xref: /freebsd/sys/contrib/openzfs/module/zfs/zfs_replay.c (revision 9729f076e4d93c5a37e78d427bfe0f1ab99bbcc6)
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 https://opensource.org/licenses/CDDL-1.0.
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 #include <sys/dmu_objset.h>
51 #include <sys/zfeature.h>
52 
53 /*
54  * NB: FreeBSD expects to be able to do vnode locking in lookup and
55  * hold the locks across all subsequent VOPs until vput is called.
56  * This means that its zfs vnops routines can't do any internal locking.
57  * In order to have the same contract as the Linux vnops there would
58  * needed to be duplicate locked vnops. If the vnops were used more widely
59  * in common code this would likely be preferable. However, currently
60  * this is the only file where this is the case.
61  */
62 
63 /*
64  * Functions to replay ZFS intent log (ZIL) records
65  * The functions are called through a function vector (zfs_replay_vector)
66  * which is indexed by the transaction type.
67  */
68 
69 static void
70 zfs_init_vattr(vattr_t *vap, uint64_t mask, uint64_t mode,
71     uint64_t uid, uint64_t gid, uint64_t rdev, uint64_t nodeid)
72 {
73 	memset(vap, 0, sizeof (*vap));
74 	vap->va_mask = (uint_t)mask;
75 	vap->va_mode = mode;
76 #if defined(__FreeBSD__) || defined(__APPLE__)
77 	vap->va_type = IFTOVT(mode);
78 #endif
79 	vap->va_uid = (uid_t)(IS_EPHEMERAL(uid)) ? -1 : uid;
80 	vap->va_gid = (gid_t)(IS_EPHEMERAL(gid)) ? -1 : gid;
81 	vap->va_rdev = zfs_cmpldev(rdev);
82 	vap->va_nodeid = nodeid;
83 }
84 
85 static int
86 zfs_replay_error(void *arg1, void *arg2, boolean_t byteswap)
87 {
88 	(void) arg1, (void) arg2, (void) byteswap;
89 	return (SET_ERROR(ENOTSUP));
90 }
91 
92 static void
93 zfs_replay_xvattr(lr_attr_t *lrattr, xvattr_t *xvap)
94 {
95 	xoptattr_t *xoap = NULL;
96 	uint64_t *attrs;
97 	uint64_t *crtime;
98 	uint32_t *bitmap;
99 	void *scanstamp;
100 	int i;
101 
102 	xvap->xva_vattr.va_mask |= ATTR_XVATTR;
103 	if ((xoap = xva_getxoptattr(xvap)) == NULL) {
104 		xvap->xva_vattr.va_mask &= ~ATTR_XVATTR; /* shouldn't happen */
105 		return;
106 	}
107 
108 	ASSERT(lrattr->lr_attr_masksize == xvap->xva_mapsize);
109 
110 	bitmap = &lrattr->lr_attr_bitmap;
111 	for (i = 0; i != lrattr->lr_attr_masksize; i++, bitmap++)
112 		xvap->xva_reqattrmap[i] = *bitmap;
113 
114 	attrs = (uint64_t *)(lrattr + lrattr->lr_attr_masksize - 1);
115 	crtime = attrs + 1;
116 	scanstamp = (caddr_t)(crtime + 2);
117 
118 	if (XVA_ISSET_REQ(xvap, XAT_HIDDEN))
119 		xoap->xoa_hidden = ((*attrs & XAT0_HIDDEN) != 0);
120 	if (XVA_ISSET_REQ(xvap, XAT_SYSTEM))
121 		xoap->xoa_system = ((*attrs & XAT0_SYSTEM) != 0);
122 	if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE))
123 		xoap->xoa_archive = ((*attrs & XAT0_ARCHIVE) != 0);
124 	if (XVA_ISSET_REQ(xvap, XAT_READONLY))
125 		xoap->xoa_readonly = ((*attrs & XAT0_READONLY) != 0);
126 	if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE))
127 		xoap->xoa_immutable = ((*attrs & XAT0_IMMUTABLE) != 0);
128 	if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK))
129 		xoap->xoa_nounlink = ((*attrs & XAT0_NOUNLINK) != 0);
130 	if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY))
131 		xoap->xoa_appendonly = ((*attrs & XAT0_APPENDONLY) != 0);
132 	if (XVA_ISSET_REQ(xvap, XAT_NODUMP))
133 		xoap->xoa_nodump = ((*attrs & XAT0_NODUMP) != 0);
134 	if (XVA_ISSET_REQ(xvap, XAT_OPAQUE))
135 		xoap->xoa_opaque = ((*attrs & XAT0_OPAQUE) != 0);
136 	if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED))
137 		xoap->xoa_av_modified = ((*attrs & XAT0_AV_MODIFIED) != 0);
138 	if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED))
139 		xoap->xoa_av_quarantined =
140 		    ((*attrs & XAT0_AV_QUARANTINED) != 0);
141 	if (XVA_ISSET_REQ(xvap, XAT_CREATETIME))
142 		ZFS_TIME_DECODE(&xoap->xoa_createtime, crtime);
143 	if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) {
144 		ASSERT(!XVA_ISSET_REQ(xvap, XAT_PROJID));
145 
146 		memcpy(xoap->xoa_av_scanstamp, scanstamp, AV_SCANSTAMP_SZ);
147 	} else if (XVA_ISSET_REQ(xvap, XAT_PROJID)) {
148 		/*
149 		 * XAT_PROJID and XAT_AV_SCANSTAMP will never be valid
150 		 * at the same time, so we can share the same space.
151 		 */
152 		memcpy(&xoap->xoa_projid, scanstamp, sizeof (uint64_t));
153 	}
154 	if (XVA_ISSET_REQ(xvap, XAT_REPARSE))
155 		xoap->xoa_reparse = ((*attrs & XAT0_REPARSE) != 0);
156 	if (XVA_ISSET_REQ(xvap, XAT_OFFLINE))
157 		xoap->xoa_offline = ((*attrs & XAT0_OFFLINE) != 0);
158 	if (XVA_ISSET_REQ(xvap, XAT_SPARSE))
159 		xoap->xoa_sparse = ((*attrs & XAT0_SPARSE) != 0);
160 	if (XVA_ISSET_REQ(xvap, XAT_PROJINHERIT))
161 		xoap->xoa_projinherit = ((*attrs & XAT0_PROJINHERIT) != 0);
162 }
163 
164 static int
165 zfs_replay_domain_cnt(uint64_t uid, uint64_t gid)
166 {
167 	uint64_t uid_idx;
168 	uint64_t gid_idx;
169 	int domcnt = 0;
170 
171 	uid_idx = FUID_INDEX(uid);
172 	gid_idx = FUID_INDEX(gid);
173 	if (uid_idx)
174 		domcnt++;
175 	if (gid_idx > 0 && gid_idx != uid_idx)
176 		domcnt++;
177 
178 	return (domcnt);
179 }
180 
181 static void *
182 zfs_replay_fuid_domain_common(zfs_fuid_info_t *fuid_infop, void *start,
183     int domcnt)
184 {
185 	int i;
186 
187 	for (i = 0; i != domcnt; i++) {
188 		fuid_infop->z_domain_table[i] = start;
189 		start = (caddr_t)start + strlen(start) + 1;
190 	}
191 
192 	return (start);
193 }
194 
195 /*
196  * Set the uid/gid in the fuid_info structure.
197  */
198 static void
199 zfs_replay_fuid_ugid(zfs_fuid_info_t *fuid_infop, uint64_t uid, uint64_t gid)
200 {
201 	/*
202 	 * If owner or group are log specific FUIDs then slurp up
203 	 * domain information and build zfs_fuid_info_t
204 	 */
205 	if (IS_EPHEMERAL(uid))
206 		fuid_infop->z_fuid_owner = uid;
207 
208 	if (IS_EPHEMERAL(gid))
209 		fuid_infop->z_fuid_group = gid;
210 }
211 
212 /*
213  * Load fuid domains into fuid_info_t
214  */
215 static zfs_fuid_info_t *
216 zfs_replay_fuid_domain(void *buf, void **end, uint64_t uid, uint64_t gid)
217 {
218 	int domcnt;
219 
220 	zfs_fuid_info_t *fuid_infop;
221 
222 	fuid_infop = zfs_fuid_info_alloc();
223 
224 	domcnt = zfs_replay_domain_cnt(uid, gid);
225 
226 	if (domcnt == 0)
227 		return (fuid_infop);
228 
229 	fuid_infop->z_domain_table =
230 	    kmem_zalloc(domcnt * sizeof (char *), KM_SLEEP);
231 
232 	zfs_replay_fuid_ugid(fuid_infop, uid, gid);
233 
234 	fuid_infop->z_domain_cnt = domcnt;
235 	*end = zfs_replay_fuid_domain_common(fuid_infop, buf, domcnt);
236 	return (fuid_infop);
237 }
238 
239 /*
240  * load zfs_fuid_t's and fuid_domains into fuid_info_t
241  */
242 static zfs_fuid_info_t *
243 zfs_replay_fuids(void *start, void **end, int idcnt, int domcnt, uint64_t uid,
244     uint64_t gid)
245 {
246 	uint64_t *log_fuid = (uint64_t *)start;
247 	zfs_fuid_info_t *fuid_infop;
248 	int i;
249 
250 	fuid_infop = zfs_fuid_info_alloc();
251 	fuid_infop->z_domain_cnt = domcnt;
252 
253 	fuid_infop->z_domain_table =
254 	    kmem_zalloc(domcnt * sizeof (char *), KM_SLEEP);
255 
256 	for (i = 0; i != idcnt; i++) {
257 		zfs_fuid_t *zfuid;
258 
259 		zfuid = kmem_alloc(sizeof (zfs_fuid_t), KM_SLEEP);
260 		zfuid->z_logfuid = *log_fuid;
261 		zfuid->z_id = -1;
262 		zfuid->z_domidx = 0;
263 		list_insert_tail(&fuid_infop->z_fuids, zfuid);
264 		log_fuid++;
265 	}
266 
267 	zfs_replay_fuid_ugid(fuid_infop, uid, gid);
268 
269 	*end = zfs_replay_fuid_domain_common(fuid_infop, log_fuid, domcnt);
270 	return (fuid_infop);
271 }
272 
273 static void
274 zfs_replay_swap_attrs(lr_attr_t *lrattr)
275 {
276 	/* swap the lr_attr structure */
277 	byteswap_uint32_array(lrattr, sizeof (*lrattr));
278 	/* swap the bitmap */
279 	byteswap_uint32_array(lrattr + 1, (lrattr->lr_attr_masksize - 1) *
280 	    sizeof (uint32_t));
281 	/* swap the attributes, create time + 64 bit word for attributes */
282 	byteswap_uint64_array((caddr_t)(lrattr + 1) + (sizeof (uint32_t) *
283 	    (lrattr->lr_attr_masksize - 1)), 3 * sizeof (uint64_t));
284 }
285 
286 /*
287  * Replay file create with optional ACL, xvattr information as well
288  * as option FUID information.
289  */
290 static int
291 zfs_replay_create_acl(void *arg1, void *arg2, boolean_t byteswap)
292 {
293 	zfsvfs_t *zfsvfs = arg1;
294 	lr_acl_create_t *lracl = arg2;
295 	char *name = NULL;		/* location determined later */
296 	lr_create_t *lr = (lr_create_t *)lracl;
297 	znode_t *dzp;
298 	znode_t *zp;
299 	xvattr_t xva;
300 	int vflg = 0;
301 	vsecattr_t vsec = { 0 };
302 	lr_attr_t *lrattr;
303 	void *aclstart;
304 	void *fuidstart;
305 	size_t xvatlen = 0;
306 	uint64_t txtype;
307 	uint64_t objid;
308 	uint64_t dnodesize;
309 	int error;
310 
311 	txtype = (lr->lr_common.lrc_txtype & ~TX_CI);
312 	if (byteswap) {
313 		byteswap_uint64_array(lracl, sizeof (*lracl));
314 		if (txtype == TX_CREATE_ACL_ATTR ||
315 		    txtype == TX_MKDIR_ACL_ATTR) {
316 			lrattr = (lr_attr_t *)(caddr_t)(lracl + 1);
317 			zfs_replay_swap_attrs(lrattr);
318 			xvatlen = ZIL_XVAT_SIZE(lrattr->lr_attr_masksize);
319 		}
320 
321 		aclstart = (caddr_t)(lracl + 1) + xvatlen;
322 		zfs_ace_byteswap(aclstart, lracl->lr_acl_bytes, B_FALSE);
323 		/* swap fuids */
324 		if (lracl->lr_fuidcnt) {
325 			byteswap_uint64_array((caddr_t)aclstart +
326 			    ZIL_ACE_LENGTH(lracl->lr_acl_bytes),
327 			    lracl->lr_fuidcnt * sizeof (uint64_t));
328 		}
329 	}
330 
331 	if ((error = zfs_zget(zfsvfs, lr->lr_doid, &dzp)) != 0)
332 		return (error);
333 
334 	objid = LR_FOID_GET_OBJ(lr->lr_foid);
335 	dnodesize = LR_FOID_GET_SLOTS(lr->lr_foid) << DNODE_SHIFT;
336 
337 	xva_init(&xva);
338 	zfs_init_vattr(&xva.xva_vattr, ATTR_MODE | ATTR_UID | ATTR_GID,
339 	    lr->lr_mode, lr->lr_uid, lr->lr_gid, lr->lr_rdev, objid);
340 
341 	/*
342 	 * All forms of zfs create (create, mkdir, mkxattrdir, symlink)
343 	 * eventually end up in zfs_mknode(), which assigns the object's
344 	 * creation time, generation number, and dnode size. The generic
345 	 * zfs_create() has no concept of these attributes, so we smuggle
346 	 * the values inside the vattr's otherwise unused va_ctime,
347 	 * va_nblocks, and va_fsid fields.
348 	 */
349 	ZFS_TIME_DECODE(&xva.xva_vattr.va_ctime, lr->lr_crtime);
350 	xva.xva_vattr.va_nblocks = lr->lr_gen;
351 	xva.xva_vattr.va_fsid = dnodesize;
352 
353 	error = dnode_try_claim(zfsvfs->z_os, objid, dnodesize >> DNODE_SHIFT);
354 	if (error)
355 		goto bail;
356 
357 	if (lr->lr_common.lrc_txtype & TX_CI)
358 		vflg |= FIGNORECASE;
359 	switch (txtype) {
360 	case TX_CREATE_ACL:
361 		aclstart = (caddr_t)(lracl + 1);
362 		fuidstart = (caddr_t)aclstart +
363 		    ZIL_ACE_LENGTH(lracl->lr_acl_bytes);
364 		zfsvfs->z_fuid_replay = zfs_replay_fuids(fuidstart,
365 		    (void *)&name, lracl->lr_fuidcnt, lracl->lr_domcnt,
366 		    lr->lr_uid, lr->lr_gid);
367 		zfs_fallthrough;
368 	case TX_CREATE_ACL_ATTR:
369 		if (name == NULL) {
370 			lrattr = (lr_attr_t *)(caddr_t)(lracl + 1);
371 			xvatlen = ZIL_XVAT_SIZE(lrattr->lr_attr_masksize);
372 			xva.xva_vattr.va_mask |= ATTR_XVATTR;
373 			zfs_replay_xvattr(lrattr, &xva);
374 		}
375 		vsec.vsa_mask = VSA_ACE | VSA_ACE_ACLFLAGS;
376 		vsec.vsa_aclentp = (caddr_t)(lracl + 1) + xvatlen;
377 		vsec.vsa_aclcnt = lracl->lr_aclcnt;
378 		vsec.vsa_aclentsz = lracl->lr_acl_bytes;
379 		vsec.vsa_aclflags = lracl->lr_acl_flags;
380 		if (zfsvfs->z_fuid_replay == NULL) {
381 			fuidstart = (caddr_t)(lracl + 1) + xvatlen +
382 			    ZIL_ACE_LENGTH(lracl->lr_acl_bytes);
383 			zfsvfs->z_fuid_replay =
384 			    zfs_replay_fuids(fuidstart,
385 			    (void *)&name, lracl->lr_fuidcnt, lracl->lr_domcnt,
386 			    lr->lr_uid, lr->lr_gid);
387 		}
388 
389 		error = zfs_create(dzp, name, &xva.xva_vattr,
390 		    0, 0, &zp, kcred, vflg, &vsec);
391 		break;
392 	case TX_MKDIR_ACL:
393 		aclstart = (caddr_t)(lracl + 1);
394 		fuidstart = (caddr_t)aclstart +
395 		    ZIL_ACE_LENGTH(lracl->lr_acl_bytes);
396 		zfsvfs->z_fuid_replay = zfs_replay_fuids(fuidstart,
397 		    (void *)&name, lracl->lr_fuidcnt, lracl->lr_domcnt,
398 		    lr->lr_uid, lr->lr_gid);
399 		zfs_fallthrough;
400 	case TX_MKDIR_ACL_ATTR:
401 		if (name == NULL) {
402 			lrattr = (lr_attr_t *)(caddr_t)(lracl + 1);
403 			xvatlen = ZIL_XVAT_SIZE(lrattr->lr_attr_masksize);
404 			zfs_replay_xvattr(lrattr, &xva);
405 		}
406 		vsec.vsa_mask = VSA_ACE | VSA_ACE_ACLFLAGS;
407 		vsec.vsa_aclentp = (caddr_t)(lracl + 1) + xvatlen;
408 		vsec.vsa_aclcnt = lracl->lr_aclcnt;
409 		vsec.vsa_aclentsz = lracl->lr_acl_bytes;
410 		vsec.vsa_aclflags = lracl->lr_acl_flags;
411 		if (zfsvfs->z_fuid_replay == NULL) {
412 			fuidstart = (caddr_t)(lracl + 1) + xvatlen +
413 			    ZIL_ACE_LENGTH(lracl->lr_acl_bytes);
414 			zfsvfs->z_fuid_replay =
415 			    zfs_replay_fuids(fuidstart,
416 			    (void *)&name, lracl->lr_fuidcnt, lracl->lr_domcnt,
417 			    lr->lr_uid, lr->lr_gid);
418 		}
419 		error = zfs_mkdir(dzp, name, &xva.xva_vattr,
420 		    &zp, kcred, vflg, &vsec);
421 		break;
422 	default:
423 		error = SET_ERROR(ENOTSUP);
424 	}
425 
426 bail:
427 	if (error == 0 && zp != NULL) {
428 #ifdef __FreeBSD__
429 		VOP_UNLOCK1(ZTOV(zp));
430 #endif
431 		zrele(zp);
432 	}
433 	zrele(dzp);
434 
435 	if (zfsvfs->z_fuid_replay)
436 		zfs_fuid_info_free(zfsvfs->z_fuid_replay);
437 	zfsvfs->z_fuid_replay = NULL;
438 
439 	return (error);
440 }
441 
442 static int
443 zfs_replay_create(void *arg1, void *arg2, boolean_t byteswap)
444 {
445 	zfsvfs_t *zfsvfs = arg1;
446 	lr_create_t *lr = arg2;
447 	char *name = NULL;		/* location determined later */
448 	char *link;			/* symlink content follows name */
449 	znode_t *dzp;
450 	znode_t *zp = NULL;
451 	xvattr_t xva;
452 	int vflg = 0;
453 	size_t lrsize = sizeof (lr_create_t);
454 	lr_attr_t *lrattr;
455 	void *start;
456 	size_t xvatlen;
457 	uint64_t txtype;
458 	uint64_t objid;
459 	uint64_t dnodesize;
460 	int error;
461 
462 	txtype = (lr->lr_common.lrc_txtype & ~TX_CI);
463 	if (byteswap) {
464 		byteswap_uint64_array(lr, sizeof (*lr));
465 		if (txtype == TX_CREATE_ATTR || txtype == TX_MKDIR_ATTR)
466 			zfs_replay_swap_attrs((lr_attr_t *)(lr + 1));
467 	}
468 
469 
470 	if ((error = zfs_zget(zfsvfs, lr->lr_doid, &dzp)) != 0)
471 		return (error);
472 
473 	objid = LR_FOID_GET_OBJ(lr->lr_foid);
474 	dnodesize = LR_FOID_GET_SLOTS(lr->lr_foid) << DNODE_SHIFT;
475 
476 	xva_init(&xva);
477 	zfs_init_vattr(&xva.xva_vattr, ATTR_MODE | ATTR_UID | ATTR_GID,
478 	    lr->lr_mode, lr->lr_uid, lr->lr_gid, lr->lr_rdev, objid);
479 
480 	/*
481 	 * All forms of zfs create (create, mkdir, mkxattrdir, symlink)
482 	 * eventually end up in zfs_mknode(), which assigns the object's
483 	 * creation time, generation number, and dnode slot count. The
484 	 * generic zfs_create() has no concept of these attributes, so
485 	 * we smuggle the values inside the vattr's otherwise unused
486 	 * va_ctime, va_nblocks, and va_fsid fields.
487 	 */
488 	ZFS_TIME_DECODE(&xva.xva_vattr.va_ctime, lr->lr_crtime);
489 	xva.xva_vattr.va_nblocks = lr->lr_gen;
490 	xva.xva_vattr.va_fsid = dnodesize;
491 
492 	error = dnode_try_claim(zfsvfs->z_os, objid, dnodesize >> DNODE_SHIFT);
493 	if (error)
494 		goto out;
495 
496 	if (lr->lr_common.lrc_txtype & TX_CI)
497 		vflg |= FIGNORECASE;
498 
499 	/*
500 	 * Symlinks don't have fuid info, and CIFS never creates
501 	 * symlinks.
502 	 *
503 	 * The _ATTR versions will grab the fuid info in their subcases.
504 	 */
505 	if ((int)lr->lr_common.lrc_txtype != TX_SYMLINK &&
506 	    (int)lr->lr_common.lrc_txtype != TX_MKDIR_ATTR &&
507 	    (int)lr->lr_common.lrc_txtype != TX_CREATE_ATTR) {
508 		start = (lr + 1);
509 		zfsvfs->z_fuid_replay =
510 		    zfs_replay_fuid_domain(start, &start,
511 		    lr->lr_uid, lr->lr_gid);
512 	}
513 
514 	switch (txtype) {
515 	case TX_CREATE_ATTR:
516 		lrattr = (lr_attr_t *)(caddr_t)(lr + 1);
517 		xvatlen = ZIL_XVAT_SIZE(lrattr->lr_attr_masksize);
518 		zfs_replay_xvattr((lr_attr_t *)((caddr_t)lr + lrsize), &xva);
519 		start = (caddr_t)(lr + 1) + xvatlen;
520 		zfsvfs->z_fuid_replay =
521 		    zfs_replay_fuid_domain(start, &start,
522 		    lr->lr_uid, lr->lr_gid);
523 		name = (char *)start;
524 		zfs_fallthrough;
525 
526 	case TX_CREATE:
527 		if (name == NULL)
528 			name = (char *)start;
529 
530 		error = zfs_create(dzp, name, &xva.xva_vattr,
531 		    0, 0, &zp, kcred, vflg, NULL);
532 		break;
533 	case TX_MKDIR_ATTR:
534 		lrattr = (lr_attr_t *)(caddr_t)(lr + 1);
535 		xvatlen = ZIL_XVAT_SIZE(lrattr->lr_attr_masksize);
536 		zfs_replay_xvattr((lr_attr_t *)((caddr_t)lr + lrsize), &xva);
537 		start = (caddr_t)(lr + 1) + xvatlen;
538 		zfsvfs->z_fuid_replay =
539 		    zfs_replay_fuid_domain(start, &start,
540 		    lr->lr_uid, lr->lr_gid);
541 		name = (char *)start;
542 		zfs_fallthrough;
543 
544 	case TX_MKDIR:
545 		if (name == NULL)
546 			name = (char *)(lr + 1);
547 
548 		error = zfs_mkdir(dzp, name, &xva.xva_vattr,
549 		    &zp, kcred, vflg, NULL);
550 		break;
551 	case TX_MKXATTR:
552 		error = zfs_make_xattrdir(dzp, &xva.xva_vattr, &zp, kcred);
553 		break;
554 	case TX_SYMLINK:
555 		name = (char *)(lr + 1);
556 		link = name + strlen(name) + 1;
557 		error = zfs_symlink(dzp, name, &xva.xva_vattr,
558 		    link, &zp, kcred, vflg);
559 		break;
560 	default:
561 		error = SET_ERROR(ENOTSUP);
562 	}
563 
564 out:
565 	if (error == 0 && zp != NULL) {
566 #ifdef __FreeBSD__
567 		VOP_UNLOCK1(ZTOV(zp));
568 #endif
569 		zrele(zp);
570 	}
571 	zrele(dzp);
572 
573 	if (zfsvfs->z_fuid_replay)
574 		zfs_fuid_info_free(zfsvfs->z_fuid_replay);
575 	zfsvfs->z_fuid_replay = NULL;
576 	return (error);
577 }
578 
579 static int
580 zfs_replay_remove(void *arg1, void *arg2, boolean_t byteswap)
581 {
582 	zfsvfs_t *zfsvfs = arg1;
583 	lr_remove_t *lr = arg2;
584 	char *name = (char *)(lr + 1);	/* name follows lr_remove_t */
585 	znode_t *dzp;
586 	int error;
587 	int vflg = 0;
588 
589 	if (byteswap)
590 		byteswap_uint64_array(lr, sizeof (*lr));
591 
592 	if ((error = zfs_zget(zfsvfs, lr->lr_doid, &dzp)) != 0)
593 		return (error);
594 
595 	if (lr->lr_common.lrc_txtype & TX_CI)
596 		vflg |= FIGNORECASE;
597 
598 	switch ((int)lr->lr_common.lrc_txtype) {
599 	case TX_REMOVE:
600 		error = zfs_remove(dzp, name, kcred, vflg);
601 		break;
602 	case TX_RMDIR:
603 		error = zfs_rmdir(dzp, name, NULL, kcred, vflg);
604 		break;
605 	default:
606 		error = SET_ERROR(ENOTSUP);
607 	}
608 
609 	zrele(dzp);
610 
611 	return (error);
612 }
613 
614 static int
615 zfs_replay_link(void *arg1, void *arg2, boolean_t byteswap)
616 {
617 	zfsvfs_t *zfsvfs = arg1;
618 	lr_link_t *lr = arg2;
619 	char *name = (char *)(lr + 1);	/* name follows lr_link_t */
620 	znode_t *dzp, *zp;
621 	int error;
622 	int vflg = 0;
623 
624 	if (byteswap)
625 		byteswap_uint64_array(lr, sizeof (*lr));
626 
627 	if ((error = zfs_zget(zfsvfs, lr->lr_doid, &dzp)) != 0)
628 		return (error);
629 
630 	if ((error = zfs_zget(zfsvfs, lr->lr_link_obj, &zp)) != 0) {
631 		zrele(dzp);
632 		return (error);
633 	}
634 
635 	if (lr->lr_common.lrc_txtype & TX_CI)
636 		vflg |= FIGNORECASE;
637 
638 	error = zfs_link(dzp, zp, name, kcred, vflg);
639 	zrele(zp);
640 	zrele(dzp);
641 
642 	return (error);
643 }
644 
645 static int
646 zfs_replay_rename(void *arg1, void *arg2, boolean_t byteswap)
647 {
648 	zfsvfs_t *zfsvfs = arg1;
649 	lr_rename_t *lr = arg2;
650 	char *sname = (char *)(lr + 1);	/* sname and tname follow lr_rename_t */
651 	char *tname = sname + strlen(sname) + 1;
652 	znode_t *sdzp, *tdzp;
653 	int error;
654 	int vflg = 0;
655 
656 	if (byteswap)
657 		byteswap_uint64_array(lr, sizeof (*lr));
658 
659 	if ((error = zfs_zget(zfsvfs, lr->lr_sdoid, &sdzp)) != 0)
660 		return (error);
661 
662 	if ((error = zfs_zget(zfsvfs, lr->lr_tdoid, &tdzp)) != 0) {
663 		zrele(sdzp);
664 		return (error);
665 	}
666 
667 	if (lr->lr_common.lrc_txtype & TX_CI)
668 		vflg |= FIGNORECASE;
669 
670 	error = zfs_rename(sdzp, sname, tdzp, tname, kcred, vflg);
671 
672 	zrele(tdzp);
673 	zrele(sdzp);
674 	return (error);
675 }
676 
677 static int
678 zfs_replay_write(void *arg1, void *arg2, boolean_t byteswap)
679 {
680 	zfsvfs_t *zfsvfs = arg1;
681 	lr_write_t *lr = arg2;
682 	char *data = (char *)(lr + 1);	/* data follows lr_write_t */
683 	znode_t	*zp;
684 	int error;
685 	uint64_t eod, offset, length;
686 
687 	if (byteswap)
688 		byteswap_uint64_array(lr, sizeof (*lr));
689 
690 	if ((error = zfs_zget(zfsvfs, lr->lr_foid, &zp)) != 0) {
691 		/*
692 		 * As we can log writes out of order, it's possible the
693 		 * file has been removed. In this case just drop the write
694 		 * and return success.
695 		 */
696 		if (error == ENOENT)
697 			error = 0;
698 		return (error);
699 	}
700 
701 	offset = lr->lr_offset;
702 	length = lr->lr_length;
703 	eod = offset + length;	/* end of data for this write */
704 
705 	/*
706 	 * This may be a write from a dmu_sync() for a whole block,
707 	 * and may extend beyond the current end of the file.
708 	 * We can't just replay what was written for this TX_WRITE as
709 	 * a future TX_WRITE2 may extend the eof and the data for that
710 	 * write needs to be there. So we write the whole block and
711 	 * reduce the eof. This needs to be done within the single dmu
712 	 * transaction created within vn_rdwr -> zfs_write. So a possible
713 	 * new end of file is passed through in zfsvfs->z_replay_eof
714 	 */
715 
716 	zfsvfs->z_replay_eof = 0; /* 0 means don't change end of file */
717 
718 	/* If it's a dmu_sync() block, write the whole block */
719 	if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
720 		uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
721 		if (length < blocksize) {
722 			offset -= offset % blocksize;
723 			length = blocksize;
724 		}
725 		if (zp->z_size < eod)
726 			zfsvfs->z_replay_eof = eod;
727 	}
728 	error = zfs_write_simple(zp, data, length, offset, NULL);
729 	zrele(zp);
730 	zfsvfs->z_replay_eof = 0;	/* safety */
731 
732 	return (error);
733 }
734 
735 /*
736  * TX_WRITE2 are only generated when dmu_sync() returns EALREADY
737  * meaning the pool block is already being synced. So now that we always write
738  * out full blocks, all we have to do is expand the eof if
739  * the file is grown.
740  */
741 static int
742 zfs_replay_write2(void *arg1, void *arg2, boolean_t byteswap)
743 {
744 	zfsvfs_t *zfsvfs = arg1;
745 	lr_write_t *lr = arg2;
746 	znode_t	*zp;
747 	int error;
748 	uint64_t end;
749 
750 	if (byteswap)
751 		byteswap_uint64_array(lr, sizeof (*lr));
752 
753 	if ((error = zfs_zget(zfsvfs, lr->lr_foid, &zp)) != 0)
754 		return (error);
755 
756 top:
757 	end = lr->lr_offset + lr->lr_length;
758 	if (end > zp->z_size) {
759 		dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
760 
761 		zp->z_size = end;
762 		dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
763 		error = dmu_tx_assign(tx, TXG_WAIT);
764 		if (error) {
765 			zrele(zp);
766 			if (error == ERESTART) {
767 				dmu_tx_wait(tx);
768 				dmu_tx_abort(tx);
769 				goto top;
770 			}
771 			dmu_tx_abort(tx);
772 			return (error);
773 		}
774 		(void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
775 		    (void *)&zp->z_size, sizeof (uint64_t), tx);
776 
777 		/* Ensure the replayed seq is updated */
778 		(void) zil_replaying(zfsvfs->z_log, tx);
779 
780 		dmu_tx_commit(tx);
781 	}
782 
783 	zrele(zp);
784 
785 	return (error);
786 }
787 
788 static int
789 zfs_replay_truncate(void *arg1, void *arg2, boolean_t byteswap)
790 {
791 	zfsvfs_t *zfsvfs = arg1;
792 	lr_truncate_t *lr = arg2;
793 	znode_t *zp;
794 	flock64_t fl = {0};
795 	int error;
796 
797 	if (byteswap)
798 		byteswap_uint64_array(lr, sizeof (*lr));
799 
800 	if ((error = zfs_zget(zfsvfs, lr->lr_foid, &zp)) != 0)
801 		return (error);
802 
803 	fl.l_type = F_WRLCK;
804 	fl.l_whence = SEEK_SET;
805 	fl.l_start = lr->lr_offset;
806 	fl.l_len = lr->lr_length;
807 
808 	error = zfs_space(zp, F_FREESP, &fl, O_RDWR | O_LARGEFILE,
809 	    lr->lr_offset, kcred);
810 
811 	zrele(zp);
812 
813 	return (error);
814 }
815 
816 static int
817 zfs_replay_setattr(void *arg1, void *arg2, boolean_t byteswap)
818 {
819 	zfsvfs_t *zfsvfs = arg1;
820 	lr_setattr_t *lr = arg2;
821 	znode_t *zp;
822 	xvattr_t xva;
823 	vattr_t *vap = &xva.xva_vattr;
824 	int error;
825 	void *start;
826 
827 	xva_init(&xva);
828 	if (byteswap) {
829 		byteswap_uint64_array(lr, sizeof (*lr));
830 
831 		if ((lr->lr_mask & ATTR_XVATTR) &&
832 		    zfsvfs->z_version >= ZPL_VERSION_INITIAL)
833 			zfs_replay_swap_attrs((lr_attr_t *)(lr + 1));
834 	}
835 
836 	if ((error = zfs_zget(zfsvfs, lr->lr_foid, &zp)) != 0)
837 		return (error);
838 
839 	zfs_init_vattr(vap, lr->lr_mask, lr->lr_mode,
840 	    lr->lr_uid, lr->lr_gid, 0, lr->lr_foid);
841 
842 	vap->va_size = lr->lr_size;
843 	ZFS_TIME_DECODE(&vap->va_atime, lr->lr_atime);
844 	ZFS_TIME_DECODE(&vap->va_mtime, lr->lr_mtime);
845 	gethrestime(&vap->va_ctime);
846 	vap->va_mask |= ATTR_CTIME;
847 
848 	/*
849 	 * Fill in xvattr_t portions if necessary.
850 	 */
851 
852 	start = (lr_setattr_t *)(lr + 1);
853 	if (vap->va_mask & ATTR_XVATTR) {
854 		zfs_replay_xvattr((lr_attr_t *)start, &xva);
855 		start = (caddr_t)start +
856 		    ZIL_XVAT_SIZE(((lr_attr_t *)start)->lr_attr_masksize);
857 	} else
858 		xva.xva_vattr.va_mask &= ~ATTR_XVATTR;
859 
860 	zfsvfs->z_fuid_replay = zfs_replay_fuid_domain(start, &start,
861 	    lr->lr_uid, lr->lr_gid);
862 
863 	error = zfs_setattr(zp, vap, 0, kcred);
864 
865 	zfs_fuid_info_free(zfsvfs->z_fuid_replay);
866 	zfsvfs->z_fuid_replay = NULL;
867 	zrele(zp);
868 
869 	return (error);
870 }
871 
872 static int
873 zfs_replay_setsaxattr(void *arg1, void *arg2, boolean_t byteswap)
874 {
875 	zfsvfs_t *zfsvfs = arg1;
876 	lr_setsaxattr_t *lr = arg2;
877 	znode_t *zp;
878 	nvlist_t *nvl;
879 	size_t sa_size;
880 	char *name;
881 	char *value;
882 	size_t size;
883 	int error = 0;
884 
885 	ASSERT(spa_feature_is_active(zfsvfs->z_os->os_spa,
886 	    SPA_FEATURE_ZILSAXATTR));
887 	if (byteswap)
888 		byteswap_uint64_array(lr, sizeof (*lr));
889 
890 	if ((error = zfs_zget(zfsvfs, lr->lr_foid, &zp)) != 0)
891 		return (error);
892 
893 	rw_enter(&zp->z_xattr_lock, RW_WRITER);
894 	mutex_enter(&zp->z_lock);
895 	if (zp->z_xattr_cached == NULL)
896 		error = zfs_sa_get_xattr(zp);
897 	mutex_exit(&zp->z_lock);
898 
899 	if (error)
900 		goto out;
901 
902 	ASSERT(zp->z_xattr_cached);
903 	nvl = zp->z_xattr_cached;
904 
905 	/* Get xattr name, value and size from log record */
906 	size = lr->lr_size;
907 	name = (char *)(lr + 1);
908 	if (size == 0) {
909 		value = NULL;
910 		error = nvlist_remove(nvl, name, DATA_TYPE_BYTE_ARRAY);
911 	} else {
912 		value = name + strlen(name) + 1;
913 		/* Limited to 32k to keep nvpair memory allocations small */
914 		if (size > DXATTR_MAX_ENTRY_SIZE) {
915 			error = SET_ERROR(EFBIG);
916 			goto out;
917 		}
918 
919 		/* Prevent the DXATTR SA from consuming the entire SA region */
920 		error = nvlist_size(nvl, &sa_size, NV_ENCODE_XDR);
921 		if (error)
922 			goto out;
923 
924 		if (sa_size > DXATTR_MAX_SA_SIZE) {
925 			error = SET_ERROR(EFBIG);
926 			goto out;
927 		}
928 
929 		error = nvlist_add_byte_array(nvl, name, (uchar_t *)value,
930 		    size);
931 	}
932 
933 	/*
934 	 * Update the SA for additions, modifications, and removals. On
935 	 * error drop the inconsistent cached version of the nvlist, it
936 	 * will be reconstructed from the ARC when next accessed.
937 	 */
938 	if (error == 0)
939 		error = zfs_sa_set_xattr(zp, name, value, size);
940 
941 	if (error) {
942 		nvlist_free(nvl);
943 		zp->z_xattr_cached = NULL;
944 	}
945 
946 out:
947 	rw_exit(&zp->z_xattr_lock);
948 	zrele(zp);
949 	return (error);
950 }
951 
952 static int
953 zfs_replay_acl_v0(void *arg1, void *arg2, boolean_t byteswap)
954 {
955 	zfsvfs_t *zfsvfs = arg1;
956 	lr_acl_v0_t *lr = arg2;
957 	ace_t *ace = (ace_t *)(lr + 1);	/* ace array follows lr_acl_t */
958 	vsecattr_t vsa = {0};
959 	znode_t *zp;
960 	int error;
961 
962 	if (byteswap) {
963 		byteswap_uint64_array(lr, sizeof (*lr));
964 		zfs_oldace_byteswap(ace, lr->lr_aclcnt);
965 	}
966 
967 	if ((error = zfs_zget(zfsvfs, lr->lr_foid, &zp)) != 0)
968 		return (error);
969 
970 	vsa.vsa_mask = VSA_ACE | VSA_ACECNT;
971 	vsa.vsa_aclcnt = lr->lr_aclcnt;
972 	vsa.vsa_aclentsz = sizeof (ace_t) * vsa.vsa_aclcnt;
973 	vsa.vsa_aclflags = 0;
974 	vsa.vsa_aclentp = ace;
975 
976 	error = zfs_setsecattr(zp, &vsa, 0, kcred);
977 
978 	zrele(zp);
979 
980 	return (error);
981 }
982 
983 /*
984  * Replaying ACLs is complicated by FUID support.
985  * The log record may contain some optional data
986  * to be used for replaying FUID's.  These pieces
987  * are the actual FUIDs that were created initially.
988  * The FUID table index may no longer be valid and
989  * during zfs_create() a new index may be assigned.
990  * Because of this the log will contain the original
991  * domain+rid in order to create a new FUID.
992  *
993  * The individual ACEs may contain an ephemeral uid/gid which is no
994  * longer valid and will need to be replaced with an actual FUID.
995  *
996  */
997 static int
998 zfs_replay_acl(void *arg1, void *arg2, boolean_t byteswap)
999 {
1000 	zfsvfs_t *zfsvfs = arg1;
1001 	lr_acl_t *lr = arg2;
1002 	ace_t *ace = (ace_t *)(lr + 1);
1003 	vsecattr_t vsa = {0};
1004 	znode_t *zp;
1005 	int error;
1006 
1007 	if (byteswap) {
1008 		byteswap_uint64_array(lr, sizeof (*lr));
1009 		zfs_ace_byteswap(ace, lr->lr_acl_bytes, B_FALSE);
1010 		if (lr->lr_fuidcnt) {
1011 			byteswap_uint64_array((caddr_t)ace +
1012 			    ZIL_ACE_LENGTH(lr->lr_acl_bytes),
1013 			    lr->lr_fuidcnt * sizeof (uint64_t));
1014 		}
1015 	}
1016 
1017 	if ((error = zfs_zget(zfsvfs, lr->lr_foid, &zp)) != 0)
1018 		return (error);
1019 
1020 	vsa.vsa_mask = VSA_ACE | VSA_ACECNT | VSA_ACE_ACLFLAGS;
1021 	vsa.vsa_aclcnt = lr->lr_aclcnt;
1022 	vsa.vsa_aclentp = ace;
1023 	vsa.vsa_aclentsz = lr->lr_acl_bytes;
1024 	vsa.vsa_aclflags = lr->lr_acl_flags;
1025 
1026 	if (lr->lr_fuidcnt) {
1027 		void *fuidstart = (caddr_t)ace +
1028 		    ZIL_ACE_LENGTH(lr->lr_acl_bytes);
1029 
1030 		zfsvfs->z_fuid_replay =
1031 		    zfs_replay_fuids(fuidstart, &fuidstart,
1032 		    lr->lr_fuidcnt, lr->lr_domcnt, 0, 0);
1033 	}
1034 
1035 	error = zfs_setsecattr(zp, &vsa, 0, kcred);
1036 
1037 	if (zfsvfs->z_fuid_replay)
1038 		zfs_fuid_info_free(zfsvfs->z_fuid_replay);
1039 
1040 	zfsvfs->z_fuid_replay = NULL;
1041 	zrele(zp);
1042 
1043 	return (error);
1044 }
1045 
1046 /*
1047  * Callback vectors for replaying records
1048  */
1049 zil_replay_func_t *const zfs_replay_vector[TX_MAX_TYPE] = {
1050 	zfs_replay_error,	/* no such type */
1051 	zfs_replay_create,	/* TX_CREATE */
1052 	zfs_replay_create,	/* TX_MKDIR */
1053 	zfs_replay_create,	/* TX_MKXATTR */
1054 	zfs_replay_create,	/* TX_SYMLINK */
1055 	zfs_replay_remove,	/* TX_REMOVE */
1056 	zfs_replay_remove,	/* TX_RMDIR */
1057 	zfs_replay_link,	/* TX_LINK */
1058 	zfs_replay_rename,	/* TX_RENAME */
1059 	zfs_replay_write,	/* TX_WRITE */
1060 	zfs_replay_truncate,	/* TX_TRUNCATE */
1061 	zfs_replay_setattr,	/* TX_SETATTR */
1062 	zfs_replay_acl_v0,	/* TX_ACL_V0 */
1063 	zfs_replay_acl,		/* TX_ACL */
1064 	zfs_replay_create_acl,	/* TX_CREATE_ACL */
1065 	zfs_replay_create,	/* TX_CREATE_ATTR */
1066 	zfs_replay_create_acl,	/* TX_CREATE_ACL_ATTR */
1067 	zfs_replay_create_acl,	/* TX_MKDIR_ACL */
1068 	zfs_replay_create,	/* TX_MKDIR_ATTR */
1069 	zfs_replay_create_acl,	/* TX_MKDIR_ACL_ATTR */
1070 	zfs_replay_write2,	/* TX_WRITE2 */
1071 	zfs_replay_setsaxattr,	/* TX_SETSAXATTR */
1072 };
1073