xref: /titanic_41/usr/src/uts/common/fs/zfs/zfs_znode.c (revision 67738a4fd5331c5c62b9fead7a7739afa2476cf9)
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  */
25 
26 /* Portions Copyright 2007 Jeremy Teo */
27 
28 #ifdef _KERNEL
29 #include <sys/types.h>
30 #include <sys/param.h>
31 #include <sys/time.h>
32 #include <sys/systm.h>
33 #include <sys/sysmacros.h>
34 #include <sys/resource.h>
35 #include <sys/mntent.h>
36 #include <sys/mkdev.h>
37 #include <sys/u8_textprep.h>
38 #include <sys/dsl_dataset.h>
39 #include <sys/vfs.h>
40 #include <sys/vfs_opreg.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/mode.h>
47 #include <sys/atomic.h>
48 #include <vm/pvn.h>
49 #include "fs/fs_subr.h"
50 #include <sys/zfs_dir.h>
51 #include <sys/zfs_acl.h>
52 #include <sys/zfs_ioctl.h>
53 #include <sys/zfs_rlock.h>
54 #include <sys/zfs_fuid.h>
55 #include <sys/dnode.h>
56 #include <sys/fs/zfs.h>
57 #include <sys/kidmap.h>
58 #endif /* _KERNEL */
59 
60 #include <sys/dmu.h>
61 #include <sys/dmu_objset.h>
62 #include <sys/refcount.h>
63 #include <sys/stat.h>
64 #include <sys/zap.h>
65 #include <sys/zfs_znode.h>
66 #include <sys/sa.h>
67 #include <sys/zfs_sa.h>
68 #include <sys/zfs_stat.h>
69 #include <sys/zfs_events.h>
70 #include <zev/zev.h>
71 
72 #include "zfs_prop.h"
73 #include "zfs_comutil.h"
74 
75 /*
76  * Define ZNODE_STATS to turn on statistic gathering. By default, it is only
77  * turned on when DEBUG is also defined.
78  */
79 #ifdef	DEBUG
80 #define	ZNODE_STATS
81 #endif	/* DEBUG */
82 
83 #ifdef	ZNODE_STATS
84 #define	ZNODE_STAT_ADD(stat)			((stat)++)
85 #else
86 #define	ZNODE_STAT_ADD(stat)			/* nothing */
87 #endif	/* ZNODE_STATS */
88 
89 /*
90  * Functions needed for userland (ie: libzpool) are not put under
91  * #ifdef_KERNEL; the rest of the functions have dependencies
92  * (such as VFS logic) that will not compile easily in userland.
93  */
94 #ifdef _KERNEL
95 /*
96  * Needed to close a small window in zfs_znode_move() that allows the zfsvfs to
97  * be freed before it can be safely accessed.
98  */
99 krwlock_t zfsvfs_lock;
100 
101 static kmem_cache_t *znode_cache = NULL;
102 
103 /*ARGSUSED*/
104 static void
znode_evict_error(dmu_buf_t * dbuf,void * user_ptr)105 znode_evict_error(dmu_buf_t *dbuf, void *user_ptr)
106 {
107 	/*
108 	 * We should never drop all dbuf refs without first clearing
109 	 * the eviction callback.
110 	 */
111 	panic("evicting znode %p\n", user_ptr);
112 }
113 
114 /*ARGSUSED*/
115 static int
zfs_znode_cache_constructor(void * buf,void * arg,int kmflags)116 zfs_znode_cache_constructor(void *buf, void *arg, int kmflags)
117 {
118 	znode_t *zp = buf;
119 
120 	ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs));
121 
122 	zp->z_vnode = vn_alloc(kmflags);
123 	if (zp->z_vnode == NULL) {
124 		return (-1);
125 	}
126 	ZTOV(zp)->v_data = zp;
127 
128 	list_link_init(&zp->z_link_node);
129 
130 	mutex_init(&zp->z_lock, NULL, MUTEX_DEFAULT, NULL);
131 	rw_init(&zp->z_parent_lock, NULL, RW_DEFAULT, NULL);
132 	rw_init(&zp->z_name_lock, NULL, RW_DEFAULT, NULL);
133 	mutex_init(&zp->z_acl_lock, NULL, MUTEX_DEFAULT, NULL);
134 
135 	mutex_init(&zp->z_range_lock, NULL, MUTEX_DEFAULT, NULL);
136 	avl_create(&zp->z_range_avl, zfs_range_compare,
137 	    sizeof (rl_t), offsetof(rl_t, r_node));
138 
139 	zp->z_dirlocks = NULL;
140 	zp->z_acl_cached = NULL;
141 	zp->z_moved = 0;
142 	return (0);
143 }
144 
145 /*ARGSUSED*/
146 static void
zfs_znode_cache_destructor(void * buf,void * arg)147 zfs_znode_cache_destructor(void *buf, void *arg)
148 {
149 	znode_t *zp = buf;
150 
151 	ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs));
152 	ASSERT(ZTOV(zp)->v_data == zp);
153 	vn_free(ZTOV(zp));
154 	ASSERT(!list_link_active(&zp->z_link_node));
155 	mutex_destroy(&zp->z_lock);
156 	rw_destroy(&zp->z_parent_lock);
157 	rw_destroy(&zp->z_name_lock);
158 	mutex_destroy(&zp->z_acl_lock);
159 	avl_destroy(&zp->z_range_avl);
160 	mutex_destroy(&zp->z_range_lock);
161 
162 	ASSERT(zp->z_dirlocks == NULL);
163 	ASSERT(zp->z_acl_cached == NULL);
164 }
165 
166 #ifdef	ZNODE_STATS
167 static struct {
168 	uint64_t zms_zfsvfs_invalid;
169 	uint64_t zms_zfsvfs_recheck1;
170 	uint64_t zms_zfsvfs_unmounted;
171 	uint64_t zms_zfsvfs_recheck2;
172 	uint64_t zms_obj_held;
173 	uint64_t zms_vnode_locked;
174 	uint64_t zms_not_only_dnlc;
175 } znode_move_stats;
176 #endif	/* ZNODE_STATS */
177 
178 static void
zfs_znode_move_impl(znode_t * ozp,znode_t * nzp)179 zfs_znode_move_impl(znode_t *ozp, znode_t *nzp)
180 {
181 	vnode_t *vp;
182 
183 	/* Copy fields. */
184 	nzp->z_zfsvfs = ozp->z_zfsvfs;
185 
186 	/* Swap vnodes. */
187 	vp = nzp->z_vnode;
188 	nzp->z_vnode = ozp->z_vnode;
189 	ozp->z_vnode = vp; /* let destructor free the overwritten vnode */
190 	ZTOV(ozp)->v_data = ozp;
191 	ZTOV(nzp)->v_data = nzp;
192 
193 	nzp->z_id = ozp->z_id;
194 	ASSERT(ozp->z_dirlocks == NULL); /* znode not in use */
195 	ASSERT(avl_numnodes(&ozp->z_range_avl) == 0);
196 	nzp->z_unlinked = ozp->z_unlinked;
197 	nzp->z_atime_dirty = ozp->z_atime_dirty;
198 	nzp->z_zn_prefetch = ozp->z_zn_prefetch;
199 	nzp->z_blksz = ozp->z_blksz;
200 	nzp->z_seq = ozp->z_seq;
201 	nzp->z_mapcnt = ozp->z_mapcnt;
202 	nzp->z_gen = ozp->z_gen;
203 	nzp->z_sync_cnt = ozp->z_sync_cnt;
204 	nzp->z_is_sa = ozp->z_is_sa;
205 	nzp->z_sa_hdl = ozp->z_sa_hdl;
206 	bcopy(ozp->z_atime, nzp->z_atime, sizeof (uint64_t) * 2);
207 	nzp->z_links = ozp->z_links;
208 	nzp->z_size = ozp->z_size;
209 	nzp->z_pflags = ozp->z_pflags;
210 	nzp->z_uid = ozp->z_uid;
211 	nzp->z_gid = ozp->z_gid;
212 	nzp->z_mode = ozp->z_mode;
213 	nzp->z_new_content = ozp->z_new_content;
214 
215 	/*
216 	 * Since this is just an idle znode and kmem is already dealing with
217 	 * memory pressure, release any cached ACL.
218 	 */
219 	if (ozp->z_acl_cached) {
220 		zfs_acl_free(ozp->z_acl_cached);
221 		ozp->z_acl_cached = NULL;
222 	}
223 
224 	sa_set_userp(nzp->z_sa_hdl, nzp);
225 
226 	/*
227 	 * Invalidate the original znode by clearing fields that provide a
228 	 * pointer back to the znode. Set the low bit of the vfs pointer to
229 	 * ensure that zfs_znode_move() recognizes the znode as invalid in any
230 	 * subsequent callback.
231 	 */
232 	ozp->z_sa_hdl = NULL;
233 	POINTER_INVALIDATE(&ozp->z_zfsvfs);
234 
235 	/*
236 	 * Mark the znode.
237 	 */
238 	nzp->z_moved = 1;
239 	ozp->z_moved = (uint8_t)-1;
240 }
241 
242 /*ARGSUSED*/
243 static kmem_cbrc_t
zfs_znode_move(void * buf,void * newbuf,size_t size,void * arg)244 zfs_znode_move(void *buf, void *newbuf, size_t size, void *arg)
245 {
246 	znode_t *ozp = buf, *nzp = newbuf;
247 	zfsvfs_t *zfsvfs;
248 	vnode_t *vp;
249 
250 	/*
251 	 * The znode is on the file system's list of known znodes if the vfs
252 	 * pointer is valid. We set the low bit of the vfs pointer when freeing
253 	 * the znode to invalidate it, and the memory patterns written by kmem
254 	 * (baddcafe and deadbeef) set at least one of the two low bits. A newly
255 	 * created znode sets the vfs pointer last of all to indicate that the
256 	 * znode is known and in a valid state to be moved by this function.
257 	 */
258 	zfsvfs = ozp->z_zfsvfs;
259 	if (!POINTER_IS_VALID(zfsvfs)) {
260 		ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_invalid);
261 		return (KMEM_CBRC_DONT_KNOW);
262 	}
263 
264 	/*
265 	 * Close a small window in which it's possible that the filesystem could
266 	 * be unmounted and freed, and zfsvfs, though valid in the previous
267 	 * statement, could point to unrelated memory by the time we try to
268 	 * prevent the filesystem from being unmounted.
269 	 */
270 	rw_enter(&zfsvfs_lock, RW_WRITER);
271 	if (zfsvfs != ozp->z_zfsvfs) {
272 		rw_exit(&zfsvfs_lock);
273 		ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_recheck1);
274 		return (KMEM_CBRC_DONT_KNOW);
275 	}
276 
277 	/*
278 	 * If the znode is still valid, then so is the file system. We know that
279 	 * no valid file system can be freed while we hold zfsvfs_lock, so we
280 	 * can safely ensure that the filesystem is not and will not be
281 	 * unmounted. The next statement is equivalent to ZFS_ENTER().
282 	 */
283 	rrm_enter(&zfsvfs->z_teardown_lock, RW_READER, FTAG);
284 	if (zfsvfs->z_unmounted) {
285 		ZFS_EXIT(zfsvfs);
286 		rw_exit(&zfsvfs_lock);
287 		ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_unmounted);
288 		return (KMEM_CBRC_DONT_KNOW);
289 	}
290 	rw_exit(&zfsvfs_lock);
291 
292 	mutex_enter(&zfsvfs->z_znodes_lock);
293 	/*
294 	 * Recheck the vfs pointer in case the znode was removed just before
295 	 * acquiring the lock.
296 	 */
297 	if (zfsvfs != ozp->z_zfsvfs) {
298 		mutex_exit(&zfsvfs->z_znodes_lock);
299 		ZFS_EXIT(zfsvfs);
300 		ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_recheck2);
301 		return (KMEM_CBRC_DONT_KNOW);
302 	}
303 
304 	/*
305 	 * At this point we know that as long as we hold z_znodes_lock, the
306 	 * znode cannot be freed and fields within the znode can be safely
307 	 * accessed. Now, prevent a race with zfs_zget().
308 	 */
309 	if (ZFS_OBJ_HOLD_TRYENTER(zfsvfs, ozp->z_id) == 0) {
310 		mutex_exit(&zfsvfs->z_znodes_lock);
311 		ZFS_EXIT(zfsvfs);
312 		ZNODE_STAT_ADD(znode_move_stats.zms_obj_held);
313 		return (KMEM_CBRC_LATER);
314 	}
315 
316 	vp = ZTOV(ozp);
317 	if (mutex_tryenter(&vp->v_lock) == 0) {
318 		ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id);
319 		mutex_exit(&zfsvfs->z_znodes_lock);
320 		ZFS_EXIT(zfsvfs);
321 		ZNODE_STAT_ADD(znode_move_stats.zms_vnode_locked);
322 		return (KMEM_CBRC_LATER);
323 	}
324 
325 	/* Only move znodes that are referenced _only_ by the DNLC. */
326 	if (vp->v_count != 1 || !vn_in_dnlc(vp)) {
327 		mutex_exit(&vp->v_lock);
328 		ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id);
329 		mutex_exit(&zfsvfs->z_znodes_lock);
330 		ZFS_EXIT(zfsvfs);
331 		ZNODE_STAT_ADD(znode_move_stats.zms_not_only_dnlc);
332 		return (KMEM_CBRC_LATER);
333 	}
334 
335 	/*
336 	 * The znode is known and in a valid state to move. We're holding the
337 	 * locks needed to execute the critical section.
338 	 */
339 	zfs_znode_move_impl(ozp, nzp);
340 	mutex_exit(&vp->v_lock);
341 	ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id);
342 
343 	list_link_replace(&ozp->z_link_node, &nzp->z_link_node);
344 	mutex_exit(&zfsvfs->z_znodes_lock);
345 	ZFS_EXIT(zfsvfs);
346 
347 	return (KMEM_CBRC_YES);
348 }
349 
350 void
zfs_znode_init(void)351 zfs_znode_init(void)
352 {
353 	/*
354 	 * Initialize zcache
355 	 */
356 	rw_init(&zfsvfs_lock, NULL, RW_DEFAULT, NULL);
357 	ASSERT(znode_cache == NULL);
358 	znode_cache = kmem_cache_create("zfs_znode_cache",
359 	    sizeof (znode_t), 0, zfs_znode_cache_constructor,
360 	    zfs_znode_cache_destructor, NULL, NULL, NULL, 0);
361 	kmem_cache_set_move(znode_cache, zfs_znode_move);
362 }
363 
364 void
zfs_znode_fini(void)365 zfs_znode_fini(void)
366 {
367 	/*
368 	 * Cleanup vfs & vnode ops
369 	 */
370 	zfs_remove_op_tables();
371 
372 	/*
373 	 * Cleanup zcache
374 	 */
375 	if (znode_cache)
376 		kmem_cache_destroy(znode_cache);
377 	znode_cache = NULL;
378 	rw_destroy(&zfsvfs_lock);
379 }
380 
381 struct vnodeops *zfs_dvnodeops;
382 struct vnodeops *zfs_fvnodeops;
383 struct vnodeops *zfs_symvnodeops;
384 struct vnodeops *zfs_xdvnodeops;
385 struct vnodeops *zfs_evnodeops;
386 struct vnodeops *zfs_sharevnodeops;
387 
388 void
zfs_remove_op_tables()389 zfs_remove_op_tables()
390 {
391 	/*
392 	 * Remove vfs ops
393 	 */
394 	ASSERT(zfsfstype);
395 	(void) vfs_freevfsops_by_type(zfsfstype);
396 	zfsfstype = 0;
397 
398 	/*
399 	 * Remove vnode ops
400 	 */
401 	if (zfs_dvnodeops)
402 		vn_freevnodeops(zfs_dvnodeops);
403 	if (zfs_fvnodeops)
404 		vn_freevnodeops(zfs_fvnodeops);
405 	if (zfs_symvnodeops)
406 		vn_freevnodeops(zfs_symvnodeops);
407 	if (zfs_xdvnodeops)
408 		vn_freevnodeops(zfs_xdvnodeops);
409 	if (zfs_evnodeops)
410 		vn_freevnodeops(zfs_evnodeops);
411 	if (zfs_sharevnodeops)
412 		vn_freevnodeops(zfs_sharevnodeops);
413 
414 	zfs_dvnodeops = NULL;
415 	zfs_fvnodeops = NULL;
416 	zfs_symvnodeops = NULL;
417 	zfs_xdvnodeops = NULL;
418 	zfs_evnodeops = NULL;
419 	zfs_sharevnodeops = NULL;
420 }
421 
422 extern const fs_operation_def_t zfs_dvnodeops_template[];
423 extern const fs_operation_def_t zfs_fvnodeops_template[];
424 extern const fs_operation_def_t zfs_xdvnodeops_template[];
425 extern const fs_operation_def_t zfs_symvnodeops_template[];
426 extern const fs_operation_def_t zfs_evnodeops_template[];
427 extern const fs_operation_def_t zfs_sharevnodeops_template[];
428 
429 int
zfs_create_op_tables()430 zfs_create_op_tables()
431 {
432 	int error;
433 
434 	/*
435 	 * zfs_dvnodeops can be set if mod_remove() calls mod_installfs()
436 	 * due to a failure to remove the the 2nd modlinkage (zfs_modldrv).
437 	 * In this case we just return as the ops vectors are already set up.
438 	 */
439 	if (zfs_dvnodeops)
440 		return (0);
441 
442 	error = vn_make_ops(MNTTYPE_ZFS, zfs_dvnodeops_template,
443 	    &zfs_dvnodeops);
444 	if (error)
445 		return (error);
446 
447 	error = vn_make_ops(MNTTYPE_ZFS, zfs_fvnodeops_template,
448 	    &zfs_fvnodeops);
449 	if (error)
450 		return (error);
451 
452 	error = vn_make_ops(MNTTYPE_ZFS, zfs_symvnodeops_template,
453 	    &zfs_symvnodeops);
454 	if (error)
455 		return (error);
456 
457 	error = vn_make_ops(MNTTYPE_ZFS, zfs_xdvnodeops_template,
458 	    &zfs_xdvnodeops);
459 	if (error)
460 		return (error);
461 
462 	error = vn_make_ops(MNTTYPE_ZFS, zfs_evnodeops_template,
463 	    &zfs_evnodeops);
464 	if (error)
465 		return (error);
466 
467 	error = vn_make_ops(MNTTYPE_ZFS, zfs_sharevnodeops_template,
468 	    &zfs_sharevnodeops);
469 
470 	return (error);
471 }
472 
473 int
zfs_create_share_dir(zfsvfs_t * zfsvfs,dmu_tx_t * tx)474 zfs_create_share_dir(zfsvfs_t *zfsvfs, dmu_tx_t *tx)
475 {
476 	zfs_acl_ids_t acl_ids;
477 	vattr_t vattr;
478 	znode_t *sharezp;
479 	vnode_t *vp;
480 	znode_t *zp;
481 	int error;
482 
483 	vattr.va_mask = AT_MODE|AT_UID|AT_GID|AT_TYPE;
484 	vattr.va_type = VDIR;
485 	vattr.va_mode = S_IFDIR|0555;
486 	vattr.va_uid = crgetuid(kcred);
487 	vattr.va_gid = crgetgid(kcred);
488 
489 	sharezp = kmem_cache_alloc(znode_cache, KM_SLEEP);
490 	ASSERT(!POINTER_IS_VALID(sharezp->z_zfsvfs));
491 	sharezp->z_moved = 0;
492 	sharezp->z_unlinked = 0;
493 	sharezp->z_atime_dirty = 0;
494 	sharezp->z_zfsvfs = zfsvfs;
495 	sharezp->z_is_sa = zfsvfs->z_use_sa;
496 
497 	vp = ZTOV(sharezp);
498 	vn_reinit(vp);
499 	vp->v_type = VDIR;
500 
501 	VERIFY(0 == zfs_acl_ids_create(sharezp, IS_ROOT_NODE, &vattr,
502 	    kcred, NULL, &acl_ids));
503 	zfs_mknode(sharezp, &vattr, tx, kcred, IS_ROOT_NODE, &zp, &acl_ids);
504 	ASSERT3P(zp, ==, sharezp);
505 	ASSERT(!vn_in_dnlc(ZTOV(sharezp))); /* not valid to move */
506 	POINTER_INVALIDATE(&sharezp->z_zfsvfs);
507 	error = zap_add(zfsvfs->z_os, MASTER_NODE_OBJ,
508 	    ZFS_SHARES_DIR, 8, 1, &sharezp->z_id, tx);
509 	zfsvfs->z_shares_dir = sharezp->z_id;
510 
511 	zfs_acl_ids_free(&acl_ids);
512 	ZTOV(sharezp)->v_count = 0;
513 	sa_handle_destroy(sharezp->z_sa_hdl);
514 	kmem_cache_free(znode_cache, sharezp);
515 
516 	return (error);
517 }
518 
519 /*
520  * define a couple of values we need available
521  * for both 64 and 32 bit environments.
522  */
523 #ifndef NBITSMINOR64
524 #define	NBITSMINOR64	32
525 #endif
526 #ifndef MAXMAJ64
527 #define	MAXMAJ64	0xffffffffUL
528 #endif
529 #ifndef	MAXMIN64
530 #define	MAXMIN64	0xffffffffUL
531 #endif
532 
533 /*
534  * Create special expldev for ZFS private use.
535  * Can't use standard expldev since it doesn't do
536  * what we want.  The standard expldev() takes a
537  * dev32_t in LP64 and expands it to a long dev_t.
538  * We need an interface that takes a dev32_t in ILP32
539  * and expands it to a long dev_t.
540  */
541 static uint64_t
zfs_expldev(dev_t dev)542 zfs_expldev(dev_t dev)
543 {
544 #ifndef _LP64
545 	major_t major = (major_t)dev >> NBITSMINOR32 & MAXMAJ32;
546 	return (((uint64_t)major << NBITSMINOR64) |
547 	    ((minor_t)dev & MAXMIN32));
548 #else
549 	return (dev);
550 #endif
551 }
552 
553 /*
554  * Special cmpldev for ZFS private use.
555  * Can't use standard cmpldev since it takes
556  * a long dev_t and compresses it to dev32_t in
557  * LP64.  We need to do a compaction of a long dev_t
558  * to a dev32_t in ILP32.
559  */
560 dev_t
zfs_cmpldev(uint64_t dev)561 zfs_cmpldev(uint64_t dev)
562 {
563 #ifndef _LP64
564 	minor_t minor = (minor_t)dev & MAXMIN64;
565 	major_t major = (major_t)(dev >> NBITSMINOR64) & MAXMAJ64;
566 
567 	if (major > MAXMAJ32 || minor > MAXMIN32)
568 		return (NODEV32);
569 
570 	return (((dev32_t)major << NBITSMINOR32) | minor);
571 #else
572 	return (dev);
573 #endif
574 }
575 
576 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)577 zfs_znode_sa_init(zfsvfs_t *zfsvfs, znode_t *zp,
578     dmu_buf_t *db, dmu_object_type_t obj_type, sa_handle_t *sa_hdl)
579 {
580 	ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs) || (zfsvfs == zp->z_zfsvfs));
581 	ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zfsvfs, zp->z_id)));
582 
583 	mutex_enter(&zp->z_lock);
584 
585 	ASSERT(zp->z_sa_hdl == NULL);
586 	ASSERT(zp->z_acl_cached == NULL);
587 	if (sa_hdl == NULL) {
588 		VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, zp,
589 		    SA_HDL_SHARED, &zp->z_sa_hdl));
590 	} else {
591 		zp->z_sa_hdl = sa_hdl;
592 		sa_set_userp(sa_hdl, zp);
593 	}
594 
595 	zp->z_is_sa = (obj_type == DMU_OT_SA) ? B_TRUE : B_FALSE;
596 
597 	/*
598 	 * Slap on VROOT if we are the root znode
599 	 */
600 	if (zp->z_id == zfsvfs->z_root)
601 		ZTOV(zp)->v_flag |= VROOT;
602 
603 	mutex_exit(&zp->z_lock);
604 	vn_exists(ZTOV(zp));
605 }
606 
607 void
zfs_znode_dmu_fini(znode_t * zp)608 zfs_znode_dmu_fini(znode_t *zp)
609 {
610 	ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zp->z_zfsvfs, zp->z_id)) ||
611 	    zp->z_unlinked ||
612 	    RW_WRITE_HELD(&zp->z_zfsvfs->z_teardown_inactive_lock));
613 
614 	sa_handle_destroy(zp->z_sa_hdl);
615 	zp->z_sa_hdl = NULL;
616 }
617 
618 /*
619  * Construct a new znode/vnode and intialize.
620  *
621  * This does not do a call to dmu_set_user() that is
622  * up to the caller to do, in case you don't want to
623  * return the znode
624  */
625 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)626 zfs_znode_alloc(zfsvfs_t *zfsvfs, dmu_buf_t *db, int blksz,
627     dmu_object_type_t obj_type, sa_handle_t *hdl)
628 {
629 	znode_t	*zp;
630 	vnode_t *vp;
631 	uint64_t mode;
632 	uint64_t parent;
633 	sa_bulk_attr_t bulk[9];
634 	int count = 0;
635 
636 	zp = kmem_cache_alloc(znode_cache, KM_SLEEP);
637 
638 	ASSERT(zp->z_dirlocks == NULL);
639 	ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs));
640 	zp->z_moved = 0;
641 
642 	/*
643 	 * Defer setting z_zfsvfs until the znode is ready to be a candidate for
644 	 * the zfs_znode_move() callback.
645 	 */
646 	zp->z_sa_hdl = NULL;
647 	zp->z_unlinked = 0;
648 	zp->z_atime_dirty = 0;
649 	zp->z_mapcnt = 0;
650 	zp->z_id = db->db_object;
651 	zp->z_blksz = blksz;
652 	zp->z_seq = 0x7A4653;
653 	zp->z_sync_cnt = 0;
654 	zp->z_new_content = 0;
655 
656 	vp = ZTOV(zp);
657 	vn_reinit(vp);
658 
659 	zfs_znode_sa_init(zfsvfs, zp, db, obj_type, hdl);
660 
661 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8);
662 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL, &zp->z_gen, 8);
663 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
664 	    &zp->z_size, 8);
665 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL,
666 	    &zp->z_links, 8);
667 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
668 	    &zp->z_pflags, 8);
669 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL, &parent, 8);
670 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
671 	    &zp->z_atime, 16);
672 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
673 	    &zp->z_uid, 8);
674 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL,
675 	    &zp->z_gid, 8);
676 
677 	if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count) != 0 || zp->z_gen == 0) {
678 		if (hdl == NULL)
679 			sa_handle_destroy(zp->z_sa_hdl);
680 		kmem_cache_free(znode_cache, zp);
681 		return (NULL);
682 	}
683 
684 	zp->z_mode = mode;
685 	vp->v_vfsp = zfsvfs->z_parent->z_vfs;
686 
687 	vp->v_type = IFTOVT((mode_t)mode);
688 
689 	switch (vp->v_type) {
690 	case VDIR:
691 		if (zp->z_pflags & ZFS_XATTR) {
692 			vn_setops(vp, zfs_xdvnodeops);
693 			vp->v_flag |= V_XATTRDIR;
694 		} else {
695 			vn_setops(vp, zfs_dvnodeops);
696 		}
697 		zp->z_zn_prefetch = B_TRUE; /* z_prefetch default is enabled */
698 		break;
699 	case VBLK:
700 	case VCHR:
701 		{
702 			uint64_t rdev;
703 			VERIFY(sa_lookup(zp->z_sa_hdl, SA_ZPL_RDEV(zfsvfs),
704 			    &rdev, sizeof (rdev)) == 0);
705 
706 			vp->v_rdev = zfs_cmpldev(rdev);
707 		}
708 		/*FALLTHROUGH*/
709 	case VFIFO:
710 	case VSOCK:
711 	case VDOOR:
712 		vn_setops(vp, zfs_fvnodeops);
713 		break;
714 	case VREG:
715 		vp->v_flag |= VMODSORT;
716 		if (parent == zfsvfs->z_shares_dir) {
717 			ASSERT(zp->z_uid == 0 && zp->z_gid == 0);
718 			vn_setops(vp, zfs_sharevnodeops);
719 		} else {
720 			vn_setops(vp, zfs_fvnodeops);
721 		}
722 		break;
723 	case VLNK:
724 		vn_setops(vp, zfs_symvnodeops);
725 		break;
726 	default:
727 		vn_setops(vp, zfs_evnodeops);
728 		break;
729 	}
730 
731 	mutex_enter(&zfsvfs->z_znodes_lock);
732 	list_insert_tail(&zfsvfs->z_all_znodes, zp);
733 	membar_producer();
734 	/*
735 	 * Everything else must be valid before assigning z_zfsvfs makes the
736 	 * znode eligible for zfs_znode_move().
737 	 */
738 	zp->z_zfsvfs = zfsvfs;
739 	mutex_exit(&zfsvfs->z_znodes_lock);
740 
741 	VFS_HOLD(zfsvfs->z_vfs);
742 	return (zp);
743 }
744 
745 static uint64_t empty_xattr;
746 static uint64_t pad[4];
747 static zfs_acl_phys_t acl_phys;
748 /*
749  * Create a new DMU object to hold a zfs znode.
750  *
751  *	IN:	dzp	- parent directory for new znode
752  *		vap	- file attributes for new znode
753  *		tx	- dmu transaction id for zap operations
754  *		cr	- credentials of caller
755  *		flag	- flags:
756  *			  IS_ROOT_NODE	- new object will be root
757  *			  IS_XATTR	- new object is an attribute
758  *		bonuslen - length of bonus buffer
759  *		setaclp  - File/Dir initial ACL
760  *		fuidp	 - Tracks fuid allocation.
761  *
762  *	OUT:	zpp	- allocated znode
763  *
764  */
765 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)766 zfs_mknode(znode_t *dzp, vattr_t *vap, dmu_tx_t *tx, cred_t *cr,
767     uint_t flag, znode_t **zpp, zfs_acl_ids_t *acl_ids)
768 {
769 	uint64_t	crtime[2], atime[2], mtime[2], ctime[2];
770 	uint64_t	mode, size, links, parent, pflags;
771 	uint64_t	dzp_pflags = 0;
772 	uint64_t	rdev = 0;
773 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
774 	dmu_buf_t	*db;
775 	timestruc_t	now;
776 	uint64_t	gen, obj;
777 	int		bonuslen;
778 	sa_handle_t	*sa_hdl;
779 	dmu_object_type_t obj_type;
780 	sa_bulk_attr_t	sa_attrs[ZPL_END];
781 	int		cnt = 0;
782 	zfs_acl_locator_cb_t locate = { 0 };
783 
784 	ASSERT(vap && (vap->va_mask & (AT_TYPE|AT_MODE)) == (AT_TYPE|AT_MODE));
785 
786 	if (zfsvfs->z_replay) {
787 		obj = vap->va_nodeid;
788 		now = vap->va_ctime;		/* see zfs_replay_create() */
789 		gen = vap->va_nblocks;		/* ditto */
790 	} else {
791 		obj = 0;
792 		gethrestime(&now);
793 		gen = dmu_tx_get_txg(tx);
794 	}
795 
796 	obj_type = zfsvfs->z_use_sa ? DMU_OT_SA : DMU_OT_ZNODE;
797 	bonuslen = (obj_type == DMU_OT_SA) ?
798 	    DN_MAX_BONUSLEN : ZFS_OLD_ZNODE_PHYS_SIZE;
799 
800 	/*
801 	 * Create a new DMU object.
802 	 */
803 	/*
804 	 * There's currently no mechanism for pre-reading the blocks that will
805 	 * be needed to allocate a new object, so we accept the small chance
806 	 * that there will be an i/o error and we will fail one of the
807 	 * assertions below.
808 	 */
809 	if (vap->va_type == VDIR) {
810 		if (zfsvfs->z_replay) {
811 			VERIFY0(zap_create_claim_norm(zfsvfs->z_os, obj,
812 			    zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS,
813 			    obj_type, bonuslen, tx));
814 		} else {
815 			obj = zap_create_norm(zfsvfs->z_os,
816 			    zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS,
817 			    obj_type, bonuslen, tx);
818 		}
819 	} else {
820 		if (zfsvfs->z_replay) {
821 			VERIFY0(dmu_object_claim(zfsvfs->z_os, obj,
822 			    DMU_OT_PLAIN_FILE_CONTENTS, 0,
823 			    obj_type, bonuslen, tx));
824 		} else {
825 			obj = dmu_object_alloc(zfsvfs->z_os,
826 			    DMU_OT_PLAIN_FILE_CONTENTS, 0,
827 			    obj_type, bonuslen, tx);
828 		}
829 	}
830 
831 	ZFS_OBJ_HOLD_ENTER(zfsvfs, obj);
832 	VERIFY(0 == sa_buf_hold(zfsvfs->z_os, obj, NULL, &db));
833 
834 	/*
835 	 * If this is the root, fix up the half-initialized parent pointer
836 	 * to reference the just-allocated physical data area.
837 	 */
838 	if (flag & IS_ROOT_NODE) {
839 		dzp->z_id = obj;
840 	} else {
841 		dzp_pflags = dzp->z_pflags;
842 	}
843 
844 	/*
845 	 * If parent is an xattr, so am I.
846 	 */
847 	if (dzp_pflags & ZFS_XATTR) {
848 		flag |= IS_XATTR;
849 	}
850 
851 	if (zfsvfs->z_use_fuids)
852 		pflags = ZFS_ARCHIVE | ZFS_AV_MODIFIED;
853 	else
854 		pflags = 0;
855 
856 	if (vap->va_type == VDIR) {
857 		size = 2;		/* contents ("." and "..") */
858 		links = (flag & (IS_ROOT_NODE | IS_XATTR)) ? 2 : 1;
859 	} else {
860 		size = links = 0;
861 	}
862 
863 	if (vap->va_type == VBLK || vap->va_type == VCHR) {
864 		rdev = zfs_expldev(vap->va_rdev);
865 	}
866 
867 	parent = dzp->z_id;
868 	mode = acl_ids->z_mode;
869 	if (flag & IS_XATTR)
870 		pflags |= ZFS_XATTR;
871 
872 	/*
873 	 * No execs denied will be deterimed when zfs_mode_compute() is called.
874 	 */
875 	pflags |= acl_ids->z_aclp->z_hints &
876 	    (ZFS_ACL_TRIVIAL|ZFS_INHERIT_ACE|ZFS_ACL_AUTO_INHERIT|
877 	    ZFS_ACL_DEFAULTED|ZFS_ACL_PROTECTED);
878 
879 	ZFS_TIME_ENCODE(&now, crtime);
880 	ZFS_TIME_ENCODE(&now, ctime);
881 
882 	if (vap->va_mask & AT_ATIME) {
883 		ZFS_TIME_ENCODE(&vap->va_atime, atime);
884 	} else {
885 		ZFS_TIME_ENCODE(&now, atime);
886 	}
887 
888 	if (vap->va_mask & AT_MTIME) {
889 		ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
890 	} else {
891 		ZFS_TIME_ENCODE(&now, mtime);
892 	}
893 
894 	/* Now add in all of the "SA" attributes */
895 	VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, NULL, SA_HDL_SHARED,
896 	    &sa_hdl));
897 
898 	/*
899 	 * Setup the array of attributes to be replaced/set on the new file
900 	 *
901 	 * order for  DMU_OT_ZNODE is critical since it needs to be constructed
902 	 * in the old znode_phys_t format.  Don't change this ordering
903 	 */
904 
905 	if (obj_type == DMU_OT_ZNODE) {
906 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs),
907 		    NULL, &atime, 16);
908 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs),
909 		    NULL, &mtime, 16);
910 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs),
911 		    NULL, &ctime, 16);
912 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs),
913 		    NULL, &crtime, 16);
914 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs),
915 		    NULL, &gen, 8);
916 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs),
917 		    NULL, &mode, 8);
918 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs),
919 		    NULL, &size, 8);
920 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs),
921 		    NULL, &parent, 8);
922 	} else {
923 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs),
924 		    NULL, &mode, 8);
925 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs),
926 		    NULL, &size, 8);
927 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs),
928 		    NULL, &gen, 8);
929 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs), NULL,
930 		    &acl_ids->z_fuid, 8);
931 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs), NULL,
932 		    &acl_ids->z_fgid, 8);
933 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs),
934 		    NULL, &parent, 8);
935 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs),
936 		    NULL, &pflags, 8);
937 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs),
938 		    NULL, &atime, 16);
939 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs),
940 		    NULL, &mtime, 16);
941 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs),
942 		    NULL, &ctime, 16);
943 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs),
944 		    NULL, &crtime, 16);
945 	}
946 
947 	SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_LINKS(zfsvfs), NULL, &links, 8);
948 
949 	if (obj_type == DMU_OT_ZNODE) {
950 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_XATTR(zfsvfs), NULL,
951 		    &empty_xattr, 8);
952 	}
953 	if (obj_type == DMU_OT_ZNODE ||
954 	    (vap->va_type == VBLK || vap->va_type == VCHR)) {
955 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_RDEV(zfsvfs),
956 		    NULL, &rdev, 8);
957 
958 	}
959 	if (obj_type == DMU_OT_ZNODE) {
960 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs),
961 		    NULL, &pflags, 8);
962 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs), NULL,
963 		    &acl_ids->z_fuid, 8);
964 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs), NULL,
965 		    &acl_ids->z_fgid, 8);
966 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PAD(zfsvfs), NULL, pad,
967 		    sizeof (uint64_t) * 4);
968 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ZNODE_ACL(zfsvfs), NULL,
969 		    &acl_phys, sizeof (zfs_acl_phys_t));
970 	} else if (acl_ids->z_aclp->z_version >= ZFS_ACL_VERSION_FUID) {
971 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_COUNT(zfsvfs), NULL,
972 		    &acl_ids->z_aclp->z_acl_count, 8);
973 		locate.cb_aclp = acl_ids->z_aclp;
974 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_ACES(zfsvfs),
975 		    zfs_acl_data_locator, &locate,
976 		    acl_ids->z_aclp->z_acl_bytes);
977 		mode = zfs_mode_compute(mode, acl_ids->z_aclp, &pflags,
978 		    acl_ids->z_fuid, acl_ids->z_fgid);
979 	}
980 
981 	VERIFY(sa_replace_all_by_template(sa_hdl, sa_attrs, cnt, tx) == 0);
982 
983 	if (!(flag & IS_ROOT_NODE)) {
984 		*zpp = zfs_znode_alloc(zfsvfs, db, 0, obj_type, sa_hdl);
985 		ASSERT(*zpp != NULL);
986 	} else {
987 		/*
988 		 * If we are creating the root node, the "parent" we
989 		 * passed in is the znode for the root.
990 		 */
991 		*zpp = dzp;
992 
993 		(*zpp)->z_sa_hdl = sa_hdl;
994 	}
995 
996 	(*zpp)->z_pflags = pflags;
997 	(*zpp)->z_mode = mode;
998 
999 	if (vap->va_mask & AT_XVATTR)
1000 		zfs_xvattr_set(*zpp, (xvattr_t *)vap, tx);
1001 
1002 	if (obj_type == DMU_OT_ZNODE ||
1003 	    acl_ids->z_aclp->z_version < ZFS_ACL_VERSION_FUID) {
1004 		VERIFY0(zfs_aclset_common(*zpp, acl_ids->z_aclp, cr, tx));
1005 	}
1006 	ZFS_OBJ_HOLD_EXIT(zfsvfs, obj);
1007 }
1008 
1009 /*
1010  * Update in-core attributes.  It is assumed the caller will be doing an
1011  * sa_bulk_update to push the changes out.
1012  */
1013 void
zfs_xvattr_set(znode_t * zp,xvattr_t * xvap,dmu_tx_t * tx)1014 zfs_xvattr_set(znode_t *zp, xvattr_t *xvap, dmu_tx_t *tx)
1015 {
1016 	xoptattr_t *xoap;
1017 
1018 	xoap = xva_getxoptattr(xvap);
1019 	ASSERT(xoap);
1020 
1021 	if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
1022 		uint64_t times[2];
1023 		ZFS_TIME_ENCODE(&xoap->xoa_createtime, times);
1024 		(void) sa_update(zp->z_sa_hdl, SA_ZPL_CRTIME(zp->z_zfsvfs),
1025 		    &times, sizeof (times), tx);
1026 		XVA_SET_RTN(xvap, XAT_CREATETIME);
1027 	}
1028 	if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
1029 		ZFS_ATTR_SET(zp, ZFS_READONLY, xoap->xoa_readonly,
1030 		    zp->z_pflags, tx);
1031 		XVA_SET_RTN(xvap, XAT_READONLY);
1032 	}
1033 	if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
1034 		ZFS_ATTR_SET(zp, ZFS_HIDDEN, xoap->xoa_hidden,
1035 		    zp->z_pflags, tx);
1036 		XVA_SET_RTN(xvap, XAT_HIDDEN);
1037 	}
1038 	if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
1039 		ZFS_ATTR_SET(zp, ZFS_SYSTEM, xoap->xoa_system,
1040 		    zp->z_pflags, tx);
1041 		XVA_SET_RTN(xvap, XAT_SYSTEM);
1042 	}
1043 	if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
1044 		ZFS_ATTR_SET(zp, ZFS_ARCHIVE, xoap->xoa_archive,
1045 		    zp->z_pflags, tx);
1046 		XVA_SET_RTN(xvap, XAT_ARCHIVE);
1047 	}
1048 	if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
1049 		ZFS_ATTR_SET(zp, ZFS_IMMUTABLE, xoap->xoa_immutable,
1050 		    zp->z_pflags, tx);
1051 		XVA_SET_RTN(xvap, XAT_IMMUTABLE);
1052 	}
1053 	if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
1054 		ZFS_ATTR_SET(zp, ZFS_NOUNLINK, xoap->xoa_nounlink,
1055 		    zp->z_pflags, tx);
1056 		XVA_SET_RTN(xvap, XAT_NOUNLINK);
1057 	}
1058 	if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
1059 		ZFS_ATTR_SET(zp, ZFS_APPENDONLY, xoap->xoa_appendonly,
1060 		    zp->z_pflags, tx);
1061 		XVA_SET_RTN(xvap, XAT_APPENDONLY);
1062 	}
1063 	if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
1064 		ZFS_ATTR_SET(zp, ZFS_NODUMP, xoap->xoa_nodump,
1065 		    zp->z_pflags, tx);
1066 		XVA_SET_RTN(xvap, XAT_NODUMP);
1067 	}
1068 	if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
1069 		ZFS_ATTR_SET(zp, ZFS_OPAQUE, xoap->xoa_opaque,
1070 		    zp->z_pflags, tx);
1071 		XVA_SET_RTN(xvap, XAT_OPAQUE);
1072 	}
1073 	if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
1074 		ZFS_ATTR_SET(zp, ZFS_AV_QUARANTINED,
1075 		    xoap->xoa_av_quarantined, zp->z_pflags, tx);
1076 		XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
1077 	}
1078 	if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
1079 		ZFS_ATTR_SET(zp, ZFS_AV_MODIFIED, xoap->xoa_av_modified,
1080 		    zp->z_pflags, tx);
1081 		XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
1082 	}
1083 	if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) {
1084 		zfs_sa_set_scanstamp(zp, xvap, tx);
1085 		XVA_SET_RTN(xvap, XAT_AV_SCANSTAMP);
1086 	}
1087 	if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
1088 		ZFS_ATTR_SET(zp, ZFS_REPARSE, xoap->xoa_reparse,
1089 		    zp->z_pflags, tx);
1090 		XVA_SET_RTN(xvap, XAT_REPARSE);
1091 	}
1092 	if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
1093 		ZFS_ATTR_SET(zp, ZFS_OFFLINE, xoap->xoa_offline,
1094 		    zp->z_pflags, tx);
1095 		XVA_SET_RTN(xvap, XAT_OFFLINE);
1096 	}
1097 	if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
1098 		ZFS_ATTR_SET(zp, ZFS_SPARSE, xoap->xoa_sparse,
1099 		    zp->z_pflags, tx);
1100 		XVA_SET_RTN(xvap, XAT_SPARSE);
1101 	}
1102 }
1103 
1104 int
zfs_zget(zfsvfs_t * zfsvfs,uint64_t obj_num,znode_t ** zpp)1105 zfs_zget(zfsvfs_t *zfsvfs, uint64_t obj_num, znode_t **zpp)
1106 {
1107 	dmu_object_info_t doi;
1108 	dmu_buf_t	*db;
1109 	znode_t		*zp;
1110 	int err;
1111 	sa_handle_t	*hdl;
1112 
1113 	*zpp = NULL;
1114 
1115 	ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num);
1116 
1117 	err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db);
1118 	if (err) {
1119 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1120 		return (err);
1121 	}
1122 
1123 	dmu_object_info_from_db(db, &doi);
1124 	if (doi.doi_bonus_type != DMU_OT_SA &&
1125 	    (doi.doi_bonus_type != DMU_OT_ZNODE ||
1126 	    (doi.doi_bonus_type == DMU_OT_ZNODE &&
1127 	    doi.doi_bonus_size < sizeof (znode_phys_t)))) {
1128 		sa_buf_rele(db, NULL);
1129 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1130 		return (SET_ERROR(EINVAL));
1131 	}
1132 
1133 	hdl = dmu_buf_get_user(db);
1134 	if (hdl != NULL) {
1135 		zp  = sa_get_userdata(hdl);
1136 
1137 
1138 		/*
1139 		 * Since "SA" does immediate eviction we
1140 		 * should never find a sa handle that doesn't
1141 		 * know about the znode.
1142 		 */
1143 
1144 		ASSERT3P(zp, !=, NULL);
1145 
1146 		mutex_enter(&zp->z_lock);
1147 		ASSERT3U(zp->z_id, ==, obj_num);
1148 		if (zp->z_unlinked) {
1149 			err = SET_ERROR(ENOENT);
1150 		} else {
1151 			VN_HOLD(ZTOV(zp));
1152 			*zpp = zp;
1153 			err = 0;
1154 		}
1155 		sa_buf_rele(db, NULL);
1156 		mutex_exit(&zp->z_lock);
1157 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1158 		return (err);
1159 	}
1160 
1161 	/*
1162 	 * Not found create new znode/vnode
1163 	 * but only if file exists.
1164 	 *
1165 	 * There is a small window where zfs_vget() could
1166 	 * find this object while a file create is still in
1167 	 * progress.  This is checked for in zfs_znode_alloc()
1168 	 *
1169 	 * if zfs_znode_alloc() fails it will drop the hold on the
1170 	 * bonus buffer.
1171 	 */
1172 	zp = zfs_znode_alloc(zfsvfs, db, doi.doi_data_block_size,
1173 	    doi.doi_bonus_type, NULL);
1174 	if (zp == NULL) {
1175 		err = SET_ERROR(ENOENT);
1176 	} else {
1177 		if (zp->z_links == 0)
1178 			zp->z_unlinked = B_TRUE;
1179 		*zpp = zp;
1180 	}
1181 	ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1182 	return (err);
1183 }
1184 
1185 int
zfs_rezget(znode_t * zp)1186 zfs_rezget(znode_t *zp)
1187 {
1188 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1189 	dmu_object_info_t doi;
1190 	dmu_buf_t *db;
1191 	uint64_t obj_num = zp->z_id;
1192 	uint64_t mode;
1193 	sa_bulk_attr_t bulk[8];
1194 	int err;
1195 	int count = 0;
1196 	uint64_t gen;
1197 
1198 	ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num);
1199 
1200 	mutex_enter(&zp->z_acl_lock);
1201 	if (zp->z_acl_cached) {
1202 		zfs_acl_free(zp->z_acl_cached);
1203 		zp->z_acl_cached = NULL;
1204 	}
1205 
1206 	mutex_exit(&zp->z_acl_lock);
1207 	ASSERT(zp->z_sa_hdl == NULL);
1208 	err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db);
1209 	if (err) {
1210 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1211 		return (err);
1212 	}
1213 
1214 	dmu_object_info_from_db(db, &doi);
1215 	if (doi.doi_bonus_type != DMU_OT_SA &&
1216 	    (doi.doi_bonus_type != DMU_OT_ZNODE ||
1217 	    (doi.doi_bonus_type == DMU_OT_ZNODE &&
1218 	    doi.doi_bonus_size < sizeof (znode_phys_t)))) {
1219 		sa_buf_rele(db, NULL);
1220 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1221 		return (SET_ERROR(EINVAL));
1222 	}
1223 
1224 	zfs_znode_sa_init(zfsvfs, zp, db, doi.doi_bonus_type, NULL);
1225 
1226 	/* reload cached values */
1227 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL,
1228 	    &gen, sizeof (gen));
1229 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
1230 	    &zp->z_size, sizeof (zp->z_size));
1231 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL,
1232 	    &zp->z_links, sizeof (zp->z_links));
1233 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
1234 	    &zp->z_pflags, sizeof (zp->z_pflags));
1235 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
1236 	    &zp->z_atime, sizeof (zp->z_atime));
1237 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
1238 	    &zp->z_uid, sizeof (zp->z_uid));
1239 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL,
1240 	    &zp->z_gid, sizeof (zp->z_gid));
1241 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
1242 	    &mode, sizeof (mode));
1243 
1244 	if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) {
1245 		zfs_znode_dmu_fini(zp);
1246 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1247 		return (SET_ERROR(EIO));
1248 	}
1249 
1250 	zp->z_mode = mode;
1251 
1252 	if (gen != zp->z_gen) {
1253 		zfs_znode_dmu_fini(zp);
1254 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1255 		return (SET_ERROR(EIO));
1256 	}
1257 
1258 	zp->z_unlinked = (zp->z_links == 0);
1259 	zp->z_blksz = doi.doi_data_block_size;
1260 
1261 	ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1262 
1263 	return (0);
1264 }
1265 
1266 void
zfs_znode_delete(znode_t * zp,dmu_tx_t * tx)1267 zfs_znode_delete(znode_t *zp, dmu_tx_t *tx)
1268 {
1269 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1270 	objset_t *os = zfsvfs->z_os;
1271 	uint64_t obj = zp->z_id;
1272 	uint64_t acl_obj = zfs_external_acl(zp);
1273 
1274 	ZFS_OBJ_HOLD_ENTER(zfsvfs, obj);
1275 	if (acl_obj) {
1276 		VERIFY(!zp->z_is_sa);
1277 		VERIFY(0 == dmu_object_free(os, acl_obj, tx));
1278 	}
1279 	VERIFY(0 == dmu_object_free(os, obj, tx));
1280 	zfs_znode_dmu_fini(zp);
1281 	ZFS_OBJ_HOLD_EXIT(zfsvfs, obj);
1282 	zfs_znode_free(zp);
1283 }
1284 
1285 void
zfs_zinactive(znode_t * zp)1286 zfs_zinactive(znode_t *zp)
1287 {
1288 	vnode_t	*vp = ZTOV(zp);
1289 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1290 	uint64_t z_id = zp->z_id;
1291 
1292 	ASSERT(zp->z_sa_hdl);
1293 
1294 	/*
1295 	 * Don't allow a zfs_zget() while were trying to release this znode
1296 	 */
1297 	ZFS_OBJ_HOLD_ENTER(zfsvfs, z_id);
1298 
1299 	mutex_enter(&zp->z_lock);
1300 	mutex_enter(&vp->v_lock);
1301 	vp->v_count--;
1302 	if (vp->v_count > 0 || vn_has_cached_data(vp)) {
1303 		/*
1304 		 * If the hold count is greater than zero, somebody has
1305 		 * obtained a new reference on this znode while we were
1306 		 * processing it here, so we are done.  If we still have
1307 		 * mapped pages then we are also done, since we don't
1308 		 * want to inactivate the znode until the pages get pushed.
1309 		 *
1310 		 * XXX - if vn_has_cached_data(vp) is true, but count == 0,
1311 		 * this seems like it would leave the znode hanging with
1312 		 * no chance to go inactive...
1313 		 */
1314 		mutex_exit(&vp->v_lock);
1315 		mutex_exit(&zp->z_lock);
1316 		ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id);
1317 		return;
1318 	}
1319 	mutex_exit(&vp->v_lock);
1320 
1321 	/*
1322 	 * If this was the last reference to a file with no links,
1323 	 * remove the file from the file system.
1324 	 */
1325 	if (zp->z_unlinked) {
1326 		mutex_exit(&zp->z_lock);
1327 		ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id);
1328 		zfs_rmnode(zp);
1329 		return;
1330 	}
1331 
1332 	mutex_exit(&zp->z_lock);
1333 	zfs_znode_dmu_fini(zp);
1334 	ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id);
1335 	zfs_znode_free(zp);
1336 }
1337 
1338 void
zfs_znode_free(znode_t * zp)1339 zfs_znode_free(znode_t *zp)
1340 {
1341 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1342 
1343 	vn_invalid(ZTOV(zp));
1344 
1345 	ASSERT(ZTOV(zp)->v_count == 0);
1346 
1347 	mutex_enter(&zfsvfs->z_znodes_lock);
1348 	POINTER_INVALIDATE(&zp->z_zfsvfs);
1349 	list_remove(&zfsvfs->z_all_znodes, zp);
1350 	mutex_exit(&zfsvfs->z_znodes_lock);
1351 
1352 	if (zp->z_acl_cached) {
1353 		zfs_acl_free(zp->z_acl_cached);
1354 		zp->z_acl_cached = NULL;
1355 	}
1356 
1357 	kmem_cache_free(znode_cache, zp);
1358 
1359 	VFS_RELE(zfsvfs->z_vfs);
1360 }
1361 
1362 void
zfs_tstamp_update_setup(znode_t * zp,uint_t flag,uint64_t mtime[2],uint64_t ctime[2],boolean_t have_tx)1363 zfs_tstamp_update_setup(znode_t *zp, uint_t flag, uint64_t mtime[2],
1364     uint64_t ctime[2], boolean_t have_tx)
1365 {
1366 	timestruc_t	now;
1367 
1368 	gethrestime(&now);
1369 
1370 	if (have_tx) {	/* will sa_bulk_update happen really soon? */
1371 		zp->z_atime_dirty = 0;
1372 		zp->z_seq++;
1373 	} else {
1374 		zp->z_atime_dirty = 1;
1375 	}
1376 
1377 	if (flag & AT_ATIME) {
1378 		ZFS_TIME_ENCODE(&now, zp->z_atime);
1379 	}
1380 
1381 	if (flag & AT_MTIME) {
1382 		ZFS_TIME_ENCODE(&now, mtime);
1383 		if (zp->z_zfsvfs->z_use_fuids) {
1384 			zp->z_pflags |= (ZFS_ARCHIVE |
1385 			    ZFS_AV_MODIFIED);
1386 		}
1387 	}
1388 
1389 	if (flag & AT_CTIME) {
1390 		ZFS_TIME_ENCODE(&now, ctime);
1391 		if (zp->z_zfsvfs->z_use_fuids)
1392 			zp->z_pflags |= ZFS_ARCHIVE;
1393 	}
1394 }
1395 
1396 /*
1397  * Grow the block size for a file.
1398  *
1399  *	IN:	zp	- znode of file to free data in.
1400  *		size	- requested block size
1401  *		tx	- open transaction.
1402  *
1403  * NOTE: this function assumes that the znode is write locked.
1404  */
1405 void
zfs_grow_blocksize(znode_t * zp,uint64_t size,dmu_tx_t * tx)1406 zfs_grow_blocksize(znode_t *zp, uint64_t size, dmu_tx_t *tx)
1407 {
1408 	int		error;
1409 	u_longlong_t	dummy;
1410 
1411 	if (size <= zp->z_blksz)
1412 		return;
1413 	/*
1414 	 * If the file size is already greater than the current blocksize,
1415 	 * we will not grow.  If there is more than one block in a file,
1416 	 * the blocksize cannot change.
1417 	 */
1418 	if (zp->z_blksz && zp->z_size > zp->z_blksz)
1419 		return;
1420 
1421 	error = dmu_object_set_blocksize(zp->z_zfsvfs->z_os, zp->z_id,
1422 	    size, 0, tx);
1423 
1424 	if (error == ENOTSUP)
1425 		return;
1426 	ASSERT0(error);
1427 
1428 	/* What blocksize did we actually get? */
1429 	dmu_object_size_from_db(sa_get_db(zp->z_sa_hdl), &zp->z_blksz, &dummy);
1430 }
1431 
1432 /*
1433  * This is a dummy interface used when pvn_vplist_dirty() should *not*
1434  * be calling back into the fs for a putpage().  E.g.: when truncating
1435  * a file, the pages being "thrown away* don't need to be written out.
1436  */
1437 /* ARGSUSED */
1438 static int
zfs_no_putpage(vnode_t * vp,page_t * pp,u_offset_t * offp,size_t * lenp,int flags,cred_t * cr)1439 zfs_no_putpage(vnode_t *vp, page_t *pp, u_offset_t *offp, size_t *lenp,
1440     int flags, cred_t *cr)
1441 {
1442 	ASSERT(0);
1443 	return (0);
1444 }
1445 
1446 /*
1447  * Increase the file length
1448  *
1449  *	IN:	zp	- znode of file to free data in.
1450  *		end	- new end-of-file
1451  *
1452  *	RETURN:	0 on success, error code on failure
1453  *
1454  *	Assumption: File is already range-locked for changing zp_size
1455  */
1456 static int
zfs_extend(znode_t * zp,uint64_t end)1457 zfs_extend(znode_t *zp, uint64_t end)
1458 {
1459 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1460 	dmu_tx_t *tx;
1461 	uint64_t newblksz;
1462 	int error;
1463 
1464 	/*
1465 	 * Nothing to do if file already at desired length.
1466 	 */
1467 	if (end <= zp->z_size)
1468 		return (0);
1469 
1470 	tx = dmu_tx_create(zfsvfs->z_os);
1471 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1472 	zfs_sa_upgrade_txholds(tx, zp);
1473 	if (end > zp->z_blksz &&
1474 	    (!ISP2(zp->z_blksz) || zp->z_blksz < zfsvfs->z_max_blksz)) {
1475 		/*
1476 		 * We are growing the file past the current block size.
1477 		 */
1478 		if (zp->z_blksz > zp->z_zfsvfs->z_max_blksz) {
1479 			/*
1480 			 * File's blocksize is already larger than the
1481 			 * "recordsize" property.  Only let it grow to
1482 			 * the next power of 2.
1483 			 */
1484 			ASSERT(!ISP2(zp->z_blksz));
1485 			newblksz = MIN(end, 1 << highbit64(zp->z_blksz));
1486 		} else {
1487 			newblksz = MIN(end, zp->z_zfsvfs->z_max_blksz);
1488 		}
1489 		dmu_tx_hold_write(tx, zp->z_id, 0, newblksz);
1490 	} else {
1491 		newblksz = 0;
1492 	}
1493 
1494 	error = dmu_tx_assign(tx, TXG_WAIT);
1495 	if (error) {
1496 		dmu_tx_abort(tx);
1497 		return (error);
1498 	}
1499 
1500 	if (newblksz)
1501 		zfs_grow_blocksize(zp, newblksz, tx);
1502 
1503 	zp->z_size = end;
1504 
1505 	VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zp->z_zfsvfs),
1506 	    &zp->z_size, sizeof (zp->z_size), tx));
1507 
1508 	dmu_tx_commit(tx);
1509 
1510 	return (0);
1511 }
1512 
1513 /*
1514  * Free space in a file.
1515  *
1516  *	IN:	zp	- znode of file to free data in.
1517  *		off	- start of section to free.
1518  *		len	- length of section to free.
1519  *
1520  *	RETURN:	0 on success, error code on failure
1521  *
1522  *	Assumption: file is already range-locked for changing zp_size
1523  */
1524 static int
zfs_free_range(znode_t * zp,uint64_t off,uint64_t len)1525 zfs_free_range(znode_t *zp, uint64_t off, uint64_t len)
1526 {
1527 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1528 	int error;
1529 
1530 	/*
1531 	 * Nothing to do if file already at desired length.
1532 	 */
1533 	if (off >= zp->z_size)
1534 		return (0);
1535 
1536 	if (off + len > zp->z_size)
1537 		len = zp->z_size - off;
1538 
1539 	error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, off, len);
1540 
1541 	return (error);
1542 }
1543 
1544 /*
1545  * Truncate a file
1546  *
1547  *	IN:	zp	- znode of file to free data in.
1548  *		end	- new end-of-file.
1549  *
1550  *	RETURN:	0 on success, error code on failure
1551  *
1552  *	Assumption: file is already range-locked for changing zp_size
1553  */
1554 static int
zfs_trunc(znode_t * zp,uint64_t end)1555 zfs_trunc(znode_t *zp, uint64_t end)
1556 {
1557 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1558 	vnode_t *vp = ZTOV(zp);
1559 	dmu_tx_t *tx;
1560 	int error;
1561 	sa_bulk_attr_t bulk[2];
1562 	int count = 0;
1563 
1564 	/*
1565 	 * Nothing to do if file already at desired length.
1566 	 */
1567 	if (end >= zp->z_size)
1568 		return (0);
1569 
1570 	error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, end,  -1);
1571 	if (error)
1572 		return (error);
1573 
1574 	tx = dmu_tx_create(zfsvfs->z_os);
1575 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1576 	zfs_sa_upgrade_txholds(tx, zp);
1577 	dmu_tx_mark_netfree(tx);
1578 	error = dmu_tx_assign(tx, TXG_WAIT);
1579 	if (error) {
1580 		dmu_tx_abort(tx);
1581 		return (error);
1582 	}
1583 
1584 	zp->z_size = end;
1585 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs),
1586 	    NULL, &zp->z_size, sizeof (zp->z_size));
1587 
1588 	if (end == 0) {
1589 		zp->z_pflags &= ~ZFS_SPARSE;
1590 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
1591 		    NULL, &zp->z_pflags, 8);
1592 	}
1593 	VERIFY(sa_bulk_update(zp->z_sa_hdl, bulk, count, tx) == 0);
1594 
1595 	dmu_tx_commit(tx);
1596 
1597 	/*
1598 	 * Clear any mapped pages in the truncated region.  This has to
1599 	 * happen outside of the transaction to avoid the possibility of
1600 	 * a deadlock with someone trying to push a page that we are
1601 	 * about to invalidate.
1602 	 */
1603 	if (vn_has_cached_data(vp)) {
1604 		page_t *pp;
1605 		uint64_t start = end & PAGEMASK;
1606 		int poff = end & PAGEOFFSET;
1607 
1608 		if (poff != 0 && (pp = page_lookup(vp, start, SE_SHARED))) {
1609 			/*
1610 			 * We need to zero a partial page.
1611 			 */
1612 			pagezero(pp, poff, PAGESIZE - poff);
1613 			start += PAGESIZE;
1614 			page_unlock(pp);
1615 		}
1616 		error = pvn_vplist_dirty(vp, start, zfs_no_putpage,
1617 		    B_INVAL | B_TRUNC, NULL);
1618 		ASSERT(error == 0);
1619 	}
1620 
1621 	return (0);
1622 }
1623 
1624 /*
1625  * Free space in a file
1626  *
1627  *	IN:	zp	- znode of file to free data in.
1628  *		off	- start of range
1629  *		len	- end of range (0 => EOF)
1630  *		flag	- current file open mode flags.
1631  *		log	- TRUE if this action should be logged
1632  *
1633  *	RETURN:	0 on success, error code on failure
1634  */
1635 int
zfs_freesp(znode_t * zp,uint64_t off,uint64_t len,int flag,boolean_t log)1636 zfs_freesp(znode_t *zp, uint64_t off, uint64_t len, int flag, boolean_t log)
1637 {
1638 	vnode_t *vp = ZTOV(zp);
1639 	dmu_tx_t *tx;
1640 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1641 	zilog_t *zilog = zfsvfs->z_log;
1642 	uint64_t mode;
1643 	uint64_t mtime[2], ctime[2];
1644 	sa_bulk_attr_t bulk[3];
1645 	int count = 0;
1646 	int error;
1647 	rl_t *rl;
1648 
1649 	if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs), &mode,
1650 	    sizeof (mode))) != 0)
1651 		return (error);
1652 
1653 	if (off > zp->z_size) {
1654 		rl = zfs_range_lock(zp, 0, UINT64_MAX, RL_WRITER);
1655 		error =  zfs_extend(zp, off+len);
1656 		if (error == 0 && log)
1657 			goto log;
1658 
1659 		zfs_range_unlock(rl);
1660 		return (error);
1661 	}
1662 
1663 	/*
1664 	 * Check for any locks in the region to be freed.
1665 	 */
1666 	if (MANDLOCK(vp, (mode_t)mode)) {
1667 		uint64_t length = (len ? len : zp->z_size - off);
1668 		if (error = chklock(vp, FWRITE, off, length, flag, NULL))
1669 			return (error);
1670 	}
1671 
1672 	if (len == 0 || off + len > zp->z_size) {
1673 		rl = zfs_range_lock(zp, 0, UINT64_MAX, RL_WRITER);
1674 	} else {
1675 		ssize_t	lock_off;
1676 		ssize_t lock_len;
1677 #ifdef _KERNEL
1678 		if (rz_zev_active()) {
1679 			/* start of this megabyte */
1680 			lock_off = P2ALIGN(off, ZEV_L1_SIZE);
1681 			/* full megabytes */
1682 			lock_len = len + (off - lock_off);
1683 			lock_len = P2ROUNDUP(lock_len, ZEV_L1_SIZE);
1684 		} else {
1685 			lock_off = off;
1686 			lock_len = len;
1687 		}
1688 #else
1689 		lock_off = off;
1690 		lock_len = len;
1691 #endif
1692 		rl = zfs_range_lock(zp, lock_off, lock_len, RL_WRITER);
1693 	}
1694 
1695 	if (len == 0) {
1696 		error = zfs_trunc(zp, off);
1697 	} else {
1698 		if ((error = zfs_free_range(zp, off, len)) == 0 &&
1699 		    off + len > zp->z_size)
1700 			error = zfs_extend(zp, off+len);
1701 	}
1702 
1703 	if (error || !log) {
1704 		zfs_range_unlock(rl);
1705 		return (error);
1706 	}
1707 log:
1708 	tx = dmu_tx_create(zfsvfs->z_os);
1709 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1710 	zfs_sa_upgrade_txholds(tx, zp);
1711 	error = dmu_tx_assign(tx, TXG_WAIT);
1712 	if (error) {
1713 		dmu_tx_abort(tx);
1714 		zfs_range_unlock(rl);
1715 		return (error);
1716 	}
1717 
1718 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, mtime, 16);
1719 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, ctime, 16);
1720 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
1721 	    NULL, &zp->z_pflags, 8);
1722 	zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, B_TRUE);
1723 	error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
1724 	ASSERT(error == 0);
1725 
1726 	zfs_log_truncate(zilog, tx, TX_TRUNCATE, zp, off, len);
1727 	zfs_range_unlock(rl);
1728 
1729 	dmu_tx_commit(tx);
1730 	return (0);
1731 }
1732 
1733 void
zfs_create_fs(objset_t * os,cred_t * cr,nvlist_t * zplprops,dmu_tx_t * tx)1734 zfs_create_fs(objset_t *os, cred_t *cr, nvlist_t *zplprops, dmu_tx_t *tx)
1735 {
1736 	uint64_t	moid, obj, sa_obj, version;
1737 	uint64_t	sense = ZFS_CASE_SENSITIVE;
1738 	uint64_t	norm = 0;
1739 	nvpair_t	*elem;
1740 	int		error;
1741 	int		i;
1742 	znode_t		*rootzp = NULL;
1743 	zfsvfs_t	*zfsvfs;
1744 	vnode_t		*vp;
1745 	vattr_t		vattr;
1746 	znode_t		*zp;
1747 	zfs_acl_ids_t	acl_ids;
1748 
1749 	/*
1750 	 * First attempt to create master node.
1751 	 */
1752 	/*
1753 	 * In an empty objset, there are no blocks to read and thus
1754 	 * there can be no i/o errors (which we assert below).
1755 	 */
1756 	moid = MASTER_NODE_OBJ;
1757 	error = zap_create_claim(os, moid, DMU_OT_MASTER_NODE,
1758 	    DMU_OT_NONE, 0, tx);
1759 	ASSERT(error == 0);
1760 
1761 	/*
1762 	 * Set starting attributes.
1763 	 */
1764 	version = zfs_zpl_version_map(spa_version(dmu_objset_spa(os)));
1765 	elem = NULL;
1766 	while ((elem = nvlist_next_nvpair(zplprops, elem)) != NULL) {
1767 		/* For the moment we expect all zpl props to be uint64_ts */
1768 		uint64_t val;
1769 		char *name;
1770 
1771 		ASSERT(nvpair_type(elem) == DATA_TYPE_UINT64);
1772 		VERIFY(nvpair_value_uint64(elem, &val) == 0);
1773 		name = nvpair_name(elem);
1774 		if (strcmp(name, zfs_prop_to_name(ZFS_PROP_VERSION)) == 0) {
1775 			if (val < version)
1776 				version = val;
1777 		} else {
1778 			error = zap_update(os, moid, name, 8, 1, &val, tx);
1779 		}
1780 		ASSERT(error == 0);
1781 		if (strcmp(name, zfs_prop_to_name(ZFS_PROP_NORMALIZE)) == 0)
1782 			norm = val;
1783 		else if (strcmp(name, zfs_prop_to_name(ZFS_PROP_CASE)) == 0)
1784 			sense = val;
1785 	}
1786 	ASSERT(version != 0);
1787 	error = zap_update(os, moid, ZPL_VERSION_STR, 8, 1, &version, tx);
1788 
1789 	/*
1790 	 * Create zap object used for SA attribute registration
1791 	 */
1792 
1793 	if (version >= ZPL_VERSION_SA) {
1794 		sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
1795 		    DMU_OT_NONE, 0, tx);
1796 		error = zap_add(os, moid, ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
1797 		ASSERT(error == 0);
1798 	} else {
1799 		sa_obj = 0;
1800 	}
1801 	/*
1802 	 * Create a delete queue.
1803 	 */
1804 	obj = zap_create(os, DMU_OT_UNLINKED_SET, DMU_OT_NONE, 0, tx);
1805 
1806 	error = zap_add(os, moid, ZFS_UNLINKED_SET, 8, 1, &obj, tx);
1807 	ASSERT(error == 0);
1808 
1809 	/*
1810 	 * Create root znode.  Create minimal znode/vnode/zfsvfs
1811 	 * to allow zfs_mknode to work.
1812 	 */
1813 	vattr.va_mask = AT_MODE|AT_UID|AT_GID|AT_TYPE;
1814 	vattr.va_type = VDIR;
1815 	vattr.va_mode = S_IFDIR|0755;
1816 	vattr.va_uid = crgetuid(cr);
1817 	vattr.va_gid = crgetgid(cr);
1818 
1819 	rootzp = kmem_cache_alloc(znode_cache, KM_SLEEP);
1820 	ASSERT(!POINTER_IS_VALID(rootzp->z_zfsvfs));
1821 	rootzp->z_moved = 0;
1822 	rootzp->z_unlinked = 0;
1823 	rootzp->z_atime_dirty = 0;
1824 	rootzp->z_is_sa = USE_SA(version, os);
1825 
1826 	vp = ZTOV(rootzp);
1827 	vn_reinit(vp);
1828 	vp->v_type = VDIR;
1829 
1830 	zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
1831 	zfsvfs->z_os = os;
1832 	zfsvfs->z_parent = zfsvfs;
1833 	zfsvfs->z_version = version;
1834 	zfsvfs->z_use_fuids = USE_FUIDS(version, os);
1835 	zfsvfs->z_use_sa = USE_SA(version, os);
1836 	zfsvfs->z_norm = norm;
1837 
1838 	error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
1839 	    &zfsvfs->z_attr_table);
1840 
1841 	ASSERT(error == 0);
1842 
1843 	/*
1844 	 * Fold case on file systems that are always or sometimes case
1845 	 * insensitive.
1846 	 */
1847 	if (sense == ZFS_CASE_INSENSITIVE || sense == ZFS_CASE_MIXED)
1848 		zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER;
1849 
1850 	mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
1851 	list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
1852 	    offsetof(znode_t, z_link_node));
1853 
1854 	for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1855 		mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL);
1856 
1857 	rootzp->z_zfsvfs = zfsvfs;
1858 	VERIFY(0 == zfs_acl_ids_create(rootzp, IS_ROOT_NODE, &vattr,
1859 	    cr, NULL, &acl_ids));
1860 	zfs_mknode(rootzp, &vattr, tx, cr, IS_ROOT_NODE, &zp, &acl_ids);
1861 	ASSERT3P(zp, ==, rootzp);
1862 	ASSERT(!vn_in_dnlc(ZTOV(rootzp))); /* not valid to move */
1863 	error = zap_add(os, moid, ZFS_ROOT_OBJ, 8, 1, &rootzp->z_id, tx);
1864 	ASSERT(error == 0);
1865 	zfs_acl_ids_free(&acl_ids);
1866 	POINTER_INVALIDATE(&rootzp->z_zfsvfs);
1867 
1868 	ZTOV(rootzp)->v_count = 0;
1869 	sa_handle_destroy(rootzp->z_sa_hdl);
1870 	kmem_cache_free(znode_cache, rootzp);
1871 
1872 	/*
1873 	 * Create shares directory
1874 	 */
1875 
1876 	error = zfs_create_share_dir(zfsvfs, tx);
1877 
1878 	ASSERT(error == 0);
1879 
1880 	for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1881 		mutex_destroy(&zfsvfs->z_hold_mtx[i]);
1882 	kmem_free(zfsvfs, sizeof (zfsvfs_t));
1883 }
1884 
1885 #endif /* _KERNEL */
1886 
1887 static int
zfs_sa_setup(objset_t * osp,sa_attr_type_t ** sa_table)1888 zfs_sa_setup(objset_t *osp, sa_attr_type_t **sa_table)
1889 {
1890 	uint64_t sa_obj = 0;
1891 	int error;
1892 
1893 	error = zap_lookup(osp, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1, &sa_obj);
1894 	if (error != 0 && error != ENOENT)
1895 		return (error);
1896 
1897 	error = sa_setup(osp, sa_obj, zfs_attr_table, ZPL_END, sa_table);
1898 	return (error);
1899 }
1900 
1901 static int
zfs_grab_sa_handle(objset_t * osp,uint64_t obj,sa_handle_t ** hdlp,dmu_buf_t ** db,void * tag)1902 zfs_grab_sa_handle(objset_t *osp, uint64_t obj, sa_handle_t **hdlp,
1903     dmu_buf_t **db, void *tag)
1904 {
1905 	dmu_object_info_t doi;
1906 	int error;
1907 
1908 	if ((error = sa_buf_hold(osp, obj, tag, db)) != 0)
1909 		return (error);
1910 
1911 	dmu_object_info_from_db(*db, &doi);
1912 	if ((doi.doi_bonus_type != DMU_OT_SA &&
1913 	    doi.doi_bonus_type != DMU_OT_ZNODE) ||
1914 	    doi.doi_bonus_type == DMU_OT_ZNODE &&
1915 	    doi.doi_bonus_size < sizeof (znode_phys_t)) {
1916 		sa_buf_rele(*db, tag);
1917 		return (SET_ERROR(ENOTSUP));
1918 	}
1919 
1920 	error = sa_handle_get(osp, obj, NULL, SA_HDL_PRIVATE, hdlp);
1921 	if (error != 0) {
1922 		sa_buf_rele(*db, tag);
1923 		return (error);
1924 	}
1925 
1926 	return (0);
1927 }
1928 
1929 void
zfs_release_sa_handle(sa_handle_t * hdl,dmu_buf_t * db,void * tag)1930 zfs_release_sa_handle(sa_handle_t *hdl, dmu_buf_t *db, void *tag)
1931 {
1932 	sa_handle_destroy(hdl);
1933 	sa_buf_rele(db, tag);
1934 }
1935 
1936 /*
1937  * Given an object number, return its parent object number and whether
1938  * or not the object is an extended attribute directory.
1939  */
1940 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)1941 zfs_obj_to_pobj(objset_t *osp, sa_handle_t *hdl, sa_attr_type_t *sa_table,
1942     uint64_t *pobjp, int *is_xattrdir)
1943 {
1944 	uint64_t parent;
1945 	uint64_t pflags;
1946 	uint64_t mode;
1947 	uint64_t parent_mode;
1948 	sa_bulk_attr_t bulk[3];
1949 	sa_handle_t *sa_hdl;
1950 	dmu_buf_t *sa_db;
1951 	int count = 0;
1952 	int error;
1953 
1954 	SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_PARENT], NULL,
1955 	    &parent, sizeof (parent));
1956 	SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_FLAGS], NULL,
1957 	    &pflags, sizeof (pflags));
1958 	SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL,
1959 	    &mode, sizeof (mode));
1960 
1961 	if ((error = sa_bulk_lookup(hdl, bulk, count)) != 0)
1962 		return (error);
1963 
1964 	/*
1965 	 * When a link is removed its parent pointer is not changed and will
1966 	 * be invalid.  There are two cases where a link is removed but the
1967 	 * file stays around, when it goes to the delete queue and when there
1968 	 * are additional links.
1969 	 */
1970 	error = zfs_grab_sa_handle(osp, parent, &sa_hdl, &sa_db, FTAG);
1971 	if (error != 0)
1972 		return (error);
1973 
1974 	error = sa_lookup(sa_hdl, ZPL_MODE, &parent_mode, sizeof (parent_mode));
1975 	zfs_release_sa_handle(sa_hdl, sa_db, FTAG);
1976 	if (error != 0)
1977 		return (error);
1978 
1979 	*is_xattrdir = ((pflags & ZFS_XATTR) != 0) && S_ISDIR(mode);
1980 
1981 	/*
1982 	 * Extended attributes can be applied to files, directories, etc.
1983 	 * Otherwise the parent must be a directory.
1984 	 */
1985 	if (!*is_xattrdir && !S_ISDIR(parent_mode))
1986 		return (SET_ERROR(EINVAL));
1987 
1988 	*pobjp = parent;
1989 
1990 	return (0);
1991 }
1992 
1993 /*
1994  * Given an object number, return some zpl level statistics
1995  */
1996 static int
zfs_obj_to_stats_impl(sa_handle_t * hdl,sa_attr_type_t * sa_table,zfs_stat_t * sb)1997 zfs_obj_to_stats_impl(sa_handle_t *hdl, sa_attr_type_t *sa_table,
1998     zfs_stat_t *sb)
1999 {
2000 	sa_bulk_attr_t bulk[4];
2001 	int count = 0;
2002 
2003 	SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL,
2004 	    &sb->zs_mode, sizeof (sb->zs_mode));
2005 	SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_GEN], NULL,
2006 	    &sb->zs_gen, sizeof (sb->zs_gen));
2007 	SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_LINKS], NULL,
2008 	    &sb->zs_links, sizeof (sb->zs_links));
2009 	SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_CTIME], NULL,
2010 	    &sb->zs_ctime, sizeof (sb->zs_ctime));
2011 
2012 	return (sa_bulk_lookup(hdl, bulk, count));
2013 }
2014 
2015 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)2016 zfs_obj_to_path_impl(objset_t *osp, uint64_t obj, sa_handle_t *hdl,
2017     sa_attr_type_t *sa_table, char *buf, int len)
2018 {
2019 	sa_handle_t *sa_hdl;
2020 	sa_handle_t *prevhdl = NULL;
2021 	dmu_buf_t *prevdb = NULL;
2022 	dmu_buf_t *sa_db = NULL;
2023 	char *path = buf + len - 1;
2024 	int error;
2025 
2026 	*path = '\0';
2027 	sa_hdl = hdl;
2028 
2029 	for (;;) {
2030 		uint64_t pobj;
2031 		char component[MAXNAMELEN + 2];
2032 		size_t complen;
2033 		int is_xattrdir;
2034 
2035 		if (prevdb)
2036 			zfs_release_sa_handle(prevhdl, prevdb, FTAG);
2037 
2038 		if ((error = zfs_obj_to_pobj(osp, sa_hdl, sa_table, &pobj,
2039 		    &is_xattrdir)) != 0)
2040 			break;
2041 
2042 		if (pobj == obj) {
2043 			if (path[0] != '/')
2044 				*--path = '/';
2045 			break;
2046 		}
2047 
2048 		component[0] = '/';
2049 		if (is_xattrdir) {
2050 			(void) sprintf(component + 1, "<xattrdir>");
2051 		} else {
2052 			error = zap_value_search(osp, pobj, obj,
2053 			    ZFS_DIRENT_OBJ(-1ULL), component + 1);
2054 			if (error != 0)
2055 				break;
2056 		}
2057 
2058 		complen = strlen(component);
2059 		path -= complen;
2060 		ASSERT(path >= buf);
2061 		bcopy(component, path, complen);
2062 		obj = pobj;
2063 
2064 		if (sa_hdl != hdl) {
2065 			prevhdl = sa_hdl;
2066 			prevdb = sa_db;
2067 		}
2068 		error = zfs_grab_sa_handle(osp, obj, &sa_hdl, &sa_db, FTAG);
2069 		if (error != 0) {
2070 			sa_hdl = prevhdl;
2071 			sa_db = prevdb;
2072 			break;
2073 		}
2074 	}
2075 
2076 	if (sa_hdl != NULL && sa_hdl != hdl) {
2077 		ASSERT(sa_db != NULL);
2078 		zfs_release_sa_handle(sa_hdl, sa_db, FTAG);
2079 	}
2080 
2081 	if (error == 0)
2082 		(void) memmove(buf, path, buf + len - path);
2083 
2084 	return (error);
2085 }
2086 
2087 int
zfs_obj_to_path(objset_t * osp,uint64_t obj,char * buf,int len)2088 zfs_obj_to_path(objset_t *osp, uint64_t obj, char *buf, int len)
2089 {
2090 	sa_attr_type_t *sa_table;
2091 	sa_handle_t *hdl;
2092 	dmu_buf_t *db;
2093 	int error;
2094 
2095 	error = zfs_sa_setup(osp, &sa_table);
2096 	if (error != 0)
2097 		return (error);
2098 
2099 	error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG);
2100 	if (error != 0)
2101 		return (error);
2102 
2103 	error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len);
2104 
2105 	zfs_release_sa_handle(hdl, db, FTAG);
2106 	return (error);
2107 }
2108 
2109 int
zfs_obj_to_stats(objset_t * osp,uint64_t obj,zfs_stat_t * sb,char * buf,int len)2110 zfs_obj_to_stats(objset_t *osp, uint64_t obj, zfs_stat_t *sb,
2111     char *buf, int len)
2112 {
2113 	char *path = buf + len - 1;
2114 	sa_attr_type_t *sa_table;
2115 	sa_handle_t *hdl;
2116 	dmu_buf_t *db;
2117 	int error;
2118 
2119 	*path = '\0';
2120 
2121 	error = zfs_sa_setup(osp, &sa_table);
2122 	if (error != 0)
2123 		return (error);
2124 
2125 	error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG);
2126 	if (error != 0)
2127 		return (error);
2128 
2129 	error = zfs_obj_to_stats_impl(hdl, sa_table, sb);
2130 	if (error != 0) {
2131 		zfs_release_sa_handle(hdl, db, FTAG);
2132 		return (error);
2133 	}
2134 
2135 	error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len);
2136 
2137 	zfs_release_sa_handle(hdl, db, FTAG);
2138 	return (error);
2139 }
2140