xref: /titanic_50/usr/src/uts/common/fs/zfs/zfs_znode.c (revision 9d7a20af7a24d9783cb149fa7a41c637935f3878)
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 
29 #ifdef _KERNEL
30 #include <sys/types.h>
31 #include <sys/param.h>
32 #include <sys/time.h>
33 #include <sys/systm.h>
34 #include <sys/sysmacros.h>
35 #include <sys/resource.h>
36 #include <sys/mntent.h>
37 #include <sys/mkdev.h>
38 #include <sys/u8_textprep.h>
39 #include <sys/dsl_dataset.h>
40 #include <sys/vfs.h>
41 #include <sys/vfs_opreg.h>
42 #include <sys/vnode.h>
43 #include <sys/file.h>
44 #include <sys/kmem.h>
45 #include <sys/errno.h>
46 #include <sys/unistd.h>
47 #include <sys/mode.h>
48 #include <sys/atomic.h>
49 #include <vm/pvn.h>
50 #include "fs/fs_subr.h"
51 #include <sys/zfs_dir.h>
52 #include <sys/zfs_acl.h>
53 #include <sys/zfs_ioctl.h>
54 #include <sys/zfs_rlock.h>
55 #include <sys/zfs_fuid.h>
56 #include <sys/dnode.h>
57 #include <sys/fs/zfs.h>
58 #include <sys/kidmap.h>
59 #endif /* _KERNEL */
60 
61 #include <sys/dmu.h>
62 #include <sys/dmu_objset.h>
63 #include <sys/refcount.h>
64 #include <sys/stat.h>
65 #include <sys/zap.h>
66 #include <sys/zfs_znode.h>
67 #include <sys/sa.h>
68 #include <sys/zfs_sa.h>
69 #include <sys/zfs_stat.h>
70 #include <sys/zfs_events.h>
71 #include <zev/zev.h>
72 
73 #include "zfs_prop.h"
74 #include "zfs_comutil.h"
75 
76 /*
77  * Define ZNODE_STATS to turn on statistic gathering. By default, it is only
78  * turned on when DEBUG is also defined.
79  */
80 #ifdef	DEBUG
81 #define	ZNODE_STATS
82 #endif	/* DEBUG */
83 
84 #ifdef	ZNODE_STATS
85 #define	ZNODE_STAT_ADD(stat)			((stat)++)
86 #else
87 #define	ZNODE_STAT_ADD(stat)			/* nothing */
88 #endif	/* ZNODE_STATS */
89 
90 /*
91  * Functions needed for userland (ie: libzpool) are not put under
92  * #ifdef_KERNEL; the rest of the functions have dependencies
93  * (such as VFS logic) that will not compile easily in userland.
94  */
95 #ifdef _KERNEL
96 /*
97  * Needed to close a small window in zfs_znode_move() that allows the zfsvfs to
98  * be freed before it can be safely accessed.
99  */
100 krwlock_t zfsvfs_lock;
101 
102 static kmem_cache_t *znode_cache = NULL;
103 
104 /*ARGSUSED*/
105 static void
znode_evict_error(dmu_buf_t * dbuf,void * user_ptr)106 znode_evict_error(dmu_buf_t *dbuf, void *user_ptr)
107 {
108 	/*
109 	 * We should never drop all dbuf refs without first clearing
110 	 * the eviction callback.
111 	 */
112 	panic("evicting znode %p\n", user_ptr);
113 }
114 
115 /*ARGSUSED*/
116 static int
zfs_znode_cache_constructor(void * buf,void * arg,int kmflags)117 zfs_znode_cache_constructor(void *buf, void *arg, int kmflags)
118 {
119 	znode_t *zp = buf;
120 
121 	ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs));
122 
123 	zp->z_vnode = vn_alloc(kmflags);
124 	if (zp->z_vnode == NULL) {
125 		return (-1);
126 	}
127 	ZTOV(zp)->v_data = zp;
128 
129 	list_link_init(&zp->z_link_node);
130 
131 	mutex_init(&zp->z_lock, NULL, MUTEX_DEFAULT, NULL);
132 	rw_init(&zp->z_parent_lock, NULL, RW_DEFAULT, NULL);
133 	rw_init(&zp->z_name_lock, NULL, RW_DEFAULT, NULL);
134 	mutex_init(&zp->z_acl_lock, NULL, MUTEX_DEFAULT, NULL);
135 
136 	mutex_init(&zp->z_range_lock, NULL, MUTEX_DEFAULT, NULL);
137 	avl_create(&zp->z_range_avl, zfs_range_compare,
138 	    sizeof (rl_t), offsetof(rl_t, r_node));
139 
140 	zp->z_dirlocks = NULL;
141 	zp->z_acl_cached = NULL;
142 	zp->z_moved = 0;
143 	return (0);
144 }
145 
146 /*ARGSUSED*/
147 static void
zfs_znode_cache_destructor(void * buf,void * arg)148 zfs_znode_cache_destructor(void *buf, void *arg)
149 {
150 	znode_t *zp = buf;
151 
152 	ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs));
153 	ASSERT(ZTOV(zp)->v_data == zp);
154 	vn_free(ZTOV(zp));
155 	ASSERT(!list_link_active(&zp->z_link_node));
156 	mutex_destroy(&zp->z_lock);
157 	rw_destroy(&zp->z_parent_lock);
158 	rw_destroy(&zp->z_name_lock);
159 	mutex_destroy(&zp->z_acl_lock);
160 	avl_destroy(&zp->z_range_avl);
161 	mutex_destroy(&zp->z_range_lock);
162 
163 	ASSERT(zp->z_dirlocks == NULL);
164 	ASSERT(zp->z_acl_cached == NULL);
165 }
166 
167 #ifdef	ZNODE_STATS
168 static struct {
169 	uint64_t zms_zfsvfs_invalid;
170 	uint64_t zms_zfsvfs_recheck1;
171 	uint64_t zms_zfsvfs_unmounted;
172 	uint64_t zms_zfsvfs_recheck2;
173 	uint64_t zms_obj_held;
174 	uint64_t zms_vnode_locked;
175 	uint64_t zms_not_only_dnlc;
176 } znode_move_stats;
177 #endif	/* ZNODE_STATS */
178 
179 static void
zfs_znode_move_impl(znode_t * ozp,znode_t * nzp)180 zfs_znode_move_impl(znode_t *ozp, znode_t *nzp)
181 {
182 	vnode_t *vp;
183 
184 	/* Copy fields. */
185 	nzp->z_zfsvfs = ozp->z_zfsvfs;
186 
187 	/* Swap vnodes. */
188 	vp = nzp->z_vnode;
189 	nzp->z_vnode = ozp->z_vnode;
190 	ozp->z_vnode = vp; /* let destructor free the overwritten vnode */
191 	ZTOV(ozp)->v_data = ozp;
192 	ZTOV(nzp)->v_data = nzp;
193 
194 	nzp->z_id = ozp->z_id;
195 	ASSERT(ozp->z_dirlocks == NULL); /* znode not in use */
196 	ASSERT(avl_numnodes(&ozp->z_range_avl) == 0);
197 	nzp->z_unlinked = ozp->z_unlinked;
198 	nzp->z_atime_dirty = ozp->z_atime_dirty;
199 	nzp->z_zn_prefetch = ozp->z_zn_prefetch;
200 	nzp->z_blksz = ozp->z_blksz;
201 	nzp->z_seq = ozp->z_seq;
202 	nzp->z_mapcnt = ozp->z_mapcnt;
203 	nzp->z_gen = ozp->z_gen;
204 	nzp->z_sync_cnt = ozp->z_sync_cnt;
205 	nzp->z_is_sa = ozp->z_is_sa;
206 	nzp->z_sa_hdl = ozp->z_sa_hdl;
207 	bcopy(ozp->z_atime, nzp->z_atime, sizeof (uint64_t) * 2);
208 	nzp->z_links = ozp->z_links;
209 	nzp->z_size = ozp->z_size;
210 	nzp->z_pflags = ozp->z_pflags;
211 	nzp->z_uid = ozp->z_uid;
212 	nzp->z_gid = ozp->z_gid;
213 	nzp->z_mode = ozp->z_mode;
214 	nzp->z_new_content = ozp->z_new_content;
215 
216 	/*
217 	 * Since this is just an idle znode and kmem is already dealing with
218 	 * memory pressure, release any cached ACL.
219 	 */
220 	if (ozp->z_acl_cached) {
221 		zfs_acl_free(ozp->z_acl_cached);
222 		ozp->z_acl_cached = NULL;
223 	}
224 
225 	sa_set_userp(nzp->z_sa_hdl, nzp);
226 
227 	/*
228 	 * Invalidate the original znode by clearing fields that provide a
229 	 * pointer back to the znode. Set the low bit of the vfs pointer to
230 	 * ensure that zfs_znode_move() recognizes the znode as invalid in any
231 	 * subsequent callback.
232 	 */
233 	ozp->z_sa_hdl = NULL;
234 	POINTER_INVALIDATE(&ozp->z_zfsvfs);
235 
236 	/*
237 	 * Mark the znode.
238 	 */
239 	nzp->z_moved = 1;
240 	ozp->z_moved = (uint8_t)-1;
241 }
242 
243 /*ARGSUSED*/
244 static kmem_cbrc_t
zfs_znode_move(void * buf,void * newbuf,size_t size,void * arg)245 zfs_znode_move(void *buf, void *newbuf, size_t size, void *arg)
246 {
247 	znode_t *ozp = buf, *nzp = newbuf;
248 	zfsvfs_t *zfsvfs;
249 	vnode_t *vp;
250 
251 	/*
252 	 * The znode is on the file system's list of known znodes if the vfs
253 	 * pointer is valid. We set the low bit of the vfs pointer when freeing
254 	 * the znode to invalidate it, and the memory patterns written by kmem
255 	 * (baddcafe and deadbeef) set at least one of the two low bits. A newly
256 	 * created znode sets the vfs pointer last of all to indicate that the
257 	 * znode is known and in a valid state to be moved by this function.
258 	 */
259 	zfsvfs = ozp->z_zfsvfs;
260 	if (!POINTER_IS_VALID(zfsvfs)) {
261 		ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_invalid);
262 		return (KMEM_CBRC_DONT_KNOW);
263 	}
264 
265 	/*
266 	 * Close a small window in which it's possible that the filesystem could
267 	 * be unmounted and freed, and zfsvfs, though valid in the previous
268 	 * statement, could point to unrelated memory by the time we try to
269 	 * prevent the filesystem from being unmounted.
270 	 */
271 	rw_enter(&zfsvfs_lock, RW_WRITER);
272 	if (zfsvfs != ozp->z_zfsvfs) {
273 		rw_exit(&zfsvfs_lock);
274 		ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_recheck1);
275 		return (KMEM_CBRC_DONT_KNOW);
276 	}
277 
278 	/*
279 	 * If the znode is still valid, then so is the file system. We know that
280 	 * no valid file system can be freed while we hold zfsvfs_lock, so we
281 	 * can safely ensure that the filesystem is not and will not be
282 	 * unmounted. The next statement is equivalent to ZFS_ENTER().
283 	 */
284 	rrm_enter(&zfsvfs->z_teardown_lock, RW_READER, FTAG);
285 	if (zfsvfs->z_unmounted) {
286 		ZFS_EXIT(zfsvfs);
287 		rw_exit(&zfsvfs_lock);
288 		ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_unmounted);
289 		return (KMEM_CBRC_DONT_KNOW);
290 	}
291 	rw_exit(&zfsvfs_lock);
292 
293 	mutex_enter(&zfsvfs->z_znodes_lock);
294 	/*
295 	 * Recheck the vfs pointer in case the znode was removed just before
296 	 * acquiring the lock.
297 	 */
298 	if (zfsvfs != ozp->z_zfsvfs) {
299 		mutex_exit(&zfsvfs->z_znodes_lock);
300 		ZFS_EXIT(zfsvfs);
301 		ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_recheck2);
302 		return (KMEM_CBRC_DONT_KNOW);
303 	}
304 
305 	/*
306 	 * At this point we know that as long as we hold z_znodes_lock, the
307 	 * znode cannot be freed and fields within the znode can be safely
308 	 * accessed. Now, prevent a race with zfs_zget().
309 	 */
310 	if (ZFS_OBJ_HOLD_TRYENTER(zfsvfs, ozp->z_id) == 0) {
311 		mutex_exit(&zfsvfs->z_znodes_lock);
312 		ZFS_EXIT(zfsvfs);
313 		ZNODE_STAT_ADD(znode_move_stats.zms_obj_held);
314 		return (KMEM_CBRC_LATER);
315 	}
316 
317 	vp = ZTOV(ozp);
318 	if (mutex_tryenter(&vp->v_lock) == 0) {
319 		ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id);
320 		mutex_exit(&zfsvfs->z_znodes_lock);
321 		ZFS_EXIT(zfsvfs);
322 		ZNODE_STAT_ADD(znode_move_stats.zms_vnode_locked);
323 		return (KMEM_CBRC_LATER);
324 	}
325 
326 	/* Only move znodes that are referenced _only_ by the DNLC. */
327 	if (vp->v_count != 1 || !vn_in_dnlc(vp)) {
328 		mutex_exit(&vp->v_lock);
329 		ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id);
330 		mutex_exit(&zfsvfs->z_znodes_lock);
331 		ZFS_EXIT(zfsvfs);
332 		ZNODE_STAT_ADD(znode_move_stats.zms_not_only_dnlc);
333 		return (KMEM_CBRC_LATER);
334 	}
335 
336 	/*
337 	 * The znode is known and in a valid state to move. We're holding the
338 	 * locks needed to execute the critical section.
339 	 */
340 	zfs_znode_move_impl(ozp, nzp);
341 	mutex_exit(&vp->v_lock);
342 	ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id);
343 
344 	list_link_replace(&ozp->z_link_node, &nzp->z_link_node);
345 	mutex_exit(&zfsvfs->z_znodes_lock);
346 	ZFS_EXIT(zfsvfs);
347 
348 	return (KMEM_CBRC_YES);
349 }
350 
351 void
zfs_znode_init(void)352 zfs_znode_init(void)
353 {
354 	/*
355 	 * Initialize zcache
356 	 */
357 	rw_init(&zfsvfs_lock, NULL, RW_DEFAULT, NULL);
358 	ASSERT(znode_cache == NULL);
359 	znode_cache = kmem_cache_create("zfs_znode_cache",
360 	    sizeof (znode_t), 0, zfs_znode_cache_constructor,
361 	    zfs_znode_cache_destructor, NULL, NULL, NULL, 0);
362 	kmem_cache_set_move(znode_cache, zfs_znode_move);
363 }
364 
365 void
zfs_znode_fini(void)366 zfs_znode_fini(void)
367 {
368 	/*
369 	 * Cleanup vfs & vnode ops
370 	 */
371 	zfs_remove_op_tables();
372 
373 	/*
374 	 * Cleanup zcache
375 	 */
376 	if (znode_cache)
377 		kmem_cache_destroy(znode_cache);
378 	znode_cache = NULL;
379 	rw_destroy(&zfsvfs_lock);
380 }
381 
382 struct vnodeops *zfs_dvnodeops;
383 struct vnodeops *zfs_fvnodeops;
384 struct vnodeops *zfs_symvnodeops;
385 struct vnodeops *zfs_xdvnodeops;
386 struct vnodeops *zfs_evnodeops;
387 struct vnodeops *zfs_sharevnodeops;
388 
389 void
zfs_remove_op_tables()390 zfs_remove_op_tables()
391 {
392 	/*
393 	 * Remove vfs ops
394 	 */
395 	ASSERT(zfsfstype);
396 	(void) vfs_freevfsops_by_type(zfsfstype);
397 	zfsfstype = 0;
398 
399 	/*
400 	 * Remove vnode ops
401 	 */
402 	if (zfs_dvnodeops)
403 		vn_freevnodeops(zfs_dvnodeops);
404 	if (zfs_fvnodeops)
405 		vn_freevnodeops(zfs_fvnodeops);
406 	if (zfs_symvnodeops)
407 		vn_freevnodeops(zfs_symvnodeops);
408 	if (zfs_xdvnodeops)
409 		vn_freevnodeops(zfs_xdvnodeops);
410 	if (zfs_evnodeops)
411 		vn_freevnodeops(zfs_evnodeops);
412 	if (zfs_sharevnodeops)
413 		vn_freevnodeops(zfs_sharevnodeops);
414 
415 	zfs_dvnodeops = NULL;
416 	zfs_fvnodeops = NULL;
417 	zfs_symvnodeops = NULL;
418 	zfs_xdvnodeops = NULL;
419 	zfs_evnodeops = NULL;
420 	zfs_sharevnodeops = NULL;
421 }
422 
423 extern const fs_operation_def_t zfs_dvnodeops_template[];
424 extern const fs_operation_def_t zfs_fvnodeops_template[];
425 extern const fs_operation_def_t zfs_xdvnodeops_template[];
426 extern const fs_operation_def_t zfs_symvnodeops_template[];
427 extern const fs_operation_def_t zfs_evnodeops_template[];
428 extern const fs_operation_def_t zfs_sharevnodeops_template[];
429 
430 int
zfs_create_op_tables()431 zfs_create_op_tables()
432 {
433 	int error;
434 
435 	/*
436 	 * zfs_dvnodeops can be set if mod_remove() calls mod_installfs()
437 	 * due to a failure to remove the the 2nd modlinkage (zfs_modldrv).
438 	 * In this case we just return as the ops vectors are already set up.
439 	 */
440 	if (zfs_dvnodeops)
441 		return (0);
442 
443 	error = vn_make_ops(MNTTYPE_ZFS, zfs_dvnodeops_template,
444 	    &zfs_dvnodeops);
445 	if (error)
446 		return (error);
447 
448 	error = vn_make_ops(MNTTYPE_ZFS, zfs_fvnodeops_template,
449 	    &zfs_fvnodeops);
450 	if (error)
451 		return (error);
452 
453 	error = vn_make_ops(MNTTYPE_ZFS, zfs_symvnodeops_template,
454 	    &zfs_symvnodeops);
455 	if (error)
456 		return (error);
457 
458 	error = vn_make_ops(MNTTYPE_ZFS, zfs_xdvnodeops_template,
459 	    &zfs_xdvnodeops);
460 	if (error)
461 		return (error);
462 
463 	error = vn_make_ops(MNTTYPE_ZFS, zfs_evnodeops_template,
464 	    &zfs_evnodeops);
465 	if (error)
466 		return (error);
467 
468 	error = vn_make_ops(MNTTYPE_ZFS, zfs_sharevnodeops_template,
469 	    &zfs_sharevnodeops);
470 
471 	return (error);
472 }
473 
474 int
zfs_create_share_dir(zfsvfs_t * zfsvfs,dmu_tx_t * tx)475 zfs_create_share_dir(zfsvfs_t *zfsvfs, dmu_tx_t *tx)
476 {
477 	zfs_acl_ids_t acl_ids;
478 	vattr_t vattr;
479 	znode_t *sharezp;
480 	vnode_t *vp;
481 	znode_t *zp;
482 	int error;
483 
484 	vattr.va_mask = AT_MODE|AT_UID|AT_GID|AT_TYPE;
485 	vattr.va_type = VDIR;
486 	vattr.va_mode = S_IFDIR|0555;
487 	vattr.va_uid = crgetuid(kcred);
488 	vattr.va_gid = crgetgid(kcred);
489 
490 	sharezp = kmem_cache_alloc(znode_cache, KM_SLEEP);
491 	ASSERT(!POINTER_IS_VALID(sharezp->z_zfsvfs));
492 	sharezp->z_moved = 0;
493 	sharezp->z_unlinked = 0;
494 	sharezp->z_atime_dirty = 0;
495 	sharezp->z_zfsvfs = zfsvfs;
496 	sharezp->z_is_sa = zfsvfs->z_use_sa;
497 
498 	vp = ZTOV(sharezp);
499 	vn_reinit(vp);
500 	vp->v_type = VDIR;
501 
502 	VERIFY(0 == zfs_acl_ids_create(sharezp, IS_ROOT_NODE, &vattr,
503 	    kcred, NULL, &acl_ids));
504 	zfs_mknode(sharezp, &vattr, tx, kcred, IS_ROOT_NODE, &zp, &acl_ids);
505 	ASSERT3P(zp, ==, sharezp);
506 	ASSERT(!vn_in_dnlc(ZTOV(sharezp))); /* not valid to move */
507 	POINTER_INVALIDATE(&sharezp->z_zfsvfs);
508 	error = zap_add(zfsvfs->z_os, MASTER_NODE_OBJ,
509 	    ZFS_SHARES_DIR, 8, 1, &sharezp->z_id, tx);
510 	zfsvfs->z_shares_dir = sharezp->z_id;
511 
512 	zfs_acl_ids_free(&acl_ids);
513 	ZTOV(sharezp)->v_count = 0;
514 	sa_handle_destroy(sharezp->z_sa_hdl);
515 	kmem_cache_free(znode_cache, sharezp);
516 
517 	return (error);
518 }
519 
520 /*
521  * define a couple of values we need available
522  * for both 64 and 32 bit environments.
523  */
524 #ifndef NBITSMINOR64
525 #define	NBITSMINOR64	32
526 #endif
527 #ifndef MAXMAJ64
528 #define	MAXMAJ64	0xffffffffUL
529 #endif
530 #ifndef	MAXMIN64
531 #define	MAXMIN64	0xffffffffUL
532 #endif
533 
534 /*
535  * Create special expldev for ZFS private use.
536  * Can't use standard expldev since it doesn't do
537  * what we want.  The standard expldev() takes a
538  * dev32_t in LP64 and expands it to a long dev_t.
539  * We need an interface that takes a dev32_t in ILP32
540  * and expands it to a long dev_t.
541  */
542 static uint64_t
zfs_expldev(dev_t dev)543 zfs_expldev(dev_t dev)
544 {
545 #ifndef _LP64
546 	major_t major = (major_t)dev >> NBITSMINOR32 & MAXMAJ32;
547 	return (((uint64_t)major << NBITSMINOR64) |
548 	    ((minor_t)dev & MAXMIN32));
549 #else
550 	return (dev);
551 #endif
552 }
553 
554 /*
555  * Special cmpldev for ZFS private use.
556  * Can't use standard cmpldev since it takes
557  * a long dev_t and compresses it to dev32_t in
558  * LP64.  We need to do a compaction of a long dev_t
559  * to a dev32_t in ILP32.
560  */
561 dev_t
zfs_cmpldev(uint64_t dev)562 zfs_cmpldev(uint64_t dev)
563 {
564 #ifndef _LP64
565 	minor_t minor = (minor_t)dev & MAXMIN64;
566 	major_t major = (major_t)(dev >> NBITSMINOR64) & MAXMAJ64;
567 
568 	if (major > MAXMAJ32 || minor > MAXMIN32)
569 		return (NODEV32);
570 
571 	return (((dev32_t)major << NBITSMINOR32) | minor);
572 #else
573 	return (dev);
574 #endif
575 }
576 
577 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)578 zfs_znode_sa_init(zfsvfs_t *zfsvfs, znode_t *zp,
579     dmu_buf_t *db, dmu_object_type_t obj_type, sa_handle_t *sa_hdl)
580 {
581 	ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs) || (zfsvfs == zp->z_zfsvfs));
582 	ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zfsvfs, zp->z_id)));
583 
584 	mutex_enter(&zp->z_lock);
585 
586 	ASSERT(zp->z_sa_hdl == NULL);
587 	ASSERT(zp->z_acl_cached == NULL);
588 	if (sa_hdl == NULL) {
589 		VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, zp,
590 		    SA_HDL_SHARED, &zp->z_sa_hdl));
591 	} else {
592 		zp->z_sa_hdl = sa_hdl;
593 		sa_set_userp(sa_hdl, zp);
594 	}
595 
596 	zp->z_is_sa = (obj_type == DMU_OT_SA) ? B_TRUE : B_FALSE;
597 
598 	/*
599 	 * Slap on VROOT if we are the root znode
600 	 */
601 	if (zp->z_id == zfsvfs->z_root)
602 		ZTOV(zp)->v_flag |= VROOT;
603 
604 	mutex_exit(&zp->z_lock);
605 	vn_exists(ZTOV(zp));
606 }
607 
608 void
zfs_znode_dmu_fini(znode_t * zp)609 zfs_znode_dmu_fini(znode_t *zp)
610 {
611 	ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zp->z_zfsvfs, zp->z_id)) ||
612 	    zp->z_unlinked ||
613 	    RW_WRITE_HELD(&zp->z_zfsvfs->z_teardown_inactive_lock));
614 
615 	sa_handle_destroy(zp->z_sa_hdl);
616 	zp->z_sa_hdl = NULL;
617 }
618 
619 /*
620  * Construct a new znode/vnode and intialize.
621  *
622  * This does not do a call to dmu_set_user() that is
623  * up to the caller to do, in case you don't want to
624  * return the znode
625  */
626 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)627 zfs_znode_alloc(zfsvfs_t *zfsvfs, dmu_buf_t *db, int blksz,
628     dmu_object_type_t obj_type, sa_handle_t *hdl)
629 {
630 	znode_t	*zp;
631 	vnode_t *vp;
632 	uint64_t mode;
633 	uint64_t parent;
634 	sa_bulk_attr_t bulk[9];
635 	int count = 0;
636 
637 	zp = kmem_cache_alloc(znode_cache, KM_SLEEP);
638 
639 	ASSERT(zp->z_dirlocks == NULL);
640 	ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs));
641 	zp->z_moved = 0;
642 
643 	/*
644 	 * Defer setting z_zfsvfs until the znode is ready to be a candidate for
645 	 * the zfs_znode_move() callback.
646 	 */
647 	zp->z_sa_hdl = NULL;
648 	zp->z_unlinked = 0;
649 	zp->z_atime_dirty = 0;
650 	zp->z_mapcnt = 0;
651 	zp->z_id = db->db_object;
652 	zp->z_blksz = blksz;
653 	zp->z_seq = 0x7A4653;
654 	zp->z_sync_cnt = 0;
655 	zp->z_new_content = 0;
656 
657 	vp = ZTOV(zp);
658 	vn_reinit(vp);
659 
660 	zfs_znode_sa_init(zfsvfs, zp, db, obj_type, hdl);
661 
662 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8);
663 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL, &zp->z_gen, 8);
664 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
665 	    &zp->z_size, 8);
666 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL,
667 	    &zp->z_links, 8);
668 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
669 	    &zp->z_pflags, 8);
670 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL, &parent, 8);
671 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
672 	    &zp->z_atime, 16);
673 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
674 	    &zp->z_uid, 8);
675 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL,
676 	    &zp->z_gid, 8);
677 
678 	if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count) != 0 || zp->z_gen == 0) {
679 		if (hdl == NULL)
680 			sa_handle_destroy(zp->z_sa_hdl);
681 		kmem_cache_free(znode_cache, zp);
682 		return (NULL);
683 	}
684 
685 	zp->z_mode = mode;
686 	vp->v_vfsp = zfsvfs->z_parent->z_vfs;
687 
688 	vp->v_type = IFTOVT((mode_t)mode);
689 
690 	switch (vp->v_type) {
691 	case VDIR:
692 		if (zp->z_pflags & ZFS_XATTR) {
693 			vn_setops(vp, zfs_xdvnodeops);
694 			vp->v_flag |= V_XATTRDIR;
695 		} else {
696 			vn_setops(vp, zfs_dvnodeops);
697 		}
698 		zp->z_zn_prefetch = B_TRUE; /* z_prefetch default is enabled */
699 		break;
700 	case VBLK:
701 	case VCHR:
702 		{
703 			uint64_t rdev;
704 			VERIFY(sa_lookup(zp->z_sa_hdl, SA_ZPL_RDEV(zfsvfs),
705 			    &rdev, sizeof (rdev)) == 0);
706 
707 			vp->v_rdev = zfs_cmpldev(rdev);
708 		}
709 		/*FALLTHROUGH*/
710 	case VFIFO:
711 	case VSOCK:
712 	case VDOOR:
713 		vn_setops(vp, zfs_fvnodeops);
714 		break;
715 	case VREG:
716 		vp->v_flag |= VMODSORT;
717 		if (parent == zfsvfs->z_shares_dir) {
718 			ASSERT(zp->z_uid == 0 && zp->z_gid == 0);
719 			vn_setops(vp, zfs_sharevnodeops);
720 		} else {
721 			vn_setops(vp, zfs_fvnodeops);
722 		}
723 		break;
724 	case VLNK:
725 		vn_setops(vp, zfs_symvnodeops);
726 		break;
727 	default:
728 		vn_setops(vp, zfs_evnodeops);
729 		break;
730 	}
731 
732 	mutex_enter(&zfsvfs->z_znodes_lock);
733 	list_insert_tail(&zfsvfs->z_all_znodes, zp);
734 	membar_producer();
735 	/*
736 	 * Everything else must be valid before assigning z_zfsvfs makes the
737 	 * znode eligible for zfs_znode_move().
738 	 */
739 	zp->z_zfsvfs = zfsvfs;
740 	mutex_exit(&zfsvfs->z_znodes_lock);
741 
742 	VFS_HOLD(zfsvfs->z_vfs);
743 	return (zp);
744 }
745 
746 static uint64_t empty_xattr;
747 static uint64_t pad[4];
748 static zfs_acl_phys_t acl_phys;
749 /*
750  * Create a new DMU object to hold a zfs znode.
751  *
752  *	IN:	dzp	- parent directory for new znode
753  *		vap	- file attributes for new znode
754  *		tx	- dmu transaction id for zap operations
755  *		cr	- credentials of caller
756  *		flag	- flags:
757  *			  IS_ROOT_NODE	- new object will be root
758  *			  IS_XATTR	- new object is an attribute
759  *		bonuslen - length of bonus buffer
760  *		setaclp  - File/Dir initial ACL
761  *		fuidp	 - Tracks fuid allocation.
762  *
763  *	OUT:	zpp	- allocated znode
764  *
765  */
766 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)767 zfs_mknode(znode_t *dzp, vattr_t *vap, dmu_tx_t *tx, cred_t *cr,
768     uint_t flag, znode_t **zpp, zfs_acl_ids_t *acl_ids)
769 {
770 	uint64_t	crtime[2], atime[2], mtime[2], ctime[2];
771 	uint64_t	mode, size, links, parent, pflags;
772 	uint64_t	dzp_pflags = 0;
773 	uint64_t	rdev = 0;
774 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
775 	dmu_buf_t	*db;
776 	timestruc_t	now;
777 	uint64_t	gen, obj;
778 	int		bonuslen;
779 	sa_handle_t	*sa_hdl;
780 	dmu_object_type_t obj_type;
781 	sa_bulk_attr_t	sa_attrs[ZPL_END];
782 	int		cnt = 0;
783 	zfs_acl_locator_cb_t locate = { 0 };
784 
785 	ASSERT(vap && (vap->va_mask & (AT_TYPE|AT_MODE)) == (AT_TYPE|AT_MODE));
786 
787 	if (zfsvfs->z_replay) {
788 		obj = vap->va_nodeid;
789 		now = vap->va_ctime;		/* see zfs_replay_create() */
790 		gen = vap->va_nblocks;		/* ditto */
791 	} else {
792 		obj = 0;
793 		gethrestime(&now);
794 		gen = dmu_tx_get_txg(tx);
795 	}
796 
797 	obj_type = zfsvfs->z_use_sa ? DMU_OT_SA : DMU_OT_ZNODE;
798 	bonuslen = (obj_type == DMU_OT_SA) ?
799 	    DN_MAX_BONUSLEN : ZFS_OLD_ZNODE_PHYS_SIZE;
800 
801 	/*
802 	 * Create a new DMU object.
803 	 */
804 	/*
805 	 * There's currently no mechanism for pre-reading the blocks that will
806 	 * be needed to allocate a new object, so we accept the small chance
807 	 * that there will be an i/o error and we will fail one of the
808 	 * assertions below.
809 	 */
810 	if (vap->va_type == VDIR) {
811 		if (zfsvfs->z_replay) {
812 			VERIFY0(zap_create_claim_norm(zfsvfs->z_os, obj,
813 			    zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS,
814 			    obj_type, bonuslen, tx));
815 		} else {
816 			obj = zap_create_norm(zfsvfs->z_os,
817 			    zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS,
818 			    obj_type, bonuslen, tx);
819 		}
820 	} else {
821 		if (zfsvfs->z_replay) {
822 			VERIFY0(dmu_object_claim(zfsvfs->z_os, obj,
823 			    DMU_OT_PLAIN_FILE_CONTENTS, 0,
824 			    obj_type, bonuslen, tx));
825 		} else {
826 			obj = dmu_object_alloc(zfsvfs->z_os,
827 			    DMU_OT_PLAIN_FILE_CONTENTS, 0,
828 			    obj_type, bonuslen, tx);
829 		}
830 	}
831 
832 	ZFS_OBJ_HOLD_ENTER(zfsvfs, obj);
833 	VERIFY(0 == sa_buf_hold(zfsvfs->z_os, obj, NULL, &db));
834 
835 	/*
836 	 * If this is the root, fix up the half-initialized parent pointer
837 	 * to reference the just-allocated physical data area.
838 	 */
839 	if (flag & IS_ROOT_NODE) {
840 		dzp->z_id = obj;
841 	} else {
842 		dzp_pflags = dzp->z_pflags;
843 	}
844 
845 	/*
846 	 * If parent is an xattr, so am I.
847 	 */
848 	if (dzp_pflags & ZFS_XATTR) {
849 		flag |= IS_XATTR;
850 	}
851 
852 	if (zfsvfs->z_use_fuids)
853 		pflags = ZFS_ARCHIVE | ZFS_AV_MODIFIED;
854 	else
855 		pflags = 0;
856 
857 	if (vap->va_type == VDIR) {
858 		size = 2;		/* contents ("." and "..") */
859 		links = (flag & (IS_ROOT_NODE | IS_XATTR)) ? 2 : 1;
860 	} else {
861 		size = links = 0;
862 	}
863 
864 	if (vap->va_type == VBLK || vap->va_type == VCHR) {
865 		rdev = zfs_expldev(vap->va_rdev);
866 	}
867 
868 	parent = dzp->z_id;
869 	mode = acl_ids->z_mode;
870 	if (flag & IS_XATTR)
871 		pflags |= ZFS_XATTR;
872 
873 	/*
874 	 * No execs denied will be deterimed when zfs_mode_compute() is called.
875 	 */
876 	pflags |= acl_ids->z_aclp->z_hints &
877 	    (ZFS_ACL_TRIVIAL|ZFS_INHERIT_ACE|ZFS_ACL_AUTO_INHERIT|
878 	    ZFS_ACL_DEFAULTED|ZFS_ACL_PROTECTED);
879 
880 	ZFS_TIME_ENCODE(&now, crtime);
881 	ZFS_TIME_ENCODE(&now, ctime);
882 
883 	if (vap->va_mask & AT_ATIME) {
884 		ZFS_TIME_ENCODE(&vap->va_atime, atime);
885 	} else {
886 		ZFS_TIME_ENCODE(&now, atime);
887 	}
888 
889 	if (vap->va_mask & AT_MTIME) {
890 		ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
891 	} else {
892 		ZFS_TIME_ENCODE(&now, mtime);
893 	}
894 
895 	/* Now add in all of the "SA" attributes */
896 	VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, NULL, SA_HDL_SHARED,
897 	    &sa_hdl));
898 
899 	/*
900 	 * Setup the array of attributes to be replaced/set on the new file
901 	 *
902 	 * order for  DMU_OT_ZNODE is critical since it needs to be constructed
903 	 * in the old znode_phys_t format.  Don't change this ordering
904 	 */
905 
906 	if (obj_type == DMU_OT_ZNODE) {
907 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs),
908 		    NULL, &atime, 16);
909 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs),
910 		    NULL, &mtime, 16);
911 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs),
912 		    NULL, &ctime, 16);
913 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs),
914 		    NULL, &crtime, 16);
915 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs),
916 		    NULL, &gen, 8);
917 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs),
918 		    NULL, &mode, 8);
919 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs),
920 		    NULL, &size, 8);
921 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs),
922 		    NULL, &parent, 8);
923 	} else {
924 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs),
925 		    NULL, &mode, 8);
926 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs),
927 		    NULL, &size, 8);
928 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs),
929 		    NULL, &gen, 8);
930 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs), NULL,
931 		    &acl_ids->z_fuid, 8);
932 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs), NULL,
933 		    &acl_ids->z_fgid, 8);
934 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs),
935 		    NULL, &parent, 8);
936 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs),
937 		    NULL, &pflags, 8);
938 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs),
939 		    NULL, &atime, 16);
940 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs),
941 		    NULL, &mtime, 16);
942 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs),
943 		    NULL, &ctime, 16);
944 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs),
945 		    NULL, &crtime, 16);
946 	}
947 
948 	SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_LINKS(zfsvfs), NULL, &links, 8);
949 
950 	if (obj_type == DMU_OT_ZNODE) {
951 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_XATTR(zfsvfs), NULL,
952 		    &empty_xattr, 8);
953 	}
954 	if (obj_type == DMU_OT_ZNODE ||
955 	    (vap->va_type == VBLK || vap->va_type == VCHR)) {
956 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_RDEV(zfsvfs),
957 		    NULL, &rdev, 8);
958 
959 	}
960 	if (obj_type == DMU_OT_ZNODE) {
961 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs),
962 		    NULL, &pflags, 8);
963 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs), NULL,
964 		    &acl_ids->z_fuid, 8);
965 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs), NULL,
966 		    &acl_ids->z_fgid, 8);
967 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PAD(zfsvfs), NULL, pad,
968 		    sizeof (uint64_t) * 4);
969 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ZNODE_ACL(zfsvfs), NULL,
970 		    &acl_phys, sizeof (zfs_acl_phys_t));
971 	} else if (acl_ids->z_aclp->z_version >= ZFS_ACL_VERSION_FUID) {
972 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_COUNT(zfsvfs), NULL,
973 		    &acl_ids->z_aclp->z_acl_count, 8);
974 		locate.cb_aclp = acl_ids->z_aclp;
975 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_ACES(zfsvfs),
976 		    zfs_acl_data_locator, &locate,
977 		    acl_ids->z_aclp->z_acl_bytes);
978 		mode = zfs_mode_compute(mode, acl_ids->z_aclp, &pflags,
979 		    acl_ids->z_fuid, acl_ids->z_fgid);
980 	}
981 
982 	VERIFY(sa_replace_all_by_template(sa_hdl, sa_attrs, cnt, tx) == 0);
983 
984 	if (!(flag & IS_ROOT_NODE)) {
985 		*zpp = zfs_znode_alloc(zfsvfs, db, 0, obj_type, sa_hdl);
986 		ASSERT(*zpp != NULL);
987 	} else {
988 		/*
989 		 * If we are creating the root node, the "parent" we
990 		 * passed in is the znode for the root.
991 		 */
992 		*zpp = dzp;
993 
994 		(*zpp)->z_sa_hdl = sa_hdl;
995 	}
996 
997 	(*zpp)->z_pflags = pflags;
998 	(*zpp)->z_mode = mode;
999 
1000 	if (vap->va_mask & AT_XVATTR)
1001 		zfs_xvattr_set(*zpp, (xvattr_t *)vap, tx);
1002 
1003 	if (obj_type == DMU_OT_ZNODE ||
1004 	    acl_ids->z_aclp->z_version < ZFS_ACL_VERSION_FUID) {
1005 		VERIFY0(zfs_aclset_common(*zpp, acl_ids->z_aclp, cr, tx));
1006 	}
1007 	ZFS_OBJ_HOLD_EXIT(zfsvfs, obj);
1008 }
1009 
1010 /*
1011  * Update in-core attributes.  It is assumed the caller will be doing an
1012  * sa_bulk_update to push the changes out.
1013  */
1014 void
zfs_xvattr_set(znode_t * zp,xvattr_t * xvap,dmu_tx_t * tx)1015 zfs_xvattr_set(znode_t *zp, xvattr_t *xvap, dmu_tx_t *tx)
1016 {
1017 	xoptattr_t *xoap;
1018 
1019 	xoap = xva_getxoptattr(xvap);
1020 	ASSERT(xoap);
1021 
1022 	if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
1023 		uint64_t times[2];
1024 		ZFS_TIME_ENCODE(&xoap->xoa_createtime, times);
1025 		(void) sa_update(zp->z_sa_hdl, SA_ZPL_CRTIME(zp->z_zfsvfs),
1026 		    &times, sizeof (times), tx);
1027 		XVA_SET_RTN(xvap, XAT_CREATETIME);
1028 	}
1029 	if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
1030 		ZFS_ATTR_SET(zp, ZFS_READONLY, xoap->xoa_readonly,
1031 		    zp->z_pflags, tx);
1032 		XVA_SET_RTN(xvap, XAT_READONLY);
1033 	}
1034 	if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
1035 		ZFS_ATTR_SET(zp, ZFS_HIDDEN, xoap->xoa_hidden,
1036 		    zp->z_pflags, tx);
1037 		XVA_SET_RTN(xvap, XAT_HIDDEN);
1038 	}
1039 	if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
1040 		ZFS_ATTR_SET(zp, ZFS_SYSTEM, xoap->xoa_system,
1041 		    zp->z_pflags, tx);
1042 		XVA_SET_RTN(xvap, XAT_SYSTEM);
1043 	}
1044 	if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
1045 		ZFS_ATTR_SET(zp, ZFS_ARCHIVE, xoap->xoa_archive,
1046 		    zp->z_pflags, tx);
1047 		XVA_SET_RTN(xvap, XAT_ARCHIVE);
1048 	}
1049 	if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
1050 		ZFS_ATTR_SET(zp, ZFS_IMMUTABLE, xoap->xoa_immutable,
1051 		    zp->z_pflags, tx);
1052 		XVA_SET_RTN(xvap, XAT_IMMUTABLE);
1053 	}
1054 	if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
1055 		ZFS_ATTR_SET(zp, ZFS_NOUNLINK, xoap->xoa_nounlink,
1056 		    zp->z_pflags, tx);
1057 		XVA_SET_RTN(xvap, XAT_NOUNLINK);
1058 	}
1059 	if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
1060 		ZFS_ATTR_SET(zp, ZFS_APPENDONLY, xoap->xoa_appendonly,
1061 		    zp->z_pflags, tx);
1062 		XVA_SET_RTN(xvap, XAT_APPENDONLY);
1063 	}
1064 	if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
1065 		ZFS_ATTR_SET(zp, ZFS_NODUMP, xoap->xoa_nodump,
1066 		    zp->z_pflags, tx);
1067 		XVA_SET_RTN(xvap, XAT_NODUMP);
1068 	}
1069 	if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
1070 		ZFS_ATTR_SET(zp, ZFS_OPAQUE, xoap->xoa_opaque,
1071 		    zp->z_pflags, tx);
1072 		XVA_SET_RTN(xvap, XAT_OPAQUE);
1073 	}
1074 	if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
1075 		ZFS_ATTR_SET(zp, ZFS_AV_QUARANTINED,
1076 		    xoap->xoa_av_quarantined, zp->z_pflags, tx);
1077 		XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
1078 	}
1079 	if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
1080 		ZFS_ATTR_SET(zp, ZFS_AV_MODIFIED, xoap->xoa_av_modified,
1081 		    zp->z_pflags, tx);
1082 		XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
1083 	}
1084 	if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) {
1085 		zfs_sa_set_scanstamp(zp, xvap, tx);
1086 		XVA_SET_RTN(xvap, XAT_AV_SCANSTAMP);
1087 	}
1088 	if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
1089 		ZFS_ATTR_SET(zp, ZFS_REPARSE, xoap->xoa_reparse,
1090 		    zp->z_pflags, tx);
1091 		XVA_SET_RTN(xvap, XAT_REPARSE);
1092 	}
1093 	if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
1094 		ZFS_ATTR_SET(zp, ZFS_OFFLINE, xoap->xoa_offline,
1095 		    zp->z_pflags, tx);
1096 		XVA_SET_RTN(xvap, XAT_OFFLINE);
1097 	}
1098 	if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
1099 		ZFS_ATTR_SET(zp, ZFS_SPARSE, xoap->xoa_sparse,
1100 		    zp->z_pflags, tx);
1101 		XVA_SET_RTN(xvap, XAT_SPARSE);
1102 	}
1103 }
1104 
1105 int
zfs_zget(zfsvfs_t * zfsvfs,uint64_t obj_num,znode_t ** zpp)1106 zfs_zget(zfsvfs_t *zfsvfs, uint64_t obj_num, znode_t **zpp)
1107 {
1108 	dmu_object_info_t doi;
1109 	dmu_buf_t	*db;
1110 	znode_t		*zp;
1111 	int err;
1112 	sa_handle_t	*hdl;
1113 
1114 	*zpp = NULL;
1115 
1116 	ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num);
1117 
1118 	err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db);
1119 	if (err) {
1120 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1121 		return (err);
1122 	}
1123 
1124 	dmu_object_info_from_db(db, &doi);
1125 	if (doi.doi_bonus_type != DMU_OT_SA &&
1126 	    (doi.doi_bonus_type != DMU_OT_ZNODE ||
1127 	    (doi.doi_bonus_type == DMU_OT_ZNODE &&
1128 	    doi.doi_bonus_size < sizeof (znode_phys_t)))) {
1129 		sa_buf_rele(db, NULL);
1130 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1131 		return (SET_ERROR(EINVAL));
1132 	}
1133 
1134 	hdl = dmu_buf_get_user(db);
1135 	if (hdl != NULL) {
1136 		zp  = sa_get_userdata(hdl);
1137 
1138 
1139 		/*
1140 		 * Since "SA" does immediate eviction we
1141 		 * should never find a sa handle that doesn't
1142 		 * know about the znode.
1143 		 */
1144 
1145 		ASSERT3P(zp, !=, NULL);
1146 
1147 		mutex_enter(&zp->z_lock);
1148 		ASSERT3U(zp->z_id, ==, obj_num);
1149 		if (zp->z_unlinked) {
1150 			err = SET_ERROR(ENOENT);
1151 		} else {
1152 			VN_HOLD(ZTOV(zp));
1153 			*zpp = zp;
1154 			err = 0;
1155 		}
1156 		mutex_exit(&zp->z_lock);
1157 		sa_buf_rele(db, NULL);
1158 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1159 		return (err);
1160 	}
1161 
1162 	/*
1163 	 * Not found create new znode/vnode
1164 	 * but only if file exists.
1165 	 *
1166 	 * There is a small window where zfs_vget() could
1167 	 * find this object while a file create is still in
1168 	 * progress.  This is checked for in zfs_znode_alloc()
1169 	 *
1170 	 * if zfs_znode_alloc() fails it will drop the hold on the
1171 	 * bonus buffer.
1172 	 */
1173 	zp = zfs_znode_alloc(zfsvfs, db, doi.doi_data_block_size,
1174 	    doi.doi_bonus_type, NULL);
1175 	if (zp == NULL) {
1176 		err = SET_ERROR(ENOENT);
1177 	} else {
1178 		if (zp->z_links == 0)
1179 			zp->z_unlinked = B_TRUE;
1180 		*zpp = zp;
1181 	}
1182 	ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1183 	return (err);
1184 }
1185 
1186 int
zfs_rezget(znode_t * zp)1187 zfs_rezget(znode_t *zp)
1188 {
1189 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1190 	dmu_object_info_t doi;
1191 	dmu_buf_t *db;
1192 	uint64_t obj_num = zp->z_id;
1193 	uint64_t mode;
1194 	sa_bulk_attr_t bulk[8];
1195 	int err;
1196 	int count = 0;
1197 	uint64_t gen;
1198 
1199 	ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num);
1200 
1201 	mutex_enter(&zp->z_acl_lock);
1202 	if (zp->z_acl_cached) {
1203 		zfs_acl_free(zp->z_acl_cached);
1204 		zp->z_acl_cached = NULL;
1205 	}
1206 
1207 	mutex_exit(&zp->z_acl_lock);
1208 	ASSERT(zp->z_sa_hdl == NULL);
1209 	err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db);
1210 	if (err) {
1211 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1212 		return (err);
1213 	}
1214 
1215 	dmu_object_info_from_db(db, &doi);
1216 	if (doi.doi_bonus_type != DMU_OT_SA &&
1217 	    (doi.doi_bonus_type != DMU_OT_ZNODE ||
1218 	    (doi.doi_bonus_type == DMU_OT_ZNODE &&
1219 	    doi.doi_bonus_size < sizeof (znode_phys_t)))) {
1220 		sa_buf_rele(db, NULL);
1221 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1222 		return (SET_ERROR(EINVAL));
1223 	}
1224 
1225 	zfs_znode_sa_init(zfsvfs, zp, db, doi.doi_bonus_type, NULL);
1226 
1227 	/* reload cached values */
1228 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL,
1229 	    &gen, sizeof (gen));
1230 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
1231 	    &zp->z_size, sizeof (zp->z_size));
1232 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL,
1233 	    &zp->z_links, sizeof (zp->z_links));
1234 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
1235 	    &zp->z_pflags, sizeof (zp->z_pflags));
1236 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
1237 	    &zp->z_atime, sizeof (zp->z_atime));
1238 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
1239 	    &zp->z_uid, sizeof (zp->z_uid));
1240 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL,
1241 	    &zp->z_gid, sizeof (zp->z_gid));
1242 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
1243 	    &mode, sizeof (mode));
1244 
1245 	if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) {
1246 		zfs_znode_dmu_fini(zp);
1247 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1248 		return (SET_ERROR(EIO));
1249 	}
1250 
1251 	zp->z_mode = mode;
1252 
1253 	if (gen != zp->z_gen) {
1254 		zfs_znode_dmu_fini(zp);
1255 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1256 		return (SET_ERROR(EIO));
1257 	}
1258 
1259 	zp->z_unlinked = (zp->z_links == 0);
1260 	zp->z_blksz = doi.doi_data_block_size;
1261 
1262 	ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1263 
1264 	return (0);
1265 }
1266 
1267 void
zfs_znode_delete(znode_t * zp,dmu_tx_t * tx)1268 zfs_znode_delete(znode_t *zp, dmu_tx_t *tx)
1269 {
1270 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1271 	objset_t *os = zfsvfs->z_os;
1272 	uint64_t obj = zp->z_id;
1273 	uint64_t acl_obj = zfs_external_acl(zp);
1274 
1275 	ZFS_OBJ_HOLD_ENTER(zfsvfs, obj);
1276 	if (acl_obj) {
1277 		VERIFY(!zp->z_is_sa);
1278 		VERIFY(0 == dmu_object_free(os, acl_obj, tx));
1279 	}
1280 	VERIFY(0 == dmu_object_free(os, obj, tx));
1281 	zfs_znode_dmu_fini(zp);
1282 	ZFS_OBJ_HOLD_EXIT(zfsvfs, obj);
1283 	zfs_znode_free(zp);
1284 }
1285 
1286 void
zfs_zinactive(znode_t * zp)1287 zfs_zinactive(znode_t *zp)
1288 {
1289 	vnode_t	*vp = ZTOV(zp);
1290 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1291 	uint64_t z_id = zp->z_id;
1292 
1293 	ASSERT(zp->z_sa_hdl);
1294 
1295 	/*
1296 	 * Don't allow a zfs_zget() while were trying to release this znode
1297 	 */
1298 	ZFS_OBJ_HOLD_ENTER(zfsvfs, z_id);
1299 
1300 	mutex_enter(&zp->z_lock);
1301 	mutex_enter(&vp->v_lock);
1302 	vp->v_count--;
1303 	if (vp->v_count > 0 || vn_has_cached_data(vp)) {
1304 		/*
1305 		 * If the hold count is greater than zero, somebody has
1306 		 * obtained a new reference on this znode while we were
1307 		 * processing it here, so we are done.  If we still have
1308 		 * mapped pages then we are also done, since we don't
1309 		 * want to inactivate the znode until the pages get pushed.
1310 		 *
1311 		 * XXX - if vn_has_cached_data(vp) is true, but count == 0,
1312 		 * this seems like it would leave the znode hanging with
1313 		 * no chance to go inactive...
1314 		 */
1315 		mutex_exit(&vp->v_lock);
1316 		mutex_exit(&zp->z_lock);
1317 		ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id);
1318 		return;
1319 	}
1320 	mutex_exit(&vp->v_lock);
1321 
1322 	/*
1323 	 * If this was the last reference to a file with no links,
1324 	 * remove the file from the file system.
1325 	 */
1326 	if (zp->z_unlinked) {
1327 		mutex_exit(&zp->z_lock);
1328 		ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id);
1329 		zfs_rmnode(zp);
1330 		return;
1331 	}
1332 
1333 	mutex_exit(&zp->z_lock);
1334 	zfs_znode_dmu_fini(zp);
1335 	ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id);
1336 	zfs_znode_free(zp);
1337 }
1338 
1339 void
zfs_znode_free(znode_t * zp)1340 zfs_znode_free(znode_t *zp)
1341 {
1342 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1343 
1344 	vn_invalid(ZTOV(zp));
1345 
1346 	ASSERT(ZTOV(zp)->v_count == 0);
1347 
1348 	mutex_enter(&zfsvfs->z_znodes_lock);
1349 	POINTER_INVALIDATE(&zp->z_zfsvfs);
1350 	list_remove(&zfsvfs->z_all_znodes, zp);
1351 	mutex_exit(&zfsvfs->z_znodes_lock);
1352 
1353 	if (zp->z_acl_cached) {
1354 		zfs_acl_free(zp->z_acl_cached);
1355 		zp->z_acl_cached = NULL;
1356 	}
1357 
1358 	kmem_cache_free(znode_cache, zp);
1359 
1360 	VFS_RELE(zfsvfs->z_vfs);
1361 }
1362 
1363 void
zfs_tstamp_update_setup(znode_t * zp,uint_t flag,uint64_t mtime[2],uint64_t ctime[2],boolean_t have_tx)1364 zfs_tstamp_update_setup(znode_t *zp, uint_t flag, uint64_t mtime[2],
1365     uint64_t ctime[2], boolean_t have_tx)
1366 {
1367 	timestruc_t	now;
1368 
1369 	gethrestime(&now);
1370 
1371 	if (have_tx) {	/* will sa_bulk_update happen really soon? */
1372 		zp->z_atime_dirty = 0;
1373 		zp->z_seq++;
1374 	} else {
1375 		zp->z_atime_dirty = 1;
1376 	}
1377 
1378 	if (flag & AT_ATIME) {
1379 		ZFS_TIME_ENCODE(&now, zp->z_atime);
1380 	}
1381 
1382 	if (flag & AT_MTIME) {
1383 		ZFS_TIME_ENCODE(&now, mtime);
1384 		if (zp->z_zfsvfs->z_use_fuids) {
1385 			zp->z_pflags |= (ZFS_ARCHIVE |
1386 			    ZFS_AV_MODIFIED);
1387 		}
1388 	}
1389 
1390 	if (flag & AT_CTIME) {
1391 		ZFS_TIME_ENCODE(&now, ctime);
1392 		if (zp->z_zfsvfs->z_use_fuids)
1393 			zp->z_pflags |= ZFS_ARCHIVE;
1394 	}
1395 }
1396 
1397 /*
1398  * Grow the block size for a file.
1399  *
1400  *	IN:	zp	- znode of file to free data in.
1401  *		size	- requested block size
1402  *		tx	- open transaction.
1403  *
1404  * NOTE: this function assumes that the znode is write locked.
1405  */
1406 void
zfs_grow_blocksize(znode_t * zp,uint64_t size,dmu_tx_t * tx)1407 zfs_grow_blocksize(znode_t *zp, uint64_t size, dmu_tx_t *tx)
1408 {
1409 	int		error;
1410 	u_longlong_t	dummy;
1411 
1412 	if (size <= zp->z_blksz)
1413 		return;
1414 	/*
1415 	 * If the file size is already greater than the current blocksize,
1416 	 * we will not grow.  If there is more than one block in a file,
1417 	 * the blocksize cannot change.
1418 	 */
1419 	if (zp->z_blksz && zp->z_size > zp->z_blksz)
1420 		return;
1421 
1422 	error = dmu_object_set_blocksize(zp->z_zfsvfs->z_os, zp->z_id,
1423 	    size, 0, tx);
1424 
1425 	if (error == ENOTSUP)
1426 		return;
1427 	ASSERT0(error);
1428 
1429 	/* What blocksize did we actually get? */
1430 	dmu_object_size_from_db(sa_get_db(zp->z_sa_hdl), &zp->z_blksz, &dummy);
1431 }
1432 
1433 /*
1434  * This is a dummy interface used when pvn_vplist_dirty() should *not*
1435  * be calling back into the fs for a putpage().  E.g.: when truncating
1436  * a file, the pages being "thrown away* don't need to be written out.
1437  */
1438 /* ARGSUSED */
1439 static int
zfs_no_putpage(vnode_t * vp,page_t * pp,u_offset_t * offp,size_t * lenp,int flags,cred_t * cr)1440 zfs_no_putpage(vnode_t *vp, page_t *pp, u_offset_t *offp, size_t *lenp,
1441     int flags, cred_t *cr)
1442 {
1443 	ASSERT(0);
1444 	return (0);
1445 }
1446 
1447 /*
1448  * Increase the file length
1449  *
1450  *	IN:	zp	- znode of file to free data in.
1451  *		end	- new end-of-file
1452  *
1453  *	RETURN:	0 on success, error code on failure
1454  *
1455  *	Assumption: File is already range-locked for changing zp_size
1456  */
1457 static int
zfs_extend(znode_t * zp,uint64_t end)1458 zfs_extend(znode_t *zp, uint64_t end)
1459 {
1460 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1461 	dmu_tx_t *tx;
1462 	uint64_t newblksz;
1463 	int error;
1464 
1465 	/*
1466 	 * Nothing to do if file already at desired length.
1467 	 */
1468 	if (end <= zp->z_size)
1469 		return (0);
1470 
1471 	tx = dmu_tx_create(zfsvfs->z_os);
1472 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1473 	zfs_sa_upgrade_txholds(tx, zp);
1474 	if (end > zp->z_blksz &&
1475 	    (!ISP2(zp->z_blksz) || zp->z_blksz < zfsvfs->z_max_blksz)) {
1476 		/*
1477 		 * We are growing the file past the current block size.
1478 		 */
1479 		if (zp->z_blksz > zp->z_zfsvfs->z_max_blksz) {
1480 			/*
1481 			 * File's blocksize is already larger than the
1482 			 * "recordsize" property.  Only let it grow to
1483 			 * the next power of 2.
1484 			 */
1485 			ASSERT(!ISP2(zp->z_blksz));
1486 			newblksz = MIN(end, 1 << highbit64(zp->z_blksz));
1487 		} else {
1488 			newblksz = MIN(end, zp->z_zfsvfs->z_max_blksz);
1489 		}
1490 		dmu_tx_hold_write(tx, zp->z_id, 0, newblksz);
1491 	} else {
1492 		newblksz = 0;
1493 	}
1494 
1495 	error = dmu_tx_assign(tx, TXG_WAIT);
1496 	if (error) {
1497 		dmu_tx_abort(tx);
1498 		return (error);
1499 	}
1500 
1501 	if (newblksz)
1502 		zfs_grow_blocksize(zp, newblksz, tx);
1503 
1504 	zp->z_size = end;
1505 
1506 	VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zp->z_zfsvfs),
1507 	    &zp->z_size, sizeof (zp->z_size), tx));
1508 
1509 	dmu_tx_commit(tx);
1510 
1511 	return (0);
1512 }
1513 
1514 /*
1515  * Free space in a file.
1516  *
1517  *	IN:	zp	- znode of file to free data in.
1518  *		off	- start of section to free.
1519  *		len	- length of section to free.
1520  *
1521  *	RETURN:	0 on success, error code on failure
1522  *
1523  *	Assumption: file is already range-locked for changing zp_size
1524  */
1525 static int
zfs_free_range(znode_t * zp,uint64_t off,uint64_t len)1526 zfs_free_range(znode_t *zp, uint64_t off, uint64_t len)
1527 {
1528 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1529 	int error;
1530 
1531 	/*
1532 	 * Nothing to do if file already at desired length.
1533 	 */
1534 	if (off >= zp->z_size)
1535 		return (0);
1536 
1537 	if (off + len > zp->z_size)
1538 		len = zp->z_size - off;
1539 
1540 	error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, off, len);
1541 
1542 	return (error);
1543 }
1544 
1545 /*
1546  * Truncate a file
1547  *
1548  *	IN:	zp	- znode of file to free data in.
1549  *		end	- new end-of-file.
1550  *
1551  *	RETURN:	0 on success, error code on failure
1552  *
1553  *	Assumption: file is already range-locked for changing zp_size
1554  */
1555 static int
zfs_trunc(znode_t * zp,uint64_t end)1556 zfs_trunc(znode_t *zp, uint64_t end)
1557 {
1558 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1559 	vnode_t *vp = ZTOV(zp);
1560 	dmu_tx_t *tx;
1561 	int error;
1562 	sa_bulk_attr_t bulk[2];
1563 	int count = 0;
1564 
1565 	/*
1566 	 * Nothing to do if file already at desired length.
1567 	 */
1568 	if (end >= zp->z_size)
1569 		return (0);
1570 
1571 	error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, end,  -1);
1572 	if (error)
1573 		return (error);
1574 
1575 	tx = dmu_tx_create(zfsvfs->z_os);
1576 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1577 	zfs_sa_upgrade_txholds(tx, zp);
1578 	dmu_tx_mark_netfree(tx);
1579 	error = dmu_tx_assign(tx, TXG_WAIT);
1580 	if (error) {
1581 		dmu_tx_abort(tx);
1582 		return (error);
1583 	}
1584 
1585 	zp->z_size = end;
1586 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs),
1587 	    NULL, &zp->z_size, sizeof (zp->z_size));
1588 
1589 	if (end == 0) {
1590 		zp->z_pflags &= ~ZFS_SPARSE;
1591 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
1592 		    NULL, &zp->z_pflags, 8);
1593 	}
1594 	VERIFY(sa_bulk_update(zp->z_sa_hdl, bulk, count, tx) == 0);
1595 
1596 	dmu_tx_commit(tx);
1597 
1598 	/*
1599 	 * Clear any mapped pages in the truncated region.  This has to
1600 	 * happen outside of the transaction to avoid the possibility of
1601 	 * a deadlock with someone trying to push a page that we are
1602 	 * about to invalidate.
1603 	 */
1604 	if (vn_has_cached_data(vp)) {
1605 		page_t *pp;
1606 		uint64_t start = end & PAGEMASK;
1607 		int poff = end & PAGEOFFSET;
1608 
1609 		if (poff != 0 && (pp = page_lookup(vp, start, SE_SHARED))) {
1610 			/*
1611 			 * We need to zero a partial page.
1612 			 */
1613 			pagezero(pp, poff, PAGESIZE - poff);
1614 			start += PAGESIZE;
1615 			page_unlock(pp);
1616 		}
1617 		error = pvn_vplist_dirty(vp, start, zfs_no_putpage,
1618 		    B_INVAL | B_TRUNC, NULL);
1619 		ASSERT(error == 0);
1620 	}
1621 
1622 	return (0);
1623 }
1624 
1625 /*
1626  * Free space in a file
1627  *
1628  *	IN:	zp	- znode of file to free data in.
1629  *		off	- start of range
1630  *		len	- end of range (0 => EOF)
1631  *		flag	- current file open mode flags.
1632  *		log	- TRUE if this action should be logged
1633  *
1634  *	RETURN:	0 on success, error code on failure
1635  */
1636 int
zfs_freesp(znode_t * zp,uint64_t off,uint64_t len,int flag,boolean_t log)1637 zfs_freesp(znode_t *zp, uint64_t off, uint64_t len, int flag, boolean_t log)
1638 {
1639 	vnode_t *vp = ZTOV(zp);
1640 	dmu_tx_t *tx;
1641 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1642 	zilog_t *zilog = zfsvfs->z_log;
1643 	uint64_t mode;
1644 	uint64_t mtime[2], ctime[2];
1645 	sa_bulk_attr_t bulk[3];
1646 	int count = 0;
1647 	int error;
1648 	rl_t *rl;
1649 
1650 	if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs), &mode,
1651 	    sizeof (mode))) != 0)
1652 		return (error);
1653 
1654 	if (off > zp->z_size) {
1655 		rl = zfs_range_lock(zp, 0, UINT64_MAX, RL_WRITER);
1656 		error =  zfs_extend(zp, off+len);
1657 		if (error == 0 && log)
1658 			goto log;
1659 
1660 		zfs_range_unlock(rl);
1661 		return (error);
1662 	}
1663 
1664 	/*
1665 	 * Check for any locks in the region to be freed.
1666 	 */
1667 	if (MANDLOCK(vp, (mode_t)mode)) {
1668 		uint64_t length = (len ? len : zp->z_size - off);
1669 		if (error = chklock(vp, FWRITE, off, length, flag, NULL))
1670 			return (error);
1671 	}
1672 
1673 	if (len == 0 || off + len > zp->z_size) {
1674 		rl = zfs_range_lock(zp, 0, UINT64_MAX, RL_WRITER);
1675 	} else {
1676 		ssize_t	lock_off;
1677 		ssize_t lock_len;
1678 #ifdef _KERNEL
1679 		if (rz_zev_active()) {
1680 			/* start of this megabyte */
1681 			lock_off = P2ALIGN(off, ZEV_L1_SIZE);
1682 			/* full megabytes */
1683 			lock_len = len + (off - lock_off);
1684 			lock_len = P2ROUNDUP(lock_len, ZEV_L1_SIZE);
1685 		} else {
1686 			lock_off = off;
1687 			lock_len = len;
1688 		}
1689 #else
1690 		lock_off = off;
1691 		lock_len = len;
1692 #endif
1693 		rl = zfs_range_lock(zp, lock_off, lock_len, RL_WRITER);
1694 	}
1695 
1696 	if (len == 0) {
1697 		error = zfs_trunc(zp, off);
1698 	} else {
1699 		if ((error = zfs_free_range(zp, off, len)) == 0 &&
1700 		    off + len > zp->z_size)
1701 			error = zfs_extend(zp, off+len);
1702 	}
1703 
1704 	if (error || !log) {
1705 		zfs_range_unlock(rl);
1706 		return (error);
1707 	}
1708 log:
1709 	tx = dmu_tx_create(zfsvfs->z_os);
1710 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1711 	zfs_sa_upgrade_txholds(tx, zp);
1712 	error = dmu_tx_assign(tx, TXG_WAIT);
1713 	if (error) {
1714 		dmu_tx_abort(tx);
1715 		zfs_range_unlock(rl);
1716 		return (error);
1717 	}
1718 
1719 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, mtime, 16);
1720 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, ctime, 16);
1721 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
1722 	    NULL, &zp->z_pflags, 8);
1723 	zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, B_TRUE);
1724 	error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
1725 	ASSERT(error == 0);
1726 
1727 	zfs_log_truncate(zilog, tx, TX_TRUNCATE, zp, off, len);
1728 	zfs_range_unlock(rl);
1729 
1730 	dmu_tx_commit(tx);
1731 	return (0);
1732 }
1733 
1734 void
zfs_create_fs(objset_t * os,cred_t * cr,nvlist_t * zplprops,dmu_tx_t * tx)1735 zfs_create_fs(objset_t *os, cred_t *cr, nvlist_t *zplprops, dmu_tx_t *tx)
1736 {
1737 	uint64_t	moid, obj, sa_obj, version;
1738 	uint64_t	sense = ZFS_CASE_SENSITIVE;
1739 	uint64_t	norm = 0;
1740 	nvpair_t	*elem;
1741 	int		error;
1742 	int		i;
1743 	znode_t		*rootzp = NULL;
1744 	zfsvfs_t	*zfsvfs;
1745 	vnode_t		*vp;
1746 	vattr_t		vattr;
1747 	znode_t		*zp;
1748 	zfs_acl_ids_t	acl_ids;
1749 
1750 	/*
1751 	 * First attempt to create master node.
1752 	 */
1753 	/*
1754 	 * In an empty objset, there are no blocks to read and thus
1755 	 * there can be no i/o errors (which we assert below).
1756 	 */
1757 	moid = MASTER_NODE_OBJ;
1758 	error = zap_create_claim(os, moid, DMU_OT_MASTER_NODE,
1759 	    DMU_OT_NONE, 0, tx);
1760 	ASSERT(error == 0);
1761 
1762 	/*
1763 	 * Set starting attributes.
1764 	 */
1765 	version = zfs_zpl_version_map(spa_version(dmu_objset_spa(os)));
1766 	elem = NULL;
1767 	while ((elem = nvlist_next_nvpair(zplprops, elem)) != NULL) {
1768 		/* For the moment we expect all zpl props to be uint64_ts */
1769 		uint64_t val;
1770 		char *name;
1771 
1772 		ASSERT(nvpair_type(elem) == DATA_TYPE_UINT64);
1773 		VERIFY(nvpair_value_uint64(elem, &val) == 0);
1774 		name = nvpair_name(elem);
1775 		if (strcmp(name, zfs_prop_to_name(ZFS_PROP_VERSION)) == 0) {
1776 			if (val < version)
1777 				version = val;
1778 		} else {
1779 			error = zap_update(os, moid, name, 8, 1, &val, tx);
1780 		}
1781 		ASSERT(error == 0);
1782 		if (strcmp(name, zfs_prop_to_name(ZFS_PROP_NORMALIZE)) == 0)
1783 			norm = val;
1784 		else if (strcmp(name, zfs_prop_to_name(ZFS_PROP_CASE)) == 0)
1785 			sense = val;
1786 	}
1787 	ASSERT(version != 0);
1788 	error = zap_update(os, moid, ZPL_VERSION_STR, 8, 1, &version, tx);
1789 
1790 	/*
1791 	 * Create zap object used for SA attribute registration
1792 	 */
1793 
1794 	if (version >= ZPL_VERSION_SA) {
1795 		sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
1796 		    DMU_OT_NONE, 0, tx);
1797 		error = zap_add(os, moid, ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
1798 		ASSERT(error == 0);
1799 	} else {
1800 		sa_obj = 0;
1801 	}
1802 	/*
1803 	 * Create a delete queue.
1804 	 */
1805 	obj = zap_create(os, DMU_OT_UNLINKED_SET, DMU_OT_NONE, 0, tx);
1806 
1807 	error = zap_add(os, moid, ZFS_UNLINKED_SET, 8, 1, &obj, tx);
1808 	ASSERT(error == 0);
1809 
1810 	/*
1811 	 * Create root znode.  Create minimal znode/vnode/zfsvfs
1812 	 * to allow zfs_mknode to work.
1813 	 */
1814 	vattr.va_mask = AT_MODE|AT_UID|AT_GID|AT_TYPE;
1815 	vattr.va_type = VDIR;
1816 	vattr.va_mode = S_IFDIR|0755;
1817 	vattr.va_uid = crgetuid(cr);
1818 	vattr.va_gid = crgetgid(cr);
1819 
1820 	rootzp = kmem_cache_alloc(znode_cache, KM_SLEEP);
1821 	ASSERT(!POINTER_IS_VALID(rootzp->z_zfsvfs));
1822 	rootzp->z_moved = 0;
1823 	rootzp->z_unlinked = 0;
1824 	rootzp->z_atime_dirty = 0;
1825 	rootzp->z_is_sa = USE_SA(version, os);
1826 
1827 	vp = ZTOV(rootzp);
1828 	vn_reinit(vp);
1829 	vp->v_type = VDIR;
1830 
1831 	zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
1832 	zfsvfs->z_os = os;
1833 	zfsvfs->z_parent = zfsvfs;
1834 	zfsvfs->z_version = version;
1835 	zfsvfs->z_use_fuids = USE_FUIDS(version, os);
1836 	zfsvfs->z_use_sa = USE_SA(version, os);
1837 	zfsvfs->z_norm = norm;
1838 
1839 	error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
1840 	    &zfsvfs->z_attr_table);
1841 
1842 	ASSERT(error == 0);
1843 
1844 	/*
1845 	 * Fold case on file systems that are always or sometimes case
1846 	 * insensitive.
1847 	 */
1848 	if (sense == ZFS_CASE_INSENSITIVE || sense == ZFS_CASE_MIXED)
1849 		zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER;
1850 
1851 	mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
1852 	list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
1853 	    offsetof(znode_t, z_link_node));
1854 
1855 	for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1856 		mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL);
1857 
1858 	rootzp->z_zfsvfs = zfsvfs;
1859 	VERIFY(0 == zfs_acl_ids_create(rootzp, IS_ROOT_NODE, &vattr,
1860 	    cr, NULL, &acl_ids));
1861 	zfs_mknode(rootzp, &vattr, tx, cr, IS_ROOT_NODE, &zp, &acl_ids);
1862 	ASSERT3P(zp, ==, rootzp);
1863 	ASSERT(!vn_in_dnlc(ZTOV(rootzp))); /* not valid to move */
1864 	error = zap_add(os, moid, ZFS_ROOT_OBJ, 8, 1, &rootzp->z_id, tx);
1865 	ASSERT(error == 0);
1866 	zfs_acl_ids_free(&acl_ids);
1867 	POINTER_INVALIDATE(&rootzp->z_zfsvfs);
1868 
1869 	ZTOV(rootzp)->v_count = 0;
1870 	sa_handle_destroy(rootzp->z_sa_hdl);
1871 	kmem_cache_free(znode_cache, rootzp);
1872 
1873 	/*
1874 	 * Create shares directory
1875 	 */
1876 
1877 	error = zfs_create_share_dir(zfsvfs, tx);
1878 
1879 	ASSERT(error == 0);
1880 
1881 	for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1882 		mutex_destroy(&zfsvfs->z_hold_mtx[i]);
1883 	kmem_free(zfsvfs, sizeof (zfsvfs_t));
1884 }
1885 
1886 #endif /* _KERNEL */
1887 
1888 static int
zfs_sa_setup(objset_t * osp,sa_attr_type_t ** sa_table)1889 zfs_sa_setup(objset_t *osp, sa_attr_type_t **sa_table)
1890 {
1891 	uint64_t sa_obj = 0;
1892 	int error;
1893 
1894 	error = zap_lookup(osp, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1, &sa_obj);
1895 	if (error != 0 && error != ENOENT)
1896 		return (error);
1897 
1898 	error = sa_setup(osp, sa_obj, zfs_attr_table, ZPL_END, sa_table);
1899 	return (error);
1900 }
1901 
1902 static int
zfs_grab_sa_handle(objset_t * osp,uint64_t obj,sa_handle_t ** hdlp,dmu_buf_t ** db,void * tag)1903 zfs_grab_sa_handle(objset_t *osp, uint64_t obj, sa_handle_t **hdlp,
1904     dmu_buf_t **db, void *tag)
1905 {
1906 	dmu_object_info_t doi;
1907 	int error;
1908 
1909 	if ((error = sa_buf_hold(osp, obj, tag, db)) != 0)
1910 		return (error);
1911 
1912 	dmu_object_info_from_db(*db, &doi);
1913 	if ((doi.doi_bonus_type != DMU_OT_SA &&
1914 	    doi.doi_bonus_type != DMU_OT_ZNODE) ||
1915 	    doi.doi_bonus_type == DMU_OT_ZNODE &&
1916 	    doi.doi_bonus_size < sizeof (znode_phys_t)) {
1917 		sa_buf_rele(*db, tag);
1918 		return (SET_ERROR(ENOTSUP));
1919 	}
1920 
1921 	error = sa_handle_get(osp, obj, NULL, SA_HDL_PRIVATE, hdlp);
1922 	if (error != 0) {
1923 		sa_buf_rele(*db, tag);
1924 		return (error);
1925 	}
1926 
1927 	return (0);
1928 }
1929 
1930 void
zfs_release_sa_handle(sa_handle_t * hdl,dmu_buf_t * db,void * tag)1931 zfs_release_sa_handle(sa_handle_t *hdl, dmu_buf_t *db, void *tag)
1932 {
1933 	sa_handle_destroy(hdl);
1934 	sa_buf_rele(db, tag);
1935 }
1936 
1937 /*
1938  * Given an object number, return its parent object number and whether
1939  * or not the object is an extended attribute directory.
1940  */
1941 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)1942 zfs_obj_to_pobj(objset_t *osp, sa_handle_t *hdl, sa_attr_type_t *sa_table,
1943     uint64_t *pobjp, int *is_xattrdir)
1944 {
1945 	uint64_t parent;
1946 	uint64_t pflags;
1947 	uint64_t mode;
1948 	uint64_t parent_mode;
1949 	sa_bulk_attr_t bulk[3];
1950 	sa_handle_t *sa_hdl;
1951 	dmu_buf_t *sa_db;
1952 	int count = 0;
1953 	int error;
1954 
1955 	SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_PARENT], NULL,
1956 	    &parent, sizeof (parent));
1957 	SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_FLAGS], NULL,
1958 	    &pflags, sizeof (pflags));
1959 	SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL,
1960 	    &mode, sizeof (mode));
1961 
1962 	if ((error = sa_bulk_lookup(hdl, bulk, count)) != 0)
1963 		return (error);
1964 
1965 	/*
1966 	 * When a link is removed its parent pointer is not changed and will
1967 	 * be invalid.  There are two cases where a link is removed but the
1968 	 * file stays around, when it goes to the delete queue and when there
1969 	 * are additional links.
1970 	 */
1971 	error = zfs_grab_sa_handle(osp, parent, &sa_hdl, &sa_db, FTAG);
1972 	if (error != 0)
1973 		return (error);
1974 
1975 	error = sa_lookup(sa_hdl, ZPL_MODE, &parent_mode, sizeof (parent_mode));
1976 	zfs_release_sa_handle(sa_hdl, sa_db, FTAG);
1977 	if (error != 0)
1978 		return (error);
1979 
1980 	*is_xattrdir = ((pflags & ZFS_XATTR) != 0) && S_ISDIR(mode);
1981 
1982 	/*
1983 	 * Extended attributes can be applied to files, directories, etc.
1984 	 * Otherwise the parent must be a directory.
1985 	 */
1986 	if (!*is_xattrdir && !S_ISDIR(parent_mode))
1987 		return (SET_ERROR(EINVAL));
1988 
1989 	*pobjp = parent;
1990 
1991 	return (0);
1992 }
1993 
1994 /*
1995  * Given an object number, return some zpl level statistics
1996  */
1997 static int
zfs_obj_to_stats_impl(sa_handle_t * hdl,sa_attr_type_t * sa_table,zfs_stat_t * sb)1998 zfs_obj_to_stats_impl(sa_handle_t *hdl, sa_attr_type_t *sa_table,
1999     zfs_stat_t *sb)
2000 {
2001 	sa_bulk_attr_t bulk[4];
2002 	int count = 0;
2003 
2004 	SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL,
2005 	    &sb->zs_mode, sizeof (sb->zs_mode));
2006 	SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_GEN], NULL,
2007 	    &sb->zs_gen, sizeof (sb->zs_gen));
2008 	SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_LINKS], NULL,
2009 	    &sb->zs_links, sizeof (sb->zs_links));
2010 	SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_CTIME], NULL,
2011 	    &sb->zs_ctime, sizeof (sb->zs_ctime));
2012 
2013 	return (sa_bulk_lookup(hdl, bulk, count));
2014 }
2015 
2016 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)2017 zfs_obj_to_path_impl(objset_t *osp, uint64_t obj, sa_handle_t *hdl,
2018     sa_attr_type_t *sa_table, char *buf, int len)
2019 {
2020 	sa_handle_t *sa_hdl;
2021 	sa_handle_t *prevhdl = NULL;
2022 	dmu_buf_t *prevdb = NULL;
2023 	dmu_buf_t *sa_db = NULL;
2024 	char *path = buf + len - 1;
2025 	int error;
2026 
2027 	*path = '\0';
2028 	sa_hdl = hdl;
2029 
2030 	for (;;) {
2031 		uint64_t pobj;
2032 		char component[MAXNAMELEN + 2];
2033 		size_t complen;
2034 		int is_xattrdir;
2035 
2036 		if (prevdb)
2037 			zfs_release_sa_handle(prevhdl, prevdb, FTAG);
2038 
2039 		if ((error = zfs_obj_to_pobj(osp, sa_hdl, sa_table, &pobj,
2040 		    &is_xattrdir)) != 0)
2041 			break;
2042 
2043 		if (pobj == obj) {
2044 			if (path[0] != '/')
2045 				*--path = '/';
2046 			break;
2047 		}
2048 
2049 		component[0] = '/';
2050 		if (is_xattrdir) {
2051 			(void) sprintf(component + 1, "<xattrdir>");
2052 		} else {
2053 			error = zap_value_search(osp, pobj, obj,
2054 			    ZFS_DIRENT_OBJ(-1ULL), component + 1);
2055 			if (error != 0)
2056 				break;
2057 		}
2058 
2059 		complen = strlen(component);
2060 		path -= complen;
2061 		ASSERT(path >= buf);
2062 		bcopy(component, path, complen);
2063 		obj = pobj;
2064 
2065 		if (sa_hdl != hdl) {
2066 			prevhdl = sa_hdl;
2067 			prevdb = sa_db;
2068 		}
2069 		error = zfs_grab_sa_handle(osp, obj, &sa_hdl, &sa_db, FTAG);
2070 		if (error != 0) {
2071 			sa_hdl = prevhdl;
2072 			sa_db = prevdb;
2073 			break;
2074 		}
2075 	}
2076 
2077 	if (sa_hdl != NULL && sa_hdl != hdl) {
2078 		ASSERT(sa_db != NULL);
2079 		zfs_release_sa_handle(sa_hdl, sa_db, FTAG);
2080 	}
2081 
2082 	if (error == 0)
2083 		(void) memmove(buf, path, buf + len - path);
2084 
2085 	return (error);
2086 }
2087 
2088 int
zfs_obj_to_path(objset_t * osp,uint64_t obj,char * buf,int len)2089 zfs_obj_to_path(objset_t *osp, uint64_t obj, char *buf, int len)
2090 {
2091 	sa_attr_type_t *sa_table;
2092 	sa_handle_t *hdl;
2093 	dmu_buf_t *db;
2094 	int error;
2095 
2096 	error = zfs_sa_setup(osp, &sa_table);
2097 	if (error != 0)
2098 		return (error);
2099 
2100 	error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG);
2101 	if (error != 0)
2102 		return (error);
2103 
2104 	error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len);
2105 
2106 	zfs_release_sa_handle(hdl, db, FTAG);
2107 	return (error);
2108 }
2109 
2110 int
zfs_obj_to_stats(objset_t * osp,uint64_t obj,zfs_stat_t * sb,char * buf,int len)2111 zfs_obj_to_stats(objset_t *osp, uint64_t obj, zfs_stat_t *sb,
2112     char *buf, int len)
2113 {
2114 	char *path = buf + len - 1;
2115 	sa_attr_type_t *sa_table;
2116 	sa_handle_t *hdl;
2117 	dmu_buf_t *db;
2118 	int error;
2119 
2120 	*path = '\0';
2121 
2122 	error = zfs_sa_setup(osp, &sa_table);
2123 	if (error != 0)
2124 		return (error);
2125 
2126 	error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG);
2127 	if (error != 0)
2128 		return (error);
2129 
2130 	error = zfs_obj_to_stats_impl(hdl, sa_table, sb);
2131 	if (error != 0) {
2132 		zfs_release_sa_handle(hdl, db, FTAG);
2133 		return (error);
2134 	}
2135 
2136 	error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len);
2137 
2138 	zfs_release_sa_handle(hdl, db, FTAG);
2139 	return (error);
2140 }
2141