1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23  * Copyright (c) 2012, 2014 by Delphix. All rights reserved.
24  * Copyright (c) 2014 Integros [integros.com]
25  */
26 
27 /* Portions Copyright 2007 Jeremy Teo */
28 /* Portions Copyright 2011 Martin Matuska <[email protected]> */
29 
30 #ifdef _KERNEL
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/mntent.h>
38 #include <sys/u8_textprep.h>
39 #include <sys/dsl_dataset.h>
40 #include <sys/vfs.h>
41 #include <sys/vnode.h>
42 #include <sys/file.h>
43 #include <sys/kmem.h>
44 #include <sys/errno.h>
45 #include <sys/unistd.h>
46 #include <sys/atomic.h>
47 #include <sys/zfs_dir.h>
48 #include <sys/zfs_acl.h>
49 #include <sys/zfs_ioctl.h>
50 #include <sys/zfs_rlock.h>
51 #include <sys/zfs_fuid.h>
52 #include <sys/dnode.h>
53 #include <sys/fs/zfs.h>
54 #endif /* _KERNEL */
55 
56 #include <sys/dmu.h>
57 #include <sys/dmu_objset.h>
58 #include <sys/dmu_tx.h>
59 #include <sys/zfs_refcount.h>
60 #include <sys/stat.h>
61 #include <sys/zap.h>
62 #include <sys/zfs_znode.h>
63 #include <sys/sa.h>
64 #include <sys/zfs_sa.h>
65 #include <sys/zfs_stat.h>
66 
67 #include "zfs_prop.h"
68 #include "zfs_comutil.h"
69 
70 /* Used by fstat(1). */
71 SYSCTL_INT(_debug_sizeof, OID_AUTO, znode, CTLFLAG_RD,
72 	SYSCTL_NULL_INT_PTR, sizeof (znode_t), "sizeof(znode_t)");
73 
74 /*
75  * Define ZNODE_STATS to turn on statistic gathering. By default, it is only
76  * turned on when DEBUG is also defined.
77  */
78 #ifdef	ZFS_DEBUG
79 #define	ZNODE_STATS
80 #endif	/* DEBUG */
81 
82 #ifdef	ZNODE_STATS
83 #define	ZNODE_STAT_ADD(stat)			((stat)++)
84 #else
85 #define	ZNODE_STAT_ADD(stat)			/* nothing */
86 #endif	/* ZNODE_STATS */
87 
88 /*
89  * Functions needed for userland (ie: libzpool) are not put under
90  * #ifdef_KERNEL; the rest of the functions have dependencies
91  * (such as VFS logic) that will not compile easily in userland.
92  */
93 #ifdef _KERNEL
94 #if !defined(KMEM_DEBUG) && __FreeBSD_version >= 1300102
95 #define	_ZFS_USE_SMR
96 static uma_zone_t znode_uma_zone;
97 #else
98 static kmem_cache_t *znode_cache = NULL;
99 #endif
100 
101 extern struct vop_vector zfs_vnodeops;
102 extern struct vop_vector zfs_fifoops;
103 extern struct vop_vector zfs_shareops;
104 
105 
106 /*
107  * This callback is invoked when acquiring a RL_WRITER or RL_APPEND lock on
108  * z_rangelock. It will modify the offset and length of the lock to reflect
109  * znode-specific information, and convert RL_APPEND to RL_WRITER.  This is
110  * called with the rangelock_t's rl_lock held, which avoids races.
111  */
112 static void
zfs_rangelock_cb(zfs_locked_range_t * new,void * arg)113 zfs_rangelock_cb(zfs_locked_range_t *new, void *arg)
114 {
115 	znode_t *zp = arg;
116 
117 	/*
118 	 * If in append mode, convert to writer and lock starting at the
119 	 * current end of file.
120 	 */
121 	if (new->lr_type == RL_APPEND) {
122 		new->lr_offset = zp->z_size;
123 		new->lr_type = RL_WRITER;
124 	}
125 
126 	/*
127 	 * If we need to grow the block size then lock the whole file range.
128 	 */
129 	uint64_t end_size = MAX(zp->z_size, new->lr_offset + new->lr_length);
130 	if (end_size > zp->z_blksz && (!ISP2(zp->z_blksz) ||
131 	    zp->z_blksz < ZTOZSB(zp)->z_max_blksz)) {
132 		new->lr_offset = 0;
133 		new->lr_length = UINT64_MAX;
134 	}
135 }
136 
137 static int
zfs_znode_cache_constructor(void * buf,void * arg,int kmflags)138 zfs_znode_cache_constructor(void *buf, void *arg, int kmflags)
139 {
140 	znode_t *zp = buf;
141 
142 	POINTER_INVALIDATE(&zp->z_zfsvfs);
143 
144 	list_link_init(&zp->z_link_node);
145 
146 	mutex_init(&zp->z_acl_lock, NULL, MUTEX_DEFAULT, NULL);
147 
148 	zfs_rangelock_init(&zp->z_rangelock, zfs_rangelock_cb, zp);
149 
150 	zp->z_acl_cached = NULL;
151 	zp->z_vnode = NULL;
152 	return (0);
153 }
154 
155 /*ARGSUSED*/
156 static void
zfs_znode_cache_destructor(void * buf,void * arg)157 zfs_znode_cache_destructor(void *buf, void *arg)
158 {
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_acl_lock);
165 	zfs_rangelock_fini(&zp->z_rangelock);
166 
167 	ASSERT(zp->z_acl_cached == NULL);
168 }
169 
170 
171 #ifdef _ZFS_USE_SMR
172 VFS_SMR_DECLARE;
173 
174 static int
zfs_znode_cache_constructor_smr(void * mem,int size __unused,void * private,int flags)175 zfs_znode_cache_constructor_smr(void *mem, int size __unused, void *private,
176     int flags)
177 {
178 
179 	return (zfs_znode_cache_constructor(mem, private, flags));
180 }
181 
182 static void
zfs_znode_cache_destructor_smr(void * mem,int size __unused,void * private)183 zfs_znode_cache_destructor_smr(void *mem, int size __unused, void *private)
184 {
185 
186 	zfs_znode_cache_destructor(mem, private);
187 }
188 
189 void
zfs_znode_init(void)190 zfs_znode_init(void)
191 {
192 	/*
193 	 * Initialize zcache
194 	 */
195 	ASSERT(znode_uma_zone == NULL);
196 	znode_uma_zone = uma_zcreate("zfs_znode_cache",
197 	    sizeof (znode_t), zfs_znode_cache_constructor_smr,
198 	    zfs_znode_cache_destructor_smr, NULL, NULL, 0, 0);
199 	VFS_SMR_ZONE_SET(znode_uma_zone);
200 }
201 
202 static znode_t *
zfs_znode_alloc_kmem(int flags)203 zfs_znode_alloc_kmem(int flags)
204 {
205 
206 	return (uma_zalloc_smr(znode_uma_zone, flags));
207 }
208 
209 static void
zfs_znode_free_kmem(znode_t * zp)210 zfs_znode_free_kmem(znode_t *zp)
211 {
212 
213 	uma_zfree_smr(znode_uma_zone, zp);
214 }
215 #else
216 void
zfs_znode_init(void)217 zfs_znode_init(void)
218 {
219 	/*
220 	 * Initialize zcache
221 	 */
222 	ASSERT(znode_cache == NULL);
223 	znode_cache = kmem_cache_create("zfs_znode_cache",
224 	    sizeof (znode_t), 0, zfs_znode_cache_constructor,
225 	    zfs_znode_cache_destructor, NULL, NULL, NULL, 0);
226 }
227 
228 static znode_t *
zfs_znode_alloc_kmem(int flags)229 zfs_znode_alloc_kmem(int flags)
230 {
231 
232 	return (kmem_cache_alloc(znode_cache, flags));
233 }
234 
235 static void
zfs_znode_free_kmem(znode_t * zp)236 zfs_znode_free_kmem(znode_t *zp)
237 {
238 
239 	kmem_cache_free(znode_cache, zp);
240 }
241 #endif
242 
243 void
zfs_znode_fini(void)244 zfs_znode_fini(void)
245 {
246 	/*
247 	 * Cleanup zcache
248 	 */
249 #ifdef _ZFS_USE_SMR
250 	if (znode_uma_zone) {
251 		uma_zdestroy(znode_uma_zone);
252 		znode_uma_zone = NULL;
253 	}
254 #else
255 	if (znode_cache) {
256 		kmem_cache_destroy(znode_cache);
257 		znode_cache = NULL;
258 	}
259 #endif
260 }
261 
262 
263 static int
zfs_create_share_dir(zfsvfs_t * zfsvfs,dmu_tx_t * tx)264 zfs_create_share_dir(zfsvfs_t *zfsvfs, dmu_tx_t *tx)
265 {
266 	zfs_acl_ids_t acl_ids;
267 	vattr_t vattr;
268 	znode_t *sharezp;
269 	znode_t *zp;
270 	int error;
271 
272 	vattr.va_mask = AT_MODE|AT_UID|AT_GID;
273 	vattr.va_type = VDIR;
274 	vattr.va_mode = S_IFDIR|0555;
275 	vattr.va_uid = crgetuid(kcred);
276 	vattr.va_gid = crgetgid(kcred);
277 
278 	sharezp = zfs_znode_alloc_kmem(KM_SLEEP);
279 	ASSERT(!POINTER_IS_VALID(sharezp->z_zfsvfs));
280 	sharezp->z_unlinked = 0;
281 	sharezp->z_atime_dirty = 0;
282 	sharezp->z_zfsvfs = zfsvfs;
283 	sharezp->z_is_sa = zfsvfs->z_use_sa;
284 
285 	VERIFY(0 == zfs_acl_ids_create(sharezp, IS_ROOT_NODE, &vattr,
286 	    kcred, NULL, &acl_ids));
287 	zfs_mknode(sharezp, &vattr, tx, kcred, IS_ROOT_NODE, &zp, &acl_ids);
288 	ASSERT3P(zp, ==, sharezp);
289 	POINTER_INVALIDATE(&sharezp->z_zfsvfs);
290 	error = zap_add(zfsvfs->z_os, MASTER_NODE_OBJ,
291 	    ZFS_SHARES_DIR, 8, 1, &sharezp->z_id, tx);
292 	zfsvfs->z_shares_dir = sharezp->z_id;
293 
294 	zfs_acl_ids_free(&acl_ids);
295 	sa_handle_destroy(sharezp->z_sa_hdl);
296 	zfs_znode_free_kmem(sharezp);
297 
298 	return (error);
299 }
300 
301 /*
302  * define a couple of values we need available
303  * for both 64 and 32 bit environments.
304  */
305 #ifndef NBITSMINOR64
306 #define	NBITSMINOR64	32
307 #endif
308 #ifndef MAXMAJ64
309 #define	MAXMAJ64	0xffffffffUL
310 #endif
311 #ifndef	MAXMIN64
312 #define	MAXMIN64	0xffffffffUL
313 #endif
314 
315 /*
316  * Create special expldev for ZFS private use.
317  * Can't use standard expldev since it doesn't do
318  * what we want.  The standard expldev() takes a
319  * dev32_t in LP64 and expands it to a long dev_t.
320  * We need an interface that takes a dev32_t in ILP32
321  * and expands it to a long dev_t.
322  */
323 static uint64_t
zfs_expldev(dev_t dev)324 zfs_expldev(dev_t dev)
325 {
326 	return (((uint64_t)major(dev) << NBITSMINOR64) | minor(dev));
327 }
328 /*
329  * Special cmpldev for ZFS private use.
330  * Can't use standard cmpldev since it takes
331  * a long dev_t and compresses it to dev32_t in
332  * LP64.  We need to do a compaction of a long dev_t
333  * to a dev32_t in ILP32.
334  */
335 dev_t
zfs_cmpldev(uint64_t dev)336 zfs_cmpldev(uint64_t dev)
337 {
338 	return (makedev((dev >> NBITSMINOR64), (dev & MAXMIN64)));
339 }
340 
341 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)342 zfs_znode_sa_init(zfsvfs_t *zfsvfs, znode_t *zp,
343     dmu_buf_t *db, dmu_object_type_t obj_type, sa_handle_t *sa_hdl)
344 {
345 	ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs) || (zfsvfs == zp->z_zfsvfs));
346 	ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zfsvfs, zp->z_id)));
347 
348 	ASSERT(zp->z_sa_hdl == NULL);
349 	ASSERT(zp->z_acl_cached == NULL);
350 	if (sa_hdl == NULL) {
351 		VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, zp,
352 		    SA_HDL_SHARED, &zp->z_sa_hdl));
353 	} else {
354 		zp->z_sa_hdl = sa_hdl;
355 		sa_set_userp(sa_hdl, zp);
356 	}
357 
358 	zp->z_is_sa = (obj_type == DMU_OT_SA) ? B_TRUE : B_FALSE;
359 
360 	/*
361 	 * Slap on VROOT if we are the root znode unless we are the root
362 	 * node of a snapshot mounted under .zfs.
363 	 */
364 	if (zp->z_id == zfsvfs->z_root && zfsvfs->z_parent == zfsvfs)
365 		ZTOV(zp)->v_flag |= VROOT;
366 
367 	vn_exists(ZTOV(zp));
368 }
369 
370 void
zfs_znode_dmu_fini(znode_t * zp)371 zfs_znode_dmu_fini(znode_t *zp)
372 {
373 	ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zp->z_zfsvfs, zp->z_id)) ||
374 	    zp->z_unlinked ||
375 	    ZFS_TEARDOWN_INACTIVE_WRITE_HELD(zp->z_zfsvfs));
376 
377 	sa_handle_destroy(zp->z_sa_hdl);
378 	zp->z_sa_hdl = NULL;
379 }
380 
381 static void
zfs_vnode_forget(vnode_t * vp)382 zfs_vnode_forget(vnode_t *vp)
383 {
384 
385 	/* copied from insmntque_stddtr */
386 	vp->v_data = NULL;
387 	vp->v_op = &dead_vnodeops;
388 	vgone(vp);
389 	vput(vp);
390 }
391 
392 /*
393  * Construct a new znode/vnode and initialize.
394  *
395  * This does not do a call to dmu_set_user() that is
396  * up to the caller to do, in case you don't want to
397  * return the znode
398  */
399 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)400 zfs_znode_alloc(zfsvfs_t *zfsvfs, dmu_buf_t *db, int blksz,
401     dmu_object_type_t obj_type, sa_handle_t *hdl)
402 {
403 	znode_t	*zp;
404 	vnode_t *vp;
405 	uint64_t mode;
406 	uint64_t parent;
407 #ifdef notyet
408 	uint64_t mtime[2], ctime[2];
409 #endif
410 	uint64_t projid = ZFS_DEFAULT_PROJID;
411 	sa_bulk_attr_t bulk[9];
412 	int count = 0;
413 	int error;
414 
415 	zp = zfs_znode_alloc_kmem(KM_SLEEP);
416 
417 #ifndef _ZFS_USE_SMR
418 	KASSERT((zfsvfs->z_parent->z_vfs->mnt_kern_flag & MNTK_FPLOOKUP) == 0,
419 	    ("%s: fast path lookup enabled without smr", __func__));
420 #endif
421 
422 #if __FreeBSD_version >= 1300076
423 	KASSERT(curthread->td_vp_reserved != NULL,
424 	    ("zfs_znode_alloc: getnewvnode without any vnodes reserved"));
425 #else
426 	KASSERT(curthread->td_vp_reserv > 0,
427 	    ("zfs_znode_alloc: getnewvnode without any vnodes reserved"));
428 #endif
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 	ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs));
438 
439 	zp->z_sa_hdl = NULL;
440 	zp->z_unlinked = 0;
441 	zp->z_atime_dirty = 0;
442 	zp->z_mapcnt = 0;
443 	zp->z_id = db->db_object;
444 	zp->z_blksz = blksz;
445 	zp->z_seq = 0x7A4653;
446 	zp->z_sync_cnt = 0;
447 	atomic_store_ptr(&zp->z_cached_symlink, NULL);
448 
449 	vp = ZTOV(zp);
450 
451 	zfs_znode_sa_init(zfsvfs, zp, db, obj_type, hdl);
452 
453 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8);
454 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL, &zp->z_gen, 8);
455 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
456 	    &zp->z_size, 8);
457 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL,
458 	    &zp->z_links, 8);
459 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
460 	    &zp->z_pflags, 8);
461 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL, &parent, 8);
462 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
463 	    &zp->z_atime, 16);
464 #ifdef notyet
465 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
466 	    &mtime, 16);
467 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
468 	    &ctime, 16);
469 #endif
470 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
471 	    &zp->z_uid, 8);
472 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL,
473 	    &zp->z_gid, 8);
474 
475 	if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count) != 0 || zp->z_gen == 0 ||
476 	    (dmu_objset_projectquota_enabled(zfsvfs->z_os) &&
477 	    (zp->z_pflags & ZFS_PROJID) &&
478 	    sa_lookup(zp->z_sa_hdl, SA_ZPL_PROJID(zfsvfs), &projid, 8) != 0)) {
479 		if (hdl == NULL)
480 			sa_handle_destroy(zp->z_sa_hdl);
481 		zfs_vnode_forget(vp);
482 		zp->z_vnode = NULL;
483 		zfs_znode_free_kmem(zp);
484 		return (NULL);
485 	}
486 
487 	zp->z_projid = projid;
488 	zp->z_mode = mode;
489 
490 	/* Cache the xattr parent id */
491 	if (zp->z_pflags & ZFS_XATTR)
492 		zp->z_xattr_parent = parent;
493 
494 	vp->v_type = IFTOVT((mode_t)mode);
495 
496 	switch (vp->v_type) {
497 	case VDIR:
498 		zp->z_zn_prefetch = B_TRUE; /* z_prefetch default is enabled */
499 		break;
500 	case VFIFO:
501 		vp->v_op = &zfs_fifoops;
502 		break;
503 	case VREG:
504 		if (parent == zfsvfs->z_shares_dir) {
505 			ASSERT(zp->z_uid == 0 && zp->z_gid == 0);
506 			vp->v_op = &zfs_shareops;
507 		}
508 		break;
509 	default:
510 			break;
511 	}
512 
513 	mutex_enter(&zfsvfs->z_znodes_lock);
514 	list_insert_tail(&zfsvfs->z_all_znodes, zp);
515 	zfsvfs->z_nr_znodes++;
516 	zp->z_zfsvfs = zfsvfs;
517 	mutex_exit(&zfsvfs->z_znodes_lock);
518 
519 	/*
520 	 * Acquire vnode lock before making it available to the world.
521 	 */
522 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
523 	VN_LOCK_AREC(vp);
524 	if (vp->v_type != VFIFO)
525 		VN_LOCK_ASHARE(vp);
526 
527 	return (zp);
528 }
529 
530 static uint64_t empty_xattr;
531 static uint64_t pad[4];
532 static zfs_acl_phys_t acl_phys;
533 /*
534  * Create a new DMU object to hold a zfs znode.
535  *
536  *	IN:	dzp	- parent directory for new znode
537  *		vap	- file attributes for new znode
538  *		tx	- dmu transaction id for zap operations
539  *		cr	- credentials of caller
540  *		flag	- flags:
541  *			  IS_ROOT_NODE	- new object will be root
542  *			  IS_XATTR	- new object is an attribute
543  *		bonuslen - length of bonus buffer
544  *		setaclp  - File/Dir initial ACL
545  *		fuidp	 - Tracks fuid allocation.
546  *
547  *	OUT:	zpp	- allocated znode
548  *
549  */
550 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)551 zfs_mknode(znode_t *dzp, vattr_t *vap, dmu_tx_t *tx, cred_t *cr,
552     uint_t flag, znode_t **zpp, zfs_acl_ids_t *acl_ids)
553 {
554 	uint64_t	crtime[2], atime[2], mtime[2], ctime[2];
555 	uint64_t	mode, size, links, parent, pflags;
556 	uint64_t	dzp_pflags = 0;
557 	uint64_t	rdev = 0;
558 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
559 	dmu_buf_t	*db;
560 	timestruc_t	now;
561 	uint64_t	gen, obj;
562 	int		err;
563 	int		bonuslen;
564 	int		dnodesize;
565 	sa_handle_t	*sa_hdl;
566 	dmu_object_type_t obj_type;
567 	sa_bulk_attr_t	*sa_attrs;
568 	int		cnt = 0;
569 	zfs_acl_locator_cb_t locate = { 0 };
570 
571 	ASSERT(vap && ((vap->va_mask & AT_MODE) == AT_MODE));
572 
573 	if (zfsvfs->z_replay) {
574 		obj = vap->va_nodeid;
575 		now = vap->va_ctime;		/* see zfs_replay_create() */
576 		gen = vap->va_nblocks;		/* ditto */
577 		dnodesize = vap->va_fsid;	/* ditto */
578 	} else {
579 		obj = 0;
580 		vfs_timestamp(&now);
581 		gen = dmu_tx_get_txg(tx);
582 		dnodesize = dmu_objset_dnodesize(zfsvfs->z_os);
583 	}
584 
585 	if (dnodesize == 0)
586 		dnodesize = DNODE_MIN_SIZE;
587 
588 	obj_type = zfsvfs->z_use_sa ? DMU_OT_SA : DMU_OT_ZNODE;
589 	bonuslen = (obj_type == DMU_OT_SA) ?
590 	    DN_BONUS_SIZE(dnodesize) : ZFS_OLD_ZNODE_PHYS_SIZE;
591 
592 	/*
593 	 * Create a new DMU object.
594 	 */
595 	/*
596 	 * There's currently no mechanism for pre-reading the blocks that will
597 	 * be needed to allocate a new object, so we accept the small chance
598 	 * that there will be an i/o error and we will fail one of the
599 	 * assertions below.
600 	 */
601 	if (vap->va_type == VDIR) {
602 		if (zfsvfs->z_replay) {
603 			VERIFY0(zap_create_claim_norm_dnsize(zfsvfs->z_os, obj,
604 			    zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS,
605 			    obj_type, bonuslen, dnodesize, tx));
606 		} else {
607 			obj = zap_create_norm_dnsize(zfsvfs->z_os,
608 			    zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS,
609 			    obj_type, bonuslen, dnodesize, tx);
610 		}
611 	} else {
612 		if (zfsvfs->z_replay) {
613 			VERIFY0(dmu_object_claim_dnsize(zfsvfs->z_os, obj,
614 			    DMU_OT_PLAIN_FILE_CONTENTS, 0,
615 			    obj_type, bonuslen, dnodesize, tx));
616 		} else {
617 			obj = dmu_object_alloc_dnsize(zfsvfs->z_os,
618 			    DMU_OT_PLAIN_FILE_CONTENTS, 0,
619 			    obj_type, bonuslen, dnodesize, tx);
620 		}
621 	}
622 
623 	ZFS_OBJ_HOLD_ENTER(zfsvfs, obj);
624 	VERIFY(0 == sa_buf_hold(zfsvfs->z_os, obj, NULL, &db));
625 
626 	/*
627 	 * If this is the root, fix up the half-initialized parent pointer
628 	 * to reference the just-allocated physical data area.
629 	 */
630 	if (flag & IS_ROOT_NODE) {
631 		dzp->z_id = obj;
632 	} else {
633 		dzp_pflags = dzp->z_pflags;
634 	}
635 
636 	/*
637 	 * If parent is an xattr, so am I.
638 	 */
639 	if (dzp_pflags & ZFS_XATTR) {
640 		flag |= IS_XATTR;
641 	}
642 
643 	if (zfsvfs->z_use_fuids)
644 		pflags = ZFS_ARCHIVE | ZFS_AV_MODIFIED;
645 	else
646 		pflags = 0;
647 
648 	if (vap->va_type == VDIR) {
649 		size = 2;		/* contents ("." and "..") */
650 		links = (flag & (IS_ROOT_NODE | IS_XATTR)) ? 2 : 1;
651 	} else {
652 		size = links = 0;
653 	}
654 
655 	if (vap->va_type == VBLK || vap->va_type == VCHR) {
656 		rdev = zfs_expldev(vap->va_rdev);
657 	}
658 
659 	parent = dzp->z_id;
660 	mode = acl_ids->z_mode;
661 	if (flag & IS_XATTR)
662 		pflags |= ZFS_XATTR;
663 
664 	/*
665 	 * No execs denied will be determined when zfs_mode_compute() is called.
666 	 */
667 	pflags |= acl_ids->z_aclp->z_hints &
668 	    (ZFS_ACL_TRIVIAL|ZFS_INHERIT_ACE|ZFS_ACL_AUTO_INHERIT|
669 	    ZFS_ACL_DEFAULTED|ZFS_ACL_PROTECTED);
670 
671 	ZFS_TIME_ENCODE(&now, crtime);
672 	ZFS_TIME_ENCODE(&now, ctime);
673 
674 	if (vap->va_mask & AT_ATIME) {
675 		ZFS_TIME_ENCODE(&vap->va_atime, atime);
676 	} else {
677 		ZFS_TIME_ENCODE(&now, atime);
678 	}
679 
680 	if (vap->va_mask & AT_MTIME) {
681 		ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
682 	} else {
683 		ZFS_TIME_ENCODE(&now, mtime);
684 	}
685 
686 	/* Now add in all of the "SA" attributes */
687 	VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, NULL, SA_HDL_SHARED,
688 	    &sa_hdl));
689 
690 	/*
691 	 * Setup the array of attributes to be replaced/set on the new file
692 	 *
693 	 * order for  DMU_OT_ZNODE is critical since it needs to be constructed
694 	 * in the old znode_phys_t format.  Don't change this ordering
695 	 */
696 	sa_attrs = kmem_alloc(sizeof (sa_bulk_attr_t) * ZPL_END, KM_SLEEP);
697 
698 	if (obj_type == DMU_OT_ZNODE) {
699 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs),
700 		    NULL, &atime, 16);
701 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs),
702 		    NULL, &mtime, 16);
703 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs),
704 		    NULL, &ctime, 16);
705 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs),
706 		    NULL, &crtime, 16);
707 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs),
708 		    NULL, &gen, 8);
709 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs),
710 		    NULL, &mode, 8);
711 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs),
712 		    NULL, &size, 8);
713 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs),
714 		    NULL, &parent, 8);
715 	} else {
716 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs),
717 		    NULL, &mode, 8);
718 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs),
719 		    NULL, &size, 8);
720 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs),
721 		    NULL, &gen, 8);
722 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs),
723 		    NULL, &acl_ids->z_fuid, 8);
724 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs),
725 		    NULL, &acl_ids->z_fgid, 8);
726 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs),
727 		    NULL, &parent, 8);
728 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs),
729 		    NULL, &pflags, 8);
730 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs),
731 		    NULL, &atime, 16);
732 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs),
733 		    NULL, &mtime, 16);
734 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs),
735 		    NULL, &ctime, 16);
736 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs),
737 		    NULL, &crtime, 16);
738 	}
739 
740 	SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_LINKS(zfsvfs), NULL, &links, 8);
741 
742 	if (obj_type == DMU_OT_ZNODE) {
743 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_XATTR(zfsvfs), NULL,
744 		    &empty_xattr, 8);
745 	}
746 	if (obj_type == DMU_OT_ZNODE ||
747 	    (vap->va_type == VBLK || vap->va_type == VCHR)) {
748 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_RDEV(zfsvfs),
749 		    NULL, &rdev, 8);
750 
751 	}
752 	if (obj_type == DMU_OT_ZNODE) {
753 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs),
754 		    NULL, &pflags, 8);
755 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs), NULL,
756 		    &acl_ids->z_fuid, 8);
757 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs), NULL,
758 		    &acl_ids->z_fgid, 8);
759 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PAD(zfsvfs), NULL, pad,
760 		    sizeof (uint64_t) * 4);
761 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ZNODE_ACL(zfsvfs), NULL,
762 		    &acl_phys, sizeof (zfs_acl_phys_t));
763 	} else if (acl_ids->z_aclp->z_version >= ZFS_ACL_VERSION_FUID) {
764 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_COUNT(zfsvfs), NULL,
765 		    &acl_ids->z_aclp->z_acl_count, 8);
766 		locate.cb_aclp = acl_ids->z_aclp;
767 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_ACES(zfsvfs),
768 		    zfs_acl_data_locator, &locate,
769 		    acl_ids->z_aclp->z_acl_bytes);
770 		mode = zfs_mode_compute(mode, acl_ids->z_aclp, &pflags,
771 		    acl_ids->z_fuid, acl_ids->z_fgid);
772 	}
773 
774 	VERIFY(sa_replace_all_by_template(sa_hdl, sa_attrs, cnt, tx) == 0);
775 
776 	if (!(flag & IS_ROOT_NODE)) {
777 		*zpp = zfs_znode_alloc(zfsvfs, db, 0, obj_type, sa_hdl);
778 		ASSERT(*zpp != NULL);
779 	} else {
780 		/*
781 		 * If we are creating the root node, the "parent" we
782 		 * passed in is the znode for the root.
783 		 */
784 		*zpp = dzp;
785 
786 		(*zpp)->z_sa_hdl = sa_hdl;
787 	}
788 
789 	(*zpp)->z_pflags = pflags;
790 	(*zpp)->z_mode = mode;
791 	(*zpp)->z_dnodesize = dnodesize;
792 
793 	if (vap->va_mask & AT_XVATTR)
794 		zfs_xvattr_set(*zpp, (xvattr_t *)vap, tx);
795 
796 	if (obj_type == DMU_OT_ZNODE ||
797 	    acl_ids->z_aclp->z_version < ZFS_ACL_VERSION_FUID) {
798 		VERIFY0(zfs_aclset_common(*zpp, acl_ids->z_aclp, cr, tx));
799 	}
800 	if (!(flag & IS_ROOT_NODE)) {
801 		vnode_t *vp;
802 
803 		vp = ZTOV(*zpp);
804 		vp->v_vflag |= VV_FORCEINSMQ;
805 		err = insmntque(vp, zfsvfs->z_vfs);
806 		vp->v_vflag &= ~VV_FORCEINSMQ;
807 		KASSERT(err == 0, ("insmntque() failed: error %d", err));
808 	}
809 	kmem_free(sa_attrs, sizeof (sa_bulk_attr_t) * ZPL_END);
810 	ZFS_OBJ_HOLD_EXIT(zfsvfs, obj);
811 }
812 
813 /*
814  * Update in-core attributes.  It is assumed the caller will be doing an
815  * sa_bulk_update to push the changes out.
816  */
817 void
zfs_xvattr_set(znode_t * zp,xvattr_t * xvap,dmu_tx_t * tx)818 zfs_xvattr_set(znode_t *zp, xvattr_t *xvap, dmu_tx_t *tx)
819 {
820 	xoptattr_t *xoap;
821 
822 	xoap = xva_getxoptattr(xvap);
823 	ASSERT(xoap);
824 
825 	ASSERT_VOP_IN_SEQC(ZTOV(zp));
826 
827 	if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
828 		uint64_t times[2];
829 		ZFS_TIME_ENCODE(&xoap->xoa_createtime, times);
830 		(void) sa_update(zp->z_sa_hdl, SA_ZPL_CRTIME(zp->z_zfsvfs),
831 		    &times, sizeof (times), tx);
832 		XVA_SET_RTN(xvap, XAT_CREATETIME);
833 	}
834 	if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
835 		ZFS_ATTR_SET(zp, ZFS_READONLY, xoap->xoa_readonly,
836 		    zp->z_pflags, tx);
837 		XVA_SET_RTN(xvap, XAT_READONLY);
838 	}
839 	if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
840 		ZFS_ATTR_SET(zp, ZFS_HIDDEN, xoap->xoa_hidden,
841 		    zp->z_pflags, tx);
842 		XVA_SET_RTN(xvap, XAT_HIDDEN);
843 	}
844 	if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
845 		ZFS_ATTR_SET(zp, ZFS_SYSTEM, xoap->xoa_system,
846 		    zp->z_pflags, tx);
847 		XVA_SET_RTN(xvap, XAT_SYSTEM);
848 	}
849 	if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
850 		ZFS_ATTR_SET(zp, ZFS_ARCHIVE, xoap->xoa_archive,
851 		    zp->z_pflags, tx);
852 		XVA_SET_RTN(xvap, XAT_ARCHIVE);
853 	}
854 	if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
855 		ZFS_ATTR_SET(zp, ZFS_IMMUTABLE, xoap->xoa_immutable,
856 		    zp->z_pflags, tx);
857 		XVA_SET_RTN(xvap, XAT_IMMUTABLE);
858 	}
859 	if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
860 		ZFS_ATTR_SET(zp, ZFS_NOUNLINK, xoap->xoa_nounlink,
861 		    zp->z_pflags, tx);
862 		XVA_SET_RTN(xvap, XAT_NOUNLINK);
863 	}
864 	if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
865 		ZFS_ATTR_SET(zp, ZFS_APPENDONLY, xoap->xoa_appendonly,
866 		    zp->z_pflags, tx);
867 		XVA_SET_RTN(xvap, XAT_APPENDONLY);
868 	}
869 	if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
870 		ZFS_ATTR_SET(zp, ZFS_NODUMP, xoap->xoa_nodump,
871 		    zp->z_pflags, tx);
872 		XVA_SET_RTN(xvap, XAT_NODUMP);
873 	}
874 	if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
875 		ZFS_ATTR_SET(zp, ZFS_OPAQUE, xoap->xoa_opaque,
876 		    zp->z_pflags, tx);
877 		XVA_SET_RTN(xvap, XAT_OPAQUE);
878 	}
879 	if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
880 		ZFS_ATTR_SET(zp, ZFS_AV_QUARANTINED,
881 		    xoap->xoa_av_quarantined, zp->z_pflags, tx);
882 		XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
883 	}
884 	if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
885 		ZFS_ATTR_SET(zp, ZFS_AV_MODIFIED, xoap->xoa_av_modified,
886 		    zp->z_pflags, tx);
887 		XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
888 	}
889 	if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) {
890 		zfs_sa_set_scanstamp(zp, xvap, tx);
891 		XVA_SET_RTN(xvap, XAT_AV_SCANSTAMP);
892 	}
893 	if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
894 		ZFS_ATTR_SET(zp, ZFS_REPARSE, xoap->xoa_reparse,
895 		    zp->z_pflags, tx);
896 		XVA_SET_RTN(xvap, XAT_REPARSE);
897 	}
898 	if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
899 		ZFS_ATTR_SET(zp, ZFS_OFFLINE, xoap->xoa_offline,
900 		    zp->z_pflags, tx);
901 		XVA_SET_RTN(xvap, XAT_OFFLINE);
902 	}
903 	if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
904 		ZFS_ATTR_SET(zp, ZFS_SPARSE, xoap->xoa_sparse,
905 		    zp->z_pflags, tx);
906 		XVA_SET_RTN(xvap, XAT_SPARSE);
907 	}
908 }
909 
910 int
zfs_zget(zfsvfs_t * zfsvfs,uint64_t obj_num,znode_t ** zpp)911 zfs_zget(zfsvfs_t *zfsvfs, uint64_t obj_num, znode_t **zpp)
912 {
913 	dmu_object_info_t doi;
914 	dmu_buf_t	*db;
915 	znode_t		*zp;
916 	vnode_t		*vp;
917 	sa_handle_t	*hdl;
918 	struct thread	*td;
919 	int locked;
920 	int err;
921 
922 	td = curthread;
923 	getnewvnode_reserve_();
924 again:
925 	*zpp = NULL;
926 	ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num);
927 
928 	err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db);
929 	if (err) {
930 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
931 		getnewvnode_drop_reserve();
932 		return (err);
933 	}
934 
935 	dmu_object_info_from_db(db, &doi);
936 	if (doi.doi_bonus_type != DMU_OT_SA &&
937 	    (doi.doi_bonus_type != DMU_OT_ZNODE ||
938 	    (doi.doi_bonus_type == DMU_OT_ZNODE &&
939 	    doi.doi_bonus_size < sizeof (znode_phys_t)))) {
940 		sa_buf_rele(db, NULL);
941 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
942 		getnewvnode_drop_reserve();
943 		return (SET_ERROR(EINVAL));
944 	}
945 
946 	hdl = dmu_buf_get_user(db);
947 	if (hdl != NULL) {
948 		zp  = sa_get_userdata(hdl);
949 
950 		/*
951 		 * Since "SA" does immediate eviction we
952 		 * should never find a sa handle that doesn't
953 		 * know about the znode.
954 		 */
955 		ASSERT3P(zp, !=, NULL);
956 		ASSERT3U(zp->z_id, ==, obj_num);
957 		if (zp->z_unlinked) {
958 			err = SET_ERROR(ENOENT);
959 		} else {
960 			vp = ZTOV(zp);
961 			/*
962 			 * Don't let the vnode disappear after
963 			 * ZFS_OBJ_HOLD_EXIT.
964 			 */
965 			VN_HOLD(vp);
966 			*zpp = zp;
967 			err = 0;
968 		}
969 
970 		sa_buf_rele(db, NULL);
971 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
972 
973 		if (err) {
974 			getnewvnode_drop_reserve();
975 			return (err);
976 		}
977 
978 		locked = VOP_ISLOCKED(vp);
979 		VI_LOCK(vp);
980 		if (VN_IS_DOOMED(vp) && locked != LK_EXCLUSIVE) {
981 			/*
982 			 * The vnode is doomed and this thread doesn't
983 			 * hold the exclusive lock on it, so the vnode
984 			 * must be being reclaimed by another thread.
985 			 * Otherwise the doomed vnode is being reclaimed
986 			 * by this thread and zfs_zget is called from
987 			 * ZIL internals.
988 			 */
989 			VI_UNLOCK(vp);
990 
991 			/*
992 			 * XXX vrele() locks the vnode when the last reference
993 			 * is dropped.  Although in this case the vnode is
994 			 * doomed / dead and so no inactivation is required,
995 			 * the vnode lock is still acquired.  That could result
996 			 * in a LOR with z_teardown_lock if another thread holds
997 			 * the vnode's lock and tries to take z_teardown_lock.
998 			 * But that is only possible if the other thread peforms
999 			 * a ZFS vnode operation on the vnode.  That either
1000 			 * should not happen if the vnode is dead or the thread
1001 			 * should also have a reference to the vnode and thus
1002 			 * our reference is not last.
1003 			 */
1004 			VN_RELE(vp);
1005 			goto again;
1006 		}
1007 		VI_UNLOCK(vp);
1008 		getnewvnode_drop_reserve();
1009 		return (err);
1010 	}
1011 
1012 	/*
1013 	 * Not found create new znode/vnode
1014 	 * but only if file exists.
1015 	 *
1016 	 * There is a small window where zfs_vget() could
1017 	 * find this object while a file create is still in
1018 	 * progress.  This is checked for in zfs_znode_alloc()
1019 	 *
1020 	 * if zfs_znode_alloc() fails it will drop the hold on the
1021 	 * bonus buffer.
1022 	 */
1023 	zp = zfs_znode_alloc(zfsvfs, db, doi.doi_data_block_size,
1024 	    doi.doi_bonus_type, NULL);
1025 	if (zp == NULL) {
1026 		err = SET_ERROR(ENOENT);
1027 	} else {
1028 		*zpp = zp;
1029 	}
1030 	if (err == 0) {
1031 		vnode_t *vp = ZTOV(zp);
1032 
1033 		err = insmntque(vp, zfsvfs->z_vfs);
1034 		if (err == 0) {
1035 			vp->v_hash = obj_num;
1036 			VOP_UNLOCK1(vp);
1037 		} else {
1038 			zp->z_vnode = NULL;
1039 			zfs_znode_dmu_fini(zp);
1040 			zfs_znode_free(zp);
1041 			*zpp = NULL;
1042 		}
1043 	}
1044 	ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1045 	getnewvnode_drop_reserve();
1046 	return (err);
1047 }
1048 
1049 int
zfs_rezget(znode_t * zp)1050 zfs_rezget(znode_t *zp)
1051 {
1052 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1053 	dmu_object_info_t doi;
1054 	dmu_buf_t *db;
1055 	vnode_t *vp;
1056 	uint64_t obj_num = zp->z_id;
1057 	uint64_t mode, size;
1058 	sa_bulk_attr_t bulk[8];
1059 	int err;
1060 	int count = 0;
1061 	uint64_t gen;
1062 
1063 	/*
1064 	 * Remove cached pages before reloading the znode, so that they are not
1065 	 * lingering after we run into any error.  Ideally, we should vgone()
1066 	 * the vnode in case of error, but currently we cannot do that
1067 	 * because of the LOR between the vnode lock and z_teardown_lock.
1068 	 * So, instead, we have to "doom" the znode in the illumos style.
1069 	 */
1070 	vp = ZTOV(zp);
1071 	vn_pages_remove(vp, 0, 0);
1072 
1073 	ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num);
1074 
1075 	mutex_enter(&zp->z_acl_lock);
1076 	if (zp->z_acl_cached) {
1077 		zfs_acl_free(zp->z_acl_cached);
1078 		zp->z_acl_cached = NULL;
1079 	}
1080 
1081 	mutex_exit(&zp->z_acl_lock);
1082 	ASSERT(zp->z_sa_hdl == NULL);
1083 	err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db);
1084 	if (err) {
1085 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1086 		return (err);
1087 	}
1088 
1089 	dmu_object_info_from_db(db, &doi);
1090 	if (doi.doi_bonus_type != DMU_OT_SA &&
1091 	    (doi.doi_bonus_type != DMU_OT_ZNODE ||
1092 	    (doi.doi_bonus_type == DMU_OT_ZNODE &&
1093 	    doi.doi_bonus_size < sizeof (znode_phys_t)))) {
1094 		sa_buf_rele(db, NULL);
1095 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1096 		return (SET_ERROR(EINVAL));
1097 	}
1098 
1099 	zfs_znode_sa_init(zfsvfs, zp, db, doi.doi_bonus_type, NULL);
1100 	size = zp->z_size;
1101 
1102 	/* reload cached values */
1103 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL,
1104 	    &gen, sizeof (gen));
1105 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
1106 	    &zp->z_size, sizeof (zp->z_size));
1107 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL,
1108 	    &zp->z_links, sizeof (zp->z_links));
1109 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
1110 	    &zp->z_pflags, sizeof (zp->z_pflags));
1111 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
1112 	    &zp->z_atime, sizeof (zp->z_atime));
1113 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
1114 	    &zp->z_uid, sizeof (zp->z_uid));
1115 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL,
1116 	    &zp->z_gid, sizeof (zp->z_gid));
1117 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
1118 	    &mode, sizeof (mode));
1119 
1120 	if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) {
1121 		zfs_znode_dmu_fini(zp);
1122 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1123 		return (SET_ERROR(EIO));
1124 	}
1125 
1126 	zp->z_mode = mode;
1127 
1128 	if (gen != zp->z_gen) {
1129 		zfs_znode_dmu_fini(zp);
1130 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1131 		return (SET_ERROR(EIO));
1132 	}
1133 
1134 	/*
1135 	 * It is highly improbable but still quite possible that two
1136 	 * objects in different datasets are created with the same
1137 	 * object numbers and in transaction groups with the same
1138 	 * numbers.  znodes corresponding to those objects would
1139 	 * have the same z_id and z_gen, but their other attributes
1140 	 * may be different.
1141 	 * zfs recv -F may replace one of such objects with the other.
1142 	 * As a result file properties recorded in the replaced
1143 	 * object's vnode may no longer match the received object's
1144 	 * properties.  At present the only cached property is the
1145 	 * files type recorded in v_type.
1146 	 * So, handle this case by leaving the old vnode and znode
1147 	 * disassociated from the actual object.  A new vnode and a
1148 	 * znode will be created if the object is accessed
1149 	 * (e.g. via a look-up).  The old vnode and znode will be
1150 	 * recycled when the last vnode reference is dropped.
1151 	 */
1152 	if (vp->v_type != IFTOVT((mode_t)zp->z_mode)) {
1153 		zfs_znode_dmu_fini(zp);
1154 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1155 		return (SET_ERROR(EIO));
1156 	}
1157 
1158 	/*
1159 	 * If the file has zero links, then it has been unlinked on the send
1160 	 * side and it must be in the received unlinked set.
1161 	 * We call zfs_znode_dmu_fini() now to prevent any accesses to the
1162 	 * stale data and to prevent automatically removal of the file in
1163 	 * zfs_zinactive().  The file will be removed either when it is removed
1164 	 * on the send side and the next incremental stream is received or
1165 	 * when the unlinked set gets processed.
1166 	 */
1167 	zp->z_unlinked = (zp->z_links == 0);
1168 	if (zp->z_unlinked) {
1169 		zfs_znode_dmu_fini(zp);
1170 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1171 		return (0);
1172 	}
1173 
1174 	zp->z_blksz = doi.doi_data_block_size;
1175 	if (zp->z_size != size)
1176 		vnode_pager_setsize(vp, zp->z_size);
1177 
1178 	ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1179 
1180 	return (0);
1181 }
1182 
1183 void
zfs_znode_delete(znode_t * zp,dmu_tx_t * tx)1184 zfs_znode_delete(znode_t *zp, dmu_tx_t *tx)
1185 {
1186 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1187 	objset_t *os = zfsvfs->z_os;
1188 	uint64_t obj = zp->z_id;
1189 	uint64_t acl_obj = zfs_external_acl(zp);
1190 
1191 	ZFS_OBJ_HOLD_ENTER(zfsvfs, obj);
1192 	if (acl_obj) {
1193 		VERIFY(!zp->z_is_sa);
1194 		VERIFY(0 == dmu_object_free(os, acl_obj, tx));
1195 	}
1196 	VERIFY(0 == dmu_object_free(os, obj, tx));
1197 	zfs_znode_dmu_fini(zp);
1198 	ZFS_OBJ_HOLD_EXIT(zfsvfs, obj);
1199 	zfs_znode_free(zp);
1200 }
1201 
1202 void
zfs_zinactive(znode_t * zp)1203 zfs_zinactive(znode_t *zp)
1204 {
1205 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1206 	uint64_t z_id = zp->z_id;
1207 
1208 	ASSERT(zp->z_sa_hdl);
1209 
1210 	/*
1211 	 * Don't allow a zfs_zget() while were trying to release this znode
1212 	 */
1213 	ZFS_OBJ_HOLD_ENTER(zfsvfs, z_id);
1214 
1215 	/*
1216 	 * If this was the last reference to a file with no links, remove
1217 	 * the file from the file system unless the file system is mounted
1218 	 * read-only.  That can happen, for example, if the file system was
1219 	 * originally read-write, the file was opened, then unlinked and
1220 	 * the file system was made read-only before the file was finally
1221 	 * closed.  The file will remain in the unlinked set.
1222 	 */
1223 	if (zp->z_unlinked) {
1224 		ASSERT(!zfsvfs->z_issnap);
1225 		if ((zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) == 0) {
1226 			ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id);
1227 			zfs_rmnode(zp);
1228 			return;
1229 		}
1230 	}
1231 
1232 	zfs_znode_dmu_fini(zp);
1233 	ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id);
1234 	zfs_znode_free(zp);
1235 }
1236 
1237 void
zfs_znode_free(znode_t * zp)1238 zfs_znode_free(znode_t *zp)
1239 {
1240 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1241 	char *symlink;
1242 
1243 	ASSERT(zp->z_sa_hdl == NULL);
1244 	zp->z_vnode = NULL;
1245 	mutex_enter(&zfsvfs->z_znodes_lock);
1246 	POINTER_INVALIDATE(&zp->z_zfsvfs);
1247 	list_remove(&zfsvfs->z_all_znodes, zp);
1248 	zfsvfs->z_nr_znodes--;
1249 	mutex_exit(&zfsvfs->z_znodes_lock);
1250 	symlink = atomic_load_ptr(&zp->z_cached_symlink);
1251 	if (symlink != NULL) {
1252 		atomic_store_rel_ptr((uintptr_t *)&zp->z_cached_symlink, (uintptr_t)NULL);
1253 		cache_symlink_free(symlink, strlen(symlink) + 1);
1254 	}
1255 
1256 	if (zp->z_acl_cached) {
1257 		zfs_acl_free(zp->z_acl_cached);
1258 		zp->z_acl_cached = NULL;
1259 	}
1260 
1261 	zfs_znode_free_kmem(zp);
1262 }
1263 
1264 void
zfs_tstamp_update_setup_ext(znode_t * zp,uint_t flag,uint64_t mtime[2],uint64_t ctime[2],boolean_t have_tx)1265 zfs_tstamp_update_setup_ext(znode_t *zp, uint_t flag, uint64_t mtime[2],
1266     uint64_t ctime[2], boolean_t have_tx)
1267 {
1268 	timestruc_t	now;
1269 
1270 	vfs_timestamp(&now);
1271 
1272 	if (have_tx) {	/* will sa_bulk_update happen really soon? */
1273 		zp->z_atime_dirty = 0;
1274 		zp->z_seq++;
1275 	} else {
1276 		zp->z_atime_dirty = 1;
1277 	}
1278 
1279 	if (flag & AT_ATIME) {
1280 		ZFS_TIME_ENCODE(&now, zp->z_atime);
1281 	}
1282 
1283 	if (flag & AT_MTIME) {
1284 		ZFS_TIME_ENCODE(&now, mtime);
1285 		if (zp->z_zfsvfs->z_use_fuids) {
1286 			zp->z_pflags |= (ZFS_ARCHIVE |
1287 			    ZFS_AV_MODIFIED);
1288 		}
1289 	}
1290 
1291 	if (flag & AT_CTIME) {
1292 		ZFS_TIME_ENCODE(&now, ctime);
1293 		if (zp->z_zfsvfs->z_use_fuids)
1294 			zp->z_pflags |= ZFS_ARCHIVE;
1295 	}
1296 }
1297 
1298 
1299 void
zfs_tstamp_update_setup(znode_t * zp,uint_t flag,uint64_t mtime[2],uint64_t ctime[2])1300 zfs_tstamp_update_setup(znode_t *zp, uint_t flag, uint64_t mtime[2],
1301     uint64_t ctime[2])
1302 {
1303 	zfs_tstamp_update_setup_ext(zp, flag, mtime, ctime, B_TRUE);
1304 }
1305 /*
1306  * Grow the block size for a file.
1307  *
1308  *	IN:	zp	- znode of file to free data in.
1309  *		size	- requested block size
1310  *		tx	- open transaction.
1311  *
1312  * NOTE: this function assumes that the znode is write locked.
1313  */
1314 void
zfs_grow_blocksize(znode_t * zp,uint64_t size,dmu_tx_t * tx)1315 zfs_grow_blocksize(znode_t *zp, uint64_t size, dmu_tx_t *tx)
1316 {
1317 	int		error;
1318 	u_longlong_t	dummy;
1319 
1320 	if (size <= zp->z_blksz)
1321 		return;
1322 	/*
1323 	 * If the file size is already greater than the current blocksize,
1324 	 * we will not grow.  If there is more than one block in a file,
1325 	 * the blocksize cannot change.
1326 	 */
1327 	if (zp->z_blksz && zp->z_size > zp->z_blksz)
1328 		return;
1329 
1330 	error = dmu_object_set_blocksize(zp->z_zfsvfs->z_os, zp->z_id,
1331 	    size, 0, tx);
1332 
1333 	if (error == ENOTSUP)
1334 		return;
1335 	ASSERT0(error);
1336 
1337 	/* What blocksize did we actually get? */
1338 	dmu_object_size_from_db(sa_get_db(zp->z_sa_hdl), &zp->z_blksz, &dummy);
1339 }
1340 
1341 /*
1342  * Increase the file length
1343  *
1344  *	IN:	zp	- znode of file to free data in.
1345  *		end	- new end-of-file
1346  *
1347  *	RETURN:	0 on success, error code on failure
1348  */
1349 static int
zfs_extend(znode_t * zp,uint64_t end)1350 zfs_extend(znode_t *zp, uint64_t end)
1351 {
1352 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1353 	dmu_tx_t *tx;
1354 	zfs_locked_range_t *lr;
1355 	uint64_t newblksz;
1356 	int error;
1357 
1358 	/*
1359 	 * We will change zp_size, lock the whole file.
1360 	 */
1361 	lr = zfs_rangelock_enter(&zp->z_rangelock, 0, UINT64_MAX, RL_WRITER);
1362 
1363 	/*
1364 	 * Nothing to do if file already at desired length.
1365 	 */
1366 	if (end <= zp->z_size) {
1367 		zfs_rangelock_exit(lr);
1368 		return (0);
1369 	}
1370 	tx = dmu_tx_create(zfsvfs->z_os);
1371 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1372 	zfs_sa_upgrade_txholds(tx, zp);
1373 	if (end > zp->z_blksz &&
1374 	    (!ISP2(zp->z_blksz) || zp->z_blksz < zfsvfs->z_max_blksz)) {
1375 		/*
1376 		 * We are growing the file past the current block size.
1377 		 */
1378 		if (zp->z_blksz > zp->z_zfsvfs->z_max_blksz) {
1379 			/*
1380 			 * File's blocksize is already larger than the
1381 			 * "recordsize" property.  Only let it grow to
1382 			 * the next power of 2.
1383 			 */
1384 			ASSERT(!ISP2(zp->z_blksz));
1385 			newblksz = MIN(end, 1 << highbit64(zp->z_blksz));
1386 		} else {
1387 			newblksz = MIN(end, zp->z_zfsvfs->z_max_blksz);
1388 		}
1389 		dmu_tx_hold_write(tx, zp->z_id, 0, newblksz);
1390 	} else {
1391 		newblksz = 0;
1392 	}
1393 
1394 	error = dmu_tx_assign(tx, TXG_WAIT);
1395 	if (error) {
1396 		dmu_tx_abort(tx);
1397 		zfs_rangelock_exit(lr);
1398 		return (error);
1399 	}
1400 
1401 	if (newblksz)
1402 		zfs_grow_blocksize(zp, newblksz, tx);
1403 
1404 	zp->z_size = end;
1405 
1406 	VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zp->z_zfsvfs),
1407 	    &zp->z_size, sizeof (zp->z_size), tx));
1408 
1409 	vnode_pager_setsize(ZTOV(zp), end);
1410 
1411 	zfs_rangelock_exit(lr);
1412 
1413 	dmu_tx_commit(tx);
1414 
1415 	return (0);
1416 }
1417 
1418 /*
1419  * Free space in a file.
1420  *
1421  *	IN:	zp	- znode of file to free data in.
1422  *		off	- start of section to free.
1423  *		len	- length of section to free.
1424  *
1425  *	RETURN:	0 on success, error code on failure
1426  */
1427 static int
zfs_free_range(znode_t * zp,uint64_t off,uint64_t len)1428 zfs_free_range(znode_t *zp, uint64_t off, uint64_t len)
1429 {
1430 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1431 	zfs_locked_range_t *lr;
1432 	int error;
1433 
1434 	/*
1435 	 * Lock the range being freed.
1436 	 */
1437 	lr = zfs_rangelock_enter(&zp->z_rangelock, off, len, RL_WRITER);
1438 
1439 	/*
1440 	 * Nothing to do if file already at desired length.
1441 	 */
1442 	if (off >= zp->z_size) {
1443 		zfs_rangelock_exit(lr);
1444 		return (0);
1445 	}
1446 
1447 	if (off + len > zp->z_size)
1448 		len = zp->z_size - off;
1449 
1450 	error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, off, len);
1451 
1452 	if (error == 0) {
1453 		/*
1454 		 * In FreeBSD we cannot free block in the middle of a file,
1455 		 * but only at the end of a file, so this code path should
1456 		 * never happen.
1457 		 */
1458 		vnode_pager_setsize(ZTOV(zp), off);
1459 	}
1460 
1461 	zfs_rangelock_exit(lr);
1462 
1463 	return (error);
1464 }
1465 
1466 /*
1467  * Truncate a file
1468  *
1469  *	IN:	zp	- znode of file to free data in.
1470  *		end	- new end-of-file.
1471  *
1472  *	RETURN:	0 on success, error code on failure
1473  */
1474 static int
zfs_trunc(znode_t * zp,uint64_t end)1475 zfs_trunc(znode_t *zp, uint64_t end)
1476 {
1477 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1478 	vnode_t *vp = ZTOV(zp);
1479 	dmu_tx_t *tx;
1480 	zfs_locked_range_t *lr;
1481 	int error;
1482 	sa_bulk_attr_t bulk[2];
1483 	int count = 0;
1484 
1485 	/*
1486 	 * We will change zp_size, lock the whole file.
1487 	 */
1488 	lr = zfs_rangelock_enter(&zp->z_rangelock, 0, UINT64_MAX, RL_WRITER);
1489 
1490 	/*
1491 	 * Nothing to do if file already at desired length.
1492 	 */
1493 	if (end >= zp->z_size) {
1494 		zfs_rangelock_exit(lr);
1495 		return (0);
1496 	}
1497 
1498 	error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, end,
1499 	    DMU_OBJECT_END);
1500 	if (error) {
1501 		zfs_rangelock_exit(lr);
1502 		return (error);
1503 	}
1504 	tx = dmu_tx_create(zfsvfs->z_os);
1505 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1506 	zfs_sa_upgrade_txholds(tx, zp);
1507 	dmu_tx_mark_netfree(tx);
1508 	error = dmu_tx_assign(tx, TXG_WAIT);
1509 	if (error) {
1510 		dmu_tx_abort(tx);
1511 		zfs_rangelock_exit(lr);
1512 		return (error);
1513 	}
1514 
1515 	zp->z_size = end;
1516 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs),
1517 	    NULL, &zp->z_size, sizeof (zp->z_size));
1518 
1519 	if (end == 0) {
1520 		zp->z_pflags &= ~ZFS_SPARSE;
1521 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
1522 		    NULL, &zp->z_pflags, 8);
1523 	}
1524 	VERIFY(sa_bulk_update(zp->z_sa_hdl, bulk, count, tx) == 0);
1525 
1526 	dmu_tx_commit(tx);
1527 
1528 	/*
1529 	 * Clear any mapped pages in the truncated region.  This has to
1530 	 * happen outside of the transaction to avoid the possibility of
1531 	 * a deadlock with someone trying to push a page that we are
1532 	 * about to invalidate.
1533 	 */
1534 	vnode_pager_setsize(vp, end);
1535 
1536 	zfs_rangelock_exit(lr);
1537 
1538 	return (0);
1539 }
1540 
1541 /*
1542  * Free space in a file
1543  *
1544  *	IN:	zp	- znode of file to free data in.
1545  *		off	- start of range
1546  *		len	- end of range (0 => EOF)
1547  *		flag	- current file open mode flags.
1548  *		log	- TRUE if this action should be logged
1549  *
1550  *	RETURN:	0 on success, error code on failure
1551  */
1552 int
zfs_freesp(znode_t * zp,uint64_t off,uint64_t len,int flag,boolean_t log)1553 zfs_freesp(znode_t *zp, uint64_t off, uint64_t len, int flag, boolean_t log)
1554 {
1555 	dmu_tx_t *tx;
1556 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1557 	zilog_t *zilog = zfsvfs->z_log;
1558 	uint64_t mode;
1559 	uint64_t mtime[2], ctime[2];
1560 	sa_bulk_attr_t bulk[3];
1561 	int count = 0;
1562 	int error;
1563 
1564 	if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs), &mode,
1565 	    sizeof (mode))) != 0)
1566 		return (error);
1567 
1568 	if (off > zp->z_size) {
1569 		error =  zfs_extend(zp, off+len);
1570 		if (error == 0 && log)
1571 			goto log;
1572 		else
1573 			return (error);
1574 	}
1575 
1576 	if (len == 0) {
1577 		error = zfs_trunc(zp, off);
1578 	} else {
1579 		if ((error = zfs_free_range(zp, off, len)) == 0 &&
1580 		    off + len > zp->z_size)
1581 			error = zfs_extend(zp, off+len);
1582 	}
1583 	if (error || !log)
1584 		return (error);
1585 log:
1586 	tx = dmu_tx_create(zfsvfs->z_os);
1587 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1588 	zfs_sa_upgrade_txholds(tx, zp);
1589 	error = dmu_tx_assign(tx, TXG_WAIT);
1590 	if (error) {
1591 		dmu_tx_abort(tx);
1592 		return (error);
1593 	}
1594 
1595 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, mtime, 16);
1596 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, ctime, 16);
1597 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
1598 	    NULL, &zp->z_pflags, 8);
1599 	zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime);
1600 	error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
1601 	ASSERT(error == 0);
1602 
1603 	zfs_log_truncate(zilog, tx, TX_TRUNCATE, zp, off, len);
1604 
1605 	dmu_tx_commit(tx);
1606 	return (0);
1607 }
1608 
1609 void
zfs_create_fs(objset_t * os,cred_t * cr,nvlist_t * zplprops,dmu_tx_t * tx)1610 zfs_create_fs(objset_t *os, cred_t *cr, nvlist_t *zplprops, dmu_tx_t *tx)
1611 {
1612 	uint64_t	moid, obj, sa_obj, version;
1613 	uint64_t	sense = ZFS_CASE_SENSITIVE;
1614 	uint64_t	norm = 0;
1615 	nvpair_t	*elem;
1616 	int		error;
1617 	int		i;
1618 	znode_t		*rootzp = NULL;
1619 	zfsvfs_t	*zfsvfs;
1620 	vattr_t		vattr;
1621 	znode_t		*zp;
1622 	zfs_acl_ids_t	acl_ids;
1623 
1624 	/*
1625 	 * First attempt to create master node.
1626 	 */
1627 	/*
1628 	 * In an empty objset, there are no blocks to read and thus
1629 	 * there can be no i/o errors (which we assert below).
1630 	 */
1631 	moid = MASTER_NODE_OBJ;
1632 	error = zap_create_claim(os, moid, DMU_OT_MASTER_NODE,
1633 	    DMU_OT_NONE, 0, tx);
1634 	ASSERT(error == 0);
1635 
1636 	/*
1637 	 * Set starting attributes.
1638 	 */
1639 	version = zfs_zpl_version_map(spa_version(dmu_objset_spa(os)));
1640 	elem = NULL;
1641 	while ((elem = nvlist_next_nvpair(zplprops, elem)) != NULL) {
1642 		/* For the moment we expect all zpl props to be uint64_ts */
1643 		uint64_t val;
1644 		char *name;
1645 
1646 		ASSERT(nvpair_type(elem) == DATA_TYPE_UINT64);
1647 		VERIFY(nvpair_value_uint64(elem, &val) == 0);
1648 		name = nvpair_name(elem);
1649 		if (strcmp(name, zfs_prop_to_name(ZFS_PROP_VERSION)) == 0) {
1650 			if (val < version)
1651 				version = val;
1652 		} else {
1653 			error = zap_update(os, moid, name, 8, 1, &val, tx);
1654 		}
1655 		ASSERT(error == 0);
1656 		if (strcmp(name, zfs_prop_to_name(ZFS_PROP_NORMALIZE)) == 0)
1657 			norm = val;
1658 		else if (strcmp(name, zfs_prop_to_name(ZFS_PROP_CASE)) == 0)
1659 			sense = val;
1660 	}
1661 	ASSERT(version != 0);
1662 	error = zap_update(os, moid, ZPL_VERSION_STR, 8, 1, &version, tx);
1663 
1664 	/*
1665 	 * Create zap object used for SA attribute registration
1666 	 */
1667 
1668 	if (version >= ZPL_VERSION_SA) {
1669 		sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
1670 		    DMU_OT_NONE, 0, tx);
1671 		error = zap_add(os, moid, ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
1672 		ASSERT(error == 0);
1673 	} else {
1674 		sa_obj = 0;
1675 	}
1676 	/*
1677 	 * Create a delete queue.
1678 	 */
1679 	obj = zap_create(os, DMU_OT_UNLINKED_SET, DMU_OT_NONE, 0, tx);
1680 
1681 	error = zap_add(os, moid, ZFS_UNLINKED_SET, 8, 1, &obj, tx);
1682 	ASSERT(error == 0);
1683 
1684 	/*
1685 	 * Create root znode.  Create minimal znode/vnode/zfsvfs
1686 	 * to allow zfs_mknode to work.
1687 	 */
1688 	VATTR_NULL(&vattr);
1689 	vattr.va_mask = AT_MODE|AT_UID|AT_GID;
1690 	vattr.va_type = VDIR;
1691 	vattr.va_mode = S_IFDIR|0755;
1692 	vattr.va_uid = crgetuid(cr);
1693 	vattr.va_gid = crgetgid(cr);
1694 
1695 	zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
1696 
1697 	rootzp = zfs_znode_alloc_kmem(KM_SLEEP);
1698 	ASSERT(!POINTER_IS_VALID(rootzp->z_zfsvfs));
1699 	rootzp->z_unlinked = 0;
1700 	rootzp->z_atime_dirty = 0;
1701 	rootzp->z_is_sa = USE_SA(version, os);
1702 
1703 	zfsvfs->z_os = os;
1704 	zfsvfs->z_parent = zfsvfs;
1705 	zfsvfs->z_version = version;
1706 	zfsvfs->z_use_fuids = USE_FUIDS(version, os);
1707 	zfsvfs->z_use_sa = USE_SA(version, os);
1708 	zfsvfs->z_norm = norm;
1709 
1710 	error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
1711 	    &zfsvfs->z_attr_table);
1712 
1713 	ASSERT(error == 0);
1714 
1715 	/*
1716 	 * Fold case on file systems that are always or sometimes case
1717 	 * insensitive.
1718 	 */
1719 	if (sense == ZFS_CASE_INSENSITIVE || sense == ZFS_CASE_MIXED)
1720 		zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER;
1721 
1722 	mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
1723 	list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
1724 	    offsetof(znode_t, z_link_node));
1725 
1726 	for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1727 		mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL);
1728 
1729 	rootzp->z_zfsvfs = zfsvfs;
1730 	VERIFY(0 == zfs_acl_ids_create(rootzp, IS_ROOT_NODE, &vattr,
1731 	    cr, NULL, &acl_ids));
1732 	zfs_mknode(rootzp, &vattr, tx, cr, IS_ROOT_NODE, &zp, &acl_ids);
1733 	ASSERT3P(zp, ==, rootzp);
1734 	error = zap_add(os, moid, ZFS_ROOT_OBJ, 8, 1, &rootzp->z_id, tx);
1735 	ASSERT(error == 0);
1736 	zfs_acl_ids_free(&acl_ids);
1737 	POINTER_INVALIDATE(&rootzp->z_zfsvfs);
1738 
1739 	sa_handle_destroy(rootzp->z_sa_hdl);
1740 	zfs_znode_free_kmem(rootzp);
1741 
1742 	/*
1743 	 * Create shares directory
1744 	 */
1745 
1746 	error = zfs_create_share_dir(zfsvfs, tx);
1747 
1748 	ASSERT(error == 0);
1749 
1750 	for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1751 		mutex_destroy(&zfsvfs->z_hold_mtx[i]);
1752 	kmem_free(zfsvfs, sizeof (zfsvfs_t));
1753 }
1754 #endif /* _KERNEL */
1755 
1756 static int
zfs_sa_setup(objset_t * osp,sa_attr_type_t ** sa_table)1757 zfs_sa_setup(objset_t *osp, sa_attr_type_t **sa_table)
1758 {
1759 	uint64_t sa_obj = 0;
1760 	int error;
1761 
1762 	error = zap_lookup(osp, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1, &sa_obj);
1763 	if (error != 0 && error != ENOENT)
1764 		return (error);
1765 
1766 	error = sa_setup(osp, sa_obj, zfs_attr_table, ZPL_END, sa_table);
1767 	return (error);
1768 }
1769 
1770 static int
zfs_grab_sa_handle(objset_t * osp,uint64_t obj,sa_handle_t ** hdlp,dmu_buf_t ** db,void * tag)1771 zfs_grab_sa_handle(objset_t *osp, uint64_t obj, sa_handle_t **hdlp,
1772     dmu_buf_t **db, void *tag)
1773 {
1774 	dmu_object_info_t doi;
1775 	int error;
1776 
1777 	if ((error = sa_buf_hold(osp, obj, tag, db)) != 0)
1778 		return (error);
1779 
1780 	dmu_object_info_from_db(*db, &doi);
1781 	if ((doi.doi_bonus_type != DMU_OT_SA &&
1782 	    doi.doi_bonus_type != DMU_OT_ZNODE) ||
1783 	    (doi.doi_bonus_type == DMU_OT_ZNODE &&
1784 	    doi.doi_bonus_size < sizeof (znode_phys_t))) {
1785 		sa_buf_rele(*db, tag);
1786 		return (SET_ERROR(ENOTSUP));
1787 	}
1788 
1789 	error = sa_handle_get(osp, obj, NULL, SA_HDL_PRIVATE, hdlp);
1790 	if (error != 0) {
1791 		sa_buf_rele(*db, tag);
1792 		return (error);
1793 	}
1794 
1795 	return (0);
1796 }
1797 
1798 static void
zfs_release_sa_handle(sa_handle_t * hdl,dmu_buf_t * db,void * tag)1799 zfs_release_sa_handle(sa_handle_t *hdl, dmu_buf_t *db, void *tag)
1800 {
1801 	sa_handle_destroy(hdl);
1802 	sa_buf_rele(db, tag);
1803 }
1804 
1805 /*
1806  * Given an object number, return its parent object number and whether
1807  * or not the object is an extended attribute directory.
1808  */
1809 static int
zfs_obj_to_pobj(objset_t * osp,sa_handle_t * hdl,sa_attr_type_t * sa_table,uint64_t * pobjp,int * is_xattrdir)1810 zfs_obj_to_pobj(objset_t *osp, sa_handle_t *hdl, sa_attr_type_t *sa_table,
1811     uint64_t *pobjp, int *is_xattrdir)
1812 {
1813 	uint64_t parent;
1814 	uint64_t pflags;
1815 	uint64_t mode;
1816 	uint64_t parent_mode;
1817 	sa_bulk_attr_t bulk[3];
1818 	sa_handle_t *sa_hdl;
1819 	dmu_buf_t *sa_db;
1820 	int count = 0;
1821 	int error;
1822 
1823 	SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_PARENT], NULL,
1824 	    &parent, sizeof (parent));
1825 	SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_FLAGS], NULL,
1826 	    &pflags, sizeof (pflags));
1827 	SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL,
1828 	    &mode, sizeof (mode));
1829 
1830 	if ((error = sa_bulk_lookup(hdl, bulk, count)) != 0)
1831 		return (error);
1832 
1833 	/*
1834 	 * When a link is removed its parent pointer is not changed and will
1835 	 * be invalid.  There are two cases where a link is removed but the
1836 	 * file stays around, when it goes to the delete queue and when there
1837 	 * are additional links.
1838 	 */
1839 	error = zfs_grab_sa_handle(osp, parent, &sa_hdl, &sa_db, FTAG);
1840 	if (error != 0)
1841 		return (error);
1842 
1843 	error = sa_lookup(sa_hdl, ZPL_MODE, &parent_mode, sizeof (parent_mode));
1844 	zfs_release_sa_handle(sa_hdl, sa_db, FTAG);
1845 	if (error != 0)
1846 		return (error);
1847 
1848 	*is_xattrdir = ((pflags & ZFS_XATTR) != 0) && S_ISDIR(mode);
1849 
1850 	/*
1851 	 * Extended attributes can be applied to files, directories, etc.
1852 	 * Otherwise the parent must be a directory.
1853 	 */
1854 	if (!*is_xattrdir && !S_ISDIR(parent_mode))
1855 		return (SET_ERROR(EINVAL));
1856 
1857 	*pobjp = parent;
1858 
1859 	return (0);
1860 }
1861 
1862 /*
1863  * Given an object number, return some zpl level statistics
1864  */
1865 static int
zfs_obj_to_stats_impl(sa_handle_t * hdl,sa_attr_type_t * sa_table,zfs_stat_t * sb)1866 zfs_obj_to_stats_impl(sa_handle_t *hdl, sa_attr_type_t *sa_table,
1867     zfs_stat_t *sb)
1868 {
1869 	sa_bulk_attr_t bulk[4];
1870 	int count = 0;
1871 
1872 	SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL,
1873 	    &sb->zs_mode, sizeof (sb->zs_mode));
1874 	SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_GEN], NULL,
1875 	    &sb->zs_gen, sizeof (sb->zs_gen));
1876 	SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_LINKS], NULL,
1877 	    &sb->zs_links, sizeof (sb->zs_links));
1878 	SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_CTIME], NULL,
1879 	    &sb->zs_ctime, sizeof (sb->zs_ctime));
1880 
1881 	return (sa_bulk_lookup(hdl, bulk, count));
1882 }
1883 
1884 static int
zfs_obj_to_path_impl(objset_t * osp,uint64_t obj,sa_handle_t * hdl,sa_attr_type_t * sa_table,char * buf,int len)1885 zfs_obj_to_path_impl(objset_t *osp, uint64_t obj, sa_handle_t *hdl,
1886     sa_attr_type_t *sa_table, char *buf, int len)
1887 {
1888 	sa_handle_t *sa_hdl;
1889 	sa_handle_t *prevhdl = NULL;
1890 	dmu_buf_t *prevdb = NULL;
1891 	dmu_buf_t *sa_db = NULL;
1892 	char *path = buf + len - 1;
1893 	int error;
1894 
1895 	*path = '\0';
1896 	sa_hdl = hdl;
1897 
1898 	uint64_t deleteq_obj;
1899 	VERIFY0(zap_lookup(osp, MASTER_NODE_OBJ,
1900 	    ZFS_UNLINKED_SET, sizeof (uint64_t), 1, &deleteq_obj));
1901 	error = zap_lookup_int(osp, deleteq_obj, obj);
1902 	if (error == 0) {
1903 		return (ESTALE);
1904 	} else if (error != ENOENT) {
1905 		return (error);
1906 	}
1907 	error = 0;
1908 
1909 	for (;;) {
1910 		uint64_t pobj;
1911 		char component[MAXNAMELEN + 2];
1912 		size_t complen;
1913 		int is_xattrdir;
1914 
1915 		if (prevdb)
1916 			zfs_release_sa_handle(prevhdl, prevdb, FTAG);
1917 
1918 		if ((error = zfs_obj_to_pobj(osp, sa_hdl, sa_table, &pobj,
1919 		    &is_xattrdir)) != 0)
1920 			break;
1921 
1922 		if (pobj == obj) {
1923 			if (path[0] != '/')
1924 				*--path = '/';
1925 			break;
1926 		}
1927 
1928 		component[0] = '/';
1929 		if (is_xattrdir) {
1930 			(void) sprintf(component + 1, "<xattrdir>");
1931 		} else {
1932 			error = zap_value_search(osp, pobj, obj,
1933 			    ZFS_DIRENT_OBJ(-1ULL), component + 1);
1934 			if (error != 0)
1935 				break;
1936 		}
1937 
1938 		complen = strlen(component);
1939 		path -= complen;
1940 		ASSERT(path >= buf);
1941 		bcopy(component, path, complen);
1942 		obj = pobj;
1943 
1944 		if (sa_hdl != hdl) {
1945 			prevhdl = sa_hdl;
1946 			prevdb = sa_db;
1947 		}
1948 		error = zfs_grab_sa_handle(osp, obj, &sa_hdl, &sa_db, FTAG);
1949 		if (error != 0) {
1950 			sa_hdl = prevhdl;
1951 			sa_db = prevdb;
1952 			break;
1953 		}
1954 	}
1955 
1956 	if (sa_hdl != NULL && sa_hdl != hdl) {
1957 		ASSERT(sa_db != NULL);
1958 		zfs_release_sa_handle(sa_hdl, sa_db, FTAG);
1959 	}
1960 
1961 	if (error == 0)
1962 		(void) memmove(buf, path, buf + len - path);
1963 
1964 	return (error);
1965 }
1966 
1967 int
zfs_obj_to_path(objset_t * osp,uint64_t obj,char * buf,int len)1968 zfs_obj_to_path(objset_t *osp, uint64_t obj, char *buf, int len)
1969 {
1970 	sa_attr_type_t *sa_table;
1971 	sa_handle_t *hdl;
1972 	dmu_buf_t *db;
1973 	int error;
1974 
1975 	error = zfs_sa_setup(osp, &sa_table);
1976 	if (error != 0)
1977 		return (error);
1978 
1979 	error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG);
1980 	if (error != 0)
1981 		return (error);
1982 
1983 	error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len);
1984 
1985 	zfs_release_sa_handle(hdl, db, FTAG);
1986 	return (error);
1987 }
1988 
1989 int
zfs_obj_to_stats(objset_t * osp,uint64_t obj,zfs_stat_t * sb,char * buf,int len)1990 zfs_obj_to_stats(objset_t *osp, uint64_t obj, zfs_stat_t *sb,
1991     char *buf, int len)
1992 {
1993 	char *path = buf + len - 1;
1994 	sa_attr_type_t *sa_table;
1995 	sa_handle_t *hdl;
1996 	dmu_buf_t *db;
1997 	int error;
1998 
1999 	*path = '\0';
2000 
2001 	error = zfs_sa_setup(osp, &sa_table);
2002 	if (error != 0)
2003 		return (error);
2004 
2005 	error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG);
2006 	if (error != 0)
2007 		return (error);
2008 
2009 	error = zfs_obj_to_stats_impl(hdl, sa_table, sb);
2010 	if (error != 0) {
2011 		zfs_release_sa_handle(hdl, db, FTAG);
2012 		return (error);
2013 	}
2014 
2015 	error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len);
2016 
2017 	zfs_release_sa_handle(hdl, db, FTAG);
2018 	return (error);
2019 }
2020 
2021 
2022 void
zfs_inode_update(znode_t * zp)2023 zfs_inode_update(znode_t *zp)
2024 {
2025 	vm_object_t object;
2026 
2027 	if ((object = ZTOV(zp)->v_object) == NULL ||
2028 	    zp->z_size == object->un_pager.vnp.vnp_size)
2029 		return;
2030 
2031 	vnode_pager_setsize(ZTOV(zp), zp->z_size);
2032 }
2033 
2034 
2035 #ifdef _KERNEL
2036 int
zfs_znode_parent_and_name(znode_t * zp,znode_t ** dzpp,char * buf)2037 zfs_znode_parent_and_name(znode_t *zp, znode_t **dzpp, char *buf)
2038 {
2039 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2040 	uint64_t parent;
2041 	int is_xattrdir;
2042 	int err;
2043 
2044 	/* Extended attributes should not be visible as regular files. */
2045 	if ((zp->z_pflags & ZFS_XATTR) != 0)
2046 		return (SET_ERROR(EINVAL));
2047 
2048 	err = zfs_obj_to_pobj(zfsvfs->z_os, zp->z_sa_hdl, zfsvfs->z_attr_table,
2049 	    &parent, &is_xattrdir);
2050 	if (err != 0)
2051 		return (err);
2052 	ASSERT0(is_xattrdir);
2053 
2054 	/* No name as this is a root object. */
2055 	if (parent == zp->z_id)
2056 		return (SET_ERROR(EINVAL));
2057 
2058 	err = zap_value_search(zfsvfs->z_os, parent, zp->z_id,
2059 	    ZFS_DIRENT_OBJ(-1ULL), buf);
2060 	if (err != 0)
2061 		return (err);
2062 	err = zfs_zget(zfsvfs, parent, dzpp);
2063 	return (err);
2064 }
2065 #endif /* _KERNEL */
2066