xref: /freebsd-head/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_znode_os.c (revision 61145dc2b94f12f6a47344fb9aac702321880e43)
1 // SPDX-License-Identifier: CDDL-1.0
2 /*
3  * CDDL HEADER START
4  *
5  * The contents of this file are subject to the terms of the
6  * Common Development and Distribution License (the "License").
7  * You may not use this file except in compliance with the License.
8  *
9  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10  * or https://opensource.org/licenses/CDDL-1.0.
11  * See the License for the specific language governing permissions
12  * and limitations under the License.
13  *
14  * When distributing Covered Code, include this CDDL HEADER in each
15  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16  * If applicable, add the following below this CDDL HEADER, with the
17  * fields enclosed by brackets "[]" replaced with your own identifying
18  * information: Portions Copyright [yyyy] [name of copyright owner]
19  *
20  * CDDL HEADER END
21  */
22 /*
23  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24  * Copyright (c) 2012, 2014 by Delphix. All rights reserved.
25  * Copyright (c) 2014 Integros [integros.com]
26  */
27 
28 /* Portions Copyright 2007 Jeremy Teo */
29 /* Portions Copyright 2011 Martin Matuska <mm@FreeBSD.org> */
30 
31 #include <sys/types.h>
32 #include <sys/param.h>
33 #include <sys/time.h>
34 #include <sys/systm.h>
35 #include <sys/sysmacros.h>
36 #include <sys/resource.h>
37 #include <sys/resourcevar.h>
38 #include <sys/mntent.h>
39 #include <sys/u8_textprep.h>
40 #include <sys/dsl_dataset.h>
41 #include <sys/vfs.h>
42 #include <sys/vnode.h>
43 #include <sys/file.h>
44 #include <sys/kmem.h>
45 #include <sys/errno.h>
46 #include <sys/unistd.h>
47 #include <sys/atomic.h>
48 #include <sys/zfs_dir.h>
49 #include <sys/zfs_acl.h>
50 #include <sys/zfs_ioctl.h>
51 #include <sys/zfs_rlock.h>
52 #include <sys/zfs_fuid.h>
53 #include <sys/dnode.h>
54 #include <sys/fs/zfs.h>
55 #include <sys/dmu.h>
56 #include <sys/dmu_objset.h>
57 #include <sys/dmu_tx.h>
58 #include <sys/zfs_refcount.h>
59 #include <sys/stat.h>
60 #include <sys/zap.h>
61 #include <sys/zfs_znode.h>
62 #include <sys/sa.h>
63 #include <sys/zfs_sa.h>
64 #include <sys/zfs_stat.h>
65 
66 #include "zfs_prop.h"
67 #include "zfs_comutil.h"
68 
69 /* Used by fstat(1). */
70 SYSCTL_INT(_debug_sizeof, OID_AUTO, znode, CTLFLAG_RD,
71 	SYSCTL_NULL_INT_PTR, sizeof (znode_t), "sizeof(znode_t)");
72 
73 /*
74  * Define ZNODE_STATS to turn on statistic gathering. By default, it is only
75  * turned on when DEBUG is also defined.
76  */
77 #ifdef	ZFS_DEBUG
78 #define	ZNODE_STATS
79 #endif	/* DEBUG */
80 
81 #ifdef	ZNODE_STATS
82 #define	ZNODE_STAT_ADD(stat)			((stat)++)
83 #else
84 #define	ZNODE_STAT_ADD(stat)			/* nothing */
85 #endif	/* ZNODE_STATS */
86 
87 #if !defined(KMEM_DEBUG)
88 #define	_ZFS_USE_SMR
89 static uma_zone_t znode_uma_zone;
90 #else
91 static kmem_cache_t *znode_cache = NULL;
92 #endif
93 
94 extern struct vop_vector zfs_vnodeops;
95 extern struct vop_vector zfs_fifoops;
96 extern struct vop_vector zfs_shareops;
97 
98 
99 /*
100  * This callback is invoked when acquiring a RL_WRITER or RL_APPEND lock on
101  * z_rangelock. It will modify the offset and length of the lock to reflect
102  * znode-specific information, and convert RL_APPEND to RL_WRITER.  This is
103  * called with the rangelock_t's rl_lock held, which avoids races.
104  */
105 static void
zfs_rangelock_cb(zfs_locked_range_t * new,void * arg)106 zfs_rangelock_cb(zfs_locked_range_t *new, void *arg)
107 {
108 	znode_t *zp = arg;
109 
110 	/*
111 	 * If in append mode, convert to writer and lock starting at the
112 	 * current end of file.
113 	 */
114 	if (new->lr_type == RL_APPEND) {
115 		new->lr_offset = zp->z_size;
116 		new->lr_type = RL_WRITER;
117 	}
118 
119 	/*
120 	 * If we need to grow the block size then lock the whole file range.
121 	 */
122 	uint64_t end_size = MAX(zp->z_size, new->lr_offset + new->lr_length);
123 	if (end_size > zp->z_blksz && (!ISP2(zp->z_blksz) ||
124 	    zp->z_blksz < ZTOZSB(zp)->z_max_blksz)) {
125 		new->lr_offset = 0;
126 		new->lr_length = UINT64_MAX;
127 	}
128 }
129 
130 static int
zfs_znode_cache_constructor(void * buf,void * arg,int kmflags)131 zfs_znode_cache_constructor(void *buf, void *arg, int kmflags)
132 {
133 	znode_t *zp = buf;
134 
135 	POINTER_INVALIDATE(&zp->z_zfsvfs);
136 
137 	list_link_init(&zp->z_link_node);
138 
139 	mutex_init(&zp->z_lock, NULL, MUTEX_DEFAULT, NULL);
140 	mutex_init(&zp->z_acl_lock, NULL, MUTEX_DEFAULT, NULL);
141 	rw_init(&zp->z_xattr_lock, NULL, RW_DEFAULT, NULL);
142 
143 	zfs_rangelock_init(&zp->z_rangelock, zfs_rangelock_cb, zp);
144 
145 	zp->z_acl_cached = NULL;
146 	zp->z_xattr_cached = NULL;
147 	zp->z_xattr_parent = 0;
148 	zp->z_vnode = NULL;
149 	zp->z_sync_writes_cnt = 0;
150 	zp->z_async_writes_cnt = 0;
151 
152 	return (0);
153 }
154 
155 static void
zfs_znode_cache_destructor(void * buf,void * arg)156 zfs_znode_cache_destructor(void *buf, void *arg)
157 {
158 	(void) arg;
159 	znode_t *zp = buf;
160 
161 	ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs));
162 	ASSERT3P(zp->z_vnode, ==, NULL);
163 	ASSERT(!list_link_active(&zp->z_link_node));
164 	mutex_destroy(&zp->z_lock);
165 	mutex_destroy(&zp->z_acl_lock);
166 	rw_destroy(&zp->z_xattr_lock);
167 	zfs_rangelock_fini(&zp->z_rangelock);
168 
169 	ASSERT3P(zp->z_acl_cached, ==, NULL);
170 	ASSERT3P(zp->z_xattr_cached, ==, NULL);
171 
172 	ASSERT0(atomic_load_32(&zp->z_sync_writes_cnt));
173 	ASSERT0(atomic_load_32(&zp->z_async_writes_cnt));
174 }
175 
176 
177 #ifdef _ZFS_USE_SMR
178 VFS_SMR_DECLARE;
179 
180 static int
zfs_znode_cache_constructor_smr(void * mem,int size __unused,void * private,int flags)181 zfs_znode_cache_constructor_smr(void *mem, int size __unused, void *private,
182     int flags)
183 {
184 	return (zfs_znode_cache_constructor(mem, private, flags));
185 }
186 
187 static void
zfs_znode_cache_destructor_smr(void * mem,int size __unused,void * private)188 zfs_znode_cache_destructor_smr(void *mem, int size __unused, void *private)
189 {
190 	zfs_znode_cache_destructor(mem, private);
191 }
192 
193 void
zfs_znode_init(void)194 zfs_znode_init(void)
195 {
196 	/*
197 	 * Initialize zcache
198 	 */
199 	ASSERT3P(znode_uma_zone, ==, NULL);
200 	znode_uma_zone = uma_zcreate("zfs_znode_cache",
201 	    sizeof (znode_t), zfs_znode_cache_constructor_smr,
202 	    zfs_znode_cache_destructor_smr, NULL, NULL, 0, 0);
203 	VFS_SMR_ZONE_SET(znode_uma_zone);
204 }
205 
206 static znode_t *
zfs_znode_alloc_kmem(int flags)207 zfs_znode_alloc_kmem(int flags)
208 {
209 	return (uma_zalloc_smr(znode_uma_zone, flags));
210 }
211 
212 static void
zfs_znode_free_kmem(znode_t * zp)213 zfs_znode_free_kmem(znode_t *zp)
214 {
215 	if (zp->z_xattr_cached) {
216 		nvlist_free(zp->z_xattr_cached);
217 		zp->z_xattr_cached = NULL;
218 	}
219 	uma_zfree_smr(znode_uma_zone, zp);
220 }
221 #else
222 void
zfs_znode_init(void)223 zfs_znode_init(void)
224 {
225 	/*
226 	 * Initialize zcache
227 	 */
228 	ASSERT3P(znode_cache, ==, NULL);
229 	znode_cache = kmem_cache_create("zfs_znode_cache",
230 	    sizeof (znode_t), 0, zfs_znode_cache_constructor,
231 	    zfs_znode_cache_destructor, NULL, NULL, NULL, KMC_RECLAIMABLE);
232 }
233 
234 static znode_t *
zfs_znode_alloc_kmem(int flags)235 zfs_znode_alloc_kmem(int flags)
236 {
237 	return (kmem_cache_alloc(znode_cache, flags));
238 }
239 
240 static void
zfs_znode_free_kmem(znode_t * zp)241 zfs_znode_free_kmem(znode_t *zp)
242 {
243 	if (zp->z_xattr_cached) {
244 		nvlist_free(zp->z_xattr_cached);
245 		zp->z_xattr_cached = NULL;
246 	}
247 	kmem_cache_free(znode_cache, zp);
248 }
249 #endif
250 
251 void
zfs_znode_fini(void)252 zfs_znode_fini(void)
253 {
254 	/*
255 	 * Cleanup zcache
256 	 */
257 #ifdef _ZFS_USE_SMR
258 	if (znode_uma_zone) {
259 		uma_zdestroy(znode_uma_zone);
260 		znode_uma_zone = NULL;
261 	}
262 #else
263 	if (znode_cache) {
264 		kmem_cache_destroy(znode_cache);
265 		znode_cache = NULL;
266 	}
267 #endif
268 }
269 
270 
271 static int
zfs_create_share_dir(zfsvfs_t * zfsvfs,dmu_tx_t * tx)272 zfs_create_share_dir(zfsvfs_t *zfsvfs, dmu_tx_t *tx)
273 {
274 	zfs_acl_ids_t acl_ids;
275 	vattr_t vattr;
276 	znode_t *sharezp;
277 	znode_t *zp;
278 	int error;
279 
280 	vattr.va_mask = AT_MODE|AT_UID|AT_GID;
281 	vattr.va_type = VDIR;
282 	vattr.va_mode = S_IFDIR|0555;
283 	vattr.va_uid = crgetuid(kcred);
284 	vattr.va_gid = crgetgid(kcred);
285 
286 	sharezp = zfs_znode_alloc_kmem(KM_SLEEP);
287 	ASSERT(!POINTER_IS_VALID(sharezp->z_zfsvfs));
288 	sharezp->z_unlinked = 0;
289 	sharezp->z_atime_dirty = 0;
290 	sharezp->z_zfsvfs = zfsvfs;
291 	sharezp->z_is_sa = zfsvfs->z_use_sa;
292 
293 	VERIFY0(zfs_acl_ids_create(sharezp, IS_ROOT_NODE, &vattr,
294 	    kcred, NULL, &acl_ids, NULL));
295 	zfs_mknode(sharezp, &vattr, tx, kcred, IS_ROOT_NODE, &zp, &acl_ids);
296 	ASSERT3P(zp, ==, sharezp);
297 	POINTER_INVALIDATE(&sharezp->z_zfsvfs);
298 	error = zap_add(zfsvfs->z_os, MASTER_NODE_OBJ,
299 	    ZFS_SHARES_DIR, 8, 1, &sharezp->z_id, tx);
300 	zfsvfs->z_shares_dir = sharezp->z_id;
301 
302 	zfs_acl_ids_free(&acl_ids);
303 	sa_handle_destroy(sharezp->z_sa_hdl);
304 	zfs_znode_free_kmem(sharezp);
305 
306 	return (error);
307 }
308 
309 /*
310  * define a couple of values we need available
311  * for both 64 and 32 bit environments.
312  */
313 #ifndef NBITSMINOR64
314 #define	NBITSMINOR64	32
315 #endif
316 #ifndef MAXMAJ64
317 #define	MAXMAJ64	0xffffffffUL
318 #endif
319 #ifndef	MAXMIN64
320 #define	MAXMIN64	0xffffffffUL
321 #endif
322 
323 /*
324  * Create special expldev for ZFS private use.
325  * Can't use standard expldev since it doesn't do
326  * what we want.  The standard expldev() takes a
327  * dev32_t in LP64 and expands it to a long dev_t.
328  * We need an interface that takes a dev32_t in ILP32
329  * and expands it to a long dev_t.
330  */
331 static uint64_t
zfs_expldev(dev_t dev)332 zfs_expldev(dev_t dev)
333 {
334 	return (((uint64_t)major(dev) << NBITSMINOR64) | minor(dev));
335 }
336 /*
337  * Special cmpldev for ZFS private use.
338  * Can't use standard cmpldev since it takes
339  * a long dev_t and compresses it to dev32_t in
340  * LP64.  We need to do a compaction of a long dev_t
341  * to a dev32_t in ILP32.
342  */
343 dev_t
zfs_cmpldev(uint64_t dev)344 zfs_cmpldev(uint64_t dev)
345 {
346 	return (makedev((dev >> NBITSMINOR64), (dev & MAXMIN64)));
347 }
348 
349 static void
zfs_znode_sa_init(zfsvfs_t * zfsvfs,znode_t * zp,dmu_buf_t * db,dmu_object_type_t obj_type,sa_handle_t * sa_hdl)350 zfs_znode_sa_init(zfsvfs_t *zfsvfs, znode_t *zp,
351     dmu_buf_t *db, dmu_object_type_t obj_type, sa_handle_t *sa_hdl)
352 {
353 	ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs) || (zfsvfs == zp->z_zfsvfs));
354 	ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zfsvfs, zp->z_id)));
355 
356 	ASSERT3P(zp->z_sa_hdl, ==, NULL);
357 	ASSERT3P(zp->z_acl_cached, ==, NULL);
358 	if (sa_hdl == NULL) {
359 		VERIFY0(sa_handle_get_from_db(zfsvfs->z_os, db, zp,
360 		    SA_HDL_SHARED, &zp->z_sa_hdl));
361 	} else {
362 		zp->z_sa_hdl = sa_hdl;
363 		sa_set_userp(sa_hdl, zp);
364 	}
365 
366 	zp->z_is_sa = (obj_type == DMU_OT_SA) ? B_TRUE : B_FALSE;
367 
368 	/*
369 	 * Slap on VROOT if we are the root znode unless we are the root
370 	 * node of a snapshot mounted under .zfs.
371 	 */
372 	if (zp->z_id == zfsvfs->z_root && zfsvfs->z_parent == zfsvfs)
373 		ZTOV(zp)->v_flag |= VROOT;
374 }
375 
376 void
zfs_znode_dmu_fini(znode_t * zp)377 zfs_znode_dmu_fini(znode_t *zp)
378 {
379 	ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zp->z_zfsvfs, zp->z_id)) ||
380 	    ZFS_TEARDOWN_INACTIVE_WRITE_HELD(zp->z_zfsvfs));
381 
382 	sa_handle_destroy(zp->z_sa_hdl);
383 	zp->z_sa_hdl = NULL;
384 }
385 
386 static void
zfs_vnode_forget(vnode_t * vp)387 zfs_vnode_forget(vnode_t *vp)
388 {
389 
390 	/* copied from insmntque_stddtr */
391 	vp->v_data = NULL;
392 	vp->v_op = &dead_vnodeops;
393 	vgone(vp);
394 	vput(vp);
395 }
396 
397 /*
398  * Construct a new znode/vnode and initialize.
399  *
400  * This does not do a call to dmu_set_user() that is
401  * up to the caller to do, in case you don't want to
402  * return the znode
403  */
404 static znode_t *
zfs_znode_alloc(zfsvfs_t * zfsvfs,dmu_buf_t * db,int blksz,dmu_object_type_t obj_type,sa_handle_t * hdl)405 zfs_znode_alloc(zfsvfs_t *zfsvfs, dmu_buf_t *db, int blksz,
406     dmu_object_type_t obj_type, sa_handle_t *hdl)
407 {
408 	znode_t	*zp;
409 	vnode_t *vp;
410 	uint64_t mode;
411 	uint64_t parent;
412 #ifdef notyet
413 	uint64_t mtime[2], ctime[2];
414 #endif
415 	uint64_t projid = ZFS_DEFAULT_PROJID;
416 	sa_bulk_attr_t bulk[9];
417 	int count = 0;
418 	int error;
419 
420 	zp = zfs_znode_alloc_kmem(KM_SLEEP);
421 
422 #ifndef _ZFS_USE_SMR
423 	KASSERT((zfsvfs->z_parent->z_vfs->mnt_kern_flag & MNTK_FPLOOKUP) == 0,
424 	    ("%s: fast path lookup enabled without smr", __func__));
425 #endif
426 
427 	KASSERT(curthread->td_vp_reserved != NULL,
428 	    ("zfs_znode_alloc: getnewvnode without any vnodes reserved"));
429 	error = getnewvnode("zfs", zfsvfs->z_parent->z_vfs, &zfs_vnodeops, &vp);
430 	if (error != 0) {
431 		zfs_znode_free_kmem(zp);
432 		return (NULL);
433 	}
434 	zp->z_vnode = vp;
435 	vp->v_data = zp;
436 
437 	/*
438 	 * Acquire the vnode lock before any possible interaction with the
439 	 * outside world.  Specifically, there is an error path that calls
440 	 * zfs_vnode_forget() and the vnode should be exclusively locked.
441 	 */
442 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
443 
444 	ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs));
445 
446 	zp->z_sa_hdl = NULL;
447 	zp->z_unlinked = 0;
448 	zp->z_atime_dirty = 0;
449 	zp->z_mapcnt = 0;
450 	zp->z_id = db->db_object;
451 	zp->z_blksz = blksz;
452 	zp->z_seq = 0x7A4653;
453 	zp->z_sync_cnt = 0;
454 	zp->z_sync_writes_cnt = 0;
455 	zp->z_async_writes_cnt = 0;
456 	atomic_store_ptr(&zp->z_cached_symlink, NULL);
457 
458 	zfs_znode_sa_init(zfsvfs, zp, db, obj_type, hdl);
459 
460 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8);
461 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL, &zp->z_gen, 8);
462 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
463 	    &zp->z_size, 8);
464 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL,
465 	    &zp->z_links, 8);
466 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
467 	    &zp->z_pflags, 8);
468 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL, &parent, 8);
469 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
470 	    &zp->z_atime, 16);
471 #ifdef notyet
472 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
473 	    &mtime, 16);
474 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
475 	    &ctime, 16);
476 #endif
477 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
478 	    &zp->z_uid, 8);
479 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL,
480 	    &zp->z_gid, 8);
481 
482 	if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count) != 0 || zp->z_gen == 0 ||
483 	    (dmu_objset_projectquota_enabled(zfsvfs->z_os) &&
484 	    (zp->z_pflags & ZFS_PROJID) &&
485 	    sa_lookup(zp->z_sa_hdl, SA_ZPL_PROJID(zfsvfs), &projid, 8) != 0)) {
486 		if (hdl == NULL)
487 			sa_handle_destroy(zp->z_sa_hdl);
488 		zfs_vnode_forget(vp);
489 		zp->z_vnode = NULL;
490 		zfs_znode_free_kmem(zp);
491 		return (NULL);
492 	}
493 
494 	zp->z_projid = projid;
495 	zp->z_mode = mode;
496 
497 	/* Cache the xattr parent id */
498 	if (zp->z_pflags & ZFS_XATTR)
499 		zp->z_xattr_parent = parent;
500 
501 	vp->v_type = IFTOVT((mode_t)mode);
502 
503 	switch (vp->v_type) {
504 	case VDIR:
505 		zp->z_zn_prefetch = B_TRUE; /* z_prefetch default is enabled */
506 		break;
507 	case VFIFO:
508 		vp->v_op = &zfs_fifoops;
509 		break;
510 	case VREG:
511 		if (parent == zfsvfs->z_shares_dir) {
512 			ASSERT0(zp->z_uid);
513 			ASSERT0(zp->z_gid);
514 			vp->v_op = &zfs_shareops;
515 		}
516 		break;
517 	default:
518 			break;
519 	}
520 
521 	mutex_enter(&zfsvfs->z_znodes_lock);
522 	list_insert_tail(&zfsvfs->z_all_znodes, zp);
523 	zp->z_zfsvfs = zfsvfs;
524 	mutex_exit(&zfsvfs->z_znodes_lock);
525 
526 #if __FreeBSD_version >= 1400077
527 	vn_set_state(vp, VSTATE_CONSTRUCTED);
528 #endif
529 	VN_LOCK_AREC(vp);
530 	if (vp->v_type != VFIFO)
531 		VN_LOCK_ASHARE(vp);
532 
533 	return (zp);
534 }
535 
536 static uint64_t empty_xattr;
537 static uint64_t pad[4];
538 static zfs_acl_phys_t acl_phys;
539 /*
540  * Create a new DMU object to hold a zfs znode.
541  *
542  *	IN:	dzp	- parent directory for new znode
543  *		vap	- file attributes for new znode
544  *		tx	- dmu transaction id for zap operations
545  *		cr	- credentials of caller
546  *		flag	- flags:
547  *			  IS_ROOT_NODE	- new object will be root
548  *			  IS_XATTR	- new object is an attribute
549  *		bonuslen - length of bonus buffer
550  *		setaclp  - File/Dir initial ACL
551  *		fuidp	 - Tracks fuid allocation.
552  *
553  *	OUT:	zpp	- allocated znode
554  *
555  */
556 void
zfs_mknode(znode_t * dzp,vattr_t * vap,dmu_tx_t * tx,cred_t * cr,uint_t flag,znode_t ** zpp,zfs_acl_ids_t * acl_ids)557 zfs_mknode(znode_t *dzp, vattr_t *vap, dmu_tx_t *tx, cred_t *cr,
558     uint_t flag, znode_t **zpp, zfs_acl_ids_t *acl_ids)
559 {
560 	uint64_t	crtime[2], atime[2], mtime[2], ctime[2];
561 	uint64_t	mode, size, links, parent, pflags;
562 	uint64_t	dzp_pflags = 0;
563 	uint64_t	rdev = 0;
564 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
565 	dmu_buf_t	*db;
566 	timestruc_t	now;
567 	uint64_t	gen, obj;
568 	int		bonuslen;
569 	int		dnodesize;
570 	sa_handle_t	*sa_hdl;
571 	dmu_object_type_t obj_type;
572 	sa_bulk_attr_t	*sa_attrs;
573 	int		cnt = 0;
574 	zfs_acl_locator_cb_t locate = { 0 };
575 
576 	ASSERT3P(vap, !=, NULL);
577 	ASSERT3U((vap->va_mask & AT_MODE), ==, AT_MODE);
578 
579 	if (zfsvfs->z_replay) {
580 		obj = vap->va_nodeid;
581 		now = vap->va_ctime;		/* see zfs_replay_create() */
582 		gen = vap->va_nblocks;		/* ditto */
583 		dnodesize = vap->va_fsid;	/* ditto */
584 	} else {
585 		obj = 0;
586 		vfs_timestamp(&now);
587 		gen = dmu_tx_get_txg(tx);
588 		dnodesize = dmu_objset_dnodesize(zfsvfs->z_os);
589 	}
590 
591 	if (dnodesize == 0)
592 		dnodesize = DNODE_MIN_SIZE;
593 
594 	obj_type = zfsvfs->z_use_sa ? DMU_OT_SA : DMU_OT_ZNODE;
595 	bonuslen = (obj_type == DMU_OT_SA) ?
596 	    DN_BONUS_SIZE(dnodesize) : ZFS_OLD_ZNODE_PHYS_SIZE;
597 
598 	/*
599 	 * Create a new DMU object.
600 	 */
601 	/*
602 	 * There's currently no mechanism for pre-reading the blocks that will
603 	 * be needed to allocate a new object, so we accept the small chance
604 	 * that there will be an i/o error and we will fail one of the
605 	 * assertions below.
606 	 */
607 	if (vap->va_type == VDIR) {
608 		if (zfsvfs->z_replay) {
609 			VERIFY0(zap_create_claim_norm_dnsize(zfsvfs->z_os, obj,
610 			    zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS,
611 			    obj_type, bonuslen, dnodesize, tx));
612 		} else {
613 			obj = zap_create_norm_dnsize(zfsvfs->z_os,
614 			    zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS,
615 			    obj_type, bonuslen, dnodesize, tx);
616 		}
617 	} else {
618 		if (zfsvfs->z_replay) {
619 			VERIFY0(dmu_object_claim_dnsize(zfsvfs->z_os, obj,
620 			    DMU_OT_PLAIN_FILE_CONTENTS, 0,
621 			    obj_type, bonuslen, dnodesize, tx));
622 		} else {
623 			obj = dmu_object_alloc_dnsize(zfsvfs->z_os,
624 			    DMU_OT_PLAIN_FILE_CONTENTS, 0,
625 			    obj_type, bonuslen, dnodesize, tx);
626 		}
627 	}
628 
629 	ZFS_OBJ_HOLD_ENTER(zfsvfs, obj);
630 	VERIFY0(sa_buf_hold(zfsvfs->z_os, obj, NULL, &db));
631 
632 	/*
633 	 * If this is the root, fix up the half-initialized parent pointer
634 	 * to reference the just-allocated physical data area.
635 	 */
636 	if (flag & IS_ROOT_NODE) {
637 		dzp->z_id = obj;
638 	} else {
639 		dzp_pflags = dzp->z_pflags;
640 	}
641 
642 	/*
643 	 * If parent is an xattr, so am I.
644 	 */
645 	if (dzp_pflags & ZFS_XATTR) {
646 		flag |= IS_XATTR;
647 	}
648 
649 	if (zfsvfs->z_use_fuids)
650 		pflags = ZFS_ARCHIVE | ZFS_AV_MODIFIED;
651 	else
652 		pflags = 0;
653 
654 	if (vap->va_type == VDIR) {
655 		size = 2;		/* contents ("." and "..") */
656 		links = (flag & (IS_ROOT_NODE | IS_XATTR)) ? 2 : 1;
657 	} else {
658 		size = links = 0;
659 	}
660 
661 	if (vap->va_type == VBLK || vap->va_type == VCHR) {
662 		rdev = zfs_expldev(vap->va_rdev);
663 	}
664 
665 	parent = dzp->z_id;
666 	mode = acl_ids->z_mode;
667 	if (flag & IS_XATTR)
668 		pflags |= ZFS_XATTR;
669 
670 	/*
671 	 * No execs denied will be determined when zfs_mode_compute() is called.
672 	 */
673 	pflags |= acl_ids->z_aclp->z_hints &
674 	    (ZFS_ACL_TRIVIAL|ZFS_INHERIT_ACE|ZFS_ACL_AUTO_INHERIT|
675 	    ZFS_ACL_DEFAULTED|ZFS_ACL_PROTECTED);
676 
677 	ZFS_TIME_ENCODE(&now, crtime);
678 	ZFS_TIME_ENCODE(&now, ctime);
679 
680 	if (vap->va_mask & AT_ATIME) {
681 		ZFS_TIME_ENCODE(&vap->va_atime, atime);
682 	} else {
683 		ZFS_TIME_ENCODE(&now, atime);
684 	}
685 
686 	if (vap->va_mask & AT_MTIME) {
687 		ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
688 	} else {
689 		ZFS_TIME_ENCODE(&now, mtime);
690 	}
691 
692 	/* Now add in all of the "SA" attributes */
693 	VERIFY0(sa_handle_get_from_db(zfsvfs->z_os, db, NULL, SA_HDL_SHARED,
694 	    &sa_hdl));
695 
696 	/*
697 	 * Setup the array of attributes to be replaced/set on the new file
698 	 *
699 	 * order for  DMU_OT_ZNODE is critical since it needs to be constructed
700 	 * in the old znode_phys_t format.  Don't change this ordering
701 	 */
702 	sa_attrs = kmem_alloc(sizeof (sa_bulk_attr_t) * ZPL_END, KM_SLEEP);
703 
704 	if (obj_type == DMU_OT_ZNODE) {
705 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs),
706 		    NULL, &atime, 16);
707 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs),
708 		    NULL, &mtime, 16);
709 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs),
710 		    NULL, &ctime, 16);
711 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs),
712 		    NULL, &crtime, 16);
713 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs),
714 		    NULL, &gen, 8);
715 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs),
716 		    NULL, &mode, 8);
717 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs),
718 		    NULL, &size, 8);
719 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs),
720 		    NULL, &parent, 8);
721 	} else {
722 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs),
723 		    NULL, &mode, 8);
724 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs),
725 		    NULL, &size, 8);
726 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs),
727 		    NULL, &gen, 8);
728 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs),
729 		    NULL, &acl_ids->z_fuid, 8);
730 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs),
731 		    NULL, &acl_ids->z_fgid, 8);
732 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs),
733 		    NULL, &parent, 8);
734 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs),
735 		    NULL, &pflags, 8);
736 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs),
737 		    NULL, &atime, 16);
738 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs),
739 		    NULL, &mtime, 16);
740 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs),
741 		    NULL, &ctime, 16);
742 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs),
743 		    NULL, &crtime, 16);
744 	}
745 
746 	SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_LINKS(zfsvfs), NULL, &links, 8);
747 
748 	if (obj_type == DMU_OT_ZNODE) {
749 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_XATTR(zfsvfs), NULL,
750 		    &empty_xattr, 8);
751 	}
752 	if (obj_type == DMU_OT_ZNODE ||
753 	    (vap->va_type == VBLK || vap->va_type == VCHR)) {
754 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_RDEV(zfsvfs),
755 		    NULL, &rdev, 8);
756 
757 	}
758 	if (obj_type == DMU_OT_ZNODE) {
759 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs),
760 		    NULL, &pflags, 8);
761 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs), NULL,
762 		    &acl_ids->z_fuid, 8);
763 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs), NULL,
764 		    &acl_ids->z_fgid, 8);
765 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PAD(zfsvfs), NULL, pad,
766 		    sizeof (uint64_t) * 4);
767 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ZNODE_ACL(zfsvfs), NULL,
768 		    &acl_phys, sizeof (zfs_acl_phys_t));
769 	} else if (acl_ids->z_aclp->z_version >= ZFS_ACL_VERSION_FUID) {
770 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_COUNT(zfsvfs), NULL,
771 		    &acl_ids->z_aclp->z_acl_count, 8);
772 		locate.cb_aclp = acl_ids->z_aclp;
773 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_ACES(zfsvfs),
774 		    zfs_acl_data_locator, &locate,
775 		    acl_ids->z_aclp->z_acl_bytes);
776 		mode = zfs_mode_compute(mode, acl_ids->z_aclp, &pflags,
777 		    acl_ids->z_fuid, acl_ids->z_fgid);
778 	}
779 
780 	VERIFY0(sa_replace_all_by_template(sa_hdl, sa_attrs, cnt, tx));
781 
782 	if (!(flag & IS_ROOT_NODE)) {
783 		*zpp = zfs_znode_alloc(zfsvfs, db, 0, obj_type, sa_hdl);
784 		ASSERT3P(*zpp, !=, NULL);
785 	} else {
786 		/*
787 		 * If we are creating the root node, the "parent" we
788 		 * passed in is the znode for the root.
789 		 */
790 		*zpp = dzp;
791 
792 		(*zpp)->z_sa_hdl = sa_hdl;
793 	}
794 
795 	(*zpp)->z_pflags = pflags;
796 	(*zpp)->z_mode = mode;
797 	(*zpp)->z_dnodesize = dnodesize;
798 
799 	if (vap->va_mask & AT_XVATTR)
800 		zfs_xvattr_set(*zpp, (xvattr_t *)vap, tx);
801 
802 	if (obj_type == DMU_OT_ZNODE ||
803 	    acl_ids->z_aclp->z_version < ZFS_ACL_VERSION_FUID) {
804 		VERIFY0(zfs_aclset_common(*zpp, acl_ids->z_aclp, cr, tx));
805 	}
806 	if (!(flag & IS_ROOT_NODE)) {
807 		vnode_t *vp = ZTOV(*zpp);
808 		vp->v_vflag |= VV_FORCEINSMQ;
809 		int err = insmntque(vp, zfsvfs->z_vfs);
810 		vp->v_vflag &= ~VV_FORCEINSMQ;
811 		(void) err;
812 		KASSERT(err == 0, ("insmntque() failed: error %d", err));
813 	}
814 	kmem_free(sa_attrs, sizeof (sa_bulk_attr_t) * ZPL_END);
815 	ZFS_OBJ_HOLD_EXIT(zfsvfs, obj);
816 }
817 
818 /*
819  * Update in-core attributes.  It is assumed the caller will be doing an
820  * sa_bulk_update to push the changes out.
821  */
822 void
zfs_xvattr_set(znode_t * zp,xvattr_t * xvap,dmu_tx_t * tx)823 zfs_xvattr_set(znode_t *zp, xvattr_t *xvap, dmu_tx_t *tx)
824 {
825 	xoptattr_t *xoap;
826 
827 	xoap = xva_getxoptattr(xvap);
828 	ASSERT3P(xoap, !=, NULL);
829 
830 	if (zp->z_zfsvfs->z_replay == B_FALSE) {
831 		ASSERT_VOP_IN_SEQC(ZTOV(zp));
832 	}
833 
834 	if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
835 		uint64_t times[2];
836 		ZFS_TIME_ENCODE(&xoap->xoa_createtime, times);
837 		(void) sa_update(zp->z_sa_hdl, SA_ZPL_CRTIME(zp->z_zfsvfs),
838 		    &times, sizeof (times), tx);
839 		XVA_SET_RTN(xvap, XAT_CREATETIME);
840 	}
841 	if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
842 		ZFS_ATTR_SET(zp, ZFS_READONLY, xoap->xoa_readonly,
843 		    zp->z_pflags, tx);
844 		XVA_SET_RTN(xvap, XAT_READONLY);
845 	}
846 	if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
847 		ZFS_ATTR_SET(zp, ZFS_HIDDEN, xoap->xoa_hidden,
848 		    zp->z_pflags, tx);
849 		XVA_SET_RTN(xvap, XAT_HIDDEN);
850 	}
851 	if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
852 		ZFS_ATTR_SET(zp, ZFS_SYSTEM, xoap->xoa_system,
853 		    zp->z_pflags, tx);
854 		XVA_SET_RTN(xvap, XAT_SYSTEM);
855 	}
856 	if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
857 		ZFS_ATTR_SET(zp, ZFS_ARCHIVE, xoap->xoa_archive,
858 		    zp->z_pflags, tx);
859 		XVA_SET_RTN(xvap, XAT_ARCHIVE);
860 	}
861 	if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
862 		ZFS_ATTR_SET(zp, ZFS_IMMUTABLE, xoap->xoa_immutable,
863 		    zp->z_pflags, tx);
864 		XVA_SET_RTN(xvap, XAT_IMMUTABLE);
865 	}
866 	if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
867 		ZFS_ATTR_SET(zp, ZFS_NOUNLINK, xoap->xoa_nounlink,
868 		    zp->z_pflags, tx);
869 		XVA_SET_RTN(xvap, XAT_NOUNLINK);
870 	}
871 	if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
872 		ZFS_ATTR_SET(zp, ZFS_APPENDONLY, xoap->xoa_appendonly,
873 		    zp->z_pflags, tx);
874 		XVA_SET_RTN(xvap, XAT_APPENDONLY);
875 	}
876 	if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
877 		ZFS_ATTR_SET(zp, ZFS_NODUMP, xoap->xoa_nodump,
878 		    zp->z_pflags, tx);
879 		XVA_SET_RTN(xvap, XAT_NODUMP);
880 	}
881 	if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
882 		ZFS_ATTR_SET(zp, ZFS_OPAQUE, xoap->xoa_opaque,
883 		    zp->z_pflags, tx);
884 		XVA_SET_RTN(xvap, XAT_OPAQUE);
885 	}
886 	if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
887 		ZFS_ATTR_SET(zp, ZFS_AV_QUARANTINED,
888 		    xoap->xoa_av_quarantined, zp->z_pflags, tx);
889 		XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
890 	}
891 	if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
892 		ZFS_ATTR_SET(zp, ZFS_AV_MODIFIED, xoap->xoa_av_modified,
893 		    zp->z_pflags, tx);
894 		XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
895 	}
896 	if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) {
897 		zfs_sa_set_scanstamp(zp, xvap, tx);
898 		XVA_SET_RTN(xvap, XAT_AV_SCANSTAMP);
899 	}
900 	if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
901 		ZFS_ATTR_SET(zp, ZFS_REPARSE, xoap->xoa_reparse,
902 		    zp->z_pflags, tx);
903 		XVA_SET_RTN(xvap, XAT_REPARSE);
904 	}
905 	if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
906 		ZFS_ATTR_SET(zp, ZFS_OFFLINE, xoap->xoa_offline,
907 		    zp->z_pflags, tx);
908 		XVA_SET_RTN(xvap, XAT_OFFLINE);
909 	}
910 	if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
911 		ZFS_ATTR_SET(zp, ZFS_SPARSE, xoap->xoa_sparse,
912 		    zp->z_pflags, tx);
913 		XVA_SET_RTN(xvap, XAT_SPARSE);
914 	}
915 }
916 
917 int
zfs_zget(zfsvfs_t * zfsvfs,uint64_t obj_num,znode_t ** zpp)918 zfs_zget(zfsvfs_t *zfsvfs, uint64_t obj_num, znode_t **zpp)
919 {
920 	dmu_object_info_t doi;
921 	dmu_buf_t	*db;
922 	znode_t		*zp;
923 	vnode_t		*vp;
924 	sa_handle_t	*hdl;
925 	int locked;
926 	int err;
927 
928 	getnewvnode_reserve();
929 again:
930 	*zpp = NULL;
931 	ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num);
932 
933 	err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db);
934 	if (err) {
935 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
936 		getnewvnode_drop_reserve();
937 		return (err);
938 	}
939 
940 	dmu_object_info_from_db(db, &doi);
941 	if (doi.doi_bonus_type != DMU_OT_SA &&
942 	    (doi.doi_bonus_type != DMU_OT_ZNODE ||
943 	    (doi.doi_bonus_type == DMU_OT_ZNODE &&
944 	    doi.doi_bonus_size < sizeof (znode_phys_t)))) {
945 		sa_buf_rele(db, NULL);
946 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
947 		getnewvnode_drop_reserve();
948 		return (SET_ERROR(EINVAL));
949 	}
950 
951 	hdl = dmu_buf_get_user(db);
952 	if (hdl != NULL) {
953 		zp = sa_get_userdata(hdl);
954 
955 		/*
956 		 * Since "SA" does immediate eviction we
957 		 * should never find a sa handle that doesn't
958 		 * know about the znode.
959 		 */
960 		ASSERT3P(zp, !=, NULL);
961 		ASSERT3U(zp->z_id, ==, obj_num);
962 		if (zp->z_unlinked) {
963 			err = SET_ERROR(ENOENT);
964 		} else {
965 			vp = ZTOV(zp);
966 			/*
967 			 * Don't let the vnode disappear after
968 			 * ZFS_OBJ_HOLD_EXIT.
969 			 */
970 			VN_HOLD(vp);
971 			*zpp = zp;
972 			err = 0;
973 		}
974 
975 		sa_buf_rele(db, NULL);
976 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
977 
978 		if (err) {
979 			getnewvnode_drop_reserve();
980 			return (err);
981 		}
982 
983 		locked = VOP_ISLOCKED(vp);
984 		VI_LOCK(vp);
985 		if (VN_IS_DOOMED(vp) && locked != LK_EXCLUSIVE) {
986 			/*
987 			 * The vnode is doomed and this thread doesn't
988 			 * hold the exclusive lock on it, so the vnode
989 			 * must be being reclaimed by another thread.
990 			 * Otherwise the doomed vnode is being reclaimed
991 			 * by this thread and zfs_zget is called from
992 			 * ZIL internals.
993 			 */
994 			VI_UNLOCK(vp);
995 
996 			/*
997 			 * XXX vrele() locks the vnode when the last reference
998 			 * is dropped.  Although in this case the vnode is
999 			 * doomed / dead and so no inactivation is required,
1000 			 * the vnode lock is still acquired.  That could result
1001 			 * in a LOR with z_teardown_lock if another thread holds
1002 			 * the vnode's lock and tries to take z_teardown_lock.
1003 			 * But that is only possible if the other thread peforms
1004 			 * a ZFS vnode operation on the vnode.  That either
1005 			 * should not happen if the vnode is dead or the thread
1006 			 * should also have a reference to the vnode and thus
1007 			 * our reference is not last.
1008 			 */
1009 			VN_RELE(vp);
1010 			goto again;
1011 		}
1012 		VI_UNLOCK(vp);
1013 		getnewvnode_drop_reserve();
1014 		return (err);
1015 	}
1016 
1017 	/*
1018 	 * Not found create new znode/vnode
1019 	 * but only if file exists.
1020 	 *
1021 	 * There is a small window where zfs_vget() could
1022 	 * find this object while a file create is still in
1023 	 * progress.  This is checked for in zfs_znode_alloc()
1024 	 *
1025 	 * if zfs_znode_alloc() fails it will drop the hold on the
1026 	 * bonus buffer.
1027 	 */
1028 	zp = zfs_znode_alloc(zfsvfs, db, doi.doi_data_block_size,
1029 	    doi.doi_bonus_type, NULL);
1030 	if (zp == NULL) {
1031 		err = SET_ERROR(ENOENT);
1032 	} else {
1033 		*zpp = zp;
1034 	}
1035 	if (err == 0) {
1036 		vnode_t *vp = ZTOV(zp);
1037 
1038 		err = insmntque(vp, zfsvfs->z_vfs);
1039 		if (err == 0) {
1040 			vp->v_hash = obj_num;
1041 			VOP_UNLOCK(vp);
1042 		} else {
1043 			zp->z_vnode = NULL;
1044 			zfs_znode_dmu_fini(zp);
1045 			zfs_znode_free(zp);
1046 			*zpp = NULL;
1047 		}
1048 	}
1049 	ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1050 	getnewvnode_drop_reserve();
1051 	return (err);
1052 }
1053 
1054 int
zfs_rezget(znode_t * zp)1055 zfs_rezget(znode_t *zp)
1056 {
1057 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1058 	dmu_object_info_t doi;
1059 	dmu_buf_t *db;
1060 	vnode_t *vp;
1061 	uint64_t obj_num = zp->z_id;
1062 	uint64_t mode, size;
1063 	sa_bulk_attr_t bulk[8];
1064 	int err;
1065 	int count = 0;
1066 	uint64_t gen;
1067 
1068 	/*
1069 	 * Remove cached pages before reloading the znode, so that they are not
1070 	 * lingering after we run into any error.  Ideally, we should vgone()
1071 	 * the vnode in case of error, but currently we cannot do that
1072 	 * because of the LOR between the vnode lock and z_teardown_lock.
1073 	 * So, instead, we have to "doom" the znode in the illumos style.
1074 	 *
1075 	 * Ignore invalid pages during the scan.  This is to avoid deadlocks
1076 	 * between page busying and the teardown lock, as pages are busied prior
1077 	 * to a VOP_GETPAGES operation, which acquires the teardown read lock.
1078 	 * Such pages will be invalid and can safely be skipped here.
1079 	 */
1080 	vp = ZTOV(zp);
1081 #if __FreeBSD_version >= 1400042
1082 	vn_pages_remove_valid(vp, 0, 0);
1083 #else
1084 	vn_pages_remove(vp, 0, 0);
1085 #endif
1086 
1087 	ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num);
1088 
1089 	mutex_enter(&zp->z_acl_lock);
1090 	if (zp->z_acl_cached) {
1091 		zfs_acl_free(zp->z_acl_cached);
1092 		zp->z_acl_cached = NULL;
1093 	}
1094 	mutex_exit(&zp->z_acl_lock);
1095 
1096 	rw_enter(&zp->z_xattr_lock, RW_WRITER);
1097 	if (zp->z_xattr_cached) {
1098 		nvlist_free(zp->z_xattr_cached);
1099 		zp->z_xattr_cached = NULL;
1100 	}
1101 	rw_exit(&zp->z_xattr_lock);
1102 
1103 	ASSERT3P(zp->z_sa_hdl, ==, NULL);
1104 	err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db);
1105 	if (err) {
1106 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1107 		return (err);
1108 	}
1109 
1110 	dmu_object_info_from_db(db, &doi);
1111 	if (doi.doi_bonus_type != DMU_OT_SA &&
1112 	    (doi.doi_bonus_type != DMU_OT_ZNODE ||
1113 	    (doi.doi_bonus_type == DMU_OT_ZNODE &&
1114 	    doi.doi_bonus_size < sizeof (znode_phys_t)))) {
1115 		sa_buf_rele(db, NULL);
1116 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1117 		return (SET_ERROR(EINVAL));
1118 	}
1119 
1120 	zfs_znode_sa_init(zfsvfs, zp, db, doi.doi_bonus_type, NULL);
1121 	size = zp->z_size;
1122 
1123 	/* reload cached values */
1124 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL,
1125 	    &gen, sizeof (gen));
1126 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
1127 	    &zp->z_size, sizeof (zp->z_size));
1128 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL,
1129 	    &zp->z_links, sizeof (zp->z_links));
1130 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
1131 	    &zp->z_pflags, sizeof (zp->z_pflags));
1132 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
1133 	    &zp->z_atime, sizeof (zp->z_atime));
1134 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
1135 	    &zp->z_uid, sizeof (zp->z_uid));
1136 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL,
1137 	    &zp->z_gid, sizeof (zp->z_gid));
1138 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
1139 	    &mode, sizeof (mode));
1140 
1141 	if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) {
1142 		zfs_znode_dmu_fini(zp);
1143 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1144 		return (SET_ERROR(EIO));
1145 	}
1146 
1147 	zp->z_mode = mode;
1148 
1149 	if (gen != zp->z_gen) {
1150 		zfs_znode_dmu_fini(zp);
1151 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1152 		return (SET_ERROR(EIO));
1153 	}
1154 
1155 	/*
1156 	 * It is highly improbable but still quite possible that two
1157 	 * objects in different datasets are created with the same
1158 	 * object numbers and in transaction groups with the same
1159 	 * numbers.  znodes corresponding to those objects would
1160 	 * have the same z_id and z_gen, but their other attributes
1161 	 * may be different.
1162 	 * zfs recv -F may replace one of such objects with the other.
1163 	 * As a result file properties recorded in the replaced
1164 	 * object's vnode may no longer match the received object's
1165 	 * properties.  At present the only cached property is the
1166 	 * files type recorded in v_type.
1167 	 * So, handle this case by leaving the old vnode and znode
1168 	 * disassociated from the actual object.  A new vnode and a
1169 	 * znode will be created if the object is accessed
1170 	 * (e.g. via a look-up).  The old vnode and znode will be
1171 	 * recycled when the last vnode reference is dropped.
1172 	 */
1173 	if (vp->v_type != IFTOVT((mode_t)zp->z_mode)) {
1174 		zfs_znode_dmu_fini(zp);
1175 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1176 		return (SET_ERROR(EIO));
1177 	}
1178 
1179 	/*
1180 	 * If the file has zero links, then it has been unlinked on the send
1181 	 * side and it must be in the received unlinked set.
1182 	 * We call zfs_znode_dmu_fini() now to prevent any accesses to the
1183 	 * stale data and to prevent automatically removal of the file in
1184 	 * zfs_zinactive().  The file will be removed either when it is removed
1185 	 * on the send side and the next incremental stream is received or
1186 	 * when the unlinked set gets processed.
1187 	 */
1188 	zp->z_unlinked = (zp->z_links == 0);
1189 	if (zp->z_unlinked) {
1190 		zfs_znode_dmu_fini(zp);
1191 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1192 		return (0);
1193 	}
1194 
1195 	zp->z_blksz = doi.doi_data_block_size;
1196 	if (zp->z_size != size)
1197 		vnode_pager_setsize(vp, zp->z_size);
1198 
1199 	ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1200 
1201 	return (0);
1202 }
1203 
1204 void
zfs_znode_delete(znode_t * zp,dmu_tx_t * tx)1205 zfs_znode_delete(znode_t *zp, dmu_tx_t *tx)
1206 {
1207 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1208 	objset_t *os = zfsvfs->z_os;
1209 	uint64_t obj = zp->z_id;
1210 	uint64_t acl_obj = zfs_external_acl(zp);
1211 
1212 	ZFS_OBJ_HOLD_ENTER(zfsvfs, obj);
1213 	if (acl_obj) {
1214 		VERIFY(!zp->z_is_sa);
1215 		VERIFY0(dmu_object_free(os, acl_obj, tx));
1216 	}
1217 	VERIFY0(dmu_object_free(os, obj, tx));
1218 	zfs_znode_dmu_fini(zp);
1219 	ZFS_OBJ_HOLD_EXIT(zfsvfs, obj);
1220 }
1221 
1222 void
zfs_zinactive(znode_t * zp)1223 zfs_zinactive(znode_t *zp)
1224 {
1225 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1226 	uint64_t z_id = zp->z_id;
1227 
1228 	ASSERT3P(zp->z_sa_hdl, !=, NULL);
1229 
1230 	/*
1231 	 * Don't allow a zfs_zget() while were trying to release this znode
1232 	 */
1233 	ZFS_OBJ_HOLD_ENTER(zfsvfs, z_id);
1234 
1235 	/*
1236 	 * If this was the last reference to a file with no links, remove
1237 	 * the file from the file system unless the file system is mounted
1238 	 * read-only.  That can happen, for example, if the file system was
1239 	 * originally read-write, the file was opened, then unlinked and
1240 	 * the file system was made read-only before the file was finally
1241 	 * closed.  The file will remain in the unlinked set.
1242 	 */
1243 	if (zp->z_unlinked) {
1244 		ASSERT(!zfsvfs->z_issnap);
1245 		if ((zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) == 0) {
1246 			ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id);
1247 			zfs_rmnode(zp);
1248 			return;
1249 		}
1250 	}
1251 
1252 	zfs_znode_dmu_fini(zp);
1253 	ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id);
1254 	zfs_znode_free(zp);
1255 }
1256 
1257 void
zfs_znode_free(znode_t * zp)1258 zfs_znode_free(znode_t *zp)
1259 {
1260 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1261 	char *symlink;
1262 
1263 	ASSERT3P(zp->z_sa_hdl, ==, NULL);
1264 	zp->z_vnode = NULL;
1265 	mutex_enter(&zfsvfs->z_znodes_lock);
1266 	POINTER_INVALIDATE(&zp->z_zfsvfs);
1267 	list_remove(&zfsvfs->z_all_znodes, zp);
1268 	mutex_exit(&zfsvfs->z_znodes_lock);
1269 
1270 	symlink = atomic_load_ptr(&zp->z_cached_symlink);
1271 	if (symlink != NULL) {
1272 		atomic_store_rel_ptr((uintptr_t *)&zp->z_cached_symlink,
1273 		    (uintptr_t)NULL);
1274 		cache_symlink_free(symlink, strlen(symlink) + 1);
1275 	}
1276 
1277 	if (zp->z_acl_cached) {
1278 		zfs_acl_free(zp->z_acl_cached);
1279 		zp->z_acl_cached = NULL;
1280 	}
1281 
1282 	zfs_znode_free_kmem(zp);
1283 }
1284 
1285 void
zfs_tstamp_update_setup_ext(znode_t * zp,uint_t flag,uint64_t mtime[2],uint64_t ctime[2],boolean_t have_tx)1286 zfs_tstamp_update_setup_ext(znode_t *zp, uint_t flag, uint64_t mtime[2],
1287     uint64_t ctime[2], boolean_t have_tx)
1288 {
1289 	timestruc_t	now;
1290 
1291 	vfs_timestamp(&now);
1292 
1293 	if (have_tx) {	/* will sa_bulk_update happen really soon? */
1294 		zp->z_atime_dirty = 0;
1295 		zp->z_seq++;
1296 	} else {
1297 		zp->z_atime_dirty = 1;
1298 	}
1299 
1300 	if (flag & AT_ATIME) {
1301 		ZFS_TIME_ENCODE(&now, zp->z_atime);
1302 	}
1303 
1304 	if (flag & AT_MTIME) {
1305 		ZFS_TIME_ENCODE(&now, mtime);
1306 		if (zp->z_zfsvfs->z_use_fuids) {
1307 			zp->z_pflags |= (ZFS_ARCHIVE |
1308 			    ZFS_AV_MODIFIED);
1309 		}
1310 	}
1311 
1312 	if (flag & AT_CTIME) {
1313 		ZFS_TIME_ENCODE(&now, ctime);
1314 		if (zp->z_zfsvfs->z_use_fuids)
1315 			zp->z_pflags |= ZFS_ARCHIVE;
1316 	}
1317 }
1318 
1319 
1320 void
zfs_tstamp_update_setup(znode_t * zp,uint_t flag,uint64_t mtime[2],uint64_t ctime[2])1321 zfs_tstamp_update_setup(znode_t *zp, uint_t flag, uint64_t mtime[2],
1322     uint64_t ctime[2])
1323 {
1324 	zfs_tstamp_update_setup_ext(zp, flag, mtime, ctime, B_TRUE);
1325 }
1326 /*
1327  * Grow the block size for a file.
1328  *
1329  *	IN:	zp	- znode of file to free data in.
1330  *		size	- requested block size
1331  *		tx	- open transaction.
1332  *
1333  * NOTE: this function assumes that the znode is write locked.
1334  */
1335 void
zfs_grow_blocksize(znode_t * zp,uint64_t size,dmu_tx_t * tx)1336 zfs_grow_blocksize(znode_t *zp, uint64_t size, dmu_tx_t *tx)
1337 {
1338 	int		error;
1339 	u_longlong_t	dummy;
1340 
1341 	if (size <= zp->z_blksz)
1342 		return;
1343 	/*
1344 	 * If the file size is already greater than the current blocksize,
1345 	 * we will not grow.  If there is more than one block in a file,
1346 	 * the blocksize cannot change.
1347 	 */
1348 	if (zp->z_blksz && zp->z_size > zp->z_blksz)
1349 		return;
1350 
1351 	error = dmu_object_set_blocksize(zp->z_zfsvfs->z_os, zp->z_id,
1352 	    size, 0, tx);
1353 
1354 	if (error == ENOTSUP)
1355 		return;
1356 	ASSERT0(error);
1357 
1358 	/* What blocksize did we actually get? */
1359 	dmu_object_size_from_db(sa_get_db(zp->z_sa_hdl), &zp->z_blksz, &dummy);
1360 }
1361 
1362 /*
1363  * Increase the file length
1364  *
1365  *	IN:	zp	- znode of file to free data in.
1366  *		end	- new end-of-file
1367  *
1368  *	RETURN:	0 on success, error code on failure
1369  */
1370 static int
zfs_extend(znode_t * zp,uint64_t end)1371 zfs_extend(znode_t *zp, uint64_t end)
1372 {
1373 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1374 	dmu_tx_t *tx;
1375 	zfs_locked_range_t *lr;
1376 	uint64_t newblksz;
1377 	int error;
1378 
1379 	/*
1380 	 * We will change zp_size, lock the whole file.
1381 	 */
1382 	lr = zfs_rangelock_enter(&zp->z_rangelock, 0, UINT64_MAX, RL_WRITER);
1383 
1384 	/*
1385 	 * Nothing to do if file already at desired length.
1386 	 */
1387 	if (end <= zp->z_size) {
1388 		zfs_rangelock_exit(lr);
1389 		return (0);
1390 	}
1391 	tx = dmu_tx_create(zfsvfs->z_os);
1392 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1393 	zfs_sa_upgrade_txholds(tx, zp);
1394 	if (end > zp->z_blksz &&
1395 	    (!ISP2(zp->z_blksz) || zp->z_blksz < zfsvfs->z_max_blksz)) {
1396 		/*
1397 		 * We are growing the file past the current block size.
1398 		 */
1399 		if (zp->z_blksz > zp->z_zfsvfs->z_max_blksz) {
1400 			/*
1401 			 * File's blocksize is already larger than the
1402 			 * "recordsize" property.  Only let it grow to
1403 			 * the next power of 2.
1404 			 */
1405 			ASSERT(!ISP2(zp->z_blksz));
1406 			newblksz = MIN(end, 1 << highbit64(zp->z_blksz));
1407 		} else {
1408 			newblksz = MIN(end, zp->z_zfsvfs->z_max_blksz);
1409 		}
1410 		dmu_tx_hold_write(tx, zp->z_id, 0, newblksz);
1411 	} else {
1412 		newblksz = 0;
1413 	}
1414 
1415 	error = dmu_tx_assign(tx, DMU_TX_WAIT);
1416 	if (error) {
1417 		dmu_tx_abort(tx);
1418 		zfs_rangelock_exit(lr);
1419 		return (error);
1420 	}
1421 
1422 	if (newblksz)
1423 		zfs_grow_blocksize(zp, newblksz, tx);
1424 
1425 	zp->z_size = end;
1426 
1427 	VERIFY0(sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zp->z_zfsvfs),
1428 	    &zp->z_size, sizeof (zp->z_size), tx));
1429 
1430 	vnode_pager_setsize(ZTOV(zp), end);
1431 
1432 	zfs_rangelock_exit(lr);
1433 
1434 	dmu_tx_commit(tx);
1435 
1436 	return (0);
1437 }
1438 
1439 /*
1440  * Free space in a file.
1441  *
1442  *	IN:	zp	- znode of file to free data in.
1443  *		off	- start of section to free.
1444  *		len	- length of section to free.
1445  *
1446  *	RETURN:	0 on success, error code on failure
1447  */
1448 static int
zfs_free_range(znode_t * zp,uint64_t off,uint64_t len)1449 zfs_free_range(znode_t *zp, uint64_t off, uint64_t len)
1450 {
1451 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1452 	zfs_locked_range_t *lr;
1453 	int error;
1454 
1455 	/*
1456 	 * Lock the range being freed.
1457 	 */
1458 	lr = zfs_rangelock_enter(&zp->z_rangelock, off, len, RL_WRITER);
1459 
1460 	/*
1461 	 * Nothing to do if file already at desired length.
1462 	 */
1463 	if (off >= zp->z_size) {
1464 		zfs_rangelock_exit(lr);
1465 		return (0);
1466 	}
1467 
1468 	if (off + len > zp->z_size)
1469 		len = zp->z_size - off;
1470 
1471 	error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, off, len);
1472 
1473 	if (error == 0) {
1474 #if __FreeBSD_version >= 1400032
1475 		vnode_pager_purge_range(ZTOV(zp), off, off + len);
1476 #else
1477 		/*
1478 		 * Before __FreeBSD_version 1400032 we cannot free block in the
1479 		 * middle of a file, but only at the end of a file, so this code
1480 		 * path should never happen.
1481 		 */
1482 		vnode_pager_setsize(ZTOV(zp), off);
1483 #endif
1484 	}
1485 
1486 	zfs_rangelock_exit(lr);
1487 
1488 	return (error);
1489 }
1490 
1491 /*
1492  * Truncate a file
1493  *
1494  *	IN:	zp	- znode of file to free data in.
1495  *		end	- new end-of-file.
1496  *
1497  *	RETURN:	0 on success, error code on failure
1498  */
1499 static int
zfs_trunc(znode_t * zp,uint64_t end)1500 zfs_trunc(znode_t *zp, uint64_t end)
1501 {
1502 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1503 	vnode_t *vp = ZTOV(zp);
1504 	dmu_tx_t *tx;
1505 	zfs_locked_range_t *lr;
1506 	int error;
1507 	sa_bulk_attr_t bulk[2];
1508 	int count = 0;
1509 
1510 	/*
1511 	 * We will change zp_size, lock the whole file.
1512 	 */
1513 	lr = zfs_rangelock_enter(&zp->z_rangelock, 0, UINT64_MAX, RL_WRITER);
1514 
1515 	/*
1516 	 * Nothing to do if file already at desired length.
1517 	 */
1518 	if (end >= zp->z_size) {
1519 		zfs_rangelock_exit(lr);
1520 		return (0);
1521 	}
1522 
1523 	error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, end,
1524 	    DMU_OBJECT_END);
1525 	if (error) {
1526 		zfs_rangelock_exit(lr);
1527 		return (error);
1528 	}
1529 	tx = dmu_tx_create(zfsvfs->z_os);
1530 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1531 	zfs_sa_upgrade_txholds(tx, zp);
1532 	dmu_tx_mark_netfree(tx);
1533 	error = dmu_tx_assign(tx, DMU_TX_WAIT);
1534 	if (error) {
1535 		dmu_tx_abort(tx);
1536 		zfs_rangelock_exit(lr);
1537 		return (error);
1538 	}
1539 
1540 	zp->z_size = end;
1541 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs),
1542 	    NULL, &zp->z_size, sizeof (zp->z_size));
1543 
1544 	if (end == 0) {
1545 		zp->z_pflags &= ~ZFS_SPARSE;
1546 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
1547 		    NULL, &zp->z_pflags, 8);
1548 	}
1549 	VERIFY0(sa_bulk_update(zp->z_sa_hdl, bulk, count, tx));
1550 
1551 	dmu_tx_commit(tx);
1552 
1553 	/*
1554 	 * Clear any mapped pages in the truncated region.  This has to
1555 	 * happen outside of the transaction to avoid the possibility of
1556 	 * a deadlock with someone trying to push a page that we are
1557 	 * about to invalidate.
1558 	 */
1559 	vnode_pager_setsize(vp, end);
1560 
1561 	zfs_rangelock_exit(lr);
1562 
1563 	return (0);
1564 }
1565 
1566 /*
1567  * Free space in a file
1568  *
1569  *	IN:	zp	- znode of file to free data in.
1570  *		off	- start of range
1571  *		len	- end of range (0 => EOF)
1572  *		flag	- current file open mode flags.
1573  *		log	- TRUE if this action should be logged
1574  *
1575  *	RETURN:	0 on success, error code on failure
1576  */
1577 int
zfs_freesp(znode_t * zp,uint64_t off,uint64_t len,int flag,boolean_t log)1578 zfs_freesp(znode_t *zp, uint64_t off, uint64_t len, int flag, boolean_t log)
1579 {
1580 	dmu_tx_t *tx;
1581 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1582 	zilog_t *zilog = zfsvfs->z_log;
1583 	uint64_t mode;
1584 	uint64_t mtime[2], ctime[2];
1585 	sa_bulk_attr_t bulk[3];
1586 	int count = 0;
1587 	int error;
1588 
1589 	if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs), &mode,
1590 	    sizeof (mode))) != 0)
1591 		return (error);
1592 
1593 	if (off > zp->z_size) {
1594 		error =  zfs_extend(zp, off+len);
1595 		if (error == 0 && log)
1596 			goto log;
1597 		else
1598 			return (error);
1599 	}
1600 
1601 	if (len == 0) {
1602 		error = zfs_trunc(zp, off);
1603 	} else {
1604 		if ((error = zfs_free_range(zp, off, len)) == 0 &&
1605 		    off + len > zp->z_size)
1606 			error = zfs_extend(zp, off+len);
1607 	}
1608 	if (error || !log)
1609 		return (error);
1610 log:
1611 	tx = dmu_tx_create(zfsvfs->z_os);
1612 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1613 	zfs_sa_upgrade_txholds(tx, zp);
1614 	error = dmu_tx_assign(tx, DMU_TX_WAIT);
1615 	if (error) {
1616 		dmu_tx_abort(tx);
1617 		return (error);
1618 	}
1619 
1620 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, mtime, 16);
1621 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, ctime, 16);
1622 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
1623 	    NULL, &zp->z_pflags, 8);
1624 	zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime);
1625 	error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
1626 	ASSERT0(error);
1627 
1628 	zfs_log_truncate(zilog, tx, TX_TRUNCATE, zp, off, len);
1629 
1630 	dmu_tx_commit(tx);
1631 	return (0);
1632 }
1633 
1634 void
zfs_create_fs(objset_t * os,cred_t * cr,nvlist_t * zplprops,dmu_tx_t * tx)1635 zfs_create_fs(objset_t *os, cred_t *cr, nvlist_t *zplprops, dmu_tx_t *tx)
1636 {
1637 	uint64_t	moid, obj, sa_obj, version;
1638 	uint64_t	sense = ZFS_CASE_SENSITIVE;
1639 	uint64_t	norm = 0;
1640 	nvpair_t	*elem;
1641 	int		error;
1642 	int		i;
1643 	znode_t		*rootzp = NULL;
1644 	zfsvfs_t	*zfsvfs;
1645 	vattr_t		vattr;
1646 	znode_t		*zp;
1647 	zfs_acl_ids_t	acl_ids;
1648 
1649 	/*
1650 	 * First attempt to create master node.
1651 	 */
1652 	/*
1653 	 * In an empty objset, there are no blocks to read and thus
1654 	 * there can be no i/o errors (which we assert below).
1655 	 */
1656 	moid = MASTER_NODE_OBJ;
1657 	error = zap_create_claim(os, moid, DMU_OT_MASTER_NODE,
1658 	    DMU_OT_NONE, 0, tx);
1659 	ASSERT0(error);
1660 
1661 	/*
1662 	 * Set starting attributes.
1663 	 */
1664 	version = zfs_zpl_version_map(spa_version(dmu_objset_spa(os)));
1665 	elem = NULL;
1666 	while ((elem = nvlist_next_nvpair(zplprops, elem)) != NULL) {
1667 		/* For the moment we expect all zpl props to be uint64_ts */
1668 		uint64_t val;
1669 		const char *name;
1670 
1671 		ASSERT3S(nvpair_type(elem), ==, DATA_TYPE_UINT64);
1672 		val = fnvpair_value_uint64(elem);
1673 		name = nvpair_name(elem);
1674 		if (strcmp(name, zfs_prop_to_name(ZFS_PROP_VERSION)) == 0) {
1675 			if (val < version)
1676 				version = val;
1677 		} else {
1678 			error = zap_update(os, moid, name, 8, 1, &val, tx);
1679 		}
1680 		ASSERT0(error);
1681 		if (strcmp(name, zfs_prop_to_name(ZFS_PROP_NORMALIZE)) == 0)
1682 			norm = val;
1683 		else if (strcmp(name, zfs_prop_to_name(ZFS_PROP_CASE)) == 0)
1684 			sense = val;
1685 	}
1686 	ASSERT3U(version, !=, 0);
1687 	error = zap_update(os, moid, ZPL_VERSION_STR, 8, 1, &version, tx);
1688 	ASSERT0(error);
1689 
1690 	/*
1691 	 * Create zap object used for SA attribute registration
1692 	 */
1693 
1694 	if (version >= ZPL_VERSION_SA) {
1695 		sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
1696 		    DMU_OT_NONE, 0, tx);
1697 		error = zap_add(os, moid, ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
1698 		ASSERT0(error);
1699 	} else {
1700 		sa_obj = 0;
1701 	}
1702 	/*
1703 	 * Create a delete queue.
1704 	 */
1705 	obj = zap_create(os, DMU_OT_UNLINKED_SET, DMU_OT_NONE, 0, tx);
1706 
1707 	error = zap_add(os, moid, ZFS_UNLINKED_SET, 8, 1, &obj, tx);
1708 	ASSERT0(error);
1709 
1710 	/*
1711 	 * Create root znode.  Create minimal znode/vnode/zfsvfs
1712 	 * to allow zfs_mknode to work.
1713 	 */
1714 	VATTR_NULL(&vattr);
1715 	vattr.va_mask = AT_MODE|AT_UID|AT_GID;
1716 	vattr.va_type = VDIR;
1717 	vattr.va_mode = S_IFDIR|0755;
1718 	vattr.va_uid = crgetuid(cr);
1719 	vattr.va_gid = crgetgid(cr);
1720 
1721 	zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
1722 
1723 	rootzp = zfs_znode_alloc_kmem(KM_SLEEP);
1724 	ASSERT(!POINTER_IS_VALID(rootzp->z_zfsvfs));
1725 	rootzp->z_unlinked = 0;
1726 	rootzp->z_atime_dirty = 0;
1727 	rootzp->z_is_sa = USE_SA(version, os);
1728 
1729 	zfsvfs->z_os = os;
1730 	zfsvfs->z_parent = zfsvfs;
1731 	zfsvfs->z_version = version;
1732 	zfsvfs->z_use_fuids = USE_FUIDS(version, os);
1733 	zfsvfs->z_use_sa = USE_SA(version, os);
1734 	zfsvfs->z_norm = norm;
1735 
1736 	error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
1737 	    &zfsvfs->z_attr_table);
1738 
1739 	ASSERT0(error);
1740 
1741 	/*
1742 	 * Fold case on file systems that are always or sometimes case
1743 	 * insensitive.
1744 	 */
1745 	if (sense == ZFS_CASE_INSENSITIVE || sense == ZFS_CASE_MIXED)
1746 		zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER;
1747 
1748 	mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
1749 	list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
1750 	    offsetof(znode_t, z_link_node));
1751 
1752 	for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1753 		mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL);
1754 
1755 	rootzp->z_zfsvfs = zfsvfs;
1756 	VERIFY0(zfs_acl_ids_create(rootzp, IS_ROOT_NODE, &vattr,
1757 	    cr, NULL, &acl_ids, NULL));
1758 	zfs_mknode(rootzp, &vattr, tx, cr, IS_ROOT_NODE, &zp, &acl_ids);
1759 	ASSERT3P(zp, ==, rootzp);
1760 	error = zap_add(os, moid, ZFS_ROOT_OBJ, 8, 1, &rootzp->z_id, tx);
1761 	ASSERT0(error);
1762 	zfs_acl_ids_free(&acl_ids);
1763 	POINTER_INVALIDATE(&rootzp->z_zfsvfs);
1764 
1765 	sa_handle_destroy(rootzp->z_sa_hdl);
1766 	zfs_znode_free_kmem(rootzp);
1767 
1768 	/*
1769 	 * Create shares directory
1770 	 */
1771 
1772 	error = zfs_create_share_dir(zfsvfs, tx);
1773 
1774 	ASSERT0(error);
1775 
1776 	for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1777 		mutex_destroy(&zfsvfs->z_hold_mtx[i]);
1778 	kmem_free(zfsvfs, sizeof (zfsvfs_t));
1779 }
1780 
1781 void
zfs_znode_update_vfs(znode_t * zp)1782 zfs_znode_update_vfs(znode_t *zp)
1783 {
1784 	vm_object_t object;
1785 
1786 	if ((object = ZTOV(zp)->v_object) == NULL ||
1787 	    zp->z_size == object->un_pager.vnp.vnp_size)
1788 		return;
1789 
1790 	vnode_pager_setsize(ZTOV(zp), zp->z_size);
1791 }
1792 
1793 int
zfs_znode_parent_and_name(znode_t * zp,znode_t ** dzpp,char * buf,uint64_t buflen)1794 zfs_znode_parent_and_name(znode_t *zp, znode_t **dzpp, char *buf,
1795     uint64_t buflen)
1796 {
1797 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1798 	uint64_t parent;
1799 	int is_xattrdir;
1800 	int err;
1801 
1802 	/* Extended attributes should not be visible as regular files. */
1803 	if ((zp->z_pflags & ZFS_XATTR) != 0)
1804 		return (SET_ERROR(EINVAL));
1805 
1806 	err = zfs_obj_to_pobj(zfsvfs->z_os, zp->z_sa_hdl, zfsvfs->z_attr_table,
1807 	    &parent, &is_xattrdir);
1808 	if (err != 0)
1809 		return (err);
1810 	ASSERT0(is_xattrdir);
1811 
1812 	/* No name as this is a root object. */
1813 	if (parent == zp->z_id)
1814 		return (SET_ERROR(EINVAL));
1815 
1816 	err = zap_value_search(zfsvfs->z_os, parent, zp->z_id,
1817 	    ZFS_DIRENT_OBJ(-1ULL), buf, buflen);
1818 	if (err != 0)
1819 		return (err);
1820 	err = zfs_zget(zfsvfs, parent, dzpp);
1821 	return (err);
1822 }
1823 
1824 int
zfs_rlimit_fsize(off_t fsize)1825 zfs_rlimit_fsize(off_t fsize)
1826 {
1827 	struct thread *td = curthread;
1828 	off_t lim;
1829 
1830 	if (td == NULL)
1831 		return (0);
1832 
1833 	lim = lim_cur(td, RLIMIT_FSIZE);
1834 	if (__predict_true((uoff_t)fsize <= lim))
1835 		return (0);
1836 
1837 	/*
1838 	 * The limit is reached.
1839 	 */
1840 	PROC_LOCK(td->td_proc);
1841 	kern_psignal(td->td_proc, SIGXFSZ);
1842 	PROC_UNLOCK(td->td_proc);
1843 
1844 	return (EFBIG);
1845 }
1846