xref: /netbsd-src/external/cddl/osnet/dist/uts/common/fs/zfs/zfs_replay.c (revision bdc22b2e01993381dcefeff2bc9b56ca75a4235c)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23  * Copyright (c) 2013, 2015 by Delphix. All rights reserved.
24  */
25 
26 #include <sys/types.h>
27 #include <sys/param.h>
28 #include <sys/systm.h>
29 #include <sys/sysmacros.h>
30 #include <sys/cmn_err.h>
31 #include <sys/kmem.h>
32 #include <sys/file.h>
33 #include <sys/fcntl.h>
34 #include <sys/vfs.h>
35 #include <sys/fs/zfs.h>
36 #include <sys/zfs_znode.h>
37 #include <sys/zfs_dir.h>
38 #include <sys/zfs_acl.h>
39 #include <sys/zfs_fuid.h>
40 #include <sys/spa.h>
41 #include <sys/zil.h>
42 #include <sys/byteorder.h>
43 #include <sys/stat.h>
44 #include <sys/acl.h>
45 #include <sys/atomic.h>
46 #include <sys/cred.h>
47 #include <sys/namei.h>
48 
49 /*
50  * Functions to replay ZFS intent log (ZIL) records
51  * The functions are called through a function vector (zfs_replay_vector)
52  * which is indexed by the transaction type.
53  */
54 
55 static void
56 zfs_init_vattr(vattr_t *vap, uint64_t mask, uint64_t mode,
57     uint64_t uid, uint64_t gid, uint64_t rdev, uint64_t nodeid)
58 {
59 	VATTR_NULL(vap);
60 	vap->va_mask = (uint_t)mask;
61 	if (mask & AT_TYPE)
62 		vap->va_type = IFTOVT(mode);
63 	if (mask & AT_MODE)
64 		vap->va_mode = mode & MODEMASK;
65 	if (mask & AT_UID)
66 		vap->va_uid = (uid_t)(IS_EPHEMERAL(uid)) ? -1 : uid;
67 	if (mask & AT_GID)
68 		vap->va_gid = (gid_t)(IS_EPHEMERAL(gid)) ? -1 : gid;
69 	vap->va_rdev = zfs_cmpldev(rdev);
70 	vap->va_nodeid = nodeid;
71 }
72 
73 /* ARGSUSED */
74 static int
75 zfs_replay_error(zfsvfs_t *zfsvfs, lr_t *lr, boolean_t byteswap)
76 {
77 	return (SET_ERROR(ENOTSUP));
78 }
79 
80 static void
81 zfs_replay_xvattr(lr_attr_t *lrattr, xvattr_t *xvap)
82 {
83 	xoptattr_t *xoap = NULL;
84 	uint64_t *attrs;
85 	uint64_t *crtime;
86 	uint32_t *bitmap;
87 	void *scanstamp;
88 	int i;
89 
90 	xvap->xva_vattr.va_mask |= AT_XVATTR;
91 	if ((xoap = xva_getxoptattr(xvap)) == NULL) {
92 		xvap->xva_vattr.va_mask &= ~AT_XVATTR; /* shouldn't happen */
93 		return;
94 	}
95 
96 	ASSERT(lrattr->lr_attr_masksize == xvap->xva_mapsize);
97 
98 	bitmap = &lrattr->lr_attr_bitmap;
99 	for (i = 0; i != lrattr->lr_attr_masksize; i++, bitmap++)
100 		xvap->xva_reqattrmap[i] = *bitmap;
101 
102 	attrs = (uint64_t *)(lrattr + lrattr->lr_attr_masksize - 1);
103 	crtime = attrs + 1;
104 	scanstamp = (caddr_t)(crtime + 2);
105 
106 	if (XVA_ISSET_REQ(xvap, XAT_HIDDEN))
107 		xoap->xoa_hidden = ((*attrs & XAT0_HIDDEN) != 0);
108 	if (XVA_ISSET_REQ(xvap, XAT_SYSTEM))
109 		xoap->xoa_system = ((*attrs & XAT0_SYSTEM) != 0);
110 	if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE))
111 		xoap->xoa_archive = ((*attrs & XAT0_ARCHIVE) != 0);
112 	if (XVA_ISSET_REQ(xvap, XAT_READONLY))
113 		xoap->xoa_readonly = ((*attrs & XAT0_READONLY) != 0);
114 	if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE))
115 		xoap->xoa_immutable = ((*attrs & XAT0_IMMUTABLE) != 0);
116 	if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK))
117 		xoap->xoa_nounlink = ((*attrs & XAT0_NOUNLINK) != 0);
118 	if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY))
119 		xoap->xoa_appendonly = ((*attrs & XAT0_APPENDONLY) != 0);
120 	if (XVA_ISSET_REQ(xvap, XAT_NODUMP))
121 		xoap->xoa_nodump = ((*attrs & XAT0_NODUMP) != 0);
122 	if (XVA_ISSET_REQ(xvap, XAT_OPAQUE))
123 		xoap->xoa_opaque = ((*attrs & XAT0_OPAQUE) != 0);
124 	if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED))
125 		xoap->xoa_av_modified = ((*attrs & XAT0_AV_MODIFIED) != 0);
126 	if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED))
127 		xoap->xoa_av_quarantined =
128 		    ((*attrs & XAT0_AV_QUARANTINED) != 0);
129 	if (XVA_ISSET_REQ(xvap, XAT_CREATETIME))
130 		ZFS_TIME_DECODE(&xoap->xoa_createtime, crtime);
131 	if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
132 		bcopy(scanstamp, xoap->xoa_av_scanstamp, AV_SCANSTAMP_SZ);
133 	if (XVA_ISSET_REQ(xvap, XAT_REPARSE))
134 		xoap->xoa_reparse = ((*attrs & XAT0_REPARSE) != 0);
135 	if (XVA_ISSET_REQ(xvap, XAT_OFFLINE))
136 		xoap->xoa_offline = ((*attrs & XAT0_OFFLINE) != 0);
137 	if (XVA_ISSET_REQ(xvap, XAT_SPARSE))
138 		xoap->xoa_sparse = ((*attrs & XAT0_SPARSE) != 0);
139 }
140 
141 static int
142 zfs_replay_domain_cnt(uint64_t uid, uint64_t gid)
143 {
144 	uint64_t uid_idx;
145 	uint64_t gid_idx;
146 	int domcnt = 0;
147 
148 	uid_idx = FUID_INDEX(uid);
149 	gid_idx = FUID_INDEX(gid);
150 	if (uid_idx)
151 		domcnt++;
152 	if (gid_idx > 0 && gid_idx != uid_idx)
153 		domcnt++;
154 
155 	return (domcnt);
156 }
157 
158 static void *
159 zfs_replay_fuid_domain_common(zfs_fuid_info_t *fuid_infop, void *start,
160     int domcnt)
161 {
162 	int i;
163 
164 	for (i = 0; i != domcnt; i++) {
165 		fuid_infop->z_domain_table[i] = start;
166 		start = (caddr_t)start + strlen(start) + 1;
167 	}
168 
169 	return (start);
170 }
171 
172 /*
173  * Set the uid/gid in the fuid_info structure.
174  */
175 static void
176 zfs_replay_fuid_ugid(zfs_fuid_info_t *fuid_infop, uint64_t uid, uint64_t gid)
177 {
178 	/*
179 	 * If owner or group are log specific FUIDs then slurp up
180 	 * domain information and build zfs_fuid_info_t
181 	 */
182 	if (IS_EPHEMERAL(uid))
183 		fuid_infop->z_fuid_owner = uid;
184 
185 	if (IS_EPHEMERAL(gid))
186 		fuid_infop->z_fuid_group = gid;
187 }
188 
189 /*
190  * Load fuid domains into fuid_info_t
191  */
192 static zfs_fuid_info_t *
193 zfs_replay_fuid_domain(void *buf, void **end, uint64_t uid, uint64_t gid)
194 {
195 	int domcnt;
196 
197 	zfs_fuid_info_t *fuid_infop;
198 
199 	fuid_infop = zfs_fuid_info_alloc();
200 
201 	domcnt = zfs_replay_domain_cnt(uid, gid);
202 
203 	if (domcnt == 0)
204 		return (fuid_infop);
205 
206 	fuid_infop->z_domain_table =
207 	    kmem_zalloc(domcnt * sizeof (char **), KM_SLEEP);
208 
209 	zfs_replay_fuid_ugid(fuid_infop, uid, gid);
210 
211 	fuid_infop->z_domain_cnt = domcnt;
212 	*end = zfs_replay_fuid_domain_common(fuid_infop, buf, domcnt);
213 	return (fuid_infop);
214 }
215 
216 /*
217  * load zfs_fuid_t's and fuid_domains into fuid_info_t
218  */
219 static zfs_fuid_info_t *
220 zfs_replay_fuids(void *start, void **end, int idcnt, int domcnt, uint64_t uid,
221     uint64_t gid)
222 {
223 	uint64_t *log_fuid = (uint64_t *)start;
224 	zfs_fuid_info_t *fuid_infop;
225 	int i;
226 
227 	fuid_infop = zfs_fuid_info_alloc();
228 	fuid_infop->z_domain_cnt = domcnt;
229 
230 	fuid_infop->z_domain_table =
231 	    kmem_zalloc(domcnt * sizeof (char **), KM_SLEEP);
232 
233 	for (i = 0; i != idcnt; i++) {
234 		zfs_fuid_t *zfuid;
235 
236 		zfuid = kmem_alloc(sizeof (zfs_fuid_t), KM_SLEEP);
237 		zfuid->z_logfuid = *log_fuid;
238 		zfuid->z_id = -1;
239 		zfuid->z_domidx = 0;
240 		list_insert_tail(&fuid_infop->z_fuids, zfuid);
241 		log_fuid++;
242 	}
243 
244 	zfs_replay_fuid_ugid(fuid_infop, uid, gid);
245 
246 	*end = zfs_replay_fuid_domain_common(fuid_infop, log_fuid, domcnt);
247 	return (fuid_infop);
248 }
249 
250 static void
251 zfs_replay_swap_attrs(lr_attr_t *lrattr)
252 {
253 	/* swap the lr_attr structure */
254 	byteswap_uint32_array(lrattr, sizeof (*lrattr));
255 	/* swap the bitmap */
256 	byteswap_uint32_array(lrattr + 1, (lrattr->lr_attr_masksize - 1) *
257 	    sizeof (uint32_t));
258 	/* swap the attributes, create time + 64 bit word for attributes */
259 	byteswap_uint64_array((caddr_t)(lrattr + 1) + (sizeof (uint32_t) *
260 	    (lrattr->lr_attr_masksize - 1)), 3 * sizeof (uint64_t));
261 }
262 
263 /*
264  * Replay file create with optional ACL, xvattr information as well
265  * as option FUID information.
266  */
267 static int
268 zfs_replay_create_acl(zfsvfs_t *zfsvfs,
269     lr_acl_create_t *lracl, boolean_t byteswap)
270 {
271 	char *name = NULL;		/* location determined later */
272 	lr_create_t *lr = (lr_create_t *)lracl;
273 	znode_t *dzp;
274 	vnode_t *vp = NULL;
275 	xvattr_t xva;
276 	int vflg = 0;
277 	vsecattr_t vsec = { 0 };
278 	lr_attr_t *lrattr;
279 	void *aclstart;
280 	void *fuidstart;
281 	size_t xvatlen = 0;
282 	uint64_t txtype;
283 	int error;
284 
285 	txtype = (lr->lr_common.lrc_txtype & ~TX_CI);
286 	if (byteswap) {
287 		byteswap_uint64_array(lracl, sizeof (*lracl));
288 		if (txtype == TX_CREATE_ACL_ATTR ||
289 		    txtype == TX_MKDIR_ACL_ATTR) {
290 			lrattr = (lr_attr_t *)(caddr_t)(lracl + 1);
291 			zfs_replay_swap_attrs(lrattr);
292 			xvatlen = ZIL_XVAT_SIZE(lrattr->lr_attr_masksize);
293 		}
294 
295 		aclstart = (caddr_t)(lracl + 1) + xvatlen;
296 		zfs_ace_byteswap(aclstart, lracl->lr_acl_bytes, B_FALSE);
297 		/* swap fuids */
298 		if (lracl->lr_fuidcnt) {
299 			byteswap_uint64_array((caddr_t)aclstart +
300 			    ZIL_ACE_LENGTH(lracl->lr_acl_bytes),
301 			    lracl->lr_fuidcnt * sizeof (uint64_t));
302 		}
303 	}
304 
305 	if ((error = zfs_zget(zfsvfs, lr->lr_doid, &dzp)) != 0)
306 		return (error);
307 
308 	xva_init(&xva);
309 	zfs_init_vattr(&xva.xva_vattr, AT_TYPE | AT_MODE | AT_UID | AT_GID,
310 	    lr->lr_mode, lr->lr_uid, lr->lr_gid, lr->lr_rdev, lr->lr_foid);
311 
312 	/*
313 	 * All forms of zfs create (create, mkdir, mkxattrdir, symlink)
314 	 * eventually end up in zfs_mknode(), which assigns the object's
315 	 * creation time and generation number.  The generic VOP_CREATE()
316 	 * doesn't have either concept, so we smuggle the values inside
317 	 * the vattr's otherwise unused va_ctime and va_nblocks fields.
318 	 */
319 	ZFS_TIME_DECODE(&xva.xva_vattr.va_ctime, lr->lr_crtime);
320 	xva.xva_vattr.va_nblocks = lr->lr_gen;
321 
322 	error = dmu_object_info(zfsvfs->z_os, lr->lr_foid, NULL);
323 	if (error != ENOENT)
324 		goto bail;
325 
326 	if (lr->lr_common.lrc_txtype & TX_CI)
327 		vflg |= FIGNORECASE;
328 	switch (txtype) {
329 	case TX_CREATE_ACL:
330 		aclstart = (caddr_t)(lracl + 1);
331 		fuidstart = (caddr_t)aclstart +
332 		    ZIL_ACE_LENGTH(lracl->lr_acl_bytes);
333 		zfsvfs->z_fuid_replay = zfs_replay_fuids(fuidstart,
334 		    (void *)&name, lracl->lr_fuidcnt, lracl->lr_domcnt,
335 		    lr->lr_uid, lr->lr_gid);
336 		/*FALLTHROUGH*/
337 	case TX_CREATE_ACL_ATTR:
338 		if (name == NULL) {
339 			lrattr = (lr_attr_t *)(caddr_t)(lracl + 1);
340 			xvatlen = ZIL_XVAT_SIZE(lrattr->lr_attr_masksize);
341 			xva.xva_vattr.va_mask |= AT_XVATTR;
342 			zfs_replay_xvattr(lrattr, &xva);
343 		}
344 		vsec.vsa_mask = VSA_ACE | VSA_ACE_ACLFLAGS;
345 		vsec.vsa_aclentp = (caddr_t)(lracl + 1) + xvatlen;
346 		vsec.vsa_aclcnt = lracl->lr_aclcnt;
347 		vsec.vsa_aclentsz = lracl->lr_acl_bytes;
348 		vsec.vsa_aclflags = lracl->lr_acl_flags;
349 		if (zfsvfs->z_fuid_replay == NULL) {
350 			fuidstart = (caddr_t)(lracl + 1) + xvatlen +
351 			    ZIL_ACE_LENGTH(lracl->lr_acl_bytes);
352 			zfsvfs->z_fuid_replay =
353 			    zfs_replay_fuids(fuidstart,
354 			    (void *)&name, lracl->lr_fuidcnt, lracl->lr_domcnt,
355 			    lr->lr_uid, lr->lr_gid);
356 		}
357 
358 #ifdef TODO
359 		error = VOP_CREATE(ZTOV(dzp), name, &xva.xva_vattr,
360 		    0, 0, &vp, kcred, vflg, NULL, &vsec);
361 #else
362 		panic("%s:%u: unsupported condition", __func__, __LINE__);
363 #endif
364 		break;
365 	case TX_MKDIR_ACL:
366 		aclstart = (caddr_t)(lracl + 1);
367 		fuidstart = (caddr_t)aclstart +
368 		    ZIL_ACE_LENGTH(lracl->lr_acl_bytes);
369 		zfsvfs->z_fuid_replay = zfs_replay_fuids(fuidstart,
370 		    (void *)&name, lracl->lr_fuidcnt, lracl->lr_domcnt,
371 		    lr->lr_uid, lr->lr_gid);
372 		/*FALLTHROUGH*/
373 	case TX_MKDIR_ACL_ATTR:
374 		if (name == NULL) {
375 			lrattr = (lr_attr_t *)(caddr_t)(lracl + 1);
376 			xvatlen = ZIL_XVAT_SIZE(lrattr->lr_attr_masksize);
377 			zfs_replay_xvattr(lrattr, &xva);
378 		}
379 		vsec.vsa_mask = VSA_ACE | VSA_ACE_ACLFLAGS;
380 		vsec.vsa_aclentp = (caddr_t)(lracl + 1) + xvatlen;
381 		vsec.vsa_aclcnt = lracl->lr_aclcnt;
382 		vsec.vsa_aclentsz = lracl->lr_acl_bytes;
383 		vsec.vsa_aclflags = lracl->lr_acl_flags;
384 		if (zfsvfs->z_fuid_replay == NULL) {
385 			fuidstart = (caddr_t)(lracl + 1) + xvatlen +
386 			    ZIL_ACE_LENGTH(lracl->lr_acl_bytes);
387 			zfsvfs->z_fuid_replay =
388 			    zfs_replay_fuids(fuidstart,
389 			    (void *)&name, lracl->lr_fuidcnt, lracl->lr_domcnt,
390 			    lr->lr_uid, lr->lr_gid);
391 		}
392 #ifdef TODO
393 		error = VOP_MKDIR(ZTOV(dzp), name, &xva.xva_vattr,
394 		    &vp, kcred, NULL, vflg, &vsec);
395 #else
396 		panic("%s:%u: unsupported condition", __func__, __LINE__);
397 #endif
398 		break;
399 	default:
400 		error = SET_ERROR(ENOTSUP);
401 	}
402 
403 bail:
404 	if (error == 0 && vp != NULL)
405 		VN_RELE(vp);
406 
407 	VN_RELE(ZTOV(dzp));
408 
409 	if (zfsvfs->z_fuid_replay)
410 		zfs_fuid_info_free(zfsvfs->z_fuid_replay);
411 	zfsvfs->z_fuid_replay = NULL;
412 
413 	return (error);
414 }
415 
416 static int
417 zfs_replay_create(zfsvfs_t *zfsvfs, lr_create_t *lr, boolean_t byteswap)
418 {
419 	char *name = NULL;		/* location determined later */
420 	char *link;			/* symlink content follows name */
421 	znode_t *dzp;
422 	vnode_t *vp = NULL;
423 	xvattr_t xva;
424 	int vflg = 0;
425 	size_t lrsize = sizeof (lr_create_t);
426 	lr_attr_t *lrattr;
427 	void *start;
428 	size_t xvatlen;
429 	uint64_t txtype;
430 	struct componentname cn;
431 	int error;
432 
433 	txtype = (lr->lr_common.lrc_txtype & ~TX_CI);
434 	if (byteswap) {
435 		byteswap_uint64_array(lr, sizeof (*lr));
436 		if (txtype == TX_CREATE_ATTR || txtype == TX_MKDIR_ATTR)
437 			zfs_replay_swap_attrs((lr_attr_t *)(lr + 1));
438 	}
439 
440 
441 	if ((error = zfs_zget(zfsvfs, lr->lr_doid, &dzp)) != 0)
442 		return (error);
443 
444 	xva_init(&xva);
445 	zfs_init_vattr(&xva.xva_vattr, AT_TYPE | AT_MODE | AT_UID | AT_GID,
446 	    lr->lr_mode, lr->lr_uid, lr->lr_gid, lr->lr_rdev, lr->lr_foid);
447 
448 	/*
449 	 * All forms of zfs create (create, mkdir, mkxattrdir, symlink)
450 	 * eventually end up in zfs_mknode(), which assigns the object's
451 	 * creation time and generation number.  The generic VOP_CREATE()
452 	 * doesn't have either concept, so we smuggle the values inside
453 	 * the vattr's otherwise unused va_ctime and va_nblocks fields.
454 	 */
455 	ZFS_TIME_DECODE(&xva.xva_vattr.va_ctime, lr->lr_crtime);
456 	xva.xva_vattr.va_nblocks = lr->lr_gen;
457 
458 	error = dmu_object_info(zfsvfs->z_os, lr->lr_foid, NULL);
459 	if (error != ENOENT)
460 		goto out;
461 
462 	if (lr->lr_common.lrc_txtype & TX_CI)
463 		vflg |= FIGNORECASE;
464 
465 	/*
466 	 * Symlinks don't have fuid info, and CIFS never creates
467 	 * symlinks.
468 	 *
469 	 * The _ATTR versions will grab the fuid info in their subcases.
470 	 */
471 	if ((int)lr->lr_common.lrc_txtype != TX_SYMLINK &&
472 	    (int)lr->lr_common.lrc_txtype != TX_MKDIR_ATTR &&
473 	    (int)lr->lr_common.lrc_txtype != TX_CREATE_ATTR) {
474 		start = (lr + 1);
475 		zfsvfs->z_fuid_replay =
476 		    zfs_replay_fuid_domain(start, &start,
477 		    lr->lr_uid, lr->lr_gid);
478 	}
479 
480 	cn.cn_cred = kcred;
481 #ifndef __NetBSD__
482 	cn.cn_thread = curthread;
483 	cn.cn_flags = SAVENAME;
484 #endif
485 
486 	vn_lock(ZTOV(dzp), LK_EXCLUSIVE | LK_RETRY);
487 	switch (txtype) {
488 	case TX_CREATE_ATTR:
489 		lrattr = (lr_attr_t *)(caddr_t)(lr + 1);
490 		xvatlen = ZIL_XVAT_SIZE(lrattr->lr_attr_masksize);
491 		zfs_replay_xvattr((lr_attr_t *)((caddr_t)lr + lrsize), &xva);
492 		start = (caddr_t)(lr + 1) + xvatlen;
493 		zfsvfs->z_fuid_replay =
494 		    zfs_replay_fuid_domain(start, &start,
495 		    lr->lr_uid, lr->lr_gid);
496 		name = (char *)start;
497 
498 		/*FALLTHROUGH*/
499 	case TX_CREATE:
500 		if (name == NULL)
501 			name = (char *)start;
502 
503 		cn.cn_nameptr = name;
504 		error = VOP_CREATE(ZTOV(dzp), &vp, &cn, &xva.xva_vattr /*,vflg*/);
505 		break;
506 	case TX_MKDIR_ATTR:
507 		lrattr = (lr_attr_t *)(caddr_t)(lr + 1);
508 		xvatlen = ZIL_XVAT_SIZE(lrattr->lr_attr_masksize);
509 		zfs_replay_xvattr((lr_attr_t *)((caddr_t)lr + lrsize), &xva);
510 		start = (caddr_t)(lr + 1) + xvatlen;
511 		zfsvfs->z_fuid_replay =
512 		    zfs_replay_fuid_domain(start, &start,
513 		    lr->lr_uid, lr->lr_gid);
514 		name = (char *)start;
515 
516 		/*FALLTHROUGH*/
517 	case TX_MKDIR:
518 		if (name == NULL)
519 			name = (char *)(lr + 1);
520 
521 		cn.cn_nameptr = name;
522 		error = VOP_MKDIR(ZTOV(dzp), &vp, &cn, &xva.xva_vattr /*,vflg*/);
523 		break;
524 	case TX_MKXATTR:
525 		error = zfs_make_xattrdir(dzp, &xva.xva_vattr, &vp, kcred);
526 		break;
527 	case TX_SYMLINK:
528 		name = (char *)(lr + 1);
529 		link = name + strlen(name) + 1;
530 		cn.cn_nameptr = name;
531 		error = VOP_SYMLINK(ZTOV(dzp), &vp, &cn, &xva.xva_vattr, link /*,vflg*/);
532 		break;
533 	default:
534 		error = SET_ERROR(ENOTSUP);
535 	}
536 	VOP_UNLOCK(ZTOV(dzp), 0);
537 
538 out:
539 	if (error == 0 && vp != NULL)
540 #ifdef __NetBSD__
541 		VN_RELE(vp);
542 #else
543 		VN_URELE(vp);
544 #endif
545 
546 	VN_RELE(ZTOV(dzp));
547 
548 	if (zfsvfs->z_fuid_replay)
549 		zfs_fuid_info_free(zfsvfs->z_fuid_replay);
550 	zfsvfs->z_fuid_replay = NULL;
551 	return (error);
552 }
553 
554 static int
555 zfs_replay_remove(zfsvfs_t *zfsvfs, lr_remove_t *lr, boolean_t byteswap)
556 {
557 	char *name = (char *)(lr + 1);	/* name follows lr_remove_t */
558 	znode_t *dzp;
559 	struct componentname cn;
560 	vnode_t *vp;
561 	int error;
562 	int vflg = 0;
563 
564 	if (byteswap)
565 		byteswap_uint64_array(lr, sizeof (*lr));
566 
567 	if ((error = zfs_zget(zfsvfs, lr->lr_doid, &dzp)) != 0)
568 		return (error);
569 
570 	if (lr->lr_common.lrc_txtype & TX_CI)
571 		vflg |= FIGNORECASE;
572 	cn.cn_nameptr = name;
573 	cn.cn_namelen = strlen(name);
574 	cn.cn_nameiop = DELETE;
575 	cn.cn_flags = ISLASTCN | SAVENAME;
576 	cn.cn_cred = kcred;
577 #ifndef __NetBSD__
578 	cn.cn_lkflags = LK_EXCLUSIVE | LK_RETRY;
579 	cn.cn_thread = curthread;
580 #endif
581 	vn_lock(ZTOV(dzp), LK_EXCLUSIVE | LK_RETRY);
582 	error = VOP_LOOKUP(ZTOV(dzp), &vp, &cn);
583 	if (error != 0) {
584 		VOP_UNLOCK(ZTOV(dzp), 0);
585 		goto fail;
586 	}
587 #ifdef __NetBSD__
588 	VOP_UNLOCK(vp, 0);
589 #endif
590 
591 	switch ((int)lr->lr_common.lrc_txtype) {
592 	case TX_REMOVE:
593 		error = VOP_REMOVE(ZTOV(dzp), vp, &cn /*,vflg*/);
594 		break;
595 	case TX_RMDIR:
596 		error = VOP_RMDIR(ZTOV(dzp), vp, &cn /*,vflg*/);
597 		break;
598 	default:
599 		error = SET_ERROR(ENOTSUP);
600 	}
601 #ifdef __NetBSD__
602 	vrele(vp);
603 #else
604 	vput(vp);
605 #endif
606 	VOP_UNLOCK(ZTOV(dzp), 0);
607 
608 fail:
609 	VN_RELE(ZTOV(dzp));
610 
611 	return (error);
612 }
613 
614 static int
615 zfs_replay_link(zfsvfs_t *zfsvfs, lr_link_t *lr, boolean_t byteswap)
616 {
617 	char *name = (char *)(lr + 1);	/* name follows lr_link_t */
618 	znode_t *dzp, *zp;
619 	struct componentname cn;
620 	int error;
621 	int vflg = 0;
622 
623 	if (byteswap)
624 		byteswap_uint64_array(lr, sizeof (*lr));
625 
626 	if ((error = zfs_zget(zfsvfs, lr->lr_doid, &dzp)) != 0)
627 		return (error);
628 
629 	if ((error = zfs_zget(zfsvfs, lr->lr_link_obj, &zp)) != 0) {
630 		VN_RELE(ZTOV(dzp));
631 		return (error);
632 	}
633 
634 	if (lr->lr_common.lrc_txtype & TX_CI)
635 		vflg |= FIGNORECASE;
636 
637 	cn.cn_nameptr = name;
638 	cn.cn_cred = kcred;
639 #ifndef __NetBSD__
640 	cn.cn_thread = curthread;
641 	cn.cn_flags = SAVENAME;
642 #endif
643 
644 	vn_lock(ZTOV(dzp), LK_EXCLUSIVE | LK_RETRY);
645 #ifndef __NetBSD__
646 	vn_lock(ZTOV(zp), LK_EXCLUSIVE | LK_RETRY);
647 #endif
648 	error = VOP_LINK(ZTOV(dzp), ZTOV(zp), &cn /*,vflg*/);
649 #ifndef __NetBSD__
650 	VOP_UNLOCK(ZTOV(zp), 0);
651 #endif
652 	VOP_UNLOCK(ZTOV(dzp), 0);
653 
654 	VN_RELE(ZTOV(zp));
655 	VN_RELE(ZTOV(dzp));
656 
657 	return (error);
658 }
659 
660 static int
661 zfs_replay_rename(zfsvfs_t *zfsvfs, lr_rename_t *lr, boolean_t byteswap)
662 {
663 	char *sname = (char *)(lr + 1);	/* sname and tname follow lr_rename_t */
664 	char *tname = sname + strlen(sname) + 1;
665 	znode_t *sdzp, *tdzp;
666 	struct componentname scn, tcn;
667 	vnode_t *svp, *tvp;
668 	kthread_t *td = curthread;
669 	int error;
670 	int vflg = 0;
671 
672 	if (byteswap)
673 		byteswap_uint64_array(lr, sizeof (*lr));
674 
675 	if ((error = zfs_zget(zfsvfs, lr->lr_sdoid, &sdzp)) != 0)
676 		return (error);
677 
678 	if ((error = zfs_zget(zfsvfs, lr->lr_tdoid, &tdzp)) != 0) {
679 		VN_RELE(ZTOV(sdzp));
680 		return (error);
681 	}
682 
683 	if (lr->lr_common.lrc_txtype & TX_CI)
684 		vflg |= FIGNORECASE;
685 	svp = tvp = NULL;
686 
687 	scn.cn_nameptr = sname;
688 	scn.cn_namelen = strlen(sname);
689 	scn.cn_nameiop = DELETE;
690 	scn.cn_flags = ISLASTCN | SAVENAME;
691 	scn.cn_cred = kcred;
692 #ifndef __NetBSD__
693 	scn.cn_lkflags = LK_EXCLUSIVE | LK_RETRY;
694 	scn.cn_thread = td;
695 #endif
696 	vn_lock(ZTOV(sdzp), LK_EXCLUSIVE | LK_RETRY);
697 	error = VOP_LOOKUP(ZTOV(sdzp), &svp, &scn);
698 	VOP_UNLOCK(ZTOV(sdzp), 0);
699 	if (error != 0)
700 		goto fail;
701 	VOP_UNLOCK(svp, 0);
702 
703 	tcn.cn_nameptr = tname;
704 	tcn.cn_namelen = strlen(tname);
705 	tcn.cn_nameiop = RENAME;
706 	tcn.cn_flags = ISLASTCN | SAVENAME;
707 	tcn.cn_cred = kcred;
708 #ifndef __NetBSD__
709 	tcn.cn_lkflags = LK_EXCLUSIVE | LK_RETRY;
710 	tcn.cn_thread = td;
711 #endif
712 	vn_lock(ZTOV(tdzp), LK_EXCLUSIVE | LK_RETRY);
713 	error = VOP_LOOKUP(ZTOV(tdzp), &tvp, &tcn);
714 	if (error == EJUSTRETURN)
715 		tvp = NULL;
716 	else if (error != 0) {
717 		VOP_UNLOCK(ZTOV(tdzp), 0);
718 		goto fail;
719 	}
720 #ifdef __NetBSD__
721 	vn_lock(tvp, LK_EXCLUSIVE | LK_RETRY);
722 #endif
723 
724 	error = VOP_RENAME(ZTOV(sdzp), svp, &scn, ZTOV(tdzp), tvp, &tcn /*,vflg*/);
725 	return (error);
726 fail:
727 	if (svp != NULL)
728 		vrele(svp);
729 	if (tvp != NULL)
730 		vrele(tvp);
731 	VN_RELE(ZTOV(tdzp));
732 	VN_RELE(ZTOV(sdzp));
733 
734 	return (error);
735 }
736 
737 static int
738 zfs_replay_write(zfsvfs_t *zfsvfs, lr_write_t *lr, boolean_t byteswap)
739 {
740 	char *data = (char *)(lr + 1);	/* data follows lr_write_t */
741 	znode_t	*zp;
742 	int error;
743 	ssize_t resid;
744 	uint64_t eod, offset, length;
745 
746 	if (byteswap)
747 		byteswap_uint64_array(lr, sizeof (*lr));
748 
749 	if ((error = zfs_zget(zfsvfs, lr->lr_foid, &zp)) != 0) {
750 		/*
751 		 * As we can log writes out of order, it's possible the
752 		 * file has been removed. In this case just drop the write
753 		 * and return success.
754 		 */
755 		if (error == ENOENT)
756 			error = 0;
757 		return (error);
758 	}
759 
760 	offset = lr->lr_offset;
761 	length = lr->lr_length;
762 	eod = offset + length;	/* end of data for this write */
763 
764 	/*
765 	 * This may be a write from a dmu_sync() for a whole block,
766 	 * and may extend beyond the current end of the file.
767 	 * We can't just replay what was written for this TX_WRITE as
768 	 * a future TX_WRITE2 may extend the eof and the data for that
769 	 * write needs to be there. So we write the whole block and
770 	 * reduce the eof. This needs to be done within the single dmu
771 	 * transaction created within vn_rdwr -> zfs_write. So a possible
772 	 * new end of file is passed through in zfsvfs->z_replay_eof
773 	 */
774 
775 	zfsvfs->z_replay_eof = 0; /* 0 means don't change end of file */
776 
777 	/* If it's a dmu_sync() block, write the whole block */
778 	if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
779 		uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
780 		if (length < blocksize) {
781 			offset -= offset % blocksize;
782 			length = blocksize;
783 		}
784 		if (zp->z_size < eod)
785 			zfsvfs->z_replay_eof = eod;
786 	}
787 
788 	error = vn_rdwr(UIO_WRITE, ZTOV(zp), data, length, offset,
789 	    UIO_SYSSPACE, 0, RLIM64_INFINITY, kcred, &resid);
790 
791 	VN_RELE(ZTOV(zp));
792 	zfsvfs->z_replay_eof = 0;	/* safety */
793 
794 	return (error);
795 }
796 
797 /*
798  * TX_WRITE2 are only generated when dmu_sync() returns EALREADY
799  * meaning the pool block is already being synced. So now that we always write
800  * out full blocks, all we have to do is expand the eof if
801  * the file is grown.
802  */
803 static int
804 zfs_replay_write2(zfsvfs_t *zfsvfs, lr_write_t *lr, boolean_t byteswap)
805 {
806 	znode_t	*zp;
807 	int error;
808 	uint64_t end;
809 
810 	if (byteswap)
811 		byteswap_uint64_array(lr, sizeof (*lr));
812 
813 	if ((error = zfs_zget(zfsvfs, lr->lr_foid, &zp)) != 0)
814 		return (error);
815 
816 top:
817 	end = lr->lr_offset + lr->lr_length;
818 	if (end > zp->z_size) {
819 		dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
820 
821 		zp->z_size = end;
822 		dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
823 		error = dmu_tx_assign(tx, TXG_WAIT);
824 		if (error) {
825 			VN_RELE(ZTOV(zp));
826 			if (error == ERESTART) {
827 				dmu_tx_wait(tx);
828 				dmu_tx_abort(tx);
829 				goto top;
830 			}
831 			dmu_tx_abort(tx);
832 			return (error);
833 		}
834 		(void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
835 		    (void *)&zp->z_size, sizeof (uint64_t), tx);
836 
837 		/* Ensure the replayed seq is updated */
838 		(void) zil_replaying(zfsvfs->z_log, tx);
839 
840 		dmu_tx_commit(tx);
841 	}
842 
843 	VN_RELE(ZTOV(zp));
844 
845 	return (error);
846 }
847 
848 static int
849 zfs_replay_truncate(zfsvfs_t *zfsvfs, lr_truncate_t *lr, boolean_t byteswap)
850 {
851 #ifdef illumos
852 	znode_t *zp;
853 	flock64_t fl;
854 	int error;
855 
856 	if (byteswap)
857 		byteswap_uint64_array(lr, sizeof (*lr));
858 
859 	if ((error = zfs_zget(zfsvfs, lr->lr_foid, &zp)) != 0)
860 		return (error);
861 
862 	bzero(&fl, sizeof (fl));
863 	fl.l_type = F_WRLCK;
864 	fl.l_whence = 0;
865 	fl.l_start = lr->lr_offset;
866 	fl.l_len = lr->lr_length;
867 
868 	error = VOP_SPACE(ZTOV(zp), F_FREESP, &fl, FWRITE | FOFFMAX,
869 	    lr->lr_offset, kcred, NULL);
870 
871 	VN_RELE(ZTOV(zp));
872 
873 	return (error);
874 #else
875 	ZFS_LOG(0, "Unexpected code path, report to pjd@FreeBSD.org");
876 	return (EOPNOTSUPP);
877 #endif
878 }
879 
880 static int
881 zfs_replay_setattr(zfsvfs_t *zfsvfs, lr_setattr_t *lr, boolean_t byteswap)
882 {
883 	znode_t *zp;
884 	xvattr_t xva;
885 	vattr_t *vap = &xva.xva_vattr;
886 	vnode_t *vp;
887 	int error;
888 	void *start;
889 
890 	xva_init(&xva);
891 	if (byteswap) {
892 		byteswap_uint64_array(lr, sizeof (*lr));
893 
894 		if ((lr->lr_mask & AT_XVATTR) &&
895 		    zfsvfs->z_version >= ZPL_VERSION_INITIAL)
896 			zfs_replay_swap_attrs((lr_attr_t *)(lr + 1));
897 	}
898 
899 	if ((error = zfs_zget(zfsvfs, lr->lr_foid, &zp)) != 0)
900 		return (error);
901 
902 	zfs_init_vattr(vap, lr->lr_mask, lr->lr_mode,
903 	    lr->lr_uid, lr->lr_gid, 0, lr->lr_foid);
904 
905 	vap->va_size = lr->lr_size;
906 	ZFS_TIME_DECODE(&vap->va_atime, lr->lr_atime);
907 	ZFS_TIME_DECODE(&vap->va_mtime, lr->lr_mtime);
908 
909 	/*
910 	 * Fill in xvattr_t portions if necessary.
911 	 */
912 
913 	start = (lr_setattr_t *)(lr + 1);
914 	if (vap->va_mask & AT_XVATTR) {
915 		zfs_replay_xvattr((lr_attr_t *)start, &xva);
916 		start = (caddr_t)start +
917 		    ZIL_XVAT_SIZE(((lr_attr_t *)start)->lr_attr_masksize);
918 	} else
919 		xva.xva_vattr.va_mask &= ~AT_XVATTR;
920 
921 	zfsvfs->z_fuid_replay = zfs_replay_fuid_domain(start, &start,
922 	    lr->lr_uid, lr->lr_gid);
923 
924 	vp = ZTOV(zp);
925 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
926 	error = VOP_SETATTR(vp, vap, kcred);
927 	VOP_UNLOCK(vp, 0);
928 
929 	zfs_fuid_info_free(zfsvfs->z_fuid_replay);
930 	zfsvfs->z_fuid_replay = NULL;
931 	VN_RELE(vp);
932 
933 	return (error);
934 }
935 
936 extern int zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
937     caller_context_t *ct);
938 
939 static int
940 zfs_replay_acl_v0(zfsvfs_t *zfsvfs, lr_acl_v0_t *lr, boolean_t byteswap)
941 {
942 	ace_t *ace = (ace_t *)(lr + 1);	/* ace array follows lr_acl_t */
943 	vsecattr_t vsa;
944 	vnode_t *vp;
945 	znode_t *zp;
946 	int error;
947 
948 	if (byteswap) {
949 		byteswap_uint64_array(lr, sizeof (*lr));
950 		zfs_oldace_byteswap(ace, lr->lr_aclcnt);
951 	}
952 
953 	if ((error = zfs_zget(zfsvfs, lr->lr_foid, &zp)) != 0)
954 		return (error);
955 
956 	bzero(&vsa, sizeof (vsa));
957 	vsa.vsa_mask = VSA_ACE | VSA_ACECNT;
958 	vsa.vsa_aclcnt = lr->lr_aclcnt;
959 	vsa.vsa_aclentsz = sizeof (ace_t) * vsa.vsa_aclcnt;
960 	vsa.vsa_aclflags = 0;
961 	vsa.vsa_aclentp = ace;
962 
963 	vp = ZTOV(zp);
964 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
965 	error = zfs_setsecattr(vp, &vsa, 0, kcred, NULL);
966 	VOP_UNLOCK(vp, 0);
967 
968 	VN_RELE(vp);
969 
970 	return (error);
971 }
972 
973 /*
974  * Replaying ACLs is complicated by FUID support.
975  * The log record may contain some optional data
976  * to be used for replaying FUID's.  These pieces
977  * are the actual FUIDs that were created initially.
978  * The FUID table index may no longer be valid and
979  * during zfs_create() a new index may be assigned.
980  * Because of this the log will contain the original
981  * doman+rid in order to create a new FUID.
982  *
983  * The individual ACEs may contain an ephemeral uid/gid which is no
984  * longer valid and will need to be replaced with an actual FUID.
985  *
986  */
987 static int
988 zfs_replay_acl(zfsvfs_t *zfsvfs, lr_acl_t *lr, boolean_t byteswap)
989 {
990 	ace_t *ace = (ace_t *)(lr + 1);
991 	vsecattr_t vsa;
992 	znode_t *zp;
993 	vnode_t *vp;
994 	int error;
995 
996 	if (byteswap) {
997 		byteswap_uint64_array(lr, sizeof (*lr));
998 		zfs_ace_byteswap(ace, lr->lr_acl_bytes, B_FALSE);
999 		if (lr->lr_fuidcnt) {
1000 			byteswap_uint64_array((caddr_t)ace +
1001 			    ZIL_ACE_LENGTH(lr->lr_acl_bytes),
1002 			    lr->lr_fuidcnt * sizeof (uint64_t));
1003 		}
1004 	}
1005 
1006 	if ((error = zfs_zget(zfsvfs, lr->lr_foid, &zp)) != 0)
1007 		return (error);
1008 
1009 	bzero(&vsa, sizeof (vsa));
1010 	vsa.vsa_mask = VSA_ACE | VSA_ACECNT | VSA_ACE_ACLFLAGS;
1011 	vsa.vsa_aclcnt = lr->lr_aclcnt;
1012 	vsa.vsa_aclentp = ace;
1013 	vsa.vsa_aclentsz = lr->lr_acl_bytes;
1014 	vsa.vsa_aclflags = lr->lr_acl_flags;
1015 
1016 	if (lr->lr_fuidcnt) {
1017 		void *fuidstart = (caddr_t)ace +
1018 		    ZIL_ACE_LENGTH(lr->lr_acl_bytes);
1019 
1020 		zfsvfs->z_fuid_replay =
1021 		    zfs_replay_fuids(fuidstart, &fuidstart,
1022 		    lr->lr_fuidcnt, lr->lr_domcnt, 0, 0);
1023 	}
1024 
1025 	vp = ZTOV(zp);
1026 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1027 	error = zfs_setsecattr(vp, &vsa, 0, kcred, NULL);
1028 	VOP_UNLOCK(vp, 0);
1029 
1030 	if (zfsvfs->z_fuid_replay)
1031 		zfs_fuid_info_free(zfsvfs->z_fuid_replay);
1032 
1033 	zfsvfs->z_fuid_replay = NULL;
1034 	VN_RELE(vp);
1035 
1036 	return (error);
1037 }
1038 
1039 /*
1040  * Callback vectors for replaying records
1041  */
1042 zil_replay_func_t *zfs_replay_vector[TX_MAX_TYPE] = {
1043 	zfs_replay_error,	/* 0 no such transaction type */
1044 	zfs_replay_create,	/* TX_CREATE */
1045 	zfs_replay_create,	/* TX_MKDIR */
1046 	zfs_replay_create,	/* TX_MKXATTR */
1047 	zfs_replay_create,	/* TX_SYMLINK */
1048 	zfs_replay_remove,	/* TX_REMOVE */
1049 	zfs_replay_remove,	/* TX_RMDIR */
1050 	zfs_replay_link,	/* TX_LINK */
1051 	zfs_replay_rename,	/* TX_RENAME */
1052 	zfs_replay_write,	/* TX_WRITE */
1053 	zfs_replay_truncate,	/* TX_TRUNCATE */
1054 	zfs_replay_setattr,	/* TX_SETATTR */
1055 	zfs_replay_acl_v0,	/* TX_ACL_V0 */
1056 	zfs_replay_acl,		/* TX_ACL */
1057 	zfs_replay_create_acl,	/* TX_CREATE_ACL */
1058 	zfs_replay_create,	/* TX_CREATE_ATTR */
1059 	zfs_replay_create_acl,	/* TX_CREATE_ACL_ATTR */
1060 	zfs_replay_create_acl,	/* TX_MKDIR_ACL */
1061 	zfs_replay_create,	/* TX_MKDIR_ATTR */
1062 	zfs_replay_create_acl,	/* TX_MKDIR_ACL_ATTR */
1063 	zfs_replay_write2,	/* TX_WRITE2 */
1064 };
1065