xref: /freebsd/sys/contrib/openzfs/module/zfs/zfs_replay.c (revision c66ec88fed842fbaad62c30d510644ceb7bd2d71)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23  * Copyright (c) 2012 Cyril Plisko. All rights reserved.
24  * Copyright (c) 2013, 2017 by Delphix. All rights reserved.
25  */
26 
27 #include <sys/types.h>
28 #include <sys/param.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/zfs_vnops.h>
42 #include <sys/spa.h>
43 #include <sys/zil.h>
44 #include <sys/byteorder.h>
45 #include <sys/stat.h>
46 #include <sys/acl.h>
47 #include <sys/atomic.h>
48 #include <sys/cred.h>
49 #include <sys/zpl.h>
50 
51 /*
52  * NB: FreeBSD expects to be able to do vnode locking in lookup and
53  * hold the locks across all subsequent VOPs until vput is called.
54  * This means that its zfs vnops routines can't do any internal locking.
55  * In order to have the same contract as the Linux vnops there would
56  * needed to be duplicate locked vnops. If the vnops were used more widely
57  * in common code this would likely be preferable. However, currently
58  * this is the only file where this is the case.
59  */
60 
61 /*
62  * Functions to replay ZFS intent log (ZIL) records
63  * The functions are called through a function vector (zfs_replay_vector)
64  * which is indexed by the transaction type.
65  */
66 
67 static void
68 zfs_init_vattr(vattr_t *vap, uint64_t mask, uint64_t mode,
69     uint64_t uid, uint64_t gid, uint64_t rdev, uint64_t nodeid)
70 {
71 	bzero(vap, sizeof (*vap));
72 	vap->va_mask = (uint_t)mask;
73 	vap->va_mode = mode;
74 #ifdef __FreeBSD__
75 	vap->va_type = IFTOVT(mode);
76 #endif
77 	vap->va_uid = (uid_t)(IS_EPHEMERAL(uid)) ? -1 : uid;
78 	vap->va_gid = (gid_t)(IS_EPHEMERAL(gid)) ? -1 : gid;
79 	vap->va_rdev = zfs_cmpldev(rdev);
80 	vap->va_nodeid = nodeid;
81 }
82 
83 /* ARGSUSED */
84 static int
85 zfs_replay_error(void *arg1, void *arg2, boolean_t byteswap)
86 {
87 	return (SET_ERROR(ENOTSUP));
88 }
89 
90 static void
91 zfs_replay_xvattr(lr_attr_t *lrattr, xvattr_t *xvap)
92 {
93 	xoptattr_t *xoap = NULL;
94 	uint64_t *attrs;
95 	uint64_t *crtime;
96 	uint32_t *bitmap;
97 	void *scanstamp;
98 	int i;
99 
100 	xvap->xva_vattr.va_mask |= ATTR_XVATTR;
101 	if ((xoap = xva_getxoptattr(xvap)) == NULL) {
102 		xvap->xva_vattr.va_mask &= ~ATTR_XVATTR; /* shouldn't happen */
103 		return;
104 	}
105 
106 	ASSERT(lrattr->lr_attr_masksize == xvap->xva_mapsize);
107 
108 	bitmap = &lrattr->lr_attr_bitmap;
109 	for (i = 0; i != lrattr->lr_attr_masksize; i++, bitmap++)
110 		xvap->xva_reqattrmap[i] = *bitmap;
111 
112 	attrs = (uint64_t *)(lrattr + lrattr->lr_attr_masksize - 1);
113 	crtime = attrs + 1;
114 	scanstamp = (caddr_t)(crtime + 2);
115 
116 	if (XVA_ISSET_REQ(xvap, XAT_HIDDEN))
117 		xoap->xoa_hidden = ((*attrs & XAT0_HIDDEN) != 0);
118 	if (XVA_ISSET_REQ(xvap, XAT_SYSTEM))
119 		xoap->xoa_system = ((*attrs & XAT0_SYSTEM) != 0);
120 	if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE))
121 		xoap->xoa_archive = ((*attrs & XAT0_ARCHIVE) != 0);
122 	if (XVA_ISSET_REQ(xvap, XAT_READONLY))
123 		xoap->xoa_readonly = ((*attrs & XAT0_READONLY) != 0);
124 	if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE))
125 		xoap->xoa_immutable = ((*attrs & XAT0_IMMUTABLE) != 0);
126 	if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK))
127 		xoap->xoa_nounlink = ((*attrs & XAT0_NOUNLINK) != 0);
128 	if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY))
129 		xoap->xoa_appendonly = ((*attrs & XAT0_APPENDONLY) != 0);
130 	if (XVA_ISSET_REQ(xvap, XAT_NODUMP))
131 		xoap->xoa_nodump = ((*attrs & XAT0_NODUMP) != 0);
132 	if (XVA_ISSET_REQ(xvap, XAT_OPAQUE))
133 		xoap->xoa_opaque = ((*attrs & XAT0_OPAQUE) != 0);
134 	if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED))
135 		xoap->xoa_av_modified = ((*attrs & XAT0_AV_MODIFIED) != 0);
136 	if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED))
137 		xoap->xoa_av_quarantined =
138 		    ((*attrs & XAT0_AV_QUARANTINED) != 0);
139 	if (XVA_ISSET_REQ(xvap, XAT_CREATETIME))
140 		ZFS_TIME_DECODE(&xoap->xoa_createtime, crtime);
141 	if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) {
142 		ASSERT(!XVA_ISSET_REQ(xvap, XAT_PROJID));
143 
144 		bcopy(scanstamp, xoap->xoa_av_scanstamp, AV_SCANSTAMP_SZ);
145 	} else if (XVA_ISSET_REQ(xvap, XAT_PROJID)) {
146 		/*
147 		 * XAT_PROJID and XAT_AV_SCANSTAMP will never be valid
148 		 * at the same time, so we can share the same space.
149 		 */
150 		bcopy(scanstamp, &xoap->xoa_projid, sizeof (uint64_t));
151 	}
152 	if (XVA_ISSET_REQ(xvap, XAT_REPARSE))
153 		xoap->xoa_reparse = ((*attrs & XAT0_REPARSE) != 0);
154 	if (XVA_ISSET_REQ(xvap, XAT_OFFLINE))
155 		xoap->xoa_offline = ((*attrs & XAT0_OFFLINE) != 0);
156 	if (XVA_ISSET_REQ(xvap, XAT_SPARSE))
157 		xoap->xoa_sparse = ((*attrs & XAT0_SPARSE) != 0);
158 	if (XVA_ISSET_REQ(xvap, XAT_PROJINHERIT))
159 		xoap->xoa_projinherit = ((*attrs & XAT0_PROJINHERIT) != 0);
160 }
161 
162 static int
163 zfs_replay_domain_cnt(uint64_t uid, uint64_t gid)
164 {
165 	uint64_t uid_idx;
166 	uint64_t gid_idx;
167 	int domcnt = 0;
168 
169 	uid_idx = FUID_INDEX(uid);
170 	gid_idx = FUID_INDEX(gid);
171 	if (uid_idx)
172 		domcnt++;
173 	if (gid_idx > 0 && gid_idx != uid_idx)
174 		domcnt++;
175 
176 	return (domcnt);
177 }
178 
179 static void *
180 zfs_replay_fuid_domain_common(zfs_fuid_info_t *fuid_infop, void *start,
181     int domcnt)
182 {
183 	int i;
184 
185 	for (i = 0; i != domcnt; i++) {
186 		fuid_infop->z_domain_table[i] = start;
187 		start = (caddr_t)start + strlen(start) + 1;
188 	}
189 
190 	return (start);
191 }
192 
193 /*
194  * Set the uid/gid in the fuid_info structure.
195  */
196 static void
197 zfs_replay_fuid_ugid(zfs_fuid_info_t *fuid_infop, uint64_t uid, uint64_t gid)
198 {
199 	/*
200 	 * If owner or group are log specific FUIDs then slurp up
201 	 * domain information and build zfs_fuid_info_t
202 	 */
203 	if (IS_EPHEMERAL(uid))
204 		fuid_infop->z_fuid_owner = uid;
205 
206 	if (IS_EPHEMERAL(gid))
207 		fuid_infop->z_fuid_group = gid;
208 }
209 
210 /*
211  * Load fuid domains into fuid_info_t
212  */
213 static zfs_fuid_info_t *
214 zfs_replay_fuid_domain(void *buf, void **end, uint64_t uid, uint64_t gid)
215 {
216 	int domcnt;
217 
218 	zfs_fuid_info_t *fuid_infop;
219 
220 	fuid_infop = zfs_fuid_info_alloc();
221 
222 	domcnt = zfs_replay_domain_cnt(uid, gid);
223 
224 	if (domcnt == 0)
225 		return (fuid_infop);
226 
227 	fuid_infop->z_domain_table =
228 	    kmem_zalloc(domcnt * sizeof (char *), KM_SLEEP);
229 
230 	zfs_replay_fuid_ugid(fuid_infop, uid, gid);
231 
232 	fuid_infop->z_domain_cnt = domcnt;
233 	*end = zfs_replay_fuid_domain_common(fuid_infop, buf, domcnt);
234 	return (fuid_infop);
235 }
236 
237 /*
238  * load zfs_fuid_t's and fuid_domains into fuid_info_t
239  */
240 static zfs_fuid_info_t *
241 zfs_replay_fuids(void *start, void **end, int idcnt, int domcnt, uint64_t uid,
242     uint64_t gid)
243 {
244 	uint64_t *log_fuid = (uint64_t *)start;
245 	zfs_fuid_info_t *fuid_infop;
246 	int i;
247 
248 	fuid_infop = zfs_fuid_info_alloc();
249 	fuid_infop->z_domain_cnt = domcnt;
250 
251 	fuid_infop->z_domain_table =
252 	    kmem_zalloc(domcnt * sizeof (char *), KM_SLEEP);
253 
254 	for (i = 0; i != idcnt; i++) {
255 		zfs_fuid_t *zfuid;
256 
257 		zfuid = kmem_alloc(sizeof (zfs_fuid_t), KM_SLEEP);
258 		zfuid->z_logfuid = *log_fuid;
259 		zfuid->z_id = -1;
260 		zfuid->z_domidx = 0;
261 		list_insert_tail(&fuid_infop->z_fuids, zfuid);
262 		log_fuid++;
263 	}
264 
265 	zfs_replay_fuid_ugid(fuid_infop, uid, gid);
266 
267 	*end = zfs_replay_fuid_domain_common(fuid_infop, log_fuid, domcnt);
268 	return (fuid_infop);
269 }
270 
271 static void
272 zfs_replay_swap_attrs(lr_attr_t *lrattr)
273 {
274 	/* swap the lr_attr structure */
275 	byteswap_uint32_array(lrattr, sizeof (*lrattr));
276 	/* swap the bitmap */
277 	byteswap_uint32_array(lrattr + 1, (lrattr->lr_attr_masksize - 1) *
278 	    sizeof (uint32_t));
279 	/* swap the attributes, create time + 64 bit word for attributes */
280 	byteswap_uint64_array((caddr_t)(lrattr + 1) + (sizeof (uint32_t) *
281 	    (lrattr->lr_attr_masksize - 1)), 3 * sizeof (uint64_t));
282 }
283 
284 /*
285  * Replay file create with optional ACL, xvattr information as well
286  * as option FUID information.
287  */
288 static int
289 zfs_replay_create_acl(void *arg1, void *arg2, boolean_t byteswap)
290 {
291 	zfsvfs_t *zfsvfs = arg1;
292 	lr_acl_create_t *lracl = arg2;
293 	char *name = NULL;		/* location determined later */
294 	lr_create_t *lr = (lr_create_t *)lracl;
295 	znode_t *dzp;
296 	znode_t *zp;
297 	xvattr_t xva;
298 	int vflg = 0;
299 	vsecattr_t vsec = { 0 };
300 	lr_attr_t *lrattr;
301 	void *aclstart;
302 	void *fuidstart;
303 	size_t xvatlen = 0;
304 	uint64_t txtype;
305 	uint64_t objid;
306 	uint64_t dnodesize;
307 	int error;
308 
309 	txtype = (lr->lr_common.lrc_txtype & ~TX_CI);
310 	if (byteswap) {
311 		byteswap_uint64_array(lracl, sizeof (*lracl));
312 		if (txtype == TX_CREATE_ACL_ATTR ||
313 		    txtype == TX_MKDIR_ACL_ATTR) {
314 			lrattr = (lr_attr_t *)(caddr_t)(lracl + 1);
315 			zfs_replay_swap_attrs(lrattr);
316 			xvatlen = ZIL_XVAT_SIZE(lrattr->lr_attr_masksize);
317 		}
318 
319 		aclstart = (caddr_t)(lracl + 1) + xvatlen;
320 		zfs_ace_byteswap(aclstart, lracl->lr_acl_bytes, B_FALSE);
321 		/* swap fuids */
322 		if (lracl->lr_fuidcnt) {
323 			byteswap_uint64_array((caddr_t)aclstart +
324 			    ZIL_ACE_LENGTH(lracl->lr_acl_bytes),
325 			    lracl->lr_fuidcnt * sizeof (uint64_t));
326 		}
327 	}
328 
329 	if ((error = zfs_zget(zfsvfs, lr->lr_doid, &dzp)) != 0)
330 		return (error);
331 
332 	objid = LR_FOID_GET_OBJ(lr->lr_foid);
333 	dnodesize = LR_FOID_GET_SLOTS(lr->lr_foid) << DNODE_SHIFT;
334 
335 	xva_init(&xva);
336 	zfs_init_vattr(&xva.xva_vattr, ATTR_MODE | ATTR_UID | ATTR_GID,
337 	    lr->lr_mode, lr->lr_uid, lr->lr_gid, lr->lr_rdev, objid);
338 
339 	/*
340 	 * All forms of zfs create (create, mkdir, mkxattrdir, symlink)
341 	 * eventually end up in zfs_mknode(), which assigns the object's
342 	 * creation time, generation number, and dnode size. The generic
343 	 * zfs_create() has no concept of these attributes, so we smuggle
344 	 * the values inside the vattr's otherwise unused va_ctime,
345 	 * va_nblocks, and va_fsid fields.
346 	 */
347 	ZFS_TIME_DECODE(&xva.xva_vattr.va_ctime, lr->lr_crtime);
348 	xva.xva_vattr.va_nblocks = lr->lr_gen;
349 	xva.xva_vattr.va_fsid = dnodesize;
350 
351 	error = dnode_try_claim(zfsvfs->z_os, objid, dnodesize >> DNODE_SHIFT);
352 	if (error)
353 		goto bail;
354 
355 	if (lr->lr_common.lrc_txtype & TX_CI)
356 		vflg |= FIGNORECASE;
357 	switch (txtype) {
358 	case TX_CREATE_ACL:
359 		aclstart = (caddr_t)(lracl + 1);
360 		fuidstart = (caddr_t)aclstart +
361 		    ZIL_ACE_LENGTH(lracl->lr_acl_bytes);
362 		zfsvfs->z_fuid_replay = zfs_replay_fuids(fuidstart,
363 		    (void *)&name, lracl->lr_fuidcnt, lracl->lr_domcnt,
364 		    lr->lr_uid, lr->lr_gid);
365 		/*FALLTHROUGH*/
366 	case TX_CREATE_ACL_ATTR:
367 		if (name == NULL) {
368 			lrattr = (lr_attr_t *)(caddr_t)(lracl + 1);
369 			xvatlen = ZIL_XVAT_SIZE(lrattr->lr_attr_masksize);
370 			xva.xva_vattr.va_mask |= ATTR_XVATTR;
371 			zfs_replay_xvattr(lrattr, &xva);
372 		}
373 		vsec.vsa_mask = VSA_ACE | VSA_ACE_ACLFLAGS;
374 		vsec.vsa_aclentp = (caddr_t)(lracl + 1) + xvatlen;
375 		vsec.vsa_aclcnt = lracl->lr_aclcnt;
376 		vsec.vsa_aclentsz = lracl->lr_acl_bytes;
377 		vsec.vsa_aclflags = lracl->lr_acl_flags;
378 		if (zfsvfs->z_fuid_replay == NULL) {
379 			fuidstart = (caddr_t)(lracl + 1) + xvatlen +
380 			    ZIL_ACE_LENGTH(lracl->lr_acl_bytes);
381 			zfsvfs->z_fuid_replay =
382 			    zfs_replay_fuids(fuidstart,
383 			    (void *)&name, lracl->lr_fuidcnt, lracl->lr_domcnt,
384 			    lr->lr_uid, lr->lr_gid);
385 		}
386 
387 		error = zfs_create(dzp, name, &xva.xva_vattr,
388 		    0, 0, &zp, kcred, vflg, &vsec);
389 		break;
390 	case TX_MKDIR_ACL:
391 		aclstart = (caddr_t)(lracl + 1);
392 		fuidstart = (caddr_t)aclstart +
393 		    ZIL_ACE_LENGTH(lracl->lr_acl_bytes);
394 		zfsvfs->z_fuid_replay = zfs_replay_fuids(fuidstart,
395 		    (void *)&name, lracl->lr_fuidcnt, lracl->lr_domcnt,
396 		    lr->lr_uid, lr->lr_gid);
397 		/*FALLTHROUGH*/
398 	case TX_MKDIR_ACL_ATTR:
399 		if (name == NULL) {
400 			lrattr = (lr_attr_t *)(caddr_t)(lracl + 1);
401 			xvatlen = ZIL_XVAT_SIZE(lrattr->lr_attr_masksize);
402 			zfs_replay_xvattr(lrattr, &xva);
403 		}
404 		vsec.vsa_mask = VSA_ACE | VSA_ACE_ACLFLAGS;
405 		vsec.vsa_aclentp = (caddr_t)(lracl + 1) + xvatlen;
406 		vsec.vsa_aclcnt = lracl->lr_aclcnt;
407 		vsec.vsa_aclentsz = lracl->lr_acl_bytes;
408 		vsec.vsa_aclflags = lracl->lr_acl_flags;
409 		if (zfsvfs->z_fuid_replay == NULL) {
410 			fuidstart = (caddr_t)(lracl + 1) + xvatlen +
411 			    ZIL_ACE_LENGTH(lracl->lr_acl_bytes);
412 			zfsvfs->z_fuid_replay =
413 			    zfs_replay_fuids(fuidstart,
414 			    (void *)&name, lracl->lr_fuidcnt, lracl->lr_domcnt,
415 			    lr->lr_uid, lr->lr_gid);
416 		}
417 		error = zfs_mkdir(dzp, name, &xva.xva_vattr,
418 		    &zp, kcred, vflg, &vsec);
419 		break;
420 	default:
421 		error = SET_ERROR(ENOTSUP);
422 	}
423 
424 bail:
425 	if (error == 0 && zp != NULL) {
426 #ifdef __FreeBSD__
427 		VOP_UNLOCK1(ZTOV(zp));
428 #endif
429 		zrele(zp);
430 	}
431 	zrele(dzp);
432 
433 	if (zfsvfs->z_fuid_replay)
434 		zfs_fuid_info_free(zfsvfs->z_fuid_replay);
435 	zfsvfs->z_fuid_replay = NULL;
436 
437 	return (error);
438 }
439 
440 static int
441 zfs_replay_create(void *arg1, void *arg2, boolean_t byteswap)
442 {
443 	zfsvfs_t *zfsvfs = arg1;
444 	lr_create_t *lr = arg2;
445 	char *name = NULL;		/* location determined later */
446 	char *link;			/* symlink content follows name */
447 	znode_t *dzp;
448 	znode_t *zp = NULL;
449 	xvattr_t xva;
450 	int vflg = 0;
451 	size_t lrsize = sizeof (lr_create_t);
452 	lr_attr_t *lrattr;
453 	void *start;
454 	size_t xvatlen;
455 	uint64_t txtype;
456 	uint64_t objid;
457 	uint64_t dnodesize;
458 	int error;
459 
460 	txtype = (lr->lr_common.lrc_txtype & ~TX_CI);
461 	if (byteswap) {
462 		byteswap_uint64_array(lr, sizeof (*lr));
463 		if (txtype == TX_CREATE_ATTR || txtype == TX_MKDIR_ATTR)
464 			zfs_replay_swap_attrs((lr_attr_t *)(lr + 1));
465 	}
466 
467 
468 	if ((error = zfs_zget(zfsvfs, lr->lr_doid, &dzp)) != 0)
469 		return (error);
470 
471 	objid = LR_FOID_GET_OBJ(lr->lr_foid);
472 	dnodesize = LR_FOID_GET_SLOTS(lr->lr_foid) << DNODE_SHIFT;
473 
474 	xva_init(&xva);
475 	zfs_init_vattr(&xva.xva_vattr, ATTR_MODE | ATTR_UID | ATTR_GID,
476 	    lr->lr_mode, lr->lr_uid, lr->lr_gid, lr->lr_rdev, objid);
477 
478 	/*
479 	 * All forms of zfs create (create, mkdir, mkxattrdir, symlink)
480 	 * eventually end up in zfs_mknode(), which assigns the object's
481 	 * creation time, generation number, and dnode slot count. The
482 	 * generic zfs_create() has no concept of these attributes, so
483 	 * we smuggle the values inside the vattr's otherwise unused
484 	 * va_ctime, va_nblocks, and va_fsid fields.
485 	 */
486 	ZFS_TIME_DECODE(&xva.xva_vattr.va_ctime, lr->lr_crtime);
487 	xva.xva_vattr.va_nblocks = lr->lr_gen;
488 	xva.xva_vattr.va_fsid = dnodesize;
489 
490 	error = dnode_try_claim(zfsvfs->z_os, objid, dnodesize >> DNODE_SHIFT);
491 	if (error)
492 		goto out;
493 
494 	if (lr->lr_common.lrc_txtype & TX_CI)
495 		vflg |= FIGNORECASE;
496 
497 	/*
498 	 * Symlinks don't have fuid info, and CIFS never creates
499 	 * symlinks.
500 	 *
501 	 * The _ATTR versions will grab the fuid info in their subcases.
502 	 */
503 	if ((int)lr->lr_common.lrc_txtype != TX_SYMLINK &&
504 	    (int)lr->lr_common.lrc_txtype != TX_MKDIR_ATTR &&
505 	    (int)lr->lr_common.lrc_txtype != TX_CREATE_ATTR) {
506 		start = (lr + 1);
507 		zfsvfs->z_fuid_replay =
508 		    zfs_replay_fuid_domain(start, &start,
509 		    lr->lr_uid, lr->lr_gid);
510 	}
511 
512 	switch (txtype) {
513 	case TX_CREATE_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_CREATE:
525 		if (name == NULL)
526 			name = (char *)start;
527 
528 		error = zfs_create(dzp, name, &xva.xva_vattr,
529 		    0, 0, &zp, kcred, vflg, NULL);
530 		break;
531 	case TX_MKDIR_ATTR:
532 		lrattr = (lr_attr_t *)(caddr_t)(lr + 1);
533 		xvatlen = ZIL_XVAT_SIZE(lrattr->lr_attr_masksize);
534 		zfs_replay_xvattr((lr_attr_t *)((caddr_t)lr + lrsize), &xva);
535 		start = (caddr_t)(lr + 1) + xvatlen;
536 		zfsvfs->z_fuid_replay =
537 		    zfs_replay_fuid_domain(start, &start,
538 		    lr->lr_uid, lr->lr_gid);
539 		name = (char *)start;
540 
541 		/*FALLTHROUGH*/
542 	case TX_MKDIR:
543 		if (name == NULL)
544 			name = (char *)(lr + 1);
545 
546 		error = zfs_mkdir(dzp, name, &xva.xva_vattr,
547 		    &zp, kcred, vflg, NULL);
548 		break;
549 	case TX_MKXATTR:
550 		error = zfs_make_xattrdir(dzp, &xva.xva_vattr, &zp, kcred);
551 		break;
552 	case TX_SYMLINK:
553 		name = (char *)(lr + 1);
554 		link = name + strlen(name) + 1;
555 		error = zfs_symlink(dzp, name, &xva.xva_vattr,
556 		    link, &zp, kcred, vflg);
557 		break;
558 	default:
559 		error = SET_ERROR(ENOTSUP);
560 	}
561 
562 out:
563 	if (error == 0 && zp != NULL) {
564 #ifdef __FreeBSD__
565 		VOP_UNLOCK1(ZTOV(zp));
566 #endif
567 		zrele(zp);
568 	}
569 	zrele(dzp);
570 
571 	if (zfsvfs->z_fuid_replay)
572 		zfs_fuid_info_free(zfsvfs->z_fuid_replay);
573 	zfsvfs->z_fuid_replay = NULL;
574 	return (error);
575 }
576 
577 static int
578 zfs_replay_remove(void *arg1, void *arg2, boolean_t byteswap)
579 {
580 	zfsvfs_t *zfsvfs = arg1;
581 	lr_remove_t *lr = arg2;
582 	char *name = (char *)(lr + 1);	/* name follows lr_remove_t */
583 	znode_t *dzp;
584 	int error;
585 	int vflg = 0;
586 
587 	if (byteswap)
588 		byteswap_uint64_array(lr, sizeof (*lr));
589 
590 	if ((error = zfs_zget(zfsvfs, lr->lr_doid, &dzp)) != 0)
591 		return (error);
592 
593 	if (lr->lr_common.lrc_txtype & TX_CI)
594 		vflg |= FIGNORECASE;
595 
596 	switch ((int)lr->lr_common.lrc_txtype) {
597 	case TX_REMOVE:
598 		error = zfs_remove(dzp, name, kcred, vflg);
599 		break;
600 	case TX_RMDIR:
601 		error = zfs_rmdir(dzp, name, NULL, kcred, vflg);
602 		break;
603 	default:
604 		error = SET_ERROR(ENOTSUP);
605 	}
606 
607 	zrele(dzp);
608 
609 	return (error);
610 }
611 
612 static int
613 zfs_replay_link(void *arg1, void *arg2, boolean_t byteswap)
614 {
615 	zfsvfs_t *zfsvfs = arg1;
616 	lr_link_t *lr = arg2;
617 	char *name = (char *)(lr + 1);	/* name follows lr_link_t */
618 	znode_t *dzp, *zp;
619 	int error;
620 	int vflg = 0;
621 
622 	if (byteswap)
623 		byteswap_uint64_array(lr, sizeof (*lr));
624 
625 	if ((error = zfs_zget(zfsvfs, lr->lr_doid, &dzp)) != 0)
626 		return (error);
627 
628 	if ((error = zfs_zget(zfsvfs, lr->lr_link_obj, &zp)) != 0) {
629 		zrele(dzp);
630 		return (error);
631 	}
632 
633 	if (lr->lr_common.lrc_txtype & TX_CI)
634 		vflg |= FIGNORECASE;
635 
636 	error = zfs_link(dzp, zp, name, kcred, vflg);
637 	zrele(zp);
638 	zrele(dzp);
639 
640 	return (error);
641 }
642 
643 static int
644 zfs_replay_rename(void *arg1, void *arg2, boolean_t byteswap)
645 {
646 	zfsvfs_t *zfsvfs = arg1;
647 	lr_rename_t *lr = arg2;
648 	char *sname = (char *)(lr + 1);	/* sname and tname follow lr_rename_t */
649 	char *tname = sname + strlen(sname) + 1;
650 	znode_t *sdzp, *tdzp;
651 	int error;
652 	int vflg = 0;
653 
654 	if (byteswap)
655 		byteswap_uint64_array(lr, sizeof (*lr));
656 
657 	if ((error = zfs_zget(zfsvfs, lr->lr_sdoid, &sdzp)) != 0)
658 		return (error);
659 
660 	if ((error = zfs_zget(zfsvfs, lr->lr_tdoid, &tdzp)) != 0) {
661 		zrele(sdzp);
662 		return (error);
663 	}
664 
665 	if (lr->lr_common.lrc_txtype & TX_CI)
666 		vflg |= FIGNORECASE;
667 
668 	error = zfs_rename(sdzp, sname, tdzp, tname, kcred, vflg);
669 
670 	zrele(tdzp);
671 	zrele(sdzp);
672 	return (error);
673 }
674 
675 static int
676 zfs_replay_write(void *arg1, void *arg2, boolean_t byteswap)
677 {
678 	zfsvfs_t *zfsvfs = arg1;
679 	lr_write_t *lr = arg2;
680 	char *data = (char *)(lr + 1);	/* data follows lr_write_t */
681 	znode_t	*zp;
682 	int error;
683 	uint64_t eod, offset, length;
684 
685 	if (byteswap)
686 		byteswap_uint64_array(lr, sizeof (*lr));
687 
688 	if ((error = zfs_zget(zfsvfs, lr->lr_foid, &zp)) != 0) {
689 		/*
690 		 * As we can log writes out of order, it's possible the
691 		 * file has been removed. In this case just drop the write
692 		 * and return success.
693 		 */
694 		if (error == ENOENT)
695 			error = 0;
696 		return (error);
697 	}
698 
699 	offset = lr->lr_offset;
700 	length = lr->lr_length;
701 	eod = offset + length;	/* end of data for this write */
702 
703 	/*
704 	 * This may be a write from a dmu_sync() for a whole block,
705 	 * and may extend beyond the current end of the file.
706 	 * We can't just replay what was written for this TX_WRITE as
707 	 * a future TX_WRITE2 may extend the eof and the data for that
708 	 * write needs to be there. So we write the whole block and
709 	 * reduce the eof. This needs to be done within the single dmu
710 	 * transaction created within vn_rdwr -> zfs_write. So a possible
711 	 * new end of file is passed through in zfsvfs->z_replay_eof
712 	 */
713 
714 	zfsvfs->z_replay_eof = 0; /* 0 means don't change end of file */
715 
716 	/* If it's a dmu_sync() block, write the whole block */
717 	if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
718 		uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
719 		if (length < blocksize) {
720 			offset -= offset % blocksize;
721 			length = blocksize;
722 		}
723 		if (zp->z_size < eod)
724 			zfsvfs->z_replay_eof = eod;
725 	}
726 	error = zfs_write_simple(zp, data, length, offset, NULL);
727 	zrele(zp);
728 	zfsvfs->z_replay_eof = 0;	/* safety */
729 
730 	return (error);
731 }
732 
733 /*
734  * TX_WRITE2 are only generated when dmu_sync() returns EALREADY
735  * meaning the pool block is already being synced. So now that we always write
736  * out full blocks, all we have to do is expand the eof if
737  * the file is grown.
738  */
739 static int
740 zfs_replay_write2(void *arg1, void *arg2, boolean_t byteswap)
741 {
742 	zfsvfs_t *zfsvfs = arg1;
743 	lr_write_t *lr = arg2;
744 	znode_t	*zp;
745 	int error;
746 	uint64_t end;
747 
748 	if (byteswap)
749 		byteswap_uint64_array(lr, sizeof (*lr));
750 
751 	if ((error = zfs_zget(zfsvfs, lr->lr_foid, &zp)) != 0)
752 		return (error);
753 
754 top:
755 	end = lr->lr_offset + lr->lr_length;
756 	if (end > zp->z_size) {
757 		dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
758 
759 		zp->z_size = end;
760 		dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
761 		error = dmu_tx_assign(tx, TXG_WAIT);
762 		if (error) {
763 			zrele(zp);
764 			if (error == ERESTART) {
765 				dmu_tx_wait(tx);
766 				dmu_tx_abort(tx);
767 				goto top;
768 			}
769 			dmu_tx_abort(tx);
770 			return (error);
771 		}
772 		(void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
773 		    (void *)&zp->z_size, sizeof (uint64_t), tx);
774 
775 		/* Ensure the replayed seq is updated */
776 		(void) zil_replaying(zfsvfs->z_log, tx);
777 
778 		dmu_tx_commit(tx);
779 	}
780 
781 	zrele(zp);
782 
783 	return (error);
784 }
785 
786 static int
787 zfs_replay_truncate(void *arg1, void *arg2, boolean_t byteswap)
788 {
789 	zfsvfs_t *zfsvfs = arg1;
790 	lr_truncate_t *lr = arg2;
791 	znode_t *zp;
792 	flock64_t fl;
793 	int error;
794 
795 	if (byteswap)
796 		byteswap_uint64_array(lr, sizeof (*lr));
797 
798 	if ((error = zfs_zget(zfsvfs, lr->lr_foid, &zp)) != 0)
799 		return (error);
800 
801 	bzero(&fl, sizeof (fl));
802 	fl.l_type = F_WRLCK;
803 	fl.l_whence = SEEK_SET;
804 	fl.l_start = lr->lr_offset;
805 	fl.l_len = lr->lr_length;
806 
807 	error = zfs_space(zp, F_FREESP, &fl, O_RDWR | O_LARGEFILE,
808 	    lr->lr_offset, kcred);
809 
810 	zrele(zp);
811 
812 	return (error);
813 }
814 
815 static int
816 zfs_replay_setattr(void *arg1, void *arg2, boolean_t byteswap)
817 {
818 	zfsvfs_t *zfsvfs = arg1;
819 	lr_setattr_t *lr = arg2;
820 	znode_t *zp;
821 	xvattr_t xva;
822 	vattr_t *vap = &xva.xva_vattr;
823 	int error;
824 	void *start;
825 
826 	xva_init(&xva);
827 	if (byteswap) {
828 		byteswap_uint64_array(lr, sizeof (*lr));
829 
830 		if ((lr->lr_mask & ATTR_XVATTR) &&
831 		    zfsvfs->z_version >= ZPL_VERSION_INITIAL)
832 			zfs_replay_swap_attrs((lr_attr_t *)(lr + 1));
833 	}
834 
835 	if ((error = zfs_zget(zfsvfs, lr->lr_foid, &zp)) != 0)
836 		return (error);
837 
838 	zfs_init_vattr(vap, lr->lr_mask, lr->lr_mode,
839 	    lr->lr_uid, lr->lr_gid, 0, lr->lr_foid);
840 
841 	vap->va_size = lr->lr_size;
842 	ZFS_TIME_DECODE(&vap->va_atime, lr->lr_atime);
843 	ZFS_TIME_DECODE(&vap->va_mtime, lr->lr_mtime);
844 	gethrestime(&vap->va_ctime);
845 	vap->va_mask |= ATTR_CTIME;
846 
847 	/*
848 	 * Fill in xvattr_t portions if necessary.
849 	 */
850 
851 	start = (lr_setattr_t *)(lr + 1);
852 	if (vap->va_mask & ATTR_XVATTR) {
853 		zfs_replay_xvattr((lr_attr_t *)start, &xva);
854 		start = (caddr_t)start +
855 		    ZIL_XVAT_SIZE(((lr_attr_t *)start)->lr_attr_masksize);
856 	} else
857 		xva.xva_vattr.va_mask &= ~ATTR_XVATTR;
858 
859 	zfsvfs->z_fuid_replay = zfs_replay_fuid_domain(start, &start,
860 	    lr->lr_uid, lr->lr_gid);
861 
862 	error = zfs_setattr(zp, vap, 0, kcred);
863 
864 	zfs_fuid_info_free(zfsvfs->z_fuid_replay);
865 	zfsvfs->z_fuid_replay = NULL;
866 	zrele(zp);
867 
868 	return (error);
869 }
870 
871 static int
872 zfs_replay_acl_v0(void *arg1, void *arg2, boolean_t byteswap)
873 {
874 	zfsvfs_t *zfsvfs = arg1;
875 	lr_acl_v0_t *lr = arg2;
876 	ace_t *ace = (ace_t *)(lr + 1);	/* ace array follows lr_acl_t */
877 	vsecattr_t vsa;
878 	znode_t *zp;
879 	int error;
880 
881 	if (byteswap) {
882 		byteswap_uint64_array(lr, sizeof (*lr));
883 		zfs_oldace_byteswap(ace, lr->lr_aclcnt);
884 	}
885 
886 	if ((error = zfs_zget(zfsvfs, lr->lr_foid, &zp)) != 0)
887 		return (error);
888 
889 	bzero(&vsa, sizeof (vsa));
890 	vsa.vsa_mask = VSA_ACE | VSA_ACECNT;
891 	vsa.vsa_aclcnt = lr->lr_aclcnt;
892 	vsa.vsa_aclentsz = sizeof (ace_t) * vsa.vsa_aclcnt;
893 	vsa.vsa_aclflags = 0;
894 	vsa.vsa_aclentp = ace;
895 
896 	error = zfs_setsecattr(zp, &vsa, 0, kcred);
897 
898 	zrele(zp);
899 
900 	return (error);
901 }
902 
903 /*
904  * Replaying ACLs is complicated by FUID support.
905  * The log record may contain some optional data
906  * to be used for replaying FUID's.  These pieces
907  * are the actual FUIDs that were created initially.
908  * The FUID table index may no longer be valid and
909  * during zfs_create() a new index may be assigned.
910  * Because of this the log will contain the original
911  * domain+rid in order to create a new FUID.
912  *
913  * The individual ACEs may contain an ephemeral uid/gid which is no
914  * longer valid and will need to be replaced with an actual FUID.
915  *
916  */
917 static int
918 zfs_replay_acl(void *arg1, void *arg2, boolean_t byteswap)
919 {
920 	zfsvfs_t *zfsvfs = arg1;
921 	lr_acl_t *lr = arg2;
922 	ace_t *ace = (ace_t *)(lr + 1);
923 	vsecattr_t vsa;
924 	znode_t *zp;
925 	int error;
926 
927 	if (byteswap) {
928 		byteswap_uint64_array(lr, sizeof (*lr));
929 		zfs_ace_byteswap(ace, lr->lr_acl_bytes, B_FALSE);
930 		if (lr->lr_fuidcnt) {
931 			byteswap_uint64_array((caddr_t)ace +
932 			    ZIL_ACE_LENGTH(lr->lr_acl_bytes),
933 			    lr->lr_fuidcnt * sizeof (uint64_t));
934 		}
935 	}
936 
937 	if ((error = zfs_zget(zfsvfs, lr->lr_foid, &zp)) != 0)
938 		return (error);
939 
940 	bzero(&vsa, sizeof (vsa));
941 	vsa.vsa_mask = VSA_ACE | VSA_ACECNT | VSA_ACE_ACLFLAGS;
942 	vsa.vsa_aclcnt = lr->lr_aclcnt;
943 	vsa.vsa_aclentp = ace;
944 	vsa.vsa_aclentsz = lr->lr_acl_bytes;
945 	vsa.vsa_aclflags = lr->lr_acl_flags;
946 
947 	if (lr->lr_fuidcnt) {
948 		void *fuidstart = (caddr_t)ace +
949 		    ZIL_ACE_LENGTH(lr->lr_acl_bytes);
950 
951 		zfsvfs->z_fuid_replay =
952 		    zfs_replay_fuids(fuidstart, &fuidstart,
953 		    lr->lr_fuidcnt, lr->lr_domcnt, 0, 0);
954 	}
955 
956 	error = zfs_setsecattr(zp, &vsa, 0, kcred);
957 
958 	if (zfsvfs->z_fuid_replay)
959 		zfs_fuid_info_free(zfsvfs->z_fuid_replay);
960 
961 	zfsvfs->z_fuid_replay = NULL;
962 	zrele(zp);
963 
964 	return (error);
965 }
966 
967 /*
968  * Callback vectors for replaying records
969  */
970 zil_replay_func_t *zfs_replay_vector[TX_MAX_TYPE] = {
971 	zfs_replay_error,	/* no such type */
972 	zfs_replay_create,	/* TX_CREATE */
973 	zfs_replay_create,	/* TX_MKDIR */
974 	zfs_replay_create,	/* TX_MKXATTR */
975 	zfs_replay_create,	/* TX_SYMLINK */
976 	zfs_replay_remove,	/* TX_REMOVE */
977 	zfs_replay_remove,	/* TX_RMDIR */
978 	zfs_replay_link,	/* TX_LINK */
979 	zfs_replay_rename,	/* TX_RENAME */
980 	zfs_replay_write,	/* TX_WRITE */
981 	zfs_replay_truncate,	/* TX_TRUNCATE */
982 	zfs_replay_setattr,	/* TX_SETATTR */
983 	zfs_replay_acl_v0,	/* TX_ACL_V0 */
984 	zfs_replay_acl,		/* TX_ACL */
985 	zfs_replay_create_acl,	/* TX_CREATE_ACL */
986 	zfs_replay_create,	/* TX_CREATE_ATTR */
987 	zfs_replay_create_acl,	/* TX_CREATE_ACL_ATTR */
988 	zfs_replay_create_acl,	/* TX_MKDIR_ACL */
989 	zfs_replay_create,	/* TX_MKDIR_ATTR */
990 	zfs_replay_create_acl,	/* TX_MKDIR_ACL_ATTR */
991 	zfs_replay_write2,	/* TX_WRITE2 */
992 };
993