xref: /illumos-gate/usr/src/uts/common/fs/zfs/zfs_replay.c (revision 241c90a0)
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/thread.h>
33 #include <sys/file.h>
34 #include <sys/fcntl.h>
35 #include <sys/vfs.h>
36 #include <sys/fs/zfs.h>
37 #include <sys/zfs_znode.h>
38 #include <sys/zfs_dir.h>
39 #include <sys/zfs_acl.h>
40 #include <sys/zfs_fuid.h>
41 #include <sys/spa.h>
42 #include <sys/zil.h>
43 #include <sys/byteorder.h>
44 #include <sys/stat.h>
45 #include <sys/mode.h>
46 #include <sys/acl.h>
47 #include <sys/atomic.h>
48 #include <sys/cred.h>
49 
50 /*
51  * Functions to replay ZFS intent log (ZIL) records
52  * The functions are called through a function vector (zfs_replay_vector)
53  * which is indexed by the transaction type.
54  */
55 
56 static void
57 zfs_init_vattr(vattr_t *vap, uint64_t mask, uint64_t mode,
58     uint64_t uid, uint64_t gid, uint64_t rdev, uint64_t nodeid)
59 {
60 	bzero(vap, sizeof (*vap));
61 	vap->va_mask = (uint_t)mask;
62 	vap->va_type = IFTOVT(mode);
63 	vap->va_mode = mode & MODEMASK;
64 	vap->va_uid = (uid_t)(IS_EPHEMERAL(uid)) ? -1 : uid;
65 	vap->va_gid = (gid_t)(IS_EPHEMERAL(gid)) ? -1 : gid;
66 	vap->va_rdev = zfs_cmpldev(rdev);
67 	vap->va_nodeid = nodeid;
68 }
69 
70 /* ARGSUSED */
71 static int
72 zfs_replay_error(void *arg1, void *arg2, boolean_t byteswap)
73 {
74 	return (SET_ERROR(ENOTSUP));
75 }
76 
77 static void
78 zfs_replay_xvattr(lr_attr_t *lrattr, xvattr_t *xvap)
79 {
80 	xoptattr_t *xoap = NULL;
81 	uint64_t *attrs;
82 	uint64_t *crtime;
83 	uint32_t *bitmap;
84 	void *scanstamp;
85 	int i;
86 
87 	xvap->xva_vattr.va_mask |= AT_XVATTR;
88 	if ((xoap = xva_getxoptattr(xvap)) == NULL) {
89 		xvap->xva_vattr.va_mask &= ~AT_XVATTR; /* shouldn't happen */
90 		return;
91 	}
92 
93 	ASSERT(lrattr->lr_attr_masksize == xvap->xva_mapsize);
94 
95 	bitmap = &lrattr->lr_attr_bitmap;
96 	for (i = 0; i != lrattr->lr_attr_masksize; i++, bitmap++)
97 		xvap->xva_reqattrmap[i] = *bitmap;
98 
99 	attrs = (uint64_t *)(lrattr + lrattr->lr_attr_masksize - 1);
100 	crtime = attrs + 1;
101 	scanstamp = (caddr_t)(crtime + 2);
102 
103 	if (XVA_ISSET_REQ(xvap, XAT_HIDDEN))
104 		xoap->xoa_hidden = ((*attrs & XAT0_HIDDEN) != 0);
105 	if (XVA_ISSET_REQ(xvap, XAT_SYSTEM))
106 		xoap->xoa_system = ((*attrs & XAT0_SYSTEM) != 0);
107 	if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE))
108 		xoap->xoa_archive = ((*attrs & XAT0_ARCHIVE) != 0);
109 	if (XVA_ISSET_REQ(xvap, XAT_READONLY))
110 		xoap->xoa_readonly = ((*attrs & XAT0_READONLY) != 0);
111 	if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE))
112 		xoap->xoa_immutable = ((*attrs & XAT0_IMMUTABLE) != 0);
113 	if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK))
114 		xoap->xoa_nounlink = ((*attrs & XAT0_NOUNLINK) != 0);
115 	if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY))
116 		xoap->xoa_appendonly = ((*attrs & XAT0_APPENDONLY) != 0);
117 	if (XVA_ISSET_REQ(xvap, XAT_NODUMP))
118 		xoap->xoa_nodump = ((*attrs & XAT0_NODUMP) != 0);
119 	if (XVA_ISSET_REQ(xvap, XAT_OPAQUE))
120 		xoap->xoa_opaque = ((*attrs & XAT0_OPAQUE) != 0);
121 	if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED))
122 		xoap->xoa_av_modified = ((*attrs & XAT0_AV_MODIFIED) != 0);
123 	if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED))
124 		xoap->xoa_av_quarantined =
125 		    ((*attrs & XAT0_AV_QUARANTINED) != 0);
126 	if (XVA_ISSET_REQ(xvap, XAT_CREATETIME))
127 		ZFS_TIME_DECODE(&xoap->xoa_createtime, crtime);
128 	if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) {
129 		ASSERT(!XVA_ISSET_REQ(xvap, XAT_PROJID));
130 
131 		bcopy(scanstamp, xoap->xoa_av_scanstamp, AV_SCANSTAMP_SZ);
132 	} else if (XVA_ISSET_REQ(xvap, XAT_PROJID)) {
133 		/*
134 		 * XAT_PROJID and XAT_AV_SCANSTAMP will never be valid
135 		 * at the same time, so we can share the same space.
136 		 */
137 		bcopy(scanstamp, &xoap->xoa_projid, sizeof (uint64_t));
138 	}
139 	if (XVA_ISSET_REQ(xvap, XAT_REPARSE))
140 		xoap->xoa_reparse = ((*attrs & XAT0_REPARSE) != 0);
141 	if (XVA_ISSET_REQ(xvap, XAT_OFFLINE))
142 		xoap->xoa_offline = ((*attrs & XAT0_OFFLINE) != 0);
143 	if (XVA_ISSET_REQ(xvap, XAT_SPARSE))
144 		xoap->xoa_sparse = ((*attrs & XAT0_SPARSE) != 0);
145 	if (XVA_ISSET_REQ(xvap, XAT_PROJINHERIT))
146 		xoap->xoa_projinherit = ((*attrs & XAT0_PROJINHERIT) != 0);
147 }
148 
149 static int
150 zfs_replay_domain_cnt(uint64_t uid, uint64_t gid)
151 {
152 	uint64_t uid_idx;
153 	uint64_t gid_idx;
154 	int domcnt = 0;
155 
156 	uid_idx = FUID_INDEX(uid);
157 	gid_idx = FUID_INDEX(gid);
158 	if (uid_idx)
159 		domcnt++;
160 	if (gid_idx > 0 && gid_idx != uid_idx)
161 		domcnt++;
162 
163 	return (domcnt);
164 }
165 
166 static void *
167 zfs_replay_fuid_domain_common(zfs_fuid_info_t *fuid_infop, void *start,
168     int domcnt)
169 {
170 	int i;
171 
172 	for (i = 0; i != domcnt; i++) {
173 		fuid_infop->z_domain_table[i] = start;
174 		start = (caddr_t)start + strlen(start) + 1;
175 	}
176 
177 	return (start);
178 }
179 
180 /*
181  * Set the uid/gid in the fuid_info structure.
182  */
183 static void
184 zfs_replay_fuid_ugid(zfs_fuid_info_t *fuid_infop, uint64_t uid, uint64_t gid)
185 {
186 	/*
187 	 * If owner or group are log specific FUIDs then slurp up
188 	 * domain information and build zfs_fuid_info_t
189 	 */
190 	if (IS_EPHEMERAL(uid))
191 		fuid_infop->z_fuid_owner = uid;
192 
193 	if (IS_EPHEMERAL(gid))
194 		fuid_infop->z_fuid_group = gid;
195 }
196 
197 /*
198  * Load fuid domains into fuid_info_t
199  */
200 static zfs_fuid_info_t *
201 zfs_replay_fuid_domain(void *buf, void **end, uint64_t uid, uint64_t gid)
202 {
203 	int domcnt;
204 
205 	zfs_fuid_info_t *fuid_infop;
206 
207 	fuid_infop = zfs_fuid_info_alloc();
208 
209 	domcnt = zfs_replay_domain_cnt(uid, gid);
210 
211 	if (domcnt == 0)
212 		return (fuid_infop);
213 
214 	fuid_infop->z_domain_table =
215 	    kmem_zalloc(domcnt * sizeof (char **), KM_SLEEP);
216 
217 	zfs_replay_fuid_ugid(fuid_infop, uid, gid);
218 
219 	fuid_infop->z_domain_cnt = domcnt;
220 	*end = zfs_replay_fuid_domain_common(fuid_infop, buf, domcnt);
221 	return (fuid_infop);
222 }
223 
224 /*
225  * load zfs_fuid_t's and fuid_domains into fuid_info_t
226  */
227 static zfs_fuid_info_t *
228 zfs_replay_fuids(void *start, void **end, int idcnt, int domcnt, uint64_t uid,
229     uint64_t gid)
230 {
231 	uint64_t *log_fuid = (uint64_t *)start;
232 	zfs_fuid_info_t *fuid_infop;
233 	int i;
234 
235 	fuid_infop = zfs_fuid_info_alloc();
236 	fuid_infop->z_domain_cnt = domcnt;
237 
238 	fuid_infop->z_domain_table =
239 	    kmem_zalloc(domcnt * sizeof (char **), KM_SLEEP);
240 
241 	for (i = 0; i != idcnt; i++) {
242 		zfs_fuid_t *zfuid;
243 
244 		zfuid = kmem_alloc(sizeof (zfs_fuid_t), KM_SLEEP);
245 		zfuid->z_logfuid = *log_fuid;
246 		zfuid->z_id = -1;
247 		zfuid->z_domidx = 0;
248 		list_insert_tail(&fuid_infop->z_fuids, zfuid);
249 		log_fuid++;
250 	}
251 
252 	zfs_replay_fuid_ugid(fuid_infop, uid, gid);
253 
254 	*end = zfs_replay_fuid_domain_common(fuid_infop, log_fuid, domcnt);
255 	return (fuid_infop);
256 }
257 
258 static void
259 zfs_replay_swap_attrs(lr_attr_t *lrattr)
260 {
261 	/* swap the lr_attr structure */
262 	byteswap_uint32_array(lrattr, sizeof (*lrattr));
263 	/* swap the bitmap */
264 	byteswap_uint32_array(lrattr + 1, (lrattr->lr_attr_masksize - 1) *
265 	    sizeof (uint32_t));
266 	/* swap the attributes, create time + 64 bit word for attributes */
267 	byteswap_uint64_array((caddr_t)(lrattr + 1) + (sizeof (uint32_t) *
268 	    (lrattr->lr_attr_masksize - 1)), 3 * sizeof (uint64_t));
269 }
270 
271 /*
272  * Replay file create with optional ACL, xvattr information as well
273  * as option FUID information.
274  */
275 static int
276 zfs_replay_create_acl(void *arg1, void *arg2, boolean_t byteswap)
277 {
278 	zfsvfs_t *zfsvfs = arg1;
279 	lr_acl_create_t *lracl = arg2;
280 	char *name = NULL;		/* location determined later */
281 	lr_create_t *lr = (lr_create_t *)lracl;
282 	znode_t *dzp;
283 	vnode_t *vp = NULL;
284 	xvattr_t xva;
285 	int vflg = 0;
286 	vsecattr_t vsec = { 0 };
287 	lr_attr_t *lrattr;
288 	void *aclstart;
289 	void *fuidstart;
290 	size_t xvatlen = 0;
291 	uint64_t txtype;
292 	uint64_t objid;
293 	uint64_t dnodesize;
294 	int error;
295 
296 	txtype = (lr->lr_common.lrc_txtype & ~TX_CI);
297 	if (byteswap) {
298 		byteswap_uint64_array(lracl, sizeof (*lracl));
299 		if (txtype == TX_CREATE_ACL_ATTR ||
300 		    txtype == TX_MKDIR_ACL_ATTR) {
301 			lrattr = (lr_attr_t *)(caddr_t)(lracl + 1);
302 			zfs_replay_swap_attrs(lrattr);
303 			xvatlen = ZIL_XVAT_SIZE(lrattr->lr_attr_masksize);
304 		}
305 
306 		aclstart = (caddr_t)(lracl + 1) + xvatlen;
307 		zfs_ace_byteswap(aclstart, lracl->lr_acl_bytes, B_FALSE);
308 		/* swap fuids */
309 		if (lracl->lr_fuidcnt) {
310 			byteswap_uint64_array((caddr_t)aclstart +
311 			    ZIL_ACE_LENGTH(lracl->lr_acl_bytes),
312 			    lracl->lr_fuidcnt * sizeof (uint64_t));
313 		}
314 	}
315 
316 	if ((error = zfs_zget(zfsvfs, lr->lr_doid, &dzp)) != 0)
317 		return (error);
318 
319 	objid = LR_FOID_GET_OBJ(lr->lr_foid);
320 	dnodesize = LR_FOID_GET_SLOTS(lr->lr_foid) << DNODE_SHIFT;
321 
322 	xva_init(&xva);
323 	zfs_init_vattr(&xva.xva_vattr, AT_TYPE | AT_MODE | AT_UID | AT_GID,
324 	    lr->lr_mode, lr->lr_uid, lr->lr_gid, lr->lr_rdev, objid);
325 
326 	/*
327 	 * All forms of zfs create (create, mkdir, mkxattrdir, symlink)
328 	 * eventually end up in zfs_mknode(), which assigns the object's
329 	 * creation time, generation number, and dnode size. The generic
330 	 * zfs_create() has no concept of these attributes, so we smuggle
331 	 * the values inside the vattr's otherwise unused va_ctime,
332 	 * va_nblocks, and va_fsid fields.
333 	 */
334 	ZFS_TIME_DECODE(&xva.xva_vattr.va_ctime, lr->lr_crtime);
335 	xva.xva_vattr.va_nblocks = lr->lr_gen;
336 	xva.xva_vattr.va_fsid = dnodesize;
337 
338 	error = dnode_try_claim(zfsvfs->z_os, objid, dnodesize >> DNODE_SHIFT);
339 	if (error)
340 		goto bail;
341 
342 	if (lr->lr_common.lrc_txtype & TX_CI)
343 		vflg |= FIGNORECASE;
344 	switch (txtype) {
345 	case TX_CREATE_ACL:
346 		aclstart = (caddr_t)(lracl + 1);
347 		fuidstart = (caddr_t)aclstart +
348 		    ZIL_ACE_LENGTH(lracl->lr_acl_bytes);
349 		zfsvfs->z_fuid_replay = zfs_replay_fuids(fuidstart,
350 		    (void *)&name, lracl->lr_fuidcnt, lracl->lr_domcnt,
351 		    lr->lr_uid, lr->lr_gid);
352 		/*FALLTHROUGH*/
353 	case TX_CREATE_ACL_ATTR:
354 		if (name == NULL) {
355 			lrattr = (lr_attr_t *)(caddr_t)(lracl + 1);
356 			xvatlen = ZIL_XVAT_SIZE(lrattr->lr_attr_masksize);
357 			xva.xva_vattr.va_mask |= AT_XVATTR;
358 			zfs_replay_xvattr(lrattr, &xva);
359 		}
360 		vsec.vsa_mask = VSA_ACE | VSA_ACE_ACLFLAGS;
361 		vsec.vsa_aclentp = (caddr_t)(lracl + 1) + xvatlen;
362 		vsec.vsa_aclcnt = lracl->lr_aclcnt;
363 		vsec.vsa_aclentsz = lracl->lr_acl_bytes;
364 		vsec.vsa_aclflags = lracl->lr_acl_flags;
365 		if (zfsvfs->z_fuid_replay == NULL) {
366 			fuidstart = (caddr_t)(lracl + 1) + xvatlen +
367 			    ZIL_ACE_LENGTH(lracl->lr_acl_bytes);
368 			zfsvfs->z_fuid_replay =
369 			    zfs_replay_fuids(fuidstart,
370 			    (void *)&name, lracl->lr_fuidcnt, lracl->lr_domcnt,
371 			    lr->lr_uid, lr->lr_gid);
372 		}
373 
374 		error = VOP_CREATE(ZTOV(dzp), name, &xva.xva_vattr,
375 		    0, 0, &vp, kcred, vflg, NULL, &vsec);
376 		break;
377 	case TX_MKDIR_ACL:
378 		aclstart = (caddr_t)(lracl + 1);
379 		fuidstart = (caddr_t)aclstart +
380 		    ZIL_ACE_LENGTH(lracl->lr_acl_bytes);
381 		zfsvfs->z_fuid_replay = zfs_replay_fuids(fuidstart,
382 		    (void *)&name, lracl->lr_fuidcnt, lracl->lr_domcnt,
383 		    lr->lr_uid, lr->lr_gid);
384 		/*FALLTHROUGH*/
385 	case TX_MKDIR_ACL_ATTR:
386 		if (name == NULL) {
387 			lrattr = (lr_attr_t *)(caddr_t)(lracl + 1);
388 			xvatlen = ZIL_XVAT_SIZE(lrattr->lr_attr_masksize);
389 			zfs_replay_xvattr(lrattr, &xva);
390 		}
391 		vsec.vsa_mask = VSA_ACE | VSA_ACE_ACLFLAGS;
392 		vsec.vsa_aclentp = (caddr_t)(lracl + 1) + xvatlen;
393 		vsec.vsa_aclcnt = lracl->lr_aclcnt;
394 		vsec.vsa_aclentsz = lracl->lr_acl_bytes;
395 		vsec.vsa_aclflags = lracl->lr_acl_flags;
396 		if (zfsvfs->z_fuid_replay == NULL) {
397 			fuidstart = (caddr_t)(lracl + 1) + xvatlen +
398 			    ZIL_ACE_LENGTH(lracl->lr_acl_bytes);
399 			zfsvfs->z_fuid_replay =
400 			    zfs_replay_fuids(fuidstart,
401 			    (void *)&name, lracl->lr_fuidcnt, lracl->lr_domcnt,
402 			    lr->lr_uid, lr->lr_gid);
403 		}
404 		error = VOP_MKDIR(ZTOV(dzp), name, &xva.xva_vattr,
405 		    &vp, kcred, NULL, vflg, &vsec);
406 		break;
407 	default:
408 		error = SET_ERROR(ENOTSUP);
409 	}
410 
411 bail:
412 	if (error == 0 && vp != NULL)
413 		VN_RELE(vp);
414 
415 	VN_RELE(ZTOV(dzp));
416 
417 	if (zfsvfs->z_fuid_replay)
418 		zfs_fuid_info_free(zfsvfs->z_fuid_replay);
419 	zfsvfs->z_fuid_replay = NULL;
420 
421 	return (error);
422 }
423 
424 static int
425 zfs_replay_create(void *arg1, void *arg2, boolean_t byteswap)
426 {
427 	zfsvfs_t *zfsvfs = arg1;
428 	lr_create_t *lr = arg2;
429 	char *name = NULL;		/* location determined later */
430 	char *link;			/* symlink content follows name */
431 	znode_t *dzp;
432 	vnode_t *vp = NULL;
433 	xvattr_t xva;
434 	int vflg = 0;
435 	size_t lrsize = sizeof (lr_create_t);
436 	lr_attr_t *lrattr;
437 	void *start;
438 	size_t xvatlen;
439 	uint64_t txtype;
440 	int error;
441 
442 	txtype = (lr->lr_common.lrc_txtype & ~TX_CI);
443 	if (byteswap) {
444 		byteswap_uint64_array(lr, sizeof (*lr));
445 		if (txtype == TX_CREATE_ATTR || txtype == TX_MKDIR_ATTR)
446 			zfs_replay_swap_attrs((lr_attr_t *)(lr + 1));
447 	}
448 
449 
450 	if ((error = zfs_zget(zfsvfs, lr->lr_doid, &dzp)) != 0)
451 		return (error);
452 
453 	uint64_t objid = LR_FOID_GET_OBJ(lr->lr_foid);
454 	int dnodesize = LR_FOID_GET_SLOTS(lr->lr_foid) << DNODE_SHIFT;
455 
456 	xva_init(&xva);
457 	zfs_init_vattr(&xva.xva_vattr, AT_TYPE | AT_MODE | AT_UID | AT_GID,
458 	    lr->lr_mode, lr->lr_uid, lr->lr_gid, lr->lr_rdev, objid);
459 
460 	/*
461 	 * All forms of zfs create (create, mkdir, mkxattrdir, symlink)
462 	 * eventually end up in zfs_mknode(), which assigns the object's
463 	 * creation time, generation number, and dnode slot count. The
464 	 * generic zfs_create() has no concept of these attributes, so
465 	 * we smuggle the values inside the vattr's otherwise unused
466 	 * va_ctime, va_nblocks and va_fsid fields.
467 	 */
468 	ZFS_TIME_DECODE(&xva.xva_vattr.va_ctime, lr->lr_crtime);
469 	xva.xva_vattr.va_nblocks = lr->lr_gen;
470 	xva.xva_vattr.va_fsid = dnodesize;
471 
472 	error = dnode_try_claim(zfsvfs->z_os, objid, dnodesize >> DNODE_SHIFT);
473 	if (error)
474 		goto out;
475 
476 	if (lr->lr_common.lrc_txtype & TX_CI)
477 		vflg |= FIGNORECASE;
478 
479 	/*
480 	 * Symlinks don't have fuid info, and CIFS never creates
481 	 * symlinks.
482 	 *
483 	 * The _ATTR versions will grab the fuid info in their subcases.
484 	 */
485 	if ((int)lr->lr_common.lrc_txtype != TX_SYMLINK &&
486 	    (int)lr->lr_common.lrc_txtype != TX_MKDIR_ATTR &&
487 	    (int)lr->lr_common.lrc_txtype != TX_CREATE_ATTR) {
488 		start = (lr + 1);
489 		zfsvfs->z_fuid_replay =
490 		    zfs_replay_fuid_domain(start, &start,
491 		    lr->lr_uid, lr->lr_gid);
492 	}
493 
494 	switch (txtype) {
495 	case TX_CREATE_ATTR:
496 		lrattr = (lr_attr_t *)(caddr_t)(lr + 1);
497 		xvatlen = ZIL_XVAT_SIZE(lrattr->lr_attr_masksize);
498 		zfs_replay_xvattr((lr_attr_t *)((caddr_t)lr + lrsize), &xva);
499 		start = (caddr_t)(lr + 1) + xvatlen;
500 		zfsvfs->z_fuid_replay =
501 		    zfs_replay_fuid_domain(start, &start,
502 		    lr->lr_uid, lr->lr_gid);
503 		name = (char *)start;
504 
505 		/*FALLTHROUGH*/
506 	case TX_CREATE:
507 		if (name == NULL)
508 			name = (char *)start;
509 
510 		error = VOP_CREATE(ZTOV(dzp), name, &xva.xva_vattr,
511 		    0, 0, &vp, kcred, vflg, NULL, NULL);
512 		break;
513 	case TX_MKDIR_ATTR:
514 		lrattr = (lr_attr_t *)(caddr_t)(lr + 1);
515 		xvatlen = ZIL_XVAT_SIZE(lrattr->lr_attr_masksize);
516 		zfs_replay_xvattr((lr_attr_t *)((caddr_t)lr + lrsize), &xva);
517 		start = (caddr_t)(lr + 1) + xvatlen;
518 		zfsvfs->z_fuid_replay =
519 		    zfs_replay_fuid_domain(start, &start,
520 		    lr->lr_uid, lr->lr_gid);
521 		name = (char *)start;
522 
523 		/*FALLTHROUGH*/
524 	case TX_MKDIR:
525 		if (name == NULL)
526 			name = (char *)(lr + 1);
527 
528 		error = VOP_MKDIR(ZTOV(dzp), name, &xva.xva_vattr,
529 		    &vp, kcred, NULL, vflg, NULL);
530 		break;
531 	case TX_MKXATTR:
532 		error = zfs_make_xattrdir(dzp, &xva.xva_vattr, &vp, kcred);
533 		break;
534 	case TX_SYMLINK:
535 		name = (char *)(lr + 1);
536 		link = name + strlen(name) + 1;
537 		error = VOP_SYMLINK(ZTOV(dzp), name, &xva.xva_vattr,
538 		    link, kcred, NULL, vflg);
539 		break;
540 	default:
541 		error = SET_ERROR(ENOTSUP);
542 	}
543 
544 out:
545 	if (error == 0 && vp != NULL)
546 		VN_RELE(vp);
547 
548 	VN_RELE(ZTOV(dzp));
549 
550 	if (zfsvfs->z_fuid_replay)
551 		zfs_fuid_info_free(zfsvfs->z_fuid_replay);
552 	zfsvfs->z_fuid_replay = NULL;
553 	return (error);
554 }
555 
556 static int
557 zfs_replay_remove(void *arg1, void *arg2, boolean_t byteswap)
558 {
559 	zfsvfs_t *zfsvfs = arg1;
560 	lr_remove_t *lr = arg2;
561 	char *name = (char *)(lr + 1);	/* name follows lr_remove_t */
562 	znode_t *dzp;
563 	int error;
564 	int vflg = 0;
565 
566 	if (byteswap)
567 		byteswap_uint64_array(lr, sizeof (*lr));
568 
569 	if ((error = zfs_zget(zfsvfs, lr->lr_doid, &dzp)) != 0)
570 		return (error);
571 
572 	if (lr->lr_common.lrc_txtype & TX_CI)
573 		vflg |= FIGNORECASE;
574 
575 	switch ((int)lr->lr_common.lrc_txtype) {
576 	case TX_REMOVE:
577 		error = VOP_REMOVE(ZTOV(dzp), name, kcred, NULL, vflg);
578 		break;
579 	case TX_RMDIR:
580 		error = VOP_RMDIR(ZTOV(dzp), name, NULL, kcred, NULL, vflg);
581 		break;
582 	default:
583 		error = SET_ERROR(ENOTSUP);
584 	}
585 
586 	VN_RELE(ZTOV(dzp));
587 
588 	return (error);
589 }
590 
591 static int
592 zfs_replay_link(void *arg1, void *arg2, boolean_t byteswap)
593 {
594 	zfsvfs_t *zfsvfs = arg1;
595 	lr_link_t *lr = arg2;
596 	char *name = (char *)(lr + 1);	/* name follows lr_link_t */
597 	znode_t *dzp, *zp;
598 	int error;
599 	int vflg = 0;
600 
601 	if (byteswap)
602 		byteswap_uint64_array(lr, sizeof (*lr));
603 
604 	if ((error = zfs_zget(zfsvfs, lr->lr_doid, &dzp)) != 0)
605 		return (error);
606 
607 	if ((error = zfs_zget(zfsvfs, lr->lr_link_obj, &zp)) != 0) {
608 		VN_RELE(ZTOV(dzp));
609 		return (error);
610 	}
611 
612 	if (lr->lr_common.lrc_txtype & TX_CI)
613 		vflg |= FIGNORECASE;
614 
615 	error = VOP_LINK(ZTOV(dzp), ZTOV(zp), name, kcred, NULL, vflg);
616 
617 	VN_RELE(ZTOV(zp));
618 	VN_RELE(ZTOV(dzp));
619 
620 	return (error);
621 }
622 
623 static int
624 zfs_replay_rename(void *arg1, void *arg2, boolean_t byteswap)
625 {
626 	zfsvfs_t *zfsvfs = arg1;
627 	lr_rename_t *lr = arg2;
628 	char *sname = (char *)(lr + 1);	/* sname and tname follow lr_rename_t */
629 	char *tname = sname + strlen(sname) + 1;
630 	znode_t *sdzp, *tdzp;
631 	int error;
632 	int vflg = 0;
633 
634 	if (byteswap)
635 		byteswap_uint64_array(lr, sizeof (*lr));
636 
637 	if ((error = zfs_zget(zfsvfs, lr->lr_sdoid, &sdzp)) != 0)
638 		return (error);
639 
640 	if ((error = zfs_zget(zfsvfs, lr->lr_tdoid, &tdzp)) != 0) {
641 		VN_RELE(ZTOV(sdzp));
642 		return (error);
643 	}
644 
645 	if (lr->lr_common.lrc_txtype & TX_CI)
646 		vflg |= FIGNORECASE;
647 
648 	error = VOP_RENAME(ZTOV(sdzp), sname, ZTOV(tdzp), tname, kcred,
649 	    NULL, vflg);
650 
651 	VN_RELE(ZTOV(tdzp));
652 	VN_RELE(ZTOV(sdzp));
653 
654 	return (error);
655 }
656 
657 static int
658 zfs_replay_write(void *arg1, void *arg2, boolean_t byteswap)
659 {
660 	zfsvfs_t *zfsvfs = arg1;
661 	lr_write_t *lr = arg2;
662 	char *data = (char *)(lr + 1);	/* data follows lr_write_t */
663 	znode_t	*zp;
664 	int error;
665 	ssize_t resid;
666 	uint64_t eod, offset, length;
667 
668 	if (byteswap)
669 		byteswap_uint64_array(lr, sizeof (*lr));
670 
671 	if ((error = zfs_zget(zfsvfs, lr->lr_foid, &zp)) != 0) {
672 		/*
673 		 * As we can log writes out of order, it's possible the
674 		 * file has been removed. In this case just drop the write
675 		 * and return success.
676 		 */
677 		if (error == ENOENT)
678 			error = 0;
679 		return (error);
680 	}
681 
682 	offset = lr->lr_offset;
683 	length = lr->lr_length;
684 	eod = offset + length;	/* end of data for this write */
685 
686 	/*
687 	 * This may be a write from a dmu_sync() for a whole block,
688 	 * and may extend beyond the current end of the file.
689 	 * We can't just replay what was written for this TX_WRITE as
690 	 * a future TX_WRITE2 may extend the eof and the data for that
691 	 * write needs to be there. So we write the whole block and
692 	 * reduce the eof. This needs to be done within the single dmu
693 	 * transaction created within vn_rdwr -> zfs_write. So a possible
694 	 * new end of file is passed through in zfsvfs->z_replay_eof
695 	 */
696 
697 	zfsvfs->z_replay_eof = 0; /* 0 means don't change end of file */
698 
699 	/* If it's a dmu_sync() block, write the whole block */
700 	if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
701 		uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
702 		if (length < blocksize) {
703 			offset -= offset % blocksize;
704 			length = blocksize;
705 		}
706 		if (zp->z_size < eod)
707 			zfsvfs->z_replay_eof = eod;
708 	}
709 
710 	error = vn_rdwr(UIO_WRITE, ZTOV(zp), data, length, offset,
711 	    UIO_SYSSPACE, 0, RLIM64_INFINITY, kcred, &resid);
712 
713 	VN_RELE(ZTOV(zp));
714 	zfsvfs->z_replay_eof = 0;	/* safety */
715 
716 	return (error);
717 }
718 
719 /*
720  * TX_WRITE2 are only generated when dmu_sync() returns EALREADY
721  * meaning the pool block is already being synced. So now that we always write
722  * out full blocks, all we have to do is expand the eof if
723  * the file is grown.
724  */
725 static int
726 zfs_replay_write2(void *arg1, void *arg2, boolean_t byteswap)
727 {
728 	zfsvfs_t *zfsvfs = arg1;
729 	lr_write_t *lr = arg2;
730 	znode_t	*zp;
731 	int error;
732 	uint64_t end;
733 
734 	if (byteswap)
735 		byteswap_uint64_array(lr, sizeof (*lr));
736 
737 	if ((error = zfs_zget(zfsvfs, lr->lr_foid, &zp)) != 0)
738 		return (error);
739 
740 top:
741 	end = lr->lr_offset + lr->lr_length;
742 	if (end > zp->z_size) {
743 		dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
744 
745 		zp->z_size = end;
746 		dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
747 		error = dmu_tx_assign(tx, TXG_WAIT);
748 		if (error) {
749 			VN_RELE(ZTOV(zp));
750 			if (error == ERESTART) {
751 				dmu_tx_wait(tx);
752 				dmu_tx_abort(tx);
753 				goto top;
754 			}
755 			dmu_tx_abort(tx);
756 			return (error);
757 		}
758 		(void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
759 		    (void *)&zp->z_size, sizeof (uint64_t), tx);
760 
761 		/* Ensure the replayed seq is updated */
762 		(void) zil_replaying(zfsvfs->z_log, tx);
763 
764 		dmu_tx_commit(tx);
765 	}
766 
767 	VN_RELE(ZTOV(zp));
768 
769 	return (error);
770 }
771 
772 static int
773 zfs_replay_truncate(void *arg1, void *arg2, boolean_t byteswap)
774 {
775 	zfsvfs_t *zfsvfs = arg1;
776 	lr_truncate_t *lr = arg2;
777 	znode_t *zp;
778 	flock64_t fl;
779 	int error;
780 
781 	if (byteswap)
782 		byteswap_uint64_array(lr, sizeof (*lr));
783 
784 	if ((error = zfs_zget(zfsvfs, lr->lr_foid, &zp)) != 0)
785 		return (error);
786 
787 	bzero(&fl, sizeof (fl));
788 	fl.l_type = F_WRLCK;
789 	fl.l_whence = 0;
790 	fl.l_start = lr->lr_offset;
791 	fl.l_len = lr->lr_length;
792 
793 	error = VOP_SPACE(ZTOV(zp), F_FREESP, &fl, FWRITE | FOFFMAX,
794 	    lr->lr_offset, kcred, NULL);
795 
796 	VN_RELE(ZTOV(zp));
797 
798 	return (error);
799 }
800 
801 static int
802 zfs_replay_setattr(void *arg1, void *arg2, boolean_t byteswap)
803 {
804 	zfsvfs_t *zfsvfs = arg1;
805 	lr_setattr_t *lr = arg2;
806 	znode_t *zp;
807 	xvattr_t xva;
808 	vattr_t *vap = &xva.xva_vattr;
809 	int error;
810 	void *start;
811 
812 	xva_init(&xva);
813 	if (byteswap) {
814 		byteswap_uint64_array(lr, sizeof (*lr));
815 
816 		if ((lr->lr_mask & AT_XVATTR) &&
817 		    zfsvfs->z_version >= ZPL_VERSION_INITIAL)
818 			zfs_replay_swap_attrs((lr_attr_t *)(lr + 1));
819 	}
820 
821 	if ((error = zfs_zget(zfsvfs, lr->lr_foid, &zp)) != 0)
822 		return (error);
823 
824 	zfs_init_vattr(vap, lr->lr_mask, lr->lr_mode,
825 	    lr->lr_uid, lr->lr_gid, 0, lr->lr_foid);
826 
827 	vap->va_size = lr->lr_size;
828 	ZFS_TIME_DECODE(&vap->va_atime, lr->lr_atime);
829 	ZFS_TIME_DECODE(&vap->va_mtime, lr->lr_mtime);
830 
831 	/*
832 	 * Fill in xvattr_t portions if necessary.
833 	 */
834 
835 	start = (lr_setattr_t *)(lr + 1);
836 	if (vap->va_mask & AT_XVATTR) {
837 		zfs_replay_xvattr((lr_attr_t *)start, &xva);
838 		start = (caddr_t)start +
839 		    ZIL_XVAT_SIZE(((lr_attr_t *)start)->lr_attr_masksize);
840 	} else
841 		xva.xva_vattr.va_mask &= ~AT_XVATTR;
842 
843 	zfsvfs->z_fuid_replay = zfs_replay_fuid_domain(start, &start,
844 	    lr->lr_uid, lr->lr_gid);
845 
846 	error = VOP_SETATTR(ZTOV(zp), vap, 0, kcred, NULL);
847 
848 	zfs_fuid_info_free(zfsvfs->z_fuid_replay);
849 	zfsvfs->z_fuid_replay = NULL;
850 	VN_RELE(ZTOV(zp));
851 
852 	return (error);
853 }
854 
855 static int
856 zfs_replay_acl_v0(void *arg1, void *arg2, boolean_t byteswap)
857 {
858 	zfsvfs_t *zfsvfs = arg1;
859 	lr_acl_v0_t *lr = arg2;
860 	ace_t *ace = (ace_t *)(lr + 1);	/* ace array follows lr_acl_t */
861 	vsecattr_t vsa;
862 	znode_t *zp;
863 	int error;
864 
865 	if (byteswap) {
866 		byteswap_uint64_array(lr, sizeof (*lr));
867 		zfs_oldace_byteswap(ace, lr->lr_aclcnt);
868 	}
869 
870 	if ((error = zfs_zget(zfsvfs, lr->lr_foid, &zp)) != 0)
871 		return (error);
872 
873 	bzero(&vsa, sizeof (vsa));
874 	vsa.vsa_mask = VSA_ACE | VSA_ACECNT;
875 	vsa.vsa_aclcnt = lr->lr_aclcnt;
876 	vsa.vsa_aclentsz = sizeof (ace_t) * vsa.vsa_aclcnt;
877 	vsa.vsa_aclflags = 0;
878 	vsa.vsa_aclentp = ace;
879 
880 	error = VOP_SETSECATTR(ZTOV(zp), &vsa, 0, kcred, NULL);
881 
882 	VN_RELE(ZTOV(zp));
883 
884 	return (error);
885 }
886 
887 /*
888  * Replaying ACLs is complicated by FUID support.
889  * The log record may contain some optional data
890  * to be used for replaying FUID's.  These pieces
891  * are the actual FUIDs that were created initially.
892  * The FUID table index may no longer be valid and
893  * during zfs_create() a new index may be assigned.
894  * Because of this the log will contain the original
895  * doman+rid in order to create a new FUID.
896  *
897  * The individual ACEs may contain an ephemeral uid/gid which is no
898  * longer valid and will need to be replaced with an actual FUID.
899  *
900  */
901 static int
902 zfs_replay_acl(void *arg1, void *arg2, boolean_t byteswap)
903 {
904 	zfsvfs_t *zfsvfs = arg1;
905 	lr_acl_t *lr = arg2;
906 	ace_t *ace = (ace_t *)(lr + 1);
907 	vsecattr_t vsa;
908 	znode_t *zp;
909 	int error;
910 
911 	if (byteswap) {
912 		byteswap_uint64_array(lr, sizeof (*lr));
913 		zfs_ace_byteswap(ace, lr->lr_acl_bytes, B_FALSE);
914 		if (lr->lr_fuidcnt) {
915 			byteswap_uint64_array((caddr_t)ace +
916 			    ZIL_ACE_LENGTH(lr->lr_acl_bytes),
917 			    lr->lr_fuidcnt * sizeof (uint64_t));
918 		}
919 	}
920 
921 	if ((error = zfs_zget(zfsvfs, lr->lr_foid, &zp)) != 0)
922 		return (error);
923 
924 	bzero(&vsa, sizeof (vsa));
925 	vsa.vsa_mask = VSA_ACE | VSA_ACECNT | VSA_ACE_ACLFLAGS;
926 	vsa.vsa_aclcnt = lr->lr_aclcnt;
927 	vsa.vsa_aclentp = ace;
928 	vsa.vsa_aclentsz = lr->lr_acl_bytes;
929 	vsa.vsa_aclflags = lr->lr_acl_flags;
930 
931 	if (lr->lr_fuidcnt) {
932 		void *fuidstart = (caddr_t)ace +
933 		    ZIL_ACE_LENGTH(lr->lr_acl_bytes);
934 
935 		zfsvfs->z_fuid_replay =
936 		    zfs_replay_fuids(fuidstart, &fuidstart,
937 		    lr->lr_fuidcnt, lr->lr_domcnt, 0, 0);
938 	}
939 
940 	error = VOP_SETSECATTR(ZTOV(zp), &vsa, 0, kcred, NULL);
941 
942 	if (zfsvfs->z_fuid_replay)
943 		zfs_fuid_info_free(zfsvfs->z_fuid_replay);
944 
945 	zfsvfs->z_fuid_replay = NULL;
946 	VN_RELE(ZTOV(zp));
947 
948 	return (error);
949 }
950 
951 /*
952  * Callback vectors for replaying records
953  */
954 zil_replay_func_t *zfs_replay_vector[TX_MAX_TYPE] = {
955 	zfs_replay_error,	/* 0 no such transaction type */
956 	zfs_replay_create,	/* TX_CREATE */
957 	zfs_replay_create,	/* TX_MKDIR */
958 	zfs_replay_create,	/* TX_MKXATTR */
959 	zfs_replay_create,	/* TX_SYMLINK */
960 	zfs_replay_remove,	/* TX_REMOVE */
961 	zfs_replay_remove,	/* TX_RMDIR */
962 	zfs_replay_link,	/* TX_LINK */
963 	zfs_replay_rename,	/* TX_RENAME */
964 	zfs_replay_write,	/* TX_WRITE */
965 	zfs_replay_truncate,	/* TX_TRUNCATE */
966 	zfs_replay_setattr,	/* TX_SETATTR */
967 	zfs_replay_acl_v0,	/* TX_ACL_V0 */
968 	zfs_replay_acl,		/* TX_ACL */
969 	zfs_replay_create_acl,	/* TX_CREATE_ACL */
970 	zfs_replay_create,	/* TX_CREATE_ATTR */
971 	zfs_replay_create_acl,	/* TX_CREATE_ACL_ATTR */
972 	zfs_replay_create_acl,	/* TX_MKDIR_ACL */
973 	zfs_replay_create,	/* TX_MKDIR_ATTR */
974 	zfs_replay_create_acl,	/* TX_MKDIR_ACL_ATTR */
975 	zfs_replay_write2,	/* TX_WRITE2 */
976 };
977