xref: /freebsd-13-stable/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_znode.c (revision e6c1e181ba7f666e02b073be104eb3e241097d83)
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 <mm@FreeBSD.org> */
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_lock, NULL, MUTEX_DEFAULT, NULL);
147 	mutex_init(&zp->z_acl_lock, NULL, MUTEX_DEFAULT, NULL);
148 	rw_init(&zp->z_xattr_lock, NULL, RW_DEFAULT, NULL);
149 
150 	zfs_rangelock_init(&zp->z_rangelock, zfs_rangelock_cb, zp);
151 
152 	zp->z_acl_cached = NULL;
153 	zp->z_xattr_cached = NULL;
154 	zp->z_xattr_parent = 0;
155 	zp->z_vnode = NULL;
156 	zp->z_sync_writes_cnt = 0;
157 	zp->z_async_writes_cnt = 0;
158 
159 	return (0);
160 }
161 
162 /*ARGSUSED*/
163 static void
zfs_znode_cache_destructor(void * buf,void * arg)164 zfs_znode_cache_destructor(void *buf, void *arg)
165 {
166 	znode_t *zp = buf;
167 
168 	ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs));
169 	ASSERT3P(zp->z_vnode, ==, NULL);
170 	ASSERT(!list_link_active(&zp->z_link_node));
171 	mutex_destroy(&zp->z_lock);
172 	mutex_destroy(&zp->z_acl_lock);
173 	rw_destroy(&zp->z_xattr_lock);
174 	zfs_rangelock_fini(&zp->z_rangelock);
175 
176 	ASSERT3P(zp->z_acl_cached, ==, NULL);
177 	ASSERT3P(zp->z_xattr_cached, ==, NULL);
178 
179 	ASSERT0(atomic_load_32(&zp->z_sync_writes_cnt));
180 	ASSERT0(atomic_load_32(&zp->z_async_writes_cnt));
181 }
182 
183 
184 #ifdef _ZFS_USE_SMR
185 VFS_SMR_DECLARE;
186 
187 static int
zfs_znode_cache_constructor_smr(void * mem,int size __unused,void * private,int flags)188 zfs_znode_cache_constructor_smr(void *mem, int size __unused, void *private,
189     int flags)
190 {
191 
192 	return (zfs_znode_cache_constructor(mem, private, flags));
193 }
194 
195 static void
zfs_znode_cache_destructor_smr(void * mem,int size __unused,void * private)196 zfs_znode_cache_destructor_smr(void *mem, int size __unused, void *private)
197 {
198 
199 	zfs_znode_cache_destructor(mem, private);
200 }
201 
202 void
zfs_znode_init(void)203 zfs_znode_init(void)
204 {
205 	/*
206 	 * Initialize zcache
207 	 */
208 	ASSERT3P(znode_uma_zone, ==, NULL);
209 	znode_uma_zone = uma_zcreate("zfs_znode_cache",
210 	    sizeof (znode_t), zfs_znode_cache_constructor_smr,
211 	    zfs_znode_cache_destructor_smr, NULL, NULL, 0, 0);
212 	VFS_SMR_ZONE_SET(znode_uma_zone);
213 }
214 
215 static znode_t *
zfs_znode_alloc_kmem(int flags)216 zfs_znode_alloc_kmem(int flags)
217 {
218 
219 	return (uma_zalloc_smr(znode_uma_zone, flags));
220 }
221 
222 static void
zfs_znode_free_kmem(znode_t * zp)223 zfs_znode_free_kmem(znode_t *zp)
224 {
225 	if (zp->z_xattr_cached) {
226 		nvlist_free(zp->z_xattr_cached);
227 		zp->z_xattr_cached = NULL;
228 	}
229 	uma_zfree_smr(znode_uma_zone, zp);
230 }
231 #else
232 void
zfs_znode_init(void)233 zfs_znode_init(void)
234 {
235 	/*
236 	 * Initialize zcache
237 	 */
238 	ASSERT3P(znode_cache, ==, NULL);
239 	znode_cache = kmem_cache_create("zfs_znode_cache",
240 	    sizeof (znode_t), 0, zfs_znode_cache_constructor,
241 	    zfs_znode_cache_destructor, NULL, NULL, NULL, 0);
242 }
243 
244 static znode_t *
zfs_znode_alloc_kmem(int flags)245 zfs_znode_alloc_kmem(int flags)
246 {
247 
248 	return (kmem_cache_alloc(znode_cache, flags));
249 }
250 
251 static void
zfs_znode_free_kmem(znode_t * zp)252 zfs_znode_free_kmem(znode_t *zp)
253 {
254 	if (zp->z_xattr_cached) {
255 		nvlist_free(zp->z_xattr_cached);
256 		zp->z_xattr_cached = NULL;
257 	}
258 	kmem_cache_free(znode_cache, zp);
259 }
260 #endif
261 
262 void
zfs_znode_fini(void)263 zfs_znode_fini(void)
264 {
265 	/*
266 	 * Cleanup zcache
267 	 */
268 #ifdef _ZFS_USE_SMR
269 	if (znode_uma_zone) {
270 		uma_zdestroy(znode_uma_zone);
271 		znode_uma_zone = NULL;
272 	}
273 #else
274 	if (znode_cache) {
275 		kmem_cache_destroy(znode_cache);
276 		znode_cache = NULL;
277 	}
278 #endif
279 }
280 
281 
282 static int
zfs_create_share_dir(zfsvfs_t * zfsvfs,dmu_tx_t * tx)283 zfs_create_share_dir(zfsvfs_t *zfsvfs, dmu_tx_t *tx)
284 {
285 	zfs_acl_ids_t acl_ids;
286 	vattr_t vattr;
287 	znode_t *sharezp;
288 	znode_t *zp;
289 	int error;
290 
291 	vattr.va_mask = AT_MODE|AT_UID|AT_GID;
292 	vattr.va_type = VDIR;
293 	vattr.va_mode = S_IFDIR|0555;
294 	vattr.va_uid = crgetuid(kcred);
295 	vattr.va_gid = crgetgid(kcred);
296 
297 	sharezp = zfs_znode_alloc_kmem(KM_SLEEP);
298 	ASSERT(!POINTER_IS_VALID(sharezp->z_zfsvfs));
299 	sharezp->z_unlinked = 0;
300 	sharezp->z_atime_dirty = 0;
301 	sharezp->z_zfsvfs = zfsvfs;
302 	sharezp->z_is_sa = zfsvfs->z_use_sa;
303 
304 	VERIFY0(zfs_acl_ids_create(sharezp, IS_ROOT_NODE, &vattr,
305 	    kcred, NULL, &acl_ids));
306 	zfs_mknode(sharezp, &vattr, tx, kcred, IS_ROOT_NODE, &zp, &acl_ids);
307 	ASSERT3P(zp, ==, sharezp);
308 	POINTER_INVALIDATE(&sharezp->z_zfsvfs);
309 	error = zap_add(zfsvfs->z_os, MASTER_NODE_OBJ,
310 	    ZFS_SHARES_DIR, 8, 1, &sharezp->z_id, tx);
311 	zfsvfs->z_shares_dir = sharezp->z_id;
312 
313 	zfs_acl_ids_free(&acl_ids);
314 	sa_handle_destroy(sharezp->z_sa_hdl);
315 	zfs_znode_free_kmem(sharezp);
316 
317 	return (error);
318 }
319 
320 /*
321  * define a couple of values we need available
322  * for both 64 and 32 bit environments.
323  */
324 #ifndef NBITSMINOR64
325 #define	NBITSMINOR64	32
326 #endif
327 #ifndef MAXMAJ64
328 #define	MAXMAJ64	0xffffffffUL
329 #endif
330 #ifndef	MAXMIN64
331 #define	MAXMIN64	0xffffffffUL
332 #endif
333 
334 /*
335  * Create special expldev for ZFS private use.
336  * Can't use standard expldev since it doesn't do
337  * what we want.  The standard expldev() takes a
338  * dev32_t in LP64 and expands it to a long dev_t.
339  * We need an interface that takes a dev32_t in ILP32
340  * and expands it to a long dev_t.
341  */
342 static uint64_t
zfs_expldev(dev_t dev)343 zfs_expldev(dev_t dev)
344 {
345 	return (((uint64_t)major(dev) << NBITSMINOR64) | minor(dev));
346 }
347 /*
348  * Special cmpldev for ZFS private use.
349  * Can't use standard cmpldev since it takes
350  * a long dev_t and compresses it to dev32_t in
351  * LP64.  We need to do a compaction of a long dev_t
352  * to a dev32_t in ILP32.
353  */
354 dev_t
zfs_cmpldev(uint64_t dev)355 zfs_cmpldev(uint64_t dev)
356 {
357 	return (makedev((dev >> NBITSMINOR64), (dev & MAXMIN64)));
358 }
359 
360 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)361 zfs_znode_sa_init(zfsvfs_t *zfsvfs, znode_t *zp,
362     dmu_buf_t *db, dmu_object_type_t obj_type, sa_handle_t *sa_hdl)
363 {
364 	ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs) || (zfsvfs == zp->z_zfsvfs));
365 	ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zfsvfs, zp->z_id)));
366 
367 	ASSERT3P(zp->z_sa_hdl, ==, NULL);
368 	ASSERT3P(zp->z_acl_cached, ==, NULL);
369 	if (sa_hdl == NULL) {
370 		VERIFY0(sa_handle_get_from_db(zfsvfs->z_os, db, zp,
371 		    SA_HDL_SHARED, &zp->z_sa_hdl));
372 	} else {
373 		zp->z_sa_hdl = sa_hdl;
374 		sa_set_userp(sa_hdl, zp);
375 	}
376 
377 	zp->z_is_sa = (obj_type == DMU_OT_SA) ? B_TRUE : B_FALSE;
378 
379 	/*
380 	 * Slap on VROOT if we are the root znode unless we are the root
381 	 * node of a snapshot mounted under .zfs.
382 	 */
383 	if (zp->z_id == zfsvfs->z_root && zfsvfs->z_parent == zfsvfs)
384 		ZTOV(zp)->v_flag |= VROOT;
385 
386 	vn_exists(ZTOV(zp));
387 }
388 
389 void
zfs_znode_dmu_fini(znode_t * zp)390 zfs_znode_dmu_fini(znode_t *zp)
391 {
392 	ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zp->z_zfsvfs, zp->z_id)) ||
393 	    zp->z_unlinked ||
394 	    ZFS_TEARDOWN_INACTIVE_WRITE_HELD(zp->z_zfsvfs));
395 
396 	sa_handle_destroy(zp->z_sa_hdl);
397 	zp->z_sa_hdl = NULL;
398 }
399 
400 static void
zfs_vnode_forget(vnode_t * vp)401 zfs_vnode_forget(vnode_t *vp)
402 {
403 
404 	/* copied from insmntque_stddtr */
405 	vp->v_data = NULL;
406 	vp->v_op = &dead_vnodeops;
407 	vgone(vp);
408 	vput(vp);
409 }
410 
411 /*
412  * Construct a new znode/vnode and initialize.
413  *
414  * This does not do a call to dmu_set_user() that is
415  * up to the caller to do, in case you don't want to
416  * return the znode
417  */
418 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)419 zfs_znode_alloc(zfsvfs_t *zfsvfs, dmu_buf_t *db, int blksz,
420     dmu_object_type_t obj_type, sa_handle_t *hdl)
421 {
422 	znode_t	*zp;
423 	vnode_t *vp;
424 	uint64_t mode;
425 	uint64_t parent;
426 #ifdef notyet
427 	uint64_t mtime[2], ctime[2];
428 #endif
429 	uint64_t projid = ZFS_DEFAULT_PROJID;
430 	sa_bulk_attr_t bulk[9];
431 	int count = 0;
432 	int error;
433 
434 	zp = zfs_znode_alloc_kmem(KM_SLEEP);
435 
436 #ifndef _ZFS_USE_SMR
437 	KASSERT((zfsvfs->z_parent->z_vfs->mnt_kern_flag & MNTK_FPLOOKUP) == 0,
438 	    ("%s: fast path lookup enabled without smr", __func__));
439 #endif
440 
441 #if __FreeBSD_version >= 1300076
442 	KASSERT(curthread->td_vp_reserved != NULL,
443 	    ("zfs_znode_alloc: getnewvnode without any vnodes reserved"));
444 #else
445 	KASSERT(curthread->td_vp_reserv > 0,
446 	    ("zfs_znode_alloc: getnewvnode without any vnodes reserved"));
447 #endif
448 	error = getnewvnode("zfs", zfsvfs->z_parent->z_vfs, &zfs_vnodeops, &vp);
449 	if (error != 0) {
450 		zfs_znode_free_kmem(zp);
451 		return (NULL);
452 	}
453 	zp->z_vnode = vp;
454 	vp->v_data = zp;
455 
456 	ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs));
457 
458 	zp->z_sa_hdl = NULL;
459 	zp->z_unlinked = 0;
460 	zp->z_atime_dirty = 0;
461 	zp->z_mapcnt = 0;
462 	zp->z_id = db->db_object;
463 	zp->z_blksz = blksz;
464 	zp->z_seq = 0x7A4653;
465 	zp->z_sync_cnt = 0;
466 	zp->z_sync_writes_cnt = 0;
467 	zp->z_async_writes_cnt = 0;
468 #if __FreeBSD_version >= 1300139
469 	atomic_store_ptr(&zp->z_cached_symlink, NULL);
470 #endif
471 
472 	vp = ZTOV(zp);
473 
474 	zfs_znode_sa_init(zfsvfs, zp, db, obj_type, hdl);
475 
476 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8);
477 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL, &zp->z_gen, 8);
478 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
479 	    &zp->z_size, 8);
480 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL,
481 	    &zp->z_links, 8);
482 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
483 	    &zp->z_pflags, 8);
484 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL, &parent, 8);
485 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
486 	    &zp->z_atime, 16);
487 #ifdef notyet
488 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
489 	    &mtime, 16);
490 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
491 	    &ctime, 16);
492 #endif
493 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
494 	    &zp->z_uid, 8);
495 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL,
496 	    &zp->z_gid, 8);
497 
498 	if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count) != 0 || zp->z_gen == 0 ||
499 	    (dmu_objset_projectquota_enabled(zfsvfs->z_os) &&
500 	    (zp->z_pflags & ZFS_PROJID) &&
501 	    sa_lookup(zp->z_sa_hdl, SA_ZPL_PROJID(zfsvfs), &projid, 8) != 0)) {
502 		if (hdl == NULL)
503 			sa_handle_destroy(zp->z_sa_hdl);
504 		zfs_vnode_forget(vp);
505 		zp->z_vnode = NULL;
506 		zfs_znode_free_kmem(zp);
507 		return (NULL);
508 	}
509 
510 	zp->z_projid = projid;
511 	zp->z_mode = mode;
512 
513 	/* Cache the xattr parent id */
514 	if (zp->z_pflags & ZFS_XATTR)
515 		zp->z_xattr_parent = parent;
516 
517 	vp->v_type = IFTOVT((mode_t)mode);
518 
519 	switch (vp->v_type) {
520 	case VDIR:
521 		zp->z_zn_prefetch = B_TRUE; /* z_prefetch default is enabled */
522 		break;
523 	case VFIFO:
524 		vp->v_op = &zfs_fifoops;
525 		break;
526 	case VREG:
527 		if (parent == zfsvfs->z_shares_dir) {
528 			ASSERT0(zp->z_uid);
529 			ASSERT0(zp->z_gid);
530 			vp->v_op = &zfs_shareops;
531 		}
532 		break;
533 	default:
534 			break;
535 	}
536 
537 	mutex_enter(&zfsvfs->z_znodes_lock);
538 	list_insert_tail(&zfsvfs->z_all_znodes, zp);
539 	zfsvfs->z_nr_znodes++;
540 	zp->z_zfsvfs = zfsvfs;
541 	mutex_exit(&zfsvfs->z_znodes_lock);
542 
543 	/*
544 	 * Acquire vnode lock before making it available to the world.
545 	 */
546 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
547 #if __FreeBSD_version >= 1400077
548 	vn_set_state(vp, VSTATE_CONSTRUCTED);
549 #endif
550 	VN_LOCK_AREC(vp);
551 	if (vp->v_type != VFIFO)
552 		VN_LOCK_ASHARE(vp);
553 
554 	return (zp);
555 }
556 
557 static uint64_t empty_xattr;
558 static uint64_t pad[4];
559 static zfs_acl_phys_t acl_phys;
560 /*
561  * Create a new DMU object to hold a zfs znode.
562  *
563  *	IN:	dzp	- parent directory for new znode
564  *		vap	- file attributes for new znode
565  *		tx	- dmu transaction id for zap operations
566  *		cr	- credentials of caller
567  *		flag	- flags:
568  *			  IS_ROOT_NODE	- new object will be root
569  *			  IS_XATTR	- new object is an attribute
570  *		bonuslen - length of bonus buffer
571  *		setaclp  - File/Dir initial ACL
572  *		fuidp	 - Tracks fuid allocation.
573  *
574  *	OUT:	zpp	- allocated znode
575  *
576  */
577 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)578 zfs_mknode(znode_t *dzp, vattr_t *vap, dmu_tx_t *tx, cred_t *cr,
579     uint_t flag, znode_t **zpp, zfs_acl_ids_t *acl_ids)
580 {
581 	uint64_t	crtime[2], atime[2], mtime[2], ctime[2];
582 	uint64_t	mode, size, links, parent, pflags;
583 	uint64_t	dzp_pflags = 0;
584 	uint64_t	rdev = 0;
585 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
586 	dmu_buf_t	*db;
587 	timestruc_t	now;
588 	uint64_t	gen, obj;
589 	int		bonuslen;
590 	int		dnodesize;
591 	sa_handle_t	*sa_hdl;
592 	dmu_object_type_t obj_type;
593 	sa_bulk_attr_t	*sa_attrs;
594 	int		cnt = 0;
595 	zfs_acl_locator_cb_t locate = { 0 };
596 
597 	ASSERT3P(vap, !=, NULL);
598 	ASSERT3U((vap->va_mask & AT_MODE), ==, AT_MODE);
599 
600 	if (zfsvfs->z_replay) {
601 		obj = vap->va_nodeid;
602 		now = vap->va_ctime;		/* see zfs_replay_create() */
603 		gen = vap->va_nblocks;		/* ditto */
604 		dnodesize = vap->va_fsid;	/* ditto */
605 	} else {
606 		obj = 0;
607 		vfs_timestamp(&now);
608 		gen = dmu_tx_get_txg(tx);
609 		dnodesize = dmu_objset_dnodesize(zfsvfs->z_os);
610 	}
611 
612 	if (dnodesize == 0)
613 		dnodesize = DNODE_MIN_SIZE;
614 
615 	obj_type = zfsvfs->z_use_sa ? DMU_OT_SA : DMU_OT_ZNODE;
616 	bonuslen = (obj_type == DMU_OT_SA) ?
617 	    DN_BONUS_SIZE(dnodesize) : ZFS_OLD_ZNODE_PHYS_SIZE;
618 
619 	/*
620 	 * Create a new DMU object.
621 	 */
622 	/*
623 	 * There's currently no mechanism for pre-reading the blocks that will
624 	 * be needed to allocate a new object, so we accept the small chance
625 	 * that there will be an i/o error and we will fail one of the
626 	 * assertions below.
627 	 */
628 	if (vap->va_type == VDIR) {
629 		if (zfsvfs->z_replay) {
630 			VERIFY0(zap_create_claim_norm_dnsize(zfsvfs->z_os, obj,
631 			    zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS,
632 			    obj_type, bonuslen, dnodesize, tx));
633 		} else {
634 			obj = zap_create_norm_dnsize(zfsvfs->z_os,
635 			    zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS,
636 			    obj_type, bonuslen, dnodesize, tx);
637 		}
638 	} else {
639 		if (zfsvfs->z_replay) {
640 			VERIFY0(dmu_object_claim_dnsize(zfsvfs->z_os, obj,
641 			    DMU_OT_PLAIN_FILE_CONTENTS, 0,
642 			    obj_type, bonuslen, dnodesize, tx));
643 		} else {
644 			obj = dmu_object_alloc_dnsize(zfsvfs->z_os,
645 			    DMU_OT_PLAIN_FILE_CONTENTS, 0,
646 			    obj_type, bonuslen, dnodesize, tx);
647 		}
648 	}
649 
650 	ZFS_OBJ_HOLD_ENTER(zfsvfs, obj);
651 	VERIFY0(sa_buf_hold(zfsvfs->z_os, obj, NULL, &db));
652 
653 	/*
654 	 * If this is the root, fix up the half-initialized parent pointer
655 	 * to reference the just-allocated physical data area.
656 	 */
657 	if (flag & IS_ROOT_NODE) {
658 		dzp->z_id = obj;
659 	} else {
660 		dzp_pflags = dzp->z_pflags;
661 	}
662 
663 	/*
664 	 * If parent is an xattr, so am I.
665 	 */
666 	if (dzp_pflags & ZFS_XATTR) {
667 		flag |= IS_XATTR;
668 	}
669 
670 	if (zfsvfs->z_use_fuids)
671 		pflags = ZFS_ARCHIVE | ZFS_AV_MODIFIED;
672 	else
673 		pflags = 0;
674 
675 	if (vap->va_type == VDIR) {
676 		size = 2;		/* contents ("." and "..") */
677 		links = (flag & (IS_ROOT_NODE | IS_XATTR)) ? 2 : 1;
678 	} else {
679 		size = links = 0;
680 	}
681 
682 	if (vap->va_type == VBLK || vap->va_type == VCHR) {
683 		rdev = zfs_expldev(vap->va_rdev);
684 	}
685 
686 	parent = dzp->z_id;
687 	mode = acl_ids->z_mode;
688 	if (flag & IS_XATTR)
689 		pflags |= ZFS_XATTR;
690 
691 	/*
692 	 * No execs denied will be determined when zfs_mode_compute() is called.
693 	 */
694 	pflags |= acl_ids->z_aclp->z_hints &
695 	    (ZFS_ACL_TRIVIAL|ZFS_INHERIT_ACE|ZFS_ACL_AUTO_INHERIT|
696 	    ZFS_ACL_DEFAULTED|ZFS_ACL_PROTECTED);
697 
698 	ZFS_TIME_ENCODE(&now, crtime);
699 	ZFS_TIME_ENCODE(&now, ctime);
700 
701 	if (vap->va_mask & AT_ATIME) {
702 		ZFS_TIME_ENCODE(&vap->va_atime, atime);
703 	} else {
704 		ZFS_TIME_ENCODE(&now, atime);
705 	}
706 
707 	if (vap->va_mask & AT_MTIME) {
708 		ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
709 	} else {
710 		ZFS_TIME_ENCODE(&now, mtime);
711 	}
712 
713 	/* Now add in all of the "SA" attributes */
714 	VERIFY0(sa_handle_get_from_db(zfsvfs->z_os, db, NULL, SA_HDL_SHARED,
715 	    &sa_hdl));
716 
717 	/*
718 	 * Setup the array of attributes to be replaced/set on the new file
719 	 *
720 	 * order for  DMU_OT_ZNODE is critical since it needs to be constructed
721 	 * in the old znode_phys_t format.  Don't change this ordering
722 	 */
723 	sa_attrs = kmem_alloc(sizeof (sa_bulk_attr_t) * ZPL_END, KM_SLEEP);
724 
725 	if (obj_type == DMU_OT_ZNODE) {
726 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs),
727 		    NULL, &atime, 16);
728 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs),
729 		    NULL, &mtime, 16);
730 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs),
731 		    NULL, &ctime, 16);
732 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs),
733 		    NULL, &crtime, 16);
734 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs),
735 		    NULL, &gen, 8);
736 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs),
737 		    NULL, &mode, 8);
738 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs),
739 		    NULL, &size, 8);
740 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs),
741 		    NULL, &parent, 8);
742 	} else {
743 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs),
744 		    NULL, &mode, 8);
745 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs),
746 		    NULL, &size, 8);
747 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs),
748 		    NULL, &gen, 8);
749 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs),
750 		    NULL, &acl_ids->z_fuid, 8);
751 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs),
752 		    NULL, &acl_ids->z_fgid, 8);
753 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs),
754 		    NULL, &parent, 8);
755 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs),
756 		    NULL, &pflags, 8);
757 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs),
758 		    NULL, &atime, 16);
759 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs),
760 		    NULL, &mtime, 16);
761 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs),
762 		    NULL, &ctime, 16);
763 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs),
764 		    NULL, &crtime, 16);
765 	}
766 
767 	SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_LINKS(zfsvfs), NULL, &links, 8);
768 
769 	if (obj_type == DMU_OT_ZNODE) {
770 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_XATTR(zfsvfs), NULL,
771 		    &empty_xattr, 8);
772 	}
773 	if (obj_type == DMU_OT_ZNODE ||
774 	    (vap->va_type == VBLK || vap->va_type == VCHR)) {
775 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_RDEV(zfsvfs),
776 		    NULL, &rdev, 8);
777 
778 	}
779 	if (obj_type == DMU_OT_ZNODE) {
780 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs),
781 		    NULL, &pflags, 8);
782 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs), NULL,
783 		    &acl_ids->z_fuid, 8);
784 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs), NULL,
785 		    &acl_ids->z_fgid, 8);
786 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PAD(zfsvfs), NULL, pad,
787 		    sizeof (uint64_t) * 4);
788 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ZNODE_ACL(zfsvfs), NULL,
789 		    &acl_phys, sizeof (zfs_acl_phys_t));
790 	} else if (acl_ids->z_aclp->z_version >= ZFS_ACL_VERSION_FUID) {
791 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_COUNT(zfsvfs), NULL,
792 		    &acl_ids->z_aclp->z_acl_count, 8);
793 		locate.cb_aclp = acl_ids->z_aclp;
794 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_ACES(zfsvfs),
795 		    zfs_acl_data_locator, &locate,
796 		    acl_ids->z_aclp->z_acl_bytes);
797 		mode = zfs_mode_compute(mode, acl_ids->z_aclp, &pflags,
798 		    acl_ids->z_fuid, acl_ids->z_fgid);
799 	}
800 
801 	VERIFY0(sa_replace_all_by_template(sa_hdl, sa_attrs, cnt, tx));
802 
803 	if (!(flag & IS_ROOT_NODE)) {
804 		*zpp = zfs_znode_alloc(zfsvfs, db, 0, obj_type, sa_hdl);
805 		ASSERT3P(*zpp, !=, NULL);
806 	} else {
807 		/*
808 		 * If we are creating the root node, the "parent" we
809 		 * passed in is the znode for the root.
810 		 */
811 		*zpp = dzp;
812 
813 		(*zpp)->z_sa_hdl = sa_hdl;
814 	}
815 
816 	(*zpp)->z_pflags = pflags;
817 	(*zpp)->z_mode = mode;
818 	(*zpp)->z_dnodesize = dnodesize;
819 
820 	if (vap->va_mask & AT_XVATTR)
821 		zfs_xvattr_set(*zpp, (xvattr_t *)vap, tx);
822 
823 	if (obj_type == DMU_OT_ZNODE ||
824 	    acl_ids->z_aclp->z_version < ZFS_ACL_VERSION_FUID) {
825 		VERIFY0(zfs_aclset_common(*zpp, acl_ids->z_aclp, cr, tx));
826 	}
827 	if (!(flag & IS_ROOT_NODE)) {
828 		vnode_t *vp = ZTOV(*zpp);
829 		vp->v_vflag |= VV_FORCEINSMQ;
830 		int err = insmntque(vp, zfsvfs->z_vfs);
831 		vp->v_vflag &= ~VV_FORCEINSMQ;
832 		(void) err;
833 		KASSERT(err == 0, ("insmntque() failed: error %d", err));
834 	}
835 	kmem_free(sa_attrs, sizeof (sa_bulk_attr_t) * ZPL_END);
836 	ZFS_OBJ_HOLD_EXIT(zfsvfs, obj);
837 }
838 
839 /*
840  * Update in-core attributes.  It is assumed the caller will be doing an
841  * sa_bulk_update to push the changes out.
842  */
843 void
zfs_xvattr_set(znode_t * zp,xvattr_t * xvap,dmu_tx_t * tx)844 zfs_xvattr_set(znode_t *zp, xvattr_t *xvap, dmu_tx_t *tx)
845 {
846 	xoptattr_t *xoap;
847 
848 	xoap = xva_getxoptattr(xvap);
849 	ASSERT3P(xoap, !=, NULL);
850 
851 	if (zp->z_zfsvfs->z_replay == B_FALSE) {
852 		ASSERT_VOP_IN_SEQC(ZTOV(zp));
853 	}
854 
855 	if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
856 		uint64_t times[2];
857 		ZFS_TIME_ENCODE(&xoap->xoa_createtime, times);
858 		(void) sa_update(zp->z_sa_hdl, SA_ZPL_CRTIME(zp->z_zfsvfs),
859 		    &times, sizeof (times), tx);
860 		XVA_SET_RTN(xvap, XAT_CREATETIME);
861 	}
862 	if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
863 		ZFS_ATTR_SET(zp, ZFS_READONLY, xoap->xoa_readonly,
864 		    zp->z_pflags, tx);
865 		XVA_SET_RTN(xvap, XAT_READONLY);
866 	}
867 	if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
868 		ZFS_ATTR_SET(zp, ZFS_HIDDEN, xoap->xoa_hidden,
869 		    zp->z_pflags, tx);
870 		XVA_SET_RTN(xvap, XAT_HIDDEN);
871 	}
872 	if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
873 		ZFS_ATTR_SET(zp, ZFS_SYSTEM, xoap->xoa_system,
874 		    zp->z_pflags, tx);
875 		XVA_SET_RTN(xvap, XAT_SYSTEM);
876 	}
877 	if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
878 		ZFS_ATTR_SET(zp, ZFS_ARCHIVE, xoap->xoa_archive,
879 		    zp->z_pflags, tx);
880 		XVA_SET_RTN(xvap, XAT_ARCHIVE);
881 	}
882 	if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
883 		ZFS_ATTR_SET(zp, ZFS_IMMUTABLE, xoap->xoa_immutable,
884 		    zp->z_pflags, tx);
885 		XVA_SET_RTN(xvap, XAT_IMMUTABLE);
886 	}
887 	if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
888 		ZFS_ATTR_SET(zp, ZFS_NOUNLINK, xoap->xoa_nounlink,
889 		    zp->z_pflags, tx);
890 		XVA_SET_RTN(xvap, XAT_NOUNLINK);
891 	}
892 	if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
893 		ZFS_ATTR_SET(zp, ZFS_APPENDONLY, xoap->xoa_appendonly,
894 		    zp->z_pflags, tx);
895 		XVA_SET_RTN(xvap, XAT_APPENDONLY);
896 	}
897 	if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
898 		ZFS_ATTR_SET(zp, ZFS_NODUMP, xoap->xoa_nodump,
899 		    zp->z_pflags, tx);
900 		XVA_SET_RTN(xvap, XAT_NODUMP);
901 	}
902 	if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
903 		ZFS_ATTR_SET(zp, ZFS_OPAQUE, xoap->xoa_opaque,
904 		    zp->z_pflags, tx);
905 		XVA_SET_RTN(xvap, XAT_OPAQUE);
906 	}
907 	if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
908 		ZFS_ATTR_SET(zp, ZFS_AV_QUARANTINED,
909 		    xoap->xoa_av_quarantined, zp->z_pflags, tx);
910 		XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
911 	}
912 	if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
913 		ZFS_ATTR_SET(zp, ZFS_AV_MODIFIED, xoap->xoa_av_modified,
914 		    zp->z_pflags, tx);
915 		XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
916 	}
917 	if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) {
918 		zfs_sa_set_scanstamp(zp, xvap, tx);
919 		XVA_SET_RTN(xvap, XAT_AV_SCANSTAMP);
920 	}
921 	if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
922 		ZFS_ATTR_SET(zp, ZFS_REPARSE, xoap->xoa_reparse,
923 		    zp->z_pflags, tx);
924 		XVA_SET_RTN(xvap, XAT_REPARSE);
925 	}
926 	if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
927 		ZFS_ATTR_SET(zp, ZFS_OFFLINE, xoap->xoa_offline,
928 		    zp->z_pflags, tx);
929 		XVA_SET_RTN(xvap, XAT_OFFLINE);
930 	}
931 	if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
932 		ZFS_ATTR_SET(zp, ZFS_SPARSE, xoap->xoa_sparse,
933 		    zp->z_pflags, tx);
934 		XVA_SET_RTN(xvap, XAT_SPARSE);
935 	}
936 }
937 
938 int
zfs_zget(zfsvfs_t * zfsvfs,uint64_t obj_num,znode_t ** zpp)939 zfs_zget(zfsvfs_t *zfsvfs, uint64_t obj_num, znode_t **zpp)
940 {
941 	dmu_object_info_t doi;
942 	dmu_buf_t	*db;
943 	znode_t		*zp;
944 	vnode_t		*vp;
945 	sa_handle_t	*hdl;
946 	int locked;
947 	int err;
948 
949 	getnewvnode_reserve_();
950 again:
951 	*zpp = NULL;
952 	ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num);
953 
954 	err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db);
955 	if (err) {
956 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
957 		getnewvnode_drop_reserve();
958 		return (err);
959 	}
960 
961 	dmu_object_info_from_db(db, &doi);
962 	if (doi.doi_bonus_type != DMU_OT_SA &&
963 	    (doi.doi_bonus_type != DMU_OT_ZNODE ||
964 	    (doi.doi_bonus_type == DMU_OT_ZNODE &&
965 	    doi.doi_bonus_size < sizeof (znode_phys_t)))) {
966 		sa_buf_rele(db, NULL);
967 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
968 		getnewvnode_drop_reserve();
969 		return (SET_ERROR(EINVAL));
970 	}
971 
972 	hdl = dmu_buf_get_user(db);
973 	if (hdl != NULL) {
974 		zp = sa_get_userdata(hdl);
975 
976 		/*
977 		 * Since "SA" does immediate eviction we
978 		 * should never find a sa handle that doesn't
979 		 * know about the znode.
980 		 */
981 		ASSERT3P(zp, !=, NULL);
982 		ASSERT3U(zp->z_id, ==, obj_num);
983 		if (zp->z_unlinked) {
984 			err = SET_ERROR(ENOENT);
985 		} else {
986 			vp = ZTOV(zp);
987 			/*
988 			 * Don't let the vnode disappear after
989 			 * ZFS_OBJ_HOLD_EXIT.
990 			 */
991 			VN_HOLD(vp);
992 			*zpp = zp;
993 			err = 0;
994 		}
995 
996 		sa_buf_rele(db, NULL);
997 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
998 
999 		if (err) {
1000 			getnewvnode_drop_reserve();
1001 			return (err);
1002 		}
1003 
1004 		locked = VOP_ISLOCKED(vp);
1005 		VI_LOCK(vp);
1006 		if (VN_IS_DOOMED(vp) && locked != LK_EXCLUSIVE) {
1007 			/*
1008 			 * The vnode is doomed and this thread doesn't
1009 			 * hold the exclusive lock on it, so the vnode
1010 			 * must be being reclaimed by another thread.
1011 			 * Otherwise the doomed vnode is being reclaimed
1012 			 * by this thread and zfs_zget is called from
1013 			 * ZIL internals.
1014 			 */
1015 			VI_UNLOCK(vp);
1016 
1017 			/*
1018 			 * XXX vrele() locks the vnode when the last reference
1019 			 * is dropped.  Although in this case the vnode is
1020 			 * doomed / dead and so no inactivation is required,
1021 			 * the vnode lock is still acquired.  That could result
1022 			 * in a LOR with z_teardown_lock if another thread holds
1023 			 * the vnode's lock and tries to take z_teardown_lock.
1024 			 * But that is only possible if the other thread peforms
1025 			 * a ZFS vnode operation on the vnode.  That either
1026 			 * should not happen if the vnode is dead or the thread
1027 			 * should also have a reference to the vnode and thus
1028 			 * our reference is not last.
1029 			 */
1030 			VN_RELE(vp);
1031 			goto again;
1032 		}
1033 		VI_UNLOCK(vp);
1034 		getnewvnode_drop_reserve();
1035 		return (err);
1036 	}
1037 
1038 	/*
1039 	 * Not found create new znode/vnode
1040 	 * but only if file exists.
1041 	 *
1042 	 * There is a small window where zfs_vget() could
1043 	 * find this object while a file create is still in
1044 	 * progress.  This is checked for in zfs_znode_alloc()
1045 	 *
1046 	 * if zfs_znode_alloc() fails it will drop the hold on the
1047 	 * bonus buffer.
1048 	 */
1049 	zp = zfs_znode_alloc(zfsvfs, db, doi.doi_data_block_size,
1050 	    doi.doi_bonus_type, NULL);
1051 	if (zp == NULL) {
1052 		err = SET_ERROR(ENOENT);
1053 	} else {
1054 		*zpp = zp;
1055 	}
1056 	if (err == 0) {
1057 		vnode_t *vp = ZTOV(zp);
1058 
1059 		err = insmntque(vp, zfsvfs->z_vfs);
1060 		if (err == 0) {
1061 			vp->v_hash = obj_num;
1062 			VOP_UNLOCK1(vp);
1063 		} else {
1064 			zp->z_vnode = NULL;
1065 			zfs_znode_dmu_fini(zp);
1066 			zfs_znode_free(zp);
1067 			*zpp = NULL;
1068 		}
1069 	}
1070 	ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1071 	getnewvnode_drop_reserve();
1072 	return (err);
1073 }
1074 
1075 int
zfs_rezget(znode_t * zp)1076 zfs_rezget(znode_t *zp)
1077 {
1078 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1079 	dmu_object_info_t doi;
1080 	dmu_buf_t *db;
1081 	vnode_t *vp;
1082 	uint64_t obj_num = zp->z_id;
1083 	uint64_t mode, size;
1084 	sa_bulk_attr_t bulk[8];
1085 	int err;
1086 	int count = 0;
1087 	uint64_t gen;
1088 
1089 	/*
1090 	 * Remove cached pages before reloading the znode, so that they are not
1091 	 * lingering after we run into any error.  Ideally, we should vgone()
1092 	 * the vnode in case of error, but currently we cannot do that
1093 	 * because of the LOR between the vnode lock and z_teardown_lock.
1094 	 * So, instead, we have to "doom" the znode in the illumos style.
1095 	 *
1096 	 * Ignore invalid pages during the scan.  This is to avoid deadlocks
1097 	 * between page busying and the teardown lock, as pages are busied prior
1098 	 * to a VOP_GETPAGES operation, which acquires the teardown read lock.
1099 	 * Such pages will be invalid and can safely be skipped here.
1100 	 */
1101 	vp = ZTOV(zp);
1102 #if __FreeBSD_version >= 1300522
1103 	vn_pages_remove_valid(vp, 0, 0);
1104 #else
1105 	vn_pages_remove(vp, 0, 0);
1106 #endif
1107 
1108 	ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num);
1109 
1110 	mutex_enter(&zp->z_acl_lock);
1111 	if (zp->z_acl_cached) {
1112 		zfs_acl_free(zp->z_acl_cached);
1113 		zp->z_acl_cached = NULL;
1114 	}
1115 	mutex_exit(&zp->z_acl_lock);
1116 
1117 	rw_enter(&zp->z_xattr_lock, RW_WRITER);
1118 	if (zp->z_xattr_cached) {
1119 		nvlist_free(zp->z_xattr_cached);
1120 		zp->z_xattr_cached = NULL;
1121 	}
1122 	rw_exit(&zp->z_xattr_lock);
1123 
1124 	ASSERT3P(zp->z_sa_hdl, ==, NULL);
1125 	err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db);
1126 	if (err) {
1127 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1128 		return (err);
1129 	}
1130 
1131 	dmu_object_info_from_db(db, &doi);
1132 	if (doi.doi_bonus_type != DMU_OT_SA &&
1133 	    (doi.doi_bonus_type != DMU_OT_ZNODE ||
1134 	    (doi.doi_bonus_type == DMU_OT_ZNODE &&
1135 	    doi.doi_bonus_size < sizeof (znode_phys_t)))) {
1136 		sa_buf_rele(db, NULL);
1137 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1138 		return (SET_ERROR(EINVAL));
1139 	}
1140 
1141 	zfs_znode_sa_init(zfsvfs, zp, db, doi.doi_bonus_type, NULL);
1142 	size = zp->z_size;
1143 
1144 	/* reload cached values */
1145 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL,
1146 	    &gen, sizeof (gen));
1147 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
1148 	    &zp->z_size, sizeof (zp->z_size));
1149 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL,
1150 	    &zp->z_links, sizeof (zp->z_links));
1151 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
1152 	    &zp->z_pflags, sizeof (zp->z_pflags));
1153 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
1154 	    &zp->z_atime, sizeof (zp->z_atime));
1155 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
1156 	    &zp->z_uid, sizeof (zp->z_uid));
1157 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL,
1158 	    &zp->z_gid, sizeof (zp->z_gid));
1159 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
1160 	    &mode, sizeof (mode));
1161 
1162 	if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) {
1163 		zfs_znode_dmu_fini(zp);
1164 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1165 		return (SET_ERROR(EIO));
1166 	}
1167 
1168 	zp->z_mode = mode;
1169 
1170 	if (gen != zp->z_gen) {
1171 		zfs_znode_dmu_fini(zp);
1172 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1173 		return (SET_ERROR(EIO));
1174 	}
1175 
1176 	/*
1177 	 * It is highly improbable but still quite possible that two
1178 	 * objects in different datasets are created with the same
1179 	 * object numbers and in transaction groups with the same
1180 	 * numbers.  znodes corresponding to those objects would
1181 	 * have the same z_id and z_gen, but their other attributes
1182 	 * may be different.
1183 	 * zfs recv -F may replace one of such objects with the other.
1184 	 * As a result file properties recorded in the replaced
1185 	 * object's vnode may no longer match the received object's
1186 	 * properties.  At present the only cached property is the
1187 	 * files type recorded in v_type.
1188 	 * So, handle this case by leaving the old vnode and znode
1189 	 * disassociated from the actual object.  A new vnode and a
1190 	 * znode will be created if the object is accessed
1191 	 * (e.g. via a look-up).  The old vnode and znode will be
1192 	 * recycled when the last vnode reference is dropped.
1193 	 */
1194 	if (vp->v_type != IFTOVT((mode_t)zp->z_mode)) {
1195 		zfs_znode_dmu_fini(zp);
1196 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1197 		return (SET_ERROR(EIO));
1198 	}
1199 
1200 	/*
1201 	 * If the file has zero links, then it has been unlinked on the send
1202 	 * side and it must be in the received unlinked set.
1203 	 * We call zfs_znode_dmu_fini() now to prevent any accesses to the
1204 	 * stale data and to prevent automatically removal of the file in
1205 	 * zfs_zinactive().  The file will be removed either when it is removed
1206 	 * on the send side and the next incremental stream is received or
1207 	 * when the unlinked set gets processed.
1208 	 */
1209 	zp->z_unlinked = (zp->z_links == 0);
1210 	if (zp->z_unlinked) {
1211 		zfs_znode_dmu_fini(zp);
1212 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1213 		return (0);
1214 	}
1215 
1216 	zp->z_blksz = doi.doi_data_block_size;
1217 	if (zp->z_size != size)
1218 		vnode_pager_setsize(vp, zp->z_size);
1219 
1220 	ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1221 
1222 	return (0);
1223 }
1224 
1225 void
zfs_znode_delete(znode_t * zp,dmu_tx_t * tx)1226 zfs_znode_delete(znode_t *zp, dmu_tx_t *tx)
1227 {
1228 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1229 	objset_t *os = zfsvfs->z_os;
1230 	uint64_t obj = zp->z_id;
1231 	uint64_t acl_obj = zfs_external_acl(zp);
1232 
1233 	ZFS_OBJ_HOLD_ENTER(zfsvfs, obj);
1234 	if (acl_obj) {
1235 		VERIFY(!zp->z_is_sa);
1236 		VERIFY0(dmu_object_free(os, acl_obj, tx));
1237 	}
1238 	VERIFY0(dmu_object_free(os, obj, tx));
1239 	zfs_znode_dmu_fini(zp);
1240 	ZFS_OBJ_HOLD_EXIT(zfsvfs, obj);
1241 	zfs_znode_free(zp);
1242 }
1243 
1244 void
zfs_zinactive(znode_t * zp)1245 zfs_zinactive(znode_t *zp)
1246 {
1247 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1248 	uint64_t z_id = zp->z_id;
1249 
1250 	ASSERT3P(zp->z_sa_hdl, !=, NULL);
1251 
1252 	/*
1253 	 * Don't allow a zfs_zget() while were trying to release this znode
1254 	 */
1255 	ZFS_OBJ_HOLD_ENTER(zfsvfs, z_id);
1256 
1257 	/*
1258 	 * If this was the last reference to a file with no links, remove
1259 	 * the file from the file system unless the file system is mounted
1260 	 * read-only.  That can happen, for example, if the file system was
1261 	 * originally read-write, the file was opened, then unlinked and
1262 	 * the file system was made read-only before the file was finally
1263 	 * closed.  The file will remain in the unlinked set.
1264 	 */
1265 	if (zp->z_unlinked) {
1266 		ASSERT(!zfsvfs->z_issnap);
1267 		if ((zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) == 0) {
1268 			ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id);
1269 			zfs_rmnode(zp);
1270 			return;
1271 		}
1272 	}
1273 
1274 	zfs_znode_dmu_fini(zp);
1275 	ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id);
1276 	zfs_znode_free(zp);
1277 }
1278 
1279 void
zfs_znode_free(znode_t * zp)1280 zfs_znode_free(znode_t *zp)
1281 {
1282 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1283 #if __FreeBSD_version >= 1300139
1284 	char *symlink;
1285 #endif
1286 
1287 	ASSERT3P(zp->z_sa_hdl, ==, NULL);
1288 	zp->z_vnode = NULL;
1289 	mutex_enter(&zfsvfs->z_znodes_lock);
1290 	POINTER_INVALIDATE(&zp->z_zfsvfs);
1291 	list_remove(&zfsvfs->z_all_znodes, zp);
1292 	zfsvfs->z_nr_znodes--;
1293 	mutex_exit(&zfsvfs->z_znodes_lock);
1294 	symlink = atomic_load_ptr(&zp->z_cached_symlink);
1295 	if (symlink != NULL) {
1296 		atomic_store_rel_ptr((uintptr_t *)&zp->z_cached_symlink, (uintptr_t)NULL);
1297 		cache_symlink_free(symlink, strlen(symlink) + 1);
1298 	}
1299 
1300 #if __FreeBSD_version >= 1300139
1301 	symlink = atomic_load_ptr(&zp->z_cached_symlink);
1302 	if (symlink != NULL) {
1303 		atomic_store_rel_ptr((uintptr_t *)&zp->z_cached_symlink,
1304 		    (uintptr_t)NULL);
1305 		cache_symlink_free(symlink, strlen(symlink) + 1);
1306 	}
1307 #endif
1308 
1309 	if (zp->z_acl_cached) {
1310 		zfs_acl_free(zp->z_acl_cached);
1311 		zp->z_acl_cached = NULL;
1312 	}
1313 
1314 	zfs_znode_free_kmem(zp);
1315 }
1316 
1317 void
zfs_tstamp_update_setup_ext(znode_t * zp,uint_t flag,uint64_t mtime[2],uint64_t ctime[2],boolean_t have_tx)1318 zfs_tstamp_update_setup_ext(znode_t *zp, uint_t flag, uint64_t mtime[2],
1319     uint64_t ctime[2], boolean_t have_tx)
1320 {
1321 	timestruc_t	now;
1322 
1323 	vfs_timestamp(&now);
1324 
1325 	if (have_tx) {	/* will sa_bulk_update happen really soon? */
1326 		zp->z_atime_dirty = 0;
1327 		zp->z_seq++;
1328 	} else {
1329 		zp->z_atime_dirty = 1;
1330 	}
1331 
1332 	if (flag & AT_ATIME) {
1333 		ZFS_TIME_ENCODE(&now, zp->z_atime);
1334 	}
1335 
1336 	if (flag & AT_MTIME) {
1337 		ZFS_TIME_ENCODE(&now, mtime);
1338 		if (zp->z_zfsvfs->z_use_fuids) {
1339 			zp->z_pflags |= (ZFS_ARCHIVE |
1340 			    ZFS_AV_MODIFIED);
1341 		}
1342 	}
1343 
1344 	if (flag & AT_CTIME) {
1345 		ZFS_TIME_ENCODE(&now, ctime);
1346 		if (zp->z_zfsvfs->z_use_fuids)
1347 			zp->z_pflags |= ZFS_ARCHIVE;
1348 	}
1349 }
1350 
1351 
1352 void
zfs_tstamp_update_setup(znode_t * zp,uint_t flag,uint64_t mtime[2],uint64_t ctime[2])1353 zfs_tstamp_update_setup(znode_t *zp, uint_t flag, uint64_t mtime[2],
1354     uint64_t ctime[2])
1355 {
1356 	zfs_tstamp_update_setup_ext(zp, flag, mtime, ctime, B_TRUE);
1357 }
1358 /*
1359  * Grow the block size for a file.
1360  *
1361  *	IN:	zp	- znode of file to free data in.
1362  *		size	- requested block size
1363  *		tx	- open transaction.
1364  *
1365  * NOTE: this function assumes that the znode is write locked.
1366  */
1367 void
zfs_grow_blocksize(znode_t * zp,uint64_t size,dmu_tx_t * tx)1368 zfs_grow_blocksize(znode_t *zp, uint64_t size, dmu_tx_t *tx)
1369 {
1370 	int		error;
1371 	u_longlong_t	dummy;
1372 
1373 	if (size <= zp->z_blksz)
1374 		return;
1375 	/*
1376 	 * If the file size is already greater than the current blocksize,
1377 	 * we will not grow.  If there is more than one block in a file,
1378 	 * the blocksize cannot change.
1379 	 */
1380 	if (zp->z_blksz && zp->z_size > zp->z_blksz)
1381 		return;
1382 
1383 	error = dmu_object_set_blocksize(zp->z_zfsvfs->z_os, zp->z_id,
1384 	    size, 0, tx);
1385 
1386 	if (error == ENOTSUP)
1387 		return;
1388 	ASSERT0(error);
1389 
1390 	/* What blocksize did we actually get? */
1391 	dmu_object_size_from_db(sa_get_db(zp->z_sa_hdl), &zp->z_blksz, &dummy);
1392 }
1393 
1394 /*
1395  * Increase the file length
1396  *
1397  *	IN:	zp	- znode of file to free data in.
1398  *		end	- new end-of-file
1399  *
1400  *	RETURN:	0 on success, error code on failure
1401  */
1402 static int
zfs_extend(znode_t * zp,uint64_t end)1403 zfs_extend(znode_t *zp, uint64_t end)
1404 {
1405 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1406 	dmu_tx_t *tx;
1407 	zfs_locked_range_t *lr;
1408 	uint64_t newblksz;
1409 	int error;
1410 
1411 	/*
1412 	 * We will change zp_size, lock the whole file.
1413 	 */
1414 	lr = zfs_rangelock_enter(&zp->z_rangelock, 0, UINT64_MAX, RL_WRITER);
1415 
1416 	/*
1417 	 * Nothing to do if file already at desired length.
1418 	 */
1419 	if (end <= zp->z_size) {
1420 		zfs_rangelock_exit(lr);
1421 		return (0);
1422 	}
1423 	tx = dmu_tx_create(zfsvfs->z_os);
1424 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1425 	zfs_sa_upgrade_txholds(tx, zp);
1426 	if (end > zp->z_blksz &&
1427 	    (!ISP2(zp->z_blksz) || zp->z_blksz < zfsvfs->z_max_blksz)) {
1428 		/*
1429 		 * We are growing the file past the current block size.
1430 		 */
1431 		if (zp->z_blksz > zp->z_zfsvfs->z_max_blksz) {
1432 			/*
1433 			 * File's blocksize is already larger than the
1434 			 * "recordsize" property.  Only let it grow to
1435 			 * the next power of 2.
1436 			 */
1437 			ASSERT(!ISP2(zp->z_blksz));
1438 			newblksz = MIN(end, 1 << highbit64(zp->z_blksz));
1439 		} else {
1440 			newblksz = MIN(end, zp->z_zfsvfs->z_max_blksz);
1441 		}
1442 		dmu_tx_hold_write(tx, zp->z_id, 0, newblksz);
1443 	} else {
1444 		newblksz = 0;
1445 	}
1446 
1447 	error = dmu_tx_assign(tx, TXG_WAIT);
1448 	if (error) {
1449 		dmu_tx_abort(tx);
1450 		zfs_rangelock_exit(lr);
1451 		return (error);
1452 	}
1453 
1454 	if (newblksz)
1455 		zfs_grow_blocksize(zp, newblksz, tx);
1456 
1457 	zp->z_size = end;
1458 
1459 	VERIFY0(sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zp->z_zfsvfs),
1460 	    &zp->z_size, sizeof (zp->z_size), tx));
1461 
1462 	vnode_pager_setsize(ZTOV(zp), end);
1463 
1464 	zfs_rangelock_exit(lr);
1465 
1466 	dmu_tx_commit(tx);
1467 
1468 	return (0);
1469 }
1470 
1471 /*
1472  * Free space in a file.
1473  *
1474  *	IN:	zp	- znode of file to free data in.
1475  *		off	- start of section to free.
1476  *		len	- length of section to free.
1477  *
1478  *	RETURN:	0 on success, error code on failure
1479  */
1480 static int
zfs_free_range(znode_t * zp,uint64_t off,uint64_t len)1481 zfs_free_range(znode_t *zp, uint64_t off, uint64_t len)
1482 {
1483 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1484 	zfs_locked_range_t *lr;
1485 	int error;
1486 
1487 	/*
1488 	 * Lock the range being freed.
1489 	 */
1490 	lr = zfs_rangelock_enter(&zp->z_rangelock, off, len, RL_WRITER);
1491 
1492 	/*
1493 	 * Nothing to do if file already at desired length.
1494 	 */
1495 	if (off >= zp->z_size) {
1496 		zfs_rangelock_exit(lr);
1497 		return (0);
1498 	}
1499 
1500 	if (off + len > zp->z_size)
1501 		len = zp->z_size - off;
1502 
1503 	error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, off, len);
1504 
1505 	if (error == 0) {
1506 #if __FreeBSD_version >= 1400032
1507 		vnode_pager_purge_range(ZTOV(zp), off, off + len);
1508 #else
1509 		/*
1510 		 * Before __FreeBSD_version 1400032 we cannot free block in the
1511 		 * middle of a file, but only at the end of a file, so this code
1512 		 * path should never happen.
1513 		 */
1514 		vnode_pager_setsize(ZTOV(zp), off);
1515 #endif
1516 	}
1517 
1518 	zfs_rangelock_exit(lr);
1519 
1520 	return (error);
1521 }
1522 
1523 /*
1524  * Truncate a file
1525  *
1526  *	IN:	zp	- znode of file to free data in.
1527  *		end	- new end-of-file.
1528  *
1529  *	RETURN:	0 on success, error code on failure
1530  */
1531 static int
zfs_trunc(znode_t * zp,uint64_t end)1532 zfs_trunc(znode_t *zp, uint64_t end)
1533 {
1534 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1535 	vnode_t *vp = ZTOV(zp);
1536 	dmu_tx_t *tx;
1537 	zfs_locked_range_t *lr;
1538 	int error;
1539 	sa_bulk_attr_t bulk[2];
1540 	int count = 0;
1541 
1542 	/*
1543 	 * We will change zp_size, lock the whole file.
1544 	 */
1545 	lr = zfs_rangelock_enter(&zp->z_rangelock, 0, UINT64_MAX, RL_WRITER);
1546 
1547 	/*
1548 	 * Nothing to do if file already at desired length.
1549 	 */
1550 	if (end >= zp->z_size) {
1551 		zfs_rangelock_exit(lr);
1552 		return (0);
1553 	}
1554 
1555 	error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, end,
1556 	    DMU_OBJECT_END);
1557 	if (error) {
1558 		zfs_rangelock_exit(lr);
1559 		return (error);
1560 	}
1561 	tx = dmu_tx_create(zfsvfs->z_os);
1562 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1563 	zfs_sa_upgrade_txholds(tx, zp);
1564 	dmu_tx_mark_netfree(tx);
1565 	error = dmu_tx_assign(tx, TXG_WAIT);
1566 	if (error) {
1567 		dmu_tx_abort(tx);
1568 		zfs_rangelock_exit(lr);
1569 		return (error);
1570 	}
1571 
1572 	zp->z_size = end;
1573 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs),
1574 	    NULL, &zp->z_size, sizeof (zp->z_size));
1575 
1576 	if (end == 0) {
1577 		zp->z_pflags &= ~ZFS_SPARSE;
1578 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
1579 		    NULL, &zp->z_pflags, 8);
1580 	}
1581 	VERIFY0(sa_bulk_update(zp->z_sa_hdl, bulk, count, tx));
1582 
1583 	dmu_tx_commit(tx);
1584 
1585 	/*
1586 	 * Clear any mapped pages in the truncated region.  This has to
1587 	 * happen outside of the transaction to avoid the possibility of
1588 	 * a deadlock with someone trying to push a page that we are
1589 	 * about to invalidate.
1590 	 */
1591 	vnode_pager_setsize(vp, end);
1592 
1593 	zfs_rangelock_exit(lr);
1594 
1595 	return (0);
1596 }
1597 
1598 /*
1599  * Free space in a file
1600  *
1601  *	IN:	zp	- znode of file to free data in.
1602  *		off	- start of range
1603  *		len	- end of range (0 => EOF)
1604  *		flag	- current file open mode flags.
1605  *		log	- TRUE if this action should be logged
1606  *
1607  *	RETURN:	0 on success, error code on failure
1608  */
1609 int
zfs_freesp(znode_t * zp,uint64_t off,uint64_t len,int flag,boolean_t log)1610 zfs_freesp(znode_t *zp, uint64_t off, uint64_t len, int flag, boolean_t log)
1611 {
1612 	dmu_tx_t *tx;
1613 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1614 	zilog_t *zilog = zfsvfs->z_log;
1615 	uint64_t mode;
1616 	uint64_t mtime[2], ctime[2];
1617 	sa_bulk_attr_t bulk[3];
1618 	int count = 0;
1619 	int error;
1620 
1621 	if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs), &mode,
1622 	    sizeof (mode))) != 0)
1623 		return (error);
1624 
1625 	if (off > zp->z_size) {
1626 		error =  zfs_extend(zp, off+len);
1627 		if (error == 0 && log)
1628 			goto log;
1629 		else
1630 			return (error);
1631 	}
1632 
1633 	if (len == 0) {
1634 		error = zfs_trunc(zp, off);
1635 	} else {
1636 		if ((error = zfs_free_range(zp, off, len)) == 0 &&
1637 		    off + len > zp->z_size)
1638 			error = zfs_extend(zp, off+len);
1639 	}
1640 	if (error || !log)
1641 		return (error);
1642 log:
1643 	tx = dmu_tx_create(zfsvfs->z_os);
1644 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1645 	zfs_sa_upgrade_txholds(tx, zp);
1646 	error = dmu_tx_assign(tx, TXG_WAIT);
1647 	if (error) {
1648 		dmu_tx_abort(tx);
1649 		return (error);
1650 	}
1651 
1652 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, mtime, 16);
1653 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, ctime, 16);
1654 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
1655 	    NULL, &zp->z_pflags, 8);
1656 	zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime);
1657 	error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
1658 	ASSERT0(error);
1659 
1660 	zfs_log_truncate(zilog, tx, TX_TRUNCATE, zp, off, len);
1661 
1662 	dmu_tx_commit(tx);
1663 	return (0);
1664 }
1665 
1666 void
zfs_create_fs(objset_t * os,cred_t * cr,nvlist_t * zplprops,dmu_tx_t * tx)1667 zfs_create_fs(objset_t *os, cred_t *cr, nvlist_t *zplprops, dmu_tx_t *tx)
1668 {
1669 	uint64_t	moid, obj, sa_obj, version;
1670 	uint64_t	sense = ZFS_CASE_SENSITIVE;
1671 	uint64_t	norm = 0;
1672 	nvpair_t	*elem;
1673 	int		error;
1674 	int		i;
1675 	znode_t		*rootzp = NULL;
1676 	zfsvfs_t	*zfsvfs;
1677 	vattr_t		vattr;
1678 	znode_t		*zp;
1679 	zfs_acl_ids_t	acl_ids;
1680 
1681 	/*
1682 	 * First attempt to create master node.
1683 	 */
1684 	/*
1685 	 * In an empty objset, there are no blocks to read and thus
1686 	 * there can be no i/o errors (which we assert below).
1687 	 */
1688 	moid = MASTER_NODE_OBJ;
1689 	error = zap_create_claim(os, moid, DMU_OT_MASTER_NODE,
1690 	    DMU_OT_NONE, 0, tx);
1691 	ASSERT0(error);
1692 
1693 	/*
1694 	 * Set starting attributes.
1695 	 */
1696 	version = zfs_zpl_version_map(spa_version(dmu_objset_spa(os)));
1697 	elem = NULL;
1698 	while ((elem = nvlist_next_nvpair(zplprops, elem)) != NULL) {
1699 		/* For the moment we expect all zpl props to be uint64_ts */
1700 		uint64_t val;
1701 		char *name;
1702 
1703 		ASSERT3S(nvpair_type(elem), ==, DATA_TYPE_UINT64);
1704 		val = fnvpair_value_uint64(elem);
1705 		name = nvpair_name(elem);
1706 		if (strcmp(name, zfs_prop_to_name(ZFS_PROP_VERSION)) == 0) {
1707 			if (val < version)
1708 				version = val;
1709 		} else {
1710 			error = zap_update(os, moid, name, 8, 1, &val, tx);
1711 		}
1712 		ASSERT0(error);
1713 		if (strcmp(name, zfs_prop_to_name(ZFS_PROP_NORMALIZE)) == 0)
1714 			norm = val;
1715 		else if (strcmp(name, zfs_prop_to_name(ZFS_PROP_CASE)) == 0)
1716 			sense = val;
1717 	}
1718 	ASSERT3U(version, !=, 0);
1719 	error = zap_update(os, moid, ZPL_VERSION_STR, 8, 1, &version, tx);
1720 
1721 	/*
1722 	 * Create zap object used for SA attribute registration
1723 	 */
1724 
1725 	if (version >= ZPL_VERSION_SA) {
1726 		sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
1727 		    DMU_OT_NONE, 0, tx);
1728 		error = zap_add(os, moid, ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
1729 		ASSERT0(error);
1730 	} else {
1731 		sa_obj = 0;
1732 	}
1733 	/*
1734 	 * Create a delete queue.
1735 	 */
1736 	obj = zap_create(os, DMU_OT_UNLINKED_SET, DMU_OT_NONE, 0, tx);
1737 
1738 	error = zap_add(os, moid, ZFS_UNLINKED_SET, 8, 1, &obj, tx);
1739 	ASSERT0(error);
1740 
1741 	/*
1742 	 * Create root znode.  Create minimal znode/vnode/zfsvfs
1743 	 * to allow zfs_mknode to work.
1744 	 */
1745 	VATTR_NULL(&vattr);
1746 	vattr.va_mask = AT_MODE|AT_UID|AT_GID;
1747 	vattr.va_type = VDIR;
1748 	vattr.va_mode = S_IFDIR|0755;
1749 	vattr.va_uid = crgetuid(cr);
1750 	vattr.va_gid = crgetgid(cr);
1751 
1752 	zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
1753 
1754 	rootzp = zfs_znode_alloc_kmem(KM_SLEEP);
1755 	ASSERT(!POINTER_IS_VALID(rootzp->z_zfsvfs));
1756 	rootzp->z_unlinked = 0;
1757 	rootzp->z_atime_dirty = 0;
1758 	rootzp->z_is_sa = USE_SA(version, os);
1759 
1760 	zfsvfs->z_os = os;
1761 	zfsvfs->z_parent = zfsvfs;
1762 	zfsvfs->z_version = version;
1763 	zfsvfs->z_use_fuids = USE_FUIDS(version, os);
1764 	zfsvfs->z_use_sa = USE_SA(version, os);
1765 	zfsvfs->z_norm = norm;
1766 
1767 	error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
1768 	    &zfsvfs->z_attr_table);
1769 
1770 	ASSERT0(error);
1771 
1772 	/*
1773 	 * Fold case on file systems that are always or sometimes case
1774 	 * insensitive.
1775 	 */
1776 	if (sense == ZFS_CASE_INSENSITIVE || sense == ZFS_CASE_MIXED)
1777 		zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER;
1778 
1779 	mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
1780 	list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
1781 	    offsetof(znode_t, z_link_node));
1782 
1783 	for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1784 		mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL);
1785 
1786 	rootzp->z_zfsvfs = zfsvfs;
1787 	VERIFY0(zfs_acl_ids_create(rootzp, IS_ROOT_NODE, &vattr,
1788 	    cr, NULL, &acl_ids));
1789 	zfs_mknode(rootzp, &vattr, tx, cr, IS_ROOT_NODE, &zp, &acl_ids);
1790 	ASSERT3P(zp, ==, rootzp);
1791 	error = zap_add(os, moid, ZFS_ROOT_OBJ, 8, 1, &rootzp->z_id, tx);
1792 	ASSERT0(error);
1793 	zfs_acl_ids_free(&acl_ids);
1794 	POINTER_INVALIDATE(&rootzp->z_zfsvfs);
1795 
1796 	sa_handle_destroy(rootzp->z_sa_hdl);
1797 	zfs_znode_free_kmem(rootzp);
1798 
1799 	/*
1800 	 * Create shares directory
1801 	 */
1802 
1803 	error = zfs_create_share_dir(zfsvfs, tx);
1804 
1805 	ASSERT0(error);
1806 
1807 	for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1808 		mutex_destroy(&zfsvfs->z_hold_mtx[i]);
1809 	kmem_free(zfsvfs, sizeof (zfsvfs_t));
1810 }
1811 #endif /* _KERNEL */
1812 
1813 static int
zfs_sa_setup(objset_t * osp,sa_attr_type_t ** sa_table)1814 zfs_sa_setup(objset_t *osp, sa_attr_type_t **sa_table)
1815 {
1816 	uint64_t sa_obj = 0;
1817 	int error;
1818 
1819 	error = zap_lookup(osp, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1, &sa_obj);
1820 	if (error != 0 && error != ENOENT)
1821 		return (error);
1822 
1823 	error = sa_setup(osp, sa_obj, zfs_attr_table, ZPL_END, sa_table);
1824 	return (error);
1825 }
1826 
1827 static int
zfs_grab_sa_handle(objset_t * osp,uint64_t obj,sa_handle_t ** hdlp,dmu_buf_t ** db,void * tag)1828 zfs_grab_sa_handle(objset_t *osp, uint64_t obj, sa_handle_t **hdlp,
1829     dmu_buf_t **db, void *tag)
1830 {
1831 	dmu_object_info_t doi;
1832 	int error;
1833 
1834 	if ((error = sa_buf_hold(osp, obj, tag, db)) != 0)
1835 		return (error);
1836 
1837 	dmu_object_info_from_db(*db, &doi);
1838 	if ((doi.doi_bonus_type != DMU_OT_SA &&
1839 	    doi.doi_bonus_type != DMU_OT_ZNODE) ||
1840 	    (doi.doi_bonus_type == DMU_OT_ZNODE &&
1841 	    doi.doi_bonus_size < sizeof (znode_phys_t))) {
1842 		sa_buf_rele(*db, tag);
1843 		return (SET_ERROR(ENOTSUP));
1844 	}
1845 
1846 	error = sa_handle_get(osp, obj, NULL, SA_HDL_PRIVATE, hdlp);
1847 	if (error != 0) {
1848 		sa_buf_rele(*db, tag);
1849 		return (error);
1850 	}
1851 
1852 	return (0);
1853 }
1854 
1855 static void
zfs_release_sa_handle(sa_handle_t * hdl,dmu_buf_t * db,void * tag)1856 zfs_release_sa_handle(sa_handle_t *hdl, dmu_buf_t *db, void *tag)
1857 {
1858 	sa_handle_destroy(hdl);
1859 	sa_buf_rele(db, tag);
1860 }
1861 
1862 /*
1863  * Given an object number, return its parent object number and whether
1864  * or not the object is an extended attribute directory.
1865  */
1866 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)1867 zfs_obj_to_pobj(objset_t *osp, sa_handle_t *hdl, sa_attr_type_t *sa_table,
1868     uint64_t *pobjp, int *is_xattrdir)
1869 {
1870 	uint64_t parent;
1871 	uint64_t pflags;
1872 	uint64_t mode;
1873 	uint64_t parent_mode;
1874 	sa_bulk_attr_t bulk[3];
1875 	sa_handle_t *sa_hdl;
1876 	dmu_buf_t *sa_db;
1877 	int count = 0;
1878 	int error;
1879 
1880 	SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_PARENT], NULL,
1881 	    &parent, sizeof (parent));
1882 	SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_FLAGS], NULL,
1883 	    &pflags, sizeof (pflags));
1884 	SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL,
1885 	    &mode, sizeof (mode));
1886 
1887 	if ((error = sa_bulk_lookup(hdl, bulk, count)) != 0)
1888 		return (error);
1889 
1890 	/*
1891 	 * When a link is removed its parent pointer is not changed and will
1892 	 * be invalid.  There are two cases where a link is removed but the
1893 	 * file stays around, when it goes to the delete queue and when there
1894 	 * are additional links.
1895 	 */
1896 	error = zfs_grab_sa_handle(osp, parent, &sa_hdl, &sa_db, FTAG);
1897 	if (error != 0)
1898 		return (error);
1899 
1900 	error = sa_lookup(sa_hdl, ZPL_MODE, &parent_mode, sizeof (parent_mode));
1901 	zfs_release_sa_handle(sa_hdl, sa_db, FTAG);
1902 	if (error != 0)
1903 		return (error);
1904 
1905 	*is_xattrdir = ((pflags & ZFS_XATTR) != 0) && S_ISDIR(mode);
1906 
1907 	/*
1908 	 * Extended attributes can be applied to files, directories, etc.
1909 	 * Otherwise the parent must be a directory.
1910 	 */
1911 	if (!*is_xattrdir && !S_ISDIR(parent_mode))
1912 		return (SET_ERROR(EINVAL));
1913 
1914 	*pobjp = parent;
1915 
1916 	return (0);
1917 }
1918 
1919 /*
1920  * Given an object number, return some zpl level statistics
1921  */
1922 static int
zfs_obj_to_stats_impl(sa_handle_t * hdl,sa_attr_type_t * sa_table,zfs_stat_t * sb)1923 zfs_obj_to_stats_impl(sa_handle_t *hdl, sa_attr_type_t *sa_table,
1924     zfs_stat_t *sb)
1925 {
1926 	sa_bulk_attr_t bulk[4];
1927 	int count = 0;
1928 
1929 	SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL,
1930 	    &sb->zs_mode, sizeof (sb->zs_mode));
1931 	SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_GEN], NULL,
1932 	    &sb->zs_gen, sizeof (sb->zs_gen));
1933 	SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_LINKS], NULL,
1934 	    &sb->zs_links, sizeof (sb->zs_links));
1935 	SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_CTIME], NULL,
1936 	    &sb->zs_ctime, sizeof (sb->zs_ctime));
1937 
1938 	return (sa_bulk_lookup(hdl, bulk, count));
1939 }
1940 
1941 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)1942 zfs_obj_to_path_impl(objset_t *osp, uint64_t obj, sa_handle_t *hdl,
1943     sa_attr_type_t *sa_table, char *buf, int len)
1944 {
1945 	sa_handle_t *sa_hdl;
1946 	sa_handle_t *prevhdl = NULL;
1947 	dmu_buf_t *prevdb = NULL;
1948 	dmu_buf_t *sa_db = NULL;
1949 	char *path = buf + len - 1;
1950 	int error;
1951 
1952 	*path = '\0';
1953 	sa_hdl = hdl;
1954 
1955 	uint64_t deleteq_obj;
1956 	VERIFY0(zap_lookup(osp, MASTER_NODE_OBJ,
1957 	    ZFS_UNLINKED_SET, sizeof (uint64_t), 1, &deleteq_obj));
1958 	error = zap_lookup_int(osp, deleteq_obj, obj);
1959 	if (error == 0) {
1960 		return (ESTALE);
1961 	} else if (error != ENOENT) {
1962 		return (error);
1963 	}
1964 	error = 0;
1965 
1966 	for (;;) {
1967 		uint64_t pobj;
1968 		char component[MAXNAMELEN + 2];
1969 		size_t complen;
1970 		int is_xattrdir;
1971 
1972 		if (prevdb) {
1973 			ASSERT3P(prevhdl, !=, NULL);
1974 			zfs_release_sa_handle(prevhdl, prevdb, FTAG);
1975 		}
1976 
1977 		if ((error = zfs_obj_to_pobj(osp, sa_hdl, sa_table, &pobj,
1978 		    &is_xattrdir)) != 0)
1979 			break;
1980 
1981 		if (pobj == obj) {
1982 			if (path[0] != '/')
1983 				*--path = '/';
1984 			break;
1985 		}
1986 
1987 		component[0] = '/';
1988 		if (is_xattrdir) {
1989 			(void) sprintf(component + 1, "<xattrdir>");
1990 		} else {
1991 			error = zap_value_search(osp, pobj, obj,
1992 			    ZFS_DIRENT_OBJ(-1ULL), component + 1);
1993 			if (error != 0)
1994 				break;
1995 		}
1996 
1997 		complen = strlen(component);
1998 		path -= complen;
1999 		ASSERT3P(path, >=, buf);
2000 		bcopy(component, path, complen);
2001 		obj = pobj;
2002 
2003 		if (sa_hdl != hdl) {
2004 			prevhdl = sa_hdl;
2005 			prevdb = sa_db;
2006 		}
2007 		error = zfs_grab_sa_handle(osp, obj, &sa_hdl, &sa_db, FTAG);
2008 		if (error != 0) {
2009 			sa_hdl = prevhdl;
2010 			sa_db = prevdb;
2011 			break;
2012 		}
2013 	}
2014 
2015 	if (sa_hdl != NULL && sa_hdl != hdl) {
2016 		ASSERT3P(sa_db, !=, NULL);
2017 		zfs_release_sa_handle(sa_hdl, sa_db, FTAG);
2018 	}
2019 
2020 	if (error == 0)
2021 		(void) memmove(buf, path, buf + len - path);
2022 
2023 	return (error);
2024 }
2025 
2026 int
zfs_obj_to_path(objset_t * osp,uint64_t obj,char * buf,int len)2027 zfs_obj_to_path(objset_t *osp, uint64_t obj, char *buf, int len)
2028 {
2029 	sa_attr_type_t *sa_table;
2030 	sa_handle_t *hdl;
2031 	dmu_buf_t *db;
2032 	int error;
2033 
2034 	error = zfs_sa_setup(osp, &sa_table);
2035 	if (error != 0)
2036 		return (error);
2037 
2038 	error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG);
2039 	if (error != 0)
2040 		return (error);
2041 
2042 	error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len);
2043 
2044 	zfs_release_sa_handle(hdl, db, FTAG);
2045 	return (error);
2046 }
2047 
2048 int
zfs_obj_to_stats(objset_t * osp,uint64_t obj,zfs_stat_t * sb,char * buf,int len)2049 zfs_obj_to_stats(objset_t *osp, uint64_t obj, zfs_stat_t *sb,
2050     char *buf, int len)
2051 {
2052 	char *path = buf + len - 1;
2053 	sa_attr_type_t *sa_table;
2054 	sa_handle_t *hdl;
2055 	dmu_buf_t *db;
2056 	int error;
2057 
2058 	*path = '\0';
2059 
2060 	error = zfs_sa_setup(osp, &sa_table);
2061 	if (error != 0)
2062 		return (error);
2063 
2064 	error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG);
2065 	if (error != 0)
2066 		return (error);
2067 
2068 	error = zfs_obj_to_stats_impl(hdl, sa_table, sb);
2069 	if (error != 0) {
2070 		zfs_release_sa_handle(hdl, db, FTAG);
2071 		return (error);
2072 	}
2073 
2074 	error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len);
2075 
2076 	zfs_release_sa_handle(hdl, db, FTAG);
2077 	return (error);
2078 }
2079 
2080 
2081 void
zfs_znode_update_vfs(znode_t * zp)2082 zfs_znode_update_vfs(znode_t *zp)
2083 {
2084 	vm_object_t object;
2085 
2086 	if ((object = ZTOV(zp)->v_object) == NULL ||
2087 	    zp->z_size == object->un_pager.vnp.vnp_size)
2088 		return;
2089 
2090 	vnode_pager_setsize(ZTOV(zp), zp->z_size);
2091 }
2092 
2093 
2094 #ifdef _KERNEL
2095 int
zfs_znode_parent_and_name(znode_t * zp,znode_t ** dzpp,char * buf)2096 zfs_znode_parent_and_name(znode_t *zp, znode_t **dzpp, char *buf)
2097 {
2098 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2099 	uint64_t parent;
2100 	int is_xattrdir;
2101 	int err;
2102 
2103 	/* Extended attributes should not be visible as regular files. */
2104 	if ((zp->z_pflags & ZFS_XATTR) != 0)
2105 		return (SET_ERROR(EINVAL));
2106 
2107 	err = zfs_obj_to_pobj(zfsvfs->z_os, zp->z_sa_hdl, zfsvfs->z_attr_table,
2108 	    &parent, &is_xattrdir);
2109 	if (err != 0)
2110 		return (err);
2111 	ASSERT0(is_xattrdir);
2112 
2113 	/* No name as this is a root object. */
2114 	if (parent == zp->z_id)
2115 		return (SET_ERROR(EINVAL));
2116 
2117 	err = zap_value_search(zfsvfs->z_os, parent, zp->z_id,
2118 	    ZFS_DIRENT_OBJ(-1ULL), buf);
2119 	if (err != 0)
2120 		return (err);
2121 	err = zfs_zget(zfsvfs, parent, dzpp);
2122 	return (err);
2123 }
2124 #endif /* _KERNEL */
2125