xref: /freebsd/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_znode_os.c (revision 0df1ee76950fc28c3c90a8aa85534fd44641c425)
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 #ifdef SYSCTL_SIZEOF
71 SYSCTL_SIZEOF(znode, znode_t);
72 #else
73 SYSCTL_INT(_debug_sizeof, OID_AUTO, znode, CTLFLAG_RD,
74 	SYSCTL_NULL_INT_PTR, sizeof (znode_t), "sizeof(znode_t)");
75 #endif
76 
77 /*
78  * Define ZNODE_STATS to turn on statistic gathering. By default, it is only
79  * turned on when DEBUG is also defined.
80  */
81 #ifdef	ZFS_DEBUG
82 #define	ZNODE_STATS
83 #endif	/* DEBUG */
84 
85 #ifdef	ZNODE_STATS
86 #define	ZNODE_STAT_ADD(stat)			((stat)++)
87 #else
88 #define	ZNODE_STAT_ADD(stat)			/* nothing */
89 #endif	/* ZNODE_STATS */
90 
91 #if !defined(KMEM_DEBUG)
92 #define	_ZFS_USE_SMR
93 static uma_zone_t znode_uma_zone;
94 #else
95 static kmem_cache_t *znode_cache = NULL;
96 #endif
97 
98 extern struct vop_vector zfs_vnodeops;
99 extern struct vop_vector zfs_fifoops;
100 extern struct vop_vector zfs_shareops;
101 
102 
103 /*
104  * This callback is invoked when acquiring a RL_WRITER or RL_APPEND lock on
105  * z_rangelock. It will modify the offset and length of the lock to reflect
106  * znode-specific information, and convert RL_APPEND to RL_WRITER.  This is
107  * called with the rangelock_t's rl_lock held, which avoids races.
108  */
109 static void
zfs_rangelock_cb(zfs_locked_range_t * new,void * arg)110 zfs_rangelock_cb(zfs_locked_range_t *new, void *arg)
111 {
112 	znode_t *zp = arg;
113 
114 	/*
115 	 * If in append mode, convert to writer and lock starting at the
116 	 * current end of file.
117 	 */
118 	if (new->lr_type == RL_APPEND) {
119 		new->lr_offset = zp->z_size;
120 		new->lr_type = RL_WRITER;
121 	}
122 
123 	/*
124 	 * If we need to grow the block size then lock the whole file range.
125 	 */
126 	uint64_t end_size = MAX(zp->z_size, new->lr_offset + new->lr_length);
127 	if (end_size > zp->z_blksz && (!ISP2(zp->z_blksz) ||
128 	    zp->z_blksz < ZTOZSB(zp)->z_max_blksz)) {
129 		new->lr_offset = 0;
130 		new->lr_length = UINT64_MAX;
131 	}
132 }
133 
134 static int
zfs_znode_cache_constructor(void * buf,void * arg,int kmflags)135 zfs_znode_cache_constructor(void *buf, void *arg, int kmflags)
136 {
137 	znode_t *zp = buf;
138 
139 	POINTER_INVALIDATE(&zp->z_zfsvfs);
140 
141 	list_link_init(&zp->z_link_node);
142 
143 	mutex_init(&zp->z_lock, NULL, MUTEX_DEFAULT, NULL);
144 	mutex_init(&zp->z_acl_lock, NULL, MUTEX_DEFAULT, NULL);
145 	rw_init(&zp->z_xattr_lock, NULL, RW_DEFAULT, NULL);
146 
147 	zfs_rangelock_init(&zp->z_rangelock, zfs_rangelock_cb, zp);
148 
149 	zp->z_acl_cached = NULL;
150 	zp->z_xattr_cached = NULL;
151 	zp->z_xattr_parent = 0;
152 	zp->z_vnode = NULL;
153 	zp->z_sync_writes_cnt = 0;
154 	zp->z_async_writes_cnt = 0;
155 
156 	return (0);
157 }
158 
159 static void
zfs_znode_cache_destructor(void * buf,void * arg)160 zfs_znode_cache_destructor(void *buf, void *arg)
161 {
162 	(void) arg;
163 	znode_t *zp = buf;
164 
165 	ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs));
166 	ASSERT3P(zp->z_vnode, ==, NULL);
167 	ASSERT(!list_link_active(&zp->z_link_node));
168 	mutex_destroy(&zp->z_lock);
169 	mutex_destroy(&zp->z_acl_lock);
170 	rw_destroy(&zp->z_xattr_lock);
171 	zfs_rangelock_fini(&zp->z_rangelock);
172 
173 	ASSERT3P(zp->z_acl_cached, ==, NULL);
174 	ASSERT3P(zp->z_xattr_cached, ==, NULL);
175 
176 	ASSERT0(atomic_load_32(&zp->z_sync_writes_cnt));
177 	ASSERT0(atomic_load_32(&zp->z_async_writes_cnt));
178 }
179 
180 
181 #ifdef _ZFS_USE_SMR
182 VFS_SMR_DECLARE;
183 
184 static int
zfs_znode_cache_constructor_smr(void * mem,int size __unused,void * private,int flags)185 zfs_znode_cache_constructor_smr(void *mem, int size __unused, void *private,
186     int flags)
187 {
188 	return (zfs_znode_cache_constructor(mem, private, flags));
189 }
190 
191 static void
zfs_znode_cache_destructor_smr(void * mem,int size __unused,void * private)192 zfs_znode_cache_destructor_smr(void *mem, int size __unused, void *private)
193 {
194 	zfs_znode_cache_destructor(mem, private);
195 }
196 
197 void
zfs_znode_init(void)198 zfs_znode_init(void)
199 {
200 	/*
201 	 * Initialize zcache
202 	 */
203 	ASSERT3P(znode_uma_zone, ==, NULL);
204 	znode_uma_zone = uma_zcreate("zfs_znode_cache",
205 	    sizeof (znode_t), zfs_znode_cache_constructor_smr,
206 	    zfs_znode_cache_destructor_smr, NULL, NULL, 0, 0);
207 	VFS_SMR_ZONE_SET(znode_uma_zone);
208 }
209 
210 static znode_t *
zfs_znode_alloc_kmem(int flags)211 zfs_znode_alloc_kmem(int flags)
212 {
213 	return (uma_zalloc_smr(znode_uma_zone, flags));
214 }
215 
216 static void
zfs_znode_free_kmem(znode_t * zp)217 zfs_znode_free_kmem(znode_t *zp)
218 {
219 	if (zp->z_xattr_cached) {
220 		nvlist_free(zp->z_xattr_cached);
221 		zp->z_xattr_cached = NULL;
222 	}
223 	uma_zfree_smr(znode_uma_zone, zp);
224 }
225 #else
226 void
zfs_znode_init(void)227 zfs_znode_init(void)
228 {
229 	/*
230 	 * Initialize zcache
231 	 */
232 	ASSERT3P(znode_cache, ==, NULL);
233 	znode_cache = kmem_cache_create("zfs_znode_cache",
234 	    sizeof (znode_t), 0, zfs_znode_cache_constructor,
235 	    zfs_znode_cache_destructor, NULL, NULL, NULL, KMC_RECLAIMABLE);
236 }
237 
238 static znode_t *
zfs_znode_alloc_kmem(int flags)239 zfs_znode_alloc_kmem(int flags)
240 {
241 	return (kmem_cache_alloc(znode_cache, flags));
242 }
243 
244 static void
zfs_znode_free_kmem(znode_t * zp)245 zfs_znode_free_kmem(znode_t *zp)
246 {
247 	if (zp->z_xattr_cached) {
248 		nvlist_free(zp->z_xattr_cached);
249 		zp->z_xattr_cached = NULL;
250 	}
251 	kmem_cache_free(znode_cache, zp);
252 }
253 #endif
254 
255 void
zfs_znode_fini(void)256 zfs_znode_fini(void)
257 {
258 	/*
259 	 * Cleanup zcache
260 	 */
261 #ifdef _ZFS_USE_SMR
262 	if (znode_uma_zone) {
263 		uma_zdestroy(znode_uma_zone);
264 		znode_uma_zone = NULL;
265 	}
266 #else
267 	if (znode_cache) {
268 		kmem_cache_destroy(znode_cache);
269 		znode_cache = NULL;
270 	}
271 #endif
272 }
273 
274 
275 static int
zfs_create_share_dir(zfsvfs_t * zfsvfs,dmu_tx_t * tx)276 zfs_create_share_dir(zfsvfs_t *zfsvfs, dmu_tx_t *tx)
277 {
278 	zfs_acl_ids_t acl_ids;
279 	vattr_t vattr;
280 	znode_t *sharezp;
281 	znode_t *zp;
282 	int error;
283 
284 	vattr.va_mask = AT_MODE|AT_UID|AT_GID;
285 	vattr.va_type = VDIR;
286 	vattr.va_mode = S_IFDIR|0555;
287 	vattr.va_uid = crgetuid(kcred);
288 	vattr.va_gid = crgetgid(kcred);
289 
290 	sharezp = zfs_znode_alloc_kmem(KM_SLEEP);
291 	ASSERT(!POINTER_IS_VALID(sharezp->z_zfsvfs));
292 	sharezp->z_unlinked = 0;
293 	sharezp->z_atime_dirty = 0;
294 	sharezp->z_zfsvfs = zfsvfs;
295 	sharezp->z_is_sa = zfsvfs->z_use_sa;
296 	sharezp->z_pflags = 0;
297 
298 	VERIFY0(zfs_acl_ids_create(sharezp, IS_ROOT_NODE, &vattr,
299 	    kcred, NULL, &acl_ids, NULL));
300 	zfs_mknode(sharezp, &vattr, tx, kcred, IS_ROOT_NODE, &zp, &acl_ids);
301 	ASSERT3P(zp, ==, sharezp);
302 	POINTER_INVALIDATE(&sharezp->z_zfsvfs);
303 	error = zap_add(zfsvfs->z_os, MASTER_NODE_OBJ,
304 	    ZFS_SHARES_DIR, 8, 1, &sharezp->z_id, tx);
305 	zfsvfs->z_shares_dir = sharezp->z_id;
306 
307 	zfs_acl_ids_free(&acl_ids);
308 	sa_handle_destroy(sharezp->z_sa_hdl);
309 	zfs_znode_free_kmem(sharezp);
310 
311 	return (error);
312 }
313 
314 /*
315  * define a couple of values we need available
316  * for both 64 and 32 bit environments.
317  */
318 #ifndef NBITSMINOR64
319 #define	NBITSMINOR64	32
320 #endif
321 #ifndef MAXMAJ64
322 #define	MAXMAJ64	0xffffffffUL
323 #endif
324 #ifndef	MAXMIN64
325 #define	MAXMIN64	0xffffffffUL
326 #endif
327 
328 /*
329  * Create special expldev for ZFS private use.
330  * Can't use standard expldev since it doesn't do
331  * what we want.  The standard expldev() takes a
332  * dev32_t in LP64 and expands it to a long dev_t.
333  * We need an interface that takes a dev32_t in ILP32
334  * and expands it to a long dev_t.
335  */
336 static uint64_t
zfs_expldev(dev_t dev)337 zfs_expldev(dev_t dev)
338 {
339 	return (((uint64_t)major(dev) << NBITSMINOR64) | minor(dev));
340 }
341 /*
342  * Special cmpldev for ZFS private use.
343  * Can't use standard cmpldev since it takes
344  * a long dev_t and compresses it to dev32_t in
345  * LP64.  We need to do a compaction of a long dev_t
346  * to a dev32_t in ILP32.
347  */
348 dev_t
zfs_cmpldev(uint64_t dev)349 zfs_cmpldev(uint64_t dev)
350 {
351 	return (makedev((dev >> NBITSMINOR64), (dev & MAXMIN64)));
352 }
353 
354 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)355 zfs_znode_sa_init(zfsvfs_t *zfsvfs, znode_t *zp,
356     dmu_buf_t *db, dmu_object_type_t obj_type, sa_handle_t *sa_hdl)
357 {
358 	ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs) || (zfsvfs == zp->z_zfsvfs));
359 	ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zfsvfs, zp->z_id)));
360 
361 	ASSERT3P(zp->z_sa_hdl, ==, NULL);
362 	ASSERT3P(zp->z_acl_cached, ==, NULL);
363 	if (sa_hdl == NULL) {
364 		VERIFY0(sa_handle_get_from_db(zfsvfs->z_os, db, zp,
365 		    SA_HDL_SHARED, &zp->z_sa_hdl));
366 	} else {
367 		zp->z_sa_hdl = sa_hdl;
368 		sa_set_userp(sa_hdl, zp);
369 	}
370 
371 	zp->z_is_sa = (obj_type == DMU_OT_SA) ? B_TRUE : B_FALSE;
372 
373 	/*
374 	 * Slap on VROOT if we are the root znode unless we are the root
375 	 * node of a snapshot mounted under .zfs.
376 	 */
377 	if (zp->z_id == zfsvfs->z_root && zfsvfs->z_parent == zfsvfs)
378 		ZTOV(zp)->v_flag |= VROOT;
379 }
380 
381 void
zfs_znode_dmu_fini(znode_t * zp)382 zfs_znode_dmu_fini(znode_t *zp)
383 {
384 	ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zp->z_zfsvfs, zp->z_id)) ||
385 	    ZFS_TEARDOWN_INACTIVE_WRITE_HELD(zp->z_zfsvfs));
386 
387 	sa_handle_destroy(zp->z_sa_hdl);
388 	zp->z_sa_hdl = NULL;
389 }
390 
391 static void
zfs_vnode_forget(vnode_t * vp)392 zfs_vnode_forget(vnode_t *vp)
393 {
394 
395 	/* copied from insmntque_stddtr */
396 	vp->v_data = NULL;
397 	vp->v_op = &dead_vnodeops;
398 	vgone(vp);
399 	vput(vp);
400 }
401 
402 /*
403  * Construct a new znode/vnode and initialize.
404  *
405  * This does not do a call to dmu_set_user() that is
406  * up to the caller to do, in case you don't want to
407  * return the znode
408  */
409 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)410 zfs_znode_alloc(zfsvfs_t *zfsvfs, dmu_buf_t *db, int blksz,
411     dmu_object_type_t obj_type, sa_handle_t *hdl)
412 {
413 	znode_t	*zp;
414 	vnode_t *vp;
415 	uint64_t mode;
416 	uint64_t parent;
417 #ifdef notyet
418 	uint64_t mtime[2], ctime[2];
419 #endif
420 	uint64_t projid = ZFS_DEFAULT_PROJID;
421 	sa_bulk_attr_t bulk[9];
422 	int count = 0;
423 	int error;
424 
425 	zp = zfs_znode_alloc_kmem(KM_SLEEP);
426 
427 #ifndef _ZFS_USE_SMR
428 	KASSERT((zfsvfs->z_parent->z_vfs->mnt_kern_flag & MNTK_FPLOOKUP) == 0,
429 	    ("%s: fast path lookup enabled without smr", __func__));
430 #endif
431 
432 	KASSERT(curthread->td_vp_reserved != NULL,
433 	    ("zfs_znode_alloc: getnewvnode without any vnodes reserved"));
434 	error = getnewvnode("zfs", zfsvfs->z_parent->z_vfs, &zfs_vnodeops, &vp);
435 	if (error != 0) {
436 		zfs_znode_free_kmem(zp);
437 		return (NULL);
438 	}
439 	zp->z_vnode = vp;
440 	vp->v_data = zp;
441 
442 	/*
443 	 * Acquire the vnode lock before any possible interaction with the
444 	 * outside world.  Specifically, there is an error path that calls
445 	 * zfs_vnode_forget() and the vnode should be exclusively locked.
446 	 */
447 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
448 
449 	ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs));
450 
451 	zp->z_sa_hdl = NULL;
452 	zp->z_unlinked = 0;
453 	zp->z_atime_dirty = 0;
454 	zp->z_mapcnt = 0;
455 	zp->z_id = db->db_object;
456 	zp->z_blksz = blksz;
457 	zp->z_seq = 0x7A4653;
458 	zp->z_sync_cnt = 0;
459 	zp->z_sync_writes_cnt = 0;
460 	zp->z_async_writes_cnt = 0;
461 	atomic_store_ptr(&zp->z_cached_symlink, NULL);
462 
463 	zfs_znode_sa_init(zfsvfs, zp, db, obj_type, hdl);
464 
465 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8);
466 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL, &zp->z_gen, 8);
467 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
468 	    &zp->z_size, 8);
469 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL,
470 	    &zp->z_links, 8);
471 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
472 	    &zp->z_pflags, 8);
473 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL, &parent, 8);
474 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
475 	    &zp->z_atime, 16);
476 #ifdef notyet
477 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
478 	    &mtime, 16);
479 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
480 	    &ctime, 16);
481 #endif
482 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
483 	    &zp->z_uid, 8);
484 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL,
485 	    &zp->z_gid, 8);
486 
487 	if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count) != 0 || zp->z_gen == 0 ||
488 	    (dmu_objset_projectquota_enabled(zfsvfs->z_os) &&
489 	    (zp->z_pflags & ZFS_PROJID) &&
490 	    sa_lookup(zp->z_sa_hdl, SA_ZPL_PROJID(zfsvfs), &projid, 8) != 0)) {
491 		if (hdl == NULL)
492 			sa_handle_destroy(zp->z_sa_hdl);
493 		zfs_vnode_forget(vp);
494 		zp->z_vnode = NULL;
495 		zfs_znode_free_kmem(zp);
496 		return (NULL);
497 	}
498 
499 	zp->z_projid = projid;
500 	zp->z_mode = mode;
501 
502 	/* Cache the xattr parent id */
503 	if (zp->z_pflags & ZFS_XATTR)
504 		zp->z_xattr_parent = parent;
505 
506 	vp->v_type = IFTOVT((mode_t)mode);
507 
508 	switch (vp->v_type) {
509 	case VDIR:
510 		zp->z_zn_prefetch = B_TRUE; /* z_prefetch default is enabled */
511 		break;
512 	case VFIFO:
513 		vp->v_op = &zfs_fifoops;
514 		break;
515 	case VREG:
516 		if (parent == zfsvfs->z_shares_dir) {
517 			ASSERT0(zp->z_uid);
518 			ASSERT0(zp->z_gid);
519 			vp->v_op = &zfs_shareops;
520 		}
521 		break;
522 	default:
523 			break;
524 	}
525 
526 	mutex_enter(&zfsvfs->z_znodes_lock);
527 	list_insert_tail(&zfsvfs->z_all_znodes, zp);
528 	zp->z_zfsvfs = zfsvfs;
529 	mutex_exit(&zfsvfs->z_znodes_lock);
530 
531 #if __FreeBSD_version >= 1400077
532 	vn_set_state(vp, VSTATE_CONSTRUCTED);
533 #endif
534 	VN_LOCK_AREC(vp);
535 	if (vp->v_type != VFIFO)
536 		VN_LOCK_ASHARE(vp);
537 
538 	return (zp);
539 }
540 
541 static uint64_t empty_xattr;
542 static uint64_t pad[4];
543 static zfs_acl_phys_t acl_phys;
544 /*
545  * Create a new DMU object to hold a zfs znode.
546  *
547  *	IN:	dzp	- parent directory for new znode
548  *		vap	- file attributes for new znode
549  *		tx	- dmu transaction id for zap operations
550  *		cr	- credentials of caller
551  *		flag	- flags:
552  *			  IS_ROOT_NODE	- new object will be root
553  *			  IS_XATTR	- new object is an attribute
554  *		bonuslen - length of bonus buffer
555  *		setaclp  - File/Dir initial ACL
556  *		fuidp	 - Tracks fuid allocation.
557  *
558  *	OUT:	zpp	- allocated znode
559  *
560  */
561 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)562 zfs_mknode(znode_t *dzp, vattr_t *vap, dmu_tx_t *tx, cred_t *cr,
563     uint_t flag, znode_t **zpp, zfs_acl_ids_t *acl_ids)
564 {
565 	uint64_t	crtime[2], atime[2], mtime[2], ctime[2];
566 	uint64_t	mode, size, links, parent, pflags;
567 	uint64_t	dzp_pflags = 0;
568 	uint64_t	projid = ZFS_DEFAULT_PROJID;
569 	uint64_t	rdev = 0;
570 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
571 	dmu_buf_t	*db;
572 	timestruc_t	now;
573 	uint64_t	gen, obj;
574 	int		bonuslen;
575 	int		dnodesize;
576 	sa_handle_t	*sa_hdl;
577 	dmu_object_type_t obj_type;
578 	sa_bulk_attr_t	*sa_attrs;
579 	int		cnt = 0;
580 	zfs_acl_locator_cb_t locate = { 0 };
581 
582 	ASSERT3P(vap, !=, NULL);
583 	ASSERT3U((vap->va_mask & AT_MODE), ==, AT_MODE);
584 
585 	if (zfsvfs->z_replay) {
586 		obj = vap->va_nodeid;
587 		now = vap->va_ctime;		/* see zfs_replay_create() */
588 		gen = vap->va_nblocks;		/* ditto */
589 		dnodesize = vap->va_fsid;	/* ditto */
590 	} else {
591 		obj = 0;
592 		vfs_timestamp(&now);
593 		gen = dmu_tx_get_txg(tx);
594 		dnodesize = dmu_objset_dnodesize(zfsvfs->z_os);
595 	}
596 
597 	if (dnodesize == 0)
598 		dnodesize = DNODE_MIN_SIZE;
599 
600 	obj_type = zfsvfs->z_use_sa ? DMU_OT_SA : DMU_OT_ZNODE;
601 	bonuslen = (obj_type == DMU_OT_SA) ?
602 	    DN_BONUS_SIZE(dnodesize) : ZFS_OLD_ZNODE_PHYS_SIZE;
603 
604 	/*
605 	 * Create a new DMU object.
606 	 */
607 	/*
608 	 * There's currently no mechanism for pre-reading the blocks that will
609 	 * be needed to allocate a new object, so we accept the small chance
610 	 * that there will be an i/o error and we will fail one of the
611 	 * assertions below.
612 	 */
613 	if (vap->va_type == VDIR) {
614 		if (zfsvfs->z_replay) {
615 			VERIFY0(zap_create_claim_norm_dnsize(zfsvfs->z_os, obj,
616 			    zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS,
617 			    obj_type, bonuslen, dnodesize, tx));
618 		} else {
619 			obj = zap_create_norm_dnsize(zfsvfs->z_os,
620 			    zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS,
621 			    obj_type, bonuslen, dnodesize, tx);
622 		}
623 	} else {
624 		if (zfsvfs->z_replay) {
625 			VERIFY0(dmu_object_claim_dnsize(zfsvfs->z_os, obj,
626 			    DMU_OT_PLAIN_FILE_CONTENTS, 0,
627 			    obj_type, bonuslen, dnodesize, tx));
628 		} else {
629 			obj = dmu_object_alloc_dnsize(zfsvfs->z_os,
630 			    DMU_OT_PLAIN_FILE_CONTENTS, 0,
631 			    obj_type, bonuslen, dnodesize, tx);
632 		}
633 	}
634 
635 	ZFS_OBJ_HOLD_ENTER(zfsvfs, obj);
636 	VERIFY0(sa_buf_hold(zfsvfs->z_os, obj, NULL, &db));
637 
638 	/*
639 	 * If this is the root, fix up the half-initialized parent pointer
640 	 * to reference the just-allocated physical data area.
641 	 */
642 	if (flag & IS_ROOT_NODE) {
643 		dzp->z_id = obj;
644 	} else {
645 		dzp_pflags = dzp->z_pflags;
646 	}
647 
648 	/*
649 	 * If parent is an xattr, so am I.
650 	 */
651 	if (dzp_pflags & ZFS_XATTR) {
652 		flag |= IS_XATTR;
653 	}
654 
655 	if (zfsvfs->z_use_fuids)
656 		pflags = ZFS_ARCHIVE | ZFS_AV_MODIFIED;
657 	else
658 		pflags = 0;
659 
660 	if (vap->va_type == VDIR) {
661 		size = 2;		/* contents ("." and "..") */
662 		links = (flag & (IS_ROOT_NODE | IS_XATTR)) ? 2 : 1;
663 	} else {
664 		size = links = 0;
665 	}
666 
667 	if (vap->va_type == VBLK || vap->va_type == VCHR) {
668 		rdev = zfs_expldev(vap->va_rdev);
669 	}
670 
671 	parent = dzp->z_id;
672 	mode = acl_ids->z_mode;
673 	if (flag & IS_XATTR)
674 		pflags |= ZFS_XATTR;
675 
676 	if (vap->va_type == VREG || vap->va_type == VDIR) {
677 		/*
678 		 * With ZFS_PROJID flag, we can easily know whether there is
679 		 * project ID stored on disk or not. See zpl_get_file_info().
680 		 */
681 		if (obj_type != DMU_OT_ZNODE &&
682 		    dmu_objset_projectquota_enabled(zfsvfs->z_os))
683 			pflags |= ZFS_PROJID;
684 
685 		/*
686 		 * Inherit project ID from parent if required.
687 		 */
688 		projid = zfs_inherit_projid(dzp);
689 		if (dzp_pflags & ZFS_PROJINHERIT)
690 			pflags |= ZFS_PROJINHERIT;
691 	}
692 
693 	/*
694 	 * No execs denied will be determined when zfs_mode_compute() is called.
695 	 */
696 	pflags |= acl_ids->z_aclp->z_hints &
697 	    (ZFS_ACL_TRIVIAL|ZFS_INHERIT_ACE|ZFS_ACL_AUTO_INHERIT|
698 	    ZFS_ACL_DEFAULTED|ZFS_ACL_PROTECTED);
699 
700 	ZFS_TIME_ENCODE(&now, crtime);
701 	ZFS_TIME_ENCODE(&now, ctime);
702 
703 	if (vap->va_mask & AT_ATIME) {
704 		ZFS_TIME_ENCODE(&vap->va_atime, atime);
705 	} else {
706 		ZFS_TIME_ENCODE(&now, atime);
707 	}
708 
709 	if (vap->va_mask & AT_MTIME) {
710 		ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
711 	} else {
712 		ZFS_TIME_ENCODE(&now, mtime);
713 	}
714 
715 	/* Now add in all of the "SA" attributes */
716 	VERIFY0(sa_handle_get_from_db(zfsvfs->z_os, db, NULL, SA_HDL_SHARED,
717 	    &sa_hdl));
718 
719 	/*
720 	 * Setup the array of attributes to be replaced/set on the new file
721 	 *
722 	 * order for  DMU_OT_ZNODE is critical since it needs to be constructed
723 	 * in the old znode_phys_t format.  Don't change this ordering
724 	 */
725 	sa_attrs = kmem_alloc(sizeof (sa_bulk_attr_t) * ZPL_END, KM_SLEEP);
726 
727 	if (obj_type == DMU_OT_ZNODE) {
728 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs),
729 		    NULL, &atime, 16);
730 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs),
731 		    NULL, &mtime, 16);
732 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs),
733 		    NULL, &ctime, 16);
734 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs),
735 		    NULL, &crtime, 16);
736 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs),
737 		    NULL, &gen, 8);
738 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs),
739 		    NULL, &mode, 8);
740 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs),
741 		    NULL, &size, 8);
742 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs),
743 		    NULL, &parent, 8);
744 	} else {
745 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs),
746 		    NULL, &mode, 8);
747 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs),
748 		    NULL, &size, 8);
749 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs),
750 		    NULL, &gen, 8);
751 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs),
752 		    NULL, &acl_ids->z_fuid, 8);
753 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs),
754 		    NULL, &acl_ids->z_fgid, 8);
755 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs),
756 		    NULL, &parent, 8);
757 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs),
758 		    NULL, &pflags, 8);
759 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs),
760 		    NULL, &atime, 16);
761 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs),
762 		    NULL, &mtime, 16);
763 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs),
764 		    NULL, &ctime, 16);
765 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs),
766 		    NULL, &crtime, 16);
767 	}
768 
769 	SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_LINKS(zfsvfs), NULL, &links, 8);
770 
771 	if (obj_type == DMU_OT_ZNODE) {
772 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_XATTR(zfsvfs), NULL,
773 		    &empty_xattr, 8);
774 	} else if (dmu_objset_projectquota_enabled(zfsvfs->z_os) &&
775 	    pflags & ZFS_PROJID) {
776 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PROJID(zfsvfs),
777 		    NULL, &projid, 8);
778 	}
779 	if (obj_type == DMU_OT_ZNODE ||
780 	    (vap->va_type == VBLK || vap->va_type == VCHR)) {
781 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_RDEV(zfsvfs),
782 		    NULL, &rdev, 8);
783 
784 	}
785 	if (obj_type == DMU_OT_ZNODE) {
786 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs),
787 		    NULL, &pflags, 8);
788 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs), NULL,
789 		    &acl_ids->z_fuid, 8);
790 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs), NULL,
791 		    &acl_ids->z_fgid, 8);
792 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PAD(zfsvfs), NULL, pad,
793 		    sizeof (uint64_t) * 4);
794 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ZNODE_ACL(zfsvfs), NULL,
795 		    &acl_phys, sizeof (zfs_acl_phys_t));
796 	} else if (acl_ids->z_aclp->z_version >= ZFS_ACL_VERSION_FUID) {
797 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_COUNT(zfsvfs), NULL,
798 		    &acl_ids->z_aclp->z_acl_count, 8);
799 		locate.cb_aclp = acl_ids->z_aclp;
800 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_ACES(zfsvfs),
801 		    zfs_acl_data_locator, &locate,
802 		    acl_ids->z_aclp->z_acl_bytes);
803 		mode = zfs_mode_compute(mode, acl_ids->z_aclp, &pflags,
804 		    acl_ids->z_fuid, acl_ids->z_fgid);
805 	}
806 
807 	VERIFY0(sa_replace_all_by_template(sa_hdl, sa_attrs, cnt, tx));
808 
809 	if (!(flag & IS_ROOT_NODE)) {
810 		*zpp = zfs_znode_alloc(zfsvfs, db, 0, obj_type, sa_hdl);
811 		ASSERT3P(*zpp, !=, NULL);
812 	} else {
813 		/*
814 		 * If we are creating the root node, the "parent" we
815 		 * passed in is the znode for the root.
816 		 */
817 		*zpp = dzp;
818 
819 		(*zpp)->z_sa_hdl = sa_hdl;
820 	}
821 
822 	(*zpp)->z_pflags = pflags;
823 	(*zpp)->z_mode = mode;
824 	(*zpp)->z_dnodesize = dnodesize;
825 	(*zpp)->z_projid = projid;
826 
827 	if (vap->va_mask & AT_XVATTR)
828 		zfs_xvattr_set(*zpp, (xvattr_t *)vap, tx);
829 
830 	if (obj_type == DMU_OT_ZNODE ||
831 	    acl_ids->z_aclp->z_version < ZFS_ACL_VERSION_FUID) {
832 		VERIFY0(zfs_aclset_common(*zpp, acl_ids->z_aclp, cr, tx));
833 	}
834 	if (!(flag & IS_ROOT_NODE)) {
835 		vnode_t *vp = ZTOV(*zpp);
836 		vp->v_vflag |= VV_FORCEINSMQ;
837 		int err = insmntque(vp, zfsvfs->z_vfs);
838 		vp->v_vflag &= ~VV_FORCEINSMQ;
839 		(void) err;
840 		KASSERT(err == 0, ("insmntque() failed: error %d", err));
841 	}
842 	kmem_free(sa_attrs, sizeof (sa_bulk_attr_t) * ZPL_END);
843 	ZFS_OBJ_HOLD_EXIT(zfsvfs, obj);
844 }
845 
846 /*
847  * Update in-core attributes.  It is assumed the caller will be doing an
848  * sa_bulk_update to push the changes out.
849  */
850 void
zfs_xvattr_set(znode_t * zp,xvattr_t * xvap,dmu_tx_t * tx)851 zfs_xvattr_set(znode_t *zp, xvattr_t *xvap, dmu_tx_t *tx)
852 {
853 	xoptattr_t *xoap;
854 
855 	xoap = xva_getxoptattr(xvap);
856 	ASSERT3P(xoap, !=, NULL);
857 
858 	if (zp->z_zfsvfs->z_replay == B_FALSE) {
859 		ASSERT_VOP_IN_SEQC(ZTOV(zp));
860 	}
861 
862 	if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
863 		uint64_t times[2];
864 		ZFS_TIME_ENCODE(&xoap->xoa_createtime, times);
865 		(void) sa_update(zp->z_sa_hdl, SA_ZPL_CRTIME(zp->z_zfsvfs),
866 		    &times, sizeof (times), tx);
867 		XVA_SET_RTN(xvap, XAT_CREATETIME);
868 	}
869 	if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
870 		ZFS_ATTR_SET(zp, ZFS_READONLY, xoap->xoa_readonly,
871 		    zp->z_pflags, tx);
872 		XVA_SET_RTN(xvap, XAT_READONLY);
873 	}
874 	if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
875 		ZFS_ATTR_SET(zp, ZFS_HIDDEN, xoap->xoa_hidden,
876 		    zp->z_pflags, tx);
877 		XVA_SET_RTN(xvap, XAT_HIDDEN);
878 	}
879 	if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
880 		ZFS_ATTR_SET(zp, ZFS_SYSTEM, xoap->xoa_system,
881 		    zp->z_pflags, tx);
882 		XVA_SET_RTN(xvap, XAT_SYSTEM);
883 	}
884 	if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
885 		ZFS_ATTR_SET(zp, ZFS_ARCHIVE, xoap->xoa_archive,
886 		    zp->z_pflags, tx);
887 		XVA_SET_RTN(xvap, XAT_ARCHIVE);
888 	}
889 	if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
890 		ZFS_ATTR_SET(zp, ZFS_IMMUTABLE, xoap->xoa_immutable,
891 		    zp->z_pflags, tx);
892 		XVA_SET_RTN(xvap, XAT_IMMUTABLE);
893 	}
894 	if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
895 		ZFS_ATTR_SET(zp, ZFS_NOUNLINK, xoap->xoa_nounlink,
896 		    zp->z_pflags, tx);
897 		XVA_SET_RTN(xvap, XAT_NOUNLINK);
898 	}
899 	if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
900 		ZFS_ATTR_SET(zp, ZFS_APPENDONLY, xoap->xoa_appendonly,
901 		    zp->z_pflags, tx);
902 		XVA_SET_RTN(xvap, XAT_APPENDONLY);
903 	}
904 	if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
905 		ZFS_ATTR_SET(zp, ZFS_NODUMP, xoap->xoa_nodump,
906 		    zp->z_pflags, tx);
907 		XVA_SET_RTN(xvap, XAT_NODUMP);
908 	}
909 	if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
910 		ZFS_ATTR_SET(zp, ZFS_OPAQUE, xoap->xoa_opaque,
911 		    zp->z_pflags, tx);
912 		XVA_SET_RTN(xvap, XAT_OPAQUE);
913 	}
914 	if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
915 		ZFS_ATTR_SET(zp, ZFS_AV_QUARANTINED,
916 		    xoap->xoa_av_quarantined, zp->z_pflags, tx);
917 		XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
918 	}
919 	if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
920 		ZFS_ATTR_SET(zp, ZFS_AV_MODIFIED, xoap->xoa_av_modified,
921 		    zp->z_pflags, tx);
922 		XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
923 	}
924 	if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) {
925 		zfs_sa_set_scanstamp(zp, xvap, tx);
926 		XVA_SET_RTN(xvap, XAT_AV_SCANSTAMP);
927 	}
928 	if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
929 		ZFS_ATTR_SET(zp, ZFS_REPARSE, xoap->xoa_reparse,
930 		    zp->z_pflags, tx);
931 		XVA_SET_RTN(xvap, XAT_REPARSE);
932 	}
933 	if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
934 		ZFS_ATTR_SET(zp, ZFS_OFFLINE, xoap->xoa_offline,
935 		    zp->z_pflags, tx);
936 		XVA_SET_RTN(xvap, XAT_OFFLINE);
937 	}
938 	if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
939 		ZFS_ATTR_SET(zp, ZFS_SPARSE, xoap->xoa_sparse,
940 		    zp->z_pflags, tx);
941 		XVA_SET_RTN(xvap, XAT_SPARSE);
942 	}
943 	if (XVA_ISSET_REQ(xvap, XAT_PROJINHERIT)) {
944 		ZFS_ATTR_SET(zp, ZFS_PROJINHERIT, xoap->xoa_projinherit,
945 		    zp->z_pflags, tx);
946 		XVA_SET_RTN(xvap, XAT_PROJINHERIT);
947 	}
948 }
949 
950 int
zfs_zget(zfsvfs_t * zfsvfs,uint64_t obj_num,znode_t ** zpp)951 zfs_zget(zfsvfs_t *zfsvfs, uint64_t obj_num, znode_t **zpp)
952 {
953 	dmu_object_info_t doi;
954 	dmu_buf_t	*db;
955 	znode_t		*zp;
956 	vnode_t		*vp;
957 	sa_handle_t	*hdl;
958 	int locked;
959 	int err;
960 
961 	getnewvnode_reserve();
962 again:
963 	*zpp = NULL;
964 	ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num);
965 
966 	err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db);
967 	if (err) {
968 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
969 		getnewvnode_drop_reserve();
970 		return (err);
971 	}
972 
973 	dmu_object_info_from_db(db, &doi);
974 	if (doi.doi_bonus_type != DMU_OT_SA &&
975 	    (doi.doi_bonus_type != DMU_OT_ZNODE ||
976 	    (doi.doi_bonus_type == DMU_OT_ZNODE &&
977 	    doi.doi_bonus_size < sizeof (znode_phys_t)))) {
978 		sa_buf_rele(db, NULL);
979 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
980 		getnewvnode_drop_reserve();
981 		return (SET_ERROR(EINVAL));
982 	}
983 
984 	hdl = dmu_buf_get_user(db);
985 	if (hdl != NULL) {
986 		zp = sa_get_userdata(hdl);
987 
988 		/*
989 		 * Since "SA" does immediate eviction we
990 		 * should never find a sa handle that doesn't
991 		 * know about the znode.
992 		 */
993 		ASSERT3P(zp, !=, NULL);
994 		ASSERT3U(zp->z_id, ==, obj_num);
995 		if (zp->z_unlinked) {
996 			err = SET_ERROR(ENOENT);
997 		} else {
998 			vp = ZTOV(zp);
999 			/*
1000 			 * Don't let the vnode disappear after
1001 			 * ZFS_OBJ_HOLD_EXIT.
1002 			 */
1003 			VN_HOLD(vp);
1004 			*zpp = zp;
1005 			err = 0;
1006 		}
1007 
1008 		sa_buf_rele(db, NULL);
1009 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1010 
1011 		if (err) {
1012 			getnewvnode_drop_reserve();
1013 			return (err);
1014 		}
1015 
1016 		locked = VOP_ISLOCKED(vp);
1017 		VI_LOCK(vp);
1018 		if (VN_IS_DOOMED(vp) && locked != LK_EXCLUSIVE) {
1019 			/*
1020 			 * The vnode is doomed and this thread doesn't
1021 			 * hold the exclusive lock on it, so the vnode
1022 			 * must be being reclaimed by another thread.
1023 			 * Otherwise the doomed vnode is being reclaimed
1024 			 * by this thread and zfs_zget is called from
1025 			 * ZIL internals.
1026 			 */
1027 			VI_UNLOCK(vp);
1028 
1029 			/*
1030 			 * XXX vrele() locks the vnode when the last reference
1031 			 * is dropped.  Although in this case the vnode is
1032 			 * doomed / dead and so no inactivation is required,
1033 			 * the vnode lock is still acquired.  That could result
1034 			 * in a LOR with z_teardown_lock if another thread holds
1035 			 * the vnode's lock and tries to take z_teardown_lock.
1036 			 * But that is only possible if the other thread peforms
1037 			 * a ZFS vnode operation on the vnode.  That either
1038 			 * should not happen if the vnode is dead or the thread
1039 			 * should also have a reference to the vnode and thus
1040 			 * our reference is not last.
1041 			 */
1042 			VN_RELE(vp);
1043 			goto again;
1044 		}
1045 		VI_UNLOCK(vp);
1046 		getnewvnode_drop_reserve();
1047 		return (err);
1048 	}
1049 
1050 	/*
1051 	 * Not found create new znode/vnode
1052 	 * but only if file exists.
1053 	 *
1054 	 * There is a small window where zfs_vget() could
1055 	 * find this object while a file create is still in
1056 	 * progress.  This is checked for in zfs_znode_alloc()
1057 	 *
1058 	 * if zfs_znode_alloc() fails it will drop the hold on the
1059 	 * bonus buffer.
1060 	 */
1061 	zp = zfs_znode_alloc(zfsvfs, db, doi.doi_data_block_size,
1062 	    doi.doi_bonus_type, NULL);
1063 	if (zp == NULL) {
1064 		err = SET_ERROR(ENOENT);
1065 	} else {
1066 		*zpp = zp;
1067 	}
1068 	if (err == 0) {
1069 		vnode_t *vp = ZTOV(zp);
1070 
1071 		err = insmntque(vp, zfsvfs->z_vfs);
1072 		if (err == 0) {
1073 			vp->v_hash = obj_num;
1074 			VOP_UNLOCK(vp);
1075 		} else {
1076 			zp->z_vnode = NULL;
1077 			zfs_znode_dmu_fini(zp);
1078 			zfs_znode_free(zp);
1079 			*zpp = NULL;
1080 		}
1081 	}
1082 	ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1083 	getnewvnode_drop_reserve();
1084 	return (err);
1085 }
1086 
1087 int
zfs_rezget(znode_t * zp)1088 zfs_rezget(znode_t *zp)
1089 {
1090 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1091 	dmu_object_info_t doi;
1092 	dmu_buf_t *db;
1093 	vnode_t *vp;
1094 	uint64_t obj_num = zp->z_id;
1095 	uint64_t mode, size;
1096 	sa_bulk_attr_t bulk[8];
1097 	int err;
1098 	int count = 0;
1099 	uint64_t gen;
1100 	uint64_t projid = ZFS_DEFAULT_PROJID;
1101 
1102 	/*
1103 	 * Remove cached pages before reloading the znode, so that they are not
1104 	 * lingering after we run into any error.  Ideally, we should vgone()
1105 	 * the vnode in case of error, but currently we cannot do that
1106 	 * because of the LOR between the vnode lock and z_teardown_lock.
1107 	 * So, instead, we have to "doom" the znode in the illumos style.
1108 	 *
1109 	 * Ignore invalid pages during the scan.  This is to avoid deadlocks
1110 	 * between page busying and the teardown lock, as pages are busied prior
1111 	 * to a VOP_GETPAGES operation, which acquires the teardown read lock.
1112 	 * Such pages will be invalid and can safely be skipped here.
1113 	 */
1114 	vp = ZTOV(zp);
1115 #if __FreeBSD_version >= 1400042
1116 	vn_pages_remove_valid(vp, 0, 0);
1117 #else
1118 	vn_pages_remove(vp, 0, 0);
1119 #endif
1120 
1121 	ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num);
1122 
1123 	mutex_enter(&zp->z_acl_lock);
1124 	if (zp->z_acl_cached) {
1125 		zfs_acl_free(zp->z_acl_cached);
1126 		zp->z_acl_cached = NULL;
1127 	}
1128 	mutex_exit(&zp->z_acl_lock);
1129 
1130 	rw_enter(&zp->z_xattr_lock, RW_WRITER);
1131 	if (zp->z_xattr_cached) {
1132 		nvlist_free(zp->z_xattr_cached);
1133 		zp->z_xattr_cached = NULL;
1134 	}
1135 	rw_exit(&zp->z_xattr_lock);
1136 
1137 	ASSERT3P(zp->z_sa_hdl, ==, NULL);
1138 	err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db);
1139 	if (err) {
1140 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1141 		return (err);
1142 	}
1143 
1144 	dmu_object_info_from_db(db, &doi);
1145 	if (doi.doi_bonus_type != DMU_OT_SA &&
1146 	    (doi.doi_bonus_type != DMU_OT_ZNODE ||
1147 	    (doi.doi_bonus_type == DMU_OT_ZNODE &&
1148 	    doi.doi_bonus_size < sizeof (znode_phys_t)))) {
1149 		sa_buf_rele(db, NULL);
1150 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1151 		return (SET_ERROR(EINVAL));
1152 	}
1153 
1154 	zfs_znode_sa_init(zfsvfs, zp, db, doi.doi_bonus_type, NULL);
1155 	size = zp->z_size;
1156 
1157 	/* reload cached values */
1158 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL,
1159 	    &gen, sizeof (gen));
1160 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
1161 	    &zp->z_size, sizeof (zp->z_size));
1162 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL,
1163 	    &zp->z_links, sizeof (zp->z_links));
1164 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
1165 	    &zp->z_pflags, sizeof (zp->z_pflags));
1166 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
1167 	    &zp->z_atime, sizeof (zp->z_atime));
1168 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
1169 	    &zp->z_uid, sizeof (zp->z_uid));
1170 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL,
1171 	    &zp->z_gid, sizeof (zp->z_gid));
1172 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
1173 	    &mode, sizeof (mode));
1174 
1175 	if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) {
1176 		zfs_znode_dmu_fini(zp);
1177 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1178 		return (SET_ERROR(EIO));
1179 	}
1180 
1181 	if (dmu_objset_projectquota_enabled(zfsvfs->z_os)) {
1182 		err = sa_lookup(zp->z_sa_hdl, SA_ZPL_PROJID(zfsvfs),
1183 		    &projid, 8);
1184 		if (err != 0 && err != ENOENT) {
1185 			zfs_znode_dmu_fini(zp);
1186 			ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1187 			return (err);
1188 		}
1189 	}
1190 
1191 	zp->z_projid = projid;
1192 	zp->z_mode = mode;
1193 
1194 	if (gen != zp->z_gen) {
1195 		zfs_znode_dmu_fini(zp);
1196 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1197 		return (SET_ERROR(EIO));
1198 	}
1199 
1200 	/*
1201 	 * It is highly improbable but still quite possible that two
1202 	 * objects in different datasets are created with the same
1203 	 * object numbers and in transaction groups with the same
1204 	 * numbers.  znodes corresponding to those objects would
1205 	 * have the same z_id and z_gen, but their other attributes
1206 	 * may be different.
1207 	 * zfs recv -F may replace one of such objects with the other.
1208 	 * As a result file properties recorded in the replaced
1209 	 * object's vnode may no longer match the received object's
1210 	 * properties.  At present the only cached property is the
1211 	 * files type recorded in v_type.
1212 	 * So, handle this case by leaving the old vnode and znode
1213 	 * disassociated from the actual object.  A new vnode and a
1214 	 * znode will be created if the object is accessed
1215 	 * (e.g. via a look-up).  The old vnode and znode will be
1216 	 * recycled when the last vnode reference is dropped.
1217 	 */
1218 	if (vp->v_type != IFTOVT((mode_t)zp->z_mode)) {
1219 		zfs_znode_dmu_fini(zp);
1220 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1221 		return (SET_ERROR(EIO));
1222 	}
1223 
1224 	/*
1225 	 * If the file has zero links, then it has been unlinked on the send
1226 	 * side and it must be in the received unlinked set.
1227 	 * We call zfs_znode_dmu_fini() now to prevent any accesses to the
1228 	 * stale data and to prevent automatically removal of the file in
1229 	 * zfs_zinactive().  The file will be removed either when it is removed
1230 	 * on the send side and the next incremental stream is received or
1231 	 * when the unlinked set gets processed.
1232 	 */
1233 	zp->z_unlinked = (zp->z_links == 0);
1234 	if (zp->z_unlinked) {
1235 		zfs_znode_dmu_fini(zp);
1236 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1237 		return (0);
1238 	}
1239 
1240 	zp->z_blksz = doi.doi_data_block_size;
1241 	if (zp->z_size != size)
1242 		vnode_pager_setsize(vp, zp->z_size);
1243 
1244 	ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1245 
1246 	return (0);
1247 }
1248 
1249 void
zfs_znode_delete(znode_t * zp,dmu_tx_t * tx)1250 zfs_znode_delete(znode_t *zp, dmu_tx_t *tx)
1251 {
1252 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1253 	objset_t *os = zfsvfs->z_os;
1254 	uint64_t obj = zp->z_id;
1255 	uint64_t acl_obj = zfs_external_acl(zp);
1256 
1257 	ZFS_OBJ_HOLD_ENTER(zfsvfs, obj);
1258 	if (acl_obj) {
1259 		VERIFY(!zp->z_is_sa);
1260 		VERIFY0(dmu_object_free(os, acl_obj, tx));
1261 	}
1262 	VERIFY0(dmu_object_free(os, obj, tx));
1263 	zfs_znode_dmu_fini(zp);
1264 	ZFS_OBJ_HOLD_EXIT(zfsvfs, obj);
1265 }
1266 
1267 void
zfs_zinactive(znode_t * zp)1268 zfs_zinactive(znode_t *zp)
1269 {
1270 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1271 	uint64_t z_id = zp->z_id;
1272 
1273 	ASSERT3P(zp->z_sa_hdl, !=, NULL);
1274 
1275 	/*
1276 	 * Don't allow a zfs_zget() while were trying to release this znode
1277 	 */
1278 	ZFS_OBJ_HOLD_ENTER(zfsvfs, z_id);
1279 
1280 	/*
1281 	 * If this was the last reference to a file with no links, remove
1282 	 * the file from the file system unless the file system is mounted
1283 	 * read-only.  That can happen, for example, if the file system was
1284 	 * originally read-write, the file was opened, then unlinked and
1285 	 * the file system was made read-only before the file was finally
1286 	 * closed.  The file will remain in the unlinked set.
1287 	 */
1288 	if (zp->z_unlinked) {
1289 		ASSERT(!zfsvfs->z_issnap);
1290 		if ((zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) == 0) {
1291 			ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id);
1292 			zfs_rmnode(zp);
1293 			return;
1294 		}
1295 	}
1296 
1297 	zfs_znode_dmu_fini(zp);
1298 	ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id);
1299 	zfs_znode_free(zp);
1300 }
1301 
1302 void
zfs_znode_free(znode_t * zp)1303 zfs_znode_free(znode_t *zp)
1304 {
1305 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1306 	char *symlink;
1307 
1308 	ASSERT3P(zp->z_sa_hdl, ==, NULL);
1309 	zp->z_vnode = NULL;
1310 	mutex_enter(&zfsvfs->z_znodes_lock);
1311 	POINTER_INVALIDATE(&zp->z_zfsvfs);
1312 	list_remove(&zfsvfs->z_all_znodes, zp);
1313 	mutex_exit(&zfsvfs->z_znodes_lock);
1314 
1315 	symlink = atomic_load_ptr(&zp->z_cached_symlink);
1316 	if (symlink != NULL) {
1317 		atomic_store_rel_ptr((uintptr_t *)&zp->z_cached_symlink,
1318 		    (uintptr_t)NULL);
1319 		cache_symlink_free(symlink, strlen(symlink) + 1);
1320 	}
1321 
1322 	if (zp->z_acl_cached) {
1323 		zfs_acl_free(zp->z_acl_cached);
1324 		zp->z_acl_cached = NULL;
1325 	}
1326 
1327 	zfs_znode_free_kmem(zp);
1328 }
1329 
1330 void
zfs_tstamp_update_setup_ext(znode_t * zp,uint_t flag,uint64_t mtime[2],uint64_t ctime[2],boolean_t have_tx)1331 zfs_tstamp_update_setup_ext(znode_t *zp, uint_t flag, uint64_t mtime[2],
1332     uint64_t ctime[2], boolean_t have_tx)
1333 {
1334 	timestruc_t	now;
1335 
1336 	vfs_timestamp(&now);
1337 
1338 	if (have_tx) {	/* will sa_bulk_update happen really soon? */
1339 		zp->z_atime_dirty = 0;
1340 		zp->z_seq++;
1341 	} else {
1342 		zp->z_atime_dirty = 1;
1343 	}
1344 
1345 	if (flag & AT_ATIME) {
1346 		ZFS_TIME_ENCODE(&now, zp->z_atime);
1347 	}
1348 
1349 	if (flag & AT_MTIME) {
1350 		ZFS_TIME_ENCODE(&now, mtime);
1351 		if (zp->z_zfsvfs->z_use_fuids) {
1352 			zp->z_pflags |= (ZFS_ARCHIVE |
1353 			    ZFS_AV_MODIFIED);
1354 		}
1355 	}
1356 
1357 	if (flag & AT_CTIME) {
1358 		ZFS_TIME_ENCODE(&now, ctime);
1359 		if (zp->z_zfsvfs->z_use_fuids)
1360 			zp->z_pflags |= ZFS_ARCHIVE;
1361 	}
1362 }
1363 
1364 
1365 void
zfs_tstamp_update_setup(znode_t * zp,uint_t flag,uint64_t mtime[2],uint64_t ctime[2])1366 zfs_tstamp_update_setup(znode_t *zp, uint_t flag, uint64_t mtime[2],
1367     uint64_t ctime[2])
1368 {
1369 	zfs_tstamp_update_setup_ext(zp, flag, mtime, ctime, B_TRUE);
1370 }
1371 /*
1372  * Grow the block size for a file.
1373  *
1374  *	IN:	zp	- znode of file to free data in.
1375  *		size	- requested block size
1376  *		tx	- open transaction.
1377  *
1378  * NOTE: this function assumes that the znode is write locked.
1379  */
1380 void
zfs_grow_blocksize(znode_t * zp,uint64_t size,dmu_tx_t * tx)1381 zfs_grow_blocksize(znode_t *zp, uint64_t size, dmu_tx_t *tx)
1382 {
1383 	int		error;
1384 	u_longlong_t	dummy;
1385 
1386 	if (size <= zp->z_blksz)
1387 		return;
1388 	/*
1389 	 * If the file size is already greater than the current blocksize,
1390 	 * we will not grow.  If there is more than one block in a file,
1391 	 * the blocksize cannot change.
1392 	 */
1393 	if (zp->z_blksz && zp->z_size > zp->z_blksz)
1394 		return;
1395 
1396 	error = dmu_object_set_blocksize(zp->z_zfsvfs->z_os, zp->z_id,
1397 	    size, 0, tx);
1398 
1399 	if (error == ENOTSUP)
1400 		return;
1401 	ASSERT0(error);
1402 
1403 	/* What blocksize did we actually get? */
1404 	dmu_object_size_from_db(sa_get_db(zp->z_sa_hdl), &zp->z_blksz, &dummy);
1405 }
1406 
1407 /*
1408  * Increase the file length
1409  *
1410  *	IN:	zp	- znode of file to free data in.
1411  *		end	- new end-of-file
1412  *
1413  *	RETURN:	0 on success, error code on failure
1414  */
1415 static int
zfs_extend(znode_t * zp,uint64_t end)1416 zfs_extend(znode_t *zp, uint64_t end)
1417 {
1418 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1419 	dmu_tx_t *tx;
1420 	zfs_locked_range_t *lr;
1421 	uint64_t newblksz;
1422 	int error;
1423 
1424 	/*
1425 	 * We will change zp_size, lock the whole file.
1426 	 */
1427 	lr = zfs_rangelock_enter(&zp->z_rangelock, 0, UINT64_MAX, RL_WRITER);
1428 
1429 	/*
1430 	 * Nothing to do if file already at desired length.
1431 	 */
1432 	if (end <= zp->z_size) {
1433 		zfs_rangelock_exit(lr);
1434 		return (0);
1435 	}
1436 	tx = dmu_tx_create(zfsvfs->z_os);
1437 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1438 	zfs_sa_upgrade_txholds(tx, zp);
1439 	if (end > zp->z_blksz &&
1440 	    (!ISP2(zp->z_blksz) || zp->z_blksz < zfsvfs->z_max_blksz)) {
1441 		/*
1442 		 * We are growing the file past the current block size.
1443 		 */
1444 		if (zp->z_blksz > zp->z_zfsvfs->z_max_blksz) {
1445 			/*
1446 			 * File's blocksize is already larger than the
1447 			 * "recordsize" property.  Only let it grow to
1448 			 * the next power of 2.
1449 			 */
1450 			ASSERT(!ISP2(zp->z_blksz));
1451 			newblksz = MIN(end, 1 << highbit64(zp->z_blksz));
1452 		} else {
1453 			newblksz = MIN(end, zp->z_zfsvfs->z_max_blksz);
1454 		}
1455 		dmu_tx_hold_write(tx, zp->z_id, 0, newblksz);
1456 	} else {
1457 		newblksz = 0;
1458 	}
1459 
1460 	error = dmu_tx_assign(tx, DMU_TX_WAIT);
1461 	if (error) {
1462 		dmu_tx_abort(tx);
1463 		zfs_rangelock_exit(lr);
1464 		return (error);
1465 	}
1466 
1467 	if (newblksz)
1468 		zfs_grow_blocksize(zp, newblksz, tx);
1469 
1470 	zp->z_size = end;
1471 
1472 	VERIFY0(sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zp->z_zfsvfs),
1473 	    &zp->z_size, sizeof (zp->z_size), tx));
1474 
1475 	vnode_pager_setsize(ZTOV(zp), end);
1476 
1477 	zfs_rangelock_exit(lr);
1478 
1479 	dmu_tx_commit(tx);
1480 
1481 	return (0);
1482 }
1483 
1484 /*
1485  * Free space in a file.
1486  *
1487  *	IN:	zp	- znode of file to free data in.
1488  *		off	- start of section to free.
1489  *		len	- length of section to free.
1490  *
1491  *	RETURN:	0 on success, error code on failure
1492  */
1493 static int
zfs_free_range(znode_t * zp,uint64_t off,uint64_t len)1494 zfs_free_range(znode_t *zp, uint64_t off, uint64_t len)
1495 {
1496 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1497 	zfs_locked_range_t *lr;
1498 	int error;
1499 
1500 	/*
1501 	 * Lock the range being freed.
1502 	 */
1503 	lr = zfs_rangelock_enter(&zp->z_rangelock, off, len, RL_WRITER);
1504 
1505 	/*
1506 	 * Nothing to do if file already at desired length.
1507 	 */
1508 	if (off >= zp->z_size) {
1509 		zfs_rangelock_exit(lr);
1510 		return (0);
1511 	}
1512 
1513 	if (off + len > zp->z_size)
1514 		len = zp->z_size - off;
1515 
1516 	error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, off, len);
1517 
1518 	if (error == 0) {
1519 #if __FreeBSD_version >= 1400032
1520 		vnode_pager_purge_range(ZTOV(zp), off, off + len);
1521 #else
1522 		/*
1523 		 * Before __FreeBSD_version 1400032 we cannot free block in the
1524 		 * middle of a file, but only at the end of a file, so this code
1525 		 * path should never happen.
1526 		 */
1527 		vnode_pager_setsize(ZTOV(zp), off);
1528 #endif
1529 	}
1530 
1531 	zfs_rangelock_exit(lr);
1532 
1533 	return (error);
1534 }
1535 
1536 /*
1537  * Truncate a file
1538  *
1539  *	IN:	zp	- znode of file to free data in.
1540  *		end	- new end-of-file.
1541  *
1542  *	RETURN:	0 on success, error code on failure
1543  */
1544 static int
zfs_trunc(znode_t * zp,uint64_t end)1545 zfs_trunc(znode_t *zp, uint64_t end)
1546 {
1547 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1548 	vnode_t *vp = ZTOV(zp);
1549 	dmu_tx_t *tx;
1550 	zfs_locked_range_t *lr;
1551 	int error;
1552 	sa_bulk_attr_t bulk[2];
1553 	int count = 0;
1554 
1555 	/*
1556 	 * We will change zp_size, lock the whole file.
1557 	 */
1558 	lr = zfs_rangelock_enter(&zp->z_rangelock, 0, UINT64_MAX, RL_WRITER);
1559 
1560 	/*
1561 	 * Nothing to do if file already at desired length.
1562 	 */
1563 	if (end >= zp->z_size) {
1564 		zfs_rangelock_exit(lr);
1565 		return (0);
1566 	}
1567 
1568 	error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, end,
1569 	    DMU_OBJECT_END);
1570 	if (error) {
1571 		zfs_rangelock_exit(lr);
1572 		return (error);
1573 	}
1574 	tx = dmu_tx_create(zfsvfs->z_os);
1575 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1576 	zfs_sa_upgrade_txholds(tx, zp);
1577 	dmu_tx_mark_netfree(tx);
1578 	error = dmu_tx_assign(tx, DMU_TX_WAIT);
1579 	if (error) {
1580 		dmu_tx_abort(tx);
1581 		zfs_rangelock_exit(lr);
1582 		return (error);
1583 	}
1584 
1585 	zp->z_size = end;
1586 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs),
1587 	    NULL, &zp->z_size, sizeof (zp->z_size));
1588 
1589 	if (end == 0) {
1590 		zp->z_pflags &= ~ZFS_SPARSE;
1591 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
1592 		    NULL, &zp->z_pflags, 8);
1593 	}
1594 	VERIFY0(sa_bulk_update(zp->z_sa_hdl, bulk, count, tx));
1595 
1596 	dmu_tx_commit(tx);
1597 
1598 	/*
1599 	 * Clear any mapped pages in the truncated region.  This has to
1600 	 * happen outside of the transaction to avoid the possibility of
1601 	 * a deadlock with someone trying to push a page that we are
1602 	 * about to invalidate.
1603 	 */
1604 	vnode_pager_setsize(vp, end);
1605 
1606 	zfs_rangelock_exit(lr);
1607 
1608 	return (0);
1609 }
1610 
1611 /*
1612  * Free space in a file
1613  *
1614  *	IN:	zp	- znode of file to free data in.
1615  *		off	- start of range
1616  *		len	- end of range (0 => EOF)
1617  *		flag	- current file open mode flags.
1618  *		log	- TRUE if this action should be logged
1619  *
1620  *	RETURN:	0 on success, error code on failure
1621  */
1622 int
zfs_freesp(znode_t * zp,uint64_t off,uint64_t len,int flag,boolean_t log)1623 zfs_freesp(znode_t *zp, uint64_t off, uint64_t len, int flag, boolean_t log)
1624 {
1625 	dmu_tx_t *tx;
1626 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1627 	zilog_t *zilog = zfsvfs->z_log;
1628 	uint64_t mode;
1629 	uint64_t mtime[2], ctime[2];
1630 	sa_bulk_attr_t bulk[3];
1631 	int count = 0;
1632 	int error;
1633 
1634 	if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs), &mode,
1635 	    sizeof (mode))) != 0)
1636 		return (error);
1637 
1638 	if (off > zp->z_size) {
1639 		error =  zfs_extend(zp, off+len);
1640 		if (error == 0 && log)
1641 			goto log;
1642 		else
1643 			return (error);
1644 	}
1645 
1646 	if (len == 0) {
1647 		error = zfs_trunc(zp, off);
1648 	} else {
1649 		if ((error = zfs_free_range(zp, off, len)) == 0 &&
1650 		    off + len > zp->z_size)
1651 			error = zfs_extend(zp, off+len);
1652 	}
1653 	if (error || !log)
1654 		return (error);
1655 log:
1656 	tx = dmu_tx_create(zfsvfs->z_os);
1657 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1658 	zfs_sa_upgrade_txholds(tx, zp);
1659 	error = dmu_tx_assign(tx, DMU_TX_WAIT);
1660 	if (error) {
1661 		dmu_tx_abort(tx);
1662 		return (error);
1663 	}
1664 
1665 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, mtime, 16);
1666 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, ctime, 16);
1667 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
1668 	    NULL, &zp->z_pflags, 8);
1669 	zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime);
1670 	error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
1671 	ASSERT0(error);
1672 
1673 	zfs_log_truncate(zilog, tx, TX_TRUNCATE, zp, off, len);
1674 
1675 	dmu_tx_commit(tx);
1676 	return (0);
1677 }
1678 
1679 void
zfs_create_fs(objset_t * os,cred_t * cr,nvlist_t * zplprops,dmu_tx_t * tx)1680 zfs_create_fs(objset_t *os, cred_t *cr, nvlist_t *zplprops, dmu_tx_t *tx)
1681 {
1682 	uint64_t	moid, obj, sa_obj, version;
1683 	uint64_t	sense = ZFS_CASE_SENSITIVE;
1684 	uint64_t	norm = 0;
1685 	nvpair_t	*elem;
1686 	int		error;
1687 	int		i;
1688 	znode_t		*rootzp = NULL;
1689 	zfsvfs_t	*zfsvfs;
1690 	vattr_t		vattr;
1691 	znode_t		*zp;
1692 	zfs_acl_ids_t	acl_ids;
1693 
1694 	/*
1695 	 * First attempt to create master node.
1696 	 */
1697 	/*
1698 	 * In an empty objset, there are no blocks to read and thus
1699 	 * there can be no i/o errors (which we assert below).
1700 	 */
1701 	moid = MASTER_NODE_OBJ;
1702 	error = zap_create_claim(os, moid, DMU_OT_MASTER_NODE,
1703 	    DMU_OT_NONE, 0, tx);
1704 	ASSERT0(error);
1705 
1706 	/*
1707 	 * Set starting attributes.
1708 	 */
1709 	version = zfs_zpl_version_map(spa_version(dmu_objset_spa(os)));
1710 	elem = NULL;
1711 	while ((elem = nvlist_next_nvpair(zplprops, elem)) != NULL) {
1712 		/* For the moment we expect all zpl props to be uint64_ts */
1713 		uint64_t val;
1714 		const char *name;
1715 
1716 		ASSERT3S(nvpair_type(elem), ==, DATA_TYPE_UINT64);
1717 		val = fnvpair_value_uint64(elem);
1718 		name = nvpair_name(elem);
1719 		if (strcmp(name, zfs_prop_to_name(ZFS_PROP_VERSION)) == 0) {
1720 			if (val < version)
1721 				version = val;
1722 		} else {
1723 			error = zap_update(os, moid, name, 8, 1, &val, tx);
1724 		}
1725 		ASSERT0(error);
1726 		if (strcmp(name, zfs_prop_to_name(ZFS_PROP_NORMALIZE)) == 0)
1727 			norm = val;
1728 		else if (strcmp(name, zfs_prop_to_name(ZFS_PROP_CASE)) == 0)
1729 			sense = val;
1730 	}
1731 	ASSERT3U(version, !=, 0);
1732 	error = zap_update(os, moid, ZPL_VERSION_STR, 8, 1, &version, tx);
1733 	ASSERT0(error);
1734 
1735 	/*
1736 	 * Create zap object used for SA attribute registration
1737 	 */
1738 
1739 	if (version >= ZPL_VERSION_SA) {
1740 		sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
1741 		    DMU_OT_NONE, 0, tx);
1742 		error = zap_add(os, moid, ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
1743 		ASSERT0(error);
1744 	} else {
1745 		sa_obj = 0;
1746 	}
1747 	/*
1748 	 * Create a delete queue.
1749 	 */
1750 	obj = zap_create(os, DMU_OT_UNLINKED_SET, DMU_OT_NONE, 0, tx);
1751 
1752 	error = zap_add(os, moid, ZFS_UNLINKED_SET, 8, 1, &obj, tx);
1753 	ASSERT0(error);
1754 
1755 	/*
1756 	 * Create root znode.  Create minimal znode/vnode/zfsvfs
1757 	 * to allow zfs_mknode to work.
1758 	 */
1759 	VATTR_NULL(&vattr);
1760 	vattr.va_mask = AT_MODE|AT_UID|AT_GID;
1761 	vattr.va_type = VDIR;
1762 	vattr.va_mode = S_IFDIR|0755;
1763 	vattr.va_uid = crgetuid(cr);
1764 	vattr.va_gid = crgetgid(cr);
1765 
1766 	zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
1767 
1768 	rootzp = zfs_znode_alloc_kmem(KM_SLEEP);
1769 	ASSERT(!POINTER_IS_VALID(rootzp->z_zfsvfs));
1770 	rootzp->z_unlinked = 0;
1771 	rootzp->z_atime_dirty = 0;
1772 	rootzp->z_is_sa = USE_SA(version, os);
1773 	rootzp->z_pflags = 0;
1774 
1775 	zfsvfs->z_os = os;
1776 	zfsvfs->z_parent = zfsvfs;
1777 	zfsvfs->z_version = version;
1778 	zfsvfs->z_use_fuids = USE_FUIDS(version, os);
1779 	zfsvfs->z_use_sa = USE_SA(version, os);
1780 	zfsvfs->z_norm = norm;
1781 
1782 	error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
1783 	    &zfsvfs->z_attr_table);
1784 
1785 	ASSERT0(error);
1786 
1787 	/*
1788 	 * Fold case on file systems that are always or sometimes case
1789 	 * insensitive.
1790 	 */
1791 	if (sense == ZFS_CASE_INSENSITIVE || sense == ZFS_CASE_MIXED)
1792 		zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER;
1793 
1794 	mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
1795 	list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
1796 	    offsetof(znode_t, z_link_node));
1797 
1798 	for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1799 		mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL);
1800 
1801 	rootzp->z_zfsvfs = zfsvfs;
1802 	VERIFY0(zfs_acl_ids_create(rootzp, IS_ROOT_NODE, &vattr,
1803 	    cr, NULL, &acl_ids, NULL));
1804 	zfs_mknode(rootzp, &vattr, tx, cr, IS_ROOT_NODE, &zp, &acl_ids);
1805 	ASSERT3P(zp, ==, rootzp);
1806 	error = zap_add(os, moid, ZFS_ROOT_OBJ, 8, 1, &rootzp->z_id, tx);
1807 	ASSERT0(error);
1808 	zfs_acl_ids_free(&acl_ids);
1809 	POINTER_INVALIDATE(&rootzp->z_zfsvfs);
1810 
1811 	sa_handle_destroy(rootzp->z_sa_hdl);
1812 	zfs_znode_free_kmem(rootzp);
1813 
1814 	/*
1815 	 * Create shares directory
1816 	 */
1817 
1818 	error = zfs_create_share_dir(zfsvfs, tx);
1819 
1820 	ASSERT0(error);
1821 
1822 	for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1823 		mutex_destroy(&zfsvfs->z_hold_mtx[i]);
1824 	kmem_free(zfsvfs, sizeof (zfsvfs_t));
1825 }
1826 
1827 void
zfs_znode_update_vfs(znode_t * zp)1828 zfs_znode_update_vfs(znode_t *zp)
1829 {
1830 	vm_object_t object;
1831 
1832 	if ((object = ZTOV(zp)->v_object) == NULL ||
1833 	    zp->z_size == object->un_pager.vnp.vnp_size)
1834 		return;
1835 
1836 	vnode_pager_setsize(ZTOV(zp), zp->z_size);
1837 }
1838 
1839 int
zfs_znode_parent_and_name(znode_t * zp,znode_t ** dzpp,char * buf,uint64_t buflen)1840 zfs_znode_parent_and_name(znode_t *zp, znode_t **dzpp, char *buf,
1841     uint64_t buflen)
1842 {
1843 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1844 	uint64_t parent;
1845 	int is_xattrdir;
1846 	int err;
1847 
1848 	/* Extended attributes should not be visible as regular files. */
1849 	if ((zp->z_pflags & ZFS_XATTR) != 0)
1850 		return (SET_ERROR(EINVAL));
1851 
1852 	err = zfs_obj_to_pobj(zfsvfs->z_os, zp->z_sa_hdl, zfsvfs->z_attr_table,
1853 	    &parent, &is_xattrdir);
1854 	if (err != 0)
1855 		return (err);
1856 	ASSERT0(is_xattrdir);
1857 
1858 	/* No name as this is a root object. */
1859 	if (parent == zp->z_id)
1860 		return (SET_ERROR(EINVAL));
1861 
1862 	err = zap_value_search(zfsvfs->z_os, parent, zp->z_id,
1863 	    ZFS_DIRENT_OBJ(-1ULL), buf, buflen);
1864 	if (err != 0)
1865 		return (err);
1866 	err = zfs_zget(zfsvfs, parent, dzpp);
1867 	return (err);
1868 }
1869 
1870 int
zfs_rlimit_fsize(off_t fsize)1871 zfs_rlimit_fsize(off_t fsize)
1872 {
1873 	struct thread *td = curthread;
1874 	off_t lim;
1875 
1876 	if (td == NULL)
1877 		return (0);
1878 
1879 	lim = lim_cur(td, RLIMIT_FSIZE);
1880 	if (__predict_true((uoff_t)fsize <= lim))
1881 		return (0);
1882 
1883 	/*
1884 	 * The limit is reached.
1885 	 */
1886 	PROC_LOCK(td->td_proc);
1887 	kern_psignal(td->td_proc, SIGXFSZ);
1888 	PROC_UNLOCK(td->td_proc);
1889 
1890 	return (EFBIG);
1891 }
1892