1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21 /*
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2012, 2014 by Delphix. All rights reserved.
24 * Copyright (c) 2014 Integros [integros.com]
25 */
26
27 /* Portions Copyright 2007 Jeremy Teo */
28 /* Portions Copyright 2011 Martin Matuska <[email protected]> */
29
30 #ifdef _KERNEL
31 #include <sys/types.h>
32 #include <sys/param.h>
33 #include <sys/time.h>
34 #include <sys/systm.h>
35 #include <sys/sysmacros.h>
36 #include <sys/resource.h>
37 #include <sys/mntent.h>
38 #include <sys/u8_textprep.h>
39 #include <sys/dsl_dataset.h>
40 #include <sys/vfs.h>
41 #include <sys/vnode.h>
42 #include <sys/file.h>
43 #include <sys/kmem.h>
44 #include <sys/errno.h>
45 #include <sys/unistd.h>
46 #include <sys/atomic.h>
47 #include <sys/zfs_dir.h>
48 #include <sys/zfs_acl.h>
49 #include <sys/zfs_ioctl.h>
50 #include <sys/zfs_rlock.h>
51 #include <sys/zfs_fuid.h>
52 #include <sys/dnode.h>
53 #include <sys/fs/zfs.h>
54 #include <sys/kidmap.h>
55 #endif /* _KERNEL */
56
57 #include <sys/dmu.h>
58 #include <sys/dmu_objset.h>
59 #include <sys/dmu_tx.h>
60 #include <sys/refcount.h>
61 #include <sys/stat.h>
62 #include <sys/zap.h>
63 #include <sys/zfs_znode.h>
64 #include <sys/sa.h>
65 #include <sys/zfs_sa.h>
66 #include <sys/zfs_stat.h>
67 #include <sys/refcount.h>
68
69 #include "zfs_prop.h"
70 #include "zfs_comutil.h"
71
72 /* Used by fstat(1). */
73 SYSCTL_INT(_debug_sizeof, OID_AUTO, znode, CTLFLAG_RD,
74 SYSCTL_NULL_INT_PTR, sizeof(znode_t), "sizeof(znode_t)");
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 extern struct vop_vector zfs_vnodeops;
116 extern struct vop_vector zfs_fifoops;
117 extern struct vop_vector zfs_shareops;
118
119 static int
zfs_znode_cache_constructor(void * buf,void * arg,int kmflags)120 zfs_znode_cache_constructor(void *buf, void *arg, int kmflags)
121 {
122 znode_t *zp = buf;
123
124 POINTER_INVALIDATE(&zp->z_zfsvfs);
125
126 list_link_init(&zp->z_link_node);
127
128 mutex_init(&zp->z_acl_lock, NULL, MUTEX_DEFAULT, NULL);
129
130 mutex_init(&zp->z_range_lock, NULL, MUTEX_DEFAULT, NULL);
131 avl_create(&zp->z_range_avl, zfs_range_compare,
132 sizeof (rl_t), offsetof(rl_t, r_node));
133
134 zp->z_acl_cached = NULL;
135 zp->z_vnode = NULL;
136 zp->z_moved = 0;
137 return (0);
138 }
139
140 /*ARGSUSED*/
141 static void
zfs_znode_cache_destructor(void * buf,void * arg)142 zfs_znode_cache_destructor(void *buf, void *arg)
143 {
144 znode_t *zp = buf;
145
146 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs));
147 ASSERT3P(zp->z_vnode, ==, NULL);
148 ASSERT(!list_link_active(&zp->z_link_node));
149 mutex_destroy(&zp->z_acl_lock);
150 avl_destroy(&zp->z_range_avl);
151 mutex_destroy(&zp->z_range_lock);
152
153 ASSERT(zp->z_acl_cached == NULL);
154 }
155
156 #ifdef ZNODE_STATS
157 static struct {
158 uint64_t zms_zfsvfs_invalid;
159 uint64_t zms_zfsvfs_recheck1;
160 uint64_t zms_zfsvfs_unmounted;
161 uint64_t zms_zfsvfs_recheck2;
162 uint64_t zms_obj_held;
163 uint64_t zms_vnode_locked;
164 uint64_t zms_not_only_dnlc;
165 } znode_move_stats;
166 #endif /* ZNODE_STATS */
167
168 #ifdef illumos
169 static void
zfs_znode_move_impl(znode_t * ozp,znode_t * nzp)170 zfs_znode_move_impl(znode_t *ozp, znode_t *nzp)
171 {
172 vnode_t *vp;
173
174 /* Copy fields. */
175 nzp->z_zfsvfs = ozp->z_zfsvfs;
176
177 /* Swap vnodes. */
178 vp = nzp->z_vnode;
179 nzp->z_vnode = ozp->z_vnode;
180 ozp->z_vnode = vp; /* let destructor free the overwritten vnode */
181 ZTOV(ozp)->v_data = ozp;
182 ZTOV(nzp)->v_data = nzp;
183
184 nzp->z_id = ozp->z_id;
185 ASSERT(ozp->z_dirlocks == NULL); /* znode not in use */
186 ASSERT(avl_numnodes(&ozp->z_range_avl) == 0);
187 nzp->z_unlinked = ozp->z_unlinked;
188 nzp->z_atime_dirty = ozp->z_atime_dirty;
189 nzp->z_zn_prefetch = ozp->z_zn_prefetch;
190 nzp->z_blksz = ozp->z_blksz;
191 nzp->z_seq = ozp->z_seq;
192 nzp->z_mapcnt = ozp->z_mapcnt;
193 nzp->z_gen = ozp->z_gen;
194 nzp->z_sync_cnt = ozp->z_sync_cnt;
195 nzp->z_is_sa = ozp->z_is_sa;
196 nzp->z_sa_hdl = ozp->z_sa_hdl;
197 bcopy(ozp->z_atime, nzp->z_atime, sizeof (uint64_t) * 2);
198 nzp->z_links = ozp->z_links;
199 nzp->z_size = ozp->z_size;
200 nzp->z_pflags = ozp->z_pflags;
201 nzp->z_uid = ozp->z_uid;
202 nzp->z_gid = ozp->z_gid;
203 nzp->z_mode = ozp->z_mode;
204
205 /*
206 * Since this is just an idle znode and kmem is already dealing with
207 * memory pressure, release any cached ACL.
208 */
209 if (ozp->z_acl_cached) {
210 zfs_acl_free(ozp->z_acl_cached);
211 ozp->z_acl_cached = NULL;
212 }
213
214 sa_set_userp(nzp->z_sa_hdl, nzp);
215
216 /*
217 * Invalidate the original znode by clearing fields that provide a
218 * pointer back to the znode. Set the low bit of the vfs pointer to
219 * ensure that zfs_znode_move() recognizes the znode as invalid in any
220 * subsequent callback.
221 */
222 ozp->z_sa_hdl = NULL;
223 POINTER_INVALIDATE(&ozp->z_zfsvfs);
224
225 /*
226 * Mark the znode.
227 */
228 nzp->z_moved = 1;
229 ozp->z_moved = (uint8_t)-1;
230 }
231
232 /*ARGSUSED*/
233 static kmem_cbrc_t
zfs_znode_move(void * buf,void * newbuf,size_t size,void * arg)234 zfs_znode_move(void *buf, void *newbuf, size_t size, void *arg)
235 {
236 znode_t *ozp = buf, *nzp = newbuf;
237 zfsvfs_t *zfsvfs;
238 vnode_t *vp;
239
240 /*
241 * The znode is on the file system's list of known znodes if the vfs
242 * pointer is valid. We set the low bit of the vfs pointer when freeing
243 * the znode to invalidate it, and the memory patterns written by kmem
244 * (baddcafe and deadbeef) set at least one of the two low bits. A newly
245 * created znode sets the vfs pointer last of all to indicate that the
246 * znode is known and in a valid state to be moved by this function.
247 */
248 zfsvfs = ozp->z_zfsvfs;
249 if (!POINTER_IS_VALID(zfsvfs)) {
250 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_invalid);
251 return (KMEM_CBRC_DONT_KNOW);
252 }
253
254 /*
255 * Close a small window in which it's possible that the filesystem could
256 * be unmounted and freed, and zfsvfs, though valid in the previous
257 * statement, could point to unrelated memory by the time we try to
258 * prevent the filesystem from being unmounted.
259 */
260 rw_enter(&zfsvfs_lock, RW_WRITER);
261 if (zfsvfs != ozp->z_zfsvfs) {
262 rw_exit(&zfsvfs_lock);
263 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_recheck1);
264 return (KMEM_CBRC_DONT_KNOW);
265 }
266
267 /*
268 * If the znode is still valid, then so is the file system. We know that
269 * no valid file system can be freed while we hold zfsvfs_lock, so we
270 * can safely ensure that the filesystem is not and will not be
271 * unmounted. The next statement is equivalent to ZFS_ENTER().
272 */
273 rrm_enter(&zfsvfs->z_teardown_lock, RW_READER, FTAG);
274 if (zfsvfs->z_unmounted) {
275 ZFS_EXIT(zfsvfs);
276 rw_exit(&zfsvfs_lock);
277 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_unmounted);
278 return (KMEM_CBRC_DONT_KNOW);
279 }
280 rw_exit(&zfsvfs_lock);
281
282 mutex_enter(&zfsvfs->z_znodes_lock);
283 /*
284 * Recheck the vfs pointer in case the znode was removed just before
285 * acquiring the lock.
286 */
287 if (zfsvfs != ozp->z_zfsvfs) {
288 mutex_exit(&zfsvfs->z_znodes_lock);
289 ZFS_EXIT(zfsvfs);
290 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_recheck2);
291 return (KMEM_CBRC_DONT_KNOW);
292 }
293
294 /*
295 * At this point we know that as long as we hold z_znodes_lock, the
296 * znode cannot be freed and fields within the znode can be safely
297 * accessed. Now, prevent a race with zfs_zget().
298 */
299 if (ZFS_OBJ_HOLD_TRYENTER(zfsvfs, ozp->z_id) == 0) {
300 mutex_exit(&zfsvfs->z_znodes_lock);
301 ZFS_EXIT(zfsvfs);
302 ZNODE_STAT_ADD(znode_move_stats.zms_obj_held);
303 return (KMEM_CBRC_LATER);
304 }
305
306 vp = ZTOV(ozp);
307 if (mutex_tryenter(&vp->v_lock) == 0) {
308 ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id);
309 mutex_exit(&zfsvfs->z_znodes_lock);
310 ZFS_EXIT(zfsvfs);
311 ZNODE_STAT_ADD(znode_move_stats.zms_vnode_locked);
312 return (KMEM_CBRC_LATER);
313 }
314
315 /* Only move znodes that are referenced _only_ by the DNLC. */
316 if (vp->v_count != 1 || !vn_in_dnlc(vp)) {
317 mutex_exit(&vp->v_lock);
318 ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id);
319 mutex_exit(&zfsvfs->z_znodes_lock);
320 ZFS_EXIT(zfsvfs);
321 ZNODE_STAT_ADD(znode_move_stats.zms_not_only_dnlc);
322 return (KMEM_CBRC_LATER);
323 }
324
325 /*
326 * The znode is known and in a valid state to move. We're holding the
327 * locks needed to execute the critical section.
328 */
329 zfs_znode_move_impl(ozp, nzp);
330 mutex_exit(&vp->v_lock);
331 ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id);
332
333 list_link_replace(&ozp->z_link_node, &nzp->z_link_node);
334 mutex_exit(&zfsvfs->z_znodes_lock);
335 ZFS_EXIT(zfsvfs);
336
337 return (KMEM_CBRC_YES);
338 }
339 #endif /* illumos */
340
341 void
zfs_znode_init(void)342 zfs_znode_init(void)
343 {
344 /*
345 * Initialize zcache
346 */
347 rw_init(&zfsvfs_lock, NULL, RW_DEFAULT, NULL);
348 ASSERT(znode_cache == NULL);
349 znode_cache = kmem_cache_create("zfs_znode_cache",
350 sizeof (znode_t), 0, zfs_znode_cache_constructor,
351 zfs_znode_cache_destructor, NULL, NULL, NULL, 0);
352 kmem_cache_set_move(znode_cache, zfs_znode_move);
353 }
354
355 void
zfs_znode_fini(void)356 zfs_znode_fini(void)
357 {
358 #ifdef illumos
359 /*
360 * Cleanup vfs & vnode ops
361 */
362 zfs_remove_op_tables();
363 #endif
364
365 /*
366 * Cleanup zcache
367 */
368 if (znode_cache)
369 kmem_cache_destroy(znode_cache);
370 znode_cache = NULL;
371 rw_destroy(&zfsvfs_lock);
372 }
373
374 #ifdef illumos
375 struct vnodeops *zfs_dvnodeops;
376 struct vnodeops *zfs_fvnodeops;
377 struct vnodeops *zfs_symvnodeops;
378 struct vnodeops *zfs_xdvnodeops;
379 struct vnodeops *zfs_evnodeops;
380 struct vnodeops *zfs_sharevnodeops;
381
382 void
zfs_remove_op_tables()383 zfs_remove_op_tables()
384 {
385 /*
386 * Remove vfs ops
387 */
388 ASSERT(zfsfstype);
389 (void) vfs_freevfsops_by_type(zfsfstype);
390 zfsfstype = 0;
391
392 /*
393 * Remove vnode ops
394 */
395 if (zfs_dvnodeops)
396 vn_freevnodeops(zfs_dvnodeops);
397 if (zfs_fvnodeops)
398 vn_freevnodeops(zfs_fvnodeops);
399 if (zfs_symvnodeops)
400 vn_freevnodeops(zfs_symvnodeops);
401 if (zfs_xdvnodeops)
402 vn_freevnodeops(zfs_xdvnodeops);
403 if (zfs_evnodeops)
404 vn_freevnodeops(zfs_evnodeops);
405 if (zfs_sharevnodeops)
406 vn_freevnodeops(zfs_sharevnodeops);
407
408 zfs_dvnodeops = NULL;
409 zfs_fvnodeops = NULL;
410 zfs_symvnodeops = NULL;
411 zfs_xdvnodeops = NULL;
412 zfs_evnodeops = NULL;
413 zfs_sharevnodeops = NULL;
414 }
415
416 extern const fs_operation_def_t zfs_dvnodeops_template[];
417 extern const fs_operation_def_t zfs_fvnodeops_template[];
418 extern const fs_operation_def_t zfs_xdvnodeops_template[];
419 extern const fs_operation_def_t zfs_symvnodeops_template[];
420 extern const fs_operation_def_t zfs_evnodeops_template[];
421 extern const fs_operation_def_t zfs_sharevnodeops_template[];
422
423 int
zfs_create_op_tables()424 zfs_create_op_tables()
425 {
426 int error;
427
428 /*
429 * zfs_dvnodeops can be set if mod_remove() calls mod_installfs()
430 * due to a failure to remove the the 2nd modlinkage (zfs_modldrv).
431 * In this case we just return as the ops vectors are already set up.
432 */
433 if (zfs_dvnodeops)
434 return (0);
435
436 error = vn_make_ops(MNTTYPE_ZFS, zfs_dvnodeops_template,
437 &zfs_dvnodeops);
438 if (error)
439 return (error);
440
441 error = vn_make_ops(MNTTYPE_ZFS, zfs_fvnodeops_template,
442 &zfs_fvnodeops);
443 if (error)
444 return (error);
445
446 error = vn_make_ops(MNTTYPE_ZFS, zfs_symvnodeops_template,
447 &zfs_symvnodeops);
448 if (error)
449 return (error);
450
451 error = vn_make_ops(MNTTYPE_ZFS, zfs_xdvnodeops_template,
452 &zfs_xdvnodeops);
453 if (error)
454 return (error);
455
456 error = vn_make_ops(MNTTYPE_ZFS, zfs_evnodeops_template,
457 &zfs_evnodeops);
458 if (error)
459 return (error);
460
461 error = vn_make_ops(MNTTYPE_ZFS, zfs_sharevnodeops_template,
462 &zfs_sharevnodeops);
463
464 return (error);
465 }
466 #endif /* illumos */
467
468 int
zfs_create_share_dir(zfsvfs_t * zfsvfs,dmu_tx_t * tx)469 zfs_create_share_dir(zfsvfs_t *zfsvfs, dmu_tx_t *tx)
470 {
471 zfs_acl_ids_t acl_ids;
472 vattr_t vattr;
473 znode_t *sharezp;
474 znode_t *zp;
475 int error;
476
477 vattr.va_mask = AT_MODE|AT_UID|AT_GID|AT_TYPE;
478 vattr.va_type = VDIR;
479 vattr.va_mode = S_IFDIR|0555;
480 vattr.va_uid = crgetuid(kcred);
481 vattr.va_gid = crgetgid(kcred);
482
483 sharezp = kmem_cache_alloc(znode_cache, KM_SLEEP);
484 ASSERT(!POINTER_IS_VALID(sharezp->z_zfsvfs));
485 sharezp->z_moved = 0;
486 sharezp->z_unlinked = 0;
487 sharezp->z_atime_dirty = 0;
488 sharezp->z_zfsvfs = zfsvfs;
489 sharezp->z_is_sa = zfsvfs->z_use_sa;
490
491 VERIFY(0 == zfs_acl_ids_create(sharezp, IS_ROOT_NODE, &vattr,
492 kcred, NULL, &acl_ids));
493 zfs_mknode(sharezp, &vattr, tx, kcred, IS_ROOT_NODE, &zp, &acl_ids);
494 ASSERT3P(zp, ==, sharezp);
495 POINTER_INVALIDATE(&sharezp->z_zfsvfs);
496 error = zap_add(zfsvfs->z_os, MASTER_NODE_OBJ,
497 ZFS_SHARES_DIR, 8, 1, &sharezp->z_id, tx);
498 zfsvfs->z_shares_dir = sharezp->z_id;
499
500 zfs_acl_ids_free(&acl_ids);
501 sa_handle_destroy(sharezp->z_sa_hdl);
502 kmem_cache_free(znode_cache, sharezp);
503
504 return (error);
505 }
506
507 /*
508 * define a couple of values we need available
509 * for both 64 and 32 bit environments.
510 */
511 #ifndef NBITSMINOR64
512 #define NBITSMINOR64 32
513 #endif
514 #ifndef MAXMAJ64
515 #define MAXMAJ64 0xffffffffUL
516 #endif
517 #ifndef MAXMIN64
518 #define MAXMIN64 0xffffffffUL
519 #endif
520
521 /*
522 * Create special expldev for ZFS private use.
523 * Can't use standard expldev since it doesn't do
524 * what we want. The standard expldev() takes a
525 * dev32_t in LP64 and expands it to a long dev_t.
526 * We need an interface that takes a dev32_t in ILP32
527 * and expands it to a long dev_t.
528 */
529 static uint64_t
zfs_expldev(dev_t dev)530 zfs_expldev(dev_t dev)
531 {
532 return (((uint64_t)major(dev) << NBITSMINOR64) | minor(dev));
533 }
534 /*
535 * Special cmpldev for ZFS private use.
536 * Can't use standard cmpldev since it takes
537 * a long dev_t and compresses it to dev32_t in
538 * LP64. We need to do a compaction of a long dev_t
539 * to a dev32_t in ILP32.
540 */
541 dev_t
zfs_cmpldev(uint64_t dev)542 zfs_cmpldev(uint64_t dev)
543 {
544 return (makedev((dev >> NBITSMINOR64), (dev & MAXMIN64)));
545 }
546
547 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)548 zfs_znode_sa_init(zfsvfs_t *zfsvfs, znode_t *zp,
549 dmu_buf_t *db, dmu_object_type_t obj_type, sa_handle_t *sa_hdl)
550 {
551 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs) || (zfsvfs == zp->z_zfsvfs));
552 ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zfsvfs, zp->z_id)));
553
554 ASSERT(zp->z_sa_hdl == NULL);
555 ASSERT(zp->z_acl_cached == NULL);
556 if (sa_hdl == NULL) {
557 VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, zp,
558 SA_HDL_SHARED, &zp->z_sa_hdl));
559 } else {
560 zp->z_sa_hdl = sa_hdl;
561 sa_set_userp(sa_hdl, zp);
562 }
563
564 zp->z_is_sa = (obj_type == DMU_OT_SA) ? B_TRUE : B_FALSE;
565
566 /*
567 * Slap on VROOT if we are the root znode unless we are the root
568 * node of a snapshot mounted under .zfs.
569 */
570 if (zp->z_id == zfsvfs->z_root && zfsvfs->z_parent == zfsvfs)
571 ZTOV(zp)->v_flag |= VROOT;
572
573 vn_exists(ZTOV(zp));
574 }
575
576 void
zfs_znode_dmu_fini(znode_t * zp)577 zfs_znode_dmu_fini(znode_t *zp)
578 {
579 ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zp->z_zfsvfs, zp->z_id)) ||
580 zp->z_unlinked ||
581 RW_WRITE_HELD(&zp->z_zfsvfs->z_teardown_inactive_lock));
582
583 sa_handle_destroy(zp->z_sa_hdl);
584 zp->z_sa_hdl = NULL;
585 }
586
587 static void
zfs_vnode_forget(vnode_t * vp)588 zfs_vnode_forget(vnode_t *vp)
589 {
590
591 /* copied from insmntque_stddtr */
592 vp->v_data = NULL;
593 vp->v_op = &dead_vnodeops;
594 vgone(vp);
595 vput(vp);
596 }
597
598 /*
599 * Construct a new znode/vnode and intialize.
600 *
601 * This does not do a call to dmu_set_user() that is
602 * up to the caller to do, in case you don't want to
603 * return the znode
604 */
605 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)606 zfs_znode_alloc(zfsvfs_t *zfsvfs, dmu_buf_t *db, int blksz,
607 dmu_object_type_t obj_type, sa_handle_t *hdl)
608 {
609 znode_t *zp;
610 vnode_t *vp;
611 uint64_t mode;
612 uint64_t parent;
613 sa_bulk_attr_t bulk[9];
614 int count = 0;
615 int error;
616
617 zp = kmem_cache_alloc(znode_cache, KM_SLEEP);
618
619 KASSERT(curthread->td_vp_reserv > 0,
620 ("zfs_znode_alloc: getnewvnode without any vnodes reserved"));
621 error = getnewvnode("zfs", zfsvfs->z_parent->z_vfs, &zfs_vnodeops, &vp);
622 if (error != 0) {
623 kmem_cache_free(znode_cache, zp);
624 return (NULL);
625 }
626 zp->z_vnode = vp;
627 vp->v_data = zp;
628
629 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs));
630 zp->z_moved = 0;
631
632 /*
633 * Defer setting z_zfsvfs until the znode is ready to be a candidate for
634 * the zfs_znode_move() callback.
635 */
636 zp->z_sa_hdl = NULL;
637 zp->z_unlinked = 0;
638 zp->z_atime_dirty = 0;
639 zp->z_mapcnt = 0;
640 zp->z_id = db->db_object;
641 zp->z_blksz = blksz;
642 zp->z_seq = 0x7A4653;
643 zp->z_sync_cnt = 0;
644
645 vp = ZTOV(zp);
646
647 zfs_znode_sa_init(zfsvfs, zp, db, obj_type, hdl);
648
649 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8);
650 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL, &zp->z_gen, 8);
651 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
652 &zp->z_size, 8);
653 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL,
654 &zp->z_links, 8);
655 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
656 &zp->z_pflags, 8);
657 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL, &parent, 8);
658 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
659 &zp->z_atime, 16);
660 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
661 &zp->z_uid, 8);
662 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL,
663 &zp->z_gid, 8);
664
665 if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count) != 0 || zp->z_gen == 0) {
666 if (hdl == NULL)
667 sa_handle_destroy(zp->z_sa_hdl);
668 zfs_vnode_forget(vp);
669 zp->z_vnode = NULL;
670 kmem_cache_free(znode_cache, zp);
671 return (NULL);
672 }
673
674 zp->z_mode = mode;
675
676 vp->v_type = IFTOVT((mode_t)mode);
677
678 switch (vp->v_type) {
679 case VDIR:
680 zp->z_zn_prefetch = B_TRUE; /* z_prefetch default is enabled */
681 break;
682 #ifdef illumos
683 case VBLK:
684 case VCHR:
685 {
686 uint64_t rdev;
687 VERIFY(sa_lookup(zp->z_sa_hdl, SA_ZPL_RDEV(zfsvfs),
688 &rdev, sizeof (rdev)) == 0);
689
690 vp->v_rdev = zfs_cmpldev(rdev);
691 }
692 break;
693 #endif
694 case VFIFO:
695 #ifdef illumos
696 case VSOCK:
697 case VDOOR:
698 #endif
699 vp->v_op = &zfs_fifoops;
700 break;
701 case VREG:
702 if (parent == zfsvfs->z_shares_dir) {
703 ASSERT(zp->z_uid == 0 && zp->z_gid == 0);
704 vp->v_op = &zfs_shareops;
705 }
706 break;
707 #ifdef illumos
708 case VLNK:
709 vn_setops(vp, zfs_symvnodeops);
710 break;
711 default:
712 vn_setops(vp, zfs_evnodeops);
713 break;
714 #endif
715 }
716
717 mutex_enter(&zfsvfs->z_znodes_lock);
718 list_insert_tail(&zfsvfs->z_all_znodes, zp);
719 membar_producer();
720 /*
721 * Everything else must be valid before assigning z_zfsvfs makes the
722 * znode eligible for zfs_znode_move().
723 */
724 zp->z_zfsvfs = zfsvfs;
725 mutex_exit(&zfsvfs->z_znodes_lock);
726
727 /*
728 * Acquire vnode lock before making it available to the world.
729 */
730 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
731 VN_LOCK_AREC(vp);
732 if (vp->v_type != VFIFO)
733 VN_LOCK_ASHARE(vp);
734
735 #ifdef illumos
736 VFS_HOLD(zfsvfs->z_vfs);
737 #endif
738 return (zp);
739 }
740
741 static uint64_t empty_xattr;
742 static uint64_t pad[4];
743 static zfs_acl_phys_t acl_phys;
744 /*
745 * Create a new DMU object to hold a zfs znode.
746 *
747 * IN: dzp - parent directory for new znode
748 * vap - file attributes for new znode
749 * tx - dmu transaction id for zap operations
750 * cr - credentials of caller
751 * flag - flags:
752 * IS_ROOT_NODE - new object will be root
753 * IS_XATTR - new object is an attribute
754 * bonuslen - length of bonus buffer
755 * setaclp - File/Dir initial ACL
756 * fuidp - Tracks fuid allocation.
757 *
758 * OUT: zpp - allocated znode
759 *
760 */
761 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)762 zfs_mknode(znode_t *dzp, vattr_t *vap, dmu_tx_t *tx, cred_t *cr,
763 uint_t flag, znode_t **zpp, zfs_acl_ids_t *acl_ids)
764 {
765 uint64_t crtime[2], atime[2], mtime[2], ctime[2];
766 uint64_t mode, size, links, parent, pflags;
767 uint64_t dzp_pflags = 0;
768 uint64_t rdev = 0;
769 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
770 dmu_buf_t *db;
771 timestruc_t now;
772 uint64_t gen, obj;
773 int err;
774 int bonuslen;
775 int dnodesize;
776 sa_handle_t *sa_hdl;
777 dmu_object_type_t obj_type;
778 sa_bulk_attr_t *sa_attrs;
779 int cnt = 0;
780 zfs_acl_locator_cb_t locate = { 0 };
781
782 ASSERT(vap && (vap->va_mask & (AT_TYPE|AT_MODE)) == (AT_TYPE|AT_MODE));
783
784 if (zfsvfs->z_replay) {
785 obj = vap->va_nodeid;
786 now = vap->va_ctime; /* see zfs_replay_create() */
787 gen = vap->va_nblocks; /* ditto */
788 dnodesize = vap->va_fsid; /* ditto */
789 } else {
790 obj = 0;
791 vfs_timestamp(&now);
792 gen = dmu_tx_get_txg(tx);
793 dnodesize = dmu_objset_dnodesize(zfsvfs->z_os);
794 }
795
796 if (dnodesize == 0)
797 dnodesize = DNODE_MIN_SIZE;
798
799 obj_type = zfsvfs->z_use_sa ? DMU_OT_SA : DMU_OT_ZNODE;
800 bonuslen = (obj_type == DMU_OT_SA) ?
801 DN_BONUS_SIZE(dnodesize) : ZFS_OLD_ZNODE_PHYS_SIZE;
802
803 /*
804 * Create a new DMU object.
805 */
806 /*
807 * There's currently no mechanism for pre-reading the blocks that will
808 * be needed to allocate a new object, so we accept the small chance
809 * that there will be an i/o error and we will fail one of the
810 * assertions below.
811 */
812 if (vap->va_type == VDIR) {
813 if (zfsvfs->z_replay) {
814 VERIFY0(zap_create_claim_norm_dnsize(zfsvfs->z_os, obj,
815 zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS,
816 obj_type, bonuslen, dnodesize, tx));
817 } else {
818 obj = zap_create_norm_dnsize(zfsvfs->z_os,
819 zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS,
820 obj_type, bonuslen, dnodesize, tx);
821 }
822 } else {
823 if (zfsvfs->z_replay) {
824 VERIFY0(dmu_object_claim_dnsize(zfsvfs->z_os, obj,
825 DMU_OT_PLAIN_FILE_CONTENTS, 0,
826 obj_type, bonuslen, dnodesize, tx));
827 } else {
828 obj = dmu_object_alloc_dnsize(zfsvfs->z_os,
829 DMU_OT_PLAIN_FILE_CONTENTS, 0,
830 obj_type, bonuslen, dnodesize, tx);
831 }
832 }
833
834 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj);
835 VERIFY(0 == sa_buf_hold(zfsvfs->z_os, obj, NULL, &db));
836
837 /*
838 * If this is the root, fix up the half-initialized parent pointer
839 * to reference the just-allocated physical data area.
840 */
841 if (flag & IS_ROOT_NODE) {
842 dzp->z_id = obj;
843 } else {
844 dzp_pflags = dzp->z_pflags;
845 }
846
847 /*
848 * If parent is an xattr, so am I.
849 */
850 if (dzp_pflags & ZFS_XATTR) {
851 flag |= IS_XATTR;
852 }
853
854 if (zfsvfs->z_use_fuids)
855 pflags = ZFS_ARCHIVE | ZFS_AV_MODIFIED;
856 else
857 pflags = 0;
858
859 if (vap->va_type == VDIR) {
860 size = 2; /* contents ("." and "..") */
861 links = (flag & (IS_ROOT_NODE | IS_XATTR)) ? 2 : 1;
862 } else {
863 size = links = 0;
864 }
865
866 if (vap->va_type == VBLK || vap->va_type == VCHR) {
867 rdev = zfs_expldev(vap->va_rdev);
868 }
869
870 parent = dzp->z_id;
871 mode = acl_ids->z_mode;
872 if (flag & IS_XATTR)
873 pflags |= ZFS_XATTR;
874
875 /*
876 * No execs denied will be deterimed when zfs_mode_compute() is called.
877 */
878 pflags |= acl_ids->z_aclp->z_hints &
879 (ZFS_ACL_TRIVIAL|ZFS_INHERIT_ACE|ZFS_ACL_AUTO_INHERIT|
880 ZFS_ACL_DEFAULTED|ZFS_ACL_PROTECTED);
881
882 ZFS_TIME_ENCODE(&now, crtime);
883 ZFS_TIME_ENCODE(&now, ctime);
884
885 if (vap->va_mask & AT_ATIME) {
886 ZFS_TIME_ENCODE(&vap->va_atime, atime);
887 } else {
888 ZFS_TIME_ENCODE(&now, atime);
889 }
890
891 if (vap->va_mask & AT_MTIME) {
892 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
893 } else {
894 ZFS_TIME_ENCODE(&now, mtime);
895 }
896
897 /* Now add in all of the "SA" attributes */
898 VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, NULL, SA_HDL_SHARED,
899 &sa_hdl));
900
901 /*
902 * Setup the array of attributes to be replaced/set on the new file
903 *
904 * order for DMU_OT_ZNODE is critical since it needs to be constructed
905 * in the old znode_phys_t format. Don't change this ordering
906 */
907 sa_attrs = kmem_alloc(sizeof (sa_bulk_attr_t) * ZPL_END, KM_SLEEP);
908
909 if (obj_type == DMU_OT_ZNODE) {
910 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs),
911 NULL, &atime, 16);
912 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs),
913 NULL, &mtime, 16);
914 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs),
915 NULL, &ctime, 16);
916 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs),
917 NULL, &crtime, 16);
918 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs),
919 NULL, &gen, 8);
920 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs),
921 NULL, &mode, 8);
922 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs),
923 NULL, &size, 8);
924 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs),
925 NULL, &parent, 8);
926 } else {
927 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs),
928 NULL, &mode, 8);
929 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs),
930 NULL, &size, 8);
931 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs),
932 NULL, &gen, 8);
933 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs),
934 NULL, &acl_ids->z_fuid, 8);
935 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs),
936 NULL, &acl_ids->z_fgid, 8);
937 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs),
938 NULL, &parent, 8);
939 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs),
940 NULL, &pflags, 8);
941 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs),
942 NULL, &atime, 16);
943 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs),
944 NULL, &mtime, 16);
945 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs),
946 NULL, &ctime, 16);
947 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs),
948 NULL, &crtime, 16);
949 }
950
951 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_LINKS(zfsvfs), NULL, &links, 8);
952
953 if (obj_type == DMU_OT_ZNODE) {
954 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_XATTR(zfsvfs), NULL,
955 &empty_xattr, 8);
956 }
957 if (obj_type == DMU_OT_ZNODE ||
958 (vap->va_type == VBLK || vap->va_type == VCHR)) {
959 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_RDEV(zfsvfs),
960 NULL, &rdev, 8);
961
962 }
963 if (obj_type == DMU_OT_ZNODE) {
964 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs),
965 NULL, &pflags, 8);
966 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs), NULL,
967 &acl_ids->z_fuid, 8);
968 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs), NULL,
969 &acl_ids->z_fgid, 8);
970 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PAD(zfsvfs), NULL, pad,
971 sizeof (uint64_t) * 4);
972 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ZNODE_ACL(zfsvfs), NULL,
973 &acl_phys, sizeof (zfs_acl_phys_t));
974 } else if (acl_ids->z_aclp->z_version >= ZFS_ACL_VERSION_FUID) {
975 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_COUNT(zfsvfs), NULL,
976 &acl_ids->z_aclp->z_acl_count, 8);
977 locate.cb_aclp = acl_ids->z_aclp;
978 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_ACES(zfsvfs),
979 zfs_acl_data_locator, &locate,
980 acl_ids->z_aclp->z_acl_bytes);
981 mode = zfs_mode_compute(mode, acl_ids->z_aclp, &pflags,
982 acl_ids->z_fuid, acl_ids->z_fgid);
983 }
984
985 VERIFY(sa_replace_all_by_template(sa_hdl, sa_attrs, cnt, tx) == 0);
986
987 if (!(flag & IS_ROOT_NODE)) {
988 *zpp = zfs_znode_alloc(zfsvfs, db, 0, obj_type, sa_hdl);
989 ASSERT(*zpp != NULL);
990 } else {
991 /*
992 * If we are creating the root node, the "parent" we
993 * passed in is the znode for the root.
994 */
995 *zpp = dzp;
996
997 (*zpp)->z_sa_hdl = sa_hdl;
998 }
999
1000 (*zpp)->z_pflags = pflags;
1001 (*zpp)->z_mode = mode;
1002 (*zpp)->z_dnodesize = dnodesize;
1003
1004 if (vap->va_mask & AT_XVATTR)
1005 zfs_xvattr_set(*zpp, (xvattr_t *)vap, tx);
1006
1007 if (obj_type == DMU_OT_ZNODE ||
1008 acl_ids->z_aclp->z_version < ZFS_ACL_VERSION_FUID) {
1009 VERIFY0(zfs_aclset_common(*zpp, acl_ids->z_aclp, cr, tx));
1010 }
1011 if (!(flag & IS_ROOT_NODE)) {
1012 vnode_t *vp;
1013
1014 vp = ZTOV(*zpp);
1015 vp->v_vflag |= VV_FORCEINSMQ;
1016 err = insmntque(vp, zfsvfs->z_vfs);
1017 vp->v_vflag &= ~VV_FORCEINSMQ;
1018 KASSERT(err == 0, ("insmntque() failed: error %d", err));
1019 }
1020 kmem_free(sa_attrs, sizeof (sa_bulk_attr_t) * ZPL_END);
1021 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj);
1022 }
1023
1024 /*
1025 * Update in-core attributes. It is assumed the caller will be doing an
1026 * sa_bulk_update to push the changes out.
1027 */
1028 void
zfs_xvattr_set(znode_t * zp,xvattr_t * xvap,dmu_tx_t * tx)1029 zfs_xvattr_set(znode_t *zp, xvattr_t *xvap, dmu_tx_t *tx)
1030 {
1031 xoptattr_t *xoap;
1032
1033 xoap = xva_getxoptattr(xvap);
1034 ASSERT(xoap);
1035
1036 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
1037 uint64_t times[2];
1038 ZFS_TIME_ENCODE(&xoap->xoa_createtime, times);
1039 (void) sa_update(zp->z_sa_hdl, SA_ZPL_CRTIME(zp->z_zfsvfs),
1040 ×, sizeof (times), tx);
1041 XVA_SET_RTN(xvap, XAT_CREATETIME);
1042 }
1043 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
1044 ZFS_ATTR_SET(zp, ZFS_READONLY, xoap->xoa_readonly,
1045 zp->z_pflags, tx);
1046 XVA_SET_RTN(xvap, XAT_READONLY);
1047 }
1048 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
1049 ZFS_ATTR_SET(zp, ZFS_HIDDEN, xoap->xoa_hidden,
1050 zp->z_pflags, tx);
1051 XVA_SET_RTN(xvap, XAT_HIDDEN);
1052 }
1053 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
1054 ZFS_ATTR_SET(zp, ZFS_SYSTEM, xoap->xoa_system,
1055 zp->z_pflags, tx);
1056 XVA_SET_RTN(xvap, XAT_SYSTEM);
1057 }
1058 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
1059 ZFS_ATTR_SET(zp, ZFS_ARCHIVE, xoap->xoa_archive,
1060 zp->z_pflags, tx);
1061 XVA_SET_RTN(xvap, XAT_ARCHIVE);
1062 }
1063 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
1064 ZFS_ATTR_SET(zp, ZFS_IMMUTABLE, xoap->xoa_immutable,
1065 zp->z_pflags, tx);
1066 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
1067 }
1068 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
1069 ZFS_ATTR_SET(zp, ZFS_NOUNLINK, xoap->xoa_nounlink,
1070 zp->z_pflags, tx);
1071 XVA_SET_RTN(xvap, XAT_NOUNLINK);
1072 }
1073 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
1074 ZFS_ATTR_SET(zp, ZFS_APPENDONLY, xoap->xoa_appendonly,
1075 zp->z_pflags, tx);
1076 XVA_SET_RTN(xvap, XAT_APPENDONLY);
1077 }
1078 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
1079 ZFS_ATTR_SET(zp, ZFS_NODUMP, xoap->xoa_nodump,
1080 zp->z_pflags, tx);
1081 XVA_SET_RTN(xvap, XAT_NODUMP);
1082 }
1083 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
1084 ZFS_ATTR_SET(zp, ZFS_OPAQUE, xoap->xoa_opaque,
1085 zp->z_pflags, tx);
1086 XVA_SET_RTN(xvap, XAT_OPAQUE);
1087 }
1088 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
1089 ZFS_ATTR_SET(zp, ZFS_AV_QUARANTINED,
1090 xoap->xoa_av_quarantined, zp->z_pflags, tx);
1091 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
1092 }
1093 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
1094 ZFS_ATTR_SET(zp, ZFS_AV_MODIFIED, xoap->xoa_av_modified,
1095 zp->z_pflags, tx);
1096 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
1097 }
1098 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) {
1099 zfs_sa_set_scanstamp(zp, xvap, tx);
1100 XVA_SET_RTN(xvap, XAT_AV_SCANSTAMP);
1101 }
1102 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
1103 ZFS_ATTR_SET(zp, ZFS_REPARSE, xoap->xoa_reparse,
1104 zp->z_pflags, tx);
1105 XVA_SET_RTN(xvap, XAT_REPARSE);
1106 }
1107 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
1108 ZFS_ATTR_SET(zp, ZFS_OFFLINE, xoap->xoa_offline,
1109 zp->z_pflags, tx);
1110 XVA_SET_RTN(xvap, XAT_OFFLINE);
1111 }
1112 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
1113 ZFS_ATTR_SET(zp, ZFS_SPARSE, xoap->xoa_sparse,
1114 zp->z_pflags, tx);
1115 XVA_SET_RTN(xvap, XAT_SPARSE);
1116 }
1117 }
1118
1119 int
zfs_zget(zfsvfs_t * zfsvfs,uint64_t obj_num,znode_t ** zpp)1120 zfs_zget(zfsvfs_t *zfsvfs, uint64_t obj_num, znode_t **zpp)
1121 {
1122 dmu_object_info_t doi;
1123 dmu_buf_t *db;
1124 znode_t *zp;
1125 vnode_t *vp;
1126 sa_handle_t *hdl;
1127 struct thread *td;
1128 int locked;
1129 int err;
1130
1131 td = curthread;
1132 getnewvnode_reserve(1);
1133 again:
1134 *zpp = NULL;
1135 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num);
1136
1137 err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db);
1138 if (err) {
1139 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1140 getnewvnode_drop_reserve();
1141 return (err);
1142 }
1143
1144 dmu_object_info_from_db(db, &doi);
1145 if (doi.doi_bonus_type != DMU_OT_SA &&
1146 (doi.doi_bonus_type != DMU_OT_ZNODE ||
1147 (doi.doi_bonus_type == DMU_OT_ZNODE &&
1148 doi.doi_bonus_size < sizeof (znode_phys_t)))) {
1149 sa_buf_rele(db, NULL);
1150 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1151 #ifdef __FreeBSD__
1152 getnewvnode_drop_reserve();
1153 #endif
1154 return (SET_ERROR(EINVAL));
1155 }
1156
1157 hdl = dmu_buf_get_user(db);
1158 if (hdl != NULL) {
1159 zp = sa_get_userdata(hdl);
1160
1161 /*
1162 * Since "SA" does immediate eviction we
1163 * should never find a sa handle that doesn't
1164 * know about the znode.
1165 */
1166 ASSERT3P(zp, !=, NULL);
1167 ASSERT3U(zp->z_id, ==, obj_num);
1168 if (zp->z_unlinked) {
1169 err = SET_ERROR(ENOENT);
1170 } else {
1171 vp = ZTOV(zp);
1172 /*
1173 * Don't let the vnode disappear after
1174 * ZFS_OBJ_HOLD_EXIT.
1175 */
1176 VN_HOLD(vp);
1177 *zpp = zp;
1178 err = 0;
1179 }
1180
1181 sa_buf_rele(db, NULL);
1182 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1183
1184 if (err) {
1185 getnewvnode_drop_reserve();
1186 return (err);
1187 }
1188
1189 locked = VOP_ISLOCKED(vp);
1190 VI_LOCK(vp);
1191 if ((vp->v_iflag & VI_DOOMED) != 0 &&
1192 locked != LK_EXCLUSIVE) {
1193 /*
1194 * The vnode is doomed and this thread doesn't
1195 * hold the exclusive lock on it, so the vnode
1196 * must be being reclaimed by another thread.
1197 * Otherwise the doomed vnode is being reclaimed
1198 * by this thread and zfs_zget is called from
1199 * ZIL internals.
1200 */
1201 VI_UNLOCK(vp);
1202
1203 /*
1204 * XXX vrele() locks the vnode when the last reference
1205 * is dropped. Although in this case the vnode is
1206 * doomed / dead and so no inactivation is required,
1207 * the vnode lock is still acquired. That could result
1208 * in a LOR with z_teardown_lock if another thread holds
1209 * the vnode's lock and tries to take z_teardown_lock.
1210 * But that is only possible if the other thread peforms
1211 * a ZFS vnode operation on the vnode. That either
1212 * should not happen if the vnode is dead or the thread
1213 * should also have a refrence to the vnode and thus
1214 * our reference is not last.
1215 */
1216 VN_RELE(vp);
1217 goto again;
1218 }
1219 VI_UNLOCK(vp);
1220 getnewvnode_drop_reserve();
1221 return (err);
1222 }
1223
1224 /*
1225 * Not found create new znode/vnode
1226 * but only if file exists.
1227 *
1228 * There is a small window where zfs_vget() could
1229 * find this object while a file create is still in
1230 * progress. This is checked for in zfs_znode_alloc()
1231 *
1232 * if zfs_znode_alloc() fails it will drop the hold on the
1233 * bonus buffer.
1234 */
1235 zp = zfs_znode_alloc(zfsvfs, db, doi.doi_data_block_size,
1236 doi.doi_bonus_type, NULL);
1237 if (zp == NULL) {
1238 err = SET_ERROR(ENOENT);
1239 } else {
1240 *zpp = zp;
1241 }
1242 if (err == 0) {
1243 vnode_t *vp = ZTOV(zp);
1244
1245 err = insmntque(vp, zfsvfs->z_vfs);
1246 if (err == 0) {
1247 vp->v_hash = obj_num;
1248 VOP_UNLOCK(vp, 0);
1249 } else {
1250 zp->z_vnode = NULL;
1251 zfs_znode_dmu_fini(zp);
1252 zfs_znode_free(zp);
1253 *zpp = NULL;
1254 }
1255 }
1256 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1257 getnewvnode_drop_reserve();
1258 return (err);
1259 }
1260
1261 int
zfs_rezget(znode_t * zp)1262 zfs_rezget(znode_t *zp)
1263 {
1264 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1265 dmu_object_info_t doi;
1266 dmu_buf_t *db;
1267 vnode_t *vp;
1268 uint64_t obj_num = zp->z_id;
1269 uint64_t mode, size;
1270 sa_bulk_attr_t bulk[8];
1271 int err;
1272 int count = 0;
1273 uint64_t gen;
1274
1275 /*
1276 * Remove cached pages before reloading the znode, so that they are not
1277 * lingering after we run into any error. Ideally, we should vgone()
1278 * the vnode in case of error, but currently we cannot do that
1279 * because of the LOR between the vnode lock and z_teardown_lock.
1280 * So, instead, we have to "doom" the znode in the illumos style.
1281 */
1282 vp = ZTOV(zp);
1283 vn_pages_remove(vp, 0, 0);
1284
1285 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num);
1286
1287 mutex_enter(&zp->z_acl_lock);
1288 if (zp->z_acl_cached) {
1289 zfs_acl_free(zp->z_acl_cached);
1290 zp->z_acl_cached = NULL;
1291 }
1292
1293 mutex_exit(&zp->z_acl_lock);
1294 ASSERT(zp->z_sa_hdl == NULL);
1295 err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db);
1296 if (err) {
1297 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1298 return (err);
1299 }
1300
1301 dmu_object_info_from_db(db, &doi);
1302 if (doi.doi_bonus_type != DMU_OT_SA &&
1303 (doi.doi_bonus_type != DMU_OT_ZNODE ||
1304 (doi.doi_bonus_type == DMU_OT_ZNODE &&
1305 doi.doi_bonus_size < sizeof (znode_phys_t)))) {
1306 sa_buf_rele(db, NULL);
1307 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1308 return (SET_ERROR(EINVAL));
1309 }
1310
1311 zfs_znode_sa_init(zfsvfs, zp, db, doi.doi_bonus_type, NULL);
1312 size = zp->z_size;
1313
1314 /* reload cached values */
1315 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL,
1316 &gen, sizeof (gen));
1317 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
1318 &zp->z_size, sizeof (zp->z_size));
1319 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL,
1320 &zp->z_links, sizeof (zp->z_links));
1321 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
1322 &zp->z_pflags, sizeof (zp->z_pflags));
1323 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
1324 &zp->z_atime, sizeof (zp->z_atime));
1325 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
1326 &zp->z_uid, sizeof (zp->z_uid));
1327 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL,
1328 &zp->z_gid, sizeof (zp->z_gid));
1329 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
1330 &mode, sizeof (mode));
1331
1332 if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) {
1333 zfs_znode_dmu_fini(zp);
1334 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1335 return (SET_ERROR(EIO));
1336 }
1337
1338 zp->z_mode = mode;
1339
1340 if (gen != zp->z_gen) {
1341 zfs_znode_dmu_fini(zp);
1342 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1343 return (SET_ERROR(EIO));
1344 }
1345
1346 /*
1347 * It is highly improbable but still quite possible that two
1348 * objects in different datasets are created with the same
1349 * object numbers and in transaction groups with the same
1350 * numbers. znodes corresponding to those objects would
1351 * have the same z_id and z_gen, but their other attributes
1352 * may be different.
1353 * zfs recv -F may replace one of such objects with the other.
1354 * As a result file properties recorded in the replaced
1355 * object's vnode may no longer match the received object's
1356 * properties. At present the only cached property is the
1357 * files type recorded in v_type.
1358 * So, handle this case by leaving the old vnode and znode
1359 * disassociated from the actual object. A new vnode and a
1360 * znode will be created if the object is accessed
1361 * (e.g. via a look-up). The old vnode and znode will be
1362 * recycled when the last vnode reference is dropped.
1363 */
1364 if (vp->v_type != IFTOVT((mode_t)zp->z_mode)) {
1365 zfs_znode_dmu_fini(zp);
1366 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1367 return (SET_ERROR(EIO));
1368 }
1369
1370 /*
1371 * If the file has zero links, then it has been unlinked on the send
1372 * side and it must be in the received unlinked set.
1373 * We call zfs_znode_dmu_fini() now to prevent any accesses to the
1374 * stale data and to prevent automatical removal of the file in
1375 * zfs_zinactive(). The file will be removed either when it is removed
1376 * on the send side and the next incremental stream is received or
1377 * when the unlinked set gets processed.
1378 */
1379 zp->z_unlinked = (zp->z_links == 0);
1380 if (zp->z_unlinked) {
1381 zfs_znode_dmu_fini(zp);
1382 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1383 return (0);
1384 }
1385
1386 zp->z_blksz = doi.doi_data_block_size;
1387 if (zp->z_size != size)
1388 vnode_pager_setsize(vp, zp->z_size);
1389
1390 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1391
1392 return (0);
1393 }
1394
1395 void
zfs_znode_delete(znode_t * zp,dmu_tx_t * tx)1396 zfs_znode_delete(znode_t *zp, dmu_tx_t *tx)
1397 {
1398 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1399 objset_t *os = zfsvfs->z_os;
1400 uint64_t obj = zp->z_id;
1401 uint64_t acl_obj = zfs_external_acl(zp);
1402
1403 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj);
1404 if (acl_obj) {
1405 VERIFY(!zp->z_is_sa);
1406 VERIFY(0 == dmu_object_free(os, acl_obj, tx));
1407 }
1408 VERIFY(0 == dmu_object_free(os, obj, tx));
1409 zfs_znode_dmu_fini(zp);
1410 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj);
1411 zfs_znode_free(zp);
1412 }
1413
1414 void
zfs_zinactive(znode_t * zp)1415 zfs_zinactive(znode_t *zp)
1416 {
1417 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1418 uint64_t z_id = zp->z_id;
1419
1420 ASSERT(zp->z_sa_hdl);
1421
1422 /*
1423 * Don't allow a zfs_zget() while were trying to release this znode
1424 */
1425 ZFS_OBJ_HOLD_ENTER(zfsvfs, z_id);
1426
1427 /*
1428 * If this was the last reference to a file with no links, remove
1429 * the file from the file system unless the file system is mounted
1430 * read-only. That can happen, for example, if the file system was
1431 * originally read-write, the file was opened, then unlinked and
1432 * the file system was made read-only before the file was finally
1433 * closed. The file will remain in the unlinked set.
1434 */
1435 if (zp->z_unlinked) {
1436 ASSERT(!zfsvfs->z_issnap);
1437 if ((zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) == 0) {
1438 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id);
1439 zfs_rmnode(zp);
1440 return;
1441 }
1442 }
1443
1444 zfs_znode_dmu_fini(zp);
1445 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id);
1446 zfs_znode_free(zp);
1447 }
1448
1449 void
zfs_znode_free(znode_t * zp)1450 zfs_znode_free(znode_t *zp)
1451 {
1452 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1453
1454 ASSERT(zp->z_sa_hdl == NULL);
1455 zp->z_vnode = NULL;
1456 mutex_enter(&zfsvfs->z_znodes_lock);
1457 POINTER_INVALIDATE(&zp->z_zfsvfs);
1458 list_remove(&zfsvfs->z_all_znodes, zp);
1459 mutex_exit(&zfsvfs->z_znodes_lock);
1460
1461 if (zp->z_acl_cached) {
1462 zfs_acl_free(zp->z_acl_cached);
1463 zp->z_acl_cached = NULL;
1464 }
1465
1466 kmem_cache_free(znode_cache, zp);
1467
1468 #ifdef illumos
1469 VFS_RELE(zfsvfs->z_vfs);
1470 #endif
1471 }
1472
1473 void
zfs_tstamp_update_setup(znode_t * zp,uint_t flag,uint64_t mtime[2],uint64_t ctime[2],boolean_t have_tx)1474 zfs_tstamp_update_setup(znode_t *zp, uint_t flag, uint64_t mtime[2],
1475 uint64_t ctime[2], boolean_t have_tx)
1476 {
1477 timestruc_t now;
1478
1479 vfs_timestamp(&now);
1480
1481 if (have_tx) { /* will sa_bulk_update happen really soon? */
1482 zp->z_atime_dirty = 0;
1483 zp->z_seq++;
1484 } else {
1485 zp->z_atime_dirty = 1;
1486 }
1487
1488 if (flag & AT_ATIME) {
1489 ZFS_TIME_ENCODE(&now, zp->z_atime);
1490 }
1491
1492 if (flag & AT_MTIME) {
1493 ZFS_TIME_ENCODE(&now, mtime);
1494 if (zp->z_zfsvfs->z_use_fuids) {
1495 zp->z_pflags |= (ZFS_ARCHIVE |
1496 ZFS_AV_MODIFIED);
1497 }
1498 }
1499
1500 if (flag & AT_CTIME) {
1501 ZFS_TIME_ENCODE(&now, ctime);
1502 if (zp->z_zfsvfs->z_use_fuids)
1503 zp->z_pflags |= ZFS_ARCHIVE;
1504 }
1505 }
1506
1507 /*
1508 * Grow the block size for a file.
1509 *
1510 * IN: zp - znode of file to free data in.
1511 * size - requested block size
1512 * tx - open transaction.
1513 *
1514 * NOTE: this function assumes that the znode is write locked.
1515 */
1516 void
zfs_grow_blocksize(znode_t * zp,uint64_t size,dmu_tx_t * tx)1517 zfs_grow_blocksize(znode_t *zp, uint64_t size, dmu_tx_t *tx)
1518 {
1519 int error;
1520 u_longlong_t dummy;
1521
1522 if (size <= zp->z_blksz)
1523 return;
1524 /*
1525 * If the file size is already greater than the current blocksize,
1526 * we will not grow. If there is more than one block in a file,
1527 * the blocksize cannot change.
1528 */
1529 if (zp->z_blksz && zp->z_size > zp->z_blksz)
1530 return;
1531
1532 error = dmu_object_set_blocksize(zp->z_zfsvfs->z_os, zp->z_id,
1533 size, 0, tx);
1534
1535 if (error == ENOTSUP)
1536 return;
1537 ASSERT0(error);
1538
1539 /* What blocksize did we actually get? */
1540 dmu_object_size_from_db(sa_get_db(zp->z_sa_hdl), &zp->z_blksz, &dummy);
1541 }
1542
1543 #ifdef illumos
1544 /*
1545 * This is a dummy interface used when pvn_vplist_dirty() should *not*
1546 * be calling back into the fs for a putpage(). E.g.: when truncating
1547 * a file, the pages being "thrown away* don't need to be written out.
1548 */
1549 /* ARGSUSED */
1550 static int
zfs_no_putpage(vnode_t * vp,page_t * pp,u_offset_t * offp,size_t * lenp,int flags,cred_t * cr)1551 zfs_no_putpage(vnode_t *vp, page_t *pp, u_offset_t *offp, size_t *lenp,
1552 int flags, cred_t *cr)
1553 {
1554 ASSERT(0);
1555 return (0);
1556 }
1557 #endif
1558
1559 /*
1560 * Increase the file length
1561 *
1562 * IN: zp - znode of file to free data in.
1563 * end - new end-of-file
1564 *
1565 * RETURN: 0 on success, error code on failure
1566 */
1567 static int
zfs_extend(znode_t * zp,uint64_t end)1568 zfs_extend(znode_t *zp, uint64_t end)
1569 {
1570 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1571 dmu_tx_t *tx;
1572 rl_t *rl;
1573 uint64_t newblksz;
1574 int error;
1575
1576 /*
1577 * We will change zp_size, lock the whole file.
1578 */
1579 rl = zfs_range_lock(zp, 0, UINT64_MAX, RL_WRITER);
1580
1581 /*
1582 * Nothing to do if file already at desired length.
1583 */
1584 if (end <= zp->z_size) {
1585 zfs_range_unlock(rl);
1586 return (0);
1587 }
1588 tx = dmu_tx_create(zfsvfs->z_os);
1589 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1590 zfs_sa_upgrade_txholds(tx, zp);
1591 if (end > zp->z_blksz &&
1592 (!ISP2(zp->z_blksz) || zp->z_blksz < zfsvfs->z_max_blksz)) {
1593 /*
1594 * We are growing the file past the current block size.
1595 */
1596 if (zp->z_blksz > zp->z_zfsvfs->z_max_blksz) {
1597 /*
1598 * File's blocksize is already larger than the
1599 * "recordsize" property. Only let it grow to
1600 * the next power of 2.
1601 */
1602 ASSERT(!ISP2(zp->z_blksz));
1603 newblksz = MIN(end, 1 << highbit64(zp->z_blksz));
1604 } else {
1605 newblksz = MIN(end, zp->z_zfsvfs->z_max_blksz);
1606 }
1607 dmu_tx_hold_write(tx, zp->z_id, 0, newblksz);
1608 } else {
1609 newblksz = 0;
1610 }
1611
1612 error = dmu_tx_assign(tx, TXG_WAIT);
1613 if (error) {
1614 dmu_tx_abort(tx);
1615 zfs_range_unlock(rl);
1616 return (error);
1617 }
1618
1619 if (newblksz)
1620 zfs_grow_blocksize(zp, newblksz, tx);
1621
1622 zp->z_size = end;
1623
1624 VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zp->z_zfsvfs),
1625 &zp->z_size, sizeof (zp->z_size), tx));
1626
1627 vnode_pager_setsize(ZTOV(zp), end);
1628
1629 zfs_range_unlock(rl);
1630
1631 dmu_tx_commit(tx);
1632
1633 return (0);
1634 }
1635
1636 /*
1637 * Free space in a file.
1638 *
1639 * IN: zp - znode of file to free data in.
1640 * off - start of section to free.
1641 * len - length of section to free.
1642 *
1643 * RETURN: 0 on success, error code on failure
1644 */
1645 static int
zfs_free_range(znode_t * zp,uint64_t off,uint64_t len)1646 zfs_free_range(znode_t *zp, uint64_t off, uint64_t len)
1647 {
1648 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1649 rl_t *rl;
1650 int error;
1651
1652 /*
1653 * Lock the range being freed.
1654 */
1655 rl = zfs_range_lock(zp, off, len, RL_WRITER);
1656
1657 /*
1658 * Nothing to do if file already at desired length.
1659 */
1660 if (off >= zp->z_size) {
1661 zfs_range_unlock(rl);
1662 return (0);
1663 }
1664
1665 if (off + len > zp->z_size)
1666 len = zp->z_size - off;
1667
1668 error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, off, len);
1669
1670 if (error == 0) {
1671 /*
1672 * In FreeBSD we cannot free block in the middle of a file,
1673 * but only at the end of a file, so this code path should
1674 * never happen.
1675 */
1676 vnode_pager_setsize(ZTOV(zp), off);
1677 }
1678
1679 zfs_range_unlock(rl);
1680
1681 return (error);
1682 }
1683
1684 /*
1685 * Truncate a file
1686 *
1687 * IN: zp - znode of file to free data in.
1688 * end - new end-of-file.
1689 *
1690 * RETURN: 0 on success, error code on failure
1691 */
1692 static int
zfs_trunc(znode_t * zp,uint64_t end)1693 zfs_trunc(znode_t *zp, uint64_t end)
1694 {
1695 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1696 vnode_t *vp = ZTOV(zp);
1697 dmu_tx_t *tx;
1698 rl_t *rl;
1699 int error;
1700 sa_bulk_attr_t bulk[2];
1701 int count = 0;
1702
1703 /*
1704 * We will change zp_size, lock the whole file.
1705 */
1706 rl = zfs_range_lock(zp, 0, UINT64_MAX, RL_WRITER);
1707
1708 /*
1709 * Nothing to do if file already at desired length.
1710 */
1711 if (end >= zp->z_size) {
1712 zfs_range_unlock(rl);
1713 return (0);
1714 }
1715
1716 error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, end,
1717 DMU_OBJECT_END);
1718 if (error) {
1719 zfs_range_unlock(rl);
1720 return (error);
1721 }
1722 tx = dmu_tx_create(zfsvfs->z_os);
1723 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1724 zfs_sa_upgrade_txholds(tx, zp);
1725 dmu_tx_mark_netfree(tx);
1726 error = dmu_tx_assign(tx, TXG_WAIT);
1727 if (error) {
1728 dmu_tx_abort(tx);
1729 zfs_range_unlock(rl);
1730 return (error);
1731 }
1732
1733 zp->z_size = end;
1734 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs),
1735 NULL, &zp->z_size, sizeof (zp->z_size));
1736
1737 if (end == 0) {
1738 zp->z_pflags &= ~ZFS_SPARSE;
1739 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
1740 NULL, &zp->z_pflags, 8);
1741 }
1742 VERIFY(sa_bulk_update(zp->z_sa_hdl, bulk, count, tx) == 0);
1743
1744 dmu_tx_commit(tx);
1745
1746 /*
1747 * Clear any mapped pages in the truncated region. This has to
1748 * happen outside of the transaction to avoid the possibility of
1749 * a deadlock with someone trying to push a page that we are
1750 * about to invalidate.
1751 */
1752 vnode_pager_setsize(vp, end);
1753
1754 zfs_range_unlock(rl);
1755
1756 return (0);
1757 }
1758
1759 /*
1760 * Free space in a file
1761 *
1762 * IN: zp - znode of file to free data in.
1763 * off - start of range
1764 * len - end of range (0 => EOF)
1765 * flag - current file open mode flags.
1766 * log - TRUE if this action should be logged
1767 *
1768 * RETURN: 0 on success, error code on failure
1769 */
1770 int
zfs_freesp(znode_t * zp,uint64_t off,uint64_t len,int flag,boolean_t log)1771 zfs_freesp(znode_t *zp, uint64_t off, uint64_t len, int flag, boolean_t log)
1772 {
1773 vnode_t *vp = ZTOV(zp);
1774 dmu_tx_t *tx;
1775 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1776 zilog_t *zilog = zfsvfs->z_log;
1777 uint64_t mode;
1778 uint64_t mtime[2], ctime[2];
1779 sa_bulk_attr_t bulk[3];
1780 int count = 0;
1781 int error;
1782
1783 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs), &mode,
1784 sizeof (mode))) != 0)
1785 return (error);
1786
1787 if (off > zp->z_size) {
1788 error = zfs_extend(zp, off+len);
1789 if (error == 0 && log)
1790 goto log;
1791 else
1792 return (error);
1793 }
1794
1795 /*
1796 * Check for any locks in the region to be freed.
1797 */
1798
1799 if (MANDLOCK(vp, (mode_t)mode)) {
1800 uint64_t length = (len ? len : zp->z_size - off);
1801 if (error = chklock(vp, FWRITE, off, length, flag, NULL))
1802 return (error);
1803 }
1804
1805 if (len == 0) {
1806 error = zfs_trunc(zp, off);
1807 } else {
1808 if ((error = zfs_free_range(zp, off, len)) == 0 &&
1809 off + len > zp->z_size)
1810 error = zfs_extend(zp, off+len);
1811 }
1812 if (error || !log)
1813 return (error);
1814 log:
1815 tx = dmu_tx_create(zfsvfs->z_os);
1816 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1817 zfs_sa_upgrade_txholds(tx, zp);
1818 error = dmu_tx_assign(tx, TXG_WAIT);
1819 if (error) {
1820 dmu_tx_abort(tx);
1821 return (error);
1822 }
1823
1824 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, mtime, 16);
1825 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, ctime, 16);
1826 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
1827 NULL, &zp->z_pflags, 8);
1828 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, B_TRUE);
1829 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
1830 ASSERT(error == 0);
1831
1832 zfs_log_truncate(zilog, tx, TX_TRUNCATE, zp, off, len);
1833
1834 dmu_tx_commit(tx);
1835 return (0);
1836 }
1837
1838 void
zfs_create_fs(objset_t * os,cred_t * cr,nvlist_t * zplprops,dmu_tx_t * tx)1839 zfs_create_fs(objset_t *os, cred_t *cr, nvlist_t *zplprops, dmu_tx_t *tx)
1840 {
1841 uint64_t moid, obj, sa_obj, version;
1842 uint64_t sense = ZFS_CASE_SENSITIVE;
1843 uint64_t norm = 0;
1844 nvpair_t *elem;
1845 int error;
1846 int i;
1847 znode_t *rootzp = NULL;
1848 zfsvfs_t *zfsvfs;
1849 vattr_t vattr;
1850 znode_t *zp;
1851 zfs_acl_ids_t acl_ids;
1852
1853 /*
1854 * First attempt to create master node.
1855 */
1856 /*
1857 * In an empty objset, there are no blocks to read and thus
1858 * there can be no i/o errors (which we assert below).
1859 */
1860 moid = MASTER_NODE_OBJ;
1861 error = zap_create_claim(os, moid, DMU_OT_MASTER_NODE,
1862 DMU_OT_NONE, 0, tx);
1863 ASSERT(error == 0);
1864
1865 /*
1866 * Give dmu_object_alloc() a hint about where to start
1867 * allocating new objects. Otherwise, since the metadnode's
1868 * dnode_phys_t structure isn't initialized yet, dmu_object_next()
1869 * would fail and we'd have to skip to the next dnode block.
1870 */
1871 os->os_obj_next = moid + 1;
1872
1873 /*
1874 * Set starting attributes.
1875 */
1876 version = zfs_zpl_version_map(spa_version(dmu_objset_spa(os)));
1877 elem = NULL;
1878 while ((elem = nvlist_next_nvpair(zplprops, elem)) != NULL) {
1879 /* For the moment we expect all zpl props to be uint64_ts */
1880 uint64_t val;
1881 char *name;
1882
1883 ASSERT(nvpair_type(elem) == DATA_TYPE_UINT64);
1884 VERIFY(nvpair_value_uint64(elem, &val) == 0);
1885 name = nvpair_name(elem);
1886 if (strcmp(name, zfs_prop_to_name(ZFS_PROP_VERSION)) == 0) {
1887 if (val < version)
1888 version = val;
1889 } else {
1890 error = zap_update(os, moid, name, 8, 1, &val, tx);
1891 }
1892 ASSERT(error == 0);
1893 if (strcmp(name, zfs_prop_to_name(ZFS_PROP_NORMALIZE)) == 0)
1894 norm = val;
1895 else if (strcmp(name, zfs_prop_to_name(ZFS_PROP_CASE)) == 0)
1896 sense = val;
1897 }
1898 ASSERT(version != 0);
1899 error = zap_update(os, moid, ZPL_VERSION_STR, 8, 1, &version, tx);
1900
1901 /*
1902 * Create zap object used for SA attribute registration
1903 */
1904
1905 if (version >= ZPL_VERSION_SA) {
1906 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
1907 DMU_OT_NONE, 0, tx);
1908 error = zap_add(os, moid, ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
1909 ASSERT(error == 0);
1910 } else {
1911 sa_obj = 0;
1912 }
1913 /*
1914 * Create a delete queue.
1915 */
1916 obj = zap_create(os, DMU_OT_UNLINKED_SET, DMU_OT_NONE, 0, tx);
1917
1918 error = zap_add(os, moid, ZFS_UNLINKED_SET, 8, 1, &obj, tx);
1919 ASSERT(error == 0);
1920
1921 /*
1922 * Create root znode. Create minimal znode/vnode/zfsvfs
1923 * to allow zfs_mknode to work.
1924 */
1925 VATTR_NULL(&vattr);
1926 vattr.va_mask = AT_MODE|AT_UID|AT_GID|AT_TYPE;
1927 vattr.va_type = VDIR;
1928 vattr.va_mode = S_IFDIR|0755;
1929 vattr.va_uid = crgetuid(cr);
1930 vattr.va_gid = crgetgid(cr);
1931
1932 zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
1933
1934 rootzp = kmem_cache_alloc(znode_cache, KM_SLEEP);
1935 ASSERT(!POINTER_IS_VALID(rootzp->z_zfsvfs));
1936 rootzp->z_moved = 0;
1937 rootzp->z_unlinked = 0;
1938 rootzp->z_atime_dirty = 0;
1939 rootzp->z_is_sa = USE_SA(version, os);
1940
1941 zfsvfs->z_os = os;
1942 zfsvfs->z_parent = zfsvfs;
1943 zfsvfs->z_version = version;
1944 zfsvfs->z_use_fuids = USE_FUIDS(version, os);
1945 zfsvfs->z_use_sa = USE_SA(version, os);
1946 zfsvfs->z_norm = norm;
1947
1948 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
1949 &zfsvfs->z_attr_table);
1950
1951 ASSERT(error == 0);
1952
1953 /*
1954 * Fold case on file systems that are always or sometimes case
1955 * insensitive.
1956 */
1957 if (sense == ZFS_CASE_INSENSITIVE || sense == ZFS_CASE_MIXED)
1958 zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER;
1959
1960 mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
1961 list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
1962 offsetof(znode_t, z_link_node));
1963
1964 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1965 mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL);
1966
1967 rootzp->z_zfsvfs = zfsvfs;
1968 VERIFY(0 == zfs_acl_ids_create(rootzp, IS_ROOT_NODE, &vattr,
1969 cr, NULL, &acl_ids));
1970 zfs_mknode(rootzp, &vattr, tx, cr, IS_ROOT_NODE, &zp, &acl_ids);
1971 ASSERT3P(zp, ==, rootzp);
1972 error = zap_add(os, moid, ZFS_ROOT_OBJ, 8, 1, &rootzp->z_id, tx);
1973 ASSERT(error == 0);
1974 zfs_acl_ids_free(&acl_ids);
1975 POINTER_INVALIDATE(&rootzp->z_zfsvfs);
1976
1977 sa_handle_destroy(rootzp->z_sa_hdl);
1978 kmem_cache_free(znode_cache, rootzp);
1979
1980 /*
1981 * Create shares directory
1982 */
1983
1984 error = zfs_create_share_dir(zfsvfs, tx);
1985
1986 ASSERT(error == 0);
1987
1988 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1989 mutex_destroy(&zfsvfs->z_hold_mtx[i]);
1990 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1991 }
1992 #endif /* _KERNEL */
1993
1994 static int
zfs_sa_setup(objset_t * osp,sa_attr_type_t ** sa_table)1995 zfs_sa_setup(objset_t *osp, sa_attr_type_t **sa_table)
1996 {
1997 uint64_t sa_obj = 0;
1998 int error;
1999
2000 error = zap_lookup(osp, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1, &sa_obj);
2001 if (error != 0 && error != ENOENT)
2002 return (error);
2003
2004 error = sa_setup(osp, sa_obj, zfs_attr_table, ZPL_END, sa_table);
2005 return (error);
2006 }
2007
2008 static int
zfs_grab_sa_handle(objset_t * osp,uint64_t obj,sa_handle_t ** hdlp,dmu_buf_t ** db,void * tag)2009 zfs_grab_sa_handle(objset_t *osp, uint64_t obj, sa_handle_t **hdlp,
2010 dmu_buf_t **db, void *tag)
2011 {
2012 dmu_object_info_t doi;
2013 int error;
2014
2015 if ((error = sa_buf_hold(osp, obj, tag, db)) != 0)
2016 return (error);
2017
2018 dmu_object_info_from_db(*db, &doi);
2019 if ((doi.doi_bonus_type != DMU_OT_SA &&
2020 doi.doi_bonus_type != DMU_OT_ZNODE) ||
2021 doi.doi_bonus_type == DMU_OT_ZNODE &&
2022 doi.doi_bonus_size < sizeof (znode_phys_t)) {
2023 sa_buf_rele(*db, tag);
2024 return (SET_ERROR(ENOTSUP));
2025 }
2026
2027 error = sa_handle_get(osp, obj, NULL, SA_HDL_PRIVATE, hdlp);
2028 if (error != 0) {
2029 sa_buf_rele(*db, tag);
2030 return (error);
2031 }
2032
2033 return (0);
2034 }
2035
2036 void
zfs_release_sa_handle(sa_handle_t * hdl,dmu_buf_t * db,void * tag)2037 zfs_release_sa_handle(sa_handle_t *hdl, dmu_buf_t *db, void *tag)
2038 {
2039 sa_handle_destroy(hdl);
2040 sa_buf_rele(db, tag);
2041 }
2042
2043 /*
2044 * Given an object number, return its parent object number and whether
2045 * or not the object is an extended attribute directory.
2046 */
2047 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)2048 zfs_obj_to_pobj(objset_t *osp, sa_handle_t *hdl, sa_attr_type_t *sa_table,
2049 uint64_t *pobjp, int *is_xattrdir)
2050 {
2051 uint64_t parent;
2052 uint64_t pflags;
2053 uint64_t mode;
2054 uint64_t parent_mode;
2055 sa_bulk_attr_t bulk[3];
2056 sa_handle_t *sa_hdl;
2057 dmu_buf_t *sa_db;
2058 int count = 0;
2059 int error;
2060
2061 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_PARENT], NULL,
2062 &parent, sizeof (parent));
2063 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_FLAGS], NULL,
2064 &pflags, sizeof (pflags));
2065 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL,
2066 &mode, sizeof (mode));
2067
2068 if ((error = sa_bulk_lookup(hdl, bulk, count)) != 0)
2069 return (error);
2070
2071 /*
2072 * When a link is removed its parent pointer is not changed and will
2073 * be invalid. There are two cases where a link is removed but the
2074 * file stays around, when it goes to the delete queue and when there
2075 * are additional links.
2076 */
2077 error = zfs_grab_sa_handle(osp, parent, &sa_hdl, &sa_db, FTAG);
2078 if (error != 0)
2079 return (error);
2080
2081 error = sa_lookup(sa_hdl, ZPL_MODE, &parent_mode, sizeof (parent_mode));
2082 zfs_release_sa_handle(sa_hdl, sa_db, FTAG);
2083 if (error != 0)
2084 return (error);
2085
2086 *is_xattrdir = ((pflags & ZFS_XATTR) != 0) && S_ISDIR(mode);
2087
2088 /*
2089 * Extended attributes can be applied to files, directories, etc.
2090 * Otherwise the parent must be a directory.
2091 */
2092 if (!*is_xattrdir && !S_ISDIR(parent_mode))
2093 return (SET_ERROR(EINVAL));
2094
2095 *pobjp = parent;
2096
2097 return (0);
2098 }
2099
2100 /*
2101 * Given an object number, return some zpl level statistics
2102 */
2103 static int
zfs_obj_to_stats_impl(sa_handle_t * hdl,sa_attr_type_t * sa_table,zfs_stat_t * sb)2104 zfs_obj_to_stats_impl(sa_handle_t *hdl, sa_attr_type_t *sa_table,
2105 zfs_stat_t *sb)
2106 {
2107 sa_bulk_attr_t bulk[4];
2108 int count = 0;
2109
2110 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL,
2111 &sb->zs_mode, sizeof (sb->zs_mode));
2112 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_GEN], NULL,
2113 &sb->zs_gen, sizeof (sb->zs_gen));
2114 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_LINKS], NULL,
2115 &sb->zs_links, sizeof (sb->zs_links));
2116 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_CTIME], NULL,
2117 &sb->zs_ctime, sizeof (sb->zs_ctime));
2118
2119 return (sa_bulk_lookup(hdl, bulk, count));
2120 }
2121
2122 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)2123 zfs_obj_to_path_impl(objset_t *osp, uint64_t obj, sa_handle_t *hdl,
2124 sa_attr_type_t *sa_table, char *buf, int len)
2125 {
2126 sa_handle_t *sa_hdl;
2127 sa_handle_t *prevhdl = NULL;
2128 dmu_buf_t *prevdb = NULL;
2129 dmu_buf_t *sa_db = NULL;
2130 char *path = buf + len - 1;
2131 int error;
2132
2133 *path = '\0';
2134 sa_hdl = hdl;
2135
2136 uint64_t deleteq_obj;
2137 VERIFY0(zap_lookup(osp, MASTER_NODE_OBJ,
2138 ZFS_UNLINKED_SET, sizeof (uint64_t), 1, &deleteq_obj));
2139 error = zap_lookup_int(osp, deleteq_obj, obj);
2140 if (error == 0) {
2141 return (ESTALE);
2142 } else if (error != ENOENT) {
2143 return (error);
2144 }
2145 error = 0;
2146
2147 for (;;) {
2148 uint64_t pobj;
2149 char component[MAXNAMELEN + 2];
2150 size_t complen;
2151 int is_xattrdir;
2152
2153 if (prevdb)
2154 zfs_release_sa_handle(prevhdl, prevdb, FTAG);
2155
2156 if ((error = zfs_obj_to_pobj(osp, sa_hdl, sa_table, &pobj,
2157 &is_xattrdir)) != 0)
2158 break;
2159
2160 if (pobj == obj) {
2161 if (path[0] != '/')
2162 *--path = '/';
2163 break;
2164 }
2165
2166 component[0] = '/';
2167 if (is_xattrdir) {
2168 (void) sprintf(component + 1, "<xattrdir>");
2169 } else {
2170 error = zap_value_search(osp, pobj, obj,
2171 ZFS_DIRENT_OBJ(-1ULL), component + 1);
2172 if (error != 0)
2173 break;
2174 }
2175
2176 complen = strlen(component);
2177 path -= complen;
2178 ASSERT(path >= buf);
2179 bcopy(component, path, complen);
2180 obj = pobj;
2181
2182 if (sa_hdl != hdl) {
2183 prevhdl = sa_hdl;
2184 prevdb = sa_db;
2185 }
2186 error = zfs_grab_sa_handle(osp, obj, &sa_hdl, &sa_db, FTAG);
2187 if (error != 0) {
2188 sa_hdl = prevhdl;
2189 sa_db = prevdb;
2190 break;
2191 }
2192 }
2193
2194 if (sa_hdl != NULL && sa_hdl != hdl) {
2195 ASSERT(sa_db != NULL);
2196 zfs_release_sa_handle(sa_hdl, sa_db, FTAG);
2197 }
2198
2199 if (error == 0)
2200 (void) memmove(buf, path, buf + len - path);
2201
2202 return (error);
2203 }
2204
2205 int
zfs_obj_to_path(objset_t * osp,uint64_t obj,char * buf,int len)2206 zfs_obj_to_path(objset_t *osp, uint64_t obj, char *buf, int len)
2207 {
2208 sa_attr_type_t *sa_table;
2209 sa_handle_t *hdl;
2210 dmu_buf_t *db;
2211 int error;
2212
2213 error = zfs_sa_setup(osp, &sa_table);
2214 if (error != 0)
2215 return (error);
2216
2217 error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG);
2218 if (error != 0)
2219 return (error);
2220
2221 error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len);
2222
2223 zfs_release_sa_handle(hdl, db, FTAG);
2224 return (error);
2225 }
2226
2227 int
zfs_obj_to_stats(objset_t * osp,uint64_t obj,zfs_stat_t * sb,char * buf,int len)2228 zfs_obj_to_stats(objset_t *osp, uint64_t obj, zfs_stat_t *sb,
2229 char *buf, int len)
2230 {
2231 char *path = buf + len - 1;
2232 sa_attr_type_t *sa_table;
2233 sa_handle_t *hdl;
2234 dmu_buf_t *db;
2235 int error;
2236
2237 *path = '\0';
2238
2239 error = zfs_sa_setup(osp, &sa_table);
2240 if (error != 0)
2241 return (error);
2242
2243 error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG);
2244 if (error != 0)
2245 return (error);
2246
2247 error = zfs_obj_to_stats_impl(hdl, sa_table, sb);
2248 if (error != 0) {
2249 zfs_release_sa_handle(hdl, db, FTAG);
2250 return (error);
2251 }
2252
2253 error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len);
2254
2255 zfs_release_sa_handle(hdl, db, FTAG);
2256 return (error);
2257 }
2258
2259 #ifdef _KERNEL
2260 int
zfs_znode_parent_and_name(znode_t * zp,znode_t ** dzpp,char * buf)2261 zfs_znode_parent_and_name(znode_t *zp, znode_t **dzpp, char *buf)
2262 {
2263 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2264 uint64_t parent;
2265 int is_xattrdir;
2266 int err;
2267
2268 /* Extended attributes should not be visible as regular files. */
2269 if ((zp->z_pflags & ZFS_XATTR) != 0)
2270 return (SET_ERROR(EINVAL));
2271
2272 err = zfs_obj_to_pobj(zfsvfs->z_os, zp->z_sa_hdl, zfsvfs->z_attr_table,
2273 &parent, &is_xattrdir);
2274 if (err != 0)
2275 return (err);
2276 ASSERT0(is_xattrdir);
2277
2278 /* No name as this is a root object. */
2279 if (parent == zp->z_id)
2280 return (SET_ERROR(EINVAL));
2281
2282 err = zap_value_search(zfsvfs->z_os, parent, zp->z_id,
2283 ZFS_DIRENT_OBJ(-1ULL), buf);
2284 if (err != 0)
2285 return (err);
2286 err = zfs_zget(zfsvfs, parent, dzpp);
2287 return (err);
2288 }
2289 #endif /* _KERNEL */
2290