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, 2018 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/dmu_tx.h>
64 #include <sys/refcount.h>
65 #include <sys/stat.h>
66 #include <sys/zap.h>
67 #include <sys/zfs_znode.h>
68 #include <sys/sa.h>
69 #include <sys/zfs_sa.h>
70 #include <sys/zfs_stat.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 /*
104 * This is used by the test suite so that it can delay znodes from being
105 * freed in order to inspect the unlinked set.
106 */
107 int zfs_unlink_suspend_progress = 0;
108
109 /*ARGSUSED*/
110 static void
znode_evict_error(dmu_buf_t * dbuf,void * user_ptr)111 znode_evict_error(dmu_buf_t *dbuf, void *user_ptr)
112 {
113 /*
114 * We should never drop all dbuf refs without first clearing
115 * the eviction callback.
116 */
117 panic("evicting znode %p\n", user_ptr);
118 }
119
120 /*
121 * This callback is invoked when acquiring a RL_WRITER or RL_APPEND lock on
122 * z_rangelock. It will modify the offset and length of the lock to reflect
123 * znode-specific information, and convert RL_APPEND to RL_WRITER. This is
124 * called with the rangelock_t's rl_lock held, which avoids races.
125 */
126 static void
zfs_rangelock_cb(locked_range_t * new,void * arg)127 zfs_rangelock_cb(locked_range_t *new, void *arg)
128 {
129 znode_t *zp = arg;
130
131 /*
132 * If in append mode, convert to writer and lock starting at the
133 * current end of file.
134 */
135 if (new->lr_type == RL_APPEND) {
136 new->lr_offset = zp->z_size;
137 new->lr_type = RL_WRITER;
138 }
139
140 /*
141 * If we need to grow the block size then lock the whole file range.
142 */
143 uint64_t end_size = MAX(zp->z_size, new->lr_offset + new->lr_length);
144 if (end_size > zp->z_blksz && (!ISP2(zp->z_blksz) ||
145 zp->z_blksz < zp->z_zfsvfs->z_max_blksz)) {
146 new->lr_offset = 0;
147 new->lr_length = UINT64_MAX;
148 }
149 }
150
151 /*ARGSUSED*/
152 static int
zfs_znode_cache_constructor(void * buf,void * arg,int kmflags)153 zfs_znode_cache_constructor(void *buf, void *arg, int kmflags)
154 {
155 znode_t *zp = buf;
156
157 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs));
158
159 zp->z_vnode = vn_alloc(kmflags);
160 if (zp->z_vnode == NULL) {
161 return (-1);
162 }
163 ZTOV(zp)->v_data = zp;
164
165 list_link_init(&zp->z_link_node);
166
167 mutex_init(&zp->z_lock, NULL, MUTEX_DEFAULT, NULL);
168 rw_init(&zp->z_parent_lock, NULL, RW_DEFAULT, NULL);
169 rw_init(&zp->z_name_lock, NULL, RW_DEFAULT, NULL);
170 rw_init(&zp->z_acl_lock, NULL, RW_DEFAULT, NULL);
171
172 rangelock_init(&zp->z_rangelock, zfs_rangelock_cb, zp);
173
174 zp->z_dirlocks = NULL;
175 zp->z_acl_cached = NULL;
176 zp->z_moved = 0;
177 return (0);
178 }
179
180 /*ARGSUSED*/
181 static void
zfs_znode_cache_destructor(void * buf,void * arg)182 zfs_znode_cache_destructor(void *buf, void *arg)
183 {
184 znode_t *zp = buf;
185
186 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs));
187 ASSERT(ZTOV(zp)->v_data == zp);
188 vn_free(ZTOV(zp));
189 ASSERT(!list_link_active(&zp->z_link_node));
190 mutex_destroy(&zp->z_lock);
191 rw_destroy(&zp->z_parent_lock);
192 rw_destroy(&zp->z_name_lock);
193 rw_destroy(&zp->z_acl_lock);
194 rangelock_fini(&zp->z_rangelock);
195
196 ASSERT(zp->z_dirlocks == NULL);
197 ASSERT(zp->z_acl_cached == NULL);
198 }
199
200 #ifdef ZNODE_STATS
201 static struct {
202 uint64_t zms_zfsvfs_invalid;
203 uint64_t zms_zfsvfs_recheck1;
204 uint64_t zms_zfsvfs_unmounted;
205 uint64_t zms_zfsvfs_recheck2;
206 uint64_t zms_obj_held;
207 uint64_t zms_vnode_locked;
208 uint64_t zms_not_only_dnlc;
209 } znode_move_stats;
210 #endif /* ZNODE_STATS */
211
212 static void
zfs_znode_move_impl(znode_t * ozp,znode_t * nzp)213 zfs_znode_move_impl(znode_t *ozp, znode_t *nzp)
214 {
215 vnode_t *vp;
216
217 /* Copy fields. */
218 nzp->z_zfsvfs = ozp->z_zfsvfs;
219
220 /* Swap vnodes. */
221 vp = nzp->z_vnode;
222 nzp->z_vnode = ozp->z_vnode;
223 ozp->z_vnode = vp; /* let destructor free the overwritten vnode */
224 ZTOV(ozp)->v_data = ozp;
225 ZTOV(nzp)->v_data = nzp;
226
227 nzp->z_id = ozp->z_id;
228 ASSERT(ozp->z_dirlocks == NULL); /* znode not in use */
229 nzp->z_unlinked = ozp->z_unlinked;
230 nzp->z_atime_dirty = ozp->z_atime_dirty;
231 nzp->z_zn_prefetch = ozp->z_zn_prefetch;
232 nzp->z_blksz = ozp->z_blksz;
233 nzp->z_seq = ozp->z_seq;
234 nzp->z_mapcnt = ozp->z_mapcnt;
235 nzp->z_gen = ozp->z_gen;
236 nzp->z_sync_cnt = ozp->z_sync_cnt;
237 nzp->z_is_sa = ozp->z_is_sa;
238 nzp->z_sa_hdl = ozp->z_sa_hdl;
239 bcopy(ozp->z_atime, nzp->z_atime, sizeof (uint64_t) * 2);
240 nzp->z_links = ozp->z_links;
241 nzp->z_size = ozp->z_size;
242 nzp->z_pflags = ozp->z_pflags;
243 nzp->z_uid = ozp->z_uid;
244 nzp->z_gid = ozp->z_gid;
245 nzp->z_mode = ozp->z_mode;
246
247 /*
248 * Since this is just an idle znode and kmem is already dealing with
249 * memory pressure, release any cached ACL.
250 */
251 if (ozp->z_acl_cached) {
252 zfs_acl_free(ozp->z_acl_cached);
253 ozp->z_acl_cached = NULL;
254 }
255
256 sa_set_userp(nzp->z_sa_hdl, nzp);
257
258 /*
259 * Invalidate the original znode by clearing fields that provide a
260 * pointer back to the znode. Set the low bit of the vfs pointer to
261 * ensure that zfs_znode_move() recognizes the znode as invalid in any
262 * subsequent callback.
263 */
264 ozp->z_sa_hdl = NULL;
265 POINTER_INVALIDATE(&ozp->z_zfsvfs);
266
267 /*
268 * Mark the znode.
269 */
270 nzp->z_moved = 1;
271 ozp->z_moved = (uint8_t)-1;
272 }
273
274 /*ARGSUSED*/
275 static kmem_cbrc_t
zfs_znode_move(void * buf,void * newbuf,size_t size,void * arg)276 zfs_znode_move(void *buf, void *newbuf, size_t size, void *arg)
277 {
278 znode_t *ozp = buf, *nzp = newbuf;
279 zfsvfs_t *zfsvfs;
280 vnode_t *vp;
281
282 /*
283 * The znode is on the file system's list of known znodes if the vfs
284 * pointer is valid. We set the low bit of the vfs pointer when freeing
285 * the znode to invalidate it, and the memory patterns written by kmem
286 * (baddcafe and deadbeef) set at least one of the two low bits. A newly
287 * created znode sets the vfs pointer last of all to indicate that the
288 * znode is known and in a valid state to be moved by this function.
289 */
290 zfsvfs = ozp->z_zfsvfs;
291 if (!POINTER_IS_VALID(zfsvfs)) {
292 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_invalid);
293 return (KMEM_CBRC_DONT_KNOW);
294 }
295
296 /*
297 * Close a small window in which it's possible that the filesystem could
298 * be unmounted and freed, and zfsvfs, though valid in the previous
299 * statement, could point to unrelated memory by the time we try to
300 * prevent the filesystem from being unmounted.
301 */
302 rw_enter(&zfsvfs_lock, RW_WRITER);
303 if (zfsvfs != ozp->z_zfsvfs) {
304 rw_exit(&zfsvfs_lock);
305 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_recheck1);
306 return (KMEM_CBRC_DONT_KNOW);
307 }
308
309 /*
310 * If the znode is still valid, then so is the file system. We know that
311 * no valid file system can be freed while we hold zfsvfs_lock, so we
312 * can safely ensure that the filesystem is not and will not be
313 * unmounted. The next statement is equivalent to ZFS_ENTER().
314 */
315 rrm_enter(&zfsvfs->z_teardown_lock, RW_READER, FTAG);
316 if (zfsvfs->z_unmounted) {
317 ZFS_EXIT(zfsvfs);
318 rw_exit(&zfsvfs_lock);
319 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_unmounted);
320 return (KMEM_CBRC_DONT_KNOW);
321 }
322 rw_exit(&zfsvfs_lock);
323
324 mutex_enter(&zfsvfs->z_znodes_lock);
325 /*
326 * Recheck the vfs pointer in case the znode was removed just before
327 * acquiring the lock.
328 */
329 if (zfsvfs != ozp->z_zfsvfs) {
330 mutex_exit(&zfsvfs->z_znodes_lock);
331 ZFS_EXIT(zfsvfs);
332 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_recheck2);
333 return (KMEM_CBRC_DONT_KNOW);
334 }
335
336 /*
337 * At this point we know that as long as we hold z_znodes_lock, the
338 * znode cannot be freed and fields within the znode can be safely
339 * accessed. Now, prevent a race with zfs_zget().
340 */
341 if (ZFS_OBJ_HOLD_TRYENTER(zfsvfs, ozp->z_id) == 0) {
342 mutex_exit(&zfsvfs->z_znodes_lock);
343 ZFS_EXIT(zfsvfs);
344 ZNODE_STAT_ADD(znode_move_stats.zms_obj_held);
345 return (KMEM_CBRC_LATER);
346 }
347
348 vp = ZTOV(ozp);
349 if (mutex_tryenter(&vp->v_lock) == 0) {
350 ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id);
351 mutex_exit(&zfsvfs->z_znodes_lock);
352 ZFS_EXIT(zfsvfs);
353 ZNODE_STAT_ADD(znode_move_stats.zms_vnode_locked);
354 return (KMEM_CBRC_LATER);
355 }
356
357 /* Only move znodes that are referenced _only_ by the DNLC. */
358 if (vp->v_count != 1 || !vn_in_dnlc(vp)) {
359 mutex_exit(&vp->v_lock);
360 ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id);
361 mutex_exit(&zfsvfs->z_znodes_lock);
362 ZFS_EXIT(zfsvfs);
363 ZNODE_STAT_ADD(znode_move_stats.zms_not_only_dnlc);
364 return (KMEM_CBRC_LATER);
365 }
366
367 /*
368 * The znode is known and in a valid state to move. We're holding the
369 * locks needed to execute the critical section.
370 */
371 zfs_znode_move_impl(ozp, nzp);
372 mutex_exit(&vp->v_lock);
373 ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id);
374
375 list_link_replace(&ozp->z_link_node, &nzp->z_link_node);
376 mutex_exit(&zfsvfs->z_znodes_lock);
377 ZFS_EXIT(zfsvfs);
378
379 return (KMEM_CBRC_YES);
380 }
381
382 void
zfs_znode_init(void)383 zfs_znode_init(void)
384 {
385 /*
386 * Initialize zcache
387 */
388 rw_init(&zfsvfs_lock, NULL, RW_DEFAULT, NULL);
389 ASSERT(znode_cache == NULL);
390 znode_cache = kmem_cache_create("zfs_znode_cache",
391 sizeof (znode_t), 0, zfs_znode_cache_constructor,
392 zfs_znode_cache_destructor, NULL, NULL, NULL, 0);
393 kmem_cache_set_move(znode_cache, zfs_znode_move);
394 }
395
396 void
zfs_znode_fini(void)397 zfs_znode_fini(void)
398 {
399 /*
400 * Cleanup vfs & vnode ops
401 */
402 zfs_remove_op_tables();
403
404 /*
405 * Cleanup zcache
406 */
407 if (znode_cache)
408 kmem_cache_destroy(znode_cache);
409 znode_cache = NULL;
410 rw_destroy(&zfsvfs_lock);
411 }
412
413 struct vnodeops *zfs_dvnodeops;
414 struct vnodeops *zfs_fvnodeops;
415 struct vnodeops *zfs_symvnodeops;
416 struct vnodeops *zfs_xdvnodeops;
417 struct vnodeops *zfs_evnodeops;
418 struct vnodeops *zfs_sharevnodeops;
419
420 void
zfs_remove_op_tables()421 zfs_remove_op_tables()
422 {
423 /*
424 * Remove vfs ops
425 */
426 ASSERT(zfsfstype);
427 (void) vfs_freevfsops_by_type(zfsfstype);
428 zfsfstype = 0;
429
430 /*
431 * Remove vnode ops
432 */
433 if (zfs_dvnodeops)
434 vn_freevnodeops(zfs_dvnodeops);
435 if (zfs_fvnodeops)
436 vn_freevnodeops(zfs_fvnodeops);
437 if (zfs_symvnodeops)
438 vn_freevnodeops(zfs_symvnodeops);
439 if (zfs_xdvnodeops)
440 vn_freevnodeops(zfs_xdvnodeops);
441 if (zfs_evnodeops)
442 vn_freevnodeops(zfs_evnodeops);
443 if (zfs_sharevnodeops)
444 vn_freevnodeops(zfs_sharevnodeops);
445
446 zfs_dvnodeops = NULL;
447 zfs_fvnodeops = NULL;
448 zfs_symvnodeops = NULL;
449 zfs_xdvnodeops = NULL;
450 zfs_evnodeops = NULL;
451 zfs_sharevnodeops = NULL;
452 }
453
454 extern const fs_operation_def_t zfs_dvnodeops_template[];
455 extern const fs_operation_def_t zfs_fvnodeops_template[];
456 extern const fs_operation_def_t zfs_xdvnodeops_template[];
457 extern const fs_operation_def_t zfs_symvnodeops_template[];
458 extern const fs_operation_def_t zfs_evnodeops_template[];
459 extern const fs_operation_def_t zfs_sharevnodeops_template[];
460
461 int
zfs_create_op_tables()462 zfs_create_op_tables()
463 {
464 int error;
465
466 /*
467 * zfs_dvnodeops can be set if mod_remove() calls mod_installfs()
468 * due to a failure to remove the the 2nd modlinkage (zfs_modldrv).
469 * In this case we just return as the ops vectors are already set up.
470 */
471 if (zfs_dvnodeops)
472 return (0);
473
474 error = vn_make_ops(MNTTYPE_ZFS, zfs_dvnodeops_template,
475 &zfs_dvnodeops);
476 if (error)
477 return (error);
478
479 error = vn_make_ops(MNTTYPE_ZFS, zfs_fvnodeops_template,
480 &zfs_fvnodeops);
481 if (error)
482 return (error);
483
484 error = vn_make_ops(MNTTYPE_ZFS, zfs_symvnodeops_template,
485 &zfs_symvnodeops);
486 if (error)
487 return (error);
488
489 error = vn_make_ops(MNTTYPE_ZFS, zfs_xdvnodeops_template,
490 &zfs_xdvnodeops);
491 if (error)
492 return (error);
493
494 error = vn_make_ops(MNTTYPE_ZFS, zfs_evnodeops_template,
495 &zfs_evnodeops);
496 if (error)
497 return (error);
498
499 error = vn_make_ops(MNTTYPE_ZFS, zfs_sharevnodeops_template,
500 &zfs_sharevnodeops);
501
502 return (error);
503 }
504
505 int
zfs_create_share_dir(zfsvfs_t * zfsvfs,dmu_tx_t * tx)506 zfs_create_share_dir(zfsvfs_t *zfsvfs, dmu_tx_t *tx)
507 {
508 zfs_acl_ids_t acl_ids;
509 vattr_t vattr;
510 znode_t *sharezp;
511 vnode_t *vp;
512 znode_t *zp;
513 int error;
514
515 vattr.va_mask = AT_MODE|AT_UID|AT_GID|AT_TYPE;
516 vattr.va_type = VDIR;
517 vattr.va_mode = S_IFDIR|0555;
518 vattr.va_uid = crgetuid(kcred);
519 vattr.va_gid = crgetgid(kcred);
520
521 sharezp = kmem_cache_alloc(znode_cache, KM_SLEEP);
522 ASSERT(!POINTER_IS_VALID(sharezp->z_zfsvfs));
523 sharezp->z_moved = 0;
524 sharezp->z_unlinked = 0;
525 sharezp->z_atime_dirty = 0;
526 sharezp->z_zfsvfs = zfsvfs;
527 sharezp->z_is_sa = zfsvfs->z_use_sa;
528 sharezp->z_pflags = 0;
529
530 vp = ZTOV(sharezp);
531 vn_reinit(vp);
532 vp->v_type = VDIR;
533
534 VERIFY(0 == zfs_acl_ids_create(sharezp, IS_ROOT_NODE, &vattr,
535 kcred, NULL, &acl_ids));
536 zfs_mknode(sharezp, &vattr, tx, kcred, IS_ROOT_NODE, &zp, &acl_ids);
537 ASSERT3P(zp, ==, sharezp);
538 ASSERT(!vn_in_dnlc(ZTOV(sharezp))); /* not valid to move */
539 POINTER_INVALIDATE(&sharezp->z_zfsvfs);
540 error = zap_add(zfsvfs->z_os, MASTER_NODE_OBJ,
541 ZFS_SHARES_DIR, 8, 1, &sharezp->z_id, tx);
542 zfsvfs->z_shares_dir = sharezp->z_id;
543
544 zfs_acl_ids_free(&acl_ids);
545 ZTOV(sharezp)->v_count = 0;
546 sa_handle_destroy(sharezp->z_sa_hdl);
547 kmem_cache_free(znode_cache, sharezp);
548
549 return (error);
550 }
551
552 /*
553 * define a couple of values we need available
554 * for both 64 and 32 bit environments.
555 */
556 #ifndef NBITSMINOR64
557 #define NBITSMINOR64 32
558 #endif
559 #ifndef MAXMAJ64
560 #define MAXMAJ64 0xffffffffUL
561 #endif
562 #ifndef MAXMIN64
563 #define MAXMIN64 0xffffffffUL
564 #endif
565
566 /*
567 * Create special expldev for ZFS private use.
568 * Can't use standard expldev since it doesn't do
569 * what we want. The standard expldev() takes a
570 * dev32_t in LP64 and expands it to a long dev_t.
571 * We need an interface that takes a dev32_t in ILP32
572 * and expands it to a long dev_t.
573 */
574 static uint64_t
zfs_expldev(dev_t dev)575 zfs_expldev(dev_t dev)
576 {
577 #ifndef _LP64
578 major_t major = (major_t)dev >> NBITSMINOR32 & MAXMAJ32;
579 return (((uint64_t)major << NBITSMINOR64) |
580 ((minor_t)dev & MAXMIN32));
581 #else
582 return (dev);
583 #endif
584 }
585
586 /*
587 * Special cmpldev for ZFS private use.
588 * Can't use standard cmpldev since it takes
589 * a long dev_t and compresses it to dev32_t in
590 * LP64. We need to do a compaction of a long dev_t
591 * to a dev32_t in ILP32.
592 */
593 dev_t
zfs_cmpldev(uint64_t dev)594 zfs_cmpldev(uint64_t dev)
595 {
596 #ifndef _LP64
597 minor_t minor = (minor_t)dev & MAXMIN64;
598 major_t major = (major_t)(dev >> NBITSMINOR64) & MAXMAJ64;
599
600 if (major > MAXMAJ32 || minor > MAXMIN32)
601 return (NODEV32);
602
603 return (((dev32_t)major << NBITSMINOR32) | minor);
604 #else
605 return (dev);
606 #endif
607 }
608
609 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)610 zfs_znode_sa_init(zfsvfs_t *zfsvfs, znode_t *zp,
611 dmu_buf_t *db, dmu_object_type_t obj_type, sa_handle_t *sa_hdl)
612 {
613 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs) || (zfsvfs == zp->z_zfsvfs));
614 ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zfsvfs, zp->z_id)));
615
616 mutex_enter(&zp->z_lock);
617
618 ASSERT(zp->z_sa_hdl == NULL);
619 ASSERT(zp->z_acl_cached == NULL);
620 if (sa_hdl == NULL) {
621 VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, zp,
622 SA_HDL_SHARED, &zp->z_sa_hdl));
623 } else {
624 zp->z_sa_hdl = sa_hdl;
625 sa_set_userp(sa_hdl, zp);
626 }
627
628 zp->z_is_sa = (obj_type == DMU_OT_SA) ? B_TRUE : B_FALSE;
629
630 /*
631 * Slap on VROOT if we are the root znode
632 */
633 if (zp->z_id == zfsvfs->z_root)
634 ZTOV(zp)->v_flag |= VROOT;
635
636 mutex_exit(&zp->z_lock);
637 vn_exists(ZTOV(zp));
638 }
639
640 void
zfs_znode_dmu_fini(znode_t * zp)641 zfs_znode_dmu_fini(znode_t *zp)
642 {
643 ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zp->z_zfsvfs, zp->z_id)) ||
644 zp->z_unlinked ||
645 RW_WRITE_HELD(&zp->z_zfsvfs->z_teardown_inactive_lock));
646
647 sa_handle_destroy(zp->z_sa_hdl);
648 zp->z_sa_hdl = NULL;
649 }
650
651 /*
652 * Construct a new znode/vnode and intialize.
653 *
654 * This does not do a call to dmu_set_user() that is
655 * up to the caller to do, in case you don't want to
656 * return the znode
657 */
658 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)659 zfs_znode_alloc(zfsvfs_t *zfsvfs, dmu_buf_t *db, int blksz,
660 dmu_object_type_t obj_type, sa_handle_t *hdl)
661 {
662 znode_t *zp;
663 vnode_t *vp;
664 uint64_t mode;
665 uint64_t parent;
666 uint64_t projid = ZFS_DEFAULT_PROJID;
667 sa_bulk_attr_t bulk[11];
668 int count = 0;
669
670 zp = kmem_cache_alloc(znode_cache, KM_SLEEP);
671
672 ASSERT(zp->z_dirlocks == NULL);
673 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs));
674 zp->z_moved = 0;
675
676 /*
677 * Defer setting z_zfsvfs until the znode is ready to be a candidate for
678 * the zfs_znode_move() callback.
679 */
680 zp->z_sa_hdl = NULL;
681 zp->z_unlinked = 0;
682 zp->z_atime_dirty = 0;
683 zp->z_mapcnt = 0;
684 zp->z_id = db->db_object;
685 zp->z_blksz = blksz;
686 zp->z_seq = 0x7A4653;
687 zp->z_sync_cnt = 0;
688
689 vp = ZTOV(zp);
690 vn_reinit(vp);
691
692 zfs_znode_sa_init(zfsvfs, zp, db, obj_type, hdl);
693
694 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8);
695 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL, &zp->z_gen, 8);
696 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
697 &zp->z_size, 8);
698 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL,
699 &zp->z_links, 8);
700 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
701 &zp->z_pflags, 8);
702 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL, &parent, 8);
703 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
704 &zp->z_atime, 16);
705 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
706 &zp->z_uid, 8);
707 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL,
708 &zp->z_gid, 8);
709
710 if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count) != 0 || zp->z_gen == 0 ||
711 (dmu_objset_projectquota_enabled(zfsvfs->z_os) &&
712 (zp->z_pflags & ZFS_PROJID) &&
713 sa_lookup(zp->z_sa_hdl, SA_ZPL_PROJID(zfsvfs), &projid, 8) != 0)) {
714 if (hdl == NULL)
715 sa_handle_destroy(zp->z_sa_hdl);
716 kmem_cache_free(znode_cache, zp);
717 return (NULL);
718 }
719
720 zp->z_projid = projid;
721 zp->z_mode = mode;
722 vp->v_vfsp = zfsvfs->z_parent->z_vfs;
723
724 vp->v_type = IFTOVT((mode_t)mode);
725
726 switch (vp->v_type) {
727 case VDIR:
728 if (zp->z_pflags & ZFS_XATTR) {
729 vn_setops(vp, zfs_xdvnodeops);
730 vp->v_flag |= V_XATTRDIR;
731 } else {
732 vn_setops(vp, zfs_dvnodeops);
733 }
734 zp->z_zn_prefetch = B_TRUE; /* z_prefetch default is enabled */
735 break;
736 case VBLK:
737 case VCHR:
738 {
739 uint64_t rdev;
740 VERIFY(sa_lookup(zp->z_sa_hdl, SA_ZPL_RDEV(zfsvfs),
741 &rdev, sizeof (rdev)) == 0);
742
743 vp->v_rdev = zfs_cmpldev(rdev);
744 }
745 /*FALLTHROUGH*/
746 case VFIFO:
747 case VSOCK:
748 case VDOOR:
749 vn_setops(vp, zfs_fvnodeops);
750 break;
751 case VREG:
752 vp->v_flag |= VMODSORT;
753 if (parent == zfsvfs->z_shares_dir) {
754 ASSERT(zp->z_uid == 0 && zp->z_gid == 0);
755 vn_setops(vp, zfs_sharevnodeops);
756 } else {
757 vn_setops(vp, zfs_fvnodeops);
758 }
759 break;
760 case VLNK:
761 vn_setops(vp, zfs_symvnodeops);
762 break;
763 default:
764 vn_setops(vp, zfs_evnodeops);
765 break;
766 }
767
768 mutex_enter(&zfsvfs->z_znodes_lock);
769 list_insert_tail(&zfsvfs->z_all_znodes, zp);
770 membar_producer();
771 /*
772 * Everything else must be valid before assigning z_zfsvfs makes the
773 * znode eligible for zfs_znode_move().
774 */
775 zp->z_zfsvfs = zfsvfs;
776 mutex_exit(&zfsvfs->z_znodes_lock);
777
778 VFS_HOLD(zfsvfs->z_vfs);
779 return (zp);
780 }
781
782 static uint64_t empty_xattr;
783 static uint64_t pad[4];
784 static zfs_acl_phys_t acl_phys;
785 /*
786 * Create a new DMU object to hold a zfs znode.
787 *
788 * IN: dzp - parent directory for new znode
789 * vap - file attributes for new znode
790 * tx - dmu transaction id for zap operations
791 * cr - credentials of caller
792 * flag - flags:
793 * IS_ROOT_NODE - new object will be root
794 * IS_XATTR - new object is an attribute
795 * bonuslen - length of bonus buffer
796 * setaclp - File/Dir initial ACL
797 * fuidp - Tracks fuid allocation.
798 *
799 * OUT: zpp - allocated znode
800 *
801 */
802 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)803 zfs_mknode(znode_t *dzp, vattr_t *vap, dmu_tx_t *tx, cred_t *cr,
804 uint_t flag, znode_t **zpp, zfs_acl_ids_t *acl_ids)
805 {
806 uint64_t crtime[2], atime[2], mtime[2], ctime[2];
807 uint64_t mode, size, links, parent, pflags;
808 uint64_t dzp_pflags = 0;
809 uint64_t projid = ZFS_DEFAULT_PROJID;
810 uint64_t rdev = 0;
811 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
812 dmu_buf_t *db;
813 timestruc_t now;
814 uint64_t gen, obj;
815 int bonuslen;
816 int dnodesize;
817 sa_handle_t *sa_hdl;
818 dmu_object_type_t obj_type;
819 sa_bulk_attr_t *sa_attrs;
820 int cnt = 0;
821 zfs_acl_locator_cb_t locate = { 0 };
822
823 ASSERT(vap && (vap->va_mask & (AT_TYPE|AT_MODE)) == (AT_TYPE|AT_MODE));
824
825 if (zfsvfs->z_replay) {
826 obj = vap->va_nodeid;
827 now = vap->va_ctime; /* see zfs_replay_create() */
828 gen = vap->va_nblocks; /* ditto */
829 dnodesize = vap->va_fsid; /* ditto */
830 } else {
831 obj = 0;
832 gethrestime(&now);
833 gen = dmu_tx_get_txg(tx);
834 dnodesize = dmu_objset_dnodesize(zfsvfs->z_os);
835 }
836
837 if (dnodesize == 0)
838 dnodesize = DNODE_MIN_SIZE;
839
840 obj_type = zfsvfs->z_use_sa ? DMU_OT_SA : DMU_OT_ZNODE;
841 bonuslen = (obj_type == DMU_OT_SA) ?
842 DN_BONUS_SIZE(dnodesize) : ZFS_OLD_ZNODE_PHYS_SIZE;
843
844 /*
845 * Create a new DMU object.
846 */
847 /*
848 * There's currently no mechanism for pre-reading the blocks that will
849 * be needed to allocate a new object, so we accept the small chance
850 * that there will be an i/o error and we will fail one of the
851 * assertions below.
852 */
853 if (vap->va_type == VDIR) {
854 if (zfsvfs->z_replay) {
855 VERIFY0(zap_create_claim_norm_dnsize(zfsvfs->z_os, obj,
856 zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS,
857 obj_type, bonuslen, dnodesize, tx));
858 } else {
859 obj = zap_create_norm_dnsize(zfsvfs->z_os,
860 zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS,
861 obj_type, bonuslen, dnodesize, tx);
862 }
863 } else {
864 if (zfsvfs->z_replay) {
865 VERIFY0(dmu_object_claim_dnsize(zfsvfs->z_os, obj,
866 DMU_OT_PLAIN_FILE_CONTENTS, 0,
867 obj_type, bonuslen, dnodesize, tx));
868 } else {
869 obj = dmu_object_alloc_dnsize(zfsvfs->z_os,
870 DMU_OT_PLAIN_FILE_CONTENTS, 0,
871 obj_type, bonuslen, dnodesize, tx);
872 }
873 }
874
875 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj);
876 VERIFY0(sa_buf_hold(zfsvfs->z_os, obj, NULL, &db));
877
878 /*
879 * If this is the root, fix up the half-initialized parent pointer
880 * to reference the just-allocated physical data area.
881 */
882 if (flag & IS_ROOT_NODE) {
883 dzp->z_id = obj;
884 }
885
886 /*
887 * If parent is an xattr, so am I.
888 */
889 if (dzp->z_pflags & ZFS_XATTR) {
890 flag |= IS_XATTR;
891 }
892
893 if (zfsvfs->z_use_fuids)
894 pflags = ZFS_ARCHIVE | ZFS_AV_MODIFIED;
895 else
896 pflags = 0;
897
898 if (vap->va_type == VDIR) {
899 size = 2; /* contents ("." and "..") */
900 links = (flag & (IS_ROOT_NODE | IS_XATTR)) ? 2 : 1;
901 } else {
902 size = links = 0;
903 }
904
905 if (vap->va_type == VBLK || vap->va_type == VCHR) {
906 rdev = zfs_expldev(vap->va_rdev);
907 }
908
909 parent = dzp->z_id;
910 mode = acl_ids->z_mode;
911 if (flag & IS_XATTR)
912 pflags |= ZFS_XATTR;
913
914 if (vap->va_type == VREG || vap->va_type == VDIR) {
915 /*
916 * With ZFS_PROJID flag, we can easily know whether there is
917 * project ID stored on disk or not. See zfs_space_delta_cb().
918 */
919 if (obj_type != DMU_OT_ZNODE &&
920 dmu_objset_projectquota_enabled(zfsvfs->z_os))
921 pflags |= ZFS_PROJID;
922
923 /*
924 * Inherit project ID from parent if required.
925 */
926 projid = zfs_inherit_projid(dzp);
927 if (dzp->z_pflags & ZFS_PROJINHERIT)
928 pflags |= ZFS_PROJINHERIT;
929 }
930
931 /*
932 * No execs denied will be deterimed when zfs_mode_compute() is called.
933 */
934 pflags |= acl_ids->z_aclp->z_hints &
935 (ZFS_ACL_TRIVIAL|ZFS_INHERIT_ACE|ZFS_ACL_AUTO_INHERIT|
936 ZFS_ACL_DEFAULTED|ZFS_ACL_PROTECTED);
937
938 ZFS_TIME_ENCODE(&now, crtime);
939 ZFS_TIME_ENCODE(&now, ctime);
940
941 if (vap->va_mask & AT_ATIME) {
942 ZFS_TIME_ENCODE(&vap->va_atime, atime);
943 } else {
944 ZFS_TIME_ENCODE(&now, atime);
945 }
946
947 if (vap->va_mask & AT_MTIME) {
948 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
949 } else {
950 ZFS_TIME_ENCODE(&now, mtime);
951 }
952
953 /* Now add in all of the "SA" attributes */
954 VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, NULL, SA_HDL_SHARED,
955 &sa_hdl));
956
957 /*
958 * Setup the array of attributes to be replaced/set on the new file
959 *
960 * order for DMU_OT_ZNODE is critical since it needs to be constructed
961 * in the old znode_phys_t format. Don't change this ordering
962 */
963 sa_attrs = kmem_alloc(sizeof (sa_bulk_attr_t) * ZPL_END, KM_SLEEP);
964
965 if (obj_type == DMU_OT_ZNODE) {
966 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs),
967 NULL, &atime, 16);
968 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs),
969 NULL, &mtime, 16);
970 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs),
971 NULL, &ctime, 16);
972 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs),
973 NULL, &crtime, 16);
974 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs),
975 NULL, &gen, 8);
976 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs),
977 NULL, &mode, 8);
978 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs),
979 NULL, &size, 8);
980 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs),
981 NULL, &parent, 8);
982 } else {
983 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs),
984 NULL, &mode, 8);
985 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs),
986 NULL, &size, 8);
987 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs),
988 NULL, &gen, 8);
989 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs),
990 NULL, &acl_ids->z_fuid, 8);
991 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs),
992 NULL, &acl_ids->z_fgid, 8);
993 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs),
994 NULL, &parent, 8);
995 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs),
996 NULL, &pflags, 8);
997 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs),
998 NULL, &atime, 16);
999 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs),
1000 NULL, &mtime, 16);
1001 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs),
1002 NULL, &ctime, 16);
1003 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs),
1004 NULL, &crtime, 16);
1005 }
1006
1007 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_LINKS(zfsvfs), NULL, &links, 8);
1008
1009 if (obj_type == DMU_OT_ZNODE) {
1010 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_XATTR(zfsvfs), NULL,
1011 &empty_xattr, 8);
1012 } else if (dmu_objset_projectquota_enabled(zfsvfs->z_os) &&
1013 pflags & ZFS_PROJID) {
1014 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PROJID(zfsvfs),
1015 NULL, &projid, 8);
1016 }
1017 if (obj_type == DMU_OT_ZNODE ||
1018 (vap->va_type == VBLK || vap->va_type == VCHR)) {
1019 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_RDEV(zfsvfs),
1020 NULL, &rdev, 8);
1021
1022 }
1023 if (obj_type == DMU_OT_ZNODE) {
1024 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs),
1025 NULL, &pflags, 8);
1026 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs), NULL,
1027 &acl_ids->z_fuid, 8);
1028 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs), NULL,
1029 &acl_ids->z_fgid, 8);
1030 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PAD(zfsvfs), NULL, pad,
1031 sizeof (uint64_t) * 4);
1032 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ZNODE_ACL(zfsvfs), NULL,
1033 &acl_phys, sizeof (zfs_acl_phys_t));
1034 } else if (acl_ids->z_aclp->z_version >= ZFS_ACL_VERSION_FUID) {
1035 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_COUNT(zfsvfs), NULL,
1036 &acl_ids->z_aclp->z_acl_count, 8);
1037 locate.cb_aclp = acl_ids->z_aclp;
1038 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_ACES(zfsvfs),
1039 zfs_acl_data_locator, &locate,
1040 acl_ids->z_aclp->z_acl_bytes);
1041 mode = zfs_mode_compute(mode, acl_ids->z_aclp, &pflags,
1042 acl_ids->z_fuid, acl_ids->z_fgid);
1043 }
1044
1045 VERIFY(sa_replace_all_by_template(sa_hdl, sa_attrs, cnt, tx) == 0);
1046
1047 if (!(flag & IS_ROOT_NODE)) {
1048 *zpp = zfs_znode_alloc(zfsvfs, db, 0, obj_type, sa_hdl);
1049 ASSERT(*zpp != NULL);
1050 } else {
1051 /*
1052 * If we are creating the root node, the "parent" we
1053 * passed in is the znode for the root.
1054 */
1055 *zpp = dzp;
1056
1057 (*zpp)->z_sa_hdl = sa_hdl;
1058 }
1059
1060 (*zpp)->z_pflags = pflags;
1061 (*zpp)->z_mode = mode;
1062 (*zpp)->z_dnodesize = dnodesize;
1063 (*zpp)->z_projid = projid;
1064
1065 if (vap->va_mask & AT_XVATTR)
1066 zfs_xvattr_set(*zpp, (xvattr_t *)vap, tx);
1067
1068 if (obj_type == DMU_OT_ZNODE ||
1069 acl_ids->z_aclp->z_version < ZFS_ACL_VERSION_FUID) {
1070 VERIFY0(zfs_aclset_common(*zpp, acl_ids->z_aclp, cr, tx));
1071 }
1072 kmem_free(sa_attrs, sizeof (sa_bulk_attr_t) * ZPL_END);
1073 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj);
1074 }
1075
1076 /*
1077 * Update in-core attributes. It is assumed the caller will be doing an
1078 * sa_bulk_update to push the changes out.
1079 */
1080 void
zfs_xvattr_set(znode_t * zp,xvattr_t * xvap,dmu_tx_t * tx)1081 zfs_xvattr_set(znode_t *zp, xvattr_t *xvap, dmu_tx_t *tx)
1082 {
1083 xoptattr_t *xoap;
1084
1085 xoap = xva_getxoptattr(xvap);
1086 ASSERT(xoap);
1087
1088 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
1089 uint64_t times[2];
1090 ZFS_TIME_ENCODE(&xoap->xoa_createtime, times);
1091 (void) sa_update(zp->z_sa_hdl, SA_ZPL_CRTIME(zp->z_zfsvfs),
1092 ×, sizeof (times), tx);
1093 XVA_SET_RTN(xvap, XAT_CREATETIME);
1094 }
1095 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
1096 ZFS_ATTR_SET(zp, ZFS_READONLY, xoap->xoa_readonly,
1097 zp->z_pflags, tx);
1098 XVA_SET_RTN(xvap, XAT_READONLY);
1099 }
1100 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
1101 ZFS_ATTR_SET(zp, ZFS_HIDDEN, xoap->xoa_hidden,
1102 zp->z_pflags, tx);
1103 XVA_SET_RTN(xvap, XAT_HIDDEN);
1104 }
1105 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
1106 ZFS_ATTR_SET(zp, ZFS_SYSTEM, xoap->xoa_system,
1107 zp->z_pflags, tx);
1108 XVA_SET_RTN(xvap, XAT_SYSTEM);
1109 }
1110 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
1111 ZFS_ATTR_SET(zp, ZFS_ARCHIVE, xoap->xoa_archive,
1112 zp->z_pflags, tx);
1113 XVA_SET_RTN(xvap, XAT_ARCHIVE);
1114 }
1115 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
1116 ZFS_ATTR_SET(zp, ZFS_IMMUTABLE, xoap->xoa_immutable,
1117 zp->z_pflags, tx);
1118 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
1119 }
1120 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
1121 ZFS_ATTR_SET(zp, ZFS_NOUNLINK, xoap->xoa_nounlink,
1122 zp->z_pflags, tx);
1123 XVA_SET_RTN(xvap, XAT_NOUNLINK);
1124 }
1125 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
1126 ZFS_ATTR_SET(zp, ZFS_APPENDONLY, xoap->xoa_appendonly,
1127 zp->z_pflags, tx);
1128 XVA_SET_RTN(xvap, XAT_APPENDONLY);
1129 }
1130 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
1131 ZFS_ATTR_SET(zp, ZFS_NODUMP, xoap->xoa_nodump,
1132 zp->z_pflags, tx);
1133 XVA_SET_RTN(xvap, XAT_NODUMP);
1134 }
1135 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
1136 ZFS_ATTR_SET(zp, ZFS_OPAQUE, xoap->xoa_opaque,
1137 zp->z_pflags, tx);
1138 XVA_SET_RTN(xvap, XAT_OPAQUE);
1139 }
1140 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
1141 ZFS_ATTR_SET(zp, ZFS_AV_QUARANTINED,
1142 xoap->xoa_av_quarantined, zp->z_pflags, tx);
1143 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
1144 }
1145 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
1146 ZFS_ATTR_SET(zp, ZFS_AV_MODIFIED, xoap->xoa_av_modified,
1147 zp->z_pflags, tx);
1148 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
1149 }
1150 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) {
1151 zfs_sa_set_scanstamp(zp, xvap, tx);
1152 XVA_SET_RTN(xvap, XAT_AV_SCANSTAMP);
1153 }
1154 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
1155 ZFS_ATTR_SET(zp, ZFS_REPARSE, xoap->xoa_reparse,
1156 zp->z_pflags, tx);
1157 XVA_SET_RTN(xvap, XAT_REPARSE);
1158 }
1159 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
1160 ZFS_ATTR_SET(zp, ZFS_OFFLINE, xoap->xoa_offline,
1161 zp->z_pflags, tx);
1162 XVA_SET_RTN(xvap, XAT_OFFLINE);
1163 }
1164 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
1165 ZFS_ATTR_SET(zp, ZFS_SPARSE, xoap->xoa_sparse,
1166 zp->z_pflags, tx);
1167 XVA_SET_RTN(xvap, XAT_SPARSE);
1168 }
1169 if (XVA_ISSET_REQ(xvap, XAT_PROJINHERIT)) {
1170 ZFS_ATTR_SET(zp, ZFS_PROJINHERIT, xoap->xoa_projinherit,
1171 zp->z_pflags, tx);
1172 XVA_SET_RTN(xvap, XAT_PROJINHERIT);
1173 }
1174 }
1175
1176 int
zfs_zget(zfsvfs_t * zfsvfs,uint64_t obj_num,znode_t ** zpp)1177 zfs_zget(zfsvfs_t *zfsvfs, uint64_t obj_num, znode_t **zpp)
1178 {
1179 dmu_object_info_t doi;
1180 dmu_buf_t *db;
1181 znode_t *zp;
1182 int err;
1183 sa_handle_t *hdl;
1184
1185 *zpp = NULL;
1186
1187 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num);
1188
1189 err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db);
1190 if (err) {
1191 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1192 return (err);
1193 }
1194
1195 dmu_object_info_from_db(db, &doi);
1196 if (doi.doi_bonus_type != DMU_OT_SA &&
1197 (doi.doi_bonus_type != DMU_OT_ZNODE ||
1198 (doi.doi_bonus_type == DMU_OT_ZNODE &&
1199 doi.doi_bonus_size < sizeof (znode_phys_t)))) {
1200 sa_buf_rele(db, NULL);
1201 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1202 return (SET_ERROR(EINVAL));
1203 }
1204
1205 hdl = dmu_buf_get_user(db);
1206 if (hdl != NULL) {
1207 zp = sa_get_userdata(hdl);
1208
1209
1210 /*
1211 * Since "SA" does immediate eviction we
1212 * should never find a sa handle that doesn't
1213 * know about the znode.
1214 */
1215
1216 ASSERT3P(zp, !=, NULL);
1217
1218 mutex_enter(&zp->z_lock);
1219 ASSERT3U(zp->z_id, ==, obj_num);
1220 if (zp->z_unlinked) {
1221 err = SET_ERROR(ENOENT);
1222 } else {
1223 VN_HOLD(ZTOV(zp));
1224 *zpp = zp;
1225 err = 0;
1226 }
1227 mutex_exit(&zp->z_lock);
1228 sa_buf_rele(db, NULL);
1229 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1230 return (err);
1231 }
1232
1233 /*
1234 * Not found create new znode/vnode
1235 * but only if file exists.
1236 *
1237 * There is a small window where zfs_vget() could
1238 * find this object while a file create is still in
1239 * progress. This is checked for in zfs_znode_alloc()
1240 *
1241 * if zfs_znode_alloc() fails it will drop the hold on the
1242 * bonus buffer.
1243 */
1244 zp = zfs_znode_alloc(zfsvfs, db, doi.doi_data_block_size,
1245 doi.doi_bonus_type, NULL);
1246 if (zp == NULL) {
1247 err = SET_ERROR(ENOENT);
1248 } else {
1249 if (zp->z_links == 0)
1250 zp->z_unlinked = B_TRUE;
1251 *zpp = zp;
1252 }
1253 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1254 return (err);
1255 }
1256
1257 int
zfs_rezget(znode_t * zp)1258 zfs_rezget(znode_t *zp)
1259 {
1260 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1261 dmu_object_info_t doi;
1262 dmu_buf_t *db;
1263 uint64_t obj_num = zp->z_id;
1264 uint64_t mode;
1265 sa_bulk_attr_t bulk[10];
1266 int err;
1267 int count = 0;
1268 uint64_t gen;
1269 uint64_t projid = ZFS_DEFAULT_PROJID;
1270
1271 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num);
1272
1273 rw_enter(&zp->z_acl_lock, RW_WRITER);
1274 if (zp->z_acl_cached) {
1275 zfs_acl_free(zp->z_acl_cached);
1276 zp->z_acl_cached = NULL;
1277 }
1278 rw_exit(&zp->z_acl_lock);
1279
1280 ASSERT(zp->z_sa_hdl == NULL);
1281 err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db);
1282 if (err) {
1283 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1284 return (err);
1285 }
1286
1287 dmu_object_info_from_db(db, &doi);
1288 if (doi.doi_bonus_type != DMU_OT_SA &&
1289 (doi.doi_bonus_type != DMU_OT_ZNODE ||
1290 (doi.doi_bonus_type == DMU_OT_ZNODE &&
1291 doi.doi_bonus_size < sizeof (znode_phys_t)))) {
1292 sa_buf_rele(db, NULL);
1293 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1294 return (SET_ERROR(EINVAL));
1295 }
1296
1297 zfs_znode_sa_init(zfsvfs, zp, db, doi.doi_bonus_type, NULL);
1298
1299 /* reload cached values */
1300 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL,
1301 &gen, sizeof (gen));
1302 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
1303 &zp->z_size, sizeof (zp->z_size));
1304 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL,
1305 &zp->z_links, sizeof (zp->z_links));
1306 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
1307 &zp->z_pflags, sizeof (zp->z_pflags));
1308 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
1309 &zp->z_atime, sizeof (zp->z_atime));
1310 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
1311 &zp->z_uid, sizeof (zp->z_uid));
1312 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL,
1313 &zp->z_gid, sizeof (zp->z_gid));
1314 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
1315 &mode, sizeof (mode));
1316
1317 if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) {
1318 zfs_znode_dmu_fini(zp);
1319 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1320 return (SET_ERROR(EIO));
1321 }
1322
1323 if (dmu_objset_projectquota_enabled(zfsvfs->z_os)) {
1324 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_PROJID(zfsvfs),
1325 &projid, 8);
1326 if (err != 0 && err != ENOENT) {
1327 zfs_znode_dmu_fini(zp);
1328 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1329 return (SET_ERROR(err));
1330 }
1331 }
1332
1333 zp->z_projid = projid;
1334 zp->z_mode = mode;
1335
1336 if (gen != zp->z_gen) {
1337 zfs_znode_dmu_fini(zp);
1338 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1339 return (SET_ERROR(EIO));
1340 }
1341
1342 zp->z_blksz = doi.doi_data_block_size;
1343
1344 /*
1345 * If the file has zero links, then it has been unlinked on the send
1346 * side and it must be in the received unlinked set.
1347 * We call zfs_znode_dmu_fini() now to prevent any accesses to the
1348 * stale data and to prevent automatical removal of the file in
1349 * zfs_zinactive(). The file will be removed either when it is removed
1350 * on the send side and the next incremental stream is received or
1351 * when the unlinked set gets processed.
1352 */
1353 zp->z_unlinked = (zp->z_links == 0);
1354 if (zp->z_unlinked)
1355 zfs_znode_dmu_fini(zp);
1356
1357 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1358
1359 return (0);
1360 }
1361
1362 void
zfs_znode_delete(znode_t * zp,dmu_tx_t * tx)1363 zfs_znode_delete(znode_t *zp, dmu_tx_t *tx)
1364 {
1365 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1366 objset_t *os = zfsvfs->z_os;
1367 uint64_t obj = zp->z_id;
1368 uint64_t acl_obj = zfs_external_acl(zp);
1369
1370 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj);
1371 if (acl_obj) {
1372 VERIFY(!zp->z_is_sa);
1373 VERIFY(0 == dmu_object_free(os, acl_obj, tx));
1374 }
1375 VERIFY(0 == dmu_object_free(os, obj, tx));
1376 zfs_znode_dmu_fini(zp);
1377 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj);
1378 zfs_znode_free(zp);
1379 }
1380
1381 void
zfs_zinactive(znode_t * zp)1382 zfs_zinactive(znode_t *zp)
1383 {
1384 vnode_t *vp = ZTOV(zp);
1385 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1386 uint64_t z_id = zp->z_id;
1387
1388 ASSERT(zp->z_sa_hdl);
1389
1390 /*
1391 * Don't allow a zfs_zget() while were trying to release this znode
1392 */
1393 ZFS_OBJ_HOLD_ENTER(zfsvfs, z_id);
1394
1395 mutex_enter(&zp->z_lock);
1396 mutex_enter(&vp->v_lock);
1397 VN_RELE_LOCKED(vp);
1398 if (vp->v_count > 0 || vn_has_cached_data(vp)) {
1399 /*
1400 * If the hold count is greater than zero, somebody has
1401 * obtained a new reference on this znode while we were
1402 * processing it here, so we are done. If we still have
1403 * mapped pages then we are also done, since we don't
1404 * want to inactivate the znode until the pages get pushed.
1405 *
1406 * XXX - if vn_has_cached_data(vp) is true, but count == 0,
1407 * this seems like it would leave the znode hanging with
1408 * no chance to go inactive...
1409 */
1410 mutex_exit(&vp->v_lock);
1411 mutex_exit(&zp->z_lock);
1412 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id);
1413 return;
1414 }
1415 mutex_exit(&vp->v_lock);
1416
1417 /*
1418 * If this was the last reference to a file with no links, remove
1419 * the file from the file system unless the file system is mounted
1420 * read-only. That can happen, for example, if the file system was
1421 * originally read-write, the file was opened, then unlinked and
1422 * the file system was made read-only before the file was finally
1423 * closed. The file will remain in the unlinked set.
1424 */
1425 if (zp->z_unlinked) {
1426 ASSERT(!zfsvfs->z_issnap);
1427 if ((zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) == 0 &&
1428 !zfs_unlink_suspend_progress) {
1429 mutex_exit(&zp->z_lock);
1430 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id);
1431 zfs_rmnode(zp);
1432 return;
1433 }
1434 }
1435
1436 mutex_exit(&zp->z_lock);
1437 zfs_znode_dmu_fini(zp);
1438 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id);
1439 zfs_znode_free(zp);
1440 }
1441
1442 void
zfs_znode_free(znode_t * zp)1443 zfs_znode_free(znode_t *zp)
1444 {
1445 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1446
1447 vn_invalid(ZTOV(zp));
1448
1449 ASSERT(ZTOV(zp)->v_count == 0);
1450
1451 mutex_enter(&zfsvfs->z_znodes_lock);
1452 POINTER_INVALIDATE(&zp->z_zfsvfs);
1453 list_remove(&zfsvfs->z_all_znodes, zp);
1454 mutex_exit(&zfsvfs->z_znodes_lock);
1455
1456 if (zp->z_acl_cached) {
1457 zfs_acl_free(zp->z_acl_cached);
1458 zp->z_acl_cached = NULL;
1459 }
1460
1461 kmem_cache_free(znode_cache, zp);
1462
1463 VFS_RELE(zfsvfs->z_vfs);
1464 }
1465
1466 void
zfs_tstamp_update_setup(znode_t * zp,uint_t flag,uint64_t mtime[2],uint64_t ctime[2],boolean_t have_tx)1467 zfs_tstamp_update_setup(znode_t *zp, uint_t flag, uint64_t mtime[2],
1468 uint64_t ctime[2], boolean_t have_tx)
1469 {
1470 timestruc_t now;
1471
1472 gethrestime(&now);
1473
1474 if (have_tx) { /* will sa_bulk_update happen really soon? */
1475 zp->z_atime_dirty = 0;
1476 zp->z_seq++;
1477 } else {
1478 zp->z_atime_dirty = 1;
1479 }
1480
1481 if (flag & AT_ATIME) {
1482 ZFS_TIME_ENCODE(&now, zp->z_atime);
1483 }
1484
1485 if (flag & AT_MTIME) {
1486 ZFS_TIME_ENCODE(&now, mtime);
1487 if (zp->z_zfsvfs->z_use_fuids) {
1488 zp->z_pflags |= (ZFS_ARCHIVE |
1489 ZFS_AV_MODIFIED);
1490 }
1491 }
1492
1493 if (flag & AT_CTIME) {
1494 ZFS_TIME_ENCODE(&now, ctime);
1495 if (zp->z_zfsvfs->z_use_fuids)
1496 zp->z_pflags |= ZFS_ARCHIVE;
1497 }
1498 }
1499
1500 /*
1501 * Grow the block size for a file.
1502 *
1503 * IN: zp - znode of file to free data in.
1504 * size - requested block size
1505 * tx - open transaction.
1506 *
1507 * NOTE: this function assumes that the znode is write locked.
1508 */
1509 void
zfs_grow_blocksize(znode_t * zp,uint64_t size,dmu_tx_t * tx)1510 zfs_grow_blocksize(znode_t *zp, uint64_t size, dmu_tx_t *tx)
1511 {
1512 int error;
1513 u_longlong_t dummy;
1514
1515 if (size <= zp->z_blksz)
1516 return;
1517 /*
1518 * If the file size is already greater than the current blocksize,
1519 * we will not grow. If there is more than one block in a file,
1520 * the blocksize cannot change.
1521 */
1522 if (zp->z_blksz && zp->z_size > zp->z_blksz)
1523 return;
1524
1525 error = dmu_object_set_blocksize(zp->z_zfsvfs->z_os, zp->z_id,
1526 size, 0, tx);
1527
1528 if (error == ENOTSUP)
1529 return;
1530 ASSERT0(error);
1531
1532 /* What blocksize did we actually get? */
1533 dmu_object_size_from_db(sa_get_db(zp->z_sa_hdl), &zp->z_blksz, &dummy);
1534 }
1535
1536 /*
1537 * This is a dummy interface used when pvn_vplist_dirty() should *not*
1538 * be calling back into the fs for a putpage(). E.g.: when truncating
1539 * a file, the pages being "thrown away* don't need to be written out.
1540 */
1541 /* ARGSUSED */
1542 static int
zfs_no_putpage(vnode_t * vp,page_t * pp,u_offset_t * offp,size_t * lenp,int flags,cred_t * cr)1543 zfs_no_putpage(vnode_t *vp, page_t *pp, u_offset_t *offp, size_t *lenp,
1544 int flags, cred_t *cr)
1545 {
1546 ASSERT(0);
1547 return (0);
1548 }
1549
1550 /*
1551 * Increase the file length
1552 *
1553 * IN: zp - znode of file to free data in.
1554 * end - new end-of-file
1555 *
1556 * RETURN: 0 on success, error code on failure
1557 */
1558 static int
zfs_extend(znode_t * zp,uint64_t end)1559 zfs_extend(znode_t *zp, uint64_t end)
1560 {
1561 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1562 dmu_tx_t *tx;
1563 locked_range_t *lr;
1564 uint64_t newblksz;
1565 int error;
1566
1567 /*
1568 * We will change zp_size, lock the whole file.
1569 */
1570 lr = rangelock_enter(&zp->z_rangelock, 0, UINT64_MAX, RL_WRITER);
1571
1572 /*
1573 * Nothing to do if file already at desired length.
1574 */
1575 if (end <= zp->z_size) {
1576 rangelock_exit(lr);
1577 return (0);
1578 }
1579 tx = dmu_tx_create(zfsvfs->z_os);
1580 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1581 zfs_sa_upgrade_txholds(tx, zp);
1582 if (end > zp->z_blksz &&
1583 (!ISP2(zp->z_blksz) || zp->z_blksz < zfsvfs->z_max_blksz)) {
1584 /*
1585 * We are growing the file past the current block size.
1586 */
1587 if (zp->z_blksz > zp->z_zfsvfs->z_max_blksz) {
1588 /*
1589 * File's blocksize is already larger than the
1590 * "recordsize" property. Only let it grow to
1591 * the next power of 2.
1592 */
1593 ASSERT(!ISP2(zp->z_blksz));
1594 newblksz = MIN(end, 1 << highbit64(zp->z_blksz));
1595 } else {
1596 newblksz = MIN(end, zp->z_zfsvfs->z_max_blksz);
1597 }
1598 dmu_tx_hold_write(tx, zp->z_id, 0, newblksz);
1599 } else {
1600 newblksz = 0;
1601 }
1602
1603 error = dmu_tx_assign(tx, TXG_WAIT);
1604 if (error) {
1605 dmu_tx_abort(tx);
1606 rangelock_exit(lr);
1607 return (error);
1608 }
1609
1610 if (newblksz)
1611 zfs_grow_blocksize(zp, newblksz, tx);
1612
1613 zp->z_size = end;
1614
1615 VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zp->z_zfsvfs),
1616 &zp->z_size, sizeof (zp->z_size), tx));
1617
1618 rangelock_exit(lr);
1619
1620 dmu_tx_commit(tx);
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 section to free.
1630 * len - length of section to free.
1631 *
1632 * RETURN: 0 on success, error code on failure
1633 */
1634 static int
zfs_free_range(znode_t * zp,uint64_t off,uint64_t len)1635 zfs_free_range(znode_t *zp, uint64_t off, uint64_t len)
1636 {
1637 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1638 locked_range_t *lr;
1639 int error;
1640
1641 /*
1642 * Lock the range being freed.
1643 */
1644 lr = rangelock_enter(&zp->z_rangelock, off, len, RL_WRITER);
1645
1646 /*
1647 * Nothing to do if file already at desired length.
1648 */
1649 if (off >= zp->z_size) {
1650 rangelock_exit(lr);
1651 return (0);
1652 }
1653
1654 if (off + len > zp->z_size)
1655 len = zp->z_size - off;
1656
1657 error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, off, len);
1658
1659 rangelock_exit(lr);
1660
1661 return (error);
1662 }
1663
1664 /*
1665 * Truncate a file
1666 *
1667 * IN: zp - znode of file to free data in.
1668 * end - new end-of-file.
1669 *
1670 * RETURN: 0 on success, error code on failure
1671 */
1672 static int
zfs_trunc(znode_t * zp,uint64_t end)1673 zfs_trunc(znode_t *zp, uint64_t end)
1674 {
1675 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1676 vnode_t *vp = ZTOV(zp);
1677 dmu_tx_t *tx;
1678 locked_range_t *lr;
1679 int error;
1680 sa_bulk_attr_t bulk[2];
1681 int count = 0;
1682
1683 /*
1684 * We will change zp_size, lock the whole file.
1685 */
1686 lr = rangelock_enter(&zp->z_rangelock, 0, UINT64_MAX, RL_WRITER);
1687
1688 /*
1689 * Nothing to do if file already at desired length.
1690 */
1691 if (end >= zp->z_size) {
1692 rangelock_exit(lr);
1693 return (0);
1694 }
1695
1696 error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, end,
1697 DMU_OBJECT_END);
1698 if (error) {
1699 rangelock_exit(lr);
1700 return (error);
1701 }
1702 tx = dmu_tx_create(zfsvfs->z_os);
1703 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1704 zfs_sa_upgrade_txholds(tx, zp);
1705 dmu_tx_mark_netfree(tx);
1706 error = dmu_tx_assign(tx, TXG_WAIT);
1707 if (error) {
1708 dmu_tx_abort(tx);
1709 rangelock_exit(lr);
1710 return (error);
1711 }
1712
1713 zp->z_size = end;
1714 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs),
1715 NULL, &zp->z_size, sizeof (zp->z_size));
1716
1717 if (end == 0) {
1718 zp->z_pflags &= ~ZFS_SPARSE;
1719 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
1720 NULL, &zp->z_pflags, 8);
1721 }
1722 VERIFY(sa_bulk_update(zp->z_sa_hdl, bulk, count, tx) == 0);
1723
1724 dmu_tx_commit(tx);
1725
1726 /*
1727 * Clear any mapped pages in the truncated region. This has to
1728 * happen outside of the transaction to avoid the possibility of
1729 * a deadlock with someone trying to push a page that we are
1730 * about to invalidate.
1731 */
1732 if (vn_has_cached_data(vp)) {
1733 page_t *pp;
1734 uint64_t start = end & PAGEMASK;
1735 int poff = end & PAGEOFFSET;
1736
1737 if (poff != 0 && (pp = page_lookup(vp, start, SE_SHARED))) {
1738 /*
1739 * We need to zero a partial page.
1740 */
1741 pagezero(pp, poff, PAGESIZE - poff);
1742 start += PAGESIZE;
1743 page_unlock(pp);
1744 }
1745 error = pvn_vplist_dirty(vp, start, zfs_no_putpage,
1746 B_INVAL | B_TRUNC, NULL);
1747 ASSERT(error == 0);
1748 }
1749
1750 rangelock_exit(lr);
1751
1752 return (0);
1753 }
1754
1755 /*
1756 * Free space in a file
1757 *
1758 * IN: zp - znode of file to free data in.
1759 * off - start of range
1760 * len - end of range (0 => EOF)
1761 * flag - current file open mode flags.
1762 * log - TRUE if this action should be logged
1763 *
1764 * RETURN: 0 on success, error code on failure
1765 */
1766 int
zfs_freesp(znode_t * zp,uint64_t off,uint64_t len,int flag,boolean_t log)1767 zfs_freesp(znode_t *zp, uint64_t off, uint64_t len, int flag, boolean_t log)
1768 {
1769 vnode_t *vp = ZTOV(zp);
1770 dmu_tx_t *tx;
1771 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1772 zilog_t *zilog = zfsvfs->z_log;
1773 uint64_t mode;
1774 uint64_t mtime[2], ctime[2];
1775 sa_bulk_attr_t bulk[3];
1776 int count = 0;
1777 int error;
1778
1779 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs), &mode,
1780 sizeof (mode))) != 0)
1781 return (error);
1782
1783 if (off > zp->z_size) {
1784 error = zfs_extend(zp, off+len);
1785 if (error == 0 && log)
1786 goto log;
1787 else
1788 return (error);
1789 }
1790
1791 /*
1792 * Check for any locks in the region to be freed.
1793 */
1794
1795 if (MANDLOCK(vp, (mode_t)mode)) {
1796 uint64_t length = (len ? len : zp->z_size - off);
1797 if (error = chklock(vp, FWRITE, off, length, flag, NULL))
1798 return (error);
1799 }
1800
1801 if (len == 0) {
1802 error = zfs_trunc(zp, off);
1803 } else {
1804 if ((error = zfs_free_range(zp, off, len)) == 0 &&
1805 off + len > zp->z_size)
1806 error = zfs_extend(zp, off+len);
1807 }
1808 if (error || !log)
1809 return (error);
1810 log:
1811 tx = dmu_tx_create(zfsvfs->z_os);
1812 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1813 zfs_sa_upgrade_txholds(tx, zp);
1814 error = dmu_tx_assign(tx, TXG_WAIT);
1815 if (error) {
1816 dmu_tx_abort(tx);
1817 return (error);
1818 }
1819
1820 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, mtime, 16);
1821 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, ctime, 16);
1822 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
1823 NULL, &zp->z_pflags, 8);
1824 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, B_TRUE);
1825 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
1826 ASSERT(error == 0);
1827
1828 zfs_log_truncate(zilog, tx, TX_TRUNCATE, zp, off, len);
1829
1830 dmu_tx_commit(tx);
1831 return (0);
1832 }
1833
1834 void
zfs_create_fs(objset_t * os,cred_t * cr,nvlist_t * zplprops,dmu_tx_t * tx)1835 zfs_create_fs(objset_t *os, cred_t *cr, nvlist_t *zplprops, dmu_tx_t *tx)
1836 {
1837 uint64_t moid, obj, sa_obj, version;
1838 uint64_t sense = ZFS_CASE_SENSITIVE;
1839 uint64_t norm = 0;
1840 nvpair_t *elem;
1841 int error;
1842 int i;
1843 znode_t *rootzp = NULL;
1844 zfsvfs_t *zfsvfs;
1845 vnode_t *vp;
1846 vattr_t vattr;
1847 znode_t *zp;
1848 zfs_acl_ids_t acl_ids;
1849
1850 /*
1851 * First attempt to create master node.
1852 */
1853 /*
1854 * In an empty objset, there are no blocks to read and thus
1855 * there can be no i/o errors (which we assert below).
1856 */
1857 moid = MASTER_NODE_OBJ;
1858 error = zap_create_claim(os, moid, DMU_OT_MASTER_NODE,
1859 DMU_OT_NONE, 0, tx);
1860 ASSERT(error == 0);
1861
1862 /*
1863 * Set starting attributes.
1864 */
1865 version = zfs_zpl_version_map(spa_version(dmu_objset_spa(os)));
1866 elem = NULL;
1867 while ((elem = nvlist_next_nvpair(zplprops, elem)) != NULL) {
1868 /* For the moment we expect all zpl props to be uint64_ts */
1869 uint64_t val;
1870 char *name;
1871
1872 ASSERT(nvpair_type(elem) == DATA_TYPE_UINT64);
1873 VERIFY(nvpair_value_uint64(elem, &val) == 0);
1874 name = nvpair_name(elem);
1875 if (strcmp(name, zfs_prop_to_name(ZFS_PROP_VERSION)) == 0) {
1876 if (val < version)
1877 version = val;
1878 } else {
1879 error = zap_update(os, moid, name, 8, 1, &val, tx);
1880 }
1881 ASSERT(error == 0);
1882 if (strcmp(name, zfs_prop_to_name(ZFS_PROP_NORMALIZE)) == 0)
1883 norm = val;
1884 else if (strcmp(name, zfs_prop_to_name(ZFS_PROP_CASE)) == 0)
1885 sense = val;
1886 }
1887 ASSERT(version != 0);
1888 error = zap_update(os, moid, ZPL_VERSION_STR, 8, 1, &version, tx);
1889
1890 /*
1891 * Create zap object used for SA attribute registration
1892 */
1893
1894 if (version >= ZPL_VERSION_SA) {
1895 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
1896 DMU_OT_NONE, 0, tx);
1897 error = zap_add(os, moid, ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
1898 ASSERT(error == 0);
1899 } else {
1900 sa_obj = 0;
1901 }
1902 /*
1903 * Create a delete queue.
1904 */
1905 obj = zap_create(os, DMU_OT_UNLINKED_SET, DMU_OT_NONE, 0, tx);
1906
1907 error = zap_add(os, moid, ZFS_UNLINKED_SET, 8, 1, &obj, tx);
1908 ASSERT(error == 0);
1909
1910 /*
1911 * Create root znode. Create minimal znode/vnode/zfsvfs
1912 * to allow zfs_mknode to work.
1913 */
1914 vattr.va_mask = AT_MODE|AT_UID|AT_GID|AT_TYPE;
1915 vattr.va_type = VDIR;
1916 vattr.va_mode = S_IFDIR|0755;
1917 vattr.va_uid = crgetuid(cr);
1918 vattr.va_gid = crgetgid(cr);
1919
1920 rootzp = kmem_cache_alloc(znode_cache, KM_SLEEP);
1921 ASSERT(!POINTER_IS_VALID(rootzp->z_zfsvfs));
1922 rootzp->z_moved = 0;
1923 rootzp->z_unlinked = 0;
1924 rootzp->z_atime_dirty = 0;
1925 rootzp->z_is_sa = USE_SA(version, os);
1926 rootzp->z_pflags = 0;
1927
1928 vp = ZTOV(rootzp);
1929 vn_reinit(vp);
1930 vp->v_type = VDIR;
1931
1932 zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
1933 zfsvfs->z_os = os;
1934 zfsvfs->z_parent = zfsvfs;
1935 zfsvfs->z_version = version;
1936 zfsvfs->z_use_fuids = USE_FUIDS(version, os);
1937 zfsvfs->z_use_sa = USE_SA(version, os);
1938 zfsvfs->z_norm = norm;
1939
1940 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
1941 &zfsvfs->z_attr_table);
1942
1943 ASSERT(error == 0);
1944
1945 /*
1946 * Fold case on file systems that are always or sometimes case
1947 * insensitive.
1948 */
1949 if (sense == ZFS_CASE_INSENSITIVE || sense == ZFS_CASE_MIXED)
1950 zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER;
1951
1952 mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
1953 list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
1954 offsetof(znode_t, z_link_node));
1955
1956 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1957 mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL);
1958
1959 rootzp->z_zfsvfs = zfsvfs;
1960 VERIFY(0 == zfs_acl_ids_create(rootzp, IS_ROOT_NODE, &vattr,
1961 cr, NULL, &acl_ids));
1962 zfs_mknode(rootzp, &vattr, tx, cr, IS_ROOT_NODE, &zp, &acl_ids);
1963 ASSERT3P(zp, ==, rootzp);
1964 ASSERT(!vn_in_dnlc(ZTOV(rootzp))); /* not valid to move */
1965 error = zap_add(os, moid, ZFS_ROOT_OBJ, 8, 1, &rootzp->z_id, tx);
1966 ASSERT(error == 0);
1967 zfs_acl_ids_free(&acl_ids);
1968 POINTER_INVALIDATE(&rootzp->z_zfsvfs);
1969
1970 ZTOV(rootzp)->v_count = 0;
1971 sa_handle_destroy(rootzp->z_sa_hdl);
1972 kmem_cache_free(znode_cache, rootzp);
1973
1974 /*
1975 * Create shares directory
1976 */
1977
1978 error = zfs_create_share_dir(zfsvfs, tx);
1979
1980 ASSERT(error == 0);
1981
1982 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1983 mutex_destroy(&zfsvfs->z_hold_mtx[i]);
1984 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1985 }
1986
1987 #endif /* _KERNEL */
1988
1989 static int
zfs_sa_setup(objset_t * osp,sa_attr_type_t ** sa_table)1990 zfs_sa_setup(objset_t *osp, sa_attr_type_t **sa_table)
1991 {
1992 uint64_t sa_obj = 0;
1993 int error;
1994
1995 error = zap_lookup(osp, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1, &sa_obj);
1996 if (error != 0 && error != ENOENT)
1997 return (error);
1998
1999 error = sa_setup(osp, sa_obj, zfs_attr_table, ZPL_END, sa_table);
2000 return (error);
2001 }
2002
2003 static int
zfs_grab_sa_handle(objset_t * osp,uint64_t obj,sa_handle_t ** hdlp,dmu_buf_t ** db,void * tag)2004 zfs_grab_sa_handle(objset_t *osp, uint64_t obj, sa_handle_t **hdlp,
2005 dmu_buf_t **db, void *tag)
2006 {
2007 dmu_object_info_t doi;
2008 int error;
2009
2010 if ((error = sa_buf_hold(osp, obj, tag, db)) != 0)
2011 return (error);
2012
2013 dmu_object_info_from_db(*db, &doi);
2014 if ((doi.doi_bonus_type != DMU_OT_SA &&
2015 doi.doi_bonus_type != DMU_OT_ZNODE) ||
2016 doi.doi_bonus_type == DMU_OT_ZNODE &&
2017 doi.doi_bonus_size < sizeof (znode_phys_t)) {
2018 sa_buf_rele(*db, tag);
2019 return (SET_ERROR(ENOTSUP));
2020 }
2021
2022 error = sa_handle_get(osp, obj, NULL, SA_HDL_PRIVATE, hdlp);
2023 if (error != 0) {
2024 sa_buf_rele(*db, tag);
2025 return (error);
2026 }
2027
2028 return (0);
2029 }
2030
2031 void
zfs_release_sa_handle(sa_handle_t * hdl,dmu_buf_t * db,void * tag)2032 zfs_release_sa_handle(sa_handle_t *hdl, dmu_buf_t *db, void *tag)
2033 {
2034 sa_handle_destroy(hdl);
2035 sa_buf_rele(db, tag);
2036 }
2037
2038 /*
2039 * Given an object number, return its parent object number and whether
2040 * or not the object is an extended attribute directory.
2041 */
2042 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)2043 zfs_obj_to_pobj(objset_t *osp, sa_handle_t *hdl, sa_attr_type_t *sa_table,
2044 uint64_t *pobjp, int *is_xattrdir)
2045 {
2046 uint64_t parent;
2047 uint64_t pflags;
2048 uint64_t mode;
2049 uint64_t parent_mode;
2050 sa_bulk_attr_t bulk[3];
2051 sa_handle_t *sa_hdl;
2052 dmu_buf_t *sa_db;
2053 int count = 0;
2054 int error;
2055
2056 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_PARENT], NULL,
2057 &parent, sizeof (parent));
2058 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_FLAGS], NULL,
2059 &pflags, sizeof (pflags));
2060 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL,
2061 &mode, sizeof (mode));
2062
2063 if ((error = sa_bulk_lookup(hdl, bulk, count)) != 0)
2064 return (error);
2065
2066 /*
2067 * When a link is removed its parent pointer is not changed and will
2068 * be invalid. There are two cases where a link is removed but the
2069 * file stays around, when it goes to the delete queue and when there
2070 * are additional links.
2071 */
2072 error = zfs_grab_sa_handle(osp, parent, &sa_hdl, &sa_db, FTAG);
2073 if (error != 0)
2074 return (error);
2075
2076 error = sa_lookup(sa_hdl, ZPL_MODE, &parent_mode, sizeof (parent_mode));
2077 zfs_release_sa_handle(sa_hdl, sa_db, FTAG);
2078 if (error != 0)
2079 return (error);
2080
2081 *is_xattrdir = ((pflags & ZFS_XATTR) != 0) && S_ISDIR(mode);
2082
2083 /*
2084 * Extended attributes can be applied to files, directories, etc.
2085 * Otherwise the parent must be a directory.
2086 */
2087 if (!*is_xattrdir && !S_ISDIR(parent_mode))
2088 return (SET_ERROR(EINVAL));
2089
2090 *pobjp = parent;
2091
2092 return (0);
2093 }
2094
2095 /*
2096 * Given an object number, return some zpl level statistics
2097 */
2098 static int
zfs_obj_to_stats_impl(sa_handle_t * hdl,sa_attr_type_t * sa_table,zfs_stat_t * sb)2099 zfs_obj_to_stats_impl(sa_handle_t *hdl, sa_attr_type_t *sa_table,
2100 zfs_stat_t *sb)
2101 {
2102 sa_bulk_attr_t bulk[4];
2103 int count = 0;
2104
2105 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL,
2106 &sb->zs_mode, sizeof (sb->zs_mode));
2107 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_GEN], NULL,
2108 &sb->zs_gen, sizeof (sb->zs_gen));
2109 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_LINKS], NULL,
2110 &sb->zs_links, sizeof (sb->zs_links));
2111 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_CTIME], NULL,
2112 &sb->zs_ctime, sizeof (sb->zs_ctime));
2113
2114 return (sa_bulk_lookup(hdl, bulk, count));
2115 }
2116
2117 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)2118 zfs_obj_to_path_impl(objset_t *osp, uint64_t obj, sa_handle_t *hdl,
2119 sa_attr_type_t *sa_table, char *buf, int len)
2120 {
2121 sa_handle_t *sa_hdl;
2122 sa_handle_t *prevhdl = NULL;
2123 dmu_buf_t *prevdb = NULL;
2124 dmu_buf_t *sa_db = NULL;
2125 char *path = buf + len - 1;
2126 int error;
2127
2128 *path = '\0';
2129 sa_hdl = hdl;
2130
2131 uint64_t deleteq_obj;
2132 VERIFY0(zap_lookup(osp, MASTER_NODE_OBJ,
2133 ZFS_UNLINKED_SET, sizeof (uint64_t), 1, &deleteq_obj));
2134 error = zap_lookup_int(osp, deleteq_obj, obj);
2135 if (error == 0) {
2136 return (ESTALE);
2137 } else if (error != ENOENT) {
2138 return (error);
2139 }
2140 error = 0;
2141
2142 for (;;) {
2143 uint64_t pobj;
2144 char component[MAXNAMELEN + 2];
2145 size_t complen;
2146 int is_xattrdir;
2147
2148 if (prevdb)
2149 zfs_release_sa_handle(prevhdl, prevdb, FTAG);
2150
2151 if ((error = zfs_obj_to_pobj(osp, sa_hdl, sa_table, &pobj,
2152 &is_xattrdir)) != 0)
2153 break;
2154
2155 if (pobj == obj) {
2156 if (path[0] != '/')
2157 *--path = '/';
2158 break;
2159 }
2160
2161 component[0] = '/';
2162 if (is_xattrdir) {
2163 (void) sprintf(component + 1, "<xattrdir>");
2164 } else {
2165 error = zap_value_search(osp, pobj, obj,
2166 ZFS_DIRENT_OBJ(-1ULL), component + 1);
2167 if (error != 0)
2168 break;
2169 }
2170
2171 complen = strlen(component);
2172 path -= complen;
2173 ASSERT(path >= buf);
2174 bcopy(component, path, complen);
2175 obj = pobj;
2176
2177 if (sa_hdl != hdl) {
2178 prevhdl = sa_hdl;
2179 prevdb = sa_db;
2180 }
2181 error = zfs_grab_sa_handle(osp, obj, &sa_hdl, &sa_db, FTAG);
2182 if (error != 0) {
2183 sa_hdl = prevhdl;
2184 sa_db = prevdb;
2185 break;
2186 }
2187 }
2188
2189 if (sa_hdl != NULL && sa_hdl != hdl) {
2190 ASSERT(sa_db != NULL);
2191 zfs_release_sa_handle(sa_hdl, sa_db, FTAG);
2192 }
2193
2194 if (error == 0)
2195 (void) memmove(buf, path, buf + len - path);
2196
2197 return (error);
2198 }
2199
2200 int
zfs_obj_to_path(objset_t * osp,uint64_t obj,char * buf,int len)2201 zfs_obj_to_path(objset_t *osp, uint64_t obj, char *buf, int len)
2202 {
2203 sa_attr_type_t *sa_table;
2204 sa_handle_t *hdl;
2205 dmu_buf_t *db;
2206 int error;
2207
2208 error = zfs_sa_setup(osp, &sa_table);
2209 if (error != 0)
2210 return (error);
2211
2212 error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG);
2213 if (error != 0)
2214 return (error);
2215
2216 error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len);
2217
2218 zfs_release_sa_handle(hdl, db, FTAG);
2219 return (error);
2220 }
2221
2222 int
zfs_obj_to_stats(objset_t * osp,uint64_t obj,zfs_stat_t * sb,char * buf,int len)2223 zfs_obj_to_stats(objset_t *osp, uint64_t obj, zfs_stat_t *sb,
2224 char *buf, int len)
2225 {
2226 char *path = buf + len - 1;
2227 sa_attr_type_t *sa_table;
2228 sa_handle_t *hdl;
2229 dmu_buf_t *db;
2230 int error;
2231
2232 *path = '\0';
2233
2234 error = zfs_sa_setup(osp, &sa_table);
2235 if (error != 0)
2236 return (error);
2237
2238 error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG);
2239 if (error != 0)
2240 return (error);
2241
2242 error = zfs_obj_to_stats_impl(hdl, sa_table, sb);
2243 if (error != 0) {
2244 zfs_release_sa_handle(hdl, db, FTAG);
2245 return (error);
2246 }
2247
2248 error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len);
2249
2250 zfs_release_sa_handle(hdl, db, FTAG);
2251 return (error);
2252 }
2253