711 	}
1712 
1713 	/*
1714 	 * If we are not given an eof variable,
1715 	 * use a local one.
1716 	 */
1717 	if (eofp == NULL)
1718 		eofp = &local_eof;
1719 
1720 	/*
1721 	 * Check for valid iov_len.
1722 	 */
1723 	if (GET_UIO_STRUCT(uio)->uio_iov->iov_len <= 0) {
1724 		zfs_exit(zfsvfs, FTAG);
1725 		return (SET_ERROR(EINVAL));
1726 	}
1727 
1728 	/*
1729 	 * Quit if directory has been removed (posix)
1730 	 */
1731 	if ((*eofp = zp->z_unlinked) != 0) {
1732 		zfs_exit(zfsvfs, FTAG);
1733 		return (0);
1734 	}
1735 
1736 	error = 0;
1737 	os = zfsvfs->z_os;
1738 	offset = zfs_uio_offset(uio);
1739 	prefetch = zp->z_zn_prefetch;
1740 	zap = zap_attribute_long_alloc();
1741 
1742 	/*
1743 	 * Initialize the iterator cursor.
1744 	 */
1745 	if (offset <= 3) {
1746 		/*
1747 		 * Start iteration from the beginning of the directory.
1748 		 */
1749 		zap_cursor_init(&zc, os, zp->z_id);
1750 	} else {
1751 		/*
1752 		 * The offset is a serialized cursor.
1753 		 */
1754 		zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
1755 	}
1756 
1757 	/*
1758 	 * Get space to change directory entries into fs independent format.
1759 	 */
1760 	iovp = GET_UIO_STRUCT(uio)->uio_iov;
1761 	bytes_wanted = iovp->iov_len;
1762 	if (zfs_uio_segflg(uio) != UIO_SYSSPACE || zfs_uio_iovcnt(uio) != 1) {
1763 		bufsize = bytes_wanted;
1764 		outbuf = kmem_alloc(bufsize, KM_SLEEP);
1765 		odp = (struct dirent64 *)outbuf;
1766 	} else {
1767 		bufsize = bytes_wanted;
1768 		outbuf = NULL;
1769 		odp = (struct dirent64 *)iovp->iov_base;
1770 	}
1771 
1772 	if (ncookies != NULL) {
1773 		/*
1774 		 * Minimum entry size is dirent size and 1 byte for a file name.
1775 		 */
1776 		ncooks = zfs_uio_resid(uio) / (sizeof (struct dirent) -
1777 		    sizeof (((struct dirent *)NULL)->d_name) + 1);
1778 		cooks = malloc(ncooks * sizeof (*cooks), M_TEMP, M_WAITOK);
1779 		*cookies = cooks;
1780 		*ncookies = ncooks;
1781 	}
1782 
1783 	/*
1784 	 * Transform to file-system independent format
1785 	 */
1786 	outcount = 0;
1787 	while (outcount < bytes_wanted) {
1788 		ino64_t objnum;
1789 		ushort_t reclen;
1790 		off64_t *next = NULL;
1791 
1792 		/*
1793 		 * Special case `.', `..', and `.zfs'.
1794 		 */
1795 		if (offset == 0) {
1796 			(void) strcpy(zap->za_name, ".");
1797 			zap->za_normalization_conflict = 0;
1798 			objnum = zp->z_id;
1799 			type = DT_DIR;
1800 		} else if (offset == 1) {
1801 			(void) strcpy(zap->za_name, "..");
1802 			zap->za_normalization_conflict = 0;
1803 			objnum = parent;
1804 			type = DT_DIR;
1805 		} else if (offset == 2 && zfs_show_ctldir(zp)) {
1806 			(void) strcpy(zap->za_name, ZFS_CTLDIR_NAME);
1807 			zap->za_normalization_conflict = 0;
1808 			objnum = ZFSCTL_INO_ROOT;
1809 			type = DT_DIR;
1810 		} else {
1811 			/*
1812 			 * Grab next entry.
1813 			 */
1814 			if ((error = zap_cursor_retrieve(&zc, zap))) {
1815 				if ((*eofp = (error == ENOENT)) != 0)
1816 					break;
1817 				else
1818 					goto update;
1819 			}
1820 
1821 			if (zap->za_integer_length != 8 ||
1822 			    zap->za_num_integers != 1) {
1823 				cmn_err(CE_WARN, "zap_readdir: bad directory "
1824 				    "entry, obj = %lld, offset = %lld\n",
1825 				    (u_longlong_t)zp->z_id,
1826 				    (u_longlong_t)offset);
1827 				error = SET_ERROR(ENXIO);
1828 				goto update;
1829 			}
1830 
1831 			objnum = ZFS_DIRENT_OBJ(zap->za_first_integer);
1832 			/*
1833 			 * MacOS X can extract the object type here such as:
1834 			 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
1835 			 */
1836 			type = ZFS_DIRENT_TYPE(zap->za_first_integer);
1837 		}
1838 
1839 		reclen = DIRENT64_RECLEN(strlen(zap->za_name));
1840 
1841 		/*
1842 		 * Will this entry fit in the buffer?
1843 		 */
1844 		if (outcount + reclen > bufsize) {
1845 			/*
1846 			 * Did we manage to fit anything in the buffer?
1847 			 */
1848 			if (!outcount) {
1849 				error = SET_ERROR(EINVAL);
1850 				goto update;
1851 			}
1852 			break;
1853 		}
1854 		/*
1855 		 * Add normal entry:
1856 		 */
1857 		odp->d_ino = objnum;
1858 		odp->d_reclen = reclen;
1859 		odp->d_namlen = strlen(zap->za_name);
1860 		/* NOTE: d_off is the offset for the *next* entry. */
1861 		next = &odp->d_off;
1862 		strlcpy(odp->d_name, zap->za_name, odp->d_namlen + 1);
1863 		odp->d_type = type;
1864 		dirent_terminate(odp);
1865 		odp = (dirent64_t *)((intptr_t)odp + reclen);
1866 
1867 		outcount += reclen;
1868 
1869 		ASSERT3S(outcount, <=, bufsize);
1870 
1871 		if (prefetch)
1872 			dmu_prefetch_dnode(os, objnum, ZIO_PRIORITY_SYNC_READ);
1873 
1874 		/*
1875 		 * Move to the next entry, fill in the previous offset.
1876 		 */
1877 		if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
1878 			zap_cursor_advance(&zc);
1879 			offset = zap_cursor_serialize(&zc);
1880 		} else {
1881 			offset += 1;
1882 		}
1883 
1884 		/* Fill the offset right after advancing the cursor. */
1885 		if (next != NULL)
1886 			*next = offset;
1887 		if (cooks != NULL) {
1888 			*cooks++ = offset;
1889 			ncooks--;
1890 			KASSERT(ncooks >= 0, ("ncookies=%d", ncooks));
1891 		}
1892 	}
1893 	zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
1894 
1895 	/* Subtract unused cookies */
1896 	if (ncookies != NULL)
1897 		*ncookies -= ncooks;
1898 
1899 	if (zfs_uio_segflg(uio) == UIO_SYSSPACE && zfs_uio_iovcnt(uio) == 1) {
1900 		iovp->iov_base += outcount;
1901 		iovp->iov_len -= outcount;
1902 		zfs_uio_resid(uio) -= outcount;
1903 	} else if ((error =
1904 	    zfs_uiomove(outbuf, (long)outcount, UIO_READ, uio))) {
1905 		/*
1906 		 * Reset the pointer.
1907 		 */
1908 		offset = zfs_uio_offset(uio);
1909 	}
1910 
1911 update:
1912 	zap_cursor_fini(&zc);
1913 	zap_attribute_free(zap);
1914 	if (zfs_uio_segflg(uio) != UIO_SYSSPACE || zfs_uio_iovcnt(uio) != 1)
1915 		kmem_free(outbuf, bufsize);
1916 
1917 	if (error == ENOENT)
1918 		error = 0;
1919 
1920 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
1921 
1922 	zfs_uio_setoffset(uio, offset);
1923 	zfs_exit(zfsvfs, FTAG);
1924 	if (error != 0 && cookies != NULL) {
1925 		free(*cookies, M_TEMP);
1926 		*cookies = NULL;
1927 		*ncookies = 0;
1928 	}
1929 	return (error);
1930 }
1931 
1932 /*
1933  * Get the requested file attributes and place them in the provided
1934  * vattr structure.
1935  *
1936  *	IN:	vp	- vnode of file.
1937  *		vap	- va_mask identifies requested attributes.
1938  *			  If AT_XVATTR set, then optional attrs are requested
1939  *		flags	- ATTR_NOACLCHECK (CIFS server context)
1940  *		cr	- credentials of caller.
1941  *
1942  *	OUT:	vap	- attribute values.
1943  *
1944  *	RETURN:	0 (always succeeds).
1945  */
1946 static int
zfs_getattr(vnode_t * vp,vattr_t * vap,int flags,cred_t * cr)1947 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr)
1948 {
1949 	znode_t *zp = VTOZ(vp);
1950 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1951 	int	error = 0;
1952 	uint32_t blksize;
1953 	u_longlong_t nblocks;
1954 	uint64_t mtime[2], ctime[2], crtime[2], rdev;
1955 	xvattr_t *xvap = (xvattr_t *)vap;	/* vap may be an xvattr_t * */
1956 	xoptattr_t *xoap = NULL;
1957 	boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
1958 	sa_bulk_attr_t bulk[4];
1959 	int count = 0;
1960 
1961 	if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0)
1962 		return (error);
1963 
1964 	zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
1965 
1966 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
1967 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
1968 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL, &crtime, 16);
1969 	if (vp->v_type == VBLK || vp->v_type == VCHR)
1970 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL,
1971 		    &rdev, 8);
1972 
1973 	if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
1974 		zfs_exit(zfsvfs, FTAG);
1975 		return (error);
1976 	}
1977 
1978 	/*
1979 	 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
1980 	 * Also, if we are the owner don't bother, since owner should
1981 	 * always be allowed to read basic attributes of file.
1982 	 */
1983 	if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
1984 	    (vap->va_uid != crgetuid(cr))) {
1985 		if ((error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
1986 		    skipaclchk, cr, NULL))) {
1987 			zfs_exit(zfsvfs, FTAG);
1988 			return (error);
1989 		}
1990 	}
1991 
1992 	/*
1993 	 * Return all attributes.  It's cheaper to provide the answer
1994 	 * than to determine whether we were asked the question.
1995 	 */
1996 
1997 	vap->va_type = IFTOVT(zp->z_mode);
1998 	vap->va_mode = zp->z_mode & ~S_IFMT;
1999 	vn_fsid(vp, vap);
2000 	vap->va_nodeid = zp->z_id;
2001 	vap->va_nlink = zp->z_links;
2002 	if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp) &&
2003 	    zp->z_links < ZFS_LINK_MAX)
2004 		vap->va_nlink++;
2005 	vap->va_size = zp->z_size;
2006 	if (vp->v_type == VBLK || vp->v_type == VCHR)
2007 		vap->va_rdev = zfs_cmpldev(rdev);
2008 	else
2009 		vap->va_rdev = 0;
2010 	vap->va_gen = zp->z_gen;
2011 	vap->va_flags = 0;	/* FreeBSD: Reset chflags(2) flags. */
2012 	vap->va_filerev = zp->z_seq;
2013 
2014 	/*
2015 	 * Add in any requested optional attributes and the create time.
2016 	 * Also set the corresponding bits in the returned attribute bitmap.
2017 	 */
2018 	if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2019 		if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2020 			xoap->xoa_archive =
2021 			    ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2022 			XVA_SET_RTN(xvap, XAT_ARCHIVE);
2023 		}
2024 
2025 		if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2026 			xoap->xoa_readonly =
2027 			    ((zp->z_pflags & ZFS_READONLY) != 0);
2028 			XVA_SET_RTN(xvap, XAT_READONLY);
2029 		}
2030 
2031 		if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2032 			xoap->xoa_system =
2033 			    ((zp->z_pflags & ZFS_SYSTEM) != 0);
2034 			XVA_SET_RTN(xvap, XAT_SYSTEM);
2035 		}
2036 
2037 		if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2038 			xoap->xoa_hidden =
2039 			    ((zp->z_pflags & ZFS_HIDDEN) != 0);
2040 			XVA_SET_RTN(xvap, XAT_HIDDEN);
2041 		}
2042 
2043 		if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2044 			xoap->xoa_nounlink =
2045 			    ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2046 			XVA_SET_RTN(xvap, XAT_NOUNLINK);
2047 		}
2048 
2049 		if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2050 			xoap->xoa_immutable =
2051 			    ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2052 			XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2053 		}
2054 
2055 		if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2056 			xoap->xoa_appendonly =
2057 			    ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2058 			XVA_SET_RTN(xvap, XAT_APPENDONLY);
2059 		}
2060 
2061 		if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2062 			xoap->xoa_nodump =
2063 			    ((zp->z_pflags & ZFS_NODUMP) != 0);
2064 			XVA_SET_RTN(xvap, XAT_NODUMP);
2065 		}
2066 
2067 		if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2068 			xoap->xoa_opaque =
2069 			    ((zp->z_pflags & ZFS_OPAQUE) != 0);
2070 			XVA_SET_RTN(xvap, XAT_OPAQUE);
2071 		}
2072 
2073 		if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2074 			xoap->xoa_av_quarantined =
2075 			    ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2076 			XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2077 		}
2078 
2079 		if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2080 			xoap->xoa_av_modified =
2081 			    ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2082 			XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2083 		}
2084 
2085 		if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2086 		    vp->v_type == VREG) {
2087 			zfs_sa_get_scanstamp(zp, xvap);
2088 		}
2089 
2090 		if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2091 			xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2092 			XVA_SET_RTN(xvap, XAT_REPARSE);
2093 		}
2094 		if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2095 			xoap->xoa_generation = zp->z_gen;
2096 			XVA_SET_RTN(xvap, XAT_GEN);
2097 		}
2098 
2099 		if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2100 			xoap->xoa_offline =
2101 			    ((zp->z_pflags & ZFS_OFFLINE) != 0);
2102 			XVA_SET_RTN(xvap, XAT_OFFLINE);
2103 		}
2104 
2105 		if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
2106 			xoap->xoa_sparse =
2107 			    ((zp->z_pflags & ZFS_SPARSE) != 0);
2108 			XVA_SET_RTN(xvap, XAT_SPARSE);
2109 		}
2110 
2111 		if (XVA_ISSET_REQ(xvap, XAT_PROJINHERIT)) {
2112 			xoap->xoa_projinherit =
2113 			    ((zp->z_pflags & ZFS_PROJINHERIT) != 0);
2114 			XVA_SET_RTN(xvap, XAT_PROJINHERIT);
2115 		}
2116 
2117 		if (XVA_ISSET_REQ(xvap, XAT_PROJID)) {
2118 			xoap->xoa_projid = zp->z_projid;
2119 			XVA_SET_RTN(xvap, XAT_PROJID);
2120 		}
2121 	}
2122 
2123 	ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime);
2124 	ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2125 	ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2126 	ZFS_TIME_DECODE(&vap->va_birthtime, crtime);
2127 
2128 
2129 	sa_object_size(zp->z_sa_hdl, &blksize, &nblocks);
2130 	vap->va_blksize = blksize;
2131 	vap->va_bytes = nblocks << 9;	/* nblocks * 512 */
2132 
2133 	if (zp->z_blksz == 0) {
2134 		/*
2135 		 * Block size hasn't been set; suggest maximal I/O transfers.
2136 		 */
2137 		vap->va_blksize = zfsvfs->z_max_blksz;
2138 	}
2139 
2140 	zfs_exit(zfsvfs, FTAG);
2141 	return (0);
2142 }
2143 
2144 /*
2145  * For the operation of changing file's user/group/project, we need to
2146  * handle not only the main object that is assigned to the file directly,
2147  * but also the ones that are used by the file via hidden xattr directory.
2148  *
2149  * Because the xattr directory may contains many EA entries, as to it may
2150  * be impossible to change all of them via the transaction of changing the
2151  * main object's user/group/project attributes. Then we have to change them
2152  * via other multiple independent transactions one by one. It may be not good
2153  * solution, but we have no better idea yet.
2154  */
2155 static int
zfs_setattr_dir(znode_t * dzp)2156 zfs_setattr_dir(znode_t *dzp)
2157 {
2158 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
2159 	objset_t	*os = zfsvfs->z_os;
2160 	zap_cursor_t	zc;
2161 	zap_attribute_t	*zap;
2162 	znode_t		*zp = NULL;
2163 	dmu_tx_t	*tx = NULL;
2164 	uint64_t	uid, gid;
2165 	sa_bulk_attr_t	bulk[4];
2166 	int		count;
2167 	int		err;
2168 
2169 	zap = zap_attribute_alloc();
2170 	zap_cursor_init(&zc, os, dzp->z_id);
2171 	while ((err = zap_cursor_retrieve(&zc, zap)) == 0) {
2172 		count = 0;
2173 		if (zap->za_integer_length != 8 || zap->za_num_integers != 1) {
2174 			err = ENXIO;
2175 			break;
2176 		}
2177 
2178 		err = zfs_dirent_lookup(dzp, zap->za_name, &zp, ZEXISTS);
2179 		if (err == ENOENT)
2180 			goto next;
2181 		if (err)
2182 			break;
2183 
2184 		if (zp->z_uid == dzp->z_uid &&
2185 		    zp->z_gid == dzp->z_gid &&
2186 		    zp->z_projid == dzp->z_projid)
2187 			goto next;
2188 
2189 		tx = dmu_tx_create(os);
2190 		if (!(zp->z_pflags & ZFS_PROJID))
2191 			dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
2192 		else
2193 			dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2194 
2195 		err = dmu_tx_assign(tx, DMU_TX_WAIT);
2196 		if (err)
2197 			break;
2198 
2199 		mutex_enter(&dzp->z_lock);
2200 
2201 		if (zp->z_uid != dzp->z_uid) {
2202 			uid = dzp->z_uid;
2203 			SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
2204 			    &uid, sizeof (uid));
2205 			zp->z_uid = uid;
2206 		}
2207 
2208 		if (zp->z_gid != dzp->z_gid) {
2209 			gid = dzp->z_gid;
2210 			SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL,
2211 			    &gid, sizeof (gid));
2212 			zp->z_gid = gid;
2213 		}
2214 
2215 		uint64_t projid = dzp->z_projid;
2216 		if (zp->z_projid != projid) {
2217 			if (!(zp->z_pflags & ZFS_PROJID)) {
2218 				err = sa_add_projid(zp->z_sa_hdl, tx, projid);
2219 				if (unlikely(err == EEXIST)) {
2220 					err = 0;
2221 				} else if (err != 0) {
2222 					goto sa_add_projid_err;
2223 				} else {
2224 					projid = ZFS_INVALID_PROJID;
2225 				}
2226 			}
2227 
2228 			if (projid != ZFS_INVALID_PROJID) {
2229 				zp->z_projid = projid;
2230 				SA_ADD_BULK_ATTR(bulk, count,
2231 				    SA_ZPL_PROJID(zfsvfs), NULL, &zp->z_projid,
2232 				    sizeof (zp->z_projid));
2233 			}
2234 		}
2235 
2236 sa_add_projid_err:
2237 		mutex_exit(&dzp->z_lock);
2238 
2239 		if (likely(count > 0)) {
2240 			err = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
2241 			dmu_tx_commit(tx);
2242 		} else if (projid == ZFS_INVALID_PROJID) {
2243 			dmu_tx_commit(tx);
2244 		} else {
2245 			dmu_tx_abort(tx);
2246 		}
2247 		tx = NULL;
2248 		if (err != 0 && err != ENOENT)
2249 			break;
2250 
2251 next:
2252 		if (zp) {
2253 			zrele(zp);
2254 			zp = NULL;
2255 		}
2256 		zap_cursor_advance(&zc);
2257 	}
2258 
2259 	if (tx)
2260 		dmu_tx_abort(tx);
2261 	if (zp) {
2262 		zrele(zp);
2263 	}
2264 	zap_cursor_fini(&zc);
2265 	zap_attribute_free(zap);
2266 
2267 	return (err == ENOENT ? 0 : err);
2268 }
2269 
2270 /*
2271  * Set the file attributes to the values contained in the
2272  * vattr structure.
2273  *
2274  *	IN:	zp	- znode of file to be modified.
2275  *		vap	- new attribute values.
2276  *			  If AT_XVATTR set, then optional attrs are being set
2277  *		flags	- ATTR_UTIME set if non-default time values provided.
2278  *			- ATTR_NOACLCHECK (CIFS context only).
2279  *		cr	- credentials of caller.
2280  *		mnt_ns	- Unused on FreeBSD
2281  *
2282  *	RETURN:	0 on success, error code on failure.
2283  *
2284  * Timestamps:
2285  *	vp - ctime updated, mtime updated if size changed.
2286  */
2287 int
zfs_setattr(znode_t * zp,vattr_t * vap,int flags,cred_t * cr,zidmap_t * mnt_ns)2288 zfs_setattr(znode_t *zp, vattr_t *vap, int flags, cred_t *cr, zidmap_t *mnt_ns)
2289 {
2290 	vnode_t		*vp = ZTOV(zp);
2291 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2292 	objset_t	*os;
2293 	zilog_t		*zilog;
2294 	dmu_tx_t	*tx;
2295 	vattr_t		oldva;
2296 	xvattr_t	tmpxvattr;
2297 	uint_t		mask = vap->va_mask;
2298 	uint_t		saved_mask = 0;
2299 	uint64_t	saved_mode;
2300 	int		trim_mask = 0;
2301 	uint64_t	new_mode;
2302 	uint64_t	new_uid, new_gid;
2303 	uint64_t	xattr_obj;
2304 	uint64_t	mtime[2], ctime[2];
2305 	uint64_t	projid = ZFS_INVALID_PROJID;
2306 	znode_t		*attrzp;
2307 	int		need_policy = FALSE;
2308 	int		err, err2;
2309 	zfs_fuid_info_t *fuidp = NULL;
2310 	xvattr_t *xvap = (xvattr_t *)vap;	/* vap may be an xvattr_t * */
2311 	xoptattr_t	*xoap;
2312 	zfs_acl_t	*aclp;
2313 	boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2314 	boolean_t	fuid_dirtied = B_FALSE;
2315 	boolean_t	handle_eadir = B_FALSE;
2316 	sa_bulk_attr_t	bulk[7], xattr_bulk[7];
2317 	int		count = 0, xattr_count = 0;
2318 
2319 	if (mask == 0)
2320 		return (0);
2321 
2322 	if (mask & AT_NOSET)
2323 		return (SET_ERROR(EINVAL));
2324 
2325 	if ((err = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0)
2326 		return (err);
2327 
2328 	os = zfsvfs->z_os;
2329 	zilog = zfsvfs->z_log;
2330 
2331 	/*
2332 	 * Make sure that if we have ephemeral uid/gid or xvattr specified
2333 	 * that file system is at proper version level
2334 	 */
2335 
2336 	if (zfsvfs->z_use_fuids == B_FALSE &&
2337 	    (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
2338 	    ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) ||
2339 	    (mask & AT_XVATTR))) {
2340 		zfs_exit(zfsvfs, FTAG);
2341 		return (SET_ERROR(EINVAL));
2342 	}
2343 
2344 	if (mask & AT_SIZE && vp->v_type == VDIR) {
2345 		zfs_exit(zfsvfs, FTAG);
2346 		return (SET_ERROR(EISDIR));
2347 	}
2348 
2349 	if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) {
2350 		zfs_exit(zfsvfs, FTAG);
2351 		return (SET_ERROR(EINVAL));
2352 	}
2353 
2354 	/*
2355 	 * If this is an xvattr_t, then get a pointer to the structure of
2356 	 * optional attributes.  If this is NULL, then we have a vattr_t.
2357 	 */
2358 	xoap = xva_getxoptattr(xvap);
2359 
2360 	xva_init(&tmpxvattr);
2361 
2362 	/*
2363 	 * Immutable files can only alter immutable bit and atime
2364 	 */
2365 	if ((zp->z_pflags & ZFS_IMMUTABLE) &&
2366 	    ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) ||
2367 	    ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
2368 		zfs_exit(zfsvfs, FTAG);
2369 		return (SET_ERROR(EPERM));
2370 	}
2371 
2372 	/*
2373 	 * Note: ZFS_READONLY is handled in zfs_zaccess_common.
2374 	 */
2375 
2376 	/*
2377 	 * Verify timestamps doesn't overflow 32 bits.
2378 	 * ZFS can handle large timestamps, but 32bit syscalls can't
2379 	 * handle times greater than 2039.  This check should be removed
2380 	 * once large timestamps are fully supported.
2381 	 */
2382 	if (mask & (AT_ATIME | AT_MTIME)) {
2383 		if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
2384 		    ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
2385 			zfs_exit(zfsvfs, FTAG);
2386 			return (SET_ERROR(EOVERFLOW));
2387 		}
2388 	}
2389 	if (xoap != NULL && (mask & AT_XVATTR)) {
2390 		if (XVA_ISSET_REQ(xvap, XAT_CREATETIME) &&
2391 		    TIMESPEC_OVERFLOW(&vap->va_birthtime)) {
2392 			zfs_exit(zfsvfs, FTAG);
2393 			return (SET_ERROR(EOVERFLOW));
2394 		}
2395 
2396 		if (XVA_ISSET_REQ(xvap, XAT_PROJID)) {
2397 			if (!dmu_objset_projectquota_enabled(os) ||
2398 			    (!S_ISREG(zp->z_mode) && !S_ISDIR(zp->z_mode))) {
2399 				zfs_exit(zfsvfs, FTAG);
2400 				return (SET_ERROR(EOPNOTSUPP));
2401 			}
2402 
2403 			projid = xoap->xoa_projid;
2404 			if (unlikely(projid == ZFS_INVALID_PROJID)) {
2405 				zfs_exit(zfsvfs, FTAG);
2406 				return (SET_ERROR(EINVAL));
2407 			}
2408 
2409 			if (projid == zp->z_projid && zp->z_pflags & ZFS_PROJID)
2410 				projid = ZFS_INVALID_PROJID;
2411 			else
2412 				need_policy = TRUE;
2413 		}
2414 
2415 		if (XVA_ISSET_REQ(xvap, XAT_PROJINHERIT) &&
2416 		    (xoap->xoa_projinherit !=
2417 		    ((zp->z_pflags & ZFS_PROJINHERIT) != 0)) &&
2418 		    (!dmu_objset_projectquota_enabled(os) ||
2419 		    (!S_ISREG(zp->z_mode) && !S_ISDIR(zp->z_mode)))) {
2420 			zfs_exit(zfsvfs, FTAG);
2421 			return (SET_ERROR(EOPNOTSUPP));
2422 		}
2423 	}
2424 
2425 	attrzp = NULL;
2426 	aclp = NULL;
2427 
2428 	if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
2429 		zfs_exit(zfsvfs, FTAG);
2430 		return (SET_ERROR(EROFS));
2431 	}
2432 
2433 	/*
2434 	 * First validate permissions
2435 	 */
2436 
2437 	if (mask & AT_SIZE) {
2438 		/*
2439 		 * XXX - Note, we are not providing any open
2440 		 * mode flags here (like FNDELAY), so we may
2441 		 * block if there are locks present... this
2442 		 * should be addressed in openat().
2443 		 */
2444 		/* XXX - would it be OK to generate a log record here? */
2445 		err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
2446 		if (err) {
2447 			zfs_exit(zfsvfs, FTAG);
2448 			return (err);
2449 		}
2450 	}
2451 
2452 	if (mask & (AT_ATIME|AT_MTIME) ||
2453 	    ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
2454 	    XVA_ISSET_REQ(xvap, XAT_READONLY) ||
2455 	    XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
2456 	    XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
2457 	    XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
2458 	    XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
2459 	    XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
2460 		need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
2461 		    skipaclchk, cr, mnt_ns);
2462 	}
2463 
2464 	if (mask & (AT_UID|AT_GID)) {
2465 		int	idmask = (mask & (AT_UID|AT_GID));
2466 		int	take_owner;
2467 		int	take_group;
2468 
2469 		/*
2470 		 * NOTE: even if a new mode is being set,
2471 		 * we may clear S_ISUID/S_ISGID bits.
2472 		 */
2473 
2474 		if (!(mask & AT_MODE))
2475 			vap->va_mode = zp->z_mode;
2476 
2477 		/*
2478 		 * Take ownership or chgrp to group we are a member of
2479 		 */
2480 
2481 		take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
2482 		take_group = (mask & AT_GID) &&
2483 		    zfs_groupmember(zfsvfs, vap->va_gid, cr);
2484 
2485 		/*
2486 		 * If both AT_UID and AT_GID are set then take_owner and
2487 		 * take_group must both be set in order to allow taking
2488 		 * ownership.
2489 		 *
2490 		 * Otherwise, send the check through secpolicy_vnode_setattr()
2491 		 *
2492 		 */
2493 
2494 		if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
2495 		    ((idmask == AT_UID) && take_owner) ||
2496 		    ((idmask == AT_GID) && take_group)) {
2497 			if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
2498 			    skipaclchk, cr, mnt_ns) == 0) {
2499 				/*
2500 				 * Remove setuid/setgid for non-privileged users
2501 				 */
2502 				secpolicy_setid_clear(vap, vp, cr);
2503 				trim_mask = (mask & (AT_UID|AT_GID));
2504 			} else {
2505 				need_policy =  TRUE;
2506 			}
2507 		} else {
2508 			need_policy =  TRUE;
2509 		}
2510 	}
2511 
2512 	oldva.va_mode = zp->z_mode;
2513 	zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
2514 	if (mask & AT_XVATTR) {
2515 		/*
2516 		 * Update xvattr mask to include only those attributes
2517 		 * that are actually changing.
2518 		 *
2519 		 * the bits will be restored prior to actually setting
2520 		 * the attributes so the caller thinks they were set.
2521 		 */
2522 		if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2523 			if (xoap->xoa_appendonly !=
2524 			    ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
2525 				need_policy = TRUE;
2526 			} else {
2527 				XVA_CLR_REQ(xvap, XAT_APPENDONLY);
2528 				XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY);
2529 			}
2530 		}
2531 
2532 		if (XVA_ISSET_REQ(xvap, XAT_PROJINHERIT)) {
2533 			if (xoap->xoa_projinherit !=
2534 			    ((zp->z_pflags & ZFS_PROJINHERIT) != 0)) {
2535 				need_policy = TRUE;
2536 			} else {
2537 				XVA_CLR_REQ(xvap, XAT_PROJINHERIT);
2538 				XVA_SET_REQ(&tmpxvattr, XAT_PROJINHERIT);
2539 			}
2540 		}
2541 
2542 		if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2543 			if (xoap->xoa_nounlink !=
2544 			    ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
2545 				need_policy = TRUE;
2546 			} else {
2547 				XVA_CLR_REQ(xvap, XAT_NOUNLINK);
2548 				XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK);
2549 			}
2550 		}
2551 
2552 		if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2553 			if (xoap->xoa_immutable !=
2554 			    ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
2555 				need_policy = TRUE;
2556 			} else {
2557 				XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
2558 				XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE);
2559 			}
2560 		}
2561 
2562 		if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2563 			if (xoap->xoa_nodump !=
2564 			    ((zp->z_pflags & ZFS_NODUMP) != 0)) {
2565 				need_policy = TRUE;
2566 			} else {
2567 				XVA_CLR_REQ(xvap, XAT_NODUMP);
2568 				XVA_SET_REQ(&tmpxvattr, XAT_NODUMP);
2569 			}
2570 		}
2571 
2572 		if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2573 			if (xoap->xoa_av_modified !=
2574 			    ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
2575 				need_policy = TRUE;
2576 			} else {
2577 				XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
2578 				XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED);
2579 			}
2580 		}
2581 
2582 		if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2583 			if ((vp->v_type != VREG &&
2584 			    xoap->xoa_av_quarantined) ||
2585 			    xoap->xoa_av_quarantined !=
2586 			    ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
2587 				need_policy = TRUE;
2588 			} else {
2589 				XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
2590 				XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED);
2591 			}
2592 		}
2593 
2594 		if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2595 			zfs_exit(zfsvfs, FTAG);
2596 			return (SET_ERROR(EPERM));
2597 		}
2598 
2599 		if (need_policy == FALSE &&
2600 		    (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
2601 		    XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
2602 			need_policy = TRUE;
2603 		}
2604 	}
2605 
2606 	if (mask & AT_MODE) {
2607 		if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr,
2608 		    mnt_ns) == 0) {
2609 			err = secpolicy_setid_setsticky_clear(vp, vap,
2610 			    &oldva, cr);
2611 			if (err) {
2612 				zfs_exit(zfsvfs, FTAG);
2613 				return (err);
2614 			}
2615 			trim_mask |= AT_MODE;
2616 		} else {
2617 			need_policy = TRUE;
2618 		}
2619 	}
2620 
2621 	if (need_policy) {
2622 		/*
2623 		 * If trim_mask is set then take ownership
2624 		 * has been granted or write_acl is present and user
2625 		 * has the ability to modify mode.  In that case remove
2626 		 * UID|GID and or MODE from mask so that
2627 		 * secpolicy_vnode_setattr() doesn't revoke it.
2628 		 */
2629 
2630 		if (trim_mask) {
2631 			saved_mask = vap->va_mask;
2632 			vap->va_mask &= ~trim_mask;
2633 			if (trim_mask & AT_MODE) {
2634 				/*
2635 				 * Save the mode, as secpolicy_vnode_setattr()
2636 				 * will overwrite it with ova.va_mode.
2637 				 */
2638 				saved_mode = vap->va_mode;
2639 			}
2640 		}
2641 		err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
2642 		    (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
2643 		if (err) {
2644 			zfs_exit(zfsvfs, FTAG);
2645 			return (err);
2646 		}
2647 
2648 		if (trim_mask) {
2649 			vap->va_mask |= saved_mask;
2650 			if (trim_mask & AT_MODE) {
2651 				/*
2652 				 * Recover the mode after
2653 				 * secpolicy_vnode_setattr().
2654 				 */
2655 				vap->va_mode = saved_mode;
2656 			}
2657 		}
2658 	}
2659 
2660 	/*
2661 	 * secpolicy_vnode_setattr, or take ownership may have
2662 	 * changed va_mask
2663 	 */
2664 	mask = vap->va_mask;
2665 
2666 	if ((mask & (AT_UID | AT_GID)) || projid != ZFS_INVALID_PROJID) {
2667 		handle_eadir = B_TRUE;
2668 		err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
2669 		    &xattr_obj, sizeof (xattr_obj));
2670 
2671 		if (err == 0 && xattr_obj) {
2672 			err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp);
2673 			if (err == 0) {
2674 				err = vn_lock(ZTOV(attrzp), LK_EXCLUSIVE);
2675 				if (err != 0)
2676 					vrele(ZTOV(attrzp));
2677 			}
2678 			if (err)
2679 				goto out2;
2680 		}
2681 		if (mask & AT_UID) {
2682 			new_uid = zfs_fuid_create(zfsvfs,
2683 			    (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
2684 			if (new_uid != zp->z_uid &&
2685 			    zfs_id_overquota(zfsvfs, DMU_USERUSED_OBJECT,
2686 			    new_uid)) {
2687 				if (attrzp)
2688 					vput(ZTOV(attrzp));
2689 				err = SET_ERROR(EDQUOT);
2690 				goto out2;
2691 			}
2692 		}
2693 
2694 		if (mask & AT_GID) {
2695 			new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid,
2696 			    cr, ZFS_GROUP, &fuidp);
2697 			if (new_gid != zp->z_gid &&
2698 			    zfs_id_overquota(zfsvfs, DMU_GROUPUSED_OBJECT,
2699 			    new_gid)) {
2700 				if (attrzp)
2701 					vput(ZTOV(attrzp));
2702 				err = SET_ERROR(EDQUOT);
2703 				goto out2;
2704 			}
2705 		}
2706 
2707 		if (projid != ZFS_INVALID_PROJID &&
2708 		    zfs_id_overquota(zfsvfs, DMU_PROJECTUSED_OBJECT, projid)) {
2709 			if (attrzp)
2710 				vput(ZTOV(attrzp));
2711 			err = SET_ERROR(EDQUOT);
2712 			goto out2;
2713 		}
2714 	}
2715 	tx = dmu_tx_create(os);
2716 
2717 	if (mask & AT_MODE) {
2718 		uint64_t pmode = zp->z_mode;
2719 		uint64_t acl_obj;
2720 		new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
2721 
2722 		if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED &&
2723 		    !(zp->z_pflags & ZFS_ACL_TRIVIAL)) {
2724 			err = SET_ERROR(EPERM);
2725 			goto out;
2726 		}
2727 
2728 		if ((err = zfs_acl_chmod_setattr(zp, &aclp, new_mode)))
2729 			goto out;
2730 
2731 		if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
2732 			/*
2733 			 * Are we upgrading ACL from old V0 format
2734 			 * to V1 format?
2735 			 */
2736 			if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
2737 			    zfs_znode_acl_version(zp) ==
2738 			    ZFS_ACL_VERSION_INITIAL) {
2739 				dmu_tx_hold_free(tx, acl_obj, 0,
2740 				    DMU_OBJECT_END);
2741 				dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
2742 				    0, aclp->z_acl_bytes);
2743 			} else {
2744 				dmu_tx_hold_write(tx, acl_obj, 0,
2745 				    aclp->z_acl_bytes);
2746 			}
2747 		} else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2748 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
2749 			    0, aclp->z_acl_bytes);
2750 		}
2751 		dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
2752 	} else {
2753 		if (((mask & AT_XVATTR) &&
2754 		    XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) ||
2755 		    (projid != ZFS_INVALID_PROJID &&
2756 		    !(zp->z_pflags & ZFS_PROJID)))
2757 			dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
2758 		else
2759 			dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2760 	}
2761 
2762 	if (attrzp) {
2763 		dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
2764 	}
2765 
2766 	fuid_dirtied = zfsvfs->z_fuid_dirty;
2767 	if (fuid_dirtied)
2768 		zfs_fuid_txhold(zfsvfs, tx);
2769 
2770 	zfs_sa_upgrade_txholds(tx, zp);
2771 
2772 	err = dmu_tx_assign(tx, DMU_TX_WAIT);
2773 	if (err)
2774 		goto out;
2775 
2776 	count = 0;
2777 	/*
2778 	 * Set each attribute requested.
2779 	 * We group settings according to the locks they need to acquire.
2780 	 *
2781 	 * Note: you cannot set ctime directly, although it will be
2782 	 * updated as a side-effect of calling this function.
2783 	 */
2784 
2785 	if (projid != ZFS_INVALID_PROJID && !(zp->z_pflags & ZFS_PROJID)) {
2786 		/*
2787 		 * For the existed object that is upgraded from old system,
2788 		 * its on-disk layout has no slot for the project ID attribute.
2789 		 * But quota accounting logic needs to access related slots by
2790 		 * offset directly. So we need to adjust old objects' layout
2791 		 * to make the project ID to some unified and fixed offset.
2792 		 */
2793 		if (attrzp)
2794 			err = sa_add_projid(attrzp->z_sa_hdl, tx, projid);
2795 		if (err == 0)
2796 			err = sa_add_projid(zp->z_sa_hdl, tx, projid);
2797 
2798 		if (unlikely(err == EEXIST))
2799 			err = 0;
2800 		else if (err != 0)
2801 			goto out;
2802 		else
2803 			projid = ZFS_INVALID_PROJID;
2804 	}
2805 
2806 	if (mask & (AT_UID|AT_GID|AT_MODE))
2807 		mutex_enter(&zp->z_acl_lock);
2808 
2809 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
2810 	    &zp->z_pflags, sizeof (zp->z_pflags));
2811 
2812 	if (attrzp) {
2813 		if (mask & (AT_UID|AT_GID|AT_MODE))
2814 			mutex_enter(&attrzp->z_acl_lock);
2815 		SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
2816 		    SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags,
2817 		    sizeof (attrzp->z_pflags));
2818 		if (projid != ZFS_INVALID_PROJID) {
2819 			attrzp->z_projid = projid;
2820 			SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
2821 			    SA_ZPL_PROJID(zfsvfs), NULL, &attrzp->z_projid,
2822 			    sizeof (attrzp->z_projid));
2823 		}
2824 	}
2825 
2826 	if (mask & (AT_UID|AT_GID)) {
2827 
2828 		if (mask & AT_UID) {
2829 			SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
2830 			    &new_uid, sizeof (new_uid));
2831 			zp->z_uid = new_uid;
2832 			if (attrzp) {
2833 				SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
2834 				    SA_ZPL_UID(zfsvfs), NULL, &new_uid,
2835 				    sizeof (new_uid));
2836 				attrzp->z_uid = new_uid;
2837 			}
2838 		}
2839 
2840 		if (mask & AT_GID) {
2841 			SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs),
2842 			    NULL, &new_gid, sizeof (new_gid));
2843 			zp->z_gid = new_gid;
2844 			if (attrzp) {
2845 				SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
2846 				    SA_ZPL_GID(zfsvfs), NULL, &new_gid,
2847 				    sizeof (new_gid));
2848 				attrzp->z_gid = new_gid;
2849 			}
2850 		}
2851 		if (!(mask & AT_MODE)) {
2852 			SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs),
2853 			    NULL, &new_mode, sizeof (new_mode));
2854 			new_mode = zp->z_mode;
2855 		}
2856 		err = zfs_acl_chown_setattr(zp);
2857 		ASSERT0(err);
2858 		if (attrzp) {
2859 			vn_seqc_write_begin(ZTOV(attrzp));
2860 			err = zfs_acl_chown_setattr(attrzp);
2861 			vn_seqc_write_end(ZTOV(attrzp));
2862 			ASSERT0(err);
2863 		}
2864 	}
2865 
2866 	if (mask & AT_MODE) {
2867 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
2868 		    &new_mode, sizeof (new_mode));
2869 		zp->z_mode = new_mode;
2870 		ASSERT3P(aclp, !=, NULL);
2871 		err = zfs_aclset_common(zp, aclp, cr, tx);
2872 		ASSERT0(err);
2873 		if (zp->z_acl_cached)
2874 			zfs_acl_free(zp->z_acl_cached);
2875 		zp->z_acl_cached = aclp;
2876 		aclp = NULL;
2877 	}
2878 
2879 
2880 	if (mask & AT_ATIME) {
2881 		ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime);
2882 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
2883 		    &zp->z_atime, sizeof (zp->z_atime));
2884 	}
2885 
2886 	if (mask & AT_MTIME) {
2887 		ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
2888 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
2889 		    mtime, sizeof (mtime));
2890 	}
2891 
2892 	if (projid != ZFS_INVALID_PROJID) {
2893 		zp->z_projid = projid;
2894 		SA_ADD_BULK_ATTR(bulk, count,
2895 		    SA_ZPL_PROJID(zfsvfs), NULL, &zp->z_projid,
2896 		    sizeof (zp->z_projid));
2897 	}
2898 
2899 	/* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
2900 	if (mask & AT_SIZE && !(mask & AT_MTIME)) {
2901 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs),
2902 		    NULL, mtime, sizeof (mtime));
2903 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
2904 		    &ctime, sizeof (ctime));
2905 		zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime);
2906 	} else if (mask != 0) {
2907 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
2908 		    &ctime, sizeof (ctime));
2909 		zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime);
2910 		if (attrzp) {
2911 			SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
2912 			    SA_ZPL_CTIME(zfsvfs), NULL,
2913 			    &ctime, sizeof (ctime));
2914 			zfs_tstamp_update_setup(attrzp, STATE_CHANGED,
2915 			    mtime, ctime);
2916 		}
2917 	}
2918 
2919 	/*
2920 	 * Do this after setting timestamps to prevent timestamp
2921 	 * update from toggling bit
2922 	 */
2923 
2924 	if (xoap && (mask & AT_XVATTR)) {
2925 
2926 		if (XVA_ISSET_REQ(xvap, XAT_CREATETIME))
2927 			xoap->xoa_createtime = vap->va_birthtime;
2928 		/*
2929 		 * restore trimmed off masks
2930 		 * so that return masks can be set for caller.
2931 		 */
2932 
2933 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) {
2934 			XVA_SET_REQ(xvap, XAT_APPENDONLY);
2935 		}
2936 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) {
2937 			XVA_SET_REQ(xvap, XAT_NOUNLINK);
2938 		}
2939 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) {
2940 			XVA_SET_REQ(xvap, XAT_IMMUTABLE);
2941 		}
2942 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) {
2943 			XVA_SET_REQ(xvap, XAT_NODUMP);
2944 		}
2945 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) {
2946 			XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
2947 		}
2948 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) {
2949 			XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
2950 		}
2951 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_PROJINHERIT)) {
2952 			XVA_SET_REQ(xvap, XAT_PROJINHERIT);
2953 		}
2954 
2955 		if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
2956 			ASSERT3S(vp->v_type, ==, VREG);
2957 
2958 		zfs_xvattr_set(zp, xvap, tx);
2959 	}
2960 
2961 	if (fuid_dirtied)
2962 		zfs_fuid_sync(zfsvfs, tx);
2963 
2964 	if (mask != 0)
2965 		zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
2966 
2967 	if (mask & (AT_UID|AT_GID|AT_MODE))
2968 		mutex_exit(&zp->z_acl_lock);
2969 
2970 	if (attrzp) {
2971 		if (mask & (AT_UID|AT_GID|AT_MODE))
2972 			mutex_exit(&attrzp->z_acl_lock);
2973 	}
2974 out:
2975 	if (err == 0 && attrzp) {
2976 		err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
2977 		    xattr_count, tx);
2978 		ASSERT0(err2);
2979 	}
2980 
2981 	if (attrzp)
2982 		vput(ZTOV(attrzp));
2983 
2984 	if (aclp)
2985 		zfs_acl_free(aclp);
2986 
2987 	if (fuidp) {
2988 		zfs_fuid_info_free(fuidp);
2989 		fuidp = NULL;
2990 	}
2991 
2992 	if (err) {
2993 		dmu_tx_abort(tx);
2994 	} else {
2995 		err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
2996 		dmu_tx_commit(tx);
2997 		if (attrzp) {
2998 			if (err2 == 0 && handle_eadir)
2999 				err = zfs_setattr_dir(attrzp);
3000 		}
3001 	}
3002 
3003 out2:
3004 	if (os->os_sync == ZFS_SYNC_ALWAYS)
3005 		zil_commit(zilog, 0);
3006 
3007 	zfs_exit(zfsvfs, FTAG);
3008 	return (err);
3009 }
3010 
3011 /*
3012  * Look up the directory entries corresponding to the source and target
3013  * directory/name pairs.
3014  */
3015 static int
zfs_rename_relock_lookup(znode_t * sdzp,const struct componentname * scnp,znode_t ** szpp,znode_t * tdzp,const struct componentname * tcnp,znode_t ** tzpp)3016 zfs_rename_relock_lookup(znode_t *sdzp, const struct componentname *scnp,
3017     znode_t **szpp, znode_t *tdzp, const struct componentname *tcnp,
3018     znode_t **tzpp)
3019 {
3020 	zfsvfs_t *zfsvfs;
3021 	znode_t *szp, *tzp;
3022 	int error;
3023 
3024 	/*
3025 	 * Before using sdzp and tdzp we must ensure that they are live.
3026 	 * As a porting legacy from illumos we have two things to worry
3027 	 * about.  One is typical for FreeBSD and it is that the vnode is
3028 	 * not reclaimed (doomed).  The other is that the znode is live.
3029 	 * The current code can invalidate the znode without acquiring the
3030 	 * corresponding vnode lock if the object represented by the znode
3031 	 * and vnode is no longer valid after a rollback or receive operation.
3032 	 * z_teardown_lock hidden behind zfs_enter and zfs_exit is the lock
3033 	 * that protects the znodes from the invalidation.
3034 	 */
3035 	zfsvfs = sdzp->z_zfsvfs;
3036 	ASSERT3P(zfsvfs, ==, tdzp->z_zfsvfs);
3037 	if ((error = zfs_enter_verify_zp(zfsvfs, sdzp, FTAG)) != 0)
3038 		return (error);
3039 	if ((error = zfs_verify_zp(tdzp)) != 0) {
3040 		zfs_exit(zfsvfs, FTAG);
3041 		return (error);
3042 	}
3043 
3044 	/*
3045 	 * Re-resolve svp to be certain it still exists and fetch the
3046 	 * correct vnode.
3047 	 */
3048 	error = zfs_dirent_lookup(sdzp, scnp->cn_nameptr, &szp, ZEXISTS);
3049 	if (error != 0) {
3050 		/* Source entry invalid or not there. */
3051 		if ((scnp->cn_flags & ISDOTDOT) != 0 ||
3052 		    (scnp->cn_namelen == 1 && scnp->cn_nameptr[0] == '.'))
3053 			error = SET_ERROR(EINVAL);
3054 		goto out;
3055 	}
3056 	*szpp = szp;
3057 
3058 	/*
3059 	 * Re-resolve tvp, if it disappeared we just carry on.
3060 	 */
3061 	error = zfs_dirent_lookup(tdzp, tcnp->cn_nameptr, &tzp, 0);
3062 	if (error != 0) {
3063 		vrele(ZTOV(szp));
3064 		if ((tcnp->cn_flags & ISDOTDOT) != 0)
3065 			error = SET_ERROR(EINVAL);
3066 		goto out;
3067 	}
3068 	*tzpp = tzp;
3069 out:
3070 	zfs_exit(zfsvfs, FTAG);
3071 	return (error);
3072 }
3073 
3074 /*
3075  * We acquire all but fdvp locks using non-blocking acquisitions.  If we
3076  * fail to acquire any lock in the path we will drop all held locks,
3077  * acquire the new lock in a blocking fashion, and then release it and
3078  * restart the rename.  This acquire/release step ensures that we do not
3079  * spin on a lock waiting for release.  On error release all vnode locks
3080  * and decrement references the way tmpfs_rename() would do.
3081  */
3082 static int
zfs_rename_relock(struct vnode * sdvp,struct vnode ** svpp,struct vnode * tdvp,struct vnode ** tvpp,const struct componentname * scnp,const struct componentname * tcnp)3083 zfs_rename_relock(struct vnode *sdvp, struct vnode **svpp,
3084     struct vnode *tdvp, struct vnode **tvpp,
3085     const struct componentname *scnp, const struct componentname *tcnp)
3086 {
3087 	struct vnode	*nvp, *svp, *tvp;
3088 	znode_t		*sdzp, *tdzp, *szp, *tzp;
3089 	int		error;
3090 
3091 	VOP_UNLOCK(tdvp);
3092 	if (*tvpp != NULL && *tvpp != tdvp)
3093 		VOP_UNLOCK(*tvpp);
3094 
3095 relock:
3096 	error = vn_lock(sdvp, LK_EXCLUSIVE);
3097 	if (error)
3098 		goto out;
3099 	error = vn_lock(tdvp, LK_EXCLUSIVE | LK_NOWAIT);
3100 	if (error != 0) {
3101 		VOP_UNLOCK(sdvp);
3102 		if (error != EBUSY)
3103 			goto out;
3104 		error = vn_lock(tdvp, LK_EXCLUSIVE);
3105 		if (error)
3106 			goto out;
3107 		VOP_UNLOCK(tdvp);
3108 		goto relock;
3109 	}
3110 	tdzp = VTOZ(tdvp);
3111 	sdzp = VTOZ(sdvp);
3112 
3113 	error = zfs_rename_relock_lookup(sdzp, scnp, &szp, tdzp, tcnp, &tzp);
3114 	if (error != 0) {
3115 		VOP_UNLOCK(sdvp);
3116 		VOP_UNLOCK(tdvp);
3117 		goto out;
3118 	}
3119 	svp = ZTOV(szp);
3120 	tvp = tzp != NULL ? ZTOV(tzp) : NULL;
3121 
3122 	/*
3123 	 * Now try acquire locks on svp and tvp.
3124 	 */
3125 	nvp = svp;
3126 	error = vn_lock(nvp, LK_EXCLUSIVE | LK_NOWAIT);
3127 	if (error != 0) {
3128 		VOP_UNLOCK(sdvp);
3129 		VOP_UNLOCK(tdvp);
3130 		if (tvp != NULL)
3131 			vrele(tvp);
3132 		if (error != EBUSY) {
3133 			vrele(nvp);
3134 			goto out;
3135 		}
3136 		error = vn_lock(nvp, LK_EXCLUSIVE);
3137 		if (error != 0) {
3138 			vrele(nvp);
3139 			goto out;
3140 		}
3141 		VOP_UNLOCK(nvp);
3142 		/*
3143 		 * Concurrent rename race.
3144 		 * XXX ?
3145 		 */
3146 		if (nvp == tdvp) {
3147 			vrele(nvp);
3148 			error = SET_ERROR(EINVAL);
3149 			goto out;
3150 		}
3151 		vrele(*svpp);
3152 		*svpp = nvp;
3153 		goto relock;
3154 	}
3155 	vrele(*svpp);
3156 	*svpp = nvp;
3157 
3158 	if (*tvpp != NULL)
3159 		vrele(*tvpp);
3160 	*tvpp = NULL;
3161 	if (tvp != NULL) {
3162 		nvp = tvp;
3163 		error = vn_lock(nvp, LK_EXCLUSIVE | LK_NOWAIT);
3164 		if (error != 0) {
3165 			VOP_UNLOCK(sdvp);
3166 			VOP_UNLOCK(tdvp);
3167 			VOP_UNLOCK(*svpp);
3168 			if (error != EBUSY) {
3169 				vrele(nvp);
3170 				goto out;
3171 			}
3172 			error = vn_lock(nvp, LK_EXCLUSIVE);
3173 			if (error != 0) {
3174 				vrele(nvp);
3175 				goto out;
3176 			}
3177 			vput(nvp);
3178 			goto relock;
3179 		}
3180 		*tvpp = nvp;
3181 	}
3182 
3183 	return (0);
3184 
3185 out:
3186 	return (error);
3187 }
3188 
3189 /*
3190  * Note that we must use VRELE_ASYNC in this function as it walks
3191  * up the directory tree and vrele may need to acquire an exclusive
3192  * lock if a last reference to a vnode is dropped.
3193  */
3194 static int
zfs_rename_check(znode_t * szp,znode_t * sdzp,znode_t * tdzp)3195 zfs_rename_check(znode_t *szp, znode_t *sdzp, znode_t *tdzp)
3196 {
3197 	zfsvfs_t	*zfsvfs;
3198 	znode_t		*zp, *zp1;
3199 	uint64_t	parent;
3200 	int		error;
3201 
3202 	zfsvfs = tdzp->z_zfsvfs;
3203 	if (tdzp == szp)
3204 		return (SET_ERROR(EINVAL));
3205 	if (tdzp == sdzp)
3206 		return (0);
3207 	if (tdzp->z_id == zfsvfs->z_root)
3208 		return (0);
3209 	zp = tdzp;
3210 	for (;;) {
3211 		ASSERT(!zp->z_unlinked);
3212 		if ((error = sa_lookup(zp->z_sa_hdl,
3213 		    SA_ZPL_PARENT(zfsvfs), &parent, sizeof (parent))) != 0)
3214 			break;
3215 
3216 		if (parent == szp->z_id) {
3217 			error = SET_ERROR(EINVAL);
3218 			break;
3219 		}
3220 		if (parent == zfsvfs->z_root)
3221 			break;
3222 		if (parent == sdzp->z_id)
3223 			break;
3224 
3225 		error = zfs_zget(zfsvfs, parent, &zp1);
3226 		if (error != 0)
3227 			break;
3228 
3229 		if (zp != tdzp)
3230 			VN_RELE_ASYNC(ZTOV(zp),
3231 			    dsl_pool_zrele_taskq(
3232 			    dmu_objset_pool(zfsvfs->z_os)));
3233 		zp = zp1;
3234 	}
3235 
3236 	if (error == ENOTDIR)
3237 		panic("checkpath: .. not a directory\n");
3238 	if (zp != tdzp)
3239 		VN_RELE_ASYNC(ZTOV(zp),
3240 		    dsl_pool_zrele_taskq(dmu_objset_pool(zfsvfs->z_os)));
3241 	return (error);
3242 }
3243 
3244 static int
3245 zfs_do_rename_impl(vnode_t *sdvp, vnode_t **svpp, struct componentname *scnp,
3246     vnode_t *tdvp, vnode_t **tvpp, struct componentname *tcnp,
3247     cred_t *cr);
3248 
3249 /*
3250  * Move an entry from the provided source directory to the target
3251  * directory.  Change the entry name as indicated.
3252  *
3253  *	IN:	sdvp	- Source directory containing the "old entry".
3254  *		scnp	- Old entry name.
3255  *		tdvp	- Target directory to contain the "new entry".
3256  *		tcnp	- New entry name.
3257  *		cr	- credentials of caller.
3258  *	INOUT:	svpp	- Source file
3259  *		tvpp	- Target file, may point to NULL initially
3260  *
3261  *	RETURN:	0 on success, error code on failure.
3262  *
3263  * Timestamps:
3264  *	sdvp,tdvp - ctime|mtime updated
3265  */
3266 static int
zfs_do_rename(vnode_t * sdvp,vnode_t ** svpp,struct componentname * scnp,vnode_t * tdvp,vnode_t ** tvpp,struct componentname * tcnp,cred_t * cr)3267 zfs_do_rename(vnode_t *sdvp, vnode_t **svpp, struct componentname *scnp,
3268     vnode_t *tdvp, vnode_t **tvpp, struct componentname *tcnp,
3269     cred_t *cr)
3270 {
3271 	int	error;
3272 
3273 	ASSERT_VOP_ELOCKED(tdvp, __func__);
3274 	if (*tvpp != NULL)
3275 		ASSERT_VOP_ELOCKED(*tvpp, __func__);
3276 
3277 	/* Reject renames across filesystems. */
3278 	if ((*svpp)->v_mount != tdvp->v_mount ||
3279 	    ((*tvpp) != NULL && (*svpp)->v_mount != (*tvpp)->v_mount)) {
3280 		error = SET_ERROR(EXDEV);
3281 		goto out;
3282 	}
3283 
3284 	if (zfsctl_is_node(tdvp)) {
3285 		error = SET_ERROR(EXDEV);
3286 		goto out;
3287 	}
3288 
3289 	/*
3290 	 * Lock all four vnodes to ensure safety and semantics of renaming.
3291 	 */
3292 	error = zfs_rename_relock(sdvp, svpp, tdvp, tvpp, scnp, tcnp);
3293 	if (error != 0) {
3294 		/* no vnodes are locked in the case of error here */
3295 		return (error);
3296 	}
3297 
3298 	error = zfs_do_rename_impl(sdvp, svpp, scnp, tdvp, tvpp, tcnp, cr);
3299 	VOP_UNLOCK(sdvp);
3300 	VOP_UNLOCK(*svpp);
3301 out:
3302 	if (*tvpp != NULL)
3303 		VOP_UNLOCK(*tvpp);
3304 	if (tdvp != *tvpp)
3305 		VOP_UNLOCK(tdvp);
3306 
3307 	return (error);
3308 }
3309 
3310 static int
zfs_do_rename_impl(vnode_t * sdvp,vnode_t ** svpp,struct componentname * scnp,vnode_t * tdvp,vnode_t ** tvpp,struct componentname * tcnp,cred_t * cr)3311 zfs_do_rename_impl(vnode_t *sdvp, vnode_t **svpp, struct componentname *scnp,
3312     vnode_t *tdvp, vnode_t **tvpp, struct componentname *tcnp,
3313     cred_t *cr)
3314 {
3315 	dmu_tx_t	*tx;
3316 	zfsvfs_t	*zfsvfs;
3317 	zilog_t		*zilog;
3318 	znode_t		*tdzp, *sdzp, *tzp, *szp;
3319 	const char	*snm = scnp->cn_nameptr;
3320 	const char	*tnm = tcnp->cn_nameptr;
3321 	int		error;
3322 
3323 	tdzp = VTOZ(tdvp);
3324 	sdzp = VTOZ(sdvp);
3325 	zfsvfs = tdzp->z_zfsvfs;
3326 
3327 	if ((error = zfs_enter_verify_zp(zfsvfs, tdzp, FTAG)) != 0)
3328 		return (error);
3329 	if ((error = zfs_verify_zp(sdzp)) != 0) {
3330 		zfs_exit(zfsvfs, FTAG);
3331 		return (error);
3332 	}
3333 	zilog = zfsvfs->z_log;
3334 
3335 	if (zfsvfs->z_utf8 && u8_validate(tnm,
3336 	    strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3337 		error = SET_ERROR(EILSEQ);
3338 		goto out;
3339 	}
3340 
3341 	/* If source and target are the same file, there is nothing to do. */
3342 	if ((*svpp) == (*tvpp)) {
3343 		error = 0;
3344 		goto out;
3345 	}
3346 
3347 	if (((*svpp)->v_type == VDIR && (*svpp)->v_mountedhere != NULL) ||
3348 	    ((*tvpp) != NULL && (*tvpp)->v_type == VDIR &&
3349 	    (*tvpp)->v_mountedhere != NULL)) {
3350 		error = SET_ERROR(EXDEV);
3351 		goto out;
3352 	}
3353 
3354 	szp = VTOZ(*svpp);
3355 	if ((error = zfs_verify_zp(szp)) != 0) {
3356 		zfs_exit(zfsvfs, FTAG);
3357 		return (error);
3358 	}
3359 	tzp = *tvpp == NULL ? NULL : VTOZ(*tvpp);
3360 	if (tzp != NULL) {
3361 		if ((error = zfs_verify_zp(tzp)) != 0) {
3362 			zfs_exit(zfsvfs, FTAG);
3363 			return (error);
3364 		}
3365 	}
3366 
3367 	/*
3368 	 * This is to prevent the creation of links into attribute space
3369 	 * by renaming a linked file into/outof an attribute directory.
3370 	 * See the comment in zfs_link() for why this is considered bad.
3371 	 */
3372 	if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) {
3373 		error = SET_ERROR(EINVAL);
3374 		goto out;
3375 	}
3376 
3377 	/*
3378 	 * If we are using project inheritance, means if the directory has
3379 	 * ZFS_PROJINHERIT set, then its descendant directories will inherit
3380 	 * not only the project ID, but also the ZFS_PROJINHERIT flag. Under
3381 	 * such case, we only allow renames into our tree when the project
3382 	 * IDs are the same.
3383 	 */
3384 	if (tdzp->z_pflags & ZFS_PROJINHERIT &&
3385 	    tdzp->z_projid != szp->z_projid) {
3386 		error = SET_ERROR(EXDEV);
3387 		goto out;
3388 	}
3389 
3390 	/*
3391 	 * Must have write access at the source to remove the old entry
3392 	 * and write access at the target to create the new entry.
3393 	 * Note that if target and source are the same, this can be
3394 	 * done in a single check.
3395 	 */
3396 	if ((error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr, NULL)))
3397 		goto out;
3398 
3399 	if ((*svpp)->v_type == VDIR) {
3400 		/*
3401 		 * Avoid ".", "..", and aliases of "." for obvious reasons.
3402 		 */
3403 		if ((scnp->cn_namelen == 1 && scnp->cn_nameptr[0] == '.') ||
3404 		    sdzp == szp ||
3405 		    (scnp->cn_flags | tcnp->cn_flags) & ISDOTDOT) {
3406 			error = EINVAL;
3407 			goto out;
3408 		}
3409 
3410 		/*
3411 		 * Check to make sure rename is valid.
3412 		 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3413 		 */
3414 		if ((error = zfs_rename_check(szp, sdzp, tdzp)))
3415 			goto out;
3416 	}
3417 
3418 	/*
3419 	 * Does target exist?
3420 	 */
3421 	if (tzp) {
3422 		/*
3423 		 * Source and target must be the same type.
3424 		 */
3425 		if ((*svpp)->v_type == VDIR) {
3426 			if ((*tvpp)->v_type != VDIR) {
3427 				error = SET_ERROR(ENOTDIR);
3428 				goto out;
3429 			} else {
3430 				cache_purge(tdvp);
3431 				if (sdvp != tdvp)
3432 					cache_purge(sdvp);
3433 			}
3434 		} else {
3435 			if ((*tvpp)->v_type == VDIR) {
3436 				error = SET_ERROR(EISDIR);
3437 				goto out;
3438 			}
3439 		}
3440 	}
3441 
3442 	vn_seqc_write_begin(*svpp);
3443 	vn_seqc_write_begin(sdvp);
3444 	if (*tvpp != NULL)
3445 		vn_seqc_write_begin(*tvpp);
3446 	if (tdvp != *tvpp)
3447 		vn_seqc_write_begin(tdvp);
3448 
3449 	vnevent_rename_src(*svpp, sdvp, scnp->cn_nameptr, ct);
3450 	if (tzp)
3451 		vnevent_rename_dest(*tvpp, tdvp, tnm, ct);
3452 
3453 	/*
3454 	 * notify the target directory if it is not the same
3455 	 * as source directory.
3456 	 */
3457 	if (tdvp != sdvp) {
3458 		vnevent_rename_dest_dir(tdvp, ct);
3459 	}
3460 
3461 	tx = dmu_tx_create(zfsvfs->z_os);
3462 	dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3463 	dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
3464 	dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3465 	dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3466 	if (sdzp != tdzp) {
3467 		dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
3468 		zfs_sa_upgrade_txholds(tx, tdzp);
3469 	}
3470 	if (tzp) {
3471 		dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
3472 		zfs_sa_upgrade_txholds(tx, tzp);
3473 	}
3474 
3475 	zfs_sa_upgrade_txholds(tx, szp);
3476 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
3477 	error = dmu_tx_assign(tx, DMU_TX_WAIT);
3478 	if (error) {
3479 		dmu_tx_abort(tx);
3480 		goto out_seq;
3481 	}
3482 
3483 	if (tzp)	/* Attempt to remove the existing target */
3484 		error = zfs_link_destroy(tdzp, tnm, tzp, tx, 0, NULL);
3485 
3486 	if (error == 0) {
3487 		error = zfs_link_create(tdzp, tnm, szp, tx, ZRENAMING);
3488 		if (error == 0) {
3489 			szp->z_pflags |= ZFS_AV_MODIFIED;
3490 
3491 			error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs),
3492 			    (void *)&szp->z_pflags, sizeof (uint64_t), tx);
3493 			ASSERT0(error);
3494 
3495 			error = zfs_link_destroy(sdzp, snm, szp, tx, ZRENAMING,
3496 			    NULL);
3497 			if (error == 0) {
3498 				zfs_log_rename(zilog, tx, TX_RENAME, sdzp,
3499 				    snm, tdzp, tnm, szp);
3500 			} else {
3501 				/*
3502 				 * At this point, we have successfully created
3503 				 * the target name, but have failed to remove
3504 				 * the source name.  Since the create was done
3505 				 * with the ZRENAMING flag, there are
3506 				 * complications; for one, the link count is
3507 				 * wrong.  The easiest way to deal with this
3508 				 * is to remove the newly created target, and
3509 				 * return the original error.  This must
3510 				 * succeed; fortunately, it is very unlikely to
3511 				 * fail, since we just created it.
3512 				 */
3513 				VERIFY0(zfs_link_destroy(tdzp, tnm, szp, tx,
3514 				    ZRENAMING, NULL));
3515 			}
3516 		}
3517 		if (error == 0) {
3518 			cache_vop_rename(sdvp, *svpp, tdvp, *tvpp, scnp, tcnp);
3519 		}
3520 	}
3521 
3522 	dmu_tx_commit(tx);
3523 
3524 out_seq:
3525 	vn_seqc_write_end(*svpp);
3526 	vn_seqc_write_end(sdvp);
3527 	if (*tvpp != NULL)
3528 		vn_seqc_write_end(*tvpp);
3529 	if (tdvp != *tvpp)
3530 		vn_seqc_write_end(tdvp);
3531 
3532 out:
3533 	if (error == 0 && zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3534 		zil_commit(zilog, 0);
3535 	zfs_exit(zfsvfs, FTAG);
3536 
3537 	return (error);
3538 }
3539 
3540 int
zfs_rename(znode_t * sdzp,const char * sname,znode_t * tdzp,const char * tname,cred_t * cr,int flags,uint64_t rflags,vattr_t * wo_vap,zidmap_t * mnt_ns)3541 zfs_rename(znode_t *sdzp, const char *sname, znode_t *tdzp, const char *tname,
3542     cred_t *cr, int flags, uint64_t rflags, vattr_t *wo_vap, zidmap_t *mnt_ns)
3543 {
3544 	struct componentname scn, tcn;
3545 	vnode_t *sdvp, *tdvp;
3546 	vnode_t *svp, *tvp;
3547 	int error;
3548 	svp = tvp = NULL;
3549 
3550 	if (is_nametoolong(tdzp->z_zfsvfs, tname))
3551 		return (SET_ERROR(ENAMETOOLONG));
3552 
3553 	if (rflags != 0 || wo_vap != NULL)
3554 		return (SET_ERROR(EINVAL));
3555 
3556 	sdvp = ZTOV(sdzp);
3557 	tdvp = ZTOV(tdzp);
3558 	error = zfs_lookup_internal(sdzp, sname, &svp, &scn, DELETE);
3559 	if (sdzp->z_zfsvfs->z_replay == B_FALSE)
3560 		VOP_UNLOCK(sdvp);
3561 	if (error != 0)
3562 		goto fail;
3563 	VOP_UNLOCK(svp);
3564 
3565 	vn_lock(tdvp, LK_EXCLUSIVE | LK_RETRY);
3566 	error = zfs_lookup_internal(tdzp, tname, &tvp, &tcn, RENAME);
3567 	if (error == EJUSTRETURN)
3568 		tvp = NULL;
3569 	else if (error != 0) {
3570 		VOP_UNLOCK(tdvp);
3571 		goto fail;
3572 	}
3573 
3574 	error = zfs_do_rename(sdvp, &svp, &scn, tdvp, &tvp, &tcn, cr);
3575 fail:
3576 	if (svp != NULL)
3577 		vrele(svp);
3578 	if (tvp != NULL)
3579 		vrele(tvp);
3580 
3581 	return (error);
3582 }
3583 
3584 /*
3585  * Insert the indicated symbolic reference entry into the directory.
3586  *
3587  *	IN:	dvp	- Directory to contain new symbolic link.
3588  *		link	- Name for new symlink entry.
3589  *		vap	- Attributes of new entry.
3590  *		cr	- credentials of caller.
3591  *		ct	- caller context
3592  *		flags	- case flags
3593  *		mnt_ns	- Unused on FreeBSD
3594  *
3595  *	RETURN:	0 on success, error code on failure.
3596  *
3597  * Timestamps:
3598  *	dvp - ctime|mtime updated
3599  */
3600 int
zfs_symlink(znode_t * dzp,const char * name,vattr_t * vap,const char * link,znode_t ** zpp,cred_t * cr,int flags,zidmap_t * mnt_ns)3601 zfs_symlink(znode_t *dzp, const char *name, vattr_t *vap,
3602     const char *link, znode_t **zpp, cred_t *cr, int flags, zidmap_t *mnt_ns)
3603 {
3604 	(void) flags;
3605 	znode_t		*zp;
3606 	dmu_tx_t	*tx;
3607 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
3608 	zilog_t		*zilog;
3609 	uint64_t	len = strlen(link);
3610 	int		error;
3611 	zfs_acl_ids_t	acl_ids;
3612 	boolean_t	fuid_dirtied;
3613 	uint64_t	txtype = TX_SYMLINK;
3614 
3615 	ASSERT3S(vap->va_type, ==, VLNK);
3616 
3617 	if (is_nametoolong(zfsvfs, name))
3618 		return (SET_ERROR(ENAMETOOLONG));
3619 
3620 	if ((error = zfs_enter_verify_zp(zfsvfs, dzp, FTAG)) != 0)
3621 		return (error);
3622 	zilog = zfsvfs->z_log;
3623 
3624 	if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
3625 	    NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3626 		zfs_exit(zfsvfs, FTAG);
3627 		return (SET_ERROR(EILSEQ));
3628 	}
3629 
3630 	if (len > MAXPATHLEN) {
3631 		zfs_exit(zfsvfs, FTAG);
3632 		return (SET_ERROR(ENAMETOOLONG));
3633 	}
3634 
3635 	if ((error = zfs_acl_ids_create(dzp, 0,
3636 	    vap, cr, NULL, &acl_ids, NULL)) != 0) {
3637 		zfs_exit(zfsvfs, FTAG);
3638 		return (error);
3639 	}
3640 
3641 	/*
3642 	 * Attempt to lock directory; fail if entry already exists.
3643 	 */
3644 	error = zfs_dirent_lookup(dzp, name, &zp, ZNEW);
3645 	if (error) {
3646 		zfs_acl_ids_free(&acl_ids);
3647 		zfs_exit(zfsvfs, FTAG);
3648 		return (error);
3649 	}
3650 
3651 	if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr, mnt_ns))) {
3652 		zfs_acl_ids_free(&acl_ids);
3653 		zfs_exit(zfsvfs, FTAG);
3654 		return (error);
3655 	}
3656 
3657 	if (zfs_acl_ids_overquota(zfsvfs, &acl_ids, ZFS_DEFAULT_PROJID)) {
3658 		zfs_acl_ids_free(&acl_ids);
3659 		zfs_exit(zfsvfs, FTAG);
3660 		return (SET_ERROR(EDQUOT));
3661 	}
3662 
3663 	getnewvnode_reserve();
3664 	tx = dmu_tx_create(zfsvfs->z_os);
3665 	fuid_dirtied = zfsvfs->z_fuid_dirty;
3666 	dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
3667 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
3668 	dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
3669 	    ZFS_SA_BASE_ATTR_SIZE + len);
3670 	dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
3671 	if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3672 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
3673 		    acl_ids.z_aclp->z_acl_bytes);
3674 	}
3675 	if (fuid_dirtied)
3676 		zfs_fuid_txhold(zfsvfs, tx);
3677 	error = dmu_tx_assign(tx, DMU_TX_WAIT);
3678 	if (error) {
3679 		zfs_acl_ids_free(&acl_ids);
3680 		dmu_tx_abort(tx);
3681 		getnewvnode_drop_reserve();
3682 		zfs_exit(zfsvfs, FTAG);
3683 		return (error);
3684 	}
3685 
3686 	/*
3687 	 * Create a new object for the symlink.
3688 	 * for version 4 ZPL datasets the symlink will be an SA attribute
3689 	 */
3690 	zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
3691 
3692 	if (fuid_dirtied)
3693 		zfs_fuid_sync(zfsvfs, tx);
3694 
3695 	if (zp->z_is_sa)
3696 		error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs),
3697 		    __DECONST(void *, link), len, tx);
3698 	else
3699 		zfs_sa_symlink(zp, __DECONST(char *, link), len, tx);
3700 
3701 	zp->z_size = len;
3702 	(void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
3703 	    &zp->z_size, sizeof (zp->z_size), tx);
3704 	/*
3705 	 * Insert the new object into the directory.
3706 	 */
3707 	error = zfs_link_create(dzp, name, zp, tx, ZNEW);
3708 	if (error != 0) {
3709 		zfs_znode_delete(zp, tx);
3710 		VOP_UNLOCK(ZTOV(zp));
3711 		zrele(zp);
3712 	} else {
3713 		zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
3714 	}
3715 
3716 	zfs_acl_ids_free(&acl_ids);
3717 
3718 	dmu_tx_commit(tx);
3719 
3720 	getnewvnode_drop_reserve();
3721 
3722 	if (error == 0) {
3723 		*zpp = zp;
3724 
3725 		if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3726 			zil_commit(zilog, 0);
3727 	}
3728 
3729 	zfs_exit(zfsvfs, FTAG);
3730 	return (error);
3731 }
3732 
3733 /*
3734  * Return, in the buffer contained in the provided uio structure,
3735  * the symbolic path referred to by vp.
3736  *
3737  *	IN:	vp	- vnode of symbolic link.
3738  *		uio	- structure to contain the link path.
3739  *		cr	- credentials of caller.
3740  *		ct	- caller context
3741  *
3742  *	OUT:	uio	- structure containing the link path.
3743  *
3744  *	RETURN:	0 on success, error code on failure.
3745  *
3746  * Timestamps:
3747  *	vp - atime updated
3748  */
3749 static int
zfs_readlink(vnode_t * vp,zfs_uio_t * uio,cred_t * cr,caller_context_t * ct)3750 zfs_readlink(vnode_t *vp, zfs_uio_t *uio, cred_t *cr, caller_context_t *ct)
3751 {
3752 	(void) cr, (void) ct;
3753 	znode_t		*zp = VTOZ(vp);
3754 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
3755 	int		error;
3756 
3757 	if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0)
3758 		return (error);
3759 
3760 	if (zp->z_is_sa)
3761 		error = sa_lookup_uio(zp->z_sa_hdl,
3762 		    SA_ZPL_SYMLINK(zfsvfs), uio);
3763 	else
3764 		error = zfs_sa_readlink(zp, uio);
3765 
3766 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
3767 
3768 	zfs_exit(zfsvfs, FTAG);
3769 	return (error);
3770 }
3771 
3772 /*
3773  * Insert a new entry into directory tdvp referencing svp.
3774  *
3775  *	IN:	tdvp	- Directory to contain new entry.
3776  *		svp	- vnode of new entry.
3777  *		name	- name of new entry.
3778  *		cr	- credentials of caller.
3779  *
3780  *	RETURN:	0 on success, error code on failure.
3781  *
3782  * Timestamps:
3783  *	tdvp - ctime|mtime updated
3784  *	 svp - ctime updated
3785  */
3786 int
zfs_link(znode_t * tdzp,znode_t * szp,const char * name,cred_t * cr,int flags)3787 zfs_link(znode_t *tdzp, znode_t *szp, const char *name, cred_t *cr,
3788     int flags)
3789 {
3790 	(void) flags;
3791 	znode_t		*tzp;
3792 	zfsvfs_t	*zfsvfs = tdzp->z_zfsvfs;
3793 	zilog_t		*zilog;
3794 	dmu_tx_t	*tx;
3795 	int		error;
3796 	uint64_t	parent;
3797 	uid_t		owner;
3798 
3799 	ASSERT3S(ZTOV(tdzp)->v_type, ==, VDIR);
3800 
3801 	if (is_nametoolong(zfsvfs, name))
3802 		return (SET_ERROR(ENAMETOOLONG));
3803 
3804 	if ((error = zfs_enter_verify_zp(zfsvfs, tdzp, FTAG)) != 0)
3805 		return (error);
3806 	zilog = zfsvfs->z_log;
3807 
3808 	/*
3809 	 * POSIX dictates that we return EPERM here.
3810 	 * Better choices include ENOTSUP or EISDIR.
3811 	 */
3812 	if (ZTOV(szp)->v_type == VDIR) {
3813 		zfs_exit(zfsvfs, FTAG);
3814 		return (SET_ERROR(EPERM));
3815 	}
3816 
3817 	if ((error = zfs_verify_zp(szp)) != 0) {
3818 		zfs_exit(zfsvfs, FTAG);
3819 		return (error);
3820 	}
3821 
3822 	/*
3823 	 * If we are using project inheritance, means if the directory has
3824 	 * ZFS_PROJINHERIT set, then its descendant directories will inherit
3825 	 * not only the project ID, but also the ZFS_PROJINHERIT flag. Under
3826 	 * such case, we only allow hard link creation in our tree when the
3827 	 * project IDs are the same.
3828 	 */
3829 	if (tdzp->z_pflags & ZFS_PROJINHERIT &&
3830 	    tdzp->z_projid != szp->z_projid) {
3831 		zfs_exit(zfsvfs, FTAG);
3832 		return (SET_ERROR(EXDEV));
3833 	}
3834 
3835 	if (szp->z_pflags & (ZFS_APPENDONLY |
3836 	    ZFS_IMMUTABLE | ZFS_READONLY)) {
3837 		zfs_exit(zfsvfs, FTAG);
3838 		return (SET_ERROR(EPERM));
3839 	}
3840 
3841 	/* Prevent links to .zfs/shares files */
3842 
3843 	if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
3844 	    &parent, sizeof (uint64_t))) != 0) {
3845 		zfs_exit(zfsvfs, FTAG);
3846 		return (error);
3847 	}
3848 	if (parent == zfsvfs->z_shares_dir) {
3849 		zfs_exit(zfsvfs, FTAG);
3850 		return (SET_ERROR(EPERM));
3851 	}
3852 
3853 	if (zfsvfs->z_utf8 && u8_validate(name,
3854 	    strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3855 		zfs_exit(zfsvfs, FTAG);
3856 		return (SET_ERROR(EILSEQ));
3857 	}
3858 
3859 	/*
3860 	 * We do not support links between attributes and non-attributes
3861 	 * because of the potential security risk of creating links
3862 	 * into "normal" file space in order to circumvent restrictions
3863 	 * imposed in attribute space.
3864 	 */
3865 	if ((szp->z_pflags & ZFS_XATTR) != (tdzp->z_pflags & ZFS_XATTR)) {
3866 		zfs_exit(zfsvfs, FTAG);
3867 		return (SET_ERROR(EINVAL));
3868 	}
3869 
3870 
3871 	owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER);
3872 	if (owner != crgetuid(cr) && secpolicy_basic_link(ZTOV(szp), cr) != 0) {
3873 		zfs_exit(zfsvfs, FTAG);
3874 		return (SET_ERROR(EPERM));
3875 	}
3876 
3877 	if ((error = zfs_zaccess(tdzp, ACE_ADD_FILE, 0, B_FALSE, cr, NULL))) {
3878 		zfs_exit(zfsvfs, FTAG);
3879 		return (error);
3880 	}
3881 
3882 	/*
3883 	 * Attempt to lock directory; fail if entry already exists.
3884 	 */
3885 	error = zfs_dirent_lookup(tdzp, name, &tzp, ZNEW);
3886 	if (error) {
3887 		zfs_exit(zfsvfs, FTAG);
3888 		return (error);
3889 	}
3890 
3891 	tx = dmu_tx_create(zfsvfs->z_os);
3892 	dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3893 	dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, name);
3894 	zfs_sa_upgrade_txholds(tx, szp);
3895 	zfs_sa_upgrade_txholds(tx, tdzp);
3896 	error = dmu_tx_assign(tx, DMU_TX_WAIT);
3897 	if (error) {
3898 		dmu_tx_abort(tx);
3899 		zfs_exit(zfsvfs, FTAG);
3900 		return (error);
3901 	}
3902 
3903 	error = zfs_link_create(tdzp, name, szp, tx, 0);
3904 
3905 	if (error == 0) {
3906 		uint64_t txtype = TX_LINK;
3907 		zfs_log_link(zilog, tx, txtype, tdzp, szp, name);
3908 	}
3909 
3910 	dmu_tx_commit(tx);
3911 
3912 	if (error == 0) {
3913 		vnevent_link(ZTOV(szp), ct);
3914 	}
3915 
3916 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3917 		zil_commit(zilog, 0);
3918 
3919 	zfs_exit(zfsvfs, FTAG);
3920 	return (error);
3921 }
3922 
3923 /*
3924  * Free or allocate space in a file.  Currently, this function only
3925  * supports the `F_FREESP' command.  However, this command is somewhat
3926  * misnamed, as its functionality includes the ability to allocate as
3927  * well as free space.
3928  *
3929  *	IN:	ip	- inode of file to free data in.
3930  *		cmd	- action to take (only F_FREESP supported).
3931  *		bfp	- section of file to free/alloc.
3932  *		flag	- current file open mode flags.
3933  *		offset	- current file offset.
3934  *		cr	- credentials of caller.
3935  *
3936  *	RETURN:	0 on success, error code on failure.
3937  *
3938  * Timestamps:
3939  *	ip - ctime|mtime updated
3940  */
3941 int
zfs_space(znode_t * zp,int cmd,flock64_t * bfp,int flag,offset_t offset,cred_t * cr)3942 zfs_space(znode_t *zp, int cmd, flock64_t *bfp, int flag,
3943     offset_t offset, cred_t *cr)
3944 {
3945 	(void) offset;
3946 	zfsvfs_t	*zfsvfs = ZTOZSB(zp);
3947 	uint64_t	off, len;
3948 	int		error;
3949 
3950 	if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0)
3951 		return (error);
3952 
3953 	if (cmd != F_FREESP) {
3954 		zfs_exit(zfsvfs, FTAG);
3955 		return (SET_ERROR(EINVAL));
3956 	}
3957 
3958 	/*
3959 	 * Callers might not be able to detect properly that we are read-only,
3960 	 * so check it explicitly here.
3961 	 */
3962 	if (zfs_is_readonly(zfsvfs)) {
3963 		zfs_exit(zfsvfs, FTAG);
3964 		return (SET_ERROR(EROFS));
3965 	}
3966 
3967 	if (bfp->l_len < 0) {
3968 		zfs_exit(zfsvfs, FTAG);
3969 		return (SET_ERROR(EINVAL));
3970 	}
3971 
3972 	/*
3973 	 * Permissions aren't checked on Solaris because on this OS
3974 	 * zfs_space() can only be called with an opened file handle.
3975 	 * On Linux we can get here through truncate_range() which
3976 	 * operates directly on inodes, so we need to check access rights.
3977 	 */
3978 	if ((error = zfs_zaccess(zp, ACE_WRITE_DATA, 0, B_FALSE, cr, NULL))) {
3979 		zfs_exit(zfsvfs, FTAG);
3980 		return (error);
3981 	}
3982 
3983 	off = bfp->l_start;
3984 	len = bfp->l_len; /* 0 means from off to end of file */
3985 
3986 	error = zfs_freesp(zp, off, len, flag, TRUE);
3987 
3988 	zfs_exit(zfsvfs, FTAG);
3989 	return (error);
3990 }
3991 
3992 static void
zfs_inactive(vnode_t * vp,cred_t * cr,caller_context_t * ct)3993 zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
3994 {
3995 	(void) cr, (void) ct;
3996 	znode_t	*zp = VTOZ(vp);
3997 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
3998 	int error;
3999 
4000 	ZFS_TEARDOWN_INACTIVE_ENTER_READ(zfsvfs);
4001 	if (zp->z_sa_hdl == NULL) {
4002 		/*
4003 		 * The fs has been unmounted, or we did a
4004 		 * suspend/resume and this file no longer exists.
4005 		 */
4006 		ZFS_TEARDOWN_INACTIVE_EXIT_READ(zfsvfs);
4007 		vrecycle(vp);
4008 		return;
4009 	}
4010 
4011 	if (zp->z_unlinked) {
4012 		/*
4013 		 * Fast path to recycle a vnode of a removed file.
4014 		 */
4015 		ZFS_TEARDOWN_INACTIVE_EXIT_READ(zfsvfs);
4016 		vrecycle(vp);
4017 		return;
4018 	}
4019 
4020 	if (zp->z_atime_dirty && zp->z_unlinked == 0) {
4021 		dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
4022 
4023 		dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4024 		zfs_sa_upgrade_txholds(tx, zp);
4025 		error = dmu_tx_assign(tx, DMU_TX_WAIT);
4026 		if (error) {
4027 			dmu_tx_abort(tx);
4028 		} else {
4029 			(void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs),
4030 			    (void *)&zp->z_atime, sizeof (zp->z_atime), tx);
4031 			zp->z_atime_dirty = 0;
4032 			dmu_tx_commit(tx);
4033 		}
4034 	}
4035 	ZFS_TEARDOWN_INACTIVE_EXIT_READ(zfsvfs);
4036 }
4037 
4038 
4039 _Static_assert(sizeof (struct zfid_short) <= sizeof (struct fid),
4040 	"struct zfid_short bigger than struct fid");
4041 _Static_assert(sizeof (struct zfid_long) <= sizeof (struct fid),
4042 	"struct zfid_long bigger than struct fid");
4043 
4044 static int
zfs_fid(vnode_t * vp,fid_t * fidp,caller_context_t * ct)4045 zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
4046 {
4047 	(void) ct;
4048 	znode_t		*zp = VTOZ(vp);
4049 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
4050 	uint32_t	gen;
4051 	uint64_t	gen64;
4052 	uint64_t	object = zp->z_id;
4053 	zfid_short_t	*zfid;
4054 	int		size, i, error;
4055 
4056 	if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0)
4057 		return (error);
4058 
4059 	if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs),
4060 	    &gen64, sizeof (uint64_t))) != 0) {
4061 		zfs_exit(zfsvfs, FTAG);
4062 		return (error);
4063 	}
4064 
4065 	gen = (uint32_t)gen64;
4066 
4067 	size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN;
4068 	fidp->fid_len = size;
4069 
4070 	zfid = (zfid_short_t *)fidp;
4071 
4072 	zfid->zf_len = size;
4073 
4074 	for (i = 0; i < sizeof (zfid->zf_object); i++)
4075 		zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
4076 
4077 	/* Must have a non-zero generation number to distinguish from .zfs */
4078 	if (gen == 0)
4079 		gen = 1;
4080 	for (i = 0; i < sizeof (zfid->zf_gen); i++)
4081 		zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
4082 
4083 	if (size == LONG_FID_LEN) {
4084 		uint64_t	objsetid = dmu_objset_id(zfsvfs->z_os);
4085 		zfid_long_t	*zlfid;
4086 
4087 		zlfid = (zfid_long_t *)fidp;
4088 
4089 		for (i = 0; i < sizeof (zlfid->zf_setid); i++)
4090 			zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
4091 
4092 		/* XXX - this should be the generation number for the objset */
4093 		for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
4094 			zlfid->zf_setgen[i] = 0;
4095 	}
4096 
4097 	zfs_exit(zfsvfs, FTAG);
4098 	return (0);
4099 }
4100 
4101 static int
zfs_pathconf(vnode_t * vp,int cmd,ulong_t * valp,cred_t * cr,caller_context_t * ct)4102 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
4103     caller_context_t *ct)
4104 {
4105 	znode_t *zp;
4106 	zfsvfs_t *zfsvfs;
4107 	int error;
4108 
4109 	switch (cmd) {
4110 	case _PC_LINK_MAX:
4111 		*valp = MIN(LONG_MAX, ZFS_LINK_MAX);
4112 		return (0);
4113 
4114 	case _PC_FILESIZEBITS:
4115 		*valp = 64;
4116 		return (0);
4117 	case _PC_MIN_HOLE_SIZE:
4118 		*valp = (int)SPA_MINBLOCKSIZE;
4119 		return (0);
4120 	case _PC_ACL_EXTENDED:
4121 #if 0		/* POSIX ACLs are not implemented for ZFS on FreeBSD yet. */
4122 		zp = VTOZ(vp);
4123 		zfsvfs = zp->z_zfsvfs;
4124 		if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0)
4125 			return (error);
4126 		*valp = zfsvfs->z_acl_type == ZFSACLTYPE_POSIX ? 1 : 0;
4127 		zfs_exit(zfsvfs, FTAG);
4128 #else
4129 		*valp = 0;
4130 #endif
4131 		return (0);
4132 
4133 	case _PC_ACL_NFS4:
4134 		zp = VTOZ(vp);
4135 		zfsvfs = zp->z_zfsvfs;
4136 		if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0)
4137 			return (error);
4138 		*valp = zfsvfs->z_acl_type == ZFS_ACLTYPE_NFSV4 ? 1 : 0;
4139 		zfs_exit(zfsvfs, FTAG);
4140 		return (0);
4141 
4142 	case _PC_ACL_PATH_MAX:
4143 		*valp = ACL_MAX_ENTRIES;
4144 		return (0);
4145 
4146 	default:
4147 		return (EOPNOTSUPP);
4148 	}
4149 }
4150 
4151 static int
zfs_getpages(struct vnode * vp,vm_page_t * ma,int count,int * rbehind,int * rahead)4152 zfs_getpages(struct vnode *vp, vm_page_t *ma, int count, int *rbehind,
4153     int *rahead)
4154 {
4155 	znode_t *zp = VTOZ(vp);
4156 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4157 	zfs_locked_range_t *lr;
4158 	vm_object_t object;
4159 	off_t start, end, obj_size;
4160 	uint_t blksz;
4161 	int pgsin_b, pgsin_a;
4162 	int error;
4163 
4164 	if (zfs_enter_verify_zp(zfsvfs, zp, FTAG) != 0)
4165 		return (zfs_vm_pagerret_error);
4166 
4167 	object = ma[0]->object;
4168 	start = IDX_TO_OFF(ma[0]->pindex);
4169 	end = IDX_TO_OFF(ma[count - 1]->pindex + 1);
4170 
4171 	/*
4172 	 * Lock a range covering all required and optional pages.
4173 	 * Note that we need to handle the case of the block size growing.
4174 	 */
4175 	for (;;) {
4176 		uint64_t len;
4177 
4178 		blksz = zp->z_blksz;
4179 		len = roundup(end, blksz) - rounddown(start, blksz);
4180 
4181 		lr = zfs_rangelock_tryenter(&zp->z_rangelock,
4182 		    rounddown(start, blksz), len, RL_READER);
4183 		if (lr == NULL) {
4184 			/*
4185 			 * Avoid a deadlock with update_pages().  We need to
4186 			 * hold the range lock when copying from the DMU, so
4187 			 * give up the busy lock to allow update_pages() to
4188 			 * proceed.  We might need to allocate new pages, which
4189 			 * isn't quite right since this allocation isn't subject
4190 			 * to the page fault handler's OOM logic, but this is
4191 			 * the best we can do for now.
4192 			 */
4193 			for (int i = 0; i < count; i++)
4194 				vm_page_xunbusy(ma[i]);
4195 
4196 			lr = zfs_rangelock_enter(&zp->z_rangelock,
4197 			    rounddown(start, blksz), len, RL_READER);
4198 
4199 			zfs_vmobject_wlock(object);
4200 			(void) vm_page_grab_pages(object, OFF_TO_IDX(start),
4201 			    VM_ALLOC_NORMAL | VM_ALLOC_WAITOK | VM_ALLOC_ZERO,
4202 			    ma, count);
4203 			zfs_vmobject_wunlock(object);
4204 		}
4205 		if (blksz == zp->z_blksz)
4206 			break;
4207 		zfs_rangelock_exit(lr);
4208 	}
4209 
4210 	zfs_vmobject_wlock(object);
4211 	obj_size = object->un_pager.vnp.vnp_size;
4212 	zfs_vmobject_wunlock(object);
4213 	if (IDX_TO_OFF(ma[count - 1]->pindex) >= obj_size) {
4214 		zfs_rangelock_exit(lr);
4215 		zfs_exit(zfsvfs, FTAG);
4216 		return (zfs_vm_pagerret_bad);
4217 	}
4218 
4219 	pgsin_b = 0;
4220 	if (rbehind != NULL) {
4221 		pgsin_b = OFF_TO_IDX(start - rounddown(start, blksz));
4222 		pgsin_b = MIN(*rbehind, pgsin_b);
4223 	}
4224 
4225 	pgsin_a = 0;
4226 	if (rahead != NULL) {
4227 		pgsin_a = OFF_TO_IDX(roundup(end, blksz) - end);
4228 		if (end + IDX_TO_OFF(pgsin_a) >= obj_size)
4229 			pgsin_a = OFF_TO_IDX(round_page(obj_size) - end);
4230 		pgsin_a = MIN(*rahead, pgsin_a);
4231 	}
4232 
4233 	/*
4234 	 * NB: we need to pass the exact byte size of the data that we expect
4235 	 * to read after accounting for the file size.  This is required because
4236 	 * ZFS will panic if we request DMU to read beyond the end of the last
4237 	 * allocated block.
4238 	 */
4239 	for (int i = 0; i < count; i++) {
4240 		int dummypgsin, count1, j, last_size;
4241 
4242 		if (vm_page_any_valid(ma[i])) {
4243 			ASSERT(vm_page_all_valid(ma[i]));
4244 			continue;
4245 		}
4246 		for (j = i + 1; j < count; j++) {
4247 			if (vm_page_any_valid(ma[j])) {
4248 				ASSERT(vm_page_all_valid(ma[j]));
4249 				break;
4250 			}
4251 		}
4252 		count1 = j - i;
4253 		dummypgsin = 0;
4254 		last_size = j == count ?
4255 		    MIN(end, obj_size) - (end - PAGE_SIZE) : PAGE_SIZE;
4256 		error = dmu_read_pages(zfsvfs->z_os, zp->z_id, &ma[i], count1,
4257 		    i == 0 ? &pgsin_b : &dummypgsin,
4258 		    j == count ? &pgsin_a : &dummypgsin,
4259 		    last_size);
4260 		if (error != 0)
4261 			break;
4262 		i += count1 - 1;
4263 	}
4264 
4265 	zfs_rangelock_exit(lr);
4266 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4267 
4268 	dataset_kstats_update_read_kstats(&zfsvfs->z_kstat, count*PAGE_SIZE);
4269 
4270 	zfs_exit(zfsvfs, FTAG);
4271 
4272 	if (error != 0)
4273 		return (zfs_vm_pagerret_error);
4274 
4275 	VM_CNT_INC(v_vnodein);
4276 	VM_CNT_ADD(v_vnodepgsin, count + pgsin_b + pgsin_a);
4277 	if (rbehind != NULL)
4278 		*rbehind = pgsin_b;
4279 	if (rahead != NULL)
4280 		*rahead = pgsin_a;
4281 	return (zfs_vm_pagerret_ok);
4282 }
4283 
4284 #ifndef _SYS_SYSPROTO_H_
4285 struct vop_getpages_args {
4286 	struct vnode *a_vp;
4287 	vm_page_t *a_m;
4288 	int a_count;
4289 	int *a_rbehind;
4290 	int *a_rahead;
4291 };
4292 #endif
4293 
4294 static int
zfs_freebsd_getpages(struct vop_getpages_args * ap)4295 zfs_freebsd_getpages(struct vop_getpages_args *ap)
4296 {
4297 
4298 	return (zfs_getpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_rbehind,
4299 	    ap->a_rahead));
4300 }
4301 
4302 static int
zfs_putpages(struct vnode * vp,vm_page_t * ma,size_t len,int flags,int * rtvals)4303 zfs_putpages(struct vnode *vp, vm_page_t *ma, size_t len, int flags,
4304     int *rtvals)
4305 {
4306 	znode_t		*zp = VTOZ(vp);
4307 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
4308 	zfs_locked_range_t		*lr;
4309 	dmu_tx_t	*tx;
4310 	struct sf_buf	*sf;
4311 	vm_object_t	object;
4312 	vm_page_t	m;
4313 	caddr_t		va;
4314 	size_t		tocopy;
4315 	size_t		lo_len;
4316 	vm_ooffset_t	lo_off;
4317 	vm_ooffset_t	off;
4318 	uint_t		blksz;
4319 	int		ncount;
4320 	int		pcount;
4321 	int		err;
4322 	int		i;
4323 
4324 	object = vp->v_object;
4325 	KASSERT(ma[0]->object == object, ("mismatching object"));
4326 	KASSERT(len > 0 && (len & PAGE_MASK) == 0, ("unexpected length"));
4327 
4328 	pcount = btoc(len);
4329 	ncount = pcount;
4330 	for (i = 0; i < pcount; i++)
4331 		rtvals[i] = zfs_vm_pagerret_error;
4332 
4333 	if (zfs_enter_verify_zp(zfsvfs, zp, FTAG) != 0)
4334 		return (zfs_vm_pagerret_error);
4335 
4336 	off = IDX_TO_OFF(ma[0]->pindex);
4337 	blksz = zp->z_blksz;
4338 	lo_off = rounddown(off, blksz);
4339 	lo_len = roundup(len + (off - lo_off), blksz);
4340 	lr = zfs_rangelock_enter(&zp->z_rangelock, lo_off, lo_len, RL_WRITER);
4341 
4342 	zfs_vmobject_wlock(object);
4343 	if (len + off > object->un_pager.vnp.vnp_size) {
4344 		if (object->un_pager.vnp.vnp_size > off) {
4345 			int pgoff;
4346 
4347 			len = object->un_pager.vnp.vnp_size - off;
4348 			ncount = btoc(len);
4349 			if ((pgoff = (int)len & PAGE_MASK) != 0) {
4350 				/*
4351 				 * If the object is locked and the following
4352 				 * conditions hold, then the page's dirty
4353 				 * field cannot be concurrently changed by a
4354 				 * pmap operation.
4355 				 */
4356 				m = ma[ncount - 1];
4357 				vm_page_assert_sbusied(m);
4358 				KASSERT(!pmap_page_is_write_mapped(m),
4359 				    ("zfs_putpages: page %p is not read-only",
4360 				    m));
4361 				vm_page_clear_dirty(m, pgoff, PAGE_SIZE -
4362 				    pgoff);
4363 			}
4364 		} else {
4365 			len = 0;
4366 			ncount = 0;
4367 		}
4368 		if (ncount < pcount) {
4369 			for (i = ncount; i < pcount; i++) {
4370 				rtvals[i] = zfs_vm_pagerret_bad;
4371 			}
4372 		}
4373 	}
4374 	zfs_vmobject_wunlock(object);
4375 
4376 	boolean_t commit = (flags & (zfs_vm_pagerput_sync |
4377 	    zfs_vm_pagerput_inval)) != 0 ||
4378 	    zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS;
4379 
4380 	if (ncount == 0)
4381 		goto out;
4382 
4383 	if (zfs_id_overblockquota(zfsvfs, DMU_USERUSED_OBJECT, zp->z_uid) ||
4384 	    zfs_id_overblockquota(zfsvfs, DMU_GROUPUSED_OBJECT, zp->z_gid) ||
4385 	    (zp->z_projid != ZFS_DEFAULT_PROJID &&
4386 	    zfs_id_overblockquota(zfsvfs, DMU_PROJECTUSED_OBJECT,
4387 	    zp->z_projid))) {
4388 		goto out;
4389 	}
4390 
4391 	tx = dmu_tx_create(zfsvfs->z_os);
4392 	dmu_tx_hold_write(tx, zp->z_id, off, len);
4393 
4394 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4395 	zfs_sa_upgrade_txholds(tx, zp);
4396 	err = dmu_tx_assign(tx, DMU_TX_WAIT);
4397 	if (err != 0) {
4398 		dmu_tx_abort(tx);
4399 		goto out;
4400 	}
4401 
4402 	if (zp->z_blksz < PAGE_SIZE) {
4403 		for (i = 0; len > 0; off += tocopy, len -= tocopy, i++) {
4404 			tocopy = len > PAGE_SIZE ? PAGE_SIZE : len;
4405 			va = zfs_map_page(ma[i], &sf);
4406 			dmu_write(zfsvfs->z_os, zp->z_id, off, tocopy, va, tx);
4407 			zfs_unmap_page(sf);
4408 		}
4409 	} else {
4410 		err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, ma, tx);
4411 	}
4412 
4413 	if (err == 0) {
4414 		uint64_t mtime[2], ctime[2];
4415 		sa_bulk_attr_t bulk[3];
4416 		int count = 0;
4417 
4418 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
4419 		    &mtime, 16);
4420 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
4421 		    &ctime, 16);
4422 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
4423 		    &zp->z_pflags, 8);
4424 		zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime);
4425 		err = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
4426 		ASSERT0(err);
4427 		/*
4428 		 * XXX we should be passing a callback to undirty
4429 		 * but that would make the locking messier
4430 		 */
4431 		zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off,
4432 		    len, commit, B_FALSE, NULL, NULL);
4433 
4434 		zfs_vmobject_wlock(object);
4435 		for (i = 0; i < ncount; i++) {
4436 			rtvals[i] = zfs_vm_pagerret_ok;
4437 			vm_page_undirty(ma[i]);
4438 		}
4439 		zfs_vmobject_wunlock(object);
4440 		VM_CNT_INC(v_vnodeout);
4441 		VM_CNT_ADD(v_vnodepgsout, ncount);
4442 	}
4443 	dmu_tx_commit(tx);
4444 
4445 out:
4446 	zfs_rangelock_exit(lr);
4447 	if (commit)
4448 		zil_commit(zfsvfs->z_log, zp->z_id);
4449 
4450 	dataset_kstats_update_write_kstats(&zfsvfs->z_kstat, len);
4451 
4452 	zfs_exit(zfsvfs, FTAG);
4453 	return (rtvals[0]);
4454 }
4455 
4456 #ifndef _SYS_SYSPROTO_H_
4457 struct vop_putpages_args {
4458 	struct vnode *a_vp;
4459 	vm_page_t *a_m;
4460 	int a_count;
4461 	int a_sync;
4462 	int *a_rtvals;
4463 };
4464 #endif
4465 
4466 static int
zfs_freebsd_putpages(struct vop_putpages_args * ap)4467 zfs_freebsd_putpages(struct vop_putpages_args *ap)
4468 {
4469 
4470 	return (zfs_putpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_sync,
4471 	    ap->a_rtvals));
4472 }
4473 
4474 #ifndef _SYS_SYSPROTO_H_
4475 struct vop_bmap_args {
4476 	struct vnode *a_vp;
4477 	daddr_t  a_bn;
4478 	struct bufobj **a_bop;
4479 	daddr_t *a_bnp;
4480 	int *a_runp;
4481 	int *a_runb;
4482 };
4483 #endif
4484 
4485 static int
zfs_freebsd_bmap(struct vop_bmap_args * ap)4486 zfs_freebsd_bmap(struct vop_bmap_args *ap)
4487 {
4488 
4489 	if (ap->a_bop != NULL)
4490 		*ap->a_bop = &ap->a_vp->v_bufobj;
4491 	if (ap->a_bnp != NULL)
4492 		*ap->a_bnp = ap->a_bn;
4493 	if (ap->a_runp != NULL)
4494 		*ap->a_runp = 0;
4495 	if (ap->a_runb != NULL)
4496 		*ap->a_runb = 0;
4497 
4498 	return (0);
4499 }
4500 
4501 #ifndef _SYS_SYSPROTO_H_
4502 struct vop_open_args {
4503 	struct vnode *a_vp;
4504 	int a_mode;
4505 	struct ucred *a_cred;
4506 	struct thread *a_td;
4507 };
4508 #endif
4509 
4510 static int
zfs_freebsd_open(struct vop_open_args * ap)4511 zfs_freebsd_open(struct vop_open_args *ap)
4512 {
4513 	vnode_t	*vp = ap->a_vp;
4514 	znode_t *zp = VTOZ(vp);
4515 	int error;
4516 
4517 	error = zfs_open(&vp, ap->a_mode, ap->a_cred);
4518 	if (error == 0)
4519 		vnode_create_vobject(vp, zp->z_size, ap->a_td);
4520 	return (error);
4521 }
4522 
4523 #ifndef _SYS_SYSPROTO_H_
4524 struct vop_close_args {
4525 	struct vnode *a_vp;
4526 	int  a_fflag;
4527 	struct ucred *a_cred;
4528 	struct thread *a_td;
4529 };
4530 #endif
4531 
4532 static int
zfs_freebsd_close(struct vop_close_args * ap)4533 zfs_freebsd_close(struct vop_close_args *ap)
4534 {
4535 
4536 	return (zfs_close(ap->a_vp, ap->a_fflag, 1, 0, ap->a_cred));
4537 }
4538 
4539 #ifndef _SYS_SYSPROTO_H_
4540 struct vop_ioctl_args {
4541 	struct vnode *a_vp;
4542 	ulong_t a_command;
4543 	caddr_t a_data;
4544 	int a_fflag;
4545 	struct ucred *cred;
4546 	struct thread *td;
4547 };
4548 #endif
4549 
4550 static int
zfs_freebsd_ioctl(struct vop_ioctl_args * ap)4551 zfs_freebsd_ioctl(struct vop_ioctl_args *ap)
4552 {
4553 
4554 	return (zfs_ioctl(ap->a_vp, ap->a_command, (intptr_t)ap->a_data,
4555 	    ap->a_fflag, ap->a_cred, NULL));
4556 }
4557 
4558 static int
ioflags(int ioflags)4559 ioflags(int ioflags)
4560 {
4561 	int flags = 0;
4562 
4563 	if (ioflags & IO_APPEND)
4564 		flags |= O_APPEND;
4565 	if (ioflags & IO_NDELAY)
4566 		flags |= O_NONBLOCK;
4567 	if (ioflags & IO_DIRECT)
4568 		flags |= O_DIRECT;
4569 	if (ioflags & IO_SYNC)
4570 		flags |= O_SYNC;
4571 
4572 	return (flags);
4573 }
4574 
4575 #ifndef _SYS_SYSPROTO_H_
4576 struct vop_read_args {
4577 	struct vnode *a_vp;
4578 	struct uio *a_uio;
4579 	int a_ioflag;
4580 	struct ucred *a_cred;
4581 };
4582 #endif
4583 
4584 static int
zfs_freebsd_read(struct vop_read_args * ap)4585 zfs_freebsd_read(struct vop_read_args *ap)
4586 {
4587 	zfs_uio_t uio;
4588 	int error = 0;
4589 	zfs_uio_init(&uio, ap->a_uio);
4590 	error = zfs_read(VTOZ(ap->a_vp), &uio, ioflags(ap->a_ioflag),
4591 	    ap->a_cred);
4592 	/*
4593 	 * XXX We occasionally get an EFAULT for Direct I/O reads on
4594 	 * FreeBSD 13. This still needs to be resolved. The EFAULT comes
4595 	 * from:
4596 	 * zfs_uio_get__dio_pages_alloc() ->
4597 	 * zfs_uio_get_dio_pages_impl() ->
4598 	 * zfs_uio_iov_step() ->
4599 	 * zfs_uio_get_user_pages().
4600 	 * We return EFAULT from zfs_uio_iov_step(). When a Direct I/O
4601 	 * read fails to map in the user pages (returning EFAULT) the
4602 	 * Direct I/O request is broken up into two separate IO requests
4603 	 * and issued separately using Direct I/O.
4604 	 */
4605 #ifdef ZFS_DEBUG
4606 	if (error == EFAULT && uio.uio_extflg & UIO_DIRECT) {
4607 #if 0
4608 		printf("%s(%d): Direct I/O read returning EFAULT "
4609 		    "uio = %p, zfs_uio_offset(uio) = %lu "
4610 		    "zfs_uio_resid(uio) = %lu\n",
4611 		    __FUNCTION__, __LINE__, &uio, zfs_uio_offset(&uio),
4612 		    zfs_uio_resid(&uio));
4613 #endif
4614 	}
4615 
4616 #endif
4617 	return (error);
4618 }
4619 
4620 #ifndef _SYS_SYSPROTO_H_
4621 struct vop_write_args {
4622 	struct vnode *a_vp;
4623 	struct uio *a_uio;
4624 	int a_ioflag;
4625 	struct ucred *a_cred;
4626 };
4627 #endif
4628 
4629 static int
zfs_freebsd_write(struct vop_write_args * ap)4630 zfs_freebsd_write(struct vop_write_args *ap)
4631 {
4632 	zfs_uio_t uio;
4633 	zfs_uio_init(&uio, ap->a_uio);
4634 	return (zfs_write(VTOZ(ap->a_vp), &uio, ioflags(ap->a_ioflag),
4635 	    ap->a_cred));
4636 }
4637 
4638 /*
4639  * VOP_FPLOOKUP_VEXEC routines are subject to special circumstances, see
4640  * the comment above cache_fplookup for details.
4641  */
4642 static int
zfs_freebsd_fplookup_vexec(struct vop_fplookup_vexec_args * v)4643 zfs_freebsd_fplookup_vexec(struct vop_fplookup_vexec_args *v)
4644 {
4645 	vnode_t *vp;
4646 	znode_t *zp;
4647 	uint64_t pflags;
4648 
4649 	vp = v->a_vp;
4650 	zp = VTOZ_SMR(vp);
4651 	if (__predict_false(zp == NULL))
4652 		return (EAGAIN);
4653 	pflags = atomic_load_64(&zp->z_pflags);
4654 	if (pflags & ZFS_AV_QUARANTINED)
4655 		return (EAGAIN);
4656 	if (pflags & ZFS_XATTR)
4657 		return (EAGAIN);
4658 	if ((pflags & ZFS_NO_EXECS_DENIED) == 0)
4659 		return (EAGAIN);
4660 	return (0);
4661 }
4662 
4663 static int
zfs_freebsd_fplookup_symlink(struct vop_fplookup_symlink_args * v)4664 zfs_freebsd_fplookup_symlink(struct vop_fplookup_symlink_args *v)
4665 {
4666 	vnode_t *vp;
4667 	znode_t *zp;
4668 	char *target;
4669 
4670 	vp = v->a_vp;
4671 	zp = VTOZ_SMR(vp);
4672 	if (__predict_false(zp == NULL)) {
4673 		return (EAGAIN);
4674 	}
4675 
4676 	target = atomic_load_consume_ptr(&zp->z_cached_symlink);
4677 	if (target == NULL) {
4678 		return (EAGAIN);
4679 	}
4680 	return (cache_symlink_resolve(v->a_fpl, target, strlen(target)));
4681 }
4682 
4683 #ifndef _SYS_SYSPROTO_H_
4684 struct vop_access_args {
4685 	struct vnode *a_vp;
4686 	accmode_t a_accmode;
4687 	struct ucred *a_cred;
4688 	struct thread *a_td;
4689 };
4690 #endif
4691 
4692 static int
zfs_freebsd_access(struct vop_access_args * ap)4693 zfs_freebsd_access(struct vop_access_args *ap)
4694 {
4695 	vnode_t *vp = ap->a_vp;
4696 	znode_t *zp = VTOZ(vp);
4697 	accmode_t accmode;
4698 	int error = 0;
4699 
4700 
4701 	if (ap->a_accmode == VEXEC) {
4702 		if (zfs_fastaccesschk_execute(zp, ap->a_cred) == 0)
4703 			return (0);
4704 	}
4705 
4706 	/*
4707 	 * ZFS itself only knowns about VREAD, VWRITE, VEXEC and VAPPEND,
4708 	 */
4709 	accmode = ap->a_accmode & (VREAD|VWRITE|VEXEC|VAPPEND);
4710 	if (accmode != 0)
4711 		error = zfs_access(zp, accmode, 0, ap->a_cred);
4712 
4713 	/*
4714 	 * VADMIN has to be handled by vaccess().
4715 	 */
4716 	if (error == 0) {
4717 		accmode = ap->a_accmode & ~(VREAD|VWRITE|VEXEC|VAPPEND);
4718 		if (accmode != 0) {
4719 			error = vaccess(vp->v_type, zp->z_mode, zp->z_uid,
4720 			    zp->z_gid, accmode, ap->a_cred);
4721 		}
4722 	}
4723 
4724 	/*
4725 	 * For VEXEC, ensure that at least one execute bit is set for
4726 	 * non-directories.
4727 	 */
4728 	if (error == 0 && (ap->a_accmode & VEXEC) != 0 && vp->v_type != VDIR &&
4729 	    (zp->z_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0) {
4730 		error = EACCES;
4731 	}
4732 
4733 	return (error);
4734 }
4735 
4736 #ifndef _SYS_SYSPROTO_H_
4737 struct vop_lookup_args {
4738 	struct vnode *a_dvp;
4739 	struct vnode **a_vpp;
4740 	struct componentname *a_cnp;
4741 };
4742 #endif
4743 
4744 static int
zfs_freebsd_lookup(struct vop_lookup_args * ap,boolean_t cached)4745 zfs_freebsd_lookup(struct vop_lookup_args *ap, boolean_t cached)
4746 {
4747 	struct componentname *cnp = ap->a_cnp;
4748 	char nm[NAME_MAX + 1];
4749 
4750 	ASSERT3U(cnp->cn_namelen, <, sizeof (nm));
4751 	strlcpy(nm, cnp->cn_nameptr, MIN(cnp->cn_namelen + 1, sizeof (nm)));
4752 
4753 	return (zfs_lookup(ap->a_dvp, nm, ap->a_vpp, cnp, cnp->cn_nameiop,
4754 	    cnp->cn_cred, 0, cached));
4755 }
4756 
4757 static int
zfs_freebsd_cachedlookup(struct vop_cachedlookup_args * ap)4758 zfs_freebsd_cachedlookup(struct vop_cachedlookup_args *ap)
4759 {
4760 
4761 	return (zfs_freebsd_lookup((struct vop_lookup_args *)ap, B_TRUE));
4762 }
4763 
4764 #ifndef _SYS_SYSPROTO_H_
4765 struct vop_lookup_args {
4766 	struct vnode *a_dvp;
4767 	struct vnode **a_vpp;
4768 	struct componentname *a_cnp;
4769 };
4770 #endif
4771 
4772 static int
zfs_cache_lookup(struct vop_lookup_args * ap)4773 zfs_cache_lookup(struct vop_lookup_args *ap)
4774 {
4775 	zfsvfs_t *zfsvfs;
4776 
4777 	zfsvfs = ap->a_dvp->v_mount->mnt_data;
4778 	if (zfsvfs->z_use_namecache)
4779 		return (vfs_cache_lookup(ap));
4780 	else
4781 		return (zfs_freebsd_lookup(ap, B_FALSE));
4782 }
4783 
4784 #ifndef _SYS_SYSPROTO_H_
4785 struct vop_create_args {
4786 	struct vnode *a_dvp;
4787 	struct vnode **a_vpp;
4788 	struct componentname *a_cnp;
4789 	struct vattr *a_vap;
4790 };
4791 #endif
4792 
4793 static int
zfs_freebsd_create(struct vop_create_args * ap)4794 zfs_freebsd_create(struct vop_create_args *ap)
4795 {
4796 	zfsvfs_t *zfsvfs;
4797 	struct componentname *cnp = ap->a_cnp;
4798 	vattr_t *vap = ap->a_vap;
4799 	znode_t *zp = NULL;
4800 	int rc, mode;
4801 
4802 #if __FreeBSD_version < 1400068
4803 	ASSERT(cnp->cn_flags & SAVENAME);
4804 #endif
4805 
4806 	vattr_init_mask(vap);
4807 	mode = vap->va_mode & ALLPERMS;
4808 	zfsvfs = ap->a_dvp->v_mount->mnt_data;
4809 	*ap->a_vpp = NULL;
4810 
4811 	rc = zfs_create(VTOZ(ap->a_dvp), cnp->cn_nameptr, vap, 0, mode,
4812 	    &zp, cnp->cn_cred, 0 /* flag */, NULL /* vsecattr */, NULL);
4813 	if (rc == 0)
4814 		*ap->a_vpp = ZTOV(zp);
4815 	if (zfsvfs->z_use_namecache &&
4816 	    rc == 0 && (cnp->cn_flags & MAKEENTRY) != 0)
4817 		cache_enter(ap->a_dvp, *ap->a_vpp, cnp);
4818 
4819 	return (rc);
4820 }
4821 
4822 #ifndef _SYS_SYSPROTO_H_
4823 struct vop_remove_args {
4824 	struct vnode *a_dvp;
4825 	struct vnode *a_vp;
4826 	struct componentname *a_cnp;
4827 };
4828 #endif
4829 
4830 static int
zfs_freebsd_remove(struct vop_remove_args * ap)4831 zfs_freebsd_remove(struct vop_remove_args *ap)
4832 {
4833 
4834 #if __FreeBSD_version < 1400068
4835 	ASSERT(ap->a_cnp->cn_flags & SAVENAME);
4836 #endif
4837 
4838 	return (zfs_remove_(ap->a_dvp, ap->a_vp, ap->a_cnp->cn_nameptr,
4839 	    ap->a_cnp->cn_cred));
4840 }
4841 
4842 #ifndef _SYS_SYSPROTO_H_
4843 struct vop_mkdir_args {
4844 	struct vnode *a_dvp;
4845 	struct vnode **a_vpp;
4846 	struct componentname *a_cnp;
4847 	struct vattr *a_vap;
4848 };
4849 #endif
4850 
4851 static int
zfs_freebsd_mkdir(struct vop_mkdir_args * ap)4852 zfs_freebsd_mkdir(struct vop_mkdir_args *ap)
4853 {
4854 	vattr_t *vap = ap->a_vap;
4855 	znode_t *zp = NULL;
4856 	int rc;
4857 
4858 #if __FreeBSD_version < 1400068
4859 	ASSERT(ap->a_cnp->cn_flags & SAVENAME);
4860 #endif
4861 
4862 	vattr_init_mask(vap);
4863 	*ap->a_vpp = NULL;
4864 
4865 	rc = zfs_mkdir(VTOZ(ap->a_dvp), ap->a_cnp->cn_nameptr, vap, &zp,
4866 	    ap->a_cnp->cn_cred, 0, NULL, NULL);
4867 
4868 	if (rc == 0)
4869 		*ap->a_vpp = ZTOV(zp);
4870 	return (rc);
4871 }
4872 
4873 #ifndef _SYS_SYSPROTO_H_
4874 struct vop_rmdir_args {
4875 	struct vnode *a_dvp;
4876 	struct vnode *a_vp;
4877 	struct componentname *a_cnp;
4878 };
4879 #endif
4880 
4881 static int
zfs_freebsd_rmdir(struct vop_rmdir_args * ap)4882 zfs_freebsd_rmdir(struct vop_rmdir_args *ap)
4883 {
4884 	struct componentname *cnp = ap->a_cnp;
4885 
4886 #if __FreeBSD_version < 1400068
4887 	ASSERT(cnp->cn_flags & SAVENAME);
4888 #endif
4889 
4890 	return (zfs_rmdir_(ap->a_dvp, ap->a_vp, cnp->cn_nameptr, cnp->cn_cred));
4891 }
4892 
4893 #ifndef _SYS_SYSPROTO_H_
4894 struct vop_readdir_args {
4895 	struct vnode *a_vp;
4896 	struct uio *a_uio;
4897 	struct ucred *a_cred;
4898 	int *a_eofflag;
4899 	int *a_ncookies;
4900 	cookie_t **a_cookies;
4901 };
4902 #endif
4903 
4904 static int
zfs_freebsd_readdir(struct vop_readdir_args * ap)4905 zfs_freebsd_readdir(struct vop_readdir_args *ap)
4906 {
4907 	zfs_uio_t uio;
4908 	zfs_uio_init(&uio, ap->a_uio);
4909 	return (zfs_readdir(ap->a_vp, &uio, ap->a_cred, ap->a_eofflag,
4910 	    ap->a_ncookies, ap->a_cookies));
4911 }
4912 
4913 #ifndef _SYS_SYSPROTO_H_
4914 struct vop_fsync_args {
4915 	struct vnode *a_vp;
4916 	int a_waitfor;
4917 	struct thread *a_td;
4918 };
4919 #endif
4920 
4921 static int
zfs_freebsd_fsync(struct vop_fsync_args * ap)4922 zfs_freebsd_fsync(struct vop_fsync_args *ap)
4923 {
4924 
4925 	return (zfs_fsync(VTOZ(ap->a_vp), 0, ap->a_td->td_ucred));
4926 }
4927 
4928 #ifndef _SYS_SYSPROTO_H_
4929 struct vop_getattr_args {
4930 	struct vnode *a_vp;
4931 	struct vattr *a_vap;
4932 	struct ucred *a_cred;
4933 };
4934 #endif
4935 
4936 static int
zfs_freebsd_getattr(struct vop_getattr_args * ap)4937 zfs_freebsd_getattr(struct vop_getattr_args *ap)
4938 {
4939 	vattr_t *vap = ap->a_vap;
4940 	xvattr_t xvap;
4941 	ulong_t fflags = 0;
4942 	int error;
4943 
4944 	xva_init(&xvap);
4945 	xvap.xva_vattr = *vap;
4946 	xvap.xva_vattr.va_mask |= AT_XVATTR;
4947 
4948 	/* Convert chflags into ZFS-type flags. */
4949 	/* XXX: what about SF_SETTABLE?. */
4950 	XVA_SET_REQ(&xvap, XAT_IMMUTABLE);
4951 	XVA_SET_REQ(&xvap, XAT_APPENDONLY);
4952 	XVA_SET_REQ(&xvap, XAT_NOUNLINK);
4953 	XVA_SET_REQ(&xvap, XAT_NODUMP);
4954 	XVA_SET_REQ(&xvap, XAT_READONLY);
4955 	XVA_SET_REQ(&xvap, XAT_ARCHIVE);
4956 	XVA_SET_REQ(&xvap, XAT_SYSTEM);
4957 	XVA_SET_REQ(&xvap, XAT_HIDDEN);
4958 	XVA_SET_REQ(&xvap, XAT_REPARSE);
4959 	XVA_SET_REQ(&xvap, XAT_OFFLINE);
4960 	XVA_SET_REQ(&xvap, XAT_SPARSE);
4961 
4962 	error = zfs_getattr(ap->a_vp, (vattr_t *)&xvap, 0, ap->a_cred);
4963 	if (error != 0)
4964 		return (error);
4965 
4966 	/* Convert ZFS xattr into chflags. */
4967 #define	FLAG_CHECK(fflag, xflag, xfield)	do {			\
4968 	if (XVA_ISSET_RTN(&xvap, (xflag)) && (xfield) != 0)		\
4969 		fflags |= (fflag);					\
4970 } while (0)
4971 	FLAG_CHECK(SF_IMMUTABLE, XAT_IMMUTABLE,
4972 	    xvap.xva_xoptattrs.xoa_immutable);
4973 	FLAG_CHECK(SF_APPEND, XAT_APPENDONLY,
4974 	    xvap.xva_xoptattrs.xoa_appendonly);
4975 	FLAG_CHECK(SF_NOUNLINK, XAT_NOUNLINK,
4976 	    xvap.xva_xoptattrs.xoa_nounlink);
4977 	FLAG_CHECK(UF_ARCHIVE, XAT_ARCHIVE,
4978 	    xvap.xva_xoptattrs.xoa_archive);
4979 	FLAG_CHECK(UF_NODUMP, XAT_NODUMP,
4980 	    xvap.xva_xoptattrs.xoa_nodump);
4981 	FLAG_CHECK(UF_READONLY, XAT_READONLY,
4982 	    xvap.xva_xoptattrs.xoa_readonly);
4983 	FLAG_CHECK(UF_SYSTEM, XAT_SYSTEM,
4984 	    xvap.xva_xoptattrs.xoa_system);
4985 	FLAG_CHECK(UF_HIDDEN, XAT_HIDDEN,
4986 	    xvap.xva_xoptattrs.xoa_hidden);
4987 	FLAG_CHECK(UF_REPARSE, XAT_REPARSE,
4988 	    xvap.xva_xoptattrs.xoa_reparse);
4989 	FLAG_CHECK(UF_OFFLINE, XAT_OFFLINE,
4990 	    xvap.xva_xoptattrs.xoa_offline);
4991 	FLAG_CHECK(UF_SPARSE, XAT_SPARSE,
4992 	    xvap.xva_xoptattrs.xoa_sparse);
4993 
4994 #undef	FLAG_CHECK
4995 	*vap = xvap.xva_vattr;
4996 	vap->va_flags = fflags;
4997 	return (0);
4998 }
4999 
5000 #ifndef _SYS_SYSPROTO_H_
5001 struct vop_setattr_args {
5002 	struct vnode *a_vp;
5003 	struct vattr *a_vap;
5004 	struct ucred *a_cred;
5005 };
5006 #endif
5007 
5008 static int
zfs_freebsd_setattr(struct vop_setattr_args * ap)5009 zfs_freebsd_setattr(struct vop_setattr_args *ap)
5010 {
5011 	vnode_t *vp = ap->a_vp;
5012 	vattr_t *vap = ap->a_vap;
5013 	cred_t *cred = ap->a_cred;
5014 	xvattr_t xvap;
5015 	ulong_t fflags;
5016 	uint64_t zflags;
5017 
5018 	vattr_init_mask(vap);
5019 	vap->va_mask &= ~AT_NOSET;
5020 
5021 	xva_init(&xvap);
5022 	xvap.xva_vattr = *vap;
5023 
5024 	zflags = VTOZ(vp)->z_pflags;
5025 
5026 	if (vap->va_flags != VNOVAL) {
5027 		zfsvfs_t *zfsvfs = VTOZ(vp)->z_zfsvfs;
5028 		int error;
5029 
5030 		if (zfsvfs->z_use_fuids == B_FALSE)
5031 			return (EOPNOTSUPP);
5032 
5033 		fflags = vap->va_flags;
5034 		/*
5035 		 * XXX KDM
5036 		 * We need to figure out whether it makes sense to allow
5037 		 * UF_REPARSE through, since we don't really have other
5038 		 * facilities to handle reparse points and zfs_setattr()
5039 		 * doesn't currently allow setting that attribute anyway.
5040 		 */
5041 		if ((fflags & ~(SF_IMMUTABLE|SF_APPEND|SF_NOUNLINK|UF_ARCHIVE|
5042 		    UF_NODUMP|UF_SYSTEM|UF_HIDDEN|UF_READONLY|UF_REPARSE|
5043 		    UF_OFFLINE|UF_SPARSE)) != 0)
5044 			return (EOPNOTSUPP);
5045 		/*
5046 		 * Unprivileged processes are not permitted to unset system
5047 		 * flags, or modify flags if any system flags are set.
5048 		 * Privileged non-jail processes may not modify system flags
5049 		 * if securelevel > 0 and any existing system flags are set.
5050 		 * Privileged jail processes behave like privileged non-jail
5051 		 * processes if the PR_ALLOW_CHFLAGS permission bit is set;
5052 		 * otherwise, they behave like unprivileged processes.
5053 		 */
5054 		if (secpolicy_fs_owner(vp->v_mount, cred) == 0 ||
5055 		    priv_check_cred(cred, PRIV_VFS_SYSFLAGS) == 0) {
5056 			if (zflags &
5057 			    (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) {
5058 				error = securelevel_gt(cred, 0);
5059 				if (error != 0)
5060 					return (error);
5061 			}
5062 		} else {
5063 			/*
5064 			 * Callers may only modify the file flags on
5065 			 * objects they have VADMIN rights for.
5066 			 */
5067 			if ((error = VOP_ACCESS(vp, VADMIN, cred,
5068 			    curthread)) != 0)
5069 				return (error);
5070 			if (zflags &
5071 			    (ZFS_IMMUTABLE | ZFS_APPENDONLY |
5072 			    ZFS_NOUNLINK)) {
5073 				return (EPERM);
5074 			}
5075 			if (fflags &
5076 			    (SF_IMMUTABLE | SF_APPEND | SF_NOUNLINK)) {
5077 				return (EPERM);
5078 			}
5079 		}
5080 
5081 #define	FLAG_CHANGE(fflag, zflag, xflag, xfield)	do {		\
5082 	if (((fflags & (fflag)) && !(zflags & (zflag))) ||		\
5083 	    ((zflags & (zflag)) && !(fflags & (fflag)))) {		\
5084 		XVA_SET_REQ(&xvap, (xflag));				\
5085 		(xfield) = ((fflags & (fflag)) != 0);			\
5086 	}								\
5087 } while (0)
5088 		/* Convert chflags into ZFS-type flags. */
5089 		/* XXX: what about SF_SETTABLE?. */
5090 		FLAG_CHANGE(SF_IMMUTABLE, ZFS_IMMUTABLE, XAT_IMMUTABLE,
5091 		    xvap.xva_xoptattrs.xoa_immutable);
5092 		FLAG_CHANGE(SF_APPEND, ZFS_APPENDONLY, XAT_APPENDONLY,
5093 		    xvap.xva_xoptattrs.xoa_appendonly);
5094 		FLAG_CHANGE(SF_NOUNLINK, ZFS_NOUNLINK, XAT_NOUNLINK,
5095 		    xvap.xva_xoptattrs.xoa_nounlink);
5096 		FLAG_CHANGE(UF_ARCHIVE, ZFS_ARCHIVE, XAT_ARCHIVE,
5097 		    xvap.xva_xoptattrs.xoa_archive);
5098 		FLAG_CHANGE(UF_NODUMP, ZFS_NODUMP, XAT_NODUMP,
5099 		    xvap.xva_xoptattrs.xoa_nodump);
5100 		FLAG_CHANGE(UF_READONLY, ZFS_READONLY, XAT_READONLY,
5101 		    xvap.xva_xoptattrs.xoa_readonly);
5102 		FLAG_CHANGE(UF_SYSTEM, ZFS_SYSTEM, XAT_SYSTEM,
5103 		    xvap.xva_xoptattrs.xoa_system);
5104 		FLAG_CHANGE(UF_HIDDEN, ZFS_HIDDEN, XAT_HIDDEN,
5105 		    xvap.xva_xoptattrs.xoa_hidden);
5106 		FLAG_CHANGE(UF_REPARSE, ZFS_REPARSE, XAT_REPARSE,
5107 		    xvap.xva_xoptattrs.xoa_reparse);
5108 		FLAG_CHANGE(UF_OFFLINE, ZFS_OFFLINE, XAT_OFFLINE,
5109 		    xvap.xva_xoptattrs.xoa_offline);
5110 		FLAG_CHANGE(UF_SPARSE, ZFS_SPARSE, XAT_SPARSE,
5111 		    xvap.xva_xoptattrs.xoa_sparse);
5112 #undef	FLAG_CHANGE
5113 	}
5114 	if (vap->va_birthtime.tv_sec != VNOVAL) {
5115 		xvap.xva_vattr.va_mask |= AT_XVATTR;
5116 		XVA_SET_REQ(&xvap, XAT_CREATETIME);
5117 	}
5118 	return (zfs_setattr(VTOZ(vp), (vattr_t *)&xvap, 0, cred, NULL));
5119 }
5120 
5121 #ifndef _SYS_SYSPROTO_H_
5122 struct vop_rename_args {
5123 	struct vnode *a_fdvp;
5124 	struct vnode *a_fvp;
5125 	struct componentname *a_fcnp;
5126 	struct vnode *a_tdvp;
5127 	struct vnode *a_tvp;
5128 	struct componentname *a_tcnp;
5129 };
5130 #endif
5131 
5132 static int
zfs_freebsd_rename(struct vop_rename_args * ap)5133 zfs_freebsd_rename(struct vop_rename_args *ap)
5134 {
5135 	vnode_t *fdvp = ap->a_fdvp;
5136 	vnode_t *fvp = ap->a_fvp;
5137 	vnode_t *tdvp = ap->a_tdvp;
5138 	vnode_t *tvp = ap->a_tvp;
5139 	int error;
5140 
5141 #if __FreeBSD_version < 1400068
5142 	ASSERT(ap->a_fcnp->cn_flags & (SAVENAME|SAVESTART));
5143 	ASSERT(ap->a_tcnp->cn_flags & (SAVENAME|SAVESTART));
5144 #endif
5145 
5146 	error = zfs_do_rename(fdvp, &fvp, ap->a_fcnp, tdvp, &tvp,
5147 	    ap->a_tcnp, ap->a_fcnp->cn_cred);
5148 
5149 	vrele(fdvp);
5150 	vrele(fvp);
5151 	vrele(tdvp);
5152 	if (tvp != NULL)
5153 		vrele(tvp);
5154 
5155 	return (error);
5156 }
5157 
5158 #ifndef _SYS_SYSPROTO_H_
5159 struct vop_symlink_args {
5160 	struct vnode *a_dvp;
5161 	struct vnode **a_vpp;
5162 	struct componentname *a_cnp;
5163 	struct vattr *a_vap;
5164 	char *a_target;
5165 };
5166 #endif
5167 
5168 static int
zfs_freebsd_symlink(struct vop_symlink_args * ap)5169 zfs_freebsd_symlink(struct vop_symlink_args *ap)
5170 {
5171 	struct componentname *cnp = ap->a_cnp;
5172 	vattr_t *vap = ap->a_vap;
5173 	znode_t *zp = NULL;
5174 	char *symlink;
5175 	size_t symlink_len;
5176 	int rc;
5177 
5178 #if __FreeBSD_version < 1400068
5179 	ASSERT(cnp->cn_flags & SAVENAME);
5180 #endif
5181 
5182 	vap->va_type = VLNK;	/* FreeBSD: Syscall only sets va_mode. */
5183 	vattr_init_mask(vap);
5184 	*ap->a_vpp = NULL;
5185 
5186 	rc = zfs_symlink(VTOZ(ap->a_dvp), cnp->cn_nameptr, vap,
5187 	    ap->a_target, &zp, cnp->cn_cred, 0 /* flags */, NULL);
5188 	if (rc == 0) {
5189 		*ap->a_vpp = ZTOV(zp);
5190 		ASSERT_VOP_ELOCKED(ZTOV(zp), __func__);
5191 		MPASS(zp->z_cached_symlink == NULL);
5192 		symlink_len = strlen(ap->a_target);
5193 		symlink = cache_symlink_alloc(symlink_len + 1, M_WAITOK);
5194 		if (symlink != NULL) {
5195 			memcpy(symlink, ap->a_target, symlink_len);
5196 			symlink[symlink_len] = '\0';
5197 			atomic_store_rel_ptr((uintptr_t *)&zp->z_cached_symlink,
5198 			    (uintptr_t)symlink);
5199 		}
5200 	}
5201 	return (rc);
5202 }
5203 
5204 #ifndef _SYS_SYSPROTO_H_
5205 struct vop_readlink_args {
5206 	struct vnode *a_vp;
5207 	struct uio *a_uio;
5208 	struct ucred *a_cred;
5209 };
5210 #endif
5211 
5212 static int
zfs_freebsd_readlink(struct vop_readlink_args * ap)5213 zfs_freebsd_readlink(struct vop_readlink_args *ap)
5214 {
5215 	zfs_uio_t uio;
5216 	int error;
5217 	znode_t	*zp = VTOZ(ap->a_vp);
5218 	char *symlink, *base;
5219 	size_t symlink_len;
5220 	bool trycache;
5221 
5222 	zfs_uio_init(&uio, ap->a_uio);
5223 	trycache = false;
5224 	if (zfs_uio_segflg(&uio) == UIO_SYSSPACE &&
5225 	    zfs_uio_iovcnt(&uio) == 1) {
5226 		base = zfs_uio_iovbase(&uio, 0);
5227 		symlink_len = zfs_uio_iovlen(&uio, 0);
5228 		trycache = true;
5229 	}
5230 	error = zfs_readlink(ap->a_vp, &uio, ap->a_cred, NULL);
5231 	if (atomic_load_ptr(&zp->z_cached_symlink) != NULL ||
5232 	    error != 0 || !trycache) {
5233 		return (error);
5234 	}
5235 	symlink_len -= zfs_uio_resid(&uio);
5236 	symlink = cache_symlink_alloc(symlink_len + 1, M_WAITOK);
5237 	if (symlink != NULL) {
5238 		memcpy(symlink, base, symlink_len);
5239 		symlink[symlink_len] = '\0';
5240 		if (!atomic_cmpset_rel_ptr((uintptr_t *)&zp->z_cached_symlink,
5241 		    (uintptr_t)NULL, (uintptr_t)symlink)) {
5242 			cache_symlink_free(symlink, symlink_len + 1);
5243 		}
5244 	}
5245 	return (error);
5246 }
5247 
5248 #ifndef _SYS_SYSPROTO_H_
5249 struct vop_link_args {
5250 	struct vnode *a_tdvp;
5251 	struct vnode *a_vp;
5252 	struct componentname *a_cnp;
5253 };
5254 #endif
5255 
5256 static int
zfs_freebsd_link(struct vop_link_args * ap)5257 zfs_freebsd_link(struct vop_link_args *ap)
5258 {
5259 	struct componentname *cnp = ap->a_cnp;
5260 	vnode_t *vp = ap->a_vp;
5261 	vnode_t *tdvp = ap->a_tdvp;
5262 
5263 	if (tdvp->v_mount != vp->v_mount)
5264 		return (EXDEV);
5265 
5266 #if __FreeBSD_version < 1400068
5267 	ASSERT(cnp->cn_flags & SAVENAME);
5268 #endif
5269 
5270 	return (zfs_link(VTOZ(tdvp), VTOZ(vp),
5271 	    cnp->cn_nameptr, cnp->cn_cred, 0));
5272 }
5273 
5274 #ifndef _SYS_SYSPROTO_H_
5275 struct vop_inactive_args {
5276 	struct vnode *a_vp;
5277 	struct thread *a_td;
5278 };
5279 #endif
5280 
5281 static int
zfs_freebsd_inactive(struct vop_inactive_args * ap)5282 zfs_freebsd_inactive(struct vop_inactive_args *ap)
5283 {
5284 	vnode_t *vp = ap->a_vp;
5285 
5286 	zfs_inactive(vp, curthread->td_ucred, NULL);
5287 	return (0);
5288 }
5289 
5290 #ifndef _SYS_SYSPROTO_H_
5291 struct vop_need_inactive_args {
5292 	struct vnode *a_vp;
5293 	struct thread *a_td;
5294 };
5295 #endif
5296 
5297 static int
zfs_freebsd_need_inactive(struct vop_need_inactive_args * ap)5298 zfs_freebsd_need_inactive(struct vop_need_inactive_args *ap)
5299 {
5300 	vnode_t *vp = ap->a_vp;
5301 	znode_t	*zp = VTOZ(vp);
5302 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5303 	int need;
5304 
5305 	if (vn_need_pageq_flush(vp))
5306 		return (1);
5307 
5308 	if (!ZFS_TEARDOWN_INACTIVE_TRY_ENTER_READ(zfsvfs))
5309 		return (1);
5310 	need = (zp->z_sa_hdl == NULL || zp->z_unlinked || zp->z_atime_dirty);
5311 	ZFS_TEARDOWN_INACTIVE_EXIT_READ(zfsvfs);
5312 
5313 	return (need);
5314 }
5315 
5316 #ifndef _SYS_SYSPROTO_H_
5317 struct vop_reclaim_args {
5318 	struct vnode *a_vp;
5319 	struct thread *a_td;
5320 };
5321 #endif
5322 
5323 static int
zfs_freebsd_reclaim(struct vop_reclaim_args * ap)5324 zfs_freebsd_reclaim(struct vop_reclaim_args *ap)
5325 {
5326 	vnode_t	*vp = ap->a_vp;
5327 	znode_t	*zp = VTOZ(vp);
5328 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5329 
5330 	ASSERT3P(zp, !=, NULL);
5331 
5332 	/*
5333 	 * z_teardown_inactive_lock protects from a race with
5334 	 * zfs_znode_dmu_fini in zfsvfs_teardown during
5335 	 * force unmount.
5336 	 */
5337 	ZFS_TEARDOWN_INACTIVE_ENTER_READ(zfsvfs);
5338 	if (zp->z_sa_hdl == NULL)
5339 		zfs_znode_free(zp);
5340 	else
5341 		zfs_zinactive(zp);
5342 	ZFS_TEARDOWN_INACTIVE_EXIT_READ(zfsvfs);
5343 
5344 	vp->v_data = NULL;
5345 	return (0);
5346 }
5347 
5348 #ifndef _SYS_SYSPROTO_H_
5349 struct vop_fid_args {
5350 	struct vnode *a_vp;
5351 	struct fid *a_fid;
5352 };
5353 #endif
5354 
5355 static int
zfs_freebsd_fid(struct vop_fid_args * ap)5356 zfs_freebsd_fid(struct vop_fid_args *ap)
5357 {
5358 
5359 	return (zfs_fid(ap->a_vp, (void *)ap->a_fid, NULL));
5360 }
5361 
5362 
5363 #ifndef _SYS_SYSPROTO_H_
5364 struct vop_pathconf_args {
5365 	struct vnode *a_vp;
5366 	int a_name;
5367 	register_t *a_retval;
5368 } *ap;
5369 #endif
5370 
5371 static int
zfs_freebsd_pathconf(struct vop_pathconf_args * ap)5372 zfs_freebsd_pathconf(struct vop_pathconf_args *ap)
5373 {
5374 	ulong_t val;
5375 	int error;
5376 
5377 	error = zfs_pathconf(ap->a_vp, ap->a_name, &val,
5378 	    curthread->td_ucred, NULL);
5379 	if (error == 0) {
5380 		*ap->a_retval = val;
5381 		return (error);
5382 	}
5383 	if (error != EOPNOTSUPP)
5384 		return (error);
5385 
5386 	switch (ap->a_name) {
5387 	case _PC_NAME_MAX:
5388 		*ap->a_retval = NAME_MAX;
5389 		return (0);
5390 #if __FreeBSD_version >= 1400032
5391 	case _PC_DEALLOC_PRESENT:
5392 		*ap->a_retval = 1;
5393 		return (0);
5394 #endif
5395 	case _PC_PIPE_BUF:
5396 		if (ap->a_vp->v_type == VDIR || ap->a_vp->v_type == VFIFO) {
5397 			*ap->a_retval = PIPE_BUF;
5398 			return (0);
5399 		}
5400 		return (EINVAL);
5401 	default:
5402 		return (vop_stdpathconf(ap));
5403 	}
5404 }
5405 
5406 static int zfs_xattr_compat = 1;
5407 
5408 static int
zfs_check_attrname(const char * name)5409 zfs_check_attrname(const char *name)
5410 {
5411 	/* We don't allow '/' character in attribute name. */
5412 	if (strchr(name, '/') != NULL)
5413 		return (SET_ERROR(EINVAL));
5414 	/* We don't allow attribute names that start with a namespace prefix. */
5415 	if (ZFS_XA_NS_PREFIX_FORBIDDEN(name))
5416 		return (SET_ERROR(EINVAL));
5417 	return (0);
5418 }
5419 
5420 /*
5421  * FreeBSD's extended attributes namespace defines file name prefix for ZFS'
5422  * extended attribute name:
5423  *
5424  *	NAMESPACE	XATTR_COMPAT	PREFIX
5425  *	system		*		freebsd:system:
5426  *	user		1		(none, can be used to access ZFS
5427  *					fsattr(5) attributes created on Solaris)
5428  *	user		0		user.
5429  */
5430 static int
zfs_create_attrname(int attrnamespace,const char * name,char * attrname,size_t size,boolean_t compat)5431 zfs_create_attrname(int attrnamespace, const char *name, char *attrname,
5432     size_t size, boolean_t compat)
5433 {
5434 	const char *namespace, *prefix, *suffix;
5435 
5436 	memset(attrname, 0, size);
5437 
5438 	switch (attrnamespace) {
5439 	case EXTATTR_NAMESPACE_USER:
5440 		if (compat) {
5441 			/*
5442 			 * This is the default namespace by which we can access
5443 			 * all attributes created on Solaris.
5444 			 */
5445 			prefix = namespace = suffix = "";
5446 		} else {
5447 			/*
5448 			 * This is compatible with the user namespace encoding
5449 			 * on Linux prior to xattr_compat, but nothing
5450 			 * else.
5451 			 */
5452 			prefix = "";
5453 			namespace = "user";
5454 			suffix = ".";
5455 		}
5456 		break;
5457 	case EXTATTR_NAMESPACE_SYSTEM:
5458 		prefix = "freebsd:";
5459 		namespace = EXTATTR_NAMESPACE_SYSTEM_STRING;
5460 		suffix = ":";
5461 		break;
5462 	case EXTATTR_NAMESPACE_EMPTY:
5463 	default:
5464 		return (SET_ERROR(EINVAL));
5465 	}
5466 	if (snprintf(attrname, size, "%s%s%s%s", prefix, namespace, suffix,
5467 	    name) >= size) {
5468 		return (SET_ERROR(ENAMETOOLONG));
5469 	}
5470 	return (0);
5471 }
5472 
5473 static int
zfs_ensure_xattr_cached(znode_t * zp)5474 zfs_ensure_xattr_cached(znode_t *zp)
5475 {
5476 	int error = 0;
5477 
5478 	ASSERT(RW_LOCK_HELD(&zp->z_xattr_lock));
5479 
5480 	if (zp->z_xattr_cached != NULL)
5481 		return (0);
5482 
5483 	if (rw_write_held(&zp->z_xattr_lock))
5484 		return (zfs_sa_get_xattr(zp));
5485 
5486 	if (!rw_tryupgrade(&zp->z_xattr_lock)) {
5487 		rw_exit(&zp->z_xattr_lock);
5488 		rw_enter(&zp->z_xattr_lock, RW_WRITER);
5489 	}
5490 	if (zp->z_xattr_cached == NULL)
5491 		error = zfs_sa_get_xattr(zp);
5492 	rw_downgrade(&zp->z_xattr_lock);
5493 	return (error);
5494 }
5495 
5496 #ifndef _SYS_SYSPROTO_H_
5497 struct vop_getextattr {
5498 	IN struct vnode *a_vp;
5499 	IN int a_attrnamespace;
5500 	IN const char *a_name;
5501 	INOUT struct uio *a_uio;
5502 	OUT size_t *a_size;
5503 	IN struct ucred *a_cred;
5504 	IN struct thread *a_td;
5505 };
5506 #endif
5507 
5508 static int
zfs_getextattr_dir(struct vop_getextattr_args * ap,const char * attrname)5509 zfs_getextattr_dir(struct vop_getextattr_args *ap, const char *attrname)
5510 {
5511 	struct thread *td = ap->a_td;
5512 	struct nameidata nd;
5513 	struct vattr va;
5514 	vnode_t *xvp = NULL, *vp;
5515 	int error, flags;
5516 
5517 	error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred,
5518 	    LOOKUP_XATTR, B_FALSE);
5519 	if (error != 0)
5520 		return (error);
5521 
5522 	flags = FREAD;
5523 #if __FreeBSD_version < 1400043
5524 	NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname,
5525 	    xvp, td);
5526 #else
5527 	NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname, xvp);
5528 #endif
5529 	error = vn_open_cred(&nd, &flags, 0, VN_OPEN_INVFS, ap->a_cred, NULL);
5530 	if (error != 0)
5531 		return (SET_ERROR(error));
5532 	vp = nd.ni_vp;
5533 	NDFREE_PNBUF(&nd);
5534 
5535 	if (ap->a_size != NULL) {
5536 		error = VOP_GETATTR(vp, &va, ap->a_cred);
5537 		if (error == 0)
5538 			*ap->a_size = (size_t)va.va_size;
5539 	} else if (ap->a_uio != NULL)
5540 		error = VOP_READ(vp, ap->a_uio, IO_UNIT, ap->a_cred);
5541 
5542 	VOP_UNLOCK(vp);
5543 	vn_close(vp, flags, ap->a_cred, td);
5544 	return (error);
5545 }
5546 
5547 static int
zfs_getextattr_sa(struct vop_getextattr_args * ap,const char * attrname)5548 zfs_getextattr_sa(struct vop_getextattr_args *ap, const char *attrname)
5549 {
5550 	znode_t *zp = VTOZ(ap->a_vp);
5551 	uchar_t *nv_value;
5552 	uint_t nv_size;
5553 	int error;
5554 
5555 	error = zfs_ensure_xattr_cached(zp);
5556 	if (error != 0)
5557 		return (error);
5558 
5559 	ASSERT(RW_LOCK_HELD(&zp->z_xattr_lock));
5560 	ASSERT3P(zp->z_xattr_cached, !=, NULL);
5561 
5562 	error = nvlist_lookup_byte_array(zp->z_xattr_cached, attrname,
5563 	    &nv_value, &nv_size);
5564 	if (error != 0)
5565 		return (SET_ERROR(error));
5566 
5567 	if (ap->a_size != NULL)
5568 		*ap->a_size = nv_size;
5569 	else if (ap->a_uio != NULL)
5570 		error = uiomove(nv_value, nv_size, ap->a_uio);
5571 	if (error != 0)
5572 		return (SET_ERROR(error));
5573 
5574 	return (0);
5575 }
5576 
5577 static int
zfs_getextattr_impl(struct vop_getextattr_args * ap,boolean_t compat)5578 zfs_getextattr_impl(struct vop_getextattr_args *ap, boolean_t compat)
5579 {
5580 	znode_t *zp = VTOZ(ap->a_vp);
5581 	zfsvfs_t *zfsvfs = ZTOZSB(zp);
5582 	char attrname[EXTATTR_MAXNAMELEN+1];
5583 	int error;
5584 
5585 	error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
5586 	    sizeof (attrname), compat);
5587 	if (error != 0)
5588 		return (error);
5589 
5590 	error = ENOENT;
5591 	if (zfsvfs->z_use_sa && zp->z_is_sa)
5592 		error = zfs_getextattr_sa(ap, attrname);
5593 	if (error == ENOENT)
5594 		error = zfs_getextattr_dir(ap, attrname);
5595 	return (error);
5596 }
5597 
5598 /*
5599  * Vnode operation to retrieve a named extended attribute.
5600  */
5601 static int
zfs_getextattr(struct vop_getextattr_args * ap)5602 zfs_getextattr(struct vop_getextattr_args *ap)
5603 {
5604 	znode_t *zp = VTOZ(ap->a_vp);
5605 	zfsvfs_t *zfsvfs = ZTOZSB(zp);
5606 	int error;
5607 
5608 	/*
5609 	 * If the xattr property is off, refuse the request.
5610 	 */
5611 	if (!(zfsvfs->z_flags & ZSB_XATTR))
5612 		return (SET_ERROR(EOPNOTSUPP));
5613 
5614 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
5615 	    ap->a_cred, ap->a_td, VREAD);
5616 	if (error != 0)
5617 		return (SET_ERROR(error));
5618 
5619 	error = zfs_check_attrname(ap->a_name);
5620 	if (error != 0)
5621 		return (error);
5622 
5623 	if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0)
5624 		return (error);
5625 	error = ENOENT;
5626 	rw_enter(&zp->z_xattr_lock, RW_READER);
5627 
5628 	error = zfs_getextattr_impl(ap, zfs_xattr_compat);
5629 	if ((error == ENOENT || error == ENOATTR) &&
5630 	    ap->a_attrnamespace == EXTATTR_NAMESPACE_USER) {
5631 		/*
5632 		 * Fall back to the alternate namespace format if we failed to
5633 		 * find a user xattr.
5634 		 */
5635 		error = zfs_getextattr_impl(ap, !zfs_xattr_compat);
5636 	}
5637 
5638 	rw_exit(&zp->z_xattr_lock);
5639 	zfs_exit(zfsvfs, FTAG);
5640 	if (error == ENOENT)
5641 		error = SET_ERROR(ENOATTR);
5642 	return (error);
5643 }
5644 
5645 #ifndef _SYS_SYSPROTO_H_
5646 struct vop_deleteextattr {
5647 	IN struct vnode *a_vp;
5648 	IN int a_attrnamespace;
5649 	IN const char *a_name;
5650 	IN struct ucred *a_cred;
5651 	IN struct thread *a_td;
5652 };
5653 #endif
5654 
5655 static int
zfs_deleteextattr_dir(struct vop_deleteextattr_args * ap,const char * attrname)5656 zfs_deleteextattr_dir(struct vop_deleteextattr_args *ap, const char *attrname)
5657 {
5658 	struct nameidata nd;
5659 	vnode_t *xvp = NULL, *vp;
5660 	int error;
5661 
5662 	error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred,
5663 	    LOOKUP_XATTR, B_FALSE);
5664 	if (error != 0)
5665 		return (error);
5666 
5667 #if __FreeBSD_version < 1400043
5668 	NDINIT_ATVP(&nd, DELETE, NOFOLLOW | LOCKPARENT | LOCKLEAF,
5669 	    UIO_SYSSPACE, attrname, xvp, ap->a_td);
5670 #else
5671 	NDINIT_ATVP(&nd, DELETE, NOFOLLOW | LOCKPARENT | LOCKLEAF,
5672 	    UIO_SYSSPACE, attrname, xvp);
5673 #endif
5674 	error = namei(&nd);
5675 	if (error != 0)
5676 		return (SET_ERROR(error));
5677 
5678 	vp = nd.ni_vp;
5679 	error = VOP_REMOVE(nd.ni_dvp, vp, &nd.ni_cnd);
5680 	NDFREE_PNBUF(&nd);
5681 
5682 	vput(nd.ni_dvp);
5683 	if (vp == nd.ni_dvp)
5684 		vrele(vp);
5685 	else
5686 		vput(vp);
5687 
5688 	return (error);
5689 }
5690 
5691 static int
zfs_deleteextattr_sa(struct vop_deleteextattr_args * ap,const char * attrname)5692 zfs_deleteextattr_sa(struct vop_deleteextattr_args *ap, const char *attrname)
5693 {
5694 	znode_t *zp = VTOZ(ap->a_vp);
5695 	nvlist_t *nvl;
5696 	int error;
5697 
5698 	error = zfs_ensure_xattr_cached(zp);
5699 	if (error != 0)
5700 		return (error);
5701 
5702 	ASSERT(RW_WRITE_HELD(&zp->z_xattr_lock));
5703 	ASSERT3P(zp->z_xattr_cached, !=, NULL);
5704 
5705 	nvl = zp->z_xattr_cached;
5706 	error = nvlist_remove(nvl, attrname, DATA_TYPE_BYTE_ARRAY);
5707 	if (error != 0)
5708 		error = SET_ERROR(error);
5709 	else
5710 		error = zfs_sa_set_xattr(zp, attrname, NULL, 0);
5711 	if (error != 0) {
5712 		zp->z_xattr_cached = NULL;
5713 		nvlist_free(nvl);
5714 	}
5715 	return (error);
5716 }
5717 
5718 static int
zfs_deleteextattr_impl(struct vop_deleteextattr_args * ap,boolean_t compat)5719 zfs_deleteextattr_impl(struct vop_deleteextattr_args *ap, boolean_t compat)
5720 {
5721 	znode_t *zp = VTOZ(ap->a_vp);
5722 	zfsvfs_t *zfsvfs = ZTOZSB(zp);
5723 	char attrname[EXTATTR_MAXNAMELEN+1];
5724 	int error;
5725 
5726 	error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
5727 	    sizeof (attrname), compat);
5728 	if (error != 0)
5729 		return (error);
5730 
5731 	error = ENOENT;
5732 	if (zfsvfs->z_use_sa && zp->z_is_sa)
5733 		error = zfs_deleteextattr_sa(ap, attrname);
5734 	if (error == ENOENT)
5735 		error = zfs_deleteextattr_dir(ap, attrname);
5736 	return (error);
5737 }
5738 
5739 /*
5740  * Vnode operation to remove a named attribute.
5741  */
5742 static int
zfs_deleteextattr(struct vop_deleteextattr_args * ap)5743 zfs_deleteextattr(struct vop_deleteextattr_args *ap)
5744 {
5745 	znode_t *zp = VTOZ(ap->a_vp);
5746 	zfsvfs_t *zfsvfs = ZTOZSB(zp);
5747 	int error;
5748 
5749 	/*
5750 	 * If the xattr property is off, refuse the request.
5751 	 */
5752 	if (!(zfsvfs->z_flags & ZSB_XATTR))
5753 		return (SET_ERROR(EOPNOTSUPP));
5754 
5755 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
5756 	    ap->a_cred, ap->a_td, VWRITE);
5757 	if (error != 0)
5758 		return (SET_ERROR(error));
5759 
5760 	error = zfs_check_attrname(ap->a_name);
5761 	if (error != 0)
5762 		return (error);
5763 
5764 	if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0)
5765 		return (error);
5766 	rw_enter(&zp->z_xattr_lock, RW_WRITER);
5767 
5768 	error = zfs_deleteextattr_impl(ap, zfs_xattr_compat);
5769 	if ((error == ENOENT || error == ENOATTR) &&
5770 	    ap->a_attrnamespace == EXTATTR_NAMESPACE_USER) {
5771 		/*
5772 		 * Fall back to the alternate namespace format if we failed to
5773 		 * find a user xattr.
5774 		 */
5775 		error = zfs_deleteextattr_impl(ap, !zfs_xattr_compat);
5776 	}
5777 
5778 	rw_exit(&zp->z_xattr_lock);
5779 	zfs_exit(zfsvfs, FTAG);
5780 	if (error == ENOENT)
5781 		error = SET_ERROR(ENOATTR);
5782 	return (error);
5783 }
5784 
5785 #ifndef _SYS_SYSPROTO_H_
5786 struct vop_setextattr {
5787 	IN struct vnode *a_vp;
5788 	IN int a_attrnamespace;
5789 	IN const char *a_name;
5790 	INOUT struct uio *a_uio;
5791 	IN struct ucred *a_cred;
5792 	IN struct thread *a_td;
5793 };
5794 #endif
5795 
5796 static int
zfs_setextattr_dir(struct vop_setextattr_args * ap,const char * attrname)5797 zfs_setextattr_dir(struct vop_setextattr_args *ap, const char *attrname)
5798 {
5799 	struct thread *td = ap->a_td;
5800 	struct nameidata nd;
5801 	struct vattr va;
5802 	vnode_t *xvp = NULL, *vp;
5803 	int error, flags;
5804 
5805 	error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred,
5806 	    LOOKUP_XATTR | CREATE_XATTR_DIR, B_FALSE);
5807 	if (error != 0)
5808 		return (error);
5809 
5810 	flags = FFLAGS(O_WRONLY | O_CREAT);
5811 #if __FreeBSD_version < 1400043
5812 	NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname, xvp, td);
5813 #else
5814 	NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname, xvp);
5815 #endif
5816 	error = vn_open_cred(&nd, &flags, 0600, VN_OPEN_INVFS, ap->a_cred,
5817 	    NULL);
5818 	if (error != 0)
5819 		return (SET_ERROR(error));
5820 	vp = nd.ni_vp;
5821 	NDFREE_PNBUF(&nd);
5822 
5823 	VATTR_NULL(&va);
5824 	va.va_size = 0;
5825 	error = VOP_SETATTR(vp, &va, ap->a_cred);
5826 	if (error == 0)
5827 		VOP_WRITE(vp, ap->a_uio, IO_UNIT, ap->a_cred);
5828 
5829 	VOP_UNLOCK(vp);
5830 	vn_close(vp, flags, ap->a_cred, td);
5831 	return (error);
5832 }
5833 
5834 static int
zfs_setextattr_sa(struct vop_setextattr_args * ap,const char * attrname)5835 zfs_setextattr_sa(struct vop_setextattr_args *ap, const char *attrname)
5836 {
5837 	znode_t *zp = VTOZ(ap->a_vp);
5838 	nvlist_t *nvl;
5839 	size_t sa_size;
5840 	int error;
5841 
5842 	error = zfs_ensure_xattr_cached(zp);
5843 	if (error != 0)
5844 		return (error);
5845 
5846 	ASSERT(RW_WRITE_HELD(&zp->z_xattr_lock));
5847 	ASSERT3P(zp->z_xattr_cached, !=, NULL);
5848 
5849 	nvl = zp->z_xattr_cached;
5850 	size_t entry_size = ap->a_uio->uio_resid;
5851 	if (entry_size > DXATTR_MAX_ENTRY_SIZE)
5852 		return (SET_ERROR(EFBIG));
5853 	error = nvlist_size(nvl, &sa_size, NV_ENCODE_XDR);
5854 	if (error != 0)
5855 		return (SET_ERROR(error));
5856 	if (sa_size > DXATTR_MAX_SA_SIZE)
5857 		return (SET_ERROR(EFBIG));
5858 	uchar_t *buf = kmem_alloc(entry_size, KM_SLEEP);
5859 	error = uiomove(buf, entry_size, ap->a_uio);
5860 	if (error != 0) {
5861 		error = SET_ERROR(error);
5862 	} else {
5863 		error = nvlist_add_byte_array(nvl, attrname, buf, entry_size);
5864 		if (error != 0)
5865 			error = SET_ERROR(error);
5866 	}
5867 	if (error == 0)
5868 		error = zfs_sa_set_xattr(zp, attrname, buf, entry_size);
5869 	kmem_free(buf, entry_size);
5870 	if (error != 0) {
5871 		zp->z_xattr_cached = NULL;
5872 		nvlist_free(nvl);
5873 	}
5874 	return (error);
5875 }
5876 
5877 static int
zfs_setextattr_impl(struct vop_setextattr_args * ap,boolean_t compat)5878 zfs_setextattr_impl(struct vop_setextattr_args *ap, boolean_t compat)
5879 {
5880 	znode_t *zp = VTOZ(ap->a_vp);
5881 	zfsvfs_t *zfsvfs = ZTOZSB(zp);
5882 	char attrname[EXTATTR_MAXNAMELEN+1];
5883 	int error;
5884 
5885 	error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
5886 	    sizeof (attrname), compat);
5887 	if (error != 0)
5888 		return (error);
5889 
5890 	struct vop_deleteextattr_args vda = {
5891 		.a_vp = ap->a_vp,
5892 		.a_attrnamespace = ap->a_attrnamespace,
5893 		.a_name = ap->a_name,
5894 		.a_cred = ap->a_cred,
5895 		.a_td = ap->a_td,
5896 	};
5897 	error = ENOENT;
5898 	if (zfsvfs->z_use_sa && zp->z_is_sa && zfsvfs->z_xattr_sa) {
5899 		error = zfs_setextattr_sa(ap, attrname);
5900 		if (error == 0) {
5901 			/*
5902 			 * Successfully put into SA, we need to clear the one
5903 			 * in dir if present.
5904 			 */
5905 			zfs_deleteextattr_dir(&vda, attrname);
5906 		}
5907 	}
5908 	if (error != 0) {
5909 		error = zfs_setextattr_dir(ap, attrname);
5910 		if (error == 0 && zp->z_is_sa) {
5911 			/*
5912 			 * Successfully put into dir, we need to clear the one
5913 			 * in SA if present.
5914 			 */
5915 			zfs_deleteextattr_sa(&vda, attrname);
5916 		}
5917 	}
5918 	if (error == 0 && ap->a_attrnamespace == EXTATTR_NAMESPACE_USER) {
5919 		/*
5920 		 * Also clear all versions of the alternate compat name.
5921 		 */
5922 		zfs_deleteextattr_impl(&vda, !compat);
5923 	}
5924 	return (error);
5925 }
5926 
5927 /*
5928  * Vnode operation to set a named attribute.
5929  */
5930 static int
zfs_setextattr(struct vop_setextattr_args * ap)5931 zfs_setextattr(struct vop_setextattr_args *ap)
5932 {
5933 	znode_t *zp = VTOZ(ap->a_vp);
5934 	zfsvfs_t *zfsvfs = ZTOZSB(zp);
5935 	int error;
5936 
5937 	/*
5938 	 * If the xattr property is off, refuse the request.
5939 	 */
5940 	if (!(zfsvfs->z_flags & ZSB_XATTR))
5941 		return (SET_ERROR(EOPNOTSUPP));
5942 
5943 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
5944 	    ap->a_cred, ap->a_td, VWRITE);
5945 	if (error != 0)
5946 		return (SET_ERROR(error));
5947 
5948 	error = zfs_check_attrname(ap->a_name);
5949 	if (error != 0)
5950 		return (error);
5951 
5952 	if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0)
5953 		return (error);
5954 	rw_enter(&zp->z_xattr_lock, RW_WRITER);
5955 
5956 	error = zfs_setextattr_impl(ap, zfs_xattr_compat);
5957 
5958 	rw_exit(&zp->z_xattr_lock);
5959 	zfs_exit(zfsvfs, FTAG);
5960 	return (error);
5961 }
5962 
5963 #ifndef _SYS_SYSPROTO_H_
5964 struct vop_listextattr {
5965 	IN struct vnode *a_vp;
5966 	IN int a_attrnamespace;
5967 	INOUT struct uio *a_uio;
5968 	OUT size_t *a_size;
5969 	IN struct ucred *a_cred;
5970 	IN struct thread *a_td;
5971 };
5972 #endif
5973 
5974 static int
zfs_listextattr_dir(struct vop_listextattr_args * ap,const char * attrprefix)5975 zfs_listextattr_dir(struct vop_listextattr_args *ap, const char *attrprefix)
5976 {
5977 	struct thread *td = ap->a_td;
5978 	struct nameidata nd;
5979 	uint8_t dirbuf[sizeof (struct dirent)];
5980 	struct iovec aiov;
5981 	struct uio auio;
5982 	vnode_t *xvp = NULL, *vp;
5983 	int error, eof;
5984 
5985 	error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred,
5986 	    LOOKUP_XATTR, B_FALSE);
5987 	if (error != 0) {
5988 		/*
5989 		 * ENOATTR means that the EA directory does not yet exist,
5990 		 * i.e. there are no extended attributes there.
5991 		 */
5992 		if (error == ENOATTR)
5993 			error = 0;
5994 		return (error);
5995 	}
5996 
5997 #if __FreeBSD_version < 1400043
5998 	NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | LOCKLEAF | LOCKSHARED,
5999 	    UIO_SYSSPACE, ".", xvp, td);
6000 #else
6001 	NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | LOCKLEAF | LOCKSHARED,
6002 	    UIO_SYSSPACE, ".", xvp);
6003 #endif
6004 	error = namei(&nd);
6005 	if (error != 0)
6006 		return (SET_ERROR(error));
6007 	vp = nd.ni_vp;
6008 	NDFREE_PNBUF(&nd);
6009 
6010 	auio.uio_iov = &aiov;
6011 	auio.uio_iovcnt = 1;
6012 	auio.uio_segflg = UIO_SYSSPACE;
6013 	auio.uio_td = td;
6014 	auio.uio_rw = UIO_READ;
6015 	auio.uio_offset = 0;
6016 
6017 	size_t plen = strlen(attrprefix);
6018 
6019 	do {
6020 		aiov.iov_base = (void *)dirbuf;
6021 		aiov.iov_len = sizeof (dirbuf);
6022 		auio.uio_resid = sizeof (dirbuf);
6023 		error = VOP_READDIR(vp, &auio, ap->a_cred, &eof, NULL, NULL);
6024 		if (error != 0)
6025 			break;
6026 		int done = sizeof (dirbuf) - auio.uio_resid;
6027 		for (int pos = 0; pos < done; ) {
6028 			struct dirent *dp = (struct dirent *)(dirbuf + pos);
6029 			pos += dp->d_reclen;
6030 			/*
6031 			 * XXX: Temporarily we also accept DT_UNKNOWN, as this
6032 			 * is what we get when attribute was created on Solaris.
6033 			 */
6034 			if (dp->d_type != DT_REG && dp->d_type != DT_UNKNOWN)
6035 				continue;
6036 			else if (plen == 0 &&
6037 			    ZFS_XA_NS_PREFIX_FORBIDDEN(dp->d_name))
6038 				continue;
6039 			else if (strncmp(dp->d_name, attrprefix, plen) != 0)
6040 				continue;
6041 			uint8_t nlen = dp->d_namlen - plen;
6042 			if (ap->a_size != NULL) {
6043 				*ap->a_size += 1 + nlen;
6044 			} else if (ap->a_uio != NULL) {
6045 				/*
6046 				 * Format of extattr name entry is one byte for
6047 				 * length and the rest for name.
6048 				 */
6049 				error = uiomove(&nlen, 1, ap->a_uio);
6050 				if (error == 0) {
6051 					char *namep = dp->d_name + plen;
6052 					error = uiomove(namep, nlen, ap->a_uio);
6053 				}
6054 				if (error != 0) {
6055 					error = SET_ERROR(error);
6056 					break;
6057 				}
6058 			}
6059 		}
6060 	} while (!eof && error == 0);
6061 
6062 	vput(vp);
6063 	return (error);
6064 }
6065 
6066 static int
zfs_listextattr_sa(struct vop_listextattr_args * ap,const char * attrprefix)6067 zfs_listextattr_sa(struct vop_listextattr_args *ap, const char *attrprefix)
6068 {
6069 	znode_t *zp = VTOZ(ap->a_vp);
6070 	int error;
6071 
6072 	error = zfs_ensure_xattr_cached(zp);
6073 	if (error != 0)
6074 		return (error);
6075 
6076 	ASSERT(RW_LOCK_HELD(&zp->z_xattr_lock));
6077 	ASSERT3P(zp->z_xattr_cached, !=, NULL);
6078 
6079 	size_t plen = strlen(attrprefix);
6080 	nvpair_t *nvp = NULL;
6081 	while ((nvp = nvlist_next_nvpair(zp->z_xattr_cached, nvp)) != NULL) {
6082 		ASSERT3U(nvpair_type(nvp), ==, DATA_TYPE_BYTE_ARRAY);
6083 
6084 		const char *name = nvpair_name(nvp);
6085 		if (plen == 0 && ZFS_XA_NS_PREFIX_FORBIDDEN(name))
6086 			continue;
6087 		else if (strncmp(name, attrprefix, plen) != 0)
6088 			continue;
6089 		uint8_t nlen = strlen(name) - plen;
6090 		if (ap->a_size != NULL) {
6091 			*ap->a_size += 1 + nlen;
6092 		} else if (ap->a_uio != NULL) {
6093 			/*
6094 			 * Format of extattr name entry is one byte for
6095 			 * length and the rest for name.
6096 			 */
6097 			error = uiomove(&nlen, 1, ap->a_uio);
6098 			if (error == 0) {
6099 				char *namep = __DECONST(char *, name) + plen;
6100 				error = uiomove(namep, nlen, ap->a_uio);
6101 			}
6102 			if (error != 0) {
6103 				error = SET_ERROR(error);
6104 				break;
6105 			}
6106 		}
6107 	}
6108 
6109 	return (error);
6110 }
6111 
6112 static int
zfs_listextattr_impl(struct vop_listextattr_args * ap,boolean_t compat)6113 zfs_listextattr_impl(struct vop_listextattr_args *ap, boolean_t compat)
6114 {
6115 	znode_t *zp = VTOZ(ap->a_vp);
6116 	zfsvfs_t *zfsvfs = ZTOZSB(zp);
6117 	char attrprefix[16];
6118 	int error;
6119 
6120 	error = zfs_create_attrname(ap->a_attrnamespace, "", attrprefix,
6121 	    sizeof (attrprefix), compat);
6122 	if (error != 0)
6123 		return (error);
6124 
6125 	if (zfsvfs->z_use_sa && zp->z_is_sa)
6126 		error = zfs_listextattr_sa(ap, attrprefix);
6127 	if (error == 0)
6128 		error = zfs_listextattr_dir(ap, attrprefix);
6129 	return (error);
6130 }
6131 
6132 /*
6133  * Vnode operation to retrieve extended attributes on a vnode.
6134  */
6135 static int
zfs_listextattr(struct vop_listextattr_args * ap)6136 zfs_listextattr(struct vop_listextattr_args *ap)
6137 {
6138 	znode_t *zp = VTOZ(ap->a_vp);
6139 	zfsvfs_t *zfsvfs = ZTOZSB(zp);
6140 	int error;
6141 
6142 	if (ap->a_size != NULL)
6143 		*ap->a_size = 0;
6144 
6145 	/*
6146 	 * If the xattr property is off, refuse the request.
6147 	 */
6148 	if (!(zfsvfs->z_flags & ZSB_XATTR))
6149 		return (SET_ERROR(EOPNOTSUPP));
6150 
6151 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6152 	    ap->a_cred, ap->a_td, VREAD);
6153 	if (error != 0)
6154 		return (SET_ERROR(error));
6155 
6156 	if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0)
6157 		return (error);
6158 	rw_enter(&zp->z_xattr_lock, RW_READER);
6159 
6160 	error = zfs_listextattr_impl(ap, zfs_xattr_compat);
6161 	if (error == 0 && ap->a_attrnamespace == EXTATTR_NAMESPACE_USER) {
6162 		/* Also list user xattrs with the alternate format. */
6163 		error = zfs_listextattr_impl(ap, !zfs_xattr_compat);
6164 	}
6165 
6166 	rw_exit(&zp->z_xattr_lock);
6167 	zfs_exit(zfsvfs, FTAG);
6168 	return (error);
6169 }
6170 
6171 #ifndef _SYS_SYSPROTO_H_
6172 struct vop_getacl_args {
6173 	struct vnode *vp;
6174 	acl_type_t type;
6175 	struct acl *aclp;
6176 	struct ucred *cred;
6177 	struct thread *td;
6178 };
6179 #endif
6180 
6181 static int
zfs_freebsd_getacl(struct vop_getacl_args * ap)6182 zfs_freebsd_getacl(struct vop_getacl_args *ap)
6183 {
6184 	int		error;
6185 	vsecattr_t	vsecattr;
6186 
6187 	if (ap->a_type != ACL_TYPE_NFS4)
6188 		return (EINVAL);
6189 
6190 	vsecattr.vsa_mask = VSA_ACE | VSA_ACECNT;
6191 	if ((error = zfs_getsecattr(VTOZ(ap->a_vp),
6192 	    &vsecattr, 0, ap->a_cred)))
6193 		return (error);
6194 
6195 	error = acl_from_aces(ap->a_aclp, vsecattr.vsa_aclentp,
6196 	    vsecattr.vsa_aclcnt);
6197 	if (vsecattr.vsa_aclentp != NULL)
6198 		kmem_free(vsecattr.vsa_aclentp, vsecattr.vsa_aclentsz);
6199 
6200 	return (error);
6201 }
6202 
6203 #ifndef _SYS_SYSPROTO_H_
6204 struct vop_setacl_args {
6205 	struct vnode *vp;
6206 	acl_type_t type;
6207 	struct acl *aclp;
6208 	struct ucred *cred;
6209 	struct thread *td;
6210 };
6211 #endif
6212 
6213 static int
zfs_freebsd_setacl(struct vop_setacl_args * ap)6214 zfs_freebsd_setacl(struct vop_setacl_args *ap)
6215 {
6216 	int		error;
6217 	vsecattr_t vsecattr;
6218 	int		aclbsize;	/* size of acl list in bytes */
6219 	aclent_t	*aaclp;
6220 
6221 	if (ap->a_type != ACL_TYPE_NFS4)
6222 		return (EINVAL);
6223 
6224 	if (ap->a_aclp == NULL)
6225 		return (EINVAL);
6226 
6227 	if (ap->a_aclp->acl_cnt < 1 || ap->a_aclp->acl_cnt > MAX_ACL_ENTRIES)
6228 		return (EINVAL);
6229 
6230 	/*
6231 	 * With NFSv4 ACLs, chmod(2) may need to add additional entries,
6232 	 * splitting every entry into two and appending "canonical six"
6233 	 * entries at the end.  Don't allow for setting an ACL that would
6234 	 * cause chmod(2) to run out of ACL entries.
6235 	 */
6236 	if (ap->a_aclp->acl_cnt * 2 + 6 > ACL_MAX_ENTRIES)
6237 		return (ENOSPC);
6238 
6239 	error = acl_nfs4_check(ap->a_aclp, ap->a_vp->v_type == VDIR);
6240 	if (error != 0)
6241 		return (error);
6242 
6243 	vsecattr.vsa_mask = VSA_ACE;
6244 	aclbsize = ap->a_aclp->acl_cnt * sizeof (ace_t);
6245 	vsecattr.vsa_aclentp = kmem_alloc(aclbsize, KM_SLEEP);
6246 	aaclp = vsecattr.vsa_aclentp;
6247 	vsecattr.vsa_aclentsz = aclbsize;
6248 
6249 	aces_from_acl(vsecattr.vsa_aclentp, &vsecattr.vsa_aclcnt, ap->a_aclp);
6250 	error = zfs_setsecattr(VTOZ(ap->a_vp), &vsecattr, 0, ap->a_cred);
6251 	kmem_free(aaclp, aclbsize);
6252 
6253 	return (error);
6254 }
6255 
6256 #ifndef _SYS_SYSPROTO_H_
6257 struct vop_aclcheck_args {
6258 	struct vnode *vp;
6259 	acl_type_t type;
6260 	struct acl *aclp;
6261 	struct ucred *cred;
6262 	struct thread *td;
6263 };
6264 #endif
6265 
6266 static int
zfs_freebsd_aclcheck(struct vop_aclcheck_args * ap)6267 zfs_freebsd_aclcheck(struct vop_aclcheck_args *ap)
6268 {
6269 
6270 	return (EOPNOTSUPP);
6271 }
6272 
6273 #ifndef _SYS_SYSPROTO_H_
6274 struct vop_advise_args {
6275 	struct vnode *a_vp;
6276 	off_t a_start;
6277 	off_t a_end;
6278 	int a_advice;
6279 };
6280 #endif
6281 
6282 static int
zfs_freebsd_advise(struct vop_advise_args * ap)6283 zfs_freebsd_advise(struct vop_advise_args *ap)
6284 {
6285 	vnode_t *vp = ap->a_vp;
6286 	off_t start = ap->a_start;
6287 	off_t end = ap->a_end;
6288 	int advice = ap->a_advice;
6289 	off_t len;
6290 	znode_t *zp;
6291 	zfsvfs_t *zfsvfs;
6292 	objset_t *os;
6293 	int error = 0;
6294 
6295 	if (end < start)
6296 		return (EINVAL);
6297 
6298 	error = vn_lock(vp, LK_SHARED);
6299 	if (error)
6300 		return (error);
6301 
6302 	zp = VTOZ(vp);
6303 	zfsvfs = zp->z_zfsvfs;
6304 	os = zp->z_zfsvfs->z_os;
6305 
6306 	if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0)
6307 		goto out_unlock;
6308 
6309 	/* kern_posix_fadvise points to the last byte, we want one past */
6310 	if (end != OFF_MAX)
6311 		end += 1;
6312 	len = end - start;
6313 
6314 	switch (advice) {
6315 	case POSIX_FADV_WILLNEED:
6316 		/*
6317 		 * Pass on the caller's size directly, but note that
6318 		 * dmu_prefetch_max will effectively cap it.  If there really
6319 		 * is a larger sequential access pattern, perhaps dmu_zfetch
6320 		 * will detect it.
6321 		 */
6322 		dmu_prefetch(os, zp->z_id, 0, start, len,
6323 		    ZIO_PRIORITY_ASYNC_READ);
6324 		break;
6325 	case POSIX_FADV_NORMAL:
6326 	case POSIX_FADV_RANDOM:
6327 	case POSIX_FADV_SEQUENTIAL:
6328 	case POSIX_FADV_DONTNEED:
6329 	case POSIX_FADV_NOREUSE:
6330 		/* ignored for now */
6331 		break;
6332 	default:
6333 		error = EINVAL;
6334 		break;
6335 	}
6336 
6337 	zfs_exit(zfsvfs, FTAG);
6338 
6339 out_unlock:
6340 	VOP_UNLOCK(vp);
6341 
6342 	return (error);
6343 }
6344 
6345 static int
zfs_vptocnp(struct vop_vptocnp_args * ap)6346 zfs_vptocnp(struct vop_vptocnp_args *ap)
6347 {
6348 	vnode_t *covered_vp;
6349 	vnode_t *vp = ap->a_vp;
6350 	zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
6351 	znode_t *zp = VTOZ(vp);
6352 	int ltype;
6353 	int error;
6354 
6355 	if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0)
6356 		return (error);
6357 
6358 	/*
6359 	 * If we are a snapshot mounted under .zfs, run the operation
6360 	 * on the covered vnode.
6361 	 */
6362 	if (zp->z_id != zfsvfs->z_root || zfsvfs->z_parent == zfsvfs) {
6363 		char name[MAXNAMLEN + 1];
6364 		znode_t *dzp;
6365 		size_t len;
6366 
6367 		error = zfs_znode_parent_and_name(zp, &dzp, name,
6368 		    sizeof (name));
6369 		if (error == 0) {
6370 			len = strlen(name);
6371 			if (*ap->a_buflen < len)
6372 				error = SET_ERROR(ENOMEM);
6373 		}
6374 		if (error == 0) {
6375 			*ap->a_buflen -= len;
6376 			memcpy(ap->a_buf + *ap->a_buflen, name, len);
6377 			*ap->a_vpp = ZTOV(dzp);
6378 		}
6379 		zfs_exit(zfsvfs, FTAG);
6380 		return (error);
6381 	}
6382 	zfs_exit(zfsvfs, FTAG);
6383 
6384 	covered_vp = vp->v_mount->mnt_vnodecovered;
6385 	enum vgetstate vs = vget_prep(covered_vp);
6386 	ltype = VOP_ISLOCKED(vp);
6387 	VOP_UNLOCK(vp);
6388 	error = vget_finish(covered_vp, LK_SHARED, vs);
6389 	if (error == 0) {
6390 		error = VOP_VPTOCNP(covered_vp, ap->a_vpp, ap->a_buf,
6391 		    ap->a_buflen);
6392 		vput(covered_vp);
6393 	}
6394 	vn_lock(vp, ltype | LK_RETRY);
6395 	if (VN_IS_DOOMED(vp))
6396 		error = SET_ERROR(ENOENT);
6397 	return (error);
6398 }
6399 
6400 #if __FreeBSD_version >= 1400032
6401 static int
zfs_deallocate(struct vop_deallocate_args * ap)6402 zfs_deallocate(struct vop_deallocate_args *ap)
6403 {
6404 	znode_t *zp = VTOZ(ap->a_vp);
6405 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
6406 	zilog_t *zilog;
6407 	off_t off, len, file_sz;
6408 	int error;
6409 
6410 	if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0)
6411 		return (error);
6412 
6413 	/*
6414 	 * Callers might not be able to detect properly that we are read-only,
6415 	 * so check it explicitly here.
6416 	 */
6417 	if (zfs_is_readonly(zfsvfs)) {
6418 		zfs_exit(zfsvfs, FTAG);
6419 		return (SET_ERROR(EROFS));
6420 	}
6421 
6422 	zilog = zfsvfs->z_log;
6423 	off = *ap->a_offset;
6424 	len = *ap->a_len;
6425 	file_sz = zp->z_size;
6426 	if (off + len > file_sz)
6427 		len = file_sz - off;
6428 	/* Fast path for out-of-range request. */
6429 	if (len <= 0) {
6430 		*ap->a_len = 0;
6431 		zfs_exit(zfsvfs, FTAG);
6432 		return (0);
6433 	}
6434 
6435 	error = zfs_freesp(zp, off, len, O_RDWR, TRUE);
6436 	if (error == 0) {
6437 		if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS ||
6438 		    (ap->a_ioflag & IO_SYNC) != 0)
6439 			zil_commit(zilog, zp->z_id);
6440 		*ap->a_offset = off + len;
6441 		*ap->a_len = 0;
6442 	}
6443 
6444 	zfs_exit(zfsvfs, FTAG);
6445 	return (error);
6446 }
6447 #endif
6448 
6449 #ifndef _SYS_SYSPROTO_H_
6450 struct vop_copy_file_range_args {
6451 	struct vnode *a_invp;
6452 	off_t *a_inoffp;
6453 	struct vnode *a_outvp;
6454 	off_t *a_outoffp;
6455 	size_t *a_lenp;
6456 	unsigned int a_flags;
6457 	struct ucred *a_incred;
6458 	struct ucred *a_outcred;
6459 	struct thread *a_fsizetd;
6460 }
6461 #endif
6462 /*
6463  * TODO: FreeBSD will only call file system-specific copy_file_range() if both
6464  * files resides under the same mountpoint. In case of ZFS we want to be called
6465  * even is files are in different datasets (but on the same pools, but we need
6466  * to check that ourselves).
6467  */
6468 static int
zfs_freebsd_copy_file_range(struct vop_copy_file_range_args * ap)6469 zfs_freebsd_copy_file_range(struct vop_copy_file_range_args *ap)
6470 {
6471 	zfsvfs_t *outzfsvfs;
6472 	struct vnode *invp = ap->a_invp;
6473 	struct vnode *outvp = ap->a_outvp;
6474 	struct mount *mp;
6475 	int error;
6476 	uint64_t len = *ap->a_lenp;
6477 
6478 	if (!zfs_bclone_enabled) {
6479 		mp = NULL;
6480 		goto bad_write_fallback;
6481 	}
6482 
6483 	/*
6484 	 * TODO: If offset/length is not aligned to recordsize, use
6485 	 * vn_generic_copy_file_range() on this fragment.
6486 	 * It would be better to do this after we lock the vnodes, but then we
6487 	 * need something else than vn_generic_copy_file_range().
6488 	 */
6489 
6490 	vn_start_write(outvp, &mp, V_WAIT);
6491 	if (__predict_true(mp == outvp->v_mount)) {
6492 		outzfsvfs = (zfsvfs_t *)mp->mnt_data;
6493 		if (!spa_feature_is_enabled(dmu_objset_spa(outzfsvfs->z_os),
6494 		    SPA_FEATURE_BLOCK_CLONING)) {
6495 			goto bad_write_fallback;
6496 		}
6497 	}
6498 	if (invp == outvp) {
6499 		if (vn_lock(outvp, LK_EXCLUSIVE) != 0) {
6500 			goto bad_write_fallback;
6501 		}
6502 	} else {
6503 #if (__FreeBSD_version >= 1302506 && __FreeBSD_version < 1400000) || \
6504 	__FreeBSD_version >= 1400086
6505 		vn_lock_pair(invp, false, LK_SHARED, outvp, false,
6506 		    LK_EXCLUSIVE);
6507 #else
6508 		vn_lock_pair(invp, false, outvp, false);
6509 #endif
6510 		if (VN_IS_DOOMED(invp) || VN_IS_DOOMED(outvp)) {
6511 			goto bad_locked_fallback;
6512 		}
6513 	}
6514 
6515 #ifdef MAC
6516 	error = mac_vnode_check_write(curthread->td_ucred, ap->a_outcred,
6517 	    outvp);
6518 	if (error != 0)
6519 		goto out_locked;
6520 #endif
6521 
6522 	error = zfs_clone_range(VTOZ(invp), ap->a_inoffp, VTOZ(outvp),
6523 	    ap->a_outoffp, &len, ap->a_outcred);
6524 	if (error == EXDEV || error == EAGAIN || error == EINVAL ||
6525 	    error == EOPNOTSUPP)
6526 		goto bad_locked_fallback;
6527 	*ap->a_lenp = (size_t)len;
6528 #ifdef MAC
6529 out_locked:
6530 #endif
6531 	if (invp != outvp)
6532 		VOP_UNLOCK(invp);
6533 	VOP_UNLOCK(outvp);
6534 	if (mp != NULL)
6535 		vn_finished_write(mp);
6536 	return (error);
6537 
6538 bad_locked_fallback:
6539 	if (invp != outvp)
6540 		VOP_UNLOCK(invp);
6541 	VOP_UNLOCK(outvp);
6542 bad_write_fallback:
6543 	if (mp != NULL)
6544 		vn_finished_write(mp);
6545 	error = ENOSYS;
6546 	return (error);
6547 }
6548 
6549 struct vop_vector zfs_vnodeops;
6550 struct vop_vector zfs_fifoops;
6551 struct vop_vector zfs_shareops;
6552 
6553 struct vop_vector zfs_vnodeops = {
6554 	.vop_default =		&default_vnodeops,
6555 	.vop_inactive =		zfs_freebsd_inactive,
6556 	.vop_need_inactive =	zfs_freebsd_need_inactive,
6557 	.vop_reclaim =		zfs_freebsd_reclaim,
6558 	.vop_fplookup_vexec = zfs_freebsd_fplookup_vexec,
6559 	.vop_fplookup_symlink = zfs_freebsd_fplookup_symlink,
6560 	.vop_access =		zfs_freebsd_access,
6561 	.vop_allocate =		VOP_EOPNOTSUPP,
6562 #if __FreeBSD_version >= 1400032
6563 	.vop_deallocate =	zfs_deallocate,
6564 #endif
6565 	.vop_lookup =		zfs_cache_lookup,
6566 	.vop_cachedlookup =	zfs_freebsd_cachedlookup,
6567 	.vop_getattr =		zfs_freebsd_getattr,
6568 	.vop_setattr =		zfs_freebsd_setattr,
6569 	.vop_create =		zfs_freebsd_create,
6570 	.vop_mknod =		(vop_mknod_t *)zfs_freebsd_create,
6571 	.vop_mkdir =		zfs_freebsd_mkdir,
6572 	.vop_readdir =		zfs_freebsd_readdir,
6573 	.vop_fsync =		zfs_freebsd_fsync,
6574 	.vop_open =		zfs_freebsd_open,
6575 	.vop_close =		zfs_freebsd_close,
6576 	.vop_rmdir =		zfs_freebsd_rmdir,
6577 	.vop_ioctl =		zfs_freebsd_ioctl,
6578 	.vop_link =		zfs_freebsd_link,
6579 	.vop_symlink =		zfs_freebsd_symlink,
6580 	.vop_readlink =		zfs_freebsd_readlink,
6581 	.vop_advise =		zfs_freebsd_advise,
6582 	.vop_read =		zfs_freebsd_read,
6583 	.vop_write =		zfs_freebsd_write,
6584 	.vop_remove =		zfs_freebsd_remove,
6585 	.vop_rename =		zfs_freebsd_rename,
6586 	.vop_pathconf =		zfs_freebsd_pathconf,
6587 	.vop_bmap =		zfs_freebsd_bmap,
6588 	.vop_fid =		zfs_freebsd_fid,
6589 	.vop_getextattr =	zfs_getextattr,
6590 	.vop_deleteextattr =	zfs_deleteextattr,
6591 	.vop_setextattr =	zfs_setextattr,
6592 	.vop_listextattr =	zfs_listextattr,
6593 	.vop_getacl =		zfs_freebsd_getacl,
6594 	.vop_setacl =		zfs_freebsd_setacl,
6595 	.vop_aclcheck =		zfs_freebsd_aclcheck,
6596 	.vop_getpages =		zfs_freebsd_getpages,
6597 	.vop_putpages =		zfs_freebsd_putpages,
6598 	.vop_vptocnp =		zfs_vptocnp,
6599 	.vop_lock1 =		vop_lock,
6600 	.vop_unlock =		vop_unlock,
6601 	.vop_islocked =		vop_islocked,
6602 #if __FreeBSD_version >= 1400043
6603 	.vop_add_writecount =	vop_stdadd_writecount_nomsync,
6604 #endif
6605 	.vop_copy_file_range =	zfs_freebsd_copy_file_range,
6606 };
6607 VFS_VOP_VECTOR_REGISTER(zfs_vnodeops);
6608 
6609 struct vop_vector zfs_fifoops = {
6610 	.vop_default =		&fifo_specops,
6611 	.vop_fsync =		zfs_freebsd_fsync,
6612 	.vop_fplookup_vexec =	zfs_freebsd_fplookup_vexec,
6613 	.vop_fplookup_symlink = zfs_freebsd_fplookup_symlink,
6614 	.vop_access =		zfs_freebsd_access,
6615 	.vop_getattr =		zfs_freebsd_getattr,
6616 	.vop_inactive =		zfs_freebsd_inactive,
6617 	.vop_read =		VOP_PANIC,
6618 	.vop_reclaim =		zfs_freebsd_reclaim,
6619 	.vop_setattr =		zfs_freebsd_setattr,
6620 	.vop_write =		VOP_PANIC,
6621 	.vop_pathconf = 	zfs_freebsd_pathconf,
6622 	.vop_fid =		zfs_freebsd_fid,
6623 	.vop_getacl =		zfs_freebsd_getacl,
6624 	.vop_setacl =		zfs_freebsd_setacl,
6625 	.vop_aclcheck =		zfs_freebsd_aclcheck,
6626 #if __FreeBSD_version >= 1400043
6627 	.vop_add_writecount =	vop_stdadd_writecount_nomsync,
6628 #endif
6629 };
6630 VFS_VOP_VECTOR_REGISTER(zfs_fifoops);
6631 
6632 /*
6633  * special share hidden files vnode operations template
6634  */
6635 struct vop_vector zfs_shareops = {
6636 	.vop_default =		&default_vnodeops,
6637 	.vop_fplookup_vexec =	VOP_EAGAIN,
6638 	.vop_fplookup_symlink =	VOP_EAGAIN,
6639 	.vop_access =		zfs_freebsd_access,
6640 	.vop_inactive =		zfs_freebsd_inactive,
6641 	.vop_reclaim =		zfs_freebsd_reclaim,
6642 	.vop_fid =		zfs_freebsd_fid,
6643 	.vop_pathconf =		zfs_freebsd_pathconf,
6644 #if __FreeBSD_version >= 1400043
6645 	.vop_add_writecount =	vop_stdadd_writecount_nomsync,
6646 #endif
6647 };
6648 VFS_VOP_VECTOR_REGISTER(zfs_shareops);
6649 
6650 ZFS_MODULE_PARAM(zfs, zfs_, xattr_compat, INT, ZMOD_RW,
6651 	"Use legacy ZFS xattr naming for writing new user namespace xattrs");
6652